Datasets:

Hendrik-a

/

arxiv

like 0

the number field sieve is the most efficient known algorithm for factoring large integers that are free of small prime factors . for the polynomial selection stage of the algorithm , montgomery proposed a method of generating polynomials which relies on the construction of small modular geometric progressions . montgomery s method is analysed in this paper and the existence of suitable geometric progressions is considered .

1412.6011

in this paper , @xmath0 denotes a positive integer that is destined to be factored . when @xmath0 is large and free of small factors , the most efficient publicly known algorithm for determining its factors is the number field sieve @xcite . such @xmath0 include rsa @xcite moduli , for which numerous record factorisations have been achieved with the number field sieve , including the current 768-bit record @xcite . the number field sieve is comprised of several stages , commonly referred to as polynomial selection , sieving , filtering , linear algebra and square root computation . the polynomial selection stage requires the selection of coprime irreducible polynomials @xmath1 $ ] that have a common root modulo @xmath0 . after polynomial selection , sieving is used to identify coprime integer pairs @xmath2 such that the prime factors of @xmath3 are below some bound @xmath4 for @xmath5 . obtaining sufficiently many pairs with this property , called _ relations _ , is the most time consuming stage of the number field sieve , with the time taken greatly influenced by the choice of polynomials @xcite . let @xmath6 denote the number of positive integers less than @xmath7 that are free of prime factors greater than @xmath8 . canfield , erds and pomerance @xcite showed that for any @xmath9 , @xmath10 for @xmath11 , uniformly in the region @xmath12 . it follows , heuristically , that in the polynomial selection stage of the number field sieve , the polynomials @xmath13 and @xmath14 should be chosen to minimise the size of the values @xmath15 and @xmath16 over the pairs @xmath2 considered in the sieve stage . thus , it is necessary for the polynomials to have small coefficients . as a result , the degrees of @xmath13 and @xmath14 should not be too small . however , the degrees should not be too large either , since @xmath3 is a homogeneous polynomial of degree @xmath17 in @xmath18 and @xmath19 . in practice , low - degree polynomials are used . for example , the two largest factorisations of rsa moduli @xcite both used a sextic polynomial together with a linear polynomial . to quantify the coefficient size of a polynomial , the skewed @xmath20-norm @xmath21 is used . the norm is defined as follows : if @xmath22 is a degree @xmath23 polynomial with real coefficients , then @xmath24 the parameter @xmath25 captures the shape of the sieve region , which is modelled by a rectangular region @xmath26\times(0,b]$ ] or an elliptic region @xmath27 such that @xmath28 . in practice , the polynomial selection stage proceeds by first generating many `` raw '' polynomial pairs with small coefficients . then various methods of optimisation @xcite are used to improve the quality of the raw pairs by taking into account additional factors that influence a pair s yield of relations , such as the presence of real roots and roots modulo small primes @xcite . the methods of polynomial selection used in all recent record factorisations @xcite produce polynomials @xmath13 and @xmath14 such that one polynomial is linear . however , it is expected that a significant advantage is gained by using two nonlinear polynomials ( * ? ? ? * section 6.2.7 ) ( see also ( * ? ? ? * section 4 ) for practical considerations relating to sieving ) . montgomery @xcite provided a method for generating two nonlinear polynomials with small coefficients . this paper extends and sharpens montgomery s original analysis of the method . a _ geometric progression _ of _ length _ @xmath29 and _ ratio _ @xmath30 modulo @xmath0 is an integer vector @xmath31 $ ] such that @xmath32 for @xmath33 . square brackets are used to distinguish geometric progressions from regular vectors , which are denoted with round brackets . montgomery @xcite showed that a length @xmath34 geometric progression @xmath35 $ ] modulo @xmath0 such that @xmath36 and @xmath37 has full rank can be used to construct polynomials @xmath1 $ ] of maximum degree @xmath23 that have a common root modulo @xmath0 . once a suitable geometric progression @xmath38 has been found , montgomery s method proceeds by computing a basis @xmath39 , where @xmath40 denotes the transpose of a matrix @xmath41 , for the free @xmath42-module that is the set of integer vectors in the kernel of the matrix @xmath43 the basis yields polynomials @xmath44 and @xmath45 . if @xmath30 is the ratio of the geometric progression modulo @xmath0 , then @xmath46 is a linear combination of the row vectors of @xmath47 modulo @xmath0 since @xmath36 . thus , @xmath30 is a root of @xmath13 and @xmath14 modulo @xmath0 . denote by @xmath48 the submatrix of @xmath47 obtained by deleting its first column . then @xmath48 has full rank since it is equal to the submatrix of @xmath49 obtained by deleting its first row . consequently , the kernel of @xmath47 is @xmath20-dimensional . moreover , and at least one of @xmath13 and @xmath14 has degree equal to @xmath23 , otherwise @xmath50 and @xmath51 are linearly independent and in the @xmath52-dimensional kernel of @xmath48 , which is absurd . finally , as was observed by montgomery @xcite and which is shown to hold in this paper , the polynomials @xmath13 and @xmath14 are coprime since @xmath49 is nonsingular . to ensure that the norms @xmath53 and @xmath54 are small for some @xmath55 , the basis is chosen such that @xmath56 is lagrange - reduced ( see @xcite ) . as a result , @xmath57 ( see ( * ? ? ? * section 3.2 ) ) , where @xmath58 is hermite s constant for dimension two , and the vector norm @xmath21 is defined as follows : if @xmath59 is a real @xmath60-dimensional vector , then @xmath61 consequently , it is a requirement of montgomery s method that @xmath62 is small . therein lies the difficulty of the method , as the basis is readily computed in polynomial time ( see ( * ? ? ? * section 3.1.2 ) ) . the problem of constructing small geometric progressions has been addressed by several authors @xcite . it is natural to consider the existence of small geometric progressions . montgomery @xcite showed that if there exist two degree @xmath23 polynomials that have small coefficients and a common root @xmath30 modulo @xmath0 , then there exists a small length @xmath34 geometric progression with ratio @xmath30 modulo @xmath0 . montgomery s proof is constructive and is generalised to two polynomials @xmath13 and @xmath14 of maximum degree @xmath23 in this paper . furthermore , it is shown that if the polynomials are coprime , then the geometric progression given by the construction is the unique vector @xmath63 , up to scalar multiple , such that the coefficient vectors of @xmath13 and @xmath14 are in the kernel of the matrix @xmath47 defined in . as a result , the analysis of the construction contributes to the analysis of montgomery s method . this paper is organised as follows : the definitions and some properties of the sylvester matrix , the bezout matrix and the resultant are reviewed in the next section ; the generalisation of montgomery s geometric progression construction is presented and analysed in section [ sec : existence ] ; and a full analysis of montgomery s method is provided in section [ sec : analysis ] . matrices with the property that each of their rows contain the coefficients of some polynomial are frequently encountered in this paper . the sylvester and bezout matrices are constructed in this manner . consequently , compact notation for such matrices is defined before introducing the protagonists of this section . for @xmath64 polynomials @xmath65 and any integer @xmath66 , denote by @xmath67 the @xmath68 matrix @xmath69 where @xmath70 is the coefficient of @xmath71 in @xmath72 . when @xmath73 , the subscript @xmath74 is drop , giving the notation @xmath75 . the parameter @xmath74 is viewed as the formal degree of the polynomials @xmath65 . for example , @xmath76 is the vector of coefficients of @xmath77 , while @xmath78 for some @xmath79 is the vector of coefficients of @xmath77 when view as a polynomial of formal degree @xmath74 . define @xmath80 to be the @xmath81 empty matrix . let @xmath82 be an integral domain . then the _ sylvester matrix _ of non - constant polynomials @xmath83 $ ] is the matrix @xmath84 the determinant of @xmath85 is called the _ resultant _ of @xmath13 and @xmath14 , and is denoted @xmath86 . the resultant of @xmath13 and @xmath14 is zero if and only if the polynomials have a nontrivial gcd over the field of fractions of @xmath82 . for non - constant polynomials @xmath1 $ ] and all real numbers @xmath55 , define @xmath87 to be the angle ( in @xmath88 $ ] ) between the row vectors of @xmath89 . the following lemma provides upper and lower bounds on the resultant of a pair of number field sieve polynomials ( see ( * ? ? ? * section 2.1.2 ) for a proof ) : [ lem : resultantbound ] suppose that @xmath1 $ ] are non - constant , coprime and have a common root modulo @xmath0 . then @xmath90 for all @xmath55 . as a consequence of lemma [ lem : resultantbound ] , a pair of number field sieve polynomials in considered to have optimal resultant if it is equal to @xmath0 , and optimal coefficient size if @xmath91 is @xmath92 . thus , inequality implies that a pair of degree @xmath23 polynomials generated by montgomery s method has optimal coefficient size if @xmath93 . let @xmath82 be an integral domain and @xmath83 $ ] be non - constant . write @xmath94 for @xmath5 such that @xmath95 and the coefficients @xmath70 are elements of @xmath82 . define polynomials @xmath96 then the _ bezout matrix _ , or _ bezoutian _ , of @xmath13 and @xmath14 is the matrix @xmath97 . denote by @xmath98 the leading coefficient of a polynomial @xmath77 . similar to the sylvester matrix , the determinant of the bezout matrix of @xmath13 and @xmath14 is related to their resultant ( see ( * ? ? ? * section 2 ) ) : @xmath99 a _ hankel matrix _ over @xmath82 is a square matrix @xmath100 such that @xmath101 for some @xmath102 . lander @xcite showed that the inverse of a bezout matrix is a hankel matrix and , conversely , that the inverse of a hankel matrix is the bezout matrix of two polynomials . for an @xmath103 matrix @xmath104 such that @xmath101 for some @xmath105 , define @xmath106 for @xmath107 . let @xmath108 denote the matrix @xmath109 . define @xmath110 . the following result of heinig and rost ( * ? ? ? * theorem 4.2 ) expresses the inverse of a real hankel matrix @xmath111 as the bezout matrix of two polynomials obtained from the kernel of @xmath108 : [ lem : hankelinverse ] let @xmath112 be a real nonsingular hankel matrix . if @xmath113 $ ] such that @xmath114 is a basis of the kernel of @xmath108 , then @xmath115 the formulation of the bezout matrix presented in this paper is different to that used by heinig and rost . thus , it is necessary to convert between the two formulations : let @xmath112 be a real nonsingular hankel matrix . suppose that @xmath113 $ ] such that @xmath114 is a basis of the kernel of @xmath108 . define @xmath116 as in the statement of the theorem . the matrix @xmath117 is equal to the submatrix of @xmath111 obtained by deleting its first row . thus , @xmath117 has full rank . it follows that @xmath118 , otherwise @xmath119 and @xmath120 would be linearly independent and in the kernel of @xmath117 . interchanging @xmath13 and @xmath14 changes @xmath121 and @xmath122 by a factor of @xmath123 . therefore , assume without loss of generality that @xmath124 . write @xmath94 for @xmath5 such that the coefficients @xmath70 are real numbers . define @xmath125 for @xmath5 . then theorem 4.2 of heinig and rost @xcite implies that @xmath126 where the matrix @xmath127 , called the ( hankel ) bezoutian of @xmath128 and @xmath129 by heinig and rost ( * ? ? ? * section 2.1 ) ( and many other authors ) , is defined as follows : @xmath130 such that @xmath131 is the coefficient of @xmath132 in the polynomial @xmath133 expanding the numerator of @xmath134 shows that @xmath135 therefore , @xmath136 where the omitted entries are zeros . let @xmath137 such that @xmath138 , and @xmath139 $ ] such that @xmath140 . then @xmath141 and @xmath142 hence , @xmath143 combining this equation with completes the proof . the _ integer kernel _ of an integer matrix @xmath41 is the free @xmath42-module consisting of all integer column vectors @xmath144 such that @xmath145 . montgomery @xcite showed that if two coprime degree @xmath146 integer polynomials @xmath13 and @xmath14 have a common root modulo @xmath0 , then a vector @xmath147 such that @xmath148 spans the integer kernel of the matrix @xmath149 is a geometric progression modulo @xmath0 . moreover , @xmath150 since the entries of @xmath38 are , up to a constant , order @xmath151 minors of the matrix in . koo , jo and kwon ( * ? ? ? * theorem 2 ) generalise montgomery s result by providing a construction which , for @xmath152 , uses @xmath153 degree @xmath23 polynomials @xmath154 with a common root modulo @xmath0 to construct a length @xmath155 geometric progression @xmath38 such that @xmath156 . in this section , another generalisation is presented , with two polynomials @xmath13 and @xmath14 of maximum degree @xmath23 used to construct geometric progressions of lengths @xmath157 . the geometric progressions are shown to have small size whenever @xmath13 and @xmath14 have small coefficients . therefore , the existence of small geometric progressions for montgomery s method , and relaxations of the method that employ shorter progressions @xcite , is established under the assumption that good nonlinear polynomial pairs exist . let @xmath82 be an integral domain . for @xmath83 $ ] such that @xmath158 , and @xmath159 , define the @xmath160 matrix @xmath161 define signed minors @xmath162 of the matrix @xmath163 as follows : for @xmath164 , @xmath165 is equal to @xmath166 times the determinant of the submatrix of @xmath163 obtained by deleting its @xmath167th column . define @xmath168 for @xmath169 . when @xmath13 and @xmath14 are clear from the context , @xmath170 is used to denote @xmath163 , @xmath171 is used to denote @xmath165 , and @xmath172 is used to denote @xmath173 . the @xmath167th entry of the vector @xmath174 is equal to the determinant of the matrix obtained by appending the @xmath167th row of @xmath163 to its top , and thus is equal to zero . therefore , @xmath175 where @xmath176 denotes the @xmath74-dimensional column vector of zeros for @xmath177 . if @xmath41 is an @xmath103 integer matrix such that @xmath178 , define @xmath179 to be the greatest common divisor of all @xmath180 minors of @xmath41 , with @xmath181 if all such minors are zero . then @xmath179 is nonzero if and only if @xmath41 has full rank . let @xmath1 $ ] be coprime degree @xmath146 polynomials that have a common root modulo @xmath0 . then @xmath182 is the matrix in . therefore , @xmath182 is the submatrix of @xmath85 obtained by first deleting its first and @xmath183th rows , giving a matrix whose first column contains zeros , then deleting the first column of the resulting matrix . thus , @xmath182 has full rank , otherwise @xmath85 is singular . therefore , @xmath184 is nonzero and in the integer kernel of @xmath182 . consequently , montgomery s result implies @xmath185 is a geometric progression modulo @xmath0 . more generally , if @xmath158 , then the vectors @xmath173 for @xmath169 are geometric progressions modulo @xmath0 : [ thm : polys - to - gp ] let @xmath13 and @xmath14 be coprime integer polynomials such that @xmath158 , @xmath13 and @xmath14 have a common root @xmath186 modulo @xmath0 , and @xmath187 is relatively prime to @xmath0 . then , for @xmath159 , the vector @xmath188 satisfies the following properties : 1 . [ isgp ] @xmath172 is a nonzero geometric progression with ratio @xmath30 modulo @xmath0 ; 2 . [ gcdcondition ] @xmath189 ; 3 . [ rankcondition ] the matrix @xmath190 has full rank ; 4 . [ kernelcondition ] the vectors @xmath191 and @xmath192 are in the kernel of @xmath193 ; and 5 . [ gpsizebounds ] the inequalities @xmath194 and @xmath195 hold for all @xmath55 . theorem [ thm : polys - to - gp ] is proved as a series of lemmas and corollaries in the next section , with some of the properties presented as part of more general statements . the requirement in the theorem that @xmath196 and @xmath0 are relatively prime has not been encountered so far in this paper , nor is it usual to require a pair of nonlinear number field sieve polynomials to satisfy it . in the next section , it is shown that @xmath197 ( see lemma [ lem : ct - upper - bnd ] and remark [ rmk : ct - upper - bnd ] ) . therefore , if @xmath13 and @xmath14 are a pair of number field sieve polynomials that are close the attaining the lower bound from lemma [ lem : resultantbound ] , then @xmath198 is expected to equal one in practice , or a factorisation of @xmath0 is possibly obtained . however , as the requirement is new , its meaning is discussed before proceeding to the proof of the theorem . let @xmath82 be an integral domain , @xmath83 $ ] such that @xmath158 , and @xmath199 for @xmath5 . then the _ first subresultant _ of @xmath13 and @xmath14 is the polynomial @xmath200 recall that @xmath201 if and only if @xmath202 , where the gcd is computed over the field of fractions of @xmath82 . this statement is refined by considering the first subresultant of the polynomials : if @xmath201 , then @xmath203 if and only if @xmath204 ( see , for instance , ( * ? ? ? * section 7.7.1 ) or @xcite ) . suppose now that @xmath13 and @xmath14 are coprime integer polynomials . write @xmath205 and @xmath206 such that the coefficients @xmath70 are integers . then using the laplace expansion ( see ( * ? ? ? * section 33 ) ) to compute the determinant of @xmath85 with respect to rows @xmath207 and @xmath208 shows that @xmath209 thus , @xmath210 divides @xmath86 and @xmath211 . therefore , if @xmath212 is a prime that divides @xmath210 and does not divide @xmath213 , then the reductions of @xmath13 and @xmath14 modulo @xmath212 have a common factor of degree greater than one . conversely , if the reductions of @xmath13 and @xmath14 modulo a prime @xmath212 are nonzero and have a common divisor of degree @xmath214 , then gomez et al . @xcite showed that @xmath215 divides @xmath86 . modifying their proof shows that @xmath216 divides the minors @xmath217 , and thus @xmath216 divides @xmath210 . hence , if @xmath212 is a prime that does not divide @xmath213 , then @xmath212 divides @xmath210 if and only if the reductions of @xmath13 and @xmath14 modulo @xmath212 have a common factor of degree greater than one . if @xmath13 and @xmath14 are a pair of number field sieve polynomials , then @xmath218 is expected to equal one in practice . therefore , the requirement that @xmath219 denies the possibility that @xmath13 and @xmath14 have a factor of degree greater than one modulo some factor of @xmath0 . in this section , @xmath13 and @xmath14 are integer polynomials such that @xmath158 . furthermore , let @xmath220 , @xmath221 for @xmath222 . the first lemma of this section proves property and property of theorem [ thm : polys - to - gp ] : [ lem : polys - to - gp ] if @xmath219 and there exists an integer @xmath30 that is a root of @xmath13 and @xmath14 modulo @xmath0 , then @xmath172 is a nonzero geometric progression with ratio @xmath30 modulo @xmath0 and @xmath223 for @xmath222 . suppose that @xmath219 and @xmath186 is a root of @xmath13 and @xmath14 modulo @xmath0 . let @xmath224 . the lemma is proved in two steps : first , it is shown that @xmath225 is a geometric progression with ratio @xmath30 modulo @xmath0 ; and second , it is shown that if @xmath226 and @xmath227 is a geometric progression with ratio @xmath30 modulo @xmath0 for some @xmath228 , then so too is @xmath172 . let @xmath229 be a @xmath230 unimodular matrix such that @xmath231 is in hermite normal form ( as defined by cohen ( * ? ? ? * definition 2.4.2 ) ) . performing elementary row operations on the columns of @xmath232 does not change @xmath233 , which is nonzero since @xmath234 . thus , @xmath235 for some @xmath236 matrix @xmath111 such that @xmath237 . the first column vector of @xmath229 is a basis for the integer kernel of @xmath232 ( see ( * ? ? ? * proposition 2.4.9 ) ) . thus , implies that @xmath238 is equal to @xmath239 times the first column vector of @xmath229 . the definition of @xmath225 implies that @xmath240 . therefore , @xmath225 is nonzero and @xmath241 . let @xmath242 and @xmath243 . then @xmath244 since @xmath30 is a root of @xmath13 and @xmath14 modulo @xmath0 . consequently , @xmath245 therefore , @xmath246 since @xmath111 is invertible modulo @xmath0 . it follows that @xmath247 since @xmath248 . hence , @xmath249 multiplying both sides of this equation on the right by @xmath250 shows that @xmath225 is a geometric progression with ratio @xmath30 modulo @xmath0 and @xmath251 . suppose that @xmath226 and @xmath227 for some @xmath228 is a nonzero geometric progression with ratio @xmath30 modulo @xmath0 and @xmath252 . let @xmath44 such that @xmath253 . then @xmath254 \vdots & \multicolumn{6}{c}{\mathrm{s}_{t-1}(f_{1},f_{2})}\\ 0 & \multicolumn{6}{c } { } \end{array}\right).\ ] ] therefore , deleting the @xmath167th column of @xmath163 and computing the determinant of the resulting matrix along its first row shows that @xmath255 it follows that @xmath172 is nonzero since @xmath256 and @xmath227 are nonzero . furthermore , @xmath257 and thus @xmath258 . by assumption , @xmath227 is a geometric progression with ratio @xmath30 modulo @xmath0 . thus , @xmath259 for @xmath260 . therefore , implies that @xmath261 for @xmath262 . the first entry of @xmath263 is equal to @xmath264 . consequently , implies that @xmath265 . thus , implies that @xmath266 hence , @xmath172 is a geometric progression with ratio @xmath30 modulo @xmath0 . define the _ volume _ of a real matrix @xmath41 , denoted @xmath267 , to be @xmath268 . for a matrix @xmath41 over a commutative ring , define @xmath269 . if @xmath41 is the @xmath81 empty matrix , then @xmath270 and @xmath271 . if @xmath41 is an @xmath103 matrix such that @xmath178 , then @xmath267 is nonzero if and only if @xmath41 has full rank . the volume function is multiplicative in the following sense : if @xmath41 is an @xmath180 matrix and @xmath272 is an @xmath103 matrix , then @xmath273 and @xmath274 . the following lemma provides the upper bound on @xmath275 in theorem [ thm : polys - to - gp ] . in particular , the lemma shows that if @xmath276 for some @xmath55 , i.e. , the polynomials are close the attaining the lower bound from lemma [ lem : resultantbound ] , then @xmath277 . [ lem : ct - upper - bnd ] the inequality @xmath278 holds for @xmath222 and all @xmath55 . if @xmath279 does not have full rank , then the recurrence relation implies that @xmath172 is the zero vector for @xmath222 . thus , if @xmath279 does not have full rank , then the lemma holds trivially . therefore , assume that @xmath279 has full rank . then the recurrence relation implies that @xmath163 has full rank for @xmath222 . let @xmath280 , @xmath25 be a positive real number and @xmath281 then the binet cauchy formula ( see ( * ? ? ? * section 36 ) ) implies that @xmath282 to complete the proof , fischer s inequality @xcite is used to derive an upper bound on @xmath283 in a manner similar to the proof of ( * ? ? ? * lemma 2.2 ) . define matrices @xmath284 as follows : @xmath285 is the @xmath286 empty matrix if @xmath287 ; @xmath288 if @xmath289 ; and @xmath290 for @xmath291 . for @xmath292 , let @xmath293 be the @xmath294 matrix obtained by arranging the matrices @xmath295 consecutively beneath each other . then @xmath296 is obtained from @xmath170 by permuting its rows . thus , @xmath297 . moreover , as @xmath163 has full rank and @xmath298 , @xmath299 and @xmath300 are positive definite hermitian for @xmath292 . therefore , fischer s inequality implies that @xmath301 for @xmath291 . hence , @xmath302 for @xmath303 , @xmath304 thus , if @xmath289 , then hadamard s determinant theorem @xcite implies that @xmath305 this inequality also holds trivially if @xmath287 . the angle between the row vectors of @xmath306 is @xmath87 and @xmath307 for @xmath291 . therefore , by computing @xmath308 or by viewing @xmath309 as the area of the parallelogram generated by the row vectors of @xmath306 , it follows that @xmath310 combining this equation with and yields the inequality @xmath311 combining this inequality with ( recalling that @xmath312 ) and computing roots yields . hence , as @xmath313 and @xmath25 were chosen arbitrarily , it follows that holds for @xmath222 and all @xmath55 . [ rmk : ct - upper - bnd ] the last equality of holds for @xmath222 and all @xmath55 . thus , the inequalities @xmath314 hold for @xmath222 and all @xmath55 . property of theorem [ thm : polys - to - gp ] is now proved . combining this property with property of theorem shows that the coefficient vectors of each pair of nonlinear number field sieve polynomials appear in the kernel of a matrix of the form for some nonzero length @xmath34 geometric progression @xmath315 $ ] , where @xmath23 is the maximum degree of the polynomials . [ lem : kernelcondition ] the vectors @xmath316 and @xmath317 belong to the kernel of @xmath193 for @xmath222 . the @xmath167th entry of @xmath318 is equal to the @xmath167th entry of @xmath319 for @xmath320 . similarly , the @xmath167th entry of @xmath321 is equal to the @xmath322th entry of @xmath319 for @xmath320 . thus , implies that @xmath316 and @xmath317 are in the kernel of @xmath193 for @xmath222 . denote by @xmath323 the adjoint ( or adjugate ) of a square matrix @xmath41 . the following lemma , from which the remaining properties of theorem [ thm : polys - to - gp ] are deduced , shows that the adjoint of the bezout matrix has entries which , up to sign , are minors of the sylvester matrix : [ lem : adjbez ] let @xmath324 and @xmath325 such that @xmath326 and @xmath327 are algebraically independent indeterminates over @xmath42 . then @xmath328 let @xmath324 and @xmath325 such that @xmath326 and @xmath327 are algebraically independent indeterminates over @xmath42 . then the algebraic independence of the coefficients implies that @xmath329 $ ] is nonzero . therefore , equation implies that it is sufficient to show that @xmath330 where @xmath331 denotes the @xmath332 identity matrix for all integers @xmath333 . following , define @xmath334 then @xmath335 . define @xmath336 the matrices @xmath337 are square of order @xmath34 since @xmath338 for @xmath5 , and @xmath339 expanding the determinant of each matrix @xmath337 along its first row shows that @xmath340 it follows from that @xmath341 for @xmath342 . therefore , for indices @xmath167 and @xmath343 such that @xmath344 , the determinant @xmath345 is zero since the polynomial @xmath346 is a linear combination of the polynomials @xmath347 and @xmath348 . similarly , for @xmath349 , @xmath350 thus , @xmath351 for @xmath349 . consequently , @xmath352 it is now shown that @xmath353 the special case @xmath354 has been proved , up to sign , by sederberg , goldman and du ( * ? ? ? * proposition 2.3 ) . their arguments are modified to obtain . in particular , the proof proceeds by computing the determinants of the following matrices two ways : @xmath355 substituting @xmath356 into shows that @xmath357 therefore , by performing elementary row operations on @xmath358 , it follows that @xmath359 for @xmath360 . the first column vector of @xmath358 contains @xmath361 in the @xmath183th coordinate and zeros elsewhere . furthermore , the submatrix of @xmath358 obtained by deleting its first column and @xmath183th row is equal to @xmath362 . therefore , expanding the determinant of @xmath358 along its first column shows that @xmath363 hence , combining and implies that holds . then combining and implies that holds . property of theorem [ thm : polys - to - gp ] is now deduced from lemma [ lem : adjbez ] by specialising the coefficients of the generic polynomials @xmath128 and @xmath129 : [ cor : adjbez ] the adjoint of @xmath121 is @xmath364 consequently , if @xmath13 and @xmath14 are coprime , then @xmath365 has full rank for @xmath222 . let @xmath366 $ ] such that @xmath326 and @xmath327 are algebraically independent indeterminates over @xmath42 . set @xmath367 $ ] and @xmath368 $ ] . then holds . define the evaluation homomorphism @xmath369 by @xmath370 and @xmath371 for @xmath372 . extend @xmath373 entry - wise to matrices . as @xmath124 , it holds that @xmath374 and @xmath375 for @xmath376 . therefore , the @xmath373-image of the each side of is equal to its respective side of . suppose now that @xmath13 and @xmath14 are coprime . then implies that @xmath121 is nonsingular . thus , implies that @xmath377 is nonsingular . if @xmath378 such that @xmath379 , then the recurrence relation implies that the submatrix of @xmath377 formed by its last @xmath313 rows is equal to @xmath380 . as @xmath381 is nonzero , it follows that @xmath365 has full rank for @xmath222 . all that remains in the proof of theorem [ thm : polys - to - gp ] is to establish the lower bound on @xmath275 stated in property of the theorem . the remainder of this section is dedicated to the proof of this property , which proceeds as follows : first , the volume of @xmath382 , where @xmath383 is an arbitrary nonsingular matrix , is computed ; then , for an appropriate choice of @xmath383 , the volume of @xmath382 is bounded above by a power of @xmath275 , providing a lower bound on @xmath275 ; and finally , by examining a special case of this bound , the lower bound stated in property is proved . let @xmath41 be an @xmath103 matrix . for all subsets @xmath384 and @xmath385 , define @xmath386 to be the @xmath387 submatrix of @xmath41 formed by the intersection of the rows that have indices in @xmath388 with the columns that have indices in @xmath389 . if @xmath390 , and @xmath391 and @xmath392 are partitions of @xmath393 such that @xmath394 , then jacobi ( see ( * ? ? ? * section 42 ) or @xcite ) showed that @xmath395 the following technical lemma is proved by repeatedly applying this identity : [ lem : vol - adjoint ] suppose that @xmath41 and @xmath383 are @xmath332 matrices such that @xmath396 and @xmath383 is invertible . then , for each partition @xmath391 of @xmath393 , @xmath397 suppose that @xmath41 and @xmath383 are @xmath332 matrices such that @xmath396 and @xmath383 is invertible . let @xmath391 be a partition of @xmath393 . set @xmath398 , @xmath399 and @xmath400 . then the binet cauchy formula implies that @xmath401 using to compute each summand on the right hand side shows that @xmath402 using the binet cauchy formula to compute the sum on the right hand side yields @xmath403 if @xmath404 and @xmath405 are @xmath332 matrices , then @xmath406 for all @xmath407 . it follows that @xmath408 and @xmath409 substituting these values and @xmath410 into completes the proof . [ lem : gp - vol ] let @xmath411 such that @xmath379 and @xmath412 , and @xmath383 be a real nonsingular @xmath413 matrix . then @xmath414 let @xmath411 such that @xmath379 and @xmath412 , and @xmath383 be a real nonsingular @xmath413 matrix . let @xmath415 $ ] where @xmath326 and @xmath327 are algebraically independent indeterminates over @xmath416 . define @xmath324 , @xmath325 and @xmath417 $ ] by . then @xmath335 . define a @xmath418 matrix @xmath419 and a @xmath413 matrix @xmath272 as follows : @xmath420 where @xmath421 denotes the @xmath103 matrix of zeros for all integers @xmath422 . the upper @xmath423 submatrix of @xmath424 is lower triangular , with each entry on its diagonal equal to @xmath425 . thus , implies that @xmath426 , which is nonzero since @xmath326 and @xmath327 are algebraically independent over @xmath416 . if @xmath427 , then implies that @xmath428 for @xmath429 . consequently , lemma [ lem : adjbez ] implies that @xmath430 as @xmath272 is nonsingular , it follows that @xmath431 therefore , on the one hand , @xmath432 on the other hand , as @xmath433 is symmetric ( which is deduced from lemma [ lem : adjbez ] by noting that @xmath434 is symmetric ) , lemma [ lem : vol - adjoint ] implies that @xmath435 write @xmath44 and @xmath45 such that the coefficients @xmath70 are integers . define the evaluation homomorphism @xmath436 by @xmath370 and @xmath371 for @xmath372 . then @xmath437 . extend @xmath373 entry - wise to matrices and let @xmath438\rightarrow{\mathbb{r}}[x]$ ] be the natural extension of @xmath373 . then @xmath439 where the final equality follows from the recurrence relation . therefore , computing the @xmath373-images of and and equating shows that @xmath440 from the definition of @xmath441 , it follows that if @xmath442 , then @xmath443 furthermore , if @xmath444 , then @xmath445 for @xmath446 . as @xmath447 , it follows that @xmath448 substituting this equation into and taking roots yields . [ lem : ct - lower - bnd ] let @xmath411 such that @xmath379 and @xmath412 . then @xmath449 for all @xmath55 . let @xmath411 such that @xmath450 and @xmath412 . for a real number @xmath55 , define @xmath451 and @xmath452 then @xmath453 thus , lemma [ lem : gp - vol ] with @xmath454 implies that @xmath455 recall that @xmath456 and @xmath457 . thus , @xmath458 therefore , the row vectors of @xmath459 each have euclidean length bounded by @xmath275 . consequently , hadamard s determinant theorem implies that @xmath460 calculating the determinants of @xmath461 , @xmath462 and @xmath463 yields @xmath464 combining , and gives , which completes the proof since @xmath25 was chosen arbitrarily . to end the section , two corollaries to lemma [ lem : ct - lower - bnd ] are given . the first corollary establishes the lower bound on @xmath275 stated in property of theorem [ thm : polys - to - gp ] , completing the proof of the theorem . the second corollary is utilised in the next section as part of the analysis of montgomery s method . [ lem : ct - lower - bnd - k0 ] the inequality @xmath465 holds for @xmath222 and all @xmath55 . for @xmath378 such that @xmath379 and @xmath55 , the @xmath466 submatrix of @xmath467 formed by columns @xmath468 is upper triangular with @xmath469 in each diagonal entry . by applying the binet cauchy formula , it follows that @xmath470 is equal to @xmath471 plus some sum of squares . thus , for all @xmath55 , @xmath472 substituting these inequalities into for @xmath222 and @xmath473 completes the proof . [ lem : ct - lower - bnd - td - k1 ] the inequality @xmath474 holds for all @xmath55 . for all @xmath55 , the euclidean length of the first row vector of the matrix @xmath475 is @xmath476 , the euclidean length of the second row vector is @xmath477 , and the angle between the two row vectors is @xmath87 . therefore , @xmath478 substituting this equation into for @xmath287 and @xmath479 completes the proof . montgomery s method is analysed in this section , providing criteria for the selection of geometric progressions that yield polynomials with optimal coefficient size and optimal resultant . in particular , the goal of this section is to prove the following theorem , which may be viewed as a converse to theorem [ thm : polys - to - gp ] : [ thm : gp - to - polys ] let @xmath146 and @xmath35\in{\mathbb{z}}^{2d-1}$ ] be a geometric progression with ratio @xmath30 modulo @xmath0 such that @xmath480 is nonsingular and @xmath481 . then , for @xmath1 $ ] such that @xmath158 and @xmath114 is a basis for the integer kernel of @xmath47 , the following properties hold : 1 . [ degcondition ] @xmath124 ; 2 . [ hascommonroot ] @xmath30 is a root of @xmath13 and @xmath14 modulo @xmath0 ; 3 . [ resformula ] @xmath13 and @xmath14 are coprime , with @xmath482 4 . [ polysizebounds ] the inequalities @xmath483 hold for all @xmath55 . property and property of theorem [ thm : gp - to - polys ] are proved in section [ sec : montgomerys - method ] . the two remaining properties of the theorem are proved in the next section . recall from section [ sec : montgomerys - method ] that in montgomery s method , the polynomials @xmath13 and @xmath14 are chosen such that @xmath484 , where @xmath95 , is a lagrange - reduced basis for some @xmath55 . it follows that @xmath485 for the chosen value of @xmath25 ( see @xcite ) . combining the inequalities from property shows that any length @xmath34 geometric progression @xmath38 that satisfies the condition of the theorem has norm satisfying @xmath486 for all @xmath55 . thus , montgomery s method is unforgiving of a poor choice of geometric progression . in particular , the method generates to two degree @xmath23 polynomials with optimal coefficient size only if a geometric progression of almost minimal size is used . property of theorem [ thm : gp - to - polys ] may aid the selection of parameters for specific geometric progression constructions by allowing parameters to be tuned so that polynomials with resultant equal to a small multiple of @xmath0 are obtained . before completing the proof of theorem [ thm : gp - to - polys ] in the next section , property is used to compute the resultant given by two existing geometric progression constructions : [ ex : resformula ] for @xmath487 , several authors @xcite propose using length @xmath488 geometric progressions of the form @xmath489=\left[\frac{am^{2}-kn}{p},am , ap\right],\ ] ] where @xmath18 , @xmath490 , @xmath212 and @xmath491 are nonzero integers such that @xmath492 and @xmath493 . letting @xmath494 , it follows that @xmath495\right)=\gcd(a , c_{2}).\ ] ] let @xmath496 and @xmath497 , with the latter being an integer since @xmath498 and @xmath493 . property of theorem [ thm : gp - to - polys ] implies that if @xmath13 and @xmath14 are quadratic polynomials whose coefficient vectors form a basis for the integer kernel of @xmath47 , then @xmath499 and @xmath500 . for @xmath501 , koo , jo and kwon @xcite and the author @xcite propose using length @xmath502 geometric progressions of the form @xmath503=\left[\frac{m(am^3 -kn)}{p^{2}},\frac{am^{3}-kn}{p},am^{2},amp , ap^{2}\right],\ ] ] where @xmath18 , @xmath490 , @xmath212 and @xmath491 are nonzero integers such that @xmath492 and @xmath493 . letting @xmath504 , it follows that @xmath505 , @xmath506 and @xmath507)=\gcd\left(m\frac{am^3 -kn}{p^{2}},p\frac{am^{3}-kn}{p^{2}},a\right)=\gcd(a , c_{3}/p).\ ] ] let @xmath508 and @xmath509 . property of theorem [ thm : gp - to - polys ] implies that if @xmath13 and @xmath14 are cubic polynomials whose coefficient vectors form a basis for the integer kernel of @xmath47 , then @xmath510 and @xmath511 . theorem [ thm : polys - to - gp ] shows that each pair of nonlinear number field sieve polynomials with maximum degree @xmath146 appears in the kernel of the matrix @xmath512 for some geometric progression @xmath315 $ ] . the following lemma shows that such a geometric progression is unique up to scalar multiple , thus allowing results from section [ sec : proof - polys - to - gp ] to be used in the proof of theorem [ thm : gp - to - polys ] : [ lem : gp - kernel ] let @xmath13 and @xmath14 be coprime integer polynomials such that @xmath158 , and @xmath220 . then the vectors @xmath316 and @xmath317 are in the kernel of @xmath512 for some vector @xmath513 if and only if @xmath514 . let @xmath13 and @xmath14 be coprime integer polynomials such that @xmath158 , @xmath220 and @xmath515 . let @xmath513 and @xmath480 . then the @xmath167th entry of @xmath516 is equal to the @xmath167th entry of @xmath517 for @xmath518 . similarly , the @xmath167th entry of @xmath519 is equal to the @xmath322th entry of @xmath517 for @xmath518 . thus , @xmath316 and @xmath317 are in the kernel of @xmath47 if and only if @xmath148 is in the kernel of @xmath182 . as @xmath13 and @xmath14 are coprime , @xmath85 is nonsingular . the matrix @xmath279 is the submatrix of @xmath85 obtained by first deleting its first and @xmath520th rows , giving a matrix whose first column contains zeros , then deleting the first column of the resulting matrix . therefore , @xmath279 has full rank . consequently , the recurrence relation implies that @xmath182 has full rank . lemma [ lem : polys - to - gp ] and imply that @xmath521 is nonzero and belongs to the kernel of @xmath182 . thus , @xmath522 is a basis of the integer kernel of @xmath182 . hence , @xmath148 is in the integer kernel of @xmath182 if and only if @xmath523 . let @xmath396 and @xmath41 be an @xmath332 matrix . for @xmath524 such that @xmath525 and @xmath526 , define @xmath527 and @xmath528 for @xmath529 . then @xmath530 which is known as sylvester s identity ( see ( * ? ? ? * theorem 4.1 ) for a proof ) . the following lemma is obtained from lemma [ lem : hankelinverse ] by computing the determinant of the matrix @xmath116 for the case where @xmath111 and the polynomials @xmath13 and @xmath14 are integral : [ lem : adjc ] let @xmath146 and @xmath531 such that @xmath480 is nonsingular . if @xmath13 and @xmath14 are integer polynomials such that @xmath114 is a basis of the integer kernel of @xmath47 , then @xmath532 let @xmath146 and @xmath531 such that @xmath480 is nonsingular . define @xmath533 matrices @xmath534 where the omitted entries are zeros , for @xmath535 . define @xmath536 for @xmath535 . then @xmath537 is the set of @xmath538 minors of @xmath47 . the matrix @xmath47 has full rank since @xmath49 is nonsingular . therefore , there exist indices @xmath490 and @xmath29 such that @xmath539 and @xmath540 . the first and second column vectors of @xmath541 are @xmath542 and @xmath543 respectively . thus , these two vectors are nonzero and in the kernel of @xmath47 . let @xmath544 let @xmath545 and @xmath546 be indices such that @xmath547 . define @xmath548 and @xmath549 for @xmath550 . then @xmath551 similarly , @xmath552 for @xmath550 . finally , @xmath553 therefore , identity implies that @xmath554 consequently , @xmath555 , which is nonzero since @xmath47 has full rank and @xmath540 . therefore , if @xmath556 is a @xmath533 unimodular matrix such that @xmath557 is in hermite normal form , then @xmath558 for some @xmath559 integer matrix @xmath111 . performing elementary column operations on @xmath272 does not change @xmath560 . thus , @xmath561 is nonzero . hence , @xmath562 has integer entries and @xmath563 . suppose that @xmath13 and @xmath14 are integer polynomials such that @xmath114 is a basis of the integer kernel of @xmath47 . then there exists a @xmath559 nonsingular integer matrix @xmath229 such that @xmath564 therefore , lemma [ lem : hankelinverse ] implies that @xmath565 where @xmath566 equation implies that @xmath567 since @xmath556 is unimodular . furthermore , @xmath568 since @xmath114 is a basis of the integer kernel of @xmath47 : if @xmath569 were greater than one , then @xmath316 and @xmath317 would only generate a proper subgroup of the integer kernel ( * ? ? ? * section 3.1 ) . thus , implies that @xmath229 is unimodular . therefore , @xmath570 expanding the determinant of @xmath49 by minors along its last row shows that @xmath571 hence , , and imply that @xmath572 . the first assertion of property of theorem [ thm : gp - to - polys ] , that @xmath13 and @xmath14 are coprime , follows from lemma [ eqn : adjc ] since @xmath49 is nonsingular by assumption . the next step in the proof of the theorem is to prove the formulae for @xmath86 and @xmath573 . [ lem : resformula ] let @xmath146 and @xmath513 such that the matrix @xmath480 is nonsingular . if @xmath1 $ ] such that @xmath158 and @xmath114 is a basis for the integer kernel of @xmath47 , then @xmath574 let @xmath146 and @xmath513 such that @xmath480 is nonsingular . suppose that @xmath1 $ ] such that @xmath158 and @xmath114 is a basis for the integer kernel of @xmath47 . then holds and computing the determinant of both sides of the equation yields @xmath575 as @xmath49 is nonsingular , @xmath576 and @xmath577 are nonzero . thus , @xmath86 is nonzero and lemma [ lem : gp - kernel ] implies that @xmath578 . therefore , on the one hand , corollary [ cor : adjbez ] implies that @xmath579 on the other hand , computing the adjoint of both side of yields @xmath580 as @xmath49 has at least one nonzero entry , it follows that @xmath581 where the right hand side is positive . the following lemma and its subsequent corollary complete the proof of property of theorem [ thm : gp - to - polys ] by showing that @xmath582 when the degree of @xmath14 is strictly less than @xmath23 : [ lem : basis_kerparc ] let @xmath146 , @xmath531 such that @xmath480 is nonsingular , and integers @xmath131 , for @xmath535 , be defined as in the proof of lemma [ lem : adjc ] . then there exist integers @xmath583 such that @xmath584 and , for any such integers , the set @xmath585 is a basis of the integer kernel of @xmath47 . let @xmath531 such that @xmath480 is nonsingular then @xmath586 is nonzero since @xmath49 is nonsingular . define @xmath131 as in the proof of lemma [ lem : adjc ] for @xmath535 . for distinct indices @xmath167 and @xmath343 , @xmath131 is up to sign equal to the determinant of the @xmath538 submatrix of @xmath47 obtained by deleting columns @xmath167 and @xmath343 . thus , @xmath587 and there exist integers @xmath583 such that @xmath584 . for such integers , define @xmath588 write @xmath589 for @xmath5 , and let @xmath590 . as @xmath591 for @xmath592 , it follows that @xmath593 and @xmath594 have integer entries . moreover , it follows from the proof of lemma [ lem : adjc ] that @xmath593 and @xmath594 belong to the integer kernel of @xmath47 . therefore , @xmath595 is a basis for the integer kernel of @xmath47 if and only if @xmath596 : @xmath593 and @xmath594 are linearly independent if and only if @xmath597 ; and if @xmath597 , then @xmath595 is a basis of an index @xmath560 subgroup of the integer kernel ( * ? ? ? * section 3.1 ) . for @xmath598 , @xmath599 therefore , implies that @xmath600 for @xmath598 . the greatest common divisor of the @xmath131 , for @xmath598 , is equal to @xmath577 . hence , @xmath596 . [ cor : basis - kerparc ] let @xmath146 and @xmath531 such that the matrix @xmath480 is nonsingular . if @xmath1 $ ] such that @xmath601 and @xmath114 is a basis for the integer kernel of @xmath47 , then @xmath602 divides @xmath381 . furthermore , if @xmath603 , then @xmath582 . let @xmath146 and @xmath531 such that @xmath480 is nonsingular . define @xmath604 and @xmath605 as in the proof of lemma [ lem : basis_kerparc ] . then @xmath595 is a basis for the integer kernel of @xmath47 , @xmath606 suppose that @xmath1 $ ] such that @xmath607 and @xmath608 is a basis of the integer kernel of @xmath47 . then there exists a unimodular matrix @xmath609 such that @xmath610 as @xmath49 is nonsingular , @xmath124 . therefore , @xmath611 since @xmath612 . if @xmath603 , then @xmath613 since @xmath614 and @xmath612 . thus , if @xmath603 , then @xmath615 since @xmath229 is unimodular . property of theorem [ thm : gp - to - polys ] is now proved , completing the proof of the theorem : [ lem : polysizebounds ] let @xmath146 and @xmath35 $ ] be a geometric progression modulo @xmath0 such that @xmath480 is nonsingular and @xmath616 . if @xmath1 $ ] such that @xmath158 and @xmath114 is a basis of the integer kernel of @xmath47 , then @xmath617 and @xmath618 for all @xmath55 . let @xmath146 and @xmath35 $ ] be a geometric progression modulo @xmath0 such that @xmath480 is nonsingular and @xmath616 . suppose that @xmath1 $ ] such that @xmath158 and @xmath114 is a basis for the integer kernel of @xmath47 . let @xmath515 . then lemma [ lem : resformula ] implies that @xmath86 and @xmath573 are nonzero . thus , @xmath13 and @xmath14 are coprime and lemma [ lem : gp - kernel ] implies that @xmath619 . therefore , lemma [ lem : resformula ] implies that @xmath620 as @xmath573 is a nonzero integer , the inequality @xmath621 holds for all @xmath55 . combining this inequality with and the upper bound on @xmath622 provided by lemma [ lem : ct - upper - bnd ] yields the first set of inequalities stated in the corollary for all @xmath55 . corollary [ lem : ct - lower - bnd - td - k1 ] , lemma [ lem : resformula ] and imply that @xmath623 where @xmath624 , for all @xmath55 . let @xmath30 be the ratio of @xmath38 modulo @xmath0 . then subtracting @xmath30 times row @xmath625 of @xmath47 from row @xmath167 for @xmath626 produces a matrix whose first @xmath627 rows contain multiples of @xmath0 . as @xmath628 divides each entry of @xmath47 and @xmath616 , it follows that @xmath629 divides @xmath577 . thus , @xmath630 since @xmath631 . combining this inequality with completes the proof of the second set of inequalities stated in the corollary . in the proof of lemma [ lem : polysizebounds ] , the assumption that @xmath38 is a geometric progression modulo @xmath0 such that @xmath616 is only used to prove the last inequality of the lemma . setting @xmath632 where the @xmath52 appears in the @xmath23th coordinate shows that the remaining inequalities can not be improved by a constant factor without using the assumption . if @xmath633 , then and corollary [ lem : ct - lower - bnd - k0 ] imply that @xmath634 if the inequality is strict , then lemma [ lem : polysizebounds ] improves upon the lower bound on @xmath635 provided by lemma [ lem : resultantbound ] . part of this work was performed while the author was employed by the university of queensland , brisbane , australia . the author is grateful to cyril bouvier for many helpful comments and discussions . georg heinig and karla rost , _ introduction to bezoutians _ , numerical methods for structured matrices and applications ( d.a . bini , v. mehrmann , v. olshevsky , e. tyrtsyhnikov , and m. van barel , eds . ) , oper . theory adv . 199 , birkhuser verlag , basel , 2010 , pp . 25118 . thorsten kleinjung , kazumaro aoki , jens franke , arjen k. lenstra , emmanuel thom , joppe w. bos , pierrick gaudry , alexander kruppa , peter l. montgomery , dag arne osvik , herman te riele , andrey timofeev , and paul zimmermann , _ factorization of a 768-bit rsa modulus _ , advances in cryptology crypto 2010 ( tal rabin , ed . ) , lecture notes in comput . 6223 , springer , berlin , 2010 , pp . 333350 . namhun koo , gooc hwa jo , and soonhak kwon , _ on nonlinear polynomial selection and geometric progression ( mod n ) for number field sieve _ , cryptology eprint archive , report 2011/292 , 2011 , http://eprint.iacr.org/2011/292.pdf . eugene tyrtyshnikov , _ hankel minors and pade approximations _ , numerical methods for structured matrices and applications ( d.a . bini , v. mehrmann , v. olshevsky , e. tyrtsyhnikov , and m. van barel , eds . ) , oper . theory adv . 199 , birkhuser verlag , basel , 2010 , pp .

over the last few years , rapid progress has been made in distance measurements for nearby galaxies based on the magnitude of the tip of red giant branch stars . current ccd surveys with hst and large ground- based telescopes bring%-accurate distances for roughly a hundred galaxies within 5 mpc . the new data on distances to galaxies situated in ( and around ) the nearest groups : the local group , m81 group , cena / m83 group , ic342/maffei group , sculptor filament , and canes venatici cloud allowed us to determine their total mass from the radius of the zero- velocity surface , , which separates a group as bound against the homogeneous cosmic expansion . the values of for the virialized groups turn out to be close each other , in the range of 0.9 1.3 mpc . as a result , the total masses of the groups are close to each other , too , yielding total mass - to - blue luminosity ratios of 10 40 . the new total mass estimates are 3 5 times lower than old virial mass estimates of these groups . because about half of galaxies in the local volume belong to such loose groups , the revision of the amount of dark matter ( dm ) leads to a low local density of matter , , which is comparable with the global baryonic fraction , but much lower than the global density of matter , . to remove the discrepancy between the global and local quantities of , we assume the existence of two different dm components : 1 ) compact dark halos around individual galaxies and 2 ) a non - baryonic dark matter `` ocean '' with and , respectively .

astro-ph0410065

in rich clusters of galaxies , virial mass estimates agree well with independent determinations of the cluster mass made from the x - ray flux of hot intracluster gas , and from weak gravitational lensing effects . a typical ratio of the total mass - to - blue luminosity for rich clusters , @xmath9 , extrapolated over the whole volume of the universe , yields the mean density of matter @xmath100.25 , in the excellent concordance with parameters of the standard @xmath11cdm model : @xmath6 = 0.27 , @xmath12 = 0.73 ( spergel et al . 2003 ) . however , about 85% of galaxies are situated outside the rich clusters . roughly a half of them belong to groups of different size and population , while the remaining half are scattered in diffuse ( unvirialized ) `` clouds '' and `` filaments '' usually called the `` field '' . until recently , application of the virial theorem to galaxy groups remained the only way to trace the dark matter distribution on scales of 0.1 1 mpc measurements of the total mass of individual groups via their x - ray flux or weak gravitational lensing have not lead yet to distinct results . in the case of our local group ( lg ) , lynden - bell ( 1981 ) and sandage ( 1986 ) proposed to determine its total mass using a method , which is based on a measurement of the radius of the `` zero - velocity surface '' , @xmath13 , where the overdensity of the group has halted expansion and infall is about to commence . under the assumption of spherical symmetry the total mass of a group can be expressed in terms of the radius @xmath13 , the age of the universe @xmath14 , and the gravitational constant @xmath15 as @xmath16 application of this method to the lg as well to other groups requires accurate distances and radial velocities for all galaxies surrounding a group . such data is now accessible for several nearby groups by using the luminosity of stars at the tip of the red giant branch ( trgb ) as standard candles . over the last 5 years , searches for additional nearby dwarf galaxies , made on the poss - ii & eso / serc plates by karachentseva & karachentsev ( 1998 ; 2000 ) , and also results of `` blind '' hi surveys of the southern sky , and the zone of avoidance by the hipass team ( kilborn et al . 2002 ) , as well as other wide - field sky surveys , lead to a doubling of the population of known galaxies in the local volume . basic observational data on 450 local volume galaxies with distances less than 10 mpc are presented in the catalog of neighboring galaxies ( karachentsev et al . . the new quantity and quality of observational data on nearby galaxy distances and radial velocities allowed us to determine virial and total masses of four of the nearest complexes : milky way+m31 ( karachentsev et al . 2002c ) , m81+ngc2403 ( karachentsev et al . 2002a ) , cena+m83 ( karachentsev et al . 2002b ) , and ic342+maffei ( karachentsev et al . 2003d ) , as well as of two other nearby scattered galaxy systems : m96 = the canes venatici i cloud ( karachentsev et al . 2003a ) and ngc253 = the sculptor filament ( karachentsev et al . below we discuss the results of mass determination for all six nearest groups made via internal and external galaxy motions . our galaxy , the milky way , and the andromeda nebulae , m31 , together with their suites of dwarf companions are usually combined into the united dynamical system , the local group ( = lg ) . the basis for this is the observed mutual approaching of two the giant galaxies with a velocity of 123 km s@xmath17 . however , such a definition has an apparent disadvantage because the dynamical center of the lg turns out to be in emptiness , and individual motions of dwarf galaxies occur around the andromeda and milky way themselves . for this reason , we prefer to consider both the giant spiral galaxies as the centers of two independent groups forming together the local complex of galaxies . different properties of companions to our galaxy are described in detail in a monograph by van den bergh ( 2000 ) `` the galaxies of the local group '' . we will discuss here only the most general parameters of the group . the list of galaxies dynamically associated with the milky way is presented in table 1 . its columns contain : ( 1 ) galaxy name ; ( 2 ) morphological type ; ( 3,4 ) radial velocities with respect to the galaxy center and regarding to the lg centroid , respectively , with the apex parameters adopted in the nasa extragalactic database ( ned ) ; ( 5 ) absolute magnitude corrected for the galactic extinction ( schlegel et al . 1998 ) and internal galaxy extinction in the manner , which takes into account the galaxy inclination , as well as luminosity ( karachentsev et al . 2004 ) ; ( 6 ) distance to the galaxy in mpc with indication of the method used : from the luminosity of cepheids ( cep ) , from the luminosity of the tip of the red giant brench ( rgb ) as given in ( karachentsev et al . 2004 ) ; ( 7 ) the so called `` tidal index '' : @xmath18 where @xmath19 is the total mass of any neighboring galaxy separated from the considered galaxy by a space distance @xmath20 ; for every galaxy `` @xmath21 '' we found its `` main disturber''(=md ) , producing the highest tidal action ; the value of the constant @xmath22 is chosen so that @xmath23 when the keplerian cyclic period of the galaxy with respect to its main disturber equals the cosmic hubble time , @xmath24 . in this sense , galaxies with @xmath25 may be considered as isolated objects of the general field . besides 15 members of the milky way group , we indicate under the horizontal line at the bottom of table another four galaxies : tucana , sagdig , sexa , and sexb , for which the milky way is also the md . all of them move away from our galaxy , taking apparently part in the general cosmic expansion . note that the lmc stands out as the md with respect to the milky way and the smc . by luminosity , the milky way appreciably dominates over the surrounding objects : the net luminosity of its 14 companions is an order magnitude lower than the milky way luminosity itself . the radius of gravitational prevalence of the milky way extends to @xmath0700 kpc , while the radial velocity dispersion of the companions is 86 km s@xmath17 with regard to the galaxy center and 76 km s@xmath17 with respect to the lg centroid . applying the virial theorem , we obtain for our group the mass estimate @xmath26 where @xmath27 is the dispersion of radial velocities with respect to the group centroid , and @xmath28 is the mean projected harmonic radius . this relation assumes a spherical symmetry of the group and a random orientation of velocity vectors for the group members . adopting @xmath29 = 68 kpc , gives us @xmath30 . for arbitrarily oriented keplerian orbits of companions with the eccentricity @xmath31 the robust estimator of mass is @xmath32 adopting @xmath31 = 0.7 as the average eccentricity , we derive for the companions of the milky way @xmath33 in excellent agreement with the previous estimate . however , two circumstances exist that force one to suppose that the two evaluations obtained overestimate the mass of our group . firstly , all the companions to milky way ( but for ngc 6822 ) are situated in an elongated volume with an axial ratio 8:3:1 , known as the polar magellanic stream . secondly , when observing orbital motions of the companions from inside , if orbits are strongly elongated then we see almost the full vectors of their velocities , i.e. the statistical relationship @xmath34 between radial and spatial velocity dispersions is not satisfied . over the last five years five new companions to m 31 : and v , cas dsph , peg dsph , cetus , and and ix ( zucker et al . 2004 ) have been discovered . as a result , the present population of m 31 group exceeds the population of the group around our galaxy . the list of 19 members of the andromeda group is given in table 2 . the following information is provided in the table : ( 1 ) galaxy name ; ( 2 ) morphological type ; ( 3 ) radial velocity with respect to the lg centroid ; ( 4 ) absolute magnitude after corrections for internal and external extinctions ; ( 5 ) distance from the observer ; ( 6 ) angular separation from the m 31 , and ( 7 ) the tidal index . as it is seen , the maximum angular separation of the most remote companions to andromeda reaches one radian . at the bottom of the table under the horizontal line another four dwarf galaxies are given : ddo 210 , kkh 98 , kkr 25 , and kk 230 , for which the m 31 is the md . however , their crossing time with respect to the andromeda exceeds the age of the universe , which does not permit us to consider them physical members of the m 31 group . for 18 companions to andromeda their mean projected linear separation is 254 kpc , while the mean harmonic radius is 42 kpc . the radial velocity dispersion in the m 31 group , 77 km s@xmath17 , is nearly the same as in the milky way group . applying the above mentioned relations , we obtain the mass estimates : @xmath35 and @xmath36 . studying the kinematics of the giant stellar stream around the m 31 , ibata et al . ( 2004 ) measured the mass of andromeda halo on the scale of 125 kpc . their estimate @xmath37 agrees well with the virial and orbital mass estimates . note also , that almost the same mass estimates have been derived by evans & wilkinson ( 2000 ) , cote et al . ( 2000 ) , and evans et al . ( 2000 ) on the scale of @xmath050 kpc based on the motions of companions and globular clusters of m 31 . kinematics of galaxies around the m 81 has been considered by karachentsev et al . ( 2002a ) . recently , several new galaxies of the group have been imaged with the advanced camera for survey at the hst . a slightly renewed parameters of the m 81 group members are summarized in table 3 , where column designations are the same as in previous tables ( in column 6 `` mem '' means that the galaxy distance was ascribed from its membership as the mean distance to the group . ) the last column contains notes regarding some objects , for which the main disturber is not m 81 , but other nearby galaxies . as it is seen , in the number of members , n = 29 , the m 81 group exceeds both the milky way group and the andromeda group . below the horizontal line in the table 3 , we give another six galaxies which may belong also to the far periphery of the m 81 group . some of them are associated with another spiral galaxy ngc 2403 rather than with m81 itself , forming a scattered system ( `` flock '' ) , which moves towards the m 81 . both from the radial velocity dispersion , @xmath38 = 91 km s@xmath17 , and from the linear projected radius , @xmath39 kpc , the m 81 group is similar to the milky way and the m 31 groups . virial and orbital mass estimates for the m 81 group are @xmath40 and @xmath41 , respectively . the structure and kinematics of this nearby galaxy complex have been discussed by karachentsev et al . ( 2002b ) . the group of galaxies around m 83 ( = ngc 5236 ) actually has the same mean radial velocity ( + 308 km s@xmath17 ) as the cen a group ( + 312 km s@xmath17 ) , but the distance from us appreciably larger ( 4.56 mpc ) than that of the cen a group ( 3.66 mpc ) . the updated lists of 28 and 14 members of both the groups are presented in tables 4 and 5 , respectively . notes in the last column indicate group members having another main disturbers ( not cen a or m 83 ) . under the horizontal lines there are some galaxies with larger projected separations which may be associated with the complex too . in the cen a group its mean linear projected radius , 290 kpc , and radial velocity dispersion , 105 km s@xmath17 , are markedly greater than in the m 83 group ( 164 kpc and 71 km s@xmath17 ) that leads to a considerable difference in mass estimates of the groups : 489 ( vir ) and 288 ( orb ) in @xmath42 for the cen a , and 109 ( vir ) and 100 ( orb ) in @xmath42 for the m 83 group , respectively . such a difference is consistent with the assumption of bahcall et al . ( 1995 ) that giant elliptical galaxies have a mass 2 3 times larger per unit luminosity than giant spiral galaxies . this nearby binary group is situated in the zone of strong galactic extinction , which complicates analysis of its dynamics . properties of the structure and kinematics of the complex have been considered by karachentsev et al . ( 2003d ) . the last summary of data on distances , radial velocities and luminosities of galaxies around ic342/maffei was published by karachentsev et al . these data are reproduced in table 6 and 7 . here , distance estimates made from luminosity of the brightest stars or via tully - fisher and faber - jackson relations are indicated as `` bs '' , `` tf '' , and `` fj '' , respectively . for three galaxies : ugca 86 , kkh 37 , and kkh 6 we use here new distance estimates obtained with the acs hst . at present , both the groups around the giant face - on spiral ic 342 , and around the pair of e+s galaxies maffei 1 and maffei 2 contain eight members each . their numbers may be increased after a careful survey of this region in the hi line . two galaxies , kkh 37 and kkh 6 , with negative tidal indexes ( in bottom of tables 6 and 7 ) do not belong , apparently , to the bound members of the complex . both the groups have a rather low dispersion of radial velocities , 54 km s@xmath17 ( ic 342 ) and 59 km s@xmath17 ( maffei ) , and quite typical mean linear projected separations , 322 kpc and 104 kpc , respectively . the virial and orbital mass estimates for the groups come to 57 and 95 ( ic 342 ) , and 65 and 135 ( maffei ) in units of @xmath42 . as it was shown by jerjen et al . ( 1998 ) , a conglomeration of bright galaxies in sculptor is a loose filament stretched along the line of sight and involved in the general cosmic expansion . according to karachentsev et al . ( 2003c ) , the giant spiral galaxy ngc 253 and its five companions form a semi - virialized core of the filament . data on ngc 253 and its companions are given in table 8 . in its lower part we indicate also five more remote galaxies associated with the group core . at a radial velocity dispersion 54 km s@xmath17 and a mean harmonic projected radius @xmath43 kpc , estimates of mass of the group are 332 ( vir ) and 153 ( orb ) in units of @xmath42 . however , both the mass estimates look extremely unreliable since the group crossing time , @xmath44 gyr , is too long for the virialization of the system . note also , that inclusion of 5 more distant galaxies under the horizontal line in table 8 brings the crossing time closer to the hubble time , @xmath013 gyr . this very loose extended system mainly inhabited by dwarf irregular galaxies was studied by karachentsev et al . ( 2003a ) . on the map of the sky region presented by the authors the cvni cloud occupies an area of about 35@xmath45 in diameter with the center near the brightest galaxy m 94 ( = ngc 4736 ) . for the sample of 34 members of the cloud , karachentsev et al . ( 2003a ) derived the radial velocity dispersion 50 km s@xmath17 , the mean linear projected radius 760 kpc , and the mass estimates : @xmath46 ( vir ) and @xmath47 ( orb).the crossing time for the cvni cloud is 15 gyr , therefore , the system is rather in the free hubble expansion than in a state of dynamical equilibrium . arrangement of galaxies by their tidal index shows that the brightest galaxy of the cvn i cloud , m 94 , is the main disturber relative only to 10 neighboring galaxies . all of them are presented in table 9 , where column designations are the same as in the previous tables . for 9 members of the m 94 group with positive tidal indices , we obtain the radial velocity dispersion 56 km s@xmath17 , the mean harmonic projected radius 346 kpc , the mass estimates @xmath48 ( vir ) , @xmath49 ( orb ) , and the crossing time 6.9 gyr . thus , even the central region of the cvni cloud can not be considered as dynamically relaxed sub - system . in all the groups , but for the strongly obscured complex ic342/maffei , there is an appreciable number of diffuse dwarf spheroidal galaxies ( dsph ) . measurement of radial velocities for these smooth objects of low surface brightness is an extremely hard observational task due to absence in them of contrast details , as well as lack of neutral hydrogen . it is only in the closest dsph galaxies that one can measure the radial velocity from a globular cluster ( when available ) or the brightest stars . the majority of such `` bald '' dwarfs are concentrated in compact groups and virgo cluster ( karachentseva & sharina , 1987 ) . this fact suggests we can assign to `` bald '' dwarfs the mean distance of the group , in whose visible perimeter they are situated . measurements of distances to dsphs with the hst from trgb generaly confirm their membership in corresponding groups . the total number of dsph+e companions in the complexes : milky way / m31 , m81 , cena / m83 , and ic342/maffei amount to 55 against 58 companions of dir+s types . `` bald '' companions show stronger concentration toward the principal galaxy as compared to gas - rich ones . thus , the medians of projected separations for them equal to 167 kpc ( dsphs ) and 210 kpc ( dirs ) . one dsph object , cetus , is at a projected distance of over 500 kpc from the main galaxy of the group . the observed difference between the median projected separations ( 167 kpc vs. 210 kps ) may be considered as insignificant . nevertheless , among 233 galaxies having distances to us within 5 mpc , there are four only isolated early - type galaxies : ngc 404 ( @xmath50 = 1.0 , md = maffei 2 ) , kks3 ( @xmath50 = 0.3 , md = ngc 1313 ) , kk 258 ( @xmath50 = 0.9 , md = ngc 253 ) , and tucana ( @xmath50 = 0.1 , md = milky way ) . all of them are peculiar objects requiring special detailed study . as an aside , it should be noted that the observed number of `` bald '' dwarfs in the group tends to increase with the luminosity of bulge of the main group member , and to decrease with crossing time of the group , which has a quite obvious evolutionary interpretation . some general characteristics of the discussed groups are listed in table 10 . the following information is provided in the table lines : ( 1,2 ) mean group distance from us and from the lg centroid ( in mpc ) ; when determining the lg centroid , we assumed that the total masses of the m 31 and the milky way are as 5 : 4 ( karachentsev & makarov , 1996 ) ; ( 3 ) distance of group center from the plane of the local supercluster ( in mpc ) ; ( 4,5 ) the known total number of members in the group and galaxies of early - type ( e+dsph ) ; ( 6,7 ) morphological type and absolute magnitude of the brightest member ; ( 8) rotation velocity of the main galaxy ( in km s@xmath17 ) ; ( 9 ) radial velocity of the principal member relative to the lg centroid ( km s@xmath17 ) ; ( 10 ) mean radial velocity of the group in the system of rest lg ( km s@xmath17 ) ; ( 11 ) dispersion of radial velocities in the group ( in km s@xmath17 ) ; ( 12 ) mean projected separation of the companions from the principal galaxy ( in kpc ) ; here the mean spatial distance of the milky way companions is multiplied by ( @xmath51/4 ) to account for projection effects ; ( 13 ) integrated blue luminosity of the group ; ( 14,15 ) virial and orbital mass estimates ( in @xmath42 units ) ; ( 16,17 ) virial and orbital mass - to - luminosity ratio in solar units ; ( 18 ) crossing time , @xmath52 , in gyr . as it was to be expected , two loose systems : sculptor filament and cvni cloud are characterized by the greatest crossing time , which exceeds half the hubble time . these two systems ( the last right columns in table 10 ) have obviously not reached an dynamical equilibrium , and we will no discuss them further . comparison of the parameters of the remaining groups allow us to make the following statements . \a ) the centers of all the groups reside in a narrow layer ( @xmath530.33 mpc ) thick with respect to the local supercluster plane , which accounts for only 10% of the volume considered . \b ) judging by the principal characteristics : dimension , luminosity , velocity dispersion , and content of dsphs , the local group is a typical representative of nearby groups , where one main galaxy dominates . \c ) virial / orbital mass - to - luminosity ratios for nearby groups lies within [ 8 88 ] @xmath2 . their median 29 @xmath2 is 3 5 times as small as the old estimates made by huchra & geller ( 1982 ) and tully ( 1987 ) for groups of the same luminosity . \d ) the binary structure looks to be a common feature of nearby groups . some paired groups : milky way + m 31 and m 81 + ngc 2403 , manifest the mutual approach of their principal galaxies , and the kinematic status of the others is open to question . \e ) the crossing times in neighboring groups are concentrated within [ 1.8 5.9 ] gyr , while the median , 2.3 gyr , is 6 times as small as the hubble time , which enables these groups to be regarded as advanced in their dynamical evolution . precise measurements of distances and radial velocities for galaxies surrounding a group permits one to determine the radius of zero- velocity surface , @xmath13 , which separates the group from the general cosmic expansion . determinations of @xmath13 for the six nearest groups were performed by karachentsev et al . ( 2002a , b , c , 2003a , b , c ) . the results are collected in table 11 . its first and second lines present the total mass of each group / complex estimated by the virial theorem or from orbital motions of companions around the principal galaxy . ( as the total mass of the lg , we use the virial / orbital mass estimates of the m31 group multiplied on the factor 1.8 , which takes into account the expected mass ratio , 4:5 , for the milky way and the m31 ) . the third line contains data on the turn - over radius @xmath13 and its error . the fourth line gives the total mass calculated from @xmath13 as @xmath54 , where the age of the universe is adopted to be ( [email protected] ) gyr ( spergel et al . 2003 ) . the last two lines indicate the total blue luminosity of the group / complex and the total mass - to - luminosity ratio , respectively . figure 1 shows the relationship between mass estimates for the groups made by two quite independent series of observational data : from internal motions of group members ( horizontal ) and from external motions of galaxies surrounding the group . two complexes which are likely unbound : the canes venatici i cloud and the sculptor filament are shown by the dotted boxes . all four likely virialized systems : milky way / m31 , m81/ngc2403 , cena / m83 , and ic342/maffei manifest quite satisfactory agreement between independent mass estimates . in two apparently unvirialized expanding complexes ( sculptor and cvni ) , their virial / orbital mass estimates turn out to be 3 8 times larger than the total masses estimated via @xmath13 . the derived total mass - to - luminosity ratios show a surprisingly low scatter , being ranked inside [ 9 37 ] with a median of 19 in solar units . the ratio of the sum of the total masses for the four complexes to the sum of their total luminosity is @xmath55 . as we have already noted , all the groups under discussion are situated in the sphere of radius 5 mpc . from the presently available data , this volume contains 233 galaxies . about half of them , 121/233 , are members of the lg , the m81/ngc2403 group , the cena / m83 group , and the ic342/maffei group . besides them , 62 more galaxies in the 5 mpc- sphere ( i.e. 27% ) belong to smaller multiple systems around ngc 3109 , ugc 8760 , ngc 784 , and ugc 3974 , and two expanding complexes in sculptor and canes venatici . thus , the relative number of the group members in the local volume accounts for about 78% , and only 22% of nearby galaxies can be considered to be the population of the general `` field '' . according to the `` catalog of neighboring galaxies '' ( karachentsev et al . 2004 ) , the integrated luminosity of all galaxies within 5 mpc is @xmath56 ( 5 mpc ) = @xmath57 . from the fifth line of table 11 , the integrated luminosity of the virialized groups is @xmath58 or 65% of the total luminosity of the volume . with allowance made for the sculptor filament and the cvni cloud , the fraction of local luminous matter in systems increases to 82% . the numbers presented reflect the known effect of segregation of galaxies by luminosity , with the concentration of dwarf galaxies in groups less pronounced than for giant galaxies . one of the remarkable properties of the spatial distribution of galaxies is its fractality . on different scales , groups / clusters look like dense knots in filaments which concentrate towards sheets , forming as a whole a fractal `` cosmic web '' pattern . this general picture is valid for the local volume too . in the local sphere of radius 5 mpc all the virialized groups are situated in a layer of @xmath53(1/3 ) mpc around the local `` pancake '' , which occupies only 10% of the sphere volume , @xmath59 = 523 mpc@xmath60 . however , the sum of 4 group volumes with radii @xmath61 mpc each add to only 20 mpc@xmath60 or @xmath04% of the local volume . thus , in the local volume of 5 mpc radius , only a small part ( 4% ) is not involved in the hubble expansion , but about 70% of luminous matter is concentrated in this volume . obviously , any theory of galaxy formation must explain these dimensionless empirical parameters of the local web . on scales of 10 100 mpc , such an approach is being developed by shandarin ( 2004 ) . giving preference to estimates of the total mass from the pattern of the velocity field of galaxies surrounding groups , we obtain the total mass of six complexes ( line 4 in table 11 ) , @xmath62 , which yields the mean density of matter inside the 5 mpc sphere @xmath63(5 mpc ) = @xmath64 mpc@xmath60 . at @xmath65 km s@xmath17 mpc@xmath17 , the critical density of matter , @xmath66 ) , is equal to @xmath67 mpc@xmath60 . consequently , the mean local density of matter is only @xmath63 ( 5 mpc ) = 0.10 @xmath68 . it follows from the data of the catalog ( karachentsev et al . 2004 ) that the mean density of luminosity within 5 mpc equals @xmath69/mpc@xmath60 . comparing this with the mean luminosity density estimated from the sloan digital sky survey ( blanton et al . 2003 ) and the millenium galaxy catalogue ( liske et al . 2003 ) , we obtain a ratio @xmath70 mpc)/ @xmath71 . here we took into account that ignoring internal extinction in galaxies in sloandss and mgc causes underestimates of @xmath72 by a factor of 1/3 . supposing that on different scales the mass density is strictly proportional to the luminosity density ( no biasing ) , then the mean density of matter contained in the groups leads to the value of global density of matter of about 0.025 of the critical density over the past few years , it has repeatedly been noted ( sandage , 1986 , karachentsev & makarov , 2001 , ekholm et al . 2001 , karachentsev et al . 2003b ) that the local hubble flow is rather cold with a dispersion of radial velocities less than 70 km s@xmath17 . as can be seen from the data of table 10 , the dispersion of virial velocities in nearby groups has approximately the same value . for the galaxies situated outside the groups , the estimates of @xmath38 are chiefly determined by errors of measurements of distances to galaxies , and the role played by these errors enhances with distance . in the immediate vicinities of the lg on scales of 2 3 mpc , the radial velocity dispersion for the field galaxies is only 25 30 km s@xmath17 ( karachentsev et al . 2002c ) . apparently , it is not always easy to distinguish between true isolated galaxies and members of loose groups . for this reason , it is interesting to examine the behaviour of group centers , but not individual galaxies on the hubble diagram . such a hubble relation is displayed in figure 2 using the data of lines 2 and 10 of table 10 . the errors of measurements of the mean velocities of the groups are not large ( @xmath73 km s@xmath17 ) , and we do not show them here . the horizontal bars correspond to the errors of the average distance of the groups being usually @xmath010% . the line corresponds to the hubble relation with @xmath65 km s@xmath17 mpc@xmath17 and a lg zero - velocity surface at 0.95 mpc . dispersion of radial velocities obtained from these data for the centroids is 25 km s@xmath17 after quadratic subtracting of distance errors . the low values of `` thermal '' velocities of the galaxies in the field and the centers of the groups are remarkably consistent with the low estimates of the mean density of matter in the local volume . note that the global value @xmath74 km s@xmath17 mpc@xmath17 fits fairly the local hubble flow . above , we considered the nearest groups only , because the quality of data on distances of galaxies situated beyond 5 mps is rather unsatisfactory . for instance , the majority of galaxies in the layer @xmath75 mpc have no direct individual distance estimates at all . nevertheless , the catalog of neighboring galaxies provide us with data on tidal indices and the names of the main disturber for each galaxy within 10 mpc , giving an idea of the location and population of more distant groups . without discussing characteristics of these groups , let us enumerate only the most representative of them containing four or more members : ngc 672 ( 5 ) , ngc 2784 ( 6 ) , ngc 3115 ( 7 ) , ngc 3368/3412/3489= leo - i group ( 37 ) , ngc 4244 ( 4 ) , ngc 4594 ( 7 ) , m101 ( 5 ) , and ngc 6946 ( 7 ) . here , every group is noted by the name of the brightest member , and the number of group members with @xmath76 0 is shown in brackets . apart from the leo - i group , all others resemble the local group by a prevalence of a single galaxy . linear dimensions , velocity dispersions and luminosities of the groups are similar to parameters indicated in table 10 . only the leo - i group is different from the others by the presence of several giant members of early ( e , s0,sa ) types , comparable to each other in luninosity . this group , resembling a mini - cluster , is characterized by the projected radius of 350 kpc , the radial velocity dispersion of 130 km s@xmath17 , and the virial mass - to - luminosity ratio of @xmath77 ( karachentsev & karachentseva , 2004 ) . the galaxy groups of the leo - i - type are much more sparse than groups of the lg - type : in the 10 mpc volume around us the ratio of their numbers is approximately 1:30 . hence , the leo - i - type groups make a contribution of minor importance to the mean density of matter . properties of groups situated within the local supercluster have been recently studied by makarov & karachentsev ( 2000 ) . they applyed a new algorithm ( similar to a criterion @xmath76 0 ) to a distribution of 6320 galaxies with radial velocities @xmath78 km s@xmath17 and selected 839 groups . for the groups with population @xmath79 the following median characteristics were derived : the radial velocity dispersion of 86 km s@xmath17 , the projected harmonic radius of 250 kpc , the crossing time of 1.5 gyr , and the virial mass - to - luminosity ratio of @xmath80 . therefore , the characteristics of the nearest groups do not differ essentially from the main dynamical characteristics of more representative sample covering a scale of @xmath030 mpc . as it has been mentioned above , determinations of mass of the galaxy clusters , made by different ways , yield mutualy concordant results with a typical mass - to - luminosity ratio about ( 250 300 ) @xmath2 in the @xmath81 band . assuming the linear proportionality between dark and luminous matter this leads to the global density of matter in the universe @xmath82 , in agreement with wmap ( spergel et al . however , only ( 10 15)% of all galaxies are concentrated in rich clusters . hence , the collective input of clusters into @xmath6 consist of 0.03 0.04 only . most of galaxies , about ( 50 70)% , are situated in groups of different size and population , with the lg and other nearest groups typical representatives . recent accurate measurements of distances to galaxies open a possibility to measure group masses by two independent manners : via internal and via external galaxy motions . applied to the well studied neighboring groups , both the methods manifest good mutual agreement . however the mass - to - luminosity ratios for groups turn out to be one order less than for clusters . it means that the hypothesis assuming linear proportionality of dark and luminous matter on different scales is not valid . since the `` external '' method estimates the total mass of group on a scale of @xmath83 mpc , but the `` internal '' method yields virial / orbital mass estimate on a lower scale , @xmath84 kpc , then the dark matter in groups is tightly tied to the luminous one . the case of the nearest group around andromeda shows that the total mass of the m31 halo , @xmath85 is reached already on the scales of ( 50 125 ) kpc , i.e. one order less than the group turn - over radius @xmath13 . as it is seen from figure 2 , centroids of the nearby groups have a low velocity dispersion with respect to the hubble flow that is an independent evidence of the low mean density of matter in the local volume ( governato et al . the principal galaxies in nearby groups ( shown by crosses in figure 2 ) have small peculiar velocities too . only in the m 81 group do we find a significant peculiar velocity of the main member . ( the situation in the maffei group is complicated by heavy galactic extinction , as well as confusion of hi emission from the group members with the galactic hi emission ) . an apparent reason of the high peculiar velocity of the m 81 is the presence nearby m 81 of another bright galaxy , m 82 , which has a peculiar velocity of opposite sign ( see table 3 ) . moreover , the peculiar velocities of m 81 and m 82 are inversly proportional to their luminosities . consequently , the galaxies m 81 and m 82 are moving in the group as heavy bodies driven by the law of conservation of motion , rather than as test particles inside the common smooth potential well . supposing the dark matter in groups follows tightly the luminous matter , we can estimate the total contribution of groups to the mean density of matter . based on the average weighted values : @xmath86 , @xmath87 , and @xmath88 in solar units , we derive the total contribution of groups of galaxies to the mean density of matter to be ( 0.03 0.04 ) in critical density units . recently , guzik & seljak ( 2002 ) and hoekstra et al . ( 2004 ) determined parameters of dark halos for field galaxies and members of loose groups from weak lensing in the sloan dss and in the red - sequence cluster survey . for a galaxy with the mean luminosity of @xmath89 km s@xmath17 mpc@xmath17 ) they derived a characteristic ratio @xmath90 = 41 @xmath2 on a scale of @xmath0250 kpc ( after correction for internal extinction ) . this independent result agrees well with our data . thus , we may conclude that groups and clusters of galaxies produce approximately the same contribution to the mean density of matter . their combined contribution is about 0.06 0.08 in the units of critical density or about ( 1/5 1/3 ) with respect to @xmath5 . to avoid a contradiction with the global matter density @xmath5 , derived from the wmap , we need to assume the existence of another ( non - baryonic ) component of dark matter with the mean density @xmath8 . the dm2-component may be distributed in space either homogeneously likes a dark cosmic `` ocean '' , or consists of many low mass halos without gas and stars , as assumed by tully et al . ( 2002 ) and tully ( 2004 ) . obviously , the presence of the dm2- component does not significantly affect the kinematics of even loose groups , because it contributes a small fraction ( about 10% within @xmath91 mpc and about 0.1% within @xmath92 mpc ) into the integrated mass of groups . ct , mateo m. , sargent w.l.w . , olszewski e.w . , 2000 , apj , 537 , l91 ekholmt . , baryshev y. , teerikorpi p. , et al . 2001 , a&a , 368 , l17 evans n.w . , wilkinson m.i . , guhathakurta p. , grebel e.k . , vogt s.s . , 2000 , apj 540 , l9 evans n.w . , wilkinson m.i . , 2000 , mnras 316 , 929 governato f. , moore b. , cen r. , et al . 1997 , new astronomy , 2 , 91 guzik j. , seljak u. , 2002 , mnras , 335 , 311 huchra j.p . , geller m.j . , 1982 , apj , 257 , 423 hoekstra h. , yee h.k . , gladders m.d . , 2004 , apj , 606 , 67 ibata r. , chapman s. , ferguson a.m. , et al . 2004 , mnras , 351 , 117 jerjen h. , freeman k.c . , binggeli b. , 1998 , aj , 116 , 2873 karachentsev , i.d . , karachentseva , v.e . , huchtmeier w.k . , makarov d.i . , 2004 , aj , 127 , 2031 karachentsev , i.d . , karachentseva , v.e . 2004 , astron . , 81 , 298 karachentsev , i.d . , karachentseva , v.e . 2002 , asp conference series , vol . 284 , p. 325 karachentsev , i.d . , sharina , m.e . , dolphin , a.e . 2003a , a&a , 398 , 467 karachentsev , i.d . , makarov , d.i . , sharina , m.e . 2003b , a&a , 398 , 479 karachentsev , i.d . , grebel , e.k . , sharina , m.e . 2003c , a&a , 404 , 93 karachentsev , i.d . , sharina , m.e . , dolphin , a.e . , & grebel , e.k . 2003d , a&a , 408 , 111 karachentsev , i.d . , dolphin , a.e . , geisler , d. et al . 2002a , a&a , 383 , 125 karachentsev , i.d . , sharina , m.e . , dolphin , a.e . 2002b , a&a , 385 , 21 karachentsev , i.d . , sharina , m.e . , makarov , d.i . 2002c , a&a , 389 , 812 karachentsev , i.d . & makarov , d.i . 2001 , afz , 44 , 5 . karachentsev , i.d . & makarov , d.i . 1996 , lett . to astron.zh . , 22 , 510 karachentseva v.e . , karachentsev , i.d . , 1998 , a&as , 127 , 409 karachentseva , v.e . , & karachentsev , i. d. 2000 , a&as , 146 , 359 karachentseva v.e . , sharina m.e . , 1987 , the catalogue of low surface brightness dwarf galaxies , publ . astroph . obs . 57 , 1 kilborn v.a . , 2002 , aj , 124 , 690 liske , j. , lemon , d.j . , driver , s.p . 2003 , mnras , 344 , 397 lynden - bell d. , 1981 , observatory , 101 , 111 makarov d.i . , karachentsev i.d . , 2000 , in iau coll . 174 , `` small galaxy groups '' , eds . m.valtonen & c.flynn , 40 sandage a. , 1986 , apj , 307 , 1 schlegel , d.j . , finkbeiner , d.p . , & davis , m. , 1998 , apj , 500 , 525 . shandarin s.f . , 2004 , ( astro - ph/0405303 ) spergel d.n . , verde l. , peiris h.v . 2003 , apjs , 148 , 175 tully , r.b . , 1987 , apj , 321 , 280 tully , r.b . , somerville r.s . , trentham n. , verheijen m.a . , 2002 , apj , 569 , 573 tully , r.b . , 2004 , ( astro - ph/0312441 ) van den bergh , s. 2000 , the galaxies of the local group , cambridge univ . press zucker d.b . , kniazev a.y . , bell e.f . , et al . 2004 ( astro - ph/0404268 ) lrrrrlr milky way & 4 & 0 & @xmath9388 & @xmath9320.80 & 0.01 cep & 2.5 + sgr dsph & @xmath933 & 171 & 161 & @xmath9312.67 & 0.024 rgb & 5.6 + lmc & 9 & 84 & 28 & @xmath9317.93 & 0.050 cep & 3.6 + smc & 9 & 17 & @xmath9322 & @xmath9316.35 & 0.063 cep & 3.5 + ursa min & @xmath933 & @xmath9385 & @xmath9344 & @xmath93 7.13 & 0.063 rgb & 3.3 + draco & @xmath933 & @xmath9398 & @xmath9348 & @xmath93 8.74 & 0.079 rgb & 3.0 + sex dsph & @xmath933 & 74 & 8 & @xmath93 7.98 & 0.086 rgb & 2.8 + sculptor & @xmath933 & 77 & 96 & @xmath93 9.77 & 0.088 rgb & 2.8 + carina & @xmath933 & 7 & @xmath9353 & @xmath93 8.97 & 0.10 rgb & 2.7 + fornax & @xmath933 & @xmath9336 & @xmath9332 & @xmath9311.50 & 0.14 rgb & 2.3 + leo ii & @xmath933 & 22 & @xmath9318 & @xmath93 9.23 & 0.21 rgb & 1.7 + leo i & @xmath933 & 177 & 128 & @xmath9310.97 & 0.25 rgb & 1.5 + phoenix & @xmath931 & @xmath93103 & @xmath93106 & @xmath9310.22 & 0.44 rgb & 0.8 + ngc 6822 & 10 & 44 & 64 & @xmath9315.22 & 0.50 cep & 0.6 + leo a & 10 & @xmath9315 & @xmath9340 & @xmath9311.36 & 0.69 rgb & 0.2 + tucana & @xmath932 & 35 & 9 & @xmath93 9.16 & 0.88 rgb & @xmath930.1 + sagdig & 10 & 9 & 23 & @xmath9311.49 & 1.04 rgb & @xmath930.3 + sex a & 10 & 163 & 94 & @xmath9313.95 & 1.32 cep & @xmath930.6 + sex b & 10 & 168 & 111 & @xmath9313.96 & 1.36 rgb & @xmath930.7 + lrrrrcr m 31 & 3 & @xmath9335 & @xmath9321.58 & 0.77 cep & @xmath940.0 & 4.6 + ngc 221=m 32 & @xmath935 & 121 & @xmath9315.96 & 0.77 rgb & @xmath940.40 & 6.8 + ngc 205 & @xmath935 & 24 & @xmath9316.15 & 0.83 rgb & @xmath940.68 & 3.7 + and ix & @xmath933 & @xmath93 & @xmath93 7.5 & 0.79 rgb & @xmath942.60 & 3.8 + and i & @xmath933 & @xmath93120 & @xmath9310.87 & 0.81 rgb & @xmath943.31 & 3.7 + and iii & @xmath933 & @xmath9392 & @xmath93 9.30 & 0.76 rgb & @xmath944.98 & 3.5 + ngc 185 & @xmath933 & 73 & @xmath9314.76 & 0.62 rgb & @xmath947.10 & 2.3 + ngc 147 & @xmath933 & 85 & @xmath9314.79 & 0.76 rgb & @xmath947.43 & 3.0 + and v & @xmath933 & @xmath93143 & @xmath93 8.41 & 0.81 rgb & @xmath948.03 & 2.8 + and ii & @xmath933 & 46 & @xmath93 9.33 & 0.68 rgb & 10.31 & 2.4 + m 33 & 5 & 36 & @xmath9318.87 & 0.85 cep & 14.72 & 2.0 + cas dsph & @xmath933 & @xmath93 5 & @xmath9311.67 & 0.79 rgb & 16.17 & 2.0 + ic 10 & 10 & @xmath9360 & @xmath9315.57 & 0.66 cep & 18.42 & 1.8 + peg dsph & @xmath933 & @xmath9394 & @xmath9310.80 & 0.82 cep & 19.77 & 1.7 + lgs 3 & @xmath931 & @xmath9374 & @xmath93 7.96 & 0.62 rgb & 19.87 & 1.7 + pegasus & 10 & 60 & @xmath9311.47 & 0.76 rgb & 31.02 & 1.2 + ic 1613 & 10 & @xmath9389 & @xmath9314.51 & 0.73 cep & 39.37 & 0.9 + cetus & @xmath932 & @xmath93 & @xmath9310.18 & 0.78 rgb & 52.4 & 0.5 + wlm & 9 & @xmath9310 & @xmath9313.59 & 0.92 rgb & 57.5 & 0.3 + ddo 210 & 10 & 13 & @xmath9311.09 & 0.94 rgb & 76.6 & @xmath930.1 + kkh 98 & 10 & 151 & @xmath9310.78 & 2.45 rgb & 11.23 & @xmath930.7 + kkr 25 & 10 & 68 & @xmath93 9.94 & 1.86 rgb & 74.3 & @xmath930.7 + kk 230 & 10 & 126 & @xmath93 8.55 & 1.90 rgb & 101.2 & @xmath931.0 + lrrrrlrl m 81 & 3 & 107 & 0 & @xmath9321.06 & 3.63 cep & 2.2 & + ho ix & 10 & 188 & 11 & @xmath9313.68 & 3.7 mem & 3.3 & + bk3n & 10 & 101 & 11 & @xmath93 9.59 & 4.02 rgb & 1.0 & + a0952 & 10 & 243 & 18 & @xmath9311.51 & 3.87 rgb & 1.9 & md= n3077 + kdg 61 & @xmath931 & 23 & 33 & @xmath9312.85 & 3.60 rgb & 3.9 & + m 82 & 8 & 347 & 39 & @xmath9319.63 & 3.53 rgb & 2.7 & + ngc 3077 & 10 & 153 & 49 & @xmath9317.76 & 3.82 rgb & 1.9 & + fm1 & @xmath933 & @xmath93 & 62 & @xmath9310.48 & 3.42 rgb & 1.8 & md= m82 + bk5n & @xmath933 & @xmath93 & 73 & @xmath9310.61 & 3.78 rgb & 2.4 & md= n3077 + ikn & @xmath933 & @xmath93 & 84 & @xmath9311.44 & 3.7 mem & 2.7 & + ngc 2976 & 5 & 139 & 87 & @xmath9317.10 & 3.56 rgb & 2.7 & + kdg 64 & @xmath933 & @xmath93 & 103 & @xmath9312.57 & 3.70 rgb & 2.5 & + kk 77 & @xmath933 & @xmath93 & 104 & @xmath9312.03 & 3.48 rgb & 2.0 & + f8d1 & @xmath933 & @xmath93 & 121 & @xmath9312.59 & 3.77 rgb & 2.0 & + hijass & 13 & 187 & 147 & @xmath93 7.9 : & 3.7 mem & 2.2 & + ho i & 10 & 291 & 156 & @xmath9314.49 & 3.84 rgb & 1.5 & + kdg 63 & @xmath933 & 0 & 169 & @xmath9312.12 & 3.50 rgb & 1.8 & + hs 117 & 10 & 116 & 190 & @xmath9311.83 & 3.7 mem & 1.9 & + ic 2574 & 9 & 197 & 193 & @xmath9317.46 & 4.02 rgb & 0.9 & + ddo 78 & @xmath933 & 191 & 201 & @xmath9312.17 & 3.72 rgb & 1.8 & + ddo 82 & 9 & 207 & 214 & @xmath9314.63 & 4.00 rgb & 0.9 & + bk6n & @xmath933 & @xmath93 & 304 & @xmath9311.08 & 3.85 rgb & 1.1 & + kdg 73 & 10 & 263 & 321 & @xmath9310.83 & 3.70 rgb & 1.3 & + kkh 57 & @xmath933 & @xmath93 & 373 & @xmath9310.19 & 3.93 rgb & 0.7 & + ddo 53 & 10 & 151 & 519 & @xmath9313.37 & 3.56 rgb & 0.7 & + kdg 52 & 10 & 268 & 506 & @xmath9311.49 & 3.55 rgb & 0.7 & + ho ii & 10 & 311 & 530 & @xmath9316.72 & 3.39 rgb & 0.6 & + ugc 4483 & 10 & 304 & 436 & @xmath9312.73 & 3.21 rgb & 0.5 & + ngc 2403 & 6 & 268 & 854 & @xmath9319.29 & 3.30 cep & 0.0 & + ugc 6456 & 10 & 89 & 799 & @xmath9314.03 & 4.34 rgb & @xmath930.3 & + ngc 4236 & 8 & 160 & 776 & @xmath9318.59 & 4.45 rgb & @xmath930.4 & + ddo 44 & @xmath933 & @xmath93 & 833 & @xmath9312.07 & 3.19 rgb & 1.7 & md = n2403 + ngc 2366 & 10 & 253 & 809 & @xmath9316.02 & 3.19 rgb & 1.0 & md = n2403 + ugc 7242 & 10 & 213 & 840 & @xmath9313.65 & 4.3 mem & 0.4 & md = n4236 + ddo 165 & 10 & 196 & 1075 & @xmath9315.09 & 4.57 rgb & 0.0 & md = n4236 + lrrrrlrl ngc 5128 & @xmath932 & 301 & 0 & @xmath9320.77 & 3.66 rgb & 0.6 & md = n4945 + kks 55 & @xmath933 & @xmath93 & 42 & @xmath93 9.91 & 3.6 mem & 3.1 & + kk 197 & @xmath933 & @xmath93 & 50 & @xmath9312.76 & 3.6 mem & 3.0 & + eso 324 - 024 & 10 & 270 & 102 & @xmath9315.45 & 3.73 rgb & 2.4 & + kk 196 & 10 & 490 & 138 & @xmath9312.00 & 3.6 mem & 2.2 & + ngc 5237 & @xmath933 & 131 & 143 & @xmath9315.00 & 3.6 mem & 2.1 & + kk 203 & @xmath933 & @xmath93 & 150 & @xmath9310.22 & 3.6 mem & 2.1 & + kk 189 & @xmath933 & @xmath93 & 167 & @xmath9310.52 & 3.6 mem & 2.0 & + eso 269 - 66,kk 190 & @xmath935 & 528 & 184 & @xmath9313.56 & 3.54 sbf & 1.7 & + kks 57 & @xmath933 & @xmath93 & 190 & @xmath9310.07 & 3.6 mem & 1.8 & + kk 213 & @xmath933 & @xmath93 & 214 & @xmath93 9.72 & 3.63 rgb & 1.7 & + kk 211 & @xmath935 & @xmath93 & 235 & @xmath9311.93 & 3.58 rgb & 1.5 & + eso 325 - 011 & 10 & 307 & 242 & @xmath9314.05 & 3.40 rgb & 1.1 & + kk 217 & @xmath933 & @xmath93 & 291 & @xmath9310.87 & 3.84 rgb & 1.1 & + eso 269 - 058 & 10 & 142 & 305 & @xmath9314.95 & 3.6 mem & 1.9 & md = n4945 + kks 53,cen7 & @xmath933 & @xmath93 & 312 & @xmath9310.86 & 3.6 mem & 1.2 & + eso 269 - 37,kk 179 & @xmath933 & @xmath93 & 334 & @xmath9312.02 & 3.48 rgb & 1.6 & md = n4945 + ngc 5206 & @xmath933 & 322 & 342 & @xmath9316.66 & 3.6 mem & 1.1 & md = n4945 + cen6,kk 182 & 10 & 360 & 350 & @xmath9311.89 & 3.6 mem & 1.2 & + kk 221 & @xmath933 & @xmath93 & 364 & @xmath9310.60 & 3.98 rgb & 0.6 & + cenn & @xmath933 & @xmath93 & 384 & @xmath9310.89 & 3.6 mem & 0.9 & + hipass 1351 & 10 & 292 & 388 & @xmath9310.90 & 3.6 mem & 0.9 & + ngc 5102 & 1 & 230 & 410 & @xmath9318.08 & 3.40 rgb & 0.7 & + hipass 1348 & 10 & 347 & 426 & @xmath9311.21 & 3.6 mem & 0.8 & + ngc 4945 & 6 & 296 & 468 & @xmath9320.51 & 3.6 mem & 0.7 & + kks 51 & @xmath933 & @xmath93 & 476 & @xmath9311.46 & 3.6 mem & 0.7 & + kks 58 & @xmath933 & @xmath93 & 492 & @xmath9310.64 & 3.6 mem & 0.6 & + eso 384 - 016 & 10 & 350 & 624 & @xmath9313.06 & 3.72 sbf & 0.3 & + eso 219 - 010 & @xmath933 & @xmath93 & 557 & @xmath9312.70 & 4.28 sbf & 0.1 & md = n4945 + pgc 51659 & 10 & 171 & 736 & @xmath9311.83 & 3.58 rgb & 0.1 & + eso 321 - 014 & 10 & 337 & 914 & @xmath9312.70 & 3.19 rgb & @xmath930.3 & + ngc 5236,m 83 & 5 & 304 & 0 & @xmath9320.43 & 4.47 cep & 0.8 & md = n5264 + kk 208 & @xmath933 & @xmath93 & 25 & @xmath9314.24 & 4.68 rgb & 1.6 & + pgc 47885 & 10 & 360 & 40 & @xmath9312.98 & 5.0 h & 0.4 & + eso 444 - 078 & 10 & 363 & 51 & @xmath9313.01 & 4.6 mem & 2.1 & + ngc 5264 & 10 & 268 & 80 & @xmath9315.90 & 4.53 rgb & 2.6 & + ic 4316 & 10 & 382 & 95 & @xmath9313.90 & 4.41 rgb & 2.4 & + eso 444 - 084 & 10 & 380 & 145 & @xmath9313.56 & 4.61 rgb & 1.7 & + ngc 5253 & 8 & 190 & 154 & @xmath9317.38 & 4.00 cep & 0.5 & + kk 218 & @xmath933 & @xmath93 & 167 & @xmath9310.97 & 4.6 mem & 1.6 & + ic 4247 & 10 & 195 & 181 & @xmath9314.18 & 4.6 mem & 1.5 & + kk 200 & 9 & 264 & 230 & @xmath9311.96 & 4.63 rgb & 1.2 & + deep 1337 - 33 & 10 & 371 & 279 & @xmath9311.18 & 4.51 rgb & 1.2 & + kks 54 & @xmath933 & @xmath93 & 308 & @xmath9310.47 & 4.6 mem & 1.0 & + kk 198 & @xmath933 & @xmath93 & 379 & @xmath9310.96 & 4.6 mem & 0.8 & + kk 195 & 10 & 338 & 302 & @xmath9311.76 & 5.22 rgb & @xmath930.2 & + eso 381@xmath93020 & 10 & 332 & 911 & @xmath9314.15 & 4.6 h & @xmath930.3 & + hipass 1337 & 10 & 258 & 792 & @xmath9312.27 & 4.90 rgb & @xmath930.3 & + ngc 5408 & 10 & 288 & 1002 & @xmath9316.50 & 4.81 rgb & @xmath930.5 & + eso 381 - 018 & 10 & 353 & 990 & @xmath9313.00 & 4.9 h & @xmath930.6 & + ic 342 & 5 & 245 & 0 & @xmath9320.69 & 3.28 cep & @xmath930.1 + kk 35 & 10 & 149 & 15 & @xmath9314.30 & 3.16 rgb & 2.4 + ua 86 & 8 & 275 & 91 & @xmath9318.06 & 3.12 rgb & 1.9 + ngc 1560 & 8 & 171 & 311 & @xmath9316.87 & 3.45 rgb & 1.0 + camb & 10 & 266 & 370 & @xmath9311.85 & 3.34 rgb & 1.0 + cama & 10 & 164 & 325 & @xmath9314.06 & 3.93 rgb & 0.1 + cas1 & 10 & 283 & 529 & @xmath9316.70 & 3.3 mem & 0.5 + ugca 105 & 9 & 279 & 615 & @xmath9316.81 & 3.15 rgb & 0.3 + kkh 37 & 10 & 204 & 932 & @xmath9311.55 & 3.34 rgb & @xmath930.3 + maffei 2 & 4 & 212 & 0 & @xmath9320.15 & 2.8 tf & 1.4 + maffei 1 & @xmath933 & 246 & 36 & @xmath9318.97 & 3.01 fj & 1.7 + mb 1 & 7 & 421 & 49 & @xmath9314.81 & 3.0 mem & 1.7 + mb 3 & 10 & 280 & 74 & @xmath9313.65 & 3.0 mem & 1.6 + dwing 2 & 10 & 316 & 88 & @xmath9314.55 & 3.0 mem & 1.6 + dwing 1 & 3 & 333 & 107 & @xmath9318.78 & 2.8 tf & 2.5 + kkh 12 & 10 & 303 & 146 & @xmath9313.03 & 3.0 mem & 1.4 + kkh 11 & 10 & 308 & 225 & @xmath9313.35 & 3.0 mem & 1.0 + kkh 6 & 10 & 270 & 630 & @xmath9312.42 & 3.8 rgb & @xmath930.8 + lrrrllr ngc 253 & 5 & 274 & 0 & @xmath9321.37 & 3.94 rgb & 0.3 + ddo 6 & 10 & 348 & 287 & @xmath9312.50 & 3.34 rgb & 0.5 + ngc 247 & 7 & 215 & 302 & @xmath9318.81 & 4.09 tf & 1.3 + sc 22 & @xmath933 & @xmath93 & 361 & @xmath9310.45 & 4.21 rgb & 0.9 + eso 540 - 032 & @xmath933 & @xmath93 & 362 & @xmath9311.32 & 3.42 rgb & 0.6 + kdg 2 & @xmath931 & @xmath93 & 482 & @xmath9311.39 & 3.40 rgb & 0.4 + ngc 7793 & 7 & 252 & 892 & @xmath9318.53 & 3.91 rgb & 0.1 + ddo 226 & 10 & 408 & 281 & @xmath9314.17 & 4.92 rgb & @xmath930.1 + ngc 625 & 9 & 335 & 1292 & @xmath9316.53 & 4.07 rgb & @xmath930.4 + ngc 59 & @xmath933 & 431 & 570 & @xmath9315.74 & 5.30 sbf & @xmath930.6 + eso 245 - 05 & 9 & 308 & 1480 & @xmath9315.59 & 4.43 rgb & @xmath930.7 + nfc 4736,m 94 & 2 & 353 & 0 & @xmath9319.83 & 4.66 rgb & @xmath930.5 + kk 160 & 10 & 346 & 205 & @xmath9311.52 & 4.8 h & 1.0 + ic 3687 & 10 & 385 & 240 & @xmath9314.64 & 4.57 rgb & 1.1 + ic 4182 & 9 & 356 & 340 & @xmath9316.40 & 4.70 cep & 0.6 + ngc 4449 & 9 & 249 & 367 & @xmath9318.27 & 4.21 rgb & 0.0 + ddo 126 & 10 & 231 & 435 & @xmath9314.38 & 4.87 rgb & 0.1 + kk 166 & @xmath933 & @xmath95 & 440 & @xmath9310.82 & 4.74 rgb & 0.3 + ddo 168 & 10 & 273 & 490 & @xmath9315.28 & 4.33 rgb & 0.0 + ngc 4244 & 6 & 255 & 560 & @xmath9318.60 & 4.49 rgb & 0.0 + ngc 5229 & 7 & 460 & 735 & @xmath9314.60 & 5.1 bs & @xmath930.6 + lccccccc|cc @xmath96 , mpc & 0.01 & 0.77 & 3.63 & 3.66 & 4.56 & 3.28 & 3.01 : & 3.94 & 4.09 + & & & & & & & & & + @xmath97 , mpc & 0.43 & 0.34 & 3.47 & 4.10 & 4.98 & 2.94 & 2.67 : & 3.79 & 4.17 + & & & & & & & & & + @xmath98 , mpc & 0.00 & 0.07 & 0.04 & @xmath930.33 & 0.08 & 0.02 & 0.08 & @xmath930.34 & 0.77 + & & & & & & & & & + @xmath99 & 15 & 19 & 29 & 28 & 14 & 8 & 8 : & 6 & 9 + & & & & & & & & & + @xmath100 & 10 & 13 & 11 & 18 & 4 & 0 & 1 : & 3 & 1 + & & & & & & & & & + ty(1 ) & 4 & 3 & 3 & @xmath932 & 5 & 5 & 4 & 5 & 2 + & & & & & & & & & + @xmath101(1 ) , mag & @xmath9320.80 & @xmath9321.58 & @xmath9321.06 & @xmath9320.77 & @xmath9320.43 & @xmath9320.69 & @xmath9320.15 & @xmath9321.37 & @xmath9319.83 + & & & & & & & & & + @xmath102(1 ) , km s@xmath17 & 220 & 255 & 232 & 398 & 211 & 162 & 163 & 199 & 164 + & & & & & & & & & + @xmath103(1 ) , km s@xmath17 & @xmath9388 & @xmath9335 & 107 & 301 & 304 & 245 & 212 & 274 & 353 + & & & & & & & & & + @xmath104 , km s@xmath17 & @xmath9379 & @xmath9316 & 193 & 312 & 308 & 229 & 302 & 279 & 306 + & & & & & & & & & + @xmath38 , km s@xmath17 & 76 & 77 & 91 & 105 & 71 & 54 & 59 & 54 & 56 + & & & & & & & & & + @xmath105 , kpc & 155 & 254 & 211 & 290 & 164 & 322 & 104 & 359 & 385 + & & & & & & & & & + @xmath106 & 3.28 & 6.83 & 6.11 & 5.55 & 2.31 & 3.21 & 2.69 & 5.58 & 2.00 + & & & & & & & & & + @xmath107 & 93 & 57 & 117 & 489 & 109 & 57 & 65 & 332 & 267 + & & & & & & & & & + @xmath108 & 96 & 111 & 197 & 288 & 100 & 95 & 135 & 153 & 322 + & & & & & & & & & + @xmath109 & 28 & 8 & 19 & 88 & 47 & 18 & 24 & 60 & 133 + & & & & & & & & & + @xmath110 & 29 & 16 & 32 & 52 & 43 & 30 & 50 & 28 & 161 + & & & & & & & & & + @xmath111 , gyr & 2.1 & 3.3 & 2.3 & 2.8 & 2.3 & 5.9 & 1.8 & 6.6 & 6.9 + lcccccc @xmath112 & 103 & 117 & 598 & 122 & 332 & 267 + @xmath113 & 200 & 197 & 388 & 230 & 153 & 322 + @xmath1 , mpc & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + @xmath114 & 121@xmath5338 & 169@xmath5334 & 292@xmath115 & 106@xmath5335 & 50@xmath5322 & 37@xmath5318 + @xmath116 & 10.1 & 6.1 & 7.9 & 5.8 & 5.6 & 2.0 + @xmath117 & 12 & 28 & 37 & 18 & 9 & 19 +

signatures of neutrino and antineutrino signals from stellar collapse in heavy water detectors are contrasted with those in water detectors . the effects of mixing , especially due to the highly dense matter in the supernova core , are studied . the mixing parameters used are those sets allowed by current understanding of available neutrino data : from solar , atmospheric and laboratory neutrino experiments . signals at a heavy water detector , especially the dominant charged current reactions on deuteron , are very sensitive to some of these sets of allowed mixing parameters . theoretical uncertainties on supernova neutrino spectra notwithstanding , a combination of supernova measurements with water and heavy water detectors may be able to distinguish many of these mixing possibilities and thus help in ruling out many of them . # 1#2

hep-ph0101093

neutrinos from stellar collapse have so far been detected only from the supernova sn1987a in the large magellanic cloud@xcite . the inital observations of the neutrino and antineutrino events by the kamiokande@xcite and imb@xcite collaborations were the subject of detailed analysis by several authors @xcite immediately following the event . the analyses confirmed the qualitative features of core collapse and subsequent neutrino emission . the effect of non - zero neutrino masses and mixing on supernova signals was first analysed in general by kuo and pantaleone@xcite . recently , several authors have looked at the possible signatures of neutrinos and antineutrinos from supernova collapse in realistic scenarios . dighe and smirnov@xcite have looked at the problem of reconstruction of the neutrino mass spectrum in a three - flavour scenario . these authors as also chiu and kuo@xcite have compared the signatures in the standard mass hierarchy and inverted mass hierarchy . while these papers incorporate the constraints from solar and atmospheric neutrino observations , the important question of the mass limits that may be obtained from the observation of time delay has been analysed by beacom and vogel@xcite and by choubey and kar@xcite ( see also the review by vogel@xcite ) . for a recent review which also discusses aspects of locating a supernova by its neutrinos in advance of optical observation , see ref . @xcite . in this paper we apply the analysis of neutrinos from stellar collapse presented in refs . @xcite to heavy water detectors . we had previously discussed in detail the signatures , in a water cerenkov detector , of neutrinos and antineutrinos from stellar collapse in both 3- and 4-flavour mixing scenarios . the 4-flavour analysis was motivated by lsnd data @xcite . the analysis was confined to type ii supernovae ( which occur when the initial mass of the star is larger than 8 solar masses ) . the choices of mixing parameters used were consistent with available data on solar , atmospheric and laboratory neutrino experiments . it turns out that different choices of ( allowed ) mixing parameters lead to drastically different supernova signals at water detectors . while certain features are common to both the 3- and 4-flavour analyses , there are important differences that may ( depending on the mixing parameters ) be able to distinguish the number of flavours . we will summarise the salient features of this analysis below . the dominant contribution is from the charged current ( cc ) scattering of @xmath0 on protons in water . mixing can increase the number of high energy events in this channel . the most dramatic effect of 3-flavour mixing is to produce a sharp increase in the cc events involving oxygen targets @xcite for most choices of mixing parameters . these will show up as a marked increase in the number of events in the backward direction with respect to the forward peaked events involving electrons as targets ( more than 90% of which lie in a @xmath1 forward cone with respect to the supernova direction for neutrinos with energies @xmath2 mev ) , both showing up over the mostly isotropic cc events on protons . in the absence of such mixing , there will be only a few events due to cc scattering on oxygen targets . these events involving oxygen targets will be seen in a heavy water detector as well ; realistically however , there will be fewer such events due to the considerably smaller size of the heavy water detector at the sudbury neutrino observatory ( sno ) as compared to the water detector at superkamiokande ( superk ) . when 4-flavour mixing is considered , the analysis becomes obviously more complex ; now , the increase due to oxygen events will be visible only for some of the allowed values of the parameters . furthermore , there is no set of allowed parameters in both the 3- and 4-flavour cases for which the signals in a water detector will be able to distinguish between adiabatic and non - adiabatic neutrino propagation in the core of the supernova . this is an important issue since it can place a lower bound on the ( 13 ) mixing angle , which is currently bounded by chooz at the upper end in the case of 3-flavour mixing . ( the only non - zero value for this angle so far comes from the lsnd data which can only be analysed in a 4-flavour frame - work ) . a zero value for this mixing angle will decouple the two sets of oscillations , @xmath3 and @xmath4 , and allow for a simple 2-flavour analysis separately of the solar and atmospheric neutrino data . an exactly zero value of this angle will also render irrelevant cp violating phases in the problem . there are also cc and neutral current ( nc ) events due to elastic scattering with electrons in water / heavy water . these interactions are the same in both detectors and have been discussed in detail in the earlier analysis on water detectors @xcite . in this paper we focus on the possible signatures of a supernova collapse in a heavy water detector . the question is of practical interest since sno is operating for more than a year now . it turns out that a combination of measurements in water and heavy water detectors has much better discriminatory power than either of them individually . hence such observations of supernova neutrinos may be a good signal to rule out some of the currently allowed parameter space in the neutrino mixing angles . ( these signals are however not very sensitive to neutrino mass squared differences . ) the most interesting signals to study at a heavy water detector are the events from cc interactions of both @xmath5 and @xmath0 on deuterons . this is the most dominant signal in contrast to the dominant cc interaction of @xmath0 alone on free protons in a water detector ( all are typically about two orders of magnitude larger than those from oxygen or electron interactions ) . we will focus mostly on these events in this paper , besides making a few remarks on the nc events on deuteron that can also easily be measured at a heavy water detector . these are therefore the `` new '' signals in a heavy water detector that would not be observable in ( or will be different from ) a water detector . as before the analysis is done assuming the standard mass hierarchy necessitated by the solar and atmospheric neutrino observations @xcite . we also impose all the known constraints on the mixing parameters and mass - squared differences including those constraints from laboratory experiments@xcite . the purpose of the calculation is to see whether different mixing scenarios give significantly different signals in the detector . in sect . ii we briefly outline the frame - work including the mixing matrix as also matter effects on mixing . we then use this to obtain expressions for the @xmath5 , @xmath6 and @xmath7 , @xmath8 fluxes reaching the detector in both 3- and 4-flavour mixing schemes . while discussions may be found in ref . @xcite we reproduce the relevant details to keep this paper self - contained . in sect . iii we list the different interaction processes relevant to both water and heavy water detectors , in particular , those with electrons ( positrons ) in the final state , as well as nc events on deuteron . the details of the supernova model used in the calculation as well as the allowed values of neutrino mixing parameters ( from already existing data ) using which the effects of mixing are computed are also listed here . results showing the effects of mixing both for the spectrum and the integrated number of events are presented in sect . sensitivity to the supernova model parameters is discussed . the nc events on deuterons in a heavy water detector are also briefly discussed here . all these results are for a @xmath9 mass hierarchy scheme in the case of 4 flavour mixing . in sect . v , we briefly discuss the other possible scenario , that of the @xmath10 scheme in 4-flavours . we present a summary and discussions in sect . we briefly review mixing among three and four flavours of neutrinos ( or antineutrinos ) and compute the neutrino ( antineutrino ) survival and conversion probabilities . these are given in more detail in refs.@xcite . the supernova neutrinos are produced mostly in the core , where the matter density is very high . hence matter effects on the propagation are important . the hierarchy of mass eigenstates is shown in fig . [ fig : level ] . the 3-flavour case is shown in fig . [ fig : level]a ; there are essentially two scales corresponding to the solution of solar and atmospheric neutrino problems assuming neutrino oscillations . in the case of four flavours ( necessitated by the non - zero result of lsnd @xcite ) , the analysis is more complicated since there is an additional sterile neutrino . one possible mass hierarchy here is a @xmath9 scenario @xcite as shown in fig . [ fig : level]b . the lsnd result implies the existence of a mass scale in the range of 0.1 ev@xmath11 to 1 ev@xmath11 . we choose two doublets separated by this mass scale . the intra - doublet separation in the lower doublet corresponds to the solar neutrino scale @xmath12 ev@xmath13 and that in the upper one to the atmospheric neutrino scale @xmath14 ev@xmath13 . yet another possibility is the so - called @xmath10 scheme @xcite , where the 3-flavour scheme is extended by adding a heavy sterile neutrino as the heaviest mass eigenstate , with a mass separation from the other three states as required by lsnd . we discuss this scheme separately later . in what follows , therefore , 4-flavour mixing always refers to the @xmath9 scheme . the mixing matrix , @xmath15 , which relates the flavour and mass eigenstates in the above scenarios has three angles in the 3-flavour case and six angles in the 4-flavour one . we shall ignore the cp violating phases . then , the mixing matrix can be parametrised in the case of 3-flavours as , @xmath16^t = u \times \left [ \nu_1 \quad \nu_2 \quad \nu_3 \right ] ^t~,\ ] ] where @xmath17 stands for transpose . here @xmath15 is parametrised by considering rotations of mass eigenstates taken two at a time : @xmath18 where @xmath19 corresponds to the rotation of mass eigenstates @xmath20 and @xmath21 . as is the convention , we denote the mixing angle relevant to the solar neutrino problem by @xmath22 and that relevant to the atmospheric neutrino problem by @xmath23 in the case of 3 flavours . the ( 13 ) angle @xmath24 is then small , due to the chooz result @xcite which translates to the constraint on the ( 13 ) mixing angle , @xmath25 , where @xmath26 . in the case of 4-flavours , we choose to work in the @xmath9 scheme : @xmath27^t = u \times \left [ \nu_1 \quad \nu_2 \quad \nu_3 \quad \nu_4 \right ] ^t~,\ ] ] where the mixing matrix is similarly defined to be , @xmath28 here the ( 13 ) and ( 14 ) angles are constrained by chooz to be small : @xmath29 @xcite . the atmospheric neutrino problem constrains @xmath30 ( the equivalent of the angle @xmath23 in the 3-flavour case ) to be maximal , when both @xmath31 become small . we shall assume they are also limited by the same small parameter , @xmath32 . the other relevant definition is the chosen mass hierarchy in the problem . we define the mass squared differences as @xmath33 , where @xmath34 run over the number of flavours . without loss of generality , we can take @xmath35 , @xmath36 ( and @xmath37 ) to be greater than zero ; this defines the standard hierarchy of masses consistent with the range of the mixing angles , as specified above . the most important consequence of the highly dense core matter is to cause mikheyev , smirnov and wolfenstein ( msw ) resonances @xcite in the neutrino sector . ( the chosen mass hierarchy prevents such resonances from occurring in the antineutrino sector ) . in both the 3- and 4-flavour cases , this causes the electron neutrino to occur essentially as a pure mass state @xcite , in fact , as the highest mass eigenstate , as can be seen from the schematic illustration in fig . [ fig : nu34 ] . nonadiabatic transitions near the resonances will alter this result , especially in the case of 4-flavours , where there are several level crossings because of the presence of the sterile neutrino . hence we will discuss the purely adiabatic and the non - adiabatic cases separately . we begin with the 3-flavour case . the average transition probability from flavour @xmath38 to flavour @xmath39 is denoted by @xmath40 where @xmath41 run over all flavours . it turns out that , independent of other mixing angles and the mass squared differences , the survival probability for the adiabatic case is , @xmath42 and hence is small . this is sufficient to determine all the relevant fluxes in the 3-flavour case . in the 4-flavour adiabatic case , we again have , @xmath42 so that the electron neutrino survival probability is flavour independent in the adiabatic case . in addition , we also need to know some other probabilities in order to compute the fluxes at the detector . we have @xmath43 because of the parametrisation it is easy to see that non - adiabatic effects in the form of landau zener ( lz ) jumps are introduced as a result of the values chosen for @xmath44 and @xmath22 . the value of @xmath44 determines whether non - adiabatic jumps are induced at the upper resonance(s ) while the value of @xmath22 determines whether the non - adiabatic jump occurs at the lower resonance . this statement holds both for three and four flavours since in both cases the non - adiabaticity in the upper resonance(s ) is controlled by @xmath44 , apart from mass squared differences . for a large range of @xmath44 , allowed by the chooz constraint , the evolution of the electron neutrino is adiabatic . as a result the lower resonance does not come into the picture at all except when @xmath44 is very close to zero , where the lz jump probability at the upper resonance(s ) , @xmath45 , abruptly changes to one @xcite . this occurs in a very narrow window : for instance , when @xmath44 changes from 0.02 to 0.01 @xmath45 changes from 0.01 to 0.34 for a mass scale in the range of @xmath46 ev@xmath13 . the subsequent discussion is therefore relevant only when @xmath44 is vanishingly small , @xmath47 . ( this is not ruled out by the known constraints except lsnd ) . the lz transition at the lower resonance is determined by the probability , @xmath48 , which is a function of the mixing angle @xmath22 and the mass squared difference @xmath35 . it turns out that @xmath48 is zero unless @xmath22 is small , @xmath49 for mass differences in the solar neutrino range @xmath50 ev@xmath13 . this case ( of small @xmath44 , small @xmath22 ) corresponds to the extreme non - adiabatic limit . when @xmath22 is large , as in the case of the large - angle msw or the vacuum solution to the solar neutrino problem , non - adiabaticity occurs only at the upper resonance and is therefore partial . in our calculations we have used the form for @xmath45 and @xmath48 discussed , as @xmath51 $ ] , where @xmath52 . the final expression for @xmath53 given in appendix b of ref . @xcite should be multiplied by the additional factor @xmath54 . however , the numerical calculation in that paper is correct . ] in the appendix of ref.@xcite ( see also @xcite ) . because of the sharpness of the transition at the upper resonance we will set @xmath55 for small enough @xmath44 and only consider the dependence of the survival probability on the lz transition at the lower resonance . in the 3-flavour non - adiabatic case the only relevant probability is , @xmath56~.\ ] ] in the non - adiabatic 4-flavour case we have , @xmath57~ , \nonumber \\ p_{es}&= & ( 1 - 2 \epsilon^2)[(1-p_l)\cos^2 \omega~ + p_l \sin^2\omega]~. \label{eq : nonad4}\end{aligned}\ ] ] since @xmath44 is small , the flux at the detector is entirely controlled by @xmath22 and @xmath48 which is also a function of @xmath22 . note that the sum @xmath58 is independent of @xmath48 in 4-flavours . since @xmath59 , this also indicates that the probability of transition of @xmath7 into @xmath5 is small in the non - adiabatic case , in contrast to the adiabatic case where @xmath60 . due to our choice of mass hierarchy , the propagation of antineutrinos is always adiabatic ( the matter dependent term has the opposite sign as compared to neutrinos ) . here also the high density of matter in the core causes the electron antineutrino to be produced as a pure mass eigenstate , @xmath61 . in the 3-flavour case , we have @xmath62 in the 4-flavour case , we have , @xmath63 hence , when @xmath22 is small , there is very little loss of the @xmath0 flux into the sterile channel . also , the sum @xmath64 is similar to the non - adiabatic neutrino propagation so that very little @xmath65 is converted to @xmath0 . for the nc events , we will also need the following : @xmath66 these probabilities can then be used to determine the observed antineutrino fluxes . following kuo and pantaleone @xcite , we denote the flux distribution , @xmath67 , of a neutrino ( or antineutrino ) of flavour @xmath39 with energy @xmath68 produced in the core of the supernova by @xmath69 . in particular we use the generic label @xmath70 for flavours other than @xmath5 and @xmath71 since @xmath72 typically , models for supernovae predict that the @xmath0 and @xmath73 have hotter spectra than @xmath5 . this is because the @xmath73 decouples first since it has only nc interactions with matter and therefore leaves the cooling supernova with the hottest thermal spectrum . the cc interactions of @xmath5 with matter are larger than those of @xmath0 and hence @xmath5 has the coldest spectrum . the model on which this work is based @xcite predicts that the average energies of the @xmath5 , @xmath0 and @xmath73 spectra are around 11 , 16 , and 25 mev . we make our observations , keeping this in mind . the @xmath5 flux on earth is given in terms of the flux of neutrinos produced in the core of the supernova by , @xmath74 where we have made use of the constraint @xmath75 . this flux is further reduced by an overall geometric factor of @xmath76 for the case of a supernova at a distance @xmath77 from the earth . since the probabilities @xmath78 and @xmath79 are known , the @xmath5 flux can be computed in terms of the original fluxes emitted by the supernova . the @xmath5 flux is independent of the ( 12 ) mixing angles @xmath22 in the adiabatic case . also , it is not very different for 3- and 4-flavours . in both , the observed flux is almost entirely due to the original @xmath73 flux since @xmath80 is small , and is therefore hotter . from eq . ( [ eq : nonad4 ] ) we see that , in contrast , the contribution from the hotter spectrum into electron neutrinos in the 4-flavour non - adiabatic case is controlled entirely by @xmath44 and is small . however , the observed flux can be depleted , depending on the value of @xmath48 . the signal in the 3-flavour case is drastically different because the contribution of the hotter spectrum now depends on @xmath44 , @xmath22 and @xmath48 . in general , the possibility of lz transitions makes the analysis more complicated . we will discuss this case numerically later . the result for @xmath0 is the same , with @xmath78 replaced by @xmath81 , etc . for example , @xmath82 there is hardly any mixing of the hotter spectrum into electron antineutrinos in the 4-flavour case since @xmath83 is small . the extent of mixing in 3-flavours depends on the value of @xmath22 through @xmath81 . for small @xmath22 , there is very little change in the electron antineutrino flux ( see eq . ( [ eq : pbar4 ] ) ) . the results are summarised in tables [ tab : adnu ] and [ tab : adnubar ] for the adiabatic neutrino and antineutrino cases , and in table [ tab : nonad ] for non - adiabatic neutrino propagation . expressions for the other flavours , needed to compute the nc events , are also given in these tables . for instance , @xmath84 a similar expression holds for @xmath85 with @xmath86 replaced by @xmath87 . while the neutral current ( nc ) combination , @xmath88 , remains unaltered in 3-flavours , there may be actual loss of spectrum into the sterile channel in the 4-flavour case . this should also be a good indicator of the number of flavours involved in the mixing . water as well as heavy water detectors will be sensitive to all these aspects of mixing . in the next section , we will discuss the inputs and constraints , both from supernova models as well as current neutrino experiments . these will then be used subsequently to predict numerically supernova event rates . we will concentrate mostly on the dominant cc and nc interactions of @xmath5 and @xmath0 on deuteron in heavy water . we will also compare the cc interactions to those at a water detector , that is , to @xmath0 on protons . interactions on electrons and oxygen nuclei in water and heavy water detectors are the same , and have been discussed in detail in refs . @xcite . both in water and heavy water detectors , the interactions we are mainly interested in are of three types : 1 . dominant events : these are due to the cc @xmath89 interaction in water . in heavy water , there are contributions from both @xmath90 and @xmath91 cc processes . at the neutrino energies relevant to this discussion , the cross - sections for these processes , and hence the number of events , are about two orders of magnitude larger than any of the others . 2 . electron events : these are due to elastic scatterings of @xmath5 , @xmath0 , @xmath73 , @xmath92 on electron targets in both water and heavy water . oxygen events : these arise from cc scattering of @xmath5 and @xmath0 on oxygen nuclei in both detectors . in all these cases , the events are identified by detecting an electron ( or positron ) in the final state . the electron events , although small in number , are highly forward peaked @xcite . the oxygen events have a high energy threshold ( @xmath93 mev ; @xmath94 mev ) and can only occur if there is substantial mixing of the hotter @xmath73 spectrum with the @xmath5 or the @xmath0 in which case these events are highly backward peaked @xcite . both these may therefore be readily separated from the mostly isotropic dominant events in water @xcite . in the case of heavy water , the angular distribution of both the dominant processes is well - known @xcite and is approximately given by @xmath95 at the energies of interest . here @xmath96 is the electron ( positron ) laboratory scattering angle . so there are typically twice as many events in the backward direction as in the forward direction . this makes the exclusive identification of oxygen events more difficult in a heavy water detector . a note on the nomenclature `` forward '' and `` backward '' . we envisage the angular dependence as being measured in typically six bins of @xmath97 each , the first bin corresponding to `` forward '' and the last one corresponding to the `` backward '' events . this obviates the need for very accurate angle measurements as well as takes into consideration effects like electron - rescattering that may smear out the scattering angle . this bin size is also typically what is available at superkamiokande . since the supernova signal is a short and well - defined signal ( lasting about 10 secs ) , it will be possible to detect all these events over background ( due to solar and other radioactive processes ) . for instance , in the relevant energy region , superkamiokande reports @xcite a background of 0.2 events / day / kton in a bin of 1 mev of scattered electron energy from the direction of the sun , and half that rate from all other scattering angles . the corresponding number for sno is not available , although the radioactive background is expected to be small compared to the solar signal @xcite . furthermore , since their angular distribution is well - known , angular information on the final state electron ( @xmath98 or @xmath99 ) may allow us to separate out these three types of events . the detailed analysis of signals due to mixing in the forward and backward event samples is given in ref.@xcite for a water cerenkov detector . these results also hold for a heavy water detector ( apart from a scaling factor of 0.9 due to the mass difference between water and heavy water ) . we will not discuss this further and concentrate on the new results for the cc events on deuterons . in addition , there are nc processes on the deuteron in heavy water that may be detected by the subsequent neutron capture and the associated gamma rays . the specific processes we will consider , for water and heavy water detectors , therefore are @xmath100 and @xmath101 common to both detectors are the processes , @xmath102 where the elastic scattering on electrons involves both cc and nc interactions . we will also discuss the interesting possibility of observing nc events on deuteron in a heavy water detector : @xmath103 the cross - sections for eqs . ( [ cr1])([cr3 ] ) are well - known @xcite . the @xmath104 cross - section is large in water cerenkov detectors , being proportional to the square of the antineutrino energy . in terms of total number of events , therefore , water cerenkov detectors are mostly dominated by @xmath105 events . the deuteron cc cross sections are comparable though the @xmath5 cc reaction on deuteron has a slightly lower threshold and a somewhat larger cross - section than the @xmath0 cc . also , the @xmath0 cc cross - section in heavy water is about 4 times smaller than the corresponding one in water due to pauli suppression . when the recoil electron ( positron ) is detected , the time integrated event rate due to neutrinos ( or antineutrinos ) of flavour @xmath39 and energy @xmath68 on target @xmath17 , as a function of the recoil electron ( or positron ) energy , @xmath106 , is as usual given by , @xmath107 the index @xmath108 refers to the time interval within which the ( original ) thermal neutrino spectrum can be assumed to be at a constant temperature @xmath109 . here @xmath110 refers to the number of scattering targets ( of @xmath77 , @xmath111 , @xmath112 , or @xmath113 ) that are available in the detector . also , for processes involving @xmath77 and @xmath111 , the hadron recoil is so small that we assume @xmath114 , where @xmath115 is the mass difference between the initial and final hadrons ( including the binding energy ) . while this results in a small threshold of a few mev in these cases , the threshold for oxygen processes is greater than 10 mev . the total number of events from a given flavour of neutrino in a given bin , @xmath116 , of electron energy ( which we choose to be of width 1 mev ) then is @xmath117 in the case of nc events in heavy water , the total number of events is calculated according to @xmath118 here again the cross - section is well - known @xcite . as in @xcite , we compute the time integrated event rate at prototype 1 kton water and heavy water detectors from neutrinos emitted by a supernova exploding 10 kpc away . results for any other supernova explosion may be obtained by scaling the event rate by the appropriate distance to the supernova and the size of the detector , as shown in eq . ( [ eq : rate ] ) . we assume the efficiency and resolution of the detectors to be perfect ; this will only slightly enhance the event rates near the detector threshold @xcite . we use the luminosity and average energy distributions ( as functions of time ) for neutrinos of flavour @xmath39 and energy @xmath68 as given in totani et al . @xcite , based on the numerical modelling of mayle , wilson and schramm @xcite . in a short time interval , @xmath119 , the temperature can be set to a constant , @xmath120 . then the neutrino number flux can be described , in this time interval , by a thermal fermi dirac distribution , @xmath121 at a time @xmath122 after the core bounce . here @xmath108 refers to the time - bin , @xmath123 . we set the time of bounce , @xmath124 . the overall normalisation , @xmath125 , is fixed by requiring that the total energy emitted per unit time equals the luminosity , @xmath126 , in that time interval . the precise values for @xmath126 and @xmath120 are taken from ref . @xcite . the total emitted energy in all flavours of neutrinos is about @xmath127 ergs . the general features of the model are as follows : the temperature is roughly constant over the entire period of emission ( lasting roughly 10 seconds ) . typical values are @xmath128 , with the average energy of each flavour , @xmath129 mev for @xmath130 and @xmath73 respectively . also , the total emitted energy is more or less equally distributed in all flavours . ( the luminosity of @xmath5 is higher than that of other flavours at early times , while that of @xmath73 is higher after 1 sec ) . the number of neutrinos emitted in each flavour , however , is not the same since their average energies are different . while we use the values for the time dependent temperature and luminosity as given by ref . @xcite in our analysis , we also examine the effects due to possible variations of these parameters and hence the sensitivity of our results to the details of the supernova model used . with large matter effects present in both the neutrino and antineutrino sector , the validity of the average energies of @xmath5 , @xmath0 and @xmath73 as 11 , 16 and 25 mev , respectively , which were calculated without mixing , may be questioned . this is especially so because re - scattering effects involve the flavour states which may then equilibrate at different temperatures . we however note that the highly dense matter projects the @xmath5 and @xmath0 states as almost pure mass eigenstates . hence thermalisation is not affected by effects of mixing . in fact , the effects of mixing are significant only when the resonant densities are reached , when the msw effect can mix different flavour states . for the parameter values as allowed from current neutrino data , this occurs only outside the neutrinosphere ( @xmath131 km ) , and not at the core ( @xmath132 km ) where most of the neutrinos are produced . this mixing therefore occurs between spectra which are already thermalised with the above - mentioned temperatures . in the case of @xmath133 and @xmath134 this argument does not go through . in the highly dense core , these are mixtures of more than one mass eigenstate . however , both mix only into each other and scatter through exactly the same processes . hence their temperatures also remain the same as in the no - mixing case . we impose the following known constraints on the mixing matrix in vacuum both for three and four flavour scenarios . consistent with the chooz constraint , namely @xmath135 which we have imposed at the level of the parametrisation itself , we choose @xmath136 for the 3-flavour adiabatic case and @xmath137 for the non - adiabatic case . in the case of the 4-flavour scheme , we set @xmath138 . the constraint from the atmospheric neutrino analysis implies that the relevant angle @xmath139 , is near maximal and the relevant mass squared difference is of the order of @xmath46 ev@xmath13 . neither of these constraints directly enter our calculations except to determine whether the upper resonance is adiabatic or not depending on the value of @xmath44 as constrained by the chooz findings . we consider both possibilities here . the value of the ( 12 ) mixing angle , @xmath22 , is not yet known . combined data on solar neutrinos give three possible values @xcite : 1 . @xmath140 ev@xmath141 ( sma ) . the small angle msw solution . @xmath142 ev@xmath141 ( lma ) . the large angle msw solution . @xmath143 ev@xmath141 ( lma - v ) . the large angle vacuum solution . this has been slightly modified @xcite in view of new data from superk ; however , it remains true , in general , that @xmath22 may be small or large , with @xmath144 ev@xmath13 . we will now discuss the results numerically for all these choices . the numerical calculations are done by following the evolution of the mass eigenstates through all the resonances including the appropriate jump probabilities when the transition is non - adiabatic ( when @xmath44 is small ) . furthermore , the lz jump at the lower resonances is significant only for small values of @xmath22 . the cc event rates on deuteron computed using the above inputs are displayed in figs . [ fig : nuead][fig : totaln ] . the time - integrated event rates per unit electron energy bin ( of 1 mev ) are shown as a function of the energy @xmath106 of the detected electron . the solid lines in all the figures refer to the case when there is no mixing and serve as a reference . results for 3- and 4-flavour mixing are displayed in each of these figures as dotted and dashed lines respectively . [ fig : nuead ] shows the predictions for the ( @xmath22-independent ) adiabatic @xmath145 cc interaction when @xmath136 . mixing enhances the high @xmath106 event rates for both 3- and 4-flavour mixing , which can not be distinguished here . furthermore , mixing shifts the peak of the spectrum to higher energies , from 15 mev to about 28 mev because of the admixture of the hotter @xmath73 spectrum and its significantly higher cross - section with deuterium . this high energy shift should be clearly observable . nonadiabaticity at the upper resonance occurs when @xmath146 . then the adiabaticity at the lower resonance is determined by the value of @xmath22 . in fact , @xmath147 unless @xmath22 is small , @xmath49 . hence this is relevant only for the small angle msw solution ( sma ) . we show the dependence of @xmath48 on the neutrino energy for the small-@xmath22 3-flavour case in fig . [ fig : pl ] . the 4-flavour result is similar , with a scale factor roughly 0.7 . it is seen that @xmath48 increases with energy , although it still does not reach unity for the relevant supernova neutrino energies . for larger @xmath22 , that is , for the large angle msw ( lma ) and the vacuum ( lma - v ) solutions , @xmath147 . in fig . [ fig : nuenonad ] we show the electron spectrum for the non - adiabatic @xmath148 cc interaction when @xmath149 is small , in fact near zero . the fully non - adiabatic case corresponding to small @xmath22 is shown in comparison with that for a larger value of @xmath22 in the figure . here , 3- and 4-flavour mixing give drastically different results , but the small-@xmath44 scenario is in general not very sensitive to the chosen values of @xmath22 . [ fig : nuebar ] we show the positron spectrum due to @xmath150 cc interactions for two different choices of @xmath22 . ( here @xmath136 , but the results are essentially the same even if @xmath44 is nearly zero since this sector is always adiabatic . ) mixing has appreciable effects only for large @xmath22 ; however , mixing does not affect the peak position , unlike in the adiabatic @xmath5 case . we would like to point out that the pure @xmath5 spectrum may not be observable in heavy water . it may be separated out from the total events sample if the @xmath0 spectrum can be reliably separated out by various detection techniques such as looking for two neutrons in coincidence with the positron . however , we will show that the total number of events will still be sensitive to the mixing parameters . we will therefore discuss both the total as well as the individual @xmath5 and @xmath0 spectra . the total event rates due to the dominant cc @xmath148 and @xmath151 processes are shown in figs . [ fig : total ] and [ fig : totaln ] for the two choices of @xmath44 . the total isotropic cc events in water , due to @xmath152 alone , are also shown for comparison . the most significant difference between the two is that of the adiabatic case with small @xmath22 ( the lower two panels of fig . [ fig : total ] ) which is independent of the number of flavours . this is because of the enhancement in the @xmath5 events , independent of @xmath22 . at all @xmath22 , the peak is at a higher energy than expected from the no - mixing case in heavy water but remains the same for a water detector . this shift may be sufficiently significant and therefore observable , in the adiabatic scenario , for all @xmath22 . finally , the upper two panels of fig [ fig : total ] indicate that a significant depletion in the observed events in water , together with an enhanced number of events in heavy water is an unambiguous signal of 4-flavour mixing with large @xmath22 ( dashed lines in fig . [ fig : total ] ) . the corresponding results for the dominant cc events in the non - adiabatic case are shown in fig . [ fig : totaln ] . here there is no significant shift in the spectral peak . also , the signals in water and heavy water are very similar , with the signals being either enhanced or depleted similarly in both . for instance , the large @xmath22 4-flavour signal shows depletion both in water and heavy water . this may be difficult to distinguish from the no - mixing case if the overall normalisation of the supernova spectrum is uncertain by more than a factor of two . in all cases , the small-@xmath44 , small-@xmath22 scenario also can not be distinguished from the no - mixing case . hence the small-@xmath44 case may be difficult to establish unambiguously , independently of the supernova model inputs . keeping in mind that the supernova dynamics may have large uncertainties , we will later also analyse the ratios of the total number of events in water and heavy water detectors . these are likely to be less sensitive to the supernova models ( although they do depend on the temperature hierarchy for @xmath5 , @xmath0 and @xmath73 ) and hence may be more robust signals of mixing . the predicted time integrated number of events resulting in a scattered electron with energy , @xmath153 mev ( which is a typical threshold for cerenkov detectors ) , are shown in tables [ tab : ccad ] and [ tab : ccnonad ] , for the adiabatic and non - adiabatic cases respectively . as before , the number is calculated assuming a supernova explosion at 10 kpc for a 1 kton detector . listed are the dominant cc events on deuterons in heavy water and free protons in water , along with the elastic scattering events on electrons and the cc events on oxygen nuclei . for heavy water , we have listed the individual contributions from @xmath5 and @xmath0 on deuteron . in water , the corresponding dominant events are from @xmath0 on @xmath111 . the elastic events are from @xmath154 , @xmath155 , @xmath156 , and @xmath157 . since they will all be detected in the extreme forward direction , they have been summed up and listed as total @xmath158 events in the tables . the oxygen events , listed as @xmath159 , include both @xmath160 and @xmath161 cc events , which will predominantly be in the backward direction , especially when enhanced by mixing . in particular we tabulate the events for 3- and 4-flavour mixing when @xmath22 is both small and large . in table [ tab : ccad ] , we show the results for the fully adiabatic case when @xmath162 . it is seen that the bulk of the events ( more than 90% ) are from the dominant events on @xmath111 or @xmath77 . mixing always enhances the @xmath148 channel by more than a factor of two ; hence adiabatic propagation always predicts an enhanced rate of total events in heavy water even though there is reduction in the @xmath151 channel ( the other dominant process ) for some parameters . in contrast , the total number of events in water may even go down as compared to the no - mixing case , depending on the parameter values . in table [ tab : ccnonad ] , we show similar predictions for the case when @xmath146 or when the upper resonance(s ) become fully non - adiabatic . the value of @xmath22 then determines whether or not the lower resonance is adiabatic . recall that the antineutrino propagation is always adiabatic . again , the contribution from the @xmath158 and @xmath163 events is small compared to the dominant events . small changes in the @xmath152 and @xmath151 events between tables [ tab : ccad ] and [ tab : ccnonad ] are due to changes in the value of @xmath44 . irrespective of the parameter values , it is seen that there is never any depletion in the 3-flavour case . as in fig . [ fig : totaln ] an interesting scenario occurs when @xmath22 is large in the 4-flavour case . here the total number of dominant events in both @xmath145 and @xmath151 in heavy water and in @xmath105 in water , is reduced by a factor proportional to @xmath164 . this is the only scenario where there is depletion in both water and heavy water . so far , we have assumed that most of the mixing parameters are known and used the supernova measurement as a potential check for self - consistency of the model parameters . this is because there is still very little known about the supernova neutrino spectrum through observations and hence there is both theoretical and experimental uncertainty about the details of the neutrino spectrum . however , it is still instructive to actually turn the question around and ask , suppose another supernova explosion is observed through its neutrino emission . will an excess over ( or a depletion from ) the expected number of events unambiguously determine some of the model mixing parameters ? the answer to this question can be obtained from table [ tab : r ] . it is seen that certain classes of models may be ruled out , depending on the observation . we define the ratio of the total number of events ( from all possible interactions with an electron ( or positron ) in the final state ) potentially observed from a future supernova ( equivalently , the prediction from a given neutrino mixing model ) to the expected number of events without mixing ( for @xmath153 mev ) : @xmath165 where @xmath166 refer to 1 kton heavy water and water detectors respectively . the denominator refers to the the expected number of events ( using a standard supernova model , with no mixing ) as computed from a monte carlo simulation that takes into account detector resolution , efficiency , etc . , consistent with the detector at which the events were observed . an observation may find @xmath167 , @xmath168 or @xmath169 . note that even though the ratio refers to the total number of events , the inferences drawn reflect mainly the behaviour expected from the dominant cc processes on protons in water and deuterons in heavy water . the mixing models ( with model parameters @xmath44 and @xmath22 , including adiabaticity ) consistent with , or predicting , such an observation are shown in table [ tab : r ] . here the non - adiabaticity , that is , the value of @xmath48 at the lower resonance has been computed assuming a typical value of @xmath170 ev@xmath13 . water detectors can not distinguish adiabatic ( a ) and non - adiabatic ( n ) scenarios , that is , whether or not @xmath44 is different from zero , but can distinguish the number of flavours when @xmath22 is large ( see the last column of table [ tab : r ] ) . in @xmath171 , however , most models predict @xmath167 . the _ only _ observation of @xmath168 in @xmath171 occurs for the 4-flavour non - adiabatic case with @xmath146 and large @xmath22 . on combining data from water and heavy water detectors , an improved discrimination of model parameters is possible , as can be seen from table [ tab : rdh ] . here the different values of @xmath172 and @xmath173 are listed , along with the models that are consistent with such a combined observation . first of all , it is seen that combining the two measurements immediately allows for a separation between the adiabatic and non - adiabatic cases and hence whether @xmath44 is different from zero , except in the 3-flavour case with large @xmath22 . it should be noted that the @xmath174 scenario is unlikely to be distinguished from the no - mixing case . it is also seen that certain combinations of @xmath173 and @xmath172 do not occur for any of the allowed parameters values . for instance a depletion in heavy water is _ only _ consistent with depletion in water as well . any other result in water indicates that the overall normalisation of the spectrum is probably in error . an occurrence of such `` forbidden '' combinations may therefore can be used as a check on the overall normalisation of the supernova spectrum . this result of course is limited to the class of models we are analysing here . the following scenario is best suited to determining the value of @xmath22 . ( 1 ) there are fewer isotropic events than expected ( @xmath175 ) in a water cerenkov detector such as superk . this reduction factor determines @xmath164 . ( 2 ) the _ same _ reduction factor ( @xmath176 ) fits the data from a heavy water detector such as sno . this can imply that the correct mixing matrix is one with 4 flavours and large @xmath22 , with non - adiabatic neutrino propagation . ( 3 ) if on the other hand there are enhanced number of events ( @xmath177 ) at the heavy water detector , it clearly indicates adiabatic neutrino propagation . this in turn implies that @xmath44 is different from zero which has so far been claimed only by lsnd . variation of supernova input parameters , which we will discuss in the next section , does not alter this result . these qualitative features can be quantified by defining the double ratio , @xmath178 which is independent of the overall normalisation of the neutrino flux and hence provides a better diagnostic . in practice , it may not be possible to directly take a ratio of the data from water and heavy water detectors since the two measurements will differ in their systematics , apart from such considerations as detector efficiency and resolution . since @xmath179 are normalised to the theoretical expectancy including these considerations , @xmath180 is not likely to be sensitive to details of detector design and can thus provide a robust , quantitative indicator of different types of mixing . this double ratio ( where @xmath181 has been calculated as before ) has been shown in figs . [ fig : newr ] and [ fig : newrc ] as a function of @xmath22 . the ratio is plotted for the total number of events with the cut on the observed electron ( positron ) energy , @xmath182 mev in fig . [ fig : newr ] . the two curves in each figure correspond to the two values of @xmath44 when the propagation at the upper resonance(s ) is purely adiabatic , @xmath136 ( solid lines ) or purely non - adiabatic , when @xmath183 ( dashed lines ) . these are the two cases that a water detector is normally not able to resolve . also shown are dotted vertical lines corresponding to the solutions allowed by the solar neutrino problem : @xmath184 and @xmath185 . non - adiabaticity at the lower resonance has been computed using @xmath170 ev@xmath13 as before . while the 3-flavour mixing case is shown on the left , the 4-flavour result is plotted on the right . obviously , a value of unity is expected for the case of no - mixing . we analyse each case in turn . 1 . we see from fig . [ fig : newr ] that the double ratio @xmath180 is always strictly greater than one for the adiabatic case , independent of @xmath22 or the number of flavours , @xmath186 . 2 . even in the non - adiabatic case , it can be less than one only when @xmath187 . note however that currently allowed values of @xmath22 lead to @xmath188 in the non - adiabatic case . the case @xmath189 occurs only for intermediate values of @xmath22 . 3 . for @xmath190 ( left panel of fig . [ fig : newr ] ) , the double ratio at small @xmath22 is different for the adiabatic and non - adiabatic cases , which may therefore be distinguished . however , it may be very difficult to distinguish these two for large values of @xmath22 . 4 . on the other hand , these two cases are easily distinguished for all @xmath22 for @xmath187 as can be seen from the right panel in the figure . while the double ratio is similar for both @xmath191 for small @xmath22 , @xmath192 , the number of flavours can be distinguished for larger values of @xmath22 , especially in the currently allowed region , only in the adiabatic case . . however , as stated before , independent of @xmath186 , the small @xmath22 non - adiabatic solution with @xmath146 can not be distinguished from the no - mixing case . keeping in mind that a thermal neutrino flux distribution such as the one we have used may overestimate the high energy spectrum , the case @xmath193 ( mev ) @xmath194 is shown separately in fig . [ fig : newrc ] . most of the features survive the cuts ; hence this ratio is likely to be a stable indicator of mixing . finally , we note that the denominator of the double ratio is dominated by @xmath0 events . hence , the _ same _ discriminatory power can be achieved using data from a heavy water detector _ alone _ if it is possible to separate the @xmath148 and @xmath151 events . as stated earlier , this may be possible , for instance , at sno , by detecting both the neutrons in coincidence with the positron emitted in the @xmath151 interaction . sno is also planning to increase the neutron detection efficiency ( to more than 80% ) by adding salt to the heavy water @xcite . in this case , the double ratio , defined for heavy water alone , @xmath195 will provide as much information as the double ratio @xmath180 . here , @xmath196 while @xmath197 is a similar ratio , defined for the total number of events from both @xmath5 and @xmath0 interaction with deuterium : @xmath198 both @xmath199 and @xmath197 are calculated for a heavy water detector . to a very good approximation , we have @xmath200 the approximation arises partly from ignoring events due to electron and oxygen targets in @xmath201 . the error also arises from the differences in the denominators of the two ratios , one involving @xmath152 and the other @xmath151 . despite a mild energy dependence of the ratio of these two cross - sections @xcite , it turns out that the ratio of the total events expected from these two processes remains in the range @xmath202 ( see table [ tab : ccad ] ) . this is true both when there is no mixing , and with mixing , for all allowed values of mixing angles @xmath203 . hence this factor cancels when the ratio @xmath204 is expressed in terms of @xmath172 . hence the approximation in eq . [ eq : sumrule ] should be valid to within a few percent . in addition , the double ratio @xmath201 will have the advantage of reduced systematic errors , since data from different experiments do not have to be combined in order to calculate it . hence it will be useful to calculate such a ratio , by separating out the @xmath0 events on deuteron . so far , we have discussed the sensitivity of the supernova neutrino spectrum to various neutrino mixing parameters . however , the supernova model parameters ( temperature and luminosity ) are themselves uncertain and still need to be experimentally established . it is therefore important to study the effect of variation of these parameters on the results we have so far obtained . supernova dynamics is a very complicated issue . here we will follow a simple - minded approach . changes in the luminosity affect the overall normalisation while changes in the temperature ( or average energy ) change in the _ shape _ of the spectrum . variations in these parameters , while being time - dependent , are not random , but systematic . for instance , the supernova model parameters depend on the protoneutron star mass ( an increase of which increases both the average energy and luminosity of neutrinos ) as well as the underlying high - density equation of state and the initial conditions . the effect of this on the total neutrino spectrum has been studied in ref . ( the temperature variation in the spectra of individual flavours is not discussed ) . the study indicates that the typical temperature variation of the total neutrino spectrum at all times does not exceed about @xmath205 mev . while the variation due to uncertainties in the initial conditions is relatively small , the average energy systematically decreases for smaller protoneutron mass stars and those evolving with a stiffer equation of state . on the other hand , the luminosities are virtually identical for all these cases until a time @xmath206 secs . , by which time most of the detectable neutrinos are emitted . we will first estimate the systematic errors due to uncertainties in the supernova temperature . in the absence of detailed information on temperature variation of the individual flavours , we shall assume a systematic time - independent increase ( or decrease ) in the temperature of spectra of all flavours by 1 mev . this will then be an estimator of the outer limits of variation of the results from the original calculation . [ fig : nueerr ] shows the expected number of events due to @xmath148 interaction in the absence of mixing and when the temperature is systematically increased or decreased by 1 mev in all time bins , @xmath207 the base - line supernova spectrum is shown in comparison as a solid line . there is of course a shift in the spectral peak ( by around 3 or 4 mev ) ; however , there is a large change in the high energy part of the spectrum ( accentuated due to the @xmath208 dependence of the cross - section ) . it will still be possible to distinguish the adiabatic mixing case since the increase at high energy in this case is substantially larger than from errors in the supernova spectrum . however , other cases , especially the non - adiabatic cases , will not be clearly distinguishable . it must be noted that in any event the spectral peak for @xmath5 events is a good index of the temperature of the spectrum , either of the unmixed @xmath5 or of the hot @xmath73 spectrum . [ fig : nuebarerr ] shows the results for the case of the @xmath151 unmixed spectrum . there is a similar dependence ( since the energy dependence of the cross - section is the same as in the @xmath148 case . since mixing does not significantly shift the @xmath0 spectral peak ( see fig . [ fig : nuebar ] ) , this will remain a good indicator of the corresponding spectral temperature . while there are large variations in the results for the individual spectra , these will be cancelled out in the double ratio @xmath180 of the events in heavy water and water ( in fact , this is the purpose of constructing such a ratio ) . this can be seen from fig . [ fig : newrerr ] , where the same double ratio , @xmath180 , is plotted for the different temperature sets , @xmath17 ( as solid lines ) and @xmath209 ( as dashed lines ) . it is likely that any time - dependence of the temperature variation that we have ignored , will affect the numerator and denominator of the ratio in the same way ; hence inclusion of time dependence should not affect this analysis . in computing this ratio , the `` observed number of events '' as required for the calculation of @xmath181 , @xmath210 , in eq . [ eq : ri ] is now determined both by the mixing parameters as well as by the modified supernova model . it is seen that there is very little sensitivity to the variations in the temperature . this is especially so in the adiabatic case . a greater sensitivity to the model parameters in the small @xmath22 , small @xmath44 ( non - adiabatic ) case occurs because of the presence of the additional energy - dependent factor , the landau - zener transition probability , @xmath48 . hence inclusion of temperature variations in the supernova model does not change the conclusions about discrimination of different mixing models . we add a note on variations in the luminosity due to , for example , choice of different initial conditions @xcite . this affects the overall normalisation which is an indicator of the total energy emitted . however , the double ratio will not be sensitive to this , unless these changes are extremely time - dependent . it is doubtful whether reasonable conclusions can be drawn in such a case , unless there are significantly large numbers of events in each time bin . in short , while the individual flavour spectra can be significantly modified by uncertainties in the supernova model parameters , the double ratio @xmath180 is largely insensitive to such variations . hence it is a good indicator of mixing . a final remark about statistical errors . as already stated , the background to a supernova signal due to both radioactivity as well as solar events is small at superkamiokande and sno . hence the signal will be clearly defined . in any case , the statistical errors ( assuming a @xmath211 error for both the numerator and the denominator and adding suitably in quadrature ) have been calculated for the double ratio @xmath180 . this has also been shown in fig . [ fig : newrerr ] . the errors are so small that they are not visible , except as a slight thickening of the lines near the @xmath212 region . of course , if the supernova is 50 kpc and not 10 kpc away , the statistical errors become 5 times larger . recall however that we have computed the events in 1 kton of detector . larger detector volumes will further reduce this error . in general , the statistical quality of the signal , while being good , will depend on the size of the detector as well as the distance to the supernova . as is well - known , heavy water detectors can directly observe nc events . this is very important in the context of supernova neutrinos since neutrino emission from supernovae is practically the only observable system where neutrinos ( and antineutrinos ) of all flavours are emitted in roughly equal proportions . note that there are also nc events on oxygen targets in both water and heavy water , with a characteristic signal of photons with energies in the range of 510 mev @xcite . however , these events are fewer in number than the nc events on the deuteron that we will discuss here . while there is no loss of nc events in the case of 3-flavour mixing , the existence of a fourth flavour will be signalled by loss of nc events into this sterile channel . this can be seen from tables [ tab : ncad ] and [ tab : ncnonad ] where the total number of nc events from neutrinos or antineutrinos of all flavours ( with @xmath213 mev ) is listed for different possible values of @xmath22 consistent with the solar neutrino expectation . it is seen that the number of nc events is not very sensitive to the value of @xmath22 ; however , from table [ tab : ncad ] it is clear that in the adiabatic case , there is about 25% depletion with 4-flavour mixing when compared to the no - mixing case . if the value of @xmath44 and the overall normalisation of the spectrum is known , nc current events can be used to discriminate between three and four flavour mixing . recall that the cc events are always enhanced by a factor of 1.52 for the adiabatic case . hence more conservatively , the nc events can be used to normalise the supernova spectrum to at least within 25% . when the upper resonance is non - adiabatic , however , part of the signal gets regenerated , especially for smaller @xmath22 . hence , as table [ tab : ncnonad ] shows , there are roughly the same number of nc events with and without mixing in the non - adiabatic case , independent of the number of flavours . the corresponding cc channel shows severe depletion only when @xmath22 is large ; otherwise , it is either enhanced , or the same as the no - mixing case . hence here also the nc events can be used to normalise the supernova spectrum . so far , all results in the 4-flavour analysis referred to the @xmath9 scheme as shown in fig . [ fig : level]b . we briefly discuss results in the @xmath10 scheme where the mixing matrix is defined through : @xmath214^t = u \times \left [ \nu_1 \quad \nu_2 \quad \nu_3 \quad \nu_4 \right ] ^t~,\ ] ] and @xmath215 here the ( 13 ) and ( 14 ) angles are constrained by chooz to be small : @xmath29 @xcite as in the @xmath9 scheme . the atmospheric neutrino problem now constrains @xmath216 ( the equivalent of the angle @xmath23 in the 3-flavour case ) to be maximal , when @xmath217 becomes small , @xmath218 . however , the ( 34 ) mixing angle @xmath219 is not constrained by any known experimental data . since @xmath5 is produced in the supernova core in essentially the @xmath220 mass eigenstate , any non - adiabaticity results in jumps near the upper msw resonances . the adiabaticity parameter here will be determined by the @xmath221 angle , which is again small , @xmath222 . however , the adiabaticity parameter at the lower resonance depends on the unknown angle @xmath219 and hence we do not comment on the non - adiabatic case here . the expressions for the fluxes as observed on earth are given in eqs . ( [ eq : fe ] ) , ( [ eq : febar ] ) and ( [ eq : fx ] ) with probabilities @xmath86 computed for the @xmath10 scheme . the relevant probabilities , @xmath78 and @xmath79 , for the cc events in the adiabatic case are _ independent _ of the unknown angle @xmath219 and are in fact the same in both the @xmath9 and @xmath10 schemes . hence all the results ( as shown in figs . [ fig : nuead ] , [ fig : nuebar ] , [ fig : total ] ) for the cc events on deuteron in the adiabatic sector are insensitive to the position of the sterile neutrino in the case of 4-flavours . this is true for the cc events on oxygen as well . in the nc case , we need the probabilities @xmath223 and @xmath224 which are different from the @xmath9 case . ignoring small terms of order @xmath225 , we have , @xmath226 where @xmath227 are @xmath228 and @xmath229 respectively . while these fluxes do depend on @xmath219 it turns out that the suppression factor in the adiabatic case is again around 75% for both large and small values of @xmath22 , again independent of the value of @xmath219 . this is because the dominant contribution to the nc sector is from the @xmath230 and @xmath85 fluxes , as in the @xmath9 case . these terms are very weakly dependent on the unknown angle @xmath219 . hence in the nc sector as well , the ( adiabatic ) @xmath10 4-flavour scheme gives almost the same predictions as the @xmath9 scheme . because of this , there will not be much difference in the elastic events on electrons as well . to summarise , we have contrasted signals from supernova neutrinos ( antineutrinos ) in water and heavy water detectors . we include the dominant charged current events from deuteron targets in heavy water , and proton targets in water , as well as the elastic scattering off electrons and charged current events on oxygen in both detectors . in all cases , an electron ( or positron ) is detected in the final state . the detailed distribution of events as a function of the scattered electron ( positron ) energy depends on the number of flavours and the mixing parameters in a complex manner . we have discussed all these cases . in particular , figs . [ fig : newr ] and [ fig : newrc ] show the combined sensitivity of water and heavy water detectors to the neutrino mixing parameters for such events by defining a double ratio of the observed to expected number of events in heavy water and water detectors respectively . these results reflect essentially the behaviour of the dominant charged current @xmath5 and @xmath0 events on deuterons which are comparable to the @xmath0 charged current interaction on protons in water . however , its dependence on the mixing parameters is very different from that for water . it turns out that a comparison of the signals from water and heavy water detectors can yield important information on not only mixing parameters but also on the number of flavours involved . while 3-flavour mixing typically results in an enhanced event rate , 4-flavour mixing can lead to substantial decrease in the number of events , depending on the mixing parameters . furthermore , we have performed a simple - minded analysis of the systematic errors involved due to uncertainties in the supernova model parameters . we have shown that the double ratio that quantifies the relative variation due to mixing in water and heavy water detectors is largely insensitive to variations in the supernova model parameters ( temperature and luminosity ) used . mohan narayan , m. v. n. murthy , g. rajasekaran , and s. uma sankar , phys . d * 53 * , 2809 ( 1996 ) ; mohan narayan , g. rajasekaran , and s. uma sankar , phys . d * 56 * , 437 ( 1997 ) ; s. goswami , k. kar , and a. raychaudhuri , int . j. mod . phys . * a 12 * , 781 ( 1997 ) ; g. l. fogli , e. lisi , a. marrone , and g. scioscia , phys . rev . d*59 * , 117303 ( 1999 ) ; _ ibid _ 033001 ; g. l. fogli , e. lisi , and d. montanino , astropart . phys . * 9 * , 119 ( 1998 ) ; g. l. fogli , e. lisi , d. montanino , and g. scioscia , phys . rev . d*56 * , 4365 ( 1997 ) . m. apollonio et al . , chooz collab . , phys . b * 420 * , 397 ( 1998 ) ; phys . b * 466 * , 415 ( 1999 ) ( hep - ex/9907037 ) ; mohan narayan , s. uma sankar , and g. rajasekaran , phys . rev . * d 58 * , 031301 ( 1998 ) .

the issue of thermal correlation functions and the associated effective statistical potential in two - dimensional moyal space , arising in the twisted approach to implement rotational symmetry , has been revisited in an operatorial formulation where no explicit star product is used initially . the corresponding results using moyal and voros star products are then easily obtained by taking the corresponding overlap with moyal and voros bases . in contrast to the moyal case where the concept of distance and , in particular , the relative separation between a pair of particles remain ambiguous when the moyal star product is used , the voros basis is more physical and the inter - particle distance can be introduced unambiguously . the forms of the correlation function and the effective potential are found to be same as the moyal case except that the thermal wavelength undergoes a non - commutative deformation , ensuring that it has a lower bound of the order of . it is shown that in a suitable basis ( called here quasi - commutative basis ) in the multi - particle sector the thermal correlation function coincides with the commutative result both in the moyal and voros cases along with the restoration of the pauli principle , except that in the voros case the thermal wavelength , again , gets a non - commutative correction . finally , we extend our result to three - dimensional non - commutative space and compute the correlation function and effective potential using both twisted and quasi - commutative bases in the moyal and voros cases . we find that there is symmetry breaking in the effective potential , which also violates the pauli principle , even for a pair of free particles , despite the fact that a deformed co - product is used to construct twisted symmetric / anti - symmetric basis . however , this symmetry , along with pauli principle , is restored once we use the quasi - commutative bases .

1306.4482

it was realized sometimes back by doplicher et.al @xcite from the consideration of both general relativity and quantum mechanics that the localization of an event in space - time with arbitrary accuracy is operationally impossible and this feature is captured by postulating a non vanishing commutation relation between the coordinates which are now promoted to the level of operators . in its simplest form they are given as @xmath3 = i\theta_{\mu\nu } \label{noncom.rel}\ ] ] where @xmath4 is taken to be an antisymmetric matrix and its entries are viewed as new fundamental constants @xcite . this form of non commutativity also follows from low energy limit of string theory @xcite . + analysis of quantum field theory in the background of such non - commutative spaces is expected to provide insight into the structure of quantum gravity as @xmath0 is expected to be of the order of planck length scale . introducing such a length scale can have some serious consequences . for example , the structure of the commutation relation ( [ noncom.rel ] ) , with @xmath4 held fixed ( i.e. not a tensor ) signals the violation of lorentz symmetry or simply the rotational @xmath2 symmetry in a non - relativistic system if the time t is taken to be the usual c - number parameter ( @xmath5 ) , rather than an operator . as has been shown in @xcite , this symmetry can be restored formally by deforming the co - product using the drinfeld twist . this in turn implies , according to the approach followed in @xcite , that the projection operator used to project multi - particle states into symmetric / antisymmetric subspaces to construct bosonic / fermionic states too should be deformed , thereby obtaining twisted bosons / fermions . in @xcite , these twisted fermions were shown to violate pauli exclusion principle by computing thermal correlation function for a pair of twisted fermions , although it preserves the fermi dirac statistics @xcite . this computation , however , was carried out using moyal star product , which is essentially associated with certain basis ( called moyal basis in @xcite ) . in fact , in the operatorial formulation of non - commutative quantum mechanics involving hilbert - schmidt operator , which was initiated in @xcite , one can bypass the use of any star product and hence can avoid the use of any associated bases and any ambiguities that may result there from @xcite . for example , in two - dimensional non - commutative moyal plane one can introduce two types of canonical star products : moyal and voros and one can identify the respective moyal and voros basis so that the representation , in any of these bases , of a composite state(obtained by a simple operator multiplication of a pair of states ) is the same as the one obtained by composing the respective representations of individual states by moyal / voros star products @xcite . however , it is only the voros basis which can be regarded as physical , as this conforms to povm ( positive operator valued measure ) , unlike the moyal basis . in particular the voros basis turns out to be a coherent state @xmath6 representing a maximally localized state in the non - commutative plane . indeed , it has been shown recently @xcite that one can compute the spectral distance , _ a la _ connes @xcite between a pair of neighboring states @xmath7 and @xmath8 to get a euclidean geometry : @xmath9 but such a distance function can not be assigned between the pair of neighboring states @xmath10 and @xmath11 corresponding to the moyal basis . the primary reason is that the moyal basis does not conform to the requirement of povm , as we have mentioned above . besides , this moyal basis turns out to be the eigenstate of commuting position - like observables : @xmath12 , which are defined as the average of left and right actions of the noncommutative position operators @xcite or equivalently obtained by a suitable linear transformation in the phase space as @xmath13 satisfying @xmath14=0 $ ] and therefore can not be interpreted as the position observables @xcite consequently , this moyal basis is a purely mathematical construct and is devoid of any physical meaning . these considerations therefore motivate us to have a re - look at this whole twisted formalism @xcite completely at the operatorial level , observe more closely the principle(s ) adopted here to enable us , to see whether this formalism where one constructs twisted bosons / fermions to accommodate deformed co - product , is an inevitable consequence of non - commutative space - time at least of the type given in ( [ noncom.rel ] ) . this is expected to shed light on other alternative formalisms existing in the literature ( see for example @xcite ) . indeed , the formalism adopted in @xcite was used to compute the thermal correlation function in @xcite . carrying out this analysis at the level of non - commutative quantum mechanics automatically paves the way for introducing second quantized non - relativistic field operators , which has a built - in tensor - product structure , where the left slot corresponds to particle creation / annihilation operator and the right slot is a momentum eigenstate . like the first quantized quantum mechanical state , one can also obtain moyal / voros space representations of this abstract second - quantized field operators by taking appropriate overlap with m / v basis . with this , the stage is set to carry out the computations of the thermal correlation function in the voros basis , where one can sensibly talk about the inter - particle distance . but , as has been discussed in @xcite , the voros wave functions @xmath15 of a state @xmath16 , besides belonging to schwartz class like its moyalian counterpart , need to satisfy an additional smoothness criterion at small length scale @xmath17 ; oscillators with wavelengths @xmath18 are suppressed exponentially . since the thermal wavelength @xmath19 occurring in the expression of correlation function in @xcite , computed in the moyal basis which can be made smaller than @xmath18 for high enough temperature , one expects to find appropriate deformation in the corresponding result in voros basis . secondly , as we show below , the twisted basis in the momentum space in the multi - particle sector is equivalent to a basis , up to an overall phase , which is symmetric / anti - symmetric under the usual exchange operation ( i.e. not the deformed one ) enabling one to define the usual bosons / fermions ( and not their twisted counterparts ) , which nevertheless retains some deformations , stemming from non - commutativity . it will thus be interesting to study the structure of the thermal correlation function in these bases as well , which we henceforth refer to as quasi - commutative basis , for reasons that will become clear in the sequel . finally , we would like to extend our computation from the planar case to the more realistic and physical voros basis in odd three - dimensional space as well . indeed in three - dimensional space such a voros basis was introduced in @xcite satisfying the over completeness relation as in the two - dimensional case . it is therefore imperative to check whether this too satisfies the povm criterion and saturates the uncertainty relation both in 3d coordinate space and 6d phase space . related to this , is the structure of 3d non - commutative parameter @xmath20 . being singular , this matrix admits a vector @xmath21 dual to @xmath20 and pointing in a particular direction which behaves like a commutative axis in a rotated frame . it thus spoils the isotropicity i.e. @xmath2 symmetry of 3d space . despite restoring the @xmath2 symmetry through the deformed co - product at the level of commutator as in @xcite , it was shown in @xcite to violate @xmath2 symmetry at the level of the action in presence of interaction . it will therefore be quite interesting to look for any signature of the violation of this symmetry for a system containing a pair of free particles , itself . as a thermal effect the structure of the 3d correlation function or the resulting statistical potential should tell us immediately about this violation of @xmath2 symmetry . yet another place where this violation can also show up , is the structure of the variance matrix which occurs in the symplectic invariant formulation of the uncertainty relation , which we use here . since , however , the williamson s theorem @xcite and the associated technique for symplectic diagonalization @xcite is not known to hold in our context , we can make use of the transformation ( [ transfm . ] ) to obtain the corresponding commutative variance matrix and its symplectic spectrum and try to see whether the saturation condition holds for the entire variance matrix and also for each distinct modes ( which are now de - coupled from each other ) in the same manner . finally , the resulting profile of the thermal effective potential may be used to study the nature of the violation of pauli exclusion principle , if any . this is expected to pave the way to study the astrophysical implications . the plan of the paper is given in the following manner : in section ii , we review the formulation of non - commutative quantum mechanics on two - dimension and three - dimension . in section iii , we provide a symplectic invariant formulation of the uncertainty relation . in section iv , we compute the non - commutative variance matrix for voros basis in three - dimension and find that voros basis states are maximally localized in phase space although it does not represent a maximally localized state in 3d non - commutative space . this single particle formulation of non - commutative quantum mechanics is extended in section v to two - particle system , where we try to formulate the twisted symmetric / anti - symmetric of @xcite through our operatorial approach . in section vi , the twisted formulation has been extended further for more than two particles and here we have introduced a quasi - commutative momentum basis which is the usual symmetric / antisymmetric basis differing from the twisted basis by only a phase factor in the momentum space . with this , in section vii , we have discussed the second quantization through the second quantized creation / annihilation operator so as to establish a contact with non - relativistic quantum field theory . introducing the creation and annihilation operators in these two types of multi - particle momentum bases in section vii , we have defined the abstract field operators in section viii and discuss its action on arbitrary state in position and momentum representations . then in section ix , we compute the two - particle correlation functions for a free gas in both two and three - dimensions and represent them in both moyal and voros bases using both twisted and quasi - commutative bases and obtain the corresponding effective potentials which is eventually plotted with respect to the inter - particle distance . finally , in section x , we conclude the paper . if @xmath38 and @xmath39 are the representations of the operators @xmath40 and @xmath41 respectively acting on @xmath33 , then a unitary representation is obtained by the following action : @xmath42\ ] ] it is easily verified that the momentum eigenstates @xmath43 are given by @xmath44 and they satisfy the usual orthonormality condition and resolution of identity @xmath45 one can then introduce the voros basis @xcite @xmath46 and the moyal basis as @xmath47 as the voros basis can also be written as @xmath48 , in terms of the coherent states @xmath49 , this represents the maximally localized state . further , the representation of any composite state @xmath50 in this voros / moyal basis automatically yields the voros / moyal star product composed expression of the corresponding representations of the individual states : @xmath51 with @xmath52 and @xmath53 , the voros basis was shown to conform to the requirement of povm , in contrast to the moyal basis @xcite as one can see that the resolution of identity for the voros basis @xmath54 has the integrand @xmath55 , which is a positive , but non - orthogonal unnormalized projection operator @xmath56 in @xmath57 , the algebra satisfied by the coordinate operators can be written as @xmath23 = i \theta _ { ij } = i\epsilon _ { ijk } \theta _ k , ~~ i , j , k = 1,2,3\ ] ] where @xmath58 is a @xmath59 anti symmetric matrix and @xmath60 is a vector dual to this . + by formally transforming the coordinate system @xmath61 by a suitable so(3 ) rotation say @xmath62 , we can orient the vector @xmath63 in the fiducial frame along say the third axis . for example , if @xmath64 then the transformation @xmath65 performed with @xmath66 then the action of the components of momenta in the original frame can be obtained through linearity as @xmath86 \end{aligned}\ ] ] we can verify that the simultaneous eigenstates of the above commuting momentum operators are given by @xmath87 note that @xmath88 is a scalar under an @xmath2 rotation . these momentum eigenstates will play an important role in the quantum hilbert space . in complete analogy with the 2-d case , these states too satisfy the orthonormality and the completeness relations @xmath89 here also , we can define the voros basis as @xmath90 which satisfies the completeness relation , if composed through voros star product : @xmath91 where @xmath92 with @xmath93 , and have non - orthogonal overlap between any pair of such states : @xmath94 the corresponding overlap of this basis with momentum basis is @xmath95 next , we define the 3d moyal basis ( the counterpart of ( [ moyal]))as @xmath96 which also satisfies the completeness relation @xmath97 where @xmath98 the moyal basis is an orthonormal basis @xmath99 and its overlap with momentum basis is @xmath100 these moyal basis states are the simultaneous eigenstates of @xmath101 given by ( [ transfm . ] ) . on the 3-d quantum hilbert space @xmath102 we can impose the additional structure of an algebra by defining the multiplication map @xcite : @xmath103 such that expanding a generic state @xmath16 in terms of momentum eigenstates and going to barred frame : @xmath104we can now prove the following composition rules ( 3d counterpart of eqn.([composed ] ) ) @xmath105where @xmath106 but here we loose any obvious connection of the voros basis to the coherent state . we can now investigate whether the uncertainty relation saturates for both phase space variables and also for just position coordinates for the voros basis ( [ voros ] ) . the standard deviation @xmath108 of any hermitian operator @xmath109 in a state @xmath110 can be written as @xmath111 where @xmath112 then using the schwarz inequality for a pair of such observables @xmath109 and @xmath113 , we can write @xmath114 splitting into real and imaginary terms , we can write @xmath115 now using the fact that @xmath116 we see that @xmath117\rangle$ ] involves commutator whereas @xmath118 involves anti - commutator . + we now introduce robertson and schrdinger uncertainty one by one : + : here we ignore the square of real part , to write @xmath119 this gives the robertson uncertainty relation @xmath120\rangle\ ] ] : here we retain both the squares of real and imaginary parts , to get , using ( [ both ] ) @xmath121\rangle\right)^2\ ] ] this finally gives the schrdinger uncertainty relation @xmath122\rangle\right)^2}\ ] ] before proceeding further with the computation in the non - commutative case ( @xmath123 ) , let us discuss how this schrdinger s form of uncertainty relation can be recast in terms of variance matrix . for this , let us rename the space and momenta operators by a single phase - space operator @xmath124 where @xmath125 with @xmath126 and @xmath127 . then the schrdinger uncertainty relation for the phase - space operators will be given by @xmath128\rangle\right)^2 } \label{sch}\ ] ] where the expectation values are to be taken in a certain state @xmath110 with @xmath129 . + the first squared term in the rhs of the above relation is the square of @xmath130 th element of the variance matrix @xmath131 , where @xmath132 and the second squared term can be identified as the square of @xmath130 th element of the symplectic matrix @xmath133 \rbrace = \frac{1}{2}\begin{bmatrix } 0&0&0&1&0&0\\ 0&0&0&0&1&0\\ 0&0&0&0&0&1\\ -1&0&0&0&0&0\\ 0&-1&0&0&0&0\\ 0&0&-1&0&0&0\\ \end{bmatrix}\ ] ] in general , the variance matrix @xmath134 will be @xmath135 square matrix corresponding to a @xmath136 dimensional phase space(in our case , @xmath137 ) . this provides an exhaustive characterization of any gaussian state . by williamson s theorem @xcite there exists a symplectic transformation @xmath138 so that any arbitrary variance matrix @xmath139 can be brought to a diagonal form @xmath140 ; @xmath141 , where @xmath142 comprises of the ( up to the orderings of @xmath143 ) the symplectic eigenvalues of @xmath139 , which are at least doubly degenerate . this symplectic spectrum is not to be confused with the ordinary spectrum , which is obtained by a similarity transformation . rather , the symplectic spectrum can be obtained through the ordinary spectrum of @xmath144 , as the composite object ( @xmath145 ) undergoes a similarity transformation , if @xmath139 undergoes a symplectic transformation @xcite . correspondingly , the density matrix @xmath146 transforms a @xmath147 , where @xmath148 is a unitary operator implementing the symplectic transformation . we can notice , at this stage , that in this diagonal form , the @xmath136dimensional phase space splits into n - copies of independent @xmath149dimensional phase space . it is therefore convenient to consider the schrdinger s uncertainty relation ( [ sch ] ) for 2d phase space re - written as , @xmath150 identifying @xmath151 and @xmath152 , this inequality is equivalent to @xmath153 ; @xmath154 whereas @xmath155 + here we have taken @xmath156 and @xmath157 without loss of generality , so that the spread @xmath158 and @xmath159 are equal . further , we have used the symplectic invariant form of @xmath160 and @xmath161 . compatibility among these three inequalities implies @xmath162 or equivalently @xmath163 thus for a bonafide variance marix @xmath139 we must have the symplectic spectrum to be such that @xmath164 or more generally @xmath165 for the general ( @xmath166)- dimensional phase space . this provides a symplectic @xmath167 invariant formulation of the uncertainty relation . finally , note that both robertson and schrdinger form of uncertainty relations become equivalent in this diagonal form . in our @xmath68 non - commutative quantum system ( @xmath168 ) , however , this formalism is not directly applicable . in particular , it is not known whether the williamson s theorem , remains valid or not , in this context . we will therefore transform this non - commutative variance matrix into a commutative one by using the transformation ( [ transfm . ] ) and apply this formalism to show that for @xmath68 voros basis the symplectic invariant form of the uncertainty relation ( [ satcon ] ) is indeed saturated indicating that like @xmath69 case , the @xmath68 voros basis also corresponds to a maximally localized state in phase space . starting with the normalized version of the voros states ( [ voros ] ) @xmath169 we can now find the expectation values of @xmath170 and the composite @xmath171 in the above normalized voros states . now , the expectation value of @xmath38 in these voros states can be rewritten in the barred frame as @xmath172 with @xmath173 being the rotation matrix ( [ rotmatrix ] ) . now , we have @xmath174 then , @xmath175 and @xmath176 with these , we have @xmath177 so , returning to the unbarred fiducial frame , we get from ( [ unbarred ] ) @xmath178 and @xmath179 having solved @xmath180 ( @xmath181 ) we now have to calculate the expectation values of the composite @xmath182 and then its symmetrized and anti - symmetrized expectation values to obtain the variance matrix and the symplectic matrix . for this , let us calculate the expectation value of a composite @xmath183 . note that , by itself @xmath184 does nt transform as a second rank tensor under @xmath2 , as was shown in @xcite . indeed , by defining @xmath185 ( @xmath186 ) one gets for the rotated composite operator @xmath187 = \hat{x}_i^r\hat{x}_j^r + \frac{i}{2}\theta_{ij } - \frac{i}{2}r_{ik}\theta_{kl}(r^t)_{lj}\ ] ] through the action of deformed co - product @xmath188 where @xmath189 is the undeformed co - product and @xmath190 is the abelian drinfeld twist . nevertheless its symmetric part ( anti - commutator ) transforms as a tensor , whereas its antisymmetric part ( commutator ) transforms as an invariant @xmath2 scalar , as was observed earlier @xcite . @xmath191 under the rotation @xmath62 , we have @xmath192 this gives after a straightforward computation @xmath193 so that upon symmetrization and anti - symmetrization , one gets @xmath194 and @xmath195 \rangle= \frac{i}{2}\theta_{ij}\]]respectively . + with this , we get the first block @xmath196 @xmath197 of the complete non - commutative variance matrix @xmath198 where the expectation values are taken in the voros states . let us now calculate the remaining matrix elements of the @xmath199 , @xmath200 and @xmath201 parts of @xmath134 . to begin with , let us consider @xmath202 again relating one of the terms in the barred frame as , @xmath203 we get for the matrix @xmath204=\begin{bmatrix } 0&-\frac{1}{2}&0\\ \frac{1}{2}&0&0\\ 0&0&0\\ \end{bmatrix}\ ] ] using the expression of @xmath205 in ( [ rotmatrix ] ) , and the fact that @xmath206 we get @xmath207 finally , for @xmath201 we get @xmath208 thus , the complete variance matrix ( [ var ] ) is @xmath209 and the non - commutative symplectic matrix @xmath210 is @xmath211\rangle&\frac{1}{2}\langle[\hat{x}_i,\hat{p}_j]\rangle\\ \frac{1}{2}\langle[\hat{p}_i,\hat{x}_j]\rangle&\frac{1}{2}\langle[\hat{p}_i,\hat{p}_j]\rangle\\ \end{bmatrix}\\ & = & \frac{1}{2}\begin{bmatrix } 0&\theta_3&-\theta_2&1&0&0\\ -\theta_3&0&\theta_1&0&1&0\\ \theta_2&-\theta_1&0&0&0&1\\ -1&0&0&0&0&0\\ 0&-1&0&0&0&0\\ 0&0&-1&0&0&0\\ \end{bmatrix}\end{aligned}\ ] ] the corresponding commutative variance matrix @xmath139 and the symplectic matrix @xmath212 can be obtained from the above respective non - commutative matrices by linear transformations @xcite @xmath213 where the matrix @xmath214 is the transformation matrix which relates the commutative position coordinates to the non - commutative position coordinates as @xmath215=0\ ] ] as we have ( [ transfm . ] ) @xcite @xmath216 the matrix @xmath214 is given by @xmath217 with this , the commutative variance matrix and the commutative symplectic matrix is found to be @xmath218 and @xmath219\rangle=\frac{1}{2}\begin{bmatrix } 0&0&0&1&0&0\\ 0&0&0&0&1&0\\ 0&0&0&0&0&1\\ -1&0&0&0&0&0\\ 0&-1&0&0&0&0\\ 0&0&-1&0&0&0\\ \end{bmatrix}\ ] ] with @xmath220 . then calculating the symplectic eigenvalues of @xmath139 i.e. the ordinary eigenvalues of @xmath144 @xcite , we get the eigenvalues as @xmath221 i.e. 6-fold degenerate . these three pairs of symplectic eigenvalues , each of the form of @xmath222 can be obtained more simply from the corresponding single mode @xmath139 of the form @xmath223 ( [ comsymmat ] ) , which occurs symmetrically in all the directions @xmath224 , by a simple canonical transformation @xmath225 for suitable @xmath226 . in any case it is simple to see that both @xmath227 ( [ noncommvar ] ) and @xmath139 ( [ comsymmat ] ) satisfy the saturation condition ( [ satcon ] ) ( since @xmath228 ) @xmath229 this indicates that the voros basis represents a maximally localized state in phase space . but note that the voros basis ( [ voros ] ) , can be factorized by going to the barred frame as @xmath230 where the first factor represents the 2d voros basis ( [ vorosbasis ] ) and the second factor @xmath231 , representing a one - dimensional gaussian state centered at @xmath232 with a spread @xmath233 . clearly , this @xmath234 can be made as small as we like by a suitable scaling factor and scaling up @xmath235 appropriately to preserve the saturation condition ( obviously this is a non - voros state ) . the generalized points , i.e. pure states of the involutive @xmath236algebra , which is just @xmath33 in this case can be described by using the barred variables by the density matrices @xmath237 . correspondingly , the pure states in the fiducial frame is obtained by applying a unitary transformation : @xmath238 . this non - voros state will be useful if we are interested in computing connes spectral distance between a pair of such points i.e. pure states ( see for example @xcite and references there - in ) . however , since our interest is to compute the thermal correlation function between a pair of identical particles in voros states , we shall not be concerned with such states in the rest of the paper . thus even if @xmath239 is squeezed to the extreme such that @xmath240 we can still have @xmath241 for such a non - voros state . more generally , we can write in this case @xmath242 it is therefore quite interesting to see the form of analogous inequality in the original fiducial frame for the voros basis . using @xmath243 from ( [ noncommvar ] ) , we have @xmath244 \label{spaceuncertainty}\ ] ] where the vector @xmath63 points in arbitrary direction . we can see that the expression ( [ spaceuncertainty ] ) attains its minimum value @xmath245 when the vector @xmath63 points in one of the three axes ( e.g. @xmath246 ) . so we have the following condition @xmath247 let us extend this formulation to two particle systems . in commutative quantum mechanics the way we proceed is to think of the two particle system in terms of wave - functions defined over @xmath248 and therefore to think of classical configuration space as a tensor product @xmath249 . one may therefore be tempted to take the same approach in non - commutative quantum mechanics and to introduce the non - commutative @xmath68 configuration space for a two particle system as a tensor product of two single particle configuration spaces , i.e. , @xmath250 and the quantum hilbert space as the space of operators generated by the @xmath251 and @xmath252 , with the subscripts referring to the non - commutative coordinates of the two particles , i.e. , the elements of the quantum hilbert space are operators @xmath253 . this is essentially the approach adopted in @xcite where a moyal - like star product between the functions @xmath254 and @xmath255 was introduced as @xmath256 which yields the following commutation relations @xmath257= i\theta_{ij } \label{commuationrelation}\ ] ] where the subscript @xmath258 are the particle labels . these types of commutation relations are also obtained in the approach of braided twisted symmetry @xcite . this way of thinking is , however , very misleading and , indeed , inappropriate in the context of the operator algebra formulation of non - commutative quantum mechanics outlined in section [ section2 ] , as we now proceed to argue . the more appropriate way to think about a two particle commutative system is to think of it as two sets of @xmath68 labels in the same @xmath68 configuration space or , more appropriately , in the language of relativity as the coordinates of two events " in @xmath68 . in non - commutative quantum mechanics the notion of coordinates does not exists in classical configuration space and should arise through the state @xmath259 or operator @xmath260 , which acts on @xmath31 , and describes the quantum state of the system . the real issue , therefore , is to identify the states in quantum hilbert space that would describe a two particle state with particles localised at points @xmath261 and @xmath262 . the answer to this question is actually quite natural and simple . to start , let us first consider one particle and , for demonstrative purposes , in @xmath69 . we want to find the state in quantum hilbert space describing a particle localised at @xmath263 . the state with this property is the one given by @xmath264 . as the action of the position operators are defined by left action , it is an eigenstate of @xmath265 and subsequently a minimum uncertainty state in classical configuration space . the right hand sector describes another property of the system . a more detailed discussion of this interpretation can be found in @xcite . this interpretation is also borne out by calculating the position representation of this state , e.g. , @xmath266 setting @xmath267 , this turns into @xmath268 clearly demonstrating that the state is localised at @xmath269 with non - local corrections deriving from an expansion in @xmath270 . restoring dimensions , @xmath270 is of the order @xmath0 , demonstrating that the non - local corrections are of the order of the length scale set by the non - commutative parameter . keeping in mind that @xmath271 is an operator on @xmath31 , and therefore an element of the algebra generated by the @xmath40 , it can be written in the form @xmath272 ( note that @xmath273 is a label and @xmath274 are operators ) . the two particle state must now also be an operator acting on the same configuration space @xmath31 and therefore be in the algebra generated by the @xmath40 , but it must also carry the indices of two particles . the most obvious , and probably only , construction of such two particle states is : @xmath275 this is a state in quantum hilbert space representing two particles localised at points @xmath261 and @xmath262 as can also be borne out by computing the position presentation of this state . indeed , the wave function of the state @xmath276 is obtained by taking the overlap in m / v basis to yield a m / v star product composed wave function @xmath277 where @xmath278 , @xmath279 and the states @xmath280 , @xmath281 are of the form ( [ posstate ] ) or ( [ posstate1 ] ) . the two particle nature of this construction can also be made more explicit by interpreting it as a map @xmath282 : @xmath283= |(\psi_1\psi_2)(\hat{\vec{x}})).\ ] ] this also brings us in contact with the philosophy of @xcite and the implementation of twisting adopted in @xcite . we can now address the question of the transformational property of a generic two - particle state ( [ 2particlestates ] ) under an infinitesimal so(3 ) rotation . this is determined by using a mathematical consistency condition following from an identity in hopf algebra . this is given in this context as @xmath284 = m\left[\delta_\theta(\hat{j}_i)\left(|\psi_1(\hat{\vec{x}}))\otimes|\psi_2(\hat{\vec{x}}))\right)\right]\ ] ] as it has been already obtained in @xcite the co product of the angular momentum operators @xmath285 get deformed in the non - commutative @xmath68 quantum hilbert space . so the action of this deformed co - product on a generic two particle state can be obtained as @xmath286 \end{aligned}\ ] ] so that , using the twist ( [ twist ] ) @xmath287 = f \delta_0 ( \hat{j}_i ) f^{-1}\ ] ] let us now introduce an exchange operation @xmath288 on the quantum hilbert space @xmath289 such that @xmath290 we find @xmath291\neq \delta_\theta(\hat{j}_i)[\sigma(\psi_1(\hat{\vec{x}})\otimes\psi_2(\hat{\vec{x}})]\ ] ] this implies that the under a transformation like rotation the statistics of the physical state can get altered ; a pure bosonic / fermionic state , obtained by projecting into symmetric / antisymmetric subspace by the projector + @xmath292 will yield an admixture of bosonic / fermionic states under rotation . but this can not be allowed as implied by the super selection rules which says that a system of fermions or bosons should remain as the one under any transformation . and for this , the exchange operation should commute with the deformed co - product . hence , the exchange operator @xmath288 should also get deformed as @xmath293 such that @xmath294 = 0 $ ] . corresponding to this deformed exchange operator , the deformed projection operator can be constructed as @xmath295 since @xmath296 , we can easily check that @xmath297 so that @xmath298 we then obtain @xmath299\label{projection}\ ] ] here , @xmath300 is referred as the twisted symmetric(+)/ antisymmetric(- ) projection operator which give the twisted symmetric/ antisymmetric states corresponding to the twisted bosons/ fermions system . + the construction of n - particle states proceed in complete analogy with the two particle states . the states in quantum hilbert space representing n - particle states are of the form @xmath301 this can again be interpreted as a map : @xmath302=|(\psi_1\psi_2\ldots\psi_n)(\hat{\vec{x}})).\ ] ] we have already mentioned that for two - particles system , the deformed co - product gives rise to the deformed permutation symmetry so that a twisted symmetric or antisymmetric two particle state is given by ( [ projection ] ) @xmath303 here , @xmath304 corresponds to the twisted symmetric state and @xmath305 corresponds to the twisted antisymmetric state . the corresponding twisted symmetric / antisymmetric momentum eigenstates can be written more compactly as @xmath306 \label{twisted2basis}\ ] ] where @xmath307 corresponds to twisted symmetric states and @xmath308 corresponds to twisted antisymmetric states . these states do not satisfy the orthonormality conditions but do satisfy the completeness relation : @xmath309 \label{twistedorthonormality}\ ] ] and @xmath310 we can extend this to n - particle system so that the twisted n - particle symmetric and antisymmetric states @xcite can be constructed as @xmath311 we just need to find the form of the deformed projection operator @xmath312 for n - particle system . for 2-particle system , it is given by ( [ 2projection ] ) . in order to extend to 3-particle system , first we should note that here we have to encounter with two deformed nearest neighbor exchange operators : @xmath313 which exchanges the first and the second slots keeping the third slot fixed in the tensor product of three operators and @xmath314 which exchanges the second and the third slots keeping the first slot fixed . thus , we can write the deformed projection operator for 3-particle physical states as @xmath315\ ] ] where we should note that @xmath316 and @xmath317 then for the n - particle system , there should be ( n-1 ) deformed nearest neighbor exchange operators @xmath318 which exchanges only the entries at the @xmath319th and @xmath320th slots , keeping all the entries at other slots fixed . that is , @xmath321 the last two relations for 3-particle system ( [ 3first ] , [ 3second ] ) can be put in a generalized form for an n - particle system as @xmath322 and @xmath323 so that the deformed projection operator for n - particle system is given by @xmath324\end{aligned}\ ] ] we should not forget that @xmath325 so that in the above expression we have a phase factor for each deformed exchange operator . thus in the last term of the above equation there are n phase factors for n deformed exchange operators . needless to say that @xmath326 , depending on whether n is even or odd . in this way , we can define a twisted symmetric/ antisymmetric state corresponding to the twisted bosons / fermions for n - particle system on the quantum hilbert space @xmath102 . it is obvious that we can get the corresponding twisted symmetric / antisymmetric momentum eigenstates as @xmath327\end{aligned}\ ] ] where we have used the wedge @xmath328 between the momenta to simply denote the following factor for simplicity , it has no relation with the wedge product of differential geometry . @xmath329 we can obtain the following relation for the twisted symmetric / antisymmetric momentum eigenstates for such n - particle system as @xmath330.\\ \nonumber\end{aligned}\ ] ] clearly , this demonstrates the non - orthogonality between these pair of states in the n - particle sector , because of the presence of relative phase factors . however , the resolution of identity is satisfied and the corresponding completeness relation is given as @xmath331 since these twisted symmetric / anti - symmetric momentum eigenstates are not orthogonal , we search for another basis ( in the next subsection ) which is just different from this former basis by a phase factor and check whether it is orthogonal , without upsetting the completeness relation . we can now introduce a new basis @xmath332 ( denoted henceforth by a double ket ) as @xmath333 \label{quasi2basis}\ ] ] this basis will be referred as quasi - commutative basis and is found to be symmetric under the undeformed exchange operation @xmath334 ( [ exchangeformalism ] ) and it is related to the twisted basis ( [ twisted2basis ] ) as @xmath335 we can easily check that this new basis satisfy the usual orthogonality and completeness relations as @xmath336\ ] ] and @xmath337 in the similar way we can define the new basis for 3-particle and so on as @xmath338\end{aligned}\ ] ] where @xmath339 satisfying the following corresponding orthogonality and completeness relations @xmath340\end{aligned}\ ] ] and @xmath341 in this way , we can write such symmetrized / anti - symmetrized quasi - commutative basis for any arbitrary number of particles which satisfy the usual orthogonality and completeness relations . let us now enlarge the physical quantum hilbert space to include states of arbitrary number of particles . we can define a full quantum hilbert space which is just the direct sum of the spaces with all possible number particle states . that is , @xmath342 where @xmath343 is the zero - particle space , the so called vacuum state and @xmath344 is the @xmath319-particle space , with the super - script @xmath319 indicating the particle number . we now introduce creation / annihilation operators for both twisted and quasi - commutative bases . if we take the twisted symmetrized / anti - symmetrized momentum basis in this full quantum hilbert space , we have identity operator in this basis as @xmath345 the overlap of any two such basis states with different number of particles on @xmath346 vanishes : @xmath347 then the creation and annihilation operators in this twisted momentum basis can be defined as @xmath348 and @xmath349 thus an arbitrary state on @xmath346 containing n - particles can be created by the n - fold action of the creation operators on the vacuum state as @xmath350 and the further action of the creation operator on an arbitrary state can be defined as @xmath351 here , we take , by convention , that the creation operator creates the new particle at the first slot which is different from creating a particle at the last slot , unlike the commutative case - even for a twisted boson , as these field operators of twisted states obey the deformed commutation relations : @xmath352 on the other hand , the action of annihilation operator of the twisted state on an arbitrary state can be written as @xmath353 now let us take the symmetrized / anti - symmetrized quasi - commutative orthonormal basis @xmath354 for which we can define the identity operator on this full quantum hilbert space @xmath346 as @xmath355again the overlap of any two states on @xmath346 with distinct number of particles vanishes : @xmath356 we can now define analogously the creation and annihilation operators in the quasi - commutative basis as @xmath357 and @xmath358 where these new operators are related to the ones of twisted basis ( [ top1 ] , [ top2 ] ) as @xmath359 where @xmath360 is the total momentum and @xmath361 stand for entirely commutative case ( @xmath107 ) , in contrast to ours , where @xmath362-dependence persists in their defining expressions ( [ qcop1 ] , [ qcop2 ] ) through the quasi - commutative basis . the fact that their ( anti ) commutation relations ( [ qcom1],[qcom3 ] ) are just like their commutative ( @xmath107 ) counterparts , which however develops @xmath362-deformation in the correlation function through deformed thermal wavelength in the more physical voros basis , as will be shown subsequently , is the reason behind adopting the terminology quasi - commutative basis . ] . the similar actions of these new creation / annihilation operators are given by @xmath363 @xmath364 and @xmath365 we can easily verify that these field operators obey the usual ( i.e. like @xmath107 ) ( anti ) commutation relations : @xmath366 now we can look for the abstract field operators without referring to any basis . clearly , this is achieved by writing @xmath367 and @xmath368 here , we should note that the first slot of tensor product is an operator acting on a particular quantum hilbert space @xmath344 to give an element of @xmath369/@xmath370 corresponding to one less ( n-1)/one more ( n+1 ) number of particles , while the second slot of the tensor product is the momentum eigenstate belonging to quantum hilbert space @xmath33 , which incidentally is also an operator acting on classical hilbert space ( [ configuration ] ) . thus the field operators in position ( moyal / voros basis ) and momentum representations for the non - commutative case can then be understood as @xmath371 @xmath372 and @xmath373 likewise , @xmath374 and the action of the field operator on any arbitrary state of @xmath346 can be understood through the action of the first slot as @xmath375 @xmath376 so that we can understand the action of field operators on any state of @xmath346 in position and momentum representations as @xmath377 @xmath378 and @xmath379 @xmath380 similarly for the new oscillators @xmath381 ( [ qcop1 ] ) and @xmath361 ( [ qcop2 ] ) , we have the field operators defined as @xmath382 and @xmath383 with their actions defined in the analogous manner . now we are interested to calculate the two - particle correlation function for a free gas in two and three dimensions using the canonical ensemble , i.e. to calculate the matrix elements @xmath384 , where @xmath385 is the canonical partition function and @xmath386 is the non - relativistic free particle hamiltonian with @xmath387 . this has been already calculated in @xcite in the twisted moyal basis . however , since moyal basis does nt conform to the povm unlike voros @xcite , our main interest will be to compute the resulting expression in voros basis . nevertheless we shall also present the corresponding computation in the moyal basis to demonstrate their structural similarity , except for deformed thermal wavelength in the voros case . thus the violation of the pauli principle seems to occur in either cases . however since the quasi - commutative basis , introduced in section vi.2 satisfy un - deformed exchange symmetry @xmath334 ( [ exchangeformalism ] ) and orthonormality condition , unlike the twisted basis , we would like to re - analyze this in this quasi - commutative basis , to find out whether the pauli principle , along with @xmath2 symmetry in 3d case , is restored or not . the twisted two - particle state in two - dimension is obtained ( in the voros / moyal basis ) by the 2-fold action of the field operator as , @xmath388 where in two dimension , it was shown in @xcite @xmath389 the two - particle correlation function in moyal / voros basis can therefore be written as @xmath390 where z stands for the partition function : @xmath391 we can now write for the free particle hamiltonian @xmath392 , by inserting the 2-particle completeness relation satisfied by the twisted momentum basis @xmath393 ( [ twistedcommcompleteness ] ) @xmath394 using ( [ twistedorthonormality ] ) and ( [ overlaps ] ) , we get for the overlap of @xmath395 with the moyal basis ( [ twistedmoyal ] ) @xmath396\ ] ] and like - wise for the voros basis . indeed , as it turns out , these two overlaps differs only by an exponential factors @xmath397 , apart from an unimportant factor of @xmath398arising from the different normalization factors for moyal and voros basis ( [ vorosbasis ] , [ moyal ] ) @xmath399 @xmath404\label{voros2}\ ] ] where @xmath405 representing a non - commutative deformation of the mean thermal wavelength @xmath406 in the corresponding commutative ( @xmath107 ) case @xcite . note that in contrast to voros case , the moyalian expression ( [ moyalian ] ) does not display any such deformation and therefore can be made arbitrarily small by taking arbitrarily large temperature @xmath407 . on the other hand , the mean thermal wavelength in the voros case ca nt be made smaller than @xmath408 which is expected , as the voros basis is supposed to capture the non - commutative feature correctly , by suppressing exponentially the oscillations with associated wavelength @xmath409 in the corresponding wave function @xmath410 @xcite . further , the relative distance @xmath411 occurring in the either expression ( [ moyalian],[voros2 ] ) can only be regarded as the true distance only in the voros case , as the spectral distance _ a la _ connes could be computed in this case - unlike in the moyal case @xcite . and this difference stems from the fact that it is only the voros basis that conforms to povm - unlike the moyal basis @xcite . finally , we would like to say that this expression ( [ voros2 ] ) is different from the one obtained in @xcite which was calculated with the aspects of braided twisted symmetry , see ( @xcite , @xcite ) . as we had already mentioned that our approach is different from that of the braided twisted symmetry , it is quite obvious that we get different results . now let us consider the usual symmetric / antisymmetric two - particle state in the quasi - commutative basis given by @xmath412 computing in the similar way as done in the twisted case , we finally obtain the two - particle correlation functions in the quasi - commutative bases as the above expressions show that the two - particle correlation function in the quasi - commutative moyal basis is just the same as the usual commutative result . however , in the quasi - commutative voros basis , although it turns out to be structurally same with the commutative or quasi - commutative moyal case , it picks up a non - commutative deformation through deformed mean thermal wavelength ( [ deformedlambda ] ) . clearly , we recover the pauli exclusion principle in both quasi - commutative moyal and voros cases . in the same way , the two - particle correlation function in the twisted momentum basis in three - dimension can be carried out . the twisted two - particle voros / moyal basis is given by @xmath414 as in three - dimension , the overlap between the momentum basis and the voros / moyal basis are given by @xmath415 we can calculate the correlation function in the same way as done in 2d case . but , as we had already mentioned that in three - dimension @xmath20 is a 3x3 matrix antisymmetric matrix and @xmath416 is a vector dual to it . so it becomes more complicated to calculate the correlation function . however , we can simplify the calculation by taking the real vector @xmath63 along the third axis so that @xmath417 and @xmath418 . with this we have , as in two - dimension , @xmath419 here in the above calculation we have chosen a specific form of @xmath20 which is indeed @xmath430 where @xmath431 and @xmath173 is the rotation in the configuration space such that @xmath432 . this rotation in the configuration space will implement a unitary transformation on the quantum hilbert space so that a one particle state will transform as @xmath433 then the two - particle state will transform as ( following ( [ 102 ] ) in section [ sec : two particle formalism ] ) @xmath434 where @xmath435 is the deformed co - product ( [ deformed co product ] ) , given as @xmath436 with @xmath437 ( [ twist ] ) being the abelian drinfeld twist and @xmath438 is the un - deformed co - product ( [ undeformed product ] ) . with this , it is clear that we can write @xmath439 the hamiltonian @xmath440 in our case is the that of a pair of free particles , given by @xmath441=0 . \label{commute}\ ] ] thus , giving overall no effect , so the above equation ( [ corr . ] ) will reduce to @xmath442 this shows that although we had taken @xmath430 to carry out the computation in the barred frame , the result will remain the same if we were to calculate in the fiducial frame taking @xmath20 . we made the choice of @xmath20 to make the calculation easier . however , it should be noted that this holds for the case of the free particle hamiltonian only . this might not be true if the hamiltonian have an interaction term , as ( [ commute ] ) may not hold any more . we can then compute the effective potential for each cases by putting the above expressions in the relation @xmath443 . for the convenience of comparison , we recast these expressions in terms of the dimensionless variables @xmath444 , @xmath445 , @xmath446 and @xmath447 , involving the un - deformed thermal wavelength @xmath448 ( [ moyalian ] ) . for the twisted moyal and voros cases , we have the thermal effective potential as @xmath449\ ] ] and @xmath450\ ] ] for the cases of quasi - commutative moyal and voros bases , in the same way , we have @xmath451\ ] ] and @xmath452\ ] ] we have the similar following expressions for the twisted moyal and voros cases in three - dimension , but now these depend both on @xmath453 and @xmath454 , as the @xmath2 symmetry is now broken to @xmath455 : @xmath456\ ] ] and @xmath457\ ] ] and for the quasi - commutative moyal and voros bases , we have the same expression as in the two - dimension case , as @xmath2 symmetry is restored and the effective potential now depends only on @xmath458 : @xmath459\ ] ] and @xmath460\ ] ] these expressions are plotted in fig . 1 ( for 2d case ) and in fig . 2 ( for 3d case ) . * , width=755,height=415 ] for each case in three - dimension . note that , in the twisted case , this depends functionally on @xmath461 and @xmath462 , in contrast to quasi commutative case , where it depends only on @xmath458 . * , title="fig:",width=302,height=207 ] for each case in three - dimension . note that , in the twisted case , this depends functionally on @xmath461 and @xmath462 , in contrast to quasi commutative case , where it depends only on @xmath458 . * , title="fig:",width=302,height=207 ] for each case in three - dimension . note that , in the twisted case , this depends functionally on @xmath461 and @xmath462 , in contrast to quasi commutative case , where it depends only on @xmath458 . * , title="fig:",width=302,height=207 ] for each case in three - dimension . note that , in the twisted case , this depends functionally on @xmath461 and @xmath462 , in contrast to quasi commutative case , where it depends only on @xmath458 . * , title="fig:",width=302,height=207 ] for each case in three - dimension . note that , in the twisted case , this depends functionally on @xmath461 and @xmath462 , in contrast to quasi commutative case , where it depends only on @xmath458 . * , title="fig:",width=264,height=188 ] for each case in three - dimension . note that , in the twisted case , this depends functionally on @xmath461 and @xmath462 , in contrast to quasi commutative case , where it depends only on @xmath458 . * , title="fig:",width=264,height=188 ] the issue of twisted symmetry in 2d/3d noncommutative moyal space has been re - visited in a completely operatorial framework using hilbert - schmidt operators to investigate whether the twisted bosons / fermions @xcite necessarily occurs in conjunction with the twisted deformed coproduct in moyal space @xcite , where the twisted fermions were shown to violate pauli principle @xcite . further , even within this scheme , it is shown that there exists a basis in the multi - particle sector called quasi - commutative basis , which satisfies orthonormality and completeness relation and is symmetric / antisymmetric under the usual i.e. un - deformed exchange operator , so that one has usual bosons / fermions and can avoid introducing twisted bosons / fermions . the correlation functions and the associated thermal effective potential is then shown to conform to pauli principle , apart from preserving the @xmath2 symmetry in 3d case , both in moyal and voros basis , in contrast to the case of twisted bosons / fermions , where there is a @xmath1 symmetry breaking . in all these cases , the resulting expressions in moyal and voros bases exhibit the same structural form , except that in the voros case , one gets a @xmath362-deformed thermal wavelength ensuring that it has a non - vanishing lower bound , which is in conformity with the requirement that wavelengths @xmath463 are suppressed exponentially . thus , in voros basis one gets a non - commutative deformation even in the quasi - commutative basis and this voros basis should be regarded as physical as one can talk sensibly about the inter - particle separation , as one can introduce spectral distance a la connes , unlike its moyalian counterpart @xcite . in this context , we would like to mention that the 3d voros basis , which was introduced earlier in @xcite , is shown here to saturate the 6d phase - space uncertainty by computing the variance matrix in a symplectic approach , although it does not correspond to maximally localised state in 3d space unlike its 2d counterpart . besides , it has an isotropic structure , in the sense that the symplectic eigenvalues of the corresponding commutative variance matrix yield the same pair of eigenvalues for three independent modes , which are now essentially decoupled from each other . 99 s.doplicher , k.fredenhagen and j.e.roberts , _ comm.math.phys . _ * 172 * 187 ( 1995 ) ; s.doplicher , k.fredenhagen and j.e.roberts , _ phys . b _ * 331 * ( 1994 ) . achieri p. , lizzi f. and vitale p. _ phys . d _ * 77 * 025037 ( 2008 ) . n.seiberg and e.witten _ jhep _ * 09*,032 ( 1999 ) . m. chaichian , p. p. kulish , k. nishijima and a. tureanu _ phys . b _ * 604 * 98 - 102 ( 2004 ) ; julius wess _ _ arxiv : hep - th/__**0408080**. a. p. balachandran , g. mangano , a. pinzul and s. vaidya _ int . j. modern phys . * 21 * 3111 - 3126 ( 2006 ) . chakraborty b.,gangopadhyay s. , a.g . hazra and f. g. scholtz _ j. phys . a _ * 39 * 9557 ( 2006 ) . p. basu , r. srivastava , s.vaidya _ phys . d _ * 82 * 025005 ( 2010 ) . p. basu , b.chakraborty and f.g . scholtz , _ j. phys . a _ * 44 * 285204 ( 2011 ) scholtz , b.chakraborty , j govaerts and s.vaidya , _ j.phys . a _ * 40 * 14581 ( 2007 ) . f.g.scholtz , l.gouba , a.hafver , c.m.rohwer _ j.phys . a _ * 42 * 175303 ( 2009 ) f g scholtz and b chakraborty _ j. phys . a : math . theor . _ * 46 * 085204 ( 2013 ) . a. connes 1994 _ non - commutative geometry _ ( new york : academic ) . balachandran a p , pinzul a , qureshi b a and vaidya s ( 2007 ) _ phys . d _ * 76 * 105025 . + balachandran a p , pinzul a , qureshi b a and vaidya s ( 2008 ) _ phys . . d _ * 77 * 025021 . sinha d. , chakraborty b. and f. g. scholtz _ j. phys . a _ * 45 * 105308 ( 2012 ) . s.adhikari , b. chakraborty , a.s . majumdar and s. vaidya , _ phys . * 79 * 042109 ( 2009 ) . j. williamson , _ am . _ * 58 * 141 ( 1936 ) . duan , g. giedke , j. i. cirac , and p. zoller , _ phys . lett . _ * 84 * , 2722 ( 2000 ) . fiore g. and wess j. _ phys . d _ * 75 * 1005022 ( 2007 ) . pathria r k 1996 _ statistical mechanics _ 2nd edn ( oxford : butterworth - heinemann ) s. gangopadhyay , a. saha and f. g. scholtz _ j. physics . a _ * 44 * 175401 ( 2011 ) . c m rohwer , l gouba , k zloschastyev , f g scholtz , _ * 43 * ( 2010 ) 345302 .

a unifying description of lattice potentials generated by aperiodic one - dimensional sequences is proposed in terms of their local reflection or parity symmetry properties . we demonstrate that the ranges and axes of local reflection symmetry possess characteristic distributional and dynamical properties which can be determined for every aperiodic binary lattice . a striking aspect of such a property is given by the return maps of sequential spacings of local symmetry axes , which typically traverse few - point symmetry orbits . this local symmetry dynamics allows for a classification of inherently different aperiodic lattices according to fundamental symmetry principles . illustrating the local symmetry distributional and dynamical properties for several representative binary lattices , we further show that the renormalized axis spacing sequences follow precisely the particular type of underlying aperiodic order . our analysis thus reveals that the long - range order of aperiodic lattices is characterized in a compellingly simple way by its local symmetry dynamics .

1307.1838

aperiodic sequences of distinct potential units have long served as a flexible model for long - range order which goes beyond conventional crystalline periodicity . as a means to approach disorder in an ordered manner , they pave the way for a fundamental characterization of condensed matter with respect to the combination of its structural and spectral properties @xcite . a decisive aspect of order in one dimension is the presence of local reflection symmetries in a potential , that is , of distinct symmetric constituents that add up to a ( typically ) globally non - symmetric structure . for infinitely extended systems , abundant local symmetries have been shown to lead to the absence of decaying eigenstates for the associated schrdinger operator @xcite . for finite structures , _ complete local symmetry _ , that is , decomposability into symmetric parts , was recently shown to underly the occurrence and enable the construction of a class of perfectly transmitting resonances ( ptrs ) in quantum scattering @xcite . a key ingredient that enables multiple ptrs in a single , globally non - symmetric potential , is the presence of nested local reflection symmetries on multiple scales . such a situation can be realized in one - dimensional ( 1d ) aperiodic lattices that are decomposable into symmetric parts in different ways , and pertains to the case of classical wave scattering in nanophotonic devices @xcite which are experimentally advantageous @xcite . independently of their physical impact , local symmetries in 1d aperiodic sequences have been extensively studied under the name palindromes in the combinatorics of words @xcite . here , the minimal reflection symmetric structural units are encoded as symbolic elements ( the letters ) of a set of given cardinality ( the alphabet ) , out of which discrete lattices ( the words ) are constructed by concatenation . aperiodically ordered sequences are constructed by the iterative action of a given substitution ( or inflation ) rule on the set of lattice elements . the presence of different palindromes ( words that are read the same forwards and backwards ) in a sequence is then expressed by its palindrome complexity function @xcite , which gives the number of contained palindromes of given length . rigorous mathematical results on the palindromicity of certain classes of ( infinite ) words have been obtained @xcite , an important part of which concerns palindromic prefixes ( factors in the beginning of a word ) @xcite . on the other hand , large effort has also been made for the computational determination of palindromes in arbitrary sequences @xcite , or even of gapped palindromes ( having a non - symmetric central part ) @xcite , which are directly related to genome structure @xcite . regarding complete local symmetry , mentioned above , the factorizability of finite words into ( maximal length ) palindromes has been demonstrated in closed form for binary words @xcite . aperiodic sequences are further studied in terms of the recurrence of factors or palindromes along given ( classes of ) words @xcite , thereby giving a dynamical aspect to their structural properties . in this context , an interesting question is whether and to what extent aperiodic sequences with arbitrary long - range order can be characterized by their _ local symmetry dynamics _ , that is , the evolution of the _ distances _ between subsequent palindromes along a symbolic sequence . implementing such a description could have two important advantages : ( i ) allow for a classification which is independent of the exact symbolic composition of the palindromes in the considered sequence and ( ii ) reveal structural complexity at a deeper level where information can be encoded in a more compact way . as a by - product , the analysis of local symmetry dynamics could offer the possibility to introduce new families of symbolic sequences by modifying only the underlying local symmetry dynamics . in this work , we address this task by considering the distribution of local reflection symmetries and the spacing of their axes in representative binary substitution sequences . in contrast to the class - specific , rigorous combinatorial statements provided in the literature , we here perform numerical experiments in order to shed light and to present a unifying viewpoint on the local symmetry properties of different classes of aperiodic lattices . the analysis reveals that the long - range order and complexity of well - known substitution sequences can be simply encapsulated within their local symmetry dynamics . in particular , we show that the return maps of palindrome spacings for given palindrome lengths consist of finite sets of strongly collinear points , with real - space trajectories which follow the original substitution rule . apart from compact characterization , this allows for an immediate recognition of qualitative but also quantitative common features and differences among various setups . the paper is organized as follows . in sec . [ notation ] the basic concepts of symbolic aperiodic sequences are introduced and defined , along with a notation suitable for our needs . in sec . [ lsdd ] the local symmetry properties of representative aperiodic sequences are individually addressed and analyzed . in particular , the different symmetry aspects including its distribution , dynamics , and scaling thereof are first introduced for the standard fibonacci lattice , and then examined for generalized fibonacci , period - doubling and thue - morse sequences , which are finally contrasted with the structurally different cantor lattice . section [ conclusion ] summarizes and concludes the present work . before starting the local symmetry analysis of aperiodic sequences , we briefly introduce here the global notation and terminology appropriate for our purposes . since we will consider , throughout this work , 1d potentials that are composed of mirror symmetric building blocks ( scatterers ) , we can make use of the basic definitions for symbolic sequences . different scatterers are thus symbolized by different _ letters _ @xmath0 , @xmath1 , which are elements of a finite _ alphabet _ @xmath2 of cardinality @xmath3 . a finite 1d lattice of scatterers thus corresponds to a combination of letters concatenated into a _ word _ @xmath4 of length @xmath5 . the set of all such words defines the _ language _ @xmath6 , which is the so called free monoid generated by @xmath2 , including the empty word @xmath7 of length zero ( the identity element of @xmath8 ) . within @xmath8 , words can thus also be concatenated to from new words . a _ factor _ @xmath9 of @xmath10 , corresponding to a number of consecutive scatterers in the lattice , is a word such that @xmath11 with @xmath12 . if @xmath13 ( @xmath14 ) , then @xmath9 is in the beginning ( end ) of the sequence and is called _ prefix _ ( _ suffix _ ) of @xmath10 . the reversal of the word @xmath10 above is defined as @xmath15 if @xmath16 , then @xmath10 is called _ palindrome _ , i.e. , a word which reads the same forwards and backwards . palindromes thus correspond to reflection symmetric sequences of scatterers , and palindromic factors to _ locally symmetric _ parts of a potential . we denote by @xmath17 the set of all palindromic factors of a ( finite or infinite ) word @xmath10 , and by @xmath18 its subset of palindromic factors of length @xmath19 . the cardinality of @xmath18 is the _ palindrome complexity function _ , @xmath20 , which thus gives the number of _ different _ palindromes of given length @xmath19 contained in @xmath10 , i.e. , irrespectively of their position . the palindromic factors of generic binary words have been extensively studied in terms of the associated palindrome complexity function @xcite , but also in relation to palindromic factorization @xcite , i.e. decomposition into locally symmetric parts . ( color online ) an aperiodic 1d lattice of two types of equidistant , arbitrary reflection symmetric scatterers ( depicted as squares and circles ) is mapped to a binary symbolic sequence of letters ( @xmath21 and @xmath22 , respectively ) , constituting the word @xmath10 ( whose shown part coincides with the @xmath23-th generation fibonacci word , see text ) . the letter symbols are set to unit length and centered at the positive integers @xmath24 on the @xmath25-axis . maximal palindromic factors of @xmath10 with more than one letters , denoted as @xmath26 $ ] , are depicted as horizontal line segments of length @xmath27 centered at @xmath28 , counted by the index @xmath29 in order of increasing @xmath28 . the index @xmath30 counts the maximal palindromes of common length @xmath31 occurring in @xmath10 , again in order of increasing position @xmath32 , as shown ( in red ) for the length @xmath33 . single - letter palindromes ( with @xmath34 , in this case @xmath21 ) are not shown . ] with the symbolic correspondence between a given 1d completely locally symmetric potential and a word @xmath10 , we map the symmetric scatterers to the letters @xmath35 ( of unit length ) of @xmath10 centered at the positive integers @xmath36 on the @xmath25-axis , and denote by @xmath26 $ ] the factor @xmath37 of @xmath10 ( see fig . [ fig.sketch ] ) . having introduced this coordinate system , any ( palindromic ) factor @xmath38 $ ] is represented as @xmath39 , where @xmath40 are its center and length , respectively ( that is , the reflection axis position and the range of this locally symmetric part in the lattice ) . the palindrome @xmath41 is called _ maximal _ if it is the largest palindrome centered at @xmath42 , i.e. , if any palindromic factor of @xmath10 centered at @xmath42 has length @xmath43 . we denote by @xmath44 the set of all pairs @xmath45 , @xmath46 , of coordinates of maximal palindromes @xmath47 with length @xmath48 , ordered in increasing @xmath28 along the @xmath25-axis . for each occurring maximal palindrome length @xmath49 , ( @xmath50 with @xmath51 ) , we define the subsets @xmath52 where the index @xmath30 counts the maximal palindromes of given length @xmath49 , again ordered in their position @xmath32 ( see fig . [ fig.sketch ] ) . mapping the elements of @xmath53 ( i.e. , word coordinates ) back to the corresponding words , @xmath54 we obtain the set @xmath55 of different maximal palindromic factors of length @xmath49 , with @xmath56 . the construction of a lattice with aperiodic order can be simply achieved by starting from one ( or more ) letter(s ) and then iteratively apply a single rule which substitutes letters ( or words ) with ( larger ) words , thereby ensuring the presence of long - range order among the scatterers of the emerging potential . such a substitution ( or _ inflation _ ) rule @xmath57 is formally defined as a map @xmath58 , supplied with the property @xmath59 as well as @xmath60 in particular , its action on the word @xmath61 , @xmath62 , simply becomes @xmath63 in the following , we will consider lattices corresponding to words on a two - letter ( binary ) alphabet @xmath64 , constructed through an inflation rule @xmath57 defined by its action on the single letters , using the shorthand notation @xmath65 to introduce the concept of _ local symmetry dynamics _ , which is the main task of this section , we focus on the space of maximal palindromes , that is , the maximal locally symmetric parts of the corresponding scatterer lattice . as a first step we classify the maximal palindromes , as they appear in the infinite word of the considered symbolic sequence , with respect to their length @xmath49 , where @xmath66 counts the possible different length values in increasing order : @xmath67 then , for a given @xmath49 , we determine the positions @xmath68 of the symmetry axes of the maximal palindromes with this length ( i.e. , those belonging to the set @xmath53 ) , and represent their distribution through the plot @xmath49 vs. @xmath68 for @xmath69 . the index @xmath70 counts the maximal palindromes of given length @xmath49 for increasing symmetry axis position @xmath68 . this local symmetry distribution characterizes the palindrome composition of the considered symbolic sequences in a symbol - independent manner , and can be used for their classification with respect to structural features . the local symmetry dynamics is then determined for each @xmath66 by the change of the distance between subsequent maximal palindrome symmetry axes , @xmath71 , as @xmath70 increases . we choose here to present this dynamics through the return map @xmath72 vs. @xmath73 , which turns out to be an excellent tool to reveal dynamical scaling properties as well as correlations at different scales . the latter are , as will be explained below , directly related to the fact that the considered symbolic sequences possess nested local symmetries , that is , systematically and multiply overlapping symmetric parts of the lattice . the detailed local symmetry dynamics , which will be shown to follow the corresponding original inflation rule of a given sequence , is finally illustrated by explicit @xmath72-trajectories for selected @xmath66 . in the following we will present results of this analysis applied to five different types of aperiodic sequences : ( a ) fibonacci , ( b ) generalized fibonacci , ( c ) period doubling , ( d ) thue - morse and ( e ) cantor lattices . let us first analyze the local symmetries of a lattice composed according to the famous fibonacci sequence , which constitutes a prominent example of quasiperiodic order @xcite . as a symbolic two - letter sequence , the @xmath74-th order fibonacci word is generated by iterative application of the inflation rule @xmath75 conventionally starting with the letter @xmath21 : @xmath76 for @xmath77 , it has the recursion property @xmath78 so that its length @xmath79 follows the numeric fibonacci sequence @xmath80 , with @xmath81 the so - called _ golden _ mean . denoting by @xmath82 the infinite fibonacci word , we note that any sufficiently long factor of @xmath83 ( i.e. , not necessarily a @xmath74-th order prefix @xmath84 ) is characterized by the aperiodic order induced by the inflation rule . ( color online ) local symmetries of the fibonacci lattice , corresponding to maximal palindromes of length @xmath27 in the letter sequence @xmath83 , represented as line segments in order @xmath29 of their occurrence along the @xmath25-axis ( see coordinates defined in fig . [ fig.sketch ] ) . those maximal palindromes of lengths @xmath85 are shown , which are contained within the @xmath86-th order fibonacci word @xmath87 . the inset shows the @xmath88-th generation word @xmath89 ( the generations are counted by the index @xmath74 ) , containing the first four prefix palindromes with @xmath90 . @xmath89 can be decomposed into a prefix maximal palindrome @xmath91 and a non - maximal palindrome @xmath92 ( distinguished by a the dashed line ) , while its last prefix palindrome @xmath93 reaches up to its two last letters @xmath94 ( see text ) . ] the local reflection symmetries of the fibonacci lattice are illustrated in fig . [ fig.fib-pal ] , where its maximal palindrome coordinates @xmath95 , with @xmath29 ordered in increasing @xmath28 , are shown ( excluding the single - letter palindromes ) . the infinite fibonacci word @xmath83 belongs to the class of _ sturmian _ words @xcite , which can in fact be defined as the ( infinite ) binary words with palindromic complexity @xcite @xmath96 note , however , that the number of different _ maximal _ palindromes of given length @xmath19 is generally _ not _ given by @xmath20 . e.g. , for @xmath97 , @xmath98 , but @xmath99 , since @xmath100 is not a factor of @xmath83 and therefore @xmath101 can not be maximal . further , sturmian aperiodicity can be geometrically connected to irrational numbers ( like @xmath102 ) by the so - called cut and project scheme @xcite : they are constructed from the projection of the vertices of a square two - dimensional lattice , lying within a stripe of given width @xcite and with irrational slope , onto the direction of the stripe . the segments between the projected vertices ( being of two possible lengths ) are mapped to letters @xmath21 and @xmath22 , thus constituting a word . the slope of the stripe is also called the slope of the word ; if it equals @xmath103 , one obtains @xmath83 . regarding their decomposability , finite sturmian words can be uniquely written as @xcite @xmath104 where @xmath105 are palindromes and @xmath106 or @xmath94 . for the fibonacci word generations @xmath107 , we have @xmath106 ( @xmath94 ) for even ( odd ) @xmath74 . this can be seen in the inset of fig . [ fig.fib-pal ] for the generations @xmath108 , with corresponding prefix palindrome lengths @xmath109 . as also seen , the palindromic decomposition can in general not be achieved solely with maximal palindromes . e.g. , for @xmath110 , the palindrome prefix with @xmath111 is maximal , but concatenated to a non - maximal palindrome with @xmath112 . further , any sufficiently large prefix of the fibonacci sequence can be factorized in multiple ways into palindromes of different lengths @xcite . therefore , any finite scatterer lattice with fibonacci aperiodic order is _ completely locally symmetric _ ( completely decomposable into locally symmetric parts ) at _ multiple _ scales . this property relies on the distribution of palindromes of each given length along the sequence , determining which palindromes can be successively concatenated to construct a given finite lattice . local symmetry distribution of the fibonacci lattice . pairs @xmath45 of axis positions and lengths of maximal palindromes @xmath47 , are plotted on base-@xmath113 logarithmic scale . for each occurring symmetry range @xmath49 ( @xmath114 , @xmath115 not plotted ) , the index @xmath70 counts the maximal palindromes @xmath116 with @xmath31 , ordered in increasing axis position @xmath32 ( horizontally collinear points ) . the inset shows the linear distribution of local symmetry axes for the first occurring ranges . ] to analyze the local symmetry distribution of the fibonacci lattice , we plot in fig . [ fig.fib-pal-cl ] the elements of @xmath117 , i.e. , the length @xmath27 of each maximal palindrome versus its axis position @xmath28 ( excluding pairs corresponding to single letter maximal palindromes ) . as we see , only certain lengths occur , which constitute the sequence @xmath118 ( see inset of fig . [ fig.fib-pal-cl ] ) . a remarkable feature of the sets @xmath119 ( horizontally collinear points in fig . [ fig.fib-pal-cl ] ) is that their first element @xmath120 always corresponds to a _ prefix _ of @xmath83 , so that @xmath121 in particular , for the fibonacci sequence we thus have @xmath122 corresponding to the palindrome @xmath93 in @xmath123 , as defined above in eq ( [ eq.sturm-dec ] ) , for each word generation @xmath124 @xcite . this means that , at these special axis positions arbitrarily deep within the lattice , there occur local symmetries which extend to the beginning ( left boundary ) of it , which we coin _ complete local symmetries_. in view of wave scattering , such complete local symmetries are important , because they can render the lattice transparent @xcite over a large part from its beginning , despite the complex , aperiodic character of the underlying medium . in particular for the fibonacci lattice , the trailing two scatterers ( @xmath125 or @xmath94 in eq . ( [ eq.sturm-dec ] ) ) could be discarded from any fibonacci word @xmath107 , in order to produce perfectly transmitting resonances @xcite in a globally symmetric system with quasiperiodic order . as anticipated from the above , the ratio of successive palindromic lengths in @xmath83 converges to the golden mean , @xmath126 in fact , the fibonacci word @xmath83 is a special case of sturmian words with _ abundant _ palindromic prefixes , studied in ref . @xcite : it is the one with the smallest ratio @xmath127 ( apart from periodic words with @xmath128 ) , thus being the most dense in prefix palindromes . furthermore , the fact that each prefix palindrome is centered within the previous one ( since @xmath129 ) , suggests that @xmath83 can be _ constructed _ in terms of its palindromic prefixes , as is indeed the general case for this class of words @xcite . ( color online ) ( a ) local symmetry spacing return maps for the fibonacci lattice . for each occurring symmetry range @xmath49 , each distance @xmath130 between axes of consecutive palindromes is plotted versus the previous distance @xmath131 , with full ( empty ) circles for even ( odd ) @xmath66 . each map consists of three possible spacings @xmath132 ( @xmath133 ) , and consecutive map points are connected by lines . note that the plot consists of multiple disconnected maps plotted together , and distinguished by the alternately filled / empty circles . the inset repeats the plot on base-@xmath113 logarithmic scale , with maps labeled by the corresponding @xmath49 ; thick arrows on the lowest map indicate possible direction of traversal of any map . ( b ) local symmetry spacing trajectory ( i.e. , the @xmath134 plotted in order of occurrence @xmath70 for fixed @xmath66 ) for range @xmath135 in the fibonacci lattice . by mapping the pairs @xmath136 and @xmath137 of consecutive spacings ( marked by dotted boxes ) to letters @xmath21 and @xmath22 , respectively , the trajectory reproduces the original sequence ( see text ) . ] let us now turn to the local symmetry dynamics of the fibonacci lattice , that is , the evolution of consecutive _ spacings _ between maximal palindromes of given length along the word @xmath83 . [ fig.fib-pal-spacing](a ) shows , for each length @xmath49 , the return map of the spacings @xmath138 between consecutive local symmetry axes ( @xmath134 is also indicated in figs . [ fig.sketch ] and [ fig.fib-pal-cl ] ) . the @xmath134 for each @xmath49 take on only three possible values @xmath139 in a succession which forms a four - point orbit in each return map . we see that , for any given symmetry range @xmath49 , none of the spacings @xmath132 is repeated consecutively as @xmath70 increases ( @xmath140 ) , and that @xmath134 returns to the minimal spacing @xmath141 for every second @xmath70 . we further observe that the spacing return maps of consecutive local symmetry ranges are correlated in the following two - fold _ nested _ sense : firstly , the middle spacing for each palindrome length coincides with the minimal spacing for the next length , @xmath142 secondly , we have that @xmath143 so that consecutive return maps have four collinear points along the lines with slope @xmath144 , as seen in the linear plot of fig . [ fig.fib-pal-spacing](a ) . with respect to the spacing return maps , there are here two properties which demonstrate how the aperiodic order of the fibonacci sequence is encoded in its local symmetry dynamics : + ( i ) the sequence of minimal spacings coincides with the numeric fibonacci sequence , @xmath145 counting here also the single - letter maximal palindromes @xmath21 ( not plotted ) . + ( ii ) the spacing return maps for different @xmath49 present a characteristic scaling , which converges @xcite to the inverse slope of @xmath83 , @xmath146 the quasiperiodicity of the fibonacci lattice is , finally , manifest also in the individual _ trajectories _ of the axis spacings for given symmetry ranges , i.e. the order in which the points of the return map orbits are traversed ( this information is obviously not available from the return maps alone ) . the spacing trajectory for @xmath135 is shown in fig . [ fig.fib-pal-spacing](b ) . we see that the sequence of spacings for odd @xmath70 , with elements @xmath147 , goes like the original fibonacci sequence , with returns to @xmath148 for every even @xmath70 . this is indeed true for any local symmetry range , yielding a third remarkable property : + ( iii ) for any maximal palindrome length @xmath49 , the sequence of spacings @xmath149 is generated as @xmath150 just like the generations of @xmath83 in eq . ( [ eq.fib ] ) , but with the fibonacci substitution rule @xmath151 now acting on the renormalized spacing alphabet @xmath152 with additional starting letter @xmath153=b$ ] ( see fig . [ fig.fib-pal-spacing](b ) ) . note that the spacings @xmath132 are here treated simply as symbols being concatenated into a sequence . equivalently , the sequence @xmath83 can itself be renormalized by mapping its letters @xmath22 and ( squares of ) @xmath21 to the occurring spacings for any given @xmath49 as follows : @xmath154 with the @xmath22 s thus mapped to the spacing returned to at every second step . renormalized according to eqs . ( [ eq.fib-renorm ] ) , the original fibonacci sequence is mapped exactly to the three - letter ( or _ trinary _ ) sequence @xmath155 : @xmath156 from the above properties ( i)(iii ) , the total local symmetry ( at different spatial scales ) of the fibonacci lattice can be seen to possess a _ dynamical self - similarity _ , which uniquely characterizes its long - range aperiodic order . the fibonacci inflation rule can be generalized @xcite to powers of @xmath21 and @xmath22 as @xmath157 where the word generations @xmath158 are equivalently given by the recursion scheme @xmath159^m[w^{(k-2)}_{m , n}]^n , ~k \geqslant 2.\ ] ] their lengths are , accordingly , @xmath160 . the generated infinite word @xmath161 then has limiting word length ratio @xcite @xmath162 for the standard fibonacci word we thus have @xmath163 . for general letter powers @xmath164 the resulting lattice is aperiodic , while for @xmath165 it is also quasiperiodic @xcite . having studied the local symmetries of the standard fibonacci lattice , we now briefly address two further examples with quasiperiodic order . ( color online ) silver fibonacci sequence , @xmath166 . ( a ) local symmetry distribution ( like in fig . [ fig.fib-pal-cl ] ) . ( b ) spacing return maps for different @xmath49 , with full ( empty ) cirles for even ( odd ) @xmath66 , repeated on logarithmic scale in the inset ( like in fig . [ fig.fib-pal-spacing](a ) ) . ( c ) spacing trajectories for even ( top ) and odd ( bottom ) @xmath66 , represented by @xmath167 and @xmath168 , respectively . ] in fig . [ fig.fib21](a ) the local symmetry distribution ( @xmath169 ) is shown for the lattice @xmath166 , with @xmath170 referred to as the _ silver _ mean . again , the first occurring palindromes @xmath171 for every given length @xmath172 ( @xmath173 not plotted ) are prefixes ; i.e. , eq . ( [ eq.prefix ] ) holds . however , the @xmath49 now form two subsequences for odd and even @xmath66 having , as we will see , different structural properties . for odd @xmath66 , the @xmath49 are given by the lengths of the @xmath66-th order generated words with the last two letters discarded , @xmath174 like in the standard fibonacci case . the ranges for even @xmath66 are asymptotically related to the above as @xmath175 thus , for both odd and even sequences the ratio of consecutive prefix palindromes converges to the silver mean , @xmath176 , in analogy to eq . ( [ eq.lambda ] ) . the local symmetry dynamics of @xmath166 is shown in fig . [ fig.fib21](b ) , as represented by the return maps of the spacings @xmath134 for each symmetry range @xmath49 . for each @xmath66 we have three possible , non - repeated spacings @xmath132 ( @xmath133 ) , as in the standard fibonacci case , now with @xmath177 which scale asymptotically as @xmath178 for even ( odd ) @xmath66 , there are consecutive returns to @xmath141 ( @xmath179 ) in each map . again , the minimal spacings follow the numeric sequence of lengths of @xmath180 , but now pairwise : @xmath181 the odd and even subsequences of symmetry ranges are further nested through the relations @xmath182 also , the spacings in consecutive return maps obey , with @xmath66 odd , @xmath183 so that the pairs @xmath66 , @xmath184 have two lines with collinear points ( see linear plot in fig . [ fig.fib21](b ) ) . the presence of the @xmath166 aperiodic order in its detailed local symmetry dynamics is illustrated in fig . [ fig.fib21](c ) for @xmath185 and @xmath186 . similarly to the golden fibonacci case , the actual spacing trajectories @xmath134 for given @xmath49 coincide with the original aperiodic sequence @xmath166 , when renormalized and shifted by @xmath187 : the corresponding sequence @xmath188 of spacing symbols @xmath132 is generated , in analogy to eq . ( [ eq.fibspacing ] ) , as @xmath189 ( @xmath190 ) by the substitution rule @xmath191 on the alphabet @xmath192 for odd ( [ eq.fib-renorm-odd ] ) or even ( [ eq.fib-renorm-even ] ) @xmath66 , respectively , again with additional starting letter @xmath193=b$ ] . the grouping of spacings into letters is depicted by the dotted lines in fig . [ fig.fib21](c ) . conversely , the letter sequence @xmath166 can be renormalized by mapping the single @xmath22 s and the squares and cubes of @xmath21 to the occurring spacings for given @xmath49 , @xmath194 for even ( odd ) @xmath66 . the original sequence @xmath166 is then renormalized into the trinary sequence @xmath188 of the spacing trajectory , @xmath195 for even ( odd ) @xmath66 ; see fig . [ fig.fib21](c ) . this latter mapping of the original aperiodic sequence onto its local symmetry spacing dynamics , which was described already for the golden fibonacci case ( eqs . ( [ eq.fib-renorm ] ) and ( [ eq.fib-renorm-seq ] ) ) , can be intuitively anticipated from the local symmetry distribution : the pattern of symmetry spacings for certain maximal range is expected to follow , in some way , the pattern of isolated @xmath22 s among powers of @xmath21 , which in turn is directly given in the original letter sequence and indeed , this way is exactly provided by eqs . ( [ eq.fib-renorm-seq ] ) and ( [ eq.fib21-renorm-seq ] ) . moreover , this renormalization scheme can be extended to larger @xmath196 in the class of quasiperiodic generalized fibonacci lattices , as will be seen in the next subsection . ( color online ) generalized fibonacci @xmath197 lattice . local symmetry ( a ) distribution and ( b ) spacing return maps ( like in fig . [ fig.fib21](a ) and ( b ) , respectively ) . the curved arrows in ( a ) distinguish the @xmath198 subsequence of symmetry ranges ( see text ) , and the shaded areas show ( a ) the @xmath199 scaling quintet bundle and ( b ) the corresponding spacing return maps . the straight vertical arrows in ( a ) , starting from @xmath200-like structures , show the jumps to larger symmetry ranges belonging to the next ( solid lines ) or next - to - next ( dashed lines ) quintet bundle ( see text ) . ] as we have seen above for the quasiperiodic case @xmath165 , the @xmath201 ( golden ) and @xmath202 ( silver ) sequences feature one and two characteristic asymptotic scaling(s ) for the spacing of local symmetries , respectively . this scheme continues for larger @xmath196 , so that @xmath196-tets of asymptotic spacing scalings arise . we will use the index @xmath203 to count the members of an @xmath196-tet : the @xmath204-th spacing scaling in an @xmath196-tet corresponds to the subsequence composed of every @xmath196-th occurring maximal local symmetry range , starting with @xmath205 ( e.g. , for @xmath206 we get the third such subsequence , @xmath207 , marked in fig . [ fig.fib51](a ) ) @xcite . equal - order members of these @xmath196 range - subsequences form @xmath196-tet bundles ( e.g. , the bundle of second members of the @xmath208 subsequences is depicted in fig . [ fig.fib51](a ) ) . for any @xmath196 , the first occurring palindromes for given ranges @xmath49 are prefixes of @xmath209 . also , although in general the @xmath196-tet subsequences have different palindrome _ spacing _ scalings , their palindrome _ lengths _ share the same scaling , given by the inverse slope of @xmath209 ( like the cases @xmath210 seen above ) : @xmath211 further , the last length in each @xmath196-tet bundle follows the length sequence of @xmath209 in the same way as seen so far for @xmath201 and @xmath212 ( with @xmath77 for @xmath201 ) , @xmath213 as an example , we briefly consider the case @xmath214 , i.e. the sequence @xmath197 . in fig . [ fig.fib51](a ) , its local symmetry distribution is plotted , showing a grouping of the ranges @xmath49 into quintets ( horizontal bundles at increasing scale , with five ranges each ) . it is evident how , due to the recurrent @xmath215-factors , axes of maximal palindromes located at the left and right ends of a larger palindrome are shifted inwards for increasing @xmath49 , thereby forming characteristic @xmath200-like structures , along the sequence ( i.e. , the @xmath28-axis ) and at increasing scales ( i.e. , the @xmath27-axis ) . at the meeting point @xmath28 of the legs of each such @xmath200 , a larger local symmetry occurs , with corresponding range @xmath49 belonging to the next or next - to - next @xmath196-tet bundle ( as indicated by vertical arrows in fig . [ fig.fib51](a ) ) ; see also fig . [ fig.fib101-fib110-pal-cl](a ) , showing the bundle structure of the @xmath216 sequence . the spacing dynamics for the different symmetry ranges become increasingly complex for higher @xmath196 , as can be seen from the return maps for @xmath208 in fig . [ fig.fib51](b ) . just as the odd - even maps in the @xmath202 case , the maps corresponding to @xmath49 s of the same @xmath196-tet bundle in fig . [ fig.fib51](a ) are collinear , giving evidence to a nested order of symmetry dynamics along the lattice . out of the @xmath196 ( different ) spacing scalings of the @xmath196-tet , the asymptotic scaling for the _ leading _ subsequence ( the sequence of smallest symmetry ranges in subsequent @xmath196-tet bundles ) is directly related to the slope of @xmath209 as ( cf . cases @xmath210 above , eqs . ( [ eq.fib-spacing ] ) , ( [ eq.fib21-spacing ] ) ) @xmath217 finally , as mentioned above , suitable renormalization of the original sequence @xmath209 , by mapping the single @xmath22 s and powers of @xmath21 to the spacings @xmath132 , again yields the spacing trajectory . in particular , for all @xmath49 contained in any @xmath204-th range - subsequence ( @xmath218 ) , there exists a common mapping @xmath219 so that @xmath220 with the different @xmath221 depending on the chosen @xmath204 ( cf . cases @xmath210 above , eqs . ( [ eq.fib-renorm-seq ] ) , ( [ eq.fib21-renorm-seq ] ) ) . ( color online ) copper fibonacci lattice , @xmath222 . local symmetry ( a ) distribution , ( b ) spacing return maps and ( c ) spacing trajectories for even and odd @xmath66 ( like in fig . [ fig.fib21](a ) , ( b ) and ( c ) , respectively ) . ] as an example of a non - quasiperiodic generalized fibonacci sequence ( i.e. , @xmath223 in eq . ( [ eq.fibgen ] ) ) we consider the case @xmath224 in eq . ( [ eq.fibgen ] ) . the lengths of the word generations @xmath225 are recursively given by @xmath226 , and have limiting ratio @xmath227 , known as the _ copper _ mean . the local symmetry distribution of @xmath222 is shown in fig . [ fig.fib12](a ) . like in the @xmath166 case , the sequence of occurring maximal palindrome lengths , @xmath228 ( @xmath173 not plotted ) , is split into two subsequences for odd and even @xmath66 . here , though , for the subsequence of even @xmath66 the first occurring maximal palindromes ( with axes @xmath229 ) are _ not _ prefixes of @xmath222 , that is , no complete local symmetries of those ranges are encountered . further , the two @xmath49 subsequences are related to the length sequence @xmath230 in a more involved way , through their _ difference _ @xmath231 , @xmath232 , as follows . for odd @xmath66 , @xmath233 coincides with the sequence of lattice generation lengths , while for even @xmath66 it oscillates around it by alternately @xmath234 : + @xmath235 as we see , the non - quasiperiodic @xmath222 sequence features more indirect connection of its palindrome distribution to the structure of the original letter sequence than the previous cases ( i.e. , the quasiperiodic @xmath209 sequences ) . its asymptotic palindrome length scaling , on the other hand , is the same for the odd and even @xmath66 subsequences , @xmath236 with asymptotic even - to - odd-@xmath66 ratio @xmath237 the spacing return maps for different local symmetry ranges in @xmath222 are shown in fig . [ fig.fib12](b ) . there are again @xmath238 possible spacings in each map , and now any map point for @xmath239 lies on an antidiagonal line with @xmath240 collinear points , @xmath23 for each subsequence ( odd and even @xmath66 ) of lengths @xmath49 . this connects the consecutive maps of each subsequence , but also fixes the subsequences to each other , as seen in the linear plot of fig . [ fig.fib12](b ) . the asymptotic scalings of the odd and even @xmath66 spacings with respect to the @xmath141 in the maps is @xmath241 for the middle spacings , and @xmath242 for the maximal spacings . also here the renormalized spacing trajectories for given @xmath49 coincide with the original aperiodic sequence @xmath222 : the sequence of spacings is given iteratively by @xmath243 on the alphabet @xmath244 for odd or even @xmath66 , respectively , with additional starting letter @xmath245=b$ ] . this is shown for the @xmath185 and @xmath186 trajectories in fig . [ fig.fib12](c ) . local symmetry distributions of the generalized fibonacci lattices ( a ) @xmath216 , with separate @xmath196-tet ( here @xmath246 ) bundles forming along the @xmath28-axis ( in analogy to fig . [ fig.fib51](a ) ) , and ( b ) @xmath247 , with intertwined @xmath248-tet ( here @xmath249 ) bundles ( see text ) . ] to illustrate the characteristic differences between the local symmetry properties of the quasiperiodic ( @xmath165 ) and non - quasiperiodic ( @xmath223 ) generalized fibonacci lattices for larger @xmath196 and @xmath248 , fig . [ fig.fib101-fib110-pal-cl ] shows the maximal palindrome distributions for the sequences @xmath216 and @xmath247 . it is clearly seen how @xmath196- and @xmath248-tets of maximal symmetry range subsequences occur ( in fig . [ fig.fib101-fib110-pal-cl](a ) and ( b ) , respectively ) , which form bundles containing consecutive ( in @xmath250-direction ) @xmath200-like structures of approaching symmetry axes for increasing @xmath49 , like in fig . [ fig.fib51](a ) . at the meeting points of the legs of the @xmath200 s larger symmetry ranges occur , belonging to a subsequent ( in @xmath27-direction ) @xmath196- or @xmath248-tet bundle . in the quasiperiodic case , [ fig.fib101-fib110-pal-cl](a ) , the first maximal palindrome of each occurring length ( i.e. , all points on the left leg of each first - occurring @xmath200 ) is a prefix of @xmath209 , and the @xmath196-tet bundles are separated . on the contrary , in the non - quasiperiodic case , [ fig.fib101-fib110-pal-cl](b ) , prefix palindromes occur only at the meeting points of first - occurring @xmath200 s , and the @xmath248-tet bundles are , in general , intertwined ( i.e. , bundles overlap with successors containing different @xmath49 s ) . ( color online ) period doubling lattice , @xmath251 . local symmetry ( a ) distribution , ( b ) spacing return maps and ( c ) spacing trajectories for even and odd @xmath66 ( like in fig . [ fig.fib21](a ) , ( b ) and ( c ) , respectively ) . ] let us now inspect the local symmetries of the period - doubling ( pd ) lattice , whose extensively studied structural and spectral properties @xcite manifest its self - similar nature , although it is not categorized as quasicrystalline @xcite . the pd letter sequence @xmath252 , with @xmath253 , is generated by the inflation rule @xmath254 that is , like a standard fibonacci rule but with a squared image of @xmath22 , and the length of the @xmath74-th generation word is now @xmath255 . its recursion scheme is @xmath256 ^ 2 $ ] , which coincides with that of the copper fibonacci sequence @xmath222 . in fact , @xmath251 is obtained from @xmath222 simply by substituting all squares @xmath100 by single @xmath22 s . we can thus expect these lattices to have certain structural similarities . in fig . [ fig.pd](a ) the local symmetry distribution of @xmath251 is shown , which , indeed , has the same characteristics as the distribution for @xmath222 described above . again , the sequence of occurring lengths , @xmath49 , consists of two subsequences for odd and even @xmath66 ( i.e. , doublets of local symmetry scaling ) , the former of which contains all palindromic prefixes . the lengths of the palindromes in the odd and even subsequences relate to the @xmath251 word generation lengths as [ eq.pd-length ] @xmath257 where @xmath258 .. from these relations it is obvious that the maximal palindrome lengths in each ( odd and even @xmath66 ) subsequence scale asymptotically as the original pd sequence , @xmath259 , with @xmath260 , just like for the copper fibonacci sequence ( eqs . ( [ eq.fib12-length-scaling ] ) and ( [ eq.fib12-length-ratio ] ) , respectively ) . moreover , it is clear that each palindrome prefix , having odd length @xmath49 , contains the previous prefix palindrome of length @xmath261 ( for odd @xmath262 ) in its left half , i.e. , up to ( but not including ) its central letter . specifically , it is composed as @xmath263 simultaneously , though , its prefix @xmath264 coincides with a word generation @xmath265 , which ends with the letter @xmath21 ( @xmath22 ) for even ( odd ) @xmath74 . this means that the pd sequence can be constructed solely from its local symmetries , by successively creating larger prefix palindromes from the repetition of the previous prefix palindrome with an additional central letter @xmath266 , which follows a period @xmath267 oscillating sequence , @xmath268 : beginning with @xmath269 , we have @xmath270 , @xmath271 , ... , and , in general , @xmath272 in this way , we obtain the relation @xmath273 as a recursive palindromic construction of the pd sequence via the directive periodic sequence @xmath274 . the palindrome complexity function of the pd sequence , i.e. the number of different contained palindromes of length @xmath19 , has been explicitly computed in ref . @xcite , and is given by @xmath275 with starting palindromes @xmath276 , @xmath277 ( this being the only palindrome with even length ) , @xmath278 , @xmath279 , and @xmath280 . nevertheless , we note that , for any odd @xmath281 , the palindromes of length @xmath282 are not _ maximal _ ( see the inset of fig . [ fig.pd](a ) ) , that is , @xmath283 further , even for lengths @xmath284 , not _ all _ palindromes are maximal ; e.g. , @xmath285 , @xmath286 , and @xmath287 ( since @xmath288 and @xmath289 ) . the return maps of the local symmetry dynamics of the pd lattice are shown in fig . [ fig.pd](b ) . their structure very much resembles , qualitatively , the maps for the copper @xmath222 sequence : the subsequences of maps for lengths @xmath49 with odd and even @xmath66 are intertwined in exactly the same way by antidiagonally collinear map points . quantitatively , however , the pd local symmetry spacings evolve more regularly along the lattice than for the copper fibonacci , with characteristics which are invariant ( as opposed to asymptotic ) in increasing @xmath66 , as follows : for all odd @xmath66 ( red circles in fig . [ fig.pd](b ) ) , there are now two occurring spacings @xmath290 ( and not _ asymptotically _ two , as in @xmath222 ) with repeated @xmath141 , and three non - repeated spacings for even @xmath66 . in each subsequence , all spacings are _ doubled _ in the subsequent symmetry range , @xmath291 further , the two map subsequences are fixed to each other by the relations @xmath292 so that each odd map has two common points with the next - to - next even one . the scaling of local symmetry spacings along the range sequences is also fixed as @xmath293 for all @xmath66 and @xmath294 for even @xmath66 . finally , the spacing trajectories for even and and odd @xmath66 are shown in fig . [ fig.pd](c ) . again the ( renormalized ) trajectories are generated by the original inflation rule @xmath295 . for even @xmath66 , we have the alphabet mapping @xmath296 for odd @xmath66 it is the same but with @xmath297 , and with additional starting spacing @xmath298 = d^{(r)}_0=d^{(r)}_1 $ ] . ( color online ) thue - morse lattice , @xmath299 . local symmetry ( a ) distribution and ( b ) spacing return maps ( like in fig . [ fig.pd](a ) and ( b ) , respectively ) . spacing trajectories coincide with the ones in fig . [ fig.pd](c ) . ] we now turn to another 1d lattice case with doublet local symmetry sequences , corresponding to the well - known ( standard ) thue - morse ( tm ) sequence , @xmath299 , which shares the non - quasicrystalline spectral characterization of the pd lattice @xcite . identifying @xmath300 with \{even , odd } , @xmath299 can be defined as the parity sequence of the sum of digits of base-@xmath212 represented integers , but is otherwise ubiquitously present in various fields of mathematics and physics @xcite . as a symbolic sequence , @xmath299 is generated by the inflation rule @xmath301 and can be generalized in a similar manner as for the fibonacci substitutions studied above @xcite , see eq . ( [ eq.fibgen ] ) . the corresponding recursive concatenation scheme now has a two - component form @xcite @xmath302 for @xmath303 , starting with @xmath304 , @xmath305 , and the length sequence coincides with that of @xmath251 : @xmath306 . the combinatorial palindromic properties of the tm sequence have , among others , been studied in refs.@xcite , where its palindrome complexity is also given , and connected to its hamiltonian spectrum in ref.@xcite . like for the pd sequence , we see that the standard tm substitution is very similar to the copper rule @xmath307 , where now @xmath22 is sent to the conjugate image @xmath94 of @xmath21 , instead of the square @xmath308 . on the other hand , @xmath309 leads to the occurrence of the conjugate square @xmath100 in @xmath299 . as a consequence , the local symmetries in @xmath299 will have ranges different from @xmath251 , but partly feature similar structural characteristics , with depleted number of symmetry ranges due to the absence of the cube @xmath310 . in fig . [ fig.tm](a ) the local symmetry distribution of @xmath299 is shown . indeed , we see again a doublet structure of two symmetry range subsequences @xmath311 , but now every second doublet bundle is missing compared to @xmath251 or @xmath222 . this is also reflected in the length sequences themselves , which are squared with respect to the @xmath251 case ( cf . ( [ eq.pd-length ] ) ) : @xmath312 with @xmath258 . there is now no residue letter(s ) to reach ( a multiple of ) the @xmath313 generation lengths ( like the @xmath144 for @xmath251 in eq . ( [ eq.pd-length ] ) ) , so that both subsequences have the same , constant scaling throughout the lattice , @xmath314 . also , in contrast to the pd case , here all @xmath49 are even except for the second one @xmath315 ( and , of course , @xmath173 ) . the depletion of local symmetry ranges in the tm lattice is also evident in the spacing return maps , shown in fig . [ fig.tm](b ) : every second return map in each subsequence is now missing with respect to the pd case , so that no overlap between maps occurs , and the spacings are accordingly multiplied by four at subsequent symmetry ranges , @xmath316 . for the remaining symmetry ranges , the ( fixed ) spacing scaling , @xmath293 , @xmath317 , is identical to the pd lattice . remarkably enough , identical are also their spacing trajectories . that is , despite the different inflation rules @xmath309 and @xmath295 , and although the @xmath49 in @xmath299 are different ( one letter larger ) from the corresponding ones @xmath251 , the spacings in the tm case are exactly the same as for the pd case , and follow the trajectories shown in fig . [ fig.pd](c ) for odd and even @xmath66 . this suggest a subtle ( renormalization ) relation between the tm and pd lattice , so that the former is also structurally conform to its maximal palindrome spacing trajectories . indeed , the tm lattice @xmath318 can be transformed to the pd lattice @xcite by the ( @xmath212-to-@xmath267 , surjective ) map @xmath319 this more or less hidden relation between the two maps is here directly ( i.e. , prior to any renormalization ) reflected in the coincidence of their local symmetry dynamics . cantor lattice , @xmath320 . ( a ) maximal palindromes in order of occurrence in the letter sequence , the beginning of which is shown in the upper inset . the magnification in the lower inset demonstrates the self - similarity of the lattice . ( b ) local symmetry distribution ( like in fig . [ fig.fib-pal-cl ] ) . ] having analyzed the local symmetry distribution and dynamics of some typical binary lattices with aperiodic order , we finally examine an extreme case with inherent palindromic self - similarity , the standard cantor lattice . in general , a 1d ( @xmath196-fold ) cantor construct is obtained by dividing a finite segment into ( @xmath196 equal ) parts , deleting a chosen number ( @xmath321 ) of them which are not consecutive , and repeating this procedure recursively on each remaining segment ad infinitum ( thus generating an @xmath196-by-@xmath248 cantor fractal ) . mapping the resulting full and empty intervals onto a binary alphabet ( @xmath21 s and @xmath22 s , respectively ) , an infinite fractal letter sequence is obtained . the sequence can be obtained equivalently by a substitution rule on the letters corresponding to the above partitioning scheme . we here consider the homogeneous @xmath238-by-@xmath267 ( or ternary ) cantor lattice @xcite , which is obtained by recursively applying the inflation rule @xmath322 with starting letter @xmath21 , yielding the sequence @xmath320 . obviously , @xmath323 inherently generates palindromes on all scales with ever increasing lengths , distributed self - similarly on a globally symmetric lattice . the hierarchical local symmetry structure is clearly evident in fig . [ fig.cantor](a ) , where the maximal palindromes of @xmath320 are plotted as horizontal segments , ordered in axis position . in contrast to the ( inhomogeneous ) sequences seen so far , @xmath320 has a well defined global center , around which the primary cluster of @xmath22 s is located , covering the central third of the total lattice . secondary @xmath22-clusters are created around the centers of the outer thirds , and then on the outer thirds of these , etc . therefore , there will be maximal palindromes of increasing length with axes approaching these hierarchical symmetry centers , at which their length jumps by a factor of three . ( color online ) local symmetry spacing ( a ) return maps and ( b ) trajectories for selected symmetry ranges in the cantor lattice . for @xmath324 ( @xmath258 ) , represented here by @xmath325 ( and distinguished by empty circles ) , the spacing trajectory starts with a boundary anomaly ( see text ) . ] the above features are seen in the ( partial ) local symmetry distribution of @xmath320 in fig . [ fig.cantor](b ) . as expected , maximal palindromes now exist for any length @xmath326 , as they are constructed explicitly by the inflation rule , so that @xmath327 . the clustering of @xmath22 s at increasing scale forms consecutive gaps along the @xmath28-axis , since palindromes of given length are maximal within the @xmath22-clusters only at their borders ( see upper inset in fig . [ fig.cantor](a ) ) . the gaps gradually become smaller with palindrome length and close at a larger @xmath27 , at which position a prefix palindrome of length @xmath328 occurs . complete local symmetries thus form naturally with every iteration of the @xmath323 inflation , with according lengths @xmath329 that is , with scaling factor @xmath238 . we here notice the characteristic similarity of the cantor local symmetry distribution with that of the @xmath247 in fig . [ fig.fib101-fib110-pal-cl](b ) . in both there are @xmath200-structures forming , with ( larger ) prefix palindromes occuring at the points where the legs would meet ; however , in the @xmath247 case the gaps do not extend to the bottom , because of the absence of pure @xmath22-clusters ( which would restrict smaller maximal palindromes to their borders ) . as can be anticipated from the symmetry distribution in fig . [ fig.cantor](b ) , also the spacings between maximal palindromes take on all possible values with changing @xmath49 . for fixed @xmath49 , we expect the spacings to feature increasing local maxima as the symmetry axis crosses the boundaries of larger @xmath22-clusters deeper in the lattice , until it reaches the globally central ( and largest ) cluster . in between , the spacing must return to smaller values , following the hierarchical structure of the lattice . in fig . [ fig.cantor-spacing](a ) , this behavior is shown in terms of the spacing return maps for selected local symmetry ranges . as we see , there are no longer few - point orbits , as in the aperiodic sequences studied so far . the number of different spacings @xmath132 for any single given range @xmath49 rather increases indefinitely with lattice size , i.e. , with word generation @xmath74 , and covers multiple orders of magnitude . there is also , in general , no fixed nestedness of consecutive maps through common or collinear points . in spite of their very different individual characteristics , however , all maps share a common asymptotic spacing scaling , given again by the cantor partitioning number : @xmath330 interestingly , there is here a boundary anomaly for every @xmath331 ( @xmath332 ) : at these ranges , which coincide with the cantor generation lengths and thereby with complete local symmetries , the very first axis distance is half the minimal distance for the rest of the @xmath134 sequence ( seen as singly connected empty circles in fig . [ fig.cantor-spacing](a ) ) . each map possesses the characteristic structure originating from the underlying self - similarity in @xmath320 . this hierarchical traversal of ever larger spacings is shown in fig . [ fig.cantor-spacing](b ) for @xmath185 and @xmath238 . for each @xmath49 , as the local symmetry axis is swept along the lattice , the spacing @xmath134 cycles recurrently on smaller return map orbits in hierarchical order , until the next ( larger ) @xmath22-cluster is reached , whereby a larger orbit is traversed . comparing the overall local symmetry properties of the cantor lattice with those of the previously examined binary sequences , we finally note the following . although the cantor lattice is generated by an inflation rule that explicitly produces palindromes ( on a minimal scale and , therefore , on all scales ) , its local symmetry dynamics is rather irregular , in the sense that its spacing return maps are unbounded , have an infinite number of points , and are quite different among symmetry ranges some even starting with the mentioned boundary anomalies . further , all possible ranges do occur , which is a characteristic shared statistically by a totally random lattice . in contrast , the aperiodic lattices of fibonacci , pd or tm type , which are generated by asymmetric inflation rules , are characterized by invariant , few - point spacing maps , with simple scaling laws along an ordered sequence of symmetry ranges . we have examined the local symmetry distribution ( positions of locally symmetric domains as a function of their range ) and dynamics ( spatial evolution of the distances between local symmetry axes for given range ) of one - dimensional aperiodic lattices generated by representative binary sequences , including ( generalized ) fibonacci , period - doubling , thue - morse and cantor . for each case , explicit scaling behaviors and correlation properties were found numerically for the sequences of occurring maximal local symmetry ranges , and for their spacings along the lattice , in relation to the original generating inflation rules or recursive schemes . in particular , the generalized fibonacci lattices were shown to incorporate multiplet symmetry range subsequences with different structural features determined by the asymptotic length ratio of the corresponding binary word generations , with a dominant subsequence containing all complete local symmetries ( i.e. , palindromic word prefixes ) . moreover , for the fibonacci , pd and tm lattices , the number of different local symmetry spacings is restricted to @xmath212 or @xmath238 for any symmetry range . in terms of the subsequences of local symmetry spacing return maps for given ranges , the ( asymptotic ) copper fibonacci lattice was shown to be structurally equivalent to the period - doubling lattice , of which in turn the thue - morse case constitutes a reduced version ( with two sub - subsequences discarded ) . remarkably , the aperiodic order of each type of lattice was found to be encoded exactly in its local symmetry dynamics : for each occurring local symmetry range , the renormalized sequence of axis spacings is generated by the inflation rule of the original aperiodic sequence . the local symmetry distribution of the above lattices was , finally , contrasted with the cantor lattice , whose inflation rule is inherently palindromic with gapped maximal symmetry distribution and unbounded symmetry spacing return maps . in conclusion , we have shown that a unified view of different classes of aperiodic lattices is provided by the distribution , and , in particular , by the fixed range spacing dynamics , of their ( maximal ) local symmetries . they yield a compellingly simple analysis tool to distinguish structural similarities but also lower level differences between given lattices . the encoding of the corresponding inflation rule within any given symmetry spacing trajectory further demonstrates that local symmetries completely characterize deterministic aperiodic order at any scale . since we perform a numerical investigation of the considered aperiodic sequences , throughout the paper the notion of asymptotic convergence is not used in a mathematically strict sense , but refers to the observed behavior for large number of corresponding iterations . the first and second range subsequences would be @xmath333 and @xmath334 , respectively . thus , in order to consider the very _ first _ @xmath196-tet bundle @xmath335 , one has to include , apart from the single - letter maximal palindromes with length @xmath173 , also the _ empty _ maximal palindromes @xmath7 of length @xmath336 . the latter are the interfaces between two different letters @xmath21 and @xmath22 in a word . since @xmath337 and @xmath115 are not plotted throughout our analysis , we describe the @xmath196-tets starting from the second bundle @xmath338 .

the dynamics in quantum magnets can often be described by effective models with bosonic excitations obeying a hard - core constraint . such models can be systematically derived by renormalization schemes such as continuous unitary transformations or by variational approaches . even in the absence of further interactions the hard - core constraint makes the dynamics of the hard - core bosons nontrivial . here we develop a systematic diagrammatic approach to the spectral properties of hard - core bosons at finite temperature . starting from an expansion in the density of thermally excited bosons in a system with an energy gap , our approach leads to a summation of ladder diagrams . conceptually , the approach is not restricted to one dimension , but the one - dimensional case offers the opportunity to gauge the method by comparison to exact results obtained via a mapping to jordan - wigner fermions . in particular , we present results for the thermal broadening of single - particle spectral functions at finite temperature . the line - shape is found to be asymmetric at elevated temperatures and the band - width of the dispersion narrows with increasing temperature . additionally , the total number of thermally excited bosons is calculated and compared to various approximations and analytic results . thereby , a flexible approach is introduced which can also be applied to more sophisticated and higher dimensional models .

1402.4359

dynamic correlations generally provide valuable information about the systems under study . in linear response , for instance , all conceivable susceptibilities are dynamic correlations . in many spectroscopic experiments dynamic correlations are measured , for instance dynamic structure factors in scattering experiments or current - current correlations in reflectivity or absorption measurements . at low temperatures , such experimental data provide valuable information about elementary excitations , their mutual interaction , and many matrix elements . theoretically , it is well established that even complex quantum systems can be described at low energies by simpler effective models . such models can result for instance from various renormalization procedures @xcite . focusing on gapped systems , the elementary excitations can be viewed generically as quasi - particles the number of which is conserved which move and interact and determine the physical properties of the system under study . the corresponding effective hamiltonian in terms of these quasi - particles can be derived systematically by unitary transformations , see for instance ref . , or by variational approaches , see for instance ref . . such approaches yield models in terms of the elementary excitations so that the dynamic response at zero temperature is captured and an enormous wealth of information is available . if such models conserve the number of quasi - particles , i.e. , the number of excitations , they are no longer plagued by quantum fluctuations : their ground state is the vacuum of the quasi - particles , cf . ref . . thus the only remaining fluctuations to worry about are thermal ones at finite temperature . we have two classes for physical realizations of such systems in mind . quantum antiferromagnets with a finite spin gap represent one wide class . in particular various antiferromagnets made from coupled spin dimers belong to this class , for instance , dimerized spin chains @xcite , spin ladders @xcite , two - dimensionally coupled spin ladders @xcite , and three - dimensionally coupled spin dimers @xcite . the elementary excitations are triplons , i.e. , @xmath0 hard - core bosons @xcite . the hard - core repulsion comes about because the presence of one excitation at a given site ( dimer ) excludes the presence of a second excitation at this site . another example of hard - core bosonic excitations are spin flips in the high - field phase of the transverse field ising model . they represent hard - core excitations of a single flavor @xcite . the other class are designed systems made from ultracold bosonic atoms trapped in suitable optical traps and lattices , for a review see ref . . the setup of ultracold atoms has the advantage that their number is conserved by construction . a disadvantage with respect to the considerations presented here is that there is no scattering process which creates or annihilates an atom as does inelastic neutron scattering with magnetic triplons or magnons . the models in terms of conserved hard - core bosons , i.e. , their number is a conserved quantity , are very successfully analyzed to extract information at zero temperature , i.e. , in the immediate vicinity of the ground state . but the models are also valid and useful at finite temperatures unless the temperature is so high that the system enters another phase . the overall goal of the present article is to benchmark a diagrammatic approach which allows one to include the thermal fluctuations in a strongly interacting model of hard - core bosons with the vacuum being the ground state , i.e. , no bosonic condensate is considered . the benchmarking is done for a simple gapped chain model in which only one kind of hard - core bosons hop from site to site . this model can be treated exactly by a jordan - wigner mapping to free fermions so that a reliable testbed is available . however , the simple model shares all the basic elements of more complicated hard - core boson models : mobile bosons with an infinite local repulsion whose number is conserved so that no quantum fluctuations occur . we stress that the one dimensional case serves only as a testbed to gauge our method while the approach is applicable also in higher dimensions and is not restricted to the one dimensional case . since only a limited toolset to calculate the finite temperature dynamics of quantum systems in higher dimensions is available , the diagrammatic approach is a promising new technique for a wide range of problems . by the careful benchmark , we validate an approach which can be applied to all kinds of particle - conserving gapped models . such effective models are no longer subject to quantum fluctuations . they can be derived systematically by various techniques , in particular by continuous unitary transformations @xcite so that a large class of problems can be tackled in this way . wide - spread examples are the excitation line - shapes in gapped quantum antiferromagnets investigated by neutron scattering in one dimension @xcite as well as in three dimensions @xcite . so far the theoretical treatments of these experiments have focused on dispersion relations @xmath1 and their temperature dependent shifts , neglecting issues of line shape and width . we highlight the one - dimensional material @xmath2 $ ] which is strongly dimerized and exhibits a large gap - to - bandwith ratio so that a particle - conserving model free of quantum fluctuations is easily derivable @xcite . the diagrammatic approach to thermal fluctuations is systematically controlled by the expansion parameter @xmath3 where @xmath4 is the inverse temperature and @xmath5 the energy gap . we present the direct calculation of the leading order in @xmath3 of the self energy and the corresponding self - consistent calculation with dressed propagators . this approach is tested for a model for which numerically exact results are available for the propagators . we show by comparison of numerical data that both approaches agree in linear order in @xmath3 . in this comparison we focus on the position of the single - boson peak in the spectral response , its broadening and the symmetry of its shape . we show that while the single - boson peak broadens , the overall bandwidth of the dispersion tends to narrow upon increasing temperature . this behavior has been observed experimentally @xcite and it is at the focus of ongoing theoretical research @xcite . this underlines its importance . in view of the constantly improved instrumentation of inelastic neutron scattering setups a growing interest in reliable theoretical techniques for the thermal effects on line shapes is to be expected . we also evaluate the average occupation due to thermal excitations and find an excellent agreement between the self - consistent diagrammatic result and the exact one . in view of these results , the self - consistent diagrammatic approach suggests itself for application to more complicated , extended models for gapped quantum antiferromagnets _ after _ their re - formulation in terms of conserved quasi - particles to take the quantum fluctuations into account . extensions may comprise longer - range hoppings , further interactions among the hard - core excitations , and excitations of several flavors . also , the application in dimensions greater than one appears promising because one may generally expect that a diagrammatic perturbative approach works even better in higher dimensions . the present article is set up as follows : in the next section , the considered model is introduced . second , in sect . iii , we present the diagrammatic approach in detail ; a particular focus lies on the pure cosine band as it results from nearest - neighbor hopping . in sect . iv we explain how the equivalent fermionic model allows the numerically exact calculation of the propagators of the hard - core bosons . in sect . v , we gauge the diagrammatic approach in quantitative comparison with the exact results and discuss its properties in detail . finally , sect . vi concludes the article . in this section we introduce the model and the general quantities , such as the spectral functions , in which we are interested . general observations on the temperature dependence of the spectral weight of hard - core bosons are also presented . to introduce the diagrammatic expansion for hard - core bosons , we choose a simple 1d model , a single chain of lattice sites . every site can either be occupied or empty , so that the local hilbert space is two dimensional . in real space the hamiltonian reads @xmath6 where @xmath7 and @xmath8 are operators at site @xmath9 , which fulfill the hard - core boson relation , @xmath10 & = \delta_{i , j } \left ( 1 - 2 b_i^\dagger b_i^{\phantom\dagger } \right).\end{aligned}\ ] ] the hamiltonian consists of a local energy term and a nearest neighbour hopping . the energy gap is given by @xmath11 , while @xmath12 is the band - width of the dispersion @xmath13.\end{aligned}\ ] ] where @xmath14 is the momentum of the excitations . the minimum of the dispersion with value @xmath5 is found at @xmath15 . the ground state of the system is given by the vacuum state . in typical physical systems , such as spin ladders or spin chains , additional interactions , non - particle - conserving terms as well as longer range hopping are present . we do not consider these terms in our simple model , but we stress that there are a variety of methods to treat such terms which can be combined with our diagrammatic approach . for example , models with non - particle - conserving terms can be mapped to effective , particle - conserving models by continuous unitary transformations ( cut ) @xcite , while additional interactions , besides the hard - core constraint , can be dealt with on a mean - field level . longer range hopping processes trivially modify the dispersion . to test our diagrammatic results , we make use of the fact that the hamiltonian can be mapped to a system of free fermions by the jordan - wigner transformation @xcite [ eq.jordan_wigner ] @xmath16 the operators @xmath17 and @xmath18 fulfill fermionic anti - commutator relations . after a fourier transformation to reciprocal space , the hamiltonian reads @xmath19 in contrast to free bosons , the spectral properties of hard - core bosons at finite temperature are not exactly known . even though the above fermionic approach can be applied , the calculation of dynamic susceptibilities remains a difficult task because of the non - locality of the jordan - wigner transformation . one example is the single - particle temperature green function @xmath20 where @xmath21 is the imaginary time and @xmath22 is the time - ordering super operator . the corresponding spectral function reads @xmath23 where we introduced the matsubara frequencies @xmath24 , the inverse temperature @xmath25 , and the total number of sites @xmath26 . the spectral function is connected to the dynamic structure factor ( dsf ) by the fluctuation - dissipation theorem @xmath27.\end{aligned}\ ] ] the dsf is the relevant quantity for many experiments . for instance , it is accessible in inelastic neutron scattering experiments . note that there are no contributions from anomalous green functions , @xmath28 since the hamiltonian is particle conserving . another important difference between hard - core bosons and normal bosons concerns sum rules for spectral functions . in general the weight of the spectral function is determined by @xmath29 \right\rangle.\end{aligned}\ ] ] for normal bosons this is a constant . in contrast , for hard - core bosons , transforming eq . into fourier space yields @xmath30 & = \delta_{p , p ' } - \frac{2}{n } \sum\limits_{q_3 } b_{q_3-p+p'}^\dagger b_{q_3 } .\end{aligned}\ ] ] as a result the sum rule reads @xmath31 where @xmath32 is the thermal occupation . at zero temperature , @xmath33 holds and the sum rule yields unity . the opposite case at infinite temperatue is @xmath34 ; a site is occupied with probability @xmath35 and the spectral weight over all positive and negative frequencies adds up to zero . in this section we introduce the diagrammatic approach to treat the hard - core repulsion at finite temperatures . the key idea is an expansion in the small parameter @xmath3 . in subsection [ ssect : gen_calcul ] the single - particle self energy is calculated for arbitrary dispersion relation . in the following subsection [ ssect : cos - band ] we specialize our results to the pure cosine band and show for this special case , that the shift of the dispersion , due to the finite density of thermal fluctuations , is a second order effect in @xmath36 . our diagrammatic approach is based on the idea to replace the hard - core boson operators by pure bosonic operators and to enforce the hard - core constraint by an infinite on - site interaction , @xmath37 this idea has long been known under the name of brckner theory for nuclear matter and he@xmath38 @xcite . for low dimensional solid state systems such as spin ladders it was first proposed in ref . in contrast to that investigation , we are not dealing with quantum fluctuations but with thermal fluctuations , i.e. , the number of bosons is a constant of motion in our case . in the original approach anomalous green functions are present and it is difficult to determine which diagrams contribute in leading order in the density of quantum fluctuations . in our application , no quantum fluctuations are present . the initial spin hamiltonian may contain non - particle - conserving terms , but we assume that they have been eliminated by some renormalizing procedure , for instance by unitary transformations . consequently , we assume that at zero temperature the propagation of the hard - core bosons is described exactly . transforming the hamiltonian @xmath39 into reciprocal space yields @xmath40 note that in higher dimensions the momenta @xmath41 are vectors . this does not change the subsequent theoretical calculations . in practice , the equations can then be solved using higher dimensional fast fourier techniques . we expand the single - particle propagator @xmath42 in terms of feynman diagrams . since the interaction is infinite , a truncated perturbative approach in the interaction strength can not succeed . instead , we use the density of thermal excitations as small expansion parameter , i.e. , we expand in @xmath36 . the leading order is given by those diagrams that have the fewest number of propagators going backwards in imaginary time . thus , in linear order in @xmath36 , we have to sum all diagrams with a single loop which amounts to the summation of the ladder diagrams , see fig . [ fig.ladder1 ] , following the arguments in refs . we treat the system as a dilute bose gas . note that the ladder approximation presented here differs from those in standard textbooks @xcite , where the leading _ quantum fluctuations _ are captured which result from the dominant interaction of the quasi - particles with the condensate . in contrast to this , our approach captures the leading effects on the spectral function due to the presence of a small density of thermally excited bosons . as a result our approximation breaks down once the density of thermally excited bosons is not small anymore . so the chosen approach represents a low - temperature approximation . ] the elementary building block in the ladder diagrams is the scattering amplitude @xmath43 , defined in fig . [ fig.ladder2 ] , which can be interpreted as a generalized effective interaction . it is easily seen that the scattering amplitude fulfills a dyson - like equation , the bethe - salpeter equation , shown in fig . [ fig.bethe_salpeter ] . to simplify the following expressions we introduce the 2-momentum @xmath44 and the corresponding summation @xmath45 the bethe - salpeter equation in fig . [ fig.bethe_salpeter ] reads in formulae @xmath46 note that the scattering amplitude only depends on the total momentum @xmath47 and not on any relative momenta . this is due to the simple structure of @xmath48 in fourier space . in the following , we assume that each propagator in the diagrams in figs . [ fig.ladder1 ] , [ fig.ladder2 ] and [ fig.bethe_salpeter ] is the bare boson propagator @xmath49 below , however , we will relax this assumption in the self - consistent evaluation . equation can be solved for @xmath50 yielding @xmath51 where @xmath52 since @xmath53 consists of convolutions of two green functions , it is of order ( at least ) @xmath54 so that a hilbert representation exists ; it reads @xmath55 which can be obtained from the imaginary part [ eq.rho_p ] @xmath56 } \nonumber \\ & \cdot \left ( \frac{1}{e^{- \beta \omega(-l ) } - 1 } - \frac{1}{e^{\beta \omega(p+l ) } - 1 } \right ) \label{eq.rho_p1 } \\ & = \frac{-1}{n } \sum\limits_{l } \delta(x-[\omega(-l ) + \omega(p+l ) ] ) \nonumber \\ \label{eq.rho_p2 } & \cdot \left ( \frac{1}{e^{\beta \omega(p+l ) } - 1 } - \frac{1}{e^{- \beta \omega(-l ) } - 1 } \right ) .\end{aligned}\ ] ] from the last expression one sees that the spectral function @xmath57 is negative so that @xmath58 is negative as well for large , positive frequencies . the spectral density @xmath57 can be calculated either analytically or numerically for a given dispersion @xmath1 . for computational details we refer the reader to appendix [ app.b ] . note that we already computed the sum over all matsubara frequencies @xmath59 appearing in eq . leading to the bose functions in eq . . the function @xmath57 consists of a two - particle continuum describing the spectral properties of two - particle scattering states . for @xmath60 the thermal factor can be expanded according to @xmath61 it is not possible to take the limit @xmath62 already in eq . , because there exists no spectral representation for @xmath63 . this is one of the main differences to ref . @xcite , where the brckner theory was applied to the double - layer heisenberg antiferromagnet at finite temperatures . in that model , quantum fluctuations are present , but the imaginary part of the self energy was neglected and therefore broadening was omitted . then , the spectral function @xmath64 remains a sharp @xmath65-function even at finite temperature , but its position in frequency depends on temperature . we know that @xmath66 for @xmath67 . thus @xmath68 holds . consequently , there also exists a spectral representation for this quantity @xmath69 which again is determined by the imaginary part of the left hand side . we calculate for finite @xmath70 and @xmath71 @xmath72 \\ & & \quad = \frac { - u^2 \rho_p(x ) } { \left[1 + u \mathcal{p } \int\limits_{-\infty}^{\infty } \frac{\rho_p(y)}{x - y } \mathrm{d}y \right]^2 + \left [ u \rho_p(x ) \pi \right]^2 } .\end{aligned}\ ] ] at this stage , one can take the limit @xmath62 and define the function @xmath73 ^ 2 + \left [ \rho_p(x ) \pi \right]^2 } \end{aligned}\ ] ] where @xmath74 stands for the principal value of the integral . again this expression can be evaluated analytically or numerically , depending on the dispersion . in addition , for large @xmath70 , the denominator in can vanish completely , because the real part of @xmath75 becomes negative in this region . the vanishing denominator yields an additional @xmath65-function in @xmath76 . thus @xmath77 is not the only contribution to @xmath76 but there also is the signature of an anti - bound state at very high energies in the range @xmath78 . the appearance of the anti - bound state in a lattice model with dominant repulsion @xmath71 between the elementary excitations is actually to be expected . the propagator of two particles acquires an additional pole at @xmath71 . in the context of our approach , however , it is a mathematical artifact because in the final limit @xmath79 the anti - bound state does not occur directly in any measurable quantity . this would be different in systems where @xmath71 is very large , but not infinite , for instance for ultracold atoms in optical lattices . we stress that in spite of the limit @xmath79 , which makes the anti - bound state vanish at infinity , it leaves traces at finite energies . to our knowledge , this has not been discussed in detail before . we will derive these effects in the following . to obtain the position and weight of the anti - bound state we expand the denominator in eq . for high frequencies @xmath80 leading to @xmath81 where @xmath82 and @xmath83 are the weight and the first moment of @xmath57 , respectively , @xmath84 based on eq . one realizes that @xmath82 actually does not depend on @xmath47 . therefore the @xmath47 dependence can be ignored in the following calculations . extracting the imaginary part from eq . leads to @xmath85 \nonumber \\ = \frac{-u}{\pi } \lim\limits_{\omega \rightarrow x + i\delta } \mathrm{im } \left [ \frac{\omega^2}{(\omega - \omega_1)(\omega - \omega_2 ) } \right],\end{aligned}\ ] ] where @xmath86 we point out that @xmath87 is the energy of the anti - bound state while @xmath88 is a spurious root due to the expansion in @xmath89 . now we use dirac s identity @xmath90 to obtain @xmath91 \nonumber \\ = u \left ( \frac{x^2}{x-\omega_2 } \delta(x-\omega_1 ) + \frac{x^2}{x-\omega_1 } \delta(x-\omega_2 ) \right ) . \label{eq.anti_bond_contribution}\end{aligned}\ ] ] in the following , we can drop the contribution of the spurious root . combination with eq . yields @xmath92 next we address the single - particle self energy . to generate all diagrams in fig . [ fig.ladder1 ] the scattering amplitude must be closed in two ways as shown in fig . [ fig.self_energy ] . if the hard - core bosons can have several flavors , the first diagram has to be counted several times , once for each flavor , because the flavor of the upper boson loop is independent of the one in the lower propagator . this is not the case for the second diagram because it does not have an independent boson loop . thus it is counted only once even if several flavors are possible . due to the simple structure of the interaction , both diagrams yield the same contribution to the self energy @xmath93 inserting the spectral representation yields @xmath94 next we sum over all matsubara frequencies @xmath95 and take the limit @xmath62 . splitting the self energy into real and imaginary parts leads to @xmath96 \frac{1}{e^{\beta \omega(k ) } - 1 } \nonumber \\ \label{eq.sigma_real } & + \mathcal{p } \int\limits_{-\infty}^{\infty } \frac{\rho_{\sigma , p}(x)}{\omega - x } \mathrm{d}x \\ \label{eq.sigma_imag } \mathrm{im } \sigma(p , \omega ) & = - \pi \rho_{\sigma , p}(\omega),\end{aligned}\ ] ] where @xmath97 is the spectral function for the self energy @xmath98 .\end{aligned}\ ] ] we draw the reader s attention to the additional contributions in the real part of the self energy _ besides _ the principal value integral over the spectral function @xmath97 . these additional terms stem from the contribution of the anti - bound state in eq . . they are subtle in nature because they represent the left - overs at finite frequencies @xmath99 of the anti - bound state in which itself tends to infinity upon @xmath79 . in this way , a frequency dependence remains in the real part of the self energy which can not be traced back to a spectral density at finite frequencies . next we can calculate our primary quantity of interest , the spectral function of the propagator [ eq.a_final ] @xmath100 the real part of the self energy describes the shift of the peak position due to the interaction with the thermally populated background . we will see that a narrowing of the dispersion will ensue . the imaginary part describes the broadening of lines of the single excitations due to the hard - core interactions . if the imaginary part of the self energy has only a negligible dependence on @xmath70 , the spectral function @xmath101 is a symmetric lorentzian with full width @xmath102 at half maximum . below , we will see that this approximation does not hold at elevated temperatures . one way to improve the results obtained from the ladder diagrams is to calculate the spectral function self - consistently . thereby , we realize a conserving approximation in the sense of baym and kadanoff @xcite . in order to do so , each bare propagator must be replaced by the fully dressed propagator @xmath103 which modifies the resulting equations slightly . on the diagrammatic level , this means that also a number of higher order diagrams are taken into account which consist of propagators with self energy insertions . we refer the reader to appendix [ app.a ] for more details on the explicit self - consistent calculation . the leading , linear contribution in @xmath3 in the above general calculations can be evaluated analytically for the cosine band corresponding to nearest - neighbor hopping . this allows us to evaluate the real part of the self energy in this order . for @xmath57 we obtain @xmath104 ^ 2 - \left[2\delta + w - x\right]^2 \right]^{-1/2 } \\ & & + \mathcal o\left(e^{-\beta \delta}\right ) .\end{aligned}\ ] ] the corresponding real part vanishes if @xmath70 lies within the band . using the expansion of @xmath57 to calculate the expansion of @xmath77 leads to @xmath105 for @xmath106 we deduce @xmath107 where @xmath108 is used for @xmath109 . we aim at the real part of the self energy describing the shift of the spectral function @xmath101 . the expansion in @xmath110 yields @xmath111 e^{-\beta \omega(k ) } \nonumber \\ & + \mathcal o\left(e^{-2\beta \delta}\right ) \label{eq.real_part_analytics } , \end{aligned}\ ] ] where we made use of @xmath112 next we consider @xmath113 and expand it in @xmath36 [ eq.rho_1_expansion ] @xmath114 where @xmath115 is the average value of the dispersion in the brillouin zone . for the cosine band this simply is @xmath116 finally , we compute the the principal value integral in for the cosine band to reach @xmath117 + \mathcal o\left(e^{-\beta \delta}\right ) \label{eq.principal_value_expansion } , \end{aligned}\ ] ] as long as @xmath118| < |w \cos(p/2+k/2)|$ ] holds . this is always true for @xmath119 . inserting eqs . , , and into the real part of the self energy yields @xmath120 since our diagrammatic expansion is correct in first order in @xmath36 we conclude that the first order corrections in @xmath36 to the real part of the self energy vanish rigorously . we will confirm this conclusion by evaluating the equivalent fermionic model in the next sections . hence the shift of the dispersion due to finite temperature for cosine bands is a second order effect @xmath121 . interestingly , this result in one dimension also holds in the case of multi - flavor hard - core bosons such as triplons because the multiplicity of the flavors only affects the prefactor of the self energy . in this section an equivalent fermionic model is introduced by means of the jordan - wigner transformation . this mapping allows us to calculate the finite temperature dynamics numerically based on the evaluation of pfaffians . this approach is limited to a certain class of one dimensional systems but allows us to gauge the diagrammatic approach . the hamiltonian can be interpreted as an anisotropic spin model , the xx chain . note that @xmath122 is identical to the hard - core boson creation operator @xmath123 , see eq . . the open - ended @xmath26-site @xmath124 xx chain in a homogeneous magnetic field @xmath125 is defined by @xmath126 @xmath127 is one of the simplest quantum many - body systems because many of its properties can be derived from those of non - interacting lattice fermions . nevertheless it shows non - trivial dynamics . @xmath127 can be mapped @xcite to a hamiltonian of noninteracting fermions , @xmath128 by means of the jordan - wigner transformation . the diagonal form of the fermion hamiltonian is @xmath129 where the operators @xmath130 and @xmath131 create and destroy a fermion in a one - particle eigenstate , respectively . the one - particle energy eigenvalues are @xmath132 and the eigenvectors are sinusoidal functions of the site index @xmath9 . obviously the dispersion relations ( [ eq:2.6 ] ) and are identical for @xmath133 and @xmath134 . in the fermionic ground state all single - particle states with negative energies are occupied while all other states are empty . for @xmath135 the ground state is either completely occupied or completely empty . in the intermediate field range , @xmath136 , the ground state contains a partially filled band of jordan - wigner fermions . we are interested in the correlation functions @xmath137 which have a rather complicated structure in the fermionic representation . with the fermionic identity @xmath138 applied to the jordan - wigner transformation , this correlation function may be expressed in terms of the auxiliary operators @xmath139 and @xmath140 as follows : @xmath141 this expectation value of a product of @xmath142 fermion operators may be expanded in terms of two - point expectation values using wick s theorem @xcite . the result is most compactly expressed as a pfaffian : @xmath143 @xmath144 the square of the pfaffian is equal to the determinant of the antisymmetric matrix with the elements of ( [ ii.8 ] ) above the diagonal . other properties of pfaffians can be found in the literature @xcite . the numerical evaluation of pfaffians proceeds along similar lines as that of determinants . many matrix elements can be reduced to zero by operations which are known to leave the value of the pfaffian invariant . after production of sufficiently many zero elements the evaluation of the pfaffian becomes trivial due to an expansion theorem . an implementation along these lines was described by derzhko and krokhmalskii @xcite . we use a similar algorithm here . an alternative recursive scheme for evaluating pfaffians was used by jia and chakravarty @xcite . the matrix elements of ( [ ii.8 ] ) can be evaluated from the expressions @xcite @xmath145 \label{ii.9 } \\ \langle a_j(t ) b_l\rangle & = & \frac{2}{n+1 } \sum_k \sin kj \sin kl [ i \sin \omega(k ) t - \cos \omega(k ) t \tanh \frac{\beta \omega(k)}{2 } ] \ , . \label{ii.10}\end{aligned}\ ] ] along with the relations @xmath146 all elements of type ( [ ii.9 ] ) with odd @xmath147 and all elements of type ( [ ii.10 ] ) with even @xmath147 vanish . in fact , for @xmath148 the elements ( [ ii.9 ] ) are zero for all @xmath149 . a word on boundary conditions is in order at this point . the mapping between the spin and fermion hamiltonians as given above is only possible for open boundary conditions . for cyclic boundary conditions the jordan - wigner transformation generates different hamiltonians for even and odd total fermion numbers , respectively this makes the calculation of the dynamic correlations @xmath150 extremely awkward , if not impossible , since every @xmath151 operator switches back and forth between subspaces of even and odd fermion numbers . we therefore stick to open boundary conditions . in order to make sure that open - chain numerical results pertain to the thermodynamic limit , only spins sufficiently far from the boundaries of sufficiently long chains may be considered . then the finite group velocity of the jordan - wigner fermions prevents the occurrence of `` echoes '' reflected from the chain boundaries in the dynamic correlations for short enough times . the quantity of interest in this study is the dsf @xmath152 , since it is directly related to the spectral function @xmath153 by the fluctuation - dissipation theorem . @xmath152 is the fourier transform with respect to space and time of the dynamic correlation @xmath150 @xmath154 \langle s_l^x ( t)s_{l+n}^x \rangle.\ ] ] @xmath152 is determined using fast fourier transform algorithms . a technical problem occurs at low temperature , where @xmath137 displays slow power - law asymptotics at long times leading to spurious oscillations in @xmath152 if the fourier transform is performed using a finite time interval . however , for the purposes of the present study only the position and width of the dominant peak ( or rather ridge ) in @xmath152 is relevant so that no filtering or asymptotic continuation techniques had to be applied . all results discussed here are for the gapped case , @xmath135 ( or @xmath11 ) . dsf data for the case @xmath155 were published some years ago @xcite as were results for dimerized xx chains @xcite . in this section we discuss the results of the diagrammatic approach in detail . first we compare the line - shapes calculated by the diagrammatic approach with the numerically exact results obtained from the equivalent fermionic model in subsection [ ssec : comparison ] . here we also verify our analytical finding concerning the shift of the peak position from subsection [ ssect : cos - band ] . next we extract the thermal occupation function from the single - particle green function and compare it to various approximations and to the exact result . in subsection [ ssec : details_diagrammatic ] , we study the real and imaginary parts of the self energy obtained by the diagrammatic approach and discuss their general features . in the final two subsections [ ssec : broadening ] and [ ssec : band_narrowing ] we show the finite temperature broadening over a wide range of temperatures for the two modes defining the minimum and the maximum of the single - boson dispersion , respectively . we also discuss how the single - particle band narrows in energy at finite temperature . and @xmath156 with @xmath157 and @xmath158 , i.e. , @xmath159 . [ fig.joachim_compare1_w05 ] and @xmath156 with @xmath160 and @xmath158 , i.e. , @xmath159 . [ fig.joachim_compare1_w4 ] here , we compare the results of the diagrammatic expansion to the results of the exact fermionic evaluation . the latter are exact except for finite size or finite time effects . thus the fermionic results will serve as testbed for the diagrammatic approach . the quantity of interest is the single particle spectral function @xmath64 . it is connected to the dsf by eq . . it turns out that the spectral function has almost no weight for @xmath161 in the parameter regime which we are focusing on . therefore , we can approximate the relation between @xmath64 and @xmath162 by @xmath163 all diagrammatic results presented are calculated self - consistently , if not denoted otherwise . in fig . [ fig.joachim_compare1_w05 ] we compare the spectral function at finite temperature @xmath158 , i.e. , @xmath164 obtained by the fermionic approach and by diagrammatic expansion for a narrow band - width @xmath157 . the results agree very well for momentum @xmath156 , while a slight difference appears for @xmath165 . to understand the main effects qualitatively , the following argument helps . the weight of the peak changes only very little as function of temperature , see for instance figs . [ fig.joachim_compare4_w05 ] and [ fig.joachim_compare4_w4 ] below . thus the height of the peak is inversely proportional to its width . the width is proportional to the imaginary part of the self energy , see eq . , which in turn is exponentially small , namely proportional to @xmath3 . thus , the height is proportional to @xmath166 . this very strong dependence on the temperature ( @xmath167 ) puts the discrepancy between the height of the two curves for @xmath165 into perspective . if @xmath166 is large , let us say @xmath168 , a slight inaccuracy of @xmath169 in the gap , which is modified by the real part of the self energy , induces a @xmath170 error in the height . for @xmath171 the relative error would even rise to @xmath172 . the shift of the peak position at finite temperature seems to be overestimated by the diagrammatic expansion . since the shift is a second order effect @xmath173 and the diagrammatic expansion is only correct in first order @xmath36 such deviations can be expected . we attribute this difference to two sources : ( i ) the ladder approximation is not able to capture all relevant physical processes at this fairly elevated temperature . ( ii ) the data obtained in the fermionic approach has a finite resolution in the time domain which implies some inaccuracies in the frequency domain . we do not include peaks at lower temperatures because they become very quickly extremely sharp so that it is difficult to evaluate their shape with appropriate numerical precision . this holds true for the numerical evaluation of both the exact and of the diagrammatic approach . we emphasize , however , that by construction the diagrammatic approach becomes better and better for lower and lower temperature . figure [ fig.joachim_compare1_w4 ] shows the spectral functions for the wide band case @xmath160 at finite temperature @xmath158 , i.e. , @xmath164 . due to the larger band - width , the group velocity of the excitations is increased and the finite system size in the fermionic evaluation induces additional errors . this is clearly visible in the additional wiggling in the spectral function obtained in the fermionic approach . next we want to study the shift of the peak position in more detail . in figs . [ fig.joachim_compare2_w05 ] and [ fig.joachim_compare2_w4 ] the shift for the narrow and for the wide band case is depicted as function of the inverse temperature . both methods indicate that the shift is a second order effect in @xmath36 , verifying our analytical argument , see sec . [ ssect : cos - band ] . the diagrammatic expansion overestimates the shift as we have already seen in figs . [ fig.joachim_compare1_w05 ] and [ fig.joachim_compare1_w4 ] . note that for @xmath160 and @xmath156 four data points obtained by the fermionic approach show the same shift . this error is caused by the finite frequency resolution used in the fermionic approach . we stress again that the exact fermionic expressions are difficult to evaluate numerically , in particular for low temperatures . as function of the inverse temperature . crosses represent results obtained in the fermionic approach while boxes and circles represent results obtained by the diagrammatic expansion . the function @xmath174 has been fitted to the data using the parameter @xmath175 . ] as function of the inverse temperature . crosses represent results obtained in the fermionic approach while boxes and circles represent results obtained by the diagrammatic expansion . the function @xmath174 has been fitted to the data using the parameter @xmath175 . ] the width of the spectral functions is also investigated . it is measured by the full width at half maximum . the data is depicted in figs . [ fig.joachim_compare3_w05 ] and [ fig.joachim_compare3_w4 ] . for low temperatures , the data of both approaches agrees well . both data sets support our analytic finding in eq . that the width is a first order effect in @xmath36 . upon increasing temperature the diagrammatic expansion underestimates the broadening of the line - shape . this can be attributed to the missing diagrams @xmath121 not included in our approach . these diagrams describe additional scattering processes increasing the decoherence and broadening the line - shape further . for @xmath157 and very high temperatures we even see that the broadening obtained from the ladder approximation decreases , which clearly indicates that the ladder diagrams no longer capture the dominant scattering processes . as function of the inverse temperature . crosses represent results obtained in the fermionic approach while boxes and circles represent results obtained by the diagrammatic expansion . the function @xmath176 has been fitted to the data using the parameter @xmath175 . ] as function of the inverse temperature . crosses represent results obtained in the fermionic approach while boxes and circles represent results obtained by the diagrammatic expansion . the function @xmath176 has been fitted to the data using the parameter @xmath175 . ] finally , we consider the temperature dependence of the weight of the spectral function . more specifically , we plot the deviation from unity in logarithmic scale @xmath177 for the narrow band in fig . [ fig.joachim_compare4_w05 ] and for the wide band in fig . [ fig.joachim_compare4_w4 ] versus the inverse temperature . for both cases an amazing agreement between the two methods is found . since we have already shown in sect . [ sec.model ] that the weight of the spectral function is directly connected to the thermal occupation , see eq . , we expect that the thermal occupation is captured very accurately by the diagrammatic expansion in agreement with the exact fermionic results , see also next subsection . as function of the inverse temperature . crosses represent results obtained in the fermionic approach while boxes and circles represent results obtained by the diagrammatic expansion . the function @xmath176 has been fitted to the data using the parameter @xmath175 . ] as function of the inverse temperature . crosses represent results obtained in the fermionic approach while boxes and circles represent results obtained by the diagrammatic expansion . the function @xmath176 has been fitted to the data using the parameter @xmath175 . ] here we focus on the thermal occupation which can be determined from the spectral function by evaluating the integral @xmath178 since the jordan - wigner transformation maps the hard - core bosons to fermions without interaction , the exact expression for the thermal occupation is easily available @xmath179 it is not possible to calculate the momentum dependent thermal occupation analytically , because it includes many - particle correlation functions in the fermionic picture , see sec . [ sec : fermionization ] . a crude estimate for the occupation function of hard - core bosons was proposed in ref . @xcite , @xmath180 where @xmath181 . it is correct for flat bands @xmath182 , but often used as a first approximation also in the case of dispersive bands , see for instance refs . . the implied approximate thermal occupation reads @xmath183 we call the thermal occupation approximate hard - core statistics below because it only captures the local aspects of repelling bosons . the approximate statistics captures the correct values for @xmath184 and @xmath185 for nonzero band - width . we therefore expect deviations from the exact expression to appear for moderate temperatures @xmath186 and wide bands @xmath187 . the thermal occupation for the narrow band @xmath157 is depicted in fig . [ fig.thermal_occupation ] . for low @xmath22 the precise statistics does not matter , so that even the free boson statistics @xmath188 captures the correct behaviour . this changes distinctively at higher temperatures . the occupation number for free bosons has no bound for rising temperature , while for hard - core bosons the limit @xmath34 has to be fulfilled . therefore the free boson approximation breaks down at @xmath189 . the non - self - consistent calculation improves the statistics beyond this point , but also breaks down once temperature reaches @xmath190 . fortunately , the self - consistent calculation stays very close to the exact result , even for temperatures far above the energy gap . for the narrow band the approximate statistics @xmath191 is indistinguishable from the exact curve . . comparison between a simple bosonic approximation , the non - self - consistent solution , the self - consistent solution , the approximate statistics from ref . @xcite , and the exact fermionic expression . ] figure [ fig.thermal_occupation_w4 ] shows the thermal occupation for the wide band @xmath160 for the same temperature range as in the narrow band case . the thermal occupation does not grow as fast as for the narrow band due to the larger energy scale @xmath192 . the free boson approximation holds up to @xmath193 while the non - self - consistent calculation is correct up to @xmath194 . again the self - consistent solution agrees excellently with the exact result , while the approximate statistics @xmath191 overestimates the occupation for temperatures above @xmath195 . for temperatures @xmath196 ( not shown ) the approximate statistics and the exact curve merge again . . comparison between a simple bosonic approximation , the non - self - consistent solution , the self - consistent solution , the approximate statistics from ref . @xcite , and the exact fermionic expression . ] in the previous subsection we gauged the diagrammatic approach by comparing it to the exact fermionic results . in the present subsection , we discuss the results of the diagrammatic expansion for various temperatures and band - widths in more detail . as before we restrict ourselves to the self - consistent solutions . even though self - consistency improves the results , we still require that the temperature is not too high , so that higher order processes can be neglected . we stress that the single particle gap @xmath5 is the most important energy scale , but also the band - width @xmath192 at zero temperature plays an important role . especially for the same temperature @xmath22 and gap @xmath5 the narrow band limit @xmath197 and the wide band limit @xmath198 can differ significantly , because narrow bands generically allow for a much larger fraction of thermal excitations , i.e. , a higher density of thermally excited hard - core bosons , than wide bands . at various temperatures for the gap mode @xmath165 for @xmath157 . the vertical line indicates the energy of the @xmath165 mode at zero temperature . ] at various temperatures for the maximum mode @xmath156 for @xmath157 . the vertical line indicates the energy of the @xmath156 mode at zero temperature . ] in figs . [ fig.imre_sigma_p0_sc_tdep ] and [ fig.imre_sigma_p1_sc_tdep ] we investigate the real and the imaginary part of the self energy for the gap mode and for the maximum mode , respectively , in case of the narrow band @xmath157 . the imaginary part is dominated by the two - particle continuum convoluted with the single particle green function , see eq . . upon increasing temperature the imaginary part gains weight . we clearly see for @xmath119 , that the imaginary part is not well approximated by a constant but rather by a linear function , leading to an asymmetric line - shape of the spectral function @xmath101 . while the gap mode shows a tail towards higher energies due to the positive slope of the imaginary part of the self energy , the maximum mode shows a tail towards lower energies , induced by the negative slope . the real part of the self energy is dominated by the term @xmath199 in eq . , for very high and very low energies . for @xmath165 the real part is positive , indicating a shift of the peak position towards higher energies , while for @xmath156 the real part is negative , leading to a shift towards lower energies . with increasing temperatures the effect is amplified due to the increased scattering from thermal excitations . at zero temperature the single - particle spectral function @xmath101 is a @xmath65-function in frequency , signaling stable quasi - particles and providing the dominant contributions to the dsf . our approach includes this basic property , because the expression for the self energy @xmath200 in eqs . and vanishes for @xmath201 . only because our numerical implementation to calculate @xmath101 is restricted to a finite frequency and momentum resolution , we require that the temperature is not too low . otherwise , we are not able to resolve the spectral function . especially the self - consistent solutions require a finite broadening as initial input . this inital broadening does not matter at all once the computation is iterated for self - consistency and convergence is indeed reached . in our case , using up to @xmath202 gb of memory to resolve the spectral function , we require that @xmath203 . at various temperatures for the gap mode @xmath165 for @xmath157 . ] first , we consider the temperature dependence of the gap mode at momentum @xmath165 for the narrow band case @xmath157 calculated self - consistently in fig . [ fig.img_p0_sc_tdep ] . for low temperatures the response is primarily of lorentzian shape and centered at the zero temperature response @xmath204 . this changes distinctively at higher temperatures . due to increased scattering with thermal excitations the height of the spectral function decreases . the decrease scales with @xmath36 . the maximum of the response shifts towards higher energies and the peak broadens asymmetrically towards higher energies . for @xmath157 . ] the degree of asymmetry is clearly visible in fig . [ fig.img_p0_sc_lorentzian ] where we compare the spectral function to a lorentzian fit . while a pure lorentzian shape provides evidence for incoherent scattering @xcite , i.e. , an exponential decay in the time domain , asymmetric deviations imply non - trivial scattering . this observation agrees with recent theoretical as well as experimental studies @xcite . at various temperatures for the maximum mode @xmath156 for @xmath157 . ] second , we consider the temperature dependence of the maximum mode at momentum @xmath156 for the narrow band case @xmath157 calculated self - consistently . this quantity is depicted in fig . [ fig.img_p1_sc_tdep ] . similar to the gap mode , the spectral function of the maximum mode has a lorentzian shape and it is centered at the zero temperature position @xmath205 for low temperatures . the peak broadens with rising temperature and becomes asymmetric . in contrast to the gap mode , the peak position of the maximum mode shifts towards lower energies . concomitantly , the asymmetric line - shape accumulates weight at lower energies . an overview plot of the self - consistent spectral function @xmath101 at fixed temperature @xmath206 is given in fig . [ fig.img_t04_sc_pdep ] . one clearly sees how the asymmetry slowly changes from the gap mode to the maximum mode in dependence of total momentum @xmath47 . in the center of the dispersion , at @xmath207 , the response remains symmetric and does not change its position with respect to the @xmath184 response . for @xmath157 . note the offset of the y - axis . ] the asymmetry and the shift of positions observed in the previous subsection indicates that the total band - width of the system decreases upon rising temperature . we examine this feature in more detail in figs . [ fig.dispersion_tdep ] and [ fig.dispersion_tdep_w4 ] for the narrow and the wide band case , respectively . the position of the maximum in the response is plotted as function of the total momentum @xmath47 for various temperatures @xmath208 . while for low temperatures the maximum is located at the dispersion @xmath209 , the band is narrowing upon increasing temperature . this effect can be explained by the thermal occupation of an increasing number of sites , blocking the propagation of an inserted particle . consequently , the energy gap is increased and the band - width is decreased . in the literature this is often called the temperature dependence of the gap @xcite and of the dispersion , although the physical parameters @xmath5 and @xmath192 of the model do not change with temperature in the strict sense . in contrast to the narrow band case @xmath157 , the wide band case @xmath160 indicates that the shift is significantly stronger for the gap mode than for the maximum mode . for various temperatures . ] for various temperatures . the inset shows a more detailed plot for the gap mode and for the maximum mode . ] in this paper , we benchmarked a diagrammatic approach to particle - conserving models which is capable of dealing with thermal fluctuations in leading order in @xmath3 . we stress that particle - conserving effective models are free from quantum fluctuations at zero temperature because the vacuum is the ground state . such effective models can be systematically derived for a large class of microscopic hamiltonians such as gapped quantum antiferromagnets , for instance dimerized spin chains @xcite , spin ladders @xcite , two - dimensionally coupled spin ladders @xcite , and three - dimensionally coupled spin dimers @xcite . for the benchmark , we study the spectral properties of an exactly solvable gapped one - dimensional hard - core boson model with nearest - neighbor hopping and no other interaction but the hard - core repulsion . we used diagrammatic perturbation theory to calculate the single - particle self energy . the hard - core constraint was enforced by an infinite on - site interaction . the small control parameter is the density of thermal excitations , which implies that the summation of ladder diagrams is necessary and sufficient to capture the leading order in @xmath36 . we emphasize that this concept applies equally in any other dimension which is a particular asset of the approach advocated and tested here . our results show how the single - particle @xmath65-peak in the spectral function broadens with increasing temperature and how band - width and momentum influence the spectral function . we used the mapping to interaction - free jordan - wigner fermions to obtain results which are numerically exact except for finite size and time effects . by this data we have gauged our diagrammatic approach and evaluated its limits . for low temperature very good agreement is reached . both methods show that the shift of the peak position is a second order effect @xmath121 for the cosine band . in contrast , the width of the peak is a first order effect @xmath210 . thus , it is captured very well by the diagrammatic approach for low enough temperatures . our findings agree with those by essler and co - workers obtained by direct computations of the leading contributions to the partition sums @xcite . the focus of the refs . @xcite is on the relevant continuum field - theoretical model while ref . @xcite studied the strong ising limit of a spin chain . the former references deal with an infinitely wide band without lattice while the latter treats a very narrow band on a lattice . both studies are specific to one dimension . with the help of the spectral function we also calculated the thermal occupation , that is , the total number of thermally excited bosons . the corresponding results of the self - consistent diagrammatic approach are in excellent agreement with the exact analytic expression available from the fermionic description , even for temperatures far above the gap @xmath5 . we found clear evidence that ( i ) the line - shapes become asymmetric and that ( ii ) the band width of the overall dispersion narrows upon increasing temperature . note that both phenomena seem to be not a specific feature of one dimension but also show up in quantum magnets which are coupled in all three space dimensions . one very recent example is given in ref . @xcite , where the asymmetric broadening is observed by inelastic neutron scattering in the three dimensional antiferromagnet @xmath211 . from a theoretical point of view , we expect that the asymmetric line - shapes as well as the narrowing of the total band width do not depend qualitatively on dimensionality . the argument in support of this expectation is that both phenomena are most pronounced in the case of relatively narrow bands of the order of the gap and at temperatures also of the order of the gap . in this regime , modes at all momenta are thermally excited and contribute , not only the modes at the lower band edge . thus the bulk of the density - of - states matters . the dimensionality , however , influences mostly the band edges and the van hove singularities at the edges of the density - of - states . a different scenario is displayed by systems with a very high or even infinite bandwidth , where the dynamics of the thermal excitations strongly depend on the lower band edge singularity . further calculations for higher dimensional systems are beyond the scope of the present paper , but they are clearly called for in the near future . we conclude that the proposed approach is successfully benchmarked against the exact solution . this finding suggests to generally extend the success of particle - conserving effective models at zero temperatures to small , but finite temperatures by this diagrammatic technique . this enlarges the applicability and thus the usefulness of such effective models considerably . especially the combination with the cut approach seems promising . by means of the cuts , the quantum fluctuations of the hard - core bosons at zero temperature are treated in high precision by mapping the microscopic models to effective , particle - conserving ones . we emphasize that the presented diagrammatic approach does not rely on integrable field theories or other properties specific to one dimension . thus , various extensions suggest themselves . one may consider more complicated dispersions and multi - flavored hard - core bosons in exactly the same way . furthermore , one may include other interactions including correlated hopping processes for the hard - core excitations at least on a mean - field level . most importantly , it is conceptually possible to extend the presented approach to higher dimensions and to different models which are currently subject of ongoing experimental research , see for instance refs . . we conclude that the diagrammatic expansion is an efficient and universal method to calculate line - shapes at not too high temperatures from the particle - conserving effective models for a wide range of gapped systems with hard - core bosonic excitations . this has been shown in the present work on the level of a testbed calculation in one dimension . various extensions can be tackled next . we thank t. fischer and f. keim for useful discussions . we acknowledge financial support of the helmholtz virtual institute `` new states of matter and their excitations '' . b.f . acknowledges the fakultt physik of the technische universitt dortmund for his funding in the program `` bestenfrderung '' . here , we modifiy the diagrammatic derivations given in sect . [ ssect : gen_calcul ] so that the dressed propagator @xmath212 is used instead of the bare one @xmath213 @xmath214 this changes the calculation of @xmath53 in , @xmath215 which leads to a modified expression for @xmath216 @xmath217 the calculation of @xmath77 in remains unchanged so that @xmath76 now reads @xmath218 where @xmath219 due to the self - consistency , the weight function @xmath82 now depends on total momentum @xmath47 . the spectral function for the self energy reads @xmath220 } - 1 } - \frac{f_{k}(x)}{e^{\beta x } - 1 } a_{k - p}(x - y ) \right ] , \end{aligned}\ ] ] which is again a two - dimensional convolution @xmath221 } - 1 } - \frac{f_{k}(x)}{e^{\beta x } - 1 } a_{-(p - k)}(-(y - x ) ) \right ] .\end{aligned}\ ] ] the real and imaginary parts of the self energy finally read @xmath222 \frac{1}{e^{\beta x'}-1 } + \mathcal{p } \int\limits_{-\infty}^{\infty } \frac{\rho_{\sigma , p}(x)}{\omega - x } \mathrm{d}x , \\ \mathrm{im } \sigma(p , \omega ) & = - \pi \rho_{\sigma , p}(\omega ) .\end{aligned}\ ] ] to make the diagrammatic expansion numerically tractable , we discretize all quantities of interest in momentum and frequency space . for momentum space we typically use up to @xmath223 points , while up to @xmath224 points are used in frequency space to ensure a good resolution . using smaller numbers does not change the results significantly . the numbers of points are powers of @xmath225 to make use of fast radix-2-algorithms from the fftw library @xcite . the necessary calculations for the diagrammatic expansion of the spectral functions can be divided into four parts . first the calculation of @xmath57 in . in the non - self - consistent we employ @xmath226 ) \nonumber \\ & \cdot \left ( \frac{1}{e^{\beta \omega(p+l ) } - 1 } - \frac{1}{e^{- \beta \omega(-l ) } - 1 } \right ) \\ & = \frac{-1}{2 \pi } \sum\limits_i \frac{1}{\left| \omega'(-l_i)+\omega'(p+l_i ) \right| } \nonumber\\ & \cdot \left ( \frac{1}{e^{\beta \omega(p+l_i ) } - 1 } - \frac{1}{e^{- \beta \omega(-l_i ) } - 1 } \right)\end{aligned}\ ] ] where @xmath227 is the derivative of the dispersion with respect to @xmath228 . the values @xmath229 are the roots of the function @xmath230 $ ] . they are calculated numerically using a one - dimensional root finding algorithm . in the self - consistent case the calculation can be carried out as given in because it represents a two - dimensional convolution in the variables @xmath228 and @xmath231 . it can be calculated using conventional , fast convolution algorithms , based on fast fourier transforms ( ffts ) . the second step is the calculation of @xmath77 in . here the main intricacy is the calculation of the principal value . this can be carried out efficiently using the tricks proposed by liu and kosloff @xcite . the third step is the calculation of @xmath232 in . in the non - self - consistent case this can be done using standard one - dimensional integration algorithms , while the self - consistent case can again be mapped to a two - dimensional convolution , see . finally , some remarks on the calculation of the real and imaginary parts of the self energy . while the imaginary part is again trivial , the real part requires another principal value integral and three one- or two - dimensional integrals . it turns out that the calculations based on two - dimensional convolutions are much faster than those in the non - self - consistent case . this can be traced back to the fact that a single two - dimensional convolution is sufficient to calculate quantities such as @xmath57 , while the same quantity in the non - self - consistent case requires the computation of a full set of roots for each pair of values @xmath231 and @xmath47 we are interested in .

a ray - based wavefield description is employed in the interpretation of measurements made during the november 1994 acoustic engineering test ( the aet experiment ) . in this experiment phase - coded pulse - like signals with 75 hz center frequency and 37.5 hz bandwidth were transmitted near the sound channel axis in the eastern north pacific ocean . the resulting acoustic signals were recorded on a moored vertical receiving array at a range of 3252 km . in our analysis both mesoscale and internal - wave - induced sound speed perturbations are taken into account . much of this analysis exploits results that relate to the subject of ray chaos ; these results follow from the hamiltonian structure of the ray equations . we present evidence that all of the important features of the measured aet wavefields , including their stability , are consistent with a ray - based wavefield description in which ray trajectories are predominantly chaotic .

nlin0301026

the chaotic dynamics of ray trajectories in ocean acoustics have been explored in a number of recent publications ( @xcite-@xcite ) , including two recent reviews @xcite . in these publications a variety of theoretical results are presented and illustrated , mostly using idealized models of the ocean sound channel . in contrast , in the present study a ray - based wavefield description is employed , in conjunction with a realistic environmental model , to interpret a set of underwater acoustic measurements . for this purpose we use both oceanographic and acoustic measurements collected during the november 1994 acoustic engineering test ( the aet experiment ) . in this experiment phase - coded pulse - like signals with 75 hz center frequency and 37.5 hz bandwidth were transmitted near the sound channel axis in the eastern north pacific ocean . the resulting acoustic signals were recorded on a moored vertical receiving array at a range of 3252 km . we present evidence that all of the important features of the measured aet wavefields are consistent with a ray - based wavefield description in which ray trajectories are predominantly chaotic . the aet measurements have previously been analyzed in refs . @xcite-@xcite . in the analysis of worcester et al . @xcite it was shown that the early aet arrivals could be temporally resolved , were stable over the duration of the experiment and could be identified with specific ray paths . these features of the aet observations are consistent with those of other long - range data sets @xcite . the aet arrivals in the last 1 to 2 sec ( the arrival finale ) could not be temporally resolved or identified with specific ray paths , however . also , significant vertical scattering of acoustic energy was observed in the arrival finale . these features of the aet arrival finale are consistent with measurements and simulations ( both ray- and parabolic - equation - based ) at 250 hz and 1000 km range @xcite-@xcite and at 133 hz and 3700 km range @xcite . the analysis of colosi et al . @xcite focused on statistical properties of the early resolved aet arrivals . measurements of time spread , travel time variance and probability density functions ( pdfs ) of peak intensity were presented and compared to theoretical predictions based on a path integral formulation as described in ref . travel time variance was well predicted by the theory , but time spreads were two to three orders of magnitude smaller than theoretical predictions , and peak intensity pdfs were close to lognormal , in marked contrast to the predicted exponential pdf . the surprising conclusion of this analysis was that the measured aet pulse statistics suggested propagation in or near the unsaturated regime ( characterized physically by the absence of micromultipaths and mathematically by use of a perturbation analysis based on the rytov approximation ) , while the theory predicted propagation in the saturated regime ( characterized by the presence of a large number of micromultipaths whose phases are random ) . in ref . @xcite it was shown that in the finale region of the measured aet wavefields , where no timefronts are resolved , the intensity pdf is close to the fully saturated exponential distribution . this result is not unexpected because the finale region is characterized by strong scattering of both rays @xcite and modes @xcite . the work reported here seeks to elucidate both the physics underlying the aet measurements , and the causes of the successes and failures of the analyses reported in refs . we show that a ray - based analysis , in which ideas associated with ray chaos play an important role , can account for the stability of the early arrivals , their small time spreads , the associated near - lognormal pdf for intensity peaks , the vertical scattering of acoustic energy in the reception finale , and the near - exponential intensity pdf in the reception finale . part of the success of this interpretation comes about due to the presence of a mixture of stable and unstable ray trajectories . also , an explanation is given for differing intensity statistics in the early and late portions of the measured wavefields , and the fairly rapid transition between these regimes . the remainder of this paper is organized as follows . in the next section we describe the aet environment and the most important features of the measured and simulated wavefields . in the simulated wavefields both measured mesoscale and simulated internal - wave - induced sound speed perturbations are taken into account . it is shown that even in the absence of internal waves , the late arrivals are not resolved and the associated ray paths are chaotic . in section ii we present results that relate to the structure of the early portion of the measured timefront and its stability . the micromultipathing process is discussed , and a quantitative explanation for the remarkably small time spreads of the early ray arrivals is given . section iii is concerned with intensity statistics . an explanation is given for the cause of the different wavefield intensity statistics in the early and late portions of the arrival pattern . in the final section we summarize and discuss our results . figure 1 shows a measured aet wavefield in the time - depth plane and ray - based simulations of such a wavefield with and without internal waves . figure 2 is a complementary display that shows measured and simulated wavefields after plane wave beamforming . in this section we describe the most important qualitative features of measured and simulated aet wavefields . we begin by reviewing fundamental ray theoretical results so that our ray simulations can be fully understood . also , this material provides a foundation for some of the later discussion . we then briefly describe the aet environment and the signal processing steps that were performed to produce the measured wavefield shown in fig . , our treatment of internal - wave - induced sound speed perturbations is briefly described . finally , we return to describing and comparing qualitative features of the measured and simulated wavefields . fixed - frequency ( cw ) acoustic wavefields satisfy the helmholtz equation , @xmath0 where @xmath1 is the angular frequency of the wavefield and @xmath2 is the sound speed . ( the extension to transient wavefields is straightforward using fourier synthesis . it is not necessary to consider this complication to present the most important results needed below . ) we assume propagation in the vertical plane with cartesian coordinates @xmath3 and @xmath4 representing depth and range , respectively . ray methods can be introduced when the wavelength @xmath5 of all waves of interest is small compared to the length scales that characterize variations in @xmath6 . substituting the geometric ansatz , @xmath7 representing a sum of locally plane waves , into the helmholtz equation gives , after collecting terms in descending powers of @xmath8 , the eikonal equation , @xmath9 and the transport equation , @xmath10 for notational simplicity , the subscripts on @xmath11 and @xmath12 have been dropped in ( [ eik ] ) and ( [ trans ] ) . the solution to the eikonal equation ( [ eik ] ) can be reduced to solving a set of ray equations . if a one - way formulation ( see , e.g. , ref . @xcite ) is invoked , with @xmath4 playing the role of the independent variable , the ray equations are @xmath13 and @xmath14 where @xmath15 these equations constitute a nonautonomous hamiltonian system with one degree of freedom ; @xmath3 and @xmath16 are canonically conjugate position and momentum variables , @xmath4 is the time - like variable , @xmath17 is the hamiltonian and the travel time @xmath12 corresponds to the classical action . note also that @xmath18 and @xmath19 are the @xmath3- and @xmath4-components , respectively , of the ray slowness vector . the ray angle relative to the horizontal @xmath20 satisfies @xmath21 , or , equivalently , @xmath22 . for a point source rays can be labeled by their initial slowness @xmath23 . the solution to the ray equations is then @xmath24 , @xmath25 , @xmath26 . the terms in ( [ geom ] ) correspond to eigenrays ; these satisfy @xmath27 where the depth and range of the receiver are @xmath28 and @xmath29 , respectively . the transport equation ( [ trans ] ) can be reduced to a statement of constancy of energy flux in ray tubes . the solution , for the @xmath30-th eigenray , can be written @xmath31 where @xmath17 , the matrix element @xmath32 and the maslov index @xmath33 are evaluated at the endpoint of the ray , and @xmath34 is a constant . the stability matrix @xmath35 quantifies ray spreading , @xmath36 elements of this matrix evolve according to @xmath37 where @xmath38 at @xmath39 is the identity matrix , and @xmath40 at caustics @xmath32 vanishes and , for waves propagating in two space dimensions , the maslov index advances by one unit . in ocean environments with realistic range - dependence , including the aet environment , ray trajectories are predominantly chaotic . chaotic rays diverge from neighboring rays at an exponential rate , on average , and are characterized by a positive lyapunov exponent , @xmath41 here @xmath42 is a measure of the separation between rays at range @xmath4 ; suitable choices for @xmath43 are any of the four matrix elements of @xmath38 or the trace of @xmath38 . the chaotic motion of ray trajectories leads to some limitations on predictability . this will be discussed in more detail below . additional background material relating to this topic can be found in @xcite and references therein . we digress now to briefly describe the aet experiment and relevant signal processing . in this experiment a source , suspended at 652 m depth from the floating instrument platform r / p flip in deep water west of san diego , transmitted sequences of a phase - coded signal whose autocorrelation is pulse - like ( with a pulse width of approximately 27 ms ) . the source center frequency was 75 hz and the bandwidth was approximately 37.5 hz . the resulting acoustic signals were recorded east of hawaii at a range of 3252.38 km on a moored 20 element receiving array between the depths of 900 m and 1600 m. after correcting for the motion of the source and receiving array , including removal of associated doppler shifts , the received acoustic signals were cross - correlated with a replica of the transmitted signal ( to achieve the desired pulse compression ) and complex demodulated . to improve the signal - to - noise ratio , 28 consecutive processed receptions , extending over a total duration of 12.7 minutes , were coherently added . the duration over which this coherent processing yields improved signal - to - noise is limited by ocean fluctuations , mostly due to internal waves . the statistics of the acoustic fluctuations were shown in ref . @xcite not to be adversely affected by the 12.7 minute coherent integration . nearly concurrent with the week - long experiment , temperature profiles in the upper 700 m were measured with xbt s every 30 km along the transmission path . an objective mapping technique was applied @xcite to these measurements to produce a map of the background sound speed structure ( including mesoscale structure ) along the transmission path . the sampling interval in range of the resulting sound speed field is 30 km so that no structure with horizontal wavelengths less than 60 km is present in this field . more details on the aet experiment , environment and processing of the acoustic signals can be found in ref . @xcite . internal - wave - induced sound speed perturbations are taken into account in most of our simulations . these are assumed to satisfy the relationship @xmath44 and the statistics of @xmath45 , the internal - wave - induced vertical displacement of a water parcel , were assumed to be described by the empirical garrett - munk spectrum ( see , e.g. , ref . the potential sound speed gradient @xmath46 , the buoyancy frequency @xmath47 , and the sound speed @xmath6 were estimated directly from hydrographic measurements using a procedure similar to that described in the appendix of ref . @xcite ; it was found that a good approximation for the aet environment is @xmath48 . the vertical displacement @xmath49 was computed using equation 19 of ref . @xcite with the variable @xmath50 replaced by @xmath4 , and @xmath51 . physically this corresponds to a frozen vertical slice of the internal wave field that includes the influence of transversely propagating internal wave modes . a fourier method , with @xmath52 km and @xmath53 km , was used to numerically generate sound speed perturbation fields . ( ray calculations were also performed using an internal wave field in which the hard upper bound on @xmath54 was replaced by an exponential damping of the large @xmath54 energy . the qualitative ray behavior described below did not depend on the manner in which the large @xmath54 cut - off was treated . ) a mode number cut - off of @xmath55 was used in all the simulations shown . contributions from approximately @xmath56 ( @xmath57 where the factors of @xmath58 are present because the assumed wavenumber cut - off was one - fifth of the assumed nyquist wavenumber ) internal wave modes were included in our simulated internal - wave - induced sound speed perturbation fields . measurements of sound speed variance made during the experiment ( see fig . 3 ) indicate that the internal wave energy strength parameter @xmath59 was close to @xmath60 , the nominal garrett - munk value . simulations were performed using both @xmath61 and @xmath62 . all of the simulations shown in this paper use @xmath61 . the difference between these simulations and those performed using @xmath62 will be discussed where appropriate . with the foregoing discussion as background , we are prepared to discuss our ray simulations in the aet environment . these are shown in figs . 1 , 2 , 4 through 11 , and 14 . the points plotted in figs . 1 , 2 and 4 were computed by integrating the coupled ray equations ( [ ray1 ] , [ dtdr ] ) from the source to the range of the receiving array . several fixed and variable step - size integration algorithms were tested . in the presence of internal waves , convergence using a fixed step fourth order runge kutta integrator required that the step size not exceed approximately one meter . all ray tracing calculations presented in this paper were performed using double precision arithmetic ( 64 bit floating point wordsize ) . the wavefield intensity is approximately proportional to the density of rays ( dots ) that are plotted in figs . 1 and 2 . a somewhat more difficult calculation in which the contributions to the wavefield from many rays are coherently added will be discussed below . in the presence of internal waves , rays with launch angles between approximately @xmath63 form the diffuse finale region of the arrival pattern arriving after approximately 2196 sec . steeper rays contribute to the earlier , mostly resolved arrivals . it is seen in fig . 1 that our ray simulation in the presence of internal waves underestimates the vertical spread of the finale region . simulations with a stronger internal wave field , @xmath62 , give a better fit to the vertical spread of energy ( but a poorer fit to other wavefield features , as described below ) . note , however , that diffractive effects , missing in our ray simulations , will contribute to enhanced broadening in depth of the arrival pattern . a striking feature of figs . 1 , 2 and 4 is the contrast between the seemingly structureless distribution of steep ( large @xmath64 ) rays in @xmath65 seen in fig . 4 when internal waves are present and the relatively structured distribution of the same rays in the time - depth ( fig . 1 ) and time - angle ( fig . 2 ) plots . the cause of the complexity seen in fig . 4 is ray chaos @xcite ; most ray trajectories diverge from initially infinitesimally perturbed rays at an exponential rate . 48,000 rays with launch angles between @xmath66 were traced to produce figs . 1 , 2 and 4 in the presence of internal waves . this fan of rays is far too sparse to resolve what should be an unbroken smooth curve a lagrangian manifold which does not intersect itself in phase space ( fig . 4 ) . under chaotic conditions the separation between neighboring rays grows exponentially , on average , in range . the complexity of the lagrangian manifold grows at the same exponential rate . the lyapunov exponent is the recriprocal of the e - folding distance ( see , e.g. , @xcite ) . finite range numerical estimates of lyapunov exponents ( hereafter referred to as stability exponents ) are shown as a function of launch angle in fig . it is seen that in this environment the flat rays ( @xmath67 ) have stability exponents of approximately 1/(100 km ) , while the steeper rays ( @xmath68 ) have stability exponents of approximately 1/(300 km ) . it follows , for example , that the complexity of the flat ray portion of the lagrangian manifold grows approximately like @xmath69 . not surprisingly , stability exponents show some sensitivity to parameters that describe the internal wave field . our simulations show that the stability exponents of rays with axial angles of approximately @xmath70 or less approximately double when @xmath71 increases from 30 to 100 , while steeper rays show very little sensitivity to @xmath71 . in contrast , stability exponents of rays with axial angles from approximately @xmath72 to @xmath73 approximately double when @xmath74 increases from @xmath75 m to @xmath76 m , while the flatter rays show little sensitivity to @xmath74 . with these comments in mind , one should not attach too much significance to the details of fig . we note , however , that rays with axial angles of approximately @xmath70 or less consistently have higher stability exponents than rays in the @xmath70 to @xmath77 band . note that what appears to be temporally resolved arrivals in the measured wavefield correspond in our simulations to contributions from an exponentially large number of ray paths . ( only a small fraction of the total number are included in our simulations , however , because of the relative sparseness of our initial set of rays . ) the observation that the travel times of chaotic ray paths may cluster and be relatively stable was first made in ref . @xcite . simmen et al . @xcite have previously produced plots similar to our figs . 1 and 4 for ray motion in a deep ocean model consisting of a background sound speed structure very similar to ours on which internal - wave - induced sound speed perturbations were superimposed . although our results are similar in many respects , it is noteworthy that the right panel of our fig . 4 shows more chaotic behavior than is present in the corresponding plot in ref . this difference persists if our 3252.38 km range simulations are replaced by simulations at the same range ( 1000 km ) that was used in ref . our rays , especially the steeper rays , are more chaotic than those in ref . we believe that this difference is primarily due to the increased complexity of our internal wave field over that of simmen et al . @xcite who included contributions from ten internal wave modes in their simulated internal wave fields . figures 6 and 7 show plots of ray depth at the aet range vs. launch angle . in fig . 6 the same rays ( excluding those with negative launch angles ) that were used to produce figs . 1 , 2 and 4 are plotted . in fig . 7 two small angular bands of the upper panel of fig . 6 are blown up ; the ray density is four times greater than that used in fig . 6 strongly suggests that , even in the absence of internal waves , the near - axial rays are chaotic . not surprisingly , in the presence of internal waves , steeper rays are also predominantly chaotic . note , however , that figs . 5 , 6 and 7 show that the near - axial rays are evidently more chaotic than the steeper rays . an explanation for this difference in behavior will be given in section iii . in this section we have briefly described and compared the gross features of measured and simulated aet wavefields . ray trajectories in our simulated wavefields in the presence of internal waves are predominantly chaotic . in spite of this , figs . 1 and 2 show that many features of our simulated wavefields are stable and in good qualitative agreement with the observations . predicted and measured spreads of acoustic energy in time , depth and angle are generally in good agreement , both in the early and late portions of the arrival pattern . we shall not discuss further the spreads of energy in depth and angle , except to note that figs . 1 and 2 show that these spreads can be accounted for using ray methods in the presence of realistic ( including internal - wave - induced sound speed perturbations ) ocean structure . in the two sections that follow we shall consider in more detail the time spreads of the early ray arrivals , and intensity statistics in both the early and late portions of the arrival pattern . in this section we consider in some detail eigenrays , timefront stability and time spreads . we focus our attention on the early branches of the timefront , where measured time spreads were , surprisingly , only approximately 2 ms , in sharp contrast to the theoretical prediction of approximately 1 sec @xcite . we focus our attention on the early arrival time spreads , primarily because quantitative estimates of the unresolved late arrivals are not available . eigenrays at a fixed depth @xmath3 correspond to the intersections of a horizontal line at depth @xmath3 with the curve @xmath78 ; these intersections are the roots of the equation @xmath79 . although the discrete samples of @xmath78 plotted in figs . 6 and 7 are too sparse to reveal what should be a smooth curve , it is evident the number of eigenrays at almost all depths is very large ; because rays are predominantly chaotic , this number grows exponentially in range . in principle , eigenrays can be found using a combination of interpolation and iteration , starting with a set of discrete samples of @xmath78 . in practice , this procedure reliably finds only those eigenrays in regions where @xmath78 has relatively little structure . these rays have the highest intensity and are the least chaotic . it is seen in fig . 1 that the early ( steeper ) ray arrivals come in clusters with small travel time spreads . the set of eigenrays within such a cluster is often referred to as a set of micromultipaths . an incomplete set of micromultipaths , found using the procedure just described , is shown in fig . an interesting and important feature of micromultipaths in our simulations is that all of the rays that make up a set of micromultipaths have the same identifier @xmath80 . here @xmath81 is the sign of the launch angle and @xmath82 is the number of ray turning points ( where @xmath16 changes sign ) following a ray . in fig . 9 rays with two fixed values of ray identifier ( + 137 and + 151 ) are plotted in the time - depth plane . in both cases the points plotted are a subset of those plotted in fig . it is seen that the relatively steep + 137 rays have a small time spread and form one of the branches of the timefront seen in fig . 1 ( similar behavior was noted in ref . @xcite ) , while the relatively flat + 151 rays have a much larger time spread and fall within the diffuse finale region of the arrival pattern seen in fig . 1 . an important ( and surprising ) feature of sets of micromultipaths is that they are highly nonlocal in the sense that interspersed ( i.e. , having intermediate launch angles ) among a group of micromultipaths are rays with @xmath82 values that differ by several units . this behavior is illustrated in fig . 10 . in view of the observation that , in the presence of internal waves in the aet environment , ray motion is strongly chaotic and the function @xmath83 has local oscillations of several units , it is remarkable that the early portion of the timefront ( see fig . 1 ) is not destroyed . the nonlocality of a set of micromultipaths in the range - depth plane is seen in fig . note that there are significant differences in the upper and lower turning depths of the plotted micromultipaths and that these rays are spread in range by a significant fraction of a ray cycle distance . the result of performing a ray - based synthesis of what appears in the measurements to be an isolated arrival is shown in fig . this involves finding a complete set of micromultipaths , including their travel times and maslov indices , and coherently summing their contributions . unfortunately , owing to the predominantly chaotic motion of ray trajectories in the aet environment in the presence of internal waves , it is extremely difficult to find a complete set of micromultipaths . indeed , the constituent ray arrivals used in the fig . 11 synthesis do not constitute a complete set of micromultipaths . this is less important than might be expected because standard eigenray finding techniques easily locate the strongest micromultipaths ; the missing micromultipaths in the fig . 11 synthesis are highly chaotic rays that have very small amplitudes . an interesting result of this synthesis is that the micromultipath - induced time spread in the synthesized pulse is only about 1 ms , which is in very good agreement with the aet measurements @xcite . note that the total time spread among the micromultipaths found about 11 ms is much greater than the spread of the synthesized pulse . this difference arises because the time spread of the dominant micromultipaths is much smaller than the total micromultipath time spread . simulations ( not shown ) using @xmath62 result in synthesized steep ray pulses that are spread in time much more than is shown in fig . 11 . with the stronger internal wave field , the total micromultipath time spread increases by approximately a factor of two to about 25 ms . more importantly , however , the dominant micromultipaths have time spreads comparable to the total micromultipath time spread , leading to synthesized pulses that are spread by 5 to 10 ms , rather than 1 to 2 ms . thus , measured early arrival time spreads are in better agreement with @xmath61 simulations than with @xmath62 simulations . the question of which value of the internal wave strength @xmath59 in our simulations gives the best fit to the observations will be revisited in the next section . an interesting and unexpected feature of our simulations is that the maslov index was consistently lower than the number of turns @xmath82 made by the same ray . the difference was typically three units in the @xmath61 simulations and four units in the @xmath62 simulations . we have no explanation for this behavior . so far in this section we have seen that associated with the chaotic motion of ray trajectories is extensive micromultipathing , and that the micromultipathing process is highly nonlocal . the many micromultipaths that add at the receiver to produce what appears to be a single arrival may sample the ocean very differently . surprisingly , on the early arrival branches the highly nonlocal micromultipathing process causes only very small time spreads and does not lead to a mixing of ray identifiers . to complete this picture it is necessary to explain why the time spreads of the early ( steep ) ray arrivals are so small . we shall now provide such an explanation . recall first , however , that the measured time spread of the early arrivals is only approximately 2 ms @xcite , in marked contrast to the theoretical prediction @xcite of approximately 1 sec . thus our quantitative explanation for the small early arrival time spreads serves both to give insight into the underlying physics , and provides an explanation for a critically importnat feature of the measured aet wavefields . computing time spreads is conceptually straightforward using ray methods . in a known environment one finds a large ensemble of rays with the same identifier that solve the same two point boundary value problem ; the spread in travel times over many ensembles of such rays , each ensemble computed using in a different realization of the environment , is the desired quantity . the constraint that all of the computed rays have the same fixed endpoints complicates this calculation . in the following we exploit an approximate form of the eigenray constraint that simplifies the calculation ; the approximate form of the eigenray constraint is first order accurate in a sense to be described below . before describing the manner in which the eigenray constraint is imposed , it is useful to describe the physical setting in which the constraint will be applied . figure 12 shows @xmath84 vs @xmath4 following a moderately steep ray ( axial angle approximately @xmath85 ) in a canonical environment @xcite on which the previously described ( based on the measured aet @xmath86 profile ) internal - wave - induced sound speed perturbation field was superimposed . it is seen that @xmath87 consists of a sequence of essentially constant @xmath17 segments , separated by near - step - like jumps . the jumps occur at the ray s upper turning points . a simple model that captures this behavior the so - called apex approximation @xcite assumes that the transition regions can be approximated as step functions . ( the validity of the apex approximation is linked to the anisotropy and inhomogeneity of oceanic internal waves ; details of the background sound speed profile are not important . for this reason we have chosen to illustrate this effect using a background canonical profile rather than the aet environment . ) figure 13 shows a schematic diagram of two rays . one ray has undergone two internal - wave - induced apex scattering events ; the other ray is an unscattered ray with the same launch angle as the scattered ray . our strategy to building the eigenray constraint into an estimate of travel time spreads can now be stated . we consider a scattered and an unperturbed ( that is , in the absence of internal - wave - induced scattering events ) ray with the same ray identifier ; the rays shown schematically in fig . 13 have identifier @xmath88 . in addition to having the same ray identifier , the scattered and unperturbed rays are assumed to start from the same point and end at the same depth , but they will generally end at different ranges . we assume that the scattered ray has travel time @xmath89 and is one of many scattered eigenrays with the same ray identifier at range @xmath4 . we estimate the travel time @xmath90 of an unperturbed eigenray at range @xmath4 whose ray identifier is the same as that of the scattered eigenray . it will be shown that @xmath91 vanishes to first order if the apex approximation is exploited . because the same result applies to all of the scattered eigenrays , it follows that the time spread vanishes to first order , independent of @xmath4 , if the apex approximation is exploited . note that this result is not expected to apply , even approximately , to the late near - axial rays where the apex approximation is known to fail . the travel time @xmath90 of the eigenray in the unperturbed ocean can , of course , be computed numerically but this provides little help in deriving an analytical estimate of @xmath92 . instead , consider a ray in the unperturbed ocean with the same launch angle as one of the scattered eigenrays , as shown in fig . 13 . in general , the range of this ray at the receiver depth , after making the appropriate number of turns , will be @xmath93 . but @xmath90 can be estimated from @xmath94 . this follows from the observation that ray travel time ( assuming a point source initial condition , for instance ) is a continuously differentible function of @xmath3 and @xmath4 with @xmath95 where @xmath96 is the ray slowness vector . because of this property @xmath97 can be expanded in a taylor series . if we consider rays with the same ray identifier and fix the final ray depth to coincide with the receiver depth , then @xmath98 where @xmath99 is the @xmath4-component of the ray slowness vector . ( more generally , @xmath97 consists of multiple smooth sheets that are joined at cusped ridges ; in spite of this complication , the property of continuous differentiability is maintained , even in a highly structured ocean . for our purposes , we need only consider one sheet of this multi - valued function . note also that our application of ( [ t(r ) ] ) in the background ocean makes this expression particularly simple to use . ) it follows from ( [ t(r ) ] ) that @xmath100 is , correct to first order , the travel time difference at range @xmath4 between eigenrays with and without internal waves . to evaluate @xmath101 we shall assume that the backgound environment is range - independent . also , consistent with the use of the apex approximation , we may treat the perturbed environment as piecewise range - independent . within each range - independent segment it is advantageous to make use of the action - angle variable formalism ( see . e.g. , @xcite or @xcite ) . in terms of the action - angle variables @xmath102 , @xmath103 and the ray equations are @xmath104 @xmath105 and @xmath106 where the physical interpretation of @xmath107 as @xmath108 is maintained . the angle variable can be defined to be zero at the upper turning depth of a ray and increases by @xmath109 over a ray cycle ( double loop ) . it follows that the frequency @xmath110 where @xmath111 is a ray cycle distance . integrating the ray equations over a complete ray cycle then gives @xmath112 and @xmath113 with @xmath114 constant following each ray . in the apex approximation @xmath114 jumps discontinuously at each ray s upper turning depth ; for the perturbed ray @xmath115 and @xmath116 where @xmath117 . note that , like these expressions , eqs . [ t(r ) ] and [ deltat ] are first order accurate in @xmath118 . consider again eq . [ deltat ] and fig . 13 with @xmath114 equal to @xmath119 , @xmath120 and @xmath121 in the left center and right ray segments , respectively . the left segment gives no contribution to @xmath92 as the ray has not yet been perturbed . in the center ray segment @xmath122 and @xmath123 . \label{pdelr}\ ] ] these terms are seen to cancel . note that if additional complete cycle ray segments are added to the center section of the ray , eqs . [ delt ] and [ pdelr ] are unchanged except that @xmath124 is replaced by @xmath125 ; again the two terms cancel . in the final ( incomplete cycle ) ray segment the difference between the terms @xmath126 and @xmath127 can be shown to be @xmath128 ; the terms do not exactly cancel because the @xmath129-values of the perturbed and unperturbed rays are generally not identical at the receiver depth . thus , to first order in @xmath118 , @xmath130 , independent of range , if the apex approximation is valid . this simple calculation provides an explanation of why , in spite of extensive ray chaos , the time spreads of the early aet ray arrivals are quite small . several comments concerning the preceeding calculation are noteworthy . first , we note that although it was assumed that the background sound speed structure is range - independent , the preceeding argument also holds in the presence of slow background range - dependence , i.e. , with structure whose horizontal scales are large relative to a typical ray double loop length . adiabatic invariance in such environments guarantees that while @xmath17 is not constant following rays between apex scattering events , @xmath114 is nearly constant . second , after @xmath131 random kicks @xmath132 in ( [ delt ] ) and ( [ pdelr ] ) is replaced by @xmath133 , and the magnitude of ( [ delt ] ) under aet conditions is @xmath134 . this is an example of a travel time spread estimate that fails to enforce the eigenray constraint . this calculation shows that the difference between travel time estimates that do and do not enforce the eigenray constraint can be quite significant . ( the constrained estimate is refined below . ) third , it should be emphasized that eqs . ( [ delt ] ) and ( [ pdelr ] ) apply ( approximately ) only to the steep rays because the apex approximation applies only to the steep rays . we have not addressed the time spreads of near - axial rays . note , however , that fig . 9 shows that time spreads are larger for flatter rays . fourth , the above calculation shows that , to lowest order in @xmath118 , there is no internal - wave - scattering - induced travel time bias if the apex approximation is strictly applied . and fifth , the arguments leading to eqs . ( [ delt ] ) and ( [ pdelr ] ) apply whether the scattered rays are chaotic or not . a relaxed form of the apex approximation in which the action jump transition region has width @xmath135 gives a nonzero travel time spread estimate . in the transition region , taken for convenience to be @xmath136 , one may choose a hamiltonian of the form @xmath137 where @xmath138 , and @xmath103 elsewhere . it follows that in the transition region @xmath139 and @xmath140 elsewhere . a simple generalization of the above calculation then gives for a complete ray cycle @xmath141 \label{delt2}\end{aligned}\ ] ] and @xmath142 , \label{pdelr2}\end{aligned}\ ] ] so the sum ( recall [ [ deltat ] ] ) is @xmath143 it should be emphasized that this expression is first order accurate in @xmath144 , but that no assumption about the smallness of @xmath135 has been made . as noted above , incomplete cycle end segment pieces give @xmath128 corrections to @xmath92 . if @xmath135 has approximately the same value at each upper turn , then one has after @xmath131 upper turns , correct to @xmath145 , @xmath146 consistent with the numerical simulations shown in fig . 12 , @xmath147 radians and @xmath148 ms . ( we , and independently f. henyey [ personal communication ] , have confirmed that these estimates also apply under aet - like conditons . ) with these numbers and @xmath149 , appropriate for aet , ( [ deltatsum ] ) gives a time spread estimate of approximately 13 ms , in approximate agreement with the numerical simulations shown in fig . note that ( [ deltatsum ] ) does not preclude a travel time bias . a cautionary remark concerning the use of ( [ deltatsum ] ) is that virovlyansky @xcite has pointed out that , owing to secular growth , the @xmath128 contribution to @xmath92 may dominate the @xmath145 contribution at long range . figure 9 shows that in the aet environment simulated near - axial ray time spreads are greater than simulated steep ray time spreads . we do not fully understand this behavior . a possible explanation for this behavior is that time spreads increase as rays become increasingly flat owing to the breakdown of the apex approximation . but one would expect that this trend should be offset , in part or whole , by the relative smallness of internal - wave - induced sound speed perturbations near the sound channel axis . in addition , we have seen some evidence that an additional factor may be important in the aet environment . namely , we have observed a positive correlation between travel time spreads and stability exponents ; stability exponents in the aet environment are shown in fig . we have chosen not to dwell on flat ray time spreads in this paper because there are no aet measurements of these spreads to which simulations can be compared . it is clear , however , that the issues just raised need to be better understood . in this section we consider the statistical distribution of the intensities of both the early and late aet arrivals . recall that experimentally the early arrival intensities have been shown @xcite to approximately fit a lognormal probability density function ( pdf ) and the late arrival intensities have been shown @xcite to fit an exponential distribution . the late arrival exponential distribution is not surprising as this distribution is characteristic of saturated statistics . the early arrival near - lognormal distribution is surprising , however , inasmuch as theory @xcite predicts saturated statistics , i.e. , an exponential intensity pdf . it has been argued @xcite that the theory can be modified in such a way as to move the early arrival prediction from saturated to unsaturated statistics . the latter regime is characterized by a lognormal intensity pdf . this fix is conceptually problematic inasmuch as , in this theory , the unsaturated regime is characterized by the absence of micromultipaths , which seriously conflicts with the numerical simulations presented in the previous section where the number of micromultipaths is very large . in this section we provide self - consistent explanations for both the late arrival exponential distribution and the early arrival lognormal distribution . the challenge is to reconcile the early arrival near - lognormal intensity pdf with the presence of a large number of micromultipaths . some of the arguments presented are heuristic , and some build on the results of numerical simulations . a complete theoretical understanding of intensity statistics has proven difficult . our approach to describing wavefield intensity statistics builds on the semiclassical construction described by eq . ( [ geom ] ) . at a fixed location it is seen that the wavefield amplitude distribution is determined by the distribution of ray amplitudes and their relative phases . note that both travel times and maslov indices influence the phases of ray arrivals . complexities associated with transient wavefields and caustic corrections will be discussed below . an important observation is that in the aet environment , including internal - wave - induced sound speed perturbations , simulated geometric amplitudes of both steep and flat ray arrivals approximately fit lognormal pdfs . this is shown in fig . previously it has been shown @xcite that ray intensities in a very different chaotic system also fit a lognormal pdf ; in that system single scale isotropic fluctuations are superimposed on a homogeneous background . ( note that all powers of a lognormally distributed variable are also lognormally distributed , so ray amplitudes have this property if and only if ray intensities have this property . ) the apparent generality of the near - lognormal ray intensity pdf suggests that it applies generally to ray systems that are far from integrable ; the arguments presented in ref . @xcite suggest that this should be the case . the intensity distributions of the early and late aet arrivals , corresponding to steep and flat rays , respectively , will be discussed separately . we consider first the early arrivals . we saw in the previous section that the micromultipaths that make up one of these arrivals have the same ray identifier and have a very small spread in travel time . the dominant micromultipaths were seen ( see fig . 11 ) to have a time spread of approximately 1 ms which is a small fraction of the approximately 13 ms period of the 75 hz carrier wave . also , our simulations show that the maslov indices of the dominant micromultipaths differ by no more than one unit . these conditions dictate that interference among the dominant micromultipaths is predominantly constructive . because travel time differences are so small , the pulse shape should have negligible influence on the distribution of peak intensities . in a model of the early aet arrivals consisting of a superposition of interfering micromultipaths , the micromultipath properties that play a critical role in controlling peak wavefield intensities are thus : 1 ) their amplitudes have a near - lognormal distribution ; 2 ) the dominant micromultipaths have negligible travel time differences ; and 3 ) the dominant micromultipaths have maslov indices that differ by no more than one unit . it should be noted that very different behavior would have been observed if : the source bandwidth were significantly more narrow as this would have caused micromultipaths with different ray identifiers to interfere with one another ; or the source center frequency were significantly higher as phase differences between interfering micromultipaths would then have been significant . an additional subtlety must be introduced now : the pdf of the intensities of the constituent micromultipaths that make up a single arrival is not identical to the pdf described in ref . @xcite and shown in the upper and middle panels of fig . the latter pdf describes the distribution of intensities of randomly ( with uniform probability ) selected rays leaving the source within some small angular band and whose range is fixed . this pdf is biased in the sense that it overcounts the micromultipaths with large intensities and undercounts those with small intensities . unbiased micromultipath intensity pdfs can be constructed from the biased pdfs shown . to do so , consider a manifold ( a smooth curve in phase space corresponding to a fan of initial rays ) which begins near @xmath150 and arrives in the neighborhood of @xmath151 at range @xmath4 . sampling in fixed steps of the differential @xmath152 ( as was done to produce the upper and middle panels of fig . 14 ; this is equivalent to uniform random sampling ) leads to a highly nonuniform density of points on the final manifold since @xmath153 propagates to @xmath154 . the greater @xmath32 , the lower the density of points locally at final range . a uniform sampling in final position is acheived instead by considering the initial conditions that would lead to @xmath155 . its density of points on the initial manifold can be deduced from its initial condition , @xmath156 . it is necessary to sample @xmath32 times more densely on the initial manifold in order to acheive uniform sampling in @xmath157 at range @xmath4 . to account for this effect , we need to know the pdf for @xmath32 with uniform initial sampling . roughly speaking , the pdf of the absolute values of the individual matrix elements of @xmath158 have the same form as for @xmath159 , apart from a shift of the centroid that is lower order in range than the leading term . from the results of ref . @xcite , the ( biased in the sense described above ) probability that @xmath32 falls in the interval between @xmath50 and @xmath160 is @xmath161 } \ ; , \ ; x \geq 0 \ ; .\end{aligned}\ ] ] here @xmath162 is a finite - range estimate of the lyapunov exponent based on an average ( over an ensemble of rays ) value of @xmath32 , and @xmath163 is the true lyapunov exponent . the new ( unbiased in the sense described above ) pdf , @xmath164 , for uniform sampling in @xmath157 is related to the previous one by @xmath165 } \ ; , \ ; x \geq 0\end{aligned}\ ] ] where the factor @xmath50 accounts for the extra counting weight of @xmath32 , and @xmath166 just preserves normalization and is calculated using eq . ( [ rhotrm ] ) . this calculation shows that the unbiased micromultipath ray intensity pdf also has a lognormal distribution ; the only change relative to the biased pdf is an increase in the mean from @xmath167 to @xmath168 . because lognormality is maintained , this correction represents only a trivial change to the problem . approximate lognormality of the constrained ( eigenray ) pdf of ray intensity is shown in the lower panel of fig . this pdf was constructed using the same eigenrays that were used to produce figs . 8 and 11 . the corresponding unconstrained ray intensity pdf is shown in the middle panel of fig . ( two constraints fixed receiver depth and fixed ray identifier are built into the lower panel pdf . a very similar constrained pdf results if only the receiver depth constraint is applied , provided ray launch angles are limited to the ` steep ' ray band used to construct the middle panel . ) it should be noted that the constrained ( eigenray ) pdf shown in the lower panel of fig . 14 was constructed from numerically found eigenrays ; because weak eigenrays are difficult to find numerically they are undercounted and the pdf is biased . in a practical sense this bias is of little consequence because the weak eigenrays that are difficult to find contribute negligibly to the wavefield . we return now to the problem of simulating the early aet arrivals . it is tempting to think that because the constituent micromultipaths have a near - lognormal distribution , the sum of their contributions should also be near - lognormally distributed . unfortunately , this is generally not the case . consider , for example , the special case in which phase , including maslov index , differences are negligible . then all micromultipaths interfere constructively and peak wavefield amplitudes can be modelled as the sum of many lognormally distributed variables . because all moments of the lognormal distribution are finite , the central limit theorem applies . under these conditions , if sufficiently many contributions are summed , the distribution of the sums the wavefield amplitudes would be a gaussian . to simulate the statistics of the early aet arrivals ( recall fig . 14 and the accompanying discussion ) we have used several variations of a simple model . an arrival was modelled as a sum of @xmath169 interfering micromultipaths whose : 1 ) amplitudes are lognormally distributed ; and 2 ) phases , @xmath170 , have a clearly identifiable peak . micromultipath contributions were coherently added . the peak intensity of the sum whose travel time is not known a priori was then recorded . using an ensemble of @xmath171 peak intensities , a peak intensity pdf was then constructed . peak intensity pdfs constructed in this fashion were found to be very close to lognormal ; a typical example is shown in fig . 15 . in this example @xmath172 hz , the @xmath173 s were identical , @xmath174 with equal probability ( note that choice of the integer @xmath30 is unimportant ) and @xmath175 . other combinations of distributions for @xmath173 ( either a gaussian or the limiting case of a delta distribution ) , @xmath176 ( taken either from @xmath177 or @xmath178 with equal probability ) , and the parameter @xmath169 ( between 2 and 100 ) were tested . these simulations showed that provided the phase constraint noted above was satisfied , a near log - normal peak intensity pdf resulted . two points regarding this simple model are noteworthy . first , this model does not constitute a theory of wavefield peak intensity statistics , but it does serve to demonstrate that our ray - based simulations of the early aet arrivals are consistent with the measured distribution of peak intensities . second , simulations ( not shown ) , performed with @xmath62 yield sets of dominant micromultipaths that violate assumption 2 ) ; phases are uniformly distributed and summing micromultipath contributions yields a distribution of peak intensities that is not close to lognormal . thus , our simulations suggest that a near - lognormal distribution of early arrival peak intensity requires a relatively weak internal wave field . a complication not accounted for in the preceeding discussion is the presence of caustics . at caustics geometric amplitudes ( [ gampl ] ) diverge and diffractive corrections must be applied . at short range ( on the order of the first focal distance a few tens of km in deep ocean environments ) we expect that intensity fluctuations will be dominated by diffractive effects . the entire wavefield should be organized by certain high - order caustics which leads to a pdf of wavefield intensity with long tails @xcite . in spite of the importance of diffractive effects at short range and probably also at very long range we believe that , in the transitional regime described above , intensity fluctuations are not dominated by diffractive effects . this somewhat counterintuitive behavior can be understood by noting that in the vicinity of caustics the importance of diffractive corrections to ( [ gampl ] ) decreases as the curvature of the caustic increases . under chaotic conditions the curvature of caustics increases , on average , with increasing range , so the fraction of the total number of multipaths that require caustic corrections decreases , on average , with increasing range . this is true even as the number of caustics grows exponentially , on average , in range . this argument leads to the somewhat paradoxical conclusion that , prior to saturation at least , we expect that the importance of caustic corrections decreases with increasing range . we turn our attention now to the late unresolved aet arrivals , corresponding to the near - axial rays . here , time spreads are sufficiently large that micromultipaths with different ray identifiers are not temporally resolved , i.e. , are not separated in time by more than @xmath179 . at each @xmath180 in the tail of the arrival pattern the wavefield can be modelled as a superposition of micromultipath contributions with random phases . the quadrature components of the wavefield have the form of sums of terms of the form @xmath181 where @xmath182 is a random variable uniformly distributed on @xmath183 . the distribution of @xmath184 is close to lognormal , but a correction must be applied to account for pulse shape as many of the interfering micromultipaths partially overlap in time . this correction is unimportant inasmuch as the central limit theorem guarantees that , provided the distributions of @xmath185 and @xmath186 have finite moments , the distributions of sums of @xmath185 and @xmath186 converge to zero mean gaussians . thus , wavefield intensity is expected to have an exponential distribution , consistent with the observations . the comments made earlier about caustics apply here as well . the question of what causes the transition from the structured early portion of the aet arrival pattern to the unstructured finale region deserves further discussion . recall that the early resolved arrivals have small time spreads and peak intensities that are near - lognormally distributed , while the finale region is characterized by unresolved arrivals and near - exponentially distributed intensities . in both regions the time spreads and intensity statistics are consistent with each other inasmuch as in our simulations a near - lognormal intensity distribution is obtained only when there is a preferred phase , while the exponential distribution is generated when phases are random , i.e. , when phases are uniformly distributed on @xmath183 . with these comments in mind , it is evident that the most important factor in causing the transition to the finale region is the increase in internal - wave - scattering - induced time spreads as rays become less steep ; as time spreads increase , neighboring timefront branches blend together and the phases of interfering micromultipaths get randomized . the trend toward increasing time spreads as rays become less steep is evident in fig . the surprising result is that the scattering - induced time spreads of the steep arrivals are so small ; we have seen that this can be explained by making use of the apex approximation . as noted at the end of the previous section , we do not fully understand the cause of the trend toward larger time spreads as rays flatten . in the finale region internal - wave - scattering - induced time spreads exceed the time difference between neighboring timefront branches that would have been observed in the absence of internal waves . 1 shows that these time gaps decrease as rays become increasingly flat . indeed , this figure shows that even in the absence of internal waves there would not have been any resolvable timefront branches in the last half second or so of the arrival pattern . some loss of temporal resolution in the measurements is of course due to the finite source bandwidth , but without internal waves ( or some other type of ocean fluctuations ) wavefield phases would not be random ; instead a stable interference pattern would be observed . finally , we note that an interesting feature of our ray simulations is that the near - axial rays have much higher stability exponents ( typically about @xmath187 ) than the steeper rays ( typically about @xmath188 ) ; see fig . 5 . also , note that figs . 6 and 7 strongly suggest that the near - axial rays in the aet environment are much more chaotic than the steeper rays . the relative lack of stability of the near - axial rays in the aet environment is , we believe , caused by the background sound speed structure . this topic will be discussed in detail elsewhere . we have chosen not to focus on this topic here becuse the measured intensity statistics in the aet finale region are not very sensitive to the near - axial ray intensity pdf ; as noted above , the argument leading to the expectation that wavefield intensity in the finale region should have an exponential distribution holds for a very large class of ray intensity distributions . finally , we note that an interesting feature of our ray simulations is that the near - axial rays have much higher stability exponents ( typically about @xmath187 ) than the steeper rays ( typically about @xmath188 ) ; see fig . 5 . also , note that figs . 6 and 7 strongly suggest that the near - axial rays in the aet environment are much more chaotic than the steeper rays . the relative lack of stability of the near - axial rays in the aet environment is , we believe , caused by the background sound speed structure . this topic is discussed in detail in ref . @xcite ; there it is shown that ray stability is largely controlled by a property of the background ( which is assumed to be range - independent ) sound speed profile , @xmath189 . large values of @xmath190 are associated with ray instability . @xmath191 vs axial ray angle in the range - averaged aet environment , and in five 650 km block range - averaged sections of the aet environment , are shown in fig ( the choice of averaging over 650 km blocks in range is , of course , arbitrary . range averaging should be done locally , however , because the local sound speed structure may be very different than that which results after averaging over the entire propagation path . ) the relatively large near - axial ray values of @xmath190 seen in this figure are consistent with the strongly chaotic nature of these rays seen in figs . 5 , 6 and 7 . unfortunately , the measured wavefield intensity statistics in the aet finale region are not very sensitive to the near - axial ray intensity pdf ; as noted above , the argument leading to the expectation that wavefield intensity in the finale region should have an exponential distribution holds for a very large class of ray intensity distributions . thus , we are not aware of any way that the aet measurements can be used to test our finding that the near - axial rays have larger stability exponents , on average , than the steeper rays . in summary , we believe that the observed near - lognormal pdf of wavefield intensity for the early resolved aet arrivals is transitional between fluctuations dominated by caustics at short range and saturation at long range , where phase differences will be larger . surprisingly , for the steep ray aet arrivals phase differences among the steep ray aet arrivals are very small so that saturation has not been reached . the late - arriving aet energy , on the other hand , is characterized by interfering micromultipaths with random phases , leading to an exponential pdf of wavefield intensity . here , the underlying near - lognormal pdf of ray intensity is obscured . we believe that diffractive effects do not control the intensity fluctuations of either the early or late aet arrivals . although our theoretical understanding of many of the issues raised in this section is clearly incomplete , it is worth emphasizing that our ray - based numerical simulations of intensity statistics , in which ray trajectories are predominantly chaotic , are in good qualitative agreement with the aet measurements . we do not believe that this agreement is accidental . in this paper we have seen that in the aet environment , including internal - wave - induced sound speed perturbations , ray trajectories are predominantly chaotic . in spite of extensive ray chaos , many features of ray - based wavefield predictions were shown to be both stable and in good agreement with the aet measurements . predicted and measured spreads of acoustic energy in time , depth and angle were shown to be in good agreement . it was shown that associated with the chaotic motion of ray trajectories is extensive micromultipathing , and that the micromultipathing process is highly nonlocal ; the many micromultipaths that add at the receiver to produce what appears to be a single arrival may sample the ocean very differently . it was shown , surprisingly , that on the early arrival branches the highly nonlocal micromultipathing process causes only very small time spreads and does not lead to a mixing of ray identifiers . a quantitative explanation for the cause of the very small time spreads of the early ray arrivals was presented . partially heuristic explanations for the near - lognormal and exponential distributions of wavefield intensities for the early and late arrivals , respectively , were provided . the fact that one is able to make apparently robust and accurate predictions of many wavefield features using ray methods under conditions in which ray trajectories are predominantly chaotic may surprise some readers . chaotic motion is , after all , intimately linked with unpredictability . this apparent paradox is reconciled by noting that while individual chaotic ray trajectories are unpredictable beyond some short predictability horizon , distributions of chaotic ray trajectories may be quite stable and have robust properties @xcite . consider , for example , the motion of the ray whose initial conditions are @xmath192 m , @xmath193 , in the aet environment including a known internal - wave - induced sound speed perturbation . at @xmath194 km , the depth @xmath3 , angle @xmath20 , travel time @xmath12 , and even the number of turns @xmath82 made by this ray are likely to be , for all practical purposes , unpredicatble . in contrast , the distribution in @xmath195 of a large ensemble of rays with @xmath192 m , @xmath196 is very stable in the sense that essentially the same distribution is seen for any large ensemble of randomly chosen ( with uniform probability ) rays inside this initial angular band . it is the latter property that allows us to make meaningful predictions using an ensemble of chaotic rays in a particular realization of the internal wave field . in addition , ray distributions with similar statistical properties are observed using different realizations of the internal wave field , suggesting that these statistical properties of rays are robust . our exploitation of results that relate to ray dynamics , including ray chaos , leads to a blurring of the distinction that is traditionally made between deterministic and stochastic wave propagation problems . the ray dynamics approach emphasizes the distinction between integrable and nonintegrable ray systems , corresponding to range - independent and range - dependent environments , respectively . in generic range - dependent environments at least some ray trajectories will exhibit chaotic motion . the oceanographic origin mesoscale variability , internal waves or something else of the range - dependent structure is not critical . an important conceptual insight that can come only from exploitation of results relating to ray chaos is that generically phase space is partitioned into chaotic and nonchaotic regions . this mixture contributes to a combination of limited determinism and constrained stochasticity . for those who wish to exploit elements of determinism in the propagation physics for the purpose of performing deterministic tomographic inverses , the results that we have presented have important implications . we have seen , for example , that , in spite of extensive ray chaos , many ray - based wavefield descriptors are stable and predictable , and should be invertible . a less encouraging but important observation is that although the travel times of the steep early arrivals are stable and can be inverted , the nonlocality of the micromultipathing process limits one s ability to invert for range - dependent ocean structure . for those who wish to understand and predict wavefield statistics , our results also have important implications . these comments are based on our analysis of the aet measurements , but are expected to apply to a large class of long - range propagation problems . first , we have seen that stable and unstable ray trajectories coexist and that this influences wavefield intensity statistics . second , we have seen that micromultipathing is extensive and highly nonlocal . the strongly nonlocal nature of the micromultipathing process is significant because : a ) this process can not be modelled using a perturbation analysis which assumes the existance of an isolated background ray path ; and b ) the effects of this process will not , in general , be eliminated as a result of local smoothing by finite frequency effects . third , we have seen that the appearance of stochastic effects is not entirely due to internal waves . we have seen , for example , that , even in the absence of internal waves , near - axial rays in the aet environment are chaotic . recall that we noted in the introduction that part of the motivation for the present study was to understand the reasons underlying the finding of colosi et al . @xcite that the theory described in ref . @xcite fails to correctly predict the aet wavefield statistics . this failure is not surprising in view of the observation that this theory is based on assumptions that either explicitly violate , or lead to the violation of , all three of the aforementioned properties of the scattered wavefield . note also in this regard that the combination of extensive micromultipathing , small time spreads , and lognormally distributed intensities that characterizes the early aet arrivals is not consistent with any of the three propagation regimes identified in this theory . although we have argued that a ray - based wavefield description which relies heavily on results relating to ray chaos can account for all of the important features of the aet measurements , it should be emphasized that our analysis has some shortcomings . recall that our simulations using @xmath61 resulted in steep arrival time spreads and peak intensity statistics in good agreement with measurements , but that simulations using @xmath62 did not , although direct measurements favor @xmath62 . we have not provided a rigorous argument to establish the connection between a near - lognormal ray intensity pdf and a near - lognormal wavefield intensity pdf . indeed , a complete theory of wavefield statistics which accounts for complexities associated with ray chaos is lacking . this is closely linked to the subject of wave chaos @xcite , which is currently not well understood . also , we have only briefly discussed the spreads of energy in depth and angle , and more work needs to be done on quantifying time spreads . finally , we wish to remark that the importance of ray methods is not diminished by recent advances , both theoretical and computational , in the development of full wave models , such as those based on parabolic equations . the latter are indispensible computational tools in many applications . in contrast , the principal virtue of ray methods is that they provide insight into the underlying wave propagation physics that is difficult , if not impossible , to obtain by any other means . the results presented in this paper illustrate this statement . we thank fred tappert and frank henyey for the benefit of discussions on many of the topics included in this paper , and the atoc group ( a. b. baggeroer , t. g. birdsall , c. clark , b. d. cornuelle , d. costa , b. d. dushaw , m. a. dzieciuch , a. m. g. forbes , b. m. howe , d. menemenlis , j. a. mercer , k. metzger , w. munk , r. c. spindel , p. f. worcester and c. wunsch ) for giving us access to the aet acoustic and environmental measurements . this work was supported by code 321 oa of the u.s . office of naval research . s. s. abdullaev and g. m. zaslavsky , `` stochastic instability of rays and the spekle structure of the field in inhomogeneous media , '' zh . * 87 * , 763 - 775 ( 1984 ) [ engl . jetp * 60 * , 435 - 441 ( 1985 ) ] . m. g. brown , f. d. tappert , g. goi and k. b. smith , `` chaos in underwater acoustics , '' in _ ocean variability and acoustic propagation _ , edited by j. potter and a. warn - varnas ( kluwer academic , dordrect , 1991 ) , 139 - 160 . m. g. brown , `` phase space structure and fractal trajectories in 1 1/2 degree of freedom hamiltonian systems whose time dependence is quasiperiodic , '' nonlinear processes in geophys . * 5 * , 69 - 74 ( 1998 ) . f. worcester , b. d. cornuelle , m. a. dzieciuch , w. h. munk , b. m. howe , j. a. mercer , r. c. spindel , j. a. colosi , k. metzger , t. birdsall and a. b. baggeroer , `` a test of basin - scale acoustic thermometry using a large- aperture vertical array at 3250-km range in the eastern north pacific ocean , '' j. acoust . 105 * , 3185 - 3201 ( 1999 ) . j. a. colosi , e. k. scheer , s. m. flatt , b. d. cornuelle , m. a. dzieciuch , w. h. munk , p. f. worcester , b. m. howe , j. a. mercer , r. c. spindel , k. metzger , t. birdsall and a. b. baggeroer , `` comparisons of measured and predicted acoustic fluctuations for a 3250-km propagation experiment in the eastern north pacific ocean , '' j. acoust . * 105 * , 3202 - 3218 ( 1999 ) . j. a. colosi , f. tappert and m. dzieciuch , `` further analysis of intensity fluctuations from a 3252-km acoustic propagation experiment in the eastern north pacific ocean , '' j. acoust . 110 * , 163 - 169 ( 2000 ) . p. f. worcester , b. d. cornuelle , j. a. hildebrand , w. s. hodgkiss , jr . , t. f. duda , j. boyd , b. m. howe , j. a. mercer and r. c. spindel , `` a comparison of measured and predicted broadband acoustic arrival patterns in travel time - depth coordinates at 1000 km range , '' j. acoust . 95 * , 3118 - 3128 ( 1994 ) . t. f. duda , s. m. flatt , j. a. colosi , b. d. cornuelle , j. a. hildebrand , w. s. hodgkiss , jr . , p. f. worcester , b. m. howe , j. a. mercer and r. c. spindel , `` measured wavefront fluctuations in 1000-km pulse propagation in the pacific ocean , '' j. acoust . 92 * , 939 - 955 ( 1992 ) .

the main purpose of this paper is to unify the theory of actions of hopf algebras , weak hopf algebras and multiplier hopf algebras to one of actions of weak multiplier hopf algebras introduced by a. van daele and s. h. wang . using such developed actions , we will define the notion of a module algebra over weak multiplier hopf algebras and construct their smash products . the main result is the duality theorem for actions and their dual actions on the smash product of weak multiplier hopf algebras . as an application , we recover the main results found in the literature for weak hopf algebras , multiplier hopf algebras and groupoids .

1701.02862

it is well known that the most well - known examples of hopf algebras ( @xcite ) are the linear spans of ( arbitrary ) groups . dually , also the vector space of linear functionals on a finite group carries the structure of a hopf algebra . in the case of infinite groups , however , the vector space of linear functionals ( with finite support ) possesses no unit . consequently , it is no longer a hopf algebra but , more generally , a multiplier hopf algebra in @xcite and @xcite . considering finite groupoids , both their linear spans and the dual vector spaces of linear functionals carry weak hopf algebra structures in @xcite and @xcite . finally , removing the finiteness constraint in this situation , both the linear spans of arbitrary groupoids and the vector spaces of linear functionals with finite support on them are examples of weak multiplier hopf algebras as introduced in @xcite and @xcite ( also see @xcite @xcite @xcite and @xcite for more references ) . especially , multiplier hopf algebras and weak hopf algebras are the special examples of weak multiplier hopf algebras . in the classical hopf algebra action theory : let @xmath0 be a finite hopf algebra and @xmath1 be a left @xmath0-module algebra . then the smash product @xmath2 is a left @xmath3-module algebra via a natural way : @xmath4 , where @xmath5 for all @xmath6 . furthermore , it follows from @xcite that there is an isomorphism @xmath7 , where @xmath8 and @xmath9 is an algebra of @xmath10-by-@xmath10 matrices over @xmath1 . for weak hopf algebras this result was proved in @xcite ( also see @xcite ) . for groupoids the result was studied in @xcite . for multiplier hopf algebras with integrals it was established in @xcite ( where infinite dimensional case was considered ) . in this paper , we will show that this result can be extended to weak multiplier hopf algebras with integrals in some form . the paper is organized as follows . first we recall some definitions and propositions related to weak multiplier hopf algebras in section [ two ] . in section [ three ] we define the notion of an action of a weak multiplier hopf algebra @xmath1 on an algebra @xmath11 and study some examples . in section [ four ] , we discuss the smash product @xmath12 for a left @xmath1-module algebra @xmath11 . in section [ five ] we consider the pairing between a weak multiplier hopf algebra @xmath1 with integrals and its dual @xmath13 . the main result is proposition [ 5.5 ] and [ 5.6 ] . these results will be basic for the theory in the next section . in section [ six ] we will get the main results of this paper . we first get some properties of bi - smash products . then we apply them to any dual pair of algebraic quantum groupoids . finally we obtain the main duality theorem in theorem [ 6.4 ] . in the paper , we always consider the non - unital associative algebra @xmath1 over @xmath14 with non - degenerate product . by this one means that : given @xmath15 we have @xmath16 if either @xmath17 for all @xmath18 or @xmath19 for all @xmath18 . we will also require the algebra @xmath1 to be idempotent ( i.e. @xmath20 ) . clearly , also the algebra @xmath21 on the same vector space @xmath1 with the opposite multiplication is idempotent and non - degenerate , whenever @xmath1 is so . for any algebra @xmath1 , recall from @xcite that a left multiplier of @xmath1 is a linear map @xmath22 such that @xmath23 for all @xmath24 . a right multiplier of @xmath1 is a linear map @xmath25 such that @xmath26 for all @xmath24 . a multiplier of @xmath1 is a pair @xmath27 of a left and a right multiplier such that @xmath28 for all @xmath24 . we denote by @xmath29 , @xmath30 , and @xmath31 the left , right , and multipliers of @xmath1 . it is clear that the composition of maps makes these vector spaces into algebras . if @xmath1 has an identity then @xmath32 ( cf.@xcite ) . we will use @xmath33 for the identity in @xmath31 and we will use @xmath34 to denote the identity map of each vector space . and if @xmath35 is another non - unital non - degenerate algebra , we have the following natural embeddings @xmath36 . in this section we will recall the definition and some properties of weak multiplier hopf algebras . recall from the definition 1.1 in @xcite that a _ coproduct _ on an algebra @xmath1 is a homomorphism @xmath37 such that @xmath38 and @xmath39 are in @xmath40 for all @xmath41 , @xmath42 is coassociative in the sense that @xmath43 for all @xmath44 . the coproduct @xmath42 is called _ full _ if the smallest subspaces @xmath45 and @xmath46 of @xmath1 satisfying @xmath47 are equal to @xmath1 . the coproduct @xmath48 is called _ regular _ if @xmath49 and @xmath50 are in @xmath40 for all @xmath41 . for a coproduct @xmath42 on @xmath1 , the _ canonical maps _ @xmath51 and @xmath52 from @xmath40 to @xmath40 are defined by @xmath53 if @xmath42 is regular , we define @xmath54 and @xmath55 on @xmath40 by @xmath56 recall from the definition 1.8 of @xcite that a linear map @xmath57 is called a _ counit _ on an algebra @xmath1 with a coproduct @xmath48 if @xmath58 for all @xmath41 . for any vector space @xmath1 , denote by @xmath59 the flip map @xmath60 . for a non - unital algebra @xmath1 with a non - degenerate multiplication , and for a coproduct @xmath37 , define a multiplicative linear map @xmath61 via @xmath62 and define @xmath63 by @xmath64 and @xmath65 recall from the definition 1.14 of @xcite that a _ weak multiplier hopf algebra _ is a pair @xmath66 of a non - degenerate idempotent algebra @xmath1 with a full coproduct @xmath48 and a counit satisfying the following conditions : there exists an idempotent @xmath67 giving the ranges of the canonical maps : @xmath68 the element @xmath69 satisfies @xmath70 the kernels of the canonical maps are of the form @xmath71 where @xmath72 and @xmath73 are the linear maps from @xmath74 to itself , given as @xmath75 and @xmath76 for all @xmath44 . a weak multiplier hopf algebra is called _ regular _ if the coproduct is regular and if also @xmath77 is a weak multiplier hopf algebra . this is the same as requiring that also @xmath78 is a weak multiplier hopf algebra . \i ) let @xmath79 be a regular weak multiplier hopf algebra . as the same as in a regular multiplier hopf algebra case ( see proposition 2.2 in @xcite ) , we know from proposition 2.21 in @xcite that @xmath1 has local units . more precisely , let @xmath80 be elements in @xmath1 . then there exist elements @xmath81 in @xmath1 such that @xmath82 , and @xmath83 , for all @xmath84 . we would like to use a formal expression for @xmath85 when @xmath86 . the problem is that @xmath85 is not in @xmath87 in general . we know however that @xmath88 for all @xmath41 . we will use the sweedler notation and write @xmath89 for any @xmath15 . then we know that @xmath88 for all @xmath90 and we will write @xmath91 . then we say that @xmath92 is _ covered _ by @xmath93 in this equation . now , we know that there is an element @xmath94 such that @xmath95 and we can think of @xmath96 to stand for @xmath97 . of course , this is still dependent on @xmath93 . but we know that for several elements @xmath93 , we can use the same @xmath98 . note that it should be careful when one uses the sweedler notation in weak multiplier hopf algebras everywhere . \ii ) if the element @xmath69 exists and satisfies the first condition , we can show that the coproduct @xmath42 has a unique extension to a homomorphism @xmath99 . we denote the extension still by @xmath42 . in a similar way , we can extend @xmath100 and @xmath101 to the homomorphisms from @xmath102 to @xmath103 . so we can give a meaning to the formulas @xmath104 and @xmath105 . we will call @xmath69 the _ canonical idempotent_. it is uniquely determined and it satisfies @xmath106 for all @xmath15 . there is a unique antipode @xmath107 from @xmath1 to @xmath31 . recall from proposition 3.5 and 3.7 in @xcite that the antipode @xmath107 is an anti - algebra and an anti - coalgebra map . moreover the antipode satisfies @xmath108 and @xmath109 for all @xmath15 . if @xmath1 is regular then the antipode is a bijective map from @xmath1 to itself ( see definition 4.5 in @xcite ) . let @xmath110 be a regular weak multiplier hopf algebra . for any @xmath15 , we have the following four linear maps @xmath111 and @xmath112 from @xmath1 to @xmath31 ( see proposition 2.1 and remark 2.22 i ) in @xcite ) : @xmath113 we will call @xmath114 the _ source algebra _ and @xmath115 the _ target algebra _ as in @xcite . they can be identified resp . with the left and the right leg of @xmath69 . we have @xmath116 and @xmath117 where @xmath90 . in the regular case , they embed in @xmath31 in such a way that their multiplier algebras @xmath118 and @xmath119 still embed in @xmath31 . these multiplier algebras are denoted by @xmath120 and @xmath121 resp . they are still commuting subalgebras of @xmath31 . for a regular weak multiplier hopf algebra , the multiplier algebras of the source and target algebras satisfy @xmath122 for element @xmath123 and @xmath124 , we have @xmath125 and @xmath126 for all @xmath15 , we also have from proposition 2.7 in @xcite that @xmath127 and @xmath128 . we also list some formulas here . for any regular weak multiplier hopf algebra @xmath1 , we have ( see @xcite ) : @xmath129 @xmath130 @xmath131 for all @xmath15 . for any @xmath132 , we have : @xmath133 we now make an important remark about the covering of the previous formulas ( see @xcite ) . \i ) first rewrite the ( images of the ) canonical maps @xmath51 and @xmath52 , and of @xmath54 and @xmath55 in the regular case , using the sweedler notation , as @xmath134 and @xmath135 where @xmath136 . in all these four expressions , either @xmath137 is covered by @xmath138 and or @xmath92 by @xmath93 . this is by the assumption put on the coproduct , requiring that the canonical maps have range in @xmath40 . if we first apply @xmath107 in the first or the second factor of the expressions in the formulas ( 2.10 ) and then multiply , we get the two elements @xmath139 where @xmath136 . this is used to define the source and target maps above . \ii ) next consider the expressions @xmath140 and @xmath141 where @xmath136 . in the first two formulas , we have a covering by the assumption that the generalized inverses @xmath142 and @xmath143 of the canonical maps exist as maps on @xmath40 with range in @xmath40 ( see @xcite ) . in the second pair of formulas , we have a good covering only in the regular case . if we simply apply multiplication on the expressions in the formulas ( 2.11 ) , we get the two elements @xmath144 where @xmath136 . this is also used to define the source and target maps above . \iii ) now , we combine the coverings obtained in the part i ) and the part ii ) . consider e.g. the two expressions : @xmath145 where @xmath136 . the first expression above is obtained by applying the canonical map @xmath51 to the first of the two expressions in ( 2.11 ) . so this gives an element in @xmath40 and we know that it is @xmath146 as we can see from the formula ( 2.6 ) above . similarly , the second expression above is obtained by applying the canonical map @xmath52 to the second of the two expressions in ( 2.11 ) . we know that this is @xmath147 as we see from the formula ( 2.7 ) above . remark that @xmath148 and @xmath147 belong to @xmath40 because by assumption @xmath149 , but that on the other hand , it is not obvious ( as we see from the above arguments ) that the expressions that we obtain for these elements belong to @xmath40 . \iv ) finally , as a consequence of the above statements , also the four expressions @xmath150 are well - defined in @xmath151 for all @xmath136 . this justifies a statement made earlier about the properties of the antipode ( see @xcite ) . and once again , in all these cases , the sweedler notation is just used as a more transparent way to denote expressions . we refer to the coverings just to indicate how the formulas with the sweedler notation can be rewritten without the use of it . in this section , we fix a weak multiplier hopf algebra @xmath152 . we do not assume that it is regular . we know that @xmath1 still admits local units in the non - regular case , see proposition 2.21 in @xcite . by a left @xmath1-module we mean a vector space @xmath11 equipped with a bilinear map @xmath153 satisfying @xmath154 for all @xmath155 and @xmath41 . it is called _ unital _ if @xmath156 . the module is called non - degenerate if @xmath155 and @xmath157 for all @xmath15 , implies @xmath158 . a unital module is automatically non - degenerate because we have local units . in this case , one can show the following . if @xmath11 is a unital left @xmath1-module , then it is non - degenerate . let us give an important remark about the unital module which is from section 3 in @xcite . we just copy it here because it will help the reader to understand the sweedler notation or the covering . let @xmath11 be a unital @xmath1-module . since @xmath1 has local units , then there exists an @xmath159 such that @xmath160 for any @xmath161 . moreover , for all @xmath162 , we have an element @xmath98 such that @xmath163 for all @xmath84 . it means that elements in a unital @xmath1-module will cover elements @xmath164 . next we show that unital @xmath1-modules can be extended to modules over @xmath31 . this will help to explain the formulas in definition [ 3.2 ] . [ 3.1 ] let @xmath11 be a unital left @xmath1-module . then there is a unique extension to a left @xmath31-module with @xmath165 for all @xmath161 , here @xmath166 . for the proof we refer to proposition 3.3 in @xcite . note that if @xmath1 does not have local units , we can also get the above result . we only need @xmath11 to be unital and non - degenerate . let @xmath167 . since @xmath11 is unital , we can write @xmath168 where @xmath169 and @xmath170 for all @xmath84 . then we can define the action of @xmath31 by @xmath171 . it is well - defined because the module is non - degenerate . if @xmath11 and @xmath107 are unital left @xmath1-modules , then we can make @xmath172 into a unital left @xmath173-module by @xmath174 . by the above theorem we can extend it to a module over the multiplier algebra @xmath102 . now we can consider the action of @xmath1 on @xmath172 by @xmath175 but the action is not unital any more since @xmath51 and @xmath52 are not surjective . fortunately we can show that the subspace @xmath176 is a non - degenerate and unital module under the action . this is because @xmath177 . [ 3.2 ] let @xmath11 be an algebra and @xmath11 a unital left @xmath1-module via @xmath178 for all @xmath179 . then @xmath11 is called a left @xmath1-module algebra if the following condition holds : @xmath180 for all @xmath181 . since @xmath11 is a unital left @xmath1-module we know that the elements @xmath182 and @xmath183 can be used to cover @xmath137 and @xmath92 . we can also explain this expression as @xmath184 where @xmath185 denotes multiplication in @xmath11 . [ 3.3 ] let @xmath11 be a left @xmath1-module algebra . the following properties hold for all @xmath186 and @xmath187 , @xmath188 and @xmath189 \(i ) note that @xmath190 , so the formula ( 3.3 ) is meaningful . for any @xmath15 , @xmath191 by the non - degeneracy of the module , we get @xmath192 . \(ii ) remark that by theorem [ 3.1 ] we have a left @xmath31-module , so the formula ( 3.4 ) is meaningful . take any element @xmath183 in @xmath114 , for any @xmath15 we have @xmath193 because the module is non - degenerate we find @xmath194 . assume that @xmath11 has a unit @xmath195 and @xmath1 is regular . then we find @xmath196 in the fourth and fifth equalities @xmath92 is covered by @xmath195 . in the case of a multiplier hopf algebra , this result means @xmath197 , which is a true and known result . the above one generalizes this to weak multiplier hopf algebras . let @xmath1 be a multiplier hopf algebra and assume that @xmath11 is a left @xmath1-module algebra . in @xcite the authors have extended the action of @xmath1 on @xmath11 to the multiplier algebra @xmath198 . now we will generalize the theory to regular weak multiplier hopf algebras . let @xmath1 be a regular weak multiplier hopf algebra . for any @xmath15 and @xmath186 , we have @xmath199 and @xmath200 first remark that @xmath92 is covered by @xmath201 in the first formula and that @xmath137 is covered by @xmath202 . for the first formula we have @xmath203 for the second equality we have @xmath204 where we use the sweedler type notation : @xmath205 . now inspired by the proposition above we can extend the action of @xmath1 from @xmath11 to @xmath198 . let @xmath1 be a regular weak multiplier hopf algebras and assume that @xmath11 is a left @xmath1-module algebra . then we can extend this action of @xmath1 from @xmath11 to @xmath198 . also @xmath206 for any @xmath15 . let @xmath207 , we will define @xmath208 by @xmath209 @xmath210 as in the above proposition , we have well coverings for the two expression . next we first show that the action is well - defined , it means that we need to prove that @xmath211 is actually a multiplier in @xmath198 . for this , let @xmath212 , we have @xmath213 on the other hand we have @xmath214 next we will show that @xmath198 is a left @xmath1-module . let @xmath215 . then for any @xmath155 , we have @xmath216 if @xmath217 , we have @xmath218 so we have @xmath206 for any @xmath15 . so far we know that we can make @xmath198 into a left @xmath1-module , but we can not make sure it is still a module algebra since @xmath198 is not unital any more . however we do know that the action is non - degenerate . if @xmath219 and @xmath220 for all @xmath221 , by the definition we have @xmath222 for all @xmath161 . we can write @xmath182 as @xmath223(@xmath224 ) and use the fullness of @xmath42 , then we can get the non - degeneracy . now let us treat some examples and special cases . let @xmath1 be a weak multiplier hopf algebra . then @xmath115 is an @xmath1-module algebra with the module action @xmath225 for all @xmath90 . it is easy to show that @xmath115 is an @xmath1-module . since @xmath1 is idempotent , we know that @xmath115 is a unital @xmath1-module . take @xmath226 , we have @xmath227 on the other hand , @xmath228 in the first equality @xmath229 is covered by @xmath183 . the third equality follows by the commutative of the two base algebras @xmath114 and @xmath115 . so @xmath115 is an @xmath1-module algebra . moreover we can check the formula ( 3.3 ) in proposition [ 3.3 ] . for any @xmath230 we have @xmath231 in the second equality @xmath137 is covered by @xmath93 . the fourth one follows because @xmath115 and @xmath114 are commuting algebras . for the second formula ( 3.4 ) , we have @xmath232 the first equality follows by @xmath233 . the third one follows by @xmath234 . now let us consider the module extension . from the above proposition and @xmath235 , we know that we can extend the action of @xmath1 from @xmath115 to @xmath121 . for any @xmath236 , we have @xmath237 where we use the sweedler type notation : @xmath205 . in the third equality we have a covering by multiplying any element in @xmath1 from the left or right . the last equality follows by the commutative of the two base algebras and the fact that the restriction of @xmath107 is an anti - isomorphism from @xmath115 to @xmath114(see proposition 2.16 in @xcite ) . from this formula we can get the non - degeneracy of the extended module . in fact , if @xmath1 is a weak hopf algebra , the formula means @xmath165 . let @xmath238 be any groupoid and consider the groupoid algebra @xmath239 . it is the space of complex functions with finite support on @xmath238 with the convolution product . denote the canonical embedding of @xmath238 in @xmath239 by @xmath240 . when @xmath241 is defined we have @xmath242 , otherwise the product is 0 . the coproduct on @xmath1 is given by @xmath243 and the counit is given by @xmath244 for all @xmath245 . the canonical idempotent @xmath69 in @xmath102 is defined as @xmath246 where the sum is taken over all units . let @xmath247 be an action of @xmath238 on an set @xmath248 . it means that for every @xmath249 there is a subset @xmath250 of @xmath248 and a map @xmath251 such that : if @xmath241 exists then @xmath252 , @xmath253 and @xmath254 for any @xmath255 , if @xmath98 is a unit in @xmath238 then @xmath256 for all @xmath257 . we also assume that the action is true . the action is called true if @xmath258 and @xmath259 imply @xmath241 is defined . for more information about the notion of an action of groupoid on a set we refer to @xcite . by the definition we have @xmath260 , moreover @xmath261 where the union is taken over the set of units . let @xmath11 be the algebra @xmath262 of complex functions with finite support on @xmath248 and pointwise product . for each @xmath249 , @xmath263 is a map from @xmath11 to @xmath11 which is defined as @xmath264 for any @xmath265 . if @xmath241 is defined , we have @xmath266 when @xmath267 , we have @xmath268 . so @xmath269 when @xmath270 , we have @xmath271 if @xmath241 is not defined and @xmath267 . since the action is true we get @xmath272 . so we can also get @xmath273 . so we finally get @xmath274 now we can associate an action of @xmath1 on @xmath11 by @xmath275 for @xmath276 . next we show that @xmath11 is an @xmath1-module algebra . it is easy to see that @xmath11 is an @xmath1-module . let @xmath98 be a unit in @xmath238 , we have @xmath277 if we take @xmath278 with support in @xmath279 , then @xmath280 . so the action is unital as @xmath261 . finally , for any @xmath281 , @xmath282 on the other hand , @xmath283 so @xmath11 is a @xmath239-module algebra . now we consider a regular weak multiplier hopf algebra @xmath1 . take @xmath15 and define a map @xmath284 by @xmath285 where @xmath286 . observe that @xmath137 is covered by @xmath182 . in what follows , we will still use the sweedler type notation : @xmath205 . let @xmath287 since we have @xmath288 and @xmath289 , so @xmath290 note that @xmath291 ( see remark 2.22 in @xcite ) . recall from proposition 2.16 in @xcite that we have @xmath292 define a linear map @xmath293 . by definition @xmath294 is surjective . @xmath295 is a subalgebra of @xmath1 . take @xmath41 . we write @xmath296 and @xmath297 where we use two copies @xmath298 and @xmath299 of @xmath69 . then we have @xmath300 in the third equality we use @xmath301 where @xmath183 is in the source algebra @xmath302 . note that we have everything well - covered here . in the first equality @xmath303 is covered by @xmath221 and @xmath304 is covered by @xmath93 . in the third equality the element @xmath221 covers @xmath303 and @xmath93 covers @xmath305 . [ 3.4 ] @xmath306 note that @xmath307 , so one gets @xmath308 let @xmath182 be the local unit of @xmath1 . then we have @xmath309 the left - hand side is @xmath310 and because @xmath311 we find that @xmath312 . similarly for the other equality . [ 3.5 ] the product in @xmath295 is non - degenerate . let @xmath313 and assume that @xmath314 for all @xmath315 , then @xmath316 for all @xmath317 . it means that @xmath318 for all @xmath286 , so @xmath319 . moreover , we have the following propositions . for any @xmath320 , we have @xmath321 . @xmath322 let @xmath15 and assume that @xmath221 commutes with @xmath114 . then @xmath323 and @xmath324 and @xmath325 for all @xmath18 . it is easy to prove it . so we have that @xmath294 is a conditional expectation of @xmath1 onto @xmath295 . now we can give a characterization of @xmath295 and @xmath326 . we have the following identities : @xmath327 and @xmath328 the first one is a consequence of proposition [ 3.4 ] and proposition [ 3.5 ] . now let us consider the multiplier algebra of @xmath295 . we first want to show the inclusion `` @xmath329 '' . so take @xmath330 and assume that @xmath187 . for any @xmath331 , we have @xmath332 if we multiply any element in @xmath151 from the left and use @xmath333 , then we get @xmath334 . so @xmath335 . next we will show the inclusion `` @xmath336 '' . take @xmath313 and @xmath187 , then @xmath337 it means that @xmath338 . similarly , we can get @xmath339 . hence @xmath340 . @xmath295 is an @xmath1-module algebra with the action @xmath341 , for any @xmath342 . obviously @xmath295 is a unital left @xmath1-module . for all @xmath343 , @xmath344 in the expression @xmath345 , we observe that @xmath137 is covered by @xmath182 . now let us consider the example associated with a separability idempotent which is studied in @xcite and @xcite . let @xmath35 and @xmath346 be non - degenerate algebras and assume that @xmath69 is a regular separability idempotent in @xmath347 . we call @xmath69 regular if it is a separability idempotent also when considered in @xmath348 . consider the algebra @xmath349 with the coproduct @xmath350 , where @xmath351 . then @xmath352 is a regular weak multiplier hopf algebra . the canonical idempotent @xmath353 is @xmath354 . let @xmath11 be a left @xmath355-module algebra with action denoted by @xmath356 . then @xmath11 can be regarded as a @xmath346-module and a @xmath35-module through the following actions @xmath357 for any @xmath358 . for any @xmath359 we also have @xmath360 and @xmath361 if we consider the extended @xmath362-module and the element @xmath363 in @xmath362 , then we get @xmath364 . similarly if we consider @xmath365 , then @xmath366 . so there exists a left multiplier @xmath367 in @xmath368 such that @xmath369 and a right multiplier @xmath370 in @xmath371 such that @xmath372 . note that @xmath373 and @xmath374 , so @xmath375 for any @xmath376 . we can rewrite it as @xmath377 . then @xmath370 is equal to @xmath378 as multipliers . now , we are ready to give the following proposition . as above , let @xmath11 be a left @xmath355-module algebra . for any @xmath358 , there exists a non - degenerate homomorphism @xmath379 such that @xmath380 , @xmath381 , @xmath382 finally let us consider the examples that come from a dual pair of regular weak multiplier hopf algebras . we will give the definition of a weak multiplier hopf algebra pairing . our idea is coming from @xcite . in fact the definition is very similar to the case of multiplier hopf algebras ( see definition 2.1 and definition 2.8 in @xcite ) . let @xmath79 be a weak multiplier hopf algebra . recall from @xcite or @xcite that a linear functional @xmath383 is called _ left invariant _ if @xmath384 for all @xmath86 . a non - zero left invariant functional is called a _ left integral _ on @xmath1 ; similarly , a linear functional @xmath385 on @xmath1 is called _ right invariant _ if @xmath386 for all @xmath86 . a non - zero right invariant functional is called a _ right integral _ on @xmath1 . recall that @xmath121 and @xmath120 are defined as subspaces of @xmath31 and so the above definition makes sense . let @xmath1 and @xmath35 be two regular weak multiplier hopf algebras with enough integrals . define two linear functions @xmath387 and @xmath388 where @xmath389 . a _ pre - pairing _ between @xmath1 and @xmath35 is a bilinear form @xmath390 from @xmath391 to @xmath14 satisfying the following : for all @xmath392 @xmath393 the pre - pairing is called non - degenerate if @xmath1 and @xmath35 are dual with respect to the bilinear form . since @xmath394 we will denote it as @xmath395 . similarly for other cases . remark that the sweedler notation here is just a notation , thus we can denote formulas in a more transparent way . for any pre - pairing we have the following four maps @xmath396 the weak multiplier hopf algebra pre - pairing @xmath397 is called a _ pairing _ if the four maps defined above are surjective and @xmath398 , for any @xmath389 . in multiplier hopf algebras theory , if one of the four maps is surjective then so do the others . we have similar results for weak multiplier hopf algebras . but the proof is not the same and it involves a long paragraph to explain . we will discuss it in a separate paper . so in the definition above we require these four maps to be surjective . we also need to give a meaning to the formula @xmath399 . we will explain it after the following proposition . and note that the formula is not involved in the following proposition . these four maps @xmath400 are actions , i.e. @xmath401 is a left @xmath1-module algebra and @xmath402 is a right @xmath1-module algebra . analogously @xmath403 is a left @xmath35-module algebra and @xmath404 is a right @xmath35-module algebra . let us check the map @xmath405 . the action will be denoted by @xmath356 . it is easy to show that @xmath35 is a left @xmath1-module . since the map is surjective we know that @xmath35 is unital . so next we have to show that @xmath406 for all @xmath407 . indeed , for any @xmath286 , @xmath408 since the module is unital so @xmath137 is covered by @xmath93 , then @xmath409 also be covered . the other cases are similar . since we have these four unital modules , the pairing on @xmath391 can be uniquely extended to @xmath410 in such a way that @xmath411 where @xmath412 . so we can give a meaning to the pairing of the form @xmath399 in @xmath410 . remark that we have to show that this is well - defined . similarly , the paring on @xmath391 can be extended to @xmath413 . so for any @xmath414 we have the following useful formula @xmath415 where @xmath416 . next we will give the definition of the dual space @xmath13 coming from definition 2.8 in @xcite . let @xmath1 be a regular weak multiplier hopf algebra with a faithful set of integrals . then we define @xmath13 as the space of linear functionals on @xmath1 spanned by the elements of the form @xmath417 where @xmath418 is a left integral of @xmath1 and @xmath15 . we say that @xmath1 has a faithful set of integrals if given an element @xmath286 we must have @xmath419 if @xmath420 for any left integral @xmath421 and element @xmath15 . similarly also if @xmath422 for any left integral @xmath421 and element @xmath15 , then @xmath419 . for more information about the integrals we also refer to @xcite . if @xmath1 is a regular weak multiplier hopf algebra with a faithful set of integrals we call it an _ algebraic quantum groupoid _ ( see definition 2.10 in @xcite ) . and the dual @xmath13 of an algebraic quantum groupoid @xmath1 is a regular weak multiplier hopf algebra ( see theorem 3.15 in @xcite ) . for the convenience of studying the natural pairing @xmath423 in section 5 and section 6 , we here list theorem 3.15 in @xcite as follows . let @xmath1 be an algebraic quantum groupoid . then the dual pair @xmath424 is a regular weak multiplier hopf algebra with the following structure : @xmath425 for all @xmath426 and @xmath427 . let @xmath1 has a left integral @xmath421 . then in the pairing @xmath423 , we note that @xmath428 means that @xmath429 where @xmath430 . in particular , if @xmath431 with @xmath286 , then @xmath432 . let @xmath1 be a regular weak multiplier hopf algebra and assume @xmath11 is a left @xmath1-module algebra . in this section we will use @xmath356 denote the action of @xmath1 on @xmath11 . define the product on @xmath433 by @xmath434 for any @xmath435 . here @xmath137 is covered by @xmath436 . then we have the following proposition . the product given by ( 4.1 ) is associative . for any @xmath437 and @xmath438 , @xmath439 so @xmath433 is an associative algebra with this product . we will use the notation @xmath12 to denote @xmath433 with the above product , and the elements @xmath440 will be denoted by @xmath441 . we know that @xmath12 can be considered as a @xmath102-module . so we can consider the space @xmath442 . the element @xmath443 in @xmath12 can be denoted by @xmath444 . the formula is well covered because @xmath11 is unital . for any @xmath445 . we have @xmath446 and @xmath447 for any @xmath445 , we have @xmath448 on the other hand , @xmath449 now let us check the final equation . @xmath450 this completes the proof . so @xmath442 is a subalgebra of @xmath451 , it is also a right ideal . now we will investigate the product in @xmath442 . [ 4.3 ] the product in @xmath442 is non - degenerate . for any @xmath452 in @xmath12 , assume that @xmath453 for all @xmath454 . then we get @xmath455 apply @xmath42 and multiply by @xmath317 , then we have @xmath456 where @xmath457 is covered by @xmath138 . since @xmath458 and @xmath459 , so @xmath460 for all @xmath461 . apply @xmath107 and replace @xmath436 by @xmath462 , this gives @xmath463 for all @xmath464 . now replace @xmath185 by @xmath465 then we get @xmath466 hence @xmath467 for all @xmath468 and @xmath359 . because @xmath11 is unital and @xmath1 is idempotent we can cancel @xmath469 and @xmath470 , then we obtain @xmath471 on the other hand , suppose that @xmath472 for all @xmath359 and @xmath18 , then @xmath473 multiply @xmath474 from the left and use the fact @xmath475 , we get @xmath476 as before , we can cancel @xmath477 to get again @xmath478 * notation . * for the algebra @xmath479 , we will use the notation @xmath480 instead and the element @xmath443 will be denoted by @xmath481 . and we call it the _ smash product algebra_. + we can also form the smash product in the following way . let @xmath1 be a regular weak multiplier hopf algebra and let @xmath11 be a left @xmath1-module algebra . then we have the following lemma . @xmath11 is a unital right @xmath115-module via @xmath482 for all @xmath483 . it is easy to know that @xmath11 is a right @xmath484-module . from proposition 2.10 in @xcite , @xmath1 can be regarded as a unital left @xmath485-module . because @xmath11 is unital , we have @xmath486 also by proposition 2.10 in @xcite , we know that @xmath1 is a unital left @xmath115-module . so we can define the tensor product @xmath487 . for concisely we will denote it by @xmath488 . let @xmath11 be a left @xmath1-module algebra . then the space @xmath488 is a non - degenerate algebra with the product @xmath489 for all @xmath490 and @xmath90 . remark that @xmath92 is covered by @xmath93 or @xmath436 can be used to cover @xmath137 . first we show that the multiplication above is well defined . it means we have to show @xmath491 for all @xmath492 we have @xmath493 the proof of the associativity of the product is straightforward . in a similar way as in proposition [ 4.3 ] we can get the non - degeneracy of the product . now we show that there is an isomorphism between @xmath488 and @xmath480 . we have that @xmath488 is isomorphic with @xmath480 . for any @xmath494 , define the map @xmath495 first we have to show that @xmath278 is well - defined . it means that we must get @xmath496 this is true since @xmath497 for any @xmath187 . obviously @xmath278 is bijective . finally it is easy to show that @xmath278 is an isomorphism . in this paper we will mainly consider the algebra @xmath480 . let us study more properties about the algebra . the product in @xmath480 is defined by the twist map @xmath498 where @xmath499 . we also know that @xmath500 is bijective and its inverse is given by @xmath501 so the product is given by @xmath502 , here @xmath503 and @xmath504 denote the multiplication in @xmath11 and @xmath1 , respectively . remark that for the last expression , @xmath137 is covered by @xmath202 . if @xmath11 has an identity @xmath195 , then we have @xmath505 it means that @xmath506 is a homomorphism of @xmath1 into @xmath480 . if @xmath1 has an identity , then @xmath1 is a weak hopf algebra with @xmath507(see proposition 4.12 in @xcite ) . and we have @xmath508 this equality also gives a homomorphism of @xmath11 into @xmath12 by @xmath509 . if @xmath1 is a regular multiplier hopf algebra , then it is the case which has been studied in @xcite . if @xmath11 and @xmath1 have identities , then it is case which has been studied in @xcite . and @xmath510 is the identity in @xmath511 . now let us consider the following useful homomorphisms which are appeared in proposition 5.7 in @xcite . we will generalize it to weak multiplier hopf algebras case . let @xmath1 and @xmath11 be as before . for all @xmath512 define @xmath513 by @xmath514 @xmath515 then @xmath516 is a unital algebra homomorphism . define @xmath517 by @xmath518 @xmath519 then @xmath520 is an algebra homomorphism . \(1 ) first we show that @xmath521 . for all @xmath186 and @xmath438 we have @xmath522 so @xmath516 is well defined . next we will show that @xmath516 is a homomorphism . for any @xmath523 , we have @xmath524 and @xmath525 since @xmath1 is idempotent and @xmath526 , we can get @xmath527 . then @xmath516 is a unital algebra homomorphism . \(2 ) the proof is similar . with the above notation . @xmath520 is not unital any more . but if @xmath528 , then @xmath520 is unital . for any @xmath186 and @xmath15 , we have @xmath519 because @xmath526 and @xmath11 is unital , so the right - hand side is the element of the form @xmath529 it is also equal to @xmath530 if @xmath528 , then the element of this form is actually equal to @xmath481 . [ 4.8 ] for all @xmath483 , we have @xmath531 and @xmath532 the proof is straightforward . it follows that @xmath533=@xmath534 , with this proposition we can extend @xmath520 from @xmath11 to @xmath198 . we also have the following universal property . let @xmath1 be a regular weak multiplier hopf algebra , assume that @xmath11 is an @xmath1-module algebra . let @xmath35 be an algebra over @xmath14 . if there exists homomorphisms @xmath535 and @xmath536 such that @xmath537 for all @xmath499 . then there is a homomorphism @xmath538 such that @xmath539 straightforward . now let us recall the definition of covariant module in @xcite . let @xmath1 be a regular weak multiplier hopf algebra and @xmath11 be a left @xmath1-module algebra . let @xmath45 be a vector space which is both a left @xmath1-module and a left @xmath11-module . we say that @xmath45 is a covariant @xmath1-@xmath11-module if @xmath540 for all @xmath541 . @xmath45 is called unital if it is unital both for @xmath1 and @xmath11 . and @xmath45 is called non - degenerate if it is non - degenerate both for @xmath1 and @xmath11 . note that if @xmath45 is unital , then @xmath45 is automatically a non - degenerate @xmath1-module . the following theorem gives the relation between the covariant modules and the smash product modules ( cf . the result is similar to the result for actions of locally compact groups on @xmath542-algebras(see section 7.6 in @xcite ) . and in @xcite the authors also obtain the result for multiplier hopf algebras . now we are going to show that it is also true for weak multiplier hopf algebras . assume that @xmath528 . then there is a one - to - one correspondence between unital @xmath480-modules and covariant @xmath11-@xmath1-modules with unital actions . and if one is non - degenerate , so is the other . suppose we have a left @xmath480-module @xmath45 which is unital . then we can extend @xmath45 to a @xmath543-module . with the homomorphisms @xmath516 and @xmath520 , we have actions of @xmath1 and @xmath11 on @xmath45 . with the formula in proposition [ 4.8 ] we have @xmath544 for any @xmath545 . here @xmath137 is covered by @xmath202 . so @xmath45 is a covariant @xmath1-@xmath11-module . because @xmath546 and @xmath547 we get @xmath45 is a unital covariant module . conversely , suppose that @xmath45 is a unital covariant @xmath11-@xmath1-module . for any @xmath545 , define the action of @xmath480 on @xmath45 by @xmath548 it is easy to show that @xmath45 is a left @xmath480-module . and since @xmath45 is unital , we have @xmath549 so @xmath45 is a unital left @xmath480-module . now let us discuss the non - degeneracy of these actions . suppose that @xmath45 is a unital non - degenerate @xmath480-module . take any @xmath550 and assume that @xmath551 for all @xmath155 . then @xmath552 for all @xmath483 . so we can get @xmath553 . it means that @xmath45 is a non - degenerate @xmath11-module . because @xmath45 is automatically a non - degenerate @xmath1-module , we get that @xmath45 is a unital non - degenerate covariant @xmath11-@xmath1-module . conversely , suppose that @xmath45 is a non - degenerate @xmath11-module . take any @xmath550 and assume that @xmath554 for all @xmath483 . then @xmath555 for all @xmath483 . because @xmath45 is a non - degenerate covariant @xmath11-@xmath1-module , it follows that @xmath553 . this means that @xmath45 is a non - degenerate @xmath480-module . in this section we will consider the dual pair of regular weak multiplier hopf algebras . for a dual pair @xmath556 of regular weak multiplier hopf algebras , we can define the smash product @xmath557 and @xmath558 . for the algebra @xmath558 we have the following faithful action . @xmath35 is a left @xmath559-module by @xmath560 for any @xmath561 . and the action is faithful . for all @xmath562 and @xmath563 , we have @xmath564 so @xmath35 is a left @xmath565-module . here @xmath566 is covered by @xmath567 . next we show that the action is faithful . let @xmath568 and assume that @xmath569 for all @xmath570 . then @xmath571 multiply by @xmath572 on the right we get that @xmath573 for any @xmath574 . so @xmath575 now we claim that @xmath576 for all @xmath389 . and if this is true , we finish the proof . indeed , for any @xmath577 , @xmath578 in the second equality and the fifth equality , we use @xmath579 note that everything is well covered here . for example , in the second equality @xmath580 is covered by @xmath581 . @xmath558 can be considered as the span of the elements @xmath582 in the algebra @xmath346 generated by @xmath1 and @xmath35 subject to the following commutation relation @xmath583 for all @xmath389 . from the definition of pairing we can get @xmath584 then @xmath346 is well - defined . define @xmath585 for @xmath389 . since @xmath35 is a faithful module and we also have an action of @xmath346 on @xmath35 , then @xmath586 is an injective homomorphism . similarly @xmath587 is the span of the elements @xmath582 in the algebra generated by @xmath1 and @xmath35 subject to the commutation relation @xmath588 for any @xmath389 . then we can get the following easy result . we have that @xmath558 is anti - isomorphic with @xmath557 . the anti - isomorphism is defined as @xmath589 . with this map we can get a right @xmath559-module structure on @xmath1 . for all @xmath590 , define @xmath591 with this action , @xmath1 is a right faithful @xmath559-module . inspired by the above module structure , @xmath1 can be also regarded as a faithful right @xmath559-module with action @xmath592 for any @xmath590 . now assume that @xmath1 is an algebraic quantum groupoid and @xmath13 denotes the dual of @xmath1 . recall that an algebraic quantum groupoid is a regular weak multiplier hopf algebra with a faithful set of left integrals . consider the dual pair @xmath593 , we have the smash product @xmath594 . and @xmath1 is a left faithful @xmath594-module with the action given by @xmath595 for all @xmath596 . [ 5.5 ] the algebra @xmath594 as acting on @xmath1 is the span of the maps of the form @xmath597 from @xmath1 to @xmath1 , where @xmath598 and @xmath421 is a left integral on @xmath1 . let @xmath599 . by remark 3.3 , we have @xmath600 here we use the property of the integral . since @xmath107 is bijective and @xmath288 , so @xmath601 where @xmath602 . this finishes the proof . consider the algebra @xmath603 with the product given by @xmath604 where @xmath605 . we will use @xmath606 to denote @xmath603 with above product and @xmath607 to denote the element @xmath608 . let @xmath609 it is a sub - algebra of @xmath606 . following the idea as in proposition 6.7 in @xcite , we have the following result . [ 5.6 ] we have that @xmath594 is isomorphic with @xmath610 . we give the sketch of the proof . the isomorphism map @xmath611 is given by @xmath612 and we know that @xmath1 is a faithful left @xmath613-module with the action @xmath614 for @xmath596 . by proposition [ 5.5 ] , we can get that @xmath615 is an isomorphism . in this section we will study the bi - smash products and obtain the duality theorem for an algebraic quantum groupoid @xmath1 . let @xmath616 be an algebraic quantum groupoid , acting on an algebra @xmath11 , so we can define the smash product @xmath480 . let @xmath35 be another algebraic quantum groupoid paired with @xmath1 . also assume the pairing is non - degenerate . given a pairing @xmath414 , there is a left action of @xmath35 on @xmath1 @xmath617 , with this action we can give @xmath480 a @xmath35-module algebra structure . @xmath480 is a left @xmath35-module algebra with the action @xmath618 given by @xmath619 for all @xmath620 . first we show it is a unital module . let @xmath621 @xmath622 since @xmath1 is a unital @xmath35-module , so @xmath480 is unital . for @xmath623 , we have @xmath624 so @xmath480 is a @xmath35-module algebra . remark that in the above calculations we always have the well coverings . now we can form the bi - smash product @xmath625 . and the following result is a consequence of proposition 5.1 . @xmath626 is a faithful left @xmath625-module with the action given by @xmath627 let @xmath11 be an @xmath1-module algebra . set @xmath628 note that @xmath629 belongs to @xmath630 . now consider the map @xmath631 which is defined by @xmath632 for all @xmath483 . in fact the map is bijective , and its inverse is given by @xmath633 remark that @xmath500 is well - defined . for @xmath499 , we get @xmath634 and with the map @xmath500 we can define a new faithful action @xmath635 of @xmath625 on @xmath636 . for all @xmath637 , define the action @xmath635 as @xmath638 with the action @xmath635 , we have that @xmath636 is a faithful @xmath625-module . first we have @xmath639 and @xmath640 then we get @xmath641 next we are going to show that the map gives a module action . @xmath642(r''\bar{\o}a'')\\ & = & \sum \pi(\<a'_{(2)},b_{(1)}\>(x(a_{(1)}x')\#_{e_a}a_{(2)}a'_{(1)})\#_{e_b}b_{(2)}b')(r''\bar{\o}a'')\\ & = & \sum \<a'_{(3)}a''_{(3 ) } , b\>\<a''_{(4 ) } , b'\>(s^{-1}(a_{(2)}a'_{(1)}a''_{(1)})(x(a_{(1)}x')))x''\bar{\o}a_{(3)}a'_{(2)}a''_{(2)}\\ & = & \sum \<a'_{(4)}a''_{(4 ) } , b\>\<a''_{(5 ) } , b'\ > ( s^{-1}(a_{(3)}a'_{(2)}a''_{(2)})x)(s^{-1}(a_{(2)}a'_{(1)}a''_{(1)})((a_{(1)}x')))x''\bar{\o}a_{(4)}a'_{(3)}a''_{(3)}\\ & = & \sum \<a'_{(4)}a''_{(4 ) } , b\>\<a''_{(5 ) } , b'\>(s^{-1}(a_{(1)}a'_{(2)}a''_{(2)})x)(s^{-1}(a'_{(1)}a''_{(1)})x')x''\bar{\o}a_{(2)}a'_{(3)}a''_{(3)}\end{aligned}\ ] ] for the last equality we use @xmath643 . and remark that we have well coverings in the above equalities . such as in the first equality @xmath137 is covered by @xmath644 , and then the others are well covered . on the other side we have @xmath645\pi [ ( r'\#_{e_a}a')\#_{e_b}b'](r''\bar{\o}a'')\\ & = & \pi(((r\#_{e_a}a)\#_{e_b}b))\sum \<a''_{(3 ) } , b'\>(s^{-1}(a'_{(1)}a''_{(1)})x')x''\bar{\o}a'_{(2)}a''_{(2)}\\ & = & \<a'_{(4)}a''_{(4)},b\>\<a''_{(5 ) } , b'\>(s^{-1}(a_{(1)}a'_{(2)}a''_{(2)})x)(s^{-1}(a'_{(1)}a''_{(1)})x')x '' \bar{\o}a_{(2)}a'_{(3)}a''_{(3)}.\end{aligned}\ ] ] so we get a left @xmath625-module @xmath636 . since @xmath500 is bijective and combine with proposition 6.2 , we have that @xmath636 is a faithful @xmath625-module . now let us specialize to the case of an algebraic quantum groupoid @xmath1 with a left integral @xmath421 . set @xmath646 following the same idea of proposition [ 5.6 ] or theorem 7.6 in @xcite , we get the following duality theorem . [ 6.4 ] let @xmath1 be an algebraic quantum groupoid with a left integral @xmath421 and with the dual @xmath13 . let @xmath11 be a left @xmath1-module algebra . then we have that @xmath647 as algebras . for @xmath648 , the resulting of the algebra @xmath649 acting on @xmath636 is the span of operators of the form @xmath650 we denote the algebra of operators by @xmath355 . since @xmath500 is bijective and write @xmath93 as @xmath651 where @xmath317 , we find that @xmath355 is spanned by the operators of the form @xmath652 we can replace @xmath653 by @xmath654 , for any @xmath655 and with the fact that @xmath500 is bijective then @xmath355 is spanned by the operators of the form @xmath656 since both modules are faithful , then we get an isomorphism . let @xmath0 be a regular weak multiplier hopf algebra with an identity . then it is a a weak hopf algebra ( see proposition 4.12 in @xcite ) . in this case we obtain the following duality theorem for actions of quantum groupoids which was proven in @xcite . let @xmath0 be a finite - dimensional weak hopf algebra and @xmath1 be a left @xmath0-module algebra . then we can form the smash product @xmath657 and @xmath658 . the map @xmath659 defined by @xmath660 for all @xmath661 is an isomorphism of algebras . in @xcite it is proved by a straightforward computation . remark that we use a different method to prove the theorem . and also we generalize it to the infinite case . if the canonical idempotent @xmath69 is equal to @xmath33 , then the regular weak multiplier hopf algebra @xmath1 is a regular multiplier hopf algebra . and now we can get the main result in @xcite . if @xmath1 is an algebraic quantum group , acting on an algebra @xmath11 and if @xmath13 is the dual of @xmath1 , acting on the smash product @xmath12 by means of the dual action , then the bi - smash product @xmath662 is isomorphic with @xmath663 . the authors are very thankful to professor a. van daele for his valuable comments on this paper . the authors are also very grateful to the anonymous referee for his / her thorough review of this work and his / her comments and suggestions which help to improve the first and the second version of this paper . the work was partially supported by the nsf of china ( no . 11371088 ) and the nsf of china ( no.11571173 ) . a. van daele , s.h.wang , weak multiplier hopf algebras . preliminaries , motivation and basic examples , in operator algebras and quantum groups , vol . 98(5 ) ( banach center publication , warsaw , 2012 ) , pp . 367415 .

in this article we consider commuting graphs of involution conjugacy classes in the affine weyl group of type . we show that where the graph is connected the diameter is at most . + msc(2000 ) : 20f55 , 05c25 , 20d60 .

1612.02185

consider @xmath2 a subset of a group @xmath3 . the elements of @xmath2 constitute the vertices of the commuting graph @xmath4 , in which @xmath5 are joined by an edge whenever @xmath6 . if @xmath2 is a set of involutions then we call @xmath4 a commuting involution graph . commuting graphs have been studied by many authors in various contexts . for example fischer @xcite studied commuting involution graphs for the case when @xmath2 is a conjugacy class of involutions , in his work on 3-transposition groups . segev and seitz looked in @xcite at commuting graphs for finite simple groups @xmath3 where @xmath2 consists of the non - identity elements of @xmath3 . more recently giudici and pope @xcite gave some results on bounding the diameters of commuting graphs of finite groups . commuting graphs for elements of order 3 have been considered in @xcite ; there have also been many papers dealing with the case where @xmath2 consists of all non - identity elements of a given group such as for example @xcite . + in @xcite , bates et al looked into the commuting involution graph @xmath4 where @xmath2 is a conjugacy class of involutions in @xmath3 and @xmath3 is @xmath7 . the remaining finite coxeter groups were analysed in @xcite . commuting involution graphs in affine coxeter groups of type @xmath8 have been considered in @xcite . in this article , we investigate the commuting involution graphs @xmath9 for type @xmath10 . + for the rest of this article , let @xmath11 be an affine weyl group of type @xmath12 , for some @xmath13 , writing @xmath3 when @xmath14 is not specified , and let @xmath2 be a conjugacy class of involutions of @xmath3 . we write @xmath15 for the diameter of @xmath4 when @xmath4 is a connected graph , in other words the maximum distance @xmath16 between any @xmath5 in the graph . since conjugation by any group element induces a graph automorphism , we can determine the diameter by fixing any @xmath17 , and then @xmath18 . our main result is the following . [ summary ] let @xmath2 be a conjugacy class of involutions in the group @xmath11 of type @xmath12 . if the commuting graph @xmath4 is connected , then its diameter is at most @xmath1 . the proof of theorem [ summary ] is broken into several cases depending on the type of the conjugacy class . to describe this further , and in order to state our results about connectedness , we need to describe a parameterisation of the involution conjugacy class . we will explain in section 2 that involutions in @xmath3 can be written as ` labelled permutations ' ( notation [ notation ] ) . these are permutations expressed as products of disjoint cycles in which every cycle has a sign and an integer written above it . for example in @xmath19 one of the involutions is @xmath20 . a cycle is positive if it has a plus sign , and negative if it has a minus sign . the _ labelled cycle type _ of an involution @xmath21 will be a quadruple @xmath22 where @xmath23 is the number of transpositions , @xmath24 is the number of negative @xmath25-cycles with an even number above them , @xmath26 is the number of negative 1-cycles with an odd number above them , and @xmath27 is the number of positive 1-cycles . for the example in @xmath19 , its labelled cycle type is @xmath28 . + we will show ( theorem [ conjclass ] ) that involution conjugacy classes in @xmath3 are parametrised by labelled cycle type . we may now give the conditions under which the graph @xmath4 is connected . [ main ] let @xmath2 be a conjugacy class of involutions in @xmath11 ( where @xmath29 ) with labelled cycle type @xmath22 . then @xmath4 is disconnected in each of the following cases . @xmath30 and @xmath31 ; @xmath32 , @xmath31 and either @xmath33 or @xmath34 ; @xmath32 and @xmath35 ; @xmath36 and @xmath37 ; @xmath38 , @xmath37 and @xmath39 . in all other cases , @xmath4 is connected . we note that the bound on diameter given in theorem [ summary ] is best possible . for example we have verified that when @xmath40 , @xmath37 and @xmath41 , the commuting involution graph has diameter 10 . in section [ sec2 ] we will establish notation and describe the conjugacy classes of involutions in @xmath3 . section 3 is dedicated to proving the main theorems . in section 4 we give examples of selected commuting involution graphs . for the remainder of this paper @xmath42 is a fixed involution with @xmath2 its conjugacy class in the affine irreducible coxeter group @xmath3 . our first job is to establish what involutions in @xmath3 look like . + let @xmath43 be a finite weyl group with root system @xmath44 and @xmath45 the set of coroots . the affine weyl group @xmath46 is the semidirect product of @xmath43 with translation group @xmath47 of the coroot lattice @xmath48 of @xmath43 . see , for example , ( * ? ? ? * chapter 4 ) for more detail . for @xmath49 and @xmath50 we have @xmath51 for the rest of the paper @xmath52 will denote a coxeter group of type @xmath53 , and set @xmath54 . we write , respectively , @xmath43 and @xmath3 whenever it is not necessary to specify @xmath14 . we may take the roots of @xmath43 to be @xmath55 and @xmath56 , for @xmath57 , where @xmath58 is the standard orthonormal basis for @xmath59 . the coroots are then @xmath60 and @xmath56 . therefore in this case @xmath61 we may view the elements of @xmath43 as signed permutations ; they act on @xmath59 by permuting the subscripts of basis vectors and changing their signs . to obtain a signed permutation we write a permutation in sym(@xmath14 ) ( including 1-cycles ) , add a plus sign or a minus sign above each @xmath62 , and say @xmath62 is positive or negative accordingly . we adopt the convention of reading the sign first ; that is , if @xmath63 , then @xmath64 , @xmath65 and @xmath66 . + expressing @xmath21 as a product of disjoint cycles , we say that a cycle @xmath67 of @xmath21 is _ positive _ if there is an even number of minus signs above its elements , and _ negative _ if the number of minus signs is odd . for example , @xmath68 is a negative cycle , whereas @xmath69 is positive . it is straightforward to check that an involution of @xmath43 only has 1-cycles ( positive or negative ) and positive 2-cycles . by the definition of group multiplication in @xmath11 , we see that the element @xmath70 of @xmath3 is an involution precisely when @xmath71 . this allows us to characterise the involutions in @xmath11 . [ invn ] a non - identity element @xmath70 of @xmath11 is an involution if and only if @xmath21 , when expressed as a product of disjoint signed cycles , has the form @xmath72 for some @xmath73 and @xmath27 ; and , writing @xmath74 , we have @xmath75 when @xmath21 contains @xmath76 , @xmath77 when @xmath21 contains @xmath76 and @xmath78 for @xmath79 . if @xmath80 is an involution , then @xmath81 . thus @xmath21 is an involution of @xmath43 , and so its cycles are all either 1-cycles or positive 2-cycles . thus @xmath21 has the form given in the statement of the lemma . write @xmath82 . we must have @xmath83 for all @xmath84 . for @xmath85 , we have @xmath86 . thus @xmath75 . for @xmath87 , we have @xmath88 and @xmath89 . so @xmath77 . for @xmath90 we have @xmath91 , so there is no restriction on @xmath92 . for @xmath93 we have @xmath94 , forcing @xmath78 . these are the necessary conditions on @xmath21 and @xmath95 ; it is clear that they are also sufficient . [ notation ] from now on we will employ a shorthand for writing involutions @xmath96 of @xmath3 : if @xmath74 , then above each signed number @xmath62 in the expression of @xmath21 as a product of disjoint signed cycles , we will write @xmath97 . however for transpositions @xmath98 of @xmath21 , where the number above @xmath42 determines the number above @xmath99 as described in lemma [ invn ] , we write @xmath100 for @xmath101 and @xmath102 for @xmath103 . we will call this the _ labelled cycle form _ of @xmath42 . where it is helpful , we adopt the convention that cycles @xmath104 are omitted , as these fix both @xmath42 and @xmath105 . let @xmath42 be an involution in @xmath11 . the _ labelled cycle type _ of @xmath42 is the tuple @xmath106 , where @xmath23 is the number of transpositions , @xmath24 is the number of negative 1-cycles with an even number above them , @xmath26 is the number of negative 1-cycles with an odd number above them , and @xmath27 is the number of positive 1-cycles ( fixed points ) , in the labelled cycle form of @xmath42 . for example , the labelled cycle type of @xmath107 is @xmath108 . + having characterised the involutions , we must now determine the conjugacy classes . a well - known result , due to richardson , gives a description of involution conjugacy classes in coxeter groups . [ equiv ] let @xmath43 be an arbitrary coxeter group , with @xmath109 the set of fundamental reflections . we say that two subsets @xmath110 and @xmath111 of @xmath109 are @xmath43-equivalent if there exists @xmath112 such that @xmath113 . in the next result , we use the notation @xmath114 for the longest element of a finite standard parabolic subgroup @xmath115 . [ richardson ] let @xmath43 be an arbitrary coxeter group , with @xmath109 the set of fundamental reflections . let @xmath116 be an involution . then there exists @xmath117 such that @xmath118 is central in @xmath115 , and @xmath119 is conjugate to @xmath118 . in addition , for @xmath120 , @xmath118 is conjugate to @xmath121 if and only if @xmath110 and @xmath111 are @xmath43-equivalent . the coxeter graphs of @xmath122 and @xmath123 , @xmath124 are as follows . + 1.00 mm ( 58.00,20)(-10,0 ) ( 10,10.80)(1,0)12 ( 10,9.00)(1,0)12 ( 10,10.00 ) ( 22,10.00 ) ( 34,10.00 ) ( 10,14)(0,0)[cc]@xmath125 ( 22,14)(0,0)[cc]@xmath126 ( 32,14)(0,0)[cc]@xmath127 ( -10,10)(0,0)[lc]@xmath128 ( 22,10.80)(1,0)12 ( 22,9.00)(1,0)12 1.00 mm ( 58.00,14)(-30,0 ) ( 10,10.80)(1,0)12 ( 10,9.00)(1,0)12 ( 22,10.00)(1,0)8 ( 50,10.00)(1,0)8 ( 58,10.80)(1,0)12 ( 58,9.02)(1,0)12 ( 58.00,10.00 ) ( 10,10.00 ) ( 22,10.00 ) ( 70,10.00 ) ( 10,14)(0,0)[cc]@xmath125 ( 22,14)(0,0)[cc]@xmath126 ( 58,14)(0,0)[cc]@xmath129 ( 70,14)(0,0)[cc]@xmath130 ( -20,10)(0,0)[lc ] @xmath131 ( 35.00,10.00)(1,0)0.4 ( 45.00,10.00)(1,0)0.4 ( 40.00,10.00)(1,0)0.4 we may set @xmath132 , @xmath133 for @xmath134 , and @xmath135 . + it is well known that in the finite coxeter group @xmath43 of type @xmath53 , elements are conjugate if and only if they have the same signed cycle type . in particular two involutions are conjugate when they have the same number of transpositions , the same number of negative 1-cycles and the same number of positive 1-cycles . + in @xmath3 ( which is of type @xmath12 ) , the element @xmath136 is conjugate to @xmath137 via some @xmath138 if and only if : @xmath139 [ conjclass ] involutions in @xmath3 are conjugate if and only if they have the same labelled cycle type . in particular , every involution is conjugate to exactly one element @xmath140 of the form @xmath141 let @xmath142 be an involution of @xmath3 . by theorem [ richardson ] , @xmath142 is conjugate to @xmath118 for some finite standard parabolic subgroup @xmath115 of @xmath3 in which @xmath118 is central . therefore the connected components of the coxeter graph for @xmath115 are of types @xmath143 or @xmath144 for some @xmath62 ( including , by a slight abuse of notation , @xmath145 , where we have connected components with just the vertex @xmath146 or @xmath130 ) . thus @xmath147 for some @xmath148 with @xmath149 , where @xmath111 is a subset of @xmath150 no two elements of which are adjacent vertices in the coxeter graph . by conjugation in @xmath151 ( which after all is isomorphic to the symmetric group @xmath152 ) , we can assume that for some @xmath148 and @xmath23 with @xmath153 we have @xmath154 this gives that @xmath142 is conjugate to @xmath118 , where @xmath155 let @xmath156 , where @size10@mathfonts @xmath157 from equation we see that @xmath158 therefore , by setting @xmath159 we see that each involution in @xmath3 is conjugate to at least one element of the required form . + now consider an involution @xmath142 in @xmath3 with labelled cycle type @xmath22 , and suppose @xmath160 for some simple reflection @xmath161 . write @xmath162 and @xmath163 . by equation , @xmath164 is conjugate to @xmath21 in the underlying weyl group @xmath43 . hence @xmath164 and @xmath21 have the same number of transpositions , negative 1-cycles and positive 1-cycles as each other . in other words , the labelled cycle type of @xmath165 is @xmath166 for some @xmath167 satisfying @xmath168 . now , from equation again , @xmath169 is a labelled 1-cycle of @xmath142 if and only if @xmath170 is a labelled 1-cycle of @xmath165 , where @xmath171 and @xmath172 . in particular this means @xmath173 is a signed cycle of @xmath164 , so that @xmath174 . now @xmath175 . so @xmath176 . therefore @xmath177 . hence @xmath178 and so @xmath142 and @xmath165 have the same labelled cycle type . in particular @xmath142 is conjugate to at most one element @xmath42 of the form stated in the theorem . conversely , any two involutions of the same labelled cycle type @xmath22 are both conjugate to @xmath179 , and hence to each other . thus conjugacy is parameterised by labelled cycle type , and the set of elements @xmath180 contains exactly one representative of each conjugacy class of involutions in @xmath11 . we next prove three preliminary lemmas which will be used repeatedly in the proofs in section [ sec3 ] . [ 1cycle ] let @xmath181 and @xmath182 . then @xmath183 commutes with @xmath184 for all @xmath185 , whereas @xmath183 commutes with @xmath186 if and only if @xmath187 . for convenience , we assume without loss of generality that @xmath188 . now @xmath189 , so @xmath190 commutes with @xmath191 . in the case of two negative 1-cycles we have @xmath192 and @xmath193 , so @xmath190 commutes with @xmath194 if and only if @xmath195 . [ 2cycle ] let @xmath196 be distinct elements of @xmath197 and let @xmath185 , @xmath198 and @xmath199 be integers . @xmath200 and @xmath201 commute if and only if @xmath187 , and @xmath202 and @xmath203 commute if and only if @xmath187 . but @xmath200 and @xmath203 commute for all @xmath185 and @xmath198 . @xmath204 and @xmath205 commute for all @xmath185 , but there is no value of @xmath198 or @xmath185 for which @xmath204 and @xmath206 or @xmath207 commute . @xmath208 and @xmath209 commute if and only if @xmath210 , whereas @xmath202 and @xmath209 commute if and only if @xmath211 . we lose no generality by assuming , for ease of notation , that @xmath188 , @xmath212 and @xmath213 . recall that @xmath214 and @xmath215 are shorthand for @xmath216 and @xmath217 respectively . involutions commute precisely when their product is an involution ( or the identity ) , which we can check using lemma [ invn ] . we calculate @xmath218 this means @xmath214 and @xmath219 commute if and only if @xmath187 . we also have @xmath220 so again @xmath215 and @xmath221 commute if and only if @xmath187 . finally a similar calculation shows that @xmath222 , which is an involution for all values of @xmath185 and @xmath198 , so @xmath214 and @xmath221 always commute . certainly @xmath223 and @xmath224 commute for all @xmath185 ; after all , @xmath224 is just the identity element when @xmath213 . moreover , already in the underlying group @xmath43 of type @xmath225 we observe that @xmath226 does not commute with @xmath227 or @xmath228 , so there is no value of @xmath198 or @xmath185 for which @xmath223 and @xmath229 or @xmath230 commute . we have @xmath231 ; this is an involution if and only if @xmath232 . thus @xmath233 and @xmath234 commute if and only if @xmath210 . similarly @xmath235 ; this is an involution if and only if @xmath236 . therefore @xmath237 and @xmath238 commute if and only if @xmath211 . [ doubletrans ] let @xmath239 , @xmath240 , @xmath241 , @xmath242 , @xmath243 and @xmath244 , for distinct @xmath245 in @xmath197 and integers @xmath246 . then @xmath247 commutes with @xmath248 if and only if @xmath249 ; @xmath247 does not commute with @xmath250 ; @xmath247 commutes with @xmath251 if and only if @xmath252 ; @xmath253 commutes with @xmath250 if and only if @xmath254 ; @xmath253 does not commute with @xmath251 ; @xmath255 commutes with @xmath251 if and only if @xmath256 . we lose no generality by assuming , for ease of notation , that @xmath36 and @xmath257 @xmath258 and @xmath259 . hence @xmath260 now @xmath247 and @xmath248 commute if and only if @xmath261 is an involution . this occurs if and only if @xmath262 and @xmath263 . rearranging gives @xmath254 , as required for part ( i ) . since @xmath264 and @xmath265 do not commute , it is impossible for @xmath247 to commute with @xmath250 . we have @xmath258 and @xmath266 , and @xmath267 now @xmath247 and @xmath251 commute if and only if @xmath268 is an involution . from the above calculation this occurs if and only if @xmath269 and @xmath270 . that is , if and only if @xmath271 . we calculate @xmath272 . thus @xmath253 commutes with @xmath250 if and only if @xmath273 . since @xmath274 and @xmath275 do not commute , @xmath253 can not commute with @xmath251 . this part follows from the fact that @xmath276 . we end this section by stating the relevant results from @xcite about commuting involution graphs in weyl groups of type @xmath277 . [ finite ] suppose that @xmath43 is of type @xmath277 , and let @xmath278 set @xmath279 and @xmath280 . then the following hold . if @xmath30 , then @xmath4 is a complete graph . if @xmath281 , then @xmath282 . if @xmath283 and @xmath284 , then @xmath4 is disconnected . if @xmath285 and @xmath286 , then @xmath287 . if @xmath288 , @xmath37 and @xmath289 then @xmath290 . if @xmath288 , @xmath37 and @xmath291 then @xmath292 . finally if @xmath36 , @xmath37 and @xmath289 then @xmath4 is disconnected we begin by looking at connectedness . for an element @xmath293 in a conjugacy class @xmath2 of @xmath3 , define @xmath294 . then let @xmath295 be the conjugacy class of @xmath296 in @xmath43 . clearly if @xmath297 , then @xmath298 . [ easy ] suppose @xmath297 . if @xmath299 , then @xmath300 . if @xmath301 is disconnected , then @xmath302 is disconnected . the result follows immediately from the observation that is @xmath119 commutes with @xmath303 in @xmath3 , then @xmath296 commutes with @xmath304 in @xmath43 . we can now prove theorem [ main ] ( which gives necessary and sufficient conditions for @xmath4 to be disconnected ) in one direction . the proof in the other direction will arise from bounding the diameters of graphs not shown in theorem [ mainhalf ] to be disconnected . [ mainhalf ] let @xmath2 be a conjugacy class of involutions in @xmath11 ( where @xmath29 ) with labelled cycle type @xmath22 . then @xmath4 is disconnected in each of the following cases . @xmath30 and @xmath31 ; @xmath32 , @xmath31 and either @xmath33 or @xmath34 ; @xmath32 and @xmath35 ; @xmath36 and @xmath37 ; @xmath38 , @xmath37 and @xmath39 . let @xmath2 be a conjugacy class of involutions in @xmath11 , with labelled cycle type @xmath305 , and @xmath306 as defined in theorem [ conjclass ] . we deal with each case in turn . suppose @xmath31 and @xmath30 . then any @xmath142 in @xmath2 is of the form @xmath307 for some @xmath308 ( where @xmath24 of the @xmath308 are even and @xmath26 are odd ) . by lemma [ 1cycle ] , @xmath142 does not commute with any other element of @xmath2 . so in fact @xmath4 is completely disconnected in this case . suppose @xmath31 and @xmath309 ( the case @xmath33 is similar ) . then @xmath310 suppose @xmath311 such that @xmath42 commutes with @xmath99 . consider the cycle of @xmath99 that contains @xmath14 . this must be a negative 1-cycle or a 2-cycle , because @xmath99 has the same labelled cycle type as @xmath42 . if it is a 2-cycle , then @xmath99 can not commute with @xmath42 , by lemma [ 2cycle](iii ) . therefore it is a negative 1-cycle @xmath312 where @xmath185 is odd , and then by lemma [ 1cycle ] , @xmath99 contains @xmath313 . the same argument shows that any element @xmath314 of @xmath2 which commutes with @xmath99 must also contain @xmath313 , and inductively all elements in the connected component of @xmath4 containing @xmath42 must contain @xmath313 . therefore @xmath4 is disconnected . suppose @xmath284 and @xmath35 . if @xmath31 then @xmath4 is disconnected by ( ii ) . so we can assume @xmath315 . if either of @xmath24 or @xmath26 is zero , then theorem [ finite](iii ) and lemma [ easy ] imply that @xmath4 is disconnected . it remains to consider the case @xmath316 . here , @xmath317 and @xmath318 suppose @xmath311 such that @xmath42 commutes with @xmath99 , and suppose @xmath319 is a 1-cycle of @xmath99 . if @xmath320 then as @xmath42 contains a transposition @xmath321 for some @xmath322 , lemma [ 2cycle ] ( ii ) and ( iii ) show that the only way @xmath42 and @xmath99 could commute is if @xmath99 contained @xmath323 or @xmath324 where @xmath325 , contradicting our assumptions about the labelled cycle type of @xmath99 . therefore the elements appearing in 1-cycles of @xmath99 are @xmath326 , @xmath327 and @xmath14 . inductively this holds for all elements in the connected component of @xmath4 containing @xmath42 . therefore @xmath4 is disconnected . suppose @xmath36 and @xmath37 . if @xmath35 then @xmath4 is disconnected by ( iii ) . if this does nt happen , then one of @xmath24 , @xmath26 or @xmath27 is 2 . by theorem [ finite](v ) and lemma [ easy ] , @xmath4 is again disconnected . suppose @xmath38 , @xmath37 and @xmath39 . for any @xmath142 in @xmath2 , we have @xmath328 , where @xmath329 , @xmath198 and @xmath330 are even , and @xmath331 are odd . associate a set @xmath332 to @xmath142 . given that , by lemma [ 2cycle ] , a transposition can only commute with a pair of negative 1-cycles if either both cycles are odd or both are even , and also with reference to lemma [ 1cycle ] , we see that if @xmath165 in @xmath2 commutes with @xmath142 , then @xmath333 . therefore , for example , @xmath334 and @xmath335 are not connected in @xmath4 . our first result bounding diameters is when @xmath30 . [ disc ] if @xmath30 and @xmath336 , then @xmath337 in all other cases @xmath338 . in this case we have @xmath339 let @xmath340 . then for appropriate @xmath341 and @xmath342 we have that @xmath343 . since @xmath344 , exactly @xmath24 of the @xmath308 must be even . if @xmath345 , that is , @xmath346 , then @xmath119 commutes with an element @xmath303 in @xmath2 which contains the cycles @xmath347 . now @xmath303 certainly commutes with @xmath42 , and so @xmath348 . + now we suppose that @xmath349 , and that one of @xmath24 and @xmath26 is zero . without loss of generality , we may choose @xmath350 . then , writing @xmath351 , we have @xmath352 . since conjugation by elements of centralizer of @xmath42 preserves distance in @xmath4 , without loss of generality we may take @xmath119 to be of the following form for some integer @xmath84 with @xmath353 and even integers @xmath308 : @xmath354 now consider the following sequence , where @xmath355 and @xmath356 , @xmath357 , @xmath358 are arbitrary even integers . @xmath359 it is clear that @xmath360 commutes with @xmath361 for @xmath362 . moreover @xmath119 commutes with @xmath363 , and @xmath364 commutes with @xmath42 . if @xmath27 divides @xmath365 , then @xmath366 , which implies that @xmath367 . if @xmath27 does not divide @xmath365 , then @xmath368 . the diameter of the graph in each case does equal this bound because at each stage of a path from @xmath119 to @xmath42 we can add at most @xmath27 to the number of correct negative 1-cycles , but this requires the element being considered to share no fixed points with @xmath42 . hence , for example the element @xmath369 must be distance at least @xmath370 from @xmath42 . + the remaining case is when @xmath24 and @xmath26 are both nonzero , and @xmath349 . + for any @xmath371 define @xmath372 to be the number of ` correct ' negative 1-cycles in @xmath142 . that is , cycles @xmath373 where @xmath374 or @xmath375 where @xmath376 . thus , for example , @xmath377 . we say that other negative 1-cycles are ` incorrect ' . let @xmath344 . then @xmath119 commutes with an element @xmath378 whose positive 1-cycles are @xmath379 . we will describe a sequence @xmath380 where at each stage @xmath381 is an element of @xmath2 such that @xmath382 and the positive 1-cycles of @xmath381 are @xmath383 for some @xmath384 where @xmath385 and @xmath386 , with @xmath387 . moreover , for @xmath388 , @xmath381 will commute with @xmath389 . + observe that the positive 1-cycles of @xmath378 have the required form , and @xmath390 . assume that we have @xmath391 and let the positive 1-cycles of @xmath381 be @xmath383 . to form @xmath392 we look for incorrect cycles @xmath393 of @xmath381 , where each @xmath394 is even , each @xmath395 is odd , @xmath396 and @xmath397 . if such cycles can be found , then @xmath381 commutes with @xmath392 where @xmath392 is given by replacing the cycles @xmath398 of @xmath381 with @xmath399 , and leaving all other cycles unchanged . it is clear that @xmath392 commutes with @xmath381 ; moreover @xmath400 , has the appropriate positive 1-cycles and @xmath401 . this sequence can continue until we have some @xmath381 with cycles @xmath383 but ( without loss of generality ) fewer than @xmath84 incorrect cycles @xmath402 where @xmath185 is even and @xmath403 . suppose there are exactly @xmath404 such cycles ( with @xmath405 ) . so @xmath381 has the cycles @xmath406 . since @xmath407 for each @xmath408 there must be @xmath409 where @xmath410 for each @xmath411 , and even numbers @xmath394 such that @xmath381 has the cycles @xmath412 . similarly @xmath381 has cycles @xmath413 for odd @xmath198 , some @xmath414 , @xmath397 and @xmath415 . we now define @xmath165 to be @xmath416 with @xmath417 replaced with @xmath418 now @xmath381 commutes with @xmath165 , and @xmath419 . we observe that @xmath420 . notice that every @xmath421 with @xmath376 is either a fixed point of @xmath165 or appears in a cycle @xmath422 with @xmath198 odd . every incorrect even negative 1-cycle of @xmath165 features a fixed point of @xmath42 , so for some @xmath423 , and conjugating @xmath165 by a suitable element of the centralizer of @xmath42 if necessary , we can assume that the even negative 1-cycles of @xmath165 are @xmath424 now @xmath165 has at least @xmath425 correct negative 1-cycles . if we ignore the even negative 1-cycles and the correct odd negative 1-cycles , then @xmath426 cycles remain ( including @xmath27 fixed points ) . we can use the result for @xmath427 on this remaining part of @xmath165 to see that @xmath165 is distance at most @xmath428 from the element @xmath429 of @xmath2 given by @xmath430 now @xmath429 commutes with @xmath42 , and so @xmath431 . if @xmath432 and @xmath433 , then in fact @xmath434 so @xmath435 . if @xmath436 then again @xmath435 . if @xmath437 then we have @xmath438 . to give a lower bound on the diameter consider @xmath439 to create @xmath27 additional correct negative 1-cycles at each stage of a path from @xmath119 to @xmath42 one requires each fixed point to be a point not fixed by @xmath42 ; moreover in this case completion of the process for , say , the even negative 1-cycles requires the recreation of at least one fixed point between @xmath440 and @xmath14 and hence fewer than @xmath27 correct negative 1-cycles being created at the next stage . thus when @xmath315 we have @xmath441 , and when @xmath437 we have @xmath442 . this completes the proof . from now on , assume that @xmath284 . then we can take @xmath443 where @xmath444 . [ m>0k=0 ] suppose @xmath445 . if @xmath446 , then @xmath447 . if @xmath448 and @xmath449 then @xmath450 . let @xmath371 . we can write @xmath451 and @xmath452 , where @xmath453 are conjugate elements of the underlying weyl group @xmath43 . define an element @xmath454 , by @xmath455 , where @xmath456 is the unique central involution of @xmath43 . the result is that every minus sign in @xmath303 corresponds to a plus sign in @xmath457 , and every plus sign corresponds to a minus sign . since @xmath281 , @xmath303 and @xmath457 are conjugate in @xmath43 . so , by theorem [ finite](i ) , @xmath458 when @xmath459 and @xmath460 otherwise . now @xmath461 and @xmath165 commutes with @xmath142 by lemmas [ 1cycle ] and [ 2cycle ] . the result follows immediately . [ mnonzerol=0 ] suppose @xmath462 , @xmath31 , one of @xmath24 and @xmath26 is zero , and either @xmath463 or @xmath464 ( or both ) . then @xmath4 is connected with diameter at most @xmath14 . if @xmath371 and @xmath465 then @xmath37 and the transposition of @xmath142 is @xmath466 for some @xmath185 . let @xmath371 . without loss of generality @xmath350 . we start with the case @xmath37 . we will show that if the transposition of @xmath142 is @xmath466 for some @xmath185 , then @xmath467 . otherwise @xmath468 . the graph for @xmath288 ( and hence @xmath37 ) is figure [ l0 ] . we can see from this graph that the inductive hypothesis holds for @xmath288 , so assume @xmath469 . suppose the transposition of @xmath142 contains some @xmath470 with @xmath471 . then by lemma [ 2cycle](iii ) @xmath142 commutes with some @xmath472 such that @xmath165 has the 1-cycle @xmath373 and such that the transposition of @xmath165 is not @xmath466 . ignoring the cycle @xmath373 we can work within @xmath473 , to see that inductively @xmath474 . hence @xmath468 . if the transposition of @xmath142 is @xmath466 then @xmath142 certainly commutes with an element of @xmath2 which does not have this transposition . so @xmath467 as required . + now we assume @xmath475 and proceed by induction on @xmath24 to show that @xmath476 . suppose @xmath477 . then @xmath142 is distance at most 2 from an element @xmath165 of @xmath2 which has the transposition @xmath478 . to see this , note that if both @xmath327 and @xmath14 appear in transpositions of @xmath142 , or if both appear in 1-cycles of @xmath142 , then lemma [ 2cycle ] or lemma [ doubletrans ] , as appropriate , implies that @xmath142 commutes with some @xmath479 in @xmath2 which contains a transposition of the form @xmath480 for some @xmath185 . then @xmath479 commutes with @xmath165 . if on the other hand , @xmath142 contains ( for example ) the 1-cycle @xmath481 and @xmath327 appears in a transposition @xmath482 for some @xmath483 less than @xmath327 , then @xmath142 commutes with @xmath484 in @xmath2 containing the transpositions @xmath485 and @xmath486 for some @xmath185 and @xmath487 . lemma [ doubletrans ] now implies that @xmath142 commutes with an appropriate @xmath165 , in particular one containing the transpositions @xmath488 and @xmath489 . now @xmath165 in turn commutes with some @xmath429 in @xmath2 with the 1-cycles @xmath490 and @xmath491 . if we ignore these cycles and work in @xmath492 , then table [ tableg4 ] implies that when @xmath38 @xmath493 , and when @xmath494 proposition [ m>0k=0 ] tells us that @xmath495 . therefore @xmath476 . finally , suppose @xmath496 and @xmath497 . suppose there is some transposition of @xmath142 containing an element @xmath470 with @xmath498 . then by lemma [ 2cycle](iii ) @xmath142 commutes with some @xmath472 such that @xmath165 has the 1-cycle @xmath373 . by induction @xmath474 . hence @xmath468 . the final possibility is that the elements of the transpositions of @xmath142 are @xmath499 . since @xmath463 we can use lemma [ doubletrans ] to show that @xmath142 commutes with some @xmath165 in @xmath2 containing the transposition @xmath500 . working in @xmath501 ( using the case @xmath37 and induction on @xmath23 ) we see that @xmath474 . hence @xmath468 , which completes the proof of proposition [ mnonzerol=0 ] . [ kois2 ] suppose @xmath37 , @xmath31 , @xmath502 , @xmath503 and @xmath371 . then @xmath302 is connected with diameter at most @xmath504 . let @xmath371 . the distance of @xmath142 from @xmath42 will largely hinge on the whereabouts of @xmath14 and @xmath327 . but first we deal with the case where the transposition of @xmath142 is @xmath505 . here @xmath142 commutes with an element @xmath165 of @xmath2 having cycles @xmath506 with @xmath198 and @xmath330 odd . using proposition [ mnonzerol=0 ] on the remaining cycles of @xmath165 , we see that @xmath165 is distance at most @xmath326 from the element @xmath99 of @xmath2 whose cycles are the same as @xmath42 except that we have @xmath507 instead of @xmath508 . clearly @xmath509 . hence @xmath510 . this also shows that any element with odd negative 1-cycles @xmath506 is distance at most @xmath14 from @xmath42 . assume from now on that the transposition of @xmath142 is not @xmath505 , and that its odd negative 1-cycles are not @xmath506 . + if the transposition of @xmath142 is @xmath511 then @xmath142 is distance 2 from some @xmath472 with cycles @xmath512 . ignoring these cycles we use proposition [ mnonzerol=0 ] in @xmath492 to see that @xmath474 if the transposition of @xmath165 is @xmath505 and @xmath513 otherwise thus @xmath467 . if @xmath142 has cycles @xmath514 where @xmath515 , then @xmath142 is distance 3 from some @xmath165 in @xmath2 with cycles @xmath512 . thus @xmath510 if the transposition of @xmath142 is @xmath505 and @xmath467 otherwise . if one of @xmath327 and @xmath14 , say @xmath14 , appears in the transposition of @xmath142 and the other appears in an even negative 1-cycle @xmath516 , then @xmath142 commutes with some @xmath517 with the cycles @xmath518 . then @xmath479 is distance 2 from some @xmath472 with the cycles @xmath512 , and such that the transposition of @xmath165 is not @xmath519 for any @xmath21 . by proposition [ mnonzerol=0 ] again , @xmath467 . if @xmath142 contains @xmath520 where @xmath185 is even , @xmath198 is odd , @xmath21 is an integer and @xmath521 , then @xmath142 commutes with an element of @xmath2 containing @xmath522 , which commutes with an element of @xmath2 containing @xmath523 , which commutes with an element of @xmath2 containing @xmath524 , which finally commutes with an element @xmath165 of @xmath2 containing @xmath525 , such that the transposition of @xmath165 is the same as the transposition of @xmath142 , namely not @xmath526 . by proposition [ mnonzerol=0 ] , hence @xmath510 . the final case to consider is where @xmath14 is contained in the transposition of @xmath142 and @xmath327 is in an odd negative 1-cycle . so @xmath142 contains @xmath527 for some @xmath196 and integers @xmath528 with @xmath198 odd . if @xmath529 then subject to appropriate conjugation we can set @xmath530 . using proposition [ mnonzerol=0 ] in @xmath531 we see that @xmath142 is distance at most @xmath532 from the element @xmath165 where @xmath533 . then @xmath165 commutes with @xmath534 , which commutes with @xmath535 which is distance 2 from @xmath42 . thus @xmath510 . the last case is where without loss of generality @xmath536 and we can assume @xmath470 is 2 or 5 . let @xmath537 be the other element of @xmath538 . then @xmath142 is distance 2 from some @xmath479 in @xmath2 containing @xmath539 where @xmath540 is determined by @xmath142 but we may choose @xmath164 arbitrarily , and @xmath541 are integers with @xmath199 odd . now let @xmath542 . then @xmath543 and @xmath544 . what is @xmath545 ? if @xmath546 then @xmath547 . set @xmath548 . now working in @xmath549 we see from proposition [ mnonzerol=0 ] that @xmath550 . if @xmath551 then @xmath552 . this time set @xmath553 . then @xmath479 commutes with @xmath554 for some @xmath555 , which commutes with @xmath556 which commutes with @xmath165 , so @xmath557 . hence in all cases @xmath510 , which completes the proof of lemma [ kois2 ] . [ connectm=1l=0 ] suppose @xmath558 , @xmath462 , @xmath31 and @xmath24 and @xmath26 are both at least 2 . then @xmath4 is connected with diameter at most @xmath1 . if @xmath496 , then @xmath559 . assume that @xmath560 . suppose first that @xmath37 and let @xmath371 . we use induction on @xmath26 to show that if the transposition of @xmath142 is not @xmath561 , then @xmath510 . otherwise @xmath562 . if @xmath502 the result holds by lemma [ kois2 ] . if @xmath563 and the transposition of @xmath142 is not @xmath561 , then @xmath142 contains cycles @xmath564 where @xmath471 , @xmath565 , @xmath566 is a cycle of @xmath42 and @xmath567 and @xmath21 are all congruent modulo 2 . now @xmath142 commutes with some @xmath517 containing the cycles @xmath568 for appropriate @xmath569 . if we ignore @xmath566 and work in @xmath570 , then inductively @xmath571 . hence @xmath510 . suppose the transposition of @xmath142 is @xmath561 . then @xmath142 commutes with some @xmath165 in @xmath2 that does not contain this transposition , and we have seen that @xmath572 . hence @xmath562 . this completes the case @xmath37 . + if @xmath463 and @xmath502 then it is easy to see that @xmath142 is distance at most 2 from an element of @xmath2 containing the transposition @xmath573 . thus @xmath142 is distance at most 3 from an element @xmath165 of @xmath2 containing @xmath574 . ignoring these 1-cycles we may work in @xmath492 and apply proposition [ mnonzerol=0 ] to see that @xmath575 . hence @xmath467 . + now suppose @xmath576 and @xmath563 . if @xmath142 has a transposition containing an element @xmath470 with @xmath498 , then @xmath142 commutes with an element @xmath165 containing @xmath373 or @xmath577 ( choose whichever of these is a cycle of @xmath42 ) . then we can ignore this cycle and work in @xmath578 . inductively , using the base case @xmath502 , we see that @xmath579 . . finally we deal with the case that every transposition of @xmath142 is of the form @xmath580 where @xmath581 . because @xmath582 , it must be the case that @xmath142 contains cycles : @xmath583 where @xmath584 , @xmath585 , @xmath586 , @xmath587 and @xmath588 . then @xmath142 is distance 2 from an element @xmath165 with the cycles @xmath589 and @xmath590 . now , working inductively in @xmath591 we see that @xmath474 . hence @xmath467 . [ n=5 ] if @xmath592 , @xmath37 , @xmath315 and @xmath477 , then @xmath593 . let @xmath371 . if the transposition is @xmath594 then using table [ tableg3 ] for @xmath595 we see that @xmath596 . suppose the transposition of @xmath142 is @xmath597 where @xmath598 , and let @xmath599 be the remaining element of @xmath600 . then @xmath142 commutes with @xmath601 for some even integers @xmath185 and @xmath602 . now @xmath479 commutes with @xmath603 , which commutes with @xmath42 . so @xmath596 . the remaining cases are ( interchanging 1 and 2 if necessary ) when @xmath604 or @xmath605 for appropriate @xmath606 and @xmath602 . the following is a path of length at most 5 from either of these to @xmath42:@xmath607 @xmath142 , @xmath608 , @xmath609 , @xmath610 , @xmath611 , @xmath42 . hence in all cases @xmath612 , which completes the proof . we observe , because we will need it for lemma [ n=6 ] later , that the proof of lemma [ n=5 ] shows that @xmath613 in all cases except where ( modulo interchanging 1 and 2 , or 3 and 4 ) the transposition of @xmath142 is @xmath614 . [ m>0l>0 ] suppose @xmath462 , @xmath615 and @xmath616 . then @xmath302 is connected with diameter at most @xmath14 . suppose @xmath14 is minimal such that @xmath302 is a counterexample , and let @xmath371 such that @xmath617 . by lemma [ n=5 ] we can assume @xmath469 . if @xmath449 then @xmath371 commutes with some @xmath472 containing @xmath618 . ignoring this 1-cycle we can work in @xmath619 to find a path to @xmath42 , which inductively is of length at most @xmath327 , which implies @xmath467 , contrary to our choice of @xmath142 . hence @xmath432 + if elements @xmath470 and @xmath421 lying between @xmath620 and @xmath621 are contained in transpositions of @xmath142 , then @xmath142 commutes with some @xmath517 having the transposition @xmath597 for some @xmath185 ( if @xmath37 then we can set @xmath622 ) . if we ignore this transposition of @xmath479 we can work in @xmath623 , which is either the case @xmath315 with a smaller @xmath23 , so inductively the graph has diameter at most @xmath326 , or ( if @xmath37 ) we can use theorem [ disc ] , in which case the graph has diameter @xmath327 . in either case , we see that @xmath142 is distance at most @xmath327 from the element @xmath99 of @xmath2 whose cycles are the same as @xmath42 except that @xmath99 has @xmath624 instead of @xmath625 . since @xmath99 commutes with @xmath42 we have @xmath467 . the same argument holds if @xmath626 . if @xmath627 then similar reasoning shows again @xmath142 is distance at most @xmath327 from an element @xmath99 containing @xmath628 that commutes with @xmath42 . so @xmath467 . thus none of these pairs @xmath196 exist in @xmath142 . this implies @xmath629 . moreover if @xmath630 then @xmath631 and without loss of generality the transpositions of @xmath142 contain @xmath632 and @xmath14 . + if there is some @xmath421 in a transposition of @xmath142 with @xmath633 and if @xmath634 , then @xmath142 commutes with some @xmath472 containing @xmath375 . inductively we can work in @xmath635 to see that @xmath579 . hence @xmath467 . similarly , as long as either @xmath497 or @xmath634 ( or both ) , if there is some @xmath470 in a transposition of @xmath142 with @xmath636 then inductively @xmath467 . this means that if @xmath630 then @xmath477 and @xmath309 . + suppose that @xmath630 , @xmath477 and @xmath309 , so that @xmath637 . we have observed that the transpositions of @xmath142 must contain @xmath638 and @xmath639 . if 6 is not contained in an even negative 1-cycle of @xmath142 , then @xmath142 commutes with some @xmath165 containing a transposition @xmath597 for some @xmath185 , where @xmath640 . as at the start of this proof , more explicitly , inductively @xmath165 is distance at most 6 from an element @xmath99 containing @xmath642 that commutes with @xmath42 . hence @xmath643 , a contradiction . thus @xmath37 . + if @xmath142 contains @xmath644 where @xmath645 then @xmath142 commutes with some @xmath517 containing @xmath577 . so , using the result for @xmath646 , we get @xmath647 . thus @xmath467 . the same reasoning holds if @xmath497 and @xmath142 contains @xmath644 for some @xmath470 where @xmath636 . suppose first that @xmath648 . since @xmath469 we must have @xmath634 or @xmath350 and @xmath503 . thus the transposition of @xmath142 can not contain any @xmath470 with @xmath649 . hence without loss of generality @xmath142 contains @xmath650 . suppose that @xmath142 contains @xmath651 where @xmath652 mod @xmath653 and either @xmath654 or @xmath655 . then , @xmath142 commutes with some @xmath472 where @xmath165 contains @xmath656 for some @xmath185 . now @xmath42 commutes with some element @xmath429 containing @xmath657 . using theorem [ disc ] on @xmath658 ( that is , removing @xmath659 , from @xmath165 and @xmath429 ) we see that @xmath660 . therefore , @xmath467 . finally , we are reduced to the possibility that @xmath142 contains @xmath661 and any pair @xmath662 where @xmath663 and @xmath664 satisfies @xmath665 . since @xmath286 and @xmath648 , the only way this can occur is when @xmath666 and @xmath552 . conjugating by an element of the centraliser of @xmath42 if necessary , we can assume that @xmath667 where @xmath21 and @xmath555 are even , and @xmath198 , @xmath330 are odd . the following is a path from @xmath142 to @xmath42 in the graph : @xmath142 , @xmath668 , @xmath669 , @xmath670 , @xmath671 , @xmath672 , @xmath42 . so @xmath673 , another contradiction . + the final case to consider is where @xmath37 and @xmath309 . assume first that @xmath674 . suppose that @xmath142 contains @xmath675 with @xmath567 both even . then @xmath142 commutes with some @xmath472 containing @xmath676 for some @xmath199 . as observed at the start of this proof , @xmath579 . hence @xmath467 . now suppose @xmath142 does not contain @xmath677 . then there is a set @xmath678 containing 1 , 2 and @xmath327 such that @xmath679 where @xmath185 and @xmath602 are both even and @xmath198 is odd . by replacing @xmath142 with a conjugate under the centraliser of @xmath42 , we can further assume that @xmath680 for some @xmath421 . let @xmath681 . now , using the case @xmath682 on @xmath683 we see that @xmath684 . next we apply theorem [ disc ] to @xmath685 and @xmath686 ( noting that here there are no odd negative 1-cycles ) to see that @xmath687 . hence @xmath688 , another contradiction . + the remaining possibility is that @xmath38 , @xmath37 , @xmath689 . but lemma [ n=6 ] immediately after this proof shows that this graph has diameter 6 , which is the final contradiction completing the proof of theorem [ m>0l>0 ] . [ n=6 ] if @xmath38 , @xmath37 , @xmath477 and @xmath309 then @xmath690 . we have @xmath691 . let @xmath371 . suppose @xmath142 contains @xmath692 for some @xmath21 . by theorem [ disc ] , the graph of @xmath693 in @xmath694 has diameter 5 . hence @xmath612 . if @xmath142 contains @xmath695 where @xmath567 are even , then @xmath142 commutes with some @xmath479 containing @xmath692 for some @xmath21 , and we have just seen that @xmath696 . hence @xmath673 . + we next consider the cases where 1 and 2 are in different types of 1-cycle of @xmath142 . because interchanging 1 and 2 does not affect the distance of @xmath142 from @xmath42 , there are just three cases to consider here : @xmath142 contains @xmath697 , @xmath698 or @xmath699 , where @xmath185 is even and @xmath198 is odd . in the first case we have @xmath700 where @xmath701 , @xmath567 are even and @xmath198 is odd . here @xmath142 commutes with @xmath702 , which commutes with the element @xmath99 given by @xmath703 . glancing at table [ tableg4 ] we observe that in @xmath694 elements at distance 4 or 5 from @xmath693 require at least one even label to be nonzero . so @xmath704 . hence @xmath612 . for the second case , where @xmath142 contains @xmath698 , note that @xmath142 commutes with some @xmath479 containing @xmath697 . hence , using the first case we get @xmath673 . the third case is where @xmath142 contains @xmath699 . here , @xmath142 commutes with some @xmath479 of the form @xmath705 . the following is a path from @xmath479 to @xmath42 : @xmath706 so @xmath673 . + it remains to take care of the possibility that exactly one of 1 and 2 lies in the transposition of @xmath142 . here , without loss of generality , we can assume @xmath142 contains @xmath707 for @xmath471 . if @xmath142 contains @xmath708 for @xmath198 odd , or @xmath709 , then @xmath142 commutes with some @xmath479 containing @xmath697 , for even @xmath185 , and as shown earlier in this proof , @xmath696 . hence @xmath673 . we may thus assume that @xmath710 . if @xmath711 then @xmath142 commutes with @xmath479 given by @xmath712 . by theorem [ n=5 ] applied to @xmath713 we see that @xmath696 and so @xmath673 . so we can assume that @xmath714 . now @xmath142 is distance 4 from the element @xmath99 given by @xmath715 , as shown by the following path . @xmath716 if @xmath717 , then @xmath718 so @xmath612 . if @xmath719 , then @xmath720 , so @xmath673 . if @xmath721 , then @xmath720 so @xmath673 . since interchanging 3 and 4 does not affect the distance from @xmath42 , there is only one case left to deal with : @xmath722 and @xmath723 . so @xmath724 . but @xmath142 is the same distance from @xmath42 as @xmath725 , and @xmath165 commutes with @xmath429 given by @xmath726 for appropriate @xmath727 and @xmath555 . by the ` @xmath717 ' case above , @xmath728 . therefore @xmath729 . this completes the proof . [ [ proof - of - theorems - summary - and - main ] ] proof of theorems [ summary ] and [ main ] + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + suppose @xmath30 . if @xmath31 then @xmath4 is disconnected by theorem [ mainhalf](i ) . otherwise @xmath730 by theorem [ disc ] . suppose @xmath37 and @xmath31 . if one of @xmath24 and @xmath26 is 1 or the largest of @xmath24 and @xmath26 is 2 , then @xmath4 is disconnected by theorem [ mainhalf ] ( ii ) , ( iv ) and ( v ) . otherwise @xmath731 by propositions [ m>0k=0 ] , [ mnonzerol=0 ] , lemma [ kois2 ] and theorem [ connectm=1l=0 ] . suppose @xmath576 and @xmath31 . if either of @xmath24 or @xmath26 is 1 , then @xmath4 is disconnected by theorem [ mainhalf](ii ) . otherwise @xmath559 by propositions [ m>0k=0 ] , [ mnonzerol=0 ] and theorem [ connectm=1l=0 ] . finally suppose @xmath462 and @xmath732 . if @xmath733 then @xmath4 is disconnected by theorem [ mainhalf](iii ) . otherwise @xmath559 by lemma [ n=5 ] , theorem [ m>0l>0 ] and lemma [ n=6 ] . we first summarise the information on commuting involution graphs for @xmath734 , @xmath735 and @xmath736 . for fixed @xmath17 , the @xmath737 disc @xmath738 is the set of elements of @xmath2 which are distance @xmath62 from @xmath42 . since a length preserving graph automorphism interchanges classes with @xmath560 and those with @xmath739 , we list here only those classes with @xmath560 . in describing the elements we omit positive 1-cycles . in @xmath734 there are two connected graphs , both of diameter 2 . if @xmath740 then @xmath741 with the remaining elements of @xmath2 comprising the second disc . if @xmath742 then @xmath743 , with all other elements of @xmath2 lying in the second disc . tables [ tableg3 ] and [ tableg4 ] give , for each connected graph and each disc , a list of orbit representatives under the action of @xmath744 . we use @xmath567 and so on to represent arbitrary even numbers , with @xmath745 $ ] being an arbitrary nonzero even number . we use @xmath746 and so on for arbitrary odd numbers , with @xmath747 $ ] being any odd number other than 1 . finally @xmath21 , @xmath555 and so on will be arbitrary integers with @xmath748 $ ] an arbitrary nonzero integer . figure [ l0 ] is the collapsed adjacency graph for @xmath750 in @xmath751 , which has diameter 5 . in the graph @xmath21 is an arbitrary integer , @xmath748 $ ] is an arbitrary non - zero integer and @xmath567 and @xmath727 are arbitrary even integers . for simplicity we have only included shortest paths that is , we have omitted edges between nodes in the same disc . 99 f. ali , m. salman , and s. huang . _ on the commuting graph of dihedral group _ , communications in algebra vol . 44 , iss . 6 ( 2016 ) 23892401 . a. nawawi and p. rowley . _ on commuting graphs for elements of order 3 in symmetric groups _ , electron . j. combin . 22 ( 2015 ) , no . 1 , paper 1.21 , 12 pp . b. fischer . _ finite group generated by 3-transpostions _ , i.invent . math.13 ( 1971 ) , 232 - 246 . c. bates , d. bundy , s. perkins and p. rowley . _ commuting involution graphs for symmetric groups _ , j. algebra , 266(1)(2003 ) , 133 - 153 . c. bates , d. bundy , s. perkins and p. rowley . _ commuting involution graphs for finite coxeter groups _ , j. group theory 6(2003),461 - 476 . m. giudici and a. pope . _ on bounding the diameter of the commuting graph of a group _ , j. group theory , 17 ( 1)(2014 ) , 131 - 149 . _ reflection groups and coxeter groups _ , cambridge studies in advanced mathematics , 29 ( 1990 ) . s. perkins . _ commuting involution graphs in the affine weyl group @xmath8 _ , arch . math . 86 ( 2006 ) , no . 1,16 - 25 . _ conjugacy classes of involutions in coxeter groups _ , bull . * 26 * ( 1982 ) , 115 . y. segev and g.m . _ anisotropic groups of type @xmath752 and the commuting graph of finite simple groups _ , pacific j. math . 202(2002 ) , 125226 .

we derive optimal order a posteriori error estimates for fully discrete approximations of linear schrdinger - type equations , in thenorm . for the discretization in time we use the crank - nicolson method , while for the space discretization we use finite element spaces that are allowed to change in time . the derivation of the estimators is based on a novel elliptic reconstruction that leads to estimates which reflect the physical properties of schrdinger equations . the final estimates are obtained using energy techniques and residual - type estimators . various numerical experiments for the one - dimensional linear schrdinger equation in the semiclassical regime , verify and complement our theoretical results . the numerical implementations are performed with both uniform partitions and adaptivity in time and space . for adaptivity , we further develop and analyze an existing time - space adaptive algorithm to the cases of schrdinger equations . the adaptive algorithm reduces the computational cost substantially and provides efficient error control for the solution and the observables of the problem , especially for small values of the planck constant .

1301.4391

in this paper we focus on the a posteriori error control and adaptivity for fully discrete crank - nicolson finite element ( cnfe ) schemes for the general form of linear schrdinger equation : @xmath1 $ , } & \\ & u=0 & & \quad\mbox{on $ \partial \varomega\times ( 0,t]$ , } & \\ & u(\cdot,0)=u_0 & & \quad\mbox{in $ \varomega$ } , & \end{aligned } \right.\ ] ] where @xmath2 is a convex `` polygonal '' domain in @xmath3 , @xmath4 , with boundary @xmath5 , and @xmath6 in , @xmath7 is a positive constant , @xmath8\to\mathbb{r}$ ] and @xmath9\to\mathbb{c}$ ] are given functions and @xmath10 is a given initial value . a special case of is the so - called linear schrdinger equation in the semiclassical regime : @xmath11 with high frequency initial data . it is clear that can be obtained from by setting @xmath12 , @xmath13 and @xmath14 . in , @xmath15 ( @xmath16 ) is the scaled planck constant , @xmath17 is an @xmath18 time - dependent potential and @xmath19 is the wave function . the wave function @xmath19 is used to define primary physical quantities , called _ observables _ ( @xcite ) , such as the _ position density _ , @xmath20 and the _ current density _ , @xmath21 problems related to are of great interest in physics and engineering . however , the solution of is complicated from the theoretical as well as the numerical analysis point of view . it is well known that for @xmath15 small ( close to zero ) , the solution of oscillates with wavelength @xmath22 , preventing @xmath19 to converge strongly as @xmath23 . because of this , standard numerical methods fail to correctly approximate @xmath19 and the observables , unless very fine mesh sizes and time steps are used . in particular , previous works ( cf . , e.g. , @xcite ) suggested that for standard finite element ( fe ) methods there is a very restrictive dispersive relation connecting the mesh sizes ( space and time ) with parameter @xmath15 ; cf . , e.g. , below . this restrictive dispersive relation can be relaxed using the so - called time - splitting spectral methods , introduced earlier by bao , jin & markowich in @xcite , for the approximation of the solution of . in this paper , our goal is to show that constructing adaptive algorithms based on rigorous a posteriori error control leads to cnfe schemes which are competitive to the best available methods for the approximation of the solution ( and the observables ) of the semiclassical schrdinger equation , and in general of linear schrdinger equations of the form . it also permits , for the first time , realistic computations for rough potentials for the linear schrdinger equation in the semiclassical regime.to achieve our goal , in the current work we : 1 . provide rigorous a posteriori error analysis for for cnfe approximations using fe spaces that are allowed to change in time ; 2 . study the advantages of adaptivity through the obtained estimators for the efficient error control of . optimal order a posteriori error estimates for the heat equation for cnfe schemes with fe spaces that are allowed to change in time have been derived very recently by bnsch , karakatsani & makridakis in @xcite . however the extension of those ideas from the simple heat equation to the linear schrdinger equation is of increased difficulty due to the complex - value and multiscale nature of the problem . because of this , novel ideas and techniques are introduced . more precisely , our main contributions are : 1 . derivation of optimal order a posteriori error bounds in the @xmath0norm for cnfe schemes for . the fact that the analysis includes time - dependent potentials , makes the problem more challenging since there are no rigorous results for schrdinger equations for such potentials . in addition the existing literature on a posteriori error analysis for problems with time - dependent operators of the form @xmath24 is quite limited . to the best of our knowledge , only in @xcite the authors consider similar operators . moreover the derived estimates hold for @xmath25type potentials as well , in contrast to the existing literature . in particular , existing results require smooth @xmath26type potentials . however , this regularity requirement on the potential is rather restrictive from applications point of view . including @xmath18 time - dependent potentials in the analysis is important for another reason : it can be considered as the first step for the a posteriori error control of nonlinear schrdinger ( nls ) equations . more precisely , the relaxation scheme introduced by besse in @xcite suggests that a posteriori error bounds for linear schrdinger equations with @xmath25type time - dependent potentials is essential for the efficient approximation of the solution of certain nls equations . introduction of a _ novel elliptic reconstruction _ leading to upper bounds that do not involve the global @xmath25norm of @xmath27 , and thus , to bounds that do reflect the physical properties of the problem.the elliptic reconstruction was developed by makridakis & nochetto in @xcite to derive optimal order @xmath28 a posteriori error bounds for fe spatial discrete schemes for the heat equation using energy techniques . a straightforward generalization of this notion of the elliptic reconstruction to schrdinger equations leads to estimates that involve the @xmath25norm of the potential . consequently , the obtained estimates are practically useless and adaptivity is inefficient , even in the simplest case of constant potentials . therefore , proposing a modified elliptic reconstruction based on the physical properties of the problem under consideration is crucial for the efficient error control of ( and so ) . additionally , the new ideas developed for this purpose might be useful for other problems as well , such as convection - diffusion or reaction - diffusion problems a detailed numerical study on the reliability and robustness of the a posteriori estimators through a _ time - space adaptive algorithm_. our starting point is the adaptive algorithm proposed in @xcite , adapted to the linear schrdinger equation , . the a posteriori estimators derived in this work are on the solution @xmath19 of . however , in many applications observables like the position density , or the current density are far more important than the solution itself . thus , we introduce an appropriate modification of the a posteriori estimators and the adaptive algorithm . this modification is based on a heuristic idea and the results concerning the observables are very encouraging . overall , the adaptive algorithm reduces the computational cost substantially and provides efficient error control of @xmath19 and the observables for small values of the planck constant @xmath15 . it is very difficult to obtain such results via standard techniques and without adaptivity . we point out that our purpose is not to prove convergence and optimality of the considered time - space adaptive algorithm , but rather to show that adaptivity based on rigorous a posteriori error control can be proven beneficial for the approximation of the solution ( and the observables ) of the linear semiclassical schrdinger equation . in addition , it is to be emphasized that as long as the adaptive algorithm converges , we can guarantee rigorously , based on the a posteriori error analysis , that _ total error remains below a given tolerance_. for parabolic problems , a number of adaptive algorithms exists in the literature ; cf . , e.g. , @xcite and the references therein . however , convergence and optimality of time - space adaptive algorithms are very delicate and difficult issues . in the literature exists only one proven convergent time - space adaptive algorithm for evolution problems and can be found in @xcite . this algorithm is appropriate for the heat equation and backward euler fe schemes and it is not clear how to generalize it to other problems and higher order in time methods . despite the fact that problem ( and thus ) is linear , a posteriori error bounds and adaptive algorithms for linear schrdinger equations are very limited in the literature . in particular , a posteriori error estimates in the @xmath0norm for fully discrete cnfe schemes have been proven earlier by drfler in @xcite ; these estimates are first order accurate in time , thus not optimal . using these estimates , drfler also proposes an adaptive algorithm in @xcite . in @xcite ( see also @xcite ) , we considered only time - discrete approximations and we managed to prove optimal order a posteriori error estimates for in the @xmath28 and @xmath29norms . this was achieved using the crank - nicolson reconstruction proposed by akrivis , makridakis & nochetto in @xcite . similar estimates for , using an alternative reconstruction , proposed by lozinski , picasso & prachittham in @xcite , can be found in @xcite . to the best of our knowledge , optimal order a posteriori error estimates for fully discrete cnfe schemes do not exist in the literature . some preliminary results to that direction can be found in @xcite . however , the a posteriori estimators derived in @xcite are scaled by the global @xmath25norm of @xmath27 . hence , as already mentioned , the derived estimators do not reflect the physical properties of the problem , which makes adaptivity through these estimates not reliable . a posteriori error estimates in the @xmath0norm have been proven earlier in @xcite for uniform partitions and the time - splitting spectral methods for the linear schrdinger equation in the semiclassical regime . in @xcite , only the one - dimensional case in space is studied and the analysis , as in @xcite , permits only time - independent potentials , without being obvious how the theory can be extended to time - dependent potentials . in addition , the time - spectral methods require smooth potentials ; the particular analysis is not applicable for @xmath25type potentials . the analysis of the current paper is based on the introduction of appropriate space - time reconstructions . such reconstructions for cnfe methods and fe spaces that are allowed to change in time were introduced , for the first time , very recently , by bnsch , karakatsani & makridakis in @xcite , for the proof of optimal order a posteriori estimates in the @xmath0norm for the heat equation . to define those time - space reconstructions , the authors combined the idea of the elliptic reconstruction in @xcite with the crank - nicolson reconstruction of @xcite . the notion of the elliptic reconstruction has also been used earlier in @xcite and @xcite for the derivation of optimal order a posteriori error estimates for backward euler fe schemes for the heat and the wave equation , respectively . the reconstruction technique is a useful tool for deriving optimal order a posteriori error bounds ; usually , this is not feasible via a direct comparison of the exact and the numerical solution ; cf . , e.g. , @xcite . in our context , time - space reconstructions can be defined through the novel elliptic reconstruction we introduce and the crank - nicolson reconstruction of @xcite . more precisely , the paper is organized as follows . in section [ prelim ] , we introduce notation , the variational formulation of problem and the fully discrete scheme . we propose the novel elliptic reconstruction and discuss its properties . with the aim of this new elliptic reconstruction , we then define appropriate time - space reconstructions . the main theoretical results are stated in section [ apost ] , where the a posteriori analysis is developed and optimal order error bounds are derived using energy techniques , residual - type error estimators and the properties of the reconstructions . the two last sections are devoted to the numerical investigation of the efficiency of the estimators . in particular , in section [ unif ] , we validate numerically the optimal order of convergence of the estimators using uniform partitions . for the linear schrdinger equation in the semiclassical regime , we verify numerically that the estimators have the expected behavior with respect to the scaled parameter @xmath15 . finally , in section [ adapt ] , we appropriately modify and apply to the one - dimensional semiclassical schrdinger equation a time - space adaptive algorithm described in @xcite ( see also @xcite ) . we further develop the algorithm and we make it applicable for the approximation not only of the exact solution @xmath19 but also for the observables , and we discuss in detail the benefits of adaptivity for equations of the form . problem can be rewritten equivalently in variational form as @xmath30 , & \\ & u(\cdot,0)=u_0 & & \mbox{in $ \overline{\varomega}$ } , & \end{aligned } \right.\ ] ] where @xmath31 denotes the @xmath32inner product , or the @xmath33 duality pairing , depending on the context . we also denote by @xmath34 the norm in @xmath35 it is well known that , if @xmath36;c^1(\overline\varomega)\big),$ ] @xmath37;l^2(\varomega)\big),\,f_t\in l^2\big([0,t];h^{-1}(\varomega)\big),$ ] and @xmath38 then problem admits a unique weak solution @xmath39;h_0 ^ 1(\varomega)\big)$ ] with @xmath40;$ ] @xmath41 ; cf . , e.g. , @xcite . we thus assume that the data of have the necessary regularity to guarantee the existence of a unique weak solution of . we emphasize that the a posteriori error estimates derived in the sequel , remain valid for @xmath42 as well , provided that is well - posed . in other words , in contrast to the existing analyses , ours includes rough potentials as well , under the knowledge of the well - posedness of . to avoid making the forthcoming analysis more technical , we further assume that @xmath27 satisfies @xmath43.\ ] ] condition is not restrictive from applications point of view , as , in most applications , @xmath27 denotes a nonnegative potential and thus is automatically satisfied . we consider a partition @xmath44 of @xmath45 $ ] , and let @xmath46 $ ] and @xmath47 denote the subintervals of @xmath45 $ ] and the time steps , respectively . let also @xmath48 . we discretize by a galerkin finite element method . to this end , we introduce a family @xmath49 of conforming shape - regular triangulations of @xmath2 . we further assume that each triangulation @xmath50 is a refinement of a macro - triangulation of @xmath2 and that @xmath51 and @xmath52 are compatible . two triangulations are said to be compatible if they are derived from the same macro - triangulation by an admissible refinement procedure . for precise definitions of these properties of the family @xmath49 , we refer to @xcite . note that the triangulations are allowed to change arbitrarily from one step to another , provided they satisfy the aforementioned compatibility conditions . these conditions are minimal and allow for heavily graded meshes and adaptivity . additionally , the forthcoming analysis is applicable without any quasiuniformity type assumptions on the mesh and without any restrictions on the sizes of neighboring elements of the triangulation . for an element @xmath53 , we denote its boundary by @xmath54 . let @xmath55 be the diameter of @xmath53 and @xmath56 . let also @xmath57 be the set of internal sides of @xmath53 ( points in @xmath58 , edges in @xmath59 and faces in @xmath60 ) and define @xmath61 to any side @xmath62 , we associate a unit vector @xmath63 on @xmath64 and for @xmath65 and a function @xmath66 , we define @xmath67(x):=\lim_{\delta\to 0}\big[\nabla { \upsilon}(x+\delta\bm{n}_e)-\nabla { \upsilon}(x-\delta\bm{n}_e)\big]\cdot\bm{n}_e.\ ] ] to each triangulation @xmath68 we associate the finite element space @xmath69 , @xmath70 where @xmath71 denotes the space of polynomials in @xmath72 variables of degree at most @xmath73 . with @xmath74 we denote the finest common coarsening triangulation of @xmath52 and @xmath51 and by @xmath75 its corresponding finite element space . finally , let @xmath76 and for @xmath77 let @xmath78 where the element @xmath79 is taken to be closed . for @xmath80 the discrete version @xmath81 of the laplace operator @xmath82 onto @xmath69 is defined as @xmath83 we now discretize problem by a _ modified crank - nicolson - galerkin _ scheme , introduced earlier for the heat equation in @xcite . given an approximation @xmath84 to the exact solution at @xmath85 we define approximation @xmath86 to the exact solution @xmath19 at the nodes @xmath87 by the numerical method : @xmath88 for @xmath89 , with @xmath90 in @xmath2 . in , @xmath91 , @xmath92 , and @xmath93 , @xmath94 are appropriate projections or interpolants . in sections [ unif ] , [ adapt ] , where we discuss the numerical experiments , @xmath95 and @xmath96 are taken to be the @xmath32projection . however , the theory is still valid for other choices of @xmath95 and @xmath96 ( cf . @xcite ) , and therefore we consider the method in this general setting . another non - standard term appearing in is @xmath97 instead of @xmath98 . as it was observed in @xcite , considering @xmath98 instead of @xmath97 may lead to oscillatory behavior of the obtained a posteriori estimators . for this reason , we consider the modified scheme instead of the standard one . the elliptic reconstruction was originally introduced by makridakis & nochetto in @xcite for the proof of optimal order a posteriori error estimates in space in the @xmath0norm for evolution problems , using energy techniques . it was also one of the main tools in the a posteriori error analysis of the heat equation for crank - nicolson fully discrete schemes ; cf . @xcite . for the linear schrdinger equation , we introduce a new type of elliptic reconstruction which reflects the physical properties of the problem , and in particular the physical properties of the semilcassical schrdinger equation . to this end , we introduce , in each @xmath99 , the constant @xmath100.\ ] ] the main reason for the choice of is that the knowledge on `` how far from @xmath101 is @xmath27 in @xmath2 '' gives qualitative information on the behavior of the exact solution , especially in the case of linear schrdinger equation in the semiclassical regime . in order for the elliptic reconstruction we introduce below to be well defined , we need @xmath102 , which is automatically satisfied due to . [ mellipticrec ] for fixed @xmath103 we define the elliptic reconstruction @xmath104 of @xmath105 to be the weak solution of the elliptic problem @xmath106 as we shall see in the sequel , the above modified elliptic reconstruction will allow us to obtain qualitatively better a posteriori error estimators compared to those obtained using the standard elliptic reconstruction ; cf . , @xcite . in fact , the @xmath107 that appears in the standard results of a priori error analysis , can now be replaced , due to , by @xmath108 leading to better constants . a very interesting question here , that needs further investigation , is whether the global constant @xmath109 can be localized in each element . this will not only lead to better constants in the final a posteriori error estimators , but also might give the inspiration of proposing appropriate adaptive strategies . using , we see that @xmath110 satisfies the orthogonality property @xmath111 let now @xmath112 be the weak solution of the following elliptic problem @xmath113 and let @xmath114 be its clment - type interpolant in @xmath69 ( for the definition of the clment - type interpolant and its properties we refer to @xcite ) . then we can prove the next auxiliary lemma . [ auxiliary ] let @xmath112 be the solution of and @xmath114 its clment - type interpolant . then , for all @xmath103 , we have the following estimate for @xmath115 @xmath116 using , we obtain @xmath117 and thus , invoking the definition of the modified elliptic reconstruction and the orthogonality property , we arrive at @xmath118 since both @xmath7 and @xmath101 are positive , follows by @xmath119 ; cf . . since we use finite element spaces that are allowed to change from @xmath120 to @xmath121 , we will need to work with quantities of the form @xmath122 for @xmath103 and @xmath123 to estimate such a quantity , we consider the elliptic problem @xmath124 with solution @xmath125 and we denote by @xmath126 its clment - type interpolant onto @xmath127 for @xmath103 and @xmath128 we have that @xmath129 the proof is similar to the proof of lemma [ auxiliary ] . to estimate a posteriori the errors @xmath130 and @xmath131 we use residual - type error estimators . to this end , for a given @xmath132 we define the following @xmath32elliptic estimator : @xmath133\|^2_{l^2({\partial}k)}\big)\bigg\}^{\frac 12}. \end{aligned}\ ] ] in case @xmath134 the term with the discontinuities in vanishes . for @xmath103 and @xmath135 we also define @xmath136\|^2_{l^2(k ) } \\&+\|h_{k}^{\frac 32}j[\nabla v_n-\nabla v_{n-1}]\|^2_{l^2(\check{\varsigma}_k^n)}\big)\bigg\}^{\frac 12}. \end{aligned}\ ] ] in view of the definition of @xmath137 and of , the lemma below is standard . its proof is based on duality arguments and the elliptic regularity estimate for the laplace operator . for details on the proof we refer , for example , to @xcite . for all @xmath138 it holds @xmath139 where the constant @xmath140 depends only on the domain @xmath2 and the shape regularity of the family of triangulations . similarly , by the estimate in the next lemma holds . for a detailed proof , we refer to @xcite . for @xmath103 and @xmath135 we have @xmath141 where the constant @xmath142 depends only on the domain @xmath2 , the shape regularity of the triangulations , and the number of bisections necessary to pass from @xmath51 to @xmath52 . we first define the continuous , piecewise linear interpolant @xmath143\to h_0 ^ 1({\varomega})$ ] between the nodal values @xmath144 and @xmath145 , i.e. , @xmath146 with @xmath147 and @xmath148 the space reconstruction of @xmath149 , that was used in @xcite to obtain of optimal order a posteriori error estimates for the backward euler - galerkin fully discrete scheme is given via @xmath150 however , as the authors note in @xcite to obtain optimal order in time a posteriori error estimates for the crank - nicolson method , a reconstruction in time is also needed . here , with the aid of the new elliptic reconstruction , we propose a two - point time - space reconstruction for linear schrdinger equations and the method . for @xmath151 we define the two - point time - space reconstruction @xmath152 of the cnfe scheme as @xmath153 where @xmath154\ ] ] and @xmath155,\ ] ] denote the linear interpolants of @xmath156 and @xmath157 , respectively , at the nodes @xmath120 and @xmath158 and @xmath159 in order to write compactly method and the reconstruction @xmath160 , we introduce the notation @xmath161 with this notation , the reconstruction @xmath160 is rewritten as @xmath162 and method as @xmath163 note that in each @xmath164,$ ] @xmath165 is a linear polynomial between the values @xmath166 and @xmath167 thus , it is straightforward to see that @xmath168 for @xmath151 there holds @xmath169 in particular , @xmath160 is continuous in time . furthermore , it satisfies @xmath170 that @xmath171 is obvious from the definition of @xmath172 moreover , @xmath173 since @xmath165 is a linear polynomial in time in @xmath99 , we have that @xmath174 and that @xmath175 follows invoking . finally , is an immediate consequence of differentiation in time of . we conclude the section by computing the difference @xmath176 for this , we introduce , for @xmath151 the notation @xmath177.\ ] ] [ the difference @xmath178 the difference @xmath179 satisfies @xmath180 using the definitions of @xmath160 and @xmath181 and the method in the form we obtain @xmath182 thus , using and the fact that @xmath183 we obtain @xmath184 equality follows now from , by noting that @xmath185 ; cf . and the definition of @xmath186 . in this section , we establish a posteriori error estimates in the @xmath0norm for problem , using the tools developed in the previous section . to this end , we denote by @xmath187 the error , where recall that @xmath149 is the piecewise linear interpolant between the nodal values @xmath144 and @xmath145 ; cf . . to achieve proving optimal order a posteriori error estimates in the @xmath0norm for we split the error as @xmath188 with @xmath189 , @xmath190 and @xmath191 we refer to @xmath192 as the main error , to @xmath193 as the time - reconstruction error and to @xmath194 as the elliptic - reconstruction error . the term @xmath193 measures the error due to the reconstruction in time . this term is of optimal order in time , cf . , but not yet an a posteriori quantity . it can be estimated a posteriori using the residual - type error estimators . the residual estimators will also be used for the direct estimation of the elliptic - reconstruction error . finally , as we shall see , the main error @xmath192 satisfies a perturbation of the original pde and it will be bounded by the perturbed terms using energy techniques . the perturbed terms are either a posteriori quantities of optimal order , or can be estimated a posteriori by estimators of optimal order . these terms will include quantities that measure the time and space errors , the effect of mesh changes and the variation of the data @xmath157 and @xmath27 . we now proceed with the estimation of @xmath193 and @xmath194 in propositions [ timerecerror ] and [ ellipterror ] , respectively . [ timerecerror ] for @xmath195 , the following estimate is valid for the time reconstruction error @xmath196 : @xmath197.\ ] ] we write @xmath198 and the desirable result now follows using and . [ ellipterror ] for the elliptic error @xmath199 we have , for @xmath195 : @xmath200 for @xmath201 @xmath202 hence , @xmath203 from where we immediately conclude , in view of . in view of we see that the reconstruction @xmath160 satisfies , for @xmath201 the equation @xmath204 with @xmath205 [ errormain ] the main error @xmath206 satisfies , for @xmath201 the equation @xmath207 where the residuals @xmath208 are given by @xmath209 @xmath210,\ ] ] @xmath211,\ ] ] and @xmath212 subtracting from we obtain , for @xmath201 @xmath213 we further write @xmath214 where we recall that @xmath215 thus , yield @xmath216,\phi\big\rangle . \end{aligned}\ ] ] similarly , in view of , we obtain @xmath217 combining , with , and using we arrive at . next , we prove the following auxiliary lemma . [ resrew ] the residual @xmath218 in can be rewritten as @xmath219 we just note , using the method in the form , that @xmath220 the result follows in light of , because @xmath221 for @xmath222 proposition [ errormain ] and lemma [ resrew ] together with energy methods , lead to the following a posteriori estimation in the @xmath0norm for the main error @xmath223 [ mainerror ] let @xmath224 then , for the main error @xmath206 and @xmath195 , it holds that @xmath225 where the time estimator @xmath226 is given by @xmath227 the space estimators @xmath228 are given by @xmath229 and the coarsening and data estimators @xmath230 and @xmath231 are @xmath232 and @xmath233\,dt,\ ] ] respectively . setting @xmath234 in and taking real parts yields @xmath235 or , @xmath236 then , it is easily seen that @xmath237 cf . , and @xmath238 cf . . going back to and plugging in we readily obtain . it is clear that the space estimators @xmath228 are expected to be of optimal order of accuracy in space . in fact , estimator @xmath239 is expected to be of optimal order in space and of order one in time , i.e. , it is a superconvergent term . as far as the first part of the time estimator @xmath226 is concerned , we note that @xmath240 so , it is expected to be of optimal order of accuracy in time . numerically , this term can be computed by invoking a quadrature in time , which is at least second order accurate ( i.e. , at least as accurate as the accuracy of the discretization method in time ) . the second part of @xmath226 is expected to be of optimal order in both time and space . on the other hand , note that estimator @xmath230 is not identically zero , only during the coarsening procedure . finally , for the estimators related to the data of the problem we have @xmath241 and @xmath242 the term @xmath243 is handled similarly . thus , it is straightforward to see that @xmath231 can be split into optimal order estimators in time and space , while @xmath244 is easily estimated a posteriori via optimal order estimators in space . for the constants @xmath245 we note that @xmath246 with @xmath247 and @xmath248.$ ] therefore , @xmath249 while @xmath250 is relatively small , provided that @xmath27 does not change much , with respect to the spatial variable . more precisely , @xmath251 when @xmath27 is constant in space , while the estimators that are multiplied by @xmath245 in vanish for constant potentials . this particular behavior of the estimators is natural from physical point of view . we conclude with the main theorem of the paper . let @xmath19 be the exact solution of and let @xmath149 be the continuous approximation of @xmath19 related to the modified crank - nicolson - galerkin method . then , the following estimate is valid for @xmath195 : @xmath252 where @xmath253 are given by , , and and @xmath254 are as in and , respectively . we write @xmath255 , whence , for @xmath256 @xmath257 estimate is now an immediate consequence of propositions [ ellipterror ] , [ timerecerror ] and [ mainerror ] . in this section , we perform various numerical experiments for the one - dimensional linear semiclassical schrdinger equation : @xmath258,\ ] ] using uniform partitions . our experiments , not only illustrate and complement our theoretical results , but also give important information in several other interesting aspects , like the behavior of the estimators with respect to the parameter @xmath15 . at the moment , the particular behavior can only be proven formally ; cf . subsection [ sensitivity ] . in all of the numerical experiments , the initial data is of the well known semiclassical wkb form : @xmath259 in , @xmath260 and @xmath261 are real and smooth functions on @xmath262 $ ] . in addition , @xmath260 is positive on @xmath263 and vanishes ( numerically ) at the endpoints @xmath264 and @xmath265 . the modified galerkin - crank - nicolson method and the corresponding a posteriori error estimators for problem - with homogeneous dirichlet boundary conditions , were implemented in a double precision c - code , using b - splines of degree @xmath266 as a basis for the finite element space @xmath267 . the involved projections @xmath96 and @xmath95 in are taken to be the @xmath32projection onto @xmath268 in what follows , we present some characteristic examples that allow us to verify the correct order of convergence of the estimators in time and space , and their dependence on the planck constant @xmath15 . we also report on the relation between the time and space mesh sizes with respect to @xmath15 in order to have convergence . we proceed by studying two different cases . the first one concerns time - independent potentials , while in the second one we consider a time - dependent potential . experiment 1 ( time - independent potentials ) . here , we consider three well - known types of potential : a constant potential , a harmonic oscillator and a double - well potential ( @xcite ) . in all three examples , the planck constant is taken to be of order 1 . more precisely , we study the following cases : a. @xmath270 , @xmath271 , @xmath272 and @xmath273 ; b. @xmath274 , @xmath275 , @xmath276 and @xmath277 ; c. @xmath278 , @xmath271 , @xmath279 and @xmath280 all computations are performed in @xmath262\times[0,t]=[-2,2]\times[0,1].$ ] our purpose is to compute the _ experimental order of convergence ( @xmath269 ) _ of the a posteriori error estimators at the final time @xmath281 . for this , we consider uniform partitions in both time and space . if we denote by @xmath73 the degree of b - splines used for the discretization in space , then in each implementation , the relation between the mesh size @xmath282 and the time step @xmath283 is taken to be @xmath284 with equality , whenever possible . we also denote by @xmath285 then , for each space estimator @xmath286 the @xmath269 is computed as @xmath287 where @xmath288 and @xmath289 denote the value of the estimators in two consecutive implementations with mesh sizes @xmath290 and @xmath291 , respectively . note that @xmath292 is expected to be of optimal order in space and of order @xmath293 in time , i.e. , it is a superconvergent term . therefore , the @xmath269 we expect to observe is @xmath294 due to and . similarly , for the time estimators @xmath295 the @xmath269 is computed as @xmath296 we are also interested in computing the _ effectivity index _ , defined as the ratio between the total a posteriori error estimator and the corresponding norm of the exact error . since we do not have at our disposal the exact solution for the three examples , we compute a reference solution @xmath297 instead , by taking very fine mesh and time step . in particular , we take as @xmath298 , while in space we discretize by b - splines of degree 5 and take as @xmath299 then , the reference error is defined as @xmath300 in addition , we define @xmath301 and we compute the effectivity index @xmath302 as @xmath303 note that for uniform partitions , the coarsening estimator @xmath304 is identically zero . our findings are reported in tables [ tbl1][tbl6 ] . in the case of constant potential @xmath270 , we discretize in space by linear b - splines . we recall that in this case @xmath305 is identically zero and does not appear in table [ tbl1 ] . as we see in tables [ tbl1 ] , [ tbl2 ] , all estimators decrease with the correct order . .space estimators @xmath306 and @xmath269 for experiment 1a . [ cols="<,^,^,^,^,^,^,^",options="header " , ] + in this section , we adjust and further develop a time - space adaptive algorithm for linear schrdinger equations , using the a posteriori error estimators derived earlier . our goal is to study numerically the behavior of the estimators under this adaptive algorithm , and investigate the benefits , in terms of computational cost and accuracy , of time - space adaptivity . to this end , we consider , as in the previous section , the one - dimensional linear schrdinger equation in the semiclassical regime , cf . , along with the wkb initial condition . the presented numerical experiments indicate that adaptivity through the a posteriori error bounds is indeed advantageous , especially for relatively small values of the planck constant @xmath15 , for both time - independent and time - dependent potentials . furthermore , by appropriately modifying the adaptive algorithm we are able to construct efficient approximations , not only to the exact solution @xmath19 , but also to observables and of problem - . as already mentioned , for small values of @xmath15 it is very difficult to approximate correctly and , unless very fine mesh sizes are used . the problem becomes harder in cases where caustics develop . this is a hard and delicate issue and adaptivity can play an important role to resolve it . we consider , modify and further develop the time - space algorithm of @xcite , introduced first in @xcite . we stress out once more that we do not claim that the particular adaptive algorithm is an optimal one . however , it appears to perform well for the problem under consideration and the estimators at hand . in that respect , it is possible to check the efficiency and robustness of the estimators . we next briefly describe the algorithm we use . to this end , we use @xmath307 to indicate the spatial grid at @xmath308 . we also use the notation @xmath309 . in addition , we can write @xmath310 where @xmath311 @xmath312 , @xmath313 and @xmath314 can readily be obtained from , , and . in all computations the constant @xmath140 , cf . , , is taken equal to @xmath293 and the involved local time integrals are computed using the midpoint quadrature rule . for @xmath315 we further define @xmath316 and let @xmath317 and @xmath318 denote the tolerances for the local time and space estimators @xmath319 and @xmath320 , respectively . the main steps of the adaptive algorithm are summarized schematically in the pseudocode below . : @xmath321 : given an initial grid @xmath322 compute @xmath323 @xmath324adaptinitialgrid(@xmath325 ) , @xmath326 more precisely , the adaptive algorithm starts by advancing the solution and computing the local space and time estimators . next , before starting the process of adapting the spatial grid , we perform a time - step refinement , if necessary , based on the local time estimator . we proceed on the spatial adaptation part of algorithm based on the local space estimator : we first mark the elements for refinement and/or coarsening and we adapt the grid appropriately , we recompute the solution and the local space and time estimators . next we perform another time - step refinement , if necessary , based on the local time estimator and then we loop back to the space estimator check . one step of the adaptive algorithm then concludes by a time - step coarsening step . reasonable choices for the parameters @xmath327 and @xmath328 are @xmath329 and @xmath330 while for @xmath331 and @xmath332 we take @xmath333 and @xmath334 . in all of the experiments , the coarsening percentage is taken to be @xmath335 . for the mesh refinement percentage , we take @xmath336 for the time - dependent potentials and @xmath337 for all the other cases . in the sequel , we denote by @xmath338 and @xmath339 the following global time and space estimators : @xmath340 respectively . finally , we define the total degrees of freedom of the adaptive algorithm at the final time @xmath341 as @xmath342 + 1,\ ] ] where @xmath343 $ ] denotes the integral part of a real number and @xmath344 denotes the degrees of freedom at time - level @xmath121 . for the first set of the numerical experiments with adaptivity , we consider two characteristic cases of time - independent potentials : a constant potential and a harmonic oscillator . in both cases , we consider the wkb initial data with @xmath345 in particular we consider : + case 1 : @xmath262\times[0,t]=[0,1]\times[0,0.1]$ ] , @xmath346 , @xmath347 and @xmath348 . + case 2 : @xmath262\times[0,t]=[-1,2]\times[0,0.54]$ ] , @xmath349 , @xmath350 and @xmath351 . for the first case , we discretize in space by b - splines of degree @xmath352 . the particular example is interesting , because caustics are formed before the final time @xmath353 . we first apply the time - space adaptive algorithm . as expected , we observe adaptivity in space but we do not observe adaptivity in time . however , in this case , we emphasize that regardless of the initial choice of the time - step , the adaptive algorithm is able to produce the required time - step for the desirable tolerance of the error . for this example , the given initial time - step was @xmath354 and adapted by the algorithm to @xmath355 , which is in agreement with ( see also ) . next , we perform the same experiment , but using uniform partitions and the same degrees of freedom as in the adaptive algorithm . the estimators are plotted in figure [ tmind1 ] in logarithmic scale , for both the adaptive algorithm and the uniform partition . we observe that the total estimator computed with the uniform partition is two orders of magnitude larger compared to the corresponding one using adaptivity . since the time - step , after its initial adaptation remains fixed , the evolution of the total time estimator is the same for both the adaptive algorithm and the corresponding uniform partition . the total space estimator dominates the time estimator in the uniform partition , and this is the reason that @xmath339 coincides with the total estimator on the right plot of figure [ tmind1 ] . [ height=6.5 cm , width=9cm]examples / ex1/t_vs_est_adapt2.pdf [ height=6.5 cm , width=9cm]examples / ex1/t_vs_est_unif2.pdf for the second case , we discretize in space by cubic b - splines and we apply again the adaptive algorithm . as initial time - step we take again @xmath354 and adapted to @xmath356 , which is larger than the expected one . this is because both , are sufficient , but not always necessary conditions for convergence for problem . the fact that the adaptive algorithm is able to compute the correct time - step size can be considered an advantage , since for the linear schrdinger equation in the semiclassical regime such a choice is crucial and delicate from the point of view of accuracy and stability of the approximations , as well as from the point of view of computational cost . in each time - slot , the mesh size varies from @xmath357 to @xmath358 , which proves that conditions can be relaxed through adaptivity in space ; very fine mesh sizes are needed only in certain areas of @xmath359 $ ] . in figure [ tmind2 ] , we plot the evolution of time , space and total estimators in logarithmic scale and the position density at the beginning and at the final time @xmath360 . as we observe from the plot of @xmath361 at @xmath362 caustics are formed for this problem as well . the a priori knowledge of such information requires very technical and tedious calculations . however this information can be obtained through the a posteriori error analysis and adaptivity . and total estimator in logarithmic scale , using the adaptive algorithm ( left ) and position density for @xmath363 ( upper right ) and at the final time @xmath360 ( lower right ) for the harmonic oscillator . [ tmind2],title="fig : " ] and total estimator in logarithmic scale , using the adaptive algorithm ( left ) and position density for @xmath363 ( upper right ) and at the final time @xmath360 ( lower right ) for the harmonic oscillator . [ tmind2],title="fig : " ] + the simplest time - dependent potentials are of the form @xmath364 , where @xmath181 denotes a smooth function in time ; @xcite . to check the efficiency of the estimators during time adaptivity , we choose two time - dependent potentials of this form , which change relatively fast with time . for the first experiment , we solve in @xmath262\times[0,t]=[1,2]\times[0,3]$ ] and we take @xmath365 and @xmath366 as @xmath367 we take the one in , while we choose @xmath368 , and we define the initial condition through . we use quadratic b - splines and we apply the time - space adaptive algorithm . in figure [ tmd1 ] , we plot the evolution of the estimators in a logarithmic scale , as well as the variation of the time - steps @xmath369 during time adaptivity . the considered potential changes faster with time in the subinterval @xmath370 $ ] , compared to @xmath371 $ ] , and this is the reason the required time - step is considerably smaller in this area . for this experiment , in each time - slot , the mesh size varies from @xmath372 to @xmath373 . and the total estimator in logarithmic scale ( left ) , and variation of time - steps @xmath369 during adaptivity ( right ) , for @xmath374[tmd1],title="fig : " ] and the total estimator in logarithmic scale ( left ) , and variation of time - steps @xmath369 during adaptivity ( right ) , for @xmath374[tmd1 ] ] for the second experiment , we solve in @xmath262\times[0,t]=[-1,2]\times[0,1]$ ] and we take @xmath375 and @xmath376 . we take the same initial condition as in the previous experiment and cubic b - splines . in figure [ tmd2 ] , we plot the evolution of the estimators in logarithmic scale and the variation in time of the time - steps and of the degrees of freedom . this is a characteristic example where intensive adaptivity is observed , in both time and space . and the dof s versus @xmath377 ( right ) during adaptivity for @xmath378 . [ tmd2],title="fig : " ] and the dof s versus @xmath377 ( right ) during adaptivity for @xmath378 . [ tmd2 ] ] .[tmd3],title="fig : " ] .[tmd3],title="fig : " ] + .[tmd3],title="fig : " ] .[tmd3 ] ] in figure [ tmd3 ] , we plot four snapshots : at the beginning , at the final time and in two intermediate times . from the plots we can also see the distribution of the grid points . at @xmath326 , we start with uniform partition . for the remaining three snapshots , we observe that the points are dense close to rough changes of the approximation . especially , in the third snapshot ( left plot from below ) , almost all the points are concentrated close to the peak , while in areas where the solution does nt change much , the grid is very sparse . this is an indicator of the robustness of the adaptive algorithm which can provide reliable results with considerably less computational cost , compared to uniform grids . we focus next on the approximation of the observables , . in particular , we propose a modification of the adaptive algorithm and we verify numerically the advantages of the modified algorithm for the approximation of the observables , in terms of computational cost and accuracy . for cnfe schemes , it is well known that the restrictive conditions between mesh sizes and the parameter @xmath15 needed for the efficient error control of the exact solution of can be relaxed for the error control of the corresponding observables . more precisely , as it was proven in @xcite , a sufficient and necessary condition for approximating well the observables is @xmath379 . moreover , the @xmath28 approximation of the exact solution implies the @xmath380 approximation of observables mean value ; @xcite . in view of all these , we modify the adaptive algorithm as follows : we multiply all estimators but @xmath381 and @xmath382 by @xmath15 , so that the new estimators will converge provided that @xmath379 , cf . , , . then , we apply the same algorithm , but with respect to these new estimators . we then perform various numerical experiments to verify whether this partially heuristic idea can be advantageous to the approximation of the observables . more precisely , we consider the constant potential @xmath346 and the wkb initial condition with @xmath383 and @xmath261 as in . we perform the experiments with adaptivity only in space . for the first two tests , we take @xmath262\times[0,t]=[-1,2]\times[0,0.54]$ ] , @xmath351 and @xmath350 or @xmath384 . recall that the particular example , considered earlier in @xcite , is interesting because caustics are formed before the final time . for the case @xmath350 , we take @xmath385 and discretize by quadratic b - splines , whereas for @xmath386 , we take @xmath387 and discretize by b - splines of degree @xmath352 . in figures [ obs11 ] , [ obs21 ] , we plot the position density using the adaptive algorithm ( left plot ) and uniform grid with the same degrees of freedom ( right plot ) . the solid line corresponds to the semiclassical limit of the exact observable which is possible to compute for constant potentials . the dotted lines correspond to the approximate observable . as we observe from these plots , the approximation using adaptivity is very good , while the one using uniform partition misses completely the angles and peaks . similar comments can be made for the plots referring to the current density . these plots can be viewed in figures [ obs12 ] and [ obs22 ] for @xmath350 and @xmath388 , respectively . in the plots concerning the approximations with space adaptivity , we also see the distribution of the grid points . it is remarkable that most of the points are concentrated close to the angles and peaks . on the contrary , very few points are placed around the endpoints , where the observables remain constant . the total number of degrees of freedom in adaptivity corresponds to @xmath389 dof s in each time - slot for @xmath350 and to @xmath390 for the case @xmath386 . the required degrees of freedom in each time - slot with uniform partition are more than @xmath391 for @xmath350 and more than @xmath392 for @xmath386 . in case @xmath393 solid line represents the semiclassical limit of the exact observable , while dotted line represents the approximation using adaptivity ( left ) and uniform partition with the same dof s ( right).[obs11],title="fig : " ] in case @xmath393 solid line represents the semiclassical limit of the exact observable , while dotted line represents the approximation using adaptivity ( left ) and uniform partition with the same dof s ( right).[obs11 ] ] in case @xmath393 solid line represents the semiclassical limit of the exact observable , while dotted line represents the approximation using adaptivity ( left ) and uniform partition with the same dof s ( right).[obs12],title="fig : " ] in case @xmath393 solid line represents the semiclassical limit of the exact observable , while dotted line represents the approximation using adaptivity ( left ) and uniform partition with the same dof s ( right).[obs12 ] ] in case @xmath394 solid line represents the semiclassical limit of the exact observable , while dotted line represents the approximation using adaptivity ( left ) and uniform partition with the same dof s ( right).[obs21],title="fig : " ] in case @xmath394 solid line represents the semiclassical limit of the exact observable , while dotted line represents the approximation using adaptivity ( left ) and uniform partition with the same dof s ( right).[obs21 ] ] in case @xmath395 solid line represents the semiclassical limit of the exact observable , while dotted line represents the approximation using adaptivity ( left ) and uniform partition with the same dof s ( right).[obs22],title="fig : " ] in case @xmath395 solid line represents the semiclassical limit of the exact observable , while dotted line represents the approximation using adaptivity ( left ) and uniform partition with the same dof s ( right).[obs22 ] ] [ obs41 ] in case @xmath396 solid line represents the semiclassical limit of the exact observable , while dot line represents the approximation using adaptivity ( left ) and uniform partition with the same dof s ( right).[obs41],title="fig : " ] in case @xmath396 solid line represents the semiclassical limit of the exact observable , while dot line represents the approximation using adaptivity ( left ) and uniform partition with the same dof s ( right).[obs41 ] ] in case @xmath396 solid line represents the semiclassical limit of the exact observable , while dotted line represents the approximation using adaptivity ( left ) and uniform partition with the same dof s ( right).[obs42],title="fig : " ] in case @xmath396 solid line represents the semiclassical limit of the exact observable , while dotted line represents the approximation using adaptivity ( left ) and uniform partition with the same dof s ( right).[obs42 ] ] for the first two tests , we take @xmath262\times[0,t]=[-1,2]\times[0,0.54]$ ] , @xmath351 and @xmath350 or @xmath384 . recall that the particular example , considered earlier in @xcite , is interesting because caustics are formed before the final time . for the case @xmath350 , we take @xmath385 and discretize by quadratic b - splines , whereas for @xmath386 , we take @xmath387 and discretize by b - splines of degree @xmath352 . in figures [ obs11 ] , [ obs21 ] , we plot the position density using the adaptive algorithm ( left plot ) and uniform grid with the same degrees of freedom ( right plot ) . the solid line corresponds to the semiclassical limit of the exact observable which is possible to compute for constant potentials . the dotted lines correspond to the approximate observable . as we observe from these plots , the approximation using adaptivity is very good , while the one using uniform partition misses completely the angles and peaks . similar comments can be made for the plots referring to the current density . these plots can be viewed in figures [ obs12 ] and [ obs22 ] for @xmath350 and @xmath388 , respectively . in the plots concerning the approximations with space adaptivity , we also see the distribution of the grid points . it is remarkable that most of the points are concentrated close to the angles and peaks . on the contrary , very few points are placed around the endpoints , where the observables remain constant . the total number of degrees of freedom in adaptivity corresponds to @xmath389 dof s in each time - slot for @xmath350 and to @xmath390 for the case @xmath386 . the required degrees of freedom in each time - slot with uniform partition are more than @xmath391 for @xmath350 and more than @xmath392 for @xmath386 . the first two tests indicate that the smaller the value of @xmath15 using adaptivity is very advantageous . to make this indication stronger , we perform a final test in which @xmath262\times[0,t]=[0,1]\times[0,0.1]$ ] , @xmath348 and @xmath397 . this is another example where caustics are formed . we use cubic b - splines and @xmath398 in figures [ obs41],[obs42 ] , we plot on the left the approximation with space adaptivity and on the right the corresponding with uniform partition and the same degrees of freedom . the result obtained using the uniform partition is very poor . the approximate solution misses the angles and peaks , and , in fact , fails to approximate the actual observables . on the other hand , those obtained by adaptivity , appear to be very good approximations . the number of total degrees of freedom in adaptivity corresponds to @xmath399 dof s in each time - slot , while the required dof s in each time slot with uniform partition is more than @xmath400 this final set of experiments , indicates that the a posteriori error estimators can appropriately be used together with adaptive strategies not only for the efficient error control of the wave function @xmath19 , but for the observables as well . the tests suggest that the computational cost is drastically reduced and the adaptive procedure gives encouraging results for small values of the planck constant @xmath15 . however , no rigorous analysis has been provided and further numerical experiments including more general potentials need to be performed in order to draw safe conclusions . this very interesting problem requires further investigation and will be the subject of a forthcoming work . some of the ideas of the theoretical part of the paper are taken from second author s ph.d thesis , @xcite . i.k . is grateful to her ph.d advisor , prof . charalambos makridakis for suggesting the problem and for his academic guidance and support . the authors thank prof.georgios akrivis for many helpful remarks . w. bao , s. jin , p. a. markowich , _ on time - splitting spectral approximations for the schrdinger equation in the semiclassical regime _ , j. comput . phys . * 175 * ( 2002 ) , 487 - 524 . besse , _ a relaxation scheme for the nonlinear schrdinger equation _ , siam j. numer . 42 * ( 2004 ) 934952 . e. bnsch , f. karakatsani , ch . makridakis , _ the effect of mesh modication in time on the error control of fully discrete approximations for parabolic equations _ , to appear in appl . , doi:10.1016/j.apnum.2011.08.008 . p. camiz , a. gerardi , c. marchioro , e. presutti , e. scacciatelli , _ exact solution of a time - dependent quantal harmonic oscillator with a singular perturbation _ , j. mathematical phys . * 12 * ( 1971 ) 20402043 . z. chen , j. feng , _ an adaptive finite element algorithm with reliable and efficient error control for linear parabolic problems _ , math . comp . * 73 * ( 2004 ) 11671193 . r. dautray , j .- l . lions , _ mathematical analysis and numerical methods for science and technology _ , vol . 5 , evolution problems i , 2@xmath401 edition , springer - verlag , berlin , 2000 . a. demlow , o. lakkis , ch . makridakis , _ a posteriori error estimates in the maximum norm for parabolic problems _ , siam j.numer . 47 * ( 2009 ) 21572176 . w. drfler , _ a time - and space - adaptive algorithm for the linear time - dependent schrdinger equation _ , numer . math . * 73 * ( 1996 ) 419448 . feit , j.a . fleck jr . , a. steiger , _ solution of the schrdinger equation by a spectral method _ , j. comput . * 47 * ( 1982 ) 412433 . e. georgoulis , o. lakkis , ch . makridakis , _ a posteriori @xmath402error bounds in finite element approximation of the wave equation_. submitted for publication . p. grard , p. a. markowich , n. j. mauser , and f. poupaud , _ homogenization limits and wigner transforms _ , comm . pure appl . math . * 50 * , ( 1997 ) , 323379 . kreuzer , c.a . mller , a. schmidt , k.g . siebert : _ design and convergence analysis for an adaptive discretization of the heat equation _ , i m a j. numer . , doi : 10.1093/imanum / drr026 . i. kyza , _ a posteriori error analysis for the crank - nicolson method for linear schrdinger equations _ , esaim math . model . * 45 * ( 2011 ) 761778 . i. kyza , _ a posteriori error estimates for approximations of semilinear parabolic and schrdinger - type equations _ , phd thesis , university of crete , 2009 . i. kyza , ch . makridakis , m. plexousakis , _ error control for time - splitting spectral approximations of the semiclassical schrdinger equation _ , i m a j. numer . anal . * 31 * ( 2011 ) 416441 . o. lakkis , ch . makridakis , _ elliptic reconstruction and a posteriori error estimates for fully discrete linear parabolic problems _ , math . comp . * 75 * ( 2006 ) 16271658 . leach , _ invariants and wavefunctions for some time - dependent harmonic oscillator type hamiltonians _ , j. mathematical phys . * 18 * ( 1977 ) 19021907 . a. lozinski , m. picasso , v. prachittham , _ an anisotropic error estimator for the crank - nicolson method : application to a parabolic problem _ , siam j. sci . comput . * 31 * ( 2009 ) 27572783 . makridakis , r.h . nochetto , _ elliptic reconstruction and a posteriori error estimates for parabolic problems _ , siam j. numer . * 41 * ( 2003 ) 15851594 . mansour , w.j.h . kirsten , _ perturbative technique as an alternative to the wkb method applied to the double - well potential _ , j. math . * 23 * ( 1982)18351845 . markowich , p. pietra , c. pohl , _ numerical approximation of quadratic observables of schrdinger - type equations in the semi - classical limit _ , * 81 * ( 1999 ) , 595630 . markowich , p. pietra , c. pohl , h .- stimming , _ a wigner - measure analysis of the dufort - frankel scheme for the schrdinger equation _ , siam j. numer . anal.*40 * ( 2002 ) 12811310 . nochetto , a. schmidt , c. verdi , _ a posteriori error estimation and adaptivity for degenerate parabolic problems _ , math . comp . * 69 * ( 2000 ) 124 . g.a . pozzi , _ problemi di cauchy e problemi ai limiti per equazioni di evoluzione del tipodi schroedinger lineari e non lineari . i. lequazione lineare astratta _ , ( italian ) , ann . mat . pura appl . * 78 * ( 1968 ) 197258 . a. schmidt , k.g . siebert , _ design of adaptive finite element software . the finite element toolbox alberta _ , lecture notes in computational science and engineering , * 42 * , springer - verlag , berlin , 2005 . l.r . scott , s. zhang , _ finite element interpolation of non - smooth functions satisfying boundary conditions _ , math . comp.*54 * ( 1990 ) 483493 . r. verfrth , _ a posteriori error estimates for finite element discretizations of the heat equation _ , calcolo * 40 * ( 2003 ) 195212 .

it is known ( , ) that there is a unique k3 surface which corresponds to a genus curve such that has a shioda - inose structure with quotient birational to the kummer surface of the jacobian of . in this paper we give an explicit realization of as an elliptic surface over with specified singular fibers of type ii and iii . we describe how the weierstrass coefficients are related to the igusa - clebsch invariants of .

math0701669

in this paper , we study some k3 surfaces of high rank . precisely , we consider k3 surfaces which have a shioda - inose structure : that is , an involution which preserves a global @xmath1-form , such that the quotient is a kummer surface . kummer surfaces are a special class of k3 surfaces which are quotients of abelian surfaces . the kummer surface thus carries algebro - geometric information about the abelian surface . it has nron - severi rank at least @xmath5 , and so therefore do the k3 surfaces with shioda - inose structure . these surfaces were first studied by shioda and inose @xcite , who give a description of _ singular _ k3 surfaces , i.e. those with rank @xmath6 , the maximum possible for a k3 surface over a field of characteristic zero . they prove that there is a natural one - to - one correspondence between the set of singular k3 surfaces up to isomorphism and the set of equivalence classes of positive definite even integral binary quadratic forms . the result follows that of shioda and mitani @xcite who show that the set of singular abelian surfaces ( that is , those having picard number @xmath7 ) is also in one - to - one correspondence with the equivalence classes of positive definite even integral binary quadratic forms . the construction of shioda and inose produces a singular k3 surface by taking a double cover of a kummer surface associated to a singular abelian surface and with a specific type of elliptic fibration . the resulting k3 surface has an involution such that the quotient is the original kummer surface . it also turns out that the lattice of transcendental cycles on the k3 surface ( i.e. the orthogonal complement of the nron - severi group in the second singular cohomology group ) is isomorphic to the lattice of transcendental cycles on the abelian surface . nikulin studied groups of automorphisms of k3 surfaces in @xcite , and morrison studied shioda - inose structures more extensively in @xcite , giving necessary and sufficient conditions for a k3 surface to have shioda - inose structure , in terms of the nron - severi group of the k3 surface . the shioda - inose setup takes a k3 surface and produces the data of a kummer surface , or of an abelian surface . one may reverse the question , and ask : given an abelian surface @xmath8 , can we produce a k3 surface with shioda - inose structure such that its quotient under the nikulin involution is birational to @xmath9 ? in particular , when @xmath8 is the jacobian of a curve @xmath2 of genus @xmath1 , galluzzi and lombardo @xcite show that there is a unique k3 surface ( up to isomorphism ) with shioda - inose structure which corresponds to @xmath8 under this construction . dolgachev , in the appendix to the paper , proves an isomorphism between the moduli space of elliptic k3 surfaces with bad fibers of type @xmath10 and @xmath11 , and the moduli space of principally polarized abelian surfaces . however , the isomorphism is not explicit , and uses a result of naruki @xcite to show that the quotient surface constructed is a kummer surface . in this paper we describe explicitly the relation between @xmath2 and the k3 surface @xmath0 with @xmath10 and @xmath11 fibers . it turns out that we can give an alternative elliptic fibration on @xmath0 , with a @xmath1-torsion section . the nikulin involution on @xmath0 is translation by the @xmath1-torsion section . the quotient elliptic surface under the @xmath1-isogeny is the kummer surface of a unique principally polarized abelian surface ( generically , the jacobian of a curve of genus @xmath1 ) . we relate the igusa - clebsch invariants of the genus @xmath1 curve to the moduli of the original k3 elliptic surface , thus giving an explicit description of the map on moduli spaces . this work generalizes a construction of elkies @xcite , who found a family of k3 elliptic surface with two @xmath10 fibers such that the quotient is birational to the kummer surface of a product of two elliptic curves @xmath12 and @xmath13 , and is related to it by a galois - invariant map of degree @xmath14 . earlier , inose @xcite had found the isogenous k3 surface to the kummer surface of @xmath15 , but with maps that were defined over an algebraically closed ground field . clingher and doran have also studied the k3 surfaces corresponding to @xmath15 in @xcite , as well as to @xmath16 more recently . such explicit formulas have some diophantine applications . elkies has used his techniques from @xcite to construct elliptic curves of high rank over @xmath17 and @xmath18 . similarly , he used a k3 surface of rank @xmath19 corresponding to a rational point on a shimura curve to find an elliptic curve of rank at least @xmath20 over @xmath18 . as an application of the results of this paper , we can find explicit parameterizations of some hilbert modular surfaces . we hope to address these techniques and results in a future publication . the outline of the paper is as follows : in section @xmath1 , we recall facts about integral lattices which appear in this context as sublattices of nron - severi groups or cohomology groups of k3 surfaces . in section @xmath21 , we cover the necessary background on k3 surfaces , whereas section @xmath7 recalls the theory of genus @xmath1 curves . in section @xmath22 , we state and prove the main theorem , and in section @xmath23 , we describe a connection to the moduli space of @xmath23 points in @xmath3 . a * lattice * will denote a finitely generated free abelian group @xmath24 equipped with a symmetric bilinear form @xmath25 . an * integral lattice * is a lattice whose form takes values in @xmath26 . we abbreviate the data @xmath27 to @xmath24 sometimes , when the form is understood , and we interchangeably write @xmath28 and @xmath29 for @xmath30 . the * signature * of the lattice is the real signature of the form @xmath31 , written @xmath32 where @xmath33 , @xmath34 and @xmath35 are the number of positive , negative and zero eigenvalues of @xmath31 , counted with multiplicity . we say that the lattice is * non - degenerate * if the form @xmath31 has zero kernel , i.e. @xmath36 . in that case , the signature is abbreviated to @xmath37 . we say @xmath24 is * even * if @xmath38 for all @xmath39 . the dual lattice to @xmath24 is denoted @xmath40 . let @xmath24 be a non - degenerate lattice . the * discriminant * of the lattice is @xmath41 . the lattice is said to be * unimodular * if its discriminant is @xmath42 . note that in the literature , the discriminant is frequently defined to the @xmath43 , in which case a lattice is unimodular iff its discriminant is @xmath44 . for a lattice @xmath24 and a real number @xmath45 , we denote by @xmath46 the lattice which has the same underlying group but with the bilinear form scaled by @xmath45 . the lattice of rank one with a generator of norm @xmath45 will be denoted @xmath47 . by a * root * of a positive definite lattice , we will mean an element @xmath48 such that @xmath49 , whereas for a negative - definite or indefinite lattice , we will mean an element @xmath48 such that @xmath50 . we note here some theorems about the structure and embeddings of lattices . a * root lattice * is a lattice that is generated ( as an abelian group ) by its roots . first , let us introduce some familiar root lattices , through their dynkin diagrams . the subscript in the name of the lattice is the dimension of the lattice , which is also the number of nodes in the dynkin diagram . ( 200,20 ) ( 10,10 ) ( 40,10 ) ( 70,10 ) ( 100,10 ) ( 130,10 ) ( 160,10 ) ( 13,10)(1,0)24 ( 43,10)(1,0)24 ( 73,10)(1,0)4 ( 93,10)(1,0)4 ( 81,10 ) ( 85,10 ) ( 89,10 ) ( 103,10)(1,0)24 ( 133,10)(1,0)24 + @xmath51 , signature @xmath52 , discriminant @xmath53 . this is the positive definite lattice with @xmath54 generators @xmath55 with @xmath56 and @xmath57 if the vertices @xmath58 and @xmath59 are connected by an edge , and @xmath60 otherwise . it may be realized as the set of integral points on the hyperplane @xmath61 . ( 170,30 ) ( 10,5 ) ( 40,5 ) ( 70,5 ) ( 100,5 ) ( 130,5 ) ( 100,25 ) ( 13,5)(1,0)24 ( 43,5)(1,0)24 ( 100,8)(0,1)14 ( 103,5)(1,0)24 ( 73,5)(1,0)4 ( 93,5)(1,0)4 ( 81,5 ) ( 85,5 ) ( 89,5 ) + @xmath62 , signature @xmath52 , discriminant @xmath7 . @xmath63 can be realized as @xmath64 . it has @xmath65 roots . ( 230,30 ) ( 10,5 ) ( 40,5 ) ( 70,5 ) ( 100,5 ) ( 130,5 ) ( 130,25 ) ( 160,5 ) ( 190,5 ) ( 13,5)(1,0)24 ( 43,5)(1,0)24 ( 73,5)(1,0)24 ( 103,5)(1,0)24 ( 133,5)(1,0)24 ( 130,8)(0,1)14 ( 163,5)(1,0)24 + @xmath10 , signature @xmath66 , discriminant @xmath42 . one realization of @xmath10 is as the span of @xmath67 and the all - halves vector @xmath68 . it has @xmath69 roots . ( 200,30 ) ( 10,5 ) ( 40,5 ) ( 70,5 ) ( 100,5 ) ( 130,5 ) ( 100,25 ) ( 160,5 ) ( 13,5)(1,0)24 ( 43,5)(1,0)24 ( 73,5)(1,0)24 ( 103,5)(1,0)24 ( 133,5)(1,0)24 ( 100,8)(0,1)14 + @xmath11 , signature @xmath70 , discriminant @xmath1 . taking the orthogonal complement of any root in @xmath10 gives us @xmath11 . it has @xmath71 roots . ( 170,30 ) ( 10,5 ) ( 40,5 ) ( 70,5 ) ( 100,5 ) ( 130,5 ) ( 70,25 ) ( 13,5)(1,0)24 ( 43,5)(1,0)24 ( 73,5)(1,0)24 ( 103,5)(1,0)24 ( 70,8)(0,1)14 + @xmath72 , signature @xmath73 , discriminant @xmath21 . taking the orthogonal complement of @xmath74 and @xmath75 in @xmath10 , where @xmath76 are roots such that @xmath77 , gives us @xmath72 . it has @xmath78 roots . we let the * nikulin lattice * @xmath79 be the lattice generated by @xmath80 and @xmath81 , with @xmath82 and @xmath83 for @xmath84 . it is isomorphic to @xmath85 . it has signature @xmath86 and discriminant @xmath87 . one checks easily that @xmath79 has 16 roots , namely @xmath88 . in particular , @xmath79 is not a root lattice . let @xmath89 be the hyperbolic plane , i.e. the indefinite rank @xmath1 lattice whose matrix is @xmath90 note that @xmath91 . indefinite lattices are sometimes easier to characterize than definite ones . for instance , one has the following theorem of milnor . [ milnor ] _ ( milnor @xcite ) _ let @xmath24 be an indefinite unimodular lattice . if @xmath24 is even , then @xmath92 for some @xmath93 and @xmath54 . if @xmath24 is odd , then @xmath94 for some @xmath93 and @xmath54 . however , computationally , indefinite lattices are difficult to work with . for instance , it is typically much easier to find roots of a definite lattice than of an indefinite one . in this section we recall the definition and basic properties of k3 surfaces . for more details we refer the reader to @xcite , @xcite and @xcite . let @xmath0 be a smooth projective surface over a field @xmath95 . we say that @xmath0 is a * k3 surface * if @xmath96 and the canonical bundle of x is trivial , i.e. @xmath97 . in the sequel , we will assume that @xmath95 is the field @xmath98 of complex numbers or some subfield of @xmath98 . k3 surfaces are simply connected . from noether s formula @xmath99 it follows that the euler characteristic of @xmath0 is @xmath100 . here @xmath101 and @xmath102 are the chern classes of ( the tangent bundle ) of @xmath0 . therefore the middle cohomology @xmath103 is @xmath104-dimensional . identifying @xmath105 with @xmath26 , we also know that @xmath106 endowed with the cup - product pairing is a unimodular lattice ( that is , the associated quadratic form has discriminant 1 ) because of poincar duality . finally , we also know that @xmath106 is an even lattice ; this follows from wu s formula involving stiefel - whitney classes . finally , using the index theorem @xmath107 we see that @xmath106 has signature @xmath108 . using the classification of milnor , theorem [ milnor ] , we use these properties to conclude that @xmath109 as lattices . now , the exact sequence of analytic sheaves @xmath110 gives , upon taking the long exact sequence of cohomology , an injective map ( the first chern class map ) @xmath111 thus , an analytic line bundle is determined by its image in @xmath112 . for an algebraic k3 surface , we can also say that an algebraic line bundle is determined by class in @xmath113 . linear equivalence , algebraic equivalence , and numerical equivalence all coincide for an algebraic k3 surface . the image of @xmath114 in @xmath115 is a sublattice of @xmath113 , which we call the * nron - severi * group of @xmath0 , and denote by @xmath116 or @xmath117 or @xmath118 . k3 surfaces are khler , and we have a hodge structure @xmath119 in fact , it can be shown that @xmath120 also , @xmath121 since the canonical bundle of @xmath0 is trivial , there exists a regular @xmath122 form on @xmath0 , and we have @xmath123 . it follows that the hodge diamond of @xmath0 is @xmath124 we will only concern ourselves with algebraic k3 surfaces , i.e. those which are nonsingular projective algebraic varieties over @xmath95 . a smooth complete quartic hypersurface in @xmath125 is a k3 surface . a smooth complete intersection of three quadric hypersurfaces in @xmath126 is a k3 surface . a * @xmath127-curve * on a surface is an irreducible rational curve with self - intersection @xmath128 . on a k3 surface , it is an easy exercise using the genus formula to see that curves of self - intersection @xmath128 are exactly the smooth rational curves . here we recall the construction of the kummer surface associated to an abelian surface , which is an example of a k3 surface . let @xmath8 be an abelian surface , and let @xmath129 be the involution which is multiplication by @xmath130 . note that @xmath129 fixes any regular algebraic @xmath1-form on @xmath8 ( the space of @xmath1-forms is one - dimensional since @xmath131 ) . now if we blow up the surface @xmath8 at its sixteen @xmath1-torsion points , we get a surface @xmath132 on which @xmath129 extends to an involution @xmath133 . the quotient surface @xmath134 is in fact a k3 surface , and the regular @xmath1-form is induced from the one on @xmath8 . the surface @xmath135 is called the kummer surface of @xmath8 . it has sixteen disjoint rational curves @xmath136 which are the images of the sixteen exceptional divisors on @xmath132 . the nron - severi lattice of a kummer surface has @xmath137 linearly independent divisor classes coming from the sixteen rational curves above . these generate a negative definite lattice , and there is also a class of a polarization on @xmath9 , since it is projective . therefore its signature is @xmath138 for some @xmath139 . in fact , the nron - severi lattice always contains a particular lattice of signature @xmath140 and discriminant @xmath87 , called the kummer lattice @xmath141 . we describe its structure . the set @xmath142 \cong ( { { \mathbb z}}/2)^4 \cong { { \mathbb f}}_2 ^ 4 $ ] of 16 elements has a natural structure of a vector space of dimension @xmath7 over @xmath143 . choose a labeling @xmath144 ( for instance , by writing @xmath145 for @xmath146 ) and let @xmath147 be the classes of the rational curves corresponding to the blowups at the @xmath1-torsion points . let @xmath148 be the set of @xmath149 elements consisting of @xmath150 affine hyperplanes ( considered as subsets of @xmath151 ) as well as the empty set and all of @xmath151 . this set has the structure of a vector space over @xmath143 , the addition operation being symmetric difference of sets . the set @xmath148 is in fact the reed - muller code @xmath152 : the characteristic functions of the sets in @xmath148 , viewed as functions from @xmath153 to @xmath143 , are exactly the polynomials in @xmath7 variables of degree at most @xmath42 . for every @xmath154 , we have an element @xmath155 of @xmath156 . these vectors actually lie in @xmath157 . the lattice generated by the @xmath158 and the @xmath159 has discriminant @xmath160 and it is called the kummer lattice . in section [ genus2 ] , we will say more about the kummer surface associated to a principally polarized abelian surface which arises as the jacobian of a genus @xmath1 curve . we now describe the data of a shioda - inose structure on a k3 surface . an involution @xmath129 on a k3 surface @xmath0 is called a * nikulin involution * if @xmath161 for every @xmath162 . in fact , a nikulin involution fixes @xmath163 pointwise . every nikulin involution has 8 isolated fixed points . as in the construction of the kummer surface , we may blow up these points to get @xmath164 , which has eight exceptional curves , and an involution @xmath133 . the quotient of @xmath165 by a nikulin involution is a k3 surface @xmath166 . the images of the exceptional curves are @xmath127-curves on the quotient k3 surface @xmath166 . the nron - severi lattice of @xmath166 contains the nikulin lattice @xmath79 : it is generated by vectors @xmath167 ( the classes of the eight disjoint @xmath128-curves ) and @xmath168 , with the form induced by @xmath169 . we say that @xmath0 admits a * shioda - inose structure * if there is a nikulin involution @xmath129 on @xmath0 with rational quotient map @xmath170 such that @xmath166 is a kummer surface and @xmath171 induces a hodge isometry @xmath172 . if @xmath0 has a shioda - inose structure , let @xmath8 be the abelian surface whose kummer surface is @xmath166 . then we have a diagram @xmath173 & & a \ar@{-->}[dl ] \\ & y & } \ ] ] of rational maps of degree @xmath1 , and hodge isometries @xmath174 , thus inducing a hodge isometry @xmath175 . the following theorem of morrison characterizes k3 surfaces with shioda - inose structures in terms of the nron - severi lattice . _ ( morrison @xcite ) _ let @xmath0 be an algebraic k3 surface . the following are equivalent : 1 . @xmath0 admits a shioda - inose structure . 2 . there exists an abelian surface @xmath8 and a hodge isometry @xmath175 . there is a primitive embedding @xmath176 . 4 . there is an embedding @xmath177 . we recall here a few facts about elliptic surfaces . references for these are @xcite and @xcite . an elliptic surface is a smooth projective algebraic surface @xmath0 with a proper morphism @xmath178 to a smooth projective algebraic curve @xmath2 , such that 1 . there exists a section @xmath179 . the generic fiber @xmath180 is an elliptic curve . @xmath181 is relatively minimal . concretely , we will be considering the case @xmath182 , since a k3 surface can not have a non - constant map to a curve of positive genus . we will choose a weierstrass equation for the generic fiber , which is an elliptic curve over the function field @xmath183 , namely @xmath184 where @xmath185 are rational functions of @xmath186 . in fact , by multiplying @xmath48 and @xmath187 by suitable rational functions , we can make @xmath188 polynomials in @xmath186 . furthermore , we may translate @xmath48 by a rational function , and @xmath187 by a @xmath17-linear combination of @xmath42 and @xmath48 . this can be done in such a way that the degree of the discriminant is minimal . we can read out some properties of the surface directly from the weierstrass equation . for instance , if @xmath189 is the arithmetic genus of @xmath0 , then @xmath190 is the minimal @xmath54 such that @xmath191 for @xmath192 . in particular , for a k3 surface @xmath0 , we need to have degree @xmath193 ( the case @xmath194 corresponds to a rational elliptic surface ) . all but finitely many of the fibers of the elliptic surface are nonsingular and hence elliptic curves . tate s algorithm @xcite allows us to compute the description of the singular fibers . each fiber is one of the types shown in the figure below . we note that the reducible fibers are unions of nonsingular rational curves , and they occur in configurations as shown below . the dual graph of the components is an extended dynkin diagram of type @xmath8 , @xmath195 or @xmath180 . the lattices labeled below are the ones generated by the non - identity components of the fiber in @xmath116 . the subscript in the root lattices indicated below is the rank of the lattice , which is also the number of non - identity components . thus , the @xmath196 fiber has @xmath54 components , whereas the @xmath197 fiber has @xmath198 components . ( 350,200 ) ( 15,170)(30,20)[l ] ( 15,180)(25,180)(40,190 ) ( 15,160)(25,160)(40,150 ) ( 0,140)@xmath199 , elliptic curve ( 0,130)(nonsingular fiber ) ( 100,170)(130,170)(140,150 ) ( 100,170)(130,170)(140,190 ) ( 100,140)@xmath200 , a cusp ( 210,160)(230,160)(240,190 ) ( 210,180)(230,180)(240,150 ) ( 210,170)(20,20)[l ] ( 200,140)@xmath201 , a node ( 305,155)(1,0)30 ( 305,185)(1,0)30 ( 315,150)(-1,1)15 ( 325,150)(1,1)15 ( 315,190)(-1,-1)15 ( 325,190)(1,-1)15 ( 305,165 ) ( 305,170 ) ( 305,175 ) ( 335,165 ) ( 335,170 ) ( 335,175 ) ( 280,140)@xmath202 ( @xmath203 ) ( 0,125)(40,105)(0,85 ) ( 40,125)(0,105)(40,85 ) ( 0,75)@xmath204 ( @xmath205 ) ( 105,120)(1,-1)30 ( 105,90)(1,1)30 ( 120,85)(0,1)40 ( 100,75)@xmath206 ( @xmath207 ) ( 200,125)(6,-1)40 ( 200,117)(6,-1)40 ( 200,85)(6,1)40 ( 200,85)(6,1)40 ( 230,85)(0,1)40 ( 200,75)@xmath208 ( @xmath209 ) ( 300,85)(1,0)40 ( 300,90)(1,0)40 ( 300,120)(1,0)40 ( 300,125)(1,0)40 ( 318,110)(1,0)22 ( 318,100)(1,0)22 ( 337,125)(0,-1)17 ( 337,85)(0,1)17 ( 320,103 ) ( 320,105 ) ( 320,107 ) ( 280,75)@xmath210 ( @xmath211 ) ( 15,20)(-1,1)15 ( 15,60)(-1,-1)15 ( 5,25)(1,0)35 ( 5,55)(1,0)35 ( 10,40)(1,0)30 ( 35,20)(0,1)40 ( 15,35)(0,1)10 ( 0,10)@xmath212 ( @xmath72 ) ( 100,30)(1,0)15 ( 100,50)(1,0)15 ( 110,20)(0,1)15 ( 110,60)(0,-1)15 ( 105,25)(1,0)35 ( 105,55)(1,0)35 ( 120,40)(1,0)20 ( 135,20)(0,1)40 ( 100,10)@xmath213 ( @xmath11 ) ( 220,23)(1,0)20 ( 235,20)(0,1)15 ( 200,32)(1,0)40 ( 220,28)(0,1)10 ( 205,28)(0,1)20 ( 200,45)(1,0)20 ( 215,40)(0,1)20 ( 210,55)(1,0)30 ( 235,50)(0,1)10 ( 200,10)@xmath214 ( @xmath10 ) the nron - severi lattice of @xmath0 is generated by the classes of all the sections of @xmath181 ( i.e. the mordell - weil group of @xmath0 ) considered as curves on the surface @xmath0 , together with the class @xmath215 of a fiber , and all the non - identity components of the reducible fibers . let @xmath216 , and for each @xmath217 , let @xmath218 , where @xmath219 is the component which intersects the identity , and the other @xmath220 are the non - identity components . let @xmath221 be the zero section . the intersection pairing satisfies : 1 . for any section @xmath222 , @xmath223 , 2 . @xmath224 , 3 . @xmath225 , 4 . @xmath226 for @xmath227 , 5 . @xmath228 for @xmath229 . the intersection pairing for @xmath220 and @xmath230 is @xmath128 if @xmath231 , and @xmath232 if @xmath84 according to the figures above ( @xmath1 occurs only for types @xmath233 and @xmath204 ) . for a general section @xmath222 , the intersection pairing with each @xmath220 can be computed locally . in particular , for each @xmath234 exactly one of the intersection numbers is @xmath42 , for some @xmath58 such that @xmath235 , and the others vanish . the rank of the nron - severi group is given by the formula @xmath236 the discriminant of the sublattice @xmath237 generated by the non - identity components of all the fibers is @xmath238 , where @xmath239 is the number of multiplicity one components of @xmath240 . here we describe the basic geometry and moduli of curves of genus @xmath1 . for more background we refer the reader to @xcite , @xcite , @xcite , @xcite . let @xmath2 be such a curve defined over a field @xmath95 of characteristic zero . then the canonical @xmath241 bundle of @xmath2 has degree @xmath1 and @xmath242 . that is , the corresponding complete linear system is a @xmath243 ( and it is the unique @xmath243 ) . we therefore have a map @xmath244 which is ramified at @xmath23 points by the riemann - hurwitz formula , and the function field of @xmath2 is a quadratic extension of @xmath245 . therefore , we may write the equation of @xmath2 as @xmath246 the roots of the sextic are the six ramification points of the map @xmath247 . their pre - images on @xmath2 are the six weierstrass points . now , the isomorphism class of @xmath2 over @xmath248 , the algebraic closure of @xmath95 , is determined by the isomorphism class of the sextic @xmath249 , where two sextics are equivalent if there is a transformation in @xmath250 which takes the set of roots ( considered inside @xmath3 ) to the roots of the other . clebsch was the first to determine the invariants of binary sextics . he defined invariants of @xmath251 of weights @xmath252 respectively . that is , @xmath253 has degree @xmath254 in the coefficients of @xmath255 , and if @xmath255 transforms to @xmath256 under the action of @xmath257 , then there is an element @xmath258 such that @xmath259 . clebsch and bolza showed that these invariants determined the sextic up to @xmath248-equivalence . therefore , the point @xmath260 in weighted projective space determines the isomorphism class of @xmath2 . in fact , @xmath2 and @xmath261 are isomorphic over @xmath95 iff there is an @xmath262 such that @xmath263 . igusa generalized clebsch s theory to hold in all characteristics by defining choosing a different algebraic equation for the curve @xmath2 ( through an embedding as a quartic in @xmath264 with one node ) and defining invariants @xmath265 and @xmath266 . he thus obtained a moduli space of genus two curves defined over @xmath267 . the invariants @xmath268 are called the igusa - clebsch invariants whereas @xmath269 are called the igusa invariants of @xmath2 . we will use the former , since we are working over a field of characteristic zero . if the igusa - clebsch invariants of a curve @xmath2 lie in a field @xmath95 , it does not necessarily mean that @xmath2 can be defined over @xmath95 : there is usually an obstruction in @xmath270 . but @xmath2 can always be defined over a quadratic extension of @xmath95 . let @xmath2 be a curve of genus @xmath1 , which we can write as @xmath271 let @xmath272 be the roots of the of the sextic , so that @xmath273 we shall concern ourselves with the embedding of the singular kummer surface as a quartic in @xmath125 , which comes from the complete linear system @xmath274 , twice the theta divisor which defines the principal polarization . we shall use the formulas from @xcite . the quartic is given by the equation @xmath275 where @xmath276 the @xmath137 singular points define ordinary double points on the quartic , which are called nodes . these are given explicitly by the coordinates @xmath277 for @xmath278 . here @xmath279 is defined as follows . let @xmath280 then @xmath281 the singular point @xmath282 comes from the @xmath60 point of the jacobian , whereas the @xmath283 comes from the @xmath1-torsion point which is the difference of divisors @xmath284 - [ ( \theta_j , 0)]$ ] corresponding to two distinct weierstrass points on @xmath2 . the sixteen singular points are called * nodes*. there are also sixteen hyperplanes in @xmath125 which are tangent to the kummer quartic . these are called * tropes*. each trope intersects the quartic in a conic with multiplicity @xmath1 , and contains @xmath23 nodes . conversely , each node is contained in exactly @xmath23 tropes . this beautiful configuration is called the @xmath285 kummer configuration . the explicit formulas for the tropes are as follows . six of the tropes are given by @xmath286 we call this trope @xmath287 . it contains the nodes @xmath282 and @xmath283 . the remaining ten tropes are labeled @xmath288 and correspond to partitions of @xmath289 into two sets of three , say @xmath290 and its complement @xmath291 . set @xmath292 @xmath293 then the equation of @xmath288 is @xmath294 the nron - severi lattice of the nonsingular kummer contains classes of rational curves @xmath295 and @xmath296 coming from the nodes , and @xmath297 and @xmath298 coming from the tropes . we will denote the lattice generated by these as @xmath299 . it has signature @xmath300 and discriminant @xmath87 and is the nron - severi lattice of the kummer surface of a generic principally polarized abelian surface . let @xmath301 be the class of a hyperplane section . furthermore , since @xmath288 only depends on the partition @xmath302 , we may assume @xmath303 and set @xmath304 for @xmath305 . also set @xmath306 for @xmath307 and @xmath308 to agree with the notation in @xcite . we have the following intersection numbers and relations in the nron - severi lattice . @xmath309 where @xmath291 is the complementary set to @xmath310 . we consider the following construction outlined in @xcite . projection to a hyperplane from @xmath282 defines a @xmath1 to @xmath42 map of the kummer to @xmath264 , and thus identifies the kummer as a double cover of @xmath264 , ramified along the union of six lines , which are the projections of the conics @xmath297 ( or the tropes @xmath287 ) . the exchange of sheets gives an involution of the lattice , which acts by @xmath311 we can explicitly write down the projection to @xmath264 as @xmath312 . the involution which is the exchange of sheets is @xmath313 . let @xmath314 be the projections of the @xmath315 . let @xmath0 be an elliptic k3 surface with bad fibers of type @xmath10 and @xmath11 at @xmath316 and @xmath60 respectively . a generic such k3 surface has a nron - severi lattice @xmath317 by shioda s explicit description of the nron - severi lattice of an elliptic surface . this lattice has rank @xmath5 , signature @xmath300 and discriminant @xmath1 . the transcendental lattice @xmath163 has rank @xmath22 , signature @xmath318 and discriminant @xmath1 . we deduce that @xmath319 . the transcendental lattice of a generic principally polarized abelian surface , that is , the jacobian @xmath16 for @xmath2 a generic curve of genus @xmath1 , also satisfies the same property , since the nron - severi of @xmath16 is generated by the theta divisor , which has self - intersection @xmath1 by the genus formula on the abelian surface @xmath320 therefore the orthogonal complement in @xmath321 is exactly @xmath322 . we expect that the elliptic k3 surface @xmath0 has a shioda - inose structure such that the quotient by the nikulin involution gives the kummer surface of a principally polarized abelian surface @xmath323 . galluzzi , lombardo and dolgachev @xcite prove that , in fact , @xmath0 corresponds to a unique @xmath2 up to isomorphism . however , an explicit identification of the quotient as a kummer surface was not known . below , we give an explicit construction of the correspondence . we begin with the k3 surface @xmath0 given by the equation @xmath324 it is easily checked that the surface @xmath0 has an @xmath214 or @xmath10 fiber at @xmath325 and a @xmath213 or @xmath11 fiber at @xmath326 . now we describe another elliptic fibration on @xmath0 . applying the transformation @xmath327 , we get @xmath328 and again replacing @xmath329 by @xmath330 for convenience gives finally @xmath331 which is an elliptic surface over the @xmath48-line with an @xmath332 or @xmath333 fiber at @xmath334 , an @xmath233 or @xmath205 fiber at @xmath335 and a @xmath1-torsion section @xmath336 . the translation by the @xmath1-torsion section is a nikulin involution . we write down the isogenous elliptic surface @xmath166 as @xmath337 this is an elliptic surface over the @xmath48-line with an @xmath338 or @xmath339 fiber at @xmath334 and @xmath233 or @xmath205 fibers at the roots of the sextic @xmath340 , and with a @xmath1-torsion section @xmath341 . the nron - severi lattice of a generic such surface has signature @xmath300 and discriminant @xmath342 . in fact , we will identify it with the nron - severi lattice of a generic kummer surface ( which we call the @xmath285 lattice ) in a later section . this will lead to the identification of the kummer surface of @xmath16 as an elliptic k3 surface with a bad fiber of type @xmath338 at @xmath316 , with @xmath233 fibers at the roots of a sextic derived from @xmath2 , and with a @xmath1-torsion section . if @xmath180 is an elliptic curve with a @xmath1-torsion point , written in the weierstrass form @xmath343 with a @xmath1-torsion point @xmath344 , then the @xmath1-isogenous curve @xmath345 is given by @xmath346 the isogeny @xmath347 is given by @xmath348 the dual isogeny @xmath349 is given by @xmath350 in this section , we state the main theorem , which gives the parameters @xmath351 of the k3 elliptic surface with @xmath10 and @xmath11 fibers ( so far conjecturally ) associated to a genus @xmath1 curve @xmath2 in terms of the igusa - clebsch invariants of @xmath2 . the proof will be given in the following sections . let @xmath2 be a curve of genus two , and @xmath352 the kummer surface of its jacobian . let @xmath251 be the igusa - clebsch invariants of @xmath166 . then there is an elliptic fibration on @xmath166 for which the weierstrass equation may be written @xmath353 there is an elliptic k3 surface @xmath0 given by @xmath354 with fibers of type @xmath10 and @xmath11 at @xmath325 and @xmath355 respectively , and a nikulin involution on @xmath0 , such that the quotient k3 surface is @xmath166 . note that the correspondence of the k3 surface @xmath0 with the genus @xmath1 curve @xmath2 is galois invariant , i.e. @xmath0 is defined over the field of definition of @xmath2 . the proof of the theorem will involve making some non - galois invariant choices ( for the level @xmath1 structure of the jacobian of @xmath2 ) but we will show that @xmath0 as well as the shioda - inose structure on @xmath0 is independent of these choices . the nikulin involution on @xmath0 may be written as follows : @xmath356 in this and the next section , we give the details of how to put an elliptic fibration on the kummer surface of a jacobian of a curve of genus @xmath1 , with a @xmath1-torsion section , a @xmath338 fiber and six @xmath233 fibers . we first identify the nron - severi lattices involved ( namely @xmath299 and @xmath357 ) . we use the construction of @xcite , which gives an embedding of the lattice @xmath358 inside @xmath299 . first , we start with the nron - severi lattice of the k3 surface @xmath0 which has @xmath10 and @xmath11 fibers . the roots of the @xmath116 which correspond to the smooth rational curves on @xmath0 are drawn below ( we use the notation from @xcite ) . these are in fact all the roots corresponding to smooth rational curves on @xmath0 , by a result of nikulin @xcite . ( 350,150 ) ( 10,10 ) ( 10,35 ) ( 10,60 ) ( 10,85 ) ( 10,110 ) ( 310,10 ) ( 310,35 ) ( 310,60 ) ( 310,85 ) ( 310,110 ) ( 40,35 ) ( 70,35 ) ( 100,35 ) ( 130,35 ) ( 160,35 ) ( 190,35 ) ( 220,35 ) ( 250,35 ) ( 280,35 ) ( 12,112)(1,0)296 ( 12,108)(1,0)296 ( 10,13)(0,1)19 ( 10,38)(0,1)19 ( 10,63)(0,1)19 ( 10,88)(0,1)19 ( 310,13)(0,1)19 ( 310,38)(0,1)19 ( 310,63)(0,1)19 ( 310,88)(0,1)19 ( 13,35)(1,0)24 ( 43,35)(1,0)24 ( 73,35)(1,0)24 ( 103,35)(1,0)24 ( 133,35)(1,0)24 ( 163,35)(1,0)24 ( 193,35)(1,0)24 ( 223,35)(1,0)24 ( 253,35)(1,0)24 ( 283,35)(1,0)24 ( 13,5)@xmath359 ( 13,25)@xmath360 ( 15,60)@xmath361 ( 15,85)@xmath362 ( 15,113)@xmath8 ( 25,40)@xmath363 ( 55,40)@xmath364 ( 85,40)@xmath365 ( 115,40)@xmath366 ( 145,40)@xmath367 ( 180,40)@xmath368 ( 205,40)@xmath369 ( 235,40)@xmath370 ( 265,40)@xmath371 ( 295,40)@xmath372 ( 315,10)@xmath373 ( 315,60)@xmath374 ( 315,85)@xmath375 ( 315,110)@xmath376 there is an elliptic fibration on @xmath0 which has @xmath377 as an @xmath214 or @xmath10 fiber , @xmath378 as a @xmath213 or @xmath11 fiber , and @xmath368 as the zero section . this is the fibration over @xmath379 . the fibration over @xmath380 has the @xmath332 or @xmath333 fiber given by @xmath381 , an @xmath233 or @xmath205 fiber @xmath382 , a @xmath1-torsion section ( say @xmath362 ) and a zero section @xmath375 . the nikulin involution @xmath383 is translation by the @xmath1-torsion section . it reflects the above picture about its vertical axis of symmetry . there are two obvious copies of @xmath384 switched by @xmath383 , namely the sublattices of @xmath116 generated by the roots @xmath385 and @xmath386 . next , we write down some roots on @xmath387 , where @xmath166 is the quotient k3 surface of @xmath0 by the involution . as we have described , @xmath166 has six @xmath233 or @xmath205 fibers @xmath388 , a @xmath338 or @xmath339 fiber , namely @xmath389 , with zero section @xmath390 and a @xmath1-torsion section @xmath391 . ( 350,170 ) ( 10,10 ) ( 10,160 ) ( 160,10 ) ( 160,160 ) ( 10,80 ) ( 40,80 ) ( 70,80 ) ( 100,80 ) ( 130,80 ) ( 160,80 ) ( 260,10 ) ( 260,160 ) ( 200,100 ) ( 200,70 ) ( 230,100 ) ( 230,70 ) ( 260,100 ) ( 260,70 ) ( 290,100 ) ( 290,70 ) ( 320,100 ) ( 320,70 ) ( 350,100 ) ( 350,70 ) ( 202,102)(1,1)56 ( 202,68)(1,-1)56 ( 231,102)(1,2)28 ( 231,68)(1,-2)28 ( 260,103)(0,1)54 ( 260,67)(0,-1)54 ( 289,102)(-1,2)28 ( 289,68)(-1,-2)28 ( 318,102)(-1,1)56 ( 318,68)(-1,-1)56 ( 348,102)(-3,2)85 ( 348,68)(-3,-2)85 ( 163,10)(1,0)94 ( 163,160)(1,0)94 ( 13,80)(1,0)24 ( 43,80)(1,0)24 ( 73,80)(1,0)24 ( 103,80)(1,0)24 ( 133,80)(1,0)24 ( 10,13)(0,1)64 ( 160,13)(0,1)64 ( 10,83)(0,1)74 ( 160,83)(0,1)74 ( 198,72)(0,1)26 ( 202,72)(0,1)26 ( 228,72)(0,1)26 ( 232,72)(0,1)26 ( 258,72)(0,1)26 ( 262,72)(0,1)26 ( 288,72)(0,1)26 ( 292,72)(0,1)26 ( 318,72)(0,1)26 ( 322,72)(0,1)26 ( 348,72)(0,1)26 ( 352,72)(0,1)26 ( 15,10)@xmath392 ( 15,160)@xmath393 ( 15,87)@xmath394 ( 35,87)@xmath395 ( 65,87)@xmath396 ( 95,87)@xmath397 ( 125,87)@xmath398 ( 145,87)@xmath399 ( 145,10)@xmath400 ( 140,160)@xmath401 ( 265,5)@xmath402 ( 265,160)@xmath403 ( 180,70)@xmath404 ( 210,70)@xmath405 ( 240,70)@xmath406 ( 270,70)@xmath407 ( 300,70)@xmath408 ( 330,70)@xmath409 ( 180,100)@xmath410 ( 210,100)@xmath411 ( 240,100)@xmath412 ( 270,100)@xmath413 ( 300,100)@xmath414 ( 330,100)@xmath415 it is easily checked that the rational components of the @xmath10 fiber described above map as following : @xmath416 ( recall that @xmath375 is the zero section of the @xmath333 fibration , on which the quotient map is an isogeny of elliptic surfaces ) , @xmath417 . hence , we see a natural copy of @xmath10 within the nron - severi lattice of @xmath166 . on the other hand , we can also see eight roots orthogonal to all the generators of @xmath10 as well as to each other , namely @xmath418 and @xmath392 . now , we use the construction of naruki @xcite which gives an explicit embedding of @xmath358 inside the nron - severi lattice of a kummer surface of a generic principally polarized abelian surface , or @xmath299 . we extend this construction to get an identification of @xmath299 with @xmath387 , i.e. the lattice generated by the roots in the diagram above . the identification is as follows : ( 350,170 ) ( 10,10 ) ( 10,160 ) ( 160,10 ) ( 160,160 ) ( 10,80 ) ( 40,80 ) ( 70,80 ) ( 100,80 ) ( 130,80 ) ( 160,80 ) ( 260,10 ) ( 260,160 ) ( 200,100 ) ( 200,70 ) ( 230,100 ) ( 230,70 ) ( 260,100 ) ( 260,70 ) ( 290,100 ) ( 290,70 ) ( 320,100 ) ( 320,70 ) ( 350,100 ) ( 350,70 ) ( 202,102)(1,1)56 ( 202,68)(1,-1)56 ( 231,102)(1,2)28 ( 231,68)(1,-2)28 ( 260,103)(0,1)54 ( 260,67)(0,-1)54 ( 289,102)(-1,2)28 ( 289,68)(-1,-2)28 ( 318,102)(-1,1)56 ( 318,68)(-1,-1)56 ( 348,102)(-3,2)85 ( 348,68)(-3,-2)85 ( 163,10)(1,0)94 ( 163,160)(1,0)94 ( 13,80)(1,0)24 ( 43,80)(1,0)24 ( 73,80)(1,0)24 ( 103,80)(1,0)24 ( 133,80)(1,0)24 ( 10,13)(0,1)64 ( 160,13)(0,1)64 ( 10,83)(0,1)74 ( 160,83)(0,1)74 ( 198,72)(0,1)26 ( 202,72)(0,1)26 ( 228,72)(0,1)26 ( 232,72)(0,1)26 ( 258,72)(0,1)26 ( 262,72)(0,1)26 ( 288,72)(0,1)26 ( 292,72)(0,1)26 ( 318,72)(0,1)26 ( 322,72)(0,1)26 ( 348,72)(0,1)26 ( 352,72)(0,1)26 ( 15,10)@xmath419 ( 15,160)@xmath420 ( 15,87)@xmath421 ( 35,87)@xmath422 ( 65,87)@xmath423 ( 95,87)@xmath424 ( 125,87)@xmath425 ( 145,87)@xmath426 ( 140,10)@xmath427 ( 140,160)@xmath428 ( 265,5)@xmath429 ( 265,160)@xmath430 ( 180,70)@xmath431 ( 210,70)@xmath432 ( 240,70)@xmath433 ( 270,70)@xmath434 ( 300,70)@xmath435 ( 330,70)@xmath436 ( 180,100)@xmath437 ( 210,100)@xmath74 ( 240,100)@xmath75 ( 270,100)@xmath438 ( 300,100)@xmath439 ( 330,100)@xmath440 here @xmath441 . the class of the fiber is @xmath442 and @xmath443 are given by @xmath444 notice that under the simple transposition @xmath445 of indices we have the permutation of fibers @xmath446 and in fact @xmath447 , @xmath448 . next , we describe how to use all this information from the nron - severi group to construct @xmath48 , @xmath187 and @xmath186 in the weierstrass equation for @xmath352 @xmath449 consider the class of the fiber @xmath450 , given by @xmath451 we can write down the parameter on the base by computing explicitly the sections of @xmath452 . this linear system consists of ( the pullback of ) quintics passing through the points @xmath453 and @xmath454 which pass through @xmath455 , having a double point at @xmath456 and a triple point at @xmath457 . this linear system is @xmath1-dimensional , and taking the ratio of two linearly independent sections gives us the parameter @xmath186 on the base , @xmath3 , for the elliptic fibration . now , @xmath186 is only determined up to the action of @xmath458 , but the first restriction we make is to put the @xmath338 fiber at @xmath325 , which fixes @xmath186 up to affine linear transformations . any elliptic k3 surface with a @xmath1-torsion section can be written in the form @xmath459 with @xmath460 of degree at most @xmath461 and @xmath462 of degree at most @xmath7 . the @xmath1-torsion section is @xmath463 . the discriminant of this elliptic surface is a multiple of @xmath464 . in fact , we see that @xmath465 must have degree exactly @xmath23 , because there are six @xmath233 fibers that the zero and @xmath1-torsion sections meet in different components . the positions @xmath466 of the @xmath233 fibers are the roots of the polynomial @xmath467 . now @xmath186 is determined up to transformations of the form @xmath468 . to have exactly a @xmath469 fiber at @xmath316 , we must have @xmath470 where @xmath462 is a monic cubic polynomial and @xmath471 is a _ linear _ polynomial in @xmath186 . we can further fix @xmath186 up to scalings @xmath472 by translating @xmath186 so that the quadratic term of @xmath462 vanishes . we notice that the top coefficient @xmath282 of @xmath460 is a square , and so by scaling @xmath473 appropriately , we may assume @xmath474 , i.e. that @xmath460 and @xmath462 are monic . now we describe how to obtain @xmath48 . it is a weil function , so that the horizontal component of its divisor equals @xmath475 , and the vertical component is uniquely determined by that fact that @xmath476 is linearly ( and hence numerically ) equivalent to zero . so we deduce that the divisor of @xmath48 is @xmath477 , where @xmath478 is the @xmath339 fiber . to convert this to formulas , we compute the functions which cut out @xmath479 , @xmath480 , @xmath237 and @xmath221 . there is a quintic @xmath481 which cuts out @xmath482 . now , notice that the @xmath339 fiber contains @xmath483 and @xmath426 . therefore @xmath481 is divisible by @xmath484 and @xmath485 . we write @xmath486 with a quadratic @xmath487 . next , we know that @xmath488 cuts out @xmath429 and @xmath489 cuts out @xmath490 . to find , for instance , the function which cuts out @xmath75 , we find the quadratic ( unique up to constants ) which passes through @xmath491 . call this function @xmath75 , by abuse of notation . similarly , we find @xmath492 . we also note that the factor of @xmath489 in the numerator of @xmath48 , which gives a zero along @xmath237 , also gives a zero along @xmath493 owing to the fact that @xmath494 intersects @xmath495 ( recall that we are working with the singular kummer surface , on which the image of the curve @xmath495 is just a single point ) . putting everything together , we can write @xmath48 up to scaling as a quotient of two homogeneous polynomials of degree @xmath137 as follows : @xmath496 finally , we have to scale @xmath48 and @xmath186 so that @xmath497 becomes a square of a function @xmath187 on the kummer . we note that in the equation of the kummer @xmath498 we can complete the square for @xmath499 to obtain @xmath500 we let @xmath187 be a constant multiple of @xmath501 a quotient of two homogenous polynomials of degree @xmath502 , and verify that this makes the weierstrass equation hold . the computation is carried out in a maxima program , which is available from the arxiv.org e - print archive . this paper is available as ` math.ag/0701669 ` . to access the auxiliary file download the source file for the paper . that will produce not only the latex files for the paper but also the computer algebra code . the code takes about half an hour to run on a @xmath503 ghz computer . we noted earlier that the permutation @xmath504 on the @xmath205 fibers by @xmath446 , and takes @xmath505 to @xmath506 . that is , it switches the components intersecting the identity and @xmath1-torsion sections as well . in addition , it switches the zero section @xmath507 and the @xmath1-torsion section @xmath508 , and on the @xmath339 fibers it switches the two near leaves @xmath428 and @xmath427 , namely , again the components intersecting @xmath237 and @xmath221 . on the other hand , consider the action on @xmath387 induced by the translation by @xmath237 . under this map , @xmath237 and @xmath221 get swapped , the @xmath1-torsion and identity components of the @xmath339 and @xmath205 fibers all get switched , and the far leaves of the @xmath339 fiber also get switched ( this can be seen , for instance , from the fact that the simple components of the special @xmath339 fiber form a group compatible with the group law on the generic elliptic curve ) . the locations of the @xmath205 fibers themselves are fixed . therefore the effect of the permutation @xmath509 is the same as translation by @xmath1-torsion composed with a pure involution @xmath510 of the @xmath205 fibers and a switch of the far leaves of the @xmath339 fiber . since the far leaves of the @xmath339 fiber are switched by the galois involution that multiplies the square root of @xmath511 by @xmath130 , this tells us that we have the correct quadratic twist , since @xmath512 is within a square factor of the discriminant of the sextic . that is , making a different choice of @xmath1-level structure while describing the fibration on the kummer surface @xmath166 would have given us the same answer for @xmath0 ! the construction above gave us a correspondence of sextics @xmath513 and @xmath514 therefore , over an algebraically closed field , we get a birational map from the moduli space of @xmath23 points in @xmath3 ( i.e. the quotient of @xmath515 under the action of @xmath458 and @xmath516 ) with the space of roots up to scaling of @xmath517 as @xmath518 vary ( we suppressed @xmath519 since it just scales @xmath520 and @xmath521 ) . this latter space is cut out inside @xmath522 by the hyperplane @xmath523 and the quartic hypersurface @xmath524 , where @xmath525 and @xmath526 are the second and fourth elementary symmetric functions of the @xmath527 . thus , we get a model as a singular quartic threefold in @xmath528 , which is known in the literature as the igusa quartic . here the igusa quartic arises naturally in the context of shioda - inose structures on k3 surfaces . there is no simple one - one correspondence between the roots of @xmath249 and @xmath529 , since the two actions of @xmath516 acting by the permutation representation on the six roots of @xmath249 on the six roots of @xmath529 are related by an outer automorphism . to see this , we recall from the last section that the permutation @xmath504 on the roots of @xmath249 ( the weierstrass points ) acts on the roots of @xmath529 ( which are the locations of the @xmath205 fibers ) by the permutation @xmath510 . by symmetry , all the transpositions of @xmath516 act by a product of three transpositions on the roots of @xmath529 . thus we get a homomorphism @xmath530 which is an outer automorphism . i thank noam elkies and barry mazur for suggesting the study of k3 surfaces as a thesis topic and for many insightful remarks and helpful conversations . i also thank tetsuji shioda for discussions and correspondence on elliptic surfaces , and igor dolgachev for helpful remarks and pointing out some useful references . i also thank charles doran for bringing his work with adrian clingher to my attention . j. w. f. cassels and e. v. flynn , _ prolegomena to a middlebrow arithmetic of curves of genus @xmath1 _ , london mathematical society lecture note series , 230 . cambridge university press , cambridge , 1996 . _ gomtrie des surfaces @xmath531 : modules et priodes . ( french ) [ geometry of @xmath531 surfaces : moduli and periods ] _ , papers from the seminar held in palaiseau , october 1981january 1982 . astrisque no . 126 ( 1985 ) . socit mathmatique de france , paris , 1985 . n. mitani and t. shioda , _ singular abelian surfaces and binary quadratic forms _ , classification of algebraic varieties and compact complex manifolds , pp . lecture notes in math . 412 , springer , berlin , 1974 . mestre , _ construction de courbes de genre @xmath1 partir de leurs modules _ , effective methods in algebraic geometry ( castiglioncello , 1990 ) , 313334 , progr . math . 94 , birkhuser boston , boston , ma , 1991 . j. milnor , _ on simply connected @xmath7-manifolds _ , symposium internacional de topologa algebraica ( international symposium on algebraic topology ) , universidad nacional autnoma de mxico and unesco , mexico ( 1958 ) pp . 122128 v. v. nikulin , _ quotient - groups of groups of automorphisms of hyperbolic forms by subgroups generated by @xmath1-reflections . algebro - geometric applications _ , current problems in mathematics , v. 18 , pp . 3114 , akad . nauk sssr , vsesoyuz . i tekhn . informatsii , moscow , 1981 ; j. soviet . * 22 * ( 1983 ) , 14011476 . j. tate , _ algorithm for determining the type of a singular fiber in an elliptic pencil _ , modular functions of one variable , iv ( proc . internat . summer school , univ . antwerp , antwerp , 1972 ) , pp . lecture notes in math . 476 , springer , berlin , 1975 .

to analyze the physical properties arising from indirect magnetic exchange between several magnetic adatoms and between complex magnetic nanostructures on metallic surfaces , the real - space extension of dynamical mean - field theory ( r - dmft ) appears attractive as it can be applied to systems of almost arbitrary geometry and complexity . while r - dmft describes the kondo effect of a single adatom exactly , indirect magnetic ( rkky ) exchange is taken into account on an approximate level only . here , we consider a simplified model system consisting of two magnetic hubbard sites ( `` adatoms '' ) hybridizing with a non - interacting tight - binding chain ( `` substrate surface '' ) . this two - impurity anderson model incorporates the competition between the kondo effect and indirect exchange but is amenable to an exact numerical solution via the density - matrix renormalization group ( dmrg ) . the particle - hole symmetric model at half - filling and zero temperature is used to benchmark r - dmft results for the magnetic coupling between the two adatoms and for the magnetic properties induced in the substrate . in particular , the dependence of the local adatom and the nonlocal adatom - adatom static susceptibilities as well as the magnetic response of the substrate on the distance between the adatoms and on the strength of their coupling with the substrate is studied . we find both , excellent agreement with the dmrg data even on subtle details of the competition between rkky exchange and the kondo effect but also complete failure of the r - dmft , depending on the parameter regime considered . r - dmft calculations are performed using the lanczos method as impurity solver . with the real - space extension of the two - site dmft , we also benchmark a simplified r - dmft variant .

1204.5111

the rapidly improving experimental techniques to probe magnetic adatoms on non - magnetic surfaces allow for direct studies of fundamental magnetic exchange mechanisms on an atomic scale . besides access to the structural and the electronic properties of such adatoms and of the underlying substrate for a given system , the construction of tailored magnetic model systems represents an exciting perspective . @xcite magnetic structures of nanometer size provide extremely small systems suitable to store and to transport information and may realize efficient nano spintronics devices . @xcite the competition between an indirect magnetic exchange of the adatoms via the substrate electrons on the one hand and the screening of the adatom magnetic moment by the conduction - band electrons of the substrate on the other represents a prominent example for a physical problem becoming accessible to new real - space techniques . the scanning tunneling microscope ( stm ) @xcite has been used to investigate the kondo physics @xcite of single magnetic adatoms @xcite and the magnetic properties of the individual magnetic islands@xcite on non - magnetic substrates . using stm , it is possible to investigate the direct magnetic interaction of atom pairs . @xcite indirect magnetic exchange , i.e. the ruderman - kittel - kasuya - yosida ( rkky ) interaction @xcite between two adatoms , was detected through the kondo effect . @xcite a direct real - space study of the rkky coupling , however , comes in reach with spin - polarized scanning - tunneling spectroscopy only . @xcite ( orange ) at a distance @xmath0 coupled via a hybridization term of strength @xmath1 to a one - dimensional `` substrate '' consisting of @xmath2 non - interacting sites ( blue ) with nearest - neighbor hopping @xmath3 . in the r - dmft , the tiam is self - consistently mapped onto two single - impurity anderson models which are solved independently by means of exact diagonalization ( ed ) to get the local self - energies . these are used to set up the tiam dyson equation the solution of which gives the local green s functions which define via the r - dmft self - consistency conditions the parameters of the impurity models ( see text for details ) . , scaledwidth=40.0% ] the most simple model which captures this competition is displayed schematically in fig . [ fig : schematic_tiam ] . here the electronic and magnetic properties of a magnetic adatom are modeled by a single non - degenerate orbital . a local magnetic moment is formed by a strong local hubbard interaction . the adatom orbital hybridizes with a valence orbital of the nearest - neighboring substrate atom . the substrate electronic structure itself is modeled by a tight - binding valence band resulting from non - degenerate and uncorrelated orbitals on a bipartite lattice with nearest - neighbor hopping . considering two adatoms yields a variant of the two - impurity anderson model ( tiam ) @xcite in a surface geometry . the main goal of our study is to benchmark the real - space variant @xcite of the dynamical mean - field theory @xcite that can be employed for theoretical studies of the electronic and magnetic properties of a single , of two and or of more magnetic adatoms in different geometries on metal surfaces . for this purpose the tiam represents a fundamental starting point . in the kondo limit of the tiam , charge fluctuations on the adatom site are largely suppressed , and the adatom spin @xmath4 couples antiferromagnetically to the local spin at the nearest - neighboring substrate site @xmath5 via a spin - spin coupling @xmath6 given by the local exchange @xmath7 . the interplay between the kondo effect and the rkky interaction has extensively been studied in the kondo limit of the tiam or in the two - impurity kondo model by different analytical as well as numerical techniques . @xcite the physics is governed by two energy scales , the nonlocal indirect magnetic interaction @xmath8 and the kondo temperature @xmath9 below which the magnetic moment of the adatom ( impurity ) is screened locally . in the kondo regime for @xmath10 , the conventional picture is that the local magnetic moments at the two impurities are individually screened by forming local singlet states with two kondo clouds of itinerant electron spins from the substrate ( conduction band ) . for large @xmath11 , on the other hand , and in the antiferromagnetic case @xmath12 , the two adatom spins form a nonlocal singlet state and there is no kondo effect . if @xmath13 is ferromagnetic and large as compared to @xmath14 , a nonlocal spin - triplet state is formed . this may subsequently be kondo screened . in the generic case and as a function of @xmath15 there is no quantum phase transition but a smooth crossover from the rkky regime at weak @xmath15 to the kondo regime at strong @xmath15 . for a dense system , i.e. the kondo or anderson lattice model , a static mean - field approach would sharpen this to a phase transition . @xcite dynamical mean - field theory is a comprehensive , thermodynamically consistent and non - perturbative approximation for correlated lattice - fermion models . @xcite dmft treats the kondo effect exactly . on the other hand , one has to tolerate an approximate treatment of the effects of the rkky interaction . it is important to note that there is no approximation of rkky coupling itself : integrating out the non - interacting substrate degrees of freedom , the effective second - order - in-@xmath15 rkky coupling , @xmath16 , is given in terms of the nonlocal static susceptibility of the substrate . it was pointed out by peters and pruschke @xcite that @xmath17 is still finite but reduces to an interaction between nearest neighbors for the case of a lattice in infinite spatial dimensions where the dmft becomes exact . for finite dimensions , it is a long - ranged and oscillating function of the distance @xmath18 . what is neglected in fact for a finite - dimensional lattice , is the feedback of nonlocal , e.g. magnetic correlations , which result from the nonlocal rkky coupling , on the self - energy and thus on the one - particle green s function . this is a rather subtle approximation the quality of which can be estimated by concrete numerical calculations only . the same argumentation holds for the real - space dmft ( r - dmft ) . @xcite r - dmft generalizes the standard dmft to systems with missing or reduced translational symmetry by self - consistently mapping the original ( lattice ) model to a set of single - impurity anderson models ( siam ) , one for each of the geometrically or electronically inequivalent sites . even for the tiam ( see fig . [ fig : schematic_tiam ] ) , this real - space generalization is necessary if one wants to apply dmft in order to test the local approximation for the self - energy . previous applications of the r - dmft concentrated on the mott metal - insulator transition at surfaces and in thin films,@xcite on surface effects in correlated fermi liquids , @xcite on multilayered nanostructures , heterostructures and interfaces , @xcite on disordered systems , @xcite as well as on ultracold atomic gases in optical lattices with harmonic confinement . @xcite it has not been employed , however , to study the effects of the indirect magnetic exchange . the main purpose of the present study is to apply the r - dmft to the particle - hole symmetric tiam at half - filling and zero temperature and to study the magnetic response , i.e. different static magnetic susceptibilities , by applying a weak local magnetic field to one of the adatoms . calculations are performed as a function of the distance between the adatoms and as a function of the hybridization strength @xmath1 to cross over from the kondo to the rkky regime . to test the reliability of the dynamical mean - field approach , the substrate electronic structure is modeled as a one - dimensional tight - binding chain ( see fig . [ fig : schematic_tiam ] ) . the resulting essentially one - dimensional model is accessible to the density - matrix renormalization group ( dmrg ) . @xcite extensive comparison with numerically exact dmrg results obtained from an implementation based on matrix - product states , @xcite and operators along the lines described in ref . helps to benchmark the mean - field approach . our intention is that , by comparing with dmrg , the strengths but also the mean - field artifacts of r - dmft become more transparent . a failure of r - dmft for the weak - coupling limit , where non - local correlations due to the rkky coupling are strong , can be expected from the very beginning . however , there are several interesting questions left , e.g. : where precisely are the limits of the mean - field approach ? how does a failure of the approach manifest itself in the observables ? which physical effects are accessible to a description on the r - dmft level ? to what extent can the physics be reproduced quantitatively in the strong - coupling limit ? such benchmarking of the r - dmft , at the level of the two - impurity anderson model , will be important for future studies of similar systems in higher spatial dimensions , with more correlated adatoms forming more complex geometries such as chains or clusters etc . by choosing the one - dimensional two - impurity anderson model at half - filling , the above - mentioned questions are tackled in a situation that is very unfavorable for r - dmft . the benchmark will thus serve as a `` lower limit '' for the applicability of r - dmft for future applications . our interest in the r - dmft approach to study magnetic nanostructures on surfaces results from its extremely large flexibility . opposed to dmrg , for example , the r - dmft is able to investigate inhomogeneous systems in arbitrary geometries in higher dimensions . while this is actually characteristic for any mean - field approach , the r - dmft is distinguished by the fact that it is non - perturbative and thermodynamically consistent . to account for the effects of short - range correlations , the theory can be improved by certain cluster extensions , such as cellular dmft . @xcite this is , in principle , also conceivable for complicated inhomogeneous geometries but requires further methodical advances as there is no straightforward tiling of the lattice in most cases . the paper is organized as the follows : the next sections introduces the model , notations and quantities of interest . sec . [ rdmft ] and sec . [ dmrg ] briefly describe our real - space dmft and our dmrg approach to the problem , respectively . results of both approaches are presented , compared and discussed in detail in sec . [ results ] . finally , sec . [ con ] concludes the paper . the hamiltonian of the two - impurity anderson model @xcite displayed in fig . [ fig : schematic_tiam ] is given by : @xmath19 here @xmath20 and @xmath21 create an electron with spin projection @xmath22 at the adatom sites @xmath23 or at the substrate sites @xmath24 , respectively . @xmath25 and @xmath26 denote the corresponding occupation - number operators . the spin - summed occupation at one of the adatom sites and at one of the substrate sites are given by @xmath27 and @xmath28 , respectively . the hopping amplitude @xmath29 between neighboring substrate lattice sites is used to fix the energy unit , i.e. @xmath3 . @xmath1 is the hybridization between an adatom site @xmath30 and the nearest - neighbored substrate lattice site which is denoted by @xmath31 . @xmath32 and @xmath33 are the on - site hubbard interaction and the local on - site energy for the adatom sites . @xmath34 is chemical potential . in all our calculations we consider the particle - hole symmetric case with @xmath35 and @xmath36 where the system is half - filled , i.e. where the average occupation numbers in thermal equilibrium are given by @xmath37 and @xmath38 for both @xmath30 and all @xmath39 . the magnetic properties of the system are best characterized by site - dependent local and nonlocal susceptibilities . we consider the adatom - adatom susceptibilities , @xmath40 i.e. the local adatom susceptibilities @xmath41 for @xmath23 and the inter - adatom susceptibility @xmath42 . these provide information on the local adatom magnetic moment and , most important , on the indirect magnetic coupling . further , we are interested in the linear magnetic response of the substrate which is accessible via the adatom - substrate susceptibilities @xmath43 here @xmath44 and @xmath45 , with @xmath46 and @xmath47 , are magnetic moments on the adatom site @xmath30 and on the substrate lattice site @xmath39 respectively . furthermore , the imaginary - time dependence of an operator @xmath48 is given by @xmath49 in our calculations the susceptibilities , eqs . ( [ imp_susceptibility ] ) and ( [ sub_susceptibility ] ) , are computed as a numerical derivative with respect to a local magnetic field of strength @xmath50 coupling as @xmath51 to the hamiltonian . calculations are done for zero temperature in the present study . nevertheless , the formalism is set up for arbitrary finite @xmath52 below . the main task is to compute , for a finite but weak field @xmath53 , the spin - dependent average occupation numbers . these can be obtained via @xmath54 from the local single - electron adatom green s function @xmath55 and the local substrate green s function @xmath56 given at the fermionic matsubara frequencies @xmath57 . the local green s functions are the diagonal elements of the green s function matrix @xmath58 . the latter can be obtained from the real - space dyson equation : @xmath59 where @xmath60 is the unity matrix , @xmath61 the diagonal local energy matrix , and @xmath62 is the hopping matrix . @xmath61 also includes the field term and is thus possibly spin - dependent . @xmath62 not only includes the hopping @xmath29 between substrate sites but also hopping @xmath1 between the substrate and the adatom sites . for a system with @xmath2 substrate sites and two adatoms , the matrix dimension is @xmath63 for each spin direction @xmath64 . as there is a local hubbard interaction on the adatom sites only , the self - energy @xmath65 is a @xmath63-dimensional matrix with non - zero elements @xmath66 in the @xmath67 adatom - sites block only . hence , eq . ( [ green1 ] ) can be written as : @xmath68 where @xmath69 , and where @xmath70 is the @xmath71 hybridization matrix including hopping between adatoms and substrate only . its non - zero elements are given by @xmath72 . further , @xmath73 is the non - interacting substrate green s function matrix . in case of periodic boundary conditions , its elements are @xmath74 where @xmath75 with @xmath76 , while for open boundary conditions @xmath77 where @xmath78 with @xmath79 . in both cases @xmath80 is non - interacting dispersion . generally , to calculate , for a given self - energy , the local green s functions , one has to numerically invert the matrix given by eq . ( [ green1 ] ) . in our case for the tiam , however , it is possible to `` integrate out '' the substrate degrees of freedom and to find analytical expressions which substantially reduce the numerical effort : using the identity @xmath81 with @xmath82 which is valid for arbitrary quadratic matrices @xmath83 and @xmath84 and arbitrary rectangular matrices @xmath85 and @xmath84 , we find @xmath86 where @xmath87 is the @xmath88 adatom green s function matrix and @xmath89 is the @xmath90 substrate green s function matrix . the remaining task thus consists in the inversion of a @xmath67 matrix @xmath91 where @xmath92 is the hybridization function . this is readily done : @xmath93 this provides us with the local adatom green s functions in particular and , using eq . ( [ green_gsub ] ) , with the local green s functions for each substrate site via : @xmath94 to complete the theory , we need the self - energy matrix @xmath66 . this requires an approximation . within real - space dmft , @xcite the self - energy is obtained by considering weak - coupling perturbation theory in @xmath32 to all orders and by summing all local diagrams in the skeleton - diagram expansion of the self - energy , @xmath95 $ ] . this implies that the resulting self - energy is local : @xmath96 but possibly site - dependent . for correlated lattice models with full translational symmetries , the approach reduces to the conventional dmft . @xcite as in the conventional dmft , the local diagrams are not summed explicitly , the problem is rather reformulated by introducing a self - consistent mapping onto an effective single - impurity problem . here , however , the self - consistent cycle is more complicated since a lattice model with @xmath97 geometrically or electronically inequivalent sites has to be self - consistently mapped onto a set of @xmath97 effective single - impurity models . in our case we have to consider at most @xmath98 single - impurity anderson models ( see fig . [ fig : schematic_tiam ] ) : we start with a guess for the local self - energies @xmath99 , i.e. for @xmath100 . this is used in the dyson equation of the lattice model to compute the green s function matrix , and in particular the local elements of the green s function matrix at the correlated sites . in our case , we can profit from eqs . ( [ gimp1 ] ) and ( [ gimp2 ] ) to get the local adatom green s functions @xmath101 directly . the r - dmft self - consistency conditions , @xmath102 then provide us with the weiss green s functions @xmath103 for @xmath104 , i.e. with the non - interacting green s functions of the @xmath97 effective impurity models . these can be written as @xmath105 . the one - particle parameters of each effective siam , the one - particle energies of the bath sites as well as the corresponding hybridization strengths , are found from the poles and the residues of the corresponding hybridization function @xmath106 ( which should not be mixed up with @xmath107 , see eq . ( [ hyb ] ) ) . once the effective impurity models are fixed , the crucial step consists in the solution of the models which can be done independently for any @xmath104 . this yields the self - energies @xmath99 and thus closes the self - consistency cycle ( fig . [ fig : schematic_tiam ] ) . this procedure is iterated until converged self - energies are obtained . as an impurity solver to get the self - energy @xmath99 of the @xmath30-th siam we use the exact - diagonalization ( ed ) method . @xcite here , a finite small number @xmath108 of auxiliary bath degrees of freedom in the effective siam is considered . we use full diagonalization with @xmath109 and the lanczos method @xcite with @xmath110 and @xmath111 . exploiting the fact that the total particle number and the @xmath112-component of the total spin are conserved quantities , the diagonalization can be done in smaller invariant subspaces of the full hilbert space . all calculations have been done , if not stated differently , with @xmath111 . for a given weiss green s function , the one - particle bath parameters of the siam are found by a minimization procedure on the imaginary - frequency axis as described in ref . using high - frequency cutoff of the order of @xmath32 and low - frequency cutoff specified by the _ temperature @xmath113 . with the latter we can formally work in the finite - temperature matsubara framework as outlined above . on the other and , the value of the fictitious temperature chosen is clearly lower than the smallest energy scale that can be accessed by means of the ed solver for @xmath114 . we have regularly checked that the results do not significantly depend on the cutoff and on @xmath115 . the computational effort of the r - dmft scheme roughly scales linearly with the number of impurity models , i.e. with the number of inequivalent sites in the original system . while here we focus on the @xmath98 case for benchmarking purposes , future applications are intended that address systems with up to @xmath116 inequivalent magnetic atoms . we expect that those applications can still be performed conveniently using ed as a solver . it might nevertheless be interesting to have a scheme at hand that is considerably faster . here , the two - site dmft @xcite represents an alternative . the main idea is the use a single bath degree of freedom only , as in the so - called linearized dmft @xcite for the mott - hubbard transition . the resulting effective two - site impurity model is readily solved . on the other hand , the dmft self - consistency condition can obviously no longer be satisfied exactly or to a high level of accuracy as in the ed approach with , say , @xmath111 sites . it has therefore been suggested @xcite to take into account the leading orders in systematic expansions of the self - consistency condition for high and for low frequencies only . this results in a simplified but extremely fast approach which is suitable to get a quick overview of magnetic phase diagrams , for example . the real - space extension of two - site dmft to systems with reduced translational symmetries is straightforward and will be discussed in detail elsewhere . the two - impurity anderson model in an essentially one - dimensional geometry ( see fig . [ fig : schematic_tiam ] ) is amenable to a numerically exact solution by using the density - matrix renormalization group . @xcite therefore , dmrg calculations can be used to benchmark the quality of the magnetic susceptibilities obtained from the r - dmft approach . the calculation of the ground state and of ground - state expectation values for the tiam is a standard problem within dmrg . here we have been following ref . and have implemented a code which is based on the variation of matrix - product states ( mps ) . @xcite the main idea is to optimize a test wave function @xmath117 of the form @xmath118 where @xmath119 is a local basis at the site @xmath120 of a one - dimensional chain with @xmath2 sites in total . the elements @xmath121 of the matrices @xmath122 are considered as variational parameters which are locally and iteratively optimized during a sweep through the chain by exploiting the ritz variational principle . several sweeps are necessary to obtain a converged ground state . in practice , the local optimization can be reformulated as a generalized eigenvalue problem which is simplified to an ordinary one by exploiting a local gauge invariance of @xmath117 to properly ( left- and right- ) orthogonalize the @xmath48-matrices . the eigenvalue problem is then efficiently solved by means of the davidson method . @xcite we profit from the so - called wave - function transformation @xcite to reduce the number of iterations necessary for convergence of the davidson algorithm and exploit the two @xmath123 symmetries of the hamiltonian corresponding to conservation of the total particle number and the @xmath112-component of the total spin . observables and the hamiltonian in particular are represented as matrix - product operators . @xcite besides an elegant and flexible coding this allows to easily consider different implementations of the hamiltonian . for the present case of the tiam there are two possibilities to treat the adatoms suggesting themselves : ( i ) an adatom orbital @xmath30 and the substrate orbital @xmath31 `` below '' @xmath30 are treated as a single `` site '' @xmath120 in the dmrg context . the disadvantage is that therewith the local hilbert - space dimension at @xmath120 is enlarged . ( ii ) the adatom orbitals @xmath30 are treated as separate sites , i.e. a chain of length @xmath63 is formed . this leaves the local hilbert - space dimension constant but introduces next - nearest - neighbor hopping terms . we have tested both variants and found the differences in computational costs and accuracy to be marginal only . routinely , variant ( ii ) is employed . an important aspect is to prevent the sweep algorithm from getting stuck in a local energy minimum . this can be circumvented by implementing a mixed single - site approach @xcite to introduce fluctuations in the reduced density matrix . the additional coupling to a larger set of states considerably improves the convergence properties when optimizing @xmath117 . it furthermore also allows to dynamically adapt the dimensions of the @xmath48-matrices . converged results for typical situations with long - range spin - spin correlations in a tiam with about @xmath124 sites are obtained with matrix dimensions of the order of @xmath125 in the largest invariant blocks of the @xmath48-matrices . a reliable error measure is the variance @xmath126 which is easily accessible within an mps - based implementation . we have checked that typically the standard deviation @xmath127 . r - dmft and dmrg calculations have been performed for the tiam at half - filling and zero temperature . we consider systems with an even number @xmath2 of substrate sites and two adatoms at positions symmetric to the chain center at a distance @xmath128 as displayed in fig . [ fig : schematic_tiam ] . as this implies an even number of electrons , there is no kramers degeneracy of the ground state . all calculations are done using open boundary conditions . [ substrate_chi ] shows the magnetic susceptibility @xmath129 for a tiam with @xmath124 as defined in eq . ( [ sub_susceptibility ] ) . because of the mirror symmetry , it is sufficient to discuss e.g. the case @xmath130 , i.e. the left adatom . for the calculations we apply a weak local magnetic field with a strength @xmath131@xmath132 at the left adatom and look for the response at substrate site @xmath39 . regularly , calculations for different @xmath133 are performed to ensure that the field strength is in the linear - response regime . let us first concentrate on distances @xmath134 with integer @xmath135 . here the rkky coupling between the magnetic adatoms is antiferromagnetic . other distances @xmath0 including those with ferromagnetic coupling will be discussed in sec . [ sec : diffdist ] . we start the discussion with @xmath136 ( lowest panel on the right in fig . [ substrate_chi ] ) . this is the case where the two adatoms are located at the edges of the substrate chain . the blue lines refer to our r - dmft calculations which have been done with @xmath111 local degrees of freedom in the effective impurity model . directly `` below '' the first adatom at @xmath137 the response is antiferromagnetic , i.e. @xmath138 . this simply reflects the antiferromagnetic kondo coupling @xmath15 . the calculations have been done for @xmath139 and @xmath140 where the nearest - neighbor hopping in the substrate @xmath3 is used to set the energy scale . this results in a negative , i.e. antiferromagnetic , local exchange interaction of intermediate strength @xmath141 . this is clearly beyond the weak - coupling limit @xmath142 but still charge fluctuations are largely suppressed : we find an average double occupancy of @xmath143 at the adatom site , and the adatom local magnetic moment @xmath144 is much closer to the localized - spin value @xmath145 than to the free fermion value @xmath146 . as a function of the distance @xmath147 to the first impurity , the response is oscillatory corresponding to the @xmath148 nesting wave vector . its absolute value is maximal at @xmath137 , decreases with increasing @xmath39 and almost saturates until there is a slight upturn for @xmath149 , i.e. at the position of the second adatom . consistent with the @xmath150 oscillation , @xmath151 is positive at @xmath152 which implies , due to the antiferromagnetic local coupling @xmath153 , that there is an antiferromagnetic ( rkky ) alignment of the two adatom moments . the corresponding dmrg results are also shown in fig . [ substrate_chi ] for comparison ( red lines ) . for the distance @xmath136 , however , there is actually no difference to the r - dmft results visible on the scale of the figure . as r - dmft accounts for the single - impurity kondo effect exactly , this perfect agreement would be plausible if a picture of two independent kondo effects applied . strictly speaking , however , this can not be the case : there is a finite nonlocal adatom - adatom susceptibility , even in this long - distance limit ( see also fig . [ dmft_vs_dmrg_d ] and corresponding discussion below ) which in principle has a non - vanishing feedback on the self - energy and generates nonlocal elements of the self - energy in particular . r - dmft is thus approximate . on the other hand , we can conclude that this feedback of the nonlocal susceptibility is apparently negligibly small and r - dmft almost exact for the present situation . upon decreasing the distance between the adatoms , this picture should change gradually . however , apart from the extreme case @xmath154 , deviations of the r - dmft from the dmrg results are extremely small , and the agreement between r - dmft and dmrg remains excellent . on the other hand , with decreasing @xmath0 , the @xmath39 dependence of the susceptibility becomes much more complicated : the response below the second adatom ( see the second maximum of @xmath155 ) becomes stronger and stronger , the response at substrate sites between the adatoms increases and its absolute value develops a pronounced minimum close to @xmath156 , whereas the response beyond the second adatom , for @xmath157 , gets very weak . furthermore , while the susceptibility changes sign between nearest neighbors , its two - site _ average _ is negative between the adatoms and also beyond the first one for @xmath158 but is found to be positive for @xmath157 . there is another subtle observation , namely the ( ferromagnetic ) response at the nearest neighbor to the right of @xmath137 is larger than the one to the left of @xmath137 for all @xmath0 down to @xmath154 , except for @xmath159 and @xmath160 . the ratio @xmath161 is decreasing with decreasing @xmath0 becomes smaller than unity for @xmath159 and @xmath160 and larger than unity again for @xmath154 . all these non - trivial features are perfectly captured by the r - dmft and in fact result from an effective adatom - adatom interaction . this becomes obvious by comparing the results for the tiam with those of a corresponding single - impurity anderson model where the second ( right ) adatom @xmath162 is missing or , equivalently where the hybridization to the second adatom is switched off . we have performed dmrg calculations for corresponding single - adatom models . to ensure a singlet ground state at half - filling , however , the substrate chain has to be shortened by one site on the right edge ( @xmath163 ) . the resulting substrate susceptibilities @xmath151 are shown in fig . [ substrate_chi ] as green lines . comparing the siam and the tiam results to each other once more demonstrates that the effects of the indirect nonlocal rkky coupling become more and more pronounced with decreasing @xmath0 . the differences between the single - adatom and the two - adatom physics visible in the susceptibilities for @xmath157 are larger by more than an order of magnitude than the differences between the r - dmft and the dmrg results . again this shows that there are sizable effects on nonlocal magnetic correlations which do not fully feed back to the one - electron self - energy . the dmrg data for @xmath151 and also for the equal - time spin - spin correlation function @xmath164 are shown in fig . [ correlation_function ] for @xmath165 on a larger scale . let us discuss the physics of this situation in detail . the response of the substrate to a static local field at @xmath130 is governed by the low - energy excitations around the fermi edge , i.e. @xmath166 . contrary , the equal - time spin - spin correlation is obtained by a frequency integration of the dynamic ( retarded ) susceptibility @xmath167 and thus includes several energy scales . nevertheless , the spin - spin correlation behaves qualitatively very similar to @xmath151 , and we will refer to this on an equal footing with the susceptibility . the ground state of the whole system is a spin singlet in all calculations discussed here . this provides us with a simple sum rule for the spin - spin correlation : exploiting rotational symmetry , we have @xmath168 where @xmath169 is the total spin and @xmath170 the total substrate spin . using @xmath171 in the ground state , we immediately find : @xmath172 for the single - adatom model we then have @xmath173 which explains why the spin - spin correlation is mainly negative : namely , if summed over all substrate sites , it just compensates the adatom local moment . for the two - adatom model at @xmath165 , the adatom - adatom spin correlation @xmath174 is negative but its absolute value is small compared to @xmath175 . looking at eq . ( [ sumrule ] ) , the overall substrate response is thus still antiferromagnetic but somewhat reduced as compared to the single - adatom model . ( upper panel ) and magnetic susceptibility @xmath151 ( lower panel ) for @xmath139 and @xmath176 as obtained by dmrg for a system with @xmath124 ( @xmath163 ) substrate sites and two adatoms ( one adatom ) as functions of the substrate site @xmath39 . red lines : results for two adatoms and @xmath124 , tiam . green lines : results for @xmath163 and a single adatom at the same position as the @xmath130 ( left ) adatom in the two - adatom model , siam . the dashed lines indicate the positions @xmath177 and @xmath178 of the substrate sites `` below '' the adatoms . , scaledwidth=45.0% ] qualitatively the same applies to the susceptibility as can be seen from the lower panel in fig . [ correlation_function ] . for any large but finite system with a non - degenerate singlet ground state , we again have a simple sum rule : a singlet ground state and a finite gap implies that the total magnetic moment must vanish for any @xmath179 up to some finite critical field : @xmath180 . taking the derivative with respect to @xmath179 then yields : @xmath181 in the same way as above , @xmath182 for a single adatom , and for two adatoms the total response of the substrate is still negative but slightly reduced in absolute magnitude due to the presence of the second adatom since @xmath183 at @xmath165 . as can be seen in fig . [ correlation_function ] by comparing with the results for the single - adatom model , the most pronounced effect due the presence of the second adatom consists in the strong enhancement of @xmath184 , i.e. the response below the second adatom . this can easily be understood by referring to the rkky limit for a system of finite size @xmath2 : for @xmath185 keeping @xmath186 fixed , charge fluctuations vanish and we are left with a kondo - type model . in the weak - coupling limit @xmath187 the substrate degrees of freedom can be integrated out , and the adatom magnetic response , i.e. @xmath188 and @xmath189 , is perfectly described by an effective rkky two - spin model @xmath190 where @xmath191 is given in terms of the static substrate susceptibility at @xmath192 . the kondo effect , on the other hand , does not interfere with this picture as it is cut by the finite - size gap : one can define a coupling strength @xmath193 at which the kondo temperature @xmath194 becomes comparable with the finite - size gap . then , for @xmath195 the kondo effect is absent as there are simply no states at the fermi energy available to screen the adatom moment . @xcite this implies that the substrate is in a singlet state for weak @xmath15 and thus @xmath196 , i.e. there is no substrate contribution to the magnetic moment induced by the field at @xmath130 . from the sum rule eq . ( [ sumrulechi ] ) we thus have @xmath197 , i.e. also the two adatom spins form a perfect singlet consistent with eq . ( [ eq : rkky ] ) . hence , applying a field @xmath179 at adatom @xmath130 induces antiferromagnetically aligned magnetic adatom moments with _ the same _ absolute magnitude . for @xmath15 beyond but close to the rkky limit we therefore expect the absolute magnitude of the substrate response at @xmath177 and @xmath178 as almost equal . for finite and actually intermediate @xmath15 , see fig . [ correlation_function ] , the effect is strongly diminished but still clearly visible . note that the above argumentation can analogously be given by referring to the spin - spin correlation . as mentioned before , looking at the sum rules ( [ sumrule ] ) and ( [ sumrulechi ] ) , we can understand that the response of the substrate is somewhat attenuated in the tiam as compared to the siam . this reduction , however , is not homogeneous : there is a comparatively strong reduction beyond the second adatom for @xmath157 while the response is nearly the same or even somewhat enhanced close to @xmath178 for @xmath198 , and there is almost no effect for @xmath158 . that the effect is least pronounced close to the first adatom , can easily be understood by referring to the extreme kondo limit where a picture of two separate kondo clouds applies . in this case the magnetic response to the field applied to the first adatom would be the same as the response in the corresponding single - adatom model . as is seen in fig . [ correlation_function ] , however , close to @xmath137 there are finite differences , i.e. the kondo clouds do overlap , but the differences are small . since according to the sum rule the total response must be weaker in the tiam , a reduced response must and in fact does show up away from @xmath137 , i.e. for @xmath157 . at the lattice site @xmath137 below the first adatom as a function of the adatom - adatom distance @xmath0 for @xmath139 , @xmath176 and @xmath124 substrate sites as obtained from r - dmft ( blue line ) and dmrg ( red line ) . , scaledwidth=48.0% ] the sum rule ( [ sumrulesiam ] ) for the siam may also be used to roughly estimate the size of the individual `` kondo clouds '' . using the dmrg data for @xmath199 , we define an integrated spin - spin correlation function , @xcite @xmath200 for the single - adatom model . we have @xmath201 . with increasing @xmath202 more and more substrate spins around @xmath177 are included in the sum , and @xmath203 essentially decreases with @xmath202 until @xmath204 if all spins are included as is obvious from the sum rule eq . ( [ sumrulesiam ] ) . the quantity gives the fraction of the adatom spin that remains unscreened by the substrate spins up to distance @xmath202 from @xmath177 . using a 90%-screening criterion , for example , i.e. @xmath205 , the extent of the cloud amounts to @xmath206 lattice sites . this is consistent with the discussion given above . a criterion based on eq . ( [ cloud ] ) can not precisely define the parameter range in which r - dmft gives reliable results . [ local_substrate_chi ] demonstrates that , using r - dmft , the deviation from the numerically exact dmrg data grows _ gradually _ when decreasing the distance between the adatoms @xmath0 . to estimate the reliability of the mean - field approach , we focus on the susceptibility @xmath207 at the substrate site below the first adatom @xmath208 where , according to the results shown in fig . [ substrate_chi ] , the deviations are the strongest . @xmath209 is shown in fig . [ local_substrate_chi ] as a function of the distance @xmath0 . there is a nice quantitative agreement of the r - dmft with the exact dmrg result for large @xmath0 . for smaller @xmath0 , r - dmft still predicts the correct trend , except for @xmath154 . and nonlocal adatom - adatom susceptibility @xmath189 as functions of the distance @xmath0 between the adatoms for @xmath139 , @xmath176 and for a system with @xmath124 substrate sites as obtained by r - dmft and dmrg ( red lines ) . r - dmft calculations are done with different numbers of bath orbitals in the effective single - impurity models : @xmath210 , as indicated . , scaledwidth=40.0% ] for the same set of parameters fig . [ dmft_vs_dmrg_d ] shows the local adatom susceptibility @xmath188 and nonlocal adatom - adatom susceptibility @xmath189 as functions of the distance @xmath134 with integer @xmath135 . in both cases the agreement of the r - dmft with the dmrg results is excellent . significant differences are found for @xmath154 only and rapidly diminish with increasing @xmath0 . [ dmft_vs_dmrg_d ] includes r - dmft results obtained with different @xmath115 . on the scale of the figure , there is no difference between the results for @xmath211 and @xmath114 bath sites in the effective single - impurity model while the results obtained for @xmath189 with @xmath212 slightly deviate for intermediate distances around @xmath213 . this comparison shows that the r - dmft results are converged with respect to @xmath115 . the differences to the dmrg data are thus intrinsic to the dynamical mean - field approach itself and not at all caused by discretization errors of the lanczos solver . it is worth to mention that the distance dependence of @xmath189 can not be explained by conventional rkky theory . for @xmath142 , the magnetic susceptibility is determined by the effective two - spin heisenberg model eq . ( [ eq : rkky ] ) which yields @xmath214 with @xmath215 at odd distances @xmath0 . the decreasing absolute magnitude of @xmath189 with increasing @xmath0 and also the fact @xmath216 just reminds us that with @xmath139 and @xmath217 the system is well beyond the perturbative - in-@xmath15 regime and that there is a strong substrate contribution @xmath218 necessary to fulfill the sum rule eq . ( [ sumrulechi ] ) . for large @xmath0 the trends can rather be understood in a picture of two independent kondo effects . clearly , @xmath219 is expected to decrease with @xmath0 . more interesting is the behavior of @xmath188 which develops a maximum around @xmath220@xmath221 . the increase of @xmath188 with @xmath0 at short @xmath0 results from a reminiscence to the rkky limit : with increasing @xmath0 the effective coupling between the adatom decreases and their magnetic moments tend to become free resulting in a higher local susceptibility @xmath188 . substrate contributions are sizable but can not outweigh this effect in @xmath188 , in contrast to @xmath189 which is 23 times smaller in absolute magnitude . in the large-@xmath0 limit , where adatom - adatom interactions can be disregarded completely , one would naively expect a saturation of the local susceptibility at the inverse kondo temperature since @xmath222 in a single - impurity model . @xcite however , @xmath188 must decrease since with increasing @xmath0 at fixed @xmath124 the adatoms move to the chain edges where we have a site - dependent kondo temperature . this increases with decreasing distance to the edge as the non - interacting substrate local density of states at the fermi energy is increasing . and @xmath189 as functions of distance @xmath0 for @xmath139 , @xmath176 . r - dmft calculations for different system sizes @xmath2 as indicated . dmrg data are shown for comparison at nearest - neighbor distance @xmath154 only . , scaledwidth=49.0% ] this interpretation is corroborated by fig . [ dmft_l_d ] which displays r - dmft results for @xmath188 and @xmath189 for different system sizes @xmath223 , @xmath124 and @xmath224 . we find the same qualitative behavior in all three cases . quantitatively , however , there are sizable differences at inter - adatom distance @xmath154 , for example , which show that even with @xmath224 substrate sites the chain center can not be regarded as bulk - like and that the center local density of states is still considerably dependent on @xmath2 . on the other hand , the susceptibilities for @xmath0 close to @xmath2 , i.e. for systems with adatoms located at or very close to the chain edges , are almost converged . note that @xmath188 for @xmath225 is almost the same for @xmath124 and @xmath224 . again this shows that , at least for the larger systems , the magnetic response is dictated by the physics of the single - site kondo effect , i.e. the presence of the second adatom has almost no effect on @xmath188 and on the kondo temperature of the first adatom . this does not exclude a finite magnetic interaction between the adatoms and in fact a non - zero @xmath189 for @xmath225 is found which , in addition , also does not depend on @xmath2 for the larger systems . these adatom - adatom magnetic interactions are correctly captured by the r - dmft . r - dmft and dmrg results coincide on the scale of the plot except for @xmath154 ( dmrg results are shown for @xmath154 only ) . this almost perfect agreement can be understood by referring to the strong - coupling limit : for @xmath226 , the two adatom magnetic moments form perfectly local kondo singlets that do not interact with each other . this limit is trivially accessible by the mean - field approach . for finite coupling , second - order perturbation theory in @xmath227 predicts spin - spin correlations to decay as @xmath228 . @xcite the inter - adatom magnetic interaction is thus expected to scale as @xmath229 . this is also accessible to the r - dmft approach while the neglected feedback of this effective interaction on local physical properties at one adatom , e.g. on @xmath14 and thus on @xmath188 , is of higher order and small in the strong-@xmath15 limit . and @xmath189 as functions of @xmath230 for @xmath139 and @xmath154 as obtained by r - dmft , real - space two - site dmft and dmrg for a system with @xmath124 substrate sites . insets : same quantities plotted on a logarithmic scale . dashed lines indicate the critical @xmath230 where @xmath188 and @xmath189 diverge . , scaledwidth=49.0% ] and @xmath189 as functions of @xmath230 for @xmath154 and @xmath124 as obtained by r - dmft . results for different @xmath32 as indicated . upper inset : inverse susceptibility @xmath231 and extrapolation ( dotted lines ) to @xmath232 . lower inset : `` phase diagram '' , separating the fermi liquid state ( fl ) from an ( artificial ) antiferromagnetic state ( af ) which shows up for weak @xmath1 and strong @xmath32 . , scaledwidth=49.0% ] the breakdown of r - dmft can be enforced , however , by decreasing @xmath15 . [ dmft_vs_2s - dmft_dmrg_j ] shows the susceptibilities for @xmath154 as a function of @xmath233 . there is again excellent agreement for strong @xmath233 even with the simplified two - site r - dmft . deviations of the two - site approach from the exact @xmath188 and @xmath189 become sizable for couplings smaller than @xmath234 . the full r - dmft is reliable down to smaller values for @xmath233 but finally also starts to significantly deviate from the dmrg data for @xmath235 . here , as compared to the strong - coupling limit , the local susceptibility is by more than an order of magnitude higher , i.e. the kondo temperature is by more than an order of magnitude smaller ( see the upper inset ) . for even smaller couplings , the mean - field approach breaks down completely and fails to maintain a fermi - liquid ground state : the small-@xmath15 limit is problematic for r - dmft as the screening of the magnetic moments is too weak to compensate the ordering tendencies induced by a comparatively strong inter - adatom interaction . the system becomes too susceptible to an artificial spontaneous symmetry breaking that is induced by the mean - field approximation itself . while the adatoms state is given by a nonlocal su(2 ) invariant singlet @xmath236 for @xmath142 , the mean - field theory predicts an incoherent mixture of degenerate ordered states @xmath237 and @xmath238 . this qualitative failure is indicated by divergencies of @xmath188 and @xmath189 which take place at coupling strengths @xmath233 that are somewhat smaller than those where first quantitative deviations from the exact data were found ( see insets in fig . [ dmft_vs_2s - dmft_dmrg_j ] ) . this also implies that the mean - field approach is able to exhibit its limitations by itself . [ dmft_u_j_pt ] shows the susceptibilities as obtained by r - dmft for @xmath154 and different @xmath32 and @xmath1 as functions of @xmath230 . we find @xmath188 and @xmath189 to diverge at the same point in parameter space . from extrapolations of the inverse local susceptibility to @xmath239 at different @xmath32 and @xmath1 , shown in the upper inset , one may derive a mean - field `` phase diagram '' . this is displayed in the lower inset . a normal fermi - liquid ground state found for large @xmath1 and small @xmath32 is separated from the su(2)-symmetry - broken antiferromagnetic state realized for small @xmath230 . this `` phase - transition line '' should actually be interpreted as a crossover from the kondo to the rkky regime or , more precisely , as the boundary up to which r - dmft is reliable . while the critical coupling @xmath230 is almost independent of @xmath32 , there is some dependence on the distance @xmath0 . however , this is weak : while a fermi - liquid ground state is obtained down to @xmath240 for @xmath154 , we find a slightly smaller critical value of @xmath241 for @xmath159 and @xmath242 for @xmath136 ( at @xmath139 ) . more important for the reliability of r - dmft is the local kondo temperature . this becomes obvious if the two adatoms are placed at a distance @xmath243 with integer @xmath135 , i.e. @xmath244 etc . at the corresponding substrate sites @xmath137 and @xmath156 ( symmetric to the chain center ) we have a low weight @xmath245 of the one - particle energy eigenstate of the non - interacting substrate at the fermi wave vector @xmath246 while @xmath245 is high for distances @xmath134 this pronounced odd - even effect is a consequence of surface friedel oscillations . the weight @xmath245 determines the local substrate density of states and thus also the local kondo temperature . consequently , @xmath194 is small for @xmath243 and the nonlocal rkky interaction much more efficient . using dmrg , we in fact find @xmath219 at @xmath247 to be more than an order of magnitude larger than at @xmath154 . this effect even increases with increasing @xmath243 since @xmath245 is decreasing if the adatoms move towards the chain edges . at the edges ( @xmath248 ) @xmath245 is suppressed by more than a factor 100 compared to the @xmath134 case ( @xmath136 ) , and the kondo temperature is essentially vanishing . this regime is not accessible to r - dmft . while the mean - field approach predicts the correct sign , the absolute value @xmath219 and also @xmath188 is strongly underestimated for @xmath243 . deviations from the dmrg results grow with increasing @xmath0 . this had to be expected , as a huge nonlocal susceptibility @xmath219 induces via the schwinger - dyson equation a sizable contribution to the nonlocal self - energy which is neglected in r - dmft . the ferromagnetic case is different . here we consider distances @xmath249 with integer @xmath135 . in the kondo limit of the model and for weak @xmath15 , the ferromagnetic rkky coupling of well - formed spins @xmath250 leads to a triplet ground state as is easily verified by means of dmrg calculations for @xmath124 , i.e. there is a nonlocal spin @xmath251 . in all our calculations this spin is not screened by the substrate electrons . this may be explained by the fact that @xmath124 is still too small to accommodate the corresponding screening cloud . however , there is a nonlocal spin @xmath251 not only in the weak-@xmath15 limit . in fact , we find a triplet ground state for any choice of @xmath252 and @xmath253 . our r - dmft calculations reproduce the spin - triplet ground state for small @xmath230 by predicting , for an infinitesimally small external magnetic field in @xmath254 direction , a spontaneously symmetry - broken ferromagnetic state @xmath255 . this corresponds to the @xmath256 state of the dmrg spin triplet . with increasing @xmath1 , however , the expectation value of the @xmath112-component of the total spin @xmath169 deviates from unity and , beyond a critical hybridization @xmath1 , even vanishes : @xmath257 . hence , in the ferromagnetic case , r - dmft is reliable in the small-@xmath15 but appears to fail in the strong - coupling limit . this requires further investigations which , however , are beyond the scope of the present paper . conventional ( rkky ) theory of indirect magnetic exchange predicts an effective exchange interaction @xmath258 where @xmath259 is the nonlocal static susceptibility of the metallic host . this interaction survives , as a nearest - neighbor coupling , @xcite even in the case of two magnetic impurities embedded in an infinite - dimensional lattice and is thus accessible by dynamical mean - field theory . in the limit of infinite spatial dimensions or , at finite dimensions , within the dynamical mean - field approximation , one can therefore expect a finite response at one magnetic impurity subject to a local magnetic field at the other one , located at nearest - neighbor but also for larger distances . on the other hand , nonlocal effective interactions do not contribute to the single - particle self - energy on the dmft level : the dmft self - energy is just defined as the sum of the _ local _ skeleton diagrams only . this is a well - known shortcoming of mean - field theory which gives rise to artifacts in the rkky limit . namely , for @xmath187 the magnetic impurities are only weakly coupled to the host and thus become extremely susceptible . a tiny weiss field within dmft is then sufficient to drive the system to an artificial symmetry - broken state , i.e. an antiferromagnetic state rather than a nonlocal singlet of the impurity magnetic moments is formed . a state with the characteristic distance dependence of the rkky interaction , e.g. @xmath260 for a one - dimensional system at half - filling , can not be recovered within dmft as it is always preempted by spontaneous symmetry breaking . clearly , solutions with a finite magnetic moment could easily be suppressed in a mean - field approach . one should note , however , that the resulting magnetic susceptibility is unphysical , i.e. negative , as it refers to a thermodynamically unstable state . therefore , in any case , the physics of the rkky limit is not accessible by dmft . the present study has shown , however , that beyond the rkky limit , ( real - space ) dmft is well suited to study even quantitatively the effects of indirect magnetic exchange . here , we have concentrated on two magnetic `` adatoms '' on a one - dimensional `` substrate surface '' a minimal model to study indirect magnetic interactions in competition with the kondo effect for magnetic atoms on metallic surfaces and a model that is amenable to an exact numerical solution by means of the density - matrix renormalization group . dmrg has been used to compute spin - spin correlation functions and static spin susceptibilities , and particularly adatom - substrate site correlations and susceptibilities . depending on the distance @xmath0 between the two adatoms and depending on the hybridization strength @xmath1 , rather complicated profiles are obtained . comparing the results for the two - adatom ( two - impurity ) anderson model with those obtained for the corresponding single - adatom ( single - impurity ) anderson model , one can easily classify the different features of those profiles as single - impurity effects or as resulting from the adatom - adatom effective interaction . in this way , clear reminiscences of the rkky interaction , i.e. of nonlocal singlet formation , are found to compete with the formation of kondo clouds and screening of the adatom magnetic moments . in addition , the profiles are strongly affected by the finite system size ( chains with typically @xmath124 have been considered here ) and by effects resulting from strong surface friedel oscillations in the local density of states , especially if the adatoms are in the vicinity to one of the chain edges . this complex interplay of different physical mechanisms is almost perfectly recovered by the real - space dmft . qualitatively , the real - space dmft is reliable as long as the model parameters , in particular the local exchange coupling @xmath261 , are in a regime well separated from the artificial symmetry - broken state . this parameter regime , where the adatom susceptibilities are not too large or where the adatom magnetic moments are predominantly interacting with the substrate moments rather than among each other , however , goes well beyond the extreme kondo limit of non - overlapping kondo clouds . the critical value for @xmath240 in units of @xmath29 at @xmath154 gives an impression of a lower bound for the applicability of r - dmft . while the present study has focused on a one - dimensional model to allow for benchmarking against numerically exact dmrg results , future applications of the r - dmft should address higher - dimensional systems . with increasing coordination number of correlated sites , the parameter space accessible to the mean - field approach is expected to be become larger or mean - field artifacts less pronounced . for the case of atoms trapped in optical lattices , there are impressing examples where r - dmft has contributed to an understanding of the physics of inhomogeneous systems with @xmath116 correlated and geometrically inequivalent sites in two dimensions , for example . let us also point out that for anderson - type multi - impurity or lattice models , one typically expects non - local magnetic correlations to diminish rapidly as the electron filling on the correlated sites is changed away from half - filling . systems off half - filling are thus expected to be more amenable to an r - dmft approach . at the same time they are also interesting physically as reducing the filling away from half - filling affects local - moment formation as well . hence , the competition between non - local rkky interaction and kondo screening must be seen as strongly filling dependent . for complex magnetic nanostructures with several magnetic adatoms in different chain or cluster geometries on two- or on semi - infinite three - dimensional metallic surfaces , a mean - field approach is inevitable anyway . here the conceptual simplicity of a single - site mean - field theory , as compared to different possibilities for cluster extensions , is important as it allows to study almost arbitrary geometries . as in ab - initio studies , the accessible system size strongly depends on the remaining e.g. lateral spatial symmetries , and the computational effort scales nearly linearly with the number of inequivalent correlated sites only . the two - impurity one - dimensional anderson model represents a model that is rather unfavorable to a single - site r - dmft approach . even for this case , as the present study has shown , r - dmft can in fact almost quantitatively predict the effects of indirect magnetic exchange in competition with the kondo and with geometrical effects as long as the approximation predicts a fermi - liquid ground state . the work was supported by the deutsche forschungsgemeinschaft within the sonderforschungsbereich 668 ( project a14 ) . n.r . acknowledges support by the forschergruppe for 1346 . m. potthoff and w. nolting , physica b * 259 - 261 * , 760 ( 1999 ) . h. ishida and a. liebsch , phys . b * 79 * , 045130 ( 2009 ) . r. nourafkan and f. marsiglio , phys . b * 83 * , 155116 ( 2011 ) . s. okamoto , phys . b * 84 * , 201305 ( 2011 ) . p. miller and j. k. freericks , j. phys . cond . mat . * 13 * , 3187 ( 2001 ) . j. k. freericks , phys . b * 70 * , 195342 ( 2004 ) . s. okamoto and a. j. millis , phys . b * 70 * , 241104 ( 2004 ) . l. chen and j. k. freericks , phys . b * 75 * , 125114 ( 2007 ) . j. k. freericks , transport in multilayered nanostructures ( imperial college press , london , 2006 ) . s. okamoto , phys . * 101 * , 116807 ( 2008 ) . h. zenia , j. k. freericks , h. r. krishnamurthy and th . pruschke , phys . lett . * 103 * , 116402 ( 2009 ) . s. t. f. hale and j. k. freericks , phys . b * 83 * , 035102 ( 2011 ) . v. dobrosavljevic and g. kotliar , phys . lett . * 78 * , 3943 ( 1997 ) ; philos . london , ser . a * 356 * , 57 ( 1998 ) . tran , phys . b * 76 * , 245122 ( 2007 ) . y. song , r. wortis and w. a. atkinson , phys . b * 77 * , 054202 ( 2008 ) . j. wernsdorfer , g. harder , u. schollwoeck , w. hofstetter , e - print arxiv:1108.6057 . helmes , t.a . costi and a. rosch , phys . lett . * 100 * , 056403 ( 2008 ) ; phys . * 101 * , 066802 ( 2008 ) . m. snoek , i. titvinidze , c. tke , k. byczuk and w. hofstetter , new j. phys . * 10 * , 093008 ( 2008 ) . a. koga , t. higashiyama , k. inaba , s. suga and n. kawakami , phys . a * 79 * , 013607 ( 2009 ) ; j. phys . . jpn . * 77 * , 073602 ( 2008 ) . k. noda , a. koga , n. kawakami and t. pruschke , phys . a * 80 * , 063622 ( 2009 ) . m. snoek , i. titvinidze and w. hofstetter , phys . b. * 83 * , 054419 ( 2011 ) . gorelik , i. titvinidze , w. hofstetter , m. snoek and n. blmer , phys . * 105 * , 065301 ( 2010 ) . n. blmer and e. v. gorelik , comput . commun . * 118 * , 115 ( 2011 ) . a. koga , j. bauer , p. werner and th . pruschke , physica e * 43 * , 697 ( 2011 ) . kim , j. j. kinnunen , j .- martikainen and p. trm , phys . 106 , 095301 ( 2011 ) . s. r. white , phys . lett . * 69 * , 2863 ( 1992 ) . u. schollwck , annals of physics * 326 * , 96 ( 2011 ) , rev . mod . phys . * 77 * , 259 ( 2005 ) . f. verstraete , v. murg and j. i. cirac , advances in physics * 57 * , 143 ( 2008 ) . s. stlund and s. rommer , phys . . lett . * 75 * , 3537 ( 1995 ) . i. p. mcculloch , j. stat . p10014 ( 2007 ) . m. caffarel and w. krauth , phys . lett . * 72 * , 1545 ( 1994 ) . q. si , m. j. rozenberg , g. kotliar and a. e. ruckenstein , phys . lett . * 72 * , 2761 ( 1994 ) . g. h. golub and c. f. van loan , _ matrix computation _ , chapter 9 third edition ( 1996 ) . e. pavarini , e. koch , d. vollhardt and a. lichtenstein ( eds . ) _ the lda+dmft approach to strongly correlated materials _ , volume 1 , chapter 8 ( 2011 ) .

a recent analysis comparing the [ na / fe ] distributions of red giant branch ( rgb ) and asymptotic giant branch ( agb ) stars in the galactic globular cluster ngc 6752 found that the ratio of na poor to na rich stars changes from 30:70 on the rgb to 100:0 on the agb . the surprising paucity of na rich stars on the agb in ngc 6752 warrants additional investigations to determine if the failure of a significant fraction of stars to ascend the agb is an attribute common to all globular clusters . therefore , we present radial velocities , [ fe / h ] , and [ na / fe ] abundances for 35 agb stars in the galactic globular cluster 47 tucanae ( 47 tuc ; ngc 104 ) , and compare the agb [ na / fe ] distribution with a similar rgb sample published previously . the abundances and velocities were derived from high resolution spectra obtained with the michigan / magellan fiber system ( m2fs ) and mspec spectrograph on the magellan clay 6.5 m telescope . we find the average heliocentric radial velocity and [ fe / h ] values to berv=18.56 km s (=10.21 km s ) and[fe / h]=0.68 (=0.08 ) , respectively , in agreement with previous literature estimates . the average [ na / fe ] abundance is 0.12 dex lower in the 47 tuc agb sample compared to the rgb sample , and the ratio of na poor to na rich stars is 63:37 on the agb and 45:55 on the rgb . however , in contrast to ngc 6752 , the two 47 tuc populations have nearly identical [ na / fe ] dispersion and interquartile range values . the data presented here suggest that only a small fraction ( ) of na rich stars in 47 tuc may fail to ascend the agb , which is a similar result to that observed in m13 . regardless of the cause for the lower average [ na / fe ] abundance in agb stars , we find that na poor stars and at least some na rich stars in 47 tuc evolve through the early agb phase . the contrasting behavior of na rich stars in 47 tuc and ngc 6752 suggests that the rgb [ na / fe ] abundance alone is insufficient for predicting if a star will ascend the agb .

1412.4108

clear evidence indicates that most galactic globular clusters host multiple , distinct stellar populations ( e.g. , see reviews by gratton et al . 2004 ; 2012 ) . for mono metallic clusters exhibiting negligible spreads in [ fe / h]@xmath7log(n@xmath8/n@xmath9)@xmath10log(n@xmath8/n@xmath9)@xmath11 and log @xmath12(a)@xmath7log(n@xmath8/n@xmath13)+12.0 for elements a and b. ] , the various stellar populations are identified spectroscopically by their light element chemistry . specifically , stars within a single globular cluster are often categorized by their [ o / fe ] and/or [ na / fe ] ratios as belonging to the primordial " , intermediate " , or extreme " populations ( e.g. , carretta et al . 2009 ) . in this categorization , primordial stars ( first generation ) are similar in composition to metal poor halo field stars ( o rich ; na poor ) , and intermediate and extreme stars ( second generation ) have lower [ o / fe ] ratios and higher [ na / fe ] ratios . the extreme stars are further distinguished from the intermediate population as having the lowest oxygen abundances ( [ o / fe]@xmath50.4 ) and highest sodium abundances ( [ na / fe]@[email protected] ) . while the intermediate population tends to dominate by number over the primordial population , the extreme stars are found only in a handful of globular clusters ( carretta et al . 2009 ; their table 5 ) . the large and often ( anti)correlated abundance variations of elements ranging from carbon through aluminum is evidence that the material in globular cluster stars atmospheres has been subjected to high temperature proton capture nucleosynthesis ( e.g. , denisenkov & denisenkova 1990 ; langer et al . 1993 ; prantzos et al . 2007 ) . while changes in [ c / fe ] , [ n / fe ] , and @xmath16c/@xmath17c ratios as a function of evolutionary state on the subgiant branch ( sgb ) and red giant branch ( rgb ) are clearly linked to _ in situ _ mixing processes ( e.g. , denissenkov & vandenberg 2003 ) , the temperatures reached near the bottom of the convective envelope in more evolved low mass rgb stars are too low to significantly alter the abundances of heavier elements ( but see also dantona & ventura 2007 for a possible exception ) . observations of similar star to star abundance variations among scarcely evolved globular cluster main sequence and sgb stars ( e.g. , briley et al . 1996 ; gratton et al . 2001 ; ramrez & cohen 2002 ; ramrez & cohen 2003 ; carretta et al . 2004 ; briley et al . 2004 ; cohen & melndez 2005 ; bragaglia et al . 2010 ; dorazi et al . 2010 ; dobrovolskas et al . 2014 ) indicate that the composition differences between the various globular cluster populations were already imprinted on the gas from which the stars formed . although there is still no consensus regarding the nucleosynthesis source(s ) driving the composition differences , possible options include intermediate mass ( @xmath1858 m@xmath19 ) asymptotic giant branch ( agb ) stars ( e.g. , karakas et al . 2006 ; ventura & dantona 2009 ) , rapidly rotating massive ( @xmath1420 m@xmath19 ) main sequence stars ( e.g. , decressin et al . 2007a ) , massive binary stars ( de mink et al . 2009 ) , and super massive ( @xmath1810@xmath20 m@xmath21 ) stars ( denissenkov & hartwick 2014 ) . the physical process by which globular clusters form and evolve remains an open question as well ( e.g. , decressin et al . 2007b ; dercole et al . 2008 ; renzini 2008 ; carretta et al . 2010a ; bekki 2011 ; conroy & spergel 2011 ; valcarce & catelan 2011 ; bastian et al . 2013 ) . regardless of the pollution source(s ) in globular clusters , the nuclear processes creating o depleted and na enhanced gas may also be concurrent with he enhancements . with the exception of a few cases , such as ngc 1851 , where the c@xmath15n@xmath15o sum may be variable ( e.g. , ventura et al . 2009 ; yong et al . 2009 ; yong et al . 2014 ) , he abundance differences ranging from @[email protected] have been invoked to explain many of the multiple photometric sequences observed in recent cluster color magnitude diagrams ( e.g. , piotto et al . 2007 ; milone et al . 2012 ) . combined evidence from photometry and spectroscopy ties the traditional light element abundance patterns to varying levels of he enhancement ( e.g. , bragaglia et al . 2010 ; dupree et al . 2011 ; pasquini et al . 2011 ; marino et al . 2014 ; mucciarelli et al . in addition to causing subtle effects in absorption line formation ( bhm vitense 1979 ) , he enhancements can significantly alter a star s position on the color magnitude diagram and shorten its evolutionary timescale , relative to a he normal star . there are some indications that the extent ( or existence ) of the agb phase may be sensitive to a star s initial he abundance such that he enhanced stars may evolve off the horizontal branch to become agb manqu stars ( e.g. , greggio & renzini 1990 ; castellani et al . 2006 ; gratton et al . 2010 ; charbonnel et al . in particular , several authors have noted a peculiar feature in many clusters that cn strong and o poor / na rich stars appear with a lower frequency on the agb compared to the rgb ( mallia 1978 ; norris et al . 1981 ; suntzeff 1981 ; smith & norris 1993 ; pilachowski et al . 1996 ; ivans et al . 1999 ; sneden et al . 2000 ; campbell et al . 2006 ; campbell et al . 2010 ; smolinski et al . 2011 ; johnson & pilachowski 2012 ; campbell et al . 2012 ; campbell et al . 2013 ) . however , it is not yet clear that the lack of cn strong , o poor , and na rich stars on the agb is a ubiquitous property of all globular clusters . gratton et al . ( 2010 ) note that the agb / rgb ratio in globular clusters is correlated with the minimum mass along the horizontal branch . in other words , clusters with redder horizontal branches , and possibly also lower levels of he enhancement , tend to retain a larger fraction of stars between the rgb and agb phases . therefore , in order to investigate this phenomenon further we have obtained spectra of rgb and agb stars in the red horizontal branch , and relatively metal rich ( [ fe / h]@xmath230.7 ) , globular cluster 47 tucanae ( 47 tuc ) , and compare the [ na / fe ] distributions between stars in the two evolutionary states . 47 tuc exhibits a relatively high agb / rgb ratio ( @xmath180.10 ; gratton et al . 2010 ) , and is suspected of having a predominantly cn strong / na rich agb population ( mallia 1978 ) . if confirmed , a dominant na rich agb population in 47 tuc would strongly contrast with the completely na poor agb population in the more metal poor blue horizontal branch cluster ngc 6752 ( campbell et al . the rgb and agb data sets for this project were taken on separate nights and with different instruments . the rgb data were obtained in 2011 november using the flames giraffe instrument on the vlt ut2 telescope at the european southern observatory on cerro paranal . additional details regarding the observation , reduction , and analysis of these data can be found in cordero et al . ( 2014 ; their section 2 ) . the new observations of 47 tuc agb stars presented here were taken on 2014 june 1 with the michigan / magellan fiber system ( m2fs ; mateo et al . 2012 ) mounted on the nasmyth east port of the magellan clay 6.5 m telescope at las campanas observatory . while further details about m2fs can be found in mateo et al . ( 2012 ) , we summarize here basic information about the instrument and the specific setup used for this project . m2fs is a wide field ( 29.3@xmath24 ) fiber fed multiobject system with two sets of 128 fiber bundles ( 256 total ) feeding identical red " and blue " spectrographs . the 1.2@xmath25 fibers refers to the aperture size at the front of the fiber . ] are mounted by hand into machine drilled plug plates , and can be placed with a minimum distance of @xmath1813@xmath25 between fibers . for the high resolution mode used here , the spectra are created by passing light through both an echelle grating and cross dispersing prism . although observing in a single order allows for up to 256 fibers to be used simultaneously , the cross disperser permits order stacking at the expense of using fewer fibers . several post fiber spectrograph slit widths are available that range from 18045@xmath26 m and provide a resolving power of r@xmath7@xmath27/@xmath22@xmath27=20,00038,000 , respectively ( mateo et al . 2012 ) . specific to this project , we developed a spectrograph setup mode that employs a wide band order blocking filter ( bulge@xmath28gc1 " ) providing continuous wavelength coverage from @xmath1861206720 over 6 consecutive orders ( 5853 ; see also table 1 ) . this reduces the maximum number of available fibers from 256 to 48 ( 24 per spectrograph channel ) , but is roughly equivalent in terms of efficiency to a system such as flames giraffe . sample object , quartz flat , and thar comparison spectrum images obtained for this project are shown in figure [ f1 ] . an internal littrow ghost reflection ( e.g. , burgh et al . 2007 ) affecting the middle columns of both ccds was discovered when using this setup ; however , the reflection only interferes with a roughly 510 wide region in some orders of some fibers and is easily avoided in the reduced spectra . our observations used a 4 amp slow readout , 2@xmath291 ( spatial@xmath29dispersion ) binning , and the 125@xmath26 m slits to achieve a resolving power of r=22,500 . we placed 42 fibers on potential 47 tuc agb stars and 5 fibers on blank sky regions . only one available fiber was left unassigned . since the agb stars in 47 tuc are relatively bright ( v@xmath514 ; see also figure [ f2 ] ) and the observing conditions were good ( fwhm@xmath51@xmath25 ) , a set of 3@xmath291200 second exposures was sufficient to produce a signal to noise ratio ( s / n ) of about 100 per resolution element . the initial target selection and rgb / agb separation was accomplished using the photometry , luminosity , and temperature values from mcdonald et al . ( 2011 ) . since the data presented in cordero et al . ( 2014 ) contained almost exclusively rgb stars , we selected only early agb stars of comparable temperature from the mcdonald et al . ( 2011 ) catalog for observation with m2fs . the data from mcdonald et al . ( 2011 ) were matched to the two micron all sky survey ( 2mass ; skrutskie et al . 2006 ) database , and the 2mass coordinates were used as input for the fiber plug plate configuration . similarly , 2mass coordinates and photometry were used to identify suitable guide , acquisition , and shack hartmann stars required by the instrument . although we obtained spectra of 42 agb stars , 7 were discarded due to a failure to adequately converge to a stable solution of the model atmosphere parameters ( see section 3 ) . a 2mass color magnitude diagram and temperature luminosity plot for the final sample of 35 agb stars , along with the comparative sample of 113 rgb stars from cordero et al . ( 2014 ) , are shown in figure [ f2 ] . the star identifications , coordinates , photometry , model atmosphere parameters , abundances , and heliocentric radial velocities for all 35 agb stars are provided in table 2 . figure [ f3 ] compares the sky positions of the agb and rgb 47 tuc samples relative to the cluster center and half light radius ( r@xmath30 ) . we prioritized agb stars residing within @xmath1812 r@xmath30 , but also sampled out to the same radial extent as the rgb data . the radial distribution of targets is relevant because one of the key science questions addressed in this paper is whether or not second generation rgb stars fail to evolve off the horizontal branch and ascend the agb . there is growing evidence that while primordial ( first generation ) stars tend to follow the underlying cluster distribution at all radii , the intermediate and extreme stars ( second generation ) are often strongly concentrated near the cluster core ( e.g. , carretta et al . 2010b ; kravtsov et al . 2010 ; lardo et al . 2011 ; johnson & pilachowski 2012 ; milone et al . 2012 ; cordero et al . 2014 ; li et al . nataf et al . ( 2011 ) also find evidence that stars near the cluster core may be he enhanced ( see also di criscienzo et al . if he and na are correlated then the most na rich agb stars should reside near the core . vesperini et al . ( 2013 ) note that the local ratio of second / first generation stars should match the global ratio at a half mass radius of @xmath181.5 . therefore , we expect that our agb target selection will adequately sample the true agb [ na / fe ] distribution and maximize our chances of finding na rich agb stars , if they exist . the basic data reduction procedures , including bias correction and trimming the overscan regions , were carried out using the iraf task _ ccdproc_. since each amplifier image on each ccd has a slightly different value for read noise and gain , the individual amplifier object and calibration files are reduced separately in a manner similar to that used for hectochelle reductions ( caldwell et al . 2009 ; szentgyorgyi et al . the individual bias corrected and trimmed exposures are then rotated , translated , and combined using the iraf tasks _ imtranspose _ and _ imjoin _ to create one file per exposure ( i.e. , images such as those in figure [ f1 ] ) . although the m2fs setup used here produces 6 orders per fiber , the more advanced data reduction procedures should in principle be similar to the default setup in which each fiber produces a single order . therefore , we used repeated calls of the iraf task _ dohydra _ to handle aperture identification and tracing , scattered light removal , flat field correction , thar wavelength calibration , cosmic ray removal , and object spectrum extraction . we ran _ dohydra _ on all fibers of each ccd , but only extracted one order from each fiber per _ dohydra _ loop . since the sky fibers are spread across both ccds , we skipped the sky subtraction routine inside _ dohydra_. instead , we used _ scombine _ to create a master sky spectrum for each order of each exposure and subtracted this from the object spectra with the _ skysub _ routine . following sky subtraction , each order of every exposure was normalized with the _ continuum _ iraf routine and then median combined with _ scombine _ to increase the s / n and remove any remaining cosmic rays . the full wavelength range spanned by each order is listed in table 1 . however , as is evident in figure [ f1 ] , the s / n decreases near the edges of each order . therefore , we removed the lower s / n regions of each order that had correspondingly higher s / n regions in an adjacent order . the final ( effective " ) wavelength coverage of each order after carrying out this procedure is given in table 1 . we also chose to remove , rather than combine , overlapping regions between orders because there is a small difference in resolution between the bluer and redder orders . a sample m2fs spectrum with all orders merged is shown in figure [ f4 ] . the model atmosphere parameters effective temperature ( t@xmath31 ) , surface gravity ( log(g ) ) , metallicity ( [ fe / h ] ) , and microturbulence ( vt ) were determined spectroscopically and in a manner identical to that used for the flames giraffe sample of cordero et al . specifically , t@xmath31 was determined by removing trends in log @xmath12(fe i ) as a function of excitation potential , log(g ) was determined by enforcing ionization equilibrium with the fe i and fe ii lines , and vt was determined by removing trends in log @xmath12(fe i ) as a function of line strength . we adopted a generic model atmosphere with t@xmath31=4600 k , log(g)=1.5 ( cgs ) , [ fe / h]=0.7 , and vt=1.5 km s@xmath3 as the starting point and then iterated to simultaneously solve all four model atmosphere parameters . the average of the derived [ fe i / h ] and [ fe ii / h ] abundances was used to update the model metallicity in each iteration . opacity and atmosphere stratification discrepancies due to the difference between [ fe / h ] and [ m / h ] were roughly accounted for by using the @xmath32enhanced atlas9 model atmospheres from castelli & kurucz ( 2004 ) . we interpolated within the available atlas9 grid to obtain the final model atmospheres used in the abundance analysis . since the range in both temperatures and luminosities spanned by the rgb and agb samples is relatively small ( see figure [ f2 ] ) , we have not adopted any corrections to either the model atmosphere parameters or abundance ratios due to departures either from plane parallel geometry or local thermodynamic equilibrium ( lte ) . however , figure [ f5 ] shows that the spectroscopic temperatures are well correlated with temperatures estimates from ( j k@xmath33)@xmath34 2mass photometry . using the color temperature relation given in gonzlez hernndez & bonifacio ( 2009 ) , we find an average difference between the spectroscopic and photometric temperatures of 14 k ( @xmath4=85 k ) . we found a similar result for the rgb flames giraffe sample in cordero et al . ( 2014 ) , with an average difference of 28 k ( @xmath4=53 k ) . the line list used to determine fe i and fe ii abundances was similar to that used by cordero et al . ( 2014 ) , but augmented to include additional lines available in the m2fs spectra . the [ fe / h ] abundances , and also model atmosphere parameters , were based on equivalent width measurements for an average of 45 fe i lines and 4 fe ii lines per star . the equivalent widths were measured manually using the semi automated code developed for johnson et al . ( 2014 ) , which fits individual or blended gaussian profiles and is aided by a simple machine learning algorithm . the final abundances were calculated using the _ abfind _ driver of the lte line analysis code moogchris / moog.html . ] ( sneden 1973 ; 2010 version ) . we adopted the same solar abundance as cordero et al . ( 2014 ) of log @xmath12(fe)@xmath19=7.52 . the list of lines and atomic data used for this work is provided in table 3 , and the log(gf ) values were redetermined using a high s / n day light sky spectrum taken with m2fs . for the lines in common between this work and cordero et al . ( 2014 ) , the average difference in adopted log(gf ) values was zero for both fe i ( @xmath4=0.09 ) and fe ii ( @xmath4=0.04 ) . the random measurement uncertainty estimated by @xmath4/@xmath35 , where @xmath4 is the line to line dispersion and n is the number of lines measured , is relatively small for both iron species . specifically , the average random measurement uncertainty for log @xmath12(fe i ) is 0.02 and for log @xmath12(fe ii ) is 0.04 . the [ fe / h ] error column listed in table 2 represents the combined uncertainty for both fe i and fe ii abundances . additional uncertainty in the iron abundance determination comes from errors in the model atmosphere parameters . we estimate that the uncertainty in t@xmath31 , log(g ) , [ m / h ] , and vt , based solely on enforcing excitation equilibrium , ionization equilibrium , and removing trends in log @xmath12(fe i ) as a function of line strength , are : 50 k , 0.10 ( cgs ) , 0.05 dex , and 0.10 km s@xmath3 , respectively . the combined sensitivity of both [ fe i / h ] and [ fe ii / h ] to these changes in model atmosphere parameters is 0.09 dex , on average . as mentioned in section 3.1 , we did not apply any corrections to the [ fe i / h ] and [ fe ii / h ] abundances for departures from lte . however , lapenna et al . ( 2014 ) used high resolution ( r@xmath1848,000 ) spectra of 24 agb and 11 rgb stars in 47 tuc to derive [ fe i / h ] and [ fe ii / h ] abundances and found that only in the rgb sample did the neutral and singly ionized abundances match . lapenna et al . ( 2014 ) further conclude that for agb stars fe i lines should not be used to derive [ fe / h ] abundances and that determining surface gravity from ionization equilibrium , which is the method used here , is also invalid for agb stars . the strong difference in fe i line formation due to departures from lte in similar temperature agb but not rgb stars is a puzzling result and does not match recent theoretical investigations ( e.g. , bergemann et al . 2012 ; lind et al . we note that ivans et al . ( 2001 ) also encountered problems deriving model atmosphere parameters for evolved stars in the more metal poor ( [ fe / h]@xmath231.2 ) globular cluster m5 , and also decided to set the [ fe / h ] scale using only fe ii and photometric log(g ) values . for the lines analyzed in this work , the bergemann et al . ( 2012 ) and lind et al . ( 2012 ) calculations suggest departures from lte should affect [ fe i / h ] at about the 0.05 dex level in both rgb and agb stars . in both populations the [ fe ii / h ] abundance is mostly unaffected because fe ii is the dominant ionization state . however , the _ difference _ in [ fe / h ] abundances and model atmosphere parameters derived for rgb and agb stars using ionization and excitation equilibrium is not expected to produce large systematic offsets when assuming lte . in fact , after reconciling a small difference in temperature scale between our rgb and agb samples ( see section 4 ) , we do not find any significant difference in [ fe / h ] for the two populations . finally , we note that lapenna et al . ( 2014 ) and this work have four agb stars in common . a comparison of the model atmosphere parameters , [ fe / h ] abundances , and radial velocities is provided in table 4 . we find average differences in t@xmath31 , log(g ) , [ fe i / h ] , [ fe ii / h ] , vt , and rv@xmath1 of 50 k , 0.12 ( cgs ) , @xmath150.12 dex , @xmath150.05 dex , 0.09 km s@xmath3 , and 0.11 km s@xmath3 , respectively . the surface gravity differences are even smaller if the most highly discrepant star ( 2m002358527206177 ) is removed . the [ fe i / h ] values from lapenna et al . ( 2014 ) are the most discrepant abundances ; however , we derive similar [ fe ii / h ] abundances ( within the stated uncertainties ) to lapenna et al . ( 2014 ) while simultaneously solving for the model atmosphere parameters via spectroscopic methods and obtaining identical [ fe i / h ] and [ fe ii / h ] ratios . the conflicting observational claims between this work and lapenna et al . ( 2014 ) suggest that a better understanding of rgb and agb atmospheres is needed before ionization equilibrium is fully dismissed as a viable method for determining [ fe / h ] and surface gravity in agb ( but not rgb ) stars . however , we can not rule out that the failure of 7 agb stars in our sample ( 17@xmath6 ) to converge to a stable spectroscopic solution is related to the discrepancies found by lapenna et al . ( 2014 ) and ivans et al . ( 2001 ) . for consistency with the cordero et al . ( 2014 ) [ na / fe ] data , we measured sodium abundances using the same line list ( see also table 3 for a summary of the transition parameters for the na i lines ) , the _ synth _ driver in moog , and adopted log @xmath12(na)@xmath19=6.33 . the dominant molecular contaminator near the 6154/6160 na i lines in the temperature and metallicity regime analyzed here is cn . while we did not have individual c , n , and o abundances for the target stars , the local cn features were fit assuming that the stars were well mixed ( e.g. , [ c / fe]=0.5 ; [ o / fe][email protected] ; @xmath16c/@xmath17c=4 ) ; the nitrogen abundance was varied as a free parameter . the cn line list was adopted from the kurucz ( 1994 ) database ( but see also a recent update by sneden et al . a sample spectrum synthesis for a typical m2fs spectrum is shown in figure [ f6 ] . the final [ na / fe ] values given in table 2 do not include any corrections for departures from lte in the analysis . however , the log @xmath12(na ) non lte corrections are @xmath50.10 dex in an absolute sense ( e.g. , lind et al . 2011 ; thygesen et al . 2014 ) , and the differential non lte corrections for [ na / fe ] due to surface gravity differences between agb and rgb stars is typically @xmath360.05 dex . the random measurement errors for log @xmath12(na i ) are only slightly larger than for @xmath12(fe i ) , with an average of 0.03 dex . this value reflects the difference in sodium abundance between the 6154 and 6160 na i lines . the [ na / fe ] error column of table 2 takes into account both the measurement uncertainty in [ fe / h ] and [ na / h ] . as with iron , additional sources of uncertainty are due to errors in the derived model atmosphere parameters . using the same estimates as in section 3.2 , the average uncertainty from model atmosphere parameters alone in the [ na / fe ] ratio is 0.05 dex . although recent color magnitude diagrams of 47 tuc suggest the cluster has a complex star formation history ( anderson et al . 2009 ; di criscienzo et al . 2010 ; nataf et al . 2011 ; milone et al . 2012 ; monelli et al . 2013 ; li et al . 2014 ) , there is no evidence supporting a substantial spread in [ fe / h ] . instead , literature work suggests the cluster has an average [ fe / h]@xmath230.60 to 0.80 , with a star to star dispersion @xmath50.10 dex ( e.g. , carretta & gratton 1997 ; alves brito et al . 2005 ; wylie et al . 2006 ; koch & mcwilliam 2008 ; carretta et al . 2009 ; gratton et al . 2013 ; cordero et al . 2014 ; dobrovolskas et al . 2014 ; thygesen et al . 2014 ) . in agreement with past work , we find an average [ fe / h]=0.68 and a star to star dispersion of @xmath4=0.08 . this result is similar to , but more metal rich than , the rgb flames giraffe sample of cordero et al . ( 2014 ) that found @xmath0[fe / h]@xmath2=0.75 ( @xmath4=0.10 ) . the 0.07 dex difference in [ fe / h ] between the agb and rgb samples is possibly tied to a systematic offset in the temperature scales between the two data sets . a comparison of the spectroscopically determined t@xmath31 values with those derived from photometric spectral energy distribution fitting in mcdonald et al . ( 2011 ) finds systematic offsets of 20 k ( @xmath4=74 k ) for the rgb sample and @xmath1535 k ( @xmath4=71 k ) for the agb sample . if the [ fe / h ] abundances are redetermined using temperatures ( and corresponding gravities ) that are 35 k lower and 20 k higher for the agb and rgb stars , respectively , then the agb [ fe / h ] values decrease by 0.05 dex and the rgb [ fe / h ] values increase by 0.02 dex . this brings both data sets into agreement at @xmath0[fe / h]@xmath2=0.73 . in contrast to [ fe / h ] , 47 tuc exhibits a significant spread in [ na / fe ] abundance . for the agb stars analyzed here , [ na / fe ] ranges from 0.11 to @xmath150.62 with @xmath0[na / fe]@[email protected] ( @xmath4=0.17 ) . the large dispersion in [ na / fe ] is typical for globular clusters ( e.g. , see reviews by kraft 1994 ; gratton et al . 2004 ; 2012 ) , and the large fraction of stars with [ na / fe]@xmath370 matches previous observations of 47 tuc rgb , horizontal branch , and main sequence turn off stars ( e.g. , carretta et al . 2004 ; alves brito et al . 2005 ; koch & mcwilliam 2008 ; carretta et al . 2009 ; dorazi et al . 2010 ; worley & cottrell 2012 ; gratton et al . 2013 ; cordero et al . 2014 ; dobrovolskas et al . 2014 ; thygesen et al . the possible temperature scale difference between the agb and rgb samples also affects sodium . however , the change in [ na / fe ] is only an increase of 0.02 dex for the agb stars and a decrease of 0.01 dex for the rgb stars . a more detailed comparison between the agb and rgb [ na / fe ] abundances is provided in section 5 . radial velocities were also measured for each star in order to verify cluster membership . we used the cross correlation code xcsao ( kurtz & mink 1998 ) , and the heliocentric correction was determined using the iraf task _ rvcorrect_. the velocities were determined relative to a synthetic rest frame spectrum with t@xmath31=4600 k , log(g)=1.5 ( cgs ) , [ fe / h]=0.70 , and vt=1.5 km s@xmath3 , and the xcsao routine was only run on order 55 ( see table 1 ) of a single exposure for each star . order 55 was selected because it exhibits several strong lines but does not contain many problematic features ( e.g. , telluric bands ; unusually strong or broad lines ) . the average measurement uncertainty returned by xcsao for the radial velocities was 0.18 km s@xmath3 . we find an average heliocentric radial velocity of 18.56 km s@xmath3 ( @xmath4=10.21 km s@xmath3 ) . the average cluster velocity determined here is in good agreement with past work , which ranges from 16.86 km s@xmath3 to 22.43 km s@xmath3 ( dinescu et al . 1999 ; carretta et al . 2004 ; alves brito et al . 2005 ; carretta et al . 2009 ; dobrovolskas et al . 2014 ; lapenna et al . 2014 ) . all of the stars observed with m2fs have velocities consistent with cluster membership . recent work mentioned in section 1 suggests that agb and rgb stars in some globular clusters may show distinctly different chemical compositions . in particular , the agb populations are suspected of having systematically lower ratios of second generation ( cn strong ; na enhanced ; he enhanced ) to first generation ( cn weak ; na poor ; he normal ) stars , compared to the rgb populations . evidence linking the minimum mass and blue extension of stars along the horizontal branch to the agb / rgb star count ratio ( gratton et al . 2010 ) indicates that the apparent composition difference between agb and rgb stars is likely due to some rgb stars not evolving through the agb phase , rather than an _ in situ _ process altering the envelope composition of some stars before or after the horizontal branch phase . the apparent loss of na rich second generation stars between the rgb and agb is most evident in the metal poor blue horizontal branch cluster ngc 6752 , which has a na poor / na rich ratio of 30:70 on the rgb and a 100:0 ratio on the agb ( campbell et al . 2013 ) . furthermore , sandquist & bolte ( 2004 ) note that ngc 6752 exhibits a low ratio of agb to horizontal branch stars , and interpret this observation as an indication that some fraction of rgb stars fail to evolve through the agb phase . since the number of dedicated studies comparing rgb and agb abundance patterns is still small ( e.g. , see campbell et al . 2006 ; their table 1 ) , evidence is insufficient to determine whether the peculiar lack of na rich agb stars in ngc 6752 is more the exception or the rule . therefore , the data provided here offer insight into a possible counter example . 47 tuc is approximately 10@xmath29 more massive than ngc 6752 ( e.g. , pryor & meylan 1993 ) , 7@xmath29 more metal rich , contains almost exclusively red horizontal branch stars ( e.g. , lee et al . 1994 ; see also figure [ f2 ] ) , and has a high agb / rgb ratio ( gratton et al . 2010 ) . in the left panels of figure [ f7 ] , we compare the [ na / fe ] abundances between the agb and rgb samples in 47 tuc as a function of t@xmath31 , as a binned distribution , and as a cumulative distribution . the agb [ na / fe ] distribution appears to be shifted to lower values than the rgb population , and also has fewer stars with [ na / fe]@[email protected] but more with [ na / fe]@[email protected] . after correcting for the agb / rgb temperature scale differences , comparing only the average values yields @xmath0[na / fe]@[email protected] for the agb sample and @xmath0[na / fe]@[email protected] for the rgb flames giraffe sample . similarly , the ratio of na poor to na rich stars is 63:37 for the agb sample and 45:55 for the rgb [email protected] , after accounting for the agb / rgb temperature scale difference , and is based on the rgb [ na / fe ] distribution ( see also figure [ f7 ] ) . ] . the 0.12 dex average [ na / fe ] abundance difference between agb and rgb stars in 47 tuc is similar to , though less extreme than , the 0.34 dex average difference found by campbell et al . ( 2013 ) for ngc 6752 ( see also the right panels of figure [ f7 ] ) . however , an examination of the dispersion ( @xmath4 ) and interquartile range ( iqr ) values of the agb and rgb stars in 47 tuc , which are mostly insensitive to systematic reduction or analysis differences between data sets , reveals that the two populations are similar . specifically , the 47 tuc agb stars have @xmath4@xmath38=0.17 and iqr@xmath38=0.23 and the rgb stars have @xmath4@xmath39=0.18 and iqr@xmath39=0.27 . these values strongly contrast with the results of ngc 6752 in which the agb has @xmath4@xmath38=0.09 and iqr@xmath38=0.10 and the rgb has @xmath4@xmath39=0.28 and iqr@xmath39=0.37 . the stark contrast in agb / rgb [ na / fe ] dispersion and iqr distributions between the two clusters suggests that only a small fraction ( @xmath3620@xmath6 ) of na rich rgb stars in 47 tuc fail to evolve through at least the early agb phase , instead of the 100@xmath6 fraction observed in ngc 6752 . color magnitude diagram analyses also support the notion that both na poor and na rich rgb stars evolve through the horizontal branch and early agb phases in 47 tuc . aside from previous spectroscopic detections of both cn strong ( na rich ) and cn weak ( na poor ) stars on 47 tuc s agb ( e.g. , mallia 1978 ; briley 1997 ) , we note that monelli et al . ( 2013 ) make use of the ( u b)(b i ) color index , which is correlated with [ na / fe ] , and find a bimodal agb color spread . these data indicate that at least two stellar populations , containing roughly equal proportions but with different light element chemistry , are present along 47 tuc s agb . furthermore , color magnitude diagrams do not show a significant population of blue horizontal branch stars that may fail to ascend the agb ( e.g. , anderson et al . 2009 ; bergbusch & stetson 2009 ; dieball et al . 2009 ; milone et al . 2012 ) , and stars spanning the entire rgb [ na / fe ] range are found on the red horizontal branch ( gratton et al . this contrasts with the color magnitude diagram of ngc 6752 , which shows a significant population of blue and extreme horizontal branch stars ( e.g. , grundahl et al . 1999 ) , and also is known to contain agb manqu stars ( e.g. , momany et al . however , the horizontal branch differences between 47 tuc and ngc 6752 are largely driven by metallicity , as is qualitatively illustrated in the isochrone tracks of figure [ f8 ] . since it is clear from figure [ f7 ] that the rgb and agb [ na / fe ] distributions are significantly different in absolute abundance _ and _ dispersion between 47 tuc and ngc 6752 , the question remains as to what ties together rgb composition and post rgb evolution . as mentioned previously , a commonly adopted solution is that na rich stars have a higher he abundance , and that the lower masses of he rich / na rich horizontal branch stars prevent them from evolving through the agb phase . at least for the extreme case of ngc 6752 , charbonnel et al . ( 2013 ) note that stars with masses less than about 0.735 m@xmath19 and [email protected] do not ascend the agb . however , the required he enhancements are significantly larger than those estimated from photometry ( @[email protected] ; milone et al . 2013 ) , but charbonnel et al . ( 2013 ) further suggest the maximum photometric he spread could be underestimated . for 47 tuc , the he spread is estimated to be even smaller ( @[email protected] ; e.g. , di criscienzo et al . 2010 ; nataf et al . 2011 ; milone et al . 2012 ) than in ngc 6752 , and as figure [ f8 ] shows even rgb stars with y=0.3 are still likely to ascend the agb at 47 tuc s metallicity ( e.g. , see also valcarce et al . 2012 and references therein ) . despite having significantly different metallicities and horizontal branch morphologies , the agb / rgb composition difference in 47 tuc may be more analagous to the case of m13 where johnson & pilachowski ( 2012 ) found that only the most na rich _ and _ o poor stars ( presumably also the most he rich stars ) failed to reach the agb abundances for many stars in the agb data set because the preferred 6300 [ o i ] line was severely contaminated by telluric bands and the 6300 sky emission feature . ] . we note that the paucity of o poor stars on m13 s agb , albeit from a sample of only @xmath1810 agb stars , is observed despite the failure ( so far ) of investigators to find a strongly he enhanced population ( sandquist et al . 2010 ; smith et al . could a different parameter be driving the agb / rgb composition disparity ? campbell et al . ( 2013 ) proposed that , in addition to he enhancement for the na rich stars , a substantial _ ad hoc _ increase in the mass loss rate ( 20@xmath29 the rgb value ) for ngc 6752 stars on the blue horizontal branch may be required to prevent further evolution up the agb . however , cassisi et al . ( 2014 ) note that the required mass loss rate of @xmath1810@xmath40 m@xmath21 year@xmath3 is several orders of magnitude higher than those allowed by current analyses of hot horizontal branch and b subdwarf stars . additionally , horizontal branch simulations by cassisi et al . ( 2014 ) that accurately reproduce the r@xmath41 parameter ( n@xmath42/n@xmath43 ) for ngc 6752 , m3 , and m13 , but do not invoke enhanced mass loss , also predict that @xmath1850@xmath6 of ngc 6752 s agb stars should be na rich ( albeit with lower y than the most extreme values ) . these results indicate that enhanced mass loss for a particular subset of horizontal branch stars is unlikely to be the key link between rgb composition and post rgb evolution . furthermore , even if processes associated with the radiative levitation of metals in blue horizontal branch stars played a role in mass loss and/or the subsequent envelope composition of agb stars , 47 tuc lacks a significant population of horizontal branch stars with t@xmath31@xmath1411,000 k that would experience this effect ( e.g. , behr et al . the data presented here indicate that the average [ na / fe ] abundance of agb stars in the galactic globular cluster 47 tuc is 0.12 dex lower than that of the rgb stars . additionally , the ratio of na poor to na rich agb stars is 63:37 compared to 45:55 for the rgb stars . however , both populations exhibit a similar dispersion and iqr of [ na / fe ] . this result strongly contrasts with that found by campbell et al . ( 2013 ) in the globular cluster ngc 6752 where the agb consists of only na poor stars with a small star to star [ na / fe ] dispersion . the 47 tuc [ na / fe ] distribution suggests that @xmath520@xmath6 of na rich rgb stars fail to reach the early agb , which is similar to the case of m13 where only the most na rich and o poor stars are missing from the agb . for 47 tuc , the cluster s relatively high metallicity leads to a predominantly red horizontal branch morphology , and no significant population of hot horizontal branch and agb manqu stars , which likely make up the missing na rich component of ngc 6752 s ( and m13 s ) agb , has been found . unlike the case for ngc 6752 , however , in 47 tuc it seems that at least some na rich rgb stars evolve through the early agb phase . we conclude that rgb [ na / fe ] abundances alone are not a unique predictor of future agb evolution in all clusters . several remaining questions must be addressed before a definitive link between stellar composition , cluster horizontal branch morphology , and agb evolution can be formed . for example , if he abundance , in addition to metallicity and age , is a significant factor in defining a globular cluster s horizontal branch morphology and a star s post rgb evolution , why do the @xmath22y values estimated from photometry appear too low in a cluster such as ngc 6752 to prevent na rich stars from ascending the agb ? are the he enhancements actually underestimated , as was suggested by charbonnel et al . what roles do additional parameters such as mass loss , he core rotation , cno abundance , and/or environment play in defining a star s horizontal branch and eventual agb evolution ? what separates clusters such as 47 tuc and m13 , which lose only a fraction of na rich stars before the agb , to those such as ngc 6752 , which lose 100@xmath6 ? a deeper understanding of the critical parameters controlling horizontal branch and agb evolution , along with additional studies comparing large samples of rgb and agb chemical compositions , seems required in order to fully place the failure ( or not ) of some na rich stars to ascend the agb into context with the developing narrative of forming multiple populations in globular clusters . we thank the anonymous referee for a thoughtful and constructive report that led to improvement of the manuscript . this research has made use of nasa s astrophysics data system bibliographic services . this publication has made 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 . c.i.j . gratefully acknowledges support from the clay fellowship , administered by the smithsonian astrophysical observatory . c.i.j . would like to thank andrea dupree and nelson caldwell for helpful discussions regarding data reduction . c.a.p . gratefully acknowledges support from the daniel kirkwood research fund at indiana university . is supported by sonderforschungs bereich sfb 881 the milky way system " ( subproject a5 ) of the german research foundation ( dfg ) . m.m . is grateful for support from the national science foundation to develop m2fs ( ast0923160 ) and carry out the observations reported here ( ast1312997 ) and to the university of michigan for its direct support of m2fs construction and operation . hinkle , k. , wallace , l. , valenti , j. , & harmer , d. 2000 , visible and near infrared atlas of the arcturus spectrum 3727 - 9300 a ed . kenneth hinkle , lloyd wallace , jeff valenti , and dianne harmer . ( san francisco : asp ) isbn : 1 - 58381 - 037 - 4 , 2000 58 & 6122@xmath446205 & 6138@xmath446190 + 57 & 6152@xmath446315 & 6190@xmath446310 + 56 & 6262@xmath446426 & 6310@xmath446410 + 55 & 6376@xmath446543 & 6410@xmath446525 + 54 & 6494@xmath446664 & 6525@xmath446650 + 53 & 6617@xmath446721 & 6650@xmath446720 + 2m00235484@xmath447210009 & 5.978539 & @xmath4472.166939 & 11.266 & 10.697 & 10.575 & 4600 & 1.71 & @xmath440.72 & 0.05 & 1.70 & @xmath150.18 & 0.05 & @xmath445.47 & 0.14 + 2m00235632@xmath447210590 & 5.984693 & @xmath4472.183060 & 11.114 & 10.514 & 10.412 & 4500 & 1.60 & @xmath440.71 & 0.06 & 1.70 & @xmath150.26 & 0.08 & @xmath4419.66 & 0.18 + 2m00235701@xmath447202185 & 5.987543 & @xmath4472.038475 & 11.145 & 10.593 & 10.445 & 4825 & 2.18 & @xmath440.57 & 0.04 & 1.95 & @xmath150.25 & 0.05 & @xmath446.03 & 0.19 + 2m00235728@xmath447207274 & 5.988676 & @xmath4472.124290 & 10.610 & 9.948 & 9.836 & 4400 & 1.47 & @xmath440.69 & 0.03 & 1.65 & @xmath150.06 & 0.07 & @xmath443.12 & 0.14 + 2m00235852@xmath447206177 & 5.993859 & @xmath4472.104919 & 11.005 & 10.436 & 10.317 & 4600 & 1.33 & @xmath440.78 & 0.02 & 1.75 & @xmath150.44 & 0.03 & @xmath441.03 & 0.14 + 2m00240062@xmath447203573 & 6.002621 & @xmath4472.065926 & 10.343 & 9.625 & 9.520 & 4350 & 1.17 & @xmath440.76 & 0.03 & 1.75 & @xmath150.39 & 0.08 & @xmath4426.60 & 0.18 + 2m00240304@xmath447202193 & 6.012702 & @xmath4472.038712 & 10.840 & 10.284 & 10.148 & 4625 & 1.93 & @xmath440.72 & 0.02 & 1.75 & @xmath150.17 & 0.06 & @xmath4428.23 & 0.20 + 2m00240310@xmath447203482 & 6.012947 & @xmath4472.063393 & 11.139 & 10.578 & 10.435 & 4600 & 1.92 & @xmath440.66 & 0.07 & 1.60 & @xmath150.08 & 0.07 & @xmath449.69 & 0.15 + 2m00240330@xmath447203075 & 6.013772 & @xmath4472.052101 & 11.139 & 10.578 & 10.435 & 4550 & 1.67 & @xmath440.76 & 0.06 & 1.95 & @xmath150.48 & 0.06 & @xmath4412.31 & 0.15 + 2m00240427@xmath447206074 & 6.017792 & @xmath4472.102081 & 10.814 & 10.124 & 10.021 & 4450 & 1.73 & @xmath440.68 & 0.06 & 1.40 & @xmath150.08 & 0.06 & @xmath4424.41 & 0.16 + 2m00241142@xmath447206126 & 6.047624 & @xmath4472.103516 & 11.184 & 10.559 & 10.440 & 4600 & 1.72 & @xmath440.72 & 0.03 & 1.55 & @xmath150.05 & 0.03 & @xmath4420.35 & 0.14 + 2m00241462@xmath447204018 & 6.060919 & @xmath4472.067184 & 11.161 & 10.568 & 10.410 & 4600 & 1.85 & @xmath440.65 & 0.05 & 1.60 & @xmath150.22 & 0.06 & @xmath4423.14 & 0.18 + 2m00241531@xmath447202231 & 6.063815 & @xmath4472.039757 & 10.552 & 9.886 & 9.771 & 4400 & 1.42 & @xmath440.58 & 0.03 & 1.60 & @xmath440.04 & 0.04 & @xmath4413.49 & 0.13 + 2m00241755@xmath447204385 & 6.073161 & @xmath4472.077370 & 11.351 & 10.733 & 10.677 & 4600 & 2.11 & @xmath440.59 & 0.05 & 0.90 & @xmath440.08 & 0.06 & @xmath4433.09 & 0.19 + 2m00241912@xmath447208360 & 6.079687 & @xmath4472.143356 & 10.992 & 10.395 & 10.240 & 4425 & 1.13 & @xmath440.82 & 0.06 & 1.55 & @xmath150.25 & 0.06 & @xmath4431.87 & 0.17 + 2m00241945@xmath447211426 & 6.081072 & @xmath4472.195183 & 11.262 & 10.646 & 10.562 & 4750 & 2.03 & @xmath440.56 & 0.05 & 1.70 & @xmath150.28 & 0.05 & @xmath4417.19 & 0.14 + 2m00242074@xmath447206332 & 6.086458 & @xmath4472.109245 & 11.157 & 10.615 & 10.472 & 4550 & 1.27 & @xmath440.75 & 0.03 & 1.45 & @xmath150.05 & 0.03 & @xmath4422.51 & 0.15 + 2m00242572@xmath447203307 & 6.107181 & @xmath4472.058540 & 10.696 & 10.017 & 9.887 & 4450 & 1.15 & @xmath440.68 & 0.05 & 1.70 & @xmath150.30 & 0.05 & @xmath4411.54 & 0.15 + 2m00242763@xmath447205213 & 6.115141 & @xmath4472.089264 & 11.004 & 10.428 & 10.245 & 4550 & 1.64 & @xmath440.67 & 0.08 & 1.70 & @xmath150.18 & 0.09 & @xmath4420.96 & 0.16 + 2m00242846@xmath447205019 & 6.118608 & @xmath4472.083862 & 11.176 & 10.581 & 10.425 & 4625 & 1.65 & @xmath440.76 & 0.03 & 1.75 & @xmath150.32 & 0.05 & @xmath4422.21 & 0.17 + 2m00243106@xmath447207311 & 6.129429 & @xmath4472.125313 & 10.898 & 10.280 & 10.189 & 4700 & 1.90 & @xmath440.61 & 0.04 & 1.70 & @xmath150.49 & 0.05 & @xmath4424.35 & 0.15 + 2m00243337@xmath447204200 & 6.139076 & @xmath4472.072243 & 10.973 & 10.362 & 10.254 & 4600 & 1.95 & @xmath440.61 & 0.06 & 1.60 & @xmath150.29 & 0.07 & @xmath152.90 & 0.13 + 2m00243771@xmath447205100 & 6.157149 & @xmath4472.086121 & 10.324 & 9.651 & 9.524 & 4300 & 1.20 & @xmath440.71 & 0.10 & 1.75 & @xmath150.31 & 0.10 & @xmath4438.05 & 0.22 + 2m00243851@xmath447203038 & 6.160488 & @xmath4472.051079 & 10.958 & 10.304 & 10.228 & 4550 & 1.48 & @xmath440.70 & 0.03 & 1.65 & @xmath150.23 & 0.04 & @xmath4422.80 & 0.17 + 2m00245127@xmath447204588 & 6.213660 & @xmath4472.083000 & 11.161 & 10.543 & 10.434 & 4550 & 1.50 & @xmath440.68 & 0.05 & 1.60 & @xmath150.44 & 0.06 & @xmath4419.99 & 0.17 + 2m00245189@xmath447200031 & 6.216213 & @xmath4472.000862 & 11.237 & 10.639 & 10.519 & 4625 & 1.63 & @xmath440.70 & 0.04 & 1.65 & @xmath150.30 & 0.06 & @xmath4430.43 & 0.23 + 2m00245600@xmath447212266 & 6.233351 & @xmath4472.207397 & 11.125 & 10.510 & 10.421 & 4675 & 2.15 & @xmath440.64 & 0.06 & 1.40 & @xmath150.36 & 0.07 & @xmath150.69 & 0.21 + 2m00250716@xmath447200415 & 6.279846 & @xmath4472.011536 & 10.537 & 9.865 & 9.761 & 4300 & 1.37 & @xmath440.66 & 0.03 & 1.65 & @xmath150.03 & 0.03 & @xmath4417.09 & 0.13 + 2m00250791@xmath447203490 & 6.282998 & @xmath4472.063629 & 11.264 & 10.658 & 10.575 & 4600 & 1.52 & @xmath440.76 & 0.06 & 1.70 & @xmath150.13 & 0.07 & @xmath4433.36 & 0.16 + 2m00250809@xmath447204092 & 6.283737 & @xmath4472.069244 & 11.027 & 10.408 & 10.296 & 4600 & 1.69 & @xmath440.59 & 0.06 & 1.65 & @xmath150.22 & 0.06 & @xmath4423.45 & 0.17 + 2m00251614@xmath447201359 & 6.317278 & @xmath4472.026649 & 10.809 & 10.187 & 10.075 & 4550 & 1.43 & @xmath440.56 & 0.05 & 1.75 & @xmath440.09 & 0.11 & @xmath4412.04 & 0.13 + 2m00253428@xmath447157352 & 6.392834 & @xmath4471.959793 & 10.549 & 9.896 & 9.763 & 4350 & 1.08 & @xmath440.79 & 0.03 & 1.75 & @xmath150.62 & 0.04 & @xmath4418.29 & 0.15 + 2m00253529@xmath447206543 & 6.397057 & @xmath4472.115105 & 11.308 & 10.727 & 10.604 & 4500 & 1.38 & @xmath440.84 & 0.06 & 1.40 & @xmath150.07 & 0.06 & @xmath4417.45 & 0.19 + 2m00254689@xmath447206494 & 6.445395 & @xmath4472.113731 & 11.238 & 10.646 & 10.508 & 4600 & 2.01 & @xmath440.57 & 0.08 & 1.70 & @xmath440.11 & 0.08 & @xmath4413.47 & 0.18 + 2m00255446@xmath447201491 & 6.476949 & @xmath4472.030312 & 10.289 & 9.558 & 9.420 & 4300 & 1.27 & @xmath440.55 & 0.08 & 1.75 & @xmath150.24 & 0.12 & @xmath4429.39 & 0.57 + 6154.23 & 11.0 & 2.101 & @xmath441.57 + 6160.75 & 11.0 & 2.103 & @xmath441.27 + 6145.41 & 26.0 & 3.368 & @xmath443.58 + 6148.65 & 26.0 & 4.320 & @xmath442.63 + 6151.62 & 26.0 & 2.176 & @xmath443.31 + 6157.73 & 26.0 & 4.076 & @xmath441.19 + 6159.37 & 26.0 & 4.607 & @xmath441.88 + 6165.36 & 26.0 & 4.143 & @xmath441.54 + 6173.33 & 26.0 & 2.223 & @xmath442.89 + 6180.20 & 26.0 & 2.727 & @xmath442.71 + 6187.99 & 26.0 & 3.943 & @xmath441.75 + 6213.43 & 26.0 & 2.223 & @xmath442.52 + 6219.28 & 26.0 & 2.198 & @xmath442.33 + 6220.78 & 26.0 & 3.881 & @xmath442.39 + 6226.73 & 26.0 & 3.883 & @xmath442.12 + 6229.23 & 26.0 & 2.845 & @xmath442.93 + 6232.64 & 26.0 & 3.654 & @xmath441.24 + 6240.65 & 26.0 & 2.223 & @xmath443.32 + 6252.56 & 26.0 & 2.404 & @xmath441.63 + 6253.83 & 26.0 & 4.733 & @xmath441.53 + 6270.22 & 26.0 & 2.858 & @xmath442.62 + 6271.28 & 26.0 & 3.332 & @xmath442.76 + 6322.69 & 26.0 & 2.588 & @xmath442.31 + 6335.33 & 26.0 & 2.198 & @xmath442.11 + 6336.82 & 26.0 & 3.686 & @xmath440.59 + 6358.70 & 26.0 & 0.859 & @xmath444.32 + 6362.88 & 26.0 & 4.186 & @xmath442.03 + 6380.74 & 26.0 & 4.186 & @xmath441.35 + 6385.72 & 26.0 & 4.733 & @xmath441.87 + 6392.54 & 26.0 & 2.279 & @xmath443.95 + 6393.60 & 26.0 & 2.433 & @xmath441.48 + 6408.02 & 26.0 & 3.686 & @xmath440.91 + 6411.65 & 26.0 & 3.654 & @xmath440.40 + 6430.85 & 26.0 & 2.176 & @xmath441.89 + 6469.19 & 26.0 & 4.835 & @xmath440.80 + 6472.15 & 26.0 & 4.371 & @xmath442.89 + 6475.62 & 26.0 & 2.559 & @xmath442.88 + 6481.87 & 26.0 & 2.279 & @xmath442.92 + 6483.94 & 26.0 & 1.485 & @xmath445.30 + 6494.98 & 26.0 & 2.404 & @xmath441.15 + 6495.74 & 26.0 & 4.835 & @xmath441.02 + 6496.47 & 26.0 & 4.795 & @xmath440.56 + 6498.94 & 26.0 & 0.958 & @xmath444.61 + 6509.62 & 26.0 & 4.076 & @xmath442.88 + 6518.37 & 26.0 & 2.831 & @xmath442.58 + 6533.93 & 26.0 & 4.558 & @xmath441.29 + 6546.24 & 26.0 & 2.758 & @xmath441.71 + 6551.68 & 26.0 & 0.990 & @xmath445.73 + 6556.79 & 26.0 & 4.796 & @xmath441.61 + 6569.21 & 26.0 & 4.733 & @xmath440.21 + 6581.21 & 26.0 & 1.485 & @xmath444.73 + 6591.31 & 26.0 & 4.593 & @xmath441.98 + 6592.91 & 26.0 & 2.727 & @xmath441.51 + 6593.87 & 26.0 & 2.433 & @xmath442.25 + 6597.56 & 26.0 & 4.795 & @xmath440.98 + 6609.11 & 26.0 & 2.559 & @xmath442.58 + 6609.68 & 26.0 & 0.990 & @xmath445.78 + 6633.41 & 26.0 & 4.835 & @xmath441.26 + 6633.75 & 26.0 & 4.558 & @xmath440.71 + 6634.11 & 26.0 & 4.795 & @xmath441.23 + 6648.08 & 26.0 & 1.011 & @xmath445.79 + 6665.43 & 26.0 & 1.557 & @xmath445.60 + 6710.32 & 26.0 & 1.485 & @xmath444.81 + 6713.74 & 26.0 & 4.795 & @xmath441.43 + 6715.38 & 26.0 & 4.607 & @xmath441.54 + 6716.24 & 26.0 & 4.580 & @xmath441.81 + 6149.26 & 26.1 & 3.889 & @xmath442.66 + 6247.56 & 26.1 & 3.892 & @xmath442.33 + 6369.46 & 26.1 & 2.891 & @xmath444.09 + 6385.45 & 26.1 & 5.553 & @xmath442.63 + 6432.68 & 26.1 & 2.891 & @xmath443.63 + 6456.38 & 26.1 & 3.903 & @xmath442.03 + 6482.20 & 26.1 & 6.219 & @xmath441.67 + 6516.08 & 26.1 & 2.891 & @xmath443.33 + 2m00235852@xmath447206177 & 4600 & 1.33 & @xmath440.78 & @xmath440.78 & 1.75 & @xmath441.03 & 100119 & 4500 & 1.65 & @xmath440.85 & @xmath440.71 & 1.80 & @xmath440.83 + 2m00241142@xmath447206126 & 4600 & 1.72 & @xmath440.72 & @xmath440.72 & 1.55 & @xmath4420.75 & 100142 & 4550 & 1.70 & @xmath440.90 & @xmath440.80 & 1.60 & @xmath4420.41 + 2m00242763@xmath447205213 & 4550 & 1.64 & @xmath440.67 & @xmath440.67 & 1.70 & @xmath4420.96 & 200021 & 4550 & 1.70 & @xmath440.83 & @xmath440.84 & 1.90 & @xmath4421.07 + 2m00242846@xmath447205019 & 4625 & 1.65 & @xmath440.76 & @xmath440.76 & 1.75 & @xmath4422.21 & 200023 & 4575 & 1.75 & @xmath440.81 & @xmath440.79 & 1.80 & @xmath4422.19 +

the pan - starrs1 survey has produced photometry and astrometry covering the 30,000 square degrees . this article describes the photometric and astrometric calibration of this survey .

1612.05242

this is the fifth in a series of seven papers describing the pan - starrs1 surveys , the data reduction techiques and the resulting data products . this paper ( paper v ) describes the final calibration process , and the resulting photometric and astrometric quality . * paper i ) provides an overview of the pan - starrs system , the design and execution of the surveys , the resulting image and catalog data products , a discussion of the overall data quality and basic characteristics , and a brief summary of important results . * paper ii ) describes how the various data processing stages are organised and implemented in the imaging processing pipeline ( ipp ) , including details of the the processing database which is a critical element in the ipp infrastructure . * paper iii ) describes the details of the pixel processing algorithms , including detrending , warping , and adding ( to create stacked images ) and subtracting ( to create difference images ) and resulting image products and their properties . * paper iv ) describes the details of the source detection and photometry , including point - spread - function and extended source fitting models , and the techniques for forced " photometry measurements . * paper vi ) describes the details of the resulting catalog data and its organization in the pan - starrs database . * paper vii ) describes the medium deep survey in detail , including the unique issues and data products specific to that survey . the medium deep survey is not part of data release 1 . ( dr1 ) the pan - starrs1 filters and photometric system have already been described in detail in @xcite . from may 2010 through march 2014 , the pan - starrs science consortium used the 1.8 m telescope to perform a set of wide - field science surveys . these surveys are designed to address a range of science goals included the search for hazardous asteroids , the study of the formation and architecture of the milky way galaxy , and the search for type ia supernovae to measure the history of the expansion of the universe . the wide - field telescope consists of a 1.8 meter diameter @xmath2/4.4 primary mirror with an 0.9 m secondary , producing a 3.3 degree field of view @xcite . the optical design yields low distortion and minimal vignetting even at the edges of the illuminated region . the optics , in combination with the natural seeing , result in generally good image quality : the median image quality for the 3@xmath3 survey is fwhm = ( 1.31 , 1.19 , 1.11 , 1.07 , 1.02 ) arcseconds for ( ) , with a floor of @xmath4 arcseconds . the camera @xcite is a mosaic of 60 edge - abutted @xmath5 pixel back - illuminated ccid58 orthogonal transfer arrays manufactured by lincoln laboratory @xcite . the ccds have 10 @xmath6 m pixels subtending 0.258 arcsec and are 70@xmath6 m thick . the detectors are read out using a stargrasp ccd controller , with a readout time of 7 seconds for a full unbinned image @xcite . the active , usable pixels cover @xmath7% of the fov . nightly observations are conducted remotely from the advanced technology research center in kula , the main facility of the university of hawaii s institute for astronomy operations on maui . during the science survey , images obtained by the system were stored first on computers at the summit , then copied with low latency via internet to the dedicated data analysis cluster located at the maui high performance computer center in kihei , maui . images obtained by are automatically processed in real time by the image processing pipeline ( ipp , * ? ? ? real - time analysis goals are aimed at feeding the discovery pipelines of the asteroid search and supernova search teams . the data obtained for the science survey has also been used in three additional complete re - processing of the data : processing versions 1 , 2 , and 3 ( pv1 , pv2 , and pv3 ) . the real - time processing of the data is considered `` pv0 '' . except as otherwise noted , the pv3 analysis of the data is used for the purpose of this article . the data processing steps are described in detail by @xcite and @xcite . in summary , individual images are detrended : non - linearity and bias corrections are applied , a dark current model is subtracted and flat - field corrections are applied . the -band images are also corrected for fringing : a master fringe pattern is scaled to match the observed fringing and subtracted . mask and variance image arrays are generated with the detrend analysis and carried forward at each stage of the ipp processing . source detection and photometry are performed for each chip independently . as discussed below , preliminary astrometric and photometric calibrations are performed for all chips in a single exposure in a single analysis . chip images are geometrically transformed based on the astrometric solution into a set of pre - defined pixel grids covering the sky , called skycells . these transformed images are called the warp images . sets of warps for a given part of the sky and in the same filter may be added together to generate deeper ` stack ' images . psf - matched difference images are generated from combinations of warps and stacks ; the details of the difference images and their calibration are outside of the scope of this article . astronomical objects are detected and characterized in the stacks images . the details of the analysis of the sources in the stack images are discussed in @xcite , but in brief these include psf photometry , along with a range of measurements driven by the goals of understanding the galaxies in the images . because of the significant mask fraction of the gpc1 focal plane , and the varying image quality both within and between exposures , the effective psf of the ps1 stack images is highly variable . the psf varies significantly on scales as small as a few to tens of pixels , making accurate psf modelling essentially infeasible . the psf photometry of sources in the stack images is thus degraded significantly compared to the quality of the photometry measured for the individual chip images . to recover most of the photometric quality of the individual chip images , while also exploiting the depth afforded by the stacks , the pv3 analysis make use of forced photometry on the individual warp images . psf photometry is measured on the warp images for all sources which are detected in the stack images images . the positions determined in the stack images are used in the warp images , but the psf model is determined for each warp independently based on brighter stars in the warp image . the only free parameter for each object is the flux , which may be insignificant or even negative for sources which are near the faint limit of the stack detections . when the fluxes from the individual warp images are averaged , a reliable measurement of the faint source flux is determined . the details of this analysis are described in detail in magnier et al @xcite . in this article , we discuss the photometric calibration of the individual exposures , the stacks , and the warp imags . we also discuss the astrometric calibration of the individual exposures and the stack images . three somewhat distinct astrometric models are employed within the ipp at different stages . the simplest model is defined independently for each chip : a simple tan projection as described by @xcite is used to relate sky coordinates to a cartesian tangent - plane coordinate system . a pair of low - order polynomials are used to relate the chip pixel coordinates to this tangent - plane coordinate system . the transforming polynomials are of the form : @xmath8 where @xmath9 are the tangent plane coordinates , @xmath10 are the coordinates on the 60 gpc1 chips , and @xmath11 are the polynomial coefficients for each order . in the analysis , @xmath12 where the order of the fit , @xmath13 , may be 1 to 3 , under the restriction that sufficient stars are needed to constrain the order . a second form of astrometry model which yields somewhat higher accuracy consists of a set of connected solutions for all chips in a single exposure . this model also uses a tan projection to relate the sky coordinates to a locally cartesian tangent plane coordinate system . a set of polynomials is then used to relate the tangent plane coordinates to a focal plane coordinate system , @xmath14 : @xmath15 this set of polynomial accounts for effects such as optical distortion in the camera and distortions due to changing atmospheric refraction across the field of the camera . since these effects are smooth across the field of the camera , a single pair of polynomials can be used for each exposure . like in the chip analysis about , the code restricts the exponents with the rule @xmath12 where the order of the fit , @xmath16 , may be 1 to 3 , under the restriction that sufficient stars are needed to constrain the order for each chip , a second set of polynomials describes the transformation from the chip coordinate systems to the focal coordinate system : @xmath17 a third form of the astrometry model is used in the context of the calibration determined within the dvo database system . we retain the two levels of transformations ( chip @xmath18 focal plane @xmath18 tangent plane ) , but the relationship between the chip and focal plane is represented with only the linear terms in the polynomial , supplemented by a course grid of displacements , @xmath19 sampled across the coordinate range of the chip . this displacement grid may have a resolution of up to @xmath20 samples across the chip . the displacement for a specific chip coordinate value is determined via bilinear interpolation between the nearest sample points . thus , the chip to focal - plane transformation may be written as : @xmath21 * wcs keywords * when this polynomial representation is written to the output files , a set of wcs keywords are used to define the astrometric transformation elements . it is necessary to transform the simply polynomials above into an alternate form : @xmath22 as images are processed by the data analysis system , every exposure is calibrated individually with respect to a photometric and astrometric database . the goal of this calibration step is to generate a preliminary astrometric calibration , to be used by the warping analysis to determine the geometric transformation of the pixels , and preliminary photometric transformation , to be used by the stacking analysis to ensure the warps are combined using consistent flux units . the program used for the real - time calibration , , loads the measurements of the chip detections from their individual -format files . it uses the header information populated at the telescope to determine an initial astrometric calibration guess based on the position of the telescope boresite right ascension , declination and position angle as reported by the telescope & camera subsystems . using the initial guess , loads astrometric and photometric data from the reference database . during the course of the ps1sc survey , several reference databases have been used . for the first 20 months of the survey , used a reference catalog with synthetic ps1 photometry generated by the pan - starrs ipp team based on based combined photometry from tycho ( b , v ) , usno ( red , blue , ir ) , and 2mass @xmath23 . the astrometry in the database was from 2mass . after 2012 may , a reference catalog generated from internal re - calibration of the pv0 analysis of ps1 photometry and astrometry was used for the reference catalog . coordinates and calibrated magnitudes of stars from the reference database are loaded by . a model for the positions of the 60 chips in the focal plane is used to determine the expected astrometry for each chip based on the boresite coordinates and position angle reported by the header . reference stars are selected from the full field of view of the gpc1 camera , padded by an additional 25% to ensure a match can be determined even in the presence of substantial errors in the boresite coordinates . it is important to choose an appropriate set of reference stars : if too few are selected , the chance of finding a match between the reference and observed stars is diminished . in addition , since stars are loaded in brightness order , a selection which is too small is likely to contain only stars which are saturated in the gpc1 images . on the other hand , if too many reference stars are chosen , there is a higher chance of a false - positive match , especially as many of the reference stars may not be detected in the gpc1 image . the seletion of the reference stars includes a limit on the brightest and fainted magnitude of the stars selected . the astrometric analysis is necessarily performed first ; after the astrometry is determined , an automatic byproduct is a reliable match between reference and observed stars , allowing a comparison of the magnitudes to determine the photometric calibration . the astrometric calibration is performed in two major stages : first , the chips are fitted independently with independent models for each chip . this fit is sufficient to ensure a reliable match between reference stars and observed sources in the image . next , the set of chip calibrations are used to define the transformation between the focal plane coordinate system and the tangent plane coordinate system . the chip - to - focal plane transformations are then determined under the single common focal plane to tangent plane transformation . the first step of the analysis is to attempt to find the match between the reference stars and the detected objects . uses 2d cross correlation to search for the match . the guess astrometry calibration is used to define a predicted set of @xmath24 values for the reference catalog stars . for all possible pairs between the two lists , the values of @xmath25 are generated . the collection of @xmath26 values are collected in a 2d histogram with sampling of 50 pixels and the peak pixel is identified . if the astrometry guess were perfect , this peak pixel would be expected to lie at ( 0,0 ) and contain all of the matched stars . however , the astrometric guess may be wrong in several ways . an error in the constant term above , @xmath27 shifts the peak to another pixel , from which @xmath27 can easily be determined . an error in the plate scale or a rotation will smear out the peak pixel potentially across many pixels in the 2d histogram . to find a good match in the face of plate scale and rotation errors , the cross correlation analysis above is performed for a series of trials in which the scale and rotation are perturbed from the nominal value by a small amount . for each trial , the peak pixel is found and a figure of merit is measured . the figure of merit is defined as @xmath28 where @xmath29 are the second moment of @xmath30 for the star pairs associated with the peak pixel , and @xmath31 is the number of star pairs in the peak . this figure of merit is thus most sensitive to a narrow distribution with many matched pairs . for the ps1 exposures , rotation offsets of ( -1.0 , -0.5 , 0.0 , 0.5 , 1.0 ) degrees , and plate scales of ( + 1% , 0 , -1% ) of the nominal plate scale are tested . the best match among these 15 cross - correlation tests is selected and used to generate a better astrometry guess for the chip . the astrometry solution from the cross correlation step above is again used to selected matches between the reference stars and observed stars in the image . the matching radius starts off quite large , and a series of fits is performed to generate the transformation between chip and tangent plane coordinates . three clipping iterations are performed , with outliers @xmath32 rejected on each pass , where here @xmath33 is determined from the distribution of the residuals in each dimension ( x , y ) independently . after each fit cycle , the matches are redetermined using a smaller radius and the fit re - tried . the astrometry solutions from the independent chip fits are used to generate a single model for the camera - wide distortion terms . the goal is to determine the two stage fit ( chip @xmath18 focal plane @xmath18 tangent plane ) . there are a number of degenerate terms between these two levels of transformation , most obviously between the parameters which define the constant offset from chip to focal plane ( @xmath34 ) and those which define the offset from focal plane to tangent plane ( @xmath35 ) . we limit ( @xmath35 ) to be 0,0 to remove this degeneracy . the initial fit of the astrometry for each chip follows the distortion introduced by the camera : the apparent plate scale for each chip is the combination of the plate scale at the optical axis of the camera , modified by the local average distortion . to isolate the effect of distortion , we choose a single common plate scale for the set of chips and re - define the chip @xmath18 sky calibrations as a set of chip @xmath18 focal plane transformation using that common pixel scale . we can now compare the observed focal plane coordinates , derived from the chip coordinates , and the tangent plane coordiantes , derived from the projection of the reference coordinates . one caveat is that the chip reference coordinates are also degenerate with the fitted distortion . in order to avoid being sensitive to the exact positions of the chips at this stage , we measure the local gradient between the focal plane and tangent plane coordinate systems . we then fit the gradient with a polynomial of order 1 less than the polynomial desired for the distortion fit . the coefficients of the gradient fit are then used to determine the coefficients for the polynomials representing the distortion . once the common distortion coming from the optics and atmosphere have been modeled , determines polynomial transformations from the 60 chips to the focal plane coordinate system . in this stage , 5 iterations of the chip fits are performed . before each iteration , the reference stars and detected objects are matched using the current best set of transformations . these fits start with low order ( 1 ) and large matching radius . as the iterations proceed , the radius is reduced and the order is allowed to increaes , up to 3rd order for the final iterations . after the astrometric calibration has finished , the photometric calibration is performed by . when the reference stars are loaded , the apparent magnitude in the filter of interest is also loaded . stars for which the reference magnitude is brighter than ( ) = ( 19 , 19 , 18.5 , 18.5 , 17.5 ) are used to determine the zero points by comparison with the instrumental magnitudes . for the pv3 analysis , an outlier - rejecting median is used to measure the zero point . for early versions of the analysis , when the reference catalog used synthetic magnitudes , it was necessary to search for the blue edge of the distribution : the synthetic magnitude poorly predicted the magnitudes of stars in the presence of significant extinction or for the very red stars , making the blue edge somewhat more reliable . note that we do not include an airmass correction in this zero point analysis : the airmass correction is folded into the observed zero point . the zero point may be measured separately for each chip or as a single value for the entire exposure ; the latter option was used for the pv3 analysis . the calibrations determined by as saved as part of the header information in the output fits tables . a single multi - extension fits table is written using the format . in these files , the measurements from each chip are written as a separate fits table . a second fits extension for each chip is used to store the header information from the original chip image . the original chip header is modified so that the extension corresponds to an image with no pixels data : is set to 0 , even though and are retained with the original dimensions of the chip . a pixel - less primary header unit ( phu ) is generated with a summary of some of the important and common chip - level keywords ( e.g. , ) . the astrometric transformation information for each chip is saved in the corresponding header using standard ( and some non - standard ) wcs keywords . for the two - level astrometric model , the phu header carries the astrometric transformation related to the projection and the camera - wide distortions . photometric calibrations are written as a set of keywords to individual chip headers , and if the calibration is performed at the exposure - level , to the phu . the photometry calibration keywords are : * : the nominal zero point for this filter * : the measured zero point for this chip / exposure * : the measured error on * : the number of stars used to measure * : minimum reference magnitude included in analysis * : maximum reference magnitude included in analysis the keyword is used to set the initial zero point when the data from the exposure are loaded into the dvo database . data from the gpc1 chip images , the stack images , and the warp images are loaded into dvo using the real - time analysis astrometric calibration to guide the association of detections into objects . after the full pv3 dvo database was constructed , including all of the chip , stack , and warp detections , several external catalogs were merged into the database . first , the complete 2mass psc was loaded into a stand - alone dvo database , which was then merged into the pv3 master database . next the dvo database of synthetic photometry in the ps1 bands ( see section [ sec : synthdb ] ) was merged in . next , the full tycho database was added , followed by the allwise database . after the gaia release in august 2016 @xcite , we generated a dvo database of the gaia positional and photometric information and merged that into the master dvo database . the photometric calibration of the dvo database starts with the `` ubercal '' analysis technique as described by @xcite . this analysis is performed by the group at harvard , loading data from the files into their instance of the large scale database ( lsd , * ? ? ? * ) , a system similar to dvo used to manage the detections and determine the calibrations . photometric nights are selected and all other exposures are ignored . each night is allowed to have a single fitted zero point and a single fitted value for the airmass extinction coefficient per filter . the zero points and extinction terms are determined as a least squares minimization process using the repeated measurements of the same stars from different nights to tie nights together . flat - field corrections are also determined as part of the minimization process . in the original ( pv1 ) ubercal analysis , @xcite determined flat - field corrections for @xmath36 sub - regions of each chip in the camera and four distinct time periods ( `` seasons '' ) . later analysis ( pv2 ) used an @xmath37 grid of flat - field corrections to good effect . the ubercal analysis was re - run for pv3 by the harvard group . for the pv3 analysis , under the pressure of time to complete the analysis , we chose to use only a @xmath36 grid per chip as part of the ubercal fit and to leave higher frequency structures to the later analysis . a 5th flat - field season consisting of nearly the last 2 years of data was also included for pv3 . in retrospect , as we show below , the data from the latter part of the survey would probably benefit from additional flat - field seasons . by excluding non - photometric data and only fitting 2 parameters for each night , the ubercal solution is robust and rigid . it is not subject to unexpected drift or sensitivity of the solution to the vagaries of the data set . the ubercal analysis is also especially aided by the inclusion of multiple medium deep field observations every night , helping to tie down overall variations of the system throughput and acting as internal standard star fields . the resulting photometric system is shown by @xcite to have reliability across the survey region at the level of ( 8.0 , 7.0 , 9.0 , 10.7 , 12.4 ) millimags in ( ) . as we discuss below , this conclusion is reinforced by our external comparison . the overall zero point for each filter is not naturally determined by the ubercal analysis ; an external constraint on the overall photometric system is required for each filter . @xcite used photometry of the md09 medium deep field to match the photometry measured by @xcite on the reference photometric night of mjd 55744 ( ut 02 july 2011 ) . @xcite and @xcite have re - examined the photometry of calspec standards observed by ps1 . @xcite reject 2 of the 7 stars used by @xcite and add photometry of 5 additional stars . @xcite further reject measurements of calspec standards obtained close to the center of the camera field of view where the psf size and shape changes very rapidly . the result of this analysis modifies the over system zero points by 20 - 35 millimags compared with the system determined by @xcite . the ubercal analysis above results in a table of zero points for all exposures considered to be photometric , along with a set of low - resolution flat - field corrections . it is now necessary to use this information to determine zero points for the remaining exposures and to improve the resolution of the flat - field correction . this analysis is done within the ipp dvo database system . the ubercal zero points and the flat - field correction data are loaded into the pv3 dvo database using the program . this program converts the reported zero point and flat field values to the dvo internal representation in which the zero point of each image is split into three main components : @xmath38 where @xmath39 and @xmath40 are static values for each filter representing respectively the nominal zero point and the slope of the trend with respect to the airmass ( @xmath41 ) for each filter . these static values are listed in table [ tab : zpts ] . when was run , these static zero points have been adjusted by the calspec offsets listed in table [ tab : zpts ] based on the analysis of calspec standards by @xcite . these offsets bring the photometric system defined by the ubercal analysis into alignment with @xcite . the value @xmath42 is the offset needed by each exposure to match the ubercal value , or to bring the non - ubercal exposures into agreement with the rest of the exposures , as discussed below . the flat - field information is encoded in a table of flat - field offsets as a function of time , filter , and camera position . each image which is part of the ubercal subset is marked with a bit in the field : when applies the ubercal information to the image tables , it also updates the individual measurements associated with those images . in the dvo database schema , the normalized instrumental magnitude , @xmath43 are stored for each measurement . the value of 25.0 is an arbitrary ( but fixed ) constant offset to place the instrumental magnitudes into approximately the correct range . associated with each measurement are two correction magnitudes : @xmath44 and @xmath45 , along with the airmass for the measurement , calculated using the altitude of the individual detection as determined from the right ascension , declination , the observatory latitude , and the sidereal time . for a camera with the field of view of the ps1 gpc1 , the airmass may vary significantly within the field of view , especially at low elevations . in the worst cases , at the celestial pole , the airmass range within a single exposure is xxx - xxx . the complete calibrated ( ` relative ' ) magnitude is determined from the stored database values as : @xmath46 the calibration offsets , @xmath44 and @xmath45 , represent the per - exposure zero point correction and the slowly - changing flat - field correction respectively . these two values are split so the flat - field corrections may be determined and applied independently from the time - resolved zero point variations . note that the above corrections are applied to each of the types of measurements stored in the database , psf , aperture , kron . the calibration math remains the same regardless of the kind of magnitude being measured . also note that for the moment , this discussion should only be considered as relevant to the chip measurements . below we discuss the implications for the stack and warp measurements . when the ubercal zero points and flat - field data are loaded , updates the @xmath44 values for all measurements which have been derived from the ubercal images . these measurements are also marked in the field with the bit . at this stage , also updates the values of @xmath45 for all gpc1 measurements in the appropriate filters . relative photometry is used to determine the zero points of the exposures which were not included in the ubercal analysis . the relative photometry analysis has been described in the past by @xcite . we review that analysis here , along with specific updates for pv3 . as described above , the instrumental magnitude and the calibrated magnitude are related by arithmetic magnitude offsets which account for effects such as the instrumental variations and atmospheric attenuation . @xmath47 from the collection of measurements , we can generate an average magnitude for a single star ( or other object ) : @xmath48 we find that the color difference of the different chips can be ignored , and set the value of @xmath49 to 0.0 . note that we only use a single mean airmass extinction term for all exposures the difference between the mean and the specific value for a given night is taken up as an additional element of the atmospheric attenuation . we write a global @xmath50 equation which we attempt to minimize by finding the best mean magnitudes for all objects and the best @xmath44 offset for each exposure : @xmath51 + zp + k \zeta + m_{clouds}[i ] - m_{ave}[j ] ) w_{i , j } / \sum_{i , j } w_{i , j}\ ] ] if everything were fitted at once and allowed to float , this system of equations would have @xmath52 unknowns . we solve the system of equations by iteration , solving first for the best set of mean magnitudes in the assumption of zero clouds , then solving for the clouds implied by the differences from these mean magnitudes . even with 1 - 2 magnitudes of extinction , the offsets converge to the milli - magnitude level within 8 iterations . only brighter , high quality measurements are used in the relative photometry analysis of the exposure zero points . we use only the brighter objects , limiting the density to a maximum of 4000 objects per square degree ( lower in areas where we have more observations ) . when limiting the density , we prefer objects which are brighter ( but not saturated ) , and those with the most measurements ( to ensure better coverage over the available images ) . there are a few classes of outliers which we need to be careful to detect and avoid . first , any single measurement may be deviant for a number of reasons ( e.g. , it lands in a bad region of the detector , contamination by a diffraction spike or other optical artifact , etc ) . we attempt to exclude these poor measurements in advance by rejecting measurements which the photometric analysis has flagged the result as suspcious . we reject detections which are excessively masked ; these include detections which are too close to other bright objects , diffraction spikes , ghost images , or the detector edges . however , these rejections do not catch all cases of bad measurements . after the initial iterations , we also perform outlier rejections based on the consistency of the measurements . for each star , we use a two pass outlier clipping process . we first define a robust median and sigma from the inner 50% of the measurements . measurements which are more than 5@xmath33 from this median value are rejected , and the mean & standard deviation ( weighted by the inverse error ) are recalculated . we then reject detections which are more than 3@xmath33 from the recalculated mean . suspicious stars are also exclude from the analsis . we exclude stars with reduced @xmath50 values more than 20.0 , or more than 2@xmath53 the median , whichever is larger . we also exclude stars with standard deviation ( of the measurements used for the mean ) greater than 0.005 mags or 2@xmath53 the median standard deviation , whichever is greater . similarly for images , we exclude those with more than 2 magnitudes of extinction or for which the deviation greater of the zero points per star are than 0.075 mags or 2@xmath53 the median value , whichever is greater . these cuts are somewhat conservative to limit us to only good measurements . the images and stars rejected above are not used to calculate the system of zero points and mean magnitudes . these cuts are updated several times as the iterations proceed . after the iterations have completed , the images which have been reject are calibrated based on their overlaps with other images . we overweight the ubercal measurements in order to tie the relative photometry system to the ubercal zero points . ubercal images and measurements from those images are not allowed to float in the relative photometry analysis . detections from the ubercal images are assigned weights of 10x their default ( inverse - variance ) weight . the calculation of the formal error on the mean magnitudes propagates this additional weight , so that the errors on the ubercal observations dominates where they are present . @xmath54 @xmath55 the calculation of the relative photometry zero points is performed for the entire @xmath0 data set in a single , highly parallelized analysis . as discussed above , the measurement and object data in the dvo database are distributed across a large number of computers in the ipp cluster : for pv3 , 100 parallel hosts are used . these machines by design control data from a large number of unconnected small patches on the sky , with the goal of speeding queries for arbitrary chunks of the sky . as a result , this parallelization is entirely inappropriate as the basis of the relative photometry analysis . for the relative photometry calculation ( and later for relative astrometry calculation ) , the sky is divided into a number of large , contiguous regions each bounded by lines of constant ra & dec , 73 regions in the case of the pv3 analysis . a separate computer , called a `` region host '' is responsible for each of these regions : that computer is responsible for calculating the mean magnitudes of the objects which land within its region and for determining the exposure zero points for exposures for which the center of the exposure lands in the region of responsibility . the iterations described above ( calculate mean magnitudes , calculate zero points , calculate new measurements ) are peformed on each of the 73 region hosts in parallel . however , between certain iteration steps , the region hosts must share some information . after mean object magnitudes are calculated , the region hosts must share the object magnitudes for the objects which are observed by exposures controlled by neighboring region hosts . after image calibrations have been determined by each region host , the image calibrations must be shared with the neighboring region hosts so measurement values associated with objects owned by a neighboring region host may be updated . the completely work flow of the all - sky relative photometry analysis starts with an instance of the program running on a master computer . this machine loads the image database table and assigns the images to the 73 region hosts . a process is then launched on each of the region hosts which is responsible for managing the image calibration analysis on that host . these processes in turn make an intial request of the photometry information ( object and measurement ) from the 100 parallel dvo partition machines . in practice , the processes on the the region hosts are launched in series by the master process to avoid overloading the dvo partition machines with requests for photometry data from all region hosts at once . once all of the photometry has been loaded , the region hosts perform their iterations , sharing the data which they need to share with their neighbors and blocking while they wait for the data they need to receive from their neighbors . the management of this stage is performed by communication between the region host . at the end of the iterations , the regions hosts write out their final image calibrations . the master machine then loads the full set of image calibrations and then applies these calibrations back to all measurements in the database , updating the mean photometry as part of this process . the calculations for this last step are performed in parallel on the dvo partition machines . with the above software , we are able to perform the entire relphot analysis for the full 3@xmath3 region at once , avoiding any possible edge effects . the region host machines have internal memory ranging from 96 gb to 192 gb . regions are drawn , and the maximum allowed density was chosen , to match the memory usage to the memory available on each machine . a total of 9.8 tb of ram was available for the analysis , allowing for up to 6000 objects per square degree in the analysis . consistency of photometry measurements across the sky . each panel shows a map of the standard deviation of photometry residuals for stars in each pixel . the median value of the measure standard deviations across the sky is @xmath56 millimags . these values reflect the typical single - measurement errors for bright stars . ] for pv3 , the relphot analysis was performed two times . the first analysis used only the flat - field corrections determined by the ubercal analysis , with a resolution of 2x2 flat - field values for each gpc1 chip ( corresponding to 2400 pixels ) , and 5 separate flat - field seasons. however , we knew from prior studies that there were significant flat - field structures on smaller scales . we used the data in dvo after the initial relphot calibration to measure the flat - field residual with much finer resolution : 124 x 124 flat - field values for each gpc1 chip ( 40x40 pixels per point ) . we then used to apply this new flat - field correction , as well as the ubercal flat - field corrections , to the data in the database . at this point , we re - ran the entire relphot analysis to determine zero points and to set the average magnitudes . figure [ fig : photflat ] shows the high - resolution photometric flat - field corrections applied to the measurements in the dvo database . these flat - fields make low - level corrections of up to 0.03 magnitudes . several features of interest are apparent in these images . first , at the center of the camera is an important structure caused by the telescope optics which we call the `` tent '' . in this portion of the focal plane , the image quality degrades very quickly . the photometry is systematically biased because the point spread function model can not follow the real changes in the psf shape on these small scales . as is evident in the image , the effect is such that the flux measured using a psf model is systematically low , as expected if the psf model is too small . the square outline surrounding the `` tent '' is due to the 2@xmath532 sampling per chip used for the ubercal flat - field corrections . the imprint of the ubercal flat - field is visible throughout this high - resolution flat - field : in regions where the underlying flat - field structure follows a smooth gradient across a chip , the ubercal flat - field partly corrects the structure , leaving behind a saw - tooth residual . the high - resolution flat - field corrects the residual structures well . especially notable in the bluer filters is a pattern of quarter circles centered on the corners of the chips . these patterns are similar to the `` tree rings '' reported by the des team and others ( g. berstein ref & refs ) . the details of these tree rings are beyond the scope of this article , and will be explored in future work . unlike the tree ring features discussed by these other authors , the features observed in the gpc1 photometry are not caused by an interaction of the flat - field with the effective pixel geometry . instead , these photometric features are due to low - level changes in the psf size which we attribute to variable charge diffusion ( magnier in prep ) . other features include some poorly responding cells ( e.g. , in xy14 ) and effects at the edges of chips , possibly where the psf model fails to follow the changes in the psf . for stacks and warps , the image calibrations were determined after the relative photometry was performed on the individual chips . each stack and each warp was tied via relative photometry to the average magnitudes from the chip photometry . in this case , no flat - field corrections were applied . for the stacks , such a correction would not be possible after the stack has been generated because multiple chip coordinates contribute to each stack pixel coordinate . for the warps , it is in principle possible to map back to the corresponding chip , but the information was not available in the dvo database , and thus it was not possible at this time to determine the flat - field correction appropriate for a given warp . this latter effect is one of several which degrade the warp photometry compared to the chip photometry at the bright end . figure [ fig : allsky.photom.sigma ] shows the standard devitions of the mean residual photometry for bright stars as a function of position across the sky . for each pixel in these images , we selected all objects with ( 14.5 , 14.5 , 14.5 , 14.0 , 13.0 ) @xmath57 ( @xmath58 ) @xmath57 ( 17 , 17 , 17 , 16.5 , 15.5 ) , with at least 3 measurements in @xmath59-band ( to reject artifacts detected in a pair of exposures from the same night ) , with @xmath60 ( to reject excessively - masked objects ) , and with @xmath61 ( to reject galaxies ) . we then generated histograms of the difference between the average magnitude and the apparent magnitude in an individual image for each filter for all stars in a given pixel in the images . from these residual histograms , we can then determine the median and the 68%-ile range to calculate a robust standard deviation . this represents the bright - end systematic error floor for a measurement from a single exposure . the standard deviations are then plotted in figure [ fig : allsky.photom.sigma ] . the 5 panels in figure [ fig : allsky.photom.sigma ] show several features . the galactic bulge is clearly seen in all five filters , with the impact strongest in the reddest bands . we attribute this to the effects of crowding and contamination of the photometry by neighbors . large - scale , roughly square features 10 degrees on a side in these images can be attributed to the vagaries of weather : these patches correspond to the observing chunks . these images include both photometric and non - photometric exposures . it seems plausible that the non - photometric images from relatively poor quality nights elevate the typical errors . on small scales , there are circular patterns 3 degrees in diameter corresponding to individual exposures ; these represent residual flat - fields structures not corrected by our stellar flat - fielding . the median of the standard deviations in the five filters are @xmath62 millimagnitudes . map of the amplitude of the koppenhfer effect on chips across the focal plane . in the affected chips , bright stars are up to 0.2 deviant from their expected positions . * bottom left * x - direction before correction . * bottom right * y - direction before correction . * top left * x - direction after correction . * top right * y - direction after correction . ] once the full pv3 dataset loaded into the master pv3 dvo database , along with supporting databases , and the photometric calibrations were performed , relative astrometry could be performed on the database to improve the overall astrometric calibration . in many respects the relative astrometric analysis is similar to the relative photometric analysis : the repeated measurements of the same object in different images are used to determine a high quality average position for the object . the new average positions are then used to determine improved astrometric calibrations for each of the images . these improved calibrations are used to set the observed coordinates of the measurements from those images , which are in turn used to improve the average positions of the objects . the whole process is repeated for several iterations . like the photometric analysis , the astrometric analysis is performed in a parallel fashion with the same concept that specific machines are responsible for exposures and objects which land within their regions of responsibility , defined on the basis of lines of constant ra and dec . between iteration steps , the astrometric calibrations are shared between the parallel machines as are the improved positions for objects controlled by one machine but detect in images controlled by another machine . like the photometric analysis , the entire sky is processed in one pass . however , there are some important differences in the details . first , the astrometric calibration has a larger number of systematic effects which must be performed . these consist of : 1 ) the koppenhffer effect , 2 ) differential chromatic refraction , 3 ) static deviations in the camera . we discuss each of these in turn below . the koppenhffer effect was first identified in february 2011 by johannes koppenhffer ( mpe ) as part of the effort to search for planet transists in the stellar transit survey data . he noticed that the astromety of bright stars and faint stars disagreed on overlapping chips at the boundary between the sts fields . after some exploration , it was determined that the x coordinate of the brightest stars was offset from the expected location based on the faint stars for a subset of the gpc1 chips . the essence of the effect was that a large charge packet could be drawn prematurely over an intervening negative serial phase into the summing well , and this leakage was proportionately worse for brighter stars . the brighter the star , the more the charge packet was pushed ahead on the serial register . the amplitude of the effect was at most @xmath63 , corresponding to a shift of about one pixel . this effect was only observed in 2-phase ota devices , with 22 / 30 of these suffering from this effect . by adjusting the summing well high voltage down from a default + 7 v to + 5.5v on the 2-phase devices , the effect was prevented in exposures after 2011 - 05 - 03 . however , this left 101,550 exposures ( 27% ) already contaminated by the effect . we measured the koppenhffer effect by accumulating the residual astrometry statistics for stars in the database . for each chip , we measured the mean x and y displacements of the astrometric residuals as function of the instrumental magnitude of the star divided by the fwhm@xmath64 . we measured the trend for all chips in a number of different time ranges and found the effect to be quite stable , in the period where it was present . the effect only appeared in the serial direction . figure [ fig : koppenhoefer ] shows the ke trend for a typical affected chip both before and after the correction . for the pv3 dataset , we re - measured the ke trends using stars in the galactic pole regions after an initial relative astrometry calibration pass : the galactic pole is necessary because the real - time astrometric calibration relies largely on the fainter stars which are not affected by the ke . the trend is then stored in a form which can be applied to the database measurements . differential chromatic refraction ( dcr ) affects astrometry because the reference stars used the calibrate the images are not the same color ( sed ) as the rest of the stars in the image . for a given star of a color different from the reference stars , as exposures are taken at higher airmass , the apparent position of the star will be biased along the parallactic angle . while it is possible to build a model for the dcr impact based on the filter response functions and atmospheric refraction , we have instead elected to use an empirical correction for the dcr present in the pv3 database . we have measured the dcr trend using the astrometric residuals of millions of stars after performing an initial relative astrometry calibration . we define a blue dcr color ( @xmath65 ) to be used when correcting the filters , , , and a red dcr color ( @xmath66 ) to be used when correcting the filters @xmath67 . in the process of performing the relative astrometry calibration , we record the median red and blue colors of the reference stars used to measure the astrometry calibration for each image . as we determine the astrometry parameters for each object in the database , we record the median red and blue reference star colors for all images used to determine the astrometry for a given object . for each star in the database , we know both the color of the star and the typical color of the reference stars used to calibrate the astrometry for that star . we measure the mean deviation of the residuals in the parallactic angle direction and the direction perpendicular to the parallactic angle . for each filter , we determine the dcr trend as a function of the difference between the star color and the reference star color , using the red or blue color approriate to the particular filter , times the tangent of the zenith distance . figure [ fig : dcr ] shows the dcr trend for the 5 filters , as well as the measured displacement in the direction perpendicular to the parallactic angle . we represent the trend with a spline fitted to this dataset . example of the dcr trend in the g - band . * top : * dcr trend in the parallactic direction * bottom : * dcr trend perpendicular to the parallactic angle . ] the amplitude of the dcr trend in the five filters is @xmath68 arcsec airmass@xmath69 magntiude@xmath69 . we saturate the dcr correction if the term @xmath70 for a given measurement is outside a range where the dcr correction is well measured . the maximum dcr correction applied to the five filters is @xmath71 arcseconds . after correction for both ke and dcr , we observe persistent residual astrometric deviations which depend on the position in the camera . we construct an astrometric `` flat - field '' response by determining the mean residual displacement in the x and y ( chip ) directions as a function of position in the focal plane . we have measured the astrometric flat using a sampling resolution of 40x40 pixels , matching the photometric flat - field correction images . figures [ fig : astroflat.gri ] and [ fig : astroflat.zy ] show the astrometric flat - field images for the five filters in each of the two coordinate directions . these plots show several types of features . the dominant pattern in the astrometric residual is roughly a series of concentric rings . the pattern is similar to the pattern of the focal surface residuals measured by ( ref ) , which also has a concentric series of rings with similar spacing . the `` tent '' in the center of the focal surface reflected in these astrometry residual plots . our interpretation of the structure is that the deviations of the focal plane from the ideal focal surface introduces small - scale psf changes , presumably coupled to the optical aberrations , which result in small changes in the centroid of the object relative to the psf model at that location . since the psf model shape parameters are only able to vary at the level of a 6x6 grid per chips , the finer structures are not included in the psf model . the pv2 analysis shows the ring structure more clearly , with a pattern much more closely following the focal surface deviations . in the pv2 analysis , the psf model used at most a 3x3 grid per chip to follow the shape variations , so any changes caused by the optical aberrations would be less well modeled in the pv2 analysis , as we observe . a second pattern which is weakly seen in several chips consists of consistent displacements in the x ( serial ) direction for certain cells . this effect can be seen most clearly in chips xy45 and xy46 . in the pv2 analysis , this pattern is also more clearly seen . in this case , the fact that the astrometric model used polynomials with a maximum of 3rd order per chip means the deviation of individual cells can not be followed by the astrometric model . a third effect is seen at the edge of the chips , where there appears to be a tendency for the residual to follow the chip edge . the origin of this is unclear , but likely caused by the astrometry model failing to follow the underlying variations because of the need to extrapolate to the edge pixels . finally , we also mention an interesting effect _ not _ visible at the resolution of these astrometric flat - field images . fine structures are observed at the 10 pixel scale similar to the `` tree rings '' reported by the des team and others ( g. berstein ref & refs ) . the details of these tree rings are beyond the scope of this article , and will be explored in future work . unfortunately , we discovered a problem with the astrometric flat - field correction too late to be repaired for dr1 . as can be seen by inspection of figures [ fig : astroflat.gri ] and [ fig : astroflat.zy ] , there is significant pixel - to - pixel noise in the the astrometric flat - field images . this pixel - to - pixel noise is caused by too few stars used in the measuremnt of the flat - field structure for the high - resolution sampling . as a result , the astrometric flat - field correction reduces systematic structures on large spatial scales , but at the expense of degrading the quality of an individual measurement . only @xmath59-band has sufficient signal - to - noise per pixel to avoid significantly increasing the per - measurement position errors . figure [ fig : allsky.astrom.sigma ] shows the standard devitions of the mean residual astrometry in @xmath72 for bright stars as a function of position across the sky . for each pixel in these images , we selected all objects with @xmath73 , with at least 3 measurements in @xmath59-band ( to reject artifacts detected in a pair of exposures from the same night ) , with @xmath60 ( to reject excessively - masked objects ) , and with @xmath74 ( to reject galaxies ) . we then generated histograms of the difference between the object position predicted for the epoch of each measurement ( based on the proper motion and parallax fit ) and the observed position of that measurement , in both the right ascension and declination directions ( in linear arcseconds ) , for all stars in a given pixel in the images . from these residual histograms , we can then determine the median and the 68%-ile range to calculate a robust standard deviation . this represents the bright - end systematic error floor for a measurement from a single exposure . the standard deviations are then plotted in figure [ fig : allsky.photom.sigma ] . the median value of the standard deviations across the sky is @xmath75 milliarcseconds . the galactic plane is clearly apparently in these images . like photometry , we attribute this to failure of the psf fitting due to crowding . the celestial north pole regions have somewhat elevated errors in both r.a . and dec . this may be due to the larger typical seeing at these high airmass regions , but without further exploration this is interpretation uncertain . several features can be seen which appear to be an effect of the tie to the gaia astrometry : the stripes near the center of the dec image and the right side of the r.a . the mesh of circular outlines is due to the outer edge of the focal plane where the astrometric calibration is poorly determined . as discussed above , the median values in the images are higher than expected based on our pv2 analysis of the astrometry : the median per - measurement error floor of 22 mas is significantly worse than the 17 mas value in that earlier analysis . we attribute this degradation to the noise introduced by the astrometric flat - field . this noise can likely be addressed before the dr2 release of the individual measurement data . consistency of photometry measurements across the sky . each panel shows a map of the standard deviation of astrometry residuals for stars in each pixel . the median value of the standard deviations across the sky is @xmath75 milliarcseconds . these values reflect the typical single - measurement errors for bright stars . see discussion regarding the astrometric flat which is likely responsible for these elevated value . ] after the initial analysis to measure the ke corrections , dcr corrections , and astrometric flat - field corrections , we applied these corrections to the entire database . within the schema of the database , each measurement has the raw chip coordinates ( ) as well as the offset for that object based on each of these three corrections : . the offsets are calculated for each measurement based on the observed instrumental chip magnitudes and fwhm for the koppenhoffer effect , on the average chip colors and the altitude & azimuth of each measurement for the dcr correction , and on the chip coordinates for the astrometric flat - field corrections . the corrections are combined and applied to the raw chip coordinates and saved back in the database in the fields . at this point , we are ready to run the full astrometric calibration . the initial analysis of the pv2 astrometry used the 2mass positions as an inertial constraint : the 2mass coordiates were included in the calculation of the mean positions for the objects in the database , with weight corresponding to the reported astrometric errors . in this analysis , the object positions used to determine the calibrations of the image parameters ignored proper motion and parallax . after the image calibrations were determined , then individual objects were fitted for proper motion and possibly parallax , as discussed in detail below . using the pv2 analysis of the astrometry calibration , we discovered large - scale systematic trends in the reported proper motions of background quasars . this motion had an amplitude of 10 - 15 milliarcseconds per year and clear trends with galactic longitude . we also observed systematic errors of the mean positions with respect to the icrf milliarcsecond radio quasar positions , with an amplitude of 60 milliarcseconds , again with trends associated with galactic longitude . since the 2mass data were believed to have minimal average deviations relative to the icrf quasars , this latter seemed to be a real effect . we realized that both the proper motion and the mean position biases could be caused by a single common effect : the proper motion of the stars used as reference stars between the 2mass epoch ( 2000 ) and ps1 epoch ( 2012 ) . since we are fitting the image calibrations without fitting for the proper motions of the stars , we are in essencence forcing those stars to have proper motions of 0.0 . the background quasars would then be observed to have proper motions corresponding to the proper motions of the reference stars , but in the opposite direction . we demonstrated that the observed quasar proper motions agreed well with the distribution expected if the median distance to our reference stars was 500 pc . for pv3 , we desired to address this bias by including our knowledge about the distances to the reference stars and the expected typical proper motions for stars at those distances . with some constraint on the distance to each star , we can determine the expected proper motion based on a model of the galactic rotation and solar motions . we can then calculate the mean positions for the objects keeping the assumed proper motion fixed . when calibrating a specific image , the reference star mean position is then translated to the expected position at the epoch of that image . the image calibration is then performed relative to these predicted postions . this process naturally accounts for the proper motion of the reference stars . in order to make the calibrations consistent with the observed coordinates of an external inertial reference , we perform the iterative fits using the technique as described , but assign very high weights in the initial iterations to the inertial reference , and reduce the weights as the astrometric calibration iterations proceed . in order to perform this analysis , we need estimated distances for every reference star used in the analysis . green et al ( ref ) performed sed fitting for 800 m stars in the 3@xmath3 region using pv2 data . the goal of this work was to determine the 3d structure of the dust in the galaxy . by fitting model seds to stars meeting a basic data quality cut , they determined the best spectral type , and thus @xmath76 , absolute @xmath77-band magnitude , distance modulus , and extinction @xmath78 ( the desired output and used to determine the dust extinction as a function of distance throughout the galaxy ) . we use the distance modulus determined in this analysis to predict the proper motions . to convert the distances to proper motions , we use the galactic rotation parameters ( @xmath79 ) = ( 14.82,-12.37 ) km sec@xmath69 pc@xmath69 and solar motion parameters ( @xmath80 ) = ( 9.32 , 11.18 , 7.61 ) km sec@xmath69 as determined by feast & whitelock ( ref ) using hipparchos data . proper motions are determined from the following : @xmath81 where @xmath82 is the distance and @xmath83 are the galactic coordintes of the star . note that the proper motion induced by the galactic rotation is independent of distance while the reflex motion induced by the solar motion decreases with increasing distance . also note that this model assumes a flat rotation curve for objects in the thin disk ; any reference stars which are part of the halo population will have proper motions which are not described by this model ; the mostly random nature of the halo motions should act to increase the noise in the measurement , but should not introduce detectable motion biases . also , if the distance modulus is not well determined , we can assume the object is simply following the galactic rotation curve and set a fixed proper motion . if we do not have a distance modulus from the green et al analysis , we assume a value of 500pc . after the full relative astrometry analysis was performed for the pv3 database , the gaia data release 1 became available . this afforded us the opportunity to constrain the astrometry on the basis of the gaia observations . gaia dr1 objects which are bright enough to have proper motion and parallax solutions are in general saturated in the ps1 observations . thus , we are limited to using the gaia mean positions reported for the fainter stars . we extracted all gaia sources not marked as a duplicate from the gaia archive and generated a dvo database from this dataset . we then merged the gaia dvo into the pv3 master dvo database . we re - ran the complete relative astrometry analysis using gaia as an additional measurement . we applied the analysis described above , applying the estimated distances to determine preliminary proper motions . the gaia mean epoch is reported as 2015.0 , so all gaia measurements were assigned this epoch . we wanted to ensure the gaia measurements dominated the astrometric solutions , so we made the weight very high for the gaia points : 1000@xmath53 the nominal weight in the initial fits ( to lock down the reference frame ) , decreasing to 100@xmath53 the nominal weight for the last fits . we also retained the 2mass measurements in the analysis , but gave them somewhat lower weights than gaia : while the 2mass data does not have the accuracy of gaia , the coverage is known to be quite complete , while the gaia dr1 has clear gaps and holes . having 2mass , even at a lower weight , helps to tile over those gaps . figure [ fig : gaia.photom ] shows a comparison between the pan - starrs photometry in @xmath84 and the gaia photometry in the @xmath85-band . to compare the ps1 photometry to the very broadband gaia g filter , we have determined a transformation based on a 3rd order polynomial fit to @xmath86 and @xmath65 colors . this transformation reproduces gaia photometry reasonably well for stars which are not too red . for a comparison , we have seleted all ps1 stars with gaia measurements meeting the following criteria : @xmath87 , with at least 10 total measurements , within a modest color range @xmath88 . we also restricted to objects with @xmath89 , using the average @xmath59 magnitudes determined from the individual exposures . for figure [ fig : gaia.photom ] , we calculate the difference between the estimated @xmath85-band magnitude based on ps1 @xmath84 photometry and the @xmath85-band photometry reported by gaia . for each pixel , we determine the histogram of these differences and calculate the median and the 68%-ile range . in figure [ fig : gaia.photom ] , these values are plotted as a color scale . the galactic plane is clearly poorly matched between the two photometry systems . this may in part be due to the difficulty of predicting @xmath85-band magnitudes for stars which are significantly extincted : the @xmath85-band includes significant flux from the ps1 @xmath90-band which was not used in our transformation . many other large scale feature in the median differences have structures similar to the gaia scanning pattern ( large arcs and long parallel lines . there are also structures related to the ps1 exposure footprint . these show up as a mottling on the 3 degree scale ( e.g. , lower right below the galactic plane ) . the amplitude of the residual structures is fairly modest . the standard devition of the median difference values is 7 millimagnitudes . this number gives an indication of the overall photometric consistency of both gaia and ps1 and implies that the systematic error floor for each survey is less than 7 millimags . figure [ fig : gaia.astrom ] shows a comparison between the pan - starrs mean astrometry positions in @xmath91 and the gaia astrometry . for this comparison , we have seleted all ps1 stars with gaia measurements with @xmath87 and with at least 10 total measurements . for figure [ fig : gaia.astrom ] , we calculate the difference between the position predicted by ps1 at the gaia epoch ( using the proper motion and parallax fit ) and the position reported by gaia . for each pixel , we determine the histogram of these differences in the r.a . and dec directions , and calculate the median and the 68%-ile range . in figure [ fig : gaia.astrom ] , these values are plotted as a color scale . there is good consistency between the ps1 and gaia astrometry . there are patterns from the galactic plane ( though not very strongly at the bulge ) . there are also clear features due to the ps1 exposure footprint ( ring structure on 3 degree scales ) . in the plots of the scatter , there are patterns which are related to the gaia scanning rule . these are presumably regions with relatively low signal to noise in gaia ; they were also apparent in the plots of the statisics of the per - exposure measurement residuals ( figure [ fig : allsky.astrom.sigma ] . the standard deviations of the median differences are ( @xmath92 milliarcseconds . the pan - starrs1 surveys ( ps1 ) have been made possible through contributions of the institute for astronomy , the university of hawaii , the pan - starrs project office , the max - planck society and its participating institutes , the max planck institute for astronomy , heidelberg and the max planck institute for extraterrestrial physics , garching , the johns hopkins university , durham university , the university of edinburgh , queen s university belfast , the harvard - smithsonian center for astrophysics , the las cumbres observatory global telescope network incorporated , the national central university of taiwan , the space telescope science institute , the national aeronautics and space administration under grant no . nnx08ar22 g issued through the planetary science division of the nasa science mission directorate , the national science foundation under grant no . ast-1238877 , the university of maryland , and eotvos lorand university ( elte ) and the los alamos national laboratory .

it has been previously proposed that some distant open clusters in the milky way may have been accreted during a dwarf galaxy merger , perhaps associated with the same event that led to the formation of the galactic anticenter stellar structure ( gass ) , also known as the `` monoceros ring '' . we have obtained and washington photometric and medium resolution ( ) multi - fiber spectroscopic data for the three distant old open clusters berkeley 29 , saurer 1 , and vdb - hagen 176 ( bh 176 ) . these clusters are spatially coincident with gass , but radial velocities and spectroscopic metallicities had not been available during previous studies of the gass candidate cluster system . similar data for the clusters berkeley 20 and berkeley 39 have been obtained for calibration purposes . we provide the first _ reliable _ radial velocity for bh 176 ( km s ) . we also find that and km s , for saurer 1(a ) and berkeley 29 , respectively . we show that-enhanced isochrones , while spectroscopically motivated , provide a poor fit to be29 in contrast to previous findings . we find that the clusters berkeley 29 and saurer 1 are consistent with the previously reported characteristics for gass candidate clusters and the gass stellar stream as derived from m - giant observations . however , the radial velocity and photometric metallicity ( [ fe / h ] dex ) for bh 176 suggests that a connection of this cluster with the putative gass cluster system is unlikely . we reassess the age - metallicity relation for the most likely members of the gass clusters system for which spectroscopic metallicities are now available .

astro-ph0509742

lccccccc vdb - hagen 176 ( bh176 ) & 15:39:05.4 & @xmath950:03:01.7 & 328.4100 & @xmath104.3418 & 2.3 & 0.70 & 1 + berkeley 20 ( be20 ) & 05:32:37.0 & @xmath1000:11:30.0 & 203.4803 & @xmath917.3711 & 2.0 & 0.16 & 2 + berkeley 29 ( be29 ) & 06:53:04.2 & @xmath1016:55:39.0 & 197.9493 & @xmath107.9802 & 2.0 & 0.10 & 3 + berkeley 39 ( be39 ) & 07:46:51.0 & @xmath904:40:30.0 & 223.5465 & @xmath1010.0915 & 7.0 & 0.11 & 4 + saurer 1 ( sa1 ) & 07:20:56.0 & @xmath1001:48:29.0 & 214.6894 & @xmath107.3862 & 1.3 & 0.18 & 5 ( * ? ? ? * hereafter f04 ) noted that the outermost open clusters in the milky way seem to lie in a string - like configuration that can be fit to an orbital plane coincident with the galactic anticenter stellar structure ( gass ) . the presence of this arc - like structure was inferred from excesses of various types of stars @xcite beyond the apparent limit of the galactic disk , and has also been used to argue for the presence of a distinct , extended stellar structure wrapping around the disk at low latitudes . the feature has been referred to as the `` monoceros ring '' @xcite or gass @xcite . however , because of its unfortunate location behind considerable extinction , the system s true shape , orientation , extent , etc . have been difficult to ascertain . even the location of the structure s center ( presumably corresponding to a `` nucleus '' ) remains uncertain and controversial . for example , this stellar stream has been argued to be associated with the postulated canis major @xcite or argo dwarf galaxies @xcite . previous work @xcite on this stellar arc has determined some characteristics including : ( 1 ) a velocity - longitude trend that indicates a slightly non - circular orbit , ( 2 ) a velocity dispersion that is smaller than even that of disk stars , and ( 3 ) a wide metallicity spread from [ fe / h]@xmath11 dex ( y03 ) to at least [ fe / h]@xmath12 dex . f04 found that at least five globular clusters and one open cluster ( tombaugh 2 ) have positions and radial velocities ( rvs ) suggesting an association with gass ( see also * ? ? ? * ; * ? ? ? * ) , but pointed to other clusters with spatial coincidence with gass that still have unknown rvs . as a result , we sought to derive the kinematics and chemistry for these other clusters . in this work , the clusters berkeley 29 ( be29 ) , saurer 1 ( sa1 ) , and vdb - hagen 176 ( bh176 ) from the f04 list of postulated gass clusters are investigated . table 1 gives the positional data for these clusters , as well as for old , outer open clusters berkeley 20 ( be20 ) and berkeley 39 ( be39 ) , which we also observed as metallicity calibrators . medium resolution spectra for all five clusters are used to determine their bulk rvs and metallicities via measurement of the infrared triplet lines . these data are used to test whether these clusters are consistent with the observed dynamical and chemical trends of the gass system . while @xcite and @xcite suggest the metallicity and rv data are insufficient to prove membership in gass , they are sufficient to identify clusters with properties inconsistent with those observed for gass . lcccccc bh176 & lco 2.5-m & 0.259 & 1998 jun 30 & & + be29 & fmo 1.0-m & 0.367 & 2003 nov 21 & @xmath13 & @xmath13 + sa1 & fmo 1.0-m & 0.367 & 2004 nov 21 & @xmath13 & @xmath14 lccccccc be20 & lco 1.0-m & 0.697 & 2000 dec 05 & @xmath15 & @xmath15 & @xmath16 + be39 & lco 1.0-m & 0.697 & 2000 dec 05 & @xmath15 & @xmath15 & @xmath16 + sa1 & fmo 1.0-m & 0.367 & 2004 feb 28 & @xmath13 & @xmath14 & @xmath17 in 2 we analyze photometry for the clusters be20 , be29 , be39 , sa1 , and bh176 . this photometry is used to select targets for spectroscopy . in 3 , we discuss the spectroscopic observations and use rvs derived from the spectra to determine the clusters bulk rvs . the derived mean cluster rvs are used to clarify the positions of the giant branches of the clusters in the color - magnitude diagram and compare our work to previous studies in 4 . finally in 5 , we explore each cluster s likelihood of belonging to gass and discuss the implications these clusters have for gass and open cluster studies in the galaxy in general . the clusters be29 and sa1 were imaged in the cousins @xmath18 and @xmath19 filters with the university of virginia s 40 inch telescope at fan mountain observatory ( fmo ) . we also make use of @xmath0 photometry of bh176 , taken with the lco dupont 2.5 m ; these data are described more fully in @xcite . the details of the @xmath0 observations are listed in table 2 . the fan mountain site ccd has 2048 @xmath20 2048 pixels that are 24 @xmath21 m square , resulting in a pixel scale of 0.367 arcsec pixel@xmath4 and a field of view of 12.67 arcmin on a side . be20 , be 39 and sa1 were also observed using washington ( @xmath22 , @xmath23)@xmath1 filters and @xmath19 filters used for this study were identical ; hereafter we will refer to these filters as @xmath19 . ] with the fmo 40 inch and with the 40 inch swope telescope at the las campanas observatory ( lco ; see table 3 ) . the swope ccd has 2048 @xmath20 2048 pixels that are 24 @xmath21 m square , yielding a pixel scale of 0.697 arcsec pixel@xmath4 , for a field of view of 23.7 arcmin on a side . all exposures were obtained under photometric conditions . during each run , standard stars ( @xmath0 from @xcite , washington@xmath1 from @xcite ) were observed at various airmasses between airmasses 1 and 2 for photometric calibration of the data ( see 2.2 below ) . the imaging data have been reduced with the @xmath24 ccdred package , following standard techniques as described in the @xmath24 ccdproc documentation . instrumental magnitudes were measured with the daophot and allstar packages , using the point - spread function method @xcite . these magnitudes were transformed to the @xcite standard - star system ( for @xmath18 and @xmath19 observations ) and to the @xcite standard system ( for data in the washington@xmath1 system ) . the data for bh176 are taken from @xcite and the errors are presented there . for the clusters be20 and be39 , observed only with washington@xmath1 photometry , transformation equations from @xcite were used to estimate @xmath18 and @xmath25 . figures 13 show the @xmath22 , @xmath26 color - magnitude diagram ( cmd ) and @xmath27 , @xmath26 color - magnitude diagram ( cmd ) for the clusters be20 , be39 and sa1 . washington @xmath22 , @xmath19 , and @xmath28 photometry can be used to discriminate between late - type dwarf and giant stars @xcite by calibrated @xmath29 color - color diagrams . however to apply this technique , an estimate of the foreground extinction is necessary to correct the photometry to standard magnitudes . literature values ( presented in table 1 ) have been adopted to implement this color - color based separation technique . we transform @xmath30 into @xmath31 using the relation derived by @xcite . finally , @xcite provides relations to obtain @xmath32 , @xmath33 and @xmath34 based on @xmath31 and the ratio @xmath35 for the selective to total absorption . again , figures 13 show the derived @xmath36 cmds with their respective @xmath29 color - color diagrams for the clusters be20 , be39 , and sa1 . the washington color - color `` giant region '' is selected to separate giant stars from dwarf stars based on their surface gravity , as described in @xcite . for this project , we define the following boundaries of the `` giant region '' for dereddened photometry : @xmath37 stars that fall within the `` giant box '' , are within @xmath38 of the cluster center ( @xmath39 for the larger be39 ) , and are brighter than the cluster main sequence turn off ( msto ) are denoted by stars ( @xmath40 ) . lcccc hd 18884 & 014 & 39.82 & -26.50 & 1.28 + ( m1.5iiia ) & 025 & 39.04 & -26.27 & 1.47 + & 028 & 30.71 & -27.14 & 1.19 + & 045 & 35.45 & -26.38 & 1.38 + & 074 & 37.98 & -26.96 & 1.44 + & 086 & 26.87 & -26.29 & 1.51 + & 099 & 41.07 & -26.33 & 1.44 + & 103 & 34.91 & -26.23 & 1.42 + & 118 & 33.42 & -26.15 & 1.20 + & & average & -26.47 & 0.34 + & & iau value & -25.8 & 0.1 + & & difference & -0.7 & + hd150798 & 008 & 46.36 & -3.41 & 1.05 + ( k2ii - iii ) & 019 & 47.99 & -3.04 & 0.96 + & 033 & 35.22 & -2.77 & 1.25 + & 074 & 40.52 & -5.10 & 1.63 + & 091 & 40.48 & -3.21 & 0.91 + & 099 & 42.19 & -3.10 & 1.13 + & 127 & 34.52 & -3.08 & 1.23 + & & average & -3.39 & 0.78 + & & iau value & -3.7 & 0.2 + & & difference & + 0.3 & + hd157457 & 020 & 43.81 & 17.78 & 0.81 + ( g8iii ) & 034 & 34.79 & 17.05 & 0.66 + & 044 & 35.51 & 16.82 & 0.73 + & 045 & 34.56 & 17.50 & 1.37 + & 116 & 34.56 & 17.88 & 0.61 + & 125 & 30.61 & 18.43 & 1.17 + & 135 & 11.41 & 22.46 & 2.09 + & & average & 17.48 & 0.32 + & & iau value & 17.4 & 0.2 + & & difference & + 0.1 & + hd161096 & 030 & 42.29 & -12.29 & 1.04 + ( k2iii ) & 038 & 35.49 & -12.29 & 0.98 + & 039 & 37.83 & -12.45 & 1.07 + & 040 & 38.85 & -11.56 & 0.97 + & 048 & 31.59 & -12.45 & 1.00 + & 057 & 28.02 & -12.41 & 0.75 + & 077 & 41.44 & -12.21 & 1.12 + & 098 & 43.22 & -12.94 & 0.88 + & 104 & 42.25 & -12.69 & 0.76 + & 110 & 39.67 & -12.03 & 1.05 + & & average & -12.33 & 0.37 + & & iau value & -12.0 & 0.1 + & & difference & -0.3 & spectroscopic data for cluster stars were collected on ut 2004 march 45 using the hydra multi - fiber spectrograph on the 4-meter blanco telescope at cerro tololo intra - american observatory . the spectra cover @xmath41@xmath42 with an instrumental spectral resolution of @xmath43 , or 1.2 per resolution element . to provide rv calibration , four rv standards were also observed , where each observation " of an rv standard entails sending the light down 7 - 10 different fibers , yielding many dozens of individual spectra of rv standards ( see table 4 ) . for wavelength calibration , the `` penray '' ( henearxe ) comparison lamp was observed for every fiber setup . to get spectra of as many members as possible in these compact star systems ( diam . @xmath44 , see table 1 ) , we selected as primary targets cluster red giant branch ( rgb ) candidates from the cluster cmd . for clusters with washington @xmath22 , @xmath19 , and @xmath28 photometry , these rgb candidates were selected based on the cmd and their location in the @xmath29 color - color diagram , as described in 2.1 ( i.e. , the stars marked with star symbols [ @xmath40 ] in figures 13 ) . these stars are the primary targets for spectroscopic observation , with stars within 2 @xmath46 of the cluster center ( @xmath47 ) having the highest priority for observation . since the hydra spectrograph has `` fiber packing '' limitations ( i.e. , there is a minimum finer spacing of 2.5 ) , some fibers were placed on `` extra '' red giant branch candidates ( ergb ) selected from @xmath48 ( see table 5 ) , and to fill yet more fibers we also observed many non - rgb - selected stars in the surrounding field in a search for additional cluster members . also , red clump ( rc ) and horizontal branch ( hb ) candidates were observed in be20 and bh176 . spectroscopic targets for be 29 and bh176 ( clusters for which we were unable to obtain washington photometry ) were selected using the cmd of the inner @xmath38 of the cluster , with the rc stars for bh176 having the highest priority . we also `` filled '' fibers for these clusters in the manner described above . the small numbers of stars observed in some fields results from a combination of poor observing conditions , which yielded low @xmath49 in the observations for some stars , as well as the limitation of the number of hydra fibers that can be packed in a small area . preliminary processing of the two - dimensional images of the fiber spectra was undertaken using standard @xmath24 techniques as described in the @xmath24 ccdred documentation . after completing the basic processing , the spectra were reduced using standard routines as listed in the dohydra documentation . lcccc bh176 & @xmath50 & 34/10 & 19/7 & 2004 mar . 4 + be20 & @xmath51 & 38/12 & 20/4 & 2004 mar . 5 + be29 & @xmath51 & 41/13 & 25/9 & 2004 mar . 4 + be39 & @xmath50 & 68/39 & 54/36 & 2004 mar . 5 + sa1 & @xmath51 & 24/12 & 11/4 & 2004 mar . 4 + all radial velocities ( @xmath52 ) were derived using iraf s fxcor package , which was first used to determine rvs for the standard stars . the rvs of standard stars were measured by cross - correlating each fourier - filtered standard star spectrum against every other . the resulting rvs for each individual spectra were averaged and the standard deviation measured ; the results are presented in table 4 . as may be seen , the derived rvs for the standard stars are all within 1 km s@xmath4 of the iau values . we determine our rv errors from the prescription described in @xcite , which is based on analysis of repeated standard star spectra ( table 4 ) . the tonry davis ratio ( * ? ? ? * tdr ) for each spectrum , measured from fxcor , scales approximately with @xmath49 , which allows us to determine the radial velocity error using the following equation : @xmath53 where @xmath7 is a constant calibrated by the standard star data . average velocity uncertainties for the individual hydra standard spectrum are @xmath54 km s@xmath4 for standard star spectra with @xmath55 , which is @xmath56 . the cumulative @xmath57 statistic for the repeat standard spectrum rv measures is @xmath58 , where @xmath59 refers to the mean velocity and @xmath60 the velocity of the standard star corresponding to the @xmath61th observation . we determine @xmath7 by finding the value of @xmath57 where the probability of exceeding it by chance is 50% ( defined as @xmath62 ) , which approximates 1@xmath63 errors . the standard star spectra ( table 4 ) provide a data set with 29 degrees of freedom . from standard tables , we find that @xmath64 for 29 degrees of freedom , which yields @xmath65 . analysis of the cluster star spectra were also measured with fxcor , and errors were determined from the measured tdr using the @xmath7 derived above . the photometric data and heliocentric rv ( @xmath52 ) and error ( error @xmath52 ) measurements for cluster target stars with a tdr greater than 5 ( @xmath66 ) are presented in tables 610 , along with information on the target selection described below in 3.3 . the measured cluster star rvs are analyzed to determine membership in each cluster . lccccccccc 140 & 15:38:46.54 & @xmath950:02:00.9 & 18.95 & 1.840 & 10.5 & 8.4 & 2.2 & fill & y + 225 & 15:38:46.92 & @xmath950:03:11.0 & 19.81 & 1.734 & 6.5 & @xmath936.5 & 3.4 & fill & n + 1793 & 15:38:52.34 & @xmath950:06:15.6 & 18.08 & 2.126 & 11.1 & @xmath91.5 & 2.1 & fill & n + 3313 & 15:38:57.42 & @xmath950:03:01.9 & 18.99 & 1.847 & 6.5 & 28.4 & 3.4 & rc & n + 3332 & 15:38:57.51 & @xmath950:02:24.5 & 18.90 & 0.977 & 17.1 & @xmath940.0 & 1.4 & hb & n + 3994 & 15:38:59.47 & @xmath950:03:50.8 & 18.80 & 1.862 & 6.5 & 12.6 & 3.4 & rc & y + 4190 & 15:39:00.03 & @xmath950:04:16.3 & 17.83 & 2.234 & 24.4 & 6.4 & 1.0 & rgb & y + 4245 & 15:39:00.24 & @xmath950:01:33.1 & 19.16 & 0.904 & 24.4 & 15.7 & 1.0 & hb & y + 4360 & 15:39:00.57 & @xmath950:02:17.0 & 18.94 & 1.850 & 8.4 & 9.9 & 2.7 & rc & y + 5278 & 15:39:02.96 & @xmath950:00:14.5 & 18.91 & 1.831 & 5.7 & 30.4 & 3.8 & ergb & n + 5988 & 15:39:04.69 & @xmath950:03:17.1 & 19.49 & 0.998 & 5.7 & 2.2 & 3.8 & hb & y + 6194 & 15:39:05.14 & @xmath950:05:05.6 & 19.25 & 0.943 & 14.9 & @xmath977.5 & 1.6 & hb & n + 8251 & 15:39:10.10 & @xmath950:02:24.0 & 18.84 & 1.793 & 9.2 & @xmath90.5 & 2.5 & rc & n + 8468 & 15:39:10.47 & @xmath950:06:33.5 & 18.96 & 1.874 & 6.1 & @xmath96.8 & 3.6 & ergb & n + 9312 & 15:39:12.70 & @xmath950:01:59.9 & 18.05 & 2.243 & 20.2 & 11.4 & 1.2 & rgb & y + 9327 & 15:39:12.72 & @xmath950:03:49.6 & 18.99 & 1.841 & 6.9 & 16.9 & 3.2 & rc & y + 10823 & 15:39:16.77 & @xmath950:02:20.4 & 18.93 & 1.901 & 7.8 & 19.0 & 2.9 & ergb & y + 14629 & 15:39:29.82 & @xmath950:00:41.2 & 18.33 & 1.906 & 20.2 & @xmath980.8 & 1.2 & fill & n + 15302 & 15:39:32.18 & @xmath949:59:51.1 & 17.53 & 2.319 & 10.5 & 34.4 & 2.2 & fill & n lcccccccccc 10308 & 5:32:13.32 & 0:10:19.1 & 17.53 & 1.213 & @xmath100.063 & 6.9 & 1.5 & 3.2 & fill & n + 10617 & 5:32:33.19 & 0:09:35.0 & 17.32 & 1.466 & @xmath90.061 & 9.2 & 0.1 & 2.5 & ergb & n + 10645 & 5:32:34.72 & 0:15:38.9 & 17.33 & 1.385 & @xmath90.048 & 10.0 & 24.4 & 2.3 & fill & n + 10741 & 5:32:36.99 & 0:12:00.8 & 17.69 & 1.456 & @xmath90.015 & 6.5 & 74.5 & 3.4 & rgb & y + 10745 & 5:32:37.66 & 0:17:04.8 & 17.95 & 1.259 & @xmath90.007 & 5.7 & 36.5 & 3.8 & fill & n + 10770 & 5:32:37.93 & 0:11:09.8 & 15.12 & 1.909 & @xmath90.064 & 35.3 & 75.5 & 0.7 & rgb & y + 10794 & 5:32:39.16 & 0:17:28.7 & 17.74 & 1.324 & @xmath90.009 & 7.8 & 77.9 & 2.9 & fill & y + 10806 & 5:32:38.98 & 0:09:27.5 & 17.08 & 0.943 & @xmath100.032 & 8.1 & 35.3 & 2.8 & hb & n + 10810 & 5:32:39.20 & 0:10:31.3 & 16.00 & 1.503 & @xmath100.006 & 20.2 & 73.7 & 1.2 & rgb & y + 10850 & 5:32:41.03 & 0:06:19.4 & 18.49 & 0.947 & @xmath100.026 & 5.7 & 17.7 & 3.8 & fill & n + 10851 & 5:32:41.53 & 0:10:02.9 & 16.25 & 1.522 & @xmath90.002 & 15.9 & 79.3 & 1.5 & rgb & y + 10864 & 5:32:42.31 & 0:12:29.7 & 17.01 & 1.050 & @xmath100.032 & 5.7 & 30.8 & 3.8 & hb & n + 10940 & 5:32:46.55 & 0:08:42.2 & 17.90 & 1.405 & @xmath90.006 & 6.3 & 40.2 & 3.5 & ergb & n + 10979 & 5:32:49.99 & 0:11:54.4 & 16.80 & 1.154 & @xmath100.013 & 13.9 & 32.5 & 1.7 & hb & n + 10980 & 5:32:50.43 & 0:16:12.7 & 15.46 & 1.736 & @xmath90.027 & 30.8 & 53.8 & 0.8 & fill & n + 11054 & 5:32:54.76 & 0:17:30.6 & 15.48 & 1.723 & @xmath90.254 & 30.8 & 52.6 & 0.8 & fill & n + 11058 & 5:32:54.62 & 0:10:49.5 & 17.62 & 1.519 & @xmath90.053 & 10.0 & 56.6 & 2.3 & ergb & n + 11094 & 5:32:57.52 & 0:15:22.7 & 17.09 & 1.428 & @xmath90.019 & 7.2 & 31.2 & 3.1 & fill & n + 11158 & 5:33:01.22 & 0:09:07.3 & 18.19 & 1.295 & @xmath90.023 & 6.1 & @xmath924.3 & 3.6 & fill & n + 11285 & 5:33:10.66 & 0:07:53.7 & 17.92 & 1.430 & @xmath90.057 & 7.5 & 56.5 & 3.0 & fill & n lccccccccc 201 & 6:52:53.40 & 16:49:12.5 & 15.73 & 1.001 & 15.9 & 39.2 & 1.5 & fill & n + 208 & 6:53:03.07 & 16:49:17.7 & 17.30 & 1.140 & 9.2 & 19.1 & 2.5 & fill & n + 246 & 6:52:50.76 & 16:49:39.2 & 17.28 & 0.991 & 7.8 & 121.2 & 2.9 & fill & n + 249 & 6:53:14.68 & 16:49:40.7 & 17.47 & 1.075 & 8.1 & @xmath90.5 & 2.8 & fill & n + 462 & 6:53:14.92 & 16:52:01.3 & 17.23 & 0.952 & 11.1 & 61.6 & 2.1 & fill & n + 482 & 6:52:50.68 & 16:52:17.5 & 16.41 & 0.960 & 13.9 & @xmath927.7 & 1.7 & fill & n + 523 & 6:53:25.94 & 16:52:38.9 & 17.67 & 1.021 & 6.1 & 49.2 & 3.6 & fill & n + 549 & 6:53:33.24 & 16:52:57.4 & 16.94 & 0.946 & 10.0 & 28.5 & 2.3 & fill & y ? + 725 & 6:53:32.80 & 16:54:26.1 & 16.43 & 1.236 & 22.1 & 31.4 & 1.1 & fill & y + 822 & 6:53:01.47 & 16:55:01.9 & 16.30 & 1.121 & 20.2 & 27.4 & 1.2 & rgb & y + 868 & 6:53:03.87 & 16:55:15.9 & 14.66 & 1.588 & 35.3 & 27.5 & 0.7 & rgb & y + 870 & 6:53:23.49 & 16:55:16.8 & 17.21 & 1.123 & 11.1 & 67.2 & 2.1 & fill & n + 934 & 6:53:02.20 & 16:55:35.3 & 15.98 & 1.077 & 20.2 & @xmath911.2 & 1.2 & rgb & n + 948 & 6:53:08.06 & 16:55:40.7 & 16.68 & 1.058 & 9.2 & 29.1 & 2.5 & rgb & y + 996 & 6:53:06.50 & 16:55:53.1 & 17.49 & 1.066 & 7.5 & 32.5 & 3.0 & rgb & y + 1000 & 6:53:04.37 & 16:55:54.3 & 14.37 & 1.781 & 30.8 & 26.4 & 0.8 & rgb & y + 1078 & 6:52:59.74 & 16:56:17.5 & 18.00 & 1.014 & 7.5 & 20.8 & 3.0 & rgb & y + 1165 & 6:53:01.67 & 16:56:45.3 & 15.97 & 1.078 & 17.1 & 53.9 & 1.4 & rgb & n + 1294 & 6:53:31.24 & 16:57:51.2 & 16.57 & 1.056 & 17.1 & 8.4 & 1.4 & fill & n + 1295 & 6:53:21.37 & 16:57:52.3 & 15.81 & 0.990 & 17.1 & 47.6 & 1.4 & fill & n + 1324 & 6:53:26.45 & 16:58:07.1 & 16.30 & 1.095 & 15.9 & 4.2 & 1.5 & fill & n + 1426 & 6:52:49.66 & 16:58:57.0 & 15.93 & 1.182 & 30.8 & 25.8 & 0.8 & fill & y + 1433 & 6:53:25.24 & 16:59:00.6 & 17.93 & 0.860 & 6.7 & @xmath917.7 & 3.3 & fill & n + 1437 & 6:53:12.25 & 16:59:01.3 & 16.77 & 1.126 & 18.5 & 24.7 & 1.3 & fill & y + 1461 & 6:53:05.25 & 16:59:13.2 & 15.31 & 1.093 & 35.3 & 33.2 & 0.7 & fill & y lccccccccccc 20289 & 7:46:12.10 & @xmath94:39:05.3 & 15.26 & 1.549 & @xmath90.024 & 27.2 & 54.7 & 0.9 & fill & y + 20332 & 7:46:13.37 & @xmath94:41:52.3 & 17.35 & 0.988 & @xmath100.024 & 7.2 & 34.7 & 3.1 & fill & n + 20362 & 7:46:14.97 & @xmath94:37:32.2 & 17.32 & 1.071 & @xmath100.044 & 5.9 & 68.0 & 3.7 & fill & n + 20429 & 7:46:16.55 & @xmath94:44:59.1 & 17.29 & 1.098 & @xmath100.037 & 8.1 & 79.1 & 2.8 & fill & n + 20552 & 7:46:21.67 & @xmath94:35:06.1 & 17.13 & 1.001 & @xmath100.034 & 5.5 & 55.2 & 3.9 & fill & y + 20685 & 7:46:25.30 & @xmath94:40:33.9 & 17.17 & 1.305 & @xmath90.032 & 11.1 & 58.8 & 2.1 & fill & y + 20729 & 7:46:27.25 & @xmath94:32:06.5 & 17.28 & 1.056 & @xmath100.031 & 6.1 & 34.4 & 3.6 & rgb & n + 20797 & 7:46:28.59 & @xmath94:38:35.4 & 17.56 & 0.947 & @xmath100.047 & 5.5 & 37.1 & 3.9 & fill & n + 20814 & 7:46:28.29 & @xmath94:46:47.8 & 17.73 & 1.081 & @xmath100.042 & 5.5 & 43.8 & 3.9 & rgb & n + 20899 & 7:46:30.39 & @xmath94:48:49.6 & 16.49 & 1.408 & @xmath100.019 & 11.1 & -0.9 & 2.1 & fill & n + 20904 & 7:46:32.17 & @xmath94:33:17.0 & 17.21 & 1.121 & @xmath90.005 & 6.1 & -3.9 & 3.6 & fill & n + 21054 & 7:46:34.82 & @xmath94:41:14.3 & 16.81 & 1.407 & @xmath90.039 & 14.9 & 53.5 & 1.6 & rgb & y + 21087 & 7:46:35.61 & @xmath94:42:09.7 & 17.31 & 1.081 & @xmath100.028 & 6.1 & 61.3 & 3.6 & rgb & n + 21091 & 7:46:35.88 & @xmath94:40:38.9 & 16.90 & 1.057 & @xmath100.016 & 11.7 & 53.4 & 2.0 & rgb & y + 21152 & 7:46:37.18 & @xmath94:40:11.4 & 14.33 & 1.626 & @xmath90.022 & 35.3 & 54.8 & 0.7 & rgb & y + 21206 & 7:46:38.58 & @xmath94:39:21.6 & 16.49 & 1.421 & @xmath100.007 & 18.5 & 54.0 & 1.3 & rgb & y + 21247 & 7:46:38.99 & @xmath94:41:56.1 & 15.09 & 1.561 & @xmath90.037 & 30.8 & 51.5 & 0.8 & rgb & y + 21305 & 7:46:40.53 & @xmath94:38:17.0 & 16.62 & 1.373 & @xmath100.029 & 13.9 & 97.7 & 1.7 & rgb & n + 21322 & 7:46:39.98 & @xmath94:45:51.3 & 17.29 & 1.235 & @xmath100.034 & 7.5 & 67.7 & 3.0 & fill & n + 21365 & 7:46:41.27 & @xmath94:40:56.9 & 14.65 & 1.620 & @xmath90.023 & 22.1 & 54.8 & 1.1 & rgb & y + 21366 & 7:46:41.45 & @xmath94:39:08.3 & 14.66 & 1.688 & @xmath90.017 & 35.3 & 50.9 & 0.7 & rgb & y + 21426 & 7:46:42.49 & @xmath94:41:28.1 & 16.95 & 1.054 & @xmath100.030 & 7.2 & 59.4 & 3.1 & rgb & y + 21432 & 7:46:43.30 & @xmath94:34:28.7 & 15.89 & 1.531 & @xmath90.138 & 30.8 & 16.0 & 0.8 & fill & n + 21471 & 7:46:43.90 & @xmath94:38:13.9 & 16.98 & 0.938 & @xmath100.035 & 9.2 & 43.6 & 2.5 & rgb & n + 21523 & 7:46:44.57 & @xmath94:40:21.9 & 16.73 & 1.405 & @xmath90.014 & 22.1 & 45.8 & 1.1 & rgb & n + 21530 & 7:46:44.52 & @xmath94:41:33.3 & 17.00 & 1.208 & @xmath100.024 & 5.3 & 55.5 & 4.0 & rgb & y + 21574 & 7:46:45.05 & @xmath94:42:55.0 & 17.05 & 1.436 & @xmath90.007 & 12.4 & 50.1 & 1.9 & rgb & y + 21589 & 7:46:45.18 & @xmath94:44:30.2 & 17.62 & 1.091 & @xmath100.015 & 8.1 & 45.3 & 2.8 & rgb & n + 21606 & 7:46:46.04 & @xmath94:38:58.4 & 16.57 & 1.423 & @xmath90.029 & 27.2 & 54.4 & 0.9 & rgb & y + 21689 & 7:46:47.35 & @xmath94:37:08.8 & 17.14 & 0.997 & @xmath90.016 & 5.2 & 42.6 & 4.1 & rgb & n + 21695 & 7:46:47.99 & @xmath94:32:35.3 & 17.72 & 1.041 & @xmath100.033 & 6.1 & -15.4 & 3.6 & fill & n + 21769 & 7:46:48.56 & @xmath94:39:09.1 & 16.96 & 1.401 & @xmath90.019 & 10.5 & 50.5 & 2.2 & rgb & y + 21792 & 7:46:48.43 & @xmath94:44:46.8 & 16.32 & 1.450 & @xmath100.002 & 15.9 & 37.6 & 1.5 & rgb & n + 21815 & 7:46:49.13 & @xmath94:41:55.5 & 15.65 & 1.540 & @xmath90.029 & 24.4 & 55.8 & 1.0 & rgb & y + 21912 & 7:46:50.77 & @xmath94:41:28.7 & 14.43 & 1.584 & @xmath90.017 & 35.3 & 58.9 & 0.7 & rgb & y + 21925 & 7:46:51.15 & @xmath94:40:29.1 & 15.03 & 1.721 & @xmath90.038 & 35.3 & 58.2 & 0.7 & rgb & y + 21977 & 7:46:51.78 & @xmath94:43:18.1 & 16.55 & 1.522 & @xmath90.004 & 22.1 & 86.4 & 1.1 & rgb & n + 21982 & 7:46:51.74 & @xmath94:44:16.5 & 16.40 & 1.599 & @xmath90.180 & 27.2 & 3.7 & 0.9 & rgb & n + 21985 & 7:46:52.21 & @xmath94:40:03.9 & 17.18 & 1.254 & @xmath100.007 & 5.9 & 36.0 & 3.7 & rgb & n + 22013 & 7:46:52.51 & @xmath94:41:14.5 & 14.71 & 1.588 & @xmath90.023 & 35.3 & 53.9 & 0.7 & rgb & y + 22135 & 7:46:54.00 & @xmath94:47:51.6 & 17.36 & 1.170 & @xmath100.063 & 7.8 & 27.4 & 2.9 & fill & n + 22138 & 7:46:53.93 & @xmath94:48:52.8 & 16.47 & 1.368 & @xmath100.004 & 9.6 & -35.3 & 2.4 & fill & n + 22144 & 7:46:55.21 & @xmath94:36:51.1 & 16.18 & 1.389 & @xmath90.003 & 13.9 & 56.3 & 1.7 & rgb & y + 22154 & 7:46:55.11 & @xmath94:39:27.2 & 15.44 & 1.563 & @xmath90.026 & 30.8 & 57.3 & 0.8 & rgb & y + 22252 & 7:46:58.01 & @xmath94:31:40.8 & 17.14 & 1.045 & @xmath100.023 & 6.7 & 170.5 & 3.3 & fill & n + 22519 & 7:47:02.87 & @xmath94:40:58.8 & 17.09 & 1.132 & @xmath100.031 & 7.8 & 72.7 & 2.9 & rgb & n + 22584 & 7:47:04.94 & @xmath94:35:42.3 & 15.83 & 1.509 & @xmath100.002 & 27.2 & 53.8 & 0.9 & fill & y + 22603 & 7:47:04.88 & @xmath94:40:35.2 & 17.40 & 1.061 & @xmath100.027 & 10.0 & 75.4 & 2.3 & rgb & n + 22704 & 7:47:06.82 & @xmath94:48:46.7 & 17.15 & 1.183 & @xmath100.064 & 8.1 & 56.5 & 2.8 & fill & y + 22869 & 7:47:12.28 & @xmath94:43:17.8 & 17.27 & 1.195 & @xmath100.014 & 7.2 & 61.9 & 3.1 & fill & n + 22943 & 7:47:15.64 & @xmath94:37:28.7 & 17.48 & 1.190 & @xmath100.028 & 8.8 & 75.8 & 2.6 & fill & n + 22989 & 7:47:16.75 & @xmath94:39:15.0 & 14.51 & 1.558 & @xmath90.004 & 30.8 & 55.8 & 0.8 & fill & y + 23040 & 7:47:17.81 & @xmath94:44:14.4 & 14.87 & 1.823 & @xmath90.072 & 35.3 & 62.5 & 0.7 & fill & n + 23082 & 7:47:19.41 & @xmath94:42:17.9 & 17.43 & 1.088 & @xmath100.003 & 5.3 & 67.1 & 4.0 & fill & n lccccccccccc 122 & 7:20:50.84 & 1:43:08.1 & 16.89 & 1.456 & @xmath90.089 & 11.7 & 11.7 & 2.0 & fill & n + 145 & 7:21:13.56 & 1:43:20.6 & 16.64 & 1.500 & @xmath90.050 & 5.3 & 88.7 & 4.0 & fill & n + 289 & 7:21:11.57 & 1:44:49.9 & 16.33 & 1.484 & @xmath90.114 & 12.4 & @xmath96.0 & 1.9 & fill & n + 290 & 7:20:46.03 & 1:44:52.3 & 17.07 & 1.601 & @xmath90.195 & 12.4 & @xmath927.2 & 1.9 & fill & n + 311 & 7:21:00.52 & 1:45:07.0 & 17.09 & 1.579 & @xmath90.001 & 11.1 & 76.1 & 2.1 & fill & n + 389 & 7:21:20.15 & 1:45:57.2 & 17.52 & 1.679 & & 8.4 & 3.2 & 2.7 & fill & n + 521 & 7:20:52.79 & 1:47:19.4 & 16.96 & 1.430 & @xmath100.005 & 11.7 & 90.3 & 2.0 & rgb & y + 538 & 7:20:58.83 & 1:47:31.2 & 17.20 & 1.577 & @xmath100.003 & 15.9 & 97.4 & 1.5 & rgb & y + 572 & 7:20:54.89 & 1:47:53.2 & 16.66 & 1.491 & @xmath90.000 & 10.5 & 98.2 & 2.2 & rgb & y + 693 & 7:20:57.23 & 1:48:45.4 & 17.14 & 1.558 & @xmath100.000 & 5.2 & 95.3 & 4.1 & rgb & y + 706 & 7:21:14.49 & 1:48:54.0 & 17.30 & 1.461 & @xmath90.047 & 7.5 & 2.4 & 3.0 & fill & n the photometrically identified , most likely rgb and rc stars unambiguously reveal the bulk rv of each cluster , as shown by the rv distributions in the left panels ( a e ) of figure 4 . since these rgb and rc stars are the most likely to provide the `` purest '' samples of cluster members , we use these samples to define one measure of the mean cluster rv and observed dispersion ( the top values of table 11 with subscript rgb " ) . with this version of the mean rv established for each cluster , we can search for other potential cluster members identified on the basis of similar rv . the rv histograms for all stars in the cluster fields with spectra are shown in the right panels ( f i ) of figure 4 . when the left and right panels of figure 4 are compared we see the usefulness of photometric pre - screening to identify the _ best _ cluster rgb candidates for these sparse , highly field star contaminated clusters : e.g. , in the cases of bh176 , be20 , and sa1 the true rv of the cluster would remain somewhat uncertain were only the rv distributions in the right panels available . we can now rederive bulk cluster rvs using all stars with rvs indicating likely cluster membership . for this purpose , we limit the cluster member range to @xmath67 7 km s@xmath4 about the mean defined by @xmath68 , values guided by the typical individual stellar rv errors and the small expected velocity dispersions of these systems ( @xmath69-@xmath70 km s@xmath4 ) . because of this imposed selection bias , the dispersions given in table 11 should be interpreted only as a guide to the observed tightness of the rvs used to define the mean rv , not as the true intrinsic dispersions of the clusters . the mean rvs for the clusters using these enlarged samples are also given in table 11 ( middle values with subscript all " ) . note that while these rv values average over a larger sample of stars and therefore might be expected to give a more accurate measure of the mean cluster rv , it is also the case that a higher fraction of interloping field stars with similar rvs may enter , and perturb , the average . the rvs we have calculated for the five clusters using the rgb " and the all " samples agree to within 2 km s@xmath4 , which is less than half the observed dispersions . these rvs also agree in general with other recent determinations for these clusters ( bottom values , denoted as pub " in table 11 ) , but an offset of unknown origin is also found for sa1 ( see 4.4 below ) . the cmds and color - color diagrams for these clusters with observed and member stars marked are shown in figures 58 . we match isochrones to our clusters while constraining the metallicity to spectroscopically determined values . we explore a range in metallicities centered on the spectroscopically determined value for each cluster . additionally , our radial velocity data help clarify the actual positions of the the cluster rgbs to improve model isochrone fits for each cluster . through these techniques , we re - derive age and distance estimates for the clusters be29 , bh 176 , and sa1 . the derived cmds for these three gass candidate clusters are compared to theoretical isochrones from @xcite . we use as free parameters the cluster s age , @xmath30 , @xmath71 , and [ fe / h ] . we must stress that given the confused appearance of the cmds for these sparse clusters in heavily contaminated backgrounds , especially for sa1 , we must be careful not to over - interpret the results of these matches . nevertheless , useful information can still be extracted from these comparisons , and it is a useful exercise to compare these results with previously published values . in the analysis that follows , visual matching of isochrones covering a range in ages and metallicities is carried out after simultaneously varying @xmath30 and the distance modulus @xmath71 to get the best fit for any given combination of age and metallicity . we have used the following equation from @xcite : @xmath72 } = { \rm log_{10}}~(z ) + 1.72125\ ] ] to convert between the observed metallicities and the total metallicity @xmath73 values used for the @xcite isochrones . additionally , we tested @xcite @xmath7-enhanced isochrone for be29 , see 4.3 below . we measure the first reliable rv for bh176 ( @xmath74 km s@xmath4 ) based on 9 stars . this measurement is significantly different and better than our single star measurement given in f04 , which was of poorer quality and lower resolution than our new data . the photometric data used for bh176 were obtained from @xcite , where these authors carried out isochrone matching using @xcite isochrones . we rederive the cluster parameters for the same photometric dataset , but now use the isochrones from @xcite . figure 9 shows the matched isochrones when we adopt @xmath75 ( [ fe / h ] = @xmath76 ) , @xmath77 ( [ fe / h ] = @xmath78 ) and @xmath79 ( [ fe / h ] = @xmath80 ) , with clusters members , determined from the spectroscopy , marked as solid triangles . the best results are achieved for the highest of these metallicities , which is also the most metal rich value available from @xcite . a range of ages ( 5.6 and 7.1 gyrs ) seems to match fairly well the msto , the slope of the rgb , and the position of the rc , suggesting the age is not well constrained photometrically . outside this age range a poor match is obtained when trying to match these three cmd features simultaneously . an @xmath30 from 0.64 to 0.68 is therefore derived , resulting in a heliocentric distance of 15.2 - 15.8 kpc ( table 12 ) . we find similar cluster parameters ( based on our new isochrone matches with updated stellar evolution models ) to those obtained by @xcite using the same photometry . we agree with the findings of @xcite that , based on current , _ photometrically_-determined [ fe / h ] , bh176 does not seem to follow the gass amr , in fact it is quite the opposite . we have also noted that the rv for bh176 is also not strongly correlated to the gass rv trend , so , collectively , the present evidence makes bh 176 unlikely to be a member of the putative gass cluster system . analysis of the rv find that it consistent with an object ( open cluster ) rotating with the flat rotation curve of the galactic disk . however , this result could also be consistent with a globular cluster passing through the disk . if bh176 is found spectroscopically to have [ fe / h ] @xmath81 , and it is actually a globular cluster ( as it has been cataloged ; * ? ? ? * ) , bh176 could provide important new constraints on milky way globular cluster formation scenarios . the rv determinations for be 29 is within the errors ( @xmath82 ) of other published findings ( see table 11 ) . for be29 , the measured spectroscopic metallicity is @xmath73=0.005 ( @xmath83}= -0.44 $ ] ; * ? ? ? * hereafter cbvmpr ) . we explored @xmath73=0.004 ( @xmath83}= -0.68 $ ] ) , @xmath73=0.008 ( @xmath83}= -0.38 $ ] ) and @xmath73=0.019 ( @xmath83}= + 0.0 $ ] ) @xcite isochrones . figure 10 shows the cmd for be29 with the matched isochrones for the different adopted metallicities and for three different ages ( 3.2 , 3.5 and 4.0 ) gyrs . it is evident from the figure that both @xmath84 and @xmath77 isochrones do a poor job matching the slope of the rgb . the best matches are obtained for @xmath75 , and we can see that the three ages seem to do similarly well matching the rather poorly defined msto , subgiant branch , and position of the rc . ages outside this range fail to reproduce the msto and the position of the rc simultaneously . together , the isochrone matches suggest a probable age range of 3.2 - 4.0 gyrs , and @xmath30 ranging from 0.08 to 0.10 . these values yield a heliocentric distance of 13.1 - 14.5 kpc . the resulting `` revised '' cluster parameters ( table 12 ) are roughly consistent with previous work . we find a similar age , reddening and distance for be29 to that found by @xcite , though we find a lower reddening and thereby farther distance when compared with ( * ? ? ? * e(@xmath25)=0.10 , @xmath85 kpc ) . using the same isochrones for be29 and sa1 , we find that be29 _ is _ the most distant known open cluster ( @xmath86 ) in the galaxy , as found by ( * ? ? * see table 12 ) . the isochrone - fitted metallicity is in agreement , within our errors , to the spectroscopic , detailed chemical abundances of cbvmpr and @xcite . however , ( * ? ? ? * hereafter aat ) pointed out in their comparison of ngc 2243 to be29 , that the findings of cbvmpr show that be29 is somewhat @xmath7-enhanced . in figure 11 , we refit the cmd using the padova @xmath7-enhanced isochrones @xcite . aat matched @xmath7-enhanced isochrones to be29 providing a reasonably good fit within previous determinations of the cluster redding and metallicity , although they do not have stars on the rgb higher than the rc . however as shown in figure 11 , our matches to @xmath7-enhanced isochrones are markedly different . while some of the differences can be attributed to the different isochrones used , the primary difference comes from trying to match our 2 spectroscopically confirmed member stars above the rc . matching the entire cmd with @xmath7-enhanced isochrones requires one to make be29 have _ solar _ [ fe / h ] . while this match is consistent with previous reddening determinations , it is very inconsistent with previously determined _ spectroscopic _ metallicities which range from @xmath83}= -0.4 $ ] ( cbvmpr ) to @xmath83}= -0.7 $ ] @xcite . an additional method that comes near to matching the cmd , is to decrease the age to @xmath87 gyr , however this match is inconsistent with the finding of aat , who find that the age difference of be29 and ngc 2243 ( @xmath88 gyr ) can not be more than 0.5 gyr . additionally the reddening determined must be 50% higher that the highest previous estimate and double that found by aat . we find that be29 can not be consistently fit with @xmath7-enhanced isochrones , therefore to account for the variable @xmath7 element enhancements found in cbvmpr more complex models are needed . we refer the reader to 6.2 in @xcite for further discussion of the high resolution studies . using the same isochrones for be29 and sa1 ( see below ) , we find that be29 _ is _ the most distant known open cluster ( @xmath89 ) in the galaxy , as found by ( * ? ? ? * see table 12 ) . our sa1 rv ( @xmath90 km s@xmath4 ) is smaller by @xmath91 9 km s@xmath4 than the sa1 velocity obtained by cbvmpr . we note that two of our sa1 stars 572 and 693 were the only stars observed by cbvmpr for this cluster ; cbvmpr find rvs of ( 104.4 , 104.8 ) km s@xmath4 for these two stars , respectively , while our survey finds ( 98.2 , 95.3 ) km s@xmath4 . thus , while both studies find small rv differences between the two stars , the relatively large rv offset between studies persists and is apparently not related to a difference in the stellar samples . while cbvmpr does observe at much higher @xmath49 than we do , the magnitude of the rv offset is significantly larger than the estimated uncertainty in the mean cluster motion for the two surveys . moreover , we have found no net zero point offset in checks of our standard star rvs by comparison to the iau values to our measurements , as shown in table 4 . the same template stars were used for the rv determination of all of the clusters and no large offset is found in be20 , 29 , 39 or bh176 . thus , the origin of the systematic offset between the surveys is not clear . however , since we observed the same stars as cbvmpr , we can determine that the four stars selected as members are indeed consistent with the work of cbvmpr . the sparse cmd of sa1 , and lack of bright giants , limits our ability to achieve a reliable isochrone match , though our rv members provide crucial benchmarks at the magnitude of the red clump . to improve the photometry depth and provide a better match , we used the photometry of @xcite within 2 @xmath46 of the cluster center and added our confirmed members . the spectroscopic metallicity for sa1 gives a @xmath75 ( @xmath83}= -0.38 $ ] ; cbvmpr ) . we fit the ages 4 , 4.5 and 5 gyrs , with the fixed @xmath75 metallicity , are studied as shown in figure 13 . an age of 4.5 - 5 gyr is derived , since the 4 gyr isochrones do not match well the msto . this match results in an @xmath30 that ranges from 0.19 to 0.24 . the corresponding heliocentric distance for these values ranges from 11.8 to 13.5 kpc ( see table 12 ) . we find that we are in relative agreement about the cluster parameters with other photometric studies of sa1 @xcite . @xcite find an older age ( 6.3 gyr ) based on the morphological age index @xcite , which is known to yield older age estimates than those derived from isochrone matching . with defined rv - member " samples , we may examine the membership reliability of stars selected to be cluster member candidates by our photometric pre - screening . this information is useful because a large number of our candidates have not been observed spectroscopically , but if we have confidence that the photometrically - selected candidates are likely to be members , they can potentially help constrain the likely positions of the rgb and rc sequences in the cluster cmds as well as point to the effectiveness of these kinds of selections for picking spectroscopic samples in future studies . the rv membership success rate among washington@xmath1 selected rgb stars with rvs is ( 100% , 53% , 100% ) for ( sa1 , be39 , and be20 ) respectively . ergb stars were added to our analysis for be20 but none were found to be members . the membership success rate for the @xmath0 selected rgb stars was 71% for bh176 ( 60% for rgb+rc+ergb ) and 75% for be29 . in previous studies of dwarf spheroidal galaxies ( e.g. , * ? ? ? * ; * ? ? ? * ) , the washington@xmath1 technique has been found to improve the selection reliability of star system rgb stars over cmd - only selection techniques by a significant factor , even with samples selected over larger areas , to the low density extremities of the clusters . the present use of the washington@xmath1 technique is not completely analogous to these dsph studies , because here we have targeted _ any _ giant star candidates for spectroscopic follow - up without regard to cmd position ( because in some cases the location of the rgb was not certain ) , whereas the dsph studies applied both washington@xmath1 selected of giant candidates _ and _ specific cmd location pre - filtering on the stars samples . nevertheless , we do find that use of the washington@xmath1 filters has apparently improved our success rate relative to simply picking targets based on cmd position . for further discussion of the washington@xmath1 selection technique the reader is directed to @xcite . lcrrrrr # members & ( rgb ) & 4 & 4 & 6 & 18 & 4 + @xmath68 & ( km s@xmath4 ) & 9.1 & 75.6 & 27.1 & 55.0 & 95.4 + @xmath92 & ( km s@xmath4 ) & 4.6 & 2.5 & 3.8 & 2.9 & 3.6 + @xmath93 & ( km s@xmath4 ) & @xmath9103.3 & @xmath922.0 & @xmath951.2 & @xmath9107.1 & @xmath941.4 + # members & ( all ) & 9 & 5 & 11 & 24 & 4 + @xmath94 & ( km s@xmath4 ) & 11.2 & 75.7 & 28.4 & 55.0 & 95.4 + @xmath95 & ( km s@xmath4 ) & 5.3 & 2.4 & 3.6 & 2.5 & 3.6 + @xmath96 & ( km s@xmath4 ) & @xmath9101.2 & @xmath921.8 & @xmath949.7 & @xmath9107.1 & @xmath941.4 + @xmath97 & ( km s@xmath4 ) & 85.0 & 78.0 & 24.7 & 58.0 & 104.0 + @xmath98 & ( km s@xmath4 ) & 30 & 5 & 0.1 & 2 & 0.1 + ref . & & 1 & 2 & 3 & 2 & 4 lcccccccc bh176 & 6.3 @xmath67 1.0 & 15.8 @xmath67 0.5 & 9.9 & 0.70 & & @xmath1011.2 & @xmath9101.2 & ? ? ? + berkeley 29 & 3.7 @xmath67 0.5 & 13.4 @xmath67 0.4 & 21.1 & 0.09 & @xmath90.44 & @xmath1028.4 & @xmath949.7 & yes + saurer 1 & 4.5 @xmath67 1.0 & 13.1 @xmath67 0.4 & 20.2 & 0.22 & @xmath90.38 & @xmath1095.4 & @xmath941.4 & yes the clusters be29 and sa1 have potentially critical leverage on radial , age and metallicity gradients in the galactic disk as determined by old open clusters ( e.g. , * ? ? ? * ; * ? ? ? even with large statistical studies , most galactic radial trends are highly dependent on the few open clusters in the extreme outer parts of the milky way @xcite . as discussed in cbvmpr and @xcite , these two outer clusters deviate from the radial - metallicity trend found for the rest of the bulk open cluster population by @xcite , which suggests either that the chemical properties of the outer disk maybe different from the rest of the disk , or that these clusters have nothing to do with the chemical evolution of the disk . we explore the latter scenario now . @xcite has argued that the among the two most plausible models for the formation of the system of outer old open clusters , their creation as part of the normal secular evolution of the disk would require an unlikely fine tuning " of formation and destruction processes . on the other hand , formation of these clusters through accretion is a natural mechanism " for forming these open clusters , which typically have large @xmath99 . we find that , when viewed together , various properties of these old , outer open clusters lend circumstantial support to their being part of a tidal debris system , and may well be `` accretion products '' from gass . we compare the derived cluster galactocentric radial velocities ( @xmath100 ) to the @xmath101 vs. @xmath100 trend found for the gass m giants and other clusters ( * ? ? ? * f04 ) in figure 14 . included are the analytical fits to the stream from figure 2 of the analysis by @xcite for comparison . the following equation is used to convert from the @xmath102 to @xmath100 trend , as used in f04 and @xcite : @xmath103 the rvs , when combined with the parameters in table 11 , yield a @xmath100 of ( @xmath104 , @xmath105 , @xmath106 ) km s@xmath4 for ( be29 , sa1 , and bh176 ) respectively . these rvs place both sa1 and be29 within the gass trend found in @xcite and , as shown in figure 14 , are generally consistent with the monoceros models of @xcite . the new bh176 rv is only , at best , marginally consistent with membership in gass . as pointed out by @xcite and @xcite , additional information , especially proper motions , are needed to test whether these clusters are part of a coherent dynamical group consistent with the gass / monoceros system . be 29 and sa1 also follow the general age - metallicity trends suggested earlier for gass ( f04 ) . we revisit ( figure 15 ) the gass age - metallicity relation ( amr ) , now including only those clusters that are both spatially and dynamically consistent with gass and for which _ spectroscopic _ metallicities have been derived : be29 , sa1 , tombaugh 2 , arp - madore 2 , palomar 1 , ngc2808 and ngc 5286 . in addition bh176 is included ( shown as @xmath20 ) to show that if the isochrone metallicity is correct that it does not fit the amr trend . figure 14 also includes the amr trend derived for the sgr system @xcite and the metal - poor `` limit '' to the old open clusters of the galaxy @xcite shown as the long dashed lines . the sgr trend is typical of that expected for an independently evolving , `` closed - box '' system with protracted star formation , and it is clear that the proposed gass cluster system more closely matches this sgr amr than the amr of the nominal milky way open cluster system at smaller galactocentric distances . we also compared to the metal poor envelope of the ( * ? ? ? * small dashed line ) amr , which shows a `` metal - poor spike '' at @xmath1074 gyr . however , all of the three clusters that are more metal - poor than @xmath108 } = -0.6 $ ] have since had their metallicity re - evaluated and are found not to be as metal - poor . still , it is interesting to consider that all of the clusters in this `` spike '' are all distant anticenter clusters , which requires one to ask whether this is due to a common formation , and if so what caused the `` spike '' we conclude that the rvs of the clusters sa1 and be29 are consistent with gass , which further points to an `` accretion '' origin for these clusters , if one believes that gass represents an accreting dwarf galaxy system . but whether this `` accretion '' origin is by the cluster being taken in by the milky way already intact or by it being formed in the process of accretion can not be distinguish with current data . the detailed abundances of @xcite and cbvmpr lend weight to outer disk clusters being formed _ from _ an accretion event , since the abundance ratios ( specifically [ @xmath7/fe ] ) are too high compared with current dwarf spheroidal galaxies . be29 and sa1 may be associated with the thick disk which also has enhanced [ @xmath7/fe ] @xcite and could point to a common origin for both . + we thank the referee for their constructive comments that helped improve this paper . the authors would like to thank eileen friel for numerous useful discussions on this project . we would also like to thank jeff crane and jorge pearrubia for help with comparisons to the gass / monoceros models . we also express gratitude to marios chatzikos , sabrina pakzad , jennifer pope , and jason buczyna for their assistance with data collection in support of this project . this work was supported by nsf grant ast-0307851 , nasa / jpl contract 1228235 , and the david and lucile packard foundation . pmf and rrm were supported by the f.h . levinson fund of the peninsula community foundation . pmf was also supported by the virginia space grant consortium . rlp acknowledges support form a california state university , sacramento research and creative activity award . frinchaboy , p. m. , muoz , r. r. , majewski , s. r. , friel , e. d. , phelps , r. l. , & kunkel , w. e. 2005 , in `` chemical abundances and mixing in stars in the milky way and its satellites '' , eds . l. pasquini & s. randich , eso astrophysics symposia ( astro - ph/0411127 )

we investigate the characteristic effects of nuclear motion on attosecond transient absorption spectra in molecules by calculating the spectrum for different model systems . two models of the hydrogen molecular ion are considered : one where the internuclear separation is fixed , and one where the nuclei are free to vibrate . the spectra for the fixed nuclei model are similar to atomic spectra reported elsewhere , while the spectra obtained in the model including nuclear motion are very different and dominated by extremely broad absorption features . these broad absorption features are analyzed and their relation to molecular dissociation investigated . the study of the hydrogen molecular ion validates an approach based on the born - oppenheimer approximation and a finite electronic basis . this latter approach is then used to study the three - dimensional hydrogen molecule including nuclear vibration . the spectrum obtained from h is compared to the result of a fixed - nuclei calculation . in the attosecond transient absorption spectra of h including nuclear motion we find a rich absorption structure corresponding to population of different vibrational states in the molecule , while the fixed - nuclei spectra again are very similar to atomic spectra . we find that light - induced structures at well - defined energies reported in atomic systems are also present in our fixed nuclei molecular spectra , but suppressed in the and h spectra with moving nuclei . we show that the signatures of light - induced structures are closely related to the nuclear dynamics of the system through the shapes and relative arrangement of the born - oppenheimer potential energy curves .

1502.05866

recent progress in laser technology has made it possible to produce isolated attosecond pulses @xcite , and with these electron dynamics can be followed on its natural time scale @xcite . one of the promising methods to follow sub - femtosecond dynamics in electronic systems is attosecond transient absorption spectroscopy ( atas ) , which offers an all - optical approach to light - matter interactions , with the high time resolution of the attosecond pulse and the high energy resolution characteristic of absorption spectroscopy . since 2010 , atas has proven itself as a very efficient tool in studying electron dynamics in atoms . by following the movement of electron distributions in a time integrated manner , atas experiments have been used to study a number of dynamical processes . examples hereof are the observation of valence electron motion in krypton ions @xcite , autoionization in argon @xcite , the observation of ac stark shifts of excited states in helium @xcite and krypton @xcite , and very recently observation and control of the dynamics in a two - electron wave packet of helium @xcite . common for these experiments is that a near - infrared ( nir ) or infrared ( ir ) pulse dresses the system , while the modification of the spectrum of a much shorter extreme ultraviolet ( xuv ) attosecond pulse is measured . there is some ambiguity in the literature about which pulse should be refereed to as the pump ( probe ) pulse in the pump - probe setup of atas . we circumvent this problem by referring to the two pulses simply as the nir pulse and the xuv pulse. a variable delay between the xuv and nir pulses makes it possible to introduce an indirect time - resolution in time - integrated atas spectra , and in this way follow the evolution of the nir field - dressed system . theoretical investigations of atas have focused on explaining some of the typical features in the spectra such as autler - townes splittings of absorption lines @xcite , light - induced structures ( lis ) @xcite , and different interference phenomena @xcite . often simplified few - level models can capture the essential physics of a process and one can benefit from the fact that the moderate intensity of the xuv pulse places its interaction with the quantum system in the perturbative regime @xcite . finally an increasing number of direct calculations of the delay - dependent atas intensity spectrum has been carried out by solving the time - dependent schrdinger equation ( tdse ) for a variety of systems . examples hereof are given in refs . @xcite where the atas intensity spectra were found by solving the tdse for helium . in the latter case the single - active - electron approximation was used . very recently atas experiments have been carried out on nitrogen @xcite and hydrogen @xcite molecules . neglecting the slow molecular rotations , the main difference between atoms and di - atomic molecules is that molecular systems vibrate in the internuclear separation . in view of the emerging experimental interest in atas of molecules it is important to characterize the main effects associated with the nuclear vibration . such a characterization is the aim of the present work . first we investigate one - dimensional ( 1d ) @xmath1 by solving the tdse exactly within the limitations of our two models : one where the nuclear distance is fixed and one where the nuclear degree of freedom is included . we note in passing that there is a long history in strong - field physics for using co - linear @xmath1 as a generic system to highlight molecular effects @xcite . the tdse calculations in @xmath1 show that the dissociative nature of the excited electronic states in the moving - nuclei model leads to major differences in the atas spectrum when compared to the fixed - nuclei case . the fixed - nuclei spectra are very similar to atomic spectra . in both cases the spectra are dominated by narrow , well - separated , absorption features centered around energies corresponding to bound states of the system . in moving - nuclei spectra , on the other hand , the absorption features are much broader and not necessarily well - separated . in the analysis of these broad moving - nuclei induced absorption lines we establish a method to determine the dissociation time of nuclear wave packets evolving on certain excited bo surfaces . this method does not depend on a pump - probe setup , and information about the dissociation time can be obtained from single - pulse experiments . we also find that the atas spectra obtained from tdse calculations can be reproduced accurately using an approach based on the born - oppenheimer ( bo ) approximation and an expansion in a finite electronic basis . this approach is then used to calculate the atas spectrum of three - dimensional ( 3d ) @xmath2 including nuclear motion . compared to @xmath1 , h@xmath0 represent a case where the lowest excited electronic states are binding . in the atas spectrum of @xmath3 we see a rich structure associated with population transfer to vibrational states of excited bo surfaces . in the spectrum of @xmath3 we recognize some of the characteristics found in fixed nuclei @xmath1 and atomic spectra , however , other effects such as autler - townes splittings and lis at well - defined energies are absent in the moving nuclei spectrum . our analysis show that the suppression of such lis in moving nuclei systems are directly related to the molecular dynamics of the system , and that lis in molecules in general are extremely dependent on the population of the involved vibrational states and thereby on the bo curves of the involved electronic states . the paper is organized as follows . in sec . [ theory ] we describe a theory for calculating the atas intensity spectrum . we use the single - system response model of ref . @xcite ( see also ref . @xcite ) . within the approximations given in sec . [ theory ] , the atas intensity spectrum is readily determined if the dipole moment of the system is known for all times . in sec . [ theory ] we also establish the relationship between our method and other methods often used to calculate the atas spectrum @xcite . in sec . [ results ] we present calculated atas spectra for our two models of @xmath1 , and discuss the atomic - like characteristics of the model with fixed nuclei . a three - level model description capturing some of the interference features is discussed in appendix [ app ] . in the model of @xmath1 where the nuclei are allowed to move , we find that the absorption lines in the atas spectrum become very broad . in sec . [ discussion ] we set up a model to explain these broad absorption lines , and use it to closer analyze the dissociation process of excited electronic states . in sec . [ h2 ] we present the atas spectrum for 3d @xmath2 for fixed and moving nuclei . the fixed nuclei spectrum is very similar to the atomic - like , fixed nuclei spectrum of @xmath1 . however , as a result of population of vibrational states , corresponding to different excited electronic states , the moving nuclei atas spectrum of @xmath2 is very complex , and we will focus on the overall structure , interference phenomena in the absorption lines and lis . finally sec . [ conclusion ] concludes the paper . atomic units ( @xmath4 ) are used throughout unless indicated otherwise . the description of a laser pulse propagating though a medium in principle requires a full quantum mechanical description of both the field and the medium . in this work , however , all pulses considered are far too intense to show any non - classical behavior , and the fields are therefore treated classically throughout . even in this semi - classical limit the full description of atas requires that we solve the tdse and maxwell s equations simultaneously , which is computationally very demanding @xcite . in ref . @xcite , it is argued that the atas intensity spectrum of helium found using a single - atom response model , does not vary qualitatively from the spectrum found using a description including light propagation if the target density is not too high . a similar conclusion can be drawn from the work presented in ref . @xcite . since our field parameters are very similar to the ones used in ref . @xcite , and our main focus is a qualitative description of characteristic molecular effects in atas , we expect the single - system response model to be sufficiently accurate for our purpose , and we will use it throughout the paper . to calculate how the xuv pulse is modified by propagation in atas experiments , we follow the approach of ref . @xcite . from maxwell s equations we obtain the wave equation for the electromagnetic field @xmath5 @xmath6 where @xmath7 is the speed of light in vacuum and @xmath8 is the time - dependent polarization of the target defined as the dipole moment density in the target region . we consider fields linearly polarized along the @xmath9-direction , and propagating in the positive @xmath10-direction . we apply the single - system response model , thereby reducing the interaction between field and matter to a single point in space and for the @xmath9-components of @xmath5 and @xmath8 we obtain @xmath11 the polarization is given in terms of the @xmath9-component of the single - atom or single - molecule time - dependent dipole moment @xmath12 where @xmath13 is the density of atoms or molecules in the sample and @xmath14 is the expectation value of the dipole operator @xmath15 , with @xmath16 the solution of the tdse . the classical field in eq . is therefore linked to the quantum evolution of the system through @xmath17 . we now split the laser field in an incoming field and a field generated by the light - matter interaction . we assume that the incoming field propagates unchanged for all times @xmath18 all modifications of the total field are thus included in the generated field @xmath19 . for the electric field of eq . , the wave equation can be solved analytically . the solution for the generated field is @xcite @xmath20 , \label{field_gen}\end{aligned}\ ] ] where @xmath21 is the heaviside function . from eq . we see that the generated field has two contributions propagating in opposite directions . in the observation point @xmath10 , the intensity spectrum corresponding to the electromagnetic field @xmath22 is expressed as @xmath23 + \left\vert e_\text{gen}(\omega ) \right\vert^2 , \label{spec_tot}\end{aligned}\ ] ] where @xmath24 and @xmath25 are the fourier components @xmath26 @xmath27 equation is obtained from eq . by partial integration , only retaining the solution propagating in the positive @xmath10-direction , and finally eq . we now consider the three terms of eq . . the first term contains only information about the incoming field , and is therefore not of any interest , since we want to investigate the modification of the xuv pulse induced by the target . the two remaining terms contain information about this modification . however , since @xmath28 , the second term dominates over the third . henceforth we therefore refer to @xmath29 , \label{response}\end{aligned}\ ] ] as the atas spectrum . as seen from eq . absorption ( emission ) is described by negative ( positive ) values of @xmath30 . in eq . we have introduced the variable time delay @xmath31 between the nir and the xuv pulses in the incoming field . it should be noted that the atas spectrum of eq . is not the full intensity spectrum , but the modification of the intensity spectrum as compared to free propagation of the laser pulses . we refer to the full intensity spectrum of atas experiments , given by eq . , as the atas intensity spectrum . for the atas spectrum to give information about the modification of the xuv pulse , it is used that the frequency profile of the nir field is well - separated from the frequency profile of the xuv field . furthermore , the frequency profile of the higher - order harmonics of the nir field must also be well separated from the spectral range of the xuv pulse . we insure that this is the case by choosing nir pulses with moderate intensities such that only few - photon transitions are induced by these . in closing this section we relate our expressions for the atas spectrum and intensity spectrum to other expressions used in the literature for similar quantities . in ref . @xcite ( see also ref . @xcite ) the response function @xmath32 \nonumber \\ & = -\frac{c}{\omega n } \tilde{s}(\omega,\tau ) \label{ref16}\end{aligned}\ ] ] is used as a measure of the modification of the atas intensity spectrum . in eq . , @xmath33 is the fourier component @xmath34 . apart from the sign convention , our approach to calculate the atas spectrum is therefore very similar to the approach of ref . another widely used approach to calculate the atas intensity spectrum is to use beers law ( see , e.g. , eq . ( 24 ) in ref . @xcite ) @xmath35 , \label{beer}\end{aligned}\ ] ] where @xmath36 is the fourier component of the xuv field when it has propagated the distance @xmath37 into a homogeneous target medium , and @xmath38 is the fourier component of the polarization . for weak absorption it is accurate to expand eq . to first - order in the argument of the exponential function , and we find that @xmath39 this means that both here , and in the theory of ref . @xcite , the atas spectrum @xmath30 of eq . is a good measure of the modification of the xuv intensity spectrum when the generated field is weak ( @xmath28 ) . some times the cross section rather than the intensity modification is reported in the literature . this is , e.g. , the case in refs . @xcite . in ref @xcite it is shown that the absorption cross section @xmath40 is directly related to @xmath30 through @xmath41 . in this section we present the atas spectrum of eq . for two different 1d models of @xmath1 . in sec . [ fixeds ] we present the atas spectrum calculated in the fixed nuclei model of @xmath1 . for this system we recognize many features from atomic spectra previously reported in the literature @xcite . in sec . [ moving ] we present results for the moving nuclei model of @xmath1 , and we find that dissociation processes completely change the spectrum . we define the @xmath9-component of the electric field @xmath42 through the 1d vector potential @xmath43 \cos \left [ \omega ( t - t_c ) \right ] , \nonumber \\ \quad t = n_c t_c = n_c\frac{2\pi}{\omega } , \label{vec_pot}\end{aligned}\ ] ] where @xmath44 . @xmath45 relates to the intensity as @xmath46 , with one atomic unit of intensity equal to @xmath47 w/@xmath48 . in eq . @xmath49 is the center of the pulse and @xmath50 is the carrier frequency . the full width at half maximum ( fwhm ) duration @xmath51 is related to @xmath52 used in eq . by @xmath53 . for the calculations reported in this section , we use a 700 nm nir pulse with @xmath54 14.0 fs [ @xmath55 6 in eq . ] and a 50 nm xuv pulse with @xmath54 330 as [ @xmath55 2 ] . the intensities of the pulses are @xmath56 w/@xmath48 and @xmath57 w/@xmath48 , respectively . at these intensities the nir pulse alone can not excite the molecule from its ground state to the spectral region of the xuv pulse , as many - photon ( @xmath58 ) processes are greatly suppressed . to propagate solutions of the tdse for the two systems we use a fast fourier transform ( fft ) split - step operator algorithm @xcite , as described in refs . @xcite . we consider a model for @xmath1 of reduced dimensionality with the fixed internuclear distance @xmath59 which is the expectation value of the internuclear distance in the ground state . the tdse in the dipole approximation and velocity gauge reads @xmath60 with the hamiltonian @xmath61 where @xmath9 is the electronic coordinate measured with respect to the center - of - mass of the nuclei . the components of the hamiltonian in eq . are @xmath62 , @xmath63 , @xmath64 , @xmath65 , @xmath66 is the reduced electron mass , @xmath67 , @xmath68 is the proton mass , and the softning parameter @xmath69 for the coulomb singularity is chosen to produce the exact three - dimensional bo electronic energy at @xmath70 . a box size of @xmath71 , @xmath72 and time steps of @xmath73 ensured converged results . at the box boundaries a complex absorbing potential was used to avoid unphysical reflections ( see ref . @xcite for details ) . following excitation by the xuv pulse , an oscillating dipole moment is established in the target molecule . in experiments the time - dependent dipole moment dephases during a time @xmath74 due to collisional broadening and finite detector resolution @xcite . to include these effects we use a window function @xmath75 to damp @xmath14 on a time scale , that produces spectra comparable to experimental results . further , we ensure that the dephasing time is long enough for the qualitative features of the atas spectrum not to vary with @xmath74 . for a window function @xmath76 \right\ } & ( \tau \leq t \leq \tau+t_0 ) \\ 0 & ( t>\tau + t_0 ) , \end{cases } \label{window}\end{aligned}\ ] ] where @xmath77 is measured with respect to the center of the nir pulse , we find that @xmath78 fs is suitable . in fig . [ fixed ] we present the atas spectrum @xmath30 [ eq . ] calculated for the fixed nuclei model of @xmath1 , with delays of the xuv pulse with respect to the nir pulse varying from -28 fs to 14 fs . ( color online ) atas spectrum @xmath30 [ eq . ] calculated for the fixed nuclei model of @xmath1 , as a function of the photon energy , for a range of delays between the xuv and nir pulses . the atas spectrum is normalized such that maximum absorption is at @xmath79 . for photon energies between 14.0 ev and 20.3 ev the atas spectrum has been multiplied by a constant factor of 50 to highlight the lis doublet in that region ( see text ) . the white line in the upper right part of the figure indicates the intensity profile of the xuv pulse . pulse parameters : @xmath80 as , @xmath81 fs , @xmath82 nm , @xmath83 nm , @xmath84 w/@xmath48 , @xmath85 w/@xmath48.,scaledwidth=48.0% ] for large positive delays , the nir pulse arrives at the target well before the xuv pulse . because the relatively weak nir pulse leaves the system in its field - free ground state , it will not have any effect on the atas spectrum , when it arrives well before the xuv pulse . the spectrum therefore shows clean absorption lines without side - bands or other modifications for delays larger than @xmath86 fs . the position of the four absorption lines in fig . [ fixed ] at approximately 10 ev , 21.5 ev , 25.5 ev , and 27 ev , correspond to resonances between the field - free ground state and states of opposite symmetry under the reflection @xmath87 with energies 9.94 ev , 21.48 ev , 25.29 ev and 26.94 ev above the ground state energy , respectively . for large negative delays , the xuv pulse arrives well before the nir pulse . the nir pulse therefore arrives when an oscillating dipole moment has already been induced . even though the nir pulse does not excite additional population from the ground state to the states reached by the xuv field , the nir pulse will modify @xmath14 . one effect of this modification is seen around the absorption peak at @xmath88 ev in fig . [ fixed ] . for delays more negative than @xmath89 fs , we see that the absorption line has several sidebands . such sidebands have previously been observed in atomic systems ( see , e.g. , ref . @xcite and references therein ) . another effect introduced by the nir field is a periodic modulation of the absorption line at @xmath90 ev in fig . [ fixed ] ( enlarged in the upper panel of fig . [ fringes ] ) . this effect extends from large negative delays and all the way to the region where the two pulses overlap in time , and is a consequence of quantum interference between populated excited states of the same symmetry @xcite . for large negative delays the xuv and nir pulses are separated in time , and the role of the xuv pulse is to populate the excited states at time @xmath31 while the nir pulse modifies the populations and coherences at a later time t@xmath91 . in appendix [ app ] we describe a ( three - level ) model which explains and reproduces the periodic modulation of the absorption line at @xmath92 25.5 ev . a result of this treatment is that oscillations of this type always have a period of @xmath93 with @xmath94 the period of a nir pulse cycle . the region @xmath95 , where the two pulses overlap in time , is dominated by intense coupling between states . as a consequence , absorption , and to a smaller extent emission , appears at certain frequencies in this region only . here we will focus on two features both visible in fig . [ fixed ] : the autler - townes splitting @xcite of the absorption line at @xmath88 ev , and the lis @xcite doublet with spectral features at @xmath96 ev and @xmath97 ev corresponding to energies one photon energy above and one photon energy below the even parity state at @xmath98 ev . when the two pulses overlap in time , we see a clear broadening and a splitting of the absorption line at @xmath88 ev . such features originate from coupling between bound states and have been observed both experimentally @xcite and in calculations @xcite . as discussed in refs . @xcite , the nature of the autler - townes splitting can be understood in few - state models . lis are signatures of population transfer , by two - photon processes , from the ground state to an excited electronic state of the same symmetry , under the reflection @xmath87 . in spectroscopy such a state is often referred to as a dark state . two different processes , one involving absorption of a nir photon and one involving emission of a nir photon , can populate the same dark state . as a result lis always appear in pairs with an energy separation of @xmath99 between the individual absorption features . the two intermediate states absorbing xuv light are always in two nir - photon resonance resulting in strong coupling between the states @xcite . this coupling leads to a @xmath100 periodic modulation of the lis . in ref . @xcite it has been shown that specific lis in helium can be reproduced using only three states in the calculation of the atas spectrum . for @xmath1 with fixed nuclei we set up a similar three - level model including the ground state , a dark state @xmath98 ev above the ground state , and a dipole allowed state @xmath101 ev above the ground state . we will refer to these field - free states as @xmath102 ( ground state ) , @xmath103 ( dark state ) , and @xmath104 ( dipole allowed , excited state ) and their corresponding energies as @xmath105 , @xmath106 , and @xmath107 . the wave function of a general state in the three - level model is given by @xmath108 we find that the three - level model reproduces the lis shown in fig . [ fixed ] at @xmath109 well . for the state the expectation value of the dipole operator is well approximated by @xmath110 , \label{2-dipole}\end{aligned}\ ] ] since the condition @xmath111 holds for field parameters used in this work . in the pursuit of a simple analytical theory , capable of explaining the characteristics of lis , we now simplify our three - level model further . this analytical result will be needed for understanding the moving nuclei results of sec . the spectral widths of lis are determined by the duration of the nir pulse . this indicates that the temporal region where the xuv field is on only has a small direct effect on the atas spectrum ( the fourier transform of the dipole moment in this region is vanishing when compared to the contribution from the remaining temporal region of the nir field ) . the main contribution from the temporal region where the xuv and nir pulses are overlapping is therefore to populate the states @xmath103 ( two - photon process ) and @xmath104 ( one - photon process ) . when the shorter xuv pulse has died out , the population transfer from @xmath102 to the excited states will stop , and the dynamics will be dominated by coupling between @xmath103 and @xmath104 induced by the nir field . when we consider instants of time after the xuv pulse ( @xmath112 ) but before the end of the nir pulse , the dynamics of the system is approximately governed by the equations @xmath113 in eqs . - the multi - photon nir - coupling between the ground state and excited states has been neglected since such multi - photon processes are greatly suppressed for the field parameters used in this work . if we for simplicity assume that the nir field is given by the monochromatic field @xmath114 , we have @xmath115.\end{aligned}\ ] ] for the field parameters used in fig . [ fixed ] the detuning between @xmath103 and @xmath104 is large ( @xmath116 ev , @xmath117 ev ) and the population transfer is therefore low . an approximate analytic expression for @xmath118 can therefore be found using perturbation theory . first order perturbation theory can reproduce the characteristics of the lis and is desirable since the result is relatively simple @xmath119 c_d ( \tau ) . \label{e_coef}\end{aligned}\ ] ] if we insert eq . into eq . , we see that @xmath14 will contain terms oscillating with frequencies @xmath109 . @xmath120 [ see eq . ] found using this dipole moment , will therefore contain spectral features at @xmath109 - the lis seen in fig . [ it should be noted that the part of the time - dependent dipole moment responsible for the lis depends critically on the initial population of the dark state , which is in compliance with the two - photon nature of lis . from eqs . and we conclude that the nir - coupling of the excited states in our three - level model gives rise to lis . the simple analytical model for lis established in this section will be the starting point of an investigation of lis in molecular systems in sec . [ h2 ] . figure [ fixed ] and the associated discussion recapitulate many of the physical processes already investigated in atoms . the detailed spectral and temporal information about electron dynamics extracted from fig . [ fixed ] is a consequence of the long dipole response of the system , allowing for spectral resolution much better than the fourier limit of the xuv pulse . with the phenomena discussed in this section in mind , we are ready to explore the effect of molecular vibrations . in this section we present the atas spectrum of eq . for the moving nuclei model of @xmath1 using the same field parameters as in sec . [ fixeds ] . we consider a simplified model for @xmath1 with reduced dimensionality that includes only the dimension that is aligned with the linearly polarized laser pulse @xcite . within this model , electronic and nuclear degrees of freedom are treated exactly . the center - of - mass motion of the molecule can be separated , such that the tdse for the relative motion in the dipole approximation and velocity gauge reads @xmath121 with the hamiltonian @xmath122 where the wave function now depends on the internuclear distance @xmath123 and the electronic coordinate @xmath9 measured with respect to the center - of - mass of the nuclei . the components of the hamiltonian in eq . were defined after eq . , except the nuclear kinetic energy @xmath124 and @xmath125 , where the softning parameter @xmath126 for the coulomb singularity is chosen to produce the exact three - dimensional @xmath127 bo potential energy curve @xcite . a box of @xmath71 , @xmath72 in the electronic coordinate together with a box of nuclear coordinates of @xmath128 , @xmath129 and time steps of @xmath73 ensured converged results . at the box boundaries complex absorbing potentials were used in both electronic and nuclear coordinates to avoid unphysical reflections ( see refs . @xcite for details ) . the atas spectrum is shown in fig . [ moving ] . ( color online ) normalized atas spectrum @xmath30 [ eq . ] , calculated using the moving nuclei model [ eqs . and ] , for 1d @xmath1 , as a function of the photon energy , for a range of delays between the xuv and nir pulses . the white line to the right shows the intensity profile of the xuv pulse . the pulse parameters are as in fig . [ fixed].,scaledwidth=48.0% ] it is clear from fig . [ moving ] that the relatively sharp lines we saw for the fixed nuclei model [ fig . [ fixed ] ] have disappeared . two very broad absorption lines corresponding to population transfer between the ground and the first excited state ( absorption feature at @xmath130 ev ) and between the ground state and the third and fifth excited states ( merged absorption feature at @xmath131 ev ) dominate the spectrum . the broad features in fig . [ moving ] persist as @xmath132 , and are therefore not a result of the nir pulse . the only effect of the nir pulse in fig . [ moving ] is the small oscillations with a period of @xmath100 clearly present between @xmath133 ev for delays @xmath134 fs @xmath135 fs , corresponding to the region where the pulses overlap the most . in sec . [ discussion ] we investigate the molecular dynamics responsible for the broad absorption lines in fig . [ moving ] . in sec . [ ns_h2 ] we also briefly discuss the origin of the @xmath100 oscillations in fig . [ moving ] . in this section we set up a simplified model to explain the broad absorption features in the atas spectrum of the moving - nuclei model of @xmath1 . to isolate the effect we first neglect the nir pulse , whose impact we found to be small [ fig . [ moving ] ] . the analysis will show that the dissociative nature of the excited electronic states in @xmath1 is responsible for the broad absorption features in the atas spectrum and will provide a link between the spectral width of the atas spectrum and the time it takes for the part of the nuclear wave packet found in an excited electronic state to dissociate . in the @xmath136-surface model the wave function of the molecule is expanded in the lowest @xmath136 electronic bo states , @xmath137 : @xmath138 where @xmath139 are the corresponding nuclear wave packets and @xmath9 denotes the electronic coordinates . in the @xmath136-surface model of eq . , the time - dependent expectation value of the @xmath9-component of the dipole operator @xmath140 , entering the evaluation of the atas spectrum as discussed in sec . [ theory ] , is given by @xmath141 where @xmath142 is the electronic dipole moment function . the evolution of the nuclear wave packet @xmath143 on the energy - surface @xmath144 , corresponding to the @xmath145th electronic state , is governed by the equation @xmath146 obtained by inserting the @xmath136-surface expansion in the tdse [ eq . ] and neglecting terms containing derivatives of the electronic states @xmath147 with respect to @xmath123 . in eq . @xmath148 where @xmath144 is the bo potential energy curve obtained by solving the electronic problem at fixed r. due to the truncation in the number of electronic states , the @xmath136-surface expansion is not gauge invariant , and the dynamics are only correctly described in the length gauge @xcite . in eq . we therefore have that @xmath149 , where @xmath150 , with @xmath151 and @xmath152 the nuclear charges , @xmath153 and @xmath154 the nuclear masses and @xmath155 the total mass of the molecule . note that an atomic version of the @xmath136-surface model was used for he in ref . @xcite . for a further analysis of the broad absorption features in fig . [ moving ] , we expand the nuclear wave packets @xmath156 in the vibrational eigenstates @xmath157 corresponding to the @xmath145th electronic state @xmath158 where @xmath159 are the eigenenergies of @xmath160 . the set of @xmath160 s for a specific @xmath145 can be continuum states , or a mixture of bound and continuum states . the continuum states are energy normalized , such that @xmath161 using the expansion , the time - dependent dipole moment of eq . reads @xmath162 where @xmath163 often the function @xmath164 varies slowly with @xmath123 , as long as @xmath123 is varied in a region where @xmath165 is non - zero . in such cases it is a good approximation to let @xmath166 and we obtain @xmath167 where the subscript - indicates that the variation in @xmath123 is neglected for the function @xmath164 . in eq . @xmath168 is the franck - condon overlap between the states @xmath160 and @xmath169 . the intensity of the xuv pulse is generally very low in atas , and it is reasonable to assume that the ground state population remains unchanged during the pulse , i.e. , @xmath170 , and that the expansion coefficients @xmath171 for vibrational states , corresponding to the bo surfaces of excited electronic states , can be found using first - order perturbation theory @xmath172 in @xmath1 all excited bo surfaces are dissociative , and we therefore expect @xmath14 to be damped when the excited nuclear wave packets dissociate [ see eq . ] this damping leads to a broadening of the atas spectrum . for @xmath1 we found that the moving nuclei atas spectrum of fig . [ moving ] can be reproduced to a very high accuracy using an @xmath136-surface expansion [ eq . ] of the total wave function followed by propagation of the nuclear wave functions @xmath143 in accordance with eq . . the nuclear wave packets were propagated using a split step algorithm , @xmath173 in eq . , @xmath128 , @xmath129 and time steps of @xmath174 ensured converged results . the fact that atas spectra can be precisely determined in the @xmath136-surface model is of importance for studies of atas in larger systems which can not be treated by direct solution of the tdse . in sec . [ h2 ] we treat h@xmath0 in three dimensions using the @xmath136-surface model . a very useful property of the @xmath136-surface expansion is that we can subtract specific surfaces from the expansion during the propagation , and in this way determine their influence on the atas spectrum . in fig . [ moving ] we saw oscillations with a period of @xmath100 in the region @xmath175 ev . by performing separate calculations , we have found that these oscillations are due to two - photon processes involving the electronic even - parity states @xmath176 and @xmath177 . in @xmath1 the coupling from the ground state to the first excited electronic state is several orders of magnitude larger than the coupling to higher excited electronic states when evaluated near @xmath178 . it is therefore expected that pulses with broad frequency distributions primarily populate the first excited state . ( color online ) atas spectrum @xmath30 [ eq . ] ( solid , black lines ) and field intensity ( dashed , red lines ) as functions of the photon energy . in the left column pulses with @xmath179 in eq . have been used in the calculations , while pulses with @xmath180 have been used in the right column . the central carrier frequencies of the pulses used in the calculations are 6 ev for ( a ) and ( b ) , 10 ev for ( c ) and ( d ) , 20 ev for ( e ) and ( f ) , 30 ev for ( g ) and ( h ) , and 50 ev for ( i ) and ( j).,scaledwidth=48.0% ] this leads to an interesting effect for ultrashort pulses ; the shape of the atas spectra shown in fig . [ multi ] for pulses with @xmath179 in eq . and carrier frequencies corresponding to energies higher than 20 ev , turn out to be remarkably unchanged as the carrier frequency is varied . as can be seen from fig . [ multi ] , this is not the case for longer pulses . the fwhm of the three pulses with @xmath179 and central carrier frequencies corresponding to 20 ev , 30 ev and 50 ev are 2.67 ev , 2.67ev and 2.64 ev , respectively . this indicates that the atas spectra for these ultrashort pulses are only determined by the system , and that only the ground state @xmath181 and the lowest excited electronic state @xmath182 are substantially populated . in the case of an ultrashort pulse the dipole signal is therefore expected to be well approximated by a simplified version of eq . @xmath183 , \label{dipole_pertb2 } \\ \nonumber & u(e_{2,k})=\int^t dt ' e_\text{in}(t ' ) e^{-i(e_{1,0 } -e_{2,k})t'},\end{aligned}\ ] ] where the superscript @xmath184 indicates that we have used first - order perturbation theory [ eq . ] to determine the coefficients in eq . . equation is a good approximation for the time - dependent dipole moment as confirmed by the results in fig . [ dipole_mov ] , which will be discussed later . after the pulse is over @xmath185 does not vary with time . in @xmath1 the fwhm @xmath186 of @xmath187 is @xmath188 ev . for pulses of duration much less than @xmath189 fs we can therefore neglect the variation of @xmath185 with respect to @xmath190 ; @xmath191 . for such short pulses we have @xmath192 , \qquad(t\gg t_c ) \label{dipole_pertb3}\end{aligned}\ ] ] where @xmath49 is the center of the pulse . @xmath193 is the inverse fourier transform of the frank - condon overlaps @xmath187 with respect to the fourier variable @xmath190 . it turns out that @xmath187 is well approximated by a gaussian function @xmath194 centered around @xmath195 , where @xmath196 is the value of the first excited bo surface , evaluated at @xmath59 . substitution with the variable @xmath197 in eq . gives @xmath198 t } \nonumber \\ & \int d\tilde{e}_{2,k } d(\tilde{e}_{2,k } ) e^{-i \tilde{e}_{2,k}t } \bigg ] . \qquad(t\gg t_c ) \label{dipole_pertb4}\end{aligned}\ ] ] since the fourier transform of a gaussian is also a gaussian , we can now conclude that the time - dependent dipole moment of , after the pulse is over , is oscillating with frequency @xmath199 and is damped by a gaussian function . the period of oscillation corresponding to the energy @xmath199 is @xmath200 as . figure [ dipole_mov ] shows the time - dependent dipole moment from @xmath1 calculated using a 30 nm pulse with an intensity of @xmath201 w/@xmath48 and @xmath179 in eq . . the fwhm duration of this pulse is 83 as , and we therefore expect the corresponding dipole signal to be well described by eq . , when the pulse is over . a gaussian fit to the absolute values of the peaks of the dipole signal , in the region where the pulse has died out ( more than 165 as , ie . , 2 fwhm durations , after the central time of the xuv attosecond pulse ) is shown in fig . [ dipole_mov ] , and we see that the damping of the dipole signal is well described by a gaussian function . we also see that the dipole signal is oscillating with a period close to the expected period of 420 as . ( color online ) time - dependent dipole moment from the full tdse calculation ( solid , black line ) and the approximation @xmath202 of eq . ( dashed , red line ) calculated for the 1d model of @xmath1 and a single - cycle 30-nm pulse . the intensity of the pulse is @xmath203 w/@xmath48 . the light blue dotted line shows a gaussian fit to the norm of the extrema in the damped dipole moment . the first data point included in the fit is the extremum at @xmath204 fs . the xuv pulse is centered at @xmath205 fs.,scaledwidth=48.0% ] assuming that the fourier transform of the incoming field is constant in the region where the atas spectrum is non - zero ( see fig . [ multi ] ) , the atas spectrum of eq . is proportional to the fourier transform of the time - dependent dipole moment [ neglecting the linear frequency dependence from eq . ] . for ultrashort pulses the atas spectrum of @xmath1 is therefore well - approximated by a gaussian function , and the half - life @xmath206 of the envelope of @xmath14 is related to the fwhm @xmath207 of the atas spectrum through the relation @xmath208 from the three ultrashort pulses discussed earlier ( left panel in fig . [ multi ] ) with @xmath209 ev we find that @xmath210 as , which is indicated on fig . [ dipole_mov ] . the time @xmath206 indicates the duration it takes the time - dependent dipole - moment to decay and it therefore sets the time scale in which the nuclear wave packet on the excited state @xmath211 travels to a region where the overlap with the ground - state wave packet @xmath212 is significantly decreased . the value of @xmath206 shows that this dynamics takes place on a subfemtosecond time scale . this very fast nuclear motion explains the extremely broad absorption features in the spectrum shown in fig . [ moving ] . in @xmath1 , all bo surfaces corresponding to excited electronic states lead to dissociation . in molecules in general there are also excited bo surfaces supporting bound nuclear motion . to have a more general description of nuclear motion effects in atas , we therefore consider the h@xmath0 molecule which is sufficiently simple to allow for an accurate numerical treatment , but still show the characteristics of excited bo surfaces with bound nuclear motion . inspired by the good performance of the @xmath136-surface model in 1d @xmath1 , we represent the total wave function of 3d h@xmath0 in the @xmath136-surface expansion [ eq . ] . the nuclear wave functions @xmath143 are propagated in accordance with eq . using a fft split - step algorithm . bo surfaces and dipole transition moments used in the propagation are given in the literature @xcite . we must ensure that the basis used in the expansion of eq . is large enough to contain all physics relevant to the atas spectrum . to reduce the number of basis states @xmath136 needed , we choose the intensity of the dressing field low enough that processes involving more than two low - energy photons are suppressed . calculations of @xmath30 for h@xmath213 with fixed nuclei did not show any signs of 3-nir - photon or 4-nir - photon processes for the field parameters we use in this section . to further reduce @xmath136 , we use a 1600 nm ir dressing pulse . under these criteria it is sufficient to use @xmath214 bo surfaces in the expansion . the propagation of @xmath143 with @xmath215 , @xmath216 and time steps of @xmath174 ensured convergenced results . to highlight the effect of nuclear dynamics on atas spectra we also preform a fixed nuclei calculation on 3d h@xmath0 using the same number of electronic states as described above . in the fixed nuclei calculation we use an internuclear separation of @xmath217 , corresponding to the separation at minimum energy in the ground state bo curve . in fig . [ 1600 ] we present the atas spectrum @xmath30 calculated for @xmath2 with delays of the xuv pulse with respect to the ir pulse varying from -84 fs to 10 fs . figure fig . [ 1600](a ) shows the fixed nuclei results , while fig . [ 1600](b ) shows the atas spectrum , when the nuclei are allowed to move . the field parameters are the same in the two calculations . the ir pulse is a 1600 nm pulse with a fwhm of @xmath218 fs [ @xmath219 in eq . ] , and the xuv pulse is an 80 nm pulse with a fwhm of @xmath220 as ( @xmath219 ) . the intensities of the pulses are @xmath221 w/@xmath48 and @xmath222 w/@xmath48 , respectively . the 11th ( singlet ) bo surface in h@xmath0 has an energy minimum of @xmath223 ev above the ground state energy @xcite , and the photon energy of the ir field is 0.775 ev . in fig . 5 regions of photon energies above @xmath224 ev @xmath225 ev @xmath226 ev are shaded indicating that we expect structures in this region of the atas spectrum not to be well described in the basis of 10 bo surfaces . in @xmath2 we , in addition to electronic dynamics , want to follow nuclear dynamics occurring on a much longer timescale . we therefore use a longer dephasing time for the time - dependent dipole moment than for the case of @xmath1 with fixed nuclei . for the calculations in this section we use the window function with @xmath227 fs , but shift the window function such that @xmath14 is undamped the first 73 fs after the end of the xuv pulse . ( color online ) normalized atas spectrum @xmath30 [ eq . ] calculated for 3d h@xmath0 . ( a ) for fixed nuclei , ( b ) for moving nuclei . the shaded areas indicate photon energy regions where the spectra are expected to be influenced by the finite electronic basis ( see text ) . the white line to the right indicates the intensity profile of the xuv pulse . the dots and triangles to the very right in ( b ) are located at the field - free energies of the vibrational states corresponding to the bo curves b@xmath228 ( black dots ) and c@xmath228 ( black triangles ) . pulse parameters : @xmath229 as , @xmath230 fs , @xmath231 nm , @xmath232 nm , @xmath84 w/@xmath48 , @xmath233 w/@xmath48.,scaledwidth=48.0% ] from fig . [ 1600 ] we immediately see that the atas spectrum of h@xmath0 is highly affected by the nuclear dynamics of the system . the effect of nuclear motion on the atas spectrum in h@xmath0 is , however , very different from what we saw in @xmath1 . in the fixed nuclei atas spectrum of fig . [ 1600](a ) we only see a single absorption line in the un - shaded region . for h@xmath0 with a fixed internuclear separation of @xmath234 , the excited states considered have energies higher than the lowest vibrational state in the corresponding bo curve for the moving nuclei system . it is therefore expected , that we can go to higher energies in the atas spectrum without expanding the basis of the calculation . for energies higher than 12.95 ev additional absorption lines are visible in the fixed nuclei atas spectrum [ one is included in the shaded area of fig . [ 1600](b ) ] . these absorption lines and the absorption line at @xmath235 ev all behave very much like the absorption lines in the fixed - nuclei spectrum of @xmath1 ; we observe interference of the type discussed in appendix a between states in two nir photon resonance , autler- townes splittings of absorption lines etc . the moving nuclei atas spectrum of h@xmath0 shown in fig . [ 1600](b ) is , on the other hand , very different from both the fixed nuclei atas spectrum of h@xmath0 , and from the moving nuclei atas spectrum of @xmath1 ( fig . [ moving ] ) . even though only three singlet bo curves ( b@xmath228 , c@xmath228 and ef@xmath236 ) are within the energy range considered in fig . [ 1600](b ) , we see rich structures in the atas spectrum . for the positive delays to the outermost right in fig . [ 1600](b ) we see that these structures converge towards absorption lines corresponding to population of the field - free vibrational states of the b@xmath228 and c@xmath228 surfaces . when the delay between the xuv pulse and the nir pulse is varied , we see that oscillations occur in the spectrum on two different time scales . throughout the spectrum there are oscillations on the time scale of half an ir period ( @xmath237 fs ) , and in the absorption lines corresponding to vibrational states of the b@xmath228 surface [ indicated by black dots in fig . [ 1600](b ) ] , we see additional oscillations on a much longer time scale . introducing the fourier transform of @xmath30 [ eq . ] @xmath238 = \frac{1}{2\pi}\int_{-\infty}^{\infty } \tilde{s}(\omega,\tau ) \exp(i\omega_\tau \tau ) d\tau , \label{f_tau}\end{aligned}\ ] ] we are able to analyze the oscillations of fig . [ 1600](b ) closer . the fourier transform eq . of the delay - dependent atas spectrum is shown in fig . [ f_freq](a ) for low and in fig . [ f_freq](b ) for higher frequencies @xmath239 . ( color online ) absolute value of the fourier transform . ( a ) for relatively low frequencies and ( b ) for higher frequencies . the dashed line in ( a ) indicates the local energy separation between vibrational states in the b@xmath228 curve corresponding to the local vibrational period of eq . . the lines in ( b ) are @xmath240 [ see eqs . ( 40 ) and ( 42 ) ] when @xmath241 is either the energy of one of the seven lowest vibrational states of the c@xmath228 surface ( dashed lines ) , one of the five lowest vibrational states of the b@xmath228 surface ( dash - dotted lines ) , or one of the three lowest vibrational states of the d@xmath228 surface ( dotted lines ) . @xcite for information on the energy levels.,scaledwidth=48.0% ] the low frequency structure in the fourier transform of the atas spectrum is a consequence of nuclear dynamics in the molecule . in the @xmath136-surface expansion , we can think of the xuv pulse as generating nuclear wave packets on bo surfaces corresponding to different excited electronic states . if the bo surfaces were harmonic , the wave packets would oscillate periodically with a period of @xmath242 , where @xmath243 is the frequency of the harmonic oscillator . however , the bo curves of @xmath2 are not perfectly harmonic , and we must account for their anharmonicity . we do this by defining a local vibrational period for the @xmath13th vibrational state as @xmath244 where @xmath245 is the energy difference between the @xmath246th and the @xmath247th vibrational states of the same electronic state . the local energy separation @xmath248 for the @xmath13th vibrational state is plotted in fig . [ f_freq](a ) . in fig . [ f_freq](a ) we can also vaguely see frequency components corresponding to the energy separation of the vibrational states in the c@xmath228 curve around 0.3 ev . in fig . [ f_freq](b ) we present the higher frequency components of the fourier transform of the delay - dependent atas spectrum . note that the structures of fig . [ f_freq](b ) are centered around the energy @xmath249 ev of two 1600 nm ir photons . much of the structure in fig . [ f_freq](b ) can be explained as interference between different paths to the same final state , just as the interference in the absorption lines for fixed - nuclei @xmath1 ( see appendix [ app ] ) . for both fixed and moving nuclei the oscillation has a period of approximately half a nir or ir field cycle . we now focus on one absorption peak in fig . [ 1600 ] and denote its energy by @xmath50 . if the field - free state corresponding to this absorption peak couples to a state with energy @xmath241 , the absorption peak at energy @xmath50 will contain a term proportional to @xmath250 $ ] , neglecting a phase . we now take the fourier transformation defined in eq . , but with @xmath251 $ ] , and see that we only obtain a nonvanishing contribution when @xmath252 in fig . [ f_freq](b ) we plot @xmath253 for a number of energies @xmath241 corresponding to vibrational states in different bo surfaces . in practice the signal @xmath250 $ ] does not exist for all @xmath31 , and the peaks in fig . [ f_freq](b ) are therefore not delta functions , but rather peaks smeared out around sets of @xmath50 s and @xmath254 s obeying the condition . if we focus on the absorption feature at @xmath255 ev we see from fig . [ f_freq](b ) that the corresponding field - free state couples to at least three other vibrational states through the ir field . from this analysis we conclude that the interference mechanism resulting in @xmath256 oscillations in the molecular absorption lines of fig . [ 1600 ] is very similar to the mechanism resulting in @xmath100 oscillations in the fixed - nuclei absorption lines discussed in appendix [ app ] . other processes such as lis and autler - townes splittings of the absorption lines have not been found in the analysis of moving nuclei atas spectra of h@xmath0 . we now investigate whether lis at well - defined energies are hidden in the rich molecular structure of atas spectra in h@xmath0 , or a more fundamental nuclear motion effect suppresses such lis in molecular systems . for this purpose we use a perturbative electronic three - level model similar to the one used in eqs . - for @xmath1 with fixed nuclei , but now accounting for nuclear motion . the model contains the electronic states @xmath257 ( ground state ) , @xmath258 ( dark state ) and @xmath259 ( essential , dipole allowed , excited state ) . the wave function of a general state in the model can be written as @xmath260 where the nuclear wave packets @xmath261 , @xmath262 and @xmath263 have been expanded in their vibrational eigenstates [ see eq . ] . in eq . the subscripts @xmath264 and @xmath265 label the three electronic states , while the subscripts @xmath246 and @xmath247 label the vibrational states corresponding to the electronic states @xmath265 and @xmath140 , respectively . in obtaining eq . we have approximated @xmath261 by its vibrational ground state wave function @xmath266 and use @xmath267 . the approximation @xmath268 is reasonable for the field parameters used in atas and is similar to the approximation @xmath269 made in sec . [ fixeds ] . under the approximation @xmath270 [ see eq . ] the expectation value of the dipole moment found using eq . is well approximated by [ see eq . ] @xmath271 . \label{dipole_three_nuc}\end{aligned}\ ] ] as in sec . [ fixeds ] we approximate the nir ( or ir ) field by a monochromatic field @xmath272 and evaluate the coefficients @xmath273 by first order perturbation theory @xmath274 \langle \chi_{e , k } \vert \chi_{d , l } \rangle c_{d , l}(\tau ) . \label{dipole_three_nuc_coef}\end{aligned}\ ] ] inserting eq . into eq . shows that the energies @xmath275 disappear in the final expression for the time - dependent dipole moment , leaving terms oscillating with frequencies @xmath276 . the distribution and population of vibrational states corresponding to the electronic state @xmath277 does therefore not affect the dephasing of the dipole moment responsible for lis . the distribution and population of vibrational states corresponding to the electronic state @xmath278 are , on the other hand , crucial for the existence of lis as spectral structures at well - defined energies . the more dark vibrational states populated , the faster the resulting dipole moment will dephase . even in the situation where a finite number of discrete vibrational states are populated , and one would expect a partial revival of the time - dependent dipole moment , the revival is often absent due to the finite duration of the nir pulse . we therefore expect lis with well - defined energies to be suppressed in the molecular atas spectrum when more than a few vibrational dark states are populated . to test this hypothesis we return to @xmath1 for a short notice . in @xmath1 the bo curve corresponding to the dark electronic state investigated in sec . [ results ] is dissociative , and a continuum of vibrational states are populated . we therefore expect the lis corresponding to this dark electronic state to be suppressed in the atas spectrum . the atas spectrum @xmath30 using the pulse parameters from fig . [ fixed ] and a fixed delay of @xmath279 is shown in fig . [ lis](a ) . normalized atas spectrum @xmath30 [ eq . ] calculated at @xmath279 for ( a ) @xmath1 and the artificial @xmath1 molecule ( see text ) with ( b ) @xmath280 , ( c ) @xmath281 and ( d ) @xmath282 . the four insets show the bo curves @xmath283 and @xmath284 . in the insets @xmath70 is indicated by a dashed gray line . the pulse parameters are as in fig . [ fixed ] , but with a fixed delay @xmath279 between nir and xuv pulses.,scaledwidth=48.0% ] we now construct an artificial @xmath1 molecule by replacing the bo curve @xmath284 corresponding to the dark state with @xmath285 $ ] , for @xmath280 this bo curve is parallel to the ground state curve @xmath283 . according to the franck-condon principle only the dark vibrational ground state will be populated when the two bo curves are parallel . we therefore expect lis in the atas spectrum of the artificial @xmath1 molecule with @xmath280 , shown in fig . [ lis](b ) . we see that this is indeed the case since spectral features of the expected width , an energy separation of two nir photon energies , and centered around the energy of the dark vibrational ground state , appear in the spectrum . it should be noted that the coupling to the dark state curve has not been changed , and the results should therefore only be interpreted qualitatively . in figs . [ lis](c ) and [ lis](d ) the atas spectrum is shown for the artificial @xmath1 molecule with non - zero @xmath286 . as @xmath286 is increased , the lis are damped . the reason for this damping is that the slope of the bo curve , corresponding to the dark state , evaluated at @xmath70 , increases with @xmath286 and more vibrational dark states are therefore populated for larger @xmath286 . in h@xmath0 the dark state curves of interest are far from parallel to the ground state curve , and hence we do not expect any significant lis in the atas spectrum . in atas experiments using a molecular gas target , we therefore predict that lis will be very difficult to detect . an exception to this general rule might be @xmath287 where the bo surfaces are remarkably parallel . it should be noted that eq . is valid only when the detuning is large , and first - order perturbation theory can be applied . the conclusions , however , remain the same if perturbation theory of a higher order is used . further , our calculations show that there is no significant difference in the nature of lis in systems with large and small detunings . the effect of nuclear motion in atas spectra has been elucidated . we performed calculations in 1d @xmath1 and 3d h@xmath0 with fixed nuclei , and found that the spectra of these systems are very similar to atomic spectra . moreover we investigated the same two systems including nuclear motion . for the systems with moving nuclei , we saw that the nature of the bo surfaces , corresponding to excited electronic states , were crucial . in @xmath1 all excited bo surfaces lead to dissociation , and as a result the atas spectrum shows very broad absorption features reflecting the damping of the time - dependent dipole moment by the dissociating nuclei . in the theoretical analysis of these broad absorption lines we found that , for specific field parameters , the width of the absorption lines were solely system dependent . the width of the lowest absorption line in the atas spectrum could therefore be linked directly to the time it takes for a nuclear wave packet to dissociate on the excited dissociative bo curve . temporal information about this process can be obtained using only one pulse , i.e. , without dressing of the system . in h@xmath0 the broad absorption features of @xmath1 were replaced by rich structures in the atas spectrum . for h@xmath0 we focused on explaining the oscillations in the absorption lines , and found that these could be put into two different categories ; one category was oscillations on the timescale of half the ir period , the other category was oscillations on a much longer timescale . we argued that the oscillations on the timescale of half the ir period were very similar in nature to the oscillations seen in the absorption lines of the atomic - like @xmath1 with fixed nuclei . we therefore conclude that these oscillations are due to interference between populated states of the same symmetry . the oscillations on longer timescales , however , are a purely molecular signature associated with transfer of population between vibrational states . finally we found that the existence of lis at well - defined energies in molecular atas spectra depends critically on the number of populated vibrational dark states . as a result clear , unambiguous detection of lis can only be expected in systems where the bo curves of the ground state and the involved dark state(s ) are nearly parallel , or if one in another way can populate only a few vibrational dark states . this work was supported by the erc - stg ( project no . 277767-tdmet ) , and the vkr center of excellence quscope . in this appendix we model the interference process between excited states of the same symmetry ( under @xmath87 ) resulting in a @xmath100 periodic modulation of the corresponding absorption lines in the atas spectrum . we focus on the interference fringes in the absorption line at @xmath288 ev in fig . [ fixed ] corresponding to an energy difference @xmath289 between the ground state and the fifth excited state . at large negative delays the population of the fifth excited state @xmath290 has two contributions for @xmath291 . a direct contribution from the xuv pulse alone , and an indirect contribution from the xuv and nir pulses acting in a combined manner . in the indirect process the xuv pulse populates the third excited state @xmath292 at time @xmath31 , and at a later time @xmath293 the nir pulse excites population to the fifth excited state through the absorption of two photons . keeping only terms that contribute to a dipole signal with frequency @xmath289 we can write up the wave function of the system as @xmath294 where @xmath295 is the ground state with energy @xmath296 , @xmath297 are the energies of @xmath292 and @xmath290 , @xmath298 and @xmath299 are the amplitudes corresponding to the direct and the indirect absorption processes , and the phase @xmath300 is given by @xmath301 this phase is similar to the phase found in ref . @xcite where constructive interference conditions are investigated . to obtain eq . we used that the population of the ground state is almost constant in time due to the low intensity of the xuv pulse . the time - dependent dipole moment in the simplified model is given by @xmath302 & ( \tau < t < t_0 ) \\ c_\text{d}\cos \left [ e_6 ( t-\tau ) \right ] + c_\text{id } \cos \left [ e_6 ( t-\tau ) -\varphi \right ] , & ( t > t_0 ) \end{cases } \label{simple_dipole}\end{aligned}\ ] ] [ b ] ( color online ) normalized atas spectrum @xmath30 [ eq . ] calculated for the fixed nuclei model of @xmath1 . in the upper panel the fourier transform of the time - dependent dipole moment is found from a full tdse calculation [ eq . ] , while in the lower panel eq . the pulse parameters are as in fig . [ fixed].,title="fig:",scaledwidth=48.0% ] in fig . [ fringes ] we compare the atas spectra constructed from dipole signals found from the full tdse calculation [ eq . ] and from eq . , respectively . following ref . @xcite we have chosen @xmath293 to be located a quarter of a nir period from the center of the nir pulse . the amplitudes @xmath298 and @xmath299 can be found using perturbation theory , but the choice @xmath303 suffices to illustrate the predictions of the model . as a result of the narrow spectral width of the nir pulse , interference of the type seen in eq . only occurs when the energy difference between two states of the same symmetry under @xmath87 is @xmath304 . consequently the associated oscillation periods are always @xmath93 with @xmath94 the period of a nir pulse cycle . we note that the interference phenomena of the type discussed here , always affect both absorption lines involved ( at @xmath305 and @xmath289 ) , and a similar pattern is found in the absorption line corresponding to the energy @xmath305 . for this absorption line , however , the phase of eq . is @xmath306 and the tilt of the fringes in the interference pattern will be opposite to that seen in fig . [ fringes ] . in sec . [ h2 ] we see that interference phenomena similar to the ones investigated here are also present between vibrational states in molecular systems , again leading to oscillations on the period of @xmath100 . 30ifxundefined [ 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 * * , ( ) in @noop _ _ ( , ) pp . in @noop _ _ ( , ) pp . link:\doibase 10.1103/physreva.53.2562 [ * * , ( ) ] @noop * * , ( ) link:\doibase 10.1103/physreva.65.013402 [ * * , ( ) ] @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) link:\doibase 10.1103/physreva.91.013414 [ * * , ( ) ] @noop * * , ( ) link:\doibase 10.1103/physreva.88.063420 [ * * , ( ) ] @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) link:\doibase 10.1103/physrevlett.109.163003 [ * * , ( ) ] @noop * * , ( ) @noop * * , ( ) @noop * * , ( )

we report the results of spectroscopic mapping observations carried out toward the herbig - haro objects hh711 and hh54 over the region using the infrared spectrograph of the _ spitzer space telescope_. these observations have led to the detection and mapping of the s(0 ) s(7 ) pure rotational lines of molecular hydrogen , together with emissions in fine structure transitions of ne , si , s , and fe . the h rotational emissions indicate the presence of warm gas with a mixture of temperatures in the range k consistent with the expected temperature behind nondissociative shocks of velocity while the fine structure emissions originate in faster shocks of velocity that are dissociative and ionizing . maps of the h line ratios reveal little spatial variation in the typical admixture of gas temperatures in the mapped regions , but show that the h ortho - to - para ratio is quite variable , typically falling substantially below the equilibrium value of 3 attained at the measured gas temperatures . the non - equilibrium ortho - to - para ratios are characteristic of temperatures as low as k , and are a remnant of an earlier epoch , before the gas temperature was elevated by the passage of a shock . correlations between the gas temperature and h ortho - to - para ratio show that ortho - to - para ratios are attained only at gas temperatures below k ; this behavior is consistent with theoretical models in which the conversion of para- to ortho - h behind the shock is driven by reactive collisions with atomic hydrogen , a process which possesses a substantial activation energy barrier ( k ) and is therefore very inefficient at low temperature . the lowest observed ortho - to - para ratios of only suggest that the shocks in hh54 and hh7 are propagating into cold clouds of temperature k in which the h ortho - to - para ratio is close to equilibrium .

astro-ph0606232

stars have a profound effect upon the interstellar medium ( ism ) that surrounds them . stellar radiation heats and sometimes photoionizes and photodissociates the interstellar gas , while supersonic outflows associated with stellar birth ( in protostellar outflows ) and death ( in supernova remnants ) send shock waves propagating through the ism . in regions of high mass star - formation such as the orion molecular cloud , for example many of these processes operate simultaneously , making their separate effects hard to disentangle and elucidate . in herbig - haro objects and supernova remnants , by contrast , the effects of shock waves can be studied in relative isolation in an environment where the stellar radiation field is not greatly enhanced . over more than three decades , the physics and chemistry of interstellar shock waves have been the subject of intensive study , both theoretical and observational , the results of which have been summarized in several review articles ( e.g. walmsley & pineau des forts 2005 ; draine & mckee 1993 ) . when shock waves propagate within molecular clouds , the shock velocity and magnetic field determine the fate of the gas . fast shocks ( with propagation velocities @xmath12 given interstellar magnetic fields of typical strength ) are dissociative : they destroy molecules and may ionize the resultant atoms . slow shocks ( @xmath13 ) , by contrast , are non - dissociative : they heat and compress the molecular gas through which they pass , and may modify its chemical composition , but molecules survive their passage . the chemical changes wrought by non - dissociative shocks typically result from elevated gas temperatures or the phenomenon of ion - neutral drift ( crucially important in non - dissociative shocks ) , which can drive chemical processes that are negligibly slow in cold clouds because they are endothermic or possess activation energy barriers . the result of these processes depends critically upon the timescale for key chemical reactions relative to the period of time for which shock elevates the temperature and ion - neutral drift velocities . in a previous study of hh54 carried out with the _ infrared space observatory ( iso ) _ , neufeld et al . ( 1998 ; hereafter nmh98 ) argued that the h@xmath2 ortho - to - para ratio , observed to lie significantly below its equilibrium value ( @xmath14 ) in the warm ( @xmath15k ) shocked gas in hh54 , provided a valuable probe of the relevant timescales . the infrared spectrograph ( irs ) on board the _ spitzer space telescope _ is a powerful tool for the study of shocked interstellar gas . with a wavelength coverage of 5.2 37 @xmath16 m , _ spitzer_/irs provides access to the s(0 ) through s(7 ) pure rotational lines of h@xmath2 , as well as fine structure emissions of several atoms and ions produced in fast , dissociative shocks : ne@xmath1 , si@xmath1 , s , and fe@xmath1 . the great sensitivity of _ spitzer_/irs , together with the multiplex advantage inherent in long slit spectroscopy , allows large regions to be mapped rapidly . in this paper we report the results of spectral line mapping in hh54 and hh@xmath17 , two well - studied herbig - haro objects known to contain warm molecular gas . hh 711 is a chain of herbig - haro objects in the perseus molecular cloud , associated with ngc 1333 ( snell & edwards 1981 ) . these objects lie in the blue - shifted part of a co outflow that originates near the protostar svs 13 . hh 711 has been studied extensively through observations of optical emission lines ( hartigan , curiel , & raymond 1989 ) , near - ir emission lines of [ fe ii ] ( stapelfeldt et al 1991 ; gredel 1996 ) , h@xmath2 vibrational emissions ( noriega - crespo et al . 2002 , khanzadyan et al . 2003 ) , far - infrared atomic and molecular lines ( molinari et al . 2000 ) ; and through high resolution studies of co emission ( bachiller et al . 2000 ) . one key result to emerge from previous studies is that the hot h@xmath2 emission exhibits a classic bow shaped morphology in hh 7 , with an emission peak offset from the peak emission of atomic fine structure lines . observations of the hh711 flow have proven valuable in constraining detailed models for shock excited gas ( e.g. hartigan , raymond , & hartmann 1987 ; smith , khanzadyan , & davis 2003 ) . the distance to hh7 - 11 is a matter of debate ; we adopt the value of 250 pc used by enoch et al . ( 2006 ; see discussion in that paper which refers to cernis 1993 ; cernis & straizys 2003 ; belikov et al . hh54 is located in the chamaeleon ii molecular cloud at an assumed distance of 200 pc ( hughes & hartigan 1992 ) , and is known to have a large column of warm ( @xmath18 2000 k ) shocked molecular and atomic gas ( schwarz & dopita 1980 ; gredel 1994 ) . while it is less well studied than hh7 - 11 , hh54 is notable because of an unusually large abundance of water vapor ( liseau et al . 1996 ) and h@xmath2 ortho - to - para ratios that are below the equilibrium value expected at the observed gas temperature . our observations and data reduction methods are discussed in 2 and appendix a , and the results are presented in 3 . the implications of these results for our understanding of interstellar shock waves are discussed in 4 . a brief summary follows in @xmath19 . one field in hh54 , and two fields in hh711 one centered on hh7 and one on the protostar svs13 were observed using the irs as part of the irac guaranteed time program . the short - low ( sl ) , short - high ( sh ) and long - high ( lh ) modules were used to provide complete spectral coverage over the wavelength range available to irs . maps covering fields of size @xmath20 were obtained by stepping the slit perpendicular , and in the case of sh and lh parallel to its long axis . the observational details are provided in table 1 . the data were processed at the spitzer science center ( ssc ) , using version 12 of the processing pipeline , and then reduced further using the smart software package ( higdon et al . 2004 ) , supplemented by additional routines that we have developed . the key capabilities provided by these additional routines were the removal of bad pixels in the lh and sh data , the calibration of fluxes obtained for extended sources , the extraction of individual spectra at each position sampled by the irs slit , and the creation of spectral line maps from the set of extracted spectra . details of our data reduction method , as well as the various tests we performed to validate the additional routines that we developed , are discussed in appendix a. widespread rotational emissions from molecular hydrogen , along with fine structure emissions from atomic sulfur and several singly - ionized species , were detected in each of the two sources we mapped . in figures 1 and 2 we present maps of the h@xmath2 line emissions detected in hh54 . the straight lines demark the regions within which each transition was mapped . the s(1 ) through s(7 ) lines all show morphological similarities , with most maps showing local maxima close to clumps e and k and to clump c ( in the designation of sandell et al . 1987 ) . however , real differences in the spatial distribution of the various transitions are apparent and provide key information about the spatial variation of the h@xmath2 excitation and ortho - to - para ratio ; these variations are discussed in detail in 4.2 and 4.4 below . emissions in several fine structure transitions have also been detected toward hh54 . figure 3 shows the distribution of the four fine structure emission lines that are strong enough to map : [ neii ] 12.8@xmath16 m , [ feii ] 26@xmath16 m , [ si ] 25@xmath16 m , and [ siii ] 35@xmath16 m . offsets between the fine structure emissions and the h@xmath2 emissions are clearly evident in figure 7 ( left panel ) , which shows contour plots that compare the distributions of the [ neii ] and h@xmath2 s(3 ) emissions , suggesting a different origin for the h@xmath2 and fine structure emissions , discussed further in 4.1 figures 4 and 5 show maps of the h@xmath2 line emissions detected in hh711 . as before , the straight lines demark the regions within which each transition was mapped ; in this case , two overlapping rectangular fields were mapped in the sl and sh modules and one field in the lh module . the dashed circles centered at offset ( 0,0 ) encircle the regions close to the bright continuum source svs13 where the line intensity estimates are unreliable . the s(1 ) , s(2 ) , s(3 ) and s(5 ) line maps are remarkably extended , with diffuse emission covering ( and extending up to 20@xmath21 on either side of ) the line joining hh7 to svs13 . a detailed discussion of the differences in the spatial distribution of the various transitions and of the inferred variations in the h@xmath2 excitation and ortho - to - para ratio is again deferred to 4.2 and 4.4 . as in hh54 , emissions in several fine structure transitions have also been detected in hh711 . figure 6 shows maps of the [ feii ] 26@xmath16 m , [ si ] 25@xmath16 m , and [ siii ] 35@xmath16 m line emissions detected with the lh module toward hh7 . unlike in hh54 , the [ neii ] 12.8 @xmath16 m line is not detected . figure 7 ( right panel ) shows that the fine structure emissions , as traced by [ siii ] in this case , are again offset from the h@xmath2 emissions , peaking several arcseconds closer to svs13 . we have averaged the observed spectra over several regions within the hh54 and hh711 fields , thereby obtaining substantial improvements in the signal - to - noise ratio . the resultant spectra , plotted in figure 8 , are the averages of all spectra within a series of circular regions , the contributing spectra being weighted by a gaussian taper ( hpbw of 15@xmath22 ) from the center of each synthesized beam . the position of each synthesized beam is demarked by the dotted circles in figure 7 , and the central positions are given in the caption to figure 8 . in hh54 , we obtained average spectra for circular regions centered on the peak of the [ neii ] emission and two peaks of the h@xmath2 emission . we denote the three circular beams hh54fs ( the peak of the [ neii ] emission ) , hh54c ( the northern of the two h@xmath2 peaks , being roughly coincident with clump c identified by sandell et al . 1987 ) , and hh54e+k ( the southern h@xmath2 peak , roughly coincident with clumps e and k ) . in hh711 , we obtained the average spectrum for a circular region centered on the hh7 bow shock . the spectra plotted in figure 8 all show very weak continua , leading to large line - to - continuum ratios even in the low - resolution module ( short - low , left panel , with @xmath23 of only @xmath2460 ) . a fairly rich spectrum of h@xmath2 rotational lines and low - excitation fine structure lines is apparent in each case . a total of eight h@xmath2 rotational lines and eight fine structure lines of ne@xmath1 , si@xmath1 , s , and fe@xmath1 ( 5 transitions ) have been detected at one or more such region . table 2 lists the detected lines together with the mean frequency - integrated intensities for each circular region . as discussed in 4.1 below , the h@xmath2 line ratios and intensities are typical of slow nondissociative shocks , while the fine structure line intensities are in good agreement with those expected from fast dissociative shocks . no fewer than five [ feii ] fine structure lines are clearly detected toward position hh54fs . the observed spectra are shown in figure 9 , and the transitions detected are marked on the grotrian diagram in figure 10 . the strength of the higher excitation transitions originating in the @xmath25 term , relative to the lower transitions within the @xmath26 term , is expected to be an increasing function of both temperature and density . as discussed in 4.1 , the [ feii ] line ratios provide a valuable probe of the physical conditions in the emitting region . upper limits are also obtained for two rotational transitions of water vapor that lie within the lh band and are discussed in 4.3 below . in addition , the r(3 ) and r(4 ) rotational lines of hd have been tentatively detected and discussed elsewhere ( neufeld et al . 2006 ) . in both sources that we observed , clear offsets between the fine - structure emissions and the molecular hydrogen emissions are evident . similar offsets have also been observed toward herbig - haro objects in the ggd37 ( cepheus a west ) region ( raines , 2000 ) . the fine structure emissions in hh7 are offset by @xmath27 from the bow shock , corresponding to a projected distance of @xmath28 for an assumed distance to the source of 250 pc . such an offset has been observed previously in hh7 ( staplefeldt et al . 1991 ) , where near - infrared emissions from [ feii ] and h@xmath2 have been mapped at subarcsecond resolution by noriega - crespo et al . ( 2002 ) using the _ hubble space telescope_. noriega - crespo et al . interpreted the [ feii ] emissions as emerging from a `` mach disk '' , in which the collimated supersonic outflow is decelerated and heated in a fast shock . in this picture , the fast shock is offset from a slower non - dissociative bow shock that is driven into the ambient molecular medium , the latter giving rise to the molecular hydrogen emissions observed from hh7 . our mid - infrared observations indicate that the [ siii ] and [ si ] fine structure emissions like the [ feii ] emissions also emerge from a region offset from the bow shock . the morphology of the [ siii ] map is most similar to the [ feii ] map . these observations support the two - shock picture in which the [ feii ] , [ si ] and [ siii ] fine structure emissions emerge from fast shock . our observations of hh54 indicate a similar offset ( @xmath29 , corresponding to a projected distance of @xmath30 for a source distance of 200 pc ) between the fine structure and molecular hydrogen emissions . as in hh7 , the centroid of the fine structure emissions lies closer to the presumed source of the outflow than the h@xmath2 emissions , consistent with the two - shock picture in which the fine structure emissions trace a fast shock that decelerates the collimated outflow . we have compared the observed fine structure line intensities listed in table 2 with predictions of the fast shock model of hollenbach & mckee ( 1989 , hereafter hm89 ) . because the beam filling factor of shocked material is not known , line ratios provide a more valuable diagnostic than absolute intensities . although the predictions of the hm89 shock models are undoubtedly somewhat uncertain given substantial uncertainties in the collisional excitation rates for the various transitions and in the gas - phase abundances of the various elements the predicted [ neii ] 12.8@xmath31m/[feii ] 26.0 @xmath31 m line ratio is a strongly increasing function of the shock velocity , values @xmath32 being needed to ionize ne significantly . the [ neii ] 12.8@xmath31m/[feii ] 26.0@xmath31 m line ratio of @xmath33 observed in hh54 requires a shock velocity @xmath34 . in hh7 , by contrast , the absence of detectable [ neii ] 12.8@xmath31 m emission places an upper limit @xmath35 , whilst our measurement of the [ feii ] 17.9@xmath31 m transition requires a shock velocity of at least @xmath36 . the hm89 shock models predict that the [ feii ] and [ siii ] line intensities show a very similar dependence upon the shock velocity , whilst the [ si]/[feii ] and [ si]/[siii ] line ratios are a decreasing function of shock velocity . this prediction provides a natural explanation for the fact that the [ siii ] map is most similar to the [ feii ] map : insofar as the fine structure emissions may arise from a superposition of shocks at varying velocities , the [ si ] emissions would be expected to show a spatial distribution that is somewhat different from [ siii ] and [ feii ] . an independent check of the two - shock picture is provided by our detection of multiple [ feii ] lines . the mimimum temperature derived from the [ feii ] line ratios ( @xmath37 k ) is larger than the maximum temperature at which h@xmath2 can survive for any appreciable time period and is a factor @xmath38 larger than that inferred from the h@xmath2 line ratios ( see 4.2 below ) lending strong support to the theory that both dissociative and nondissociative shocks are present in the source . the h@xmath2 s(0 ) s(7 ) emissions provide a valuable probe of the temperature distribution in the shocked gas in which h@xmath2 is the dominant constituent . the observed line intensities for these h@xmath2 quadrupole transitions provide a direct measurement of the column density , @xmath39 in each upper state , @xmath40 , because all the lines involved are optically - thin . as in nmh98 , we applied extinction corrections based upon the interstellar extinction curves of weingartner & draine ( 2001 ; milky way , @xmath41 " ) . for hh54 , we adopted gredel s ( 1994 ) estimate of 0.3 for @xmath42 , obtained from observations of the [ feii ] 1.257 and 1.644 @xmath16 m transitions toward hh54e+k , whilst for hh7 we adopted the estimate @xmath43 obtained by gredel ( 1996 ) for the nearby source hh8 . in figure 11 , based upon the line intensities tabulated in table 2 , we present rotational diagrams for each of the four circular regions for which we computed an average spectrum ( 3.2 ) . two features are common to all observed positions . first , as discussed further in 4.4 below , the rotational diagrams all show the zigzag " patterns characteristic of non - equilibrium ortho - to - para ratios , in which a fit to the para- ( even-@xmath44 ) states lies above that for the ortho- ( odd-@xmath44 states ) . second , the rotational diagrams all exhibit a degree of curvature , with the slope of the rotational diagram decreasing with increasing @xmath44 . this curvature indicates that each beam is sampling material at a range of temperatures . furthermore , the absence of any turnover in the curvature at high @xmath44 suggests that the rotational level populations are close to local thermodynamic equilibrium ( lte ) even for @xmath44 as high as 9 . for each circular region in table 2 , we determined an excitation temperature , @xmath45 , from the ratio of the @xmath46 and @xmath47 column densities and defined according to @xmath48}\ ] ] where @xmath49 and @xmath50 are the energy and degeneracy of state @xmath44 . because the ortho - para ratio shows departures from equilibrium , the quantity @xmath45 is most useful when @xmath51 and @xmath52 are either both odd or both even . the excitation temperatures @xmath53 , @xmath54 , @xmath55 , @xmath56 , @xmath57 and @xmath58 are given in table 3 ; the inequalities @xmath59 and @xmath60 observed in every case are a simple manifestation of the curvature apparent in the rotational diagrams . we have obtained two - component fits to the rotational diagrams , in which a least squares fit was obtained for a combination of warm and hot gas components , assumed to be in lte except with non - lte ortho - to - para ratios , with temperatures @xmath61 and @xmath62 ( typically found to be @xmath63 k and @xmath64 k , respectively ) . the solid lines in figure 11 show the two - component fits to the rotational diagrams , while the points indicate the measured values . the two - component fits provide an excellent fit to the data , although the solution is clearly non - unique . indeed , there is no specific evidence for a bimodal temperature distribution , and the data could be fit equally well by models that invoke a continuous range of gas temperatures . the assumption of lte is clearly valid for the lower rotational states , but the data would allow modest departures from lte for the highest states that we have observed . indeed , calculations by le bourlot et al . ( 1999 ) indicate that the critical density for @xmath65 ( at which the probabilities of collisional and radiative de - excitation are equal ) is @xmath66 , a factor of several larger than the densities inferred from observations of co in some of these sources ( see discussion below ) . departures from lte , if present , would be an increasing function of the rotational quantum number , @xmath44 , and would therefore require higher gas temperatures than the values inferred under the assumption of lte . the best - fit temperatures , computed under the assumption of lte , together with the beam - averaged column densities ( @xmath67 and @xmath68 ) and the h@xmath2 ortho - to - para ratios ( opr@xmath69 and opr@xmath70 ) , are listed in table 3 . table 3 also shows the sum of the @xmath71 though @xmath72 column densities , @xmath73 , all of which are derived directly from observations with the short - low module ; not surprising , @xmath73 falls a factor of several below @xmath74 because it only includes states with @xmath75 . the estimated gas temperatures , @xmath61 and @xmath62 , are characteristic of low - velocity nondissociative shocks ( which are of c "- type in the designation of draine ( 1980 ) ; see also appendix b ) . by means of a simple analytic treatment of energy conservation in such shocks ( appendix b ) , constrained by the results of the detailed shock models of kaufman & neufeld ( 1996 ; hereafter kn96 ) , we find the characteristic gas temperature within such shocks to be given by @xmath76 ( eqn . b8 ) , where @xmath77 is the shock velocity and @xmath78 is the assumed preshock magnetic field for a shock propagating in material of preshock h@xmath2 density @xmath79 . given our standard assumption about the magnetic field , @xmath80 , the warm and hot gas components typically correspond to shock velocities of @xmath81 and @xmath82 @xmath83 respectively . the h@xmath2 rotational diagrams predicted by kn96 and other theoretical shock models ( e.g. wilgenbus et al . 2000 ) all exhibit less curvature than our observed rotational diagrams , implying the presence of a range of shock velocities within each of the circular apertures demarked in figure 7 . the preshock h@xmath2 density , @xmath79 , is less well constrained by the h@xmath2 line fluxes . the best estimates for the density of the warm shocked gas have been obtained by observations of high - lying rotational lines of co , for which non - lte line ratios provide a valuable probe . based upon _ iso _ observations of such transitions toward hh54 , liseau et al . ( 1996 ) estimated the h@xmath2 density in the warm shocked gas as @xmath84 . since most of the high-@xmath44 co emission occurs before the compression of the neutral species becomes large , the corresponding preshock density is only a factor @xmath85 smaller ( appendix b , eqn . b6 ) : @xmath86 . similar values were obtained for hh711 , based upon measurements of co line ratios ( molinari et al . 2000 ) , although measurements of the co / h@xmath2 line ratios argued for smaller values ( @xmath87 ) toward hh7 . the observed column density of shocked h@xmath2 is predicted to be @xmath88 ( appendix b , eqn . b7 ) , where @xmath89 is a geometric factor defined to be the ratio of the projected beam size to the surface area of the shock . in principle , the quantity @xmath89 can be either larger or smaller than unity . for a planar shock with a beam covering factor of unity that is viewed at angle @xmath90 to the shock normal , the observed value of @xmath89 is @xmath91 . on the other hand , @xmath89 can be smaller than unity when the beam is incompletely covered by shocked material . for our two component fits to the circular apertures presented in table 3 , the values of @xmath89 required for the warm and hot components are typically @xmath92 and @xmath93 , respectively , implying that the shocked material has a covering factor considerably smaller than unity and possesses substructure that is not resolved by these observations . in hh7 , the subarcsecond resolution maps of near - ir h@xmath2 vibrational emissions presented by noreiga - crespo et al . ( 2002 ) clearly indicate that @xmath94 for observations with the irs instrument . the spectral line maps obtained for hh54 and hh7 also permit us to obtain a rotational diagram at each spatial position , and thereby to derive detailed maps of the gas temperature , ortho - to - para ratio , and column density of the shocked gas . while the average spectra presented in figure 8 are of sufficient signal - to - noise ratio ( snr ) to permit the two - component fits described above , the lower snr obtained in a single spectrum does not warrant a two - component fit to the rotational diagram . accordingly , in constructing maps of the gas temperature , we obtained a least - squares fit to the @xmath71 through @xmath95 column densities ( all measured with the sl module ) for a _ single _ gas component with a temperature and h@xmath2 ortho - to - para ratio , maps of which are presented in figure 12 ( middle and lower panels respectively ) . here , the white contours show the h@xmath2 s(3 ) line intensity , and regions of low snr have been masked and appear in black . the upper panel in figure 12 shows the sum of the @xmath71 though @xmath72 column densities , @xmath73 . figure 12 shows that the fitted gas temperature is remarkably constant over the mapped region , lying generally within the range @xmath96 k , and indicating that a similar admixture of shock velocities is present along each line - of - sight . the ortho - to - para ratio , however , shows substantial variations as noted previously by wilgenbus et al . ( 2001 ) and is discussed further in 4.4 the increase in temperature and density behind a shock front can alter significantly the composition of the gas in its wake . specifically , chemical reactions not favored at temperatures characteristic of the preshock quiescent material ( i.e. , @xmath9750 k ) can proceed rapidly in the warm post - shocked gas . one such example are the reactions that lead to the production of gas - phase h@xmath2o . in the cold preshock gas , cosmic - ray driven ion - neutral reactions are capable of converting o and h into water yielding a peak water abundance relative to h@xmath2 , @xmath98 , of @xmath2410@xmath99 in approximately 10@xmath100 years ( bergin et al . when the gas temperature exceeds @xmath24300 k , a series of neutral - neutral reactions , @xmath101 and @xmath102 , dominate and chemical models ( e.g. , elitzur & de jong 1978 ; neufeld et al . 1995 ) predict that most of the gas - phase oxygen not bound as co will be processed into water vapor . these reactions are rapid in warm gas , producing @xmath1032@xmath10410@xmath105 in less than a few hundred years ( bergin et al . 1998 ) , but are negligibly slow at low temperatures because they possess significant activation barriers . observations of both hh54 and hh7 have been conducted using esa s _ infrared space observatory ( iso)_. _ iso _ s long wavelength spectrometer ( lws ) was used in its grating mode to acquire low resolution ( @xmath106200 ) spectra between 43 and 197@xmath107 m within an 80@xmath22 diameter beam . toward hh54 , liseau _ _ ( 1996 ) detected a number of low - lying ( i.e. , @xmath109300 k above the ground state ) h@xmath2o and oh lines , along with co emission lines from the rotational upper level @xmath11014 to 19 . these data are best fit by a model that assumes collisional excitation behind a 10 - 15 km s@xmath111 nondissociative @xmath112-type shock and @xmath11310@xmath114 . toward hh7 , molinari et al . ( 2000 ) find evidence for the presence of both fast dissociative @xmath44-type and slower @xmath112-type shocks . in particular , molinari et al . find that a 15 - 20 km s@xmath111 @xmath112-type shock accounts well for the observed strengths of the co @xmath1101413 to @xmath1101716 transitions plus the h@xmath2 ( 0 - 0 ) s(0)s(5 ) transitions , which were measured with the _ iso _ short wavelength spectrometer . assuming that the _ iso_-detected h@xmath2o 179.5@xmath107 m 2@xmath1151@xmath116 and 174.6@xmath107 m 3@xmath1172@xmath118 lines arise predominately from behind the same @xmath112-type shock , molinari _ et@xmath108al . _ derive @xmath1196@xmath10410@xmath99 . molinari et al.suggest that the lower - than - expected post - shock h@xmath2o abundance is due to the rapid condensation of water onto dust grains in the post - shocked gas . a similar effect is observed toward ic@xmath108443 , in which a low water abundance ( 2@xmath10410@xmath1203@xmath10410@xmath121 ) is inferred . here we make the assumption that @xmath122 . ] in the presence of both @xmath112- and @xmath44-type shocks ( snell et al . snell et al . ascribe the low water abundance to a combination of two effects . first , more than half of the preshock oxygen not in co remains trapped as water - ice on grain surfaces despite the passage of a weak @xmath112-type shock , suppressing the production of gas - phase h@xmath2o in the post - shocked gas . second , ultraviolet radiation generated by the nearby fast @xmath44-type shock photodissociates much of the gas - phase water produced behind the @xmath112-type shock . eight h@xmath2o transitions with upper state energies @xmath123 k lie within the wavelength range accessible to _ spitzer_/irs . prior to the measurements reported here , none of these transitions had been observationally investigated toward either hh54 or hh7 . to assess the extent to which the upper limits to the h@xmath2o line strengths set here constrain the water abundance , the equilibrium level populations of all ortho and para rotational levels of the ground h@xmath124o vibrational state with energies _ e / k _ up to 7700 k have been calculated using an escape probability method described by neufeld & melnick ( 1991 ) . only two updates have been made in the calculations undertaken here : ( 1 ) for collisionally induced transitions among the lowest 45 levels of ortho- and para - water , the rate coefficients calculated by green , maluendes , & mclean ( 1993 ) for he - h@xmath2o collisions were used ( multiplied by 1.348 to account for the different reduced mass when h@xmath2 is the collision partner ) , and ( 2 ) the expression for the photon escape probability used by neufeld & kaufman ( 1993 ) was adopted . we obtain level populations for a plane - parallel emitting region in which the limit of large velocity gradient applies . a two - temperature component model is used , consistent with the results of our h@xmath2 measurements ( which are summarized in table 3 ) . in addition , preshock densities of 10@xmath100 and 10@xmath125 @xmath126 are considered . the h@xmath2 column densities and velocity gradients are assumed to be those consistent with the simple shock model presented in appendix b , i.e. : and where @xmath129 is the preshock gas density and @xmath130 is the gas temperature . finally , we assume that the fraction of the beam filled by the warm and hot h@xmath2o is given by the ratio of the measured values of @xmath131 ( given in table 3 ) to those derived above . for these conditions , the strongest water lines within the _ spitzer_/irs wavelength range are predicted to be the 29.837@xmath107m7@xmath132 - 6@xmath133 and 30.899@xmath107 m 6@xmath134 - 5@xmath135 ortho - h@xmath2otransitions . table 4 lists the _ spitzer_-measured 3@xmath136 upper limits to the line intensities for each transition along with the derived upper limits to the gas - phase water abundance . several things are apparent . first , the upper limits to the h@xmath2oabundance given in table 4 are higher than , but nonetheless consistent with the water abundances derived previously from the _ iso _ data . in particular , for a preshock density of 10@xmath125 @xmath126 favored by the h@xmath2 data toward hh54 ( liseau _ et@xmath108al._1996 ) the inferred abundance limits are less than would be expected from the full conversion of elemental oxygen not locked in co into h@xmath2o . unfortunately , the present 29.837 and 30.899@xmath107 m line flux limits are insufficient to allow us to comment further upon the previous hh54 models . second , the h@xmath2o 174 and 179@xmath107 m line fluxes measured by liseau _ et@xmath108al._provide an additional constraint on @xmath137 within the warm and hot components discussed here . in particular , the _ iso_-measured flux in the h@xmath2o 174@xmath107 m line , @xmath243@xmath10410@xmath138 w @xmath139 , requires that @xmath1402@xmath10410@xmath114 in the warm and hot h@xmath2 components or these components will produce a 174@xmath107 m line flux in excess of that measured . this result is in agreement with the water abundance determined by liseau et al . of 10@xmath114 and suggests that the size of the h@xmath2o 174 and 179@xmath107 m emitting region is larger than that inferred for the _ spitzer_-measured h@xmath2 emission ( or that assumed for the mid - infrared h@xmath2oemission ) . this is plausible since the 174 and 179@xmath107 m lines probe cooler gas than either the h@xmath2 or mid - infrared h@xmath2o lines . toward hh7 , molinari et al.estimate the preshock density to be @xmath2410@xmath100 @xmath126 . for this density , rather high h@xmath2o abundances are required to reproduce the 29.837 and 30.899@xmath107 m line intensity limits . however , the strength of the h@xmath2o 179@xmath107 m line observed by molinari et al.does set a more stringent upper limit of @xmath1406@xmath10410@xmath105 within the warm and hot components measured here . unfortunately , this limit remains significantly greater than the water abundance inferred from the _ iso _ measurements and , thus , does not permit a test of their model . the h@xmath2 ortho - to - para ratios ( table 3 ) derived for the two - component fits , opr@xmath69 and opr@xmath70 , are uniformly less than the local thermodynamic ratio , @xmath141 at the temperatures of relevance . this non - equilibrium behavior , first observed in the atmospheres of giant planets ( e.g. conrath & gierasch 1983 ) , was noted in _ iso _ observations of hh54 ( nmh98 ; wilgenbus et al . 2001 ) and subsequently observed in several other interstellar regions ( e.g. fuente et al . 1999 , rodrguez - fernndez et al . 2000 ) . as discussed in nmh98 , the h@xmath2 ortho - to - para ratio equilibrates very slowly , exhibiting memory effects " in which the opr can reflect the temperature of a previous epoch in its thermal history , reflecting the frozen - in " equilibrium value at room temperature , and greatly exceeding the value of 0.02 obtained at equilibrium at 21 k , the boiling point of lh2 . equilibration occurs over a timescale of 6.5 days , resulting in the eventual release of a total heat content that exceeds the latent heat of vaporization of lh2 : without continued refrigeration , the lh2 simply boils away ! the solution to this problem involves the introduction of a paramagnetic catalyst to facilitate ortho - to - para conversion during the original liquification process and thereby prevent the room temperature opr from being frozen - in . ] . thus nmh98 argued that the opr of @xmath142 observed toward hh54 indicated ( 1 ) that the warm ( @xmath143 k ) shocked h@xmath2 that they observed in hh54 must have previously acquired an opr characteristic of equilibrium at a temperature @xmath144 k and ( 2 ) that the observed gas has not been warm long enough for a high - temperature opr to have been established . the equilibration of ortho- and para - h@xmath2 has been considered in the detailed shock models of timmermann ( 1998 ) and wilgenbus et al . both studies concluded that reactive collisions with atomic hydrogen are the dominant para - to - ortho conversion process in non - dissociative molecular shocks : @xmath145 because reaction ( r1 ) has a substantial activitation energy barrier ( @xmath146 k ) , the efficiency of para - to - ortho conversion is a strong function of temperature ( and , consequently , shock velocity ) . this strong temperature dependence is entirely consistent with the results in table 3 , which show that for each region the hot gas component has a much higher opr than the warm gas component . this behavior is also apparent in the h@xmath2 rotational diagrams ( figure 11 ) , in which the departures from lte ( i.e. the degree of zigzag ) are plainly larger for small @xmath44 than for large @xmath44 . in figure 13 , the ortho - to para ratios , opr@xmath69 and opr@xmath70 , for the two - component fits in table 3 , are plotted as a function of the fitted gas temperatures , @xmath61 and @xmath62 ( open triangles ) . we have also obtained two - component fits to the gas temperature and opr elsewhere in the mapped regions . averaging the observed spectra in a set of @xmath147 subregions to attain adequate snr to yield meaningful two - component fits , we determined opr@xmath69 , opr@xmath70 , @xmath61 and @xmath62 estimates for each such subregion . green and red squares in figure 13 indicate respectively the ( @xmath148opr@xmath69 ) and ( @xmath62 , opr@xmath70 ) values obtained in subregions within the hh54 and hh7 maps . green and red crosses apply to the discontiguous regions northwest of hh7 ( figure 12 ; these are close to hh8 ) . the orange curve denotes the ortho - to - para ratio expected in lte . the green and red symbols in figure 13 confirm that the ortho - to - para ratio tends to increase with gas temperature . the black curves show the expected behavior for a simple model in which all the observed gas started with the same initial ortho - to - para ratio opr@xmath149 ( corresponding to equilibrium at temperature 55 k ) for hh54 and opr@xmath150 ( corresponding to equilibrium at temperature 48 k ) for hh7 and has been at the currently - observed temperature @xmath151 for the same time period , @xmath152 . in that case , the current opr is given by @xmath153 where @xmath154 is the atomic hydrogen density and @xmath155 is the rate coefficient for para - to - ortho conversion , estimated as @xmath156 by schofield et al . ( 1967 ) , based upon the laboratory experiments of schulz & le roy ( 1965 ) . the black curve applies to the case @xmath157 yr , where @xmath158 . if , following nmh98 , we assume that h atoms are produced primarily by reaction of o with h@xmath2 to form oh , followed by reactions of oh with h@xmath2 to form water , we expect an @xmath159 ratio @xmath160 , corresponding to @xmath161 for a preshock h@xmath2 density @xmath162 . the steep rise in the black curve at @xmath163 k is a consequence of the strong temperature - dependence of @xmath155 . to within the likely errors , all the points lie between the black and orange curves in figure 13 . this behavior then implies that all the shocked gas had an initial opr between 0.4 and 3 and has been warm for at least @xmath164 yr . the foregoing analysis is premised on the assumption that the rotational temperature inferred for the hot gas component is equal to the kinetic temperature . as discussed in 4.2 , above , subthermal excitation of the s(7 ) transition is allowed by the data and would mean that the actual gas temperature is higher than the rotational temperature for the hot component , @xmath62 . such effects , in turn , would decrease the required timescale for ortho - para conversion . for a postshock atomic h abundance @xmath160 , the timescale of @xmath164 yr inferred above is broadly consistent with the time period for which gas temperature is elevated upon passing through a nondissociative shock . for @xmath162 and @xmath165 , the simple treatment of appendix b implies a shock column density , @xmath166 , corresponding to a timescale @xmath167 yr . however , the opr values plotted in figure 13 are entirely inconsistent with a shock model with a large h abundance , as has been invoked by smith et al . ( 2003 ) , who proposed an @xmath159 ratio of 3 for the bow shock in hh7 . such a large h abundance would lead to very rapid para - to - ortho conversion , producing opr values significantly in excess of those observed here . we may also compare the results plotted in figure 13 with the predictions of detailed theoretical models for para - to - ortho conversion in non - dissociative molecular shocks . the cyan loci in figure 13 represent the predictions of wilgenbus et al . ( 2000 ) for a shock propagating in gas with an initial opr of 0.01 . the loci correspond to preshock h@xmath2 densities of @xmath168 , @xmath169 , 10@xmath125 and @xmath170 , respectively from left to right , and the asterisks apply to shock velocities of 30 and 40 @xmath83 ( @xmath171 case , from lower left to upper right ) ; 20 , 30 , and 40 @xmath83 ( @xmath169 and @xmath172 cases , from lower left to upper right ) ; or 10 , 15 , 20 , 25 , and 30 @xmath83 ( @xmath173 case , again from lower left to upper right ) . as in the case of the toy model ( black curve ) , the curves for 10@xmath125 or @xmath173 are a reasonable approximation to the lower envelope of the observed data points . the absence of sight - lines with high temperature and low opr is consistent with shock models in which para - to - ortho conversion is driven by reactive collisions with atomic hydrogen . on the other hand , the presence of sight - lines with _ low _ temperature and _ high _ opr suggests that some of the shocked material has a high _ initial _ opr . such material is clearly present for the sight - lines represented by the crosses in figure 13 . these regions lie behind the working surface " of the jet in hh711 ; they may , perhaps , have been heated by relatively fast shocks during a previous epoch of shock activity , thereby acquiring a larger opr , and are currently being reheated by a slower shock . the bow shock in hh7 , by contrast , appears to be propagating into fresh material in which the opr is @xmath174 , corresponding to equilibrium at temperature @xmath6 k. we have carried out spectroscopic observations toward the herbig - haro objects hh711 and hh54 over the @xmath0 region using the infrared spectrograph of the _ spitzer space telescope _ , and have thereby mapped the s(0 ) s(7 ) pure rotational lines of molecular hydrogen , together with emissions in fine structure transitions of ne@xmath1 , si@xmath1 , s , and fe@xmath1 . 1 . the h@xmath2 rotational emissions indicate the presence of warm gas with a mixture of temperatures in the range @xmath3 k , consistent with the expected temperature behind nondissociative shocks of velocity @xmath4 . 2 . the fine structure emissions are spatially offset from the h@xmath2 emissions , lying closer to the source of the outflow that drives the shocks in these sources . they originate in faster shocks of velocity @xmath5 that are dissociative and ionizing . such shocks likely result from the deceleration of the outflow , and are offset from a slower non - dissociative bow shock that is driven into the ambient molecular medium . maps of the h@xmath2 line ratios reveal little spatial variation in the typical admixture of gas temperatures in the mapped regions , but show that the h@xmath2 ortho - to - para ratio is quite variable , typically falling substantially below the equilibrium value of 3 attained at the measured gas temperatures . 4 . the non - equilibrium ortho - to - para ratios are characteristic of temperatures as low as @xmath6 k , and are a remnant of an earlier epoch , before the gas temperature was elevated by the passage of a shock . correlations between the gas temperature and h@xmath2 ortho - to - para ratio show that ortho - to - para ratios @xmath7 are attained only at gas temperatures below @xmath8 k ; this behavior is consistent with theoretical models in which the conversion of para- to ortho - h@xmath2 behind the shock is driven by reactive collisions with atomic hydrogen , a process which possesses a substantial activation energy barrier ( @xmath9 k ) and is therefore very inefficient at low temperature . the lowest observed ortho - to - para ratios of only @xmath10 suggest that the shocks in hh54 and hh7 are propagating into cold clouds of temperature @xmath11 k in which the h@xmath2 ortho - to - para ratio is close to equilibrium . upper limits on the 29.837@xmath107m7@xmath132 - 6@xmath133 and 30.899@xmath107 m 6@xmath134 - 5@xmath135 transitions of ortho - h@xmath2o place upper limits on the h@xmath2oabundance that are higher than and therefore consistent with abundances derived previously from _ iso _ observations of transitions at longer wavelength . this work is based on observations made with the _ spitzer space telescope _ , which is operated by the jet propulsion laboratory , california institute of technology , under a nasa contract . d.a.n . and p.s . gratefully acknowledge the support of grant nag5 - 13114 from nasa s long term space astrophysics ( ltsa ) research program . \1 ) _ removal of bad pixels in the basic calibrated data ( bcd ) " files that are generated by the spitzer science center for the short - high and long - high modules . _ bad pixels were defined as those flagged in either the ssc bmask " file or in a grand rogue mask " generated at university of rochester from an analysis of dark fields acquired at the start of each irs campaign . as a rogue might not appear in every campaign dark field , we assumed that any pixel identified as a rogue in any campaign would be corrected for all campaigns ; thus , the grand rogue mask contains 25@xmath175 more rogue pixels than a single campaign rogue mask . they were removed with the imclean " routine , which interpolates between nearby good pixels . \2 ) _ extraction of the individual spectra at each position along the slit . _ these spectral extractions were accomplished using an augmentation to the smart software package that has been developed at university of rochester . this augmentation generates a spectrum for each resolution element along the slit , resulting in a series of spectra at defined positions in r.a . and the spectra were trimmed to eliminate overlap between different orders of the irs . \3 ) _ regridding of the spectra onto a grid that is regular in r.a . and declination . _ because there are positions that are multiply observed with more than one slit position , this regridding sometimes involves the averaging of spectra obtained at nearly identical positions . \4 ) _ application of the slit loss correction function ( slcf ) . _ because the ssc - provided data are flux - calibrated using point source calibrators , a downwards correction must be provided to obtain intensities for extended sources , the correction factor ( i.e. the slcf " ) being the fraction of the point source flux that falls within the slit width . we used the slcfs kindly provided by j. d. smith from an analysis of the instrumental point spread function . \5 ) _ determination of line and continuum fluxes at each spatial position within the mapped region . _ here , we used a levenberg - marquardt ( l - m ) fitting routine simultaneously to fit a gaussian line and second order polynomial baseline at each position . line fits were obtained for a series of spectral lines identified in the co - averaged spectra for each field . in fitting the co - averaged spectra , which are of high signal - to - noise ratio ( snr ) , the line centroids and widths were treated as free parameters to be optimized in the l - m fit . the resultant best - fit centroids and widths were then used as _ fixed _ parameters in the l - m fits to the lower snr spectra obtained at individual spatial positions ; for these , only the line intensities and continuum parameters were optimized . this procedure is justified by the fact that the expected doppler shifts associated with gas motions in the source are very small compared to the spectral resolution , @xmath23 being only @xmath176 and @xmath177 respectively for the low ( sl ) and high ( sh and lh ) resolution modules . the data reduction procedure was accomplished using the smart software package and a set of associated idl procedures that we have written . the contour maps presented in figures 1 7 were also created using idl procedures . \1 ) _ continuum mapping of svs13 . _ figure 14 shows contour maps obtained for several continuum wavelengths for hh711 field 1 ( see table 1 ) , which contains the strong continuum source svs13 at its center ( i.e. at position ( 0,0 ) in figure 14 ) . the intensity maps shown in figure 14 use logarithmic contours spaced by a factor @xmath178 , with the second - highest contour corresponding to an intensity one - half the maximum . to the extent that svs13 can be regarded as an unresolved point source at these wavelengths , the maps in figure 14 represent the beam profile resulting at the end of the entire data reduction procedure . these maps allow the astrometric accuracy and the half - power - beam width ( hpbw ) to be estimated for the final data product . table 5 shows the results of gaussian fits to the spatial profiles , and indicates that the astrometric accuracy is better than @xmath179 in the final data product . the map at 5.8 @xmath16 m clearly exhibits a sixfold rotational symmetry at intensities levels less than @xmath180 of the maximum , a behavior that is also evident in infrared array camera ( irac ) observations ( hora et al . 2004 ) and reflects the diffraction pattern produced by the telescope . \2 ) _ spectral extraction of svs13 and comparison with _ infrared space observatory _ ( iso ) observations . _ as a check of the flux calibration , we have extracted spectra of the strong continuum source svs13 , obtaining a co - average from the entire map of hh711 field 1 . for comparison , we have obtained an archival spectrum of svs13 from the _ iso _ short wavelength spectrometer ( sws ) , co - averaging multiple scans and removing bad pixels with the use of the isap software package . figure 15 compares the spectrum obtained with _ ( heavy black line ) with that obtained with _ ( red , blue , green and magenta lines , different colors being used to show the different sws orders ) . overall , the agreement is very good , with the absolute fluxes typically agreeing to within @xmath181 . in the spectral region longward of 27.5 @xmath16 m , the fluxes obtained with _ iso_/sws exceed those measured by _ spitzer _ by as much as @xmath182 . however , a significant order mismatch at 27.5 @xmath16 m ( between detectors 11 and 13 ) in the _ iso _ spectrum suggests that the _ iso_-determined flux is an overestimate longward of 27.5 @xmath16 m . here we present a simple analytic treatment of nondissociative c - type shocks , which is limited in its applicability to steady - state shocks with plane parallel geometry . detailed models of planar nondissociative c - type shocks ( e.g. kaufman & neufeld 1996 , hereafter kn96 ; wilgenbus et al . 2000 ) , predict line strengths for the h@xmath2 pure rotational transitions that yield rotational diagrams with remarkably little curvature . this behavior suggests that the warm gas behind such shocks can be approximated as an isothermal , isobaric slab . in this appendix , we show that such a description yields predictions for the emission line spectrum of the shocked gas that are in good agreement with those of detailed models , and we derive analytic expressions for the appropriate slab thickness , density , temperature and velocity gradient as a function of the preshock h@xmath2 density , @xmath183 , shock velocity , @xmath184 , and preshock magnetic field , @xmath185 . nondissociative c - type shocks have been discussed extensively in the literature ( e.g. draine & mckee 1993 , and references therein ) and will not be reviewed here . they were originally defined by draine ( 1980 ) , who distinguished between j- ( or jump")-type shocks in which the fluid velocity changes suddenly and the neutral and ionized species share a common velocity , and c- or ( continuous")-type shocks in which the fluid velocity varies slowly . the basic physics of c - type shocks had been discussed previously by mullan ( 1971 ) , who recognized ion - neutral drift to be an essential feature of magnetic shocks propagating in weakly - ionized media . ions entering the shock , being strongly coupled to the magnetic field , are decelerated fairly rapidly and quickly reach their maximum compression . neutral species entering the shock , by contrast , are uncoupled to the magnetic field and are slowly decelerated in collisions with the ions through which they drift . thus the maximum heating rate within the shocked material occurs where the compression of the ions is large and that of the neutral species is small . accordingly , we make the following simplifying approximations : \1 ) the neutral fluid is at a constant compression , @xmath186 , which is slightly greater than unity ; the exact value is an adjustable parameter to be determined below by a comparison with the line flux predictions of the kn96 shock models . \2 ) the ionized fluid is at a constant compression equal to the maximum that is attained behind the shock @xmath187 where @xmath188 is the alfven mach number and @xmath189 is the preshock ion density . additional ionization within the shock ( self - ionization " ) is neglected . \3 ) the thickness of emitting slab is given by @xmath190 where @xmath191 , v_d)$ ] is the ion - neutral coupling length ( kn96 ) for h@xmath2 density @xmath192 and ion - neutral drift velocity @xmath193 , and @xmath194 is a dimensionless constant of order unity , the exact value of which is an adjustable parameter to be determined below by a comparison with the line flux predictions of the kn96 shock models . \4 ) the slab is at constant temperature , @xmath151 , determined such that the radiative output is equal to the mechanical luminosity , implying @xmath195^{-2 } ) \sim \onehalf \mu n_0 v_s^3 , \eqno(b3)\ ] ] where @xmath196)$ ] is the cooling rate coefficient for temperature , @xmath197k , and h@xmath2 density , @xmath198 . \5 ) the velocity gradient is @xmath199 where @xmath200 is a dimensionless constant of order unity , the exact value of which is an adjustable parameter to be determined below by a comparison with the line flux predictions of the kn96 shock models . \6 ) we adopt as our standard assumptions an ion - neutral coupling length @xmath201 based upon a fit to results of kn96 ( obtained for the case where polycyclic aromatic hydrocarbons ( pahs ) are assumed to be present ) , and a cooling rate coefficient @xmath202 ) = 1.4 \times 10^{-26 } n_5^{-0.5 } t_3^{2.8}\,\rm erg \ , cm^3 \,s^{-1 } \eqno(b5)\ ] ] based upon a fit to the results of neufeld et al . the superscripts on @xmath203 and @xmath204 are used to denote the standard values " adopted for our treatment . assumptions ( 1 ) ( 6 ) above define the required properties of the isothermal , isobaric slab , once the three parameters @xmath205 , @xmath194 , and @xmath200 have been specified . in determining the best - fit values for these parameters , we have compared the emission predicted for the slab model with the detailed predictions of kn96 . here we assumed fixed co and h@xmath2o abundances , corresponding to the assumption that co and h@xmath2o account for all the o and c nuclei in the gas - phase . because the o and c abundances assumed by kn96 were somewhat larger than those adopted by neufeld et al . ( 1995 ) in computing the cooling rate coefficient , @xmath206)$ ] , we adopted @xmath207 ) = 1.5 \ , \lambda^0(t , n[{\rm h}_2])$ ] for the purposes of this comparison . the optimum fit to the kn96-predicted emission line spectrum is obtained for @xmath208 , @xmath209 , and @xmath210 . combining the expressions given above , we find that the slab h@xmath2 density is fit best by @xmath211 the h@xmath2 column density by @xmath212 the gas temperature by @xmath213 and the velocity gradient by @xmath214 in figure 16 , we compare the h@xmath2 s(5)/s(3 ) line ratio and the total h@xmath2 , h@xmath2o and co surface luminosities derived for the simple slab model ( red ) with those obtained in the detailed calculations of kn96 ( blue ) . in general , the agreement is remarkably good . almost invariably , either ( a ) the shock velocity corresponding to a given flux or flux ratio agrees to within @xmath215 ; and/or ( b ) the predicted line flux or flux ratio for a given shock velocity agrees to within a factor 2 . often , the agreement is considerably better . the main exception is for water emissions at low shock velocity ( @xmath216 ) , where the detailed kn96 models predict considerably less h@xmath2o than the slab models . this behavior results because for the slab models we assumed that water vapor accounts for all gas - phase oxygen nuclei not locked in co , whereas the treatment of oxygen chemistry in kn96 predicts considerably lower water abundances at low shock velocity . llccccc hh711 field 1 & 2004 feb 15 & sl & ... & 3h 29 m 3.71s & + 31d 16@xmath217 01.9@xmath218@xmath218 & @xmath219 + hh711 field 1 & 2004 feb 15 & sh & ... & 3h 29 m 3.66s & + 31d 16@xmath217 03.6@xmath218@xmath218 & @xmath220 + hh711 field 1 & 2004 feb 15 & lh & 10243@xmath221 & 3h 29 m 3.66s & + 31d 16@xmath217 03.6@xmath218@xmath218 & @xmath222 + hh711 field 2 & 2004 feb 16 & sl & ... & 3h 29 m 8.61s & + 31d 15@xmath217 26.6@xmath218@xmath218 & @xmath219 + hh711 field 2 & 2004 feb 16 & sh & ... & 3h 29 m 8.56s & + 31d 15@xmath217 28.3@xmath218@xmath218 & @xmath220 + hh711 field 2 & 2004 feb 16 & lh & 10246@xmath221 & 3h 29 m 8.56s & + 31d 15@xmath217 28.3@xmath218@xmath218 & @xmath222 + hh54 & 2004 apr 7 & sl & 1133 & 12h 55 m 51.52s & 76d 56@xmath217 19.1@xmath218@xmath218 & @xmath223 + hh54 & 2004 apr 7 & sh & 3260 & 12h 55 m 50.87s & 76d 56@xmath217 19.3@xmath218@xmath218 & @xmath224 + hh54 & 2004 apr 7 & lh & 931 & 12h 55 m 50.82s & 76d 56@xmath217 18.0@xmath218@xmath218 & @xmath225 + lrcccc h@xmath2 s(0 ) @xmath226 & 28.2188 & 0.52 & 0.44 & 0.47 & 0.78 + h@xmath2 s(1 ) @xmath227 & 17.0348 & 3.68 & 3.18 & 3.99 & 3.07 + h@xmath2 s(2 ) @xmath228 & 12.2786 & 14.1 & 21.4 & 17.2 & 18.2 + h@xmath2 s(3 ) @xmath229 & 9.6649 & 24.7 & 20.7 & 29.5 & 20.6 + h@xmath2 s(4 ) @xmath230 & 8.0251 & 24.5 & 28.9 & 22.9 & 25.6 + h@xmath2 s(5 ) @xmath231 & 6.9095 & 33.0 & 24.4 & 32.9 & 29.0 + h@xmath2 s(6 ) @xmath232 & 6.1086 & 17.0 & 14.0 & 14.3 & 12.0 + h@xmath2 s(7 ) @xmath233 & 5.5112 & 15.6 & 12.7 & 17.2 & 16.0 + ne@xmath1 @xmath234 & 12.8149 & 1.68 & 1.03 & 0.82 & @xmath235 + s @xmath236 & 25.2490 & 1.75 & 0.75 & 0.75 & 1.73 + si@xmath1 @xmath237 & 34.8141 & 6.07 & 2.95 & 3.43 & 4.18 + fe@xmath1 @xmath238 & 5.3403 & @xmath239 & @xmath240 & @xmath241 & + fe@xmath1 @xmath242 & 17.9363 & @xmath243 & @xmath244 & @xmath245 & @xmath246 + fe@xmath1 @xmath247 & 24.5186 & @xmath248 & @xmath249 & @xmath250 & @xmath251 + fe@xmath1 @xmath252 & 25.9882 & @xmath253 & @xmath254 & @xmath255 & @xmath254 + fe@xmath1 @xmath256 & 35.3491 & @xmath257 & @xmath258 & @xmath259 & @xmath260 + lcccc rotational state & + j=2 & 19.51 & 19.44 & 19.47 & 19.70 + j=3 & 18.94 & 18.88 & 18.98 & 18.89 + j=4 & 18.62 & 18.81 & 18.71 & 18.77 + j=5 & 18.25 & 18.17 & 18.33 & 18.26 + j=6 & 17.70 & 17.77 & 17.67 & 17.75 + j=7 & 17.40 & 17.27 & 17.40 & 17.36 + j=8 & 16.78 & 16.69 & 16.70 & 16.65 + j=9 & 16.45 & 16.36 & 16.49 & 16.48 + + @xmath53 & 445 & 571 & 501 & 428 + @xmath54 & 729 & 717 & 764 & 779 + @xmath55 & 716 & 650 & 647 & 658 + @xmath56 & 919 & 871 & 852 & 877 + @xmath57 & 986 & 857 & 946 & 841 + @xmath58 & 1076 & 1120 & 1120 & 1156 + + @xmath61 ( k ) & 382 & 558 & 470 & 422 + log@xmath261/@xmath262 & 19.89 & 19.80 & 19.84 & 19.99 + opr@xmath69 & 0.43 & 0.41 & 0.48 & 0.21 + + @xmath62 ( k ) & 1025 & 1150 & 1062 & 1029 + log@xmath263/@xmath262 & 19.22 & 18.86 & 19.13 & 19.17 + opr@xmath70 & 1.53 & 1.55 & 1.98 & 1.88 + log@xmath264)/@xmath139 ) & 19.97 & 19.84 & 19.92 & 20.05 + @xmath73 ( @xmath139 ) & 18.83 & 18.88 & 18.91 & 18.93 +

we present high resolution n - body / sph simulations of the interacting cluster 1e0657 . the main and the sub - cluster are modeled using extended cuspy dark matter halos and isothermal-profiles for the collisional component . the hot gas is initially in hydrostatic equilibrium inside the global potential of the clusters . we investigate the x - ray morphology and derive the most likely impact parameters , mass ratios and initial relative velocities . we find that the observed displacement between the x - ray peaks and the associated mass distribution , the morphology of the bow shock , the surface brightness and projected temperature profiles across the shock discontinuity can be well reproduced by offset 1:6 encounters where the sub - cluster has initial velocity ( in the rest frame of the main cluster ) close to 2 times the virial velocity of the main cluster dark matter halo . a model with the same mass ratio and lower velocity ( times the main cluster virial velocity ) matches quite well most of the observations . however , it does not reproduce the morphology of the main cluster peak . dynamical friction strongly affects the kinematics of the sub - cluster so that the low velocity bullet is actually bound to the main system at the end of the simulation . we find that a relatively high concentration ( c=6 ) of the main cluster dark matter halo is necessary in order to prevent the disruption of the associated x - ray peak . for a selected sub - sample of runs we perform a detailed three dimensional analysis following the past , present and future evolution of the interacting systems . in particular , we investigate the kinematics of the gas and dark matter components as well as the changes in the density profiles and the motion of the system in the diagram . [ firstpage ] methods : n - body simulations galaxies : clusters : individual : 1e0657 - 56 dark matter x - rays : galaxies : clusters

0711.0967

the `` bullet '' cluster 1e0657 - 56 represents one of the most complex and unusual large - scale structures ever observed . located at a redshift @xmath5 it has the highest x - ray luminosity and temperature of all known clusters as a result of overheating due to a recent supersonic mach @xmath6 @xcite central encounter of a sub - cluster ( the bullet ) with its main cluster . the 500 ks chandra acis - i image of 1e0657 - 56 ( fig . 1 in markevitch 2006 ) shows two plasma concentrations with the bullet sub - cluster on the right of the image being deformed in a classical bow shock on the western side as a result of its motion through the hot gas of the main cluster . the analysis of the shock structure leads to the conclusion that the bullet is now moving away from the main cluster with a velocity of @xmath7 km s@xmath8 @xcite . the line - of - sight velocity difference between the two systems is only 600 km s@xmath8 suggesting that the encounter is seen nearly in the plane of the sky @xcite . as the core passage must have occurred @xmath9 gyr ago we have the unique opportunity to study this interaction in a very special short - lived stage , far away from thermal and dynamical equilibrium . as a result of the encounter , the collision - dominated hot plasma and the collisionless stellar and dark matter components have been separated . the galaxy components of both clusters are clearly offset from the associated x - ray emitting cluster gas @xcite . in addition , weak and strong lensing maps @xcite show that the gravitational potential does not trace the distribution of the hot cluster gas that dominates the baryonic mass but follows approximately the galaxy distribution as expected for a collisionless dark matter component . the likelihood to find such a high velocity cluster encounter in a @xmath10cdm cosmology has recently been investigated by @xcite using the millenium run simulation . according to the newest estimates from x - ray and gravitational lensing results the @xcite likelihood becomes @xmath11 @xcite which means that 1e0657 - 56 represents an extremely rare system in a @xmath10cdm universe . recent numerical works @xcite have demonstrated that the relative velocity of the dark matter components associated to the main and the sub - cluster are not necessarily coincident with the bullet velocity inferred from the shock analysis . in details , @xcite using a 2-d eulerian code well reproduced the observed increase in temperature across the shock front with a dark halo velocity @xmath12 lower than that of the shock , while @xcite found even a larger difference between the shock velocity ( @xmath13 km s@xmath8 in their best model ) and the speed of the halo ( only @xmath14 km s@xmath8 ) . moreover , according to @xcite , due to a drop in ram - pressure after the cores interaction the gas component of the sub - cluster can eventually be larger than that of its dark matter counterpart . [ cols="<,^,^,^,^,^,^,^,^,^ " , ] [ runsresults ] the main cluster is initially at rest and the sub - cluster moves in the @xmath15 direction with a velocity which ranges from 2000 to 5000 km s@xmath8 . the initial conditions of the different runs are summarized in table [ runs ] . the velocities of the sub - cluster relative to the center of mass of the system are listed in the last but one column of the table . all the simulations were carried out using gasoline , a parallel sph tree - code with multi - stepping @xcite . most of the runs are adiabatic , with @xmath16 . radiative cooling for a primordial mixture of hydrogen and helium in collisional equilibrium is implemented in 1:6c and 1:3lfgc . the consequences of assuming a lower concentration for the main dark halo are investigated in run 1:6c4 , where @xmath17 . the main cluster is modeled with @xmath18 particles , @xmath19 sph and @xmath20 collisionless . the sub - cluster , with the exception of run 1:8 ( where the number of gas particles in the bullet is @xmath21 ) , has @xmath22 particles , @xmath23 collisional and the remainder dark matter particles . the gravitational spline softening is set equal to 5 kpc for the gaseous and dark component . each numerical work which aims to simulate the bullet cluster should be able to reproduce simultaneously the main features observed in x - ray maps ( the bow shock , the relative surface brightness of the bullet and the main cluster ) , and the observed surface brightness and temperature profiles across the shock discontinuity . an additional constraint is provided by the observed displacement between x - ray and lensing maps , which is not negligible in both the main and the sub - cluster ( @xmath24 and @xmath25 kpc respectively , according to clowe et al . 2006 ) . a first indication about the validity of a model arises from the qualitative comparison of our simulated x - ray surface brightness maps with the x - ray 500 ks _ acis - i images provided by @xcite . the impact parameter @xmath26 is not strictly constrained by observations . nevertheless a head - on merger , with @xmath27 , seems to be excluded by comparing deep x - ray observations and weak lensing maps . in particular in the top panel of fig . 1 which is fig . 1b of @xcite the brightest gas associated with the main cluster is not located along the line which connects the centers of the two total mass distributions . moreover , the x - ray emission from the main cluster is asymmetric , with a peak in the north of the image and an extended tail of less bright material pointing south . these features are hardly associable with a zero impact parameter interaction , as shown in the middle and bottom panels of fig . [ b=0 ] which illustrate two simulated 1:6 head - on encounters where the sub - cluster moves from the left to the right of the image ( @xmath15 axis of the simulation ) with decreasing initial velocities ( the middle panel corresponds to run 1:6vb0 of table [ runs ] , the bottom one to 1:6v3000b0 ) . images are projected along an axis perpendicular to the collision plane ( the encounter is seen face - on ) and the selected snapshot is the one which most closely matches the observed distance between the centers of the total mass distributions , associated with the two clusters about 720 kpc from @xcite once the sub - cluster has passed through the core of the main system ( hereafter the present time ) . in the lower two panels of fig . [ b=0 ] colours represent x - ray maps in the _ chandra _ energy band ( 0.8 - 4 kev ) generated using the theoretical image processing system ( tipsy ) , which produces projected x - ray surface brightness maps with the appropriate variable sph kernel applied individually to the flux represented by each particle . assuming complete ionization and zero metallicity ( metal lines are expected to provide a significant contribution to emissivity only at relatively low temperatures , less than 2 kev ) the x - ray luminosity in a given energy band is defined as @xcite @xmath28 where @xmath29 is the cooling function in the specific band , @xmath30 , @xmath31 and @xmath32 are temperature , density and mass associated with the @xmath33 hot ( @xmath34k ) gas particle , respectively , @xmath35 is the proton mass and @xmath36= 0.6 the mean molecular weight . the sum runs over all the @xmath37 particles within an oblong of base equal to the pixel size and major axis oriented along the line of sight . when the x - ray luminosity of the entire cluster is calculated @xmath37 is the number of hot particles within the virial radius @xmath38 . the cooling function is computed using a raymond - smith code @xcite for a gas of primordial composition . the energy band ( 0.8 - 4 kev ) is chosen in such a way to reproduce the 500 ks _ acis - i image of the bullet cluster @xcite . the entire energy band , used to calculate bolometric x - ray luminosities in the following of the paper , goes from 5 ev to @xmath39 kev . the lower two panels of fig . [ b=0 ] show that after an encounter with zero impact parameter the displacement of the main cluster s x - ray peak is aligned with the @xmath15 axis . a large relative velocity ( central panel ) induces a significant offset between the dark and baryonic component of the main cluster ( see table [ runsresults ] for details ) but it also leads to substantial disruption of the main cluster gaseous core . moreover , the displacement of the bullet from its dark halo ( 278 kpc ) is much larger than observed . decreasing the relative velocity between the two clusters ( bottom panel ) two x - ray peaks are clearly visible but the displacement of the main cluster gas is now negligible due to the lower ram - pressure experienced by the main cluster core . @xcite ( fig . 7 in their article ) provide further examples of head - on encounters with even lower mass ratios and relative velocities ( @xmath40 km s@xmath8 in the center of mass rest frame ) . even assuming extremely low concentrations ( @xmath41 ) for the main halo , the authors never reproduce the displacement observed in the two systems . increasing the concentration strongly increases the luminosity of the main cluster , which appears much brighter than the bullet , contrary to what is observed . a bow shock is clearly visible on the right of each image . the shape of the shock front is only marginally dependent on the kinematics of the model while the distance between the edge of the bullet ( the so called contact discontinuity ) and the shock front becomes larger for decreasing bullet velocities . the contact discontinuity itself is much flatter in the case of 1:6b0 than in the low velocity encounter 1:6v3000b0 and clearly not comparable with observations , which show a more narrow structure . in general , a more efficient ram - pressure during the phase of core - core interaction is associated with a larger opening angle of the contact discontinuity at the present time @xcite . the rest of the runs listed in table [ runs ] have an impact parameter @xmath26 equal to 150 kpc , comparable with the core radius @xmath42 of the main cluster gas distribution for most of the models . for a plasma distributed according to a @xmath2 profile like the one adopted in this paper the assumption @xmath43 kpc implies that the maximum external density crossed by the core of the sub - cluster is 1.5 times smaller than the one it would pass through for @xmath27 kpc . for the same initial velocity , the sub - cluster also sees a weaker potential with respect to the case of a head - on encounter and experiences a smaller maximum orbital velocity . a choice of a much larger value of @xmath26 would further decrease the mutual ram - pressure between the two systems and require much higher relative velocities in order to explain the observed offset between the gas and dark matter . in fig . [ b>0 ] and [ b>02 ] we illustrate the projected x - ray surface brightness maps of some interesting models with impact parameter @xmath44 kpc . the encounters are shown face - on . the box size and the surface mass density contours are the same as in the last two panels of fig . [ b=0 ] . in order to underline the morphological details of the high emission regions the upper limit of the surface brightness scale varies in the different images . individual values are indicated in the captions . among this sub - sample of runs , 1:3 produces the largest displacement of the x - ray peak associated with the main cluster , but the x - ray map differs from the observations . in particular a large strip of strongly emitting gas still connects the two x - ray peaks while the morphology of the main cluster peak is much more elongated than observed . decreasing the mass ratio between the two interacting systems ( runs 1:6 and 1:8 ) the displacement in the bullet becomes larger than that in the main cluster . the run 1:6 is characterized by an initial sub - cluster velocity of 5000 km s@xmath8 ( as well as runs 1:3 and 1:8 ) in the system of reference where the main cluster is at rest , that corresponds to a present time velocity of @xmath45 km s@xmath8 in the center of mass rest frame . the same model is simulated assuming lower relative velocities ( 1:6v3000 and 1:6v2000 ) . with decreasing velocities the offset in the sub - cluster becomes smaller and the x - ray emission from the bullet less pronounced with respect to the bright x - ray emitting region at the center of the main cluster . at the same time the shape of the contact discontinuity changes , getting progressively more narrow while the distance between the contact discontinuity and the shock front increases as will be shown more quantitatively in the next section . the displacement associated with the main - cluster is determined by the distance of closest approach between the centers of the two clusters , which becomes smaller assuming the same initial impact parameter @xmath26 with decreasing bullet velocities . indeed , the separation between the main cluster x - ray emission peak and its dark matter counterpart is maximum in the case of the low velocity run 1:6v2000 . [ future ] a not negligible fraction of the x - ray emission visible at the present time near the center of the main - cluster is actually associated with hot gas stripped from the external regions of the bullet . fig . [ mixing ] refers to run 1:6v3000 . it shows the individual distribution of gas originating from the main ( top ) and sub - cluster ( bottom ) and lying at the present time within 1 mpc from the center of the system . comparing these images with the middle right panel of fig . [ b>0 ] it appears evident that the bright elongated x - ray feature crossing the second innermost isodensity contour is associated with the displaced gaseous center of the main cluster while the surrounding more diffuse region hosts a significant amount of sub - cluster gas . indeed , as it shown in the upper panel of fig . [ mixing ] , the motion of the bullet across the inner regions of the main system creates a low density `` tunnel '' in the main cluster gas distribution . at the same time the sub - cluster looses a large amount of baryonic material during the phase of core - core interaction . this material , which fills the tunnel , falls back into the gravitational center of the main cluster and resides now at more than 500 kpc distance from the x - ray bullet . as we will show in the next section , the amount of gas deposited by the sub - cluster in the central regions of the main system increases with decreasing relative velocities . this trend explains the relative increase in luminosity of the diffuse strongly emitting component if compared to the peak associated to the main cluster core gas ( always in the upper right region with respect to the center of the mass density distribution ) , when we compare run 1:6 with the low velocity run 1:6v2000 where it becomes the primary peak of x - ray emission . as shown in the bottom right panel of fig . [ b>0 ] a main halo with low concentration ( @xmath17 ) does not survive a 1:6 sub - cluster encounter with velocity @xmath46 km s@xmath8 and its x - ray peak is destroyed . as previously noticed by @xcite the choice of a lower gas fraction ( @xmath47 in three of the last four runs of table [ runs ] ) does not affect significantly the x - ray map morphology although the displacement of the two luminosity peaks with respect to their dark matter counterparts slightly changes . finally , we tested the consequences caused by including radiative cooling and choosing a larger main halo model ( fig . [ b>02 ] ) . cooling makes the contact discontinuity more narrow and the amount of diffuse x - ray gas around the peak associated with the main cluster smaller . the offset in the main and sub - cluster remains unaltered . run 1:6v3000big is characterized by a main cluster total mass of @xmath48 , which is closer to the value adopted by @xcite and predicted by fitting the large field weak lensing data with extremely low concentrated ( @xmath49 ) nfw halos . although the initial relative velocity is only 3000 km s@xmath8 , due to the large mass of the host halo the present time velocity of the bullet in the center of mass rest frame is much higher than in the corresponding 1:6v3000 run . consequently the offset of the x - ray peak associated with the main cluster is less than 100 kpc and the amount of gas lost by the bullet in the core of the main system closer to that observed in 1:6 than in 1:6v3000 . summarizing , run 1:6v3000 , with mass ratio 1:6 , initial relative velocity @xmath50 km s@xmath8 and present time sub - cluster velocity @xmath51 km s@xmath8 in the center of mass rest frame , best reproduces the main features observed in x - ray maps , in particular the peculiar morphology of the x - ray emission associated with the main cluster , the relative surface brightness between the main and the sub - cluster , the shape of the shock front and of the contact discontinuity . although the low velocity run 1:6v2000 leads to a x - ray displacement closer to that observed by @xcite both in the main and in the sub - cluster , this model seems to be excluded on the basis of a pure morphological comparison with the observational data . indeed , the bullet seems to be much less bright than the center of the main cluster . [ masses ] compares the integrated mass profiles of the main and the sub - cluster for different mass models with the gravitational lensing results from @xcite and previous simulations by @xcite . the profiles inferred from observations are obtained by measuring the enclosed mass in cylinders centered on the southern cd of the main cluster and on the bcg of the sub - cluster . here we calculated the present time projected mass within cylindrical bins centered on the center of mass of the two clusters . the massive cluster run 1:6v3000big ( green lines ) , overestimates the cumulative profile of the main cluster , while an initial main cluster mass of @xmath52 ( 1:6v3000 , in red ) fits the observational points better than the large - mass - low - concentration model of @xcite in the central regions and underestimates the projected mass by less than @xmath53 between 250 and 450 kpc from the center . blue lines refer to run 1:3lfg where the main halo has the same dark matter mass than 1:6v3000 but lower baryonic fraction . it is interesting to notice that using almost the same initial sub - cluster mass ( the sub - cluster model in run 1:6v3000big and 1:3lfg only differs in the baryonic fraction ) we can nearly reproduce the observed projected mass in the bullet only in the case of a very massive host halo ( green dashed line , 1:6v3000big run ) which overpredicts the main cluster profile ( green solid line ) . on the other hand a smaller main halo which better fits the observational data ( blue solid line ) associated with a similar sub - cluster mass ( 1:3lfg ) underpredicts the lensing results ( blue dashed line ) . our favourite run 1:6v3000 gives even lower projected sub - cluster mass estimates ( red dashed line ) and lies close to the model proposed by @xcite ( their sub - cluster mass is similar to the one used in our 1:6 models ) . these results seem to indicate that the mass of the sub - cluster is a significant part of the main cluster mass , with initial mass ratio between the sub and the main cluster much larger than 1:3 . @xcite successfully reproduces the projected mass associated with the sub - cluster only assuming very massive halos and high mass ratios . in particular the smallest mass ratio allowed by their models is 1:2.7 , with a main cluster and bullet mass of @xmath54 and @xmath55 , respectively . such high masses are however clearly incompatible with the values inferred by galaxy kinematics . in particular the virial mass derived by @xcite for the main cluster is @xmath56 ( comparable with the value obtained by girardi & mezzetti 2001 ) . they also measure the main cluster total r - band luminosity and find @xmath57 using @xmath58 to estimate the cluster mass @xcite we obtain @xmath59 . the bullet mass determination of @xcite is less reliable since it is based on only seven galaxies and on the assumption of equilibrium while the sub - cluster seems to be tidally perturbed beyond @xmath60 kpc from its center . they find @xmath61 . from the total r - band luminosity ( @xmath62 ) we get a value almost five times larger ( @xmath63 ) . the bullet mass proposed by @xcite is even one order of magnitude larger than that obtained from weak lensing analysis ( @xmath64 ) of a new larger lensing field which covers most of the area occupied by the system ( clowe et al . in prep . ) . with such low mass ratios we can not reproduce the morphology of the x - ray maps ( upper left panel of fig . [ b>0 ] ) since the luminosity peak associated with the main cluster core is almost completely destroyed by the interaction . actually any attempt to derive the mass of the main and sub - cluster simply by projecting the mass of two isolated nfw halos and fitting the observed lensing data like the analysis of @xcite is going to provide wrong results . indeed , as we will show in section 4 , the dark matter profiles of the main and sub - cluster change significantly during the central phases of the interaction , with a density increase in the inner regions and a decrement beyond 0.3 @xmath38 . this effect , which characterizes both the main and the sub - cluster , is not related to stripping processes but can be explained by the fact that the two halos are strongly perturbed and can not be idealized as the sum of two isolated equilibrium models . fitting the projected cumulative mass around the centers of the two interacting clusters with nfw models we would overestimate either the virial mass or the concentration of the original clusters . = 8truecm according to @xcite the line of sight velocity difference between the main and the sub - cluster is relatively small , about 600 km s@xmath8 , which implies that the encounter must be occurring nearly in the plane of the sky . in fig . [ inclination ] we show the x - ray surface brightness map of run 1:6v3000 when the orbital plane is viewed inclined by 45 degrees left ( top panel ) and down ( bottom panel ) . in the upper image the bullet ( which is closer to the observer ) appears to be much brighter than the main cluster while the contact discontinuity and the shock front are much rounder than in the case where the encounter is seen perpendicular to the orbital plane . such a large inclination has to be excluded by a comparison with observations . moreover , the line of sight velocity difference between the two clusters ( we assume the line of sight kinematics of the dark matter to be coincident with that of the stellar component observed by barrena et al . 2002 ) is much larger than the observed one ( @xmath65 km s@xmath8 ) due to the fact that the main component ( @xmath66 km s@xmath8 in the center of mass rest frame ) of the sub - cluster velocity is oriented in the direction of the positive @xmath15 axis in the original face - on encounter . a velocity difference comparable to the observed one is reached for a right to left inclination of @xmath67 degrees . on the opposite , a top - down inclination of the plane of the encounter with respect to the line of sight ( bottom panel ) does not produce drastic changes in the morphology of the bow shock and of the edge of the bullet . indeed the main and the sub - cluster still lie close to the plane of the sky , but the relative surface brightness between the dense and the diffuse strongly emitting components at the center of the main system changes . in particular the peak of x - ray emission due to the presence of hot gas deposited by the sub - cluster within the core of the main cluster is now comparable with the main cluster x - ray peak . the difference in line of sight velocity in this case is only @xmath68 km s@xmath8 and would probably not be distinguishable with respect to a pure face - on encounter . [ future ] = 8truecm fig . [ sb ] represents the x - ray brightness profile measured in a narrow slit ( with thickness 20 kpc ) parallel to the @xmath15-axis across the shock front . the bullet is located at @xmath69 with @xmath15 increasing towards the pre - shock region . the general trend does not depend on the model and is comparable with the profile suggested by observations @xcite : all the simulations show an inner bump associated with the bullet and an outer one ( between @xmath70 and @xmath71 kpc ) caused by the shock . the abrupt jump at @xmath72 kpc is the shock front while the pre - shock region is well fitted by a two dimensional @xmath2-profile @xmath73^{-3\beta + 1/2}$ ] . in the case of run 1:6 @xmath74 kpc , @xmath75 and @xmath76 erg s@xmath8 kpc@xmath77 . a different orientation ( @xmath78 , bottom - up inclination of the plane of the encounter ) of the line of sight does not affect substantially the surface brightness profile . if cooling is activated the surface brightness of the pre and post - shock region is smaller , but the thickness of the shock - front and the jump in surface brightness are similar . in order to calculate projected temperatures we need to define a weighting function . the emission weighted temperature @xmath79 was originally introduced to provide a better comparison between simulations and observations with respect to a simple mass weighted temperature definition and has been commonly used in the analysis of simulations @xcite . it assumes a weighting function proportional to the emissivity of each hot gas particle and is defined as @xmath80 recently @xcite have demonstrated that for clusters with a complex thermal structure the emission weighted temperature always overestimates the spectroscopic temperature obtained from x - ray observations due to fact that the source is not a single temperature plasma . for clusters with a temperature @xmath81 kev the discrepancy is @xmath82 . this difference becomes particularly large in the presence of strong temperature gradients like shocks which appear to be much weaker in observations than what is predicted by emission - weighted temperature maps ( mathiesen & evrard 2001 , gardini et al . 2004 , rasia et al . 2005 ) . @xcite proposed a new definition of temperature , the spectroscopic - like temperature @xmath83 : @xmath84 when applied to clusters hotter than 2 - 3 kev this equation , with @xmath85 , gives a good approximation ( within few percent ) of the spectroscopic temperature obtained from data analysis of chandra . the bullet cluster 1e0657 - 56 is characterized by the highest luminosity , temperature and the strongest bow shock of all known clusters @xcite and seems indeed to be the ideal candidate for adopting the spectroscopic - like temperature definition . therefore in the remainder of this paper @xmath83 will be analyzed and @xmath79 only indicated for a comparison . [ temppro1 ] illustrates the projected spectroscopic - like temperature profiles across the shock for the different runs of table [ runs ] . all the values refer to the present time . in particular , the upper left panel shows the temperature jumps associated with different relative velocities of the two clusters . decreasing the initial relative velocity from @xmath86 to @xmath87 km s@xmath8 ( in the system of reference where the main cluster is at rest ) reduces the temperature peak by @xmath88 kev while the peak itself becomes broader since the thickness of the shock increases by almost a factor of two due to the lower pressure exercised by the pre - shock gas . both the 1:6v3000 and 1:6v2000 models seem to fit quite nicely the observed height ( @xmath89 kev ) and thickness ( 150 - 200 kpc ) of the shock front @xcite while the 1:6 run produces a peak which is too narrow ( @xmath68 kpc ) and pronounced ( @xmath90 kev ) . the upper right panel refers to different mass ratios . if the encounter is characterized by the same initial relative velocity and gas fraction the strongest shock is associated with the most massive sub - cluster . a 1:3 adiabatic encounter produces a @xmath91 kev temperature peak with thickness @xmath92 kpc while the 1:8 temperature profile is not substantially different from the one of the 1:6 run . in the same plot we also show the shock created in the massive run 1:6v3000big . clearly the maximum temperature is much higher than the observed one . including radiative cooling ( bottom left panel ) has the effect of reducing the peak in temperature but does not influence the thickness of the shock region . the adoption of a simple cooling model is actually questionable . indeed , although the estimate of the temperature jump is reasonable , the entire temperature profile drops by 5 kev due to the fact that once cooling is activated the main cluster gas component becomes thermally unstable in the early phases of the interaction and overcools in the central regions . if cooling is important , models with higher mass ratios and relative velocities ( like the 1:3lfg run in plot ) have still to be taken in account and can not be excluded as the high temperature peaks could actually cool significantly . the choice of a different line of sight ( bottom right plot ) does not affect significantly the temperature profile along the shock . even decreasing the baryonic fraction in the clusters and assuming a much less concentrated main halo , the height and thickness of the temperature peak do not change . = 8truecm in fig . [ temppro2 ] we compare the spectroscopic - like temperature profile across the shock region with the emission weighted one for our favourite run 1:6v3000 . while the projected temperature profile calculated according to the two definitions is similar ( @xmath79 is only slightly higher than @xmath83 ) in the pre and post - shock regions , the emission weighted temperature @xmath79 in the region @xmath93 kpc is @xmath94 higher than @xmath83 . @xmath95 represents the actual temperature along the@xmath15-axis through the shock . indeed the blue curve in fig . [ temppro2 ] gives the exact temperature jump across the shock ( the `` true temperature '' of the shock ) which is characterized by an even higher peak with respect to the projected ones . this deprojected temperature profile is actually the one shown by @xcite in their fig . 9 where they compare their model with observations which on the other hand refer to projected quantities @xcite . as the calculated local temperatures of @xcite fit the projected observed temperature profile very well we conclude that their projected temperature profiles are actually inconsistent with the observations by @xcite . in fig . [ sequencexray ] , [ sequencetemp ] , [ sequenceden ] we show for our selected adiabatic run 1:6v3000 the evolution with time of the @xmath96 kev x - ray surface brightness , spectroscopic - like temperature and gas surface density during the central phases of the interaction . all the quantities are projected along an axis perpendicular to the plane of the collision . on top of the color maps we draw in black the projected isodensity contours of the total mass distribution , which is dominated by the dark matter component . time increases from the left to the right and from the top to the bottom . the sequence of six panels covers an interval in time of 400 myr , between 1.3 and 1.7 gyr from the beginning of the simulation . the snapshot corresponding to the present time is the one on the bottom left . the bullet approaches the main cluster from the left , with an initial velocity of 3000 km s@xmath8 in the main cluster rest - frame . the shock front has an arc - like shape @xcite and becomes progressively more asymmetric as the bullet moves closer to the core of the main cluster . when the centers of the two clusters are less than 250 kpc separated , ram - pressure becomes effective in producing a displacement ( visible in the projected density maps as well as in surface brightness ) between the gaseous core of the bullet and the peak of its associated mass distribution . the offset in the main cluster is evident only when the bow shock passes through its core . in the central panels of fig . [ sequencexray ] and [ sequenceden ] the x - ray luminosity and surface density saturate in order to distinguish features in the maps at later times . nevertheless , it is visible how the main cluster core gets compressed and displaced from the center of the potential towards the top right of the image and appears in the x - ray maps ( last two panels ) as an elongated structure , characterized by high surface brightness . temperature maps better describe the evolution of the shock region , which gets compressed and hotter during the core passage while in a later phase it cools down and becomes thicker due to lower pressure of the pre - shock gas . the bullet itself expands as it leaves the central regions of the main cluster . as observed by @xcite and previously noticed by @xcite , despite of its strong x - ray emission ( it is the brightest feature in the post core - core interaction x - ray maps ) the bullet remains relatively cold . even if radiative cooling is not activated , the core of the sub - cluster is heated only to a maximum temperature of @xmath97 k , while the shock front is much hotter ( @xmath98 k ) . the projected temperature associated with the bullet is higher when it passes through the core of the main system ( central panels of fig . [ sequencetemp ] ) due to the line of sight overlap with the hot gas from the main cluster and the shock heated material stripped from the bullet itself which surrounds it . as soon as the sub - cluster moves out into the cool external regions of the main cluster and loses part of the hot envelope of stripped gas , its projected temperature decreases to values comparable to the observed ones . another peculiar feature in the temperature maps is the high temperature region next to the innermost total density contour of the main cluster and visible in the middle right and bottom left panels of fig . [ sequencetemp ] . this area could be associated with the southeastern high temperature region observed by @xcite ( regions @xmath99 , @xmath100 and @xmath101 in their fig . 2 ) in the main cluster x - ray map and assumed to be coincident with the main merger site . in our simulations the high temperature region is filled with hot gas stripped from the external regions of the sub - cluster and deposited within the core of the main system ( compare with fig . [ mixing ] ) . this high temperature material , combined with the main cluster gas which lies in the same projected region , produces the diffuse x - ray high emission feature visible at the present time in the main cluster below the primary peak ( bottom left panel of fig . [ sequencexray ] or middle right panel of fig . [ b>0 ] for a better colour contrast ) . the eastern side of the high temperature region is less bright in x - ray emission and lies beyond the luminosity peak associated with the main cluster . as the sub - cluster moves to larger radii , the high temperature region expands and cools . shortly ( @xmath102 myr ) after the present time this gas is not anymore clearly distinguishable in the x - ray and density maps . in this section we will focus on a sample of runs from table [ runs ] namely 1:6 , 1:6v3000 , 1:6v2000 and 1:3clfg and investigate in details their three - dimensional characteristics . three of these simulations are adiabatic models characterized by different initial relative velocities and will permit us to study the effects of the sub - cluster speed on the present and future state of the encounter . the remaining run 1:3clfg is the one which better reproduces together with 1:6v3000 and 1:6v2000 the observed jump in temperature across the shock front . [ shock ] illustrates the behavior of hydrodynamical quantities across the shock discontinuity for a snapshot which corresponds to the present time . the horizontal axis is centered on the gaseous bullet and oriented perpendicular to the bow shock nose . the bow shock location is indicated by a vertical line while the edge of the bullet corresponds to the peak in density . the sub - cluster is moving outward from the main cluster core towards positive values of @xmath15 . all the physical quantities are calculated and mass averaged on a one dimensional grid where each grid element has a volume of @xmath103 kpc@xmath104 . @xmath105 is the component of the velocity perpendicular to the shock front and is calculated with respect to the system center of mass . the mach number @xmath106 is determined from the temperature jump which shows a better defined discontinuity compared with the density jump using the rankine - hugoniot conditions . we find values in good agreement with observations @xcite for 1:6v3000 and the cooling run 1:3clfg ( both with @xmath107 ) . 1:6v2000 is characterized by a slightly lower value of @xmath106 ( @xmath108 ) while the high velocity run 1:6 has a stronger shock , with @xmath109 . as seen in the previous section , the pre - shock temperature ( @xmath11011 - 12 kev for the adiabatic runs ) is slightly higher than the one obtained by projection along the line of sight ( fig . [ temppro2 ] ) . assuming @xmath111=11 kev and @xmath112 we predict a pre - shock sound speed @xmath113 km s@xmath8 and a shock velocity @xmath114 km s@xmath8 . this value reduces to the observed shock velocity @xmath115 km s@xmath8 if we use the projected average pre - shock temperature ( @xmath116 kev ) adopted by @xcite . as previously noticed by other authors @xcite the velocity jump shown in the last panel of each plot is much smaller than the theoretically inferred shock velocity . actually the pre - shock gas is not at rest but shows a negative velocity along the @xmath15-axis which however can only be partially explained by the fact that the center of mass of the system is moving in the positive direction of the @xmath15-axis following the bullet . indeed the upstream velocity maintains a negative sign even with respect to the rest frame of the parent cluster , indicating a pre - shock infall towards the bullet . this effect is explained in @xcite by studying the dynamical evolution of the system s global potential , which becomes deeper after the core - core interaction and induces infall of material from the region ahead of the shock . nevertheless the infall velocity which characterizes our models is a significantly smaller than the sub - cluster velocity in contrast to the values found by @xcite and ranges only between 500 and 900 km s@xmath8 in the different runs . = 8truecm = 8truecm figs . [ vel3000 ] and [ vel5000_2000 ] illustrate the characteristic velocities of the sub - cluster in the orbital plane for the three runs 1:6v3000 , 1:6 and 1:6v2000 . all the velocities are calculated in the center of mass rest frame and the time corresponding to the present position is indicated by a vertical line . the velocity of the dark matter component peaks at the moment of closest approach between the two cores and then decreases faster than for a ballistic orbit as a result of dynamical friction . the escape velocity at a given sub - cluster position is calculated assuming a spherical unperturbed host potential and is indicated by a black dotted curve . all the runs have initially unbound sub - clusters . due to the effects of dynamical friction after the phase of core - core interaction the 1:6v2000 sub - cluster is actually bound to the main system , while sub - clusters with initial velocities @xmath50 km s@xmath8 and 5000 km s@xmath8 have velocities slightly or much larger than the escape velocity from the main cluster . the gaseous bullet initially follows its dark matter counterpart but before the point of closest approach it is slowed down by ram - pressure . at the same time the morphology of the sub - cluster gas distribution changes . the contact discontinuity assumes an arc - like shape ( partially reducing the effect of ram - pressure ) and the bow shock forms . it is interesting to notice that after the point of closest approach the evolution of the relative velocity between gas and dark matter in the sub - cluster strongly depends on the intensity of the ram - pressure force . in particular for relatively low ram - pressure values ( 1:6v2000 and 1:6v3000 ) the gaseous bullet is accelerated towards its dark matter counterpart as soon as it leaves the core of the host cluster where it experienced the largest external densities and ram - pressure . as a result the gravitational acceleration relative velocity between the gaseous and dark component of the sub - cluster ( whose two components in the orbital plane are represented for 1:6v3000 by the blue solid and dashed curves in fig . [ vel3000 ] ) is larger than zero . at the time corresponding to the present position the two velocities look comparable in the case of 1:6v3000 while in 1:6v2000 , where the ram - pressure acting on the bullet is lower , the acceleration starts earlier and the gaseous bullet is already @xmath117 km s@xmath8 faster than its dark counterpart . for larger impact velocities as in the case of 1:6 ram - pressure is effective in slowing down the gaseous bullet even at large distances from the center of the main - cluster . the velocity of the gaseous bullet is therefore smaller than that of its dark counterpart until when the sub - cluster is well outside of the virial radius of the main system . the green solid curve in each plot represents the velocity of the front shock obtained by differentiating the positions of the shock front at increasing times . the shock velocity rapidly increases after the core - core interaction and at the present time it is always larger than the velocity of the gaseous bullet . in order to calculate the bullet and shock velocities in the system of reference of the pre - shock gas ( red and green dashed curve , respectively ) the infall velocity of the upstream gas is calculated at different times before and after the present one . the shock velocity @xmath118 at the present position is 4100 , 4500 and 5100 km s@xmath8 in the case of 1:6v2000 , 1:6v3000 and 1:6 , respectively , consistently with the shock velocity @xmath119 inferred from the rankine - hugoniot jump conditions and with the value provided by observations . only 1:6v2000 shows a two - dimensional shock velocity well below the observational uncertainties . as found by @xcite , after the point of closest interaction the shock velocity is always larger than the velocity of the sub - cluster mass centroid , but the amount of the difference strongly depends on the model . in particular , in the case of the two low velocity runs 1:6v2000 and 1:6v3000 @xmath118 is @xmath120 and @xmath121 larger than the velocity of the dark matter component , while in 1:6 the difference is almost negligible ( only @xmath122 ) . + the cluster collision produces a drastic increase in luminosity and temperature . [ lvst ] and [ tvst ] show the bolometric x - ray luminosity @xmath123 and average spectroscopic - like temperature @xmath83 as functions of time . both quantities are calculated for the entire simulated box and scaled to their initial values . = 8truecm = 8truecm the first phase of the interaction which involves only the external low density regions of the two clusters is characterized by an identical slow increment of luminosity and temperature in all the adiabatic runs . the jump is associated with the phase of core - core interaction and peaks right after the time of closest approach . the present time is indicated with an empty squared and sits on the downturning curve . the high velocity run 1:6 is associated with the largest increase in temperature ( @xmath124 3.5 ) and with the smallest jump in luminosity ( @xmath1254 ) . for decreasing sub - cluster velocities the peak in temperature becomes slightly less pronounced while the luminosity jump rises by a factor of 1.5 . the loss of baryonic material from the sub - cluster within the high density core of the main system is indeed larger for low velocity encounters and leads to higher luminosities even if the increment in temperature is smaller with respect to the high velocity runs . excluding the bound run 1:6v2000 , the amount of gas stripped from the bullet and lying within the virial radius of the main cluster is @xmath126 in 1:6 and @xmath127 in 1:6v3000 , which in both cases corresponds to almost @xmath128 of the initial baryonic content of the sub - cluster . the difference becomes more pronounced if we consider only the core ( @xmath129 ) of the main cluster , where the mass of sub - cluster gas is @xmath130 in the 1:6v3000 run and one order of magnitude less in the case of 1:6 . at 4 gyrs after the beginning of the simulation the total luminosity of the system is similar to the initial one in the case of the bound system 1:6v2000 where the center of mass of the bullet does not move out to distances beyond the virial radius of the main system . the luminosity drops to @xmath131 or even less of the initial luminosity for the runs 1:6v3000 and 1:6 respectively . this decrease in luminosity is motivated by the fact that at the final stages of the simulations a significant fraction of the gas is unbound and very extended . due to its low density it does not contribute to the luminosity of the system despite the high temperature . in particular , for the same intracluster distance , the high velocity encounter 1:6 is associated to the highest fraction of unbound material , as will be shown later in this section . on the other hand , the final temperature of the system is higher than the one associated to the initially isolated clusters and converges to a value @xmath132 almost independently of the sub - cluster velocity . the cooling run 1:3clfg shows a somehow different behaviour : both , luminosity and temperature profiles have an initial decrement due to the cooling of the central regions of the two approaching clusters . already during the early phases of the interaction the cooling run moves out of the equilibrium . the peaks in luminosity and temperature are much smaller than the corresponding adiabatic ones ( not illustrated ) . the final luminosity approaches zero . = 8truecm = 15truecm @xcite found that the bullet cluster lies exactly on the @xmath133 relation for nearby clusters @xcite but its temperature is much higher than the one expected according to weak lensing mass estimates . [ tl ] illustrates the drift of the simulated systems along the @xmath133 diagram . the position of 1e0657 - 560 is indicated by a star . @xmath123 and @xmath83 are calculated in a cylindrical region centered on the center of mass of the main cluster and radius equal to its initial virial radius . the luminosity of each model is normalized in such a way that the initial main cluster lies on the @xmath133 relation for local clusters despite of the different initial gas fractions . the starting time of the simulations is indicated with a black solid square . all the adiabatic runs present a similar evolution and move roughly parallel to the @xmath133 relation , as previously noticed by @xcite for major mergers in cosmological simulations . during the early stages of the encounter the cluster moves along a curve which is flatter than the observed @xmath133 relation : the compression of the low density gas at the outskirts of the cluster produces an increase in temperature which is only marginally accompanied by a luminosity growth . the time when the core of the sub - cluster enters the virial radius of the main system represents an inversion point in the diagram : despite the formation of a bow - shock the temperature decreases due to the expansion of the main - cluster gas and the presence of low temperature baryons belonging to the sub - cluster within the virial radius of the main system . at the same time the luminosity rises as a result of the shock and the cluster moves perpendicularly towards the @xmath133 relation . this phase is actually quite short ( on average @xmath134 gyr ) but it characterizes all the adiabatic runs . later on the cluster moves almost parallel to the @xmath133 relation toward larger values of @xmath83 , with the peak in temperature being reached at the point of closest approach . most of the runs show a small delay ( @xmath135 myr ) between the time characterized by the highest temperature and the time with highest luminosity with the curve making a knot in the diagram . the branch of the curve associated with the post core - core interaction is parallel to the increasing one but shifted to smaller luminosities : during the strongest phase of the interaction some of the hot material is lost beyond the virial radius of the main cluster and indeed the largest shift in luminosity is observable in the high velocity run 1:6 . both , luminosity and temperature now decrease until they reach a second inversion point in the curve ( in the case of 1:6 it is only a change in slope ) , associated with the egress of the bullet from the virial radius of the main system . the cooling run 1:3clfg is characterized by a first decrease in temperature which corresponds to the initial phase of thermal instability and central cooling of the main cluster . later on it moves in the @xmath133 diagram similarly to the adiabatic runs although the peaks in luminosity and temperature are much less prominent . nevertheless in a pure cooling model the main cluster does not return to a state of thermal equilibrium at the end of the interaction since nothing prevents the central regions from cooling and the system moves toward extremely low values of luminosity and temperature . [ profiles ] illustrates the final structure of the remnants . in the case of the two highest velocities the final time is chosen in such a way that the distance between the centers of the two systems is about @xmath136 kpc larger than the sum of the virial radii at @xmath137 . this occurs at @xmath138 and @xmath139 gyr in the case of 1:6 and 1:6v3000 , respectively . the low velocity run 1:6v2000 is analyzed at the time corresponding to the first apocenter when the core of the sub - cluster is close to the virial radius of the main cluster . the collisionless component of the main cluster ( top left panel ) in all cases is not substantially affected by the interaction while the sub - cluster dark matter halo appears to be strongly perturbed and retains the original spherical symmetry only within its scale radius . the sub - cluster central density profile ( top right ) is vertically shifted downward without a significant change of slope while for @xmath140 the loss of material becomes more significant and in the case of low velocity encounters ( @xmath141 and @xmath142 km s@xmath8 ) with mass ratio 1:6 the profile shows a large jump of more than one order of magnitude between @xmath143 and @xmath144 . beyond the scale radius the isodensity contours appear elongated and show a large plateau ( see also fig . [ future ] ) associated to tidally stripped material . the evolution of gas density profiles is represented by the two panels on the bottom row of fig . [ profiles ] . in the 1:6 runs the interaction affects the central slope of the main cluster which becomes shallower while the 1:3 encounter produces the largest deviation from the initial values , with the final density profile shifted down by @xmath145 . as mentioned earlier , a not negligible part of the gas within the virial radius of the main cluster originally belonged to the sub - cluster and was subsequently stripped by ram - pressure during the central phases of the interaction . the density profiles of the stripped gas are drawn in blue for the different runs . in the case of 1:6v2000 only the gas outside the virial radius of the sub - cluster is considered . both 1:6v3000 and 1:6v2000 are characterized by a large fraction of sub - cluster gas lost to the core of the main system , with a flat ( 1:6v2000 ) or even positive ( 1:6v3000 ) central slope , while the high velocity run 1:6 , despite the larger ram - pressure values , has less time to deposit gas in the central regions and shows a clear cut - off for @xmath146 . at larger radii , baryonic material is still accreting onto the remnant . indeed , part of the main cluster gas has been pushed out by the bow shock and is now falling back into the cluster potential together with a fraction of the material lost by the sub - cluster . curves in the bottom right panel of fig . [ profiles ] represent the sub - cluster gas density profiles . the low velocity sub - clusters 1:6v3000 and 1:6v2000 retain less than one tenth of their initial gas within @xmath147 while the density profiles in the external regions drop by a factor @xmath148 . in general , the encounter flattens the gas density profile of the bullet core and this effect is evident in the cooling simulation as well . the 1:6 bullet at @xmath138 gyr has lost almost all the gas within @xmath38 . half of this baryonic material was stripped from its dark halo by ram - pressure during the central phases of the interaction and now accelerates towards its dark matter counterpart . it will be accreted by the sub - cluster halo at later times . [ profiles ] compares dark matter and gas density profiles at times when the main and sub - cluster are close to a state of virial equilibrium . despite the changes observable in the gas density profiles and the tidal stripping affecting the bullet beyond its core radius , the central slope of the collisionless component seems to be not strongly perturbed by the interaction @xcite the situation changes comparing the density profiles of the initial systems with those of the main and sub - cluster at the present time , when the bullet is located at almost one third of the host virial radius . as illustrated in the two upper panels of fig . [ profiles2 ] , the density profile of the interacting dark matter halos increases in the inner regions and shows a decrement beyond @xmath149 . this behavior is similar for different models and therefore independent of the orbital details , of the mass ratios and is seen for both , the main and the sub - cluster . the two - body relaxation time scale beyond the softening radius is longer than @xmath150 gyr @xcite , thus implying that dark matter density profiles are not affected by numerical relaxation . as noticed in section 3 , fitting the projected masses around the centers of main and sub - cluster at the present time we would get more concentrated or more massive systems with respect to the real ones . the last two panels of fig . [ profiles2 ] represent gas density profiles at the present time . the loss of gas in the central regions of the sub - cluster is expected as a result of ram - pressure stripping . but interestingly also the inner 200 kpc of the main cluster are completely devoid of gas originally belonging to the main cluster . later on , after the displacement of the main cluster gaseous core , the baryonic material stripped from the bullet replenishes the central regions of the host system . as soon as the bullet moves toward the outskirts of the main system , the main cluster gas has time to collapse again into the center of the potential ( bottom left panel of fig . [ profiles ] ) . + according to @xcite the probability to find a cosmological configuration where the most massive sub - halo has a velocity larger than @xmath151 is well fitted by the function : @xmath152 where @xmath153 is the virial velocity of the main cluster . adopting for @xmath151 the present time velocity of the sub - cluster dark matter component ( in table [ runs ] ) we find a probability of @xmath154 , @xmath155 and @xmath156 for runs 1:6 , 1:6v3000 and 1:6v2000 respectively . we used high resolution n - body / sph simulations to perform an extensive parameter study of the `` bullet '' cluster system 1e0657 - 56 . collisions of two nfw halos with hot isothermal gas components in hydrostatic equilibrium were studied , adopting initial relative velocities of 5000 , 3000 and 2000 km s@xmath8 , which is 3.8 , 2.3 and 1.5 times the main cluster s virial velocity , respectively . we varied masses , mass ratios , impact parameter , baryonic fraction and concentrations . most of the runs are adiabatic . radiative cooling is included in two cases , adopting a standard cooling function for primordial gas . we analyzed the projected properties of the system at a time where the distance between the centers of the mass distribution associated with the main and sub - cluster is comparable to the value provided by lensing observations . we have shown that : * most of the main features in the observed x - ray maps are not well reproduced by encounters with zero impact parameter . indeed , depending on the relative velocity of the bullet , a perfectly head - on encounter either destroys the x - ray peak associated with the main cluster or does not produce a significant displacement between gas and dark matter in the main system . moreover , in a pure head - on collision , the gaseous core of the main cluster is displaced along the line which connects the centers of the two total mass distributions , contrary to observations . an impact parameter corresponding to the core radius of the main cluster gas distribution provides enough ram - pressure to produce a displacement comparable to observations and introduces asymmetries in the main cluster emissivity map , similar to those detected in x - ray . * a low concentrated ( @xmath17 ) main cluster does not survive the collision and its x - ray emissivity peak is destroyed . * encounters with mass ratios as large as 1:3 do not match the observed x - ray morphology and the size of the projected temperature jump across the shock discontinuity . introducing cooling in the simulations , the temperature peak is cooled to a value comparable with the observed one . on the other hand , lensing mass reconstructions seem to suggest even larger mass ratios . this remains an open question . gravitational lensing analysis of the simulations presented in this paper will be part of a follow - up paper ( mastropietro , maccio & burkert in prep ) . * a pure cooling model neglecting energetic feedback still does not give a realistic description of the interacting system . what we are witnessing is overcooling in the central regions following a thermal instability in the early phases of the interaction . nevertheless , cooling simulations provide a significant reduction of the temperature peak across the shock indicating that a more realistic treatment of the gas physics make models with higher mass ratios more realistic . * the choice of a different gas fraction does not affect significantly the results . * the morphology of x - ray maps is best simulated by adiabatic 1:6 encounters . in particular the run 1:6v3000 , with initial velocity @xmath46 km s@xmath8 ( @xmath157 km s@xmath8 in the center of mass system of reference ) and present time dark matter velocity @xmath158 km s@xmath8 ( again in the center of mass rest - frame ) , reproduces most of the main x - ray features : the peculiar morphology of the x - ray emission associated with the main cluster , the relative surface brightness between the main and the sub - cluster , the shape of the bow shock and of the contact discontinuity . decreasing the relative velocity ( 1:6v2000 has a present time dark matter velocity which is almost 300 km s@xmath8 smaller than that used by springel & farrar 2007 ) the bullet becomes much less bright with respect to the center of the host system in disagreement with observations , although the displacement in the two clusters is better reproduced . the high velocity run 1:6 produces a contact discontinuity much broader than observed . * a significant fraction of the x - ray emission next to the center of the main cluster is associated with gas stripped from the external regions of the bullet . in the 1:6 runs the main cluster x - ray peak presents two distinct components : a compact strongly emitting region associated with the displaced core of the host system and a more diffuse component spatially coincident with gas stripped from the sub - cluster during the central phases of the interaction . the relative luminosity of this secondary component increases with decreasing bullet velocities . * the observed difference in line of sight velocity is compatible with an inclination of about 10 degrees of the plane of the encounter . a bottom - down inclination of our simulations keeps both the clusters close to the plane of the sky and does not produce essential changes in the the x - ray maps . but it affects the morphology of the main cluster peak and the luminosity of the secondary peak with respect to that associated with the main cluster core . * the projected temperature jump across the shock discontinuity gives important indications about the nature of the encounter . indeed , both , the height and the thickness of the peak change , with the peak becoming broader for decreasing bullet velocities due to the lower pressure , exercised by the pre - shock gas after the central phases of the encounter . assuming an adiabatic equation of state , both runs 1:6v3000 and 1:6v2000 , match quite well the observed temperature jump , while for larger velocities ( 1:6 ) the peak becomes too narrow and pronounced compared to observations . * temperature maps reveal some interesting features . in particular , as already found by @xcite , the bullet remains relatively cold despite of its dominant x - ray emission . for a short time , a high temperature region appears next to the center of the main cluster mass distribution immediately after the central phases of the interaction . this feature still visible at the present time is not spatially coincident with the main cluster gaseous core . it is associated with hot gas stripped from the sub - cluster within the core of the host system . its location partially corresponds to the x - ray diffuse peak , while its eastern component could be related to the high temperature region observed by @xcite southeast of the main - cluster peak . for a selected sub - sample of runs we performed a detailed three dimensional analysis following their past , present and future evolution for four gyr . * only the low velocity bullet 1:6v2000 is actually bound to the main cluster at the end of the simulation while in 1:6v3000 and 1:6 the sub - cluster has a velocity larger than the escape velocity from the host system . * the velocity of the gaseous component of the bullet starts diverging from that of its dark matter counterpart before the point of closest approach between the two clusters , when the gas is slowed down by ram - pressure . as predicted by @xcite , due to the drop in ram - pressure after the central phases of the interaction and gravitational acceleration by its dark halo counterpart , at the present time the gaseous bullet is not necessarily slower than the dark matter halo at the present time . we explored this question in greater details and found that the behavior of the sub - cluster gas after the point of closest approach strongly depends on the initial velocity of the bullet . for relatively low velocities ( and ram - pressure values , run 1:6v3000 and 1:6v2000 ) the gaseous bullet is accelerated towards its dark matter counterpart as it leaves the core of the main cluster . at the present time the gaseous bullet moves as fast as its dark matter halo in 1:6v3000 and even 500 km s@xmath8 faster in 1:6v2000 . for larger encounter velocities ( 1:6 ) ram - pressure is more effective in slowing down the bullet even beyond the core radius of host cluster and the gaseous bullet is always slower than dark matter . * it has been already noticed by @xcite and @xcite that the sub - cluster velocities do not coincide with the shock velocity @xmath159 as measured by observers . in the case of the three runs ( 1:6v3000 , 1:6v2000 and 1:6clfg ) which best reproduce the projected temperature jump , the mach number and shock velocity determined using the rankine - hugoniot conditions across the shock discontinuity show a good agreement with the values provided by @xcite . the two dimensional shock velocity calculated by tracking the shock position as function of time is consistent with these values . we find that , although after the point of closest approach @xmath159 is always larger than the velocity of the sub - cluster dark matter halo , the difference depends on the model and becomes less significant for higher velocities . in particular in the high velocity run 1:6 the present time shock velocity is only @xmath122 higher than the dark matter bullet . * the collision produces a drastic increase in luminosity and temperature . the highest velocity impacts are associated with the largest increase in temperature ( and shocks ) and the smallest peaks in luminosity due to the fact that in high velocity encounters a significant fraction of the baryons stripped from the bullet are deposited at large radii within the main cluster . @xcite found that the bullet cluster lies on the @xmath4 relation for local clusters @xcite at very high temperature . we followed the evolution of the main cluster in the @xmath4 diagram during the different phases of the interaction and find that it moves roughly parallel to the @xmath4 relation for nearby clusters . the maximum temperature is reached at the point of closest approach and the peak in luminosity immediately afterwards , with a small delay which produces a knot in the diagram . the branch of the curve associated with the post core - core interaction is still parallel to the @xmath4 relation but shifted towards smaller luminosities with respect to the early increasing branch , due to the loss of gas beyond the virial radius of the main cluster . * after the encounter , as soon as the bullet is close or beyond the virial radius of its host system , the dark matter density profile of the main cluster does not deviate anymore significantly from the original one , while the gas profile becomes shallower in the central regions , with a large decrement in density observed only in run 1:3clfg . the situation changes drastically if we compare the density profiles of the initial systems with those at the present time , when the bullet is still located well within the virial radius of the main cluster . in this case the interacting systems are not in virial equilibrium and the dark matter densities of both , the main and sub - cluster , increase in the inner regions and show a decrement beyond 0.4 @xmath38 . comparing these present time profiles with nfw halos would give wrong estimates of the halo parameters . at a time corresponding to the present configuration the center of the host system is completely devoid of main cluster gas as a result of the displacement of the main cluster gaseous core by the bullet . on the other hand , the baryons stripped from the bullet have replenished the central regions of the host . this is only a temporary situation however since a few gyrs later the main cluster gas falls back and becomes again the dominant component in the central regions , although a significant fraction of gas stripped from the bullet is still present . the final sub - cluster dark matter density profiles seem significantly affected by the interaction beyond their scale radius , where the isodensity contours are elongated and show a plateau . we would like to thank r. piffaretti , s. borgani , a. biviano , t. naab , a. maccio and m. girardi for useful discussions . we acknowledge t. quinn for support with the tipsy x - ray package and d. clowe for providing the upper panel of fig . the work was partly supported by the dfg sonderforschungsbereich 375 `` astro - teilchenphysik '' . the numerical simulations were performed on a local sgi - altix 3700 bx2 which was partly funded by the cluster of excellence `` origin and structure of the universe '' .

in this paper we analyze the possibility of persistent currents of a two - species bosonic mixture in the one - dimensional ring geometry . we extend the arguments used by bloch to obtain a criterion for the stability of persistent currents for the two - species system . if the mass ratio of the two species is a rational number , persistent currents can be stable at multiples of a certain total angular momenta . we show that the bloch criterion can also be viewed as a landau criterion involving the elementary excitations of the system . our analysis reveals that persistent currents at higher angular momenta are more stable for the two - species system than previously thought .

1207.3449

the hallmark of superfluidity is the possibility of dissipationless flow in situations where the flow of a normal fluid would degrade as a result of viscosity . the textbook example of this is the flow of a superfluid through a narrow capillary @xcite . according to the landau criterion @xcite , the superfluid component flows without dissipation provided the superfluid velocity does not exceed some critical value . in this situation , the normal component remains locked to the walls of the capillary whereas the superfluid component , carrying zero entropy , flows as if the walls of the capillary behaved as a perfectly smooth conduit . if the capillary is now bent into a torus , one can imagine that a flow , once established , could persist indefinitely . the conditions under which persistent currents can occur for a bosonic mixture in the ring geometry is the subject of this paper . the usual analysis @xcite leading to the landau criterion is not obviously applicable since one can not invoke galilean invariance for this closed system . however , for a system having a single component , bloch @xcite presented general arguments based on an analysis of the quantum mechanical many - body wave function which provided a criterion for persistent currents . he considered an idealized one - dimensional ring of radius @xmath0 in which the particles interact via an arbitrary pair - wise interaction . since the total angular momentum commutes with the hamiltonian of the system , the stationary states have energies @xmath1 which are functions of the angular momentum quantum number @xmath2 ; all other quantum numbers are subsumed in the index @xmath3 . bloch showed that these energy eigenvalues take the form e_(l)=+e_(l ) [ spec_e ] where @xmath4 is the total mass of the system containing @xmath5 particles of mass @xmath6 . the first term on the right hand side of eq . ( [ spec_e ] ) is interpreted as the kinetic energy of a rigid ring rotating with angular velocity @xmath7 . the second term , @xmath8 , corresponds to internal excitations of the system ; it has the periodicity property e_(l+n)=e_(l ) . [ e_period ] this implies that the system can find itself in the same internal state for angular momenta that differ from each other by multiples of @xmath9 . in addition , @xmath8 has the inversion property e_(-l)=e_(l ) , [ e_inversion ] which reflects the fact that the energy does not depend on the sense of the angular momentum . the state with the lowest energy for a given @xmath2 will be given the label @xmath10 . in the noninteracting limit , @xmath11 has a local minimum at @xmath12 @xcite ; one expects this property to persist with repulsive interactions . the periodicity of this function then implies that @xmath13 can exhibit local minima at certain multiples of @xmath9 . if and when such minima occur , bloch argued that the system is capable of sustaining persistent currents . conversely , if @xmath13 is not at a local minimum , nonidealities will induce transitions which change the angular momentum and hence the flow of the superfluid around the ring . in sec . [ bloch ] , we extend bloch s analysis to a two - species gas containing @xmath14 particles of type @xmath15 and @xmath16 particles of type @xmath17 . here the term `` species '' can refer either to different kinds of atoms or to atoms distinguished by their hyperfine states . when the masses of the two species are different , we find that the energy can still be written in the form of eq . ( [ spec_e ] ) but in general , @xmath11 is no longer a periodic function of @xmath2 . however , if the masses are equal , @xmath11 is found to have the same periodicity as for the single - species case with @xmath18 . in the case that the mass ratio @xmath19 is a rational number , @xmath11 remains a periodic function of @xmath2 but with a periodicity that differs from @xmath9 . for these special cases , bloch s arguments for the possibility of persistent currents goes through as for the single - species case . for arbitrary mass ratios , @xmath13 may still exhibit a local minimum at some finite value of @xmath2 but there is no general argument which can be used to determine where such a local minimum might occur . we go on to show that bloch s criterion for persistent currents can be phrased in terms of the more familiar landau criterion . for @xmath20 , @xmath11 is periodic and a landau criterion can be formulated at the discrete set of angular momenta @xmath21 , with @xmath22 an integer , where the system can be taken to be in its internal ground state . the landau criterion then imposes a constraint on the spectrum of the elementary excitations with angular momentum @xmath23 and energy @xmath24 . in sec . [ bogoliubov ] , these excitation energies are determined for the two - species system in the bogoliubov approximation . in general there are two bogoliubov modes which are usually phonon - like at long wavelengths . for the case @xmath20 , the landau criterion then suggests that persistent currents may be stable for certain values of @xmath22 . however , if the interaction parameters satisfy a certain relation ( given in sec . [ bogoliubov ] ) , one of the bogoliubov modes has a particle - like dispersion and the landau criterion leads to the conclusion that persistent currents are unstable for all @xmath22 . the above conclusion was arrived at earlier by smyrnakis _ et al . _ @xcite based on an analysis of the mean - field gross - pitaevskii ( gp ) energy functional for the two - species system . with the assumption that all interaction parameters are equal , these authors determine @xmath13 by minimizing the gp energy functional subject to the constraint that the average angular momentum of the system is @xmath2 . although persistent currents are destabilized at @xmath25 , the authors find that @xmath13 can exhibit local minima at non - integral values of @xmath26 . in particular , they show that persistent currents are stable at @xmath27 , provided the interactions are sufficiently strong . furthermore , their analysis leads to the conclusion that persistent currents are unstable for @xmath28 even when the concentration of the minority component is arbitrarily small . this latter conclusion seems at odds with what one might expect in the pure single - species limit ( @xmath29 ) . in sec . [ gp - theory ] , we present the analysis of the gp energy functional in somewhat more detail than was provided by smyrnakis _ et al . _ @xcite this analysis essentially confirms all of their analytical results , however , we find that the information regarding the behaviour of @xmath11 in the vicinity of @xmath30 is not sufficient to establish whether or not persistent currents are actually stable . in fact , a more global analysis of @xmath11 shows that persistent currents can exist when @xmath28 . our work also clarifies how the single - species results are recovered in the @xmath29 limit . in this section we extend bloch s analysis to a two - species system consisting of @xmath14 particles of type @xmath15 and @xmath16 particles of type @xmath17 . the masses of the particles are @xmath31 and @xmath32 . in addition , we assume an idealized one - dimensional ring geometry . the hamiltonian @xmath33 for this system is taken to be @xmath34 where the angular momentum operator of the @xmath35-th particle about the centre of the ring is @xmath36 the index @xmath35 denotes an @xmath15-type particle for @xmath37 and a @xmath17-type particle for @xmath38 . the subscripts on the interaction potential @xmath39 allow for the interactions between the particles to be species - dependent . for the pair - wise interactions assumed , the total angular momentum @xmath40 commutes with the hamiltonian . the stationary states @xmath41 of the hamiltonian can thus be chosen to be simultaneous eigenstates of the total angular momentum . a suitable basis of states can be constructed from the following product states for noninteracting particles : @xmath42 here @xmath43 is an integer and _ m ( ) = . [ s_p_basis ] the wave function in eq . ( [ wave_function_phi ] ) is an eigenfunction of @xmath44 with eigenvalue @xmath45 . it can be written in different ways . one possibility is @xmath46\ ] ] where @xmath47 is the angular momentum per particle in units of @xmath48 and @xmath49 is the mean angular coordinate . the above wave function is identical in form to that of a single species system . by construction , the first exponential in eq . ( [ wave_function_phi_ii ] ) is an eigenfunction of @xmath44 with eigenvalue @xmath50 . the second exponential is a function of the coordinate differences @xmath51 and as such is a zero total angular momentum wave function . properly symmetrized functions are obtained from eq . ( [ wave_function_phi_ii ] ) with the application of the symmetrization operator @xmath52 where @xmath53 the operator @xmath54 permutes the coordinates of the @xmath15 particles , whereas @xmath55 does the same for @xmath17 particles . applying the symmetrization operator to the wave function @xmath56 yields @xmath57 where @xmath58 is a normalized function of the coordinate differences @xmath59 . the functions in eq . ( [ phi_two - species ] ) provide a basis of properly symmetrized @xmath5-particle states , with @xmath60 . the stationary state solutions of @xmath61 with angular momentum @xmath2 will be denoted @xmath62 , where @xmath3 indicates the rest of the quantum numbers . these states can be expanded in terms of the basis functions eq . ( [ phi_two - species ] ) as @xmath63 \tilde \chi_{l\alpha}(\theta_1 , ... , \theta_n),\end{aligned}\ ] ] where the prime on the summation implies the restriction @xmath64 . it is clear from the way @xmath65 is defined that it is a function of the relative angular coordinates @xmath51 . substituting eq . ( [ psi_l ] ) into the schrdinger equation for @xmath66 , we find that @xmath67 satisfies the equation @xmath68 where @xmath69 and @xmath70 since @xmath71 , @xmath72 in eq . ( [ chi_l_se ] ) can be expressed equivalently as @xmath73 we observe that this hamiltonian is in general @xmath2-dependent which has important consequences for @xmath74 . ( [ chi_l_se ] ) must be solved with appropriate boundary conditions . since the wave function @xmath66 is required to be single - valued with respect to each of the angular variables , it satisfies @xmath75 eq . ( [ periodicity_psi ] ) then implies @xmath76\tilde \chi_{l\alpha } ( ... , \theta_i , ... ) \ ] ] for @xmath77 . from this we see that the boundary conditions are periodic as a function of @xmath2 with period @xmath9 . with the basis functions written in the form given in eq . ( [ wave_function_phi_ii ] ) , they are the _ same _ boundary conditions that apply in the single - species case . in the @xmath78 limit , @xmath79 with @xmath80 . in addition , the hamiltonian @xmath72 reduces to @xmath33 since @xmath81 and @xmath71 . as a result , @xmath82 with @xmath83 satisfies the same schrdinger equation and boundary conditions as @xmath67 . this implies that the eigenvalue spectrum for these two functions is identical . as concluded by bloch @xcite , the eigenvalues @xmath8 for the single - component system are then periodic functions of @xmath2 with period @xmath9 . in particular , the ground state energy is given by @xmath84 the same considerations apply to the two - species situation for the special case @xmath20 since @xmath72 in eq . ( [ h_l ] ) also reduces to @xmath33 in this limit and eq . ( [ e_0 ] ) is still valid . in both of the above situations , the periodicity of the eigenvalues @xmath8 means physically that the internal " excitations can be the same for distinct macroscopic flows whose angular momenta differ by some multiple of @xmath9 . however , @xmath85 is no longer periodic when @xmath86 , since the hamiltonian @xmath72 in eq . ( [ chi_l_se ] ) depends explicitly on @xmath2 . an alternative analysis is provided by writing the wave function in ( [ psi_l ] ) as @xmath87 where @xmath88 is the ` centre - of - mass ' angular coordinate defined as @xmath89 and @xmath90\tilde \chi_{l\alpha}(\theta_1 , ... , \theta_n).\ ] ] here and in the following , @xmath91 is equal to @xmath31 for @xmath92 and @xmath32 for @xmath93 ; @xmath94 is the total mass . we observe that the exponential in eq . ( [ psi_l_cm ] ) is still an eigenfunction of @xmath44 with eigenvalue @xmath2 and that @xmath95 is a function of coordinate differences and therefore a zero - angular momentum function . ( [ psi_l_cm ] ) amounts to a separation of the centre - of - mass motion from the internal degrees of freedom . indeed , substitution of eq . ( [ psi_l_cm ] ) into the schrdinger equation for @xmath66 yields @xmath96 where @xmath97 eqs . ( [ tilde_chi_l_se ] ) and ( [ tilde_eps ] ) suggest that @xmath95 and @xmath98 can be viewed , respectively , as the `` internal '' wave function and internal " excitation energy . the boundary conditions imposed on @xmath99 can be derived from eq . ( [ psi_l_cm ] ) and are given by [ tilde_chi_bc ] _ l(,_i+2,)= ( -i2 ) _ l(,_i , ) , where @xmath100 . when @xmath101 , these boundary conditions revert to those of the single - species case where _ l(,_i+2,)=(-i2l ) _ l(,_i , ) . this , together with eq . ( [ tilde_chi_l_se ] ) implies that @xmath102 and @xmath103 . in fact , in this case @xmath8 and @xmath95 coincide with @xmath104 and @xmath67 , respectively . when @xmath105 , @xmath8 is not in general a periodic function of @xmath2 . however , it can be if the boundary conditions in eq . ( [ tilde_chi_bc ] ) remain unaltered when @xmath106 is augmented by some number @xmath107 ( i.e. , @xmath108 ) such that = p , and = q , where @xmath109 and @xmath110 are both integers . this implies that @xmath19 must be equal to the rational number @xmath111 . the lowest possible value of @xmath112 is obtained when @xmath109 and @xmath110 have no common divisor and is then given by = pn_a+qn_b . with this choice of @xmath112 , @xmath8 is a periodic function of @xmath2 with periodicity @xmath113 . in this situation , it is possible to impart a definite angular momentum to the two - species system without altering its internal " state . for two different atomic species , the mass ratio @xmath19 is never strictly a rational number and thus @xmath8 can not be strictly periodic . however , if m_a / m_b p / q + where @xmath114 , one would expect @xmath8 , by continuity , to be quasi - periodic with a periodicity close to @xmath115 . for example , a mixture of @xmath116 ( @xmath15 ) and @xmath117 ( @xmath17 ) has a mass ratio m_a / m_b 2 + 0.07 , in which case the quasi - periodicity of @xmath8 would be @xmath118 . in the rest of this section we discuss the close connection between bloch s argument on persistent currents and landau s criterion for superfluidity . our analysis mainly concerns the single - species and equal - mass two - species systems , where there is strict periodicity for @xmath8 . however , it also applies to the two - species system with unequal masses , insofar as it is a good approximation to regard @xmath11 as quasi - periodic . according to bloch , persistent currents can occur at the angular momenta @xmath119 , for integral @xmath22 , if @xmath13 has a local minimum at @xmath120 . we thus examine the behaviour of @xmath13 in the neighbourhood of @xmath121 . from eqs . ( [ spec_e ] ) and ( [ e_period ] ) one has e_0(l_n+l)&=&+e_0(l_n+ l ) + & = & + _ nl+e_0(l ) , [ e_0(l ) ] where @xmath122 is the angular velocity of the centre of mass of the system at @xmath121 . this expression for the energy is analogous to the expression obtained via a galilean transformation for a homogeneous system in which an excitation is produced in the rest frame of the superfluid @xcite . to make this correspondence evident , we define the velocity @xmath123 and write the energy in eq . ( [ e_0(l ) ] ) as e_0(l_n+l)= m_t v_n^2 + ( ) v_n+ e_0(l ) . the first term on the right hand side is identified as the kinetic energy of the superfluid moving with velocity @xmath124 . likewise , the last term is identified as the energy of a stationary superfluid containing an excitation with momentum " @xmath125 . it should be noted , however , that the analogy is not complete since for a homogeneous system the superfluid velocity @xmath124 can take arbitrary values whereas for the ring geometry the angular momentum is restricted to the discrete values @xmath121 . with this correspondence in mind , we take @xmath126 to be the energy of the system with a single quasi - particle excitation with angular momentum @xmath127 and energy @xmath24 , i.e. , e_0(l)=e_0(0)+(m ) . we thus have e_0(l_n+l)=e_0(l_n)+ ( m)+ _ n m. the stability of the state with energy @xmath128 is then assured if the excitations lead to an increase in energy . in other words , the system will sustain persistent currents at @xmath121 for an arbitrary excitation of the system if ( m)+_n m>0 [ lc1 ] for all @xmath129 . since @xmath130 , the left hand side has a minimum for negative values of @xmath129 and we thus require _ n < ( ) _ min . [ lc2 ] we have thus shown that bloch s argument for persistent currents in the one - dimensional ring geometry naturally leads to the more familiar landau criterion for superfluidity . if @xmath24 has a positive curvature as a function of @xmath129 , which precludes a roton - like minimum , the inequality in eq . ( [ lc2 ] ) can be replaced by _ n<. [ lc3 ] it is clear from this expression that the inequality must eventually fail when @xmath22 exceeds some critical value @xmath131 . the landau criterion derived in the previous section focuses attention on the elementary excitations of the system . in this section , we obtain these excitations for a two - species gas in a one - dimensional ring geometry in the bogoliubov approximation . we then apply eq . ( [ lc3 ] ) to discuss persistent currents at integer values of angular momentum per particle for an equal - mass two - species system . in the following , we assume that the particles interact via contact interactions with strengths @xmath132 , where @xmath133 specify the species . using the single - particle basis in eq . ( [ s_p_basis ] ) , the hamiltonian in eq . ( [ two - comp hamiltonian ] ) can be written in the second - quantized form h & = & _ s_m _ s a^_s , m a_s , m+_s , s_m , m,n a^_s , ma^_s,n - ma_s,ma_s , n - m , where @xmath129 is the angular momentum quantum number and @xmath134 . assuming both species to be bose - condensed in the @xmath135 state , the corresponding bogoliubov hamiltonian can be written as [ h_bog_0 ] h_bog&=&_s , s g_ss + _ s_m0 + & & 1.45truein + _ ss_m0g_ss , where @xmath136 . the diagonalization of a hamiltonian similar to eq . ( [ h_bog_0 ] ) for a three - dimensional system was carried out in @xcite . here we present a different method of determining the bogoliubov quasiparticle operators . this is done in three steps . first , we perform a bogoliubov transformation for each of the species treated individually . the transformation is defined by [ bog_tran_0 ] & & a_s , m = u^(0)_s , m_s , m - v^(0)_s , m ^_s ,- m + & & a_s ,- m = u^(0)_s , m _ s ,- m - v^(0)_s , m^_s , m , with [ bare_bog_amp ] ( u^(0)_s , m ) ^2= ( + 1 ) = , ( v^(0)_s , m)^2= ( -1 ) = , where e_s=. [ bog_freq ] @xmath137 is the bogoliubov excitation energy for independent components . substituting eq . ( [ bog_tran_0 ] ) into eq . ( [ h_bog_0 ] ) and dropping all constant terms , we obtain [ h_bog_1 ] h_bog=_s_m0e_s^_s , m _ s , m + _ ss_m0 g , where @xmath138 . the second term in this hamiltonian describes the coupling between the bogoliubov excitations defined for each of the species . it is convenient to write the hamiltonian ( again to within a constant ) in the matrix form [ h_bog_1_matr ] h_bog=_m > 0^_m_m , where & & _ m(_a , m ^_a ,- m_b , m^_b ,- m ) ^t + & & ^_m(^_a , m _ a ,- m^_b , m_b ,- m ) and = ( cccc e_a & 0 & & + 0 & e_a & & + & & e_b & 0 + & & 0 & e_b ) . to complete the diagonalization process we introduce the following transformations & & _ s , m=^(+)_s , m _ + , m-^(+)_s , m^_+,-m+^(- ) _ s , m_-,m-^(-)_s , m^_-,-m + & & _ s ,- m=^(+)_s , m _ + , -m-^(+)_s , m^_+,m+^(- ) _ s , m_-,-m-^(-)_s , m^_-,m , where the amplitudes are chosen to be real . the hamiltonian is reduced to the diagonalized form [ h_bog_2 ] h_bog=_m0e_+^_+,m _ + , m + _ m0e_-^_-,m_-,m , if the amplitudes satisfy the matrix equation [ bogo_equ ] _ z_= _ _ with the normalization condition [ normalization ] ^t__z_=1 . here , @xmath139 and the matrix @xmath140 is defined as _ z= ( cccc 1&0&0&0 + 0&-1&0&0 + 0&0&1&0 + 0&0&0&-1 ) . it should be noted that eqs . ( [ bogo_equ ] ) and ( [ normalization ] ) guarantee that the bose commutation relations of the new operators @xmath141 and @xmath142 are preserved . the bogoliubov excitation energies @xmath143 are determined by the characteristic equation [ char_equ1 ] det(_z - e ) = 0 , which yields @xmath144 this quadratic equation in @xmath145 has the two roots @xcite @xmath146 the dispersion of these modes is ` phonon - like ' for small @xmath129 ( @xmath147 ) and ` particle - like ' for large @xmath129 ( @xmath148 ) . the upper branch has the higher sound speed and evolves continuously into @xmath149 where @xmath150 signifies the smaller of the two masses . the bogoliubov excitations of the two - component system are dynamically stable provided @xmath151 . since only @xmath152 can become negative , the criterion for dynamic stability is @xmath153 in view of eqs . ( [ char_equ1 ] ) and ( [ char_equ2 ] ) , this is equivalent to the condition ( _ z ) = det ( ) > 0 , since @xmath154 . using the definition of @xmath155 in eq . ( [ bog_freq ] ) and defining _ ss g_ss , [ gamma ] eq . ( [ dynamic stability condition ] ) becomes @xmath156 for repulsive interactions , this inequality is satisfied for all @xmath129 if it is satisfied for @xmath157 . this limiting case gives the condition @xmath158 a criterion of this form was obtained in @xcite for @xmath159 but is also seen to be valid for @xmath86 with the definition of @xmath160 given in eq . ( [ gamma ] ) . to complete our discussion of the bogoliubov excitations we present the results for the bogoliubov amplitudes . it is straightforward to show that eqs . ( [ bogo_equ ] ) and ( [ normalization ] ) lead to ( ^()_s , m)^2&= & [ interm_bog_amp_u ] + ( ^()_s , m)^2&=&. [ interm_bog_amp_v ] where @xmath161 denotes the species complementary to @xmath162 . finally , the relation of the original creation and annihilation operators to the bogoliubov quasiparticle operators is defined via & & a_s , m = u^(+)_s , m _ + , m - v^(+)_s , m^_+,-m+u^(-)_s , m _ -,m - v^(-)_s , m^_-,-m + & & a_s ,- m = u^(+)_s , m _ + , -m - v^(+)_s , m^_+,m+u^(-)_s , m _ -,-m - v^(-)_s , m^_-,m . these amplitudes can be obtained from eq . ( [ bare_bog_amp ] ) and eqs . ( [ interm_bog_amp_u ] ) and ( [ interm_bog_amp_v ] ) with the result [ bog_amp ] ( u^()_s , m ) ^2= + ( v^()_s , m)^2=. it can be shown that these expressions are equivalent to those given in ref . @xcite in the one - dimensional limit . the amplitudes can be used to evaluate the mode density fluctuations @xmath163 of each species . we find that the @xmath15 and @xmath17 density fluctuations are _ in - phase _ for the ( + ) mode and _ out - of - phase _ for the ( ) mode . we now make use of these results in eq . ( [ lc3 ] ) to investigate the possibility of persistent currents at the angular momenta @xmath164 for the equal - mass system . the lower of the two branches in eq . ( [ bog_spec ] ) is the branch relevant to determining the stability of the current . for @xmath101 , the energy of this branch reads e_-(m)=. [ lower_branch ] according to eq . ( [ lc3 ] ) , the stability of persistent currents at @xmath121 requires < . this inequality is satisfied if the following two inequalities ( _ aa- ) ( _ bb-)&>&^2_ab [ cri_g1 ] + _ aa+_bb&>&4n^2 - 1 , [ cri_g2 ] are simultaneously satisfied . in the limit @xmath165 , we have two independent components and we observe that the inequalities are satisfied if @xmath166 satisfies _ min > ( 2n+1)(2n-1 ) . [ gamma_crit ] for @xmath167 this gives the critical interaction strength @xmath168 which is the value quoted in ref . @xcite . for the two - species system with equal masses , the inequalities in eqs . ( [ cri_g1 ] ) and ( [ cri_g2 ] ) can usually be satisfied for suitable choices of the interaction parameters , implying the possible stability of persistent currents at any @xmath121 . the only exception occurs when _ aa_bb=^2_ab , or equivalently u_aau_bb = u_ab^2 . [ u_ij_equality ] in this case , the coefficient of @xmath169 in eq . ( [ lower_branch ] ) vanishes and the lower branch has a free particle dispersion which destabilizes persistent currents for any value of @xmath22 . this conclusion was arrived at earlier by smyrnakis _ et al . _ @xcite for the special case @xmath170 ; we see here how it follows from the landau criterion for the more general relation in eq . ( [ u_ij_equality ] ) . however , this does not preclude the possibility of persistent currents at non - integral values of angular momentum per particle . in the next section we reconsider the problem from the point of view of mean - field theory , following closely the work of smyrnakis _ et al . _ the analysis in this section is based on the mean - field gross - pitaevksii energy functional for the two - component system in the ring geometry : @xmath171 = \int_{0}^{2\pi } d\theta \left ( \frac{n_a\hbar^2}{2m_ar^2 } \left| \frac{d \psi_a}{d \theta } \right|^2 + \frac{n_b\hbar^2}{2m_br^2 } \left| \frac{d \psi_b}{d \theta } \right|^2 \right ) \nonumber \\ & & \hspace{18mm}+\hspace{1 mm } \frac{1}{2}u_{aa}n_a^2\int_{0}^{2\pi } d\theta |\psi_a|^4 + \frac{1}{2}u_{bb}n_b^2\int_{0}^{2\pi } d\theta |\psi_b|^4 + u_{ab}n_a n_b\int_{0}^{2\pi } d\theta functions @xmath172 and @xmath173 are normalized as @xmath174 as discussed in the previous section , bloch s argument allows for persistent currents at integral values of @xmath175 when @xmath101 except when eq . ( [ u_ij_equality ] ) is true . here , following smyrnakis _ et al . _ @xcite , we consider the special case @xmath176 . in units of the energy @xmath177 , eq . ( [ 4.2_energy_1 ] ) becomes @xmath178 = \int_{0}^{2\pi } d\theta \left ( x_a \left| \frac{d \psi_a}{d \theta } \right|^2 + x_b \left| \frac{d \psi_b}{d \theta } \right|^2 \right ) \nonumber \\ & & \hspace{18mm}+ \hspace{1 mm } x_a^2\pi\gamma \int_{0}^{2\pi } d\theta + \hspace{1mm}2x_ax_b\pi\gamma \int_{0}^{2\pi}d\theta \label{gp_functional}\end{aligned}\ ] ] where @xmath179 , @xmath180 are the relative fractions of the two species in the system and @xmath181 is a dimensionless interaction parameter . for definiteness , we take @xmath182 . the objective is to minimize the energy functional in eq . ( [ gp_functional ] ) with the constraint that that the average value of the total angular momentum has a fixed value @xmath183 . this is achieved by expanding the condensate wave functions as @xmath184 where the basis functions @xmath185 are given in eq . ( [ s_p_basis ] ) . the normalization of the wave functions requires @xmath186 such a superposition implies that the wave functions are in general nonuniform around the ring . in addition , the expansion coefficients @xmath187 and @xmath188 must satisfy the angular momentum constraint @xmath189 @xmath190 ( @xmath191 ) represents the average angular momentum in units of @xmath48 of an @xmath15 ( @xmath17)-species particle . the minimization of the energy with respect to the expansion coefficients in eqs . ( [ 4.2_wfa ] ) and ( [ 4.2_wfb ] ) was first considered by smyrnakis _ et al . _ it will be clear from the following that much of our analysis closely follows theirs . however , we have expanded on their discussion in order to obtain a number of results that are not given explicitly in their paper . substituting the wave functions in eqs . ( [ 4.2_wfa ] ) and ( [ 4.2_wfb ] ) into eq . ( [ 4.2_energy_2 ] ) , we obtain @xmath192 according to bloch s argument , @xmath193 should exhibit the periodicity @xmath194 where @xmath22 is an integer . this periodicity is ensured if the expansion coefficients satisfy the periodicity conditions @xmath195 the fact that @xmath196 must remain unchanged when the wave functions @xmath197 with angular momenta @xmath198 are used to evaluate the energy functional leads to the relations @xmath199 these two conditions are the mean - field counterparts of eqs . ( [ e_period ] ) and ( [ e_inversion ] ) . the function @xmath193 is the central quantity determining the possibility of persistent currents and its detailed evaluation is taken up next . to begin , we consider wave functions @xmath172 and @xmath173 containing only two components , that is , @xmath200 the coefficients @xmath187 and @xmath188 are normalized according to eq . ( [ 4.2_normalization ] ) and the angular momentum constraint becomes @xmath201 expressing the complex coefficients in the form @xmath202 the gp energy becomes @xmath203 where @xmath204 . the choice of @xmath205 which minimizes @xmath206 is @xmath207 and we then have @xmath208 the lowest possible value of this energy is @xcite @xmath209 which occurs for @xmath210 this relation , together with the normalization and angular momentum constraints , yields the coefficients @xmath211 these quantities are positive provided @xmath212 is in the range @xmath213 or @xmath214 . assuming the validity of eq . energy simplified ] ) for @xmath212 in these ranges , we see that @xmath196 does not have a local minimum at @xmath215 . thus , persistent currents are not possible at @xmath216 , and by virtue of the periodicity of @xmath193 , at all integral values of @xmath212 . these conclusions are consistent with our earlier discussion based on the landau criterion ; the validity of eq . ( [ u_ij_equality ] ) implies the existence of particle - like excitations and the absence of persistent currents at integral values of @xmath212 . although eq . energy simplified ] ) was obtained for the simplest possible variational wave function , it in fact is exact when @xmath212 is restricted to the above ranges @xcite . to show this , we consider normalized wave functions of the form @xmath217 where @xmath172 and @xmath173 are defined by eqs . ( [ 4.2_wfa_01 ] ) and ( [ 4.2_wfb_01 ] ) with the coefficients given in eqs . ( [ 4.2_coeff_csq])-([4.2_coeff_dsq ] ) . if the deviations are expressed in the form @xmath218 the angular momentum constraint in eq . ( [ ang_mom_constraint ] ) leads to @xmath219 we next observe that the density @xmath220 is in fact uniform , that is , @xmath221 . using these results , the energy is found to be given by @xmath222 = \bar e_0(l ) + x_a\sum_{m}(m^2-m)|\delta c_m|^2+x_b\sum\limits_{m}(m^2-m)|\delta d_m|^2 + \frac{\pi\gamma}{n^2}\int_0^{2\pi } d\theta |\delta n(\theta)|^2,\ ] ] where @xmath223 . we thus see that @xmath224 > \bar e_0(l)$ ] , implying that the state defined by eqs . ( [ 4.2_wfa_01 ] ) and ( [ 4.2_wfb_01 ] ) is indeed the ground state of the system for the assumed ranges of the angular momentum . it should be noted that this result depends crucially on the assumption of equal interaction parameters between all components . the weaker condition in eq . ( [ u_ij_equality ] ) still precludes the possibility of persistent currents at integral values of @xmath212 , but the energy does not have the simple form shown in eq . energy simplified ] ) . we next analyze the energy for @xmath225 . in particular we consider the situation when @xmath212 is close to @xmath226 , that is @xmath227 , where @xmath228 is a small positive quantity . for @xmath229 we see from eqs . ( [ 4.2_coeff_csq])-([4.2_coeff_dsq ] ) that @xmath230 as @xmath228 increases from zero , we therefore expect deviations from these limiting values and additional components in the expansion of the @xmath172 and @xmath173 wave functions . to be specific , we consider the three - component wave functions @xmath231 we anticipate that @xmath232 , @xmath233 , @xmath234 and @xmath235 are all of order @xmath228 . with this assumption , the energy to first order in @xmath228 is found to be @xmath236,\end{aligned}\ ] ] where we have defined the phase angles @xmath237 , @xmath238 and @xmath239 . this energy is an extremum with respect to the phase angles if they are all 0 or @xmath207 . if we choose them arbitrarily to be 0 , we obtain @xmath240 ^ 2,\ ] ] which must now be minimized with respect to the coefficients @xmath241 , @xmath242 , @xmath243 and @xmath244 subject to the angular momemtum constraint @xmath245 ) to eliminate @xmath246 from eq . e new 2 ] ) , and introducing a lagrange multiplier @xmath247 to account for the angular momentum constraint , the functional to be minimized is @xmath248 ^ 2 \nonumber \\ & & \hspace{27.5mm}+ \hspace{1mm}\lambda \big [ x_a\left(1- |c_0|^2 + where the variations of the coefficients are now unconstrained . this variation leads to the results @xmath249 where the lagrange multiplier @xmath247 is the solution of the cubic equation @xcite @xmath250 the roots of this equation are to be determined for @xmath251 and @xmath252 . in fig . [ fig : cubic_gamma_2 ] , @xmath253 is plotted for @xmath254 , 0.5 and 1 and for @xmath255 ; fig.[fig : cubic_gamma_8 ] is a similar plot for @xmath256 . for @xmath254 , @xmath257 , which has the roots @xmath258 and @xmath259 . for @xmath260 , @xmath261 $ ] , which has the roots @xmath262 and @xmath263 . the latter two values are the lagrange multipliers in the single - species limit as obtained from the minimization of eq . lagrange 00 ] ) for @xmath264 . since the term @xmath265 in @xmath253 simply shifts the curves in figs . [ fig : cubic_gamma_2 ] and [ fig : cubic_gamma_8 ] vertically , it is clear that there are always three real roots for the physical range of @xmath266 values . for any positive value of @xmath267 , one root is always less than @xmath268 , a second lies in the range @xmath269 ( more precisely in the range @xmath270 ) and a third in the range @xmath271 . substituting the coefficients given in eq . lambda cd1 ] ) into eq . ang mom new ] ) we find @xmath272}.\ ] ] it is clear from this expression that the @xmath273 root makes @xmath232 negative . this root is therefore physically inadmissible and only the positive @xmath247 roots are relevant . c0 as eps ] ) together with eq . lambda cd1 ] ) can be used in eq . e new 2 ] ) to evaluate the energy . one finds the remarkably simple result @xmath274 we now see that the smaller of the two positive @xmath247 roots gives the lowest possible energy . this thus identifies the root in the range @xmath275 as the one that is physically relevant @xcite . for @xmath247 in this range we observe that the ratios in eq . lambda cd1 ] ) are negative , indicating that the phases in eq . ( [ e_3_component ] ) were chosen incorrectly . the proper phases are @xmath276 , @xmath277 and @xmath278 . the criterion for the existence of persistent currents at @xmath229 used in ref . @xcite is that the slope of @xmath196 in eq . e new 5 simple ] ) at @xmath279 is negative , i.e. , @xmath280 . the critical condition is thus @xmath281 , which from eq . ( [ 4.2 f of lambda ] ) gives the critical interaction strength @xcite @xmath282 in the @xmath283 limit this reduces to @xmath284 which is the value obtained at @xmath216 for the single - species system . to obtain the critical coupling at @xmath285 , where @xmath286 , 2 , .. , we write @xmath287 and use the fact that @xmath193 is periodic . the slope at @xmath288 is thus found to be @xmath289 if the root in the range @xmath290 is used , the slope can not be zero for any @xmath291 . this is the basis of the claim made in ref . @xcite that persistent currents are not possible for @xmath28 ; seemingly , an arbitrarily small amount of the minority component @xmath17 has a profound effect on the possibility of persistent currents . for the single - species case , the energy is given by eq . e new 5 simple ] ) with @xmath292 , but the appropriate value of @xmath247 is @xmath293 , which is not bounded as a function of @xmath267 . using this value in eq . denergy prime ] ) , one finds that persistent currents are possible for all @xmath22 in this case , with a critical interaction strength of @xmath294 . this is the result found earlier ( eq . ( [ gamma_crit ] ) ) using the landau criterion . this comparison indicates an inconsistency . on the one hand , eq . denergy prime ] ) does allow for persistent currents for @xmath28 in the single - species limit if the appropriate value of @xmath247 is used . however , the two - species analysis requires that the root in the range @xmath295 be used , which precludes the possibility of persistent currents for @xmath28 for any nonzero value of @xmath296 . since the energy functional in eq . ( [ 4.2_energy_2 ] ) reduces to the single - species case when @xmath297 , it would appear that taking the @xmath298 limit of the two - species analysis is problematic . in order to explain this discrepancy it is useful to examine the behaviour of the coefficients in eqs . lambda cd1 ] ) and ( [ 4.2 . c0 as eps ] ) in the @xmath299 limit in more detail . these coefficients are determined by the root @xmath247 that lies in the range @xmath300 . if @xmath301 , the limiting value of this root for @xmath302 is @xmath303 . this is the @xmath247 value for the single - species case . thus for this range of @xmath267 , one recovers the single - species values for all the coefficients . however , for @xmath304 , the root in the range @xmath300 has the limiting value of 2 which is less than the @xmath305 root . the limiting values of the coefficients do not correspond to the single - species values in this case . the distinction between @xmath301 and @xmath306 is revealed more clearly by plotting the coefficients in these two cases as a function of @xmath296 . we observe that the angular momenta carried by each of the species is given by @xmath307 the change in angular momentum as @xmath212 is reduced from @xmath226 is associated with the transfer of weight from one angular momentum component to another . for example , for the @xmath15 species , the transfer takes place from the @xmath157 state to the @xmath135 or @xmath308 states , with respectively , a decrease or increase in angular momentum . for the @xmath17 species , the transfer takes place from the @xmath135 state to the @xmath309 and @xmath157 states . of interest is the relative magnitude of the angular momentum change @xmath310 that is associated with each angular momentum component . we therefore define the ratios @xmath311 where for example , @xmath312 . these ratios represent the fraction of the angular momentum change @xmath313 attributable to each of the angular momentum components . in fig . [ fig : coeff_gamma_2 ] we plot these ratios as a function of @xmath296 for @xmath314 ; fig . [ fig : coeff_gamma_8 ] gives similar plots for @xmath315 . for @xmath316 , we see that species @xmath17 carries a relatively small contribution of the angular momentum change . this contribution vanishes in the @xmath317 limit and the situation reverts to that of the single species which , as discussed above , is generally the case for @xmath301 . the situation for @xmath318 , however , is quite different . [ fig : coeff_gamma_8 ] for @xmath256 shows that the angular momentum change is carried entirely by the @xmath309 component of the @xmath17 species in the @xmath317 limit . the reason for this surprising result is that the relevant @xmath247 root approaches 2 for @xmath317 when @xmath304 . c0 as eps ] ) then gives @xmath319 and from eq . lambda cd1 ] ) we find @xmath320 for @xmath317 , i.e. @xmath321 in this limit . the divergence of @xmath234 as @xmath317 is indicating that the result can only be valid for a decreasingly smaller range of @xmath228 since the normalization @xmath322 must be preserved . in other words , the energy @xmath196 , as given by eq . e new 5 simple ] ) , is meaningful in an interval of @xmath212 of decreasing size as @xmath298 . the above results call into question any conclusion regarding the possibility of persistent currents at higher angular momenta when @xmath226 approaches 1 . in this limit , a more global perspective regarding the behaviour of the energy as a function of @xmath212 in the interval @xmath323 is required . we now give a general argument for the possibility of persistent currents at @xmath28 based on the assumption of continuity of the gp energy as a function of @xmath296 . to exhibit this dependence we write @xmath324 and consider this function in the limit of small @xmath296 . in particular , we have @xmath325 and @xmath326 where @xmath327 is the energy of the single - species system . the assumption of continuity implies that @xmath328 and @xmath329 approach 0 as @xmath298 . we then have @xmath330 . by choosing @xmath331 , @xmath332 will have some fixed _ positive _ value . thus , we can say that @xmath333 for @xmath296 sufficiently small . since @xmath334 , we conclude that @xmath324 must have a local minimum between @xmath335 and @xmath336 . this argument can be used for any @xmath22 and shows that persistent currents must be stable in the vicinity of @xmath337 if @xmath296 is sufficiently small and @xmath267 is sufficiently large . although it is difficult to evaluate @xmath324 for arbitrary @xmath212 , the above general argument can be illustrated quantitatively by evaluating the energy at @xmath338 . to do so , it is sufficient to assume four - component wave functions of the form @xmath339 substituting these wave functions into eq . ( [ 4.2_energy_2 ] ) , we have @xmath340 and @xmath341 , with analogous relations for the @xmath188 amplitudes . these relations reduce the number of variational parameters by half . we have in particular @xmath342 using these definitions , the normalization constraints reduce to @xmath343 furthermore , the angular momentum of each species is given by @xmath344 we thus see that normalization ensures that the total angular momentum has the required value of 1/2 . using these results , the expression for the energy becomes @xmath345 + x_b^2\gamma\big[u^4+v^4 + 8u^2v^2 + 4u^3v\cos\xi\big ] \nonumber \\ & & \hspace{2.3mm}+\hspace{1mm}x_ax_b\gamma\big [ 8xyuv \left \ { \cos(\theta-\beta+\xi)+ \cos(2\theta-\beta+\xi)\right \ } + \hspace{1mm}4xyu^2\cos(\theta-\beta ) + 4x^2uv\cos(\theta+\xi ) \big . \nonumber \\ \big.&&\hspace{2.5cm}+\hspace{1mm}2x^2u^2\cos\theta + 2y^2v^2\cos(3\theta-2\beta+2\xi)\big],\end{aligned}\ ] ] where we have defined the phase angles @xmath346 , @xmath347 and @xmath348 . we see that the energy depends on these three independent phases and the two amplitudes @xmath349 and @xmath350 . it clearly reduces to the single - species result in the @xmath317 limit . for @xmath292 , the energy is minimized for @xmath351 and a value of @xmath349 which is close to @xmath352 .. we do not expect this conclusion to change when @xmath226 is close to , but not exactly equal to 1 . for these values of @xmath226 , the term in eq . ( [ 4.2.2 energy final ] ) proportional to @xmath353 is small and can be neglected . setting @xmath351 , the energy is approximately @xmath354 \nonumber \\ & & \hspace{2.3mm}+\hspace{1mm}x_ax_b\gamma\big [ -8xyuv \left \ { \cos(\theta+\xi)+ \cos(2\theta+\xi)\right \ } -\hspace{1mm}4xyu^2\cos\theta + 4x^2uv\cos(\theta+\xi ) \big . \nonumber \\ \big.&&\hspace{2.5cm}+\hspace{1mm}2x^2u^2\cos\theta + 2y^2v^2\cos(3\theta+2\xi)\big ] , \label{e_0_approx}\end{aligned}\ ] ] from this we see that the phases @xmath355 and @xmath356 only appear in the last term proportional to @xmath296 . it is clear that @xmath357 is stationary with respect to these phases when they take the values 0 and @xmath207 . to explore the various possibilities , we define the function @xmath358 - 4xyu^2 \cos\theta \\ & & \hspace{0.05cm}+\hspace{1 mm } 4x^2uv\cos(\theta+\xi ) + 2x^2u^2\cos\theta + 2y^2v^2\cos(3\theta+2\xi),\end{aligned}\ ] ] which is the quantity multiplying @xmath359 in eq . ( [ e_0_approx ] ) . this function is tabulated in table [ table ] for various values of @xmath356 and @xmath355 . .the function @xmath360 defined in eq . ( [ 4.2.2 f function ] ) tabulated for various values of @xmath356 and @xmath355 . [ cols="<,<,<",options="header " , ] from this table it is clear that @xmath361 , @xmath362 will give a lower energy than @xmath363 , @xmath362 . for @xmath361 , @xmath364 we have @xmath365 since @xmath226 is close to 1 , eq . ( [ e_0_approx ] ) is minimized for a value of @xmath349 close to @xmath352 which is much larger than @xmath366 . this implies that any minima of the function @xmath367 will occur for _ negative _ values of @xmath350 . but @xmath350 must be positive ( recall @xmath368 ) , so this case must be rejected . finally , for @xmath363 , @xmath364 , we have @xmath369 the same argument implies that minima of @xmath370 must occur at _ positive _ @xmath350 . we are thus left with the two possibilities @xmath371 , @xmath372 and @xmath373 and @xmath374 . a comparison of the contour plots of @xmath375 and @xmath376 shows that the latter is the one that provides the lowest energy . for @xmath377 and @xmath378 , @xmath376 is minimized for @xmath379 and @xmath380 . the value of @xmath381 found here is close to the value of 0.696 found for @xmath382 . not surprisingly , the @xmath383 coefficients are close to the values obtained in the single - species limit . we will now use the value of @xmath384 to show that persistent currents are possible for @xmath385 . to be specific , we consider @xmath386 with @xmath387 . using the periodicity of @xmath388 , we have @xmath389 at @xmath229 , eq . ( [ 4.2 . energy simplified ] ) gives @xmath390 . we then find that @xmath391 for @xmath377 . as explained earlier , this value is exact within the mean - field analysis . we next use eq . ( [ 4.2.2 . e0 l ] ) to obtain @xmath392 in fig . [ fig : e_0_vs_gamma ] we show the behaviour of @xmath393 as a function of @xmath267 for @xmath394 . we see that @xmath395 becomes larger than @xmath396 at a value of @xmath397 . this implies the existence of a local minimum in the range @xmath398 and hence the possibility of persistent currents . the value @xmath397 is clearly an upper bound to @xmath399 for this value of @xmath226 . the approximate behaviour of @xmath196 as a function of @xmath212 can be obtained by generating approximations to @xmath400 . for @xmath401 and @xmath402 , @xmath403 . from eq . e new 5 simple ] ) we have @xmath404 . the simplest approximation to @xmath193 in the range @xmath405 consistent with this information is @xmath406 an improved approximation is a fit that reproduces the value of @xmath193 at @xmath338 . it takes the form @xmath407 where @xmath408 $ ] . a third approximation ignores the information about the slope of @xmath388 at @xmath229 but includes the value at @xmath409 . this approximation gives @xmath410 where @xmath411 . these various approximations are plotted in fig . [ fig : epsilon ] for @xmath412 . we expect the correct variation of @xmath193 to be bounded by the @xmath413 and @xmath414 curves ; for @xmath415 , @xmath193 should be closer to the @xmath413 curve but for @xmath416 it should be closer to the @xmath414 curve . we note that @xmath414 must give the correct behaviour in the @xmath302 limit . these different approximations can be used to determine corresponding approximations to @xmath417 , which is plotted in fig . [ fig : energy ] in the range @xmath418 for @xmath419 . the red curve based on @xmath420 does not show a local minimum at @xmath421 as predicted by considerations of the slope of @xmath196 at this point . on the other hand , the black curve based on @xmath413 which includes the information about @xmath422 shows a local minimum below @xmath421 and demonstrates that persistent currents should be possible for @xmath212 between 1.5 and 2 . regarding persistent currents at @xmath423 , the critical interaction strength according to eq . . gamma cr ] ) for @xmath394 is @xmath424 . for this value of @xmath267 , @xmath425 is actually quite close to to the true value @xmath422 as determined by the four - component wave function analysis . thus the prediction of the critical interaction strength based on the slope of @xmath417 remains quite accurate in this case . however , it is clear that the slope of @xmath206 calculated at @xmath288 , although correct , is not sufficient to provide a criterion for the existence of persistent currents for @xmath291 . we have also extended the analysis to slightly smaller values of @xmath226 and arrive at similar conclusions . however , increasingly larger values of @xmath267 are then required to achieve a local minimum between @xmath426 and @xmath427 . we finally mention the behaviour of @xmath428 when @xmath429 . in this limit the expression for @xmath206 given in eq . energy simplified ] ) is correct for _ all _ @xmath212 and gives in particular @xmath430 . this value is reproduced by eq . ( [ 4.2.2 energy final ] ) at @xmath429 irrespective of the phases @xmath431 and @xmath356 since the minimum occurs for @xmath362 and @xmath432 , where all the @xmath431 and @xmath356 dependent terms have no effect since @xmath433 . we note that the minimizing value of @xmath355 is the same as in the @xmath302 limit and anticipate that this will remain true for intermediate values of @xmath226 between @xmath429 and @xmath292 . however , a more careful analysis of eq . ( [ 4.2.2 energy final ] ) would be required to confirm this and to determine the remaining variational parameters that minimize the gp energy . in this paper we have extended to the two - species system bloch s original argument regarding the possibility of persistent currents in the idealized one - dimensional ring geometry . strict periodicity of the energy @xmath8 defined in eq . ( [ spec_e ] ) is found to arise when the mass ratio @xmath19 is a rational number . by making a connection to the landau criterion for the special case @xmath159 , we show that persistent currents are in general possible at the discrete set of total angular momenta @xmath434 , except when the interaction parameters satisfy the condition in eq . ( [ u_ij_equality ] ) . the underlying reason for this limitation is the existence of excitations with a particle - like dispersion . this conclusion is consistent with the predictions of a mean - field analysis based on the gp energy functional . a detailed analysis of the gp energy in the vicinity of @xmath435 , first carried out by smyrnakis _ et al . @xcite , indicates that persistent currents are possible at this angular momentum per particle if the interaction parameter exceeds the critical value given in eq . . gamma cr ] ) . these authors go on to claim that persistent currents can not arise for @xmath28 in the two - species system . however , a more detailed analysis of the global behaviour of the gp energy demonstrates that this conclusion is not valid . quite generally , the properties of the two - species system evolve continuously to those of the single - species system as the concentration of the minority component is reduced . it would of course be of interest to verify these theoretical predictions experimentally . the recent experimental realization of toroidal bose - einstein condensates @xcite would suggest that experiments on two - species systems may soon be feasible . 99 f. bloch , phys . rev . a * 7 * , 2187 ( 1973 ) . j. wilks , _ an introduction to liquid helium _ , ( clarendon press , oxford , 1970 ) . e. m. lifshitz andl . p. pitaevskii , _ statistical physics , part 2 _ , ( butterworth - heinemann , oxford , 1980 ) . j. smyrnakis , s. bargi , g. m. kavoulakis , m. magiropoulos , k. krkkinen , and s. m. reimann , phys . * 103 * , 100404 ( 2009 ) . p. tommasini , e. j. v. de passos , a. f. r. de toledo piza , m. s. hussein and e. timmermans , phys . a * 67 * , 023606 ( 2003 ) . p. ao and s. t. chui , j. phys . b * 33 * , 535 ( 2000 ) . c. j. pethick and h. smith , _ bose - einstein condensation in dilute gases _ , 2nd ed . ( cambridge , cambridge , 2008 ) . c. ryu , m. f. andersen , p. clad , vasant natarajan , k. helmerson , and w. d. phillips , phys . lett * 99 * 260401 ( 2007 ) . a. ramanathan , k. c. wright , s. r. muniz , m. zelan , w. t. hill iii , c. j. lobb , k. helmerson , w. d. phillips , and g. k. campbell , phys . rev * 106 * , 130401 ( 2011 ) .

we discuss the chiral phase diagram in the parameter space of lattice qcd with minimal - doubling fermions , which can be seen as lattice fermions with flavored chemical potential terms . we study strong - coupling lattice qcd with the karsten - wilczek formulation , which has one relevant parameter as well as gauge coupling and a mass parameter . we find a nontrivial chiral phase structure with a second - order phase transition between chiral symmetric and broken phases . to capture the whole structure of the phase diagram , we study the related lattice gross - neveu model . the result indicates that the chiral phase transition also exists in the weak - coupling region . from these results we speculate on the- chiral phase diagram in lattice qcd with minimal - doubling fermions , and discuss their application to numerical simulations

1206.0969

the famous no - go theorem @xcite states that lattice fermion actions with chiral symmetry , locality and other common features must produce massless degrees of freedom in multiples of two in a continuum limit . this is contradictory with a phenomenological fact that there exist only three quarks with masses below the qcd scale . by now several fermion constructions to bypass the no - go theorem have been developed , although all of them have their individual shortcomings : the explicit chiral symmetry breaking with the wilson fermion approach @xcite results in an additive mass renormalization , which in turn requires a fine - tuning of the mass parameter for qcd simulations . domain - wall @xcite and overlap fermions @xcite produce a single fermion mode by modifying the definition of chiral symmetry or introducing the momentum - dependent chiral charge , although they lead to rather expensive simulations algorithms .. these approaches attempt to realize single fermionic degrees of freedom by breaking the requisite conditions for the no - go theorem . on the other hand , there is another direction to approach numerical simulations . according to @xcite , hypercubic symmetry and reflection positivity of actions result in @xmath2 species of fermions where @xmath3 stands for the dimension . thus it is potentially possible to reduce the number of species by breaking hypercubic symmetry properly . actually , the staggered fermion approach @xcite , with only 4 species of fermions does this and possesses flavored - hypercubic symmetry instead . however this requires rooting procedures for the physical @xmath4 or @xmath5-flavor qcd simulation . a possible goal in this direction is a lattice fermion with @xmath4 species , the minimal number required by the no - go theorem . such a minimal - doubling action was first proposed by karsten , and later by wilczek @xcite . other than the original type , two more types are known as creutz - borici type @xcite and twisted - ordering type @xcite . these fermions all possess one exact chiral symmetry and ultra locality . as such they could be faster for simulation , at least for two - flavor qcd , than other chirally symmetric lattice fermions . however it has been shown @xcite that we need to fine - tune several parameters for a continuum limit with these actions . this is because they lack sufficient discrete symmetry to prohibit redundant operators from being generated through loop corrections @xcite . thus the minimally doubled fermions have not been extensively used so far . nevertheless , there is the possibility to apply them to simulations if one can efficiently perform the necessary fine - tuning of parameters . in this paper we pursue the chiral phase structure in the parameter plane for minimal - doubling lattice qcd for two purposes : one purpose is to understand properties of these formulations theoretically . the other purpose is to show their applicability to lattice qcd simulations since understanding a nontrivial phase diagram , as with the aoki phase in wilson fermion @xcite , can be useful in showing the applicability to lattice qcd . we first show that minimal - doubling fermions can be seen as a special case of lattice fermions with species - dependent ( imaginary ) chemical potential terms . we focus on the karsten - wilczek formulation with one relevant parameter @xmath0 , which corresponds to a mass parameter in the analogy of wilson fermion . we next investigate the chiral phase structure in the space of the gauge coupling and this relevant parameter . we analyze strong - coupling lattice qcd , and show that chiral symmetry is spontaneously broken in a certain range of the parameter while the chiral condensate is zero outside the range . we find a second - order phase transition between these chiral symmetric and broken phases . we also show that pion becomes massless as a nambu - goldstone boson in the chiral - broken phase while the sigma meson becomes massless on the second - order phase boundary due to the critical behavior of the second - order phase transition . we investigate the lattice gross - neveu model to capture an entire phase structure . from these results we suppose a similar chiral phase structure in 4d qcd with flavored - chemical - potential lattice fermions , and discuss their applicability to lattice qcd simulations . in sec . [ sec : fcp ] , we study lattice fermions with flavored chemical potential , or minimal - doubling fermions . in sec . [ sec : slq ] , we investigate a chiral phase structure in the framework of strong - coupling lattice qcd . in sec . [ sec : gn ] , we study the gross - neveu model to obtain information of the whole phase diagram . in sec . [ sec : qcd ] , we discuss a phase structure in 4d qcd from the results of the last two sections . section [ sec : sd ] is devoted to a summary and discussion . in this section we study minimal - doubling fermions and their generalization . before going to the main theme , let us remind ourselves of wilson s prescription to shirk the no - go theorem . wilson fermion extracts one light fermion by introducing a species - dependent mass term , which we call a flavored - mass term " . a free action and a dirac operator of wilson fermion are given by @xmath6 @xmath7 we here exhibits a lattice spacing @xmath8 to manifest mass dimensions of each term . in this formulation , 15 out of 16 species have @xmath9 mass and are decoupled in the naive continuum limit . as shown in @xcite , this is not the only case of flavored - mass terms . there are four types of nontrivial flavored - mass terms ( @xmath10 , @xmath11 , @xmath12 , @xmath13 ) , which satisfy @xmath14 hermiticity and give second - derivative terms up to @xmath15 errors as with the usual wilson term . a general form of wilson - type fermions are written as @xmath16 where @xmath17 stands for flavored - mass terms . details of species - splitting depends on explicit forms of @xmath17 @xcite . we now consider further deformation of the fermion action . we multiply @xmath18 by @xmath19 or @xmath20 as @xmath21 in these cases , degeneracy of species is lifted by specie - dependent real or imaginary chemical potential terms , not by species - dependent mass . it means that we can get rid of some doublers by this method too . we name such terms as flavored - chemical - potential terms " , and name lattice fermions with them as flavored - chemical - potential ( fcp ) fermions " . ( we will later discuss problems with this kind of naive introduction of chemical potential @xcite . ) it is obvious that a real type of fcp terms breaks down @xmath14 hermiticity and leads to a sign problem while an imaginary type keeps it and has no sign problem . an outstanding point in this formulation is that ultra - locality and one exact @xmath22 chiral symmetry remains intact as @xmath23 ( see refs . @xcite for details of chiral symmetry in this type of lattice fermions . ) in principle , by using this deformation we can reduce 16 species to smaller multiple numbers of two without losing all chiral symmetries . we note that the chemical potential term , of course , breaks hypercubic symmetry into cubic symmetry and breaks c , p , t into ct and p @xcite as far as @xmath24 is cubic - symmetric . it means that this formulation automatically corresponds to finite - density systems unless we tune several parameters . in this paper we concentrate on the following explicit form of imaginary - type fcp fermions , @xmath25 @xmath26 here 14 species is decoupled in the naive continuum limit while two species at @xmath27 and @xmath28 has zero mass and zero imaginary chemical potential .. in the above case it is given by @xmath29 . this means the chiral symmetry possessed by this action is identified as a flavored one given by @xmath30 . ] more precisely , among 16 species , two species have zero imaginary chemical potential , six have @xmath31 , six have @xmath32 and two have @xmath33 . in fig . [ md ] we compares specie - splitting of kw fermion in chemical potential space to that of wilson fermion in mass space . it is notable that two - flavor is the minimal number allowed by the no - go theorem . this form has been known as the karsten - wilczek ( kw ) fermion @xcite , which is the first known type of minimal - doubling fermions " @xcite . it has one exact chiral symmetry , ultra - locality , cubic symmetry , ct and p. since the chemical potential term breaks discrete symmetries into the subgroup , we need to fine - tune three parameters for one dimension-3 ( @xmath34 ) and two dimension-4 ( @xmath35 , @xmath36 ) counterterms in order to take a lorentz - symmetric continuum limit for the zero-(@xmath37,@xmath38 ) lattice qcd simulations @xcite . among the three counterterms , in this paper we mainly deal with the dimension-3 term @xmath39 with a relevant parameter @xmath0 and the dimension-4 term @xmath40 with a marginal parameter @xmath41 since we study the strong - coupling lattice qcd and the gross - neveu model , which contain no plaquette action . in particular the parameter @xmath0 is of special importance : it changes the number of flavors and plays an important role in the chiral phase structure . furthermore the quantum effects produce @xmath9 additive chemical potential renormalization in this case instead the additive mass renormalization , and we need to cancel it by adjusting @xmath0 even for the application to the imaginary - chemical - potential lattice qcd . this necessity of parameter tuning is also understood from the well - known fact that the naive introduction of chemical potential into lattice fermions leads to divergence of energy density and requires a counterterm due to the violation of the abelian gauge invariance as shown in ref . @xcite . we here write the kw fermion action of the interacting theory as @xmath42 , \label{smd } \end{aligned}\ ] ] where we introduce a parameter @xmath43 in analogy with the wilson parameter . we introduce the dimension-3 counterterm with the parameter @xmath0 . although we mainly focus on @xmath0 in this work , we also introduce the dimension-4 counterterm with the parameter @xmath41 in eq . ( [ smd ] ) which is also relevant for the strong - coupling study . we make all the quantities dimensionless . now let us look into how the number of flavors depend on @xmath0 . for a free theory , the associated massless dirac operator in momentum space is @xmath44 we first look into a minimal - doubling range for @xmath0 and the speed of light in the range . we for simplicity take @xmath45 . for @xmath46 , we have only two zeros , both of which have the form as @xmath47 . by expanding the momentum as @xmath48 around the zeros , the coefficient of @xmath20 in the dirac operator is given as @xmath49 + \mu_{3}+3-\sum_{i=1}^{3}\cos ( \bar{p}_{i}+q_{i } ) \nonumber\\ & = [ ( 1+d_{4})\cos a\bar{p}_{4}]\ , aq_{4 } + o(a^{2}q^{2 } ) . \label{dpeq}\end{aligned}\ ] ] therefore , for general values of @xmath0 in this range , the speed of light is modified as @xmath50 we can fix this speed of light to a correct value up to the @xmath51 discretization errors by taking @xmath52 . we will discuss details of lorentz symmetry restoration in sec.[sec : qcd ] . now , let us classify @xmath0 parameter regions by the number of physical flavors in the free theory . since @xmath41 just gives a shift of the parameter regions , we now consider @xmath53 for simplicity . the above minimal - doubling range is given just by @xmath54 . for @xmath55 , we again have two zeros . for @xmath56 and @xmath57 , there is no zero of the dirac operator , which means that there are no physical fermions . for @xmath58 and @xmath59 , six zeros exist . @xmath60 have one zero , but the dispersion relations have a unphysical form as @xmath61 . @xmath62 have four zeros , but the dispersions are unphysical . @xmath63 has six zeros , but the dispersion is again unphysical . in the end , the parameter ranges where physical fermions can be described are @xmath55 ( 2 flavors ) , @xmath58 ( 6 flavors ) , @xmath59 ( 6 flavors ) , @xmath54 ( 2 flavors ) . we summarize it in fig . [ p - range ] . we note that nonzero @xmath41 gives larger minimal - doubling range as shown above . . on the boundaries between different sectors , the dispersion relation of fermions becomes unphysical.,height=264 ] what we want to study in this work is how these parameter ranges evolve in the finite gauge coupling direction . the most important question for practical use of this formulation in lattice qcd is how the two - flavor range , or the minimal - doubling range changes in the interacting theory . our question is deeply related to a possible chiral phase structure with respect to the @xmath22 chiral symmetry . it is because spontaneous chiral symmetry breaking is expected to occur only in parameter ranges with physical fermions . we thus speculate that boundaries between @xmath64ssb and non-@xmath64ssb phases starts from @xmath65 and @xmath66 in the weak - coupling limit . from next section , we will elucidate the @xmath0-@xmath1 chiral phase diagram by using strong - coupling lattice qcd and the gross - neveu model . in this section we employ the strong coupling analysis to investigate the chiral phase structure in lattice qcd with karsten - wilczek ( kw ) fermion . the first step is to derive an effective potential of meson fields corresponding to the fermion action in eq . ( [ smd ] ) . the strong - coupling study for kw fermion was first performed by the present author and collaborators in @xcite . we here take the same approach . lattice fermion action is generally written as following by using hopping operators @xmath67 and an onsite operator @xmath68 @xmath69 by using these operators an effective action for mesons in the strong coupling limit can be written @xcite as @xmath70 \ , , \\ \lambda_{n,\mu}&=&\frac{v_{n,\mu } \bar v_{n,\mu}}{n^2_c } , \quad \mathcal{m}(n)^{\alpha\beta } = \frac{\sum_a \bar \psi_n^{a,\alpha}\psi_n^{a,\beta}}{n_c}\ , , \nonumber\end{aligned}\ ] ] @xmath71 where @xmath72 is the number of colors , @xmath73 ( @xmath74 ) means a trace over color(spinor ) index , and @xmath75 is a meson field . @xmath76 are indices for colors while @xmath77 for spinors . the explicit form of the function @xmath78 is determined by performing a one - link integral of the gauge field . in the large @xmath72 limit , it is known that @xmath79 can be analytically evaluated @xcite as @xmath80 for most cases of studying phase structure , we can approximate it as @xmath81 , which corresponds to a large - dimension limit @xcite . since the phase transition is expected to be second - order for a massless case , this approximation at least works well near the phase boundary . we however note that it becomes less valid for large @xmath82 . in the case of the karsten - wilczek fermion , we have @xmath83 and @xmath84 we here assume a form of meson condensate with chiral and 4th vector condensates as @xmath85 it is because the flavored chemical potential term is expected to produce 4th vector condensate , which is related to quark density . ( we will discuss possibility of other condensates in the end of this section . ) the explicit form of the effective action for @xmath82 and @xmath86 is given by @xmath87\sigma^{2 } -{1\over{4}}[3(1-r^2)-(1+d_{4})^{2}]\pi_{4}^{2}.\end{aligned}\ ] ] we now find saddle points of @xmath88 from @xmath89 then gap equations are given by @xmath90 it is notable that these gap equations have a particle - hole symmetry as @xmath91 , which is reflected by chiral phase structure as we will see later . we first consider @xmath92 , and solve the equations analytically . one of the main purposes here is to find a boundary between chiral symmetric and broken phases . for this purpose we take @xmath93 after dividing the first equation by @xmath82 since @xmath82 is an order parameter of chiral symmetry breaking . then we have @xmath94 these equations give chiral boundaries for @xmath0 as @xmath95 therefore we have two ranges of @xmath0 with different chiral properties , * i * and * ii * : @xmath96 and @xmath45 . we define @xmath97 and depict the chiral condensate as a function of @xmath98 for @xmath99 . this result indicates the second - order chiral transition although the approximation @xmath81 becomes less reliable for larger @xmath82.,height=302 ] the question is which corresponds to the chiral symmetric or broken phases . we for a while look into the @xmath45 case to show details of chiral phase structure . since the change of @xmath41 just shift the chiral phase boundary ( [ bound ] ) , we for a while take @xmath53 to catch the chiral phase structure although we note that we need to take care of it for a lorentz symmetric continuum limit . for @xmath45 and @xmath53 the boundaries are given by @xmath100 ( @xmath101 ) . here we defined shifted @xmath0 as @xmath102 . by solving the gap equations in eq . ( [ g1])([g2 ] ) for @xmath45 and @xmath53 we derive two solutions of chiral and @xmath86 condensates as @xmath103 and @xmath104 by comparing effective potentials of the two solutions , we find which corresponds to vacua for the two parameter ranges , * i * and * ii*. we show the following by substituting the two solutions : @xmath105 for @xmath106 ( * ii * ) while @xmath107 for @xmath108 ( * i * ) . to sum up , the chiral symmetric and broken phases are given by @xmath109 the behavior of chiral condensate is depicted as a function of @xmath102 in fig . [ chicon ] , which shows that the transition between chiral symmetric and broken phases is second - order . we also depict parameter regions for the two phases in the strong - coupling limit for @xmath45 , @xmath53 and @xmath92 in fig . [ stp ] . as seen from eq . ( [ bound ] ) , a nonzero @xmath41 gives a larger physical range ( * ii * ) with ssb of chiral symmetry . , @xmath53 and @xmath92.,height=188 ] we show that the order of transition becomes a crossover for @xmath110 . fig.[try ] shows the chiral condensate as a function of @xmath102 and @xmath111 . it is obvious that the second - order phase transition changes into a crossover for @xmath110 . and @xmath53 as a function of @xmath98 and @xmath111.,height=264 ] we note that @xmath86 condensate is in general non - zero in the chiral - broken phase . @xmath86 becomes zero only for @xmath112 , which corresponds to a theory with unphysical dispersions . we next look into mass of mesons within the same framework . we here consider @xmath92 but general values of @xmath43 and @xmath41 . to calculate meson masses we expand the meson field as , @xmath113 where @xmath114 and @xmath115 stand for scalar , pseudo - scalar , vector , axial - vector and tensor respectively . here @xmath116 is the vacuum expectation value of @xmath75 . we note that @xmath117 then the effective action at the second order of @xmath118 is given by @xmath119 \nonumber \\ & = & n_c\int\frac{d^4 p}{(2\pi)^4 } \pi^{x}(-p ) d_{{x}{y}}(p ) \pi^{y}(p ) \ , , \end{aligned}\ ] ] where @xmath120 \widetilde{d}_{{x}{y}}(p ) & = & \frac{1}{2}{\rm tr}\,(\mathcal{m}_0^{-1 } \gamma_{x}\mathcal{m}_0^{-1}\gamma_{y } ) + \sum_\mu { \rm tr}\ , ( \gamma_{x}p^-_\mu \gamma_{y}p^+_\mu ) e^{i p_\mu } .\end{aligned}\ ] ] in our case @xmath121 gives @xmath122 we now write the whole inverse meson propagator matrix in the @xmath123-@xmath124-@xmath37-@xmath125-@xmath126 sector as @xmath127 \ , , \nonumber\\\end{aligned}\ ] ] where components are given by @xmath128 , \label{ds } \\ d_{v_{4}}&={\sigma^{2}-\pi_{4}^{2}\over{2(\sigma^{2}+\pi_{4}^{2})^{2 } } } -{1\over{4}}[(1-r^{2})(c_{1}+c_{2}+c_{3})-(1+d_{4})^{2}c_{4}],\\ d_{v_{3}}&= { 1\over{2(\sigma^{2}+\pi_{4}^{2 } ) } } - { 1\over{4 } } [ ( 1+r^{2})(c_{1}+c_{2})+(r^{2}-1)c_{3 } + ( 1+d_{4})^{2}c_{4 } ] , \\ d_{v_{2}}&= { 1\over{2(\sigma^{2}+\pi_{4}^{2 } ) } } - { 1\over{4 } } [ ( 1+r^{2})(c_{1}+c_{3})+(r^{2}-1)c_{2 } + ( 1+d_{4})^{2}c_{4 } ] , \\ d_{v_{1}}&= { 1\over{2(\sigma^{2}+\pi_{4}^{2 } ) } } - { 1\over{4}}[(1+r^{2})(c_{2}+c_{3 } ) + ( r^{2}-1)c_{1}+(1+d_{4})^{2}c_{4 } ] , \\ d_{t_{14}}&= { 1\over{2(\sigma^{2}+\pi_{4}^{2 } ) } } - { 1\over{4 } } [ ( r^{2}-1)(c_{2}+c_{3})+(1+r^{2})c_{1 } + ( 1+d_{4})^{2}c_{4 } ] , \\ d_{t_{24}}&= { 1\over{2(\sigma^{2}+\pi_{4}^{2 } ) } } - { 1\over{4 } } [ ( r^{2}-1)(c_{1}+c_{3})+(1+r^{2})c_{2 } + ( 1+d_{4})^{2}c_{4 } ] , \\ d_{t_{34}}&= { 1\over{2(\sigma^{2}+\pi_{4}^{2 } ) } } - { 1\over{4 } } [ ( r^{2}-1)(c_{1}+c_{2})+(1+r^{2})c_{3 } + ( 1+d_{4})^{2}c_{4 } ] , \\ d_{t_{12}}&= { \sigma^{2}-\pi_{4}^{2}\over{2(\sigma^{2}+\pi_{4}^{2})^{2 } } } + { 1\over{4 } } [ ( r^{2}-1)(c_{1}+c_{2})+(1+r^{2})c_{3 } + ( 1+d_{4})^{2}c_{4 } ] , \\ d_{t_{13}}&= { \sigma^{2}-\pi_{4}^{2}\over{2(\sigma^{2}+\pi_{4}^{2})^{2 } } } + { 1\over{4 } } [ ( r^{2}-1)(c_{1}+c_{3})+(1+r^{2})c_{2 } + ( 1+d_{4})^{2}c_{4 } ] , \\ d_{t_{23}}&= { \sigma^{2}-\pi_{4}^{2}\over{2(\sigma^{2}+\pi_{4}^{2})^{2 } } } + { 1\over{4 } } [ ( r^{2}-1)(c_{2}+c_{3})+(1+r^{2})c_{1 } + ( 1+d_{4})^{2}c_{4 } ] , \\ d_{a_{1}}&= { \sigma^{2}-\pi_{4}^{2}\over{2(\sigma^{2}+\pi_{4}^{2})^{2 } } } + { 1\over{4}}[(1+r^{2})(c_{2}+c_{3 } ) + ( r^{2}-1)c_{1}+(1+d_{4})^{2}c_{4 } ] , \\ d_{a_{2}}&= { \sigma^{2}-\pi_{4}^{2}\over{2(\sigma^{2}+\pi_{4}^{2})^{2 } } } + { 1\over{4 } } [ ( 1+r^{2})(c_{1}+c_{3})+(r^{2}-1)c_{2 } + ( 1+d_{4})^{2}c_{4 } ] , \\ d_{a_{3}}&= { \sigma^{2}-\pi_{4}^{2}\over{2(\sigma^{2}+\pi_{4}^{2})^{2 } } } + { 1\over{4 } } [ ( 1+r^{2})(c_{1}+c_{2})+(r^{2}-1)c_{3 } + ( 1+d_{4})^{2}c_{4 } ] , \\ d_{a_{4}}&={1\over{2(\sigma^{2}+\pi_{4}^{2 } ) } } + { 1\over{4}}[(1-r^{2})(c_{1}+c_{2}+c_{3})-(1+d_{4})^{2}c_{4 } ] , \label{da4 } \\ d_{p}&= { 1\over{2(\sigma^{2}+\pi_{4}^{2 } ) } } - { 1\over{4 } } [ ( 1+r^{2})(c_{1}+c_{2}+c_{3 } ) + ( 1+d_{4})^{2}c_{4 } ] , \\ c&= { i\sigma\pi_{4}\over{(\sigma^{2}+\pi_{4}^{2})^{2}}},\end{aligned}\ ] ] with @xmath129 and @xmath130 . by diagonalizing this matrix , we can derive an explicit form of physical meson propagators . we note that @xmath131 gets zero in the chiral symmetric phase while it has nonzero values in the chiral broken phase . we first check that pion mass becomes zero in the chiral - broken phase . for this purpose we substitute @xmath132 into the propagator . then @xmath133 is decoupled and the calculation gets simplified . we note that , in the chiral broken phase , the gap equation eq . ( [ g1 ] ) gives @xmath134 $ ] . pion mass in the chiral broken phase with @xmath45 and @xmath92 is derived from a pole of @xmath133 with @xmath135 as @xmath136 this mode corresponds to a massless ng boson associated with ssb of chiral @xmath22 symmetry , which is consistent with realistic qcd . on the second - order phase boundary , a divergent correlation length in the critical behavior should produce another massless mode . we here show that the scalar meson becomes massless on the critical line . an inverse meson propagator matrix in general has off - diagonal components , but substitution of @xmath137 again decouples the scalar sector in the chiral - symmetric phase ( @xmath138 ) . an explicit form of the scalar inverse propagator is given by eq . ( [ ds ] ) . then we derive the scalar mass from a pole of this propagator with @xmath137 as @xmath139 .\ ] ] as seen from this , scalar meson has nonzero mass in general . on the phase boundary , however , we have @xmath93 , @xmath140}$ ] and @xmath141 with @xmath142 . it leads to zero @xmath82 mass as @xmath143 we can also show this from the chiral - broken phase . we have two massless modes only on the 2nd - order phase boundary , @xmath82-meson and @xmath144-meson , which is inconsistent with the meson mass spectrum in qcd . we thus need to avoid this point in the simulation of qcd . in the first place this boundary corresponds to the boundary between the two - flavor and no - flavor ranges , which has the unphysical dispersion relation as shown in sec . [ sec : fcp ] . from the theoretical viewpoint , it is however an attractive topic . further study including numerical simulations can elucidate detailed properties of this point . now let us discuss possibility of restoration of lorentz symmetry . in the strong - coupling limit , what we can do is tuning of @xmath0 and @xmath41 since we can ignore tuning for the plaquette action in this limit . however , we can not restore the lorentz symmetry correctly in this limit because of the large lattice artifacts . the @xmath145 meson propagator explicitly manifests this point as following : the @xmath145-sector propagator , which is diagonal even with finite spacial momentum , is given by eq . ( [ da4 ] ) . in the chiral - broken phase , we have @xmath146/2 $ ] . by substituting @xmath147 , the pole of the propagator determines the axial vector meson mass as @xmath148 for small but finite spatial momentum as @xmath149 , the pole of the propagator up to @xmath150 gives @xmath151 it apparently indicates that we need to tune @xmath41 as @xmath152 . however , for @xmath45 , the dependence on @xmath153 itself disappears and the lorentz symmetry can not be restored . it is a typical strong - coupling artifact . moreover , the tuned value of @xmath41 does not depend on @xmath0 , but it is also a lattice artifact . in the first place , the equation ( [ g1 ] ) indicates that @xmath86 always has a nonzero value in the physical parameter range ( @xmath154 ) even if we tune @xmath0 and @xmath41 independently except for @xmath155 . all these results show that it is difficult to restore the symmetry in the physical parameter range within the framework of the strong - coupling qcd . we expect that it is just a strong - coupling artifact , and we can make @xmath86 zero and restore lorentz symmetry in the weak coupling by tuning the three parameters appropriately . in this section we have assumed the form of condensation @xmath156 . as shown in the appendix.a , we can also consider possibility of other condensations as @xmath157 or @xmath158 . the results show that the solution with nonzero @xmath159 or @xmath160 condensates can not be a vacuum , and our solutions in this section are likely to be true vacua . we thus consider that the parity breaking phase will not appear in the kw fermion unless we introduce the flavored - mass terms shown in @xcite . at the weak coupling there may be a more subtle competition between the discretization error and the counterterm . we thus need further study to conclude whether the parity breaking exist or not when we take into account all the three counterterms in the weak coupling . in the end of this section , we discuss the other type of minimal - doubling fermions , called the creutz - borici type @xcite . we can analyze it in a parallel way . we note that this type specifies the diagonal direction characterized by @xmath161 , instead of the time direction . appendix . b is devoted to detailed analysis for this case . the result is qualitatively the same . we find two chiral boundaries and chiral condensate is nonzero between the boundaries . we investigate the whole phase diagram for karsten - wilczek ( kw ) fermion by using the two - dimensional lattice gross - neveu model @xcite , which has common features with 4d lattice qcd . in two dimensions , massless kw fermion action is given by @xmath162 . \label{2dd}\ ] ] in this section we concentrate on the case with @xmath45 . in this section we denote the relevant parameter @xmath0 as just @xmath163 and denote the maginal parameter @xmath41 as @xmath164 . to look into the number of flavors in a free theory , we for a while consider @xmath165 . for @xmath166 and @xmath167 , there are only two zeros , and it becomes minimal - doubling . for @xmath168 and @xmath169 , there is no zero , and it becomes a fermion - less theory . for @xmath170 , there is one zero , but the dispersion relation becomes unphysical @xmath171 . for @xmath172 , there two zeros but whose dispersion relation is again unphysical . the main difference from four - dimensional cases is that there is no 6-flavor range . the lattice gross - neveu model with kw fermion is given by @xmath173 \nonumber\\ & + \mu\sum_{n}\bar{\psi}_{n}i\gamma_{2}\psi_{n } + d\sum_{n}\bar{\psi}_{n}i\gamma_{2}(\psi_{n+\hat{2}}-\psi_{n-\hat{2 } } ) + m\sum_{n}\bar{\psi}_{n}\psi_{n } , \label{sgns}\end{aligned}\ ] ] where @xmath174 stands for @xmath175 , @xmath176 are the two dimensional coordinates and @xmath177 stands for a @xmath178-component dirac fermion field @xmath179(@xmath180 ) . the bilinear @xmath181 means @xmath182 , and ( @xmath183 , @xmath184 ) corresponds to the t hooft couplings for the two types of four - fermi interactions . here we define the two dimensional gamma matrices as @xmath185 , @xmath186 and @xmath187 . we make all the quantities dimensionless in this equation . we consider scalar and time - direction vector four - fermi interactions , which are natural choices since kw fermion specifies the time direction . we note that , if we drop a mass term ( @xmath92 ) , the action has @xmath188 discrete chiral symmetry , which can be spontaneously broken due to chiral condensate . by introducing auxiliary bosonic fields @xmath189 , @xmath190 we remove the four - point interactions as @xmath191 + \sum_{n}\bar{\psi}_{n}[\sigma(n)+i\gamma_{2}\pi_{2}(n)]\psi_{n}. \label{sgnssp}\end{aligned}\ ] ] by solving the equations of motion , we show the following relation between these auxiliary fields and the bilinears of the fermion fields @xmath192 these relations indicate that @xmath82 and @xmath193 stand for the scalar and vector mesons . after integrating the fermion fields , the partition function and the effective action with these auxiliary fields are given by @xmath194-{\rm tr}\,\log d_{n , m } , \label{nseff}\end{aligned}\ ] ] with @xmath195\delta_{n.m}&+ { \gamma_{\mu}\over{2}}(\delta_{n+\mu , m}-\delta_{n-\mu , m } ) \nonumber\\ & -{i\gamma_{2}\over{2}}(\delta_{n+\hat{1},m}+\delta_{n-\hat{1},m } ) + d{\gamma_{2}\over{2}}(\delta_{n+\hat{2},m}-\delta_{n-\hat{2},m } ) . \label{nd}\end{aligned}\ ] ] here @xmath73 stands for the trace both for the position and spinor spaces . as is well - known , the partition function in the gross - neveu model is given by the saddle point of this effective action in the large @xmath178 limit . we denote as @xmath196 , @xmath197 solutions satisfying the saddle - point conditions @xmath198\over{\delta \sigma(n)}}\,=\ , { \delta s_{\rm eff}[\sigma(n),\pi_{2}(n)]\over{\delta \pi_{2}(n)}}\,=\ , 0 . \label{nsadeq1}\ ] ] then the partition function is given by @xmath199}. \label{neffpar}\ ] ] by assuming the translational invariance we define the position - independent solutions as @xmath200 and @xmath201 then we can factorize a volume factor @xmath202 in the effective action as @xmath203 we can write @xmath204 in a simple form by the fourier transformation to momentum space @xmath205 \nonumber\\ & = \ , v\int{d^{2}k\over{(2\pi)^{2}}}\log[\sigma^{2}+(\pi_{2}+(1+d)\sin k_{2}-\cos k_{1})^{2}+(\sin k_{1})^{2 } ] , \label{ni}\end{aligned}\ ] ] with @xmath206 being the determinant in the spinor space . now saddle - point equations are written as @xmath207 here the values of @xmath82 and @xmath193 in the vacuum are determined as @xmath208 , @xmath209 from the saddle - point equations once @xmath38 , @xmath3 , @xmath183 , @xmath184 and @xmath111 are fixed . let us look into the phase diagram with respect to chiral symmetry . we here consider a massless case as @xmath92 to have the exact discrete chiral symmetry in the action . to capture rough structure of the phase diagram , we first take the simplest case with @xmath165 and @xmath210 . nonzero @xmath3 just gives slight change of the phase diagram . since a single coupling constant works when we study the aoki phase in the wilson fermion @xcite , we expect that the above condition for the couplings works at least for deriving rough phase structure . the order parameter is @xmath82 , which can be zero or non - zero depending on values of @xmath38 and @xmath1 . the phase boundary is determined by imposing @xmath93 on eq . ( [ ncond1])([ncond2 ] ) after the overall @xmath82 being removed in eq . ( [ ncond1 ] ) . then the conditions for the phase boundary are given by gap equations as @xmath211 with @xmath212 being a critical value of @xmath38 . here we derive the chiral phase boundary @xmath213 as a function of the coupling @xmath1 by getting rid of vector condensate @xmath193 from these equations . the phase diagram is depicted in fig . [ fig1 ] . a stands for the chiral symmetric phase @xmath93 and b for chiral broken phase @xmath154 . for @xmath214 the integral in eq.([ncricon2 ] ) diverges , which leads to the critical lines with zero gauge coupling for @xmath215 as shown in fig . this is reasonable since the weak - coupling limit should have zero chiral condensate . as we expected , the chiral critical line is connected to boundaries between two - flavor and no - flavor phases in the weak - coupling limit ( @xmath216 ) . it is consistent with our intuition that the fermion - less theory can not cause spontaneous breaking of chiral symmetry . -@xmath1 space . a stands for the chiral symmetric phase . b stands for the chiral - broken phase . the chiral boundaries are connected to the edges of minimal - doubling ranges.,height=264 ] we can check the mass of the scalar meson mass becomes zero on the critical line @xmath213 . the mass of @xmath82 is calculated analytically on the boundary , and is shown to be zero as @xmath217_{\mu_{c } } \nonumber\\ & = 0.\end{aligned}\ ] ] a massless scalar meson indicates that the phase boundary we derived is a second - order critical line . this result is consistent with the strong - coupling lattice qcd in the previous section . next , we discuss more general cases with @xmath218 and @xmath219 . nonzero values of @xmath3 change @xmath220 in ( [ ncricon1])([ncricon2 ] ) and give just qualitative changes of the phase diagram : since the minimal - doubling ranges are given by @xmath221 and @xmath222 for nonzero @xmath3 , it gives a larger physical range in the phase diagram . as an example , we depict the @xmath38-@xmath1 phase diagram for @xmath223 in fig . [ 0.5 ] . in the case with two independent coupling constants @xmath219 , the equations for the phase boundary are given by @xmath224 in this case the phase diagram is deformed to some extent : if we fix one of the coupling constants and depict the phase diagram , the physical phase gets larger toward the strong - coupling limit while it gets narrower at the weak - coupling . as an example , we depict the @xmath38-@xmath183 phase diagram for @xmath225 in fig . [ g1 ] . since we have already shown that the physical phase gets rather narrower in the strong - coupling limit in 4d lattice qcd in sec . [ sec : slq ] . it is natural to consider that the single - coupling gn model in fig . [ fig1 ] is sufficient to mimic 4d qcd at least for investigating the chiral phase diagram . we note that we can introduce two more types of four - point interactions in the gn action ( [ sgns ] ) as @xmath226 and @xmath227 , which give @xmath228 and @xmath144 meson fields . both vacuum expectation values of @xmath228 and @xmath144 , if they are nonzero , cause spontaneous parity symmetry breaking . however , by comparing the effective potentials or solving four gap equations ( @xmath229 , @xmath230 , @xmath231 , @xmath232 ) , we can clearly show that they have zero vacuum expectation values as in the case of the strong - coupling lattice qcd in appendix.a . instead of solving all the gap equations numerically , we here show the consistency check on @xmath233 . from two saddle - point equations @xmath229 and @xmath230 we derive one equation , @xmath234 if we take @xmath233 , the right - hand side becomes an odd function of @xmath235 , and the above equation holds identically . we can also show from the gap equations that @xmath236 is preferred . these results mean that the parity - broken vacuum is not preferred , and as long as we consider the vacuum of the theory , the two four - fermi interactions in ( [ sgns ] ) are sufficient . we next show that , however , we have to take into account the excitation of @xmath228 and @xmath144 to investigate the lorentz symmetry restoration in the gn model . -@xmath1 space with @xmath223 ( @xmath237 ) . the physical ( minimal - doubling ) phase with chiral condensate is enlarged by nonzero values of @xmath3.in the weak - coupling limit , the minimal - doubling phases are given by @xmath238 and @xmath239.,height=264 ] -@xmath183 space with @xmath225 ( @xmath165 ) . the physical ( minimal - doubling ) phase with chiral condensate gets larger toward the strong - coupling limit.,height=264 ] by now we have considered only the vacuum of the theory . although it works to elucidate the chiral phase structure , we need look into excitations from the vacuum in order to investigate renormalization for the rotation symmetry restoration . by expanding the mesonic action up to the 2nd order , we can derive the meson propagator matrix for @xmath240 as in the strong - coupling lattice qcd in sec . [ sec : slq ] . by diagonalizing this @xmath241 matrix , we obtain the proper propagators . then , we can discuss how the rotation symmetry can be restored by tuning the parameters including @xmath38 , @xmath3 , @xmath183 and @xmath184 . although we can perform this procedure in principle , all the matrix components include complicated momentum integrals in this case . moreover , we need to substitute the vev of meson fields derived from the gap equations , which also require numerical integrals . we consider that these numerical calculations are beyond the scope of this study , but the future work should be devoted to it . instead , we here show a process of deriving the dispersion relations in details in the following . the meson excitations from the vacuum are given by @xmath242 where vevs ( @xmath243 , @xmath244 , @xmath245 , @xmath246 ) are determined by the gap equations . we have already shown @xmath247 while @xmath248 and @xmath249 depends on the parameters . the dirac operator with these excitations are written as @xmath250\delta_{n.m}+ { \gamma_{\mu}\over{2}}(\delta_{n+\mu , m}-\delta_{n-\mu , m})/a \nonumber\\ & \,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,-{i\gamma_{2}\over{2}}(\delta_{n+\hat{1},m}+\delta_{n-\hat{1},m})/a + d{\gamma_{2}\over{2}}(\delta_{n+\hat{2},m}-\delta_{n-\hat{2},m})/a \\ \mathcal{m}(n)&=\sigma(n)+i\gamma_{1}\pi_{1}(n)+i\gamma_{2}\pi_{2}(n)+i\gamma_{3}\pi(n),\end{aligned}\ ] ] where we make the lattice spacing @xmath8 manifest . note @xmath251 . for completeness , we consider the following action including four independent coupling constants , @xmath252-{\rm tr}\,\log [ \mathcal{s}_{0}^{-1}(1+\mathcal{s}_{0}\mathcal{m})]_{n , m}.\ ] ] @xmath253 is a trace for coordinate and spinor spaces . by expanding this expression up to second order of mesonic fluctuations , we derive the effective action up to the quadratic order as @xmath254 with @xmath255,\ ] ] with @xmath256 is a trace for the spinor space . by fourier transformation we can derive the form in the momentum space as @xmath257,\end{aligned}\ ] ] with @xmath258 , \\ & \det(\mathcal{s}_{0}^{-1}(k ) ) = \bar{\sigma}^{2 } + \left(\bar{\pi}_{2 } + ( 1+d){\sin ak_{2}\over{a}}-{\cos ak_{1}\over{a}}\right)^{2}+ \left({\sin ak_{1}\over{a}}\right)^{2}.\end{aligned}\ ] ] finally , the mesonic propagator matrix is given by @xmath259 where @xmath260 and @xmath261 stand for one of channels @xmath82 , @xmath228 , @xmath193 and @xmath144 . for example , @xmath262 is given by @xmath263 } } \times\big[\bar{\sigma}^2 \nonumber\\ & -\big(\bar{\pi}_{2}+(1+d){\sin ak_{2}\over{a}}+{\cos ak_{1}\over{a}}\big ) \big(\bar{\pi}_{2}+(1+d){\sin a(p_{2}+k_{2})\over{a}}+{\cos a(p_{1}+k_{1})\over{a}}\big ) \nonumber\\ & -{\sin ak_{1}\sin a(p_{1}+k_{1})\over{a^{2}}}\big ] . \label{dead}\end{aligned}\ ] ] in the same way we can derive all other 16 components of the matrix , all of which take nonzero values in general . what we are interested in is terms of the components up to @xmath150 as far as we consider the rotation symmetry up to @xmath51 discretization errors . the coefficients can be extracted through numerical integrals of the equations as ( [ dead ] ) . we note that we also need to substitute into the integral the values of vev of @xmath82 and @xmath193 derived from the gap equations . then we diagonalize the @xmath241 meson matrix @xmath264 to derive the proper meson propagator . by introducing @xmath265 into the propagators , we find the dispersion relations . the question is which parameters among @xmath38 , @xmath3 , @xmath183 , @xmath266 , @xmath184 , @xmath267 need to be tuned to recover lorentz symmetry . it can give an important suggestion to lattice qcd with the minimal - doubling fermions . we devote a future work to this analysis . from the study of strong - coupling lattice qcd and the gross - neveu model we speculate on the whole chiral phase structure in lattice qcd with karsten - wilczek fermion . [ al ] is a conjectured chiral phase structure with the number of flavors in the @xmath0-@xmath1 space for @xmath45 . there are roughly two phases with and without chiral condensate , or equivalently with and without ssb of chiral symmetry . as was shown in the previous section , the boundary between chiral symmetric and broken phases starts from the edge of the two - flavor region of the free theory . we expect that the chiral boundaries are connected to the two - flavor and no - flavor phases also in 4d qcd as shown in fig . [ al ] . -@xmath1 chiral phase structure for karsten - wilczek fermion with r=1 . the width of the minimal - doubling range determines how hard it is to tune @xmath0.,height=264 ] the question is a boundary between two - flavor and six - flavor ranges . in the weak - coupling limit ( @xmath268 ) we analytically know the number of physical flavors : there are four sectors with two , six , six and two flavors . there are only no flavors of fermions outside these ranges . toward the strong coupling , these ranges will change with @xmath1 as shown in fig . we have seen that we can not distinguish two - flavor and six flavor ranges in the strong - coupling limit , which means that the number of species becomes an ambiguous notion in this limit . we thus expect that the boundary disappears at a certain gauge coupling , and the two - flavor and six - flavor regions become undistinguishable as shown in fig . [ al ] . from the viewpoints of practical application to two - flavor qcd , the relevant parameter @xmath0 has to be tuned to cancel the @xmath9 imaginary chemical potential renormalization for the two flavors . one necessary condition ( but not a sufficient condition ) for this purpose is to set @xmath0 in the minimal - doubling range to realize the two - flavor qcd . as we conjectured in fig . [ al ] , the minimal - doubling range in the middle gauge coupling should have some width . one possible indicator of the minimal - doubling phase is the pion spectrum : if @xmath0 is in the no - flavor range , there is no ssb of chiral symmetry and no massless pion . if @xmath0 gets into the six - flavor region , the number of pseudo nambu - goldstone bosons increases . however , setting @xmath0 in the minimal - doubling range is not sufficient for physical qcd to be described : the lorentz symmetric dispersion relation is broken down for general values of @xmath0 in the minimal - coupling range as shown in eq . ( [ dp1 ] ) and below for a free theory . @xmath269 since this free - theory argument indicates that the rotation symmetry can be restored by tuning @xmath41 as @xmath270 , we may be able to restore lorentz symmetry just by tuning @xmath41 with @xmath0 being set in the minimal - doubling range . note that the minimal - doubling range gets larger with nonzero @xmath41 as shown in sec.iii and iv , thus it seems that proper tuning of @xmath41 can be done for any value of @xmath0 without breaking down minimal - doubling . ( we need one more parameter tuning for the plaquette action in any case . ) we also emphasize that the same relative tuning of @xmath0 and @xmath41 makes tree level couplings of the gauge field to the fermions have a correct lorentz - symmetric form . to show this , we look into the quark - quark - gluon vertex at the tree level . for the case of @xmath271 and @xmath53 it is given by @xmath272 as shown in @xcite . for nonzero @xmath0 and @xmath41 , it is modified as @xmath273 \right ) , \label{mdv}\ ] ] where we have no direct emergence of @xmath0 since it is a parameter for onsite ( non - hopping ) terms . however , as we discussed , the zeros of the dirac operator for nonzero @xmath0 and @xmath41 is given by a function of @xmath0 and @xmath41 as @xmath274 . now we expand both @xmath275 and @xmath276 about the zeros as @xmath277 and @xmath278 . in particular , the coefficient of @xmath19 in ( [ mdv ] ) is expanded as @xmath279 + \sum_{i=1}^{3}\sin { a(p_{i}+k_{i})\over{2 } } \nonumber\\ = & ( 1+d_{4})\big[\sqrt{1-{\mu_{3}\over{(1+d_{4})^{2 } } } } \cos { a(p_{4}+k_{4})\over{2 } } + { \mu_{3}\over{1+d_{4 } } } \sin { a(p_{4}+k_{4})\over{2}}\big ] + \sum_{i=1}^{3}\sin { a(p_{i}+k_{i})\over{2 } } \nonumber\\ = & ( 1+d_{4})\sqrt{1-{\mu_{3}\over{(1+d_{4})^{2}}}}+o(ap , ak),\end{aligned}\ ] ] then , the vertex surviving in the naive continuum limit is given by @xmath280 here we omit the @xmath281 sign in front of @xmath19 for the doubler pairs for simplicity . it is now obvious that the tuning condition for the speed of light @xmath270 also fixes the couplings of the gauge fields to the fermion fields in the tree level up to the discretization errors . at least in the naive continuum limit , we can have a correct set of the feynman rules for fermion fields with the condition . however it is too early to conclude that this condition is sufficient for lorentz symmetry restoration since all the other ward identities may not be corrected by it in the interacting theory including the loop effects : to discuss details , we consider the quark self - energy in lattice qcd with minimal - doubling fermions following @xcite as @xmath282 where @xmath283 , @xmath284 , @xmath285 and @xmath286 can be calculated in the perturbative analysis . it is clear that @xmath0 and @xmath41 corresponds to counter parameters for @xmath286 and @xmath285 respectively , and the dirac operator ( [ dp2 ] ) is renormalized as @xmath287 and @xmath288 in the interacting theory . the last term with @xmath286 , or the @xmath9 renormalization , causes a shift of the poles of the dirac propagator away from their original positions as well as the change of the speed of light as shown in ( [ dp2 ] ) . in the present work , we have also shown that this contribution changes the size of the minimal - doubling range at the finite gauge coupling and provides the non - trivial phase structure as shown in fig . the question is whether or not we need to move the poles of the propagator back to the tree - level positions by tuning @xmath0 for a correct continuum limit . as far as the dispersion relation can be restored by @xmath41 , it seems that the position of poles is not relevant to physics . however , in practical use of the minimal - doubling fermion , @xmath41 should be also tuned to make the conserved charge unity as shown in @xcite . it is not obvious whether this condition can also fix the speed of light non - perturbatively . more generally speaking , it is very nontrivial whether all the ward identities are fixed only by one tuning condition in the interacting theory . if we can not restore lorentz symmetry only with @xmath41 tuning unlike the free theory , it means that we still need to fine - tune the three parameters independently for a correct continuum limit of lattice qcd simulations @xcite . further study is needed to figure out this point . in the end of this section , we comment on another possibility for studying minimal - doubling fermions . one interesting possibility is the chiral perturbation theory for minimal - doubling fermions . although we expect that it is quite tedious to construct the minimal - doubling chpt with the lower discrete symmetry than wilson and staggered fermions , the process could have some similarities with that of the in - medium chpt @xcite . if we succeed to construct the minimal - doubling chpt , it is intriguing to consider the lorentz symmetry restoration within the theory and discuss the parameter tuning for the symmetry restoration . we can also investigate the vacuum and the phase structure in the theory . we devote future works to the study on the minimal - doubling chpt . in this paper we investigate the chiral phase structure in the parameter space for lattice qcd with minimal - doubling fermions , which can be seen as lattice fermions with a species - dependent naive chemical potential term . we study the phase structure with karsten - wilczek fermion by using strong - coupling lattice qcd and the gross - neveu model , and find out the nontrivial chiral phase structure in the @xmath289 plane . in sec . [ sec : fcp ] , we have proposed flavored - chemical - potential lattice fermions , where some of doublers are eliminated by a species - dependent chemical potential term without losing all chiral symmetries . minimal - doubling fermions are shown to be a special case of this type . in sec . [ sec : slq ] we investigate the chiral phase structure of lattice qcd with karsten - wilczek fermions in the strong - coupling regime . we derive an effective action for the scalar and 4th vector fields , and find that chiral symmetry is spontaneously broken in a certain range of the relevant parameter @xmath0 while the chiral condensate becomes zero outside the range . we show that there is a 2nd - order phase transition between chiral symmetric and broken phases as a function of @xmath0 . we also show that pion becomes massless as a nambu - goldstone boson in the chiral - broken phase while the scalar meson becomes massless only on the second - order phase boundary due to the critical behavior . in sec . [ sec : gn ] we obtain information on the whole chiral phase structure in the @xmath289 space by using the gross - neveu model with kw fermion in large n limit . from the gap equations we derive a chiral phase diagram . in sec [ sec : qcd ] we discuss the whole phase structure in 4-dimensional lattice qcd with kw fermions . we conjecture the chiral phase structure and numbers of massless flavors in the phase diagram . we also discuss the fine - tuning process of @xmath0 in lattice qcd from the viewpoint of the minimal - doubling range of the conjectured phase diagram . in this paper we have investigated whether or not chiral symmetry is spontaneously broken depending on the parameters . unlike wilson parity - flavor breaking phase with the width @xmath290 @xcite , it seems that we do not have fine phase structures in the weak - coupling regime as shown in fig . it is because the chiral symmetry breaking is physical , and there is no competition between the relevant parameter and the lattice artifact , which did cause the aoki phase or the creutz - sharpe - singleton phase in wilson fermion @xcite . although our analysis on chiral symmetry and that on parity symmetry in wilson have some common features , they are on the different levels . in this work , we do not take much care of @xmath86 condensate nor its physical implication . as we discussed in sec . [ sec : slq ] this condensate is likely to be related to the ( pseudo ) quark density . on the other hand , eq . ( [ g1 ] ) in the strong - coupling limit shows that @xmath86 is nonzero for any values of @xmath0 and @xmath41 except for the unphysical point @xmath155 . we consider that it is just a strong - coupling artifact , and the @xmath86 condensate can be eliminated by fine - tuning all the three parameters appropriately in the weak coupling . there is possibility that another way of introducing the dimension-4 counterterm as an o(1 ) imaginary chemical potential term @xmath291 may work since the flavored - chemical - potential term could generate @xmath292 effective imaginary chemical potential too . in the end of this paper , we refer to a future work . we can apply the flavored - chemical - potential ( fcp ) fermions to finite - temperature and finite - density lattice qcd , and obtain finite-(@xmath37,@xmath38 ) qcd phase diagram . in our next work @xcite , we will propose a new method to study lattice qcd in medium by using fcp fermion formulations . tm is thankful to m. creutz , t. kimura and a. ohnishi for fruitful discussion and hearty encouragement . the author also thanks f. karsch for the discussion . the author appreciates the technical help by t. kawanai . tm is supported by grant - in - aid for japan society for the promotion of science ( jsps ) postdoctoral fellows for research abroad(no.24 - 8 ) . we here assume a form of meson condensate with chiral , 4th vector and pion condensates as @xmath293 the effective potential for @xmath82 , @xmath86 and @xmath159 is given by @xmath294\sigma^{2 } -{1\over{4}}[3(1-r^2)-(1+d_{4})^{2}]\pi_{4}^{2 } - { 1\over{4}}[3(1+r^2)+(1+d_{4})^{2}]\pi_{5}^{2}. \label{v5}\end{aligned}\ ] ] we now find saddle points of @xmath295 given by @xmath296 we now consider a case for @xmath45 , @xmath53 and @xmath92 . we have two types of solutions as @xmath297 and @xmath298 . for @xmath297 , we have the two solutions @xmath299 and @xmath300 in ( [ chis])([chib ] ) as we discussed in sec . [ sec : slq ] . for @xmath298 , ( [ p4eq])([pi5eq ] ) give @xmath301 and @xmath302 . by substituting them into ( [ v5 ] ) the effective potential as a function of @xmath98 is given by @xmath303 on the other hand , for example , the effective potential for one of @xmath297 solutions @xmath300 is given by @xmath304 it is obvious @xmath305 for any value of @xmath98 . it indicates that the @xmath298 solution is unlikely to be a vacuum . we thus conclude that , at least in this framework , there is no pion condensate or no spontaneous parity symmetry breaking . we also perform the same analysis for the condensate form @xmath306 , and will find that @xmath307 is not a vacuum of the theory . a free action of creutz - borici fermion is given by @xmath308 \,,\\ \gamma & = \frac{1}{2}\sum_\mu \gamma_\mu , \quad \gamma_\mu^\prime = \gamma\gamma_\mu\gamma = \gamma-\gamma_\mu,\end{aligned}\ ] ] where @xmath309 corresponds to @xmath0 in karsten - wilczek fermion and @xmath310 satisfies @xmath311 and @xmath312 . in this case we have onsite operator and projection operators as @xmath313 and @xmath314 the strong - coupling analysis is done in a parallel manner to kw fermion . by taking @xmath315 , the corresponding gap equations become @xmath316 for @xmath92 , equations for second - order phase boundaries are given by @xmath317 then chiral boundaries are given by @xmath318 for @xmath45 chiral condensate is nonzero and chiral symmetry is spontaneously broken for @xmath319 .

we investigate thermodynamic and dynamical stability of a family of six - dimensional braneworld solutions with de sitter branes . first , we investigate thermodynamic stability in terms of de sitter entropy . we see that the family of solutions is divided into two distinct branches : the high - entropy branch and the low - entropy branch . by analogy with ordinary thermodynamics , the high - entropy branch is expected to be stable and the low - entropy branch to be unstable . next , we investigate dynamical stability by analysing linear perturbations around the solutions . perturbations are decomposed into scalar , vector and tensor sectors according to the representation of the 4d de sitter symmetry , and each sector is analysed separately . it is found that when the hubble expansion rates on the branes are too large , there appears a tachyonic mode in the scalar sector and the background solution becomes dynamically unstable . we show analytically that the onset of the thermodynamic instability and that of the dynamical instability exactly coincide . therefore , the braneworld model provides a new example illustrating close relations between thermodynamic and dynamical instability .

hep-th0703271

the braneworld scenario , in which matter fields are confined on a brane in higher dimensional spacetime , has been one of the most exciting subjects in the research into the early universe . as the early universe is supposed to have experienced a rather high - energy epoch , studies of the dynamics of the early universe might reveal as yet undiscovered imprints of extra dimensions to us . in our previous studies @xcite we considered a @xmath0-dimensional braneworld model with one or two @xmath1-branes . the shape and the size of the extra dimensions are stabilized by magnetic flux of a @xmath2 gauge field and the geometry on each brane is either minkowski or de sitter . since all moduli are stabilized , it is expected that the @xmath3-dimensional einstein gravity should be recovered on each brane . in ref . @xcite we analyzed how the hubble expansion rate on each brane changes when the brane tension changes . the resulting relation between the hubble expansion rate and the brane tension can be considered as an effective friedmann equation , which in turn defines an effective newton s constant . it was shown that at sufficiently low hubble expansion rates , the effective newton s constant resulting from this analysis agrees with that inferred by simply integrating extra dimensions out . the stability of this model was investigated in ref . @xcite , where it was shown that the background with minkowski branes is always stable against linear perturbations with axisymmetry in the extra dimensions . the purpose of this paper is to study stability of backgrounds with de sitter branes . this is not just a trivial extension of the previous analysis but actually results in a qualitatively different picture . indeed , the background becomes unstable when the hubble expansion rate becomes too large . ( for other models , see for example @xcite . ) we shall obtain the critical expansion rate at which the background becomes unstable and interpret this behavior in terms of de sitter thermodynamics . spacetimes with horizons appear mysterious in many respects . an observer does not have causal contact with any events beyond a horizon and , in this sense , the horizon separates the region behind it from the observer s side . thus , from the observer s viewpoint , information beyond the horizon is completely irrelevant ( as far as classical evolution in the observable region is concerned ) . this , nothing but a restatement of the definition of horizons , suggests a coarse - grained description of a physical system in or of a spacetime with horizons . it is therefore not totally surprising that a spacetime with horizons exhibits properties analogous to those of thermodynamics . indeed , in the case of black holes such properties are known as black hole thermodynamics . black hole entropy , defined as one quarter of horizon area in planck units , is central to black hole thermodynamics . the second law of black hole thermodynamics states that black hole entropy never decreases classically . its semi - classical extension , known as the generalized second law , states that the sum of black hole entropy and matter entropy does not decrease . these second laws are direct analogues of the second law of ordinary thermodynamics and restrict directions in which a physical system involving black holes can evolve . in ordinary thermodynamics it is due to the second law that entropy can be used to analyze stability of a thermodynamic system . therefore , the validity of the second law of black hole thermodynamics suggests potential use of black hole entropy to guess stability of a system including black holes or objects ( branes , strings , rings , and so on ) @xcite . ( also , for a review see e.g. @xcite and references therein . ) the close connection between black hole event horizons and thermodynamics can be extended to event horizons in cosmological models with a positive cosmological constant @xcite . the de sitter thermodynamics includes concepts of de sitter temperature and de sitter entropy . in this paper we shall extend the de sitter entropy to our braneworld set - up with two extra dimensions and use it to `` predict '' the stability condition for the backgrounds with de sitter branes . intriguingly , the simple `` prediction '' by de sitter entropy completely agrees with the result of rather involved analysis of linear perturbations . this is really non - trivial and may be considered as an evidence showing the usefulness of de sitter entropy and , more generally , holographic principle @xcite . the rest of this paper is organized as follows . in sec . [ sec:6dwfc ] we briefly review our @xmath0-dimensional braneworld model with de sitter branes . in sec . [ sec : thermodynamics ] we study thermodynamic stability of the spacetime by using entropy . in sec . [ sec : dynamical ] , in order to investigate dynamical stability we consider linear perturbation around the background solution . in sec . [ sec : summary ] we summarize the results and discuss them . we consider a 6d einstein maxwell system described by the action @xmath4 where @xmath5 is the 6d bulk cosmological constant @xmath6 and @xmath7 is the field strength of the @xmath2 gauge field @xmath8 introduced to stabilize the extra dimensions . ( we use units with @xmath9 where @xmath10 is the 6d newton s constant unless otherwise noted . ) the braneworld solution is @xmath11 where @xmath12 and @xmath13 is the metric of the 4-dimensional spacetime with a constant curvature @xmath14 , i.e. de sitter ( @xmath15 ) , minkowski ( @xmath16 ) , or anti - de sitter ( ads ) ( @xmath17 ) spacetime . without loss of generality , we can normalize @xmath14 to @xmath18 except for the minkowski ( @xmath16 ) case . we consider the range of parameters in which @xmath19 has two positive roots , and locate two 3-branes at the positions of the two roots @xmath20 @xmath21 . we call the branes at @xmath22 the @xmath23-branes , respectively . the tension of each brane determines a conical deficit in the extra dimensions . to be precise , the period of the angular coordinate ( @xmath24 ) is given by @xmath25 where @xmath26 are tensions of the branes at @xmath20 . this braneworld model includes two compact extra dimensions and two 3-branes at @xmath20 with the tensions @xmath26 . for @xmath27 , the geometries on the brane at @xmath20 are the 4d de sitter spacetimes with the hubble expansion rates @xmath28 , respectively . for @xmath16 , the brane geometry is the minkowski spacetime . in each case the energy scale on each brane is scaled by the warp factor @xmath29 depending on the position of the brane . the role of the @xmath2 field is to stabilize extra dimensions with magnetic flux . now , for later convenience we re - scale coordinates and parameters of the geometry with the 4d hubble parameter @xmath30 on the @xmath31-brane in the following manner : @xmath32 where we have omitted the subscript `` @xmath33 '' of @xmath30 and the tilde denotes re - scaled quantities . after this re - scaling , we obtain @xmath34 where @xmath35 here @xmath36 is the metric of the 4d de sitter spacetime with the hubble parameter @xmath37 . the metric function now satisfies @xmath38 , where @xmath39 is the ratio of the warp factors at the branes and , at the same time , describes how the geometry of the 2d extra dimensions is warped . note that in the special case @xmath40 the geometry is not warped but locally a round sphere . advantages of this re - scaled parameterization are as follows . first , the new parameters and coordinates remain finite and regular in the limit where the geometries on the branes become minkowski . second , since the warp factor evaluated at the @xmath31-brane is unity ( @xmath41 ) , many quantities rooted in these coordinates have physical meaning as observables measured by observers on the @xmath31-brane . as we will see later , a kaluza klein ( kk ) mass measured on the @xmath31-brane remains finite in the @xmath42 limit , while that on the @xmath43-brane diverges . hence , it is useful ( in particular for the purpose of numerical calculations ) to use quantities measured on the @xmath31-brane . of course , if we know the value of a quantity measured on the @xmath31 brane then we can easily obtain the corresponding value measured on the @xmath43-brane by multiplying the warp factor @xmath44 , e.g. , @xmath45 , @xmath46 , and so on , where @xmath47 is the kk mass observed on the @xmath48-brane . we introduce another convenient parameter @xmath49 . from @xmath38 we obtain the relation @xmath50 since @xmath51 , the hubble parameter @xmath37 has a maximum value @xmath52 for a given @xmath44 . hence we define @xmath49 by @xmath53 which is the squared hubble parameter @xmath54 on the @xmath31-brane normalized by @xmath52 . alternatively , @xmath49 could be defined as @xmath55 normalized by its maximum value , but the two definitions are actually equivalent . thus , @xmath49 describes how much the @xmath3d part of the metric is curved . when @xmath56 , the 4d spacetime becomes flat then the solution represents a braneworld with ads branes . in this paper , however , we shall not consider this case , focusing only on the de sitter ( @xmath57 ) and minkowski ( @xmath56 ) branes . ] . in contrast , in the maximal case ( @xmath58 ) , the flux disappears : @xmath59 . for a fixed bulk cosmological constant @xmath5 and a fixed discrete parameter @xmath14 ( @xmath60 ) , the solution is locally parameterized by the two parameters @xmath61 and @xmath62 . however it is more convenient to use the pair of dimension - less parameters @xmath63 , both of which run over the finite interval @xmath64 $ ] . this parameterization includes both @xmath65 ( @xmath57 ) and @xmath16 ( @xmath56 ) cases , and treats the de sitter and the minkowski branes in the common way . in the following sections we use the parameters ( @xmath44 , @xmath49 ) and the coordinates ( @xmath66 , @xmath67 , @xmath68 ) to describe the background geometry . as stated in the introduction , one of the main subjects in this paper is the close connection between thermodynamic and dynamical properties of spacetimes . in particular , we shall see exact agreement between thermodynamic and dynamical stabilities for the @xmath0-dimensional brane world solutions described in the previous section . in this section we discuss thermodynamic stability by using the de sitter entropy . we also point out the close relation between the specific heat and the effective friedmann equation . in the @xmath3-dimensional einstein gravity , a de sitter space has entropy given by one quarter of the area of the horizon @xcite . this is an analogue of the well - known black hole entropy and has been the basis of holographic arguments for de sitter spaces @xcite . in the 6d braneworld background solution , each point in the 2d extra dimensions corresponds to a 4d de sitter space . therefore , we shall define the de sitter entropy in our context by one quarter of the area of the cosmological horizon integrated over the extra dimensions , i.e. , one quarter of the volume of the 4-dimensional surface foliated by de sitter horizons : @xmath69 where @xmath70 is the 4-dimensional volume given by @xmath71 and @xmath10 is the 6d newton s constant . note that , for the convenience of the discussion , we have temporarily restored @xmath10 in this paragraph . actually , it is easy to show that this definition agrees with the definition of the 4d de sitter entropy that observers on each brane would adopt . the area of the de sitter horizon and the newton s constant on the @xmath72-brane are , respectively , given by @xmath73 and @xmath74 thus , the 4d de sitter entropies on the @xmath72-branes are @xmath75 respectively . these do indeed agree with the above definition based on the horizon area integrated over the extra dimensions : @xmath76 in the 4d picture , the integration over extra dimensions is taken care of by the above formula of the 4d newton s constant ( [ eqn : gn ] ) . since the three different definitions agree , we have a unique definition of de sitter entropy in our braneworld set - up . to investigate the thermodynamic stability of the system , it is useful to consider the de sitter entropy as a function of conserved quantities . the family of solutions described in the previous section has three conserved quantities : the total magnetic flux and the tensions of two branes . the total magnetic flux @xmath77 is given by @xmath78 by eq . ( [ eq : delta_phi ] ) , tensions @xmath79 of the @xmath72-branes are expressed as @xmath80 respectively , where @xmath81 are defined by @xmath82 since there is one - to - one correspondence between @xmath79 and @xmath81 , we can consider the de sitter entropy @xmath83 as a function of either ( @xmath77 , @xmath84 , @xmath85 ) or ( @xmath77 , @xmath86 , @xmath87 ) . variables @xmath83 , @xmath77 and @xmath88 defined above are proportional to @xmath89 . because of this trivial dependence , we can eliminate @xmath89 from all thermodynamic considerations by properly normalizing the variables . in particular , in appendix [ sec : thermodynamic_relatios ] it is shown that the entropy normalized by @xmath87 , @xmath90 , satisfies the differential relation @xmath91 where @xmath92 , and @xmath93 is the normalized total flux . this means that the normalized entropy @xmath94 is a function of the two variables @xmath95 and @xmath96 : @xmath97 . hereafter , @xmath95 and @xmath96 are to be considered as two independent thermodynamic variables . this accords with the observation made in the previous section : the family of solutions are parameterized by two parameters ( @xmath44 , @xmath49 ) after proper scaling of variables . the angular period @xmath89 has been eliminated so that @xmath94 , @xmath95 and @xmath96 do not depend on it . note that the normalization factor @xmath98 is just a fixed constant from the viewpoint of observers on the @xmath31-brane since these observers are to see the response of the 4d geometry induced on the @xmath31-brane to the change of the tension @xmath84 or equivalently , @xmath86 . in particular , with @xmath87 fixed , we have @xmath99 thus , the relation ( [ eq : first_law ] ) is relevant for observers on the @xmath31-brane ( but not for those on the @xmath43-brane ) . alternatively , we could write down a differential relation relevant for observers on the @xmath43-brane by normalizing the variables by the factor @xmath100 , which is a constant from the viewpoint of observers on the @xmath43-brane . in this sense , we have two conceptually different pictures : one from @xmath31-brane perspectives and the other from @xmath43-brane perspectives . the result of the thermodynamic stability analysis does not depend on the picture : the result from one picture completely agrees with that from the other picture . in the following , we shall see the thermodynamic stability from the @xmath31-brane perspectives . thus , unless there is a possibility of confusion , we omit the subscripts `` @xmath101 '' . as a function of @xmath95 and @xmath96 . different points on the surface @xmath102 represent different background solutions , and there is no background solution outside the domain of @xmath103 . it is easily seen that @xmath103 is not single - valued but actually double - valued in some part of its domain . thus , the family of background solutions has two branches , divided by a critical curve on which some derivatives of @xmath103 are singular . the high - entropy branch is thermodynamically preferred to the low - entropy branch . ] [ fig : entropy ] shows the graph of the normalized de sitter entropy @xmath94 as a function of the conserved quantities @xmath95 and @xmath96 . as we can easily see , @xmath103 is not single - valued but actually double - valued in some part of its domain . it is also easy to see that derivatives of @xmath103 are singular on a part of the boundary of the domain . we call this part of the boundary of the domain the _ critical curve_. different points on the surface @xmath102 represent different background solutions , i.e. different spacetimes . ( there is no background solution outside the domain of @xmath103 . ) therefore , the critical curve divides the family of solutions into two distinct branches : one with larger @xmath94 and the other with smaller @xmath94 . we call them the _ high - entropy branch _ and the _ low - entropy branch _ , respectively . by definition , these two branches merge on the critical curve . note that in some region of the low - entropy branch , @xmath103 is single - valued . if the de sitter entropy has any physical meaning like entropy in ordinary thermodynamics then the high - entropy branch must be thermodynamically preferred in some sense and must be more stable than the low - entropy branch . the spacetime is marginally stable on the critical curve . these observations are based only on our intuition that the de sitter entropy should play a role similar to that of entropy in ordinary thermodynamics . nonetheless , surprisingly , we shall see in the next section that these observations correctly `` predict '' the result of rather detailed analysis of dynamical stability against linear perturbations . we now derive an explicit equation describing the critical curve . for this purpose , we consider the coordinate transformation from @xmath104 defined in the previous section to @xmath105 defined above . as we shall see below , the critical curve is the set of points where this coordinate transformation becomes singular . as noticed in sec . [ sec:6dwfc ] , @xmath44 ( @xmath106$])is the ratio of the warp factors of two branes and describes the shape of the two - dimensional extra space . on the other hand @xmath49 ( @xmath107 $ ] ) represents the curvature of 4d de sitter space normalized by the maximum value @xmath108 for a given @xmath44 . since the background solution is uniquely specified by @xmath104 up to the scaling explained in the previous section , any ( properly normalized ) quantities can be rewritten as functions of @xmath104 . in particular , we have two single - valued functions @xmath109 and @xmath110 as a map @xmath111 . the entropy @xmath94 is also a single - valued function of @xmath104 and , thus , the critical curve is determined by vanishing jacobian determinant of this map : @xmath112 this condition can be solved with respect to @xmath49 to give the following expression for the critical value of @xmath49 : @xmath113 . \end{aligned } \label{eq : critical_curve}\ ] ] this is the analytic expression for the critical curve . as we see later , @xmath114 ( or , @xmath115 and @xmath116 ) is nothing but the critical value for stabilizing the extra dimensions like in freund rubin compactification @xcite . we have seen that the ( normalized ) de sitter entropy is not a single - valued function but actually a double - valued function of the conserved quantities @xmath105 in some region of the domain . this observation has led us to a rather natural stability criterion that the high - entropy branch should be stable and that the low - entropy branch should be unstable . this is a global statement : we would not be able to say one of the two branches should be stable or unstable without knowing the existence of the other branch . in this subsection , we look at the stability from a different viewpoint , considering the effective friedmann equation investigated in ref . @xcite . in particular , we shall see that the thermodynamic stability is equivalent to the positivity of an _ effective newton s constant_. note that , while the thermodynamic stability is a global statement , the positivity of the effective newton s constant is a local statement . in our previous work @xcite we showed that the 4d friedmann equation is recovered on the brane at sufficiently low energy as the response of the 4d hubble expansion rate to the change of the brane tension . to be more precise , for small hubble expansion rates @xmath37 , @xmath54 is expanded as @xmath117 where @xmath118 is the brane tension and the 4d newton s constant @xmath119 is obtained by simply integrating extra dimensions out as ( [ eqn : gn ] ) . the constant @xmath120 is the value of @xmath118 for the minkowski brane ( @xmath121 ) and depends on @xmath77 and @xmath85 , which are actually constants from the viewpoint of observers on the @xmath31-brane . note that the 4d newton s constant @xmath119 defined as ( [ eqn : gn ] ) is always positive . the terms of order @xmath122 and higher are small compared with the linear term if the hubble expansion rate is sufficiently lower than the energy scale set by the bulk cosmological constant . however , for larger @xmath54 , the higher order corrections become relevant and @xmath54 is no longer approximately linear . moreover , as we can see from fig . [ fig : hubble ] , @xmath54 is not necessarily an increasing function of the brane tension . indeed , if @xmath54 is larger than a critical value then it is a decreasing function of the brane tension . to make this peculiar behavior more quantitative , let us define the effective newton s constant @xmath123 on the @xmath31-brane by @xmath124 and characterize the critical value by @xmath125 . for @xmath37 smaller ( or larger ) than the critical value , @xmath123 is positive ( or negative , respectively ) . note that , since this effective newton s constant runs as the hubble expansion rate @xmath37 changes , the exact agreement with @xmath119 is met only in the @xmath126 limit . the agreement in this limit itself is a rather non - trivial consequence of the dynamics of bulk spacetime as explicitly shown in @xcite , but physical interpretation is simple . while @xmath119 was determined via the coefficient of the contribution of the graviton zero mode to the higher dimensional action , @xmath123 here incorporates contributions of all kaluza klein modes as well as the zero mode . since kaluza klein contributions are irrelevant at low energies , these two definitions should agree . on the other hand , at high energies the kaluza klein modes become relevant and make @xmath123 deviate from @xmath119 . , as a function of @xmath95 and @xmath96 . for small @xmath54 , @xmath127 is negative . however , as @xmath54 increases , the slope turns over and @xmath127 becomes positive . ( this figure is identical to fig . 3 of ref . @xcite . ) ] in order to examine the curve @xmath125 , the following observation is useful : @xmath128 since @xmath37 and @xmath96 are single - valued smooth functions of @xmath63 , the jacobian @xmath129 does not diverge . hence , @xmath123 diverges if and only if the denominator @xmath130 vanishes . this condition is nothing but eq . ( [ eqn : critical - curve - jacobian ] ) characterizing the critical curve , which divides the parameter space of background solutions into the high - entropy branch and the low - entropy branch . therefore , @xmath131 additionally , it is worth mentioning that the effective newtion s constant has a connection with thermodynamic quantities on the brane . in the 4d case we can define a temperature of the de sitter space by the hubble parameter @xmath132 as @xmath133 . then , in a straightforward manner , the specific heat is given by @xmath134 now , concerning the 4d de sitter brane , we consider a specific heat regarding @xmath135 as its temperature . from eq . ( [ eq : first_law ] ) we obtain @xmath136 this expression is the same as that of the 4d de sitter space except for its gravitational `` constant . '' for the 4d de sitter brane the sign of the specific heat will change on the critical curve also . hence , from the viewpoint of the 4d effective theory , we conclude that the effective newton s constant on the brane determines thermodynamic properties of the braneworld . in the previous section we have discussed the thermodynamic stability of the 6d braneworld model as a property of the background solution . now let us investigate the dynamical stability of this spacetime . for this purpose , we consider linear perturbations around the background solution . perturbations are decomposed into scalar- , vector- and tensor - sectors according to the representation of the 4d de sitter symmetry . if the mass squared of the kaluza klein modes @xmath137 is negative , then the corresponding mode is tachyonic and unstable . since the background spacetime forms a two - parameter family of solutions , the system of the perturbation equations depends on two parameters ( @xmath44,@xmath49 ) . we investigate the lowest mass squared @xmath138 in each sector as a function of the two parameters in the square domain @xmath139 , @xmath140 . our strategy for attacking the problem is as follows . first , we analytically solve the perturbation equations in the @xmath40 case for arbitrary @xmath141 $ ] , and obtain @xmath142 on the side @xmath143 of the square domain . next , by successively applying the relaxation method , we numerically calculate @xmath138 as @xmath44 changes from @xmath144 to @xmath145 for every @xmath49 . in each step of the second procedure , the result of the previous step with a slightly larger value of @xmath44 is used as an initial guess for the relaxation method . as shown in appendix [ app : gauge ] , the scalar perturbation with an appropriate gauge choice is given by @xmath146\frac{\mathrm dr^2}{f } + [ 1 - ( \phi_1 + 3\phi_2)y]f\mathrm d\phi^2,\\ a_m\mathrm dx^m & = ( a + a_\phi y)\mathrm d\phi , \end{aligned}\ ] ] where @xmath147 is the scalar harmonics on the 4d de sitter space satisfying @xmath148 here , @xmath149 is the covariant derivative associated with the 4-dimensional de sitter spacetime with the hubble @xmath37 . the einstein equation and the maxwell equation are reduced to the following two perturbation equations for @xmath150 and @xmath151 : @xmath152 where the prime denotes the derivative with respect to @xmath66 , and @xmath137 is the eigenvalue of the harmonics . the boundary conditions at @xmath153 are obtained by setting the coefficients of @xmath154 in ( [ eq : scalar_perturbation - original ] ) to zero at the brane positions , as in @xcite . hence we have @xmath155 an alternative and more rigorous derivation of the boundary conditions is to use the formalism developed in @xcite . the result is @xmath156_{r=\alpha,1 } = 0 . \label{eq : scalar_bc - original}\ ] ] these two sets of boundary conditions lead to the same taylor expansion of @xmath150 and @xmath151 in the neighborhood of the boundaries and , thus , are equivalent . in the original coordinate system @xmath157 the geometry in the @xmath143 limit appears to be singular because two positive roots of @xmath158 corresponding to the brane positions coincide . actually , the proper distance between the @xmath48-branes remains finite and it turns out that the apparent singularity due to coordinate artifacts is removed by appropriate coordinate transformations . indeed , in the coordinate system ( @xmath159,@xmath160 ) defined by @xmath161 the geometry is obviously regular in the @xmath143 limit . note that the coordinate @xmath159 ranges over the interval @xmath162 $ ] . since the branes are always located at @xmath163 for any @xmath44 ( even for @xmath40 ) , the new coordinate @xmath159 is more useful for numerical calculation than the original one @xmath66 . therefore we rewrite the perturbation equation in terms of @xmath159 as @xmath164}\right\}\partial_w\phi_1 - \frac{1}{\bar{f}}(\phi_1 + \phi_2 ) + \frac{2\mu^2 + 18h^2/\lambda_6}{[(1-\alpha)w+(1+\alpha)]^2\bar{f}}\phi_1 & = 0,\\ \partial_w^2\phi_2 + \frac{4(1-\alpha)}{[(1-\alpha)w+(1+\alpha)]}\partial_w\phi_2 + \frac{\mu^2}{[(1-\alpha)w+(1+\alpha)]^2\bar{f}}(\phi_1 + 2\phi_2 ) & = 0 , \label{eq : scalar_perturbation } \end{aligned}\ ] ] where @xmath165 ^ 2 + \frac{4 ( 1+\alpha+\alpha^2+\alpha^3+\alpha^4 ) \beta } { 1+\alpha+\alpha^2 } \right.\\ & \left . + \frac{32 ( 1+\alpha ) [ 1+\alpha^2 ( 1+\alpha^2+\alpha^4 ) ( 1-\beta)-(1+\alpha^3 ) \beta ] } { [ ( 1-\alpha)w + ( 1+\alpha)]^3 ( 1+\alpha+\alpha^2 ) } - \frac{256 \alpha^3 ( 1+\alpha+\alpha^2+\alpha^3+\alpha^4 ) ( 1-\beta ) } { [ ( 1-\alpha)w + ( 1+\alpha)]^6 ( 1+\alpha+\alpha^2 ) } \right\ } \end{aligned}\ ] ] and @xmath166 the boundary condition is written as @xmath167 ^ 2}\phi_1 \right|_{w=\pm 1}&=0 , \\ \left . ( \phi_1 + 2\phi_2)\right|_{w=\pm 1 } & = 0 . \end{aligned } \label{eq : scalar_bc}\ ] ] now we show the analytic solution for @xmath40 . in this case the geometry of the extra dimensions becomes locally a round 2-sphere as in ref . @xcite and the warp factor is a constant . hence , the background spacetime for @xmath40 corresponds to the football - shaped extra dimensions @xcite . in the @xmath40 limit , the perturbation equations ( [ eq : scalar_perturbation ] ) are reduced to @xmath168 and the boundary conditions are given by @xmath169 the solution satisfying these boundary conditions is @xmath170 , \quad \phi_2^{(\pm ) } = p_n(w),\ ] ] up to an overall constant , where @xmath171 , and @xmath172 is the legendre function of the first kind . we obtain the mass spectrum of the scalar perturbation as @xmath173 the mass of the lowest mode is given by @xmath174 thus , for @xmath175 , the lowest @xmath176 becomes negative and the background spacetime is destabilized@xcite . now we numerically solve the perturbation equations and obtain the lowest kk mass of the scalar mode for general @xmath44 . using the analytic solution for @xmath40 as a first initial guess , we solve the problem for a slightly smaller value of @xmath44 by the relaxation method . then , as we change @xmath44 to smaller values towards @xmath145 step by step , we in turn use the numerical solution of the previous step as an initial guess for the relaxation method and obtain the mass squared for each @xmath44 and @xmath49 . for the lowest mode of the scalar perturbation . when @xmath56 the 4d spacetime is the minkowski . as @xmath49 becomes larger ( i.e. the 4d hubble becomes larger ) , @xmath137 decreases and eventually becomes negative beyond some value of @xmath49 . the line represents the boundary between the stable and the unstable region , where the massless modes appear . ] [ fig : scalar_kkmass ] shows @xmath137 for the lowest mode of the scalar perturbation . for @xmath56 , which means the 4d spacetime is the minkowski , we have reproduced the result of our previous study @xcite that there is no unstable ( i.e. neither massless nor tachyonic ) mode . we find that as @xmath49 becomes larger , namely as the hubble parameter on the brane becomes larger , @xmath137 decreases and eventually becomes negative . therefore we conclude that when the hubble on the brane is too large the extra dimensions are destabilized and such configurations are unstable for general @xmath44 . in addition , in the case of vanishing flux ( @xmath58 ) the configurations are always unstable . supported by the result shown in fig . [ fig : scalar_kkmass ] , the assumption that @xmath137 is a continuous function of @xmath44 and @xmath49 is made . hence , it must vanish on the boundary between the stable and unstable regions . in other words , there should appear a massless mode as one moves from the stable / unstable region to the unstable / stable region . we can actually obtain this threshold massless mode analytically . the general solution of ( [ eq : scalar_perturbation - original ] ) with @xmath177 is given by @xmath178 where @xmath179 are arbitrary constants . by imposing the boundary conditions ( [ eq : scalar_bc - original ] ) , we will be able to obtain the boundary between the stable and the unstable region on the @xmath63 plane , where the massless modes appear . the boundary conditions lead to @xmath180 where @xmath62 and @xmath61 are functions of @xmath104 determined by @xmath38 . in order for this equation to allow a non - trivial solution for @xmath181 , the determinant of the matrix must vanish . from this condition , we obtain the relation between @xmath63 as @xmath182 . \end{aligned}\ ] ] this relation is exactly identical with ( [ eq : critical_curve ] ) , which defines the critical curve on which the high - entropy branch and the low - entropy branch merge . the threshold massless mode divides the whole square @xmath63-domain into positive and negative @xmath137 regions , while the critical curve in sec . [ sec : thermodynamics ] divides the whole square @xmath63-domain into high- and low - entropy regions . therefore , the high - entropy and the low - entropy branches in fig . [ fig : entropy ] exactly agree with the positive and negative @xmath137 regions in fig . [ fig : scalar_kkmass ] , respectively . this fact implies that in this spacetime thermodynamically preferred configurations are dynamically stable , while thermodynamically non - preferred ones have tachyonic modes and become dynamically unstable . the onset of dynamical instability coincides with the onset of thermodynamic instability . this is similar to the gubser mitra conjecture for extended black objects @xcite . as shown in appendix [ app : gauge ] , the vector perturbation with an appropriate gauge choice is given by @xmath183 where @xmath184 is the vector harmonics on the 4d de sitter space , which satisfies the transverse condition , @xmath185 . the einstein equation and the maxwell equation are reduced to two perturbation equations with respect to @xmath186 and @xmath187 , @xmath188 ' - \frac{2b}{r^4}a'_{(\mathrm t ) } + \frac{m^2 + 6h^2}{r^2f}h_{(\mathrm t)\phi } & = 0,\\ ( r^2f a'_{(\mathrm t ) } ) ' + b\left ( \frac{h_{(\mathrm t)\phi}}{r^2}\right ) ' + m^2 a_{(\mathrm t ) } & = 0 , \end{aligned}\ ] ] where @xmath137 is the eigenvalue of vector harmonics on the de sitter space with the hubble constant @xmath37 : @xmath189 it is again convenient to use the coordinates ( @xmath159,@xmath160 ) defined in eq . ( [ eq : new_coordinates ] ) which do not have any singular behaver in the limit @xmath190 . the perturbation equations and the boundary conditions requiring regularity at the brane positions are written in the coordinates ( @xmath159,@xmath160 ) as @xmath191 - \frac{1}{r^4}\partial_w\psi_1 + \frac{\mu^2 + 3h^2/\lambda_6}{2r^2\bar{f}}\psi_2 & = 0,\\ \partial_w(r^2\bar{f } \partial_w\psi_1 ) + \frac{b^2}{2}\partial_w\left(\frac{\psi_2}{r^2}\right ) + \frac{\mu^2}{2 } \psi_1 & = 0 , \end{aligned}\ ] ] and @xmath192 where @xmath193 note that the factor @xmath194 in the definition of @xmath195 reflects the fact that the relation between the original angular coordinate @xmath67 and the new coordinate @xmath160 includes the factor @xmath196 : @xmath197 . on the other hand , the factor @xmath61 in the definition of @xmath198 is not essential but is just to absorb the factor @xmath61 in the perturbation equations . for @xmath40 the perturbation equations become @xmath199 + \frac{8(1-\beta)}{4 - 3\beta}\partial_w\psi_2 + \frac{4\mu^2}{4 - 3\beta}\psi_1 & = 0,\\ ( 1-w^2 ) [ \partial_w^2\psi_2-\partial_w\psi_1 ] + \frac{4\mu^2 + 2\beta}{4 - 3\beta}\psi_2 & = 0 , \end{aligned}\ ] ] where @xmath200 . the solution satisfying the boundary conditions is given by @xmath201 up to an overall constant , and the mass spectrum of the vector perturbation @xmath176 is @xmath202 where @xmath203 for @xmath204 and @xmath205 for @xmath206 . there are two lowest mass modes for @xmath207 and @xmath208 , and both are massless . in other words the vector perturbations have two zero modes : @xmath209 where @xmath210 and @xmath211 are arbitrary constants . as shown below for general @xmath44 , these two zero modes represent physical degrees of freedom of the maxwell field and the gravi - photon . for general @xmath44 we find two zero modes corresponding to the above ones as @xmath212 where @xmath210 and @xmath211 are arbitrary constants . it is obvious that the degree of freedom represented by @xmath210 originates from the original 6d maxwell field and , thus , is associated with the 4d part of the @xmath2 gauge symmetry . on the other hand , as seen below , the other degree of freedom represented by @xmath211 is associated with coordinate transformation of the angular coordinate @xmath67 and , thus , can be regarded as a gravi - photon . we now see this explicitly . let us consider an infinitesimal @xmath2 gauge transformation represented by a parameter @xmath213 and an infinitesimal coordinate transformation @xmath214 , where @xmath213 and @xmath215 depend only on the 4d coordinates . under these , the @xmath216-components of the perturbation of the u(1 ) gauge potential and the ( @xmath217)-components of the metric perturbation transform as @xmath218 and all other components are unchanged . thus , if we define two @xmath3-vectors @xmath219 and @xmath220 as @xmath221 then the above transformation law is rewritten as two separate @xmath2 gauge transformations : @xmath222 clearly , @xmath210 and @xmath211 above are coefficients of the transverse components of @xmath219 and @xmath220 , respectively . thus , any instability will not occur for these two modes . having seen that the two zero modes of the vector perturbation remain massless for any @xmath44 , we will now examine the 1st kk modes for general @xmath44 numerically . for the first two kk modes of the vector perturbation . when @xmath40 , the values correspond to @xmath204 for @xmath223 ( left ) and @xmath206 for @xmath224 ( right ) . these are non - negative for the entire region of @xmath104.,title="fig : " ] for the first two kk modes of the vector perturbation . when @xmath40 , the values correspond to @xmath204 for @xmath223 ( left ) and @xmath206 for @xmath224 ( right ) . these are non - negative for the entire region of @xmath104.,title="fig : " ] fig . [ fig : vector_kk ] shows @xmath137 for the first two kk modes of the vector perturbation . we find that @xmath137 is non - negative for the entire region of @xmath63 . therefore , there are no unstable modes in the vector sector . the tensor perturbation is given by @xmath225 where @xmath226 is the tensor harmonics on the 4d de sitter space , which satisfies the transverse and traceless conditions , @xmath227 , and @xmath228 the perturbed einstein equation becomes @xmath229 and there is no relevant equation coming from the maxwell equation . with the coordinate @xmath159 , this is written as @xmath230 ^ 2}\partial_w\left\{[(1-\alpha)w+(1+\alpha)]^4\bar{f}\partial_w h_{(\mathrm t)}\right\ } + 2\mu^2 h_{(\mathrm t ) } = 0.\ ] ] for @xmath143 the perturbation equation becomes @xmath231 + \frac{4\mu^2}{4 - 3\beta } h_{(\mathrm t ) } = 0.\ ] ] the regular solution is given by @xmath232 where @xmath172 is the legendre function of the first kind . the kk mass squared @xmath176 is @xmath233 where @xmath171 . it is obvious that the lowest - mass mode is @xmath234 and it is massless , representing the 4d graviton . for the tensor perturbation it is clear that the zero mode is @xmath235 , i.e. , a homogeneous mode . for the first kk mode of the tensor perturbation . the lower plane describes the particular mass , @xmath236 , for the tensor mode on the de sitter space . ] [ fig : tensor_kk ] shows @xmath137 for the first kk mode of the tensor perturbation . we find that @xmath137 for the kk modes is non - negative is forbidden by unitarity @xcite . also , the masses in the region @xmath237 are called _ complementary series _ @xcite . now , the masses for the kk modes satisfy @xmath238 . ] for the entire region of @xmath63 . therefore , there is no unstable mode in the tensor sector . ( see e.g. @xcite . ) in this paper we have studied stability of the 6d braneworld solution with 4d de sitter branes from two different perspectives . one is thermodynamic stability of the braneworld solution . we have defined the de sitter entropy in the six - dimensional braneworld by one quarter of the area of the cosmological horizon integrated over the extra dimensions . we have seen that this definition agrees with the @xmath3-dimensional de sitter entropy defined on each brane . as shown in fig . [ fig : entropy ] , the de sitter entropy as a function of the conserved quantities @xmath95 and @xmath96 is not single - valued but double - valued . therefore , the de sitter entropy divides the family of solutions into two branches , one with higher entropy ( the high - entropy branch ) and the other with lower entropy ( the low - entropy branch ) , and defines the boundary between them ( the critical curve ) . the other is dynamical stability of the background solutions against linear perturbations . perturbations are decomposed into the scalar- , vector- and tensor - sectors , according to the @xmath3-dimensional de sitter symmetry , and we have calculated the lowest mass squared in each sector . we have found that when the hubble expansion rate on the brane is larger than a critical value , there is a tachyonic mode in the scalar sector and thus the background is unstable . at the critical value , there appears a threshold massless mode in the scalar sector . on the other hand , there is no unstable mode in the vector- and tensor - sectors . we have found that the critical value at which the threshold massless mode appears is exactly on the critical curve dividing the family of solutions into the high - entropy branch and the low - entropy branch . therefore , we have shown that the low - entropy branch is dynamically unstable while the high - entropy branch is dynamically stable . moreover , we have also seen that the @xmath3-dimensional effective newton s constant is positive in the high - entropy branch and negative in the low - entropy branch . in summary , we have shown the equivalence of the following three conditions : * thermodynamic stability . * dynamical stability . * positivity of the effective newton s constant . the close connection between thermodynamic stability and dynamical stability has already been pointed out in the literature @xcite . the result of the present paper , thus , adds yet another example to the list of systems exhibiting the close connection between thermodynamic and dynamical properties . in the previous examples in the literature , however , one has to rely on numerical analysis to show the equivalence and the boundary between stable and unstable regions is not obtained rigorously . on the other hand , our system was simple enough to obtain the boundary analytically . in this sense , we may say that the analysis in the present paper is the first example in which the equivalence between thermodynamic stability and dynamical stability was proved . moreover , we have shown that for our gravitational system the thermodynamic and dynamical properties of the spacetime is closely related to the sign of the effective newton s constant . non - linear evolution of the unstable solutions is one of the important subjects for the near future . while we have shown that the low - entropy branch is unstable , our analysis is limited by the linearized approximation and we do not know to what final state the spacetime will evolve from the low - entropy branch . a naive guess is that the system should evolve to the corresponding solution in the high - entropy branch with the same values of the conserved quantities , but there remains the possibility that the system might evolve towards a big crunch singularity or evolve to another unknown configuration @xcite . indeed , as seen from fig . [ fig : entropy ] , for a solution in the low - entropy branch with a very large hubble parameter close to the maximum value , there is no corresponding solution in the high - entropy branch . however , one can show that there is a solution with ads branes for the same conserved quantities @xmath95 and @xmath96 . this indicates that the frw cosmology on the brane in this regime should evolve towards a big crunch singularity unless there is another unknown stable solution without 4d maximal symmetry . it is worthwhile seeking such solutions without 4d maximal symmetry and also analyzing properties of solutions with ads branes @xcite . the whole phase structure including the high- and low - entropy branches , the ads branch and possibly other new branches is unexplored . inclusion of general matter on the brane is also an interesting subject as a future work . in the present set - up , the 4d spacetime on the branes has the maximal symmetry and the stress energy tensor on the brane is restricted to the form of vacuum energy , or brane tension . it is certainly interesting to investigate dynamical stability and thermodynamic properties of the braneworld with more general friedmann universe and matter contents on the brane . for this purpose , we need some extensions of the model to include arbitrary matter on the codimension-2 brane @xcite . it is also interesting to consider supergravity extensions . we expect that the close connection between thermodynamic stability and dynamical stability could be extended to models with branes of higher codimensions , i.e. models with more extra dimensions . we hope to investigate this subject in the future . here , as the first step , let us briefly discuss extensions to models with more extra dimensions but without branes . we consider general freund rubin flux compactifications with @xmath239 . the ( @xmath240)-dimensional action is @xmath241 where @xmath242 is the @xmath243-form field for stabilizing the @xmath243-sphere . the metric and the @xmath243-form flux are given by @xmath244 and @xmath245 where @xmath246 is the volume element of the @xmath243-sphere . the einstein and maxwell equations reduce to two algebraic equations : @xmath247 the entropy @xmath83 is given by @xmath248 where @xmath77 is the total flux defined as @xmath249 and @xmath250 are respectively the volume of the unit @xmath251-sphere and @xmath243-sphere . from ( [ eq : algebraic_relations ] ) we can express @xmath83 as a function of one parameter . in this case we can also see that the entropy @xmath252 as a function of @xmath77 is not single - valued and splits into a high - entropy branch and a low - entropy branch . the critical point dividing into two branches becomes @xmath253 which is determined by @xmath254 . these values are nothing but the threshold at which a tachyonic mode appears in the scalar sector @xcite . therefore , we have shown the equivalence of thermodynamic stability and dynamical stability in the general freund rubin flux compactifications with @xmath239 . it is worthwhile investigating a similar relation in models with branes of codimensions higher than two . we would like to thank kei - ichi maeda for valuable comments . the work was in part supported by jsps through a grant - in - aid for the 21st century coe program `` quantum extreme systems and their symmetries '' ( s.k . ) and a grant - in - aid for jsps fellows ( y.s . ) and by mext through a grant - in - aid for young scientists ( b ) no . 17740134 ( s.m . ) . in this appendix , we show the thermodynamic relations in the 6d braneworld solution with the 4d de sitter space . the brane positions are determined by the two positive roots of @xmath255 , where @xmath256 eliminating @xmath62 from the conditions @xmath257 and @xmath258 , we obtain @xmath259 and this yields @xmath260 this is the relation between the total area of the horizon @xmath261 , the total magnetic flux of @xmath2 field @xmath77 , and the proper periods at the brane positions @xmath88 which are respectively defined as @xmath262 and @xmath263 here @xmath89 is a given period of the angular coordinate @xmath67 of extra dimensions . recalling eq . ( [ eq : delta_phi ] ) , we find @xmath264 where @xmath26 are tensions of the @xmath48-brane . we note that eq . ( [ eq : smarr_relation ] ) is similar to the thermodynamic relation for the 6d rnds black hole . furthermore we can obtain the differential relation for the above quantities which is similar to the first law of black hole mechanics . the parameters of this spacetime are determined by two equations , @xmath265 and @xmath266 ; then the variation of parameters in the two equations becomes @xmath267 eliminating @xmath268 , we have @xmath269 although the period of the angular coordinate , @xmath89 , is necessary for determining the global geometry of the spacetime , locally it is irrelevant . in other words , @xmath261 , @xmath77 and @xmath88 in eq . ( [ eq : smarr_relation ] ) are `` extensive '' quantities with respect to the normalization of @xmath89 . therefore we express the above relations in terms of following `` intensive '' quantities , @xmath270 , @xmath271 , and @xmath272 . using these quantities normalized by @xmath87 it is convenient to discuss the case with the tension of the @xmath43-brane fixed . from eqs . ( [ eq : smarr_relation ] ) and ( [ eq : differential_relation ] ) , we have @xmath273 and @xmath274 combining these two equations we obtain another expression , @xmath275 when the tension of the @xmath43-brane is fixed , the variation of @xmath95 is directly related to the variation of the tension of the @xmath31-brane , i.e. , @xmath276 . in this sense the above expression is important . in this appendix , we revisit the derivation of the thermodynamic relations from the euclidean action . ( see , for example , @xcite . ) we take an ansatz for the euclidean 6d metric with topology @xmath277 , @xmath278,\ ] ] and for the @xmath2 field , @xmath279 the 4-dimensional part of the metric is a 4-sphere with the radius @xmath37 , which is the euclidean continuation of the 4d de sitter space with the hubble parameter @xmath37 by the wick rotation of the time coordinate @xmath280 . the euclidean time @xmath281 has a period @xmath282 related with the de sitter temperature . we obtain a component of the einstein tensor , @xmath283 ' , \ ] ] and the non - vanishing components of the energy - momentum tensor are @xmath284 the nontrivial maxwell equation is @xmath285 which is integrated to yield @xmath286 where @xmath61 is the integration constant . the component of einstein equation , @xmath287 is @xmath288 ' + \lambda_6 - \frac{b^2}{2r^8 } = 0,\ ] ] and it is easily integrated to obtain @xmath289 where @xmath62 is the integration constant . the ricci scalar and the field strength are @xmath290 ' - \frac{2}{r^4xy } \left(\frac{r^4x'}{y}\right)',\ ] ] and @xmath291 the euclidean action is given by @xmath292 ' - 2 \left(\frac{r^4x'}{y}\right ) ' - 2\lambda_6 r^4xy - \frac{r^4a'^2}{xy } \right\}\\ & = - \frac{\delta\phi\beta}{6h^3 } \int_\alpha^1 \mathrm dr \left\{ba ' - \left(\frac{r^4x'}{y}\right)'\right\}\\ & = -\frac{\beta}{6h^3}(b\phi + \kappa_+ + \alpha^4\kappa_- ) , \end{aligned}\ ] ] where @xmath293 and @xmath294 note that @xmath88 and @xmath77 are the boundary data held fixed in the current action . here , @xmath49 is the inverse of the temperature of the de sitter horizon , i.e. , @xmath295 . two 3-branes are located at @xmath153 , so we require @xmath296 the regularities at @xmath153 lead to @xmath297 , and we obtain @xmath298 from these two conditions we can determine the two integration constants @xmath299 : @xmath300 the reduced action @xmath301 for fixed @xmath88 and @xmath77 is @xmath302 extremizing this action with respect to @xmath44 and @xmath37 gives us @xmath303 these two conditions are reduced to @xmath304 where @xmath305 is defined as @xmath306 we denote @xmath307 satisfying the conditions ( [ eq : equilibrium_condition ] ) for given @xmath88 and @xmath77 as @xmath308 and @xmath309 , and we have a classical action @xmath310 & \equiv i^*(\phi , \kappa_+ , \kappa_- ; \alpha_{\mathrm e } , h_{\mathrm e } ) = - \frac{\pi}{3h_{\mathrm e}^4}(b_{\mathrm e}\phi + \kappa_+ + \alpha_{\mathrm e}^4\kappa_-)\\ & = - s^*(\phi ; \alpha_{\mathrm e } , h_{\mathrm e } ) \equiv - s[\phi , \kappa_+ , \kappa_- ] , \end{aligned}\ ] ] where @xmath311 . moreover , the differential relation is given by @xmath312 or by using `` intensive '' variables @xmath94 , @xmath313 and @xmath95 we obtain the alternative expression @xmath314 from the conditions ( [ eq : equilibrium_condition ] ) we obtain the stationary points of @xmath301 , in other words , a series of classical equilibrium solutions . to see the ( thermodynamic ) stability of the solutions , we have to know which of the equilibrium solutions has the minimum of @xmath315 ( or the maximum of @xmath83 ) for given @xmath77 and @xmath88 . for example , requiring one of the conditions , @xmath316 , we regard @xmath301 as an one - parameter function of @xmath37 . ( note that @xmath77 and @xmath88 are fixed . ) then , we have @xmath317 at a stationary point they become @xmath318 = -\frac{4}{h}\left(\frac{\partial s}{\partial h}\right)_{\phi,\kappa_-}. \end{aligned}\ ] ] hence , the condition for taking a minimum value of @xmath315 ( a maximum value of @xmath83 ) for fixed @xmath77 and @xmath88 is given by @xmath319 we give definitions of scalar , vector , and tensor harmonics on the 4d de sitter space . @xmath36 is the metric of the 4d de sitter space with the hubble parameter @xmath37 , namely , @xmath321 $ ] , and @xmath149 is the covariant derivative associated with @xmath36 . the tensor harmonics @xmath326 satisfies @xmath327t_{(\mathrm t)\mu\nu } = m^2 t_{(\mathrm t)\mu\nu } , \quad \nabla^\mu t_{(\mathrm t)\mu\nu } = t_{(\mathrm t)}{}^\mu{}_\mu = 0,\ ] ] and @xmath328 is defined by @xmath329 we expand the perturbed metric in harmonics of the 4d de sitter space with the hubble parameter @xmath37 as @xmath330 where @xmath147 , @xmath331 , and @xmath332 are scalar , vector , and tensor harmonics , respectively . greek indices @xmath333 run over the @xmath3d de sitter space and latin indices @xmath334 run over the @xmath335d extra dimensions @xmath336 . the coefficients @xmath337 , @xmath338 , @xmath339 , @xmath340 , @xmath341 , @xmath342 , and @xmath343 depend only on @xmath66 assuming that the perturbations are axisymmetric . the perturbations of the @xmath2 gauge field can be expanded as @xmath344 the vector - type components transform as @xmath348 where the prime denotes the derivative with respect to @xmath66 . choosing @xmath349 , we have @xmath350 . with this gauge choice we use new variables . using a part of perturbed einstein equation we obtain algebraically @xmath351 where @xmath352 is an arbitrary constant for @xmath353 , or @xmath354 for @xmath355 . when @xmath353 , this constant corresponds to the pure gauge mode for tensor perturbations because @xmath356v_{(\mathrm t)\nu } = 0 $ ] and @xmath357 . hence we can choose @xmath354 . the scalar - type components transform as @xmath358 choosing @xmath359 where @xmath360 is an arbitrary constant , we have @xmath361 . with this gauge choice we redefine each component as follows : @xmath362 moreover , using parts of the perturbed einstein and maxwell equations we algebraically obtain the following relations for four variables : @xmath363,\ ] ] where @xmath352 is an arbitrary constant for @xmath364 , or @xmath365 for @xmath366 . this constant corresponds to the residual ( homogeneous ) gauge freedom , @xmath367 and @xmath368 . we can set @xmath354 , i.e. , @xmath369 . s. mukohyama , y. sendouda , h. yoshiguchi and s. kinoshita , jcap * 0507 * , 013 ( 2005 ) [ arxiv : hep - th/0506050 ] . h. yoshiguchi , s. mukohyama , y. sendouda and s. kinoshita , jcap * 0603 * , 018 ( 2006 ) [ arxiv : hep - th/0512212 ] . y. sendouda , s. kinoshita and s. mukohyama , class . * 23 * , 7199 ( 2006 ) [ arxiv : hep - th/0607189 ] . a. v. frolov and l. kofman , phys . d * 69 * , 044021 ( 2004 ) [ arxiv : hep - th/0309002 ] . c. r. contaldi , l. kofman and m. peloso , jcap * 0408 * , 007 ( 2004 ) [ arxiv : hep - th/0403270 ] . r. bousso , o. dewolfe and r. c. myers , found . phys . * 33 * , 297 ( 2003 ) [ arxiv : hep - th/0205080 ] . j. u. martin , jcap * 0504 * , 010 ( 2005 ) [ arxiv : hep - th/0412111 ] . c. krishnan , s. paban and m. zanic , jhep * 0505 * , 045 ( 2005 ) [ arxiv : hep - th/0503025 ] . h. w. braden , j. d. brown , b. f. whiting and j. w. york , phys . d * 42 * , 3376 ( 1990 ) . k. i. maeda , t. tachizawa , t. torii and t. maki , phys . * 72 * , 450 ( 1994 ) [ arxiv : gr - qc/9310015 ] . t. prestidge , phys . d * 61 * , 084002 ( 2000 ) [ arxiv : hep - th/9907163 ] . s. s. gubser and i. mitra , jhep * 0108 * , 018 ( 2001 ) [ arxiv : hep - th/0011127 ] . s. s. gubser and i. mitra , arxiv : hep - th/0009126 . h. s. reall , phys . d * 64 * , 044005 ( 2001 ) [ arxiv : hep - th/0104071 ] . g. arcioni and e. lozano - tellechea , phys . d * 72 * , 104021 ( 2005 ) [ arxiv : hep - th/0412118 ] . t. harmark , v. niarchos and n. a. obers , arxiv : hep - th/0701022 . a. higuchi , nucl . b * 282 * , 397 ( 1987 ) . s. deser and a. waldron , phys . b * 508 * , 347 ( 2001 ) [ arxiv : hep - th/0103255 ] . t. garidi , j. p. gazeau and m. v. takook , j. math . phys . * 44 * , 3838 ( 2003 ) [ arxiv : hep - th/0302022 ] . m. peloso , l. sorbo and g. tasinato , phys . d * 73 * , 104025 ( 2006 ) [ arxiv : hep - th/0603026 ] . b. himmetoglu and m. peloso , arxiv : hep - th/0612140 . e. papantonopoulos , a. papazoglou and v. zamarias , arxiv : hep - th/0611311 . t. kobayashi and m. minamitsuji , arxiv : hep - th/0703029 .

we develop a microscopic theory for the relaxation dynamics of an optically pumped two - level system ( tls ) coupled to a bath of weakly interacting bose gas . using keldysh formalism and diagrammatic perturbation theory , expressions for the relaxation times of the tls rabi oscillations are derived when the boson bath is in the normal state and the bose - einstein condensate ( bec ) state . we apply our general theory to consider an irradiated quantum dot coupled with a boson bath consisting of a two - dimensional dipolar exciton gas . when the bath is in the bec regime , relaxation of the rabi oscillations is due to both condensate and non - condensate fractions of the bath bosons for weak tls - light coupling and dominantly due to the non - condensate fraction for strong tls - light coupling . our theory also shows that a phase transition of the bath from the normal to the bec state strongly influences the relaxation rate of the tls rabi oscillations . the tls relaxation rate is approximately independent of the pump field frequency and monotonically dependent on the field strength when the bath is in the low - temperature regime of the normal phase . phase transition of the dipolar exciton gas leads to a non - monotonic dependence of the tls relaxation rate on both the pump field frequency and field strength , providing a characteristic signature for the detection of bec phase transition of the coupled dipolar exciton gas .

1701.01847

the dynamics of a quantum two - level system ( tls ) is a topic of fundamental importance . its sustained influence is evident in the continual interest in the dynamics of spin or pseudospin systems ranging from quantum optics @xcite to quantum information @xcite . the spin - boson model @xcite captures the interaction between the tls and its environment by a spin @xmath0 degree of freedom coupled linearly to an oscillator bath @xcite . despite the simplicity of such a model , it exhibits a rich variety of behavior and describes a diverse array of physical systems and phenomena @xcite . one of the quantum systems that is well described by a tls is the quantum dot ( qd ) . fueled by interests in quantum information processing , coherent optical control of quantum dots has seen substantial development in the past decade @xcite . new functionalities or tuning capabilities can be achieved with hybrid systems by further coupling qds to other materials , such as nano - sized cavity @xcite , graphene @xcite , and superconductor @xcite . hybrid quantum systems comprising a fermion gas coupled to a boson gas constitute the condensed matter analogue of @xmath1he-@xmath2he mixtures . in systems where an electron gas is coupled with excitons or exciton - polaritons , it was recently predicted that the transition of the excitonic subsystem to the bose einstein condensate ( bec ) phase strongly modifies the properties of the electronic subsystem , resulting in polariton - mediated superconductivity and supersolidity @xcite . the topic of phase transition of the exciton or exciton - polariton bose system into the bec state is itself an intriguing topic that has garnered much attention @xcite . for dipolar exciton systems realized in gaas double quantum well ( dqw ) structures , the critical temperature to reach the condensate phase is about @xmath3 . recent works have demonstrated that double - layer structures based on transition metal dichalcogenides ( tmd ) monolayers @xcite can further push the transition temperature to @xmath4 @xcite . motivated by recent interest in hybrid fermion - boson systems and exciton - polariton physics mentioned above , in this paper we consider a radiation - driven quantum dot coupled to a dipolar exciton gas and study the influence of the latter s bec phase transition on the dynamics and relaxation of the qd states . the problem of tls relaxation coupled to a fermionic bath transitioning to a superconducting state was studied in the context of metallic glasses @xcite ; however , the question of tls relaxation coupled to a _ bosonic _ bath transitioning to a bec state has not , up to the authors knowledge , been considered before . it is noteworthy to mention that our current work is closely connected to the problem of a mobile impurity moving in a bec @xcite , since the renormalization of physical properties of a moving electron that strongly interacts with the surrounding medium ( polaron problem ) can be described by a quantum particle coupled with a bath @xcite . our theory consists of a tls modeling the ground and lowest excited states of the qd , which is coupled @xcite to a bath of weakly - interacting bose gas modeling the dipolar exciton system . in contrast to the simple spin - boson model , the interaction between the qd and the 2d dipolar exciton gas in our system is described by a _ nonlinear _ coupling hamiltonian . we take the bose gas to be weakly interacting , exhibiting a normal phase as well as a bec phase described by the bogoliubov model @xcite . our results demonstrate that the damping of the rabi oscillations of the tls is highly sensitive to the phase transition of the bosonic bath . the rest of our paper is organized as follows . the second section is devoted to the development of general theory for the relaxation rate of an illuminated tls coupled to a bosonic bath . we then apply our general results to the situation of a qd coupled with a dipolar exciton gas in the third section . finally , in the fourth section we present numerical results of the relaxation rates and discuss their behavior as a function of the optical pump field s parameters . in the appendix we present details of our calculations . first , we consider dynamics of isolated tls system under strong external electromagnetic field and describe the system s response using the non - equilibrium keldysh green function technique . the tls hamiltonian is given by @xmath5 where @xmath6 are the energies of the upper and lower states of the tls , and quantities with an overhead caret ( @xmath7 ) symbol denotes a matrix quantity . the interaction hamiltonian with the electromagnetic field is written here in the rotating wave approximation ( rwa ) . @xmath8 is the interaction matrix element and @xmath9 the frequency of the electromagnetic field . it is also assumed in eq . ( [ eq1 ] ) that the wavelength of the electromagnetic field is much larger than the geometrical size of the tls so that the field is uniform on our scale of interest . in this work , we denote quantities in the laboratory frame and the rotating frame , respectively , with and without an overhead tilde . the dynamics of the tls is described by the time - ordered green s function @xmath10 satisfying the equation of motion @xmath11 to remove the explicit time dependence , it is convenient to transform this equation to the rotating frame using the unitary transformation @xmath12 as a result , the green s function in the rotating frame , @xmath13 , is described by the equation @xmath14 where @xmath15 . to find the self - energies and lifetimes , we need the retarded and the lesser components of the non - equilbrium green s function . the retarded green s function is derived as ( see appendix a ) , @xmath16 where @xmath17 is the rabi frequency , and @xmath18 and @xmath19 are matrices defined as @xmath20 with @xmath21 , @xmath22 and @xmath23 . ( [ eq5])-([eq5_1 ] ) imply that new quasiparticles emerge from the light - matter coupling that renormalizes the original tls states into dressed states with energies @xmath24 . @xmath18 and @xmath19 are the projection operators to these dressed states . it is instructive to recover the result for rabi oscillations using the above retarded green s function eq . ( [ eq5 ] ) . the tls wave function at a latter time @xmath25 is obtained by propagating the initial time ( @xmath26 ) wave function , @xmath27_{ij}\psi_j(0 ) \nonumber \\ & = & \left[\hat{s}^{-1}(t)\int\frac{d\varepsilon}{2\pi}\hat{g}^r(\varepsilon)e^{-i\varepsilon t}\right]_{ij}\psi_j(0 ) , \label{eq6}\end{aligned}\ ] ] where @xmath28 are the wave functions of the tls upper @xmath29 and lower @xmath30 levels . here @xmath31 is the retarded green s function in the laboratory frame . when only the lower level is initially occupied , @xmath32 and @xmath33 . using eqs.([eq3])-([eq5_1 ] ) , the transition probability to the upper level is then given by @xmath34 which is the rabi oscillations @xcite . the lesser green s function can be expressed in terms of the distribution functions @xmath35 of the upper and lower dressed states as @xmath36 \nonumber \\ & = & 2\pi i n_{\varepsilon}\left[\hat{a}\delta(\varepsilon-\omega)+\hat{b}\delta(\varepsilon+\omega)\right ] \nonumber \\ & = & 2\pi i \left[\hat{a}n_{\omega}\delta(\varepsilon-\omega)+\hat{b}n_{-\omega}\delta(\varepsilon+\omega)\right ] . \label{glesser1}\end{aligned}\ ] ] note that @xmath37 . assuming the radiation is turned on adiabatically , we can obtain @xmath35 in the following . the density matrix @xmath38 in the original basis of tls upper and lower levels satisfies the kinetic equation ( see appendix a ) : @xmath39 = 0 , \label{ke } \\ & & \hat{f } = \left(\begin{array}{cc } f_{11 } & f_{12 } \\ f_{21 } & f_{22 } \\ \end{array } \right ) , \label{distfcn } % \end{equation}\end{aligned}\ ] ] where the subscripts @xmath40 respectively denotes the original ( _ i.e. _ , unrenormalized by light ) upper and lower levels of the tls , and @xmath41 is the hamiltonian in the rotating frame . writing the hamiltonian as @xmath42 , we can define an effective magnetic field @xmath43 that drives the tls pseudospin degrees of freedom , where @xmath44 are the real and imaginary parts of @xmath8 and @xmath45 the unit vectors along the @xmath46 directions . then , decomposing the density matrix @xmath38 as @xmath47 , the kinetic equation can be written as a bloch equation : @xmath48 from the definition of @xmath38 in eq . ( [ distfcn ] ) , we can relate the distribution functions in the two representations as @xmath49 and @xmath50 . with the laser field switched on adiabatically , the optical response follows adiabatically the driving field and is therefore stationary in the rotating frame , _ i.e. _ , @xmath51 . before laser is turned on , the tls initial state is in the lower level so that @xmath52 . since @xmath53 is a constant of motion , this implies that @xmath54 for all times @xmath25 . here we focus on the regime without population inversion , so that @xmath49 is always less than zero . we obtain @xmath55 as @xmath56 using @xmath57 , we also find the density matrix @xmath38 in the original basis of the tls upper and lower levels : @xmath58 , \label{fpp } \\ f_{12 } & = & -\frac{\mathrm{sgn}(2\delta-\omega)2\lambda}{\sqrt{(2\delta-\omega)^2 + 4\lambda^2}}. \label{fpm}\end{aligned}\ ] ] on the other hand , eq . ( [ glesser1 ] ) gives the density matrix in the basis of the dressed quasiparticles as follows @xmath59 we can determine @xmath60 by comparing the expressions of the density matrix in eq . ( [ glesser2 ] ) and eq . ( [ fpp ] ) . the @xmath61 element , for instance , gives @xmath62 or @xmath63 , from which we determine @xmath64 . \label{nomega } % \frac{1}{2}\left[1-\mathrm{sgn}\left(\delta-\omega/2\right)\right ] . \label{nomega}\ ] ] it follows that @xmath60 takes on the values @xmath65 or @xmath66 depending on whether the light frequency @xmath9 is smaller or larger , respectively , than the energy difference @xmath67 of the tls . in the absence of bath coupling , the tls is described by bare green s functions with a vanishing level broadening . to focus on the effects of bosonic bath coupling , we ignore the effects of spontaneous and stimulated emission due to electrons coupling to light . the only damping effects on the tls dynamics , once coupled to the bosonic bath , will be due to interlevel transitions caused by absorption or emission of the bosons . to analyze the tls - bath coupling , we add to the bare tls hamiltonian eq . ( [ eq1 ] ) the tls interaction term with the bath and the bath hamiltonian @xmath68 & 0 \\ 0 & w_{22}[\varphi ] \\ \end{array } \right)+\hat{h}_{bath}[\varphi].\end{gathered}\ ] ] here the first term is tls - bath coupling hamiltonian where the matrix elements describe the interaction of the upper and lower levels with the bath bosons . both terms in eq . ( [ eq8 ] ) are functionals of the quantum field @xmath69 , which describes the dynamics of bath degrees of freedom . the structure of the bath hamiltonian depends on whether it is in the normal or bose - condensed phase and will be specified later on . we assume short - range interaction between the tls and the bosonic bath @xmath70=g_i\int d\textbf{r}|\psi_{i}(\textbf{r})|^2|\varphi(\textbf{r},t)|^2,\end{gathered}\ ] ] with the coupling constant @xmath71 . the form of this interaction contains @xmath72 and is markedly different from the conventional coupling to a phonon - like bath , which is linear in @xmath69 @xcite . anticipating further application of our general theory to consider a qd interacting with a 2d exciton gas , we note that @xmath73 in eq . ( [ eq9 ] ) takes into account the most important direct contribution to the coulomb interaction energy between electrons in the qd and the 2d exciton gas . to elucidate the influence of the bosonic bath on the tls dynamics , we use the diagrammatic perturbation theory . within this approach the retarded green function can be found from the dyson equation , @xmath74 where the bare green s function @xmath75 is given by eq . ( [ eq5 ] ) . the solution of eq . ( [ eq10 ] ) gives the interacting green s function as follows @xmath76 , \label{eq11}\ ] ] where @xmath77 , @xmath78 is the matrix transpose of @xmath79 , and @xmath80 is the pauli matrix . we arrive at the following form of the green s function ( see appendix b ) @xmath81 } % \nonumber \\ & = & \frac{\hat{a}}{\epsilon-\omega+i/2\tau_u } + \frac{\hat{b}}{\epsilon+\omega+i/2\tau_l } , \label{eq12 } % \bar{g}^r_{ij}(\epsilon)=\frac{\bar{g}^{0r}_{ij}(\epsilon)}{1-tr\left[\bar{g}^{0r}(\epsilon)\bar{\sigma}^r(\epsilon)\right]}\approx\\ % \approx\frac{a_{ij}}{\epsilon-\omega+i/2\tau_u } % + \frac{b_{ij}}{\epsilon+\omega+i/2\tau_l},\end{aligned}\ ] ] where we have introduced the relaxation times @xmath82 of the upper ( subscript @xmath83 ) and lower ( subscript @xmath84 ) dressed quasiparticles @xmath85,\label{eq13_1 } \\ \frac{1}{2\tau_l } & = & -\textmd{im } \ , \mathrm{tr}\left[\hat{b}\hat{\sigma}^r(-\omega)\right ] , \label{eq13_2}\end{aligned}\ ] ] and neglected the shift of the levels due to the real part of self - energy . to obtain expressions of the relaxation times , we apply the diagrammatic perturbation theory in the leading order of the tls - bath interaction potential and account for the lowest - order non - vanishing diagrams for the self - energy . the explicit form of such diagrams depends on whether the bath is in the normal or the condensate state . first we consider the normal state . in the normal state of the bath , we assume that the bosons are non - interacting with a kinetic energy @xmath86 and chemical potential @xmath87 . to lowest order in the tls - bath interaction , the self - energy diagram is shown in fig . [ figure1 ] . detailed calculation of the self - energy is presented in the appendix c. the result reads @xmath88+n_b(\xi_{\textbf{p}+\textbf{k } } ) [ 1+n_b(\xi_{\textbf{p}})]}{\omega-\omega+e_{\textbf{p}}-e_{\textbf{p}+\textbf{k}}+i\delta}\right . \nonumber \\ & & \left.+\hat{m}_{\textbf{k}}\hat{b}\hat{m}_{-\textbf{k}}\frac{(1-n_{-\omega } ) [ n_b(\xi_{\textbf{p}})-n_b(\xi_{\textbf{p}+\textbf{k}})]+n_b(\xi_{\textbf{p}+\textbf{k } } ) [ 1+n_b(\xi_{\textbf{p}})]}{\omega+\omega+e_{\textbf{p}}-e_{\textbf{p}+\textbf{k}}+i\delta } \right],\end{aligned}\ ] ] where @xmath89 is the energy of the bath bosons rendered from the chemical potential , @xmath90^{-1}$ ] is the bose distribution function , and @xmath91 taking the imaginary part of eq . ( [ eq14 ] ) , integrating over the angle between @xmath92 and @xmath93 , and substituting the resulting expression into eqs . ( [ eq13_1])-([eq13_2 ] ) we find the relaxation times @xmath94,\nonumber\\ \frac{1}{2\tau_l } & = & \frac{m}{(2\pi)^2}\int_0^\infty dk\left[\int_{k/2}^\infty \frac{\gamma_kf_{-\omega}^l(p)pdp}{\sqrt{p^2-(k/2)^2}}\right . \label{eq15_2 } \\ & & \left.+\int_{|{k}/{2}+{2m\omega}/{k}|}^\infty \frac{\beta_kf_{\omega}^l(p)pdp}{\sqrt{p^2-\left({k}/{2}+{2m\omega}/{k}\right)^2}}\right ] . \nonumber\end{aligned}\ ] ] here we have introduced the coefficients @xmath95 and @xmath96,\nonumber\\ f^u_{-\omega}(p ) & = & n_{\omega}[n_b(\xi_{\textbf{p}})-n_b(\xi_{\textbf{p}}+2\omega)]\nonumber \\ & & + n_b(\xi_{\textbf{p}}+2\omega ) [ 1+n_b(\xi_{\textbf{p}})],\nonumber\\ f^l_{\omega}(p ) & = & \theta(e_{\textbf{p}}-2\omega)\left\{n_{-\omega}[n_b(\xi_{\textbf{p}})-n_b(\xi_{\textbf{p}}-2\omega)]\right . \nonumber \\ & & \left.+n_b(\xi_{\textbf{p}}-2\omega ) [ 1+n_b(\xi_{\textbf{p}})]\right\ } , \label{eq15_4 } % f^u_{\omega}(p)=f^l_{-\omega}(p)=n_{\textbf{p}}(1+n_{\textbf{p}}),\\\nonumber % f^u_{-\omega}(p)=n_{\omega}(n_{\textbf{p}}-n^+_{\textbf{p}})+n^+_{\textbf{p}}(1+n_{\textbf{p}}),\\\nonumber % f^l_{\omega}(p)=\theta(e_{\textbf{p}}-2\omega)[n_{-\omega}(n_{\textbf{p}}-n^-_{\textbf{p } } ) % + n^-_{\textbf{p}}(1+n_{\textbf{p}})],\end{aligned}\ ] ] we now consider the bath to be in the bose condensed phase and obtain the tls self - energy . the elementary excitations of the bose - condensed system are bogoliubov quasi - particles . an explicit form of the dispersion law of bogoliubov excitations depends on the model used to describe the system of interacting bosons . in the case of small particle density an appropriate theoretical model is the bogoluibov model of weakly - interacting bose gas . in the framework of this model , the dispersion law of elementary excitations has the form of @xmath97 , where @xmath98 is particle density in the condensate and @xmath99 the strength of inter - particle interaction . in the low - energy , long - wavelength limit @xmath100 the elementary excitations comprise sound quanta , with a dispersion @xmath101 where @xmath102 is the sound velocity . in the bose - condensed state , most of the particles are in the condensate but there are also noncondensate particles , due to both interaction and finite temperature effects ( thermal - excited particles ) . all three fractions of particles contribute to relaxation times of tls . we consider the quantum limit @xmath103 when thermal excitations are not important and the theory can be developed for @xmath104 . in the present dilute boson gas limit , the density of the noncondensate particles is small , and one can neglect the interaction term due to fluctuations of the condensate density and the noncondensate density . thus , the contribution to relaxation times due to the condensate and noncondensate particles can be calculated independently . [ h ] , and ( b ) is due to the non - condensate particles , @xmath105 . double blue line represents the tls green function and dashed lines the tls electron - bath interaction . red lines denote the normal or anomalous green functions of the non - condensate particles , and the zigzag lines stand for @xmath106 corresponding to the condensate particles.,title="fig:",width=321 ] the self - energy diagrams in the lowest order with respect to the tls - bath coupling are depicted in fig . [ figure2 ] . [ figure2](a ) corresponds to the contribution to the self - energy from the condensate particles and describes virtual processes in which a condensate particle is scattered by a tls electron through an intermediate non - condensate state . [ figure2](b ) corresponds to the contribution from the non - condensate particles and describes polarization of the non - condensate particles induced by the tls electrons . these self - energy contributions due to condensate particles @xmath107 and non - condensate @xmath108 particles have the following form ( see appendix c ) @xmath109,\label{eq16_1}\end{aligned}\ ] ] @xmath110 . \label{eq16_2}\end{aligned}\ ] ] similar steps leading to eqs . ( [ eq15_1])-([eq15_2 ] ) yields @xmath111 explicit expressions for relaxation times depend on the shape of the wave functions of the tls upper and lower states through the matrix elements eq . ( [ eq15 ] ) . in the next section we propose and analyze an experimental setup in which explicit expressions of the relaxation time can be obtained . applying our theory developed in the previous sections , we consider the nanostructure depicted in fig . 3 . a double quantum well ( dqw ) with closely separated electron - doped and hole doped wells realizes a 2d gas of indirect excitons ( also called dipolar excitons ) . a self - assembled qd is positioned above the dqw and is irradiated with a frequency close to the lowest exciton energy of the qd . in the electron - hole representation of the qd , the lower quantum state describes the unexcited qd state ( _ i.e. _ vaccum ) @xmath112 , while the lowest excited qd state describes the state @xmath113 of an excited electron - hole pair upon irradiation @xcite . with a shift in energy that does not affect the system s dynamics , we can assume that these two qd states correspond exactly to the lower and upper tls levels in eq . ( [ eq1 ] ) with @xmath114 , where @xmath115 are the lowest energies of electrons and holes in the qd ( corresponding to the energies at the conduction and valence band edges ) , and @xmath8 represents the dipole matrix element weighted by the electron - hole pair envelope wave function integrated over all space . [ h ] let us discuss the exciton gas model we use for our calculations . the indirect excitons have a dipole moment @xmath93 in the out - of - plane direction of the dqw . the excitons are modeled as rigid dipole molecules that are free to move on the dqw plane as described by the center - of - mass motion of the dipoles , a valid assumption as long as the dipole s internal degrees of freedom are not excited . the exciton density @xmath116 is assumed to be small so that @xmath117 , where @xmath118 is an exciton bohr radius . for simplicity , we assume that there is no particle tunneling between the qd and the dqw so that the exchange contribution to the interaction potential is negligible . this can be guaranteed using a large - band gap dielectric spacer as a substrate between the qd and the dqw . the direct contribution to the interaction potential between the electron - hole pair in the qd and the excitons in the dqw is given by @xmath119 where @xmath120 are the electron ( e ) and hole ( h ) wave functions in the qd , @xmath121 is the qw exciton center - of - mass wave function , @xmath122 is an effective dielectric constant taking account of the dielectric environment of the spacer and the dqw , @xmath123 and @xmath124 are the distance of the qd to the dqw surface and the separation between the positive charges and negative charges of the dipole layer ( see fig . [ figure3 ] ) . the fourier transform of @xmath125 yields @xmath126 typical values of the wave vector @xmath127 here are determined by the bath excitons of dqw . at temperatures above the condensation temperature @xmath128 , the bath in the normal state , the dqw exciton energy is of the order of @xmath129 and thus the wave vector @xmath130 , where @xmath131 is the exciton mass . at @xmath104 , the bath is in the condensed phase , the typical value of the bath excitation momentum is of the order of @xmath132 . hereafter we assume the distance of the qd from the dqw as well as the inter - well distance in the dqw to be sufficiently small so that @xmath133 and @xmath134 . this allows us to simplify the expression eq . ( [ eq18 ] ) assuming that @xmath135 . thus we have @xmath136 under this approximation the interaction eq . ( [ eq17 ] ) becomes contact - like , @xmath137 in our coupled qd - exciton gas system , since the lower level is the vacuum , we have in eqs.([eq8 ] ) and ( [ eq9 ] ) that the coupling constants @xmath138 and their respective matrix elements @xmath139 . the qd is described by a two - dimensional system of electrons and holes confined by a parabolic potential . for strongly confined qds where the qd size @xmath140 is small compared to the bohr radius of the electron - hole pair , the lowest - energy electrons and holes are characterized by the wavefunctions @xcite @xmath141 , where @xmath142 is an electron @xmath143 or a hole @xmath144 characteristic length determined by the qd confinement potential . using these wavefunctions , we find the matrix element @xmath145-\mathrm{exp}[{-(ka_e/2)^2}]$ ] . with @xmath146 , we perform the integration over @xmath127 in eqs . ( [ eq15_1])-([eq15_2 ] ) and obtain the following expressions for the relaxation times @xmath147pdp\nonumber \\ & & + \frac{m\beta_0}{(2\pi)^2}\int_0^\infty f_+(p , p_0)f^u_{-\omega}(p)pdp , \label{eq21_1 } \\ \frac{1}{2\tau_l}&=&\frac{m\gamma_0}{(2\pi)^2}\int_0^\infty f_-(p,0)n_b(\xi_{\textbf{p}})[1+n_b(\xi_{\textbf{p}})]pdp\nonumber\\ & & + \frac{m\beta_0}{(2\pi)^2}\int_0^\infty f_-(p , p_0)f^l_{\omega}(p)pdp , \label{eq21_2 } % \frac{1}{2\tau_u}&=&\frac{m\alpha_0}{(2\pi)^2}\int_0^\infty f_+(p,0)n_{\textbf{p}}(1+n_{\textbf{p}})pdp+\\\nonumber % & & + \frac{m\beta_0}{(2\pi)^2}\int_0^\infty % f_+(p , p_0)f^u_{-\omega}(p)pdp , \label{eq21_1 } \\ % \frac{1}{2\tau_l}&=&\frac{m\gamma_0}{(2\pi)^2}\int_0^\infty f_-(p,0)n_{\textbf{p}}(1+n_{\textbf{p}})pdp+\\\nonumber % & & + \frac{m\beta_0}{(2\pi)^2}\int_0^\infty % f_-(p , p_0)f^l_{\omega}(p)pdp , \label{eq21_2}\end{aligned}\ ] ] where @xmath148 with @xmath149 , and @xmath150 to simplify eqs . ( [ eq21_1])-([eq21_2 ] ) we consider two limiting cases . at large temperatures @xmath151 [ while still small compared to @xmath152 , we can keep up to the zeroth order in @xmath153 with @xmath154 and @xmath155 $ ] in eqs . ( [ eq21_1])-([eq21_2 ] ) . the relaxation rates for @xmath156 then read @xmath157 where the @xmath158 signs apply for @xmath82 respectively , and @xmath159 and @xmath160 is the polylogarithm function of order @xmath161 . in the opposite limit of low temperatures @xmath162 , we keep up to the zeroth order in @xmath163 . then the respective first terms in eqs . ( [ eq21_1])-([eq21_2 ] ) drop out , and we have @xmath164 and @xmath165 in the second terms . the result is @xmath166 where @xmath116 is the exciton density in the bath . we have restored @xmath167 in the expressions for the relaxation rates here and in the following section . straightforward but cumbersome integration in eqs . ( [ b10.1])-([b10.2 ] ) results in the following expressions for relaxation times due to the condensate and non - condensate particles in the bath @xmath168 ^ 2 , \label{eq24}\ ] ] and @xmath169 , \nonumber\end{aligned}\ ] ] where @xmath170 is the dawson integral @xmath171 simple analysis of rabi oscillations given above accounting for the finite values of relaxation times yields @xmath172 we focus our discussion on the low temperature regime @xmath162 . first we note that the relaxation rates for the upper and lower levels coincide both in the normal phase [ eq . ( [ eq23 ] ) ] and in the bose - condensed phase [ eqs . ( [ eq24])-([eq25 ] ) ] . with relaxation times for the upper and lower levels being equal , eq . ( [ eq26 ] ) is simplified to @xmath173 , with @xmath174 given by eq . ( [ eq23 ] ) in the normal phase and by @xmath175 from eqs . ( [ eq24])-([eq25 ] ) in the bec phase . secondly , the relaxation rates are all proportional to the distribution function @xmath60 of the dressed quasiparticle states given in eq . ( [ nomega ] ) . since @xmath176 only if the upper dressed quasiparticle state @xmath177 is occupied and vanishes otherwise , finite relaxation of the tls rabi oscillations occurs only when the pump field frequency exceeds the tls energy level difference @xmath178 @xcite . in our following discussion , therefore , we focus on the regime @xmath179 . in the normal phase , we note that the low - temperature relaxation rates eq . ( [ eq23 ] ) are independent of the driving frequency and increases monotonically with the tls - light coupling as @xmath180 . in the bec phase , we find that the relaxation rates eqs . ( [ eq24])-([eq25 ] ) exhibit strong non - monotonic dependence on the driving frequency through the rabi frequency and the tls - light coupling . [ h ] due to condensate particles ( normalized by @xmath181 ) versus frequency detuning @xmath182 . red ( dot - dashed ) , blue ( dashed ) and black ( solid ) lines correspond to @xmath183 , @xmath184 and @xmath185 respectively.,title="fig:",width=321 ] [ h ] due to non - condensate particles ( normalized by @xmath181 ) versus frequency detuning @xmath182 . red ( dot - dashed ) , blue ( dashed ) and black ( solid ) lines correspond to @xmath183 , @xmath184 and @xmath185 respectively.,title="fig:",width=321 ] [ h ] at frequency detuning @xmath186 . solid ( black ) line corresponds to @xmath187 and dashed ( red ) line to @xmath188 , respectively.,title="fig:",width=321 ] [ h ] and @xmath188 as a function of tls - light coupling @xmath189 and frequency detuning @xmath182 . grey ( red ) surface corresponds to @xmath187 and black ( blue ) surface to @xmath188 , respectively.,title="fig:",width=321 ] below , we proceed to analyze the numerical dependence of the relaxation rates on the driving frequency and tls - light coupling in the bec phase . for the tls , we take the following parameters @xmath190 , @xmath191 and @xmath192 ( @xmath193 is the electron mass ) typical for gaas - based qds @xcite . the characteristic lengths of the hole and the electron wavefunctions are taken as @xmath194 and @xmath195 . with the dipole matrix element of the qd @xmath196 ( @xmath197 ) and optical field strength @xmath198 , the tls - light coupling constant takes the range of values @xmath199 . for the dipolar exciton gas , we take @xmath200 , @xmath201 , @xmath202 and @xmath203 typical for gaas dqw structures . to provide an estimate for the inter - particle interaction @xmath99 , we assume a simple point - charge treatment of the dipolar excitons , and the exciton - exciton interaction potential takes the form @xcite @xmath204 . fourier transform of @xmath205 then gives @xmath206 $ ] and the coupling constant @xmath207 . the bogoliubov speed of sound @xmath208 is thus also fixed from @xmath209 . for convenience , we display the frequency @xmath9 in terms of the dimensionless frequency detuning @xmath210 . figs . [ figure4]-[figure5 ] show the relaxation rates @xmath187 and @xmath188 as a function of @xmath182 for relatively small values of @xmath211 . we find that both relaxation rates behave non - monotonically as a function of detuning , reaching maximum values at @xmath212 and then becoming exponentially suppressed at larger values of @xmath182 . @xmath187 is more strongly suppressed than @xmath188 . secondly , we observe that , for the present values of @xmath211 , the condensate and non - condensate fractions contribute to the relaxation rate by the same order of magnitude , with @xmath187 exceeding @xmath188 . this trend is maintained until @xmath8 reaches @xmath213 of @xmath181 ( corresponding to @xmath214 ) , when @xmath187 starts to drop signifcantly faster than @xmath188 . [ figure6 ] shows both quantities plotted versus @xmath189 at a frequency detuning @xmath186 , from which we observe that @xmath187 decreases much more abruptly than @xmath188 . when @xmath8 is increased beyond @xmath215 , it is seen that @xmath188 now overtakes @xmath187 . for @xmath8 values beyond @xmath216 , @xmath187 has dropped essentially to zero and the non - condensate fraction constitutes the dominant contribution to relaxation . although the relaxation rates vanish expectedly when @xmath217 , they do not vanish at zero frequency detuning @xmath218 , as one might conclude by inspecting figs . [ figure4]-[figure5 ] . to examine more fully the behavior of @xmath187 and @xmath188 , we plot them in fig . [ figure7 ] in the full range of @xmath8 and @xmath182 . at @xmath218 , both relaxation rates become small only when @xmath219 ; in addition @xmath187 also become small when @xmath220 . around @xmath221 , both relaxation rates as a function of @xmath182 reach maximum at @xmath218 . conventionally , the bec phase transition of a dipolar exciton gas is detected using optical spectroscopy . in the bec phase , the excitons or exciton - polaritons are described by a single coherent wave function and emit light coherently . the resulting luminescence peak becomes much narrower in comparison with that in the normal phase , signaling formation of the condensate state . another way to detect the bec phase transition has been theoretically suggested recently @xcite . bec phase transition strongly influences the non - equilibrium properties of a dipolar exciton gas driven by an external surface acoustic waves ( saw ) . under phase transition , the saw attenuation effect and the saw - exciton drag current become strongly modified , allowing one to detect the bec phase transition using acoustic spectroscopy . on top of the foregoing , our findings in principle provide a new strategy to detect the bec phase transition of the dipolar exciton gas . while the qd s relaxation rate displays only a monotonic linear dependence on the light intensity when the exciton gas is in the normal phase , it becomes strongly non - monotonic as a function of both the pump field s frequency and intensity once the exciton gas is in the bec phase . thus , by monitoring the rabi oscillation dynamics of the qd , the normal and condensed phases of the exciton gas can be distinguished by the dependence of the relaxation rate on the frequency and intensity of the driving field . to conclude , we have developed a theory for the relaxation of optically pumped two - level systems coupled to a bosonic bath using the nonequilibrium keldysh technique and the diagrammatic perturbation theory . to elucidate the effects of bath phase transition , we have considered the cases when the bosonic bath is in the normal state and in the bose - condensed state . we then apply our theory to study the scenario of an illuminated quantum dot coupled to a dipolar exciton gas . the condensate and non - condensate fractions of the bath particles contribute to the relaxation rate by variable proportions depending on the value of pump field amplitude . when the pump field is weak , both fractions contribute by about the same order of magnitude ; while for strong pump field , the non - condensate fraction becomes the dominant contribution . our findings also show that the phase transition of the dipolar exciton gas to the bec regime results in a strong dependence of the relaxation rate on the optical pump field . the relaxation rate then exhibits a strong non - monotonic behavior , reaching a maximum and then becoming exponentially suppressed as a function of both the pump field s frequency and amplitude . such a non - monotonic dependence could in principle serve as a smoking gun for detecting bec phase transition of the coupled dipolar exciton gas . finally , we point out that despite our focus on dipolar exciton gas in this work , the theory we have developed is also applicable to other types of bose gas , such as 2d exciton - polaritons @xcite , magnons @xcite and cold atoms @xcite . v.m.k . acknowledges the support from rfbr grant @xmath222 . w.k . acknowledges the support by a startup fund from the university of alabama . because of the time - dependent perturbation from light , we employ the keldysh formalism to calculate the green s function and distribution function of the system . following established routes in non - equilibrium green s function formalism , the left - multiplied and right - mulitplied dyson equations for the contour - ordered green s function @xmath223 are @xmath224 we are interested in the green s function of the tls under irradiation , therefore the self - energy @xmath225 due to interaction with the bath is set to zero . in the rotating frame , we already find @xmath226 first we derive the retarded green s function . applying langreth s rules @xcite to the two equations in eq . ( [ dyson ] ) and summing them together , we have @xmath227 , @xmath228 we transform the time variables @xmath229 , into the wigner coordinates with the average time @xmath230 and relative time @xmath231 . ( [ dyson2 ] ) becomes @xmath232 performing fourier transformation with respect to @xmath233 gives @xmath234 solving this matrix equation yields the retarded green s function in eq . ( [ eq5 ] ) : @xmath235 \nonumber \\ & = & \hat{a}\frac{1}{\varepsilon-\omega+i\delta}+\hat{b}\frac{1}{\varepsilon+\omega+i\delta } , \label{7}\end{aligned}\ ] ] with @xmath236 defined in eq . ( [ eq5_1 ] ) . now from the contour - ordered dyson s equations eq . ( [ dyson ] ) and applying langreth s rule , then subtracting the two equations , we get @xmath237 , @xmath238 transforming into the wigner coordinates , we obtain the kinetic equation @xmath239 = 0\ ] ] the density matrix @xmath240 is given by the equal - time keldysh green s function @xmath241 , which satisfies @xmath242 = 0.\ ] ] we start from eq . ( [ eq11 ] ) in the main text @xmath76 . \nonumber % \label{eq11}\ ] ] from eqs . ( [ eq5])-([eq5_1 ] ) it can be easily evaluated that @xmath243 . upon substitution of eq . ( [ eq5 ] ) , the first term of eq . ( [ eq11 ] ) can be written as follows in the vicinity of the poles we have @xmath245 since @xmath246 where @xmath233 is the quasiparticle lifetime , @xmath247 is satisfied for our perturbative calculations . we see that the last terms in the denominators above @xmath248 are a factor of @xmath249 smaller than @xmath174 and thus can be neglected . next we consider the second term of eq . ( [ eq11 ] ) , whereupon substituting eq . ( [ eq5 ] ) becomes in the vicinity of the upper level where @xmath251 , @xmath252 we see that the expression in the last line above is a factor of @xmath253 smaller than the corresponding @xmath251 contribution ( _ i.e. _ , the term @xmath254 ) in eq . ( [ eqsb1 ] ) , and hence can be neglected . a similar analysis shows that the same is true for the @xmath255 contribution in eq . ( [ eqsb2 ] ) . therefore we find @xmath256 which gives eq . ( [ eq12 ] ) with eqs . ( [ eq13_1])-([eq13_2 ] ) in the main text . here we derive the expression for the tls self - energy . analytic expression of the diagram depicted in fig . [ figure1 ] reads @xmath257 where times @xmath229 are located on the keldysh contour , @xmath258 and @xmath259 are the polarization operator and green s function of the bath s particles . to proceed further , let us first perform an analytic continuation to the real time domain . using the langreth s rules @xcite we find @xmath260\hat{m}_{-\textbf{k } } , \label{app2_2}\end{aligned}\ ] ] @xmath261.\label{app2_4}\end{aligned}\ ] ] lets consider the bath in normal phase state , @xmath128 . in this case the bare green s functions of the bath s particles are @xmath262\delta(\varepsilon - e_{\textbf{k}}),\label{a3_2}\\ \mathcal{g}^r(\textbf{k},\varepsilon)&=&\frac{1}{\varepsilon - e_{\textbf{k}}+i\delta},\label{a3_3 } % \mathcal{g}^<(\textbf{k},\epsilon)=-2\pi in_{\textbf{k}}\delta(\epsilon - e_{\textbf{k}})\\\nonumber % \mathcal{g}^>(\textbf{k},\epsilon)=-2\pi i[n_{\textbf{k}}+1]\delta(\epsilon - e_{\textbf{k}})\\\nonumber % \mathcal{g}^r(\textbf{k},\epsilon)=\frac{1}{\epsilon - e_{\textbf{k}}+i\delta}\end{aligned}\ ] ] where @xmath263 is equilibrium bose distribution and @xmath264 is the energy of the bath s particles ( which corresponds to the kinetic energy of the exciton s center - of - mass motion for the excitonic bath we consider in section [ sec : qd ] . using these functions we find the following polarization operators @xmath265 \nonumber \\ & & \times\delta(\omega+e_{\textbf{p}}-e_{\textbf{p}+\textbf{k } } ) , \label{a4_1}\\ \pi^r(\textbf{k},\omega)&=&\sum_{\textbf{p}}\frac{n_b(\xi_{\textbf{p}})-n_b(\xi_{\textbf{k}+\textbf{p}})}{\omega+e_{\textbf{p}}-e_{\textbf{p}+\textbf{k}}+i\delta}. \label{a4_2 } % \pi^<(\textbf{k},\omega)=-2\pi i\sum_{\textbf{p}}n_{\textbf{p}+\textbf{k}}(n_{\textbf{p}}+1)\delta(\omega+e_{\textbf{p}}-e_{\textbf{p}+\textbf{k}})\\ % \pi^r(\textbf{k},\omega)=\sum_{\textbf{p}}\frac{n_{\textbf{p}}-n_{\textbf{p}+\textbf{k}}}{\omega+e_{\textbf{p}}-e_{\textbf{p}+\textbf{k}}+i\delta}\end{aligned}\ ] ] substituting these expressions into eq . ( [ app2_1 ] ) , we obtain the self - energy expression eq . ( [ eq14 ] ) . if the bath is in the bose - condensed state , the elementary excitations in the bogoliubov s theory of weakly - interacting bose gas have the energy dispersion @xmath266 where @xmath267 is the bogolubov quasiparticle s speed of sound , @xmath99 is the inter - particle interaction strength , and @xmath98 particles density in the condensate . the green functions are given by @xmath268 and @xmath269 \nonumber \\ & = & -\frac{2\pi i}{2\epsilon_{\textbf{k}}}\left ( \begin{array}{cc } \varepsilon+k^2/2m+ms^2 & -ms^2 \\ -ms^2 & -\varepsilon+k^2/2m+ms^2 \\ \end{array}\right)\nonumber \\ & & \times\left\{n_b(\epsilon_{\textbf{k}})\delta(\varepsilon-\epsilon_{\textbf{k}})+[1+n_b(\epsilon_{\textbf{k}})]\delta(\varepsilon+\epsilon_{\textbf{k}})\right\},\label{a8_1}\end{aligned}\ ] ] @xmath270[\hat{\mathfrak{g}}^r(\textbf{k},\varepsilon)-\hat{\mathfrak{g}}^a(\textbf{k},\varepsilon)]\nonumber \\ & = & -\frac{2\pi i}{2\epsilon_{\textbf{k}}}\left ( \begin{array}{cc } \varepsilon+k^2/2m+ms^2 & -ms^2 \\ -ms^2 & -\varepsilon+k^2/2m+ms^2 \\ \end{array}\right)\nonumber \\ & & \times\left\{[1+n_b(\epsilon_{\textbf{k}})]\delta(\varepsilon-\epsilon_{\textbf{k}})+n_b(\epsilon_{\textbf{k}})\delta(\varepsilon+\epsilon_{\textbf{k}})\right\}.\label{a8_2 } % \hat{\mathfrak{g}}^<(\textbf{k},\epsilon)=n_\epsilon[\hat{\mathfrak{g}}^r(\textbf{k},\epsilon)-\hat{\mathfrak{g}}^a(\textbf{k},\epsilon)]\\\nonumber=-\frac{2\pi i}{2\epsilon_k}\left ( % \begin{array}{cc } % \epsilon+k^2/2m+ms^2 & -ms^2 \\ % -ms^2 & -\epsilon+k^2/2m+ms^2 \\ % \end{array}\right)\times\\\nonumber % \times[n_k\delta(\epsilon-\epsilon_k)+(1+n_k)\delta(\epsilon+\epsilon_k)],\\\nonumber % \hat{\mathfrak{g}}^>(\textbf{k},\epsilon)=(1+n_\epsilon)[\hat{\mathfrak{g}}^r(\textbf{k},\epsilon)-\hat{\mathfrak{g}}^a(\textbf{k},\epsilon)]\\\nonumber=-\frac{2\pi i}{2\epsilon_k}\left ( % \begin{array}{cc } % \epsilon+k^2/2m+ms^2 & -ms^2 \\ % -ms^2 & -\epsilon+k^2/2m+ms^2 \\ % \end{array}\right)\times\\\nonumber % \times[(1+n_k)\delta(\epsilon-\epsilon_k)+n_k\delta(\epsilon+\epsilon_k)],\\\nonumber % n_k=\frac{1}{e^{\epsilon_k / t}-1},\end{aligned}\ ] ] the polarization operator has two contributions . the first one comes from the condensate particles and the other one from non - condensate particles , fig . [ figure1 ] . if the bose bath is a two - dimensional system , the condensate occurs at zero temperature only . in this case one assumes that @xmath271 . the contribution @xmath272 to the polarization operator from the condensate particles is @xmath273 \nonumber \\ & = & n_c\frac{k^2/m}{(\omega+i\delta)^2-\epsilon_{\textbf{k}}^2 } , \label{a9_1}\\ p^<_c(\textbf{k},\omega ) & = & n_c[\mathfrak{g}^<+\mathfrak{f}^{+<}+\mathfrak{f}^<+\tilde{\mathfrak{g}}^ < ] \nonumber \\ & = & -\pi in_c\frac{k^2/m}{\epsilon_{\textbf{k}}}\delta(\omega+\epsilon_{\textbf{k}}).\label{a9_2 } % p^r_c(\textbf{k},\omega ) & = & % n_c[\mathfrak{g}^r+\mathfrak{f}^{+r}+\mathfrak{f}^r+\tilde{\mathfrak{g}}^r ] \nonumber \\ % & = & n_c\frac{k^2/m}{(\varepsilon+i\delta)^2-\epsilon_{\textbf{k}}^2 } , \label{a9_1}\\ % p^<_c(\textbf{k},\omega ) & = & % n_c[\mathfrak{g}^<+\mathfrak{f}^{+<}+\mathfrak{f}^<+\tilde{\mathfrak{g}}^ < ] % \nonumber \\ % & = & -\pi in_c\frac{k^2/m}{\epsilon_k}\delta(\varepsilon+\epsilon_{\textbf{k}}).\label{a9_2 } % % p^r_c(\textbf{k},\omega)=n_c[\mathfrak{g}^r+\mathfrak{f}^{+r}+\mathfrak{f}^r+\tilde{\mathfrak{g}}^r]=n_c\frac{k^2/m}{(\epsilon+i\delta)^2-\epsilon_k^2},\\\nonumber % % p^<_c(\textbf{k},\omega)=n_c[\mathfrak{g}^<+\mathfrak{f}^{+<}+\mathfrak{f}^<+\tilde{\mathfrak{g}}^<]\\=-2\pi in_c\frac{k^2/m}{2\epsilon_k}\delta(\epsilon+\epsilon_k).\end{aligned}\ ] ] using this functions one finds the condensate contribution to the retarded self - energy , eq . ( [ eq16_1 ] ) . now let us find the contribution to the self - energy from the non - condensate particles . first , we need the retarded and lesser polarization operators . in the regime @xmath274 where linear dispersion @xmath275 of the bogoliubov quasiparticles holds , using eqs . ( [ a7])-([a8_2 ] ) we find for the retarded and lesser polarization operators of non - condensate particles @xmath276\nonumber \\ & = & \frac{(ms^2)^2}{2}\sum_{\textbf{p}}\frac{1}{\epsilon_{\textbf{p}}\epsilon_{\textbf{p}+\textbf{k}}}\left(\frac{1}{\omega-\epsilon_{\textbf{p}}-\epsilon_{\textbf{p}+\textbf{k}}+i\delta}\right . \nonumber\\ & & \left.-\frac{1}{\omega+\epsilon_{\textbf{p}}+\epsilon_{\textbf{p}+\textbf{k}}+i\delta}\right),\nonumber\end{aligned}\ ] ] @xmath277\\\nonumber % = \frac{(ms^2)^2}{2}\sum_{\textbf{p}}\frac{1}{\epsilon_{\textbf{p}}\epsilon_{\textbf{p}+\textbf{k}}}\left[\frac{1}{\omega-\epsilon_{\textbf{p}}-\epsilon_{\textbf{p}+\textbf{k}}+i\delta}-\frac{1}{\omega+\epsilon_{\textbf{p}}+\epsilon_{\textbf{p}+\textbf{k}}+i\delta}\right],\\\nonumber % p_n^<(\textbf{k},\omega)=2i\sum_{\textbf{p},\epsilon}\mathfrak{f}^<(\textbf{p}+\textbf{k},\epsilon+\omega)\mathfrak{f}^{+>}(\textbf{p},\epsilon)\\\nonumber % = -2\pi i\frac{(ms^2)^2}{2}\sum_{\textbf{p}}\frac{1}{\epsilon_{\textbf{p}}\epsilon_{\textbf{p}+\textbf{k}}}\delta(\omega+\epsilon_{\textbf{p}}+\epsilon_{\textbf{p}+\textbf{k}}).\end{aligned}\ ] ] now the calculation of the non - condensate particles contribution to the self - energy of tls is simple , and we arrive at the expression eq . ( [ eq16_2 ] ) of the main text . p. rivera , j. r. schaibley , a. m. jones , j. s. ross , s. wu , g. aivazian , p. klement , k. seyler , g. clark , n. j. ghimire , j. yan , d. g. mandrus , w. yao , and x. xu , _ et al . . commun . * 6 * , 6242 ( 2015 ) . this follows from our assumption of low - temperatures : the relaxation rate in the normal state obtained in the limit of low temperatures @xmath162 is a zero - temperature result , while the relaxation rate in the bose - condensed phase is obtained for temperatures below the bec critical temperature ( typically a few kelvins ) where the bogoliubov theory at @xmath279 applies . at higher temperatures , the relaxation rates become non - zero when @xmath280 , as indicated by our result eq . ( [ eq22 ] ) .

suppose is a real univariate polynomial of degree with exactly monomial terms . we present an algorithm , with complexity polynomial in on average ( relative to the stable log - uniform measure ) , for counting the number of real roots of . the best previous algorithms had complexity super - linear in . we also discuss connections to sums of squares and-discriminants , including explicit obstructions to expressing positive definite sparse polynomials as sums of squares of few sparse polynomials . our key tool is the introduction of efficiently computable _ chamber cones _ , bounding regions in coefficient space where the number of real roots of can be computed easily . much of our theory extends to-variate-nomials . @matrix[1][c]- ifnextchar@ifnextchar [ section ] [ thm]algorithm [ thm]corollary [ thm]lemma [ thm]proposition [ thm]question [ thm]conjecture [ thm]problem [ thm]definition [ thm]remark [ thm]example [ thm]remark

1101.2642

counting the real solutions of polynomial equations in one variable is a fundamental ingredient behind many deeper tasks and applications involving the topology of real algebraic sets . however , the intrinsic complexity of this basic enumerative problem becomes a mystery as soon as one considers the input representation in a refined way . such complexity questions have practical impact for , in many applications such as geometric modelling or the discretization of physically motivated partial differential equations , one encounters polynomials that have sparse expansions relative to some basis . so we focus on new , exponential speed - ups for counting the real roots of certain sparse univariate polynomials of high degree . sturm sequences @xcite , and their later refinements @xcite , have long been a centrally important technique for counting real roots of univariate polynomials . in combination with more advanced algebraic tools such as a grbner bases or resultants @xcite , sturm sequences have even been applied to algorithmically study the topology of real algebraic sets in arbitrary dimension ( see , e.g. , ( * ? ? ? * chapters 2 , 5 , 11 , and 16 ) ) . however , as we will see below ( cf . examples [ ex : tri ] and [ ex : tetra1 ] ) , there are obstructions to attaining polynomial intrinsic complexity , for _ sparse _ polynomials , via sturm sequences . so we must seek alternatives . more recently , relating multivariate positive polynomials to sums of squares has become an important algorithmic tool in optimizing real polynomials over semi - algebraic domains @xcite . however , there are also obstructions to the use of sums of squares toward speed - ups for sparse polynomials ( see theorem [ thm : neg ] below ) . discriminants have a history nearly as long as that of sturm sequences and sums of squares , but their algorithmic power has not yet been fully exploited . our main result is that _ @xmath4-discriminants _ @xcite yield an algorithm for counting real roots , with average - case complexity polynomial in the _ logarithm _ of the degree , for certain choices of probability distributions on the input ( see theorem [ thm : pos ] below ) . the use of randomization is potentially inevitable in light of the fact that even detecting real roots becomes @xmath7-hard already for moderately sparse multivariate polynomials @xcite . the classical technique of sturm sequences @xcite reduces counting the roots of a polynomial @xmath0 in a half - open interval @xmath8 to a gcd - like computation , followed by sign evaluations for a sequence of polynomials . a key difficulty in these methods , however , is their apparent super - linear dependence on the degree of the underlying polynomial . consider the following two examples ( see also ( * ? ? ? * example 1 ) ) . [ ex : tri ] _ setting @xmath9 , the realroot command in maple 14 ( which is an implementation of sturm sequences ) results in an out of memory error after about @xmath10 seconds . the polynomials in the underlying computation , while quite sparse , have coefficients with hundreds of thousands of digits , thus causing this failure . on the other hand , via more recent work @xcite , one can show that when @xmath11 and @xmath12 , @xmath13 has exactly @xmath14 , @xmath15 , or @xmath16 positive roots according as @xmath17 is less than , equal to , or greater than @xmath18 . in particular , our @xmath0 has exactly @xmath16 positive roots . ( we discuss how to efficiently decide the size of monomials in rational numbers with rational exponents in algorithm [ algor : sign ] of section [ sub : bit ] below . ) @xmath19 _ [ ex : tetra1 ] _ going to tetranomials , consider @xmath20 with @xmath21 . then ( via the classical descartes rule of signs ) such an @xmath0 has exactly @xmath15 or @xmath22 positive roots , but the inequalities characterizing which @xmath23 yield either possibility are much more unwieldy than in our last example : there are at least @xmath16 , involving polynomials in @xmath24 and @xmath25 having tens of thousands of terms . in particular , for @xmath26 , sturm sequences on maple 14 result in an out of memory error after about 122 seconds . @xmath19 _ we have discovered that @xmath4-discriminants , reviewed in section [ sec : back ] , enable algorithms with complexity polynomial in the _ logarithm _ of the degree . _ for any @xmath27 , we define @xmath28 and let the _ stable log - uniform measure _ on @xmath29 ( resp . @xmath30 ) be the probability measure @xmath31 ( resp . @xmath32 ) defined as follows : @xmath33^d)}{(2m)^d}$ ] ( resp . @xmath34 ) , where @xmath35 denotes the standard lebesque measure on @xmath36 and @xmath37 denotes set cardinality . @xmath19 _ note that the stable log - uniform measure is finitely additive ( but not countably additive ) , and is invariant under reflection across coordinate hyperplanes . [ thm : pos ] let @xmath38 be positive integers , + @xmath39 + with @xmath40 being independent stable log - uniform random variables chosen from @xmath41 ( resp . @xmath42 ) , and define @xmath43 . then there is a deterministic algorithm , with complexity polynomial in @xmath3 ( resp . @xmath44 and @xmath3 ) , that computes a number in @xmath45 that , with probability @xmath15 , is exactly the number of real roots of @xmath0 . the underlying computational model is the bss model over @xmath41 ( resp . the turing model ) . the key idea is that while the regions of coefficients determining polynomials with a constant number of real roots become more complicated as the number of monomial terms increases , one can nevertheless efficiently characterize large subregions _ chamber cones _ where the number of real roots is very easy to compute . this motivates the introduction of probability and average - case complexity . the @xmath4-discriminant allows one to make this approach completely precise and algorithmic . in fact , our framework enables us to transparently extend theorem [ thm : pos ] to @xmath5-variate @xmath6-nomials ( see theorem [ thm : nplus3 ] of section [ sub : can ] ) . our focus on the stable log - uniform measure simplifies our development and has some practical motivation : when one considers @xmath46-bit floating - point numbers with uniformly random exponent and mantissa , taking @xmath47 and suitably rescaling yields exactly the stable log - uniform measure on @xmath42 . the stable log - uniform measure has also been used in work of avendao and ibrahim to study the expected number of roots of sparse polynomial systems over local fields other than @xmath41 @xcite . it is of course quite natural to ask how the expected complexity in theorem [ thm : pos ] behaves under other well - known measures , e.g. , uniform or gaussian . unfortunately , the underlying calculations become much more complicated . on a deeper level , it is far from clear what a truly `` natural '' probability measure on the space of tetranomials is . for instance , for non - sparse polynomials , it is popular to use specially weighted independent gaussian coefficients since the resulting measure becomes invariant under a natural orthogonal group action ( see , e.g. , @xcite ) . however , we are unaware of any study of the types of distributions occuring for the coefficients of polynomials actually occuring in physical applications . the speed - ups we derive actually hold in far greater generality : see @xcite for the case of @xmath5-variate @xmath48-nomials with @xmath49 , section [ sec : chambercones ] here for connections to @xmath5-variate @xmath6-nomials , the forthcoming paper @xcite for the general univariate case , and the forthcoming paper @xcite for chamber cone theory for @xmath50 sparse polynomial systems . one of the main goals of our paper is thus to illustrate and clarify the underlying theory in a non - trivial special case . we now state our second main result . recent advances in semidefinite programming have produced efficient algorithms for finding sum of squares representations of certain nonnegative polynomials , thus enabling efficient polynomial optimization under certain conditions . when the input is a sparse polynomial it is then natural to ask if there is a sum of squares representation that also respects sparseness . indeed , it is well - known that a nonnegative univariate polynomial can always be written as a sum of two squares of , usually non - sparse , polynomials ( see , e.g. , @xcite for refinements ) . the following result demonstrates that a sparse analogue is either unlikely or much more subtle . [ thm : neg ] there do _ not _ exist absolute constants @xmath51 and @xmath52 with the following property : any trinomial @xmath53 $ ] that is positive on @xmath41 can be written in the form @xmath54 , for some @xmath55 $ ] with @xmath56 having at most @xmath52 terms for all @xmath57 . were there to be a sufficiently efficient representation of positive sparse polynomials as sums of squares , one could then try to use semidefinite programming to find such a representation explicitly for a given polynomial . this in turn could yield an efficient reduction from deciding the existence of real roots to a ( small ) semidefinite programming problem , similar to the techniques of @xcite . our last theorem thus reveals an obstruction to this sums of squares approach . the best known algorithms for real root counting lack speed - ups for sparse polynomials like the average - case complexity bound from our first main result . for example , in the notation of theorem [ thm : pos ] , @xcite gives an arithmetic complexity bound of @xmath58 which , via the techniques of @xcite , yields a bit complexity bound super - linear in @xmath59 . no algorithm with complexity polynomial in @xmath3 ( deterministic , randomized , or average - case ) appears to have been known before for tetranomials . ( see @xcite for recent speed benchmarks of univariate real solvers . ) as for alternative approaches , softening our concept of sparse sum of squares representation may still enable speed - ups similar to theorem [ thm : pos ] via semidefinite programming . for instance , one could ask if a positive trinomial of degree @xmath1 always admits a representation as a sum of @xmath60 squares of polynomials with @xmath60 terms . this question appears to be completely open . [ ex : log ] _ observe that a quick derivative computation yields that + @xmath61 + attains a unique minimum value of @xmath14 at @xmath62 . so this @xmath0 is nonnegative . on the other hand , one can prove easily by induction that @xmath63 , thus yielding an expression for @xmath0 as a sum of @xmath64 _ bi_nomials with @xmath65 . @xmath19 _ note also that while we focus on speed - ups that replace the polynomial degree @xmath1 by @xmath3 in this paper , other practically important speed - ups combining semidefinite programming and sparsity are certainly possible ( see , e.g. , @xcite ) . let us first brieflyreview two important constructions by gelfand , kapranov , and zelevinsky . [ dfn : first ] _ let @xmath66 , let @xmath67 have cardinality @xmath52 , and define the corresponding family of ( laurent ) polynomials + @xmath68 , + where the notation @xmath69 is understood . when @xmath70 for all @xmath71 then we call @xmath4 the _ support _ of @xmath72 , also using the notation @xmath73 . @xmath19 _ [ dfn : amoeba ] _ for any field @xmath74 we let @xmath75 . given any @xmath76 $ ] , we then define its _ amoeba _ , @xmath77 , to be @xmath78 . @xmath19 _ ( weaker version of ( * ? ? ? * cor . 1.8 ) ) _ following the notation of definition [ dfn : amoeba ] , the complement of @xmath77 in @xmath79 is a finite disjoint union of open convex sets . @xmath80 _ an example of an amoeba of a bivariate polynomial ( see example [ ex : tetra2 ] below ) appears in the right - hand illustration . while the complement of the amoeba ( in white ) appears to have @xmath22 convex connected components , there are in fact @xmath2 : the fourth component is a thin sliver emerging further below from the downward pointing tentacle . _ following the notation of definition [ dfn : first ] and letting @xmath81 , we define @xmath82 the _ @xmath4-discriminant variety _ * chs . 1 & 911 ) to be the closure of the set of all @xmath83\!\in\!{{\mathbb{p}}}^{m-1}_{\mathbb{c}}$ ] such that + @xmath84 + has a solution in @xmath85 . we then define ( up to sign ) the _ @xmath4-discriminant _ , @xmath86 $ ] , to be the ( irreducible ) defining polynomial of @xmath82 when @xmath82 is a hypersurface . finally , we let @xmath87 denote the real part of @xmath82 . @xmath19 _ [ rem : gen ] _ the @xmath82 considered in this paper will all ultimately be hypersurfaces . @xmath19 _ [ ex : tetra2 ] _ taking @xmath88 , we see that + _ @xmath89 consists simply of polynomials of the form @xmath90@xmath91 . the underlying @xmath4-discriminant is then a polynomial in the @xmath92 having @xmath93 monomial terms and degree @xmath94 . however , while @xmath95 is unwieldy , we can still easily plot the real part of its zero set @xmath87 via the _ horn - kapranov uniformization _ ( see its statement below , and the illustration to the right ) . @xmath19 ( 100,10)(0,0 ) ( 15,-3 ) the plotted curve above is the image of the real roots of @xmath96 under the @xmath97 map , i.e. , the amoeba of @xmath98 . amoebae give us a convenient way to introduce polyhedral / tropical methods into our setting . for our last example , the boundary of @xmath99 is contained in the curve above . @xmath4-discriminants are notoriously large in all but a few restricted settings . for instance , the polynomial @xmath100 defining the curve above has the following coefficient for @xmath101 : 903947086576700909448402875044761267196347419431440828445529608410806270 + ... [ 2062 digits omitted ] ... ... 93441588472666704061962310429908170311749217550336 . + fortunately , we have the following theorem , describing a one - line parametrization of @xmath82 . _ [ 1](see @xcite , @xcite , and ( * ? ? ? [ 1]given @xmath102 with @xmath82 a hypersurface , the discriminant locus @xmath82 is exactly the closure of + [ 1 ] @xmath103\ ; \left| \ ; u\!:=\!(u_1,\ldots , u_m)\!\in\!{\mathbb{c}}^m,\ { { \mathcal{a}}}u \!=\!{\mathbf{o } } , \ \sum^m_{i=1}u_i\!=\!0 , \ \right\}$ ] . @xmath80 _ thus , once we know the null - space of an @xmath104 matrix , we have a formula parametrizing @xmath82 . recall that for any two subsets @xmath105 , their _ minkowski sum _ @xmath106 is@xmath107 . also , for any matrix @xmath108 , we let @xmath109 denote its transpose . [ cor : mink ] following the notation above , let @xmath110 denote the @xmath104 matrix whose @xmath111 column has coordinates corresponding to @xmath112 , let @xmath113 be any real matrix whose columns are a basis for the right null - space of @xmath110 , and define @xmath114 via @xmath115 . then @xmath116 is the minkowski sum of the row space of @xmath110 and @xmath117 . @xmath80 if one is familiar with elimination theory , then it is evident from the horn - kapranov uniformization that discriminant amoebae are subspace bundles over a lower - dimensional amoeba . this is a geometric reformulation of the homogeneities satisfied by the polynomial @xmath95 . [ ex : tetracell ] _ continuing example [ ex : tetra2 ] , we observe that @xmath118 has right null - space generated by the respective transposes of @xmath119 and @xmath120 . the horn - kapranov uniformization then tells us that @xmath82 is simply the closure of the rational surface @xmath121 \ ; \right| \ ; t_1,t_2\!\in\!{\mathbb{c } } , \lambda\!\in\!{{\mathbb{c}}^*}\right\}\subset { { \mathbb{p}}}^3_{\mathbb{c}}$ ] . note that @xmath0 and @xmath122 have the same roots and that @xmath123 is a well - defined bijection on @xmath41 that preserves sign . note also that the roots of @xmath0 and @xmath124 { $ \left(\frac{c_1}{c_3}\right)^{1/405}$}}y\right)$ ] differ only by a scaling when @xmath0 has real coefficients , and that @xmath125 is of the form @xmath126 . it then becomes clear that we can reduce the study of @xmath87 to a lower - dimensional slice : intersecting @xmath82 with the plane defined by @xmath127 yields the following parametrized curve in @xmath128 : @xmath129 . in other words , the preceding curve is the closure of the set of all @xmath130 such that @xmath131 has a degenerate root in @xmath132 . our preceding illustration of the image of @xmath133 within @xmath99 ( after taking log absolute values of coordinates ) thus has the following explicit parametrization : + [ 1]@xmath134\in{{\mathbb{p}}}^1_{\mathbb{r}}}$ ] @xmath19 _ a geometric fact about amoebae that will prove quite useful here is the following elegant quantitative result of passare and rullgrd . recall that the _ newton polytope _ of a laurent polynomial @xmath135 $ ] is the _ convex hull _ of the exponent vectors appearing in the monomial term expansion of @xmath0 . * cor . 1 ) _ suppose @xmath136 $ ] has newton polygon @xmath137 . then @xmath138 . @xmath80 _ @xmath4-discriminants are central in real root counting because the real part of @xmath82 determines where , in coefficient space , the real zero set of a polynomial changes topology . recall that a _ cone _ in @xmath79 is any subset closed under addition and nonnegative linear combinations . the dimension of a cone @xmath139 is the dimension of the smallest flat containing @xmath139 . [ dfn : reduction ] _ suppose @xmath67 has cardinality @xmath52 and @xmath82 is a hypersurface . we then call any connected component @xmath140 of the complement of @xmath82 in @xmath141 a _ ( real ) discriminant chamber_. also let @xmath110 denote the @xmath104 matrix whose @xmath111 column has coordinates corresponding to @xmath112 and let @xmath142 be any real matrix @xmath143\!\in\!{\mathbb{r}}^{m\times p}$ ] with @xmath144 invertible . if @xmath145 contains an @xmath52-dimensional cone then we call @xmath140 an _ outer _ chamber ( of @xmath82 ) . all other chambers of @xmath82 are called _ inner _ chambers ( of @xmath82 ) . finally , we call the formal expression @xmath146 a monomial change of variables , and we refer to images of the form @xmath147 ( with @xmath140 an inner or outer chamber ) as _ reduced _ chambers . @xmath19 _ it is easily verified that @xmath148 . the latter notation simply means the image of @xmath149 under right multiplication by the matrix @xmath142 . _ the illustration from example [ ex : tetra2 ] shows @xmath150 partitioned into what appear to be @xmath22 convex and unbounded regions , and @xmath15 non - convex unbounded region . there are in fact @xmath2 convex and unbounded regions , with the fourth visible only if the downward pointed spike were allowed to extend much farther down . so @xmath88 results in exactly @xmath2 reduced outer chambers . @xmath19 _ note that exponentiating by any @xmath142 as above yields a well - defined multiplicative homomorphism from @xmath151 to @xmath152 when @xmath142 has rational entries with all denominators odd . in particular , the definition of outer chamber is independent of @xmath142 , since ( for the @xmath142 considered above ) @xmath153 is unbounded and convex iff @xmath154 is unbounded and convex , where @xmath155 is any matrix whose columns are a basis for the orthogonal complement of the row space of @xmath110 . one can in fact reduce the study of the topology of sparse polynomial real zero sets to representatives coming from reduced discriminant chambers . a special case of this reduction is the following result . ( see ( * ? ? ? [ lemma : cell ] suppose @xmath156 has @xmath157 elements , @xmath4 is not contained in any @xmath158-flat , @xmath159 has cardinality @xmath5 for all facets @xmath160 of @xmath161 , all the entries of @xmath162 have odd denominator , and @xmath163 is invertible . also let @xmath164 have respective real coefficient vectors @xmath17 and @xmath165 with @xmath166 and @xmath167 lying in the same reduced discriminant chamber . then all the complex roots of @xmath0 and @xmath13 are non - singular , and the respective zero sets of @xmath0 and @xmath13 in @xmath168 are diffeotopic . in particular , for @xmath169 , we have that either @xmath170 and @xmath171 have the same number of positive roots or @xmath170 and @xmath172 have the same number of positive roots . @xmath80 certain quantitative results on integer matrix factorizations and linear forms in logarithms will prove crucial for our main algorithmic results . [ dfn : smith ] _ let @xmath173 denote the set of @xmath174 matrices with all entries integral , and let @xmath175 denote the set of all matrices in @xmath176 with determinant @xmath177 ( the set of _ unimodular _ matrices ) . recall that any @xmath174 matrix @xmath178 $ ] with @xmath179 for all @xmath180 is called _ upper triangular_. _ _ given any @xmath181 , we then call an identity of the form @xmath182 , with @xmath183\!\in\!{\mathbb{z}}^{n\times m}$ ] upper triangular and @xmath184 , a _ hermite factorization _ of @xmath108 . also , if we have the following conditions in addition : _ 1 . @xmath185 for all @xmath186 . [ 1]for all @xmath57 , if @xmath187 is the smallest @xmath188 such that @xmath189 then @xmath190 for all @xmath191 . then we call @xmath192 _ hermite normal form _ of @xmath108 . @xmath19 [ prop : mono ] we have that @xmath193 for any @xmath194 . also , for any field @xmath74 , the map defined by @xmath195 , for any unimodular matrix @xmath196 , is an automorphism of @xmath197 . @xmath80 [ thm : unimod ] ( * ? ? ? * ch . 6 , table 6.2 , pg . 94 ) for any @xmath198\!\in\!{\mathbb{z}}^{n\times m}$ ] with @xmath199 , the hermite factorization of @xmath200 can be computed within @xmath201 bit operations . furthermore , the entries of all matrices in the hermite factorization have bit size @xmath202 . @xmath80 the following result is a very special case of work of nesterenko that dramatically refines baker s famous theorem on linear forms in logarithms @xcite . [ thm : log ] ( see ( * ? ? ? * thm . 2.1 , pg . for any integers @xmath203 with @xmath204 for all @xmath57 , define @xmath205 . then @xmath206 is bounded above by @xmath207 . @xmath80 the most obvious consequence of lower bounds for linear forms in logarithms is an efficient way to determine the signs of monomials in integers . [ algor : sign ] + * input : * positive integers @xmath208 and @xmath209 with @xmath210 for all @xmath57 . + * output : * the sign of @xmath211 . + * description : * + 1 . check via gcd - free bases ( see , e.g. , ( * ? ? ? 8.4 ) ) whether @xmath212 . if so , output `` they are equal . '' and stop . 2 . let @xmath213 and + @xmath214 . 3 . for all @xmath215 $ ] ( resp . @xmath216 $ ] ) , let @xmath217 ( resp . @xmath218 ) be a rational number agreeing with @xmath219 ( resp . @xmath220 ) in its first @xmath221 ( resp . @xmath222 ) leading bits . output the sign of @xmath223 and stop . [ lemma : bin ] algorithm [ algor : sign ] is correct and , following the preceding notation , runs within a number of bit operations asymptotically linear in + @xmath224 , + where @xmath225 . @xmath80 lemma [ lemma : bin ] follows immediately from theorem [ thm : log ] , the well - known fast iterations for approximating log ( see , e.g. , @xcite ) , and the known refined bit complexity estimates for fast multiplication ( see , e.g. , ( * ? ? ? * table 3.1 , pg . _ suppose @xmath226 is convex and @xmath227 is the polyhedral cone consisting of all @xmath228 with @xmath229 . we call @xmath160 the _ recession cone _ for @xmath230 and , if @xmath231 satisfies(1 ) @xmath232 and ( 2 ) @xmath233 for any @xmath234 , then we call @xmath235 the _ placed _ recession cone . in particular , the placed recession cone for @xmath236 for @xmath140 an outer chamber ( resp . a reduced outer chamber ) is called a _ chamber cone _ ( resp . a _ reduced chamber cone _ ) of @xmath82 . we call the facets of the ( reduced ) chamber cones of @xmath82 _ ( reduced ) walls of @xmath82_. we also refer to walls of dimension @xmath15 as _ rays_. @xmath19 _ _ returning to example [ ex : tetra2 ] , let us draw the rays that are the boundaries of the @xmath2 reduced chamber cones : while there appear to be just @xmath22 reduced chamber cones , there are in fact @xmath2 , since there is an additional ( slender ) reduced chamber cone with vertex placed much farther down . ( the magnified illustration to the right actually shows @xmath16 nearly parallel rays going downward , very close together . ) note also that reduced chamber cones need not share vertices . @xmath19 _ [ 1 ] chamber cones are well - defined since chambers are _ log - convex _ , being the domains of convergence of a particular class of hypergeometric series ( see , e.g. , ( * ? ? ? * ch . 6 ) ) . a useful corollary of the horn - kapranov uniformization is a surprisingly simple and explicit description for chamber cones . [ dfn : radiant ] _ suppose @xmath237 has cardinality @xmath157 , is not contained in any @xmath158-flat , and is not a pyramid . such that @xmath238 lies in an @xmath158-flat . ] also let @xmath142 be any real @xmath239 matrix whose columns are a basis for the right null space of @xmath110 and let @xmath240 be the rows of @xmath142 . we then call any set of indices @xmath241 satisfying : + ( a ) @xmath242_{i\in { \mathcal{i}}}$ ] is a maximal rank @xmath15 submatrix of @xmath142 . + ( b ) @xmath243 is _ not _ the zero vector . + a _ radiant subset corresponding to @xmath4_. @xmath19 _ [ thm : walls ] suppose @xmath237 has cardinality @xmath157 , is not contained in any @xmath158-flat , is not a pyramid , and @xmath82 is a hypersurface . also let @xmath142 be any real @xmath239 matrix whose columns are a basis for the right null space of @xmath110 and let @xmath240 be the rows of @xmath142 . finally , let @xmath244\!:=\!b\begin{bmatrix}0 & -1\\ 1 & 0 \end{bmatrix } b^t$ ] and let @xmath245 denote the row vector whose @xmath246 coordinate is @xmath14 or @xmath247 according as @xmath248 is @xmath14 or not . then each wall of @xmath82 is the minkowski sum of the row - space of @xmath249 and a ray of the form @xmath250 for some unique radiant subset @xmath251 of @xmath4 and any @xmath252 . in particular , the number of walls of @xmath82 , the number of chamber cones of @xmath82 , and the number of radiant subsets corresponding to @xmath4 are all identical , and at most @xmath157 . note that the hypotheses on @xmath4 are trivially satisfied when @xmath253 has cardinality @xmath2 . note also that the definition of a radiant subset corresponding to @xmath4 is independent of the chosen basis @xmath142 since the definition is invariant under column operations on @xmath142 . _ theorem [ thm : walls ] thus refines an earlier result of dickenstein , feichtner , and sturmfels ( ( * ? ? ? 1.2 ) ) where , in essence , _ un_shifted variants of chamber cones ( all going through the origin ) were computed for nonpyramidal @xmath156 of arbitrary cardinality . a version of theorem [ thm : walls ] for @xmath4 of arbitrary cardinality will appear in @xcite . @xmath19 _ _ it is easy to show that a generic @xmath4 satisfying the hypotheses of our theorem will have exactly @xmath157 chamber cones , e.g. , example [ ex : tetra2 ] . it is also almost as easy to construct examples having fewer chamber cones . for instance , taking @xmath254 , we see that @xmath255{$\begin{bmatrix}[r ] -1 & 0\\ 2 & 1\\ -2 & -2\\ 0 & 1\\ 1 & 0\end{bmatrix}$}}$ ] satisfies the hypotheses of theorem [ thm : walls ] and that @xmath256 is a _ non_-radiant subset . so the underlying @xmath82 has only @xmath22 chamber cones . @xmath19 _ * proof of theorem [ thm : walls ] : * first note that by corollary [ cor : mink ] , @xmath116 is the minkowski sum of @xmath257 and the row space of @xmath110 , where @xmath258 . so then , determining the walls reduces to determining the directions orthogonal to the row space of @xmath110 in which @xmath259 becomes unbounded . since @xmath260 is in the row space of @xmath110 , we have @xmath261 and thus @xmath262 for all @xmath263 . so we can restrict to the compact subset @xmath264 for some @xmath57 . in particular , we see that there are no more than @xmath157 reduced walls . note also that @xmath265 iff @xmath266 tends to a suitable ( nonzero ) multiple of @xmath267 , in which case the coordinates of @xmath259 becoming unbounded are exactly those with index @xmath268 where @xmath251 is the unique radiant subset corresponding to those rows of @xmath142 that are nonzero multiples of @xmath269 . ( the assumption that @xmath4 not be a pyramid implies that @xmath142 can have no zero rows . ) furthermore , the coordinates of @xmath259 that become unbounded each tend to @xmath270 . note that condition ( b ) of the radiance condition comes into play since we are looking for directions _ orthogonal to the row - space of @xmath110 _ in which @xmath259 becomes unbounded . we thus obtain that each wall is of the asserted form . however , we still need to account for the coordinates of @xmath259 that remain bounded . now , if @xmath266 tends to a suitable ( nonzero ) multiple of @xmath271 , then it clear that any coordinate of @xmath259 of index @xmath272 tends to @xmath248 ( modulo a multiple of @xmath273 added to @xmath259 ) . so we have indeed described every wall , and given a bijection between radiant subsets corresponding to @xmath4 and the walls of @xmath82 . to conclude , note that the row space of @xmath110 has dimension @xmath274 by construction , so the walls are all actually ( parallel ) @xmath5-plane bundles over rays . so by the archimedean amoeba theorem , each outer chamber of @xmath82 must be bounded by @xmath16 walls and the walls have a natural cyclic ordering . thus the number of chamber cones is the same as the number of rays and we are done . @xmath80 an important consequence of theorem [ thm : walls ] is that while the underlying @xmath4-discriminant polynomial @xmath95 may have huge coefficients , the _ rays _ of a linear projection of @xmath116 admit a concise description involving few bits , save for the transcendental coordinates coming from the `` shifts '' @xmath245 . by applying our quantitative estimates from section [ sub : bit ] we can then quickly find which chamber cone contains a given @xmath5-variate @xmath6-nomial . [ thm : lp ] following the notation of theorem [ thm : walls ] , suppose @xmath275 $ ] and let @xmath276 denote the maximum bit size of any coordinate of @xmath4 . then we can determine the unique chamber cone containing @xmath0 or correctly decide if @xmath0 is contained in @xmath16 or more chamber cones within a number of arithmetic operations polynomial in @xmath277 . furthermore , if @xmath278 $ ] , @xmath279 is the maximum bit size of any coefficient of @xmath0 , and @xmath5 is fixed , then we can also obtain a _ bit _ complexity bound polynomial in @xmath280 . @xmath80 theorem [ thm : lp ] is the central tool behind our complexity results and follows immediately upon proving the correctness of ( and giving suitable complexity bounds for ) the following algorithm : [ algor : conesort ] + * input : * a subset @xmath156 of cardinality @xmath157 ( with @xmath4 not a pyramid and not contained in any@xmath158-flat , and @xmath82 a hypersurface ) and the coefficient vector @xmath17 of an @xmath281 $ ] . + * output : * radiant subsets @xmath251 and @xmath282 ( corresponding to @xmath4 ) generating the walls of the uniquechamber cone containing @xmath0 , or a true declaration that @xmath0 is contained in at least @xmath16 chamber cones . + * description : * + 1 . ( preprocessing ) compute the hermite factorization @xmath283 and let @xmath142 be the submatrix defined by the rightmost @xmath16 columns of @xmath284 . 2 . ( preprocessing ) let @xmath240 be the rows of @xmath142 , @xmath244\!:=\!b\begin{bmatrix}0 & -1\\ 1 & 0 \end{bmatrix } b^t$ ] , and let @xmath245 denote the row vector whose @xmath246 coordinate is @xmath14 or @xmath247 according as @xmath248 is @xmath14 or not . 3 . ( preprocessing ) find all radiant subsets @xmath285 corresponding to @xmath4 . 4 . ( preprocessing ) for any radiant subset @xmath251 let @xmath286 and let @xmath287 denote the row vector @xmath288 for any fixed @xmath252 . 5 . ( preprocessing ) sort the @xmath289 in order of increasing counter - clockwise angle with the @xmath290-coordinate ray and let @xmath291 denote the resulting ordered collection of @xmath289 . ( preprocessing ) for any radiant subset @xmath251 let @xmath292 denote the intersection of the lines @xmath293 and @xmath294 , where @xmath295 is the counter - clockwise neighbor of @xmath296 . let @xmath297 . 8 . via binary search , attempt to find a pair of adjacent rays of the form + @xmath298 containing @xmath299 . if ( @xmath300 and there is no such pair of rays ) + or + ( @xmath301 and there is such a pair of rays ) + then output `` your @xmath0 lies in at least @xmath16 distinct chamber cones . '' and stop . [ 1]if @xmath300 and there is such a pair of rays , set @xmath302 , delete @xmath289 and @xmath295 from @xmath291 , and goto step ( 2 ) . output `` your @xmath0 lies in the unique chamber cone determined by @xmath251 and @xmath282 . '' and stop . _ an important detail for large scale computation is that the preprocessing steps ( -5)(0 ) need only be done _ once _ per support @xmath4 . this can significantly increase efficiency in applications where one has just one ( or a few ) @xmath4 and one needs to answer chamber cone membership queries for numerous @xmath0 with the same support . @xmath19 _ * proof of correctness of algorithm [ algor : conesort ] : * first note that the computed matrix @xmath142 indeed has columns that form a basis for the right null - space of @xmath4 . this is because our assumptions on @xmath4 ensure that the rank of @xmath110 is @xmath274 and thus the last @xmath16 rows of @xmath303 consist solely of zeroes . by construction , theorem [ thm : walls ] then implies that the @xmath289 are exactly the reduced rays for @xmath82 , modulo an invertible linear map . ( the invertible map arises because right - multiplication by @xmath142 induces an injective but non - orthogonal projection of the right null - space of @xmath110 onto @xmath150 . ) it is then clear that the preprocessing steps do nothing more than give us a @xmath142 suitable for theorem [ thm : walls ] and a sorted set of reduced rays ready for chamber cone membership queries via binary search . in particular , since the reduced chamber cones cover @xmath150 , the correctness of steps ( 1)(3 ) is clear and we are done . @xmath80 in what follows , we will use the `` soft - oh '' notation @xmath304 to abbreviate bounds of the form @xmath305 . * complexity analysis of algorithm [ algor : conesort ] : * let us first approach our analysis from the more involved point of view of bit complexity . our arithmetic complexity bound will then follow upon a quick revisit . by theorem [ thm : unimod ] ( which quotes bounds from @xcite ) , step ( -5 ) takes @xmath306 bit operations . also , the resulting bit size for the entries of @xmath142 is @xmath307 . the complexity of step ( -4 ) is negligible , save for the approximation of certain logarithms . the latter wo nt come into play until we start checking on which side of a ray a point lies , so let us analyze the remaining preprocessing steps . step ( -3 ) can be done easily through a greedy approach : one simply goes through the rows @xmath308 to see which rows are multiples of @xmath309 . once this is done , one checks if the resulting set of indices is indeed radiant or not , and then one repeats this process with the remaining rows of @xmath142 . in summary , this entails @xmath310 arithmetic operations on number of bit - size @xmath307 , yielding a total of @xmath311 bit operations . step ( -2 ) has negligible complexity . the comparisons in step ( -1 ) can be accomplished by computing the cosine and sine of the necessary angles via dot products and cross products . via the well - known asymptotically optimal sorting algorithms , it is then clear that step ( -1 ) requires @xmath312 arithmetic operations on integers of bit size @xmath307 , contributing a total of @xmath313 bit operations . step ( 0 ) has negligible complexity . at this point , we see that the complexity of the preprocessing steps ( -5)(0 ) is @xmath314 bit operations . continuing on to steps ( 1)(3 ) , we now see that we are faced with @xmath315 sidedness comparisons between a point and an oriented line . more precisely , we need to evaluate @xmath315 signs of determinants of matrices of the form @xmath316 . each such sign evaluation , thanks to algorithm [ algor : sign ] and lemma [ lemma : bin ] , takes @xmath317 bit operations . we have thus proved our desired bit complexity bound which , while polynomial in @xmath280 for fixed @xmath5 , is visibly exponential in @xmath5 . note however that the exponential bottleneck occurs only in the sidedness comparisons of step ( 2 ) . to obtain an improved arithmetic complexity bound , we then simply observe that the sidedness comparisons can be replaced by computations of signs of differences of monomials , simply by exponentiating the resulting linear forms in logarithms . via recursive squaring ( * ? ? ? * thm . 5.4.1 , pg . 103 ) , it is then clear that each such comparison requires only @xmath318 arithmetic operations . so the overall number of arithmetic operations drops to polynomial in @xmath277 and we are done . let us now state some final combinatorial constructions before fully describing how chamber cones apply to real root counting . our use of outer chambers and chamber cones enables us to augment an earlier construction of viro . let us first recall that a _ triangulation _ of a point set @xmath4 is simply a simplicial complex @xmath319 whose vertices lie in @xmath4 . _ we say that a triangulation of @xmath4 is _ coherent _ iff its maximal simplices are exactly the domains of linearity for some function @xmath51 that is convex , continuous , and piecewise linear on the convex hull of @xmath4 . in particular , we will sometimes define such an @xmath51 by fixing the values @xmath320 for just those @xmath321 and then employing the convex hull of the points @xmath322 as @xmath24 ranges over @xmath4 . the resulting graph is known as the _ lower hull _ of the _ lifted _ point set @xmath323 . @xmath19 _ [ dfn : viro ] _ ( see proposition 5.2 and theorem 5.6 of ( * ? ? ? * ch . 5 , pp . 378393 ) . ) suppose @xmath156 is finite and the convex hull of @xmath4 has positive volume and boundary @xmath324 . suppose also that @xmath4 is equipped with a coherent triangulation @xmath319 and a function @xmath325 which we will call a _ distribution of signs for @xmath4_. we then call any edge with vertices of opposite sign an _ alternating edge _ , and we define a piece - wise linear manifold the _ viro diagram _ @xmath326 in the following local manner : for any @xmath5-cell @xmath327 , let @xmath328 be the convex hull of the set of all midpoints of alternating edges of @xmath139 , and then define @xmath329 . when @xmath330 and @xmath331 is the corresponding sequence of coefficient signs , then we also call @xmath332 the _ viro diagram of @xmath0 corresponding to @xmath319_. @xmath19 _ [ ex : penta ] _ consider @xmath333 . then @xmath334@xmath335 and has convex hull a pentagon . so then there are exactly @xmath336 coherent triangulations , yielding @xmath336 possible viro diagrams for @xmath0 ( drawn in thicker green lines ) : + + note that all these diagrams have exactly @xmath16 connected components , with each component isotopic to an open interval . note also that our @xmath0 here is a @xmath16-variate @xmath337-nomial . @xmath19 _ [ dfn : lift ] _ suppose @xmath237 has cardinality @xmath157 , is not a pyramid , is not contained in any @xmath158-flat , and @xmath82 is a hypersurface . also let @xmath142 be any real @xmath239 matrix whose columns are a basis for the right null space of @xmath110 . for any @xmath338 let us then define @xmath339 where @xmath251 and @xmath340 are the unique radiant subsets corresponding to the unique chamber cone containing @xmath341 . ( we set @xmath342 should there not be a unique such chamber cone . ) let us then define @xmath343 to be the convex hull of @xmath344 and let @xmath345 denote the triangulation of @xmath4 induced by the lower hull of @xmath343 . also , we call any polynomial of the form @xmath346 with @xmath160 a cell of @xmath345 a _ canonical lower polynomial _ for @xmath0 . finally , we call @xmath347 the _ canonical viro diagram _ of @xmath0 . @xmath19 _ _ let @xmath348 , @xmath349 , and @xmath350 . a routine calculation then reveals that @xmath351 is the pair of radiant subsets corresponding to the unique chamber cone containing @xmath341 . we then obtain that @xmath352 and thus @xmath343 is ( modulo some artistic stretching ) . in particular , @xmath345 thus has the single cell @xmath353 $ ] , which is an alternating cell , and thus @xmath354 consists of a single point . more than coincidentally , @xmath0 has exactly @xmath15 positive root . @xmath19 _ _ returning to example [ ex : penta ] , let @xmath355 . a routine calculation then reveals that the unique chamber cone containing @xmath341 is defined by the pair of radiant subsets @xmath351 . so then @xmath356 and @xmath345 is the exactly the upper middle triangulation from the illustration of example [ ex : penta ] . @xmath354 then consists of @xmath16 disjoint open intervals and , more than coincidentally , the positive zero set of @xmath0 has exactly @xmath16 connected components , each homeomorphic to an open interval . @xmath19 _ [ thm : vtx ] following the notation above , set @xmath357 and assume in addition that @xmath159 has cardinality @xmath5 for all facets @xmath160 of @xmath161 . then @xmath358 lies in an outer chamber @xmath359 the positive zero sets of @xmath360 , as @xmath266 ranges over @xmath361 $ ] , are each diffeotopic to the positive zero set of @xmath362 . in particular , @xmath363 and thus , when @xmath17 lies in an outer chamber , the positive zero set of @xmath0 is isotopic to @xmath354 . _ for @xmath169 we thus obtain that the number of positive roots of the tetranomial @xmath0 is exactly the cardinality of its canonical viro diagram . @xmath19 _ * proof : * by construction , the image of @xmath364 as @xmath266 ranges over @xmath361 $ ] is simply a ray entirely contained in a unique chamber cone . moreover , by assumption ( and since outer chambers are log convex ) , the ray is also contained entirely in @xmath97 of an outer chamber . so the first part of our theorem follows from lemma [ lemma : cell ] . the final part of our theorem is then just a reformulation of viro s theorem on the isotopy type of toric deformations of real algebraic sets ( see , e.g. , ( * ? ? ? * thm . 5.6 ) ) . @xmath80 the main contribution of our paper is thus an efficient method to associate a _ canonical _ viro diagram to the positive zero set of a given @xmath0 so that both @xmath365 manifolds have the same topology . such a method appears to be new , although the necessary ingredients have existed in the literature since at least the 1990s . in particular , to the best of our knowledge , all earlier applications of viro s method designed clever @xmath0 having the same topology as some specially tailored viro diagram , thus going in the opposite direction of our construction . we state up front that our method does _ not _ work for all @xmath0 . however , our development yields a sufficient condition outer chamber membership that holds with high probability under the stable log - uniform measure . [ thm : nplus3 ] suppose @xmath237 has cardinality @xmath157 , is not a pyramid , is not contained in any @xmath158-flat , and @xmath82 is a hypersurface . suppose also that the coefficients of an @xmath281 $ ] are independently chosen via the stable log - uniform measure over @xmath41 ( resp . @xmath42 ) . then with probability @xmath15 , @xmath0 lies in some outer chamber . in particular , if we assume in addition that @xmath159 has cardinality @xmath5 for all facets @xmath160 of @xmath161 , then the positive zero set of @xmath0 is isotopic to @xmath354 with probability @xmath15 . * proof : * by theorem [ thm : walls ] , @xmath116 is an @xmath5-plane bundle over @xmath366 , where @xmath367 $ ] and @xmath368 for suitable monomials @xmath369 and @xmath370 in @xmath17 . furthermore , thanks to corollary 8 of @xcite , the complement of @xmath99 has no bounded convex connected components . letting @xmath17 denote the coefficient vector of @xmath0 we thus obtain that @xmath0 lies in an outer chamber iff @xmath371 . in particular , it is clear by the passare - rullgrd theorem that in any large centered cube @xmath139 , the volume of @xmath372 occupies a vanishingly small fraction of @xmath139 . so the first assertion is proved . the final assertion is an immediate consequence of the first assertion and theorem [ thm : vtx ] . @xmath80 theorem [ thm : pos ] then follows easily from theorems [ thm : vtx ] and [ thm : nplus3 ] . the applications of theorems [ thm : vtx ] and [ thm : nplus3 ] to computational real topology will be pursued in another paper . let us first consider the following algorithm for counting the positive roots of `` most '' real univariate tetranomial . [ algor : tetra ] + * input : * a tetranomial @xmath53 $ ] with support @xmath4 . + * output : * a number in @xmath45 that , for any @xmath0 in an outer chamber of @xmath82 , is exactly the number of positive roots of @xmath0 . + * description : * + 1 . via algorithm [ algor : conesort ] , and any sub - quadratic planar convex hull algorithm ( see , e.g. , @xcite ) , compute the canonical viro diagram @xmath354 . 2 . if @xmath0 did not lie in a unique chamber cone then output `` your @xmath0 does not lie in an outer chamber so please use an alternative method . '' and stop . 3 . output the cardinality of @xmath354 and stop . assuming algorithm [ algor : tetra ] is correct , we can count the real roots of @xmath0 simply by applying algorithm [ algor : tetra ] to @xmath373 and @xmath374 . theorem [ thm : pos ] thus follows immediately upon proving the correctness of our last algorithm and a suitable complexity bound . we now do this . * proof of correctness of algorithm [ algor : tetra ] : * by theorem [ thm : vtx ] , the number of positive roots of @xmath0 is exactly the cardinality of @xmath354 whenever @xmath0 is in an outer chamber . so we indeed have correctness . @xmath80 * complexity analysis of algorithm [ algor : tetra ] : * let us first observe that algorithm [ algor : tetra ] indeed gives a correct answer with probability @xmath15 ( relative to the stable log - uniform measure ) : this is immediate from theorem [ thm : nplus3 ] . so now we need only prove a suitable complexity bound . let us begin with the more refined setting of bit complexity : from our earlier complexity analysis of algorithm [ algor : conesort ] , it is clear that step ( 1 ) needs @xmath375 bit operations , neglecting the computation of @xmath354 . the complexity of computing @xmath354 ( which is essentially dominated by computing the convex hull of @xmath2 points with coordinates of bit size @xmath64 ) is clearly negligible in comparison . so we obtain a final bit complexity bound of @xmath376 . ( the complexity of steps ( 2 ) and ( 3 ) is negligible . ) for arithmetic complexity , our earlier complexity analysis of algorithm [ algor : conesort ] specializes easily to an upper bound of @xmath377 . ( the speed - up arises from the easiness of checking inequalities involving integral powers of real numbers in the bss model over @xmath41 . ) so we are done . @xmath80 _ it is important to note that when @xmath0 lies in a chamber cone but _ not _ in any outer chamber , algorithm [ algor : tetra ] can give a wrong answer . however , thanks to theorem [ thm : nplus3 ] , such an occurence has probability @xmath14 under the stable log - uniform measure . @xmath19 _ to prepare for the proof , we first set up some notation . fix positive integers @xmath51 and @xmath52 . let @xmath378 be a matrix of natural numbers ordered as @xmath379 for @xmath380 and @xmath381 . also , let @xmath382 be indeterminates over the same indexing set . consider now the following polynomial : @xmath383[x ] \\ = \ & g_d(p ) x^d + \cdots + g_1(p ) x + g_0(p ) , \\ \end{split}\ ] ] in which each nonzero @xmath384 is a homogeneous ( quadratic ) polynomial in @xmath385 $ ] . notice that at most @xmath386 of these @xmath56 are nonzero . we will refer to the integer @xmath387 as the _ exponent _ corresponding to the _ coefficient _ @xmath382 . suppose now that @xmath391 $ ] has degree @xmath392 and is a sum of @xmath51 squares , each involving at most @xmath52 terms . then , there is a set of exponents @xmath137 and an assignment @xmath393 for the coefficients @xmath382 such that @xmath394 . if we fix a set of exponents @xmath137 for @xmath0 , then conversely , any real point in the variety determined by the @xmath56 gives a representation of @xmath0 as a sum of @xmath51 squares , each involving at most @xmath52 terms . we will prove theorem [ thm : neg ] using contradiction by showing that a certain infinite family of trinomials can not all have sparse representations of the form in equality ( [ supposedsos ] ) . for this approach to work , however , we will need to find a universal set of coefficients @xmath395 that represents " all of them . [ universalaij ] let @xmath396 $ ] be an infinite collection of polynomials which are sums of @xmath51 squares , each involving at most @xmath52 terms . moreover , suppose that the nonzero coefficients of polynomials @xmath397 come from a finite set @xmath139 . then , there is an infinite subset @xmath398 of them with corresponding exponents @xmath399 and a single set of real coefficients @xmath395 such that @xmath400 for all @xmath401 . * proof : * let @xmath398 be an infinite sequence of polynomials with corresponding exponents @xmath399 . also let @xmath402 be the set of all possible ( polynomial ) coefficients of polynomials @xmath403 ; from lemma [ finiteglemma ] , this set is finite . by assumption , each @xmath404 gives rise to a set of equations involving a subset of @xmath402 and real numbers coming from the finite set @xmath139 . the set of all such equations is again finite , and therefore , by the infinite pigeon - hole principle , there is a subsequence of the @xmath404 which have the same set of equations governing their coefficients @xmath382 . picking any real solution to such a set of equations finishes the proof . we now have all the tools we need to prove theorem [ thm : neg ] . however , to make the argument less cumbersome , we first recall little - oh " notation : for a function @xmath405 , we say that @xmath406 if @xmath407 it is easy to see that the sum of any finite number of such functions is also @xmath408 . moreover , if @xmath409 for some constant @xmath410 , then @xmath411 . * proof of theorem [ thm : neg ] : * suppose , by way of contradiction , that every positive definite trinomial can be written sparsely as the sum of @xmath51 squares , each involving at most @xmath52 terms . consider the infinite sequence of positive definite trinomials , @xmath412 using lemma [ universalaij ] , we can find a subsequence @xmath413 with corresponding exponents @xmath414 and a single set of real numbers @xmath395 such that latexmath:[$f_{k_s } = s_{p_{k_s}}(x ) assume that the @xmath395 are chosen such that the number of them that are zero is maximal among all such sets of coefficients @xmath395 . moreover , we shall assume that this maximality still holds if we restrict to any infinite subsequence of the @xmath413 . for clarity of exposition in what follows , we will not keep updating the subscripting of indices when taking subsequences . given an exponent matrix @xmath416 , define a new matrix @xmath417 this corresponds naturally to the transformation @xmath418 applied to both sides of an equation @xmath419 . since @xmath420 , each matrix @xmath421 has entries in the interval @xmath422 $ ] . by compactness , we may choose a subsequence @xmath423 such that @xmath421 converges in the ( entry - wise ) euclidean norm to a matrix @xmath424^{\ell \times m}$ ] . clearly , we have @xmath425 and also that some entry of @xmath426 is @xmath14 ; it turns out that these are the only possible numbers for entries of @xmath426 which play a role in ( [ supposedsos ] ) . suppose , on the contrary , that @xmath426 contains @xmath429 real numbers , @xmath430 with @xmath431 . each power of @xmath290 occurring in a summand of equality ( [ supposedsos ] ) , after squaring , is of the form @xmath432 thus , for all sufficiently large @xmath331 , the powers of @xmath290 occurring in equality ( [ supposedsos ] ) are partitioned into classes determined by the number of distinct values , @xmath433 notice that the numbers in formula ( [ exponentssqring ] ) all become strictly smaller ( resp . larger ) than ( resp . @xmath14 ) as @xmath434 unless @xmath435 ( resp . @xmath436 ) . in particular , with @xmath331 large , the polynomials @xmath437 $ ] for @xmath438 do not involve the indeterminates @xmath382 coming from exponents of the form @xmath439 with @xmath440 . conversely , each monomial in every polynomial @xmath441 with @xmath442 not in one of the classes from ( [ beginendgs ] ) contains at least one such indeterminate @xmath382 . since the only nonzero coefficients of the sequence ( [ trinomialfamily ] ) come from the classes from ( [ beginendgs ] ) , it follows that we may replace with @xmath14 all coefficients @xmath395 corresponding to exponents @xmath439 with @xmath440 and still have an equality @xmath443 the claim therefore follows from the maximality property of the chosen set of coefficients @xmath395 . we next examine what all this says about the limiting behavior of the expressions from equality ( [ supposedsos ] ) . from the claim , it follows that when @xmath331 is large , we need only consider those exponents from the matrices @xmath423 that are on the order @xmath444 fix such a large @xmath331 and consider the smallest exponent @xmath410 on the order @xmath445 that occurs with a nonzero coefficient in ( [ supposedsos ] ) after substituting the @xmath395 for the @xmath382 . when the sum from ( [ supposedsos ] ) is expanded , the term @xmath446 will appear with positive coefficient ; i.e. , @xmath447 it follows from the form of the sequence ( [ trinomialfamily ] ) that @xmath448 ( since the other nonzero term is odd ) . in particular , it is not possible to obtain a nonzero coefficient for @xmath449 in @xmath413 . this contradiction completes the proof . @xmath80 baker , alan , _ `` the theory of linear forms in logarithms , '' _ in transcendence theory : advances and applications : proceedings of a conference held at the university of cambridge , cambridge , jan.feb . , 1976 , academic press , london , 1977 . dickenstein , alicia ; rojas , j. maurice ; rusek , korben ; shih , justin , _ `` extremal real algebraic geometry and @xmath4-discriminants , '' _ moscow mathematical journal , vol . 7 , no . 3 , ( july september , 2007 ) , pp . 425452 . hemmer , michael ; tsigaridas , elias p. ; zafeirakopolous , zafeirakis ; emiris , ioannis z. ; karavelas , menelaos i. ; and mourrain , bernard , _ `` experimental evaluation and cross - benchmarking of univariate real solvers , '' _ proceedings of snc 2009 ( symbolic - numeric computation , kyoto , japan , august 25 ) , pp . 4554 , acm press , 2009 . kojima , m. and muramatsu , m. , _ `` a note on sparse sos and sdp relaxations for polynomial optimization problems over symmetric cones , '' _ computational optimization and applications vol . 42 ( 1 ) , pp . 3141 ( 2009 ) . malajovich , gregorio and rojas , j. maurice , _ `` high probability analysis of the condition number of sparse polynomial systems , '' _ theoretical computer science , special issue on algebraic and numerical algorithms , vol . 315 , no . 23 , ( may 2004 ) , pp . 525555 . parrilo , pablo a. , _ `` semidefinite programming relaxations for semialgebraic problems , '' _ algebraic and geometric methods in discrete optimization , math . program . 96 ( 2003 ) , no . 2 , ser . 293320 . passare , mikael and tsikh , august , _ `` amoebas : their spines and their contours , '' _ idempotent mathematics and mathematical physics , contemp . , v. 377 , amer . soc . , providence , ri , 2005 , pp . 275288 . pbay , philippe ; rojas , j. maurice ; thompson , david c. , _ `` optimization and @xmath450-completeness of certain fewnomials , '' _ proceedings of snc 2009 ( august 35 , 2009 , kyoto , japan ) , pp . 133142 , acm press , 2009 . rojas , j. maurice and ye , yinyu , _ `` on solving sparse polynomials in logarithmic time , '' _ journal of complexity , special issue for the 2002 foundations of computation mathematics ( focm ) meeting , february 2005 , pp . 87110 .

a long standing problem in weak lensing is about how to construct cosmic shear estimators from galaxy images . conventional methods average over a single quantity per galaxy to estimate each shear component . we show that any such shear estimators must reduce to a highly nonlinear form when the galaxy image is described by three parameters ( pure ellipse ) , even in the absence of the point spread function ( psf ) . in the presence of the psf , we argue that this class of shear estimators do not likely exist . alternatively , we propose a new way of measuring the cosmic shear : instead of averaging over a single value from each galaxy , we average over two numbers , and then take the ratio to estimate the shear component . in particular , the two numbers correspond to the numerator and denominators which generate the quadrupole moments of the galaxy image in fourier space , as proposed in zhang ( 2008 ) . this yields a statistically unbiased estimate of the shear component . consequently , measurements of the n - point spatial correlations of the shear fields should also be modified : one needs to take the ratio of two correlation functions to get the desired , unbiased shear correlation . [ firstpage ] cosmology : gravitational lensing - methods : data analysis - techniques : image processing : large scale structure

1002.3615

weak gravitational lensing refers to the weak and systematic shape distortions of background source images ( galaxies , cmb , etc . ) by the foreground inhomogeneous density distributions on cosmological scales . since this effect only involves gravity , it has been widely used as a direct probe of matter density fluctuations of our universe ( see , , @xcite for a recent review ) . the weak lensing effect can only be probed statistically due to the fact that the intrinsic projected shape of each galaxy is always somewhat random and anisotropic . a central theme in the study of weak lensing is to find unbiased cosmic shear estimators on galaxy images . this is indeed very challenging because the shape distortion due to weak lensing is generally much weaker than the intrinsic variations of the galaxy shapes . in the early stage of this field , most of the work focused on issues regarding the use of the quadrupole moments of a galaxy image as a shear estimator ( @xcite ) . ever since then , a number of other shear estimators have been considered in the literature , including moments defined by a certain set of orthogonal functions ( @xcite ) , the spatial derivatives of the galaxy surface brightness field ( @xcite ) , etc .. conventionally , for each shear component , the shear estimator is simply one number derived from a galaxy image , whose statistical mean is supposed to be equal to the true shear value , provided that the intrinsic galaxy image is statistically isotropic . unfortunately , even in the absence of the psf , we show that such shear estimators at least do not exist in simple forms , making them hard to use in practice for precise shear measurements ( , in the presence of noise ) . in the presence of the psf , we argue that such shear estimators do not likely exist . we give reasons for the above statements in [ not_exist ] . ( the readers who are just interested in our new way of measuring shears may skip this section . ) in [ alternatives ] , we present a new form of shear estimators : instead of having only one number from each galaxy image for each shear component , one can keep _ two _ numbers , and use the ratio of their averages over many galaxies to accurately measure the cosmic shear . we find that this new way of measuring the shear can be easily implemented by using the method of zhang ( 2008 ) ( z08 hereafter ) . the new type of shear estimator requires weak lensing statistics such as the n - point correlation functions of the shear field to be carried out in a slightly unusual way , but with little additional cost . this is discussed in [ statistics ] . in [ examples ] , we give numerical examples . finally , we summarize in [ summary ] . conventional shear estimators are defined as a class of shear estimators , which average over a single quantity per galaxy to estimate each shear component . most of the existing shear measurement methods belong to this class . for example , the method by kaiser et al . ( 1995 ) and its extensions basically use the quadrupole moments of each gaussian - profile - weighted galaxy image to measure the shear components ; in the shapelets method ( @xcite ) , the value of each shear component is estimated from best - fitting a shapelets model to each observed galaxy image ; the method of bernstein and jarvis ( 2002 ) evaluate the shear components by fitting each galaxy shape ( also the psf ) with a series of orthogonal 2d gaussian - based functions ( see , , @xcite for more examples ) . a common feature of these methods is that they all generate one quantity per galaxy for each shear component . in [ classic ] , we start our discussion with the well - known examples of shear estimators consisting of quadrupole moments of galaxy images , and show what the issues are . in [ no_psf ] , we show that , even in the absence of the psf , any conventional shear estimator has to reduce to a _ highly _ nonlinear form , making it hard to use in practice . in [ prove ] , we provide arguments as to why we think that conventional shear estimators do not likely exist when a psf is present . the use of galaxy quadrupole moments as shear estimators has been a central topic in weak lensing for many years . it is therefore easier to start our discussion with the quadrupole moments . to present the issues clearly , let us first consider the case without the psf or any photon noise . for convenience , we use @xmath0 or @xmath1 instead of @xmath2 for coordinates in 2d in this paper . suppose that the surface brightness field of the lensed galaxy image is @xmath3 on the image plane , and that of the original ( pre - lensing ) galaxy image is @xmath4 on the source plane , where @xmath5 and @xmath6 are the position angles on the image and source planes , respectively . we have the following relations : @xmath7 where @xmath8 , and @xmath9 , which are the spatial derivatives of the lensing deflection angle . @xmath10 can also be written as @xmath11 , where @xmath12 is sometimes called the lensing potential . matrix @xmath13 can be alternatively written in terms of the convergence @xmath14 and the two shear components @xmath15 and @xmath16 . the quadrupole moments of the lensed galaxy image are defined as follows : @xmath17 where the origin of the coordinates has been chosen to be the center of the light , , @xmath18 let us also define the ellipticities of the image as : @xmath19 in the absence of the psf , the quantities @xmath20 and @xmath21 are often thought to be good estimators for @xmath22 and @xmath23 up to the first order in the shear . let us find out if they are indeed unbiased shear estimators . the observed quadrupole can be rewritten from using eq.([fifstits ] ) in eq.([q_moments ] ) : @xmath24 note that the last step of the above equation is achieved by redefining @xmath25 as @xmath26 . keeping up to first order in @xmath27 , @xmath22 , and @xmath23 in eq.([q2 ] ) , we get : @xmath28 where @xmath29 are the quadrupole moments of the original galaxy image defined as : @xmath30 note that the two light centers defined in the image and source planes coincide . based on eq.([q3 ] ) , we find : @xmath31 - 2\gamma_2\epsilon_1^s\epsilon_2^s\\ \nonumber \epsilon_2&=&\frac{\epsilon_2^s+2\gamma_2}{1 + 2\gamma_1\epsilon_1^s+2\gamma_2\epsilon_2^s}\\ \nonumber & = & \epsilon_2^s+2\gamma_2\left[1-\left(\epsilon_2^s\right)^2\right]-2\gamma_1\epsilon_1^s\epsilon_2^s\end{aligned}\ ] ] where @xmath32 given that the surface brightness distribution of the original galaxy image is statistically isotropic , we have @xmath33 and @xmath34 . therefore , we find @xmath35\\ \nonumber \langle\epsilon_2\rangle&=&2\gamma_2\left[1-\langle\left(\epsilon_2^s\right)^2\rangle\right]\end{aligned}\ ] ] this result , eq.([q5 ] ) , clearly shows that @xmath20 and @xmath21 are _ not _ unbiased shear estimators , as @xmath36 and @xmath37 in the multiplicative factors depend on the galaxy morphology distribution , and can not be reduced to constant factors . ( also see eq.(3.29 ) of bernstein & jarvis ( 2002 ) , or eq.(9.5.26 ) of weinberg ( 2008 ) . ) one can construct an unbiased estimator of the shear , if one keeps three quantities from each lensed galaxy , @xmath38 , and @xmath39 , and use the ratios of their averages . assuming statistical isotropy of intrinsic galaxy shapes in eq.([q3 ] ) , and keeping up to first order in shear / convergence , we have ( also see eq.(9.5.30 ) of weinberg ( 2008 ) ) : @xmath40 this form of shear estimators is not conventional , as one has to keep more than one quantities from each galaxy image for each shear component . it is this class of estimators we shall discuss in this paper in detail . one may wonder whether unbiased shear estimators in the conventional form ever exist . the answer is yes , at least when the psf is absent . for example , we find the following unbiased shear estimators : @xmath41 eq.([q7 ] ) can be checked by applying taylor expansion of @xmath42 $ ] ( @xmath43 ) to the first order in shear / convergence using eq.([q4 ] ) . eq.([q7 ] ) defines a special type of conventional shear estimators that are accidentally found by us . it is now immediately interesting to ask if there exist other types of unbiased shear estimators in the conventional form . we study this issue specifically in the next two sections . if the readers wish to go directly to the relevant sections on the new estimator , read on from [ alternatives ] . for notational convenience , we shall abbreviate `` conventional shear estimator '' as `` cse '' in the rest of the paper . once again , by cse we mean the shear estimators that are made of just one number measured from a galaxy image for each shear component . in preparation for our main theme of this section , we discuss the spin properties of cosmic shears and their estimators in [ coor_spin ] . we then study the forms of cses in the absence of the psf in [ spin_2_esti ] . to study the forms of the cses , it is useful to first consider their properties under coordinate rotations . suppose we rotate the coordinates @xmath1 clockwise by an angle @xmath44 . the new coordinates @xmath45 are related to the old one via the following relation : @xmath46 if we write the position vector as a complex number of the form @xmath47 , where @xmath48 is the complex unit , the coordinate transformation under rotation can then be written as : @xmath49 for notational brevity , we shall generally use @xmath50 to denote the value of any quantity @xmath51 in the new coordinates that are rotated clockwise by an angle @xmath44 with respect to the original coordinates . let us now discuss how cosmic shears and their cses transform under coordinate rotation . the definitions of the shear components ( shown in the beginning of [ classic ] ) involve spatial derivatives ; thus , their transformation rules under coordinate rotation can be found from the chain rule : @xmath52 from eq.([coor_trans ] ) , we get : @xmath53 therefore , we have : @xmath54 taking the square of eq.([coor_trans_complex_2 ] ) , we find : @xmath55\exp(\ima 2\theta)\end{aligned}\ ] ] therefore , the shear components , which are 2nd order derivatives of the lensing potential @xmath12 , also transform under coordinate rotation as : @xmath56 because of this property , we usually say that cosmic shears form a spin-2 quantity . in general , a complex quantity , say @xmath57 , is called a spin-@xmath58 quantity if it transforms as @xmath59 under a clockwise coordinate rotation of angle @xmath44 . now let us discuss shear estimators . it is straightforward to see that the cses defined in eq.([q7 ] ) do not form a spin-2 quantity . more generally , assuming that @xmath60 and @xmath61 are the cses for @xmath22 and @xmath23 , respectively , then , unlike @xmath62 , @xmath63 is not necessarily a spin-2 quantity . however , it turns out that we can regularize any cses by turning them into components of a spin-2 quantity . we have the following lemma : [ turn_spin_2 ] based on any pair of cses @xmath64 , one can build a new pair of cses @xmath65 to form a spin-2 quantity through the following procedure : @xmath66 firstly , under a clockwise coordinate rotation by angle @xmath67 , we have : @xmath68\left(\gamma_1^{\theta'}+\ima\gamma_2^{\theta'}\right)\\ \nonumber & = & \left(\gamma_1'+\ima\gamma_2'\right)\exp(\ima 2\theta_0)\end{aligned}\ ] ] therefore , @xmath69 form a spin-2 quantity . to show that @xmath65 are a pair of cses , let us take the ensemble average on both sides of eq . ( [ trans_spin_2_rot ] ) : @xmath70 since by definition , @xmath71 and @xmath72 measure the shear values in the rotated coordinates , we have , @xmath73 in a more compact form , we can write eq.([gamma_theta ] ) as : @xmath74 using eq . ( [ gamma_theta2 ] ) in eq . ( [ trans_ideal_esti ] ) , we get : @xmath75 which proves that @xmath65 are indeed also a pair of cses . due to the invariance under a coordinate rotation of angle @xmath76 , one can always decompose any shear estimator into components of integer spins . eq.([trans_spin_2 ] ) essentially defines a procedure of isolating the spin-2 components of any cses using fourier transformation . since the cosmic shears form a spin-2 quantity , only the spin-2 components of any cses are the relevant / principle components of the estimators . this point is further supported by the fact that the ensemble averages of spin-@xmath58 ( @xmath77 ) components of any cses are zero , , @xmath78 therefore , we only need to focus on spin-2 cses from here on . in the weak lensing limit , , when the cosmic shear parameters @xmath79 are small , any spin-2 cses @xmath64 can be taylor expanded to the first order in shear as follows : @xmath80 where @xmath81 ( @xmath51 is any quantity ) denotes the value of @xmath51 at @xmath82 . since @xmath83 , @xmath84 , @xmath85 , and @xmath86 are all spin-2 quantities , their ensemble average must vanish . on the other hand , the coefficients associated with @xmath22 and @xmath23 in eq.([taylor_weak_lensing ] ) can be decomposed into spin-0 and spin-4 components as follows : @xmath87 where @xmath88 as shown in appendix a , @xmath89 is a scalar , @xmath90 is a pseudo scalar , @xmath91 is a spin-4 quantity . the ensemble averages of @xmath92 and @xmath93 must vanish . the ensemble average of @xmath90 vanishes if galaxy images have parity symmetry along any direction in the plane of the sky statistically , which is assumed to be true in this paper . consequently , for @xmath64 to be spin-2 cses , we only require @xmath94 . this actually implies that for any individual galaxy , @xmath95 . the reason is that @xmath89 of any single galaxy does not change under coordinate rotation of random angles , and is equal to @xmath96 because the galaxies generated by rotations of a single galaxy form a complete set of statistically isotropic samples ( , there are no special directions ) . as a result , any spin-2 cses @xmath64 must satisfy the following necessary condition : @xmath97 this is also a sufficient condition , because @xmath95 directly implies that @xmath94 . in general , the cses are functions of a certain number of shape parameters ( , the multipole moments of an image ) . the functions can be very complicated , and are certainly not fixed by the requirement given by eq.([requirement ] ) . however , for galaxies whose shapes are described by only three parameters ( perfect ellipses ) , any cse should reduce to a function of just three variables . in this case , we find that any pair of spin-2 cses must reduce to a unique form , which is sufficient for us to judge whether cses are convenient in practice : namely , if we find that the resulting form is highly non - linear even for such a simple case , then it is reasonable to conclude that cses are not so useful for accurate shear measurements from more realistic galaxy shapes as well . this is shown in the rest of this section . for clarity , we refer the readers to appendix b for the mathematical details / proofs for some of the statements made hereafter in this section . let us consider a set of galaxies whose surface brightness profiles can be parametrized as @xmath98 with @xmath99 , where @xmath100 and @xmath101 are the coordinates , @xmath98 is a function of a fixed form , and @xmath102 are the three parameters determining galaxy shapes . for the images to be ellipses , we require the following three things : 1 . @xmath98 decays sufficiently fast when @xmath103 becomes large ; 2 . @xmath104 ; 3 . @xmath105 . for example , if @xmath106 ( @xmath107 is the step function ) and @xmath102 satisfy the above conditions , the galaxy surface brightness is then distributed evenly inside the ellipse defined by @xmath108 . when such images are weakly lensed , the three conditions are not violated , and @xmath102 becomes @xmath109 without changing the form of @xmath110 . in other words , weak lensing does not introduce additional degrees of freedom to the galaxy shapes . note that otherwise , one has to consider using more than 3 parameters to construct shear estimators . among the three parameters , there are indeed only two degrees of freedom useful for shear measurement : the ratios of the parameters . this is because the overall amplitudes of @xmath102 only change the galaxy size , not its shape . as shown in appendix b , the ellipticities @xmath111 defined in [ classic ] are directly equal to @xmath112 , therefore , we can write the shear estimators as functions of only @xmath20 and @xmath21 . we can further show that @xmath64 must take the following form : @xmath113 where @xmath114 , and @xmath115 is a one - variable complex function , whose form is to be determined later in this section . to see why eq.([g_1_g_2_form ] ) is true , one can use the taylor expansion to write @xmath60 and @xmath61 as power series of @xmath20 and @xmath21 . consequently , one can write @xmath63 as power series of @xmath116 and @xmath117 , whose spins are @xmath118 and @xmath119 respectively . since @xmath63 is a spin-2 quantity , each term in the power series must also be a spin-2 quantity . therefore , in each term of the power series , the power on @xmath116 must be larger than that on @xmath117 by exactly one , , each term must take the form of @xmath120 , where @xmath58 is any non - negative integer , and @xmath121 is a coefficient which can be any complex number at this point . as a result , the shear estimators must have the form defined in eq.([g_1_g_2_form ] ) . to find out what @xmath122 is , let us write it as @xmath123 . ( [ g_1_g_2_form ] ) then becomes : @xmath124 using the constraint in eq.([requirement ] ) , we find : @xmath125 it is interesting to note that eq.([constraint ] ) does not place any constraints on @xmath126 , , it can be any real function . this is because @xmath126 simply adds unnecessary even - parity terms into odd parity ones , and vice versa , without affecting the ensemble averages and the spin of the shear estimators . for convenience , we set @xmath127 hereafter . eq.([constraint ] ) is a typical first - order ordinary differential equation . it can be solved by introducing an integrating factor @xmath128 which satisfies : @xmath129\ ] ] multiplying both sides of eq.([constraint ] ) with @xmath128 , we get : @xmath130=k(u)\ ] ] it is now straightforward to solve both eq.([ku ] ) and eq.([g_1_u ] ) . the results are : @xmath131 where @xmath90 is a real number constant . to guarantee that @xmath60 and @xmath61 do not diverge when @xmath20 and @xmath21 approach zero , we need @xmath132 . finally , we find the unique form for the spin-2 cses : @xmath133 regarding the uniqueness , it is useful to note that if we transform the cses defined in eq.([q7 ] ) into spin-2 shear estimators using the procedure given in eq.([trans_spin_2 ] ) , we achieve the same shear estimators as those shown in eq.([g_1_g_2_form_final ] ) . we have shown that the principle components ( spin-2 ) of any pair of cses have to take specific and _ highly _ nonlinear forms for galaxies of elliptical shapes . this feature makes cses not convenient in practice ( , in the presence of noise ) . any cses which correct for the psf effect also have to reduce to the forms given in eq.([g_1_g_2_form_final ] ) in the limit of zero psf size when the galaxy images have pure elliptical shapes . for this reason , the conclusion in the previous section is already sufficient to argue against the usefulness of cses in practice . for academic interests , we provide the following arguments for why cses may not even exist in the presence of the psf : in the presence of the point spread function , structural details of galaxy images on scales smaller than the size of the psf are smeared out . this implies that there are only a finite number of shape parameters ( , multipole moments up to some order ) available for constructing shear estimators . on the other hand , the derivatives of the lower order shape parameters ( , lower order multipole moments ) with respect to the cosmic shears depend on the higher order shape parameters because of the psf , suggesting the requirement for an infinite number of shape parameters to form the shear estimators . combining the above two reasons , we find it unlikely to form cses when a psf is present . the mathematical details of the above statements are given in appendix c. as searching for optimal shear estimators is actively ongoing nowadays ( @xcite ) , it is important to realize that cses ( `` conventional shear estimators , '' by which we mean the shear estimators that are made of just one number from a galaxy image for each shear component ) are hard to use in practice due to their unavoidable complex forms even in the absence of the psf ( simpler forms , such as the quadrupole moments , are biased estimators , as shown in 2.1 ) . therefore , existing shear estimators of the conventional type must quantify the bias factor when estimating the shear , which can be achieved numerically ( see , , erben et al . 2001 , bacon et al . 2001 , or most recently , @xcite ) or estimated analytically ( , shear susceptibility in ksb [ kaiser et al . 1995 ] and derived methods , or responsivity factor in bernstein & jarvis 2002 and similar methods ) , although most people have been mainly focusing on the systematic errors caused by the photon noise and the psf . however , to achieve percent or even sub - percent level accuracy in cosmic shear measurements , it does not seem enough to completely rely on numerical tests using computer - generated galaxies of limited morphology richness , or approximate analytical methods . unfortunately , in the presence of psf , most of the existing shear measurement methods are too complicated or too model - dependent ( @xcite ) to allow for an accurate analytic analysis of the systematic errors in their shear estimators . the method of z08 ( see also @xcite for the treatment of photon noise and the pixelation effect ) is easily amenable to the corrections described in eq.([q5 ] ) , and can also account for the psf correction . not only is it simple , but also well defined regardless of the morphologies of galaxies and the psf . we show here how to properly use this method ( instead of using it as cses ) to recover the cosmic shear in an unbiased way . the basic idea of z08 is to use the spatial derivatives of the galaxy surface brightness field to measure the cosmic shears . it relies on the fact that gravitational lensing does not only distort the overall shape of the object , but also locally modifies the anisotropy of the gradient field of the surface brightness . as it allows for using the shape information from galaxy substructures , the method of z08 can potentially improve on the signal - to - noise ratio of the shear measurements . it is shown in z08 that the shear measurement should be carried out in the fourier space , in which any psf can be transformed into the desired isotropic gaussian form through multiplications , and the spatial derivatives of the surface brightness field can be easily measured . the cosmic shear can be estimated using the following relations : @xmath134 where @xmath135 @xmath136 is the scale radius of the isotropic gaussian psf @xmath137 , which is defined as : @xmath138 @xmath139 is the surface brightness field . @xmath140 denotes @xmath141 . as shown in appendix d , the method of z08 effectively utilizes the quadrupole moments in the fourier space to measure the cosmic shears . now , here is an important point : in order to implement this method , we must make it clear what we mean by the angular brackets in eq.([shear12psf ] ) . first , we need to measure the derivatives of the surface brightness and average them within a single galaxy . let us denote this averaging by @xmath142 , and write : @xmath143 of course , these are still extremely noisy as they use only one galaxy . the question is then , `` how do we average these quantities over many galaxies to obtain an unbiased estimator of the shears ? '' if one uses these quantities as if they were the cses , then one would simply average them over many galaxies . however , this will produce a biased estimator : ) used for the average was still too small . ] @xmath144 where @xmath145 and @xmath146 are the ensemble averages of functions of multipole moments of the galaxy images in fourier space , and @xmath147 denotes the ensemble average over many galaxies . the derivation of the forms of @xmath145 and @xmath146 is given in appendix d. @xmath145 and @xmath146 are generally nonzero and dependent on the galaxy morphology . instead , we need to take the ensemble averages of the numerator and the denominator separately first , and then divide them to obtain an unbiased estimator : @xmath148 this is the main result of this paper , and the form of the unbiased estimator that we propose to use for the actual analysis of the weak lensing data . of course , one could divide the left hand sides of eq.([shear12psf_wrong ] ) by @xmath149 and @xmath150 to obtain an unbiased estimator . this is similar to correcting the measured shear for a multiplicative bias that is evaluated from the same ensemble of galaxies . in this sense , eq.([shear12psf_wrong ] ) provides the exact definitions for the multiplicative biases for @xmath22 and @xmath23 . however , since @xmath145 and @xmath146 in eq.([shear12psf_wrong ] ) involve many high order fourier - space multipole moments of the surface brightness field , evaluation of these terms from simulations ( which are incomplete anyway ) can be highly uncertain . even worse , the multiplicative bias mentioned here is not even a constant , but depends on the morphological distribution of the galaxies . this makes the conventional way of measuring shear correlation functions even more challenging , as one must take into account the _ correlations _ of the multiplicative biases , as will be shown in [ statistics ] . in summary , according to eq.([shear12psf2 ] ) , for each shear component , two quantities from each galaxy should be kept , and the ratios of their ensemble averages yield unbiased estimates for the corresponding shear components . finally , it is important to note that , to efficiently use eq.([shear12psf2 ] ) , the surface brightness of each participating galaxy should be normalized to have roughly the same maximum value , so that faint galaxies are not much less weighted than their brighter counterparts . the details regarding the optimal weighting scheme as a function of the galaxy luminosity should also take into account the photon noise . this is a separate topic , and will be studied in a future work . strictly speaking , eq.([shear12psf2 ] ) holds when we average over an infinite number of galaxies ; however , as we shall show in this section , the error that we make by having a finite number of galaxies for averaging is much smaller than the statistical errors , and thus the estimator remains unbiased for practical applications . for simplicity , we use eq.([q6 ] ) rather than our main equation [ eq.([shear12psf2 ] ) ] in the following discussion , but the conclusion will be the same for eq.([shear12psf2 ] ) . let us use @xmath151 to denote the average of the quantity @xmath51 over @xmath152 galaxies . from eq.([q3 ] ) , we get : @xmath153 consequently , we have : @xmath154 where @xmath155 here , the terms @xmath156 and @xmath157 contribute to random errors because their ensemble averages vanish , whereas the terms @xmath158 and @xmath159 lead to systematic biases because their ensemble averages do not vanish . fortunately , as @xmath160 scales as @xmath161 and @xmath162 , the amplitudes of such systematic biases are always much smaller than the sizes of the statistical errors . therefore , the results from this new type of shear estimators may be regarded as unbiased for practical applications . numerical verifications will be given in [ examples ] . the cosmic shear field can only be probed statistically . this is mainly due to the intrinsic variations of the galaxy shapes and the spatial fluctuations of the shear components . as a result , the shear statistics is usually studied in the form of n - point spatial correlation functions of the shear field . the previous discussions and measurements in the literature are based on `` conventional '' shear estimators ( cses ) , , one often assumes that the following is true : @xmath163 where @xmath164 can be either @xmath22 or @xmath23 , and @xmath165 is a cse for @xmath164 . for individual galaxies , eq.([assume_wr ] ) implies : @xmath166 where @xmath167 satisfies @xmath168 . ] . it is usually assumed that @xmath167 s of different galaxies do not correlate with each other s belong to the topic of `` galaxy intrinsic alignment '' , which is beyond the scope of this paper . ] . therefore , the n - point correlation functions of the shear field can be directly measured by the correlations of @xmath165 s . however , in [ no_psf ] , we have shown that such a @xmath165 at least does not exist in a convenient form . instead , as proposed in [ m_zhang08 ] , we can use the new form of shear estimators defined in eq.([shear12psf2 ] ) to probe the cosmic shear in an unbiased way . let us now find out how to measure the n - point shear correlation functions with the new form of shear estimators . numerical examples are given in [ examples ] . for notational convenience , the type of shear measurement in eq.([shear12psf2 ] ) can be symbolized as follows : @xmath169 where @xmath164 can be either @xmath22 or @xmath23 , and @xmath89 and @xmath170 are properties of a galaxy , such as those defined in eq.([shear12psf2 ] ) . similar to eq.([assume_wr2 ] ) , eq.([symbol ] ) implies the following : @xmath171 where @xmath90 satisfies @xmath172 . if we assume that the @xmath90 of any galaxy does not correlate with the @xmath170 s and @xmath90 s of other galaxies , the n - point correlation functions of the shear field can be probed using the following relation : @xmath173 the ensemble averages are taken over a large number of galaxies whose relative positions @xmath174 ( @xmath175 ) are fixed . in practice , the n - point shear correlation functions can be measured using : @xmath176 where the sum is taken over all the galaxy groups that satisfy the positional constraints . note that _ the ratio is taken after the summations_. the standard deviation ( @xmath177 ) of the correlation function in such a measurement can be calculated as follows : @xmath178 ^ 2\right\rangle_{en}\\ \nonumber & \doteq&\left\langle\frac{\sum a^2(\vec{x}_{i_1})a^2(\vec{x}_{i_2})\ldots a^2(\vec{x}_{i_n})}{\left[\sum b(\vec{x}_{i_1})b(\vec{x}_{i_2})\ldots b(\vec{x}_{i_n})\right]^2}\right\rangle_{en}\\ \nonumber & = & \frac{n\left\langle a^2(\vec{x}_1)a^2(\vec{x}_2)\ldots a^2(\vec{x}_n)\right\rangle_{en}}{n^2\left\langle b(\vec{x}_1)b(\vec{x}_2)\ldots b(\vec{x}_n)\right\rangle_{en}^2}\\ \nonumber & = & \frac{1}{n}\left(\frac{\langle a^2\rangle_{en}}{\langle b\rangle_{en}^2}\right)^n\end{aligned}\ ] ] where @xmath152 is the total number of galaxy groups ( , the number of galaxy pairs for 2-point correlations ) used . to summarize , in the new type of shear measurement , the shear correlation function should be measured using the ratio of two ensemble averages , as shown in eq.([symbol3 ] ) . if @xmath170 in eq.([symbol ] ) is viewed as a multiplicative bias , we need to measure the _ correlations _ of these multiplicative biases as well in order to get the correct shear correlation functions . in this section , we show how accurately one can recover the cosmic shears and their 2-point correlation functions with the method proposed in [ alternatives ] and [ statistics ] . for a comparison , we also show the results using the method of z08 in the `` conventional '' ( but wrong ) way , , eq.([shear12psf_wrong ] ) , but without taking into account the biases , @xmath145 and @xmath146 . ) in the numerical tests . here , we use @xmath179 galaxies ( or pairs of galaxies for the 2-point correlation function ) for each test . ] since our focus is to demonstrate principles , we do not include photon noise or the pixelation effect in this paper , and we approximate the psf as an isotropic gaussian . including these effects is straightforward . ( note also that the conventional estimators yield biased results even in this idealized case . ) more comprehensive tests of the method of z08 have been presented in @xcite , which further improves the accuracy of shear measurement by including the second order shear / convergence terms . as shown in that paper , the second order correction is proportional to the convergence @xmath27 . we simply set @xmath180 in the numerical simulations here . the residual systematic error on the recovered shear ( @xmath22 or @xmath23 ) therefore should be expected to have a magnitude comparable to the third order terms in shear ( , @xmath181 , @xmath182 , @xmath183 , @xmath184 ) . the errors shown in the rest of this section are all at the @xmath185 confidence level . the mock galaxy images we use in our numerical tests are generated by the algorithms introduced in z08 and zhang ( 2010a ) , , each galaxy is generated as a collection of point sources . the reason is simple : one can accurately and easily mimic the lensing effect by displacing the points . it also allows us to generate galaxies of complex morphologies . there are two types of galaxies we use in this paper : 1 . randomly oriented regular galaxies , each of which contains an exponential disk in the galactic plane ( no bulge ) ; 2 . irregular galaxies being made of points generated by the trajectories of 2d random walks . for simplicity , the psf is always an isotropic gaussian function , whose scale radius is four times the grid size to avoid the pixelation problem . all the lengths in our simulations are in units of the grid size in the rest of this section . the dimension of the grid is @xmath186 . as our first example , we study how accurately a single input cosmic shear can be recovered by a large number of mock galaxies , , the 1-point statistics . we use the regular type mock galaxies as introduced in [ image ] . each disk galaxy is composed of ten point sources which are randomly distributed within a radius of 7 . the intensity of a point is an exponentially decaying function of its distance to the center of the disk with a decay length equal to 7 . the galactic disk is then projected onto the source plane in a random direction . for each input shear value , we use @xmath179 mock galaxies to recover the shear . to quantify the accuracy of shear recovery , we adopt the standard technique in the weak lensing community by using the `` multiplicative bias '' @xmath187 and the `` additive bias '' @xmath188 , which are defined as : @xmath189 our simulations use six sets of input shear values ( @xmath22 , @xmath23 ) . they are : ( @xmath190 , @xmath191 ) , ( @xmath192 , @xmath193 ) , ( @xmath194 , @xmath195 ) , ( @xmath195 , @xmath194 ) , ( @xmath193 , @xmath192 ) , ( @xmath191 , @xmath190 ) . the recovered shear values as well as the linear fitting results for @xmath22 and @xmath23 are shown in table [ one_p_gamma1 ] and table [ one_p_gamma2 ] respectively . note that we also list the values of @xmath196 s and @xmath197 s for the goodness of linear fitting ( see @xcite for details ) . for a comparison , we show in the last column of each table the quality of the shear recoveries using the `` conventional '' ( but wrong ) way , given by eq.([shear12psf_wrong ] ) . the tables show that the shear recovery can be very accurate if we use the method of z08 in the proper way , , eq.([shear12psf2 ] ) . on the other hand , if z08 is used in the `` conventional '' way , , eq.([shear12psf_wrong ] ) , we find nonzero multiplicative biases ( @xmath198 , @xmath199 ) for both @xmath22 and @xmath23 . the additive biases @xmath200 and @xmath201 are consistent with being zero . these results agree with our conclusions in [ m_zhang08 ] . c|c|c input @xmath22 & @xmath22 measured in & @xmath22 measured in + & the proper way & the `` conventional '' way + 0.05 & [email protected] & [email protected] + 0.03 & [email protected] & [email protected] + 0.01 & [email protected] & [email protected] + @xmath195 & @xmath2030.00008 & @xmath2040.00009 + @xmath193 & @xmath2050.00008 & @xmath2060.00009 + @xmath191 & @xmath2070.00008 & @xmath2080.00009 + & & + & & + linear fitting & using @xmath22 s from & using @xmath22 s from + results & the proper way & the `` conventional '' way + @xmath198 & @xmath209 & @xmath210 + @xmath200 & @xmath211 & @xmath212 + ( @xmath213 , @xmath214 ) & ( 0.58 , 0.96 ) & ( 0.58 , 0.96 ) + c|c|c input @xmath23 & @xmath23 measured in & @xmath23 measured in + & the proper way & the `` conventional '' way + 0.05 & [email protected] & [email protected] + 0.03 & [email protected] & [email protected] + 0.01 & [email protected] & [email protected] + @xmath195 & @xmath2150.00008 & @xmath2160.00009 + @xmath193 & @xmath2170.00008 & @xmath2060.00009 + @xmath191 & @xmath2180.00008 & @xmath2190.00009 + & & + & & + linear fitting & using @xmath23 s from & using @xmath23 s from + results & the proper way & the `` conventional '' way + @xmath199 & @xmath220 & @xmath221 + @xmath201 & @xmath222 & @xmath223 + ( @xmath224 , @xmath225 ) & ( 3.3 , 0.51 ) & ( 3.4 , 0.50 ) + let us now test the recovery of the 2-point shear correlations using the method of z08 in the way that is proposed in [ statistics ] . in each test , we use two large groups of galaxies that are given ( @xmath22 , @xmath23 ) and ( @xmath226 , @xmath227 ) , respectively . each group of each test contains @xmath179 galaxies . the values of @xmath22 , @xmath23 , @xmath226 , and @xmath227 vary from galaxy to galaxy . they are assumed to be normally distributed with the following covariance matrix : @xmath228 , @xmath229 , and @xmath230 and @xmath231 are to be specified in each test . the purpose of the test is to find out how accurately the 2-point correlations @xmath230 and @xmath231 can be recovered . for the tests presented here , we use the irregular type of mock galaxies , each of which is made of ten point sources generated by the 2d random walks . each step size of the random walks is a random number between 0 and 2 . the radius of each galaxy is limited to be less than 7 . in table [ two_p_gamma1 ] and [ two_p_gamma2 ] , we report the results of six tests with six different sets of ( @xmath230 , @xmath231 ) : ( 0.001,0.001 ) , ( 0.0006 , 0.0006 ) , ( 0.0002 , 0.0002 ) , ( @xmath232 , @xmath232 ) , ( @xmath233 , @xmath233 ) , ( @xmath234 , @xmath234 ) . as in [ one_point ] , to characterize the accuracy of the method , we again use the multiplicative biases ( @xmath235 , @xmath236 ) and additive biases ( @xmath237 , @xmath238 ) that are defined as follows : @xmath239 we also show the results from the `` conventional '' way of using z08 for a comparison . our results again show no systematic errors for the proper way of using z08 . in contrast , the `` conventional '' way tends to underestimate the amplitudes of the shear correlations ( negative multiplicative bias ) . the signs of the multiplicative biases in the `` conventional '' cases are opposite to those found in [ one_point ] . this is because we have used two different types of mock galaxies . c|c|c input @xmath230 & @xmath230 measured in & @xmath230 measured in + & the proper way & the `` conventional '' way + @xmath240 & @xmath241 & @xmath242 + @xmath243 & @xmath244 & @xmath245 + @xmath246 & @xmath247 & @xmath248 + @xmath249 & @xmath250 & @xmath251 + @xmath252 & @xmath253 & @xmath254 + @xmath255 & @xmath256 & @xmath257 + & & + & & + linear fitting & using @xmath230 s from & using @xmath230 s from + results & the proper way & the `` conventional '' way + @xmath235 & [email protected] & [email protected] + @xmath237 & @xmath258 & @xmath259 + ( @xmath260 , @xmath261 ) & ( 2.5 , 0.64 ) & ( 2.3 , 0.68 ) + c|c|c input @xmath231 & @xmath231 measured in & @xmath231 measured in + & the proper way & the `` conventional '' way + @xmath240 & @xmath241 & @xmath242 + @xmath243 & @xmath262 & @xmath263 + @xmath246 & @xmath264 & @xmath265 + @xmath249 & @xmath266 & @xmath267 + @xmath252 & @xmath268 & @xmath269 + @xmath255 & @xmath270 & @xmath271 + & & + & & + linear fitting & using @xmath231 s from & using @xmath231 s from + results & the proper way & the `` conventional '' way + @xmath236 & [email protected] & [email protected] + @xmath238 & @xmath272 & @xmath273 + ( @xmath274 , @xmath275 ) & ( 1.8 , 0.77 ) & ( 1.6 , 0.82 ) + conventionally , in the studies of weak lensing , for each shear component ( @xmath22 or @xmath23 ) , one hopes to construct a single quantity from each background galaxy image , whose ensemble average is equal to the true value of a component of the true shear field . we have shown that such conventional shear estimators ( cses ) _ do not exist in convenient forms _ even in the absence of the psf . based on the method of zhang ( 2008 ) , we have proposed to measure the cosmic shear in a new way : using the ratio of the ensemble averages of two galaxy properties to estimate each shear component . ( also see 9.2 of weinberg ( 2008 ) for a similar study . ) we have shown that , using both analytic analyses and numerical examples , the new way of estimating cosmic shears is unbiased , and does not contain systematic errors to the first order in shear at least . the new type of shear measurement demands shear statistics such as n - point correlation functions to be measured in an unconventional way as well , but with little additional cost . the author acknowledges gary bernstein , toshifumi futamase , yi mao , pengjie zhang for helpful discussions , and shanghai astronomical observatory ( shao ) , national astronomical observatories of china ( naoc ) for their hospitality . in particular , jz would like to thank gary bernstein for pointing out a mistake in the proof for the nonexistence of cses in the presence of psf in the previous version of this manuscript . jz is currently supported by the tcc fellowship of texas cosmology center of the university of texas at austin , and was previously supported by the tac fellowship of the theoretical astrophysics center of uc berkeley , where part of this work was done . ek is supported in part by nsf grants ast-0807649 and phy-0758153 and nasa grant nnx08al43 g . bacon d. , refregier a. , clowe d. , ellis r. , 2001 , mnras , 325 , 1065 bernstein g. , 2010 , mnras , 406 , 2793 bernstein g. & jarvis m. , 2002 , aj , 123 , 583 bonnet h. & mellier y. , 1995 , a&a , 303 , 331 bridle s. , gull s. , bardeau s. , kneib j. , 2001 , in scientific n. w. , ed . , proceedings of the yale cosmology workshop bridle s. et al . , 2009 , annals of applied statistics , vol.3 , no.1 , 6 - 37 , arxiv : 0802.1214 bridle s. et al . , 2010 , mnras , 405 , 2044 erben t. , van waerbeke l. , bertin e. , mellier y. , schneider p. , 2001 , a&a , 366 , 717 heymans c. et al . , 2006 , mnras , 368 , 1323 hoekstra h. , franx m. , kuijken k. , squires g. , 1998 , apj , 504 , 636 hoekstra h. & jain b. , 2008 , annual review of nuclear and particle science , 58 , 99 , arxiv:0805.0139 kaiser n. , 2000 , apj , 537 , 555 kaiser n. , squires g. & broadhurst t. , 1995 , apj , 449 , 460 luppino g. & kaiser n. , 1997 , apj , 475 , 20 massey r. et al . , 2007 , mnras , 376 , 13 massey r. & refregier a. , 2005 , mnras , 363 , 197 nakajima r. & bernstein g. , 2007 , aj , 133 , 1763 press w. , flannery b. , teukolsky s. , vetterling w. , 1992 , _ numerical recipes _ , cambridge univ . press , 2nd ed . refregier a. , 2003 , mnras , 338 , 35 refregier a. & bacon d. , 2003 , mnras , 338 , 48 rhodes j. , refregier a. & groth e. , 2000 , apj , 536 , 79 tyson j. , wenk r. & valdes f. , 1990 , apjl , 349 , l1 voigt l. & bridle s. , 2010 , mnras , 404 , 458 weinberg s. , 2008 , _ cosmology _ , oxford university press , 1st ed . zhang j. , 2008 , mnras , 383 , 113 zhang j. , 2010a , mnras , 403 , 673 zhang j. , 2010b , submitted to mnras , arxiv:1002.3614 under a clockwise coordinate rotation of angle @xmath44 , the cosmic shear components transform according to the following rule : @xmath276 using the chain rule , we then find : @xmath277 on the other hand , we know that the spin-2 shear estimators @xmath64 transform as : @xmath278 apply eq.([shear_deriv_rotation ] ) onto eq . ( [ estimator_transform ] ) , we get : @xmath279\exp(\ima 4\theta)\end{aligned}\ ] ] eq.([spin-4 ] ) clearly shows that the derivatives @xmath280 form a spin-4 quantity . we can also apply eq.([shear_deriv_rotation ] ) onto the complex conjugate of eq . ( [ estimator_transform ] ) , and get : @xmath281 therefore , @xmath282 is a scalar , and @xmath283 is a pseudo - scalar ( it has odd parity ) . the surface brightness profiles of the galaxies used in [ spin_2_esti ] are parametrized as @xmath98 , where @xmath99 , and @xmath100 and @xmath101 are the coordinate variables . curves of constant values of @xmath103 can be ellipses , hyperbolas , parabolas , or lines depending on the values of @xmath284 , @xmath285 , and @xmath286 . for our purposes , we only need ellipses . this requires @xmath104 and @xmath105 according to the matrix theory . since linear coordinate transformations do not spoil these relations , weakly lensed images of these galaxies are still ellipses . for the same reason , one can also easily show that the lensed images are still parametrized by the same function @xmath98 with @xmath287 , where @xmath288 , @xmath289 , and @xmath290 . let us now show that the ellipticity parameters @xmath111 for galaxies of the form @xmath291 $ ] are equal to @xmath112 . according to the definitions in [ classic ] , we have : @xmath292\ ] ] this integration can be carried out using the following linear coordinate transformation : @xmath293 where @xmath294 using @xmath295 defined in eq.([ct_e1e2 ] ) to replace @xmath1 in eq.([e1e2 ] ) , we get : @xmath296f_s(x'^2+y'^2)\\ \nonumber & & q_{11}+q_{22}\\ \nonumber & = & \frac{\beta^2-\alpha^2}{\sqrt{a^2-b^2}}\int d^2\vec{x'}\left[-\frac{b(\beta^2-\alpha^2)}{a^2-b^2}(x'^2-y'^2)\right.\\ \nonumber & + & \left.\frac{a(\alpha^2+\beta^2)}{a^2-b^2}(x'^2+y'^2)+\frac{4a\alpha\beta}{a^2-b^2}x'y'\right]f_s(x'^2+y'^2)\\ \nonumber & & 2q_{12}\\ \nonumber & = & \frac{\beta^2-\alpha^2}{\sqrt{a^2-b^2}}\int d^2\vec{x'}\left[\frac{2\alpha\beta}{\sqrt{a^2-b^2}}(x'^2+y'^2)\right.\\ \nonumber & + & \left.\frac{2(\alpha^2+\beta^2)}{\sqrt{a^2-b^2}}x'y'\right]f_s(x'^2+y'^2)\end{aligned}\ ] ] given that the function @xmath110 only depends on @xmath297 , we have : @xmath298 it is now straightforward to calculate @xmath111 : @xmath299 for technical convenience , let us work in fourier space . according to the notations in [ classic ] , we use @xmath300 and @xmath301 to denote the fourier transformations of the lensed galaxy image @xmath3 and the original galaxy image @xmath4 respectively . their relations are given by the following equations : @xmath302 using the relations defined in eq.([fifstits ] ) , we get : @xmath303 eq . ( [ fourier2 ] ) simply means that under lensing , the fourier transformation of the galaxy image is changed from @xmath304 to @xmath305 . due to the presence of the psf , the fourier transformation of the observed image @xmath306 is related to that of the lensed image @xmath307 via : @xmath308 where @xmath309 is the fourier transformation of the psf . without loss of generality , in the rest of our discussion , we use the isotropic gaussian psf , , @xmath310 . the advantage of working in fourier space is that the psf is included as a multiplicative factor , rather than a convolution as in real space . combining eq.([fourier2 ] ) and eq.([f_o ] ) , we get : @xmath311 since the psf profile in fourier space typically falls off quickly when the wave number exceeds the inverse of the size of the psf , it strongly suppresses the power of the observed images on small scales . therefore , only a finite number of fourier modes are available for providing shape information . recovering information on arbitrarily small scales is never feasible in practice due to noise and numerical problems . to form cses , let us use the multipole moments of galaxy images to represent the shape information , which are defined as : @xmath312 where @xmath313 and @xmath314 are non - negative integers . note that due to the finite degrees of freedom of the shape information , one can equivalently choose other basis ( , shapelets ) to study the same issue without affecting the conclusion . for simplicity but without loss of generality , let us only consider galaxies that are invariant under the parity transformation @xmath315 . note that weak lensing does not change this property . for this type of galaxies , the imaginary part of @xmath316 is always zero , and @xmath317 s are real . furthermore , @xmath317 is zero when @xmath318 is an odd number . in this case , the shear estimators @xmath60 and @xmath61 can be written as functions of the @xmath317 with @xmath318 being even numbers only . for some of the lowest order @xmath317 s , we can find out how they transform under lensing using eq.([f_o2 ] ) : @xmath319 the last step in the above equation is achieved by redefining @xmath320 as @xmath321 . by keeping the terms in matrix @xmath13 up to the first order in shear , one can straightforwardly show the following : @xmath322\\ \nonumber \\ \nonumber & & m_{11}\\ \nonumber & = & ( 1 - 2\kappa)m_{11}^s-\gamma_2\left(m_{20}^s+m_{02}^s-2\beta^2 m_{22}^s\right)\\ \nonumber & + & \kappa\beta^2\left(m_{31}^s+m_{13}^s\right)+\gamma_1\beta^2\left(m_{31}^s - m_{13}^s\right)\\ \nonumber \\ \nonumber & & m_{20}+m_{02}\\ \nonumber & = & ( 1 - 2\kappa)(m_{20}^s+m_{02}^s)+\kappa\beta^2\left(m_{40}^s+2m_{22}^s+m_{04}^s\right)\\ \nonumber & -&\gamma_1\left[2(m_{20}^s - m_{02}^s)-\beta^2\left(m_{40}^s - m_{04}^s\right)\right]\\ \nonumber & -&\gamma_2\left[4m_{11}^s-2\beta^2\left(m_{31}^s+m_{13}^s\right)\right]\end{aligned}\ ] ] where @xmath323 the conventional cosmic shear estimators ( @xmath324 ) are functions of @xmath317 . according to our discussion in [ spin_2_esti ] , the two functions have to satisfy the following relation : @xmath325 in the presence of psf , we are unable to find out whether one can find cses that satisfy the most general requirement given in eq.([requirement_again ] ) . however , we can show that there do not exist cses satisfying a slightly stronger condition : @xmath326 in addition to eq . ( [ requirement_again ] ) , eq.([requirement_stronger ] ) simply imposes another requirement that the spin-4 parts of the derivatives of the shear estimators with respect to the shears are zero . we can rewrite eq.([requirement_stronger ] ) for , , @xmath60 , using the chain rule as : @xmath327 since there are only a finite number of multipole moments available for constructing the shear estimators , we assume the maximum value of @xmath318 is @xmath152 ( n is an even integer ) , , we have : @xmath328 because the right side of eq.([gamma_1_chain_rule2 ] ) is evaluated at @xmath82 , both @xmath329 and @xmath330 are functions of only @xmath331 s , , the multipole moments in the absence of lensing . as in eq.([mij_trans ] ) , one can show in general that @xmath330 involves higher order multipole moments , , @xmath332 s with @xmath333 . in particular , when @xmath334 , @xmath330 depends linearly on @xmath332 s with @xmath335 . to satisfy the constraint in eq.([gamma_1_chain_rule2 ] ) , the coefficients in front of the terms proportional to @xmath332 s ( @xmath335 ) must vanish because they are independent of the multipole moments with @xmath336 . as a result , we find that @xmath329 has to vanish when @xmath334 . in other words , we have : @xmath337 we can now recursively use the above reasoning to show that there does not exist an @xmath152 that can satisfy eq.([gamma_1_chain_rule2 ] ) . therefore , we can never find cses of the type defined in eq.([requirement_stronger ] ) . though , one may still expect to find cses based on the most general requirement defined in eq.([requirement_again ] ) . however , even in this case , cses _ must _ reduce to highly nonlinear forms for galaxies of pure elliptical shapes when the psf effect is small , as we have shown in [ spin_2_esti ] . this feature is already sufficient for arguing against their usefulness in practice . let us now calculate the terms @xmath145 and @xmath146 defined in eq.([shear12psf_wrong ] ) . the averages of the spatial derivatives of the surface brightness field of a single galaxy can be related to the fourier modes of the image . the fourier transformation has been defined in eq.([fourier ] ) in appendix c. following the notations of appendix c , we find : @xmath338 and @xmath339 where @xmath340 is the total area of the map containing the galaxy . similarly , one can derive the following relation : @xmath341 eq.([rf_derivative_aves ] ) and eq.([rf_derive2 ] ) allow us to transform eq.([shear12psf_wrong ] ) into its version in fourier space : @xmath342 where @xmath343 note that @xmath344 . it is now clear that the method of z08 basically utilizes the quadrupole moments in the fourier space to measure the cosmic shear . using eq.([f_o2 ] ) , we can find out how @xmath345 transform under lensing : @xmath346 the last step in the above equation is achieved by redefining @xmath320 as @xmath321 . for the isotropic gaussian psf , @xmath347 . by keeping the terms in matrix @xmath13 up to the first order in shear , one can straightforwardly show the following : @xmath348\\ \nonumber \\ \nonumber & & p_{11}\\ \nonumber & = & p_{11}^s-\gamma_2\left(p_{20}^s+p_{02}^s-4\beta^2 p_{22}^s\right)+2\kappa\beta^2\left(p_{31}^s+p_{13}^s\right)\\ \nonumber & + & 2\gamma_1\beta^2\left(p_{31}^s - p_{13}^s\right)\\ \nonumber \\ \nonumber & & p_{20}+p_{02}\\ \nonumber & = & p_{20}^s+p_{02}^s-2\gamma_1\left[p_{20}^s - p_{02}^s-\beta^2\left(p_{40}^s - p_{04}^s\right)\right]\\ \nonumber & -&4\gamma_2\left[p_{11}^s-\beta^2\left(p_{31}^s+p_{13}^s\right)\right]+2\kappa\beta^2d_4^s\\ \nonumber \\ \nonumber & & d_4\\ \nonumber & = & d_4^s-2\kappa\left(d_4^s-\beta^2d_6^s\right)\\ \nonumber & -&2\gamma_1\left[2\left(p_{40}^s - p_{04}^s\right)-\beta^2\left(p_{60}^s+p_{42}^s - p_{24}^s - p_{06}^s\right)\right]\\ \nonumber & -&4\gamma_2\left[2\left(p_{31}^s+p_{13}^s\right)-\beta^2\left(p_{51}^s+2p_{33}^s+p_{15}^s\right)\right]\\ \nonumber\end{aligned}\ ] ] where @xmath349 by keeping the terms up to the first order in shear , and using the fact that the ensemble averages are taken over statistically isotropic galaxy samples , one can now directly find expressions for @xmath145 and @xmath146 of eq.([shear12psf_wrong ] ) : @xmath350

isolated swift gamma - ray burst ( grb ) pulses , like their higher - energy batse counterparts , emit the bulk of their pulsed emission as a hard - to - soft component that can be fitted by the empirical pulse model . this signal is overlaid by a fainter , three - peaked signal that can be modeled by an empirical wave - like function : the two fits combine to reproduce grb pulses with distinctive three - peaked shapes . the precursor peak appears on or before the pulse rise and is often the hardest component , the central peak is the brightest , and the decay peak converts exponentially decaying emission into a long , soft , power - law tail . accounting for systematic instrumental differences , the general characteristics of the fitted pulses are remarkably similar . isolated grb pulses are dominated by hard - to - soft evolution ; this is more pronounced for asymmetric pulses than for symmetric ones . isolated grb pulses can also exhibit intensity tracking behaviors that , when observed , are tied to the timing of the three peaks : pulses with the largest maximum hardnesses are hardest during the precursor , those with smaller maximum hardnesses are hardest during the central peak , and all pulses can re - harden during the central peak and/or during the decay peak . since these behaviors are essentially seen in all isolated pulses , the distinction between `` hard - to - soft '' and `` intensity - tracking '' pulses really no longer applies . additionally , the triple - peaked nature of isolated grb pulses seems to indicate that energy is injected on three separate occasions during the pulse duration : theoretical pulse models need to account for this .

1511.03695

gamma - ray burst ( grb ) pulse characteristics remain poorly understood even though pulses are the most common structures in grb light curves : there is little agreement concerning the physical mechanisms responsible for producing them . pulse characteristics are difficult to extract due to a mixture of strong spectral - temporal evolution , instrumental and sampling biases , confusion caused by overlapping pulses , and low signal - to - noise measurements . understanding pulse behaviors is important , as it is difficult to adequately model grb progenitors and outflow without first understanding the mechanism by which pulses liberate energy , and it is difficult to understand pulse mechanisms without first identifying pulse behaviors . data driven analysis approaches are providing new insights into grb pulse characteristics . these approaches have both advantages and disadvantages over standard theoretical modeling techniques : they can identify behaviors ( typically clustering and/or correlations ) that theoretical models are not trying to explain and thus may not recognize , but by asking questions about pulse behaviors they do not necessarily couch these questions in the form of specific theoretical models . data driven approaches have allowed the identification of important pulse characteristics even as theoretical models have been unable to explain these behaviors . as an example , consider empirical light curve models , which can be used to provide an accurate representation of grb pulse behaviors in the absence of identified pulse physics . several empirical models are able to adequately explain the general shapes of grb pulse light curves ( e.g. see the introduction in @xcite ) . among these , the simple model of @xcite is useful because it has only four free parameters which allow reasonable fits over the wide range of observed pulse characteristics . empirical approaches demonstrate that grb pulse light curves can be generally represented by a _ hard - to - soft _ spectral evolution . pulses are generally hardest at the first instant they appear , and they continue to soften during the pulse rise , past the pulse peak , and through the pulse decay @xcite . although hard - to - soft evolution is the most common pulse evolution ( e.g. @xcite ) , _ intensity tracking _ pulse evolution has been identified in a smaller number of pulses ( e.g. @xcite ) . it is difficult to estimate what fraction of pulses exhibit intensity tracking behaviors , since there are many cases where overlapping hard - to - soft pulses appear to have been misclassified as intensity tracking @xcite . hard - to - soft spectral evolution is so central to grb pulse structure that it can be used as a defining characteristic in rigorous statistical modeling techniques @xcite . the harder a pulse is , the more pronounced the evolution typically is . a small percentage of pulses appear to harden during the pulse rise , but these _ intensity tracking _ pulses are typically among the very softest , suggesting that instrumental biases might be in part responsible for the initial soft emission . several of the well - known correlated spectral - temporal grb pulse properties are simply a result of this hard - to - soft spectral decay : rapid hardness evolution produces short duration pulses with short spectral lags while slow hardness evolution produces long duration ones with long lags . the relationship between pulse asymmetry and hard - to - soft evolution is not as clear as the one between duration , lag , and evolution . asymmetry is anti - correlated with hardness and correlated with duration across a large batse grb pulse sample such that asymmetric pulses are generally longer and softer than symmetric pulses @xcite . this implies that asymmetric pulses in general do not undergo a significant hardness evolution , since the softening is mild and occurs over a long time interval . however , some of the faintest , softest , longest grb pulses are symmetric and undergo very little hard - to - soft evolution , while some asymmetric pulses are quite hard . to further confuse matters , the bright , short pulses in classical short grbs are also symmetric yet very hard ; undergoing a rapid hard - to - soft evolution . without an obvious connection between hard - to - soft evolution and pulse asymmetry , the possibility exists that asymmetry and hard - to - soft evolution are independent or semi - independent pulse characteristics . grb pulses ( at least those belonging to the long and intermediate grb classes ) exhibit additional intensity fluctuations that overlay the hard - to - soft monotonic pulse component @xcite ; these fluctuations can be seen in the residuals obtained by subtracting the @xcite modeled monotonic pulse shape from the light curve . these fluctuations , which have the appearance of three peaks separated by two valleys , are more easily observed in high signal - to - noise pulses ; this makes bright pulses look like they have variable , multi - peaked light curves , in contrast to the smooth monotonic light curves of fainter pulses . when combined with spectral softening , bright pulses appear to be composed of at least three peaks , with the first one ( during or immediately preceding the pulse rise ) generally being the hardest , the second ( and brightest ) one being of intermediate hardness , and the last one ( during the long decay ) being soft . hard - to - soft evolution and phased non - monotonic intensity fluctuations can be used to help determine if a grb emission episode is composed of a single pulse or of several merged pulses . a single pulse , at least in the energy ranges of batse and fermi , has pulse residual parameters that correlate with the pulse asymmetry . the residual light curve waves can be fitted by a simple four - parameter empirical model @xcite this empirical model has several interesting features : 1 ) the basis of the model is a truncated bessel function of the first kind , @xmath0 that begins at the true pulse peak , 2 ) the wave can be modeled by a mirror image of itself in time , with the time - reflected part occurring during the pulse rise and the time - forward part occurring during the pulse decay , 3 ) the pulse decay portion is a time - stretched version of the mirrored part on the pulse rise , and 4 ) most if not all of the pulse residual fit parameters correlate with the pulse asymmetry . at the present time , no physical mechanism is known that can provide these observed model characteristics . it is the reproducibility of the light curve residual fluctuations , combined with the overall hard - to - soft spectral evolution , which demonstrates that the fluctuations all combine to form a single evolving pulse with a non - monotonic shape . all pulse light curves seem to undergo multiple intensity increases and decreases . however , the definition of a pulse being composed of several monotonically rising and falling regions may be a semantic one : a single pulse might indeed be composed of three separate , linked , smaller pulses . if we do choose to believe that each fluctuation identifies a separate pulse , then the coupling between these fluctuations and the overall hard - to - soft evolution indicates that multiple pulses always combine to form similar , temporally - linked , complex shapes that need to be described by a single smooth hardness evolution . the grbs used to measure pulse residual fluctuations have thus far all been those observed by batse and fermi s gbm experiment . since these instruments have spectral responses favoring detection of higher energy photons ( 25 kev to 2 mev ) over lower ones , it is reasonable to wonder whether these residual features are present at lower energies observed by swift , or if they are primarily a high energy phenomenon . to this end , we examine a sample of single episodic bursts observed by swift . the swift grbs analyzed here were observed between august 1 , 2005 and september 1 , 2009 , and were selected by eye from a sample having measured redshifts as being those having single emission episodes . the initial sample contains 30 grbs . similar to batse , single - pulsed bursts appear to comprise roughly one - third of all swift grbs ( one - quarter if the sample is limited to to long bursts ) . light curves are composed of 64 ms , mask - weighted , background - subtracted , 4-channel data available from swift . the swift four - channel data are similar to those used for batse and fermi gbm , but optimized to the swift energy range ( channel 1 @xmath1 kev , channel 2 @xmath2 kev , channel 3 @xmath3 kev , and channel 4@xmath4 kev ) . the technique used for extracting pulses from grb light curves has been described elsewhere ( e.g. @xcite and references contained therein ) . intervals in the light curve where pulses may reside are first identified using the bayesian blocks approach of @xcite , followed by a pulse - fitting technique that fits the empirical model of @xcite to each interval using a nonlinear least squares routine @xcite . for batse and fermi gbm data , poisson statistics are assumed . for swift data , intensity uncertainties are gaussian , and this requires a modification of the bayesian blocks code ( norris 2014 , private communication ) . the @xcite pulse model is defined as @xmath5},\ ] ] where @xmath6 is time since trigger , and the four norris pulse fit parameters are the pulse amplitude @xmath7 , the pulse start time @xmath8 , the pulse rise parameter @xmath9 , and the pulse decay parameter @xmath10 . the normalization constant @xmath11 is defined in terms of these parameters as @xmath12}$ ] . similarly , the pulse peak time occurs at time @xmath13 . in obtaining the pulse fit , the start and end time values of each bayesian block are used to obtain initial estimates of @xmath8 , @xmath9 , and @xmath10 , while the maximum amplitude occurring during the block is used * to * estimate @xmath7 . observable pulse parameters are defined from the four free pulse fit parameters . we have previously defined the pulse duration @xmath14 as the time interval between instances at which the intensity is @xmath15 ( @xmath16 ) of its maximum value @xcite , @xmath17^{1/2},\ ] ] where @xmath18 . the pulse _ asymmetry _ @xmath19 is also defined from the intensities at @xmath15 of the maximum value , and it has a value of @xmath20^{-1/2};\ ] ] it ranges from a value of @xmath21 for a symmetric pulse to @xmath22 for an asymmetric pulse with a rapid rise and slow decay . the @xcite model provides a good general fit to batse and fermi gbm pulses , but close examination have identified small yet distinct deviations in the residuals that are phased to the pulse duration @xcite . these variations have been found to be contained within the fiducial time interval @xmath23 defined by : @xmath24 with the fiducial end time @xmath25 given by @xmath26 and the fiducial start time @xmath27 given by @xmath28.\ ] ] in the fiducial time interval , the residual variations can be fitted with the empirical four parameter function described by @xcite : @xmath29 } ) & \textrm{if $ t < t_0 - 0.005 $ } \\ a & \textrm{if $ t_0 - 0.005 \le t \le t_0 + 0.005 $ } \\ a j_0(\sqrt{s \omega [ t - t_0 - 0.005 ] } ) & \textrm{if $ t > t_0 + 0.005$. } \end{array } \right.\ ] ] here , @xmath0 is an integer bessel function of the first kind , @xmath30 is the central time of the residuals peak , @xmath31 is the amplitude of the residual peak , @xmath32 is the variability factor the bessel function s angular frequency which defines the timescales of residual wave ( a large @xmath32 corresponds to a rapid rise and fall ) , and @xmath33 is a scaling factor that relates the fraction of time which the function before @xmath30 has been compressed relative to its time - inverted form after @xmath30 ( or a stretching of the function after the peak time relative to that before it ) . using this model , the time during which the pulse intensity is a maximum needs to be a plateau instead of a peak , with a duration of @xmath34 . since there is no evidence that the bessel function continues beyond the third zero ( following the second half - wave ) , the function is truncated at the third zeros @xmath35 . in case it is not intuitively obvious what roles each of these four free parameters plays in the residual fit , we have included examples in which we vary each parameter individually within the fiducial time interval . the results are shown in figure 1a ( varying @xmath31 ) , figure 1b ( varying @xmath30 ) , figure 2a ( varying @xmath32 ) , and figure 2b ( varying @xmath33 ) . for these residual models we have assumed that the underlying pulse has an asymmetry of @xmath36 , and that the fiducial timescale is defined to encapsulate a pulse of this asymmetry . the fiducial values can be converted back to measured values using the following relationships : @xmath37 @xmath38 @xmath39 and @xmath40 where @xmath27 and @xmath25 are the real time values corresponding to the start and end of the fiducial duration . a convenient way to describe the pulse residual amplitudes is to normalize them to the pulse fit amplitudes , because the result is independent of the instrument s signal - to - noise , allowing amplitude comparisons between different instruments . this produces a relative amplitude @xmath41 defined by @xmath42 we apply these pulse - fitting techniques to our sample of 30 swift grbs having single emission episodes , with our initial hypothesis being that each emission episode represents a single pulse . we feel justified in making this assumption because the @xcite results suggest that the residual pattern underlies all isolated pulse structures . this hypothesis is testable ; a pulse exhibiting the residual wave structure should have residual wave properties that correlate with the pulse asymmetry in the manner found previously for batse pulses . thus , we fit the swift 4-channel grb data using the @xcite model and the techniques described by @xcite , but we increase the ncp prior variable ( a prior value for the number of change points occurring in the light curve ) in the bayesian blocks code @xcite and/or the level of dual timescale peak flux significance @xcite so that only a single fitted pulse is returned for each grb in the sample . even though the swift grb sample has been selected from what appear to be bright single - pulsed bursts observed during a predetermined time period , many of the light curves are relatively noisy with large uncertainties . this is not surprising , as grb emission typically decreases at lower energies where background noise increases . as a result , the four - channel pulse fits are generally not as good as those obtained for batse and fermi gbm data , and two bursts ( grb 080129 and grb 090313 ) can not be fitted using the @xcite pulse model . the results of the @xcite fits to the other 28 pulses in the swift sample are summarized in table 1 . we attempt to fit the remaining 28 pulses with both the @xcite model and the @xcite residual fit model . we use a _ @xmath43 difference test _ @xcite to determine if the fit has been improved with the addition of the @xcite residual fit . in this test , the difference in the @xmath43 fit parameters between the two models @xmath44 is given as @xmath45 where the difference in degrees of freedom @xmath46 is @xmath47 here ` p ' refers to the pulse fit and ` r ' refers to the residual fit . a resulting probability of @xmath48 is obtained for a @xmath44 value having @xmath46 degrees of freedom . pulse plus residual fits having @xmath49 do not show significant improvement with the additional model parameters and are excluded , while pulse plus residual fits having @xmath50 are significant enough to be retained . as shown in table 2 , only eleven pulses in our swift sample satisfy the @xmath50 criterion : grb 051111 , grb 060206 , grb 060708 , grb 060912a , grb 061004 , grb 061021 , grb 070318 , grb 070117 , grb 080413b , grb 081222 , and grb 091018 . the residual fits and combined fits of these pulses are summarized in table 3 and are shown in figures 3 through 13 ( panels on the left side indicate the fit to the residuals while panels on the right side indicate the overall fit ) . in order to compare the results of the swift fits with those previously obtained for batse and fermi pulses , we also apply the @xmath43 difference test to the pulses listed in @xcite . six batse pulses ( batse 0673 , batse 0711 , batse 0727 , batse 1301 , batse 7567 , and batse 7775 ) and one fermi pulse ( grb 120326a ) are not found to exhibit @xmath44 fit improvements and have been subsequently excluded from the analysis . it seems possible that _ all _ grb pulses might be characterized by a pulse model that includes the same residual wave pattern found in the batse data , but that this residual pattern becomes harder to measure for faint pulses . to test this hypothesis on our isolated pulse sample , we check to see if the residual pattern is easier to identify when pulses are bright . we use the signal - to - noise ratio @xmath51 defined as the peak 1 s pulse intensity divided by the peak 1 s pulse intensity uncertainty ; these intensities are available for swift in table 1 and for batse and gbm from published online grb catalogs . the @xmath51 values depend on both instrumental sensitivity and instrumental energy response . in figure 14 we compare our measure of the fit improvement ( @xmath52 $ ] ) to the @xmath51 for each pulse ( a large @xmath53 value represents an improved fit ) . a spearman rank order test applied to batse pulses ( crosses ) yields a p - value of @xmath54 that @xmath55 and @xmath51 are uncorrelated while a similar test applied to swift pulses ( filled circles ) yields a p - value of @xmath56 . the trend also appears to hold for the three fermi pulses ( open diamonds ) . these results demonstrate that the fits improve significantly as the signal - to - noise increases , which is consistent with our hypothesis . the pulses not substantially improved by the residual fits also support the hypothesis that the residual wave is present in all isolated pulses : these unimproved pulses are typically among the faintest pulses in the sample . the mean @xmath51 of the pulses whose fits have been improved ( @xmath50 ) is @xmath57 while the the mean @xmath51 of the unimproved pulses ( @xmath49 ) is @xmath58 a student s t - test indicates only a @xmath59 likelihood that these distributions have the same mean . the characteristics of the wave patterns extracted from the improved swift grb pulse residuals are found to be similar to those of batse grb pulse residuals in that they exhibit similar correlations with pulse asymmetry @xmath19 . figure 15 demonstrates the relationship between the variability parameter @xmath32 with @xmath19 , figure 16 demonstrates the relationship between the expansion parameter @xmath33 and @xmath19 , and figure 17 demonstrates the relationship between the fiducial residual peak time @xmath30 and @xmath19 . these correlations show a clear linkage between the shape of the episode s bulk emission and the more rapidly - varying component contained within the episode . since the peak of the residuals @xmath30 occurs close in time to the fitted @xcite pulse peak @xmath60 , both @xmath30 and @xmath60 correlate with @xmath19 . since the residual peaks are closely associated with pulse rise and decay , the stretching parameter @xmath33 correlates with @xmath19 . and , since an asymmetric pulse has a short rise time relative to a longer decay time , it has a large variability and therefore a large @xmath32 to go with a large @xmath19 . there is no observed correlation between @xmath41 and @xmath19 , nor is there any clear correlation between @xmath41 and @xmath51 . figure 18 shows that the mean value of @xmath41 is @xmath61 , which indicates that the residual peak intensity @xmath31 is generally @xmath62 of the fitted peak intensity @xmath7 ; this is true for all pulses in which the residuals can be fitted , regardless of the instrument making the measurements . pulse residuals describing the pulse shape are just as identifiable in symmetric pulses as they are in asymmetric pulses . this does not mean that symmetric pulses are as easy to observe as asymmetric pulses : the fitted pulse amplitude @xmath7 has been found to correlate with @xmath19 for batse pulses @xcite , indicating that asymmetric pulses are generally brighter than symmetric pulses . the residual pattern is not only easier to detect in bright pulses ; it is also easier to detect at higher energies , in pulses with intrinsically larger @xmath63 values , and in asymmetric pulses ( @xmath64 . figure 19 contrasts the characteristics of pulses that have not been improved by the residual fits ( @xmath65 ; in blue ) with the characteristics of pulses that have been improved by the residual fits ( @xmath66 ; in black ) . improved swift fits generally require higher @xmath51 ratios than improved batse fits . since batse and swift pulses have similar minimum @xmath51 ratios indicative of comparable instrumental sensitivities , this difference between residual intensities most likely results from the different spectral responses of the instruments , and suggests that the residual structures are more apparent at high energies than at low energies . the left panel of figure 19 demonstrates an anti - correlation between the normalized residual amplitude @xmath63 and the signal - to - noise @xmath51 : faint pulses with small @xmath51 ratios tend to have larger values of @xmath41 . unimproved pulses near the threshold tend to balance this anti - correlation somewhat ( this is especially true for batse pulses ) by having smaller @xmath41-values . the right panel of figure 19 demonstrates that symmetric pulses tend to be fainter and rarer than asymmetric pulses ; most of the symmetric pulses in the sample have low @xmath51 ratios . overall , we find that eleven out of thirty isolated mask - weighted swift grb pulses are improved by the addition of the residual model . these pulses exhibit complex shapes consistent with those detected in batse and fermi data , compared to 43 out of 50 batse grb pulses and two out of three fermi grb pulses . when the potential model improvement resulting from the residual fits is considered as functions of @xmath51 ( figure 14 ) , @xmath41 ( figure 19 , left panel ) , and @xmath19 ( figure 19 , right panel ) , it appears that residual curves may be present in all isolated pulses , but are harder to identify in faint pulses . it also appears that the residual features are harder to identify at swift s lower energies , even though swift and batse have similar instrumental sensitivities : _ a reasonable interpretation is that the residual features are less pronounced at lower energies . _ finally , residuals are easier to extract when pulses have large normalized residual peaks ( @xmath63 ) ; since symmetric pulses seem to have smaller @xmath41 values than asymmetric pulses , this suggests that there might be an observational bias favoring asymmetric pulses . figures 20 through 25 summarize the hardness evolutions of the eleven swift grb pulses having measurable pulse shapes . the swift hardness is defined to be the ratio of counts in the 50 to 300 kev energy range relative to the counts in the 15 to 50 kev energy range . in order to explore how hardness changes with pulse shape , each hardness plot is marked with the five maxima and minima in the residual light curve : the precursor peak ( downward facing red arrow ) , the dip between the precursor peak and the central peak ( upward facing red arrow ) , the central plateau or peak ( downward facing black arrow ) , the dip between the central peak and the decay peak ( upward facing blue arrow ) , and the decay peak ( downward facing blue arrow ) . these figures demonstrate the same general hard - to - soft evolution that characterizes batse and gbm grb pulses : the pulse rise phase is spectrally harder than the pulse decay phase . furthermore , there seems to be a relationship between the times at which pulses harden and the three intensity peaks . for example , two pulses ( grb 061004 and grb 080413b ) and possibly two others ( grb 070318 and grb 091018 ) have hardnesses that start soft and increase ( reaching maximum hardness at roughly the time of the central peak ) before subsequently decreasing in the normal fashion ; it is interesting to note that these three or four pulses are among the softest in the sample . additionally , many of the pulses either undergo a brief re - hardening near the time of the decay peak : this is easily noticed for hard pulses grb 051111 , grb 060912a , and grb 061021 yet harder to discern for soft pulses . table 4 summarizes the visual pulse structure as it relates to the maximum pulse hardness for swift pulses . we wonder if similar coupled spectral and temporal behaviors exist for the previously - analyzed gbm and batse pulses , and therefore we have expanded the time periods of all plots from @xcite to correspond to the fiducial duration . even though this redistribution of counts might result in low signal - to - noise , we recognize that the hardness evolution may be important over the entire life cycle of a grb pulses , and particularly in the time intervals corresponding to the residual wave features ( for this analysis , we note that the hardness ratio used batse data is the ratio of counts in the 100 kev to 1 mev energy range relative to the counts in the 25 to 100 kev energy range ) . individual pulse results for forty batse pulses having well - measured hardness evolutions have been obtained for the entire fiducial timescale @xcite . the hardness - related temporal structures for the two measurable gbm pulses are summarized in table 5 , while the 44 measurable batse pulse properties are shown in table 6 . tables 4 , 5 , and 6 demonstrate interesting similarities between swift , gbm , and batse grb pulse evolutions : 1 ) hard pulses tend to have hard precursor peaks relative to the central peak , while soft pulses do not , and 2 ) hard pulses typically demonstrate a re - hardening of the decay peak , while soft pulses do not . these results appear to be generally valid even though the energy ranges of the experiments are different . thus , the complex pulse shapes uncovered for batse grb pulses are apparently ubiquitous , and span a larger energy range than just of that detected by batse . we note the rather remarkable result that the hardness evolution seems to differ for hard pulses and soft pulses . specifically , hard pulses seem more likely to evolve hard - to - soft while soft pulses are less likely to show this behavior . furthermore , the times at which pulse evolution changes its decay rate appear linked to the times of the precursor peak , the central peak , and the decay peak . to explore in greater detail the hypothesis that pulse spectral evolution is associated with the residual structures , we examine the hardness evolution of batse pulses by grouping together those with similar maximum hardness ratios . since the batse pulses in this sample all have different amplitudes , durations , and background count rates , it is difficult to average their time - dependent hardnesses in a reasonable yet unbiased way . the approach we choose is to sum together the 64 ms photon counts found in like parts of the fiducial window for similar pulses , then ratio these in different energy channels to obtain an evolving counts hardness ratio . figure 26 indicates the resulting typical hardness evolution of the batse grb pulses , sorted into five groups using individual maximum hardness ratios . hardnesses are taken by dividing the summed high energy counts ( 100 kev to 1 mev ) in each partition by the summed low energy counts ( 25 kev to 100 kev ) . hardness groups are hr@xmath67 ( dark blue ) , @xmath68hr@xmath69 ( light blue ) , @xmath70hr@xmath71 ( dark green ) , @xmath72hr@xmath73 ( light green ) , and hr@xmath74 ( orange ) . similar to the arrows shown in figures 20 through 25 , the down - facing arrows indicate the averaged fiducial positions of the precursor peak , the central peak , and the decay peak , while the up - facing arrows indicate the averaged fiducial positions of the gaps in the residual function between the peaks . _ this figure demonstrates that all grb pulse groups evolve from hard to soft in that they are softer at the end of their fiducial time intervals than at the beginning ; in this sense all grb pulses evolve from hard to soft_. within this general hard - to - soft evolution , it is also true that grb pulse hardness is reinvigorated at the times of the three pulse peaks . in particular , the precursor peak seems to have a profound influence on the subsequent pulse evolution , since the pulses with the largest maximum hardness ratios have the most abrupt hard - to - soft evolutions and are hardest during the precursor peak , while those with the smallest maximum hardness ratios reach these values during the central peak . there is weaker evidence in figure 25 for a decrease in the hardness decay during the decay peak than there is for the other two peaks . @xcite have shown that hardness evolution is a function of pulse asymmetry : the short rise phases of asymmetric pulses cause them to have shorter lags than their symmetric counterparts , and lags are indicative of the rate of hardness decay . in this study we have found that pulses with larger maximum hardnesses also decay faster , suggesting that maximum hardness should correlate with asymmetry . figure 27 demonstrates that this correlation indeed exists and is in fact fairly robust : a spearman rank order test yields a p - value of only @xmath75 that these parameters are uncorrelated . since asymmetry correlates with peak hardness , we expect to see that batse pulses grouped by asymmetry should exhibit hardness evolution that is also related to the timing of the three pulse peaks . figure 28 demonstrates the effects of this grouping in a manner similar to figure 26 , but using @xmath19 as the key parameter rather than maximum hardness . asymmetry groups are @xmath76 ( dark blue ) , @xmath77 ( light blue ) , @xmath78 ( dark green ) , @xmath79 ( light green ) , and @xmath80 ( orange ) . the same general hard - to - soft evolution can be seen in the figure , although grouping the pulses by asymmetry may cause the precursor peak to be weaker than it is in figure 26 , but the rapid decay starting from the central peak is more pronounced . maximum hardness is potentially a fragile classification characteristic , as is depends on how the hardnesses are binned in time ( e.g. bin size , quantity , and bin center time ) . to assure ourselves that the correlation of pulse evolution with hardness is robust , we group the pulses by hardness ( hr@xmath81 ) as measured from the ratio of the published energy fluences ( @xmath82 , where @xmath83 represents some combination of batse energy channels ) : @xmath84 the results , with hardness groups @xmath85 ( dark blue ) , @xmath86 ( light blue ) , @xmath87 ( dark green ) , @xmath88 ( light green ) , and @xmath89 ( orange ) , are shown in figure 29 . these hardness groupings are entirely consistent with those shown in figure 26 . the fact that they are also consistent with those shown in figure 28 demonstrates that grb pulse evolution has a similar functional dependence on both spectral hardness and asymmetry . isolated swift grb pulses , like their batse and gbm counterparts , exhibit three - peaked pulse shapes even though swift observes the sky at lower energies than batse or gbm . it is becoming increasingly clear that grb pulse structure , particularly at high signal - to - noise , can not be adequately modeled by simple monotonically increasing and decreasing functions , and spectrally - dependent variations must be treated as an integral part of each pulse . isolated pulses generally evolve from hard to soft , although re - hardening often occurs coincident with each of the three peaks . hard pulses are more likely to have a hard precursor peak and exhibit hard - to - soft evolution throughout . soft pulses are more likely to have a soft precursor peak , causing them to reach maximum hardness at the time of the central pulse peak . all isolated pulses are capable of having an additional hardening at the time of the decay peak ; this hardening is most often seen as a change in the rate of hardness decay . _ these results strongly suggest that the commonly - used nomenclature of ` hard - to - soft ' and ` intensity tracking ' pulses is not correct ; all isolated grb pulses exhibit both behaviors simultaneously and to different degrees . _ the re - hardening of grb pulse spectra coincident with the three temporal peaks suggests that a single - component evolving spectral model might be insufficient for producing characteristics of isolated grb pulses : it is reasonable to suspect that three separate but overlapping evolving components might exist . @xcite have recently demonstrated through modeling that overlapping hard - to - soft pulses produce a composite hard - to - soft pulse exhibiting intensity tracking behaviors at the peaks . if each of the three pulse components in isolated grb pulses evolves hard - to - soft , then our simple residual model is inadequate for describing these because it assumes that the first and third peaks are mirror images of one another , whereas the hardness evolution of these events would not be time - symmetric . more detailed , physical models are needed to understand the true nature of the underlying complex grb pulse structure . the standard model for grb prompt spectral emission is kinematic energy injection into a medium via collisionless relativistic shocks ; the preferred mechanism by which radiative energy is thought to be released is in the form of a synchrotron spectrum ( e.g. @xcite ) . the physical internal shock model of @xcite , naturally accounts for three separate but temporally overlapping emission regions via a forward shock , a reverse shock , and the region containing the contact discontinuity , and this model may provide a physical basis for the underlying physical structures . the model also predicts that the shock interactions overlap to a considerable extent , which can make them difficult to unambiguously observed separately . kinematic models such as this one need to be merged with multi - component evolving spectral models , as recent multiple component spectral models ( e.g. @xcite ) , which thus far have not addressed the possibility that spectral evolution might be associated with specific light curve elements within an individual pulse . @xcite have considered the radiation mechanisms of kinematically evolving synchrotron shocks , but find that the model better explains observed pulse spectral evolution if ( i ) the distribution of shock - accelerated electrons is steeper than generally assumed , ( ii ) the peak energy dependence is reduced by altering microphysics parameters , ( iii ) the lorentz factor undergoes more pronounced initial variations than previously thought , and ( iv ) the relativistic ejection proceeds with a constant mass flux rather than a constant kinetic energy flux . @xcite indicate that these conditions are hard to justify , but that their ability to test microphysics conditions are limited by existing data . further modeling is needed to see under what conditions the synchrotron shock model is consistent with observations . single - episodic models of grb pulse shapes , such as those in which the asymmetric pulse shape is produced by curvature in a relativistic outflow ( e.g. @xcite ) predict neither the multi - episodic shapes nor the spectral rehardenings observed here . these models likely need significant modifications in order to explain the observations . complex grb pulses containing many more than three peaks may be simply the result of observing overlapping light curves from several isolated pulses . however , it is also possible that there are types of grb pulses different than those described here . _ this latter hypothesis can not even be adequately explored without first having a working definition of a ` gamma - ray burst pulse ' such as has been summarized here for isolated grb pulses . _ the results obtained in this study allow us to make general predications about the spectral evolution of grb - like pulses at other intrinsic energies . for example , low - energy bursting events have been observed by transient experiments and include xrfs ( x - ray flashes ) and xrrs ( x - ray rich grbs ) : these appear to be low - energy counterparts to grbs ( e.g. @xcite ) . similarly , the correlated behaviors of x - ray flares ( e.g. @xcite ) and optical flares ( e.g. @xcite ) observed in grb afterglows suggest that these are simply late , low - energy grb pulses . _ the correlated pulse behaviors identified in this manuscript suggest that low - energy pulses should exhibit time histories dominated by intensity tracking spectral behaviors , with a soft precursor peak and potentially a soft or non - existent decay peak . _ we are currently exploring these possibilities in follow - up studies involving x - ray data and optical / infrared observations using etelman observatory at the university of the virgin islands @xcite . in this paper and two prior ones @xcite we have re - conceived the observational definition of an isolated gamma - ray burst pulse . we have been able to do this because isolated grb pulses have complex but well - defined and easily - replicated characteristics . pulse spectral and temporal evolutions are intertwined in a way suggesting that the underlying physics is orderly and reproducible , even though titanic energies are unleashed through each grb in a relatively short period of time . this study finds that swift grb pulses exhibit the same general characteristics as batse grb pulses : individual pulse light curves are characterized by triple - peaked profiles that generally evolve from hard to soft . the three peaks ( the precursor peak , the central peak , and the decay peak ) are aligned to coincide with the asymmetry of the pulse shape . the precursor and decay peaks can exhibit separate spectral evolution from that of the central peak , causing some pulses to follow intensity tracking evolution overlaying the natural hard - to - soft evolution . the specific form that the evolution takes depends on the maximum spectral hardness of the pulse : hard pulses begin hard , and the overall hardness decays rapidly throughout the pulse . however , during the decay , a re - hardening can occur as the decay peak intensity increases . pulses of intermediate hardness have precursor peaks having hardnesses similar to those of the central peak which decay at a moderate rate . soft pulses can begin soft and harden at the central peak , after which the overall hardness decays slowly throughout the pulse . the hardness of any pulse can remain constant or even increase during the time leading up to the decay peak . _ as a result , ` hard - to - soft ' and ` intensity matching ' grb pulse definitions no longer appear to be valid , since all isolated pulses show both of these behaviors in ways that depend upon asymmetry and/or maximum hardness . _ there is a strong interplay between a grb pulse s hardness and the symmetry of its temporal evolution : asymmetric pulses are generally hard and show a strong hard - to - soft evolution . symmetric pulses are more often soft and exhibit intensity tracking spectral evolution . a pulse s duration indicates the timescale over which a system is radiating ; this can be due to radiative cooling and/or to the timescale over which kinematic events occur . a pulse s asymmetry and hard - to - soft evolution indicate that the rate of cooling is changing and/or that the kinematic events have more energy to release initially than at later times . the triple - peaked nature and spectral re - hardening of isolated grb pulses seems to indicate that energy is injected on as many as three separate occasions during the pulse duration . theoretical models of grb prompt emission need to account for these observed pulse spectrotemporal characteristics . we would like to acknowledge the helpful comments of the anonymous referee , as these greatly helped to improve the manuscript . this work has been supported by nasa epscor grant nnx13ad28a . bonjak , . , & daigne , f. 2014 , , 568 , a45 broadbent , m. e. 2014 , ph.d . burgess , j. m. , preece , r. d. , baring , m. g. , et al . 2011 , , 741 , 24 chincarini , g. , et al . 2010 , , 406 , 2113 crider , a. , liang , e. p. , preece , r. d. , briggs , m. s. , pendleton , g. n. , paciesas , w. s. , band , d. l. , & matteson , j. l. 1999 , , 138 , 401 dermer , c. d. 2004 , , 614 , 284 fenimore , e. e. , madras , c. d. , & nayakshin , s. 1996 , , 473 , 998 ford , l. a. , et al . 1995 , , 439 , 307 hakkila , j. , & preece , r. d. 2011 , , 740 , 104 hakkila , j. , & preece , r. d. 2014 , , 783 , 88 kaneko , y. , preece , r. d. , briggs , m. s. , paciesas , w. s. , meegan , c. a. , & band , d. l. 2006 , , 166 , 298 kino , m. , mizuta , a. , & yamada , s. 2004 , , 611 , 1021 li , l. , liang , e .- w . , tang , q .- w . , et al . 2012 , , 758 , 27 liang , e. , & kargatis , v. 1996 , , 381 , 49 lu , r .- j . , hou , s .- j . , & liang , e .- w . 2010 , , 720 , 1146 margutti , r. , guidorzi , c. , chincarini , g. , bernardini , m. g. , genet , f. , mao , j. , & pasotti , f. 2010 , , 406 , 2149 markwardt , c. b. 2009 , astronomical data analysis software and systems xviii , 411 , 251 morris , d. c. , neff , j. e. , & hakkila , j. e. 2014 , american astronomical society meeting abstracts # 223 , 223 , # 148.42 norris , j. p. , bonnell , j. t. , kazanas , d. , scargle , j. d. , hakkila , j. , & giblin , t. w. 2005 , , 627 , 324 peng , z. y. , ma , l. , lu , r. j. , fang , l. m. , bao , y. y. , & yin , y. 2009 , , 14 , 311 peng , z. y. , zhao , x. h. , yin , y. , bao , y. y. , & ma , l. 2012 , , 752 , 132 preece , r. d. , goldstein , a. , bhat , n. , stanbro , m. , hakkila , j. & blalock , d. 2015 , submitted qin , y .- p . 2002 , , 396 , 705 ramirez - ruiz , e. , & fenimore , e. e. 2000 , , 539 , 712 rees , m. j. , & meszaros , p. 1994 , , 430 , l93 sakamoto , t. , lamb , d. q. , kawai , n. , et al . 2005 , , 629 , 311 scargle , j. d. 1998 , , 504 , 405 steiger , j. h. , shapiro , a. , & browne , m. b. 1985 , psychometrika , 50 , 253 . willingale , r. , genet , f. , granot , j. , & obrien , p. t. 2010 , , 403 , 1296 grb 050801 & 5.8 & 4.1 & 0.999 & 0.100 & 1.46 & 0.09 & 0.08 & 0.12 & 0.21 & 1.35 & 2.92e-06 & 1.50e-02 & 1.94e+00 & 2.12e-01 & 1.00 & 1556 + grb 050908 & 19.8 & 56.0 & 0.053 & 0.286 & 0.70 & 0.09 & -77.87 & 134.36 & 0.06 & 0.01 & 8.15e+03 & 4.10e+04 & 7.91e-01 & 1.32e+00 & 1.01 & 6233 + grb 051111 & 43.4 & 1.9 & 0.708 & 0.032 & 2.66 & 0.13 & -7.88 & 0.36 & 0.28 & 0.01 & 5.69e+00 & 1.03e+00 & 1.02e+01 & 4.91e-01 & 1.02 & 6243 + grb 060206 & 8.3 & 1.4 & 0.209 & 0.064 & 2.79 & 0.10 & -7.18 & 1.95 & 0.35 & 0.01 & 1.55e+02 & 9.76e+01 & 5.75e-01 & 1.19e-01 & 1.01 & 6243 + grb 060512 & 18.3 & 111.3 & 0.032 & 0.373 & 0.88 & 0.12 & -140.92 & 1103.85 & 0.06 & 0.01 & 1.06e+05 & 2.44e+06 & 1.94e-01 & 1.50e+00 & 0.95 & 697 + grb 060708 & 9.2 & 1.2 & 0.518 & 0.099 & 1.94 & 0.09 & -1.85 & 0.51 & 0.18 & 0.01 & 6.63e+00 & 3.43e+00 & 1.58e+00 & 2.54e-01 & 1.11 & 774 + grb 060912a & 5.1 & 0.3 & 0.750 & 0.042 & 8.58 & 0.27 & -0.44 & 0.05 & 0.97 & 0.03 & 4.28e-01 & 1.09e-01 & 1.26e+00 & 7.68e-02 & 1.08 & 774 + grb 060926 & 9.7 & 1.5 & 0.840 & 0.094 & 1.09 & 0.09 & -0.25 & 0.13 & 0.13 & 0.01 & 2.63e-01 & 2.00e-01 & 2.71e+00 & 4.55e-01 & 1.06 & 618 + grb 061004 & 7.1 & 2.3 & 0.161 & 0.099 & 2.54 & 0.10 & -7.49 & 4.45 & 0.26 & 0.01 & 3.04e+02 & 3.79e+02 & 3.83e-01 & 1.58e-01 & 1.04 & 6243 + grb 061021 & 11.5 & 0.4 & 0.844 & 0.024 & 6.11 & 0.16 & 1.66 & 0.04 & 0.66 & 0.01 & 2.95e-01 & 5.85e-02 & 3.23e+00 & 1.19e-01 & 1.17 & 3118 + grb 070318 & 51.1 & 2.2 & 0.921 & 0.026 & 1.76 & 0.09 & -0.83 & 0.10 & 0.17 & 0.00 & 2.84e-01 & 1.04e-01 & 1.57e+01 & 7.14e-01 & 1.01 & 4681 + grb 070506 & 9.2 & 10.5 & 0.114 & 0.245 & 0.96 & 0.08 & -13.75 & 28.46 & 0.09 & 0.01 & 1.14e+03 & 4.87e+03 & 3.52e-01 & 5.02e-01 & 1.00 & 3899 + grb 070611 & 18.6 & 327.4 & 0.012 & 0.413 & 0.82 & 0.13 & -377.70 & 8441.82 & 0.06 & 0.01 & 1.90e+06 & 1.26e+08 & 7.60e-02 & 1.69e+00 & 1.04 & 696 + grb 070810a & 13.1 & 2.3 & 0.444 & 0.119 & 1.90 & 0.12 & -5.10 & 1.24 & 0.18 & 0.01 & 1.82e+01 & 1.20e+01 & 1.94e+00 & 4.04e-01 & 1.01 & 9986 + grb 071010b & 18.3 & 0.4 & 0.624 & 0.017 & 7.70 & 0.18 & -2.40 & 0.11 & 0.85 & 0.01 & 5.28e+00 & 4.63e-01 & 3.81e+00 & 9.47e-02 & 1.08 & 4680 + grb 071031 & 44.7 & 12.2 & 0.585 & 0.204 & 0.50 & 0.06 & -10.89 & 3.87 & 0.04 & 0.00 & 1.82e+01 & 1.90e+01 & 8.70e+00 & 2.79e+00 & 1.00 & 3119 + grb 071117 & 5.4 & 0.2 & 0.659 & 0.029 & 11.3 & 0.4 & -0.80 & 0.05 & 1.44 & 0.02 & 1.13e+00 & 1.72e-01 & 1.19e+00 & 4.84e-02 & 1.10 & 462 + grb 080129 & & & & & 0.50 & 0.12 + grb 080413b & 4.8 & 0.3 & 0.450 & 0.040 & 18.70 & 0.49 & -1.94 & 0.15 & 2.07 & 0.04 & 6.33e+00 & 1.39e+00 & 7.25e-01 & 4.90e-02 & 1.03 & 3274 + grb 080430 & 14.8 & 0.8 & 0.725 & 0.040 & 2.60 & 0.12 & -0.94 & 0.14 & 0.32 & 0.01 & 1.65e+00 & 3.80e-01 & 3.58e+00 & 2.18e-01 & 1.05 & 1087 + grb 080710 & 23.0 & 5.0 & 0.853 & 0.132 & 1.00 & 0.12 & -0.70 & 0.39 & 0.08 & 0.01 & 5.13e-01 & 5.81e-01 & 6.56e+00 & 1.50e+00 & 0.99 & 2337 + grb 080804 & 36.4 & 2.8 & 0.781 & 0.060 & 3.10 & 0.24 & -1.64 & 0.41 & 0.28 & 0.01 & 2.17e+00 & 8.45e-01 & 9.47e+00 & 8.12e-01 & 0.98 & 2337 + grb 080805 & 74.4 & 5.4 & 0.787 & 0.043 & 1.10 & 0.06 & -5.62 & 0.57 & 0.11 & 0.00 & 4.16e+00 & 1.18e+00 & 1.95e+01 & 1.57e+00 & 1.02 & 2337 + grb 081007 & 9.9 & 1.5 & 0.672 & 0.117 & 2.60 & 0.24 & -0.69 & 0.35 & 0.27 & 0.02 & 1.85e+00 & 1.17e+00 & 2.23e+00 & 3.84e-01 & 1.01 & 2337 + grb 081118 & 98.4 & 41.0 & 0.276 & 0.198 & 0.60 & 0.12 & -38.57 & 40.86 & 0.04 & 0.00 & 7.52e+02 & 1.15e+03 & 9.05e+00 & 4.70e+00 & 0.99 & 3274 + grb 081222 & 15.6 & 0.5 & 0.453 & 0.023 & 7.70 & 0.12 & -3.23 & 0.26 & 0.94 & 0.01 & 1.98e+01 & 2.48e+00 & 2.35e+00 & 9.24e-02 & 1.51 & 1087 + grb 090313 & & & & & 0.80 & 0.18 + grb 090809a & 11.9 & 11.1 & 0.191 & 0.331 & 1.10 & 0.12 & -12.14 & 17.55 & 0.10 & 0.01 & 2.97e+02 & 1.05e+03 & 7.54e-01 & 8.86e-01 & 0.97 & 696 + grb 090927 & 3.0 & 0.5 & 0.920 & 0.131 & 2.00 & 0.12 & 0.01 & 0.02 & 0.31 & 0.04 & 1.71e-02 & 3.18e-02 & 9.35e-01 & 1.57e-01 & 1.08 & 2337 + grb 091018 & 5.4 & 0.3 & 0.399 & 0.034 & 10.30 & 0.24 & -1.55 & 0.18 & 1.14 & 0.02 & 1.11e+01 & 2.24e+00 & 7.13e-01 & 4.52e-02 & 1.03 & 2336 + columns : ( 1 ) burst i d , ( 2 ) pulse duration @xmath14 , ( 3 ) duration uncertainty @xmath90 , ( 4 ) asymmetry @xmath19 , ( 5 ) asymmetry uncertainty @xmath91 , ( 6 ) peak flux @xmath92 ( @xmath93 ) , ( 7 ) peak flux uncertainty @xmath94 ( @xmath93 ) , ( 8) start time @xmath8 ( s ) , ( 9 ) start time uncertainty @xmath95 ( s ) , ( 10 ) amplitude @xmath7 , ( 11 ) amplitude uncertainty @xmath96 ( s ) , ( 12 ) rise parameter @xmath9 , ( 13 ) rise parameter uncertainty @xmath97 , ( 14 ) decay parameter @xmath10 , ( 15 ) decay parameter uncertainty @xmath98 , ( 16 ) @xmath99 per degree of freedom @xmath100 , and ( 17 ) degrees of freedom @xmath100 . grb080129 and grb 090313 could not be fitted . lrrrrrrcc grb pulse & @xmath27 & @xmath25 & @xmath101 & @xmath102 & @xmath103 & @xmath104 & p for @xmath44 & significant ? + grb 050801 & -1.501 & 15.890 & 1.00 & 267 & 1.00 & 263 & @xmath105 & + grb 050908 & -80.452 & 28.326 & 0.98 & 1695 & 0.97 & 1691 & @xmath106 & + grb 051111 & -16.495 & 84.408 & 1.04 & 1572 & 1.01 & 1568 & @xmath107 & yes + grb 060206 & -8.108 & 11.548 & 1.14 & 303 & 1.07 & 299 & @xmath108 & yes + grb 060512 & -142.254 & 15.719 & 0.95 & 2504 & 0.95 & 2500 & @xmath109 & + grb 060708 & -3.384 & 16.702 & 1.34 & 308 & 1.17 & 304 & @xmath110 & yes + grb 060912a & -1.519 & 11.104 & 1.17 & 187 & 0.93 & 183 & @xmath111 & yes + grb 060926 & -2.495 & 23.078 & 1.12 & 394 & 1.12 & 390 & 1.0 & + grb 061004 & -8.235 & 10.784 & 1.58 & 235 & 1.21 & 231 & @xmath112 & yes + grb 061021 & -0.760 & 26.887 & 1.87 & 427 & 1.44 & 423 & @xmath113 & yes + grb 070318 & -13.591 & 128.84 & 1.08 & 2221 & 1.06 & 2217 & @xmath114 & yes + grb 070506 & -14.649 & 15.275 & 1.02 & 463 & 1.02 & 459 & @xmath115 & + grb 070611 & -379.247 & 17.260 & 1.00 & 4013 & 1.00 & 4009 & 1.0 & + grb 070810a & -7.116 & 20.954 & 0.99 & 463 & 0.98 & 459 & @xmath116 & + grb 071010b & -5.841 & 36.544 & 1.19 & 621 & 1.18 & 617 & @xmath117 & + grb 071031 & -18.935 & 82.172 & 1.00 & 1546 & 1.00 & 1542 & @xmath118 & + grb 071117 & -1.857 & 10.901 & 1.36 & 194 & 1.31 & 190 & @xmath119 & yes + grb 080129 & & & & & & & & + grb 080413b & -2.668 & 7.669 & 1.94 & 156 & 1.45 & 152 & @xmath120 & yes + grb 080430 & -4.023 & 32.346 & 1.05 & 563 & 1.05 & 559 & @xmath121 & + grb 080710 & -6.119 & 55.364 & 1.03 & 956 & 1.03 & 952 & @xmath122 & + grb 080804 & -9.643 & 82.956 & 1.01 & 1442 & 1.00 & 1438 & @xmath123 & + grb 080805 & -22.089 & 168.045 & 1.06 & 2965 & 1.05 & 2961 & @xmath124 & + grb 081007 & -2.651 & 20.996 & 1.01 & 365 & 1.00 & 361 & @xmath125 & + grb 081118 & -46.324 & 121.500 & 1.00 & 3372 & 1.00 & 3368 & 1.0 & + grb 081222 & -5.638 & 27.706 & 2.14 & 516 & 1.85 & 512 & @xmath126 & yes + grb 090313 & & & & & & & & + grb 090809a & -13.438 & 15.806 & 1.01 & 452 & 1.01 & 450 & 1.0 & + grb 090927 & -0.750 & 7.742 & 1.10 & 128 & 1.04 & 124 & @xmath127 & + grb 091018 & -2.324 & 9.042 & 1.49 & 172 & 1.15 & 168 & @xmath110 & yes + lrrrrrrrrrrrrr grb pulse & @xmath30 & @xmath128 & @xmath31 & @xmath129 & @xmath32 & @xmath130 & @xmath33 & @xmath131 & r + grb 051111 & -0.01 & 0.56 & 0.053 & 0.003 & 9.5 & 0.9 & 0.18 & 0.02 & 0.19 + grb 060206 & 2.93 & 0.12 & 0.047 & 0.004 & 9.7 & 0.8 & 1.00 & 0.13 & 0.13 + grb 060708 & 1.55 & 0.04 & 0.097 & 0.005 & 22.1 & 1.4 & 0.53 & 0.04 & 0.55 + grb 060912a & 0.33 & 0.02 & 0.189 & 0.010 & 73.0 & 10.3 & 0.15 & 0.02 & 0.20 + grb 061004 & 2.97 & 0.05 & 0.098 & 0.004 & 21.9 & 1.4 & 0.53 & 0.04 & 0.38 + grb 061021 & 2.78 & 0.04 & 0.186 & 0.007 & 32.2 & 2.4 & 0.19 & 0.02 & 0.28 + grb 070318 & 2.52 & 0.18 & 0.031 & 0.002 & 11.9 & 2.0 & 0.11 & 0.02 & 0.19 + grb 071117 & 0.61 & 0.10 & 0.090 & 0.020 & 54.8 & 16.3 & 0.70 & 0.29 & 0.06 + grb 080413b & 0.40 & 0.02 & 0.423 & 0.020 & 35.8 & 2.5 & 0.37 & 0.03 & 0.20 + grb 081222 & 4.10 & 0.08 & 0.113 & 0.005 & 18.0 & 1.7 & 0.20 & 0.02 & 0.12 + grb 091018 & 1.02 & 0.04 & 0.146 & 0.008 & 35.6 & 3.1 & 0.31 & 0.035 & 0.13 + lccc swift pulse & max cts hardness & precursor peak & decay peak re - hardening ? + grb 081222 & @xmath132 & hard & yes + grb 071117 & @xmath133 & hard & yes + grb 051111 & @xmath134 & hard & yes + grb 061021 & @xmath135 & moderate & maybe + grb 060912a & @xmath136 & hard & yes + grb 060708 & @xmath137 & moderate & yes + grb 080413b & @xmath138 & soft & maybe + grb 070318 & @xmath139 & moderate & no + grb060206 & @xmath140 & hard & maybe + grb 061004 & @xmath141 & soft & no + grb 091018 & @xmath142 & moderate & maybe + lccc batse pulse & max cts hardness & precursor peak & decay peak re - hardening ? + batse 3040 & @xmath145 & hard & yes + batse 1883 & @xmath146 & hard & yes + batse 563 & @xmath147 & moderate & maybe + batse 907 & @xmath148 & moderate & maybe + batse 3257 & @xmath149 & hard & no + batse 7711 & @xmath150 & hard & maybe + batse 1406 & @xmath151 & hard & maybe + batse 469 & @xmath152 & moderate & yes + batse 2662 & @xmath153 & moderate & maybe + batse 2665 & @xmath154 & hard & yes + batse 3143 & @xmath155 & hard & maybe + batse 3003 & @xmath156 & hard & maybe + batse 658 & @xmath157 & hard & yes + batse 3164 & @xmath158 & soft & yes + batse 1580 & @xmath159 & moderate & maybe + batse 1446 & @xmath160 & soft & maybe + batse 332 & @xmath161 & hard & yes + batse 1467 & @xmath162 & hard & maybe + batse 1806 & @xmath163 & hard & yes + batse 7614 & @xmath164 & moderate & no + batse 501 & @xmath165 & moderate & no + batse 2958p1 & @xmath166 & hard & maybe + batse 7903 & @xmath167 & moderate & yes + batse 7638 & @xmath168 & soft & yes + batse 7843 & @xmath169 & soft & maybe + batse 7989 & @xmath170 & soft & yes + batse 3168 & @xmath171 & soft & no + batse 914 & @xmath172 & soft & no + batse 1145 & @xmath173 & moderate & maybe + batse 1306 & @xmath174 & moderate & no + batse 2958p2 & @xmath175 & soft & no + batse 680 & @xmath176 & soft & maybe + batse 795 & @xmath177 & soft & no + batse 1039 & @xmath178 & moderate & no + batse 540 & @xmath179 & moderate & maybe + batse 2862 & @xmath180 & soft & maybe + batse 3026 & @xmath181 & moderate & maybe + batse 8112 & @xmath182 & soft & no + batse 493 & @xmath141 & soft & no + batse 1432 & @xmath183 & moderate & maybe + batse 1200 & @xmath184 & moderate & maybe + batse 8121 & @xmath185 & soft & no + batse 1319 & @xmath186 & soft & no + batse 1379 & @xmath187 & soft & no + $ ] ) is shown as a function of the signal - to - noise ratio ( @xmath51 ) . a large @xmath188 $ ] value indicates improvement consistent with a better model fit , while a small one suggests only slight improvement . batse fits ( crosses ) and fermi pulses ( open diamonds ) improve systematically with increasing signal - to - noise , suggesting that complex pulse shapes are normal at high signal - to - noise but get harder to recognize at lower signal - to - noise . swift pulses ( filled circles ) have similar @xmath51 ratios to batse and fermi pulses . [ fig14 ] ] vs. @xmath19 for the pulses in this analysis ; there is no obvious correlation . swift pulses are indicated by filled circles , batse pulses by crosses , and fermi pulses by diamonds . * the horizontal line shows the mean value @xmath189 * [ fig18 ] ] ( dark blue ) , eight have @xmath68hr@xmath69 ( light blue ) , nine have @xmath190hr@xmath71 ( dark green ) , ten have @xmath72hr@xmath73 ( light green ) , and ten have hr@xmath74 ( orange ) . the effect of the precursor peak is clearly important in defining whether a pulse undergoes hard - to - soft or intensity matching evolution . [ fig26 ] ] . individual pulses are partitioned using residual structures ( arrows are as described in figure 26 ) , then the counts in each partition are summed for all pulses having similar fluence hardnesses . hardnesses are obtained for these binned groups as described for figure 26 . eight pulses have hr@xmath193 ( dark blue ) , nine have @xmath194hr@xmath195 ( light blue ) , eight have @xmath196hr@xmath197 ( dark green ) , eight have @xmath198hr@xmath199 ( light green ) , and eleven have @xmath200hr@xmath201 ( orange ) . [ fig29 ] ]

this paper reports on our effort in modeling realistic astrophysical neutron star binaries in general relativity . we analyze under what conditions the conformally flat quasiequilibrium ( cfqe ) approach can generate `` astrophysically relevant '' initial data , by developing an analysis that determines the violation of the cfqe approximation in the evolution of the binary described by the full einstein theory . we show that the cfqe assumptions significantly violate the einstein field equations for corotating neutron stars at orbital separations nearly double that of the innermost stable circular orbit ( isco ) separation , thus calling into question the astrophysical relevance of the isco determined in the cfqe approach . with the need to start numerical simulations at large orbital separation in mind , we push for stable and long term integrations of the full einstein equations for the binary neutron star system . we demonstrate the stability of our numerical treatment and analyze the stringent requirements on resolution and size of the computational domain for an accurate simulation of the system .

gr-qc0312030

the analysis of general relativistic binary neutron star processes is an important , yet challenging , endeavor . the importance of understanding these processes is rooted in observational astronomy , both in gravitational wave astronomy and high - energy electromagnetic wave astronomy . neutron star binaries could be the central engines for some classes of gamma - ray bursts , and they are definitely strong candidates as sources of gravitational radiation detectable by the up and coming generation of gravitational wave detectors such as ligo , virgo , tama , geo , and lisa . the challenge of understanding binary neutron star processes is rooted in the complexity of both the nonlinear einstein field equations and the physical properties of the super - nuclear density matter which make up the neutron stars . due to the complexity of the binary neutron star system , various levels of approximation have been employed ( with various levels of success ) as aids in understanding the details of the different stages of the inspiral of comparable mass binary neutron stars , from the quasistationary inspiral stage through to the plunge and merger of the binary stars to the ring - down of the final merged object . these range from the approximation of the structure of the neutron stars themselves ( e.g. , from point particle to finite sized perfect fluid models and the equations of state with different physical approximations ) to the approximation of general relativistic effects ( e.g. , from newtonian gravity , to post - newtonian , to full general relativity ) . based on insights from point particle mechanics in general relativity and orbits of finite size bodies in newtonian gravity , one expects that the early part of the inspiral process will be quasistationary , with the secular ( i.e. on timescales larger than the orbital timescale ) shrinkage of the orbit driven by gravitational radiation . when the orbit shrinks to a small enough radius ( but before touching ) , there may or may not exist an innermost stable orbit ( iso ) beyond which dynamical processes drive the quasistationary inspiral into a plunge phase . whether there is , in fact , a `` phase change '' from quasistationary inspiral to a plunge phase in the fully relativistic theory , and whether or not this happens before any other ( hydrodynamical ) dynamical instabilities , is an unsolved problem in full general relativity ( see , e.g. , @xcite for answers in the newtonian case ) . a recent approach for the investigation of the later part of the neutron star inspiral which has been studied in detail @xcite has drawn much attention . this treatment , which we refer to as the conformally flat quasiequilibrium ( cfqe ) approach , is a procedure for constructing general relativistic configurations that correspond to compact , equal mass binary neutron stars in a quasiequilibrium , circular orbit . these individual configurations , which we refer to as `` cfqe configurations '' , are by themselves solutions to the constraints of general relativity in 3 + 1 form , i.e. , the hamiltonian and momentum constraints . for a given equation of state , each equal mass binary cfqe configuration can be characterized by two parameters : the separation of the two neutron stars and the baryonic ( rest ) mass of each of the neutron stars ( we only consider equal mass , corotating binary configurations here ) . taking the cfqe approximation one step further , one can construct an entire 4-dimensional spacetime by `` gluing '' these configurations together as a time sequence of different cfqe configurations . this construction , which we will refer to as a cfqe - sequence , produces a spacetime that solves 5 of the einstein field equations ( 4 constraint equations and trace of the extrinsic curvature equation ) @xcite . since the total baryonic mass will be constant during the inspiral , the cfqe approach takes constant baryonic cfqe - sequences as a representation of the evolutionary sequence of the binary neutron star inspiral process ; secular evolution of the orbit due to gravitational radiation brings one equilibrium configuration into another with the same rest mass , forming the cfqe - sequence @xcite . in this cfqe - sequence approach , one finds that for a corotational neutron star binary system ( which is the focus of this paper ) , the adm mass of each of the individual cfqe configurations decreases along a constant rest mass cfqe - sequence as the separation between the binary stars decreases until a minimum is attained , after which the adm mass increases as the separation decreases further . using turning point theorems for exact equilibrium configurations in general relativity @xcite , this minimum point signals a secular instability in the evolutionary sequence , and is commonly referred to as the innermost stable circular orbit ( isco ) configuration . in the newtonian case , it has been shown that this secular instability point is encountered in the evolutionary process _ before _ any dynamical instability is reached @xcite . while it is reasonable to assume that the cfqe approximation holds reasonably well for highly separated neutron stars , it is not clear at which point along the evolutionary sequence the cfqe approximation breaks down . it is certainly not clear whether or not the cfqe - sequence approximation is good for all neutron star separations larger than the isco separation , due to the fact that one main assumption in the turning - point theorems @xcite used to interpret these cfqe sequences is the assumption of exact equilibrium . while cfqe configurations are very close to equilibrium for large neutron star separations , they become less so as the separation decreases . unfortunately , it is exactly the small separation regime , i.e. near the isco configuration , in which these theorems are being applied . also , the cfqe - sequence approximation to full general relativity does not provide an estimate of the error of its solutions , unlike , e.g. , the post - newtonian approach where one can compute the next order post - newtonian terms to estimate the error of any post - newtonian approximation . the fact that each cfqe configuration satisfies the constraints of general relativity suggests the use of these configurations as initial data to full general relativistic calculations . the setting of initial data is obviously an important issue in numerical general relativistic astrophysical simulations . while all initial data configurations satisfying the constraints of general relativity ( i.e. , the hamiltonian and momentum constraints ) are in principle legitimate initial data sets , they may not be acceptable for the study of coalescence of astrophysical neutron star binaries . in order for the results of the numerical evolutions to be relevant to observations , e.g. , the gravitational waves emitted in actual neutron star coalescences , we have to make sure that the initial data actually corresponds to a configuration in an astrophysically realistic inspiral . it is not straightforward how one can go about evaluating the usefulness of the cfqe approach in approximating astrophysically relevant phenomena . we note that while the cfqe approach leads to solutions of the constraint equations , the cfqe - sequence approach is _ not _ consistent with the full set of einstein equations . this paper is divided into four main sections . in section [ sec : formulation ] , we describe our fully consistent general relativistic hydrodynamics code used in this paper . we describe in detail the 3 + 1 formulation of the einstein field equations that we couple to the relativistic hydrodynamics equations , along with details of the gauge conditions and discretization techniques used . in section [ sec : cfqe ] , we describe the cfqe - sequence approximation , whose individual configurations will be used as initial data for our fully consistent general relativistic calculations . we demonstrate that the cfqe - sequence definition of the isco for corotating neutron star binaries may not be a relevant concept by showing that if one takes into account the spin energy of the neutron stars when constructing the effective binding energy in the cfqe - sequence approximation , then there is no longer a minimum in the effective binding energy . in section [ sec : short_evolve ] , we analyze the key assumptions of the cfqe - sequence approximation by comparing them with fully general relativistic simulations using cfqe configurations as initial data . we focus on the question of how well the cfqe - sequence approximation for corotating neutron stars approximates full general relativity . this is done by comparing the assumptions in building the cfqe - sequence to fully consistent general relativistic calculations using cfqe configurations as initial data to our numerical evolution code . specifically , we analyze the conformal flatness assumption and the assumption of the existence of a killing vector field . we devise a number of invariant measures of these assumptions , and monitor them in our fully consistent , general relativistic simulations . we find that , as expected , the accuracy of the cfqe - sequence approximation increases for increasing neutron star separation . we present a general algorithm for evaluating whether or not any cfqe configuration can be thought of as astrophysically realistic initial data by analyzing how much the cfqe - sequence approximation violates the einstein field equations . we demonstrate this method by showing that , for any given tolerance , one can find a cfqe configuration whose subsequent evolution in full general relativity will not violate the einstein field equations within a small fraction of an orbit . in section [ sec : long_evolve ] , we analyze the long timescale ( i.e. multiple orbits of the binary system ) numerical evolutions of cfqe initial data configurations using our general relativistic hydrodynamics code . in particular , we use a cfqe configuration which has a @xmath0 larger proper separation than that of the isco cfqe configuration . we analyze the orbital decay rate on multiple orbit timescales . using care in estimating the numerical truncation error as well as the errors introduced by placing the computational boundaries at a finite distance from the neutron star binaries , we find that while the computational resources at our disposal are sufficient to get a reasonable handle on the truncation error , the errors introduced by the placement of the boundary of the computational domain can have a significant impact on the dynamics of the neutron star inspiral . we demonstrate that this is true even if the linear dimensions of the computational domain are as large as one half the size of the gravitational waves being emitted , which is a fairly large computational domain by today s numerical relativity standards . we conjecture that presently available computational resources will not allow a unigrid finite difference code to decrease the discretization parameter @xmath1 sufficiently and simultaneously increase the distance from the binary system to the computational domain boundary sufficiently in order to guarantee that the induced numerical errors will not significantly affect the details of the inspiral process . mesh refinement techniques and/or better outer boundary conditions will be needed in order to accurately simulate the physics on the length scales of the compact object as well as the length scales of the gravitational radiation . our code numerically solves the einstein field equations coupled to a relativistic perfect fluid . the gravitational degrees of freedom are geometrically encoded in the 4-metric , @xmath2 , which are governed by the einstein field equations @xmath3 where @xmath4 is the einstein tensor and @xmath5 is the stress - energy tensor of the perfect fluid , given as @xmath6 here , we have set the gravitational constant @xmath7 and the speed of light @xmath8 to be identically @xmath9 . the 4-velocity of the perfect fluid is denoted as @xmath10 , and @xmath11 , @xmath12 , and @xmath13 are the baryonic mass density , pressure , and specific enthalpy , respectively , of the fluid . the equations of motion governing the perfect fluid are the conservation of stress - energy and baryonic mass @xmath14 these represent five equations governing the five degrees of freedom of the perfect fluid ( the mass density , energy density , and velocity ) . the entire system of equations is closed by choosing an equation of state for the pressure @xmath12 as a function of the baryonic mass density and internal energy density of the fluid . in order to numerically solve the einstein field equations , [ eq : einstein ] , we must cast the equations as an initial value problem . in order to facilitate this , we introduce a foliation of the spacetime into spacelike hypersurfaces where the coordinate @xmath15 labels each spacial hypersurface . we furthermore introduce cartesian coordinates @xmath16 on each spacelike hypersurface . the line element can now be written as @xmath17 where the shift vector @xmath18 is a 3-vector defined on each spacelike hypersurface , @xmath19 is the lapse function , and @xmath20 is the 3-metric . we denote @xmath21 as the inverse of the 3-metric @xmath20 , such that @xmath22 . there are many ways to formulate the einstein equations in an initial value , 3 + 1 form . the standard `` adm '' 3 + 1 formulation @xcite writes the six space - space components of the einstein equations , eq . [ eq : einstein ] , as 12 equations that are first order in time @xmath23 where @xmath24 is the extrinsic curvature of a spacelike hypersurface . here , @xmath25 is the lie derivative operator , @xmath26 is the covariant derivative operator compatible with the 3-metric @xmath20 , @xmath27 is the trace of the extrinsic curvature , @xmath28 is the 3-ricci tensor , while @xmath29 are the components of the 4-ricci tensor that represent the perfect fluid source terms to the einstein equations . the remaining four einstein equations are the constraint equations , which are analytically satisfied on each of the spacelike hypersurfaces as long as they are satisfied on the initial slice . the hamiltonian and momentum constraints are @xmath30 while one could base a numerical evolution code on the adm equations , recent results ( both empirical @xcite and analytical studies @xcite ) indicate that a more suitable choice would be the so - called bssn formulation @xcite . one feature of this formulation of the einstein equations is that the 3-metric is decomposed into a conformal factor @xmath31 and a conformal 3-metric @xmath32 as @xmath33 where the determinant of the conformal 3-metric @xmath32 is identically @xmath9 . similarly , the extrinsic curvature is decomposed into its trace and trace - free parts @xmath34 where @xmath35 is referred to as the conformal trace - free extrinsic curvature , such that @xmath36 . in addition to the decomposition of the traditional adm variables , a key ingredient in the bssn formulation is the introduction of three new evolved variables , namely the three conformal connection functions @xmath37 @xmath38 the final form of the evolution equations which we use in the numerical evolution of the einstein field equations is given as @xmath39 we also perform a 3 + 1 decomposition of the hydrodynamics equations , eq . [ eq : hydro ] . note that the 4-velocity @xmath10 is normalized @xmath40 , so that its components can be written in terms of the three spatial velocity components @xmath41 as @xmath42 where @xmath43 is the lorentz factor @xmath44 . the specific enthalpy , @xmath13 , is given as @xmath45 where @xmath46 is the specific internal energy density . the general relativistic hydrodynamics equations , eq . [ eq : hydro ] , can be written in first order , flux conservative form as @xmath47 where the conservative hydrodynamical variables @xmath48 are written in terms of the primitive variables @xmath49 as @xmath50 = \left [ \begin{array}{c } \sqrt{\gamma } w \rho \\ \sqrt{\gamma } \rho h w^2 v_j \\ \sqrt{\gamma}(\rho h w^2 - p - w\rho ) \end{array } \right ] .\ ] ] the flux vector @xmath51 is written as @xmath52 , \label{eq : hydro_flux}\ ] ] and the source vector @xmath53 is written as @xmath54\ ] ] we discretize each of the 3 spatial coordinate variables @xmath55 using a constant spacing @xmath56 , e.g. , @xmath57 we discretize the time coordinate @xmath15 as @xmath58 where we set @xmath59 for all dynamical simulations performed in this paper . due to the fundamental differences in the phenomena being described by the einstein field equations , eq . [ eq : einstein ] , and the relativistic hydrodynamics equations , eq . [ eq : hydro ] , the discretization methods that we employ for the two sets of equations are very different . in the case of the einstein field equations , we expect the dynamical degrees of freedom to remain smooth and continuous . in the case of the relativistic hydrodynamical equations , we know that shocks ( discontinuities ) can easily form in the physical degrees of freedom . thus , the discretization method used for the hydrodynamical equations will be more complicated in order to allow for the accurate treatment of shock propagation . the approach we use will be based on a finite differencing scheme employing high resolution shock capturing ( hrsc ) techniques . in order to use these techniques , a complete knowledge of the characteristic information is needed . we therefore require the eigenstructure of the jacobian matrices in eq . [ eq : hydro_fosh ] , namely @xmath60 for the flux vector @xmath51 defined in eq . [ eq : hydro_flux ] . this is not a straightforward task , since the flux @xmath51 is expressed as a function of both the primitive and evolved hydrodynamical variables . what we require are a complete set of eigenvectors @xmath61 $ ] and corresponding eigenvalues @xmath62 such that @xmath63 [ \vec{r}_i ] = { \lambda}_i [ \vec{r}_i ] , \;\;\ ; i=1,\cdots,5.\ ] ] ( here , we present the spectral decomposition for the @xmath64-component of the jacobian , since the decomposition for the other two spatial components of the jacobian can be obtained by a straightforward permutation of the spatial coordinates @xmath55 ) . it turns out that the spectral decomposition contains a triply degenerate eigenvalue @xmath65 a set of linearly independent vectors that span this degenerate space is given by @xmath66}^t,\ ] ] @xmath67}^t,\ ] ] @xmath68}^t.\ ] ] the other two eigenvalues are given by @xmath69 } \right \ } - { \beta}^x,\ ] ] with corresponding eigenvectors @xmath70 } ^t,\ ] ] where the relativistic speed of sound in the fluid @xmath71 is given by ( see , e.g. , @xcite ) @xmath72 we have set @xmath73 and @xmath74 . @xmath75 is the entropy per particle and @xmath76 is the total rest energy density which in our case is @xmath77 . we use the above characteristic information to calculate the numerical fluxes @xmath78 using the piecewise - parabolic method ( ppm ) , described in @xcite . while the ppm method has been extended to special relativistic applications ( see , e.g. , @xcite ) , this is the first fully general relativistic application of the method . the discretization of the flux terms in eq . [ eq : hydro_fosh ] are written @xmath79 in order to update the discretized hydrodynamical variables , we simply perform a two - step predictor - corrector method , in order that the entire hydrodynamical update is done in a fully second order manner in _ both _ space and time ( modulo the points where the hydrodynamical variables obtain local extrema , where the accuracy of the spatial derivatives drop down to first order in space . this is a well known property of the so - called godunov schemes ; see , e.g. , @xcite ) . as previously stated , since the fields describing the gravitational degrees of freedom are expected to remain smooth , we simply perform centered - in - space discretizing of the spatial derivatives in eqs . [ eq : bssn_phi]- [ eq : bssn_gamma ] . for discrete time evolution , we use the iterated crank - nicholson method @xcite . in order to achieve a completely second order method in both space and time for the coupled system of equations ( the einstein field equations and the relativistic hydrodynamics equation ) , we use the time stepping method described in figure [ fig : icn_couple ] . [ fig : icn_couple ] in the 3 + 1 initial value formulation of general relativity , one is free to specify the slicing and spatial coordinate conditions by specifying the lapse @xmath19 and shift @xmath18 , respectively . the code described in the previous subsections has been written allowing for an arbitrary choice of these gauge variables . as described in the next section , each configuration in the cfqe - sequence approximation has a vanishing trace of the extrinsic curvature @xmath27 . as we would like to compare our full general relativistic simulations to the cfqe - sequence approximation in an invariant manner , it is desirable to use the same slicing condition during the full numerical simulation as that of the cfqe - sequence approximation . to this end , we would like to select the lapse function @xmath19 such that the trace of the extrinsic curvature @xmath27 remains 0 . notice that if one sets @xmath80 in eq . [ eq : bssn_trk ] , then the equation becomes an elliptic equation for the lapse function @xmath19 . we have thus implemented a multigrid solver @xcite for efficiently solving this equation , also known as the maximal slicing condition @xcite . however , solving an elliptic equation at every timestep can be numerically expensive . we therefore also implement a variant of the so - called `` 1+log '' slicing condition for the lapse , @xmath81 note that this is a completely local condition , and is therefore computationally inexpensive . we use both the maximal slicing condition and the `` 1+log '' slicing condition for simulations presented in this paper . for each result , we indicate which slicing condition is used . for the conditions on the shift , we use a slight modification of the `` gamma - freezing '' shift equation @xcite . specifically , we implement the first integral form of the hyperbolic gamma - driver ( eq . 46 of reference @xcite ) , @xmath82 where we set the constants @xmath83 for all numerical simulations in this paper . we find that the gamma - freezing condition in this modified form , despite its simplicity , results in enhanced stability for the type of problems studied in this paper , and was also very computationally efficient . the problem of boundary conditions in numerical relativity is only now gaining the attention it deserves @xcite . here , we implement a simple , yet empirically stable , boundary condition for the fields describing the gravitational degrees of freedom , eqs . [ eq : bssn_phi]- [ eq : bssn_gamma ] , due to alcubierre @xcite . let @xmath84 represent a field on which we wish to impose this boundary condition in , say , the @xmath64-face of our computational boundary ( the @xmath85- and @xmath86-face are similar ; averaging this method yields boundary conditions for the edge and corner points of the computational boundary ) . we take as an ansatz the form for @xmath87 as @xmath88 @xmath89 taking the derivative of this expression with respect to @xmath15 and @xmath64 , and eliminating @xmath90 and @xmath91 yields @xmath92 where the function @xmath93 . [ eq : miguel_boundary ] is finite differenced in a second order fashion to obtain a boundary condition on the field @xmath87 , where we have interpolated the function @xmath94 from interior points , assuming a falloff of @xmath95 for @xmath94 . we adopt this boundary condition for all of the fields describing the gravitational degrees of freedom . the boundary condition used for the hydrodynamical variables is a simple outflow boundary condition . note that all of the hydrodynamical fields are trivially small at the computational boundary ( the atmosphere has a baryonic mass density that is @xmath96 times smaller than the central baryonic mass density of the neutron stars ) , thus the boundary conditions used for these fields are relativity unimportant . the mathematical assumptions that go into the conformally flat , quasiequilibrium approximation to general relativity for binary , corotating neutron stars are as follows ( for the physical motivation behind the assumptions , see , e.g. , @xcite ) . * the physical 3-metric @xmath97 is assumed to be conformally flat @xmath98 * the lie derivative of the conformal metric @xmath99 with respect to the time variable @xmath15 is identically zero : @xmath100 * the trace of the extrinsic curvature and it s time derivative vanishes : @xmath101 * there exists an approximate timelike helical killing vector field @xmath102 @xmath103 for some constant @xmath104 . * the 4-velocity of the fluid @xmath105 is proportional to the approximate killing vector field @xmath102 : @xmath106 from eq . [ eq : cfqe_confmet_vanish ] , the extrinsic curvature @xmath24 takes the form @xmath107 using this , along with the assumptions of the conformally flat , quasiequilibrium approximation above , the hamiltonian constraint , eq . [ eq : ham_constraint ] , and the momentum constraints , eqs . [ eq : mom_constraint ] , can be written as @xmath108 and @xmath109 respectively , where we define the conformal extrinsic curvature @xmath110 as @xmath111 using this form of the hamiltonian constraint , the maximal slicing condition , @xmath112 , can be written as @xmath113 since the quasiequilibrium approximation assumes the existence of a timelike killing vector , we can analytically find the first integral of the relativistic bernoulli equation , which is simply @xmath114 along with the normalization condition for the 4-velocity , @xmath115 , this equation can be explicitly written as @xmath116 we use the algorithm detailed in @xcite , along with a parallel multigrid solver , to simultaneously solve the algebraic bernoulli integral equation , eq . [ eq : bernoulli ] , and the five elliptic equations corresponding to the hamiltonian constraint ( eq . [ eq : cfqe_ham ] ) , the three momentum constraints ( eqs . [ eq : cfqe_mom ] ) , and the maximal slicing equation ( eq . [ eq : cfqe_maximal ] ) , for the conformal factor @xmath117 , the three components of the shift vector @xmath18 , the lapse @xmath19 , and the matter fields , thus producing a single cfqe configuration . in numerically solving these elliptic equations , we use identical boundary conditions to those in @xcite . we have also assumed a polytropic equation of state , @xmath118 where @xmath119 is the polytropic constant and @xmath120 is the adiabatic index . the baryonic mass of each neutron star in a cfqe configuration , @xmath121 , is defined as half of the total rest mass of the configuration , @xmath122 in this paper , we set @xmath123 and @xmath124 , where @xmath125 is nuclear density ( approximately @xmath126 x @xmath127 ) . for these values of parameters , a single static neutron star configuration that is stable has a maximum adm mass of @xmath128 and a baryonic mass of @xmath129 ( @xmath130 is 1 solar mass ) . for the studies in this paper , we use neutron stars with a baryonic mass of @xmath131 each , which is approximately @xmath132 that of the maximum stable configuration . the adm mass of a single static neutron star for this configuration is @xmath133 ( see figure [ fig : tov ] ) . [ fig : tov ] once the baryonic mass of each of the stars is fixed , the only remaining degree of freedom in the specification of a cfqe corotating configuration is the separation of the two stars . a natural invariant way of specifying the separation of the two neutron stars is to calculate the geodesic distance on the constant @xmath15 hypersurface between the points in each of the neutron stars that corresponds to the maximum baryonic mass density . specifically , if @xmath134 and @xmath135 are the spatial coordinates of the points of maximum baryonic mass density in the first and second star , respectively , then the geodesic whose length will represent the ( spatially ) invariant separation of the neutron stars is the curve @xmath136,\ ] ] where @xmath137 such that @xmath138 the geodesic distance @xmath139 between the two stars is now defined as @xmath140 we use a relaxation technique for numerically solving the geodesic ode , eq . [ eq : geodesic ] , and then use a standard quadrature formula for evaluating the integral for @xmath139 , eq . [ eq : geolength ] . we choose the discretization for solving eq . [ eq : geodesic ] and eq . [ eq : geolength ] to be 10 times as fine as the 3d discretization , and use a third order accurate interpolator to calculate the christoffel symbols @xmath141 from the 3d grid . the maximum baryonic mass points , @xmath134 and @xmath135 , are located by finding the maximum of a 3d third order accurate interpolation polynomial , the interpolation being centered on the discrete point on the 3d grid that has the largest value of the baryonic mass density in each star . [ fig : cfqe_restmass ] in figure [ fig : cfqe_restmass ] , we plot logarithmically spaced contours of the baryonic mass density in the equatorial ( @xmath142 ) plane for two cfqe configurations representing the smallest and largest separations used as initial data for the 3d numerical simulations performed in this paper . a cfqe - sequence is constructed by stringing together several constant baryonic mass cfqe configurations , each configuration differing in only the separation of the neutron stars . the idea is that , due to the fact that the time scale of the gravitational radiation , and thus , that of the orbital decay , is much longer than the orbital timescales , the binary neutron stars are considered to be in `` quasiequilibrium '' . thus , at any particular time , a binary neutron star system can be described by one particular cfqe configuration ; the effect of the gravitational radiation is , over the timescales of one orbit , to alter the configuration to a new cfqe configuration with a slightly smaller separation . it is typical in cfqe approximation studies to calculate an effective binding energy for each configuration . following @xcite , we define the ( dimensionless ) effective binding energy @xmath143 of a single configuration to be @xmath144 where @xmath145 is the adm mass of the configuration , and @xmath146 is the adm mass of a single static neutron star in isolation with rest mass @xmath121 . the usual expression for the adm mass of an asymptotically flat spatial slice is @xmath147 where @xmath148 is the unit outward normal to the sphere of constant radius @xmath149 , which is the domain of integration . using the fact that the cfqe configurations are conformally flat , the adm mass of any configuration reduces to the volume integral @xmath150 [ fig : cfqe_seq_data ] in figure [ fig : cfqe_seq_data ] , we plot the effective binding energy @xmath143 as a function of the angular velocity parameter @xmath104 , for a constant baryonic mass cfqe - sequence ( each star has a baryonic mass of @xmath131 ) . the highest value of @xmath104 for each of our calculated cfqe - sequences corresponds to a cfqe configuration where the neutron stars are touching . the configuration corresponding to the minimum effective binding energy @xmath143 is the isco configuration . it is this point along the cfqe - sequence approximation where the quasiequilibrium configuration goes secularly unstable ; the subsequent evolution of the system is thought to enter a `` plunge phase '' where the neutron stars coalesce within an orbital timescale . of course , when using a computer code to solve differential equations , results produced at one single resolution are meaningless due to the fact that we have no way of knowing how big the numerical errors are . it is always important to make numerical calculations at different resolutions , in order to be able to assess the numerical errors that are inherent to any discretization . not only do we have the usual truncation errors ( which are due to the fact that higher order terms in the taylor expansion of functions have been dropped in our finite difference approximation ) of any finite difference approximation , we also have boundary errors , as we are solving a set of 5 coupled elliptic equations to arrive at any cfqe configuration . therefore , we must make at least three numerical experiments in order to assess the effect of both resolution and boundary placement on the numerical results . we show our cfqe configuration results for three combinations of resolution and outer boundary placement , alongside the results obtained by baumgarte , et . al . @xcite . our cfqe sequence results are in fairly good agreement with those of @xcite , considering the two terms that are subtracted in the construction of @xmath143 , eq . [ eq : bindingenergy ] , are the same to the first two or three significant digits . the resolution used in the two studies are comparable . however , in @xcite , the number of gridpoints across the neutron star was kept fixed . we , on the other hand , keep ( for each sequence calculated ) the boundary fixed , and thus have different numbers of gridpoints across the star in each sequence . notice that , as we both increase the resolution and increase the distance from the center of mass to the boundary of the computational domain , the shape of the curves in figure [ fig : cfqe_seq_data ] remain roughly the same . in fact , we can perform a richardson extrapolation on our numerical results by fitting each point on the curve to an error function . for example , we can use the error function @xmath151 where @xmath152 is the numerical value for the binding energy @xmath143 obtained using a specific discretization @xmath1 and coordinate distance from the center of mass to the outer boundary @xmath153 ( we use @xmath153 to denote this distance in solving the elliptic equations for the cfqe configurations , and we will use @xmath154 to denote the coordinate distance from the center of mass to the outer boundary of the computational domain used during the dynamical evolutions . of course , we always have @xmath155 . ) . note that the first nonzero term for the expansion of the boundary error does not contain a @xmath156 term . this is due to that fact that the boundary conditions we use in solving the elliptic equations for the cfqe configurations , which are identical to those used in @xcite , are exact to this order . @xmath157 is the binding energy in the limit as @xmath158 and @xmath159 , i.e. that given by an exact solution to the differential equations . of course , it is @xmath157 that we are interested in . we use our three cfqe - sequences shown in figure [ fig : cfqe_seq_data ] to solve for the three unknowns @xmath157 , @xmath160 , and @xmath161 from eq . [ eq : cfqe_errfunc ] . a generous estimate of the total error , which will be used for the size of the error bars , is @xmath162 the results are plotted in figure [ fig : cfqe_richard ] . [ fig : cfqe_richard ] note in figure [ fig : cfqe_richard ] that the effective binding energy attains a minimum as the separation between the neutron stars decreases . as explained in the introduction , the cfqe - sequence approximation thus predicts a secular instability at this orbital separation , at which point the evolution of the system will change from quasistationary into a plunge phase , where the neutron stars would merge on timescales of the orbital period . several types of arguments have been used to support this claim @xcite . on the one hand , it is intuitively clear that , in the presence of a dissipation mechanism ( gravitational waves are slowly dissipating the binding energy of the binary system ) , if the binding energy would increase with decreasing separation , this would energetically be an unstable situation . there are also turning point theorems @xcite in full nonlinear general relativity which state that if a sequence of equilibrium configurations attain a minimum in the adm mass - energy , then this equilibrium configuration is secularly unstable . hence , to the accuracy of the cfqe - sequence approximation , the minimum in figure [ fig : cfqe_richard ] is often referred to as the innermost stable circular orbit ( isco ) configuration . one problem with the view that these cfqe - sequences somehow approximate fully consistent solutions to the einstein equations coupled to the relativistic hydrodynamic equations has to do with the spin of the the individual neutron stars . note that in each of the cfqe configurations , the 4-velocity of the fluid is assumed proportional to the timelike helical killing vector field . the neutron stars are thus spinning at the exact same frequency as the orbital frequency , @xmath104 . the neutron stars are said to be corotating with the orbital motion . however , it has been known for some time that realistic binary neutron stars can not be tidally locked during the late stages of inspiral ( i.e. , less than 1000 orbits until the final plunge ) @xcite . this raises a question : if a cfqe configuration is used as initial data , should we expect the subsequent solution to the einstein equations to follow the cfqe - sequence approximation , and keep the stars tidally locked as the stars slowly inspiral ? in light of the results of @xcite , the answer must be no . to first order , we would expect the stars to retain approximately the same spin during the dynamical evolutions as is given in the cfqe configuration that was used as initial data . the resolution of this first question presents a second : what is the status of the isco ? if the stars do _ not _ , in fact , stay tidally locked during dynamical evolution , will this affect the location and/or existence of a turning point in the binding energy curve , i.e. the cfqe isco configuration ? one can gain insight into this question by comparing the spin kinetic energy of relativistic , uniformly rotating neutron stars to the energy scales involved in calculating the binding energy , eq . [ eq : bindingenergy ] . to this end , we define a new effective binding energy that explicitly takes the spin kinetic energy of the neutron stars into account : @xmath163 where @xmath164 is the difference between the adm mass of an isolated stationary neutron star with baryonic mass @xmath121 uniformly rotating with angular velocity @xmath104 and the adm mass of an isolated static non - rotating neutron star with baryonic mass @xmath121 . thus , @xmath164 represents the spin kinetic energy of a uniformly rotating neutron star rotating with an angular velocity @xmath104 . we plot this new effective binding energy @xmath165 in figure [ fig : cfqe_richard_new ] . as can be seen , the binding energy where the spin kinetic energy of each of the neutron stars has been manually factored out no longer attains a minimum ( the data point with the highest value of @xmath104 corresponds to the cfqe configuration in which the two stars are touching ) . this indicates that the secular stability of very close binary neutron stars could depend very sensitively on the spin - orbit coupling of the binary system . the two extreme cases , that of tidally locked neutron stars and non - rotating neutron stars , are depicted by figures [ fig : cfqe_richard ] and [ fig : cfqe_richard_new ] , respectively . the fact that we expect a weak spin - orbit coupling suggests that , in the fully consistent general relativistic hydrodynamics simulations , where we do not expect the neutron stars to remain tidally locked , there may be no turning point and that there may not , in fact , be any isco configuration . [ fig : cfqe_richard_new ] using our numerical evolution code described in section [ sec : formulation ] , along with the cfqe configurations described in section [ sec : cfqe ] as initial data , we analyze to what degree the cfqe configurations actually correspond to quasiequilibrium configurations in full general relativity . we do this by computing fully consistent general relativistic evolutions numerically using cfqe configurations as initial data , and comparing the resulting solution of the einstein field equations with the spacetime obtained by the cfqe approximation . recall that one major assumption in the cfqe approximation method is that the change of the binary separation is small on timescales less than one orbital period . if the solution to the einstein equations obtained by using any particular cfqe configuration as initial data deviates significantly from the cfqe - sequence spacetime on suborbital timescales , then that particular cfqe configuration can not be thought of as representing a quasiequilibrium configuration . of course , the quasiequilibrium approximation becomes better as the separation of the neutron stars increases . that question then becomes : how close can the neutron stars be before the cfqe approximation breaks down ? in this section , we describe a generic method that we subsequently use to answer this question . in short , we use cfqe configurations of differing initial separation as initial data for our fully consistent general relativistic code , and compare the resulting solutions to the einstein field equations to the cfqe - sequence spacetime in a coordinate independent fashion . the magnitude of these differences , which we find to decrease as the initial separation of the neutron stars increases ( as expected ) , will therefore quantify the degree of failure of the cfqe configurations to be truly quasiequilibrium configurations in full general relativity . recall that the cfqe - sequence approximation is done in a maximally sliced @xmath166 coordinate system . in order to facilitate a meaningful comparison between the fully general relativistic calculation and the cfqe - sequence approach , we adopt this slicing condition for all numerical computations done in this section . thus , we determine the lapse function @xmath19 by solving the maximal slicing condition , which is the elliptic equation obtained by setting @xmath167 to @xmath168 in eq . [ eq : bssn_trk ] . in order to decrease the computational cost involved in solving this elliptic equation , we only solve the maximal slicing condition every eight timesteps . we use `` 1 + log '' evolution , eq . [ eq:1pluslog ] , for the lapse @xmath19 for the timesteps where we do not solve the maximal slicing condition . we have performed several numerical tests where we do , in fact , solve the maximal slicing condition for every timestep , and have seen only a negligible affect on the results ( in fact , the measures of the violation of the einstein equations by the cfqe approximation that we find in this section slightly _ increase _ when we solve the maximal slicing condition for the lapse @xmath19 at every timestep ) . in order to meaningfully compare our fully consistent general relativistic simulations with the cfqe - sequence approximation , we must compare quantities that are independent of our choice of coordinates . since we are using the same slicing condition as that assumed by the cfqe - sequence approximation ( namely , maximal slicing ) , we must compare our numerical results with the cfqe - sequence approximation using quantities that are either fully 4-invariant quantities , or 3-invariant quantities ( quantities that are invariant under spatial coordinate transformations ) . in the following , we will construct measures for comparing our numerically constructed spacetime with the cfqe - sequence spacetime based on both the conformal flatness assumption and the assumption of the existence of a timelike helical killing vector field . both of these measures will be exactly zero for the cfqe - sequence spacetime , but will not necessarily be zero for the full solution to the einstein field equations using cfqe configurations as initial data . the magnitude of these measures will thus quantify the magnitude of the cfqe configuration s failure to correspond to a true quasiequilibrium configuration in full general relativity . in table [ tab : shortconfig ] , we show the parameters for the cfqe configurations that are used as initial data for the simulations done in this section . .cfqe configuration parameters used as initial data for fully consistent general relativistic simulations in section [ sec : short_evolve ] . for reference , the isco configuration has a proper geodesic separation of @xmath169 . [ cols="^,^,^,^,^,^ " , ] [ tab : longconfig ] in this section we analyze the numerical evolutions of one particular cfqe initial configuration which has a larger geodesic separation @xmath139 than that of the cfqe isco configuration . specifically , the angular orbital velocity of the cfqe configuration we use exclusively in this section is @xmath170 . this configuration has a geodesic separation of @xmath171 , and corresponds to the second smallest @xmath104 data point shown in figure [ fig : cfqe_richard ] . according to the study in section [ sec : short_evolve ] , this configuration has a violation of the qe and cf assumptions at the @xmath172 and @xmath173 levels , respectively , soon after the evolution starts . we numerically evolve this cfqe initial data configuration using our fully consistent general relativistic treatment . the gauge conditions used for these simulations are the `` 1+log '' equation for the lapse @xmath19 ( eq . [ eq:1pluslog ] ) and eq . [ eq : shiftcondition ] for the shift vector @xmath18 . there is no need to use maximal slicing in this section as comparing to the cfqe - sequence is no longer the point . in table [ tab : longconfig ] , we list the properties of the various computational domains , varying both the resolution and outer boundary placement , used for our long timescale numerical evolutions . our numerical implementation allows us to use a different location for the outer boundary of the computational domain for our initial data solve of the cfqe configuration as that used for the dynamical evolution . we denote @xmath153 as the shortest coordinate distance between the center of our computational domain and the computational boundary of our cubical domain used in solving the initial data problem for the cfqe configuration . we denote @xmath154 as the shortest coordinate distance between the center of our computational domain and the computational boundary of our cubical domain used in our fully consistent general relativistic numerical simulations . note that in table [ tab : longconfig ] , the number of grid points refers to that used in the full dynamical evolution . for those computational domains where @xmath174 , we have used the same resolution @xmath1 with a larger number of grid points to solve for the cfqe initial data configuration . [ fig : geodist ] in figure [ fig : geodist ] , we plot the geodesic separation @xmath139 as a function of coordinate time @xmath15 using this initial data for the various computational domains listed in table [ tab : longconfig ] . qualitatively , we observe a number of interesting features . first of all , note that all of the simulations in figure [ fig : geodist ] display an eccentricity in the orbit . some level of eccentricity is expected due to the fact that the cfqe configuration is constructed with the explicit assumption that the time derivative of the separation @xmath139 exactly vanishes , as argued in @xcite . that is , the assumption of the radial velocity of the binaries to be exactly zero instantaneously ( which is done by assuming the existence of a timelike helical killing vector field in the construction of cfqe configurations ) is not consistent with the astrophysically relevant scenario of quasicircular binary evolution , where the magnitude of the radial velocity , while small , never vanishes . one can obtain some idea regarding the level of orbital eccentricity intrinsic to cfqe configurations by analyzing the point - particle dynamics in the post - newtonian approximation @xcite . in @xcite , it is shown in the context of the post@xmath175-newtonian approximation that if the assumption of circular motion is used to construct the initial conditions ( that is , the conditions @xmath176 and @xmath177 are used to specify the initial conditions ) , then the resulting orbit will have a nonzero eccentricity whose value depends on the initial separation . ideally , we would like to directly compare our general relativistic results for the eccentricity of the orbits apparent in figure [ fig : geodist ] with the post - newtonian results in @xcite . the reason we are unable to do this at the present time is obvious from figure [ fig : geodist ] . note the large difference between the eccentricity for computational domain configuration ns - a as compared to the other configurations . in fact , it is obvious from figure [ fig : geodist ] that the ns - a configuration is qualitatively different from the other configurations . recall that all configurations ns - a through ns - e use the same cfqe configuration as initial data ; note from table [ tab : longconfig ] that the only parameters that differ in producing the results from figure [ fig : geodist ] are the spatial resolution of the discretized computational domain , the location of the outer boundary of the computational domain during numerical evolution where the dynamical boundary conditions are imposed , and the location of the outer boundary of the computational domain used in solving for the cfqe initial data configuration . a cursory study of table [ tab : longconfig ] reveals the major difference between computational domain configuration ns - a and the others , namely , the location of the outer boundary @xmath154 of the computational domain during numerical evolution . the boundary distance @xmath154 of the computational domain configuration ns - a is slightly larger than @xmath178 of the gravitational wavelength @xmath179 characterized by the cfqe initial data configuration , while all of the other configurations have @xmath180 . we must always quantify the errors of any numerical calculation . here , our numerical errors originate from two distinct sources . the first source of error is the truncation error associated with our discretization parameter @xmath1 , while the second source of error is induced by the outer boundary conditions imposed on our numerical calculation . while the behavior of the truncation error is well understood ( it is a local error which scales as an integer power of the discretization parameter @xmath1 ) , the assessment of the effect of the boundary conditions is not as straightforward . theoretically , one could place the outer boundary of the computational domain sufficiently far away so that the outer boundary condition would not be causally connected to the compact objects , and thus would have no effect on them during numerical evolution . however , this is not a practical solution , due to the limitations of computational resources ( especially for a unigrid code ; adaptive mesh refinement could be used in this direction ) . as our outer boundary _ is _ causally connected to the neutron stars in our simulations , we must attempt to assess the errors introduced by the outer boundary conditions in our numerical simulations . we assume that the error induced by the outer boundary conditions can be expanded in terms of powers of @xmath181 , and that the error goes to @xmath168 as @xmath182 . we therefore assume an error function for the eccentricity @xmath183 as @xmath184 where @xmath185 denotes the measured value of eccentricity from our numerical solution using discretization parameter @xmath1 and outer boundary location @xmath154 . using the definition of eccentricity defined in @xcite in which the eccentricity of the orbit is calculated from the orbital separation as a function of time , we compute the eccentricity associated with each simulation shown in figure [ fig : geodist ] . we find that @xmath186 , @xmath187 , @xmath188 , @xmath189 , and @xmath190 . we can then solve for the unknown quantities @xmath191 , @xmath160 , @xmath161 , @xmath192 , and @xmath193 in eq . [ eq : eccentricerror ] . we find that the richardson extrapolated value of the eccentricity is @xmath194 , and that the leading error terms for the truncation error and boundary error are @xmath195 and @xmath196 , respectively . above and beyond the fact that the richardson extrapolated value of the eccentricity @xmath191 is negative , the obvious sign that we are not in the convergence regime ( i.e. that the higher order terms neglected in the error expansion of eq . [ eq : eccentricerror ] are not relatively small ) is that @xmath197 is larger than the error terms in eq . [ eq : eccentricerror ] . as the computational resources available at the present time do not currently allow us to use our unigrid code to simultaneously decrease the discretization parameter @xmath1 further and increase the distance @xmath154 from the center of mass to the outer boundary , we must admit that we can , at this time , make no definite conclusion as to the inherent eccentricity in cfqe configurations used as initial data in numerical relativity . however , the prospect of being able to determine this point in the near future is good ; we can expect both mesh refinement techniques and better outer boundary conditions to greatly aid in reducing the errors induced by the outer boundary in our numerical calculations . recall from sections [ sec : cfqesequence ] and [ sec : nsspin ] that the binding energies @xmath143 and @xmath165 ( eqs . [ eq : bindingenergy ] and [ eq : newbindingenergy ] , respectively ) shown in figures [ fig : cfqe_richard ] and [ fig : cfqe_richard_new ] represent the binding energy of each neutron star binary cfqe configuration as a function of geodesic separation @xmath139 ( the orbital angular velocity @xmath104 is monotonically increasing with decreasing geodesic separation @xmath139 ) . in the cfqe - sequence approximation , this binding energy is slowly converted to gravitational wave energy , and figures [ fig : cfqe_richard ] and [ fig : cfqe_richard_new ] tell us how much gravitational radiation energy is produced for changes in geodesic separation . we can approximate the rate of energy loss at any specific point in this sequence using the standard quadrupole formula ( see , e.g. , @xcite ) , which reduces to @xmath198 for two point particles of mass @xmath199 in circular orbit with radius @xmath200 and orbital angular velocity @xmath104 . we interpolate the data represented in figures [ fig : cfqe_richard ] and [ fig : cfqe_richard_new ] with a cubic spline to obtain the effective binding energies @xmath143 ( eq . [ eq : bindingenergy ] ) and @xmath165 ( eq . [ eq : newbindingenergy ] ) as a function of geodesic separation @xmath139 . we can then easily find @xmath201 as a function of geodesic separation . an estimate of the time rate of change of geodesic separations is then @xmath202 which can be numerically integrated to produce the geodesic separation @xmath139 as a function of time predicted from the cfqe - sequence approximation . we plot these functions in figure [ fig : geodist_cfqe ] , along with results from our fully consistent general relativistic calculation ns - a and ns - b . [ fig : geodist_cfqe ] note that the standard cfqe - sequence prediction of the evolution of the geodesic separation @xmath139 , labeled `` cfqe sequence '' in figure [ fig : geodist_cfqe ] , terminates at a neutron star geodesic separation @xmath169 , whereas the modified cfqe - sequence prediction , labeled `` cfqe sequence @xmath203 '' , terminates at a smaller geodesic separation of @xmath204 . this is due to the fact that the standard effective binding energy @xmath143 defined by eq . [ eq : bindingenergy ] actually attains a minimum before the neutron stars touch ( see figure [ fig : cfqe_richard ] ) , whereas the effective binding energy for the cfqe - sequence where the spin of the neutron stars do not change ( @xmath165 defined by eq . [ eq : newbindingenergy ] ) is monotonically decreasing as the neutron star separation decreases , right through to the point where the neutron stars are touching ( see figure [ fig : cfqe_richard_new ] ) . therefore , the first `` cfqe - sequence '' prediction in figure [ fig : geodist_cfqe ] terminates at this minimum point ( defined as the isco configuration of the sequence ) , whereas the second `` cfqe sequence @xmath203 '' prediction in the figure terminates at the cfqe configuration where the neutron stars are touching . one might expect that a full solution to the einstein equations using a cfqe configuration as initial data might , in fact , have an evolution of the geodesic separation of the neutron star binaries @xmath139 that would lie somewhere in between the two cfqe - sequence approximations shown in figure [ fig : geodist_cfqe ] . after all , these two cfqe - sequence approximations represent the extreme cases of the evolution of the neutron star spin ; the first represents complete tidal locking during the entire evolution , the second represents absolutely no change in the spin state of the individual neutron stars during the entire evolution . of course , the actual solution to the full einstein equations would be expected to couple , to some amount , the increasing orbital angular frequency of the binary neutron stars to the spin of the individual stars . the magnitude of this coupling would determine which of the two cfqe - sequence approximations shown in figure [ fig : geodist_cfqe ] would be considered to be `` more '' correct . it is exactly this question which could be answered with our fully consistent general relativistic calculations . however , in figure [ fig : geodist_cfqe ] we once again see the dramatic effect of the outer boundary conditions on our numerical simulations . note that computational domain configurations used in the calculations ns - a and ns - b have the same spatial resolution @xmath205 and outer boundary location @xmath206 used for solving the cfqe initial data configuration . the only difference between the configurations ns - a and ns - b is the outer boundary placement used during the dynamical evolution : the outer boundary of the computational domain used during numerical simulation ns - a is twice as far away from the center of mass as that used in simulation ns - b . one may be tempted to conclude that the calculation ns - a could be said to be `` more correct '' than ns - b , since the outer boundary is located farther away for the ns - a configuration . however , one must be careful when trying to apply physical intuition to numerical results that are not in the convergence regime . in this case , for instance , there is no reason to believe that the error induced by the outer boundary on our calculation is a monotonic function of @xmath154 ( whereas , e.g. , it is true in general that the truncation error is monotonic in @xmath1 as @xmath158 ) . due to the wave nature of the gravitational radiation being emitted by the binary neutron stars , the boundary errors could have an oscillating component . this makes it even more difficult to try to do a richardson extrapolation type of error analysis in realistic compact object simulations in numerical relativity . while it should be expected that the extraction of gravitational radiation for a numerically generated spacetime could be highly sensitive to the location of the outer boundary ( see @xcite ) , there have been , up to now , no results showing what effect the outer boundary conditions can have on the details of the orbits of compact binaries on timescales larger than one orbital period . here , we see that the dynamical outer boundary conditions can , and do , significantly affect the orbital parameters of compact binaries during numerical evolution . to date , the only fully general relativistic simulations of corotating binary , @xmath123 polytropes is @xcite . the study in @xcite differs from the present study in several important ways . one basic difference is that while we focus on the capability of simulating astrophysical realistic neutron star binaries , @xcite focuses on the dynamics of the final merger of the two neutron stars . thus , the initial data used in @xcite are cfqe configurations either at isco separation , or closer than isco separation ( isco separation here means simply the neutron star separation of the unique configuration that corresponds to a minimum of the binding energy in the constant rest mass cfqe - sequence , e.g. , in figure [ fig : cfqe_richard ] ) . also , in the study @xcite , it was found necessary to manually decrease the orbital angular momentum of each cfqe configuration at the level of several percent , in order to precipitate the binary merger more quickly . complementary to that approach , we are trying to use numerical relativity as a tool to assess the fidelity of the cfqe - sequence approximation itself . we found that in order to perform simulations of the neutron star binary systems compatible with realistic astrophysical scenarios , one must perform simulations using initial data at a distance considerably larger than the isco separation when using corotating cfqe configurations as initial data . we have outlined a generic method for analyzing the regime of validity of the cfqe - sequence approximation and have applied this method to the case of equal mass , corotating binary neutron stars . we have found that , for corotating neutron stars , the violation of the timelike helical killing vector field existence assumption was an order of magnitude larger than the violation of the assumption of conformal flatness . specifically , we have demonstrated that initial data specified by a cfqe configuration with neutron stars having an initial geodesic separation of less than @xmath207 ( which is slightly more than @xmath208 neutron star radii , or roughly twice the isco configuration separation ) would produce a solution to the einstein field equations that violates the killing vector field assumption by more than @xmath9 part in @xmath209 ; the conformal flatness assumption would be violated by more than @xmath9 part in @xmath210 . we thus conclude that , in the corotating case , the cfqe - sequence approximation for neutron star separations of @xmath207 and less violates the einstein field equations at a level larger than @xmath211 , and thus numerical simulations starting with similar cfqe configurations as initial data can not , therefore , be considered as approximating a realistic neutron star binary inspiral . we note that the violations of the assumptions of the cfqe - sequence approximation that we observe in our general relativistic calculations for the corotating binary systems occur on timescales that are two orders of magnitude shorter than the orbital timescale . we suspect that this may be due to interactions between the spin assumption of the individual neutron stars ( the corotating assumption ) and the cfqe assumptions . we have shown that the characteristic shape of the effective binding energy curve within the cfqe approximation is highly sensitive to the spin kinetic energy of the individual neutron stars . we have shown that if we subtract out the spin kinetic energy of the neutron stars in the construction of the effective binding energy ( which approximates the case where the neutron star spin does not increase as the orbital angular velocity increases ) , then the resulting binding energy curve will have no minimum , and thus the cfqe - sequence approximation would not predict the existence of an isco configuration . we speculate that specifying neutron stars with irrotational spin states in the cfqe - sequence approximation may yield a smaller violation of the einstein field equations for a fixed neutron star separation . the analysis we have developed in this paper can be used for a detailed investigation of this effect . more interestingly , the analysis we have developed might provide a way to determine a spin state most consistent with the cfqe approximations , and hence provide a more realistic set of initial data that can be used to start simulations at a smaller initial separations . we have shown that , for our specific neutron star models , we require a resolution of approximately @xmath212 in order to adequately resolve the neutron stars ( `` adequately resolve '' here refers to verifying that we are in the convergence regime through an appropriate richardson extrapolation technique ; this is a much more stringent condition than has been typically used in numerical relativity studies to date involving neutron stars and black holes ) . this resolution scale is over three orders of magnitude smaller than the characteristic wavelength of the gravitational radiation emitted during the last five to ten orbits of the neutron star inspiral process . we have also shown that the location of the outer boundary of the computational domain can have a significant impact on the details of the evolution of the compact objects on timescales of the orbital period . specifically , we have seen that changing the linear dimensions of our computational domain from @xmath213 to @xmath214 can significantly impact the dynamics of binary neutron stars during the first several orbits . this should serve as a warning to the numerical relativity community studying simulations of compact binaries with the hope of extracting gravitational wave information : not only will the outer boundary inhibit the actual process of extracting the gravitational waves , but they also directly affect the sources of the gravitational waves themselves . while our dynamical boundary conditions are not the best choice , and a more consistent treatment , e.g. constraint preserving boundary conditions , would most likely improve the situation , it may be that numerical relativists will be forced to push the outer boundary of the computational domain to the `` local wave zone '' ( which in this case means @xmath215 ) in order to provide realistic gravitational waveforms suitable for use as templates in gravitational wave detectors . unfortunately , this makes the numerical study of orbiting compact objects in numerical relativity particularly hard . note that every time we increase the resolution in our 3d simulations by a factor of two , keeping the outer boundary @xmath154 fixed , we must use 16 times the amount of computational resources to perform any particular simulation . also , every time we increase the outer boundary distance @xmath154 by a factor of two , we must use 8 times the amount of computational resources . therefore , if we wanted to simultaneously increase both the resolution and the outer boundary distance by a factor of 2 , we would require over two orders of magnitude more computational resources . while we have shown in this paper that it is possible to track the details of finite sized compact objects in full numerical relativity , what remains is to be able to do so in such a way that all of the numerical errors ( both truncation and boundary errors ) can be demonstrated to be small over a timescale of several orbital periods . this will be an extremely challenging task , given the current level of computational resources available . it may be necessary to employ mesh refinement in order to accurately simulate all of the physical degrees of freedom that we are interested in . it is a pleasure to thank abhay ashtekar , comer duncan , david garfinkle , lee lindblom , david meier , peter miller , thierry mora , masaru shibata , kip thorne , and clifford will for useful discussions and comments . we also thank nikolaos stergioulas for providing us with code for calculating the adm mass of stationary , uniformly rotating polytropic stars . our application code which solves the einstein equations coupled to the relativistic hydrodynamic equations , along with our various multigrid elliptic solvers , uses the cactus computational toolkit @xcite for parallelization and high performance i / o . financial support for this research has been provided by the asc project ( nsf phy 99 - 79985 ) and the jet propulsion laboratory ( account 100581-a.c.02 ) under contract with the national aeronautics and space administration . computational resource support has been provided by the nsf nrac projects mca02n022 and mca93s025 , and the nas at ames , nasa .

this study presents the mass distribution for a sample of 18 late - type galaxies in nine hickson compact groups . we used rotation curves from high resolution 2d velocity fields of fabry - perot observations and j - band photometry from the 2mass survey , in order to determine the dark halo and the visible matter distributions . the study compares two halo density profile , an isothermal core - like distribution and a cuspy one . we also compare their visible and dark matter distributions with those of galaxies belonging to cluster and field galaxies coming from two samples : 40 cluster galaxies of and 35 field galaxies of . the central halo surface density is found to be constant with respect to the total absolute magnitude similar to what is found for the isolated galaxies . this suggests that the halo density is independent to galaxy type and environment . we have found that core - like density profiles fit better the rotation curves than cuspy - like ones . no major differences have been found between field , cluster and compact group galaxies with respect to their dark halo density profiles .

0909.3936

hickson compact groups of galaxies ( hcgs ) are collections of three to seven galaxies inside a three magnitude interval , where the members have a projected separation of the order of the galaxy diameters and a low velocity dispersion ( @xmath0200 km.s@xmath1 , hickson 1992 ) . these galaxies , located in the denser environments of the local universe , experience a high rate of galaxy - galaxy interactions and merging . they constitute a privileged laboratory to study different stages of interactions from violent merging to systems with no apparent signs of interaction . @xcite analyzed rotation curves for 32 hickson compact group galaxies and found that two - thirds of them had peculiar rotation curves . for the sub - sample of galaxies for which rotation curves could be derived , they found that spiral galaxies in compact groups have low mass - to - light ratios compared to field galaxies by about 30% , which could be explained if compact group galaxies have smaller dark halos than their field counterparts . given that compact groups are environments where tidal encounters are common , it may be expected that interactions have stripped or disrupted the galaxy dark halos at some level . these conclusions have important consequences for the determination of group lifetimes and understanding how compact groups evolve and eventually merge . in order to address these questions , we have launched an observational program to obtain and analyze 2d velocity fields of hcgs using scanning fabry - perot techniques . fabry - perot imaging of about 30 groups ( about 100 galaxies ) have be obtained to date . specific groups have been studied such as hcg 16 @xcite , hcg 92 @xcite , hcg 31@xcite , hcg 90 @xcite , hcg 18 @xcite and hcg 79 @xcite . hcg 07 , hcg 10 , hcg 19 , hcg 87 , hcg 91 and hcg 96 have been studied by @xcite and @xcite studied hcg 88 , hcg 89 , hcg 100 . @xcite studied the kinematics and the star formation in the intragroup medium in hcg 02 , hcg 22 and hcg 23 . a determination of a group evolutionary sequence was done by @xcite and the study of the tully - fisher ( t - f ) relation of the galaxies in these dense environments was done by @xcite . they have shown , for a sample of 25 rotation curves of late - type galaxies , that hcg galaxies follow the ( t - f ) relation but a few galaxies are brighter than expected from the relation with their mass or , alternatively , present a mass that is too low for their luminosities . @xcite favor a scenario in which these outliers have been brightened because of either enhanced star formation or merging . alternatively , the cg galaxies may have undergone truncation of their dark halo due to interactions . the fact that the b - band t - f relation is similar for compact group and field galaxies tells us that these galaxies show common mass - to - size relations and that the halos of compact group galaxies have not been significantly stripped inside r@xmath2 . in this paper , we address the question of the mass distribution ( luminous and dark ) of spiral galaxies in compact groups in order to determine whether galaxies in dense environments have experienced a different history than galaxies in less dense environments . this work extends the study by @xcite on the dynamics of galaxies in compact groups by investigating the distribution of the mass , taking into account the contribution of the bulge , the disk and the dark halo . to quantity the mass distribution of galaxies we have used mass models based on h@xmath3 rotation curves and surface brightness profiles . we could not derive extended rotation curves , with combined h@xmath3 and hi data , as done by @xcite ( hereafter s08 ) for galaxies in looser environments , given that there are no hi rotation curves available for the compact group galaxies in our sample . different profiles have been tested to model dark halos . the mass - to - light fraction of the disk is computed together with the dark halo parameters using the best fit model ( hereafter bfm ) and the maximum disk model ( hereafter mdm ) . the mdm approach uses the exponential disk in order to fit the rotation curve . with the mdm approch we use the iso hypothesis to model the halo . the paper is structured as follows . in section 2 , we present the data and models ( photometric and kinematic ) . in section 3 , we review the mass distribution of galaxies in the field and in clusters . in section 4 we present , analyze and discuss our results on hcgs and compare them to results in less dense environments . finally in section 5 , some concluding remarks are given . in appendix ( a ) we describe individually the rcs for each galaxy and in appendix ( b ) , from figure [ figure04 ] to figure [ figure21 ] , we present the j - band surface brightness profile and the rcs for the galaxies studies . we have used the 2mass survey @xcite to compute the photometric surface brightness profiles . images from the 2mass survey are already calibrated with a given zero point for each field and with a pixel size of 1 on the sky . this pixel size is well matched to the pixel size of the fabry - perot maps of 0.91 @xcite . we have built surface brightness profiles using the j - band because it probes the dominant stellar component out to the outskirts . we have selected a sub - sample of 18 galaxies among the 25 galaxies used in the @xcite study . indeed , 2mass data are not exploitable for seven of them : four galaxies ( hcg19c , hcg79d , hcg89d , hcg96d ) show j - images that are too weak to be exploited and three others ( hcg100c , hcg100d , hcg96a ) prove to be too difficult to derive satisfactory surface brightness profiles . surface brightness profiles have been built by fitting ellipses to the isophotes of the j - band images using the _ ellipse task _ of the _ sdsdas package _ with iraf . we only fixed the center of the ellipse in the fitting parameters and we usually used a 0.2 space between two ellipses . in a few cases ( for hcg07c , hcg16c , hcg19a , hcg88d ) , we had to extrapolate the outer profile using an exponential disk profile in order to reach the same extension as the rotation curves since the sky background contaminates the surface brightness profile . in order to use the mass model , we have performed a decomposition of the surface brightness profile into classical two components : an exponential disk @xmath4 where @xmath5 is the central surface brightness and @xmath6 is the disk scale parameter , and a bulge with an r@xmath7 law @xmath8\ ] ] where @xmath9 is the central surface brightness , and @xmath10 and @xmath11 are scale factors . to perform the decomposition we used a home - made program based on a minimized @xmath12 routine from the minuit package @xcite . the program first fits the disk using a visual estimate of the profile . it then subtracts the fitted disk and adjusts the bulge . the operation is repeated in order to minimize the @xmath12 for each disk and bulge parameter . using this method , the structures seen in the disk surface brightness profiles are carried on to the mass profiles . we then have five parameters , two for the disk ( @xmath5 and @xmath6 ) and three for the bulge ( @xmath9 , @xmath10 and @xmath11 ) . table 1 shows the results of the fit . we also list the values of inclination used to derive the rotation curves , from velocity fields , published in @xcite , @xcite and @xcite . in order to compute the mass distribution , we have combined the stellar light contribution derived from the photometry with the dark halo contribution from the rotation curves using the model developed by @xcite and revised by @xcite . the surface brightness profile is transformed into a mass distribution for the stellar disk and the stellar bulge assuming a variable but radially constant mass - to - light ratio m / l @xcite . mass decompositions into luminous and dark components are not unique , mainly due to the uncertainties of the stellar disk m / l @xcite but also due to our ignorance of the actual dark halo density profile . we minimized this problem by using j - band photometry in order to narrow down the stellar component to the old stellar populations . this mass model estimates the contribution of each velocity component ( bulge , disk and halo ) using the photometric profile decomposition for the luminous contribution ( bulge and disk ) and a predetermined density profile for the halo . the luminosity profile , in the j - band , is used to probe the mass - dominant stellar component and this is transformed into a mass distribution for the stellar disk assuming a variable but radially constant m / l ratio for the disk and the bulge . for this study we used two different density distributions to represent the dark halo . first we used a spherical distribution given by an isothermal sphere with a density profile given by : @xmath13^{3/2}}\ ] ] where @xmath14 is the central halo density and @xmath15 the core radius . we also used the navarro , frenk and white @xcite ( hereafter nfw ) density profile : @xmath16 the result of the quadratic sum of each velocity component ( bulge , disk and halo ) is fitted to the observational rotation curve by minimizing the reduced is the @xmath17 divided by the number of degrees of freedom . ] @xmath12 in the four dimensions space : @xmath18 , @xmath19 , @xmath20 and @xmath15 . the reduced @xmath17 can be affected by small and large scale variations in the measured velocities that can have various origins ( bars , other non - aximmetric motions , star formation , projection effects , etc ) that the models does not take into account . a high @xmath17 does not therefore necessarily means a bad fit . a reduced @xmath17 around 1 is an indication of a good fit , but because of the degeneracies between parameters the result may not be unique . we looked if errors listed in table 1 and 2 for the halo and m / l parameters are correlated with the reduced @xmath17 . it appears that the reduced @xmath17 remains more or less constant with the errors , showing no correlations . in appendix ( a ) we give comments on each galaxy analysis . in appendix ( b ) ( from figure [ figure04 ] to figure [ figure21 ] ) we present the fits to the j - band surface brightness profiles and the rotation curves using the isothermal sphere to model the dark halo . the velocity uncertainties , presented on the plots , are the differences between the two sides of the rotation curve ( receeding and approaching ) . in appendix ( a ) we also give a description of the fit for both the nfw and maximum disk hypothesis for the dark halo , but the corresponding plots are not presented in the paper . the main goal of this work is to compare the mass distributions for galaxies in dense environments with the ones found in less dense or isolated environments . two different works can be found in the literature to which we could compare our results : @xcite ( hereafter b04 ) studies on galaxies in clusters and @xcite ( hereafter s08 ) studied field galaxies . b04 shows rotation curves for 40 spiral galaxies from the @xcite survey , observed with a fp interferometer . b04 followed a similar technique to ours using both the photometry ( in the i - band ) and the rcs in order to fit the halo parameters and determine the mass distribution . the photometric profiles used by b04 come from @xcite . two dark halo models were used by b04 , same as those used by us , a pseudo - isothermal and the nfw density profile . the sample from @xcite is not very homogenous with respect to galaxy environment . more than half the galaxies present comes from clusters . in order to compare our results with cluster galaxies , we had to convert the nfw dark halo parameters they used into our set of parameters . we deduced the central density ( @xmath14 ) and the core radius ( @xmath15 ) from the @xmath21 and @xmath22 from b04 using definitions found in nfw . the parameters for the iso - thermal halo are the same as ours . b04 found that stars - only model fits most of the rcs of their sample over a significant radial range . but the addition of a dark halo generally leads to low values of m / ls . they also confirm a scaling law between the central halo density and the core radius . @xcite used i - band photometry and the @xmath23 indicated in b04 were determined in the same band . in order to compare the results from our sample with the cluster galaxies , it was necessary to transform our surface brightness profile from the j - band to the i - band . we first made the hypothesis that the profile in the j - band has the same shape and slope as in the i - band ( same scale length for the exponential disk and same r@xmath24 for the bulge ) . only the @xmath25 and @xmath26 were modified . we used the apparent magnitude found in ned in the sdss i - band for our galaxies and determined the absolute magnitude in the johnson i - band using @xcite formula and the distances noted in the rc3 @xcite . we scaled the surface brightness profile of our galaxies in order to get the calculated i - band absolute magnitude using our mass model program . hereafter we perform the comparison with all the cluster galaxies sample from b04 and our hcg galaxies and we discuss the results in the conclusion but we did not plot it in the paper . we also wanted to compare with the field galaxies from the ghasp survey ( galaxies halpha survey project , @xcite , @xcite , @xcite , @xcite , @xcite , @xcite ) where late - type field galaxies have been observed using a fabry - perot interferometer . s08 present the mass model for a sub - sample of 36 galaxies of the ghasp survey . for this survey , we had to convert our j - band surface brightness profiles into r - band . for hcg 16a , hcg 88a , hcg 88c , hcg 89a and hcg 89c , we directly used the surface brightness profiles from @xcite and for hcg 96c we used the profile from @xcite . we estimated a mean shift between @xmath25 in the j - band and @xmath25 in the r - band for the remaining galaxies and we verified that the @xmath27 magnitudes obtained with the mass model program were consistent with what we found in the litterature @xcite . s08 confirm the halo scaling laws showing that low luminosity galaxies have small core radius and high central halo density and that the halo surface density is constant with the absolute magnitude . they also found that a trend can be seen between the core radius and the color or the disk scale length . the galaxies from the cluster galaxies sample have absolute b magnitudes between : @xmath28 and the field galaxies b magnitudes are between @xmath29 . our sample of hcg have @xmath30 . we performed a ks test between our sample and the cluster galaxies sample plus a sub - sample of the field galaxies with an @xmath31 representing 16 galaxies . the results show they are similar at the 85% level or higher . we then used this sub - sample of the field galaxies sample with @xmath31 to perform the comparison . in this section we discuss the relationships between halo parameters , disk m / l ratios , halo and disk masses using the different halo models and also comparing with other galaxies samples ( figures [ figure01 ] to [ figure03 ] ) . in order to point out differences in the halo shape parameters between galaxies in different environment , we compared the two samples , cluster and field galaxies , using both the iso and the nfw dark halo profiles . thus we analyse the disk m / l and the halo mass fraction as a function of the b absolute magnitude ( see figure [ figure2 ] ) and we compare the halo mass fraction and the disk m / l using a iso or a nfw halo ( see figure [ figure03 ] ) . we only compare with field galaxies and we derived the m / l within the r - band . for the halo central density and the core radius , we estimated the errors of the fit by taking the 1@xmath32 error of the @xmath12 minimization . we plot the maximum error bars determined for different figures in order not to overfill the plots . finally we show differences between the central halo density and core radius using iso and nfw halo models and we compare the @xmath17 for those two halo models ( figure [ figure03 ] ) . in order to make a fair comparison of the halo parameters , we evaluated these parameters using the r and i photometry for the hcg galaxies . coefficients of the linear regression are quite similar and differences are inside the error bars of the linear fit . for this reason we present in figure [ figure01 ] , the halo parameters derived using the j - band photometry . figures [ figure01]a , b show the tight correlation between the central halo density @xmath14 and the core radius @xmath15 using the iso and nfw models for galaxies in three different environments . figure [ figure01]a shows the strong correlation between the central density and the core radius for all samples using the iso model . table 3 summarises coefficients of the linear regression with the correspondent correlation coefficients . these fits were obtained excluding 5 galaxies from the cluster galaxies sample because they were several orders of magnitude out of range . for the cluster galaxies sample , 5 galaxies show highly peaked halos in comparison with the other galaxies . cluster galaxies galaxies show that part of the sample has lower core radii , when compared to hcg and field galaxies . this is mainly due to the fact that b04 uses a pseudo - isothermal halo density profile instead a strict isothermal model . with the mdm model ( not shown on the plot ) , the relation is also very well defined with a correlation of : @xmath34 ( see table 3 ) . figure [ figure01]b shows the same correlation as before but using nfw model . it is clear that the relation between the central halo density and the core radius is much scattered using the nfw than the iso model . the correlation found using the nfw model is @xmath35 for the hcg galaxies and the central density of the regression is lower than the iso model . the correlation between the central halo density and the core radius is very strong for the three samples using iso or nfw models . the correlation coefficient is almost the same for the cluster galaxies sample using either the iso or the nfw models and quite different for the field and hcg samples using these two models . instead of plotting the disk scale length versus halo core radius as s08 and @xcite , we present the disk scale length normalised to the halo core radius vs the absolute magnitude . the result is presented in figure [ figure01]c . we used the all field galaxies sample in order to display the less luminous galaxies . for the field galaxies it is clear that the ratio is constant , showing that the core radius and the disk scale length are connected somehow and that it is independent from galaxy luminosity . the result for the three samples is consistent . we perform the plot using the iso model only , but the result is almost the same if we use nfw . from @xcite and @xcite we know that the disk scale length is proportional to the galaxy luminosity for a large variety of galaxy types . this relation is also found for field and hcg galaxies . we verified that the core radius increases with the galaxy luminosity . in figure [ figure01]d , we present the halo central surface density ( @xmath38 ) versus the b absolute magnitude for the three samples using the isothermal sphere . hcg galaxies from our sample are brighter than the field galaxies , but the halo surface density is almost the same between the three samples using the iso model . s08 found , from the central halo density and the core radius as a function of the absolute magnitude , that faint galaxies show a more concentrated dark halo than massive late type galaxies . s08 confirmed the results of @xcite , with a slightly higher constant halo density . however , with the cluster galaxies sample and our hcg galaxies , it is difficult to confirm this trend because of the lack of faint galaxies in both samples . the results using nfw model are very similar to what can be observed with iso . @xcite have investigated the evolution of galactic dark matter haloes through collisionless high resolution n - body simulations . they found that haloes evolve through a series of quiescient phases , well fitted by a nfw profile . they showed that the characteristic density @xmath39 and the radius @xmath40 are strongly correlated . in this section , we discuss the fitted disk m / l parameter using the iso , the nfw models ( see figure [ figure02]a and [ figure03]a ) . the dependence of disk m / l on the absolute magnitude for both samples ( field and hcg galaxies ) is analysed in figure [ figure02]a using only the iso model . more than half the hcg galaxies show a very low disk mass - to - light ratio ( inferior to one ) . if we limit the field galaxies sample to galaxies with similar magnitudes ( @xmath31 ) , we found that slightly less than half field galaxies have disk m / l larger than 4.0 when the maximum m / l for hcg galaxies is 3.9 . _ hcg galaxies have a lower m / l disk ratio than isolated galaxies in the same range of absolute blue magnitude . _ this may be due to galaxies in hickson groups having luminosities boosted by star formation triggered by interactions or their maximum velocities diminished by tidal stripping . the disk mass - to - light ratio between iso and nfw halo models for both the hcg galaxies and the field galaxies of ghasp are analysed in figure [ figure03]a . we first can notice that m / l values for the hcg galaxies are spread over a large range . this situation is clearer for field galaxies . almost all galaxies show a greater m / l using the iso model . ( a ) with hcg 87a , the nfw model ca nt fit the rotation curve with a minimum of halo and a little disk , while iso puts the disk to almost zero and extend the dark halo . ( b ) for hcg 91c , nfw puts the bulge to zero and gives a larger disk than iso . the result is that the central part is better fitted with iso and the outer part of the rotation curve is better using nfw and the @xmath17 is very similar for both . ( c ) the situation is similar with hcg 007c because of a bump that neither model can fit . the fact that nfw gives lower m / l disk ratios is consistent with what b04 found for their sample . in their case , the photometry was done in i - band , but it does not change significantly the trend . we also compared the disk m / l for hcg between iso and the maximum disk model ( not shown here ) . the agreement is high even if mdm has , by definition , a natural tendency to give an higher disk m / l . this means that the bfm gives a solution that is very close to mdm and thus mdm appears to be the favourite solution . in conclusion , using the iso model instead of the nfw model results in a greater disk m / l and that , at least for hcg galaxies , the disk mass is independent of the galaxy luminosity . this section presents relations between halo and disk masse as a function of absolute magnitude for two different models : iso and nfw . table 4 presents the disk mass , halo mass and halo mass fraction using the different models iso , nfw and mdm . in the determination of the total mass , the hi mass was not taken into account . figure [ figure02]b shows the halo mass fraction versus the absolute magnitude for both hcg and field galaxies . the plot shows that only three hcg galaxies , hcg 10d , hcg 96c and hcg 88a , have a halo mass fraction lower than 0.8 within the range of absolute b magnitude and 66% of field galaxies have halo mass fraction larger than 0.8 for a larger range of b magnitude . this can be explained by the fact that the disk is predominent . the others galaxies ( hcg and isolated galaxies ) are totally dominated by the dark halo and the dark halo is not related to the luminosity of the galaxy . we were interested in a possible relation between @xmath41 and the disk mass for hcg and field galaxies . figure [ figure02]c shows a clear correlation between the disk @xmath42 and the disk mass for both hcg and field galaxies . as for all figures in that section , we have plot the all sample of field galaxies , if we restrict the field galaxies sample to those with an absolute b magnitude inferior to -18.5 , we noticied that hcg galaxies and field galaxies follow the same relation . this is confirmed with a work by @xcite that analyse disk masses for 18 spirals by both modelling their rotation curve and by fitting spectral energy distributions . the authors found a good agreement between the two methods . beside their discussion about the best way to calculate the disk mass , they present the relationship between the stellar mass , the color and the stellar mass - to - light ratio . they found that the m / l is increasing with the disk mass . based on a technique of mass decomposition of spiral galaxies from disk kinematics @xcite , found that the ratio of the disk mass to halo mass fraction shows a strong dependency with respect to the luminosity . their study points out a higher disk mass compared to our sample or field galaxies . figure [ figure03]b compares the relation between the ratios halo mass fraction / total mass estimated with iso and nfw models for both hcg and field galaxies . a ks test between the two halo mass fractions from field and compact group galaxies for the same range of absolute b magnitude , proved that both samples have 90% chance to be drawn from the parent distribution , indicating that there is no statistical difference between the two four galaxies ( hcg 007c , hcg 10d , hcg 87c and hcg 91c ) show lower halo masses using the nfw profile . the halo is much more important for hcg 007c using the iso model with respect to nfw , but both fit shows a poor @xmath17 . even if hcg 87c does not show a disturbed rc , nfw halo model provides a poor fit . no fit is possible to mimic the rotation curve with a nfw halo model . for hcg 91c , both models lead to almost the same amount of disk , but nfw uses 10 times less halo than iso for almost the same @xmath17 . in conclusion , for both field and hcg galaxies , the mass is dominated by the halo , as b04 showed , and no differences are found between galaxies in the field and galaxies in dense environment with respect to disk mass . we have compared the halo parameters ( core radius and central density ) for both models ( nfw and iso ) for both hcg and field galaxies . figures [ figure03]c , d show that the central halo density with iso is higher than nfw and , in the mean time , the halo core radius is smaller with iso than with nfw . the isothermal model gives apparently a more concentrated dark halo shape than the nwf96 model . this seems in contradiction with the `` natural '' cuspy shape of nfw dark halo profile . indeed iso profile is shallow in the center while nfw is more peaked . ( see @xcite for the comparison between the two halo profiles ) . the different parameters of the halo @xmath14 and @xmath15 are derived together with the m / l of the disk using the best fit model on the rcs . since the slope of the rc is not high enough to be fitted by the cuspy nfw profile , it favours the disk and thus underestimates the halo . this is not the case with the core - like iso model which matches the relatively faint slope of the rc and thus minimizes the disk and favours the halo . figure [ figure03]e compares the quality of the fits using a @xmath17 parameter between the isothermal sphere and the nfw models for both hcg and field galaxies . to display the result , we have limited the @xmath12 range between 0 and 15 . two galaxies ( hcg 19a and hcg 91a ) show a high @xmath17 ( @xmath43 ) with both models . two galaxies ( hcg 087c and hcg 88d ) show a very poor @xmath17 with the nfw model only . three other galaxies ( hcg 16c , hcg 19b and hcg 89c ) show a higher @xmath17 for the nfw than for the isothermal sphere . s08 noted that nfw gives higher @xmath17 than the iso model . figure [ figure03]e confirms this for hcg galaxies . we can conclude that , as for field galaxies , hcg iso dark halo fits the rc better than the nfw model . based on a sample of @xmath44 100 velocity fields of spiral galaxies belonging to 25 hcgs , our previous study has demonstrated that : 40% ( 38 galaxies ) of the hcg galaxies analysed in our sample do not allow the computation of the rc . this is because these galaxies are strongly perturbed by interaction and mergers . 33% ( 31 galaxies ) of hcg galaxies are regular enough to calculate a rc but the rc is not symmetric to allow the derivation of a mass model . these perturbed galaxies are mildly interacting . the remaining 27% ( 25 galaxies ) of them are suitable to derive mass models . for a variety reasons only 18 galaxies have been studied in the present work . the rcs have been combined with 2mass j - band surface brightness profiles to derive their mass models . two shapes of dark halos have been considered : a core - like density profile ( isothermal sphere or iso ) and a cuspy - like one ( nfw ) . best fit models and maximum disk models have been computed for the two dark halo profiles . we have compared the hcg galaxies with two samples of galaxies in different environments : field and clusters galaxies . the three samples ( hcg , isolated and cluster ) have been analysed using the same tools . the results are summarized below : * no obvious differences can be found between the halo parameters for the hcg galaxies and galaxies in other environments . the strong correlation between @xmath14 and @xmath15 is present for the three samples and the slopes of the linear regressions between them are very similar ( between -1.62 and -1.00 ) , well inside the uncertainties . in a study based on late - type and dwarf spheroidal galaxies , @xcite confirms this correlation giving a slope of -1.038 . on the other hand , the linear regression constant found by @xcite is different in comparison with those found with the three samples . this scale difference is may be due to the fact that @xcite compiled his sample using several sources using different methods to obtain halo parameters . * the use of the nfw model gives less satisfactory results than the iso model . as mentioned by s08 , the @xmath17 coefficient is usually larger when using the nfw . nfw model gives a worse correlation between the halo central density and the core radius than iso . this is also true if we use the field galaxies sample . in contrast , the nfw model gives a consistent result for the cluster galaxies sample . halo profiles are closer to isothermal spheres than nfw profile as already found by other authors @xcite . the slope of the linear regression between @xmath14 and @xmath15 is higher for thecluster galaxies sample than for the two other samples when the nfw model is used . the disagreement between iso and nfw is smaller with cluster galaxies sample . 61% of hcg galxies shows higher disk m / l using the iso model compared to nfw . no relation between the disk scale length and the halo central density is seen using either iso or nfw models . * we explored the possible connection between the halo parameters , the halo mass fraction and the disk m / l . the halo mass is high for both field galaxies and compact groups galaxies ( 75 to 95% of the total mass ) , leaving only modest room for the disk mass . the halo surface density is independent of the absolute b magnitude and no clear relation is seen between the disk m / l and @xmath36 , meaning that the halo is independent of the galaxy luminosity . because of the lack of low luminosity hgc galaxies , we could not investigate properl the possible relation between the disk scale length and the halo core radius as s08 and @xcite . we just can say that , for our sample of bright hcg galaxies ( m@xmath45 ) no clear trend could be found . hp would like to thank the scientific council of the laboratoire dastrophysique de marseille ( lam ) for its financial support during the visits in february 2007 and 2008 . hp thanks the capes foundation for its financial support during hp fellowship at lam during the year 2009 ( process 3656/08 - 0 ) . hp would like to thank dr . d. russeil for its help in the surface brightness decomposition . hp wishes to thank niraj welikala for the english proof reading of this paper . the authors wish to thank the anonymous referee for its comments . 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 . 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 . amram , p. , plana , h. , mendes de oliveira , c. , balkowski , c. , boulesteix , j. 2003 , a&a , 402 , 865 amram , p. , mendes de oliveira , c. , plana , h. , balkowski , c. , hernandez , o. , carignan , c. , et al . , 2004 apj , 612 , l5 amram , p. , mendes de oliveira , c. , plana , h. , balkowski , c. , hernandez , o. 2007 , a&a . , 471 , 75 barnes e.i . , sellwood j.a . , kosowsky a. 2004 , aj , 128 , 2724 blais - ouellette s. , 2000 , ph.d . thesis , universit de montral . blais - ouellette s. , amram p. , carignan c. , 2001 , aj , 121 , 1952 . carignan , c. , freeman , k.c . 1985 , apj , 294 , 494 casernato , s. 1983 , mnras , 203 , 735 de block w.j.g . & bosma a. , 2002 , a&a , 385 , 816 de vaucouleurs , g. ; de vaucouleurs , a. ; corwin , h. g. ; buta , r. j. ; paturel , g. ; fouque , p. vizier on - line data catalog : vii/155 . originally published in : springer - verlag : new york , ( 1991 ) , third reference cat . of bright galaxies ( rc3 ) donato f. , gentile g. , salucci p. , 2004 , mnras , 353 , l17 . durbala , a. , del olmo , a. , yun , m. s. , rosado , m. , sulentic , j. w. , plana , h. , iovino , a. et al . 2008 , aj , 135 , 130 epinat , b. ; amram , p. ; marcelin , m. ; balkowski , c. ; daigle , o. ; hernandez , o. ; chemin , l. ; carignan , c. ; gach , j .- l . ; balard , p. 2008 mnras , 388 , 500 epinat , b. ; amram , p. ; marcelin , m. 2008 mnras , 390 , 466 fletcher , r. 1970 , comp . j. , 13 , 317 freeman , k.c . 1970 , apj , 160 , 811 fukugita , m. , shimasaku , k. , ichikawa , t. 1995 , pasp , 107 , 945 garrido o. , marcelin m. , amram p. , balkowski c. , gach j.l . , boulesteix j. 2005 , mnras , 362 , 127 garrido o. , marcelin m. , amram p. , 2004 , mnras , 349 , 225 garrido o. , marcelin m. , amram p. , boissin o. , 2003 , a&a , 399 , 51 garrido o. , marcelin m. , amram p. , boulesteix j. , 2002 , a&a , 387 , 821 hickson , p. , kindl , e. , auman , j.r . 1989 , apjs , 70 , 687 hickson , p. , mendes de oliveira , c. , huchra , j.p . , palumbo , g. 1992 apj , 399 , 353 kassin , susan , a. , de jong , roelof s. , weiner , benjamin , j. 2006 , apj , 643 , 804 kormendy j. , freeman k.c . 2004 , iau symposium 220 . s. ryder , d.j . pisano , m. walker & k.c . freeman editors , p. 377 mendes de oliveira , c. , plana , h. , amram , p. , bolte , m. , boulesteix , j. 1998 apj , 507 691 mendes de oliveira , c. , plana , h. , amram , p. , balkowski , c. , bolte , m. 2001 , aj , 121 , 2524 mendes de oliveira , c. , amram , p. , plana , h. , balkowski , c. 2003 , aj , 126 , 2635 navarro , j.f . , frenk , c.s . , white , s.d . 1996 , apj , 462 , 563 palunas p. , williams t.b . 2000 , aj , 120 , 2884 parodi , b. r. , barazza , f. d. , binggeli , b. 2002 , a&a , 388 , 29 persic , m. , salucci , p. 1990 , , 245 , 577 plana , h. , mendes de oliveira , c. , amram , p. , boulesteix , j. 1998 , , 116 , 2123 plana , h. , mendes de oliveira , c. , amram , p. , bolte , m. , balkowski , c. , boulesteix , j. 1999 apj , 516 , l69 plana , h. , amram , p. , mendes de oliveira , c. , balkowski , c. 2000 , , 120 , 621 plana , h. , amram , p. , mendes de oliveira , c. , balkowski , c. , boulesteix , j. 2003 , aj , 125 , 1736 plana , h. , amram , p. , balkowski , c. , mendes de oliveira , c. _ galaxies : the third dimension _ , asp conference proceedings , vol . 282 . edited by margarita rosado , luc binette , and lorena arias . san francisco : astronomical society of the pacific , 2002 . , p238 romano - daz e. , hoffman y. , heller c. , faltenbacher a. , jones d. , shlosman i. , 2007 , apj , 657 , 56 rubin v.c . , hunter d.a . , ford w. , kent jr . 1991 , apjs , 76 , 153 salucci p. , yegorova i.a . , drory n. 2008 , , 388 , 159 schommer r.a . , bothun g.d . , williams t.b . , mould j.r . 1993 , aj , 105 , 97 spano m. , marcelin m. , amram p. , carignan c. , epinat b. , hernandez o. , 2008 mnras , 383 , 297 skrutskie , m.f . , cutri , r.m . , stiening , r. , weinberg , m.d . , schneider , s. , carpenter , j.m . , beichman , c. et al . 2006 , aj , 131 , 1163 swaters r.a . , madore b.f . , van der bosch f.c . , balcells m. , 2003 , apj , 583 , 732 torres - flores , s. , mendes de oliveira , c. , de mello , d. , amram , p. , epinat , b. , plana , h. and iglesias - pramo , j. 2009 , a&a , submitted verdes - montenegro , l. , del olmo , a. , perea , j. , athanassoula , e. ; marquez , i. , augarde , r. 1997 a&a 321 , 409 zhao h. , 1996 , 278 , 488 * hcg07c * : hcg07 is composed of four galaxies @xcite . we only analyse the kinematics of hcg07c , a galaxy with morphological type sbc . the rotation curve rises to a velocity of about 170 km / s , at a radius of r @xmath0 35(10 kpc ) and then goes down to a plateau , at a lower velocity of @xmath0 150 km / s , out to a radius of 52(14.5 kpc ) . at small radius , the curve is not regular : a plateau is present between 5(1 kpc ) and 8(2 kpc ) . because of the irregularity of the rc , the fit is not good , especially in the center . the bulge component is rather significant in the fit , in order to take into account the first 5(1 kpc ) of the curve . the halo has , by far , the largest contribution of the fitted rcs . it reaches a plateau at @xmath0 35(10 kpc ) . the disk , on the other hand , is weak and its contribution is no more than 10% of the total rotation velocity . when using nfw model and the maximum disk hypothesis , we found a more prominent disk and the halo shows a plateau around 20(6kpc ) . the photometric profile is well fitted with a bulge and an exponential disk . note that the disk profile presents wiggles ( most probably due to the spiral arms ) . in addition , the fit of the bulge within the first 4(0.8 kpc ) is not good , as can be seen from the well visible deviations in figure 1 . + * hcg10d * : hcg10 is composed of four galaxies . we applied the mass model to hcg10d , an sc - type galaxy . the rotation curve rises regularly to 149 km s@xmath1 , out to 20(6 kpc ) . no bulge has been used to fit the rc . the fit is very good and both the disk and the halo almost have an equal contribution in the rc . the halo shows a plateau around 10(3 kpc ) and its contribution is more regular than the disk . as for the nfw and maximum disk models , the contribution for the disk is larger than for the isothermal halo . in particular , in the case of nfw , the disk is responsible for almost 90% of the velocity of the rc . the photometric profile shows that an exponential disk alone is enough to fit it . + * hcg016a * : hcg16 is a group formed of four galaxies . hcg16a is an sbab type galaxy . the rotation curve is a typical example of a late - type rotation curve with a sharp rise in the first 4(3 kpc ) till the maximum rotation velocity of 247 km.s@xmath1 and a plateau reaching out to 30(8 kpc ) . the general fit is good ( ignoring the wiggles in the profile due to the spiral arms ) . the bulge has the largest contribution for any single galaxy in our sample . the contribution of the halo and the disk are comparable , except for the fact that the halo reaches a constant value around @xmath0 14(3.5 kpc ) at 140 km s@xmath1 and the disk rises constantly . the use of the nfw model for the dark halo shows a much larger contribution for the halo ( more than 65% of the fitted rc ) with a plateau starting at 8(2 kpc ) . the maximum disk model gives a larger contribution of the disk ( as expected ) and the halo represents only 30% of the maximum rotational velocity . the photometric profile shows that the disk profile suffers a drop starting at 22(5.5 kpc ) , below the theoretical exponential disk . in that case , for the mass model we used a theoretical exponential disk from 2(0.5 kpc ) to 30(8 kpc ) . + * hcg016c * : hcg16c is an s0 galaxy . the rotation curve is slowly rising to 200 km s@xmath1 . the fit of the rotation curve does not fit well an observed depression between 10(2 kpc ) and 22(5 kpc ) , but the rotation curve is almost entirely dominated by the halo , except the first 5(1.2 kpc ) , where the disk and the bulge allow a good fit the result is almost identical for the other two models we used . the halo totally dominates the rc . the photometric profile is well fitted by a rather important bulge and an exponential disk . in this case the wiggles are probably not related to spiral arms but instead with the merger nature of this galaxy @xcite . + * hcg19a * : hcg19 is a group formed of four galaxies and its brightest member , hcg 19a , is an s0 galaxy . the rotation curve of this galaxy shows a flat part within a radius of 2(0.5 kpc ) and a rise outward , reaching a velocity of 125 km s@xmath1 at 20(5.5 kpc ) . the fit is rather poor and only the halo has a contribution . no disk has been used to fit the rc . the other two models do not fit the rotation curve any better . in particular , the nfw model is a very poor fit to the data , since it can not reproduce the maximum velocity at 35(9.5 kpc ) . the photometric profile is well fitted with only a bulge contribution . + * hcg19b * : hcg19b is classified as an scd galaxy and its rotation curve is quite irregular . it presents a small plateau between 6(1.8 kpc ) and 12(3 kpc ) . it then reaches a maximum velocity of 90 km s@xmath1 at around 30(8 kpc ) . due to these irregularities , the fit of the rotation curve is not good . the halo is the major contribution of the rotation curve and the disk has only a contribution of 20% of the total rotational velocity . the results of the fit using the two other models are better . in these case the contribution of the halo is more important than using the iso model . the photometric profile shows a drop at 12.5(3.2 kpc ) . we replaced the part of the profile between 12.5(3.2 kpc ) and 30 ( 8 kpc ) by an exponential disk . + * hcg87a * : hcg87 is a group formed of three galaxies , and its brightest member , hcg87a is an sbc galaxy . the rotation curve shows a solid body rotation with rotational velocity rises up to 400 km s@xmath1 at 25(13 kpc ) . this is certainly due to the fact that this object is almost edge on . fits of the rcs with the two others models show that the largest contribution comes from the disk , but the fits are not as good as with the isothermal halo . in both cases the halo does not represent more than 45% of the total rotational velocity . the photometric profile is not well fitted by an exponential between 12(6 kpc ) and 26(14 kpc ) where a bump is present . the first 7(3.5 kpc ) of the profile are well fitted by a bulge profile . + * hcg87c * : hcg87c is classified as an sd galaxy . the rotation curve is regular and shows a plateau at 10(5.5 kpc ) with a maximum velocity of 160 km s@xmath1 . the halo fits almost completely the rc . no bulge is present and the disk shows a plateau beginning at 6(3.5 kpc ) . the result of the fit using the maximum disk is the same as for the isothermal halo . on the other hand , the fit using the nfw model is very poor and can not reach the maximum velocity . the photometric profile is fit only with an exponential disk . + * hcg88a * : hcg88 is formed of four galaxies , and its brightest member , hcg88a is classified as an sb galaxy . the rotation curve rises regularly without reaching a plateau . the disk contribution is prominent . the halo contribution rises linearly and the bulge contribution fits well the beginning of the curve . the fit of this rotation curve is one of the best of our sample . the two other models show a very good fit of the rotation curve but the nfw model indicates a larger contribution of the halo ( @xmath0 38% ) in comparison with the isothermal halo . the disk contribution is lower in , almost , the same proportion . the photometric profile is very well fitted with a bulge and an exponential disk . + * hcg88b * : hcg88b is an sbb galaxy . the rotation curve rises regularly . it presents a small plateau between 4(1.5 kpc ) and 10(4 kpc ) and between 30(11.5 kpc ) and 40(15.5 kpc ) . the rotation curve rises to 300 km s@xmath1 and then drops to 190 km s@xmath1 . due to these irregularities , the `` best fit '' is clearly not appropriate . the largest contribution comes from the disk . it shows a plateau at 210 kms s@xmath1 . the bulge also has a large contribution in order to fit the inner part of the curve and the halo rises linearly to 190 km s@xmath1 . the result of the fit using the nfw model shows a major contribution of the halo and the result for the maximum disk is almost identical to that of the isothermal halo . the photometric profile is well fitted by a combination of a bulge an exponential disk . however the disk shows some small oscillations which obviously can not be reproduced with an exponential disk . + * hcg88c * : hcg88c is classified as an scd galaxy . the rotation curve reaches a maximum velocity of 130 km s@xmath1 . the fit performed using only the halo and disk contributions is good . the halo rises regularly and the disk profile shows an unusual shape . the disk profile shows two peaks , at 7(2.5 kpc ) and 15(5.8 kpc ) , and a small depression in between . the maximum disk shows a similar result to that given by the isothermal halo model . the fit with the nfw model shows that the halo alone can almost account for the whole rc . the disk has only a marginal contribution . we tested the fit of the rotation curve using a theoretical exponential but this fit did not represent well the depression of the rotation curve at small radii , we decide then to keep photometric profile . + * hcg88d * : hcg88d is classified as an sm galaxy . the rotation curve rises regularly without showing any plateau , out to a radius of 30(11 kpc ) . the fit with a halo and an exponential disk contributions is very good . the halo rises linearly and the disk rises to a maximum , at 12(4.5 kpc ) and then slowly decreases . the maximum disk model presents similar results to that given by the isothermal sphere , but the nfw model is not able to fit the curve at all . the photometric profile is very well fitted with only a disk . + * hcg89a * : hcg89a is classified as an sc galaxy . the rotation curve shows a bump between 4(2 kpc ) and 8(4 kpc ) and then the velocities rise regularly up to 220 km s@xmath1 . this bump can not be fitted with a larger contribution from the bulge without spoiling the fit of the curve beyond 10(5 kpc ) . the halo profile is increasing almost linearly . the disk contribution rises to 90 km s@xmath46 and then slowly decreases . we could not fit properly the entire curve even using an exponential disk instead of the original disk photometric profile . the maximum disk model gives similar results even if the bump present between 4 and 8is not as well fitted . the nfw model shows that the halo alone is responsible for almost the whole rc , but the model fails to take the bulge into account . the photometric profile is well fitted with a bulge and an exponential disk , for radii up to 12(6 kpc ) , after which the profile drops . + * hcg89b * : hcg89b is an sbc galaxy . the rotation curve shows a plateau starting at 6(3.2 kpc ) , with a maximum velocity of 120 km s@xmath1 . the halo presents the largest contribution , showing a plateau also around 6(3.2 kpc ) . the nwf96 model and the maximum disk basically yield the same result . we replaced the photometric profile between 0 and 6(3.2 kpc ) and with a theoretical exponential disk , to obtain the disk contribution of the rc we tested the fit of the rotation curve using the original brightness profile , but the fit is not significantly different . the surface brightness profile is well fitted using an exponential disk profile except for the first 6(3.2 kpc ) where the profile is flat . this is certainly due to an apparent double nucleus , present on the 2mass and dss images . + * hcg89c * : hcg89c is an scd galaxy . the fit is very good with a larger contribution of the dark halo in comparison with the disk . the use of the maximum disk does not show much of a difference in the result of the fit . the nfw fit , on the other hand , yields the worse fit . in that case the halo occupies almost the entire fitted rc . as for hcg89b , the photometric profile is fitted with an exponential disk only , except at the very beginning , where the profile is flat . + * hcg91a * : hcg 91 contains four galaxies , the brightest being , hcg 91a , classified as an sbc galaxy . the rotation curve for this galaxy is very irregular and , in particular , for a radius greater than 20 ( 9 kpc ) , the rotation curve is very asymmetric . the major contribution comes from the bulge , the disk contribution being much less important . the halo presents a plateau at 10(4 kpc ) . the nfw model gives an even larger contribution from the bulge and the disk is marginal . the maximum disk model yields similar results to those given with the isothermal halo . the photometric profile is well fitted with a bulge and an exponential disk . + * hcg91c * : hcg91c is classified as an sc galaxy . the rotation curve is regular , with a plateau at 16(7 kpc ) , and with a velocity of about 120 km s@xmath1 . the fitted rotation curve is completely dominated by the disk . the halo only represents 30% of the total rotation velocity and the bulge contribution is marginal . the fit with the nfw hardly uses a dark halo and for the maximum disk model the result is similar to that for the isothermal halo , only with an even higher disk contribution . the photometric profile is well fitted by a large bulge and an exponential disk . since at 13.5(6 kpc ) the profile drops drastically , we replaced this part with a theoretical exponential disk to derive the disk contribution for the rc . + * hcg96c * : hcg96c , one of the four galaxies of the compact group hcg 96 , is classified as an sa . its rotation curve is not very extended . it shows a plateau at 4(2.2 kpc ) , with a velocity of 80 km s@xmath1 . no bulge was used , the halo rises linearly and the disk has the largest contribution to the rc . the result of the fit using the maximum disk is similar and the fit using the nfw model shows a larger contribution of the dark halo ( of about @xmath0 15% ) in the first 4(2 kpc ) . after which the halo increases at a slower rate . the photometric profile is well fitted with a disk and a very small bulge , except at the outskirts of the galaxy . lccccccccccc h007c & @xmath47&19.12 & 7.21 & 14.70 & 9.02 & 2.81 & 0.55@xmath48 & 0.05@xmath49 & 6.00@xmath50 & 0.05@xmath51 & 4.29 + h010d & @xmath52 & 17.80 & 2.47 & 0.00 & 0.00 & 0.00 & 1.00@xmath53 & 1.60@xmath54 & 1.80@xmath55 & 0.18@xmath56 & 0.39 + h016a & @xmath57 & 17.64 & 3.98 & 9.91 & 3.17 & 3.12 & 0.20@xmath58 & 0.40@xmath59 & 1.60@xmath60 & 0.50@xmath48 & 7.61 + h016c & @xmath61 & 16.60 & 1.44 & 12.35 & 9.92 & 2.86 & 0.11@xmath62 & 0.03@xmath63 & 6.00@xmath50 & 0.13@xmath64 & 3.00 + h019a & @xmath65 & 0.00 & 1.00 & 9.95 & 1.88 & 25.06 & 0.00@xmath66 & 0.10@xmath63 & 9.00@xmath67 & 0.02@xmath68 & 39.53 + h019b & @xmath69 & 18.75 & 3.12 & 11.90 & 0.84 & 2.71 & 0.14@xmath51 & 0.05@xmath70 & 8.00@xmath71 & 0.01@xmath72 & 3.13 + h087a & @xmath73 & 17.62 & 6.78 & 12.80 & 10.58 & 5.32 & 0.00@xmath66 & 0.00@xmath74 & 23.00@xmath50 & 0.03@xmath72 & 8.22 + h087c & @xmath75 & 18.26 & 2.85 & 0.00 & 0.00 & 0.00 & 0.10@xmath63 & 0.10@xmath58 & 3.50@xmath76 & 0.13@xmath77 & 2.07 + h088a & @xmath78 & 17.21 & 4.56 & 13.01 & 11.02 & 2.65 & 1.05@xmath79 & 1.75@xmath80 & 21.00@xmath81 & 0.02@xmath72 & 0.90 + h088b & @xmath82 & 18.13 & 4.56 & 9.10 & 1.12 & 4.27 & 0.20@xmath58 & 1.50@xmath83 & 45.00@xmath84 & 0.01@xmath72 & 3.64 + h088c & @xmath85 & 17.95 & 2.09 & 19.72 & 3882.00 & 3.54 & 0.30@xmath86 & 0.45@xmath87 & 19.50@xmath88 & 0.01@xmath72 & 5.89 + h088d & @xmath89 & 17.94 & 2.85 & 0.00 & 0.00 & 0.00 & 0.10@xmath63 & 0.97@xmath53 & 14.50@xmath90 & 0.01@xmath72 & 1.20 + h089a & @xmath91 & 19.62 & 12.48 & 13.30 & 6.81 & 5.31 & 1.20@xmath60 & 0.80@xmath92 & 18.00@xmath93 & 0.01@xmath72 & 1.28 + h089b & @xmath94 & 18.17 & 4.24 & 0.00 & 0.00 & 0.00 & 1.00@xmath53 & 0.70@xmath53 & 2.00@xmath95 & 0.19@xmath96 & 1.28 + h089c & @xmath97 & 18.51 & 2.12 & 0.00 & 0.00 & 0.00 & 1.00@xmath53 & 1.10@xmath98 & 6.00@xmath99 & 0.05@xmath100 & 0.28 + h091a & @xmath75 & 18.13 & 7.42 & 13.15 & 20.69 & 8.44 & 1.00@xmath53 & 0.10@xmath58 & 2.00@xmath101 & 0.35@xmath102 & 33.14 + h091c & @xmath103 & 18.76 & 5.08 & 14.80 & 12.78 & 4.26 & 0.00@xmath72 & 0.70@xmath104 & 3.00@xmath101 & 0.01@xmath72 & 1.66 + h096c & @xmath105 & 16.30 & 1.29 & 12.43 & 3.83 & 2.50 & 0.05@xmath62 & 0.16@xmath64 & 18.00@xmath81 & 0.01@xmath72 & 0.81 + h007c & 0.45 & 1.45 & 5.00 & 0.03 & 5.13 & 0.00@xmath106 & 1.95@xmath107 & 6.00@xmath108 & 0.01@xmath72 & 5.50 + h010d & 1.00 & 1.90 & 1.70 & 0.15 & 0.43 & 1.00@xmath53 & 2.70@xmath109 & 2.50@xmath110 & 0.03@xmath51 & 1.42 + h016a & 0.20 & 0.75 & 1.10 & 0.45 & 9.04 & 0.15@xmath56 & 0.35@xmath111 & 1.70@xmath55 & 0.75@xmath59 & 7.54 + h016c & 0.05 & 0.10 & 5.00 & 0.16 & 3.52 & 0.00@xmath112 & 0.01@xmath113 & 35.00@xmath114 & 0.01@xmath72 & 14.27 + h019a & 0.00 & 0.10 & 9.00 & 0.03 & 45.26 & 0.00@xmath66 & 0.10@xmath63 & 12.00@xmath115 & 0.00@xmath66 & 189.79 + h019b & 0.03 & 0.21 & 18.00 & 0.01 & 3.65 & 0.00@xmath112 & 0.00@xmath116 & 32.00@xmath117 & 0.00@xmath66 & 7.53 + h087a & 0.00 & 0.39 & 5.00 & 0.01 & 1.19 & 0.00@xmath66 & 0.01@xmath118 & 54.00@xmath119 & 0.01@xmath72 & 59.36 + h087c & 0.10 & 0.20 & 4.00 & 0.11 & 2.47 & 0.10@xmath63 & 1.40@xmath120 & 1.00@xmath53 & 0.01@xmath72 & 63.94 + h088a & 1.05 & 1.75 & 36.00 & 0.01 & 1.03 & 0.80@xmath121 & 1.00@xmath53 & 130.00@xmath122 & 0.00@xmath66 & 0.72 + h088b & 0.20 & 1.80 & 76.00 & 0.01 & 4.33 & 0.10@xmath63 & 0.50@xmath123 & 16.00@xmath124 & 0.03@xmath51 & 4.54 + h088c & 0.30 & 0.49 & 23.50 & 0.01 & 6.91 & 0.30@xmath86 & 0.03@xmath113 & 13.00@xmath125 & 0.01@xmath72 & 5.33 + h088d & 0.10 & 1.00 & 14.50 & 0.01 & 1.44 & 0.10@xmath63 & 1.39@xmath126 & 19.95@xmath127 & 0.00@xmath66 & 61.71 + h089a & 1.10 & 0.90 & 20.00 & 0.01 & 1.43 & 0.15@xmath128 & 0.11@xmath129 & 67.00@xmath130 & 0.00@xmath66 & 3.76 + h089b & 1.00 & 1.00 & 1.50 & 0.28 & 1.51 & 1.00@xmath53 & 0.80@xmath131 & 3.00@xmath132 & 0.10@xmath133 & 3.22 + h089c & 1.00 & 1.20 & 8.00 & 0.04 & 0.33 & 1.00@xmath53 & 0.01@xmath134 & 97.00@xmath135 & 0.00@xmath66 & 9.88 + h091a & 1.00 & 0.20 & 2.00 & 0.32 & 37.54 & 1.60@xmath136 & 0.01@xmath72 & 10.00@xmath137 & 0.01@xmath72 & 36.49 + h091c & 0.01 & 0.78 & 2.00 & 0.01 & 1.95 & 0.00@xmath138 & 0.79@xmath139 & 1.00@xmath140 & 0.00@xmath106 & 1.74 + h096c & 0.05 & 0.16 & 36.00 & 0.01 & 0.98 & 0.05@xmath62 & 0.11@xmath77 & 9.00@xmath141 & 0.01@xmath106 & 0.82 + a & @xmath142 & @xmath143 & @xmath144 & @xmath145 & @xmath146 & @xmath147 & + b & @xmath148 & @xmath149 & @xmath150 & @xmath151 & @xmath152 & @xmath153 & + corr & @xmath154 & @xmath155 & @xmath156 & @xmath157 & @xmath158 & @xmath159 & @xmath160 + h007c & 58.00 & 0.18 & 9.5 & 0.98 & 5.1 & 3.3 & 0.39 & 6.8 & 1.40 & 0.17 + h010d & 62.00 & 1.5 & 1.5 & 0.50 & 1.8 & 1.2 & 0.40 & 2.5 & 0.32 & 0.11 + h016a & 54.00 & 2.6 & 3.5 & 0.57 & 5.0 & 1.3 & 0.21 & 2.3 & 4.30 & 0.65 + h016c & 51.00 & 0.081 & 12.0 & 0.99 & 0.27 & 11.0 & 0.98 & 0.027 & 12.00 & 1.00 + h019a & 57.00 & 0.45 & 4.3 & 0.91 & 0.45 & 5.0 & 0.92 & 0.45 & 9.0 & 0.67 + h019b & 56.00 & 0.050 & 2.6 & 0.98 & 0.32 & 2.7 & 0.90 & 0.0 & 1.70 & 1.00 + h087a & 112.00 & 0.0 & 86 & 1.00 & 2.8 & 0.94 & 0.25 & 0.072 & 59.00 & 1.00 + h087c & 119.00 & 0.16 & 5.8 & 0.97 & 0.32 & 6.6 & 0.95 & 2.2 & 0.019 & 0.01 + h088a & 80.00 & 14.0 & 9.9 & 0.42 & 14.0 & 11.0 & 0.45 & 7.8 & 18.00 & 0.70 + h088b & 80.00 & 12.0 & 18.0 & 0.61 & 14.0 & 14.0 & 0.50 & 3.9 & 24.00 & 0.86 + h088c & 80.00 & 1.0 & 5.4 & 0.84 & 1.1 & 5.2 & 0.82 & 0.069 & 5.30 & 0.99 + h088d & 80.00 & 1.2 & 4.2 & 0.78 & 1.2 & 4.2 & 0.78 & 1.7 & 0.09 & 0.05 + h089a & 117.00 & 2.2 & 15.0 & 0.87 & 2.5 & 15.0 & 0.86 & 0.31 & 18.00 & 0.98 + h089b & 120.00 & 0.84 & 2.6 & 0.76 & 1.6 & 2.0 & 0.56 & 1.3 & 2.6 & 0.67 + h089c & 117.00 & 1.1 & 9.1 & 0.90 & 1.4 & 11.0 & 0.88 & 0.012 & 15.00 & 1.00 + h091a & 92.50 & 1.1 & 6.3 & 0.85 & 2.2 & 5.8 & 0.72 & 0.11 & 3.0 & 0.96 + h091c & 97.60 & 2.8 & 0.43 & 0.13 & 3.1 & 0.084 & 0.03 & 3.2 & 0.002 & 0.00 + h096c & 118.00 & 0.45 & 0.40 & 0.47 & 0.45 & 0.54 & 0.55 & 0.31 & 0.52 & 0.63 +

we outline a method by which the angular radii of giant and main sequence stars located in the galactic bulge can be measured to a few percent accuracy . the method combines comprehensive ground - based photometry of caustic - crossing bulge microlensing events , with a handful of precise ( ) astrometric measurements of the lensed star during the event , to measure the angular radius of the source , . dense photometric coverage of one caustic crossing yields the crossing time scale . less frequent coverage of the entire event yields the einstein timescale and the angle of source trajectory with respect to the caustic . the photometric light curve solution predicts the motion of the source centroid up to an orientation on the sky and overall scale . a few precise astrometric measurements therefore yield , the angular einstein ring radius . then the angular radius of the source is obtained by . we argue that the parameters , and , and therefore , should all be measurable to a few percent accuracy for galactic bulge giant stars using ground - based photometry from a network of small ( 1m - class ) telescopes , combined with astrometric observations with a precision of to measure . we find that a factor of times fewer photons are required to measure to a given precision for binary - lens events than single - lens events . adopting parameters appropriate to the _ space interferometry mission _ ( sim ) , we find that minutes of sim time is required to measure to accuracy for giant sources in the bulge . for main - sequence sources , can be measured to accuracy in hours . thus , with access to a network of 1m - class telescopes , combined with 10 hours of sim time , it should be possible to measure to for giant stars , or to for main sequence stars . we also discuss methods by which the distances and spectral types of the source stars can be measured . a byproduct of such a campaign is a significant sample of precise binary - lens mass measurements . # 1 # 1equation ( [ # 1 ] )

astro-ph0208210

although of fundamental importance to stellar astrophysics , precise measurements of angular radii are generically difficult to acquire routinely and in a model - independent way . classical direct methods of measuring stellar radii include lunar occultations , interferometry , and eclipsing binaries . lunar occultation measurements yield precise angular radii ( see richichi et al . 1999 and references therein ) , but the number of stars to which this technique can be applied is limited . the number of direct measurements using interferometers has recently increased dramatically with advent of , e.g. the palomar testbed interferometer ( van belle et al . 1999a , colavita et al . 1999 ) , and the navy prototype optical interferometer ( armstrong et al . 1998 , nordgren et al . 1999 ) , and is likely to continue to increase as more technologically advanced interferometers come online . unfortunately , both lunar occultation and interferometric angular diameter measurements have traditionally been primarily limited to nearby , evolved stars . angular radii of main - sequence stars can be determined using detached eclipsing binaries ( i.e. popper 1998 ) , however the large amount of data ( both photometric and spectroscopic ) required to yield accurate radii determinations makes this method prohibitive . thus , of the @xmath17 direct , precise angular diameter measurements compiled by van belle ( 1999 ) , the overwhelming majority , @xmath18 , are of evolved stars . finally , it will be difficult to acquire a large sample of angular radii determinations of stars with metallicity considerably smaller than solar using these methods , due to the paucity of metal - poor stars in the local neighborhood . here we present a method , based on a suggestion by paczyski ( 1998 ) , of measuring angular radii of stars that overcomes some of the difficulties inherent in the classical methods . this method employs the extraordinary angular resolution provided by caustics in gravitational microlensing events , and as such is yet another in the growing list of applications of microlensing to the study of stellar astrophysics ( see gould 2001 for a review ) . the original suggestion of paczyski ( 1998 ) was to invert the method of gould ( 1994 ) for measuring the relative source - lens proper motion @xmath19 in microlensing events . if the lens transits the source in a microlensing event , precise photometry can be used to determine the time it takes for the lens to transit one source radius , @xmath20 , where @xmath1 is angular radius of the source . an estimate of @xmath1 , using an empirical color - surface brightness relation , together with a measurement of the flux of the source , can then be used to estimate @xmath19 , which gould ( 1994 ) argued could be used to constrain the location of the lens . however , as paczyski ( 1998 ) pointed out , it is possible to independently measure the angular einstein ring radius of the lens , @xmath21 by making precise astrometric measurements of the centroid shift of the source during the microlensing event using , i.e. , the _ space interferometry mission _ ( sim ) . here @xmath22 is the mass of the lens , @xmath23 is defined by , @xmath24 , and @xmath25 , @xmath26 , and @xmath27 are the distances between the observer - source , observer - lens , and lens - source , respectively . since @xmath28 , by combining the measurement of @xmath5 with the einstein timescale @xmath3 of the event determined from the light curve , it is possible to measure @xmath1 for the source stars of microlensing events . we show that , with reasonable expenditure of resources , it should be possible to measure angular radii of a significant sample ( @xmath29 ) of giant stars in the bulge to an accuracy of @xmath30 , or @xmath10 main - sequence stars to an accuracy of @xmath31 . limb - darkening determinations should also be possible for the majority of the sources , and most will be relatively metal poor as compared to those for which angular radii determinations are currently available . although measurements of @xmath1 can be made using single - lens events , in [ sec : bvs ] we argue that this method is better suited to caustic - crossing binary - lens events , which are more common , easier to plan for , and considerably less resource - intensive than source - crossing single - lens events . we describe in some detail the basic method of measuring @xmath1 for the source stars of caustic - crossing binary - lens events in [ sec : method ] , including a discussion of the expected errors on the individual parameters that enter into the measurement . we discuss various subtleties , complications , and extensions to the method in [ sec : discussion ] , and also present an estimate of the number of @xmath1 measurements that might be made in this way . finally , we summarize and conclude in [ sec : conclusion ] . the primary requirement to be able to measure @xmath5 in a microlensing event is that the source should be resolved by the gravitational lens . this effectively means that the source must cross a caustic in the source plane . caustics are the set of positions in the source plane where the determinant of the jacobian of the lens mapping from source to lens plane vanishes , and where the magnification therefore is formally infinite . large gradients ( with respect to source position ) in the magnification exist near caustics , enabling the resolution of the source . generically , microlenses come in two classes : single and binary lenses . here ` binary lens ' means a lens systems composed of two masses with angular separation of order the angular einstein ring radius of the system . very close and very wide binaries act essentially as single lenses . single lenses have a caustic that consists of a single point at the position of the lens . in these cases , the magnification close to the caustic diverges as the inverse of the distance to the caustic . in contrast , the caustics of binary lenses are extended , and can cover a significant fraction of the einstein ring . the caustics of binary lenses generically consist of two types of singularities , folds and cusps .. however , folds and cusps are the only stable singularities of any lens system ( petters , levine , & wambsganss 2000 ) . beak - to - beak singularities are unstable in the sense that , for infinitesimally small variations in the lens parameters , a beak - to - beak singularity disintegrates into two cusp - type singularities . thus the set of parameters where beak - to - beak singularities are expected is formally sparse , and practically small . interestingly , alcock et al . ( 2000a ) suggested that an observed event may have been due to a source crossing a beak - to - beak singularity in a binary - lens , although alcock et al . ( 2000b ) seem to favor the interpretation that this event is due to a background supernova . ] near a fold , the caustic is well - described by generic linear fold singularities , for which the magnification locally diverges inversely as the square - root of the distance to the caustic ( schneider & weiss 1986 , schneider , ehlers , & falco 1992 , gaudi & petters 2002a ) . cusps are points where two fold caustics meet , and the magnification for cusps locally diverges roughly inversely as the distance to the cusp point , similar to the magnification pattern near the point caustic of a single lens ( schneider & weiss 1992 , schneider , ehlers , & falco 1992 , gaudi & petters 2002b ) . the fact that the magnification near the point caustic of a single lens or near a cusp diverges inversely as the distance to the singularity , rather than as the square root of the distance to the singularity as with folds , means that for a given source size , the ` resolving power ' of fold caustic crossings is less than single lens or cusp crossings . although single - lens events are more well suited to studies of stellar atmospheres , they are much less useful for measuring the sizes of stars for three main reasons . they are generally rarer than fold caustic crossings , their crossings can not be predicted in advance , and their centroid motion is more difficult to measure . caustic - crossing binaries comprise roughly @xmath32 of all events toward the galactic bulge ( alcock et al . 2000a ; udalski et al . 2000 ) . of the remaining @xmath33 events , which we will conservatively assume due to single lenses , only a fraction @xmath34 will exhibit caustic crossings . therefore , the expected ratio of binary - to - single events for which the source is resolved ( and thus measurement of @xmath1 is possible ) is @xmath35 where @xmath36 is the physical radius of the source , and we adopted @xmath37 , @xmath38 , and @xmath39 for the scaling relation on the extreme right hand side . although not overwhelming for giant sources , for main sequence sources we expect at least an order of magnitude more binary lensing events for which the source is resolved . the ratio of the number of fold - to - cusp crossings is roughly @xmath40 where @xmath41 is the number of cusps , which is either 6 , 8 , or 10 for a binary lens , depending primarily on @xmath42 . thus the overwhelming majority of events for which it will be possible to measure @xmath1 will be fold caustic crossing binary - lens events . binary - lens fold caustic - crossing events also have the advantage that the second caustic crossing can be predicted in advance . this is because fold caustic crossings always come in pairs , and it is typically easy to tell , even with sparse sampling , that the first caustic crossing has occurred . then more frequent sampling can be used to monitor the rise to the second caustic , in principle enabling the prediction of the time of the second crossing a day or more in advance ( jaroszy ' nski & mao 2001 ) , and allowing the marshaling of the resources necessary to obtain the dense coverage of the second crossing needed to measure the crossing time @xmath2 of the source ( see [ sec : dt ] ) . in contrast , a single lens caustic crossing can only reliably be ` predicted ' at about the time it begins . finally , it is considerably harder to measure @xmath5 for caustic - crossing single - lens events than binary - lensing events . single lens events have a maximum absolute centroid shift relative to the unlensed source position of @xmath43 , whereas binary lensing events can exhibit large variations of size @xmath44 or more when the source crosses the caustic ( han , chung , & chang 1999 ) . therefore considerably more time will generally be required to determine @xmath5 to given accuracy for single - lens events than for binary - lens events . since the astrometric measurements essentially require the capabilities of sim ( or some similarly precious instrument ) , it is highly desirable to minimize the amount of time spent on this step . given the above arguments , we conclude that fold caustic crossing binary lensing events are the most suitable for use in routinely measuring @xmath1 . we will therefore focus on this case for the remainder of the discussion , however we will briefly revisit single - lens and cusp - crossing events in [ sec : cusp ] . consider a binary - lens event in which the source crosses a simple linear caustic . , and when the angle of incidence of the source trajectory to the caustic is not small . this approximation will break down when the source is large compared to the overall size of the caustic , when the source crosses near a cusp , or when the source ` straddles ' the caustic for a long time due to a small incidence angle . although it will still be possible to measure @xmath1 for such events , the relation between the observables and @xmath1 is less straightforward . we discuss such cases in [ sec : cusp ] . ] defining @xmath2 as the ( one - half ) the time it takes for the source to completely traverse the caustic , and @xmath4 as the angle between the source trajectory and the tangent to the caustic at the crossing point , then the time for the lens to cross the angular radius of the source is @xmath45 . however , we also have that @xmath46 , where again @xmath19 is the relative source - lens proper motion . combining these expressions , we have that @xmath47 . using the definition of @xmath19 , we can write the angular source radius @xmath1 as the following function of observables , @xmath48 the process of @xmath1 measurement can therefore be subdivided into three basic steps : ( 1 ) : : measurement of the caustic crossing timescale @xmath2 from a single photometrically well - resolved caustic crossing . ( 2 ) : : measurement of the angle @xmath4 and timescale @xmath3 from the global fit to the binary - lens light curve . ( 3 ) : : measurement of the angular einstein ring radius @xmath5 using precise astrometric measurements of the source centroid . each of these steps are illustrated schematically in figure 1 . in the following subsections , we consider each of these steps in more detail . we outline the basic requirements for the measurement of each of the four parameters ( @xmath49 , and @xmath5 ) , and the expected accuracy with which each can be determined assuming reasonable expenditure of observing resources . when the source is interior to a binary - lens caustic , five images are created . as the source approaches a fold caustic , two of these images brighten and merge in a characteristic way , and eventually disappear when the source completely exits the caustic . in contrast to the significant brightening of the two images associated with the fold , the remaining three images generally vary only slowly over the timescale of the crossing . all fold caustics locally have this generic behavior , and the magnification @xmath50 of the source near a caustic crossing as a function of time is typically well - fitted by the functional form ( albrow et al . 1999b ) , @xmath51 where @xmath52 is the effective rise time of the caustic , which is related to the local derivatives of the lens mapping ( petters et al.2001 , gaudi & petters 2002 ) , @xmath53 is the time when the center of the source crosses the caustic , @xmath54 is the magnification of all the images unrelated to the fold caustic at @xmath55 , @xmath56 is the slope of the magnification of these images as a function of time , and @xmath57 can be expressed in terms of complete elliptic integrals of the first and second kind ( schneider & weiss 1987 ) . note that is only formally appropriate for a simple linear fold caustic . thus , for a well - sampled fold caustic crossing , @xmath2 can be determined essentially independently of the global geometry of the event , and indeed without reference to the photometric data away from the crossing itself . in practice , the magnification is not directly observable , but rather the flux @xmath58 as a function of time . the form for @xmath58 takes on a similar form as , but with a slightly different parameterization ( see albrow et al . 1999b ) . equation ( [ eqn : foft ] ) assumes a uniform source . this will likely be a poor approximation in optical bands , and assuming uniform source in the presence of limb - darkening may result in a systematic underestimate of @xmath2 , and therefore @xmath1 since , the effect of limb - darkening can partially compensated for by a smaller ( dimensionless ) source size , at least for poorly - sampled light curves . however , for well - sampled caustic crossings and @xmath59 photometric accuracy , it should generally be possible to accurately measure both the source size and limb - darkening coefficient(s ) ( rhie & bennett 1999 ) . such data quality is readily achievable ; indeed , independent limb - darkening and ( dimensionless ) source size measurements have been made for the source stars of at least five microlensing events ( albrow et al . 1999a , afonso et al . 2000 , albrow et al . 2000 , albrow et al . 2001 , an et al . a generalized form of that includes simple limb - darkening can be found in albrow et al . ( 1999b ) and afonso et al . ( 2001 ) . by fitting ( or a generalized form of it ) to the light curve near a well - sampled fold caustic crossing , one can derive the parameters @xmath60 , and @xmath2 . the parameters @xmath61 , and @xmath52 can subsequently be used to constrain the global solution to the entire light curve ( see albrow et al . however , of primary interest here is the parameter @xmath2 , whose value is essentially independent of the global solution . see figure 1(b ) . this means that @xmath2 and the parameters determined from the global solution , @xmath3 and @xmath4 , will be essentially uncorrelated . in order to be able to determine @xmath2 from the caustic crossing data alone , the caustic crossing must be well - sampled , so that the parameters in be well - constrained . practically , this requires forewarning of the caustic crossing . fortunately , this is generally possible with only photometry available from the collaborations which survey the galactic bulge and find the alert the microlensing events real - time ( eros , afonso et al . 2001 ; moa , bond et al . 2001 ; ogle , udalski et al . 2000 , woziak et al . 2001 ) , although improved predictions would be possible with continuous photometry ( jaroszy ' nski & mao 2001 ) . thus no additional resources need to be invested to predict caustic crossings . however , as we discuss in [ sec : teandphi ] , some additional sampling of the overall light curve may be needed to constrain @xmath3 and @xmath4 and determine a unique global solution . the accuracy with which @xmath2 can be determined for a given caustic crossing will depend not only on the intrinsic parameters of the caustic crossing , but also on the sampling rate and photometric accuracy near the caustic crossing , which in turn will depend on weather , blending , etc . , which tend to vary in a stochastic manner . therefore it is not very useful to attempt to quantify the expected errors on @xmath2 for an idealized observing setup . however we can obtain an order - of - magnitude estimate for @xmath62 by examining measurements of @xmath2 from published analyses of observed caustic - crossing events . we discuss these determinations more thoroughly in [ sec : errors ] . typically , well - covered caustic crossings yield fractional errors of @xmath63 . the parameters @xmath3 and @xmath4 are determined from the global solution to the overall geometry of the binary - lens light curve . the relationship between the salient features of an observed light curve , and the canonical parameters of a binary - lensing event , are generally not obvious or straightforward . this fact generally makes fitting an observed light curve , and thus inferring the parameters @xmath3 and @xmath4 , quite difficult ( see albrow et al . 1999b for a thorough discussion ) . although many methods have been proposed to overcome these difficulties , the lack of obvious correspondence between these parameters of interest and the light curve morphology generally implies that it is difficult to make general statements about the kinds of observations that are needed to reliably measure @xmath3 and @xmath4 . however , we can make some generic comments . for caustic - crossing binary lenses , one potential observable is the time @xmath64 between caustic crossings . between caustic crossings , the magnification is typically considerably larger than outside the caustic , and therefore even with sparse ( @xmath65 day ) sampling , it should be possible to determine to reasonable precision the time of the first caustic crossing , provided that data exist before the first crossing . the requisite photometry will generally be acquired by the survey collaboration(s ) . once the binary - lens event is alerted , more frequent photometry can be acquired by follow - up collaborations ( planet , albrow et al . 1998 ; mps , rhie et al . 2000 ) , thus mapping the `` u''-shaped curve between the caustic crossings . this shape , combined with information from cusp - approaches ( or lack thereof ) just outside the caustic , provides information about the shape of the caustic and the trajectory of the source through it . from this , the angle @xmath4 of the trajectory with respect to the caustic can be derived , and also the distance @xmath66 between the caustic crossings in units of @xmath5 . then , the timescale is given by @xmath67 . although the above analysis is highly trivialized , it does suggest that the following steps should be taken to ensure an accurate measurement of @xmath3 and @xmath4 . first , it is important to constrain the time of both caustic crossings to reasonable precision . this means that the survey collaborations should sample on timescales no less than a few days . second , followup photometry should be initiated relatively soon after the first crossing , to measure the shape of the intra - caustic light curve reasonably well . this is also necessary in order to predict the second caustic crossing ( jaroszy ' nski & mao 2001 ) . furthermore , the follow - up photometry should continue past the second caustic crossing , to detect cusp - approaches or cusp - crossings ( or the lack thereof ) . due to the difficulties inherent in fitting binary - lens light curves , it would be extremely difficult to attempt to predict the expected errors on @xmath3 and @xmath4 for a hypothetical observing scenario . furthermore , due to the complicated relation between observables and parameters implies that these errors are likely to depend strongly on the geometry of the event and light curve coverage , and therefore such ` predictions ' would not be very useful . however , we would like to have an order - of - magnitude estimate for the expected errors given reasonable light curve coverage . from determinations of these parameters in published caustic - crossing events , we can expect fractional accuracies of a few percent , provided that the light curve is well - covered , in the sense outlined above . however , if the light curve coverage is incomplete , then errors of @xmath68 are expected . see [ sec : errors ] . cccccccc & & & & & & & + macho 98-smc-1 & @xmath69 & @xmath70 ( @xmath71 ) & ... & @xmath72 & 22.1 & 1,2 & fold crossing + ogle-1999-bul-23 & @xmath73 & @xmath74 ( @xmath75 ) & ... & @xmath76 & 18.1 & 3 & fold crossing + macho 95-blg-30 & ... & ... & @xmath77 & ... & 13.4 & 4 & single lens + macho 97-blg-28 & ... & ... & @xmath78 & ... & 15.6 & 5 & cusp crossing + macho 97-blg-41 & ... & ... & @xmath14 & ... & 16.8 & 6 & rotating binary + eros 2000-blg-5 & ... & ... & @xmath79 & ... & 16.6 & 7 & parallax effects + fractional errors on parameters from observed events . + [ tbl : table1 ] a global solution to the entire light curve effectively requires the specification of the vector position of the source @xmath80 as a function of time in units of @xmath5 , and the topology of the lens , i.e. the mass ratio @xmath81 and projected separation @xmath42 in units of @xmath5 . these parameters yield not only the total magnification @xmath82 of all the images as a function of time . but also the individual image positions @xmath83 and magnifications @xmath84 as a function of time , therefore , it is also possible to predict the centroid @xmath85 of all the individual microimages , @xmath86 note that @xmath85 is the centroid shift with respect to the lens position . is it customary to consider the centroid shift with respect to the unlensed source position , @xmath87 . the unlensed source position @xmath88 , which is comprised of the parallax and proper motion of the source in some astrometric frame , can be determined via measurements of the unlensed motion of the source . since the astrometric effects falls off very slowly ( as @xmath89 ) , these must be obtained many @xmath3 after the event is over . in fact , such measurements are not strictly needed in order to measure @xmath5 ( but may be desirable for other reasons , see [ sec : dos ] ) . rather , one can simultaneously fit astrometric measurements during the course of the event for the relative position and proper motion of the source ( in order to establish a local astrometric reference frame ) , and the offset induced by microlensing . the centroid shift @xmath85 is in units of @xmath5 , and its components are oriented with respect to the projected binary - lens axis , whose orientation @xmath90 on the sky is unknown . the observable centroid is @xmath91 alternatively , one can combine @xmath90 and @xmath5 and simply consider the vector @xmath92 . thus the global solution to the photometric light curve yields not only @xmath3 and @xmath4 , but also yields a prediction for the astrometric curve , up to an unknown orientation and scale @xmath5 .. ] therefore , by making a series of astrometric measurements @xmath93 at several different times during the course of the event , one can determine @xmath5 and @xmath90 via , and using the predictions for @xmath94 from the photometric solution . the accuracy with which @xmath5 can determined will depend on the geometry of the event , the time of the astrometric measurements , and the time span between the measurements . furthermore , as we discuss , there are , in reality , + additional parameters that must be determined from the astrometric data . we therefore perform a monte carlo simulation to estimate the accuracy with which @xmath5 can be recovered . we explore whether the various parameters can be measured independently , or are degenerate with the measurement of @xmath5 , and study the range of fractional uncertainties in measuring @xmath5 for a full ensemble of binary lenses . our simulation consists of the following elements : a galactic model and model of the ensemble of lenses to generate an ensemble of microlensing events , an observational strategy , and a fit of these observations to a set of parameters including @xmath5 . our ensemble of lenses is identical to that of graff & gould ( 2002 ) . briefly , we draw sources and lenses from self - lensing isothermal sphere 8 kpc from the observer , and with two - dimensional velocity dispersion of @xmath95 . both masses in the binary lens are chosen from the remnant mass function of gould ( 2000 ) . we pick a flat distribution in @xmath96 , the logarithm of the dimensionless binary separation . the path of the source through the lens geometry is chosen randomly with a uniform distribution of angular impact parameter @xmath97 , and we only consider paths that cross caustics . , but are weighted towards large @xmath5 events by multiplying the mass function by @xmath98 . ] as discussed in graff & gould ( 2002 ) , in this ensemble of events there are many more events with a short time between caustic crossings , the _ caustic interior time @xmath99 _ , than is observationally detected by the macho and ogle observing groups . this led these authors to suggest that most events with a short caustic interior time are not detected as caustic crossing binaries , and to define a caustic crossing detection efficiency @xmath100 akin to the standard single lens detection efficiency @xmath101 . the observational strategy is relatively unimportant , as long as there are astrometric measurement on either side of a caustic . although it is conceivable that observations might be scheduled at particularly favorable times , such as times of maximum magnification or maximum displacement of the image centroid , it is likely that the telescope measuring the astrometric displacement will be oversubscribed . a simpler strategy would be to schedule periodic observations in advance . we have assumed that observations will be made every four days , with a 24 hour delay after the event is recognized as a caustic crossing binary , i.e. , after the first caustic crossing . that is , the first observation comes @xmath102 days after the first caustic crossing . we assume observations are made for a total of 36 days , which corresponds to approximately @xmath103 for the median event timescale . practically , observations should continue until after the second caustic crossing , and the total number of observations should be at least as large as the number of parameters to be constrained . as we discuss in [ sec : errors ] ( see also table 1 ) , well covered binary lenses can be fit photometrically with small errors on the parameters . thus , we have assumed that all the parameters which can be fit from a single photometric telescope are determined . given a binary microlensing event from our ensemble , we use our observational strategy to create a series of photometric and astrometric measurements , respectively @xmath50 and @xmath93 which we can combine into a single list of measurements @xmath104 each with uncertainty @xmath105 . using the fisher matrix technique , e.g. , gould & welch ( 1996 ) , we determine the covariance matrix @xmath106 of the errors @xmath107 here the @xmath108 are the various parameters being fit . the error in parameter @xmath108 is simply @xmath109 . we assumed that the photometric uncertainty , @xmath105 , of the interferometric telescope , is photon - noise - dominated , and that the total telescope time , aperture , efficiency , filter width , and source brightness are such that a total of @xmath110 photons would be detected from an unmagnified source for a total photometric signal to noise of 250 . and diameter @xmath111 collects @xmath112 photons per second at @xmath113 . ] we assume that the fractional photometric accuracy is simply @xmath114 , and that the astrometric uncertainty is @xmath115 , where @xmath116 is the width of the point spread function , or in the case of an interferometer , the fringe separation . here we have assumed @xmath117mas . it is trivial to scale our results to brighter sources or larger telescopes : the fractional uncertainty in @xmath5 is simply proportional to @xmath114 . we always fit for the four parameters required to establish a local astrometric frame , and @xmath5 and @xmath90 , the size scale and orientation of the microlensing excursion . we also assumed that the satellite which measures the astrometric motion is 0.2 au from the ( ground based ) photometric measurements which fix the lens parameters . thus , we can simultaneously fit for @xmath118 , the projected einstein ring radius , in the manner of graff & gould ( 2002 ) . in addition to this basic fit , we also considered blending from luminous lenses and binary sources , which requires several additional parameters . we discuss parallax and blending in [ sec : parallax ] and [ sec : lumlens ] , respectively . our basic results are summarized in the leftmost curve in figure 2 . we see that in the absence of blending , @xmath5 can be determined with @xmath119 uncertainty in 85% of events . the median error is @xmath120 . this is comparable to the uncertainty found by gould & salim ( 1999 ) for single star events , but with 50 times as many photons as we have assumed here . thus , it is much easier to measure @xmath5 for caustic crossing binary lenses than for single lenses . from the expression for @xmath1 ( eq . [ eqn : thetas ] ) , and assuming the errors in @xmath5 , @xmath3 , @xmath2 , and @xmath4 are small and uncorrelated , the fractional error in @xmath1 is given by , @xmath121^{1/2 } \label{eqn : errths}\ ] ] where @xmath122 , @xmath123 , @xmath62 , and @xmath124 are the uncertainties in @xmath5 , @xmath3 , @xmath2 and @xmath4 , respectively , and @xmath124 is in radians . to date , there exist in the published literature 10 microlensing events for which the source star was well - resolved . unfortunately , for four of the events , those presented in alcock et al . ( 2000a ) , no estimate of the errors of the derived parameters is given . we therefore can not use these events to explore the expected magnitudes of @xmath123 , @xmath62 , and @xmath124 . for the six events for which errors on the relevant fit parameters were given or derivable , only two of them are generic binary - lens fold caustic crossing events . the errors @xmath123 , @xmath62 , and @xmath124 for these two events , macho 98-smc-1 and ogle-1999-bul-23 , are presented in table 1 . for macho 98-smc-1 , these errors have been determined from the ensemble of solutions presented in albrow et al . ( 1999b ) , which were fits to the planet collaboration photometry , which only covered the last half of the event , whereas the analysis of the combined photometry of the eros , macho , mps , ogle , and planet collaborations ( afonso et al . 2000 ) yields considerably smaller errors . we concentrate on the results of albrow et al . ( 1999b ) here in order to demonstrate the kinds of errors that result from incomplete light curve coverage . note that @xmath125 , which is not surprising , since the accuracy with which @xmath2 can be determined depends almost exclusively on the photometric coverage near the caustic crossing . however , the errors on the global parameters @xmath3 and @xmath4 are quite large , @xmath126 and @xmath127 . this is due to the fact that the planet photometry only covered the latter half of the event , and contained no data prior to or during the first caustic crossing . therefore the global geometry was quite poorly constrained with their data alone . such incomplete coverage would clearly jeopardize a precise measurement of @xmath1 . this is in contrast to ogle-1999-bul-23 , for which both planet and ogle obtained data before the first caustic crossing . in this case , it was possible to measure @xmath2 , @xmath3 , and @xmath4 to better than @xmath76 ( albrow et al . 2001a ) . the remaining four include a single lens event ( macho 95-blg-30 ; alcock et al . 1997 ) , a binary - lens event in which the source crossed a cusp ( macho 97-blg-28 ; albrow et al . 1999a ) , a binary - lens event in which the rotation of the binary was detected ( macho 97-blg-41 ; albrow et al . 2000 ) , and a binary - lens event for which both rotation and parallax effects were detected ( eros blg-2000 - 5 ; an et al . 2002 ) . in general , the previous discussion is not directly applicable to these events , as the information in the light curves is not easily decomposed into the parameters @xmath2 , @xmath4 and @xmath3 . nevertheless , in all four cases the dimensionless source size @xmath128 was determined . in three of the cases , @xmath129 . in one case , macho 97-blg-41 , @xmath130 , due primarily to the fact that there exists only a handful of data points in which the source was resolved . thus , although it is difficult to draw any general conclusions from these unique events , it is does seem likely that errors of @xmath131 are achievable for most types of events in which the source is resolved . also presented in table 1 are the determined values of the @xmath132-magnitude of the source , for all six events . at @xmath133 , the source star for macho 98-smc-1 would be too faint to target with most upcoming interferometers , including sim . the other events are primarily bulge clump giants ( @xmath134 ) , for which accuracies of @xmath135 should be achievable with a reasonable amount of exposure time with upcoming interferometers , and in particular with @xmath136 minutes of sim time ( see [ sec : number ] for a estimate of the required exposure times for sim ) . in these cases , the expected error on @xmath1 is typically dominated by @xmath122 , and therefore we can expect @xmath137 . the source of ogle-1999-bul-23 is g / k subgiant ( @xmath138 ) , and thus dimmer . the time required to achieve an accuracy of @xmath139 will be larger by a factor of @xmath140 , or @xmath141 for sim . our goal in [ sec : method ] was to capture the essence of the method of measuring @xmath1 , and the discussions were therefore somewhat oversimplified , and glossed over several important points . in particular , we concentrated on fold caustic crossing binary - lens events toward the bulge , whereas measurements of @xmath1 should be possible in other , rarer , types of events , such as cusp - crossings and single - lens events , and possibly events toward the magellanic clouds . we also ignored various higher - order effects which could , in principle , complicate the measurements . we therefore briefly discuss some of these complications and extensions . we also discuss the prospects for measuring the spectral type of the source , and also its distance , in order to convert from angular radius @xmath1 to physical radius @xmath36 . finally , we present an example observing campaign aimed at measuring angular radii for a significant sample of sources , outlining the resources required , and estimating the number of @xmath1 measurements that might be made per year for such a campaign . although we have focussed on fold caustic - crossing binary - lens events toward the bulge , it is important to emphasize that @xmath1 can , in principle , be measured for other types of caustic - crossing events such as cusp - crossing events , single - lens events , and all types of caustic - crossing events toward the magellanic clouds . indeed , in [ sec : errors ] we discussed examples in the literature of a single - lens and two cusp - crossing events for which a @xmath11 measurement of @xmath1 would have been feasible . in general , isolated cusp crossings , such as in macho 98-blg-28 , can not be predicted in advance , and thus planning for such events is difficult , if not impossible . however , one will still have some advance warning of those cusp events which occur just after or in place of second fold caustic crossings . for such events , sufficient photometric coverage of the crossing should routinely be possible . in all cases , it is more difficult to disentangle the information arising from the cusp itself with the information from the global light curve . this generally implies that the analysis of these light curves will be more complicated , however this does not necessarily preclude an accurate measurement of @xmath1 . single - lens events are less desirable simply because they require a factor of @xmath9 times more astrometric observing time to achieve the same fractional accuracy in @xmath5 as binary - lens events . since the astrometric observations are likely to be the most limited resource , this makes single lens events considerably less attractive . if it were possible to measure angular radii of stars in the magellanic clouds ( mcs ) , this would be quite interesting , due to the metal - poor nature of the stars . unfortunately , there are several major hindrances to measuring @xmath1 for a substantial number of stars in the mcs . first , the event rates toward both the mcs are small , and a large number of stars must be monitored just to detect a few events per year . therefore , the number of caustic - crossing events is quite low . to date , there have been only two caustic crossing events toward the mcs : macho 98-smc-1 , which we discussed in [ sec : errors ] , and macho lmc-9 . these events have source magnitudes of @xmath142 ( afonso et al . 2000 ) and @xmath143 ( alcock et al . 2000a ) , respectively , which brings up a second difficulty : sim can not follow source stars fainter than @xmath144 , so these two events could not have been used to measure the angular radii of their source stars . in fact , even if the entire lmc were monitored for microlensing , only @xmath65 event per year would have @xmath145 , and this event would be from an evolved star . the probability of a caustic - crossing event ( either binary or single lens ) is smaller by at least an order of magnitude . the paucity of events and faintness of the source stars might be circumvented if sufficiently rapid target of opportunity times are available . in this case , it might be possible to use intrinsically fainter source stars , for which caustic - crossing event will be more common , and measure the astrometric displacement during the brief period of time when the source is highly magnified as it crosses the caustic . the maximum magnification of a source of dimensionless size @xmath146 crossing a fold caustic is @xmath147 . for main - sequence sources , @xmath148 , or more than three magnitudes , and thus sources with @xmath149 can briefly be brightened to sim detectability . for example , the source star of macho 98-smc-1 was brighter than @xmath150 for about 7 hours during the second caustic crossing . finally , even if the source does attain a sufficient brightness to be measurable by sim , it remains to be seen whether the centroid varies sufficiently during this time to provide an accurate measurement of @xmath5 . this is especially difficult in light of the fact that typical value of @xmath5 for self - lensing events toward the mcs are only an order of magnitude larger than sim s accuracy ( paczy ' nski 1998 , gould & salim 1999 ) . in summary , it appears that it will be quite difficult to measure angular radii of stars in the mcs using this method , especially if the majority of the events seen toward these targets are due to self - lensing ( sahu 1994 ) . the method we have presented here is only interesting if it can feasibly be used to make precise @xmath1 measurements for a large number of sources with reasonable expenditure of resources . since the requisite astrometric instruments are likely to be the most limited resource , it is crucial that accurate and unambiguous determinations of @xmath5 be generically possible using a few astrometric measurements , when combined with the photometric light curve solution . we have explained how a complete photometric solution _ generally _ leads to a prediction for the astrometric centroid shift up to an unknown scale @xmath5 and orientation @xmath90 on the sky . however , this is true only under a number of simplifying assumptions , including uniform motion of the observer , source , and lens , dark lenses , isolated sources , and unique global solutions . if one or more of these assumptions are violated , then the prediction for shape of the astrometric curve may not be unique , and thus the measurement of @xmath5 may be compromised . we therefore discuss each of these complications and under what conditions they may be important . binary lenses are characterized by two quantities : @xmath81 , the mass ratio , and @xmath42 , the instantaneous projected separation in units of @xmath5 . it has been demonstrated both theoretically ( dominik 1999a ) and observationally ( afonso et al . 2000 , albrow et al . 2002 ) that certain limiting cases of binary lenses can exhibit extremely similar observable properties . in particular , dominik ( 1999 ) showed that the binary - lens equation can be approximated by an single lens with external shear , or chang - refsdal ( cr ) lens ( chang & refsdal 1979,1984 ) , near the individual masses for widely - separated binaries ( @xmath151 ) , and near the secondary ( least massive ) lens when @xmath152 . furthermore , near the center - of - mass of a close binary , the lens equation is well - approximated by a quadrupole lens , and both the quadrupole lens and cr - lens can exhibit extremely similar magnifications when the quadrupole moment is equated to the shear ( albrow et al . 2002 ) . thus there can exist multiple degenerate solutions to an observed photometric light curve , even with extremely accurate photometry . however the astrometric behavior of these degenerate solutions is very similar both in the shape and overall scale of the astrometric curves , at least for the close / wide degeneracy ( gould & han 2000 ) . therefore this degeneracy should not affect the determination of @xmath5 using the prediction from the light curve . it is likely that the other intrinsic denegeracies will also not affect the determination of @xmath5 , since the degeneracy arises from the lens equation itself , and thus affects both the photometric and astrometric curve in the same manner . note that it is important that the normalization of @xmath5 be consistent for the two degenerate solutions . for example , consider the case of the close / wide degeneracy in macho 98-smc-1 ( afonso et al 2000 , gould & han 2000 ) . if one normalizes to the total mass of the binary , the close solution implies a value of @xmath153 , whereas the wide solution has @xmath154 . since the astrometric curves are essentially identical ( both in shape and scale ) , one might therefore suspect the inversion of this process would yield two equally likely values of @xmath1 that differed by a factor of @xmath155 of course , this ` ambiguity ' is wholly artificial , and arises because the value of @xmath156 for the wide binary is normalized to the entire mass of the binary , whereas the lensing effects are basically caused by the least massive lens , since @xmath157 . normalizing to the mass of the single lens , @xmath158 , where here @xmath159 , and thus @xmath160 , essentially identical to the close - binary solution . note that as @xmath42 approaches unity , the identification of the ` proper ' @xmath5 normalization becomes more nebulous , since the lenses can no longer be considered independent . however , the degeneracies also become less severe as @xmath161 . dominik ( 1999b ) has also shown that poorly - sampled binary lens light curve can also yield distinct degenerate solutions . note that these solutions are ` accidental ' in the sense that they do not arise from degeneracies in the lens equation itself . thus han et al . ( 1999 ) found that such degenerate photometric light curves yield astrometric curves which are widely different . such degeneracies would prohibit the measurement of @xmath5 using a few astrometric measurements . therefore well - sampled photometric light curves are essential for reliable measurements of @xmath1 . if the two observers are displaced by a significant fraction of @xmath162 , the angular einstein ring radius projected onto the observer plane , then the source position @xmath88 as seen by the two observers will be significantly different . since sim will be in an earth - trailing orbit , it will drift away from the earth at a rate of @xmath163 . thus after 2.5 years ( half way through the sim mission ) , it will be displaced from the earth by @xmath164 , which corresponds to a displacement in the einstein ring of , @xmath165 where @xmath166 is the angle between the the line of sight and the earth - sim vector . for typical bulge parameters , @xmath167 and therefore @xmath168 . this implies that both the magnification @xmath82 and the centroid shift @xmath169 will be significantly different ( at a given time ) as seen from the earth and sim . since the value of @xmath170 is not known _ a priori _ , @xmath171 can not be predicted from the ground - based photometry alone , and must be estimated from the astrometric data itself . fortunately , sim will likely have excellent photometric capabilities ( see gould & salim 1999 ) , and thus the relative magnifications between the light curves from the ground and sim will provide additional constraints . indeed , in our monte carlo simulations , we assumed that the astrometric observer was displaced by @xmath172 from the earth , and simultaneously fit for both @xmath5 and @xmath170 ( among other parameters ) , and found that @xmath5 could still be constrained quite accurately . this is because the information about @xmath170 comes primarily from the photometry , while the information about @xmath5 comes primarily from the astrometry . therefore , the two parameters are not degenerate . note that a ` byproduct ' of these measurements is a determination of the total mass of the binary lens ( gould & salim 1999 ; han & kim 2000 ; graff & gould 2002 ) there are two additional parallax effects . one is due to the motion of the earth ( or sim ) around the sun , and will become significant on timescales that are a substantial fraction of a year , which corresponds to many @xmath3 for typical bulge events . there is also a second order effect that arises from the difference in projected velocities between the earth and sim . this effect is @xmath173 , where @xmath174 is the transverse velocity of the lens projected on the observer plane , and is @xmath175 for typical bulge self - lensing events . since the velocities and positions of the earth and sim will be known , both of these effects can easily be included in the fit for the microlensing parallax , and so do not present any additional difficulties . with its planned 10 meter baseline , sim will have a resolution of @xmath176 , sufficient to resolve the majority of unassociated nearby stars that are blended with the source in ground - based photometry ( han & kim 1999 ) . since the photometric blending is well - constrained in binary - lens events , unambiguous prediction of the unblended astrometric behavior of the source is possible . thus , blending will typically not affect the measurement of @xmath5 . however , luminous lenses and companions to the source star with separations @xmath177 will not be automatically resolved by sim ( jeong , han & park 1999 ) . dalal & griest ( 2001 ) have shown that , using two pointings , this limit may be lowered to @xmath178 , however it is essentially impossible to resolve multiple sources with separations below this limit ( e.g. binary source companions ) . in these cases , all that will be measured is the total centroid of all the sources in the resolution element . the centroid in the presence of luminous lenses , @xmath179 , is related to the centroid in the absence of blending , @xmath85 , by @xmath180 \left(1 + \frac{f_t}{a}\right)^{-1 } , \label{eqn : dbpclt}\ ] ] where @xmath181 is the sum of the flux of all unlensed sources ( blends ) relative to @xmath182 , the baseline flux of the lensed source , @xmath82 is the magnification of the source , and @xmath183 is the centroid of the blends relative to the origin of the lens . from , it is clear blending is more complicated in astrometric microlensing than in photometric microlensing : whereas photometric blending can be described by one parameter , the blend fraction @xmath184 , astrometric blending requires two additional parameters , the centroid of light of the blend @xmath183 . a special case of astrometric blending is bright lens blending . in the single lens case , this eliminates the blend location parameters , the location of the centroid of light is the moving lens ( i.e. @xmath185 ) . in bright binary lens blending , only one parameter is eliminated , the centroid of light is somewhere on the lens axis between the two stars in the lens . however , it will generally not be known _ a priori _ which case one is dealing with , and therefore @xmath183 must be included as a parameter in the astrometric fit . blending is problematic because it effectively ` dilutes ' the astrometric shift between two points in the lightcurve , which is qualitatively similar to the effect of changing @xmath5 . if the event is not well covered , these two effects can be quite degenerate . in order to determine how degenerate blending is with the @xmath5 , we have included in our monte carlo simulations a fixed amount of blended light of @xmath186 , and @xmath187 . we assume that @xmath184 is known ( i.e. from photometry ) , but @xmath183 is not . the results are shown in figure 2 . we find that , if the blending fraction is close to 1 , then the two effects are nearly degenerate , and our fractional uncertainty in @xmath5 increases by two orders of magnitude to of order unity . however , for @xmath188 , the median error increases by less than a factor of two . in most cases , the blending will be known to be small from the photometric data . in these cases , the fractional uncertainty in @xmath5 will not be seriously degraded . the few events with known large blending can be easily jettisoned from the sample . the photometric effects of lens rotation in binary microlensing events has been explored theoretically by dominik ( 1998 ) and has been detected in event macho 1997-bul-41 ( albrow et al . the astrometric effects of rotating binary - lenses have not been explored , and it is therefore difficult to draw any general conclusions as to the importance of this effect . however , to the extent that it is detectable in the photometric light curve , lens rotation poses no difficulties , as its astrometric effect should be predictable from the global solution . effects that are photometrically undetectable but astrometrically significant are potentially problematic . the amount that a binary - lens rotates during @xmath3 is given by , @xmath189 where @xmath190 , and assuming circular , face - on orbits . because the caustic cross - section is maximized for binaries with separations of order @xmath5 , the majority of detected caustic - crossings will have @xmath191 ( baltz & gondolo 2001 ) . therefore , for typical events , the effects of binary rotation should be small if astrometric observations are closely spaced with respect to the event timescale @xmath3 . this is generally also advantageous for the accurate recovery of @xmath5 ( see [ sec : thetae ] ) . in this paper , we have emphasized the measurement of the angular radius @xmath1 , rather than the physical radius , @xmath36 . however , it may also be interesting to measure @xmath36 for some events . in order to do this , the distance to the source star must be measured independently . fortunately , the astrometric accuracy needed to measure @xmath1 is generally sufficient to measure the parallax @xmath192 of the source stars , @xmath193 in order to measure @xmath36 to a similar accuracy as @xmath1 ( @xmath11 ) , @xmath192 must be measured to somewhat better accuracy , which implies an astrometric error of @xmath194 . for sim and an @xmath113 source , this is achievable with @xmath195 hours of integration , which is considerably more time than is needed for the @xmath1 measurement alone . however , it is important to note that these measurements can be made after the microlensing event is over . therefore , it should be possible to employ ground - based interferometers for the measurement of @xmath192 , rather than spend precious sim resources , at least for brighter sources . a measurement of @xmath1 is essentially useless if the spectral type and luminosity class of the source is not known . the source stars of microlensing events can be typed in two ways . the first is to simply measure the color and apparent magnitude of the source . this information is generally acquired automatically from the fit to the photometric data of the microlensing event . by positioning the source star on a color - magnitude diagram of other stars in the field , one can generally type the source to reasonable accuracy . the primary pitfalls of this method are differential reddening and projection effects ( i.e. the source may in the foreground or background of the bulk of the stars in the field ) . a more robust way of typing the source star is to acquire spectra . this is best done when the source is highly magnified as it crosses a caustic , as this minimizes the effects of blended light and increases the signal - to - noise . thus such measurements require target - of - opportunity observations . for highly - magnified events , spectra with @xmath196 per resolution element can be achieved with exposure times of tens of minutes for low - resolution spectra ( lennon et al . 1996 ) , or a couple of hours for high - resolution spectra ( minniti et al . 1998 ) , using 8 m or 10m - class telescopes . although low - resolution spectra are sufficient for accurate + spectral typing , high resolution spectra are desirable for a number of other applications , including resolution of the atmosphere of the source star ( gaudi & gould 1999 , castro et al . 2001 , albrow et al . 2001b ) , detailed abundance analysis ( minniti et al . 1998 ) , and detection of a luminous lens ( mao , reetz , & lennon 1998 ) . note also that as a byproduct , true space velocities of a sample of stars in the bulge will be obtained by combining the proper motions and parallaxes of the sources acquired from astrometric measurements with radial velocities determined from the spectra . in this section , we review the requirements for measuring @xmath1 for the source stars of galactic bulge microlensing events , and outline the resources needed for a campaign aimed at measuring @xmath1 for a significant number of sources . the first requirement is a large sample of caustic - crossing binary - lens events from which to choose targets , which in turn requires an even larger sample of microlensing events . a large sample is important in that it ensures that only interesting and promising sources and events are followed . currently , both the ogle and moa collaborations monitor many millions of stars in the galactic bulge . both reduce their data real - time , enabling them to issue ` alerts , ' notification of ongoing microlensing events ( udalski et al . 1994 , bond et al . 2002a).ftp / ogle / ogle3/ews / ews.html ( ogle ) and http://www.roe.ac.uk/@xmath15iab/alert/alert.html ( moa ) . ] combined , these two collaborations should alert about @xmath197 events per year ( with the majority of alerts from ogle ) . extrapolating from previous results ( alcock et al . 2000a , udalski et al . 2000 ) , approximately @xmath14 of these will be caustic - crossing binaries , or 25 events per year . figure 3 shows the cumulative distribution of apparent ( i.e.uncorrected for redenning ) @xmath132 magnitudes of the 438 independent bulge microlensing alerts in 2002 for which baseline magnitudes were available . of these events , 382 ( @xmath198 ) were alerted by ogle , 61 ( @xmath199 ) were alerted by moa , and 5 were alerted by both collaborations . for the typical colors of sources in the bulge ( @xmath200 ) , only about @xmath201 of the alerts have @xmath202 , and thus would have been accessible to sim . the boundary between dwarfs and giants will occur at an apparent magnitude that depends on the color of the source , the distance to the source , and the reddening . however , for definiteness we will simply assume that the boundary between giants and main sequence stars occurs at roughly @xmath203 . with this assumption , we can therefore expect that approximately @xmath204 of all alerted events will be due to giant stars . therefore , we can expect approximately @xmath205 of all alerts , or @xmath206 events ( assuming 500 alerts ) , to be caustic crossing events with giant sources , and @xmath207 , or @xmath208 events , to be caustic crossing events with main sequence sources , @xmath209 of which will be bright enough to monitor with sim .- magnitude of the source . in fact , deviations from the single - lens form will generally be easier to detect in brighter sources , however this bias is likely to be relatively small for caustic - crossing events , which generally exhibit dramatic and easily - detectable deviations from the point lens form . this may seem in contradiction with the fact that none of the five events toward the bulge presented in table 1 are main sequence sources . however , this almost certainly a selection effect : bright binary - lens events are currently preferentially monitored by the follow - up collaborations , in order to achieve higher signal - to - noise during the second caustic crossing . ] these numbers are likely to remain valid at least for the next several years . however , in the more distant future , and in particular by the time sim is launched , it is likely that the next generation of microlensing survey collaborations will have come online . thus we can expect that , when sim time is operational , a considerably larger sample of caustic - crossing events will be available . survey - type experiments are needed to discover microlensing events toward the bulge , and survey - quality data is generally sufficient to uncover the caustic - crossing nature of the target events . however , as we discussed in [ sec : dt ] , more accurate and densely - sampled photometry is generally needed during the caustic crossing in order to measure @xmath2 . currently , there are several collaborations with dedicated ( or substantial ) access to 1 - 2 m class telescopes distributed throughout the southern hemisphere that closely monitor alerted microlensing events with the goal of discovering deviations from the single - lens form , with emphasis on search for extrasolar planets ( albrow et al . 1998 , rhie et al . 2000 , tsapras et al . 2001 , bond et al . these collaborations have also been quite successful in predicting and monitoring binary - lens caustic crossings . it seems likely that these collaborations , or similar ones , will still be in place when the next generation of interferometers , or even sim , come online . in our monte carlo simulations we derived the expected precisions @xmath210 assuming that the photometric errors were dominated by photon statistics , and that a total of @xmath110 photons where collected over the entire exposure time for each event this corresponds to total exposure time of @xmath211 for sim on an @xmath113 source . we assume the fractional photometric uncertainty is @xmath114 , and that the astrometric uncertainty was related to the photometric uncertainty via the expression , @xmath212 , with @xmath213mas , as appropriate for sim . since we assumed that the photometric and astrometric errors are given simply by photon statistics , it is trivial to scale our results for other total exposure times @xmath214 and source brightnesses assuming the characteristics of sim : @xmath215 , and @xmath216 . for the purposes of planning observations , and providing an order - of - magnitude estimate for the number of angular radii that can be measured for a given amount of sim time , it is useful to derive an expression for the exposure time required to achieve a given median photometric precision . to be conservative , we assume that the blending is small , but non - negligible . specifically , we adopt the median error found for the monte carlo simulations assuming a blend fraction of @xmath217 , which is @xmath218 . then , @xmath219 thus , for giant sources ( @xmath220 ) , @xmath221 minutes are required to achieve @xmath14 precision , whereas for main - sequence sources ( @xmath222 ) , @xmath223 hours are required for @xmath14 precision , or @xmath224 hours for @xmath16 precision . we have focussed here primarily on astrometric observations with sim because its capabilities are well - suited to this application . the basic requirements to be able to measure @xmath5 accurately for the events we have discussed are relatively high astrometric precisions , @xmath225 , and high sensitivity ( via , i.e. large apertures ) , as the sources we are considering are faint , @xmath226 . these faint sources are inaccessible to current ground - based interferometers . upcoming large - aperture , ground - based interferometers , such as the very large telescope interferometer or the keck interferometer , should be able to achieve the requisite astrometric precisions on all of the bright ( @xmath227 ) giant events . if the target microlensing source happens to have a bright star within the isoplanatic angle , it may be possible to employ phase referencing to extend sensitivity to very faint @xmath228 sources . this would allow one to measure @xmath5 for main - sequences sources from the ground as well . finally , it may be possible to determine @xmath5 from single - epoch measurements of the visibility and/or closure phase ( delplancke , g ' orski , & richichi 2001 , dalal & lane 2003 ) . in this way , sensitivity could plausibly be extended to main - sequences sources by making carefully - timed interferometric measurements of the source when it is highly - magnified during a caustic crossing . however , it is not clear if there exists enough structure in the image positions during this time to extract @xmath5 . this remains an interesting topic for future study . non - targeted space - based astrometric surveys , such as the global astrometric interferometer for astrophysics , are generally not well - suited to this application , due to the relatively sparse sampling of the source stars . finally , access to target - of - opportunity time on 8 - 10 m class telescopes would allow for accurate spectral typing of the source stars . several nights per bulge season would likely be adequate to type the @xmath229 caustic - crossing events per year . however , more time would be required to perform some of the auxiliary science discussed in [ sec : dos ] , such as resolution of the source - star atmospheres . thus , by combining alerts from survey collaborations , with comprehensive ground - based photometry from follow - up collaborations with access to dedicated ( or semi - dedicated ) 1m - class telescopes , and a modest allocation of a total 10 hours of sim time , it should be possible to measure the angular radii of @xmath29 giant stars in the bulge to @xmath14 , or @xmath10 main sequence stars to @xmath16 . several nights of target - of - opportunity time on @xmath230 m telescopes should allow for accurate spectral typing of the sources via high or low - resolution spectroscopy . we have outlined a method to measure the angular radii @xmath1 of giant and main sequence source stars of fold caustic - crossing binary microlensing events toward the galactic bulge . our method to measure @xmath1 consists of four steps . first , survey - quality data can be used to discover and alert caustic - crossing binary - lensing events . such data is sufficient to characterize the event timescale @xmath3 and the angle @xmath4 of source trajectory with respect to the caustic . dense sampling of one of the caustic crossings yields the caustic - crossing timescale @xmath231 . the global solution to the binary - lens light curve yields a prediction for the trajectory of the centroid of the source up to an unknown angle @xmath90 , and the scale , @xmath5 . thus a few , precise astrometric measurements during the course of the event yield @xmath5 . the angular source radius is then simply given by @xmath6 . we argued , based on past experience with modeling binary - lens events , that the parameters @xmath231 , @xmath4 , and @xmath3 should be measurable to a few percent accuracy , provided one caustic - crossing is densely and accurately sampled , and the entire event is reasonably well - covered . we then performed a series of monte carlo experiments that demonstrated that astrometric measurements during the course of the binary - lens event should allow for the determination of @xmath5 to @xmath232 accuracy , assuming photon - limited statistics and a total of 60,000 photons per event . this is a factor of @xmath9 fewer photons than are required to measure @xmath5 to the same precision in single - lens events and corresponds to an exposure time of @xmath233 hour with sim on an @xmath113 source . therefore , it should be possible to measure @xmath1 for a significant sample of giant and main - sequence stars in the bulge with reasonable expenditure of resources . we would like to thank neal dalal for helpful conversations . we would also like to thank the anonymous referee for useful comments and suggestions . this work was supported by nasa through a hubble fellowship grant from the space telescope science institute , which is operated by the association of universities for research in astronomy , inc . , under nasa contract nas5 - 26555 , by jpl contract 1226901 , and by the science research center ( src ) of the korean science and engineering foundation ( kosef ) .

we present results from a systematic study of star formation in local galaxy clusters using 22 m data from the wide - field infrared survey explorer ( wise ) . the 69 systems in our sample are drawn from the cluster infall regions survey ( cirs ) , and all have robust mass determinations . the all - sky wise data enables us to quantify the amount of star formation , as traced by 22 m , as a function of radius well beyond , and investigate the dependence of total star formation rate upon cluster mass . we find that the fraction of star - forming galaxies increases with cluster radius but remains below the field value even at 3 . we also find that there is no strong correlation between the mass - normalized total specific star formation rate and cluster mass , indicating that the mass of the host cluster does not strongly influence the total star formation rate of cluster members .

1108.4693

it is well - established that the fraction of star - forming galaxies declines as a function of increasing local galaxy density in the low redshift universe . also known as the star formation - density relation @xcite , this correlation has been confirmed in many studies , primarily using optical and uv data to trace star formation in massive galaxy clusters and field environments . mid - infrared data from the infrared satellite observatory ( iso ) and the multi- band imaging photometer for spitzer ( mips ) have also revealed the presence of highly obscured , dusty star forming galaxies , previously undetected by optical or uv surveys ( e.g. * ? ? ? * ; * ? ? ? * ) . while the sensitivity of mips has enabled detailed studies of obscured star formation in individual local and distant galaxy clusters ( e.g. * * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , there are still only a small number of low redshift clusters that have been systematically surveyed for dusty star - forming galaxies out to the virial radius . there remain many uncertainties in the relationship between star formation in clusters and their global cluster properties . in particular , several studies have tried to understand the correlation between cluster mass and the mass - normalized cluster star formation rate ( sfr ) . while results from @xcite suggest that there is no strong correlation between cluster specific sfr and cluster mass , others such as @xcite , @xcite , and @xcite argue that cluster specific sfr decreases with cluster mass . the large spatial coverage required to observe dusty star - forming galaxies in low redshift clusters out to the cluster infall regions has thus far hindered our ability to understand how star formation is affected by global cluster properties such as cluster mass . in this paper we exploit data from the wide - field infrared survey explorer ( wise ; * ? ? ? * ) to overcome this observational challenge and present results on obscured star formation and how it relates to cluster mass and radius out to 3 in a sample of 69 clusters at @xmath1 . and are commonly used interchangeably with virial radius and total cluster mass , respectively . is the radius within which the average density is 200 times the critical density of the universe and is the mass enclosed within that radius . wise is a medium - class explorer mission funded by nasa and has completed observations of the entire sky in four infrared bands : 3.4 , 4.6 , 12 , and 22 ( w1 to w4 , respectively ) . wise scanned the sky with 8.8 second exposures , each with a 47 arcmin field of view , providing at least eight exposures per position on the ecliptic and increasing depth towards the ecliptic poles . the individual frames were combined into coadded images with a pixel scale of 1.375 arcsec per pixel . cosmic - rays and other transient features were removed via outlier pixel rejection . the photometry used for our analyses is point spread function ( psf ) fitted magnitudes from the `` first - pass operations coadd source working database '' created by the wise data reduction pipeline . galaxies in our cluster sample have a diffraction limited resolution of @xmath2 ( full width half maximum ) in the 22@xmath0 m band . we have confirmed from w4 coadded images that all star - forming galaxies considered in our analyses appear unresolved in the 22 band , and have psf photometry reduced @xmath3 values less than 1.5 . therefore we use the psf magnitudes from the first - pass photometric catalog to obtain estimates of total flux . for the minimum coverage of 8 overlapping frames , the sensitivity for @xmath4 in the w4 band is 6 mjy , including uncertainty due to source confusion @xcite . to ensure an unbiased comparison of global sfrs and total ir luminosities of clusters at different redshifts , we impose a lower limit of sfr=4.6 on our entire cluster sample , which is equivalent to a total ir luminosity of @xmath5 , and corresponds to the 6 mjy flux limit at @xmath6 . we hereafter refer to our sample of star - forming galaxies as demi - lirgs , which have nearly half the total ir luminosity of a luminous infrared galaxy or lirg . however we note for future extragalactic studies using wise data that most coadded observations will have at least 12 overlapping frames and hence better sensitivity than the conservative 6 mjy limit we adopt in this paper . additional information regarding wise data processing is available from the preliminary data release explanatory supplement . we use the cluster infall regions ( cirs ; rines & diaferio 2006 ) sample because it provides high - fidelity mass estimates , is at sufficiently low redshift to enable detection with wise of strongly star - forming galaxies , and has extensive spectroscopy for membership determination . the cirs sample consists of 72 low - redshift x - ray galaxy clusters identified from the rosat all - sky survey that are within the spectroscopic footprint of sdss data release 4 . the redshift range of the cirs clusters is @xmath7 , with a median of @xmath8 . cluster masses are available from @xcite , who utilize the caustics infall pattern to determine total dynamical cluster mass and @xcite . the clusters in this paper consist of the entire cirs sample , excluding three clusters at @xmath9 , which leaves 69 remaining clusters with a minimum redshift of @xmath10 . optical photometric and spectroscopic data are obtained from the sloan digital sky survey data release 7 ( sdss dr7 ) @xcite , which are 90% spectroscopically complete for galaxies with @xmath11 and half - light surface brightness @xmath12 mag arcsec@xmath13 . however , the spectroscopic completeness is lower in high - density regions such as in the core of galaxy clusters , due to fiber collisions . adjacent fibers can not be placed closer than 55 arcsec from each other , which corresponds to a separation of 63 kpc at @xmath8 . to verify that the spectroscopic incompleteness arising from constraints on fiber placement has a negligible effect on radial trends of star formation in clusters , we look at the fraction of w4-bright sources as a function of distance from the cluster center . among the @xmath11 sample , the spectroscopic completeness of w4-bright sources ( w4 @xmath14 ) is @xmath1570% within the central 5 arcmin radial bin , and increases to @xmath1580% in the 5 - 10 arcmin bin . in other words , we are not missing a significant fraction of star - forming galaxies in the cluster core due to fiber collisions . the spectroscopic completeness for all galaxies ( regardless of ir emission ) with @xmath11 shows a similar behavior with distance from the cluster center , and on average reaches at least @xmath1580% completeness beyond the central 5 arcmin radius . since the spectroscopic incompleteness of both w4-bright and optically bright galaxies reaches @xmath1580% beyond the cluster core , our results on radial trends of star formation are negligibly affected by spectroscopic incompleteness . in addition , we are measuring trends on scales of hundreds of kpc , and therefore not significantly affected by small differences in spectroscopic completeness near the cluster core . we use photometric data from sdss dr7 to estimate stellar masses for cluster members , using the tight correlation between stellar mass - to - light ( m / l ) ratio and optical colors , as determined by @xcite . with a sample of more than 10,000 optically bright galaxies ( @xmath16 ) , @xcite construct a grid of stellar population models with a range of metallicities and star formation histories , then compare the best - fit galaxy templates with evolved zero redshift templates to determine present - day m / l ratios . we use the relation between @xmath17 color and @xmath18-band stellar m / l ratio to derive stellar masses for our sample . the estimated uncertainty of the color - based stellar m / l ratios , including random and systematic errors , is @xmath1545% @xcite . spectroscopic redshifts from sdss dr7 are used to determine membership for each of our clusters . figure [ fig : caustic ] illustrates an example of using the caustics infall method to determine cluster membership for abell 1377 , the most massive cluster in our sample . it shows the difference in radial velocity ( @xmath19 ) of galaxies in the cluster field with respect to the cluster systemic velocity , as a function of projected distance from the cluster center . galaxies that are dynamically bound to the cluster form a well - defined region that decreases in velocity offset as a function of projected radius . we define the galaxies that are within the edge of this envelope and within a projected radius of @xmath203 , as cluster members . while the average turnaround radius for the cirs sample is @xmath155 , we restrict our cluster galaxy sample to within 3 , since cluster infall patterns are generally better defined closer in to the cluster center . only galaxies with sdss spectroscopic redshifts are considered in the following analyses . we also limit all cluster and field galaxies to be brighter than @xmath21 , which corresponds to the 90% spectroscopic completeness limit of @xmath22 at @xmath6 . star formation rates and total infrared luminosities ( ) are determined from the wise 22 photometry , using the relations presented in @xcite that are calibrated for mips 24 data . @xcite constructed model average spectral energy distribution ( sed ) templates from a sample of local lirgs and ulirgs and derive correlations between 24 flux , sfr , and . the flux - sfr relation given in their equation 14 can be used for both mips 24 and wise 22 data @xcite . the similarity of the two bandpasses are confirmed by @xcite who quantify the correlation between and the inferred mips 24 and wise 22 luminosities . the mips and wise luminosities are inferred by applying a color - correction to the 18 akari flux , and is derived from fitting the ir sed templates of @xcite to all six akari bandpasses ( 9 , 18 , 65 , 90 , 140 , and 160 ) for @xmath15600 galaxies at @xmath1 . the resulting correlations between the @xmath23 and @xmath24 are nearly identical , with only a @xmath154% offset . therefore we proceed by using the flux - sfr and @xmath23 relations of @xcite , assuming the calibration determined from mips 24 data . the total error attributed to the sfr is @xmath150.2 dex , which is dominated by scatter in the @xmath25- relation , and does not include uncertainties inherent in the assumed stellar initial mass function ( imf ) . the imf adopted by @xcite is similar to the @xcite and @xcite imfs , which have relatively fewer low mass stars compared to a salpeter imf , and is more applicable for extragalactic star - forming regions @xcite . a robust estimate of the cluster star formation rates requires that we first identify and exclude all 22@xmath0 m sources that are dominated by agn rather than star formation . as demonstrated in @xcite , even a few ir - bright agn can significantly bias the inferred global star formation rate of a cluster . there are several methods by which we can identify agn in our data set . the first method relies upon wise colors . similar to the agn wedge in spitzer / irac data @xcite , agn - dominated sources will have red colors in w1-w2 ( [ 3.4]-[4.6 ] ) . specifically , we use the criteria [email protected] ( vega ) to identify candidate agn in the wise data set . this color selection is similar to , though slightly bluer than , the agn selection determined in jarrett et al . 2011 ( submitted ) and stern et al . 2011 ( in prep ) . while this single color cut is in general less robust than the full agn wedge , at the low redshifts that are the focus of this work , star - forming galaxies should have colors uniformly blueward of this threshold . we illustrate in figure [ fig : bpt ] that our [email protected] criterion works well in selecting out agn from star - forming galaxies . profile - fit photometry is used to determine the w1-w2 colors and agn exclusion . while some of the nearest galaxies will be resolved in these bands , we are interested only in the w1-w2 color , rather than in single - band photometry . comparison of agn selection based on w1-w2 color from aperture photometry and profile - fit photometry show that the two methods are similarly successful in isolating agn . however , the color cut based on profile - fit magnitudes has a slightly better overlap with the optically detected agn , discussed below . a second approach to agn identification is use of optical spectroscopy to identify sources that lie in the agn region of the baldwin - phillips - terlevich ( bpt ) diagram ( bpt ; * ? here we use the bpt diagram as a cross - check on the wise color selection for the subset of sources . figure [ fig : bpt ] shows the emission line ratios of [ oiii]/h@xmath27 and [ sii]/h@xmath28 , for a sample of 136 cluster members with @xmath29 , of which 27 have [email protected] . boundaries from @xcite are shown as dotted lines and separate regions where narrow emission lines arise from the presence of seyferts , liners ( low ionization narrow emission line regions ) , and hii regions ( star - forming galaxies ) . galaxies that are selected as agn candidates based purely on having a red wise color ( [email protected] ) are indicated as filled ( red ) circles . figure [ fig : bpt ] shows that out of 22 optically identified seyferts , fourteen ( 65% ) are also identified as agn based on the wise [email protected] selection . sources flagged as quasars ( or qsos ) in the sdss catalog are highlighted with ( purple ) triangle symbols , with all ten quasars being independently identified as agn using the wise color selection . since the bpt diagnostic is meant to classify galaxies based only on _ narrow _ emission line ratios , it is not surprising that nearly all of the sdss quasars are not properly diagnosed in the bpt diagram . the h@xmath30 and h@xmath31 emission lines in eight out of the ten sdss quasars have broad wings relative to their neighboring [ oiii ] or [ sii ] emission lines . overall , we conclude from figure [ fig : bpt ] that the wise color selection is an effective method of excluding agn from our sample of bright w4 sources , identifying 85% of the optically detected seyferts and quasars . the two central questions that we aim to address are how the mean specific star formation rate of cluster galaxies ( mssfr ) depends upon location within a cluster , and how the total integrated star formation rate of a cluster depends upon cluster mass . specifically , the mssfr is defined as the total star formation rate in cluster galaxies _ at _ a given projected radius as inferred from 22@xmath0 m photometry , divided by the total stellar mass of all cluster galaxies _ at _ that radius . for probing the total integrated star formation rate we also define the integrated cluster quantity @xmath32 which is a useful mass - normalized measure of the total star formation rate within the infall region , and referred to as the cluster specific sfr ( cssfr ) . among the 69 clusters , a total of 136 demi - lirgs are detected within 3 , of which 27 are determined to be agn based on their w1-w2 color . in the following sections , all sfr quantities are determined from the remaining 109 star - forming demi - lirgs . eight seyferts identified from the bpt diagram are included in the cluster star - forming galaxy sample because their w1-w2 colors are not indicative of agn activity , and we prefer to maintain a uniform wise selection of agn among all field and cluster galaxies . we note that all results presented in this paper are negligibly affected by the exclusion / inclusion of these eight seyferts . the coordinates , redshift , w4 magnitude , sfr , , projected distance from cluster center , and agn flags are listed for the 136 demi - lirgs in table 1 . there is a long history in the literature demonstrating the existence of a strong radial dependence for star formation ( or color ) in galaxy clusters ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? the main strengths of the current analysis are the uniform 22@xmath0 m wise data , spectroscopic completeness , and existence of measurements for the full sample , which together provide us with an infrared - based view of star formation for a homogeneous sample extending well beyond the virial radius . a common approach in the literature has been to look at the radial dependence of the fraction of star - forming galaxies . here we investigate both the dependence of the star - forming fraction , and also the mssfr out to 3 . for comparison , the identical quantities for a field population are calculated , with the field sample chosen from a catalog of galaxies located within a projected radius of 5 mpc from the cluster center , at the cluster redshift . among these galaxies , we choose those with a radial velocity greater than 5000 away from the cluster systemic velocity , redshift @xmath1 , and absolute r magnitude @xmath33 . these are the same redshift and magnitude limits of the cluster galaxy sample . the highest velocity dispersion of our cluster sample is @xmath15960 , which means that the chosen field galaxies are more than @xmath34 away from the cluster redshift . figure [ fig : caustic ] shows delta - velocity ( @xmath19 ) versus projected distance from the cluster center for galaxies within a 5 mpc projected radius of abell 119 . the field galaxies indicated with ( blue ) cross symbols are clearly not associated with the cluster galaxies , which appear distinctly confined to a trumpet - shaped region . by gathering field galaxies within a projected 5 mpc radius from 69 different regions of the sky , we have compiled a large enough sample to obtain a representative field value of mssfr and the fraction of star - forming galaxies with @xmath29 . there are a total of 11180 field galaxies , of which 566 are demi - lirgs . figure [ fig : ssfr_radius ] shows the mssfr as a function of r/ for our ensemble of 69 clusters , including star formation only for galaxies with @xmath29 l@xmath35 ( sfr@xmath36 m@xmath35 yr@xmath37 ) . we emphasize that by construction our sfr limit therefore means that the observed mssfr is a lower bound but a consistent lower bound across the sample . as expected , the mean specific sfr increases with projected radius , with the most central bin containing the fewest star - forming demi - lirgs twelve out of a total of 109 demi - lirgs found in 69 clusters . the mssfr displays a steep increase from the central bin to , then continues to increase monotonically and nearly flattens out below the field value at larger radii . this is consistent with a low redshift study of h@xmath30 star - forming galaxies by @xcite , who found that the median sfr ( normalized by l@xmath38 ) of cluster galaxies reaches the field value beyond 3 . such a radial trend can be driven by two factors an increase in the ssfr of the sub - population of star - forming galaxies , or an increase in the fraction of star - forming galaxies with radius . we find that there is no statistically significant change in the ssfr of individual star - forming galaxies with radius , implying that the trend is driven primarily by a radial gradient in the star forming fraction . this can be seen in figure [ fig : lirg_fraction ] , where we directly plot the fraction of star - forming galaxies as a function of radius . more than one third of the star - forming galaxies reside at 2@xmath39r@xmath393 . what is striking is that even at these large radii the mssfr and star - forming fraction remain below the field values . in figure [ fig : lirg_fraction ] , the demi - lirg fraction is higher in the field relative to the cluster population at 3 by a factor of @xmath151.5 . one possible interpretation of this result is that these are infalling galaxies that may already be `` pre - processed '' in intermediate - density environments , such as galaxy groups or filaments ( e.g. zabludoff & mulchaey 1998 ) , which would explain why they are suppressed relative to a true field population . there is both observational and theoretical support in favor of the suppressed star - forming galaxy fraction out to 3 . @xcite found that the mean cluster sfr , which they derived from [ oii ] emission of galaxies in 15 clusters , is lower than the mean field sfr by more than a factor of two at 2 . results from @xcite and @xcite similarly conclude that cluster star formation is suppressed at several times the virial radius relative to the field . in addition to galaxies being pre - processed at large radii , another explanation for the suppressed demi - lirg fraction in figure [ fig : lirg_fraction ] could be the contribution of a `` backsplash '' population beyond . simulations from @xcite and @xcite show that up to @xmath1550% of galaxies currently located between and 2 may have previously traveled inward of the virial radius of the cluster . therefore the suppressed demi - lirg fraction at large radii may , to some degree , reflect quenched star formation that occurred within the cluster core . however our galaxy sample , which has an absolute magnitude limit of @xmath33 , is sensitive to massive galaxies brighter than m@xmath40 @xcite , whereas backsplash galaxies are on average significantly less massive than first infall galaxies in the cluster outskirts @xcite . as such , our sample is biased against the lower mass backsplash galaxies relative to first infall galaxies . nonetheless , a contribution of backsplash galaxies may indeed be partially responsible for the suppressed star formation activity at large distances from the cluster core . total mass is , both for astrophysics and cosmology , the most fundamental physical galaxy cluster parameter . there are multiple physical processes that depend on the depth of the cluster potential well ( e.g. ram pressure , harassment , tidal interactions ) which can significantly alter the morphologies , gas content , and star formation rates in cluster galaxies ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? furthermore , it has been observed that while the total baryon fraction ( stellar and gas content ) remains roughly constant with cluster mass , the stellar mass decreases and gas content increases in galaxy clusters as a function of cluster mass @xcite . this implies that the integrated star formation efficiency of the history of a cluster is directly tied to cluster mass . in this section we examine the relation between cluster mass and total current cluster star formation to better understand how strongly the present - day star formation rate depends upon cluster mass . our approach is to compute the cluster sfr normalized by cluster mass ( cssfr ; equation 1 ) as a function of cluster mass . for each cluster , the cluster sfr is the sum of sfrs for all member galaxies that have [email protected] and lie within a projected radius of 3 . the left panel of figure [ fig : ssfr_m200 ] shows the cssfr for 62 clusters , which consists of the sample of 69 clusters , excluding seven clusters that have incomplete spatial coverage in the sdss dr7 spectroscopic survey within 3 . the dotted curve shows the limiting detectable cssfr as a function of corresponding to a single demi - lirg with a sfr of 4.6 in a cluster . clusters that have no members above the [email protected] limit are assigned a cssfr of zero . the binned data ( large squares ) indicate the average cssfr in each bin , including clusters with a cssfr of zero . thus the cssfr of the lowest bin is strongly biased by incompleteness , since nearly all of the clusters at low mass fall below the [email protected] limit and have a cluster sfr set to zero . to ensure that our results are not significantly biased by this incompleteness at low cluster mass , we also show the cluster specific sfr versus for a sub - sample of clusters at @xmath41 in the right panel of figure [ fig : ssfr_m200 ] . for these 22 low - z clusters , we are sensitive to star - forming galaxies down to [email protected] , which corresponds to the w4 band 5@xmath42 detection limit of 6 mjy at @xmath8 , and is illustrated with a dotted curve . using the sample of 62 clusters , the left panel of figure [ fig : ssfr_m200 ] shows no significant correlation between cluster specific sfr and cluster mass , over more than one order of magnitude in @xmath43 . the first pass coadd processing used here has the same calibration as in the wise preliminary data release , but was run every other day using only one day of data , creating gaps in coverage at low ecliptic latitude . these gaps should not significantly impact this result , since location of the gaps are random with respect to the cluster centers . we confirm this expectation by examining the cluster specific sfrs of a sub - sample of 24 clusters with ecliptic latitude @xmath44 deg . these clusters are at sufficiently high ecliptic latitude such that they have fairly complete spatial coverage with the first - pass wise coadd data . the average cluster specific sfrs of the high ecliptic latitude sub - sample are overplotted in figure [ fig : ssfr_m200 ] in small purple diamonds . these high ecliptic latitude clusters show no significant correlation between cluster specific sfr and cluster mass , similar to the lack of correlation seen with the full cluster sample . the lack of correlation between cssfr and is also confirmed with the @xmath41 sub - sample , for which we have applied a significantly lower detection limit of [email protected] ( figure [ fig : ssfr_m200 ] , right panel ) . with the lower sfr limit , the clusters in figure [ fig : ssfr_m200 ] have an average of 5.2 star - forming galaxies within r@xmath393 , with a large scatter . five of the low redshift clusters have between 10 to 16 star - forming galaxies . in contrast , for the sample of 62 clusters with the demi - lirg cut ( [email protected] ) , there is an average of 1.6 star - forming galaxies per cluster within r@xmath393 . while the average number of star - forming galaxies per cluster is small in this case , the cssfrs calculated using only the demi - lirgs are well correlated with the cssfrs using the [email protected] cut , with a linear pearson correlation coefficient of 0.9 . therefore , even with an average of only 1.6 star - forming galaxies per cluster , the cssfrs of the full cluster sample are qualitatively representative of cssfrs that sample the fainter end of the infrared luminosity function . the binned data in both panels of figure [ fig : ssfr_m200 ] show cluster specific sfrs that are consistent within 1@xmath42 of each other , across the full range of . the error bars associated with the average cssfr are the poisson uncertainty on the number of star - forming galaxies detected above the sfr limit in each bin . the lack of a significant trend between cssfr and indicates that transformation mechanisms which scale strongly with cluster mass may not play a dominant role in the evolution of star formation in clusters . while figure [ fig : ssfr_m200 ] illustrates the integrated cluster sfr per by taking the sum of all star - forming galaxies within r@xmath393 , our results remain robust when considering only those galaxies within r@xmath39 . since galaxies within are more susceptible to global cluster processes such as ram pressure , the lack of correlation between cssfr ( r@xmath39 ) and cluster mass re - affirms the main conclusion drawn from figure [ fig : ssfr_m200 ] mechanisms that are strongly dependent on cluster mass do not play a significant role in the evolution of star formation in clusters . two cluster processes that strongly scale with cluster mass are galaxy harassment and ram pressure . ram pressure is a mechanism that acts to compress and/or strip cold gas reservoirs in galaxies . the impact of ram pressure is proportional to @xmath45 , where @xmath46 is the density of the intracluster medium ( icm ) and @xmath47 is the velocity of a galaxy with respect to the icm @xcite . galaxy harassment refers to the cumulative effect of high velocity close encounters between galaxies and has the potential to disturb morphologies and quench star formation @xcite . the effects of galaxy harassment , like ram pressure , are expected to scale up with cluster mass , since both cluster velocity dispersion and icm gas density increase with cluster mass . while there has been evidence in support of both ram pressure and harassment having an impact on star formation in individual cluster galaxies ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , the lack of a correlation between cluster specific sfr and cluster mass in figure [ fig : ssfr_m200 ] indicates that neither of these mechanisms significantly trigger / quench star formation in local galaxy clusters , within the mass range of @xmath48 to @xmath49 . our results are consistent with those of @xcite and @xcite , who use a sample of @xmath15100 clusters at @xmath1 and find no significant correlation between h@xmath30-derived cluster specific sfr and cluster mass . both @xcite and @xcite use sdss dr2 data , with the same limiting r - band magnitude as used in this paper . the limiting h@xmath30 sfr is @xmath153 . similarly , @xcite demonstrate that the passive galaxy fraction is roughly constant as a function of system mass , from @xmath50 to @xmath51 . assuming that the non - passive galaxy fraction is dominated by star - forming galaxies rather than agn , the results of @xcite support our finding that cluster specific sfr is not strongly dependent on cluster mass . in contrast , several other studies have found evidence for an anti - correlation between cluster specific sfr and cluster mass . @xcite and @xcite obtain cluster sfrs by integrating h@xmath30 derived sfrs for members out to 0.5 , and cluster masses from observed velocity dispersions for a sample of eight and nine clusters , respectively . their clusters span an approximate redshift range of 0.2 to 0.8 , and the h@xmath30 sfrs are sensitive to @xmath151 . the anti - correlation observed between cluster specific sfr and cluster mass is significantly weakened when only clusters of the same epoch are considered , indicating that evolutionary effects may be the dominating factor . in addition to interactions between galaxies and the cluster environment , studies have demonstrated that tidal interactions between galaxies or galaxy - galaxy mergers can trigger a burst of star formation ( e.g. * ? ? ? * ; * ? ? ? galaxy - galaxy interactions are generally optimized in intermediate density regions with low velocity dispersions , such as in galaxy groups , whereas the more massive clusters are less likely to host galaxy mergers . figure [ fig : ssfr_m200 ] includes a substantial number of low mass systems ( 57 clusters with @xmath52 ) , yet shows no sign of significantly elevated cluster specific sfr in low mass clusters relative to intermediate mass clusters . any signature of enhanced star formation in low mass clusters due to galaxy - galaxy interactions should be particularly evident in our data , because we integrate the sfr of cluster members out to three times the virial radius , where galaxy - galaxy mergers and tidal interactions are more like to occur . a comparison of the distance to nearest neighbor among the demi - lirg population and the general cluster population as a function of projected radius shows that the two populations have mean nearest neighbor distances that are consistent within 1@xmath42 . this indicates that galaxy - galaxy mergers or close tidal interactions do not play a dominant role in the star formation properties of these clusters . our results imply that even in the low mass @xmath53 groups , star formation has already been quenched to similar levels observed in clusters that are more massive by over an order of magnitude . the lack of correlation between cssfr and ( figure [ fig : ssfr_m200 ] ) , combined with the suppressed demi - lirg fraction at 3 relative to the field ( figure [ fig : lirg_fraction ] , would suggest a scenario in which star formation is quenched in a significant population of galaxies within small groups or filaments . then once the remaining star - forming galaxies are accreted into the cluster environment , a mechanism that is not strongly dependent on cluster mass must operate . we suggest strangulation as a plausible candidate for quenching star formation in cluster star - forming galaxies . as a galaxy enters the cluster icm , its hot halo gas is removed , thereby cutting off its resource for future cold gas supplies , since the halo gas would otherwise eventually cool and settle onto the disk @xcite . the effectiveness of strangulation is less dependent on cluster mass than ram pressure or harassment . simulations from @xcite have shown that strangulation can occur in low mass groups , where the icm - galaxy halo interaction is weak relative to clusters . strangulation has been suggested by several cluster studies as an important cluster mechanism ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , including by @xcite who discovered `` mild gradients '' in the morphological fractions of galaxies in cl0024 + 16 out to large cluster radii , which were best explained by a slow - working gentle mechanism such as strangulation . while strangulation may quench star formation in massive clusters , our results suggest that the dominant process for regulating star formation in dense environments does not depend upon cluster mass , and hence must be efficient in small groups or filaments . we use data from wise and sdss dr7 to study the dependence of star formation on cluster mass and location within a cluster in 69 local ( @xmath1 ) galaxy clusters . cluster membership is determined from sdss dr7 spectroscopic redshifts for galaxies brighter than @xmath21 , and star formation rates are determined with 22 photometry from wise using the relations outlined in @xcite . out of 69 clusters , we find a total of 109 star - forming demi - lirgs with sfr@xmath36 within 3 . both the fraction of demi - lirgs and the mean specific sfr of cluster galaxies increases with projected distance from the cluster center . however , the fraction of demi - lirgs remains below the field value even at 3 times the cluster virial radius . one plausible explanation for the suppressed demi - lirg fraction at 3 is that even galaxies that reside significantly beyond the virial radius have already been quenched in their previous environments , such as small groups or filaments . we also investigate the impact of cluster mass on star formation by presenting the total cluster sfr normalized by as a function of . we find no evidence of a correlation between the cluster specific sfr and cluster mass in this first uniform dataset to detect obscured star formation out to several times the virial radius in a large sample of low redshift clusters . our result indicates that cluster mechanisms which scale with cluster mass , such as ram pressure or harassment , are not likely to play a dominant role in the evolution of star formation in local clusters . this publication makes use of data products from the wide - field infrared survey explorer , which is a joint project of the university of california , los angeles , and the jet propulsion laboratory / california institute of technology , funded by the national aeronautics and space administration . the authors would like to thank emilio donoso for his help and advice on navigating the sdss database . we also thank the anonymous referee for a careful reading and comments which improved the paper . wisepc j235654.30 - 101605.2 & 359.22626 & -10.26812 & 0.074 & 4.12 & 0.025 & 90.3 & 7.2e+11 & 1.1 & agn + wisepc j155850.42 + 272324.5 & 239.71010 & 27.39013 & 0.093 & 4.76 & 0.024 & 86.7 & 7.0e+11 & 1.7 & agn + wisepc j121635.79 + 040709.2 & 184.14912 & 4.11922 & 0.076 & 4.62 & 0.027 & 58.5 & 4.9e+11 & 2.1 & agn + wisepc j130534.25 - 021119.1 & 196.39270 & -2.18864 & 0.088 & 5.41 & 0.039 & 38.8 & 3.4e+11 & 0.2 & agn + wisepc j121742.00 + 034631.2 & 184.42499 & 3.77535 & 0.080 & 5.25 & 0.032 & 35.6 & 3.1e+11 & 1.8 & agn + wisepc j160003.13 + 263707.8 & 240.01305 & 26.61884 & 0.093 & 5.67 & 0.033 & 35.1 & 3.1e+11 & 2.0 & agn + wisepc j111519.95 + 542316.6 & 168.83313 & 54.38794 & 0.070 & 4.98 & 0.026 & 34.2 & 3.0e+11 & 1.5 & agn + wisepc j171031.49 + 643914.6 & 257.63123 & 64.65405 & 0.079 & 5.41 & 0.026 & 29.7 & 2.6e+11 & 0.3 & sf + wisepc j134236.22 + 592324.7 & 205.65091 & 59.39019 & 0.071 & 5.15 & 0.030 & 29.7 & 2.6e+11 & 0.5 & agn + wisepc j170226.20 + 341117.5 & 255.60916 & 34.18821 & 0.105 & 6.16 & 0.035 & 28.9 & 2.6e+11 & 0.1 & agn + wisepc j111448.38 + 401749.3 & 168.70160 & 40.29702 & 0.076 & 5.47 & 0.032 & 25.8 & 2.3e+11 & 1.2 & sf * + wisepc j162021.16 + 301020.7 & 245.08818 & 30.17243 & 0.096 & 6.14 & 0.037 & 23.7 & 2.2e+11 & 2.6 & agn + wisepc j160105.12 + 272539.2 & 240.27135 & 27.42756 & 0.087 & 5.91 & 0.034 & 23.3 & 2.1e+11 & 2.0 & sf + wisepc j170100.39 + 342042.9 & 255.25163 & 34.34524 & 0.101 & 6.27 & 0.039 & 23.1 & 2.1e+11 & 0.2 & sf + wisepc j232514.19 + 151442.1 & 351.30911 & 15.24503 & 0.043 & 4.28 & 0.021 & 21.8 & 2.0e+11 & 0.3 & sf + wisepc j165948.48 + 335944.1 & 254.95201 & 33.99559 & 0.085 & 5.96 & 0.036 & 20.4 & 1.9e+11 & 0.1 & sf + wisepc j152138.79 + 305037.4 & 230.41161 & 30.84372 & 0.081 & 6.07 & 0.040 & 16.2 & 1.5e+11 & 1.0 & sf + wisepc j104319.01 + 050818.0 & 160.82922 & 5.13833 & 0.068 & 5.70 & 0.041 & 15.4 & 1.5e+11 & 0.8 & sf + wisepc j151941.93 + 312905.5 & 229.92471 & 31.48486 & 0.080 & 6.11 & 0.031 & 15.2 & 1.4e+11 & 1.1 & sf + wisepc j221445.86 - 092300.8 & 333.69110 & -9.38356 & 0.082 & 6.24 & 0.051 & 14.2 & 1.4e+11 & 0.4 & agn + wisepc j102200.74 + 382914.4 & 155.50308 & 38.48734 & 0.057 & 5.39 & 0.030 & 14.0 & 1.3e+11 & 0.5 & sf + wisepc j130330.80 - 021400.0 & 195.87834 & -2.23334 & 0.084 & 6.38 & 0.049 & 13.2 & 1.3e+11 & 0.1 & agn + wisepc j152021.87 + 485222.0 & 230.09113 & 48.87279 & 0.074 & 6.09 & 0.033 & 12.8 & 1.2e+11 & 1.5 & sf + wisepc j170957.92 + 335507.3 & 257.49133 & 33.91869 & 0.084 & 6.42 & 0.042 & 12.5 & 1.2e+11 & 0.2 & sf + wisepc j004336.31 - 092547.6 & 10.90130 & -9.42988 & 0.050 & 5.29 & 0.030 & 11.3 & 1.1e+11 & 0.3 & sf + wisepc j155857.04 + 273758.0 & 239.73769 & 27.63278 & 0.090 & 6.70 & 0.056 & 11.2 & 1.1e+11 & 1.7 & sf + wisepc j170231.67 + 335135.6 & 255.63196 & 33.85988 & 0.087 & 6.61 & 0.051 & 11.2 & 1.1e+11 & 2.6 & sf + wisepc j110018.01 + 100256.9 & 165.07504 & 10.04914 & 0.036 & 4.57 & 0.027 & 10.7 & 1.1e+11 & 0.8 & agn + wisepc j101346.84 - 005450.9 & 153.44518 & -0.91414 & 0.042 & 4.96 & 0.028 & 10.6 & 1.0e+11 & 0.7 & agn + wisepc j163032.67 + 392303.2 & 247.63612 & 39.38421 & 0.030 & 4.21 & 0.024 & 10.5 & 1.0e+11 & 2.2 & agn + wisepc j122940.27 + 121743.6 & 187.41780 & 12.29546 & 0.087 & 6.72 & 0.155 & 10.2 & 1.0e+11 & 0.1 & sf + wisepc j114623.86 + 552422.2 & 176.59941 & 55.40616 & 0.053 & 5.53 & 0.024 & 10.0 & 9.9e+10 & 1.7 & sf + wisepc j005656.90 - 011242.4 & 14.23707 & -1.21176 & 0.050 & 5.45 & 0.036 & 9.6 & 9.6e+10 & 0.3 & sf + wisepc j162143.25 + 294332.6 & 245.43019 & 29.72572 & 0.098 & 7.09 & 0.070 & 9.5 & 9.4e+10 & 2.8 & sf + wisepc j102126.42 + 381747.5 & 155.36008 & 38.29652 & 0.055 & 5.72 & 0.038 & 9.1 & 9.1e+10 & 1.2 & sf * + wisepc j165903.80 + 334854.6 & 254.76585 & 33.81516 & 0.086 & 6.81 & 0.062 & 8.9 & 8.8e+10 & 0.2 & sf + wisepc j171447.38 + 643541.1 & 258.69742 & 64.59475 & 0.080 & 6.67 & 0.050 & 8.7 & 8.7e+10 & 0.4 & sf + wisepc j132632.23 + 002800.7 & 201.63428 & 0.46685 & 0.086 & 6.84 & 0.060 & 8.7 & 8.7e+10 & 0.3 & sf + wisepc j075225.73 + 283040.0 & 118.10722 & 28.51112 & 0.062 & 6.08 & 0.044 & 8.5 & 8.5e+10 & 0.3 & agn + wisepc j103342.72 + 392926.9 & 158.42801 & 39.49082 & 0.068 & 6.32 & 0.049 & 8.3 & 8.4e+10 & 1.2 & sf + wisepc j132944.92 - 014239.7 & 202.43716 & -1.71104 & 0.089 & 6.98 & 0.077 & 8.2 & 8.2e+10 & 0.8 & sf + wisepc j162345.89 + 410456.5 & 245.94122 & 41.08235 & 0.034 & 4.70 & 0.016 & 8.1 & 8.1e+10 & 2.1 & agn + wisepc j125830.21 - 015835.4 & 194.62585 & -1.97650 & 0.080 & 6.75 & 0.073 & 8.1 & 8.1e+10 & 0.2 & sf * + wisepc j115720.93 + 051506.0 & 179.33719 & 5.25168 & 0.081 & 6.80 & 0.064 & 8.0 & 8.0e+10 & 1.6 & sf + wisepc j152003.58 + 304350.6 & 230.01491 & 30.73073 & 0.082 & 6.84 & 0.208 & 7.9 & 7.9e+10 & 1.7 & sf + wisepc j155637.07 + 270043.1 & 239.15446 & 27.01196 & 0.091 & 7.11 & 0.079 & 7.7 & 7.7e+10 & 1.6 & sf + wisepc j154351.50 + 363136.9 & 235.96458 & 36.52691 & 0.067 & 6.38 & 0.033 & 7.6 & 7.7e+10 & 1.4 & agn + wisepc j121754.98 + 040117.5 & 184.47910 & 4.02153 & 0.082 & 6.87 & 0.108 & 7.6 & 7.7e+10 & 2.1 & sf + wisepc j102946.81 + 401913.7 & 157.44505 & 40.32048 & 0.067 & 6.39 & 0.047 & 7.6 & 7.6e+10 & 1.3 & agn + wisepc j162637.09 + 390739.3 & 246.65456 & 39.12757 & 0.035 & 4.89 & 0.016 & 7.5 & 7.6e+10 & 2.6 & sf + wisepc j123011.93 + 120632.8 & 187.54971 & 12.10911 & 0.085 & 6.96 & 0.180 & 7.5 & 7.6e+10 & 0.1 & sf + wisepc j112344.58 + 030018.9 & 170.93575 & 3.00525 & 0.051 & 5.75 & 0.036 & 7.5 & 7.5e+10 & 1.8 & sf + wisepc j152255.30 + 305905.4 & 230.73042 & 30.98484 & 0.080 & 6.83 & 0.073 & 7.4 & 7.5e+10 & 1.4 & sf * + wisepc j170102.30 + 340400.7 & 255.25958 & 34.06686 & 0.094 & 7.23 & 0.069 & 7.4 & 7.5e+10 & 0.2 & agn + wisepc j155842.84 + 270736.5 & 239.67848 & 27.12680 & 0.086 & 7.03 & 0.060 & 7.3 & 7.4e+10 & 1.6 & sf + wisepc j115604.23 + 050150.7 & 179.01761 & 5.03075 & 0.079 & 6.81 & 0.075 & 7.3 & 7.4e+10 & 2.1 & sf * + wisepc j152420.43 + 295719.7 & 231.08514 & 29.95546 & 0.076 & 6.72 & 0.049 & 7.3 & 7.3e+10 & 1.5 & sf + wisepc j075213.93 + 292023.5 & 118.05803 & 29.33986 & 0.061 & 6.22 & 0.034 & 7.2 & 7.3e+10 & 0.4 & sf + wisepc j133742.55 + 585209.8 & 204.42731 & 58.86938 & 0.074 & 6.68 & 0.078 & 7.2 & 7.2e+10 & 0.9 & sf + wisepc j130344.14 - 030652.2 & 195.93390 & -3.11451 & 0.081 & 6.91 & 0.081 & 7.1 & 7.2e+10 & 0.4 & sf + wisepc j123018.51 + 113811.4 & 187.57713 & 11.63651 & 0.083 & 6.99 & 0.101 & 7.0 & 7.1e+10 & 2.5 & sf + wisepc j155711.60 + 274455.2 & 239.29832 & 27.74868 & 0.088 & 7.12 & 0.073 & 7.0 & 7.1e+10 & 2.1 & sf + wisepc j011242.10 + 010839.7 & 18.17543 & 1.14435 & 0.044 & 5.51 & 0.033 & 6.9 & 7.0e+10 & 0.5 & sf + wisepc j161241.86 + 484805.8 & 243.17441 & 48.80160 & 0.059 & 6.16 & 0.034 & 6.9 & 7.0e+10 & 2.0 & sf + wisepc j133940.58 + 590307.8 & 204.91910 & 59.05216 & 0.073 & 6.70 & 0.081 & 6.9 & 6.9e+10 & 0.8 & sf + wisepc j165915.56 + 331019.5 & 254.81482 & 33.17208 & 0.087 & 7.13 & 0.072 & 6.8 & 6.9e+10 & 0.0 & sf + wisepc j171424.22 + 633937.8 & 258.60092 & 63.66050 & 0.081 & 6.95 & 0.110 & 6.8 & 6.9e+10 & 0.1 & sf + wisepc j135135.91 + 045100.4 & 207.89963 & 4.85010 & 0.076 & 6.78 & 0.069 & 6.8 & 6.9e+10 & 1.1 & sf + wisepc j132633.84 + 000924.7 & 201.64101 & 0.15686 & 0.086 & 7.09 & 0.061 & 6.8 & 6.9e+10 & 0.5 & sf + wisepc j111459.27 + 402142.0 & 168.74695 & 40.36167 & 0.073 & 6.70 & 0.060 & 6.8 & 6.9e+10 & 0.8 & sf + wisepc j112016.54 + 540624.1 & 170.06892 & 54.10668 & 0.072 & 6.68 & 0.047 & 6.7 & 6.8e+10 & 1.4 & sf + wisepc j170522.43 + 334145.0 & 256.34344 & 33.69582 & 0.090 & 7.23 & 0.077 & 6.6 & 6.7e+10 & 0.0 & sf + wisepc j152043.23 + 304122.7 & 230.18015 & 30.68964 & 0.077 & 6.86 & 0.058 & 6.5 & 6.6e+10 & 0.9 & agn + wisepc j154256.63 + 351602.8 & 235.73595 & 35.26744 & 0.070 & 6.64 & 0.041 & 6.5 & 6.6e+10 & 1.7 & sf + wisepc j170132.94 + 331901.1 & 255.38724 & 33.31697 & 0.088 & 7.18 & 0.069 & 6.4 & 6.6e+10 & 2.5 & agn + wisepc j152227.36 + 303446.0 & 230.61401 & 30.57945 & 0.078 & 6.90 & 0.051 & 6.4 & 6.6e+10 & 1.6 & sf + wisepc j134227.74 + 585930.3 & 205.61560 & 58.99174 & 0.072 & 6.73 & 0.057 & 6.3 & 6.5e+10 & 0.6 & sf + wisepc j235425.88 - 095923.4 & 358.60782 & -9.98984 & 0.076 & 6.88 & 0.067 & 6.2 & 6.4e+10 & 0.9 & sf + wisepc j151933.94 + 303057.6 & 229.89142 & 30.51601 & 0.078 & 6.93 & 0.071 & 6.2 & 6.3e+10 & 1.0 & sf + wisepc j152418.93 + 295638.8 & 231.07889 & 29.94411 & 0.076 & 6.89 & 0.056 & 6.1 & 6.3e+10 & 1.2 & sf + wisepc j111250.48 + 015616.5 & 168.21034 & 1.93791 & 0.076 & 6.91 & 0.077 & 6.1 & 6.2e+10 & 1.5 & sf + wisepc j114951.78 + 560852.8 & 177.46574 & 56.14800 & 0.050 & 5.93 & 0.040 & 6.0 & 6.2e+10 & 1.5 & sf + wisepc j162549.24 + 402042.8 & 246.45515 & 40.34522 & 0.029 & 4.70 & 0.028 & 5.9 & 6.0e+10 & 2.6 & sf + wisepc j111925.05 + 545849.5 & 169.85439 & 54.98043 & 0.071 & 6.78 & 0.056 & 5.9 & 6.0e+10 & 1.6 & sf + wisepc j122747.50 + 120322.3 & 186.94794 & 12.05620 & 0.088 & 7.28 & 0.073 & 5.9 & 6.0e+10 & 0.1 & sf + wisepc j111459.91 + 024551.0 & 168.74963 & 2.76418 & 0.077 & 6.95 & 0.056 & 5.9 & 6.0e+10 & 1.3 & sf + wisepc j162253.90 + 402947.0 & 245.72458 & 40.49639 & 0.031 & 4.85 & 0.026 & 5.9 & 6.0e+10 & 2.6 & sf + wisepc j155849.70 + 272639.2 & 239.70708 & 27.44421 & 0.085 & 7.20 & 0.076 & 5.9 & 6.0e+10 & 1.8 & sf + wisepc j152308.37 + 304847.1 & 230.78488 & 30.81309 & 0.072 & 6.82 & 0.064 & 5.9 & 6.0e+10 & 1.4 & sf + wisepc j102815.22 + 400800.0 & 157.06342 & 40.13334 & 0.067 & 6.65 & 0.064 & 5.8 & 5.9e+10 & 0.8 & sf + wisepc j141309.46 + 442850.4 & 213.28941 & 44.48066 & 0.090 & 7.37 & 0.082 & 5.7 & 5.9e+10 & 1.1 & sf + wisepc j011703.58 + 000027.4 & 19.26492 & 0.00762 & 0.046 & 5.76 & 0.029 & 5.7 & 5.9e+10 & 0.5 & agn + wisepc j111341.36 + 412319.9 & 168.42235 & 41.38885 & 0.073 & 6.88 & 0.062 & 5.7 & 5.9e+10 & 0.8 & sf + wisepc j102134.50 + 410605.2 & 155.39377 & 41.10143 & 0.092 & 7.42 & 0.103 & 5.7 & 5.8e+10 & 0.5 & sf + wisepc j132459.25 + 110431.7 & 201.24689 & 11.07547 & 0.084 & 7.21 & 0.142 & 5.7 & 5.8e+10 & 0.5 & sf + wisepc j111512.08 + 542741.4 & 168.80035 & 54.46151 & 0.066 & 6.64 & 0.057 & 5.7 & 5.8e+10 & 1.5 & sf + wisepc j133450.77 + 593713.3 & 203.71155 & 59.62035 & 0.076 & 6.99 & 0.075 & 5.6 & 5.8e+10 & 0.9 & sf + wisepc j132054.04 + 113912.6 & 200.22517 & 11.65351 & 0.095 & 7.51 & 0.147 & 5.6 & 5.8e+10 & 0.6 & sf + wisepc j103130.85 + 401450.9 & 157.87852 & 40.24746 & 0.066 & 6.66 & 0.058 & 5.6 & 5.7e+10 & 1.3 & sf + wisepc j155844.97 + 270812.4 & 239.68739 & 27.13678 & 0.095 & 7.53 & 0.095 & 5.5 & 5.7e+10 & 1.8 & sf + wisepc j104115.63 + 045313.8 & 160.31512 & 4.88717 & 0.068 & 6.76 & 0.063 & 5.5 & 5.6e+10 & 1.3 & agn + wisepc j162036.15 + 335643.1 & 245.15063 & 33.94529 & 0.031 & 4.97 & 0.026 & 5.4 & 5.6e+10 & 2.1 & sf + wisepc j130306.96 - 031846.9 & 195.77899 & -3.31304 & 0.086 & 7.30 & 0.112 & 5.4 & 5.6e+10 & 0.2 & sf + wisepc j115809.58 + 555140.0 & 179.53992 & 55.86110 & 0.063 & 6.57 & 0.084 & 5.4 & 5.5e+10 & 1.8 & sf + wisepc j171248.41 + 634258.0 & 258.20169 & 63.71612 & 0.075 & 7.01 & 0.067 & 5.4 & 5.5e+10 & 0.3 & sf + wisepc j115938.06 + 561702.5 & 179.90857 & 56.28402 & 0.070 & 6.85 & 0.080 & 5.3 & 5.4e+10 & 2.2 & sf * + wisepc j121734.75 + 035019.8 & 184.39478 & 3.83884 & 0.073 & 6.96 & 0.110 & 5.3 & 5.4e+10 & 1.9 & sf + wisepc j155745.93 + 274043.6 & 239.44136 & 27.67878 & 0.090 & 7.46 & 0.096 & 5.2 & 5.4e+10 & 1.7 & sf + wisepc j101051.19 - 003842.5 & 152.71327 & -0.64514 & 0.043 & 5.75 & 0.076 & 5.2 & 5.4e+10 & 0.8 & sf * + wisepc j130510.25 - 023516.0 & 196.29272 & -2.58778 & 0.088 & 7.41 & 0.133 & 5.2 & 5.4e+10 & 0.4 & sf + wisepc j162858.00 + 391909.3 & 247.24168 & 39.31924 & 0.034 & 5.18 & 0.027 & 5.2 & 5.3e+10 & 2.4 & sf + wisepc j151848.57 + 305449.2 & 229.70239 & 30.91367 & 0.078 & 7.13 & 0.064 & 5.2 & 5.3e+10 & 1.2 & sf + wisepc j123001.03 + 113940.5 & 187.50427 & 11.66124 & 0.082 & 7.24 & 0.124 & 5.2 & 5.3e+10 & 2.2 & sf + wisepc j133105.70 + 584939.1 & 202.77376 & 58.82752 & 0.073 & 6.97 & 0.065 & 5.1 & 5.3e+10 & 0.8 & sf + wisepc j152428.76 + 310750.4 & 231.11983 & 31.13067 & 0.078 & 7.14 & 0.094 & 5.1 & 5.2e+10 & 1.1 & agn + wisepc j103954.63 + 051546.3 & 159.97762 & 5.26285 & 0.064 & 6.69 & 0.069 & 5.1 & 5.2e+10 & 1.7 & sf + wisepc j132422.33 + 105047.7 & 201.09305 & 10.84658 & 0.093 & 7.59 & 0.175 & 5.0 & 5.2e+10 & 0.7 & sf + wisepc j082926.09 + 302528.8 & 127.35872 & 30.42466 & 0.051 & 6.17 & 0.047 & 5.0 & 5.2e+10 & 0.7 & sf + wisepc j111503.18 + 542856.5 & 168.76323 & 54.48237 & 0.067 & 6.80 & 0.076 & 5.0 & 5.2e+10 & 1.5 & sf + wisepc j155957.42 + 272745.9 & 239.98924 & 27.46274 & 0.095 & 7.63 & 0.124 & 5.0 & 5.1e+10 & 1.8 & agn + wisepc j171351.15 + 640012.7 & 258.46313 & 64.00354 & 0.088 & 7.45 & 0.179 & 5.0 & 5.1e+10 & 0.2 & sf + wisepc j155930.98 + 273257.4 & 239.87907 & 27.54927 & 0.091 & 7.53 & 0.132 & 4.9 & 5.1e+10 & 1.9 & sf + wisepc j235345.66 - 103353.6 & 358.44025 & -10.56490 & 0.078 & 7.19 & 0.083 & 4.9 & 5.1e+10 & 0.7 & sf + wisepc j115610.78 + 045943.3 & 179.04491 & 4.99536 & 0.079 & 7.24 & 0.104 & 4.8 & 5.0e+10 & 1.6 & agn + wisepc j122915.88 + 113858.6 & 187.31615 & 11.64962 & 0.088 & 7.49 & 0.160 & 4.8 & 4.9e+10 & 2.5 & sf + wisepc j235514.75 - 102946.9 & 358.81146 & -10.49636 & 0.082 & 7.33 & 0.105 & 4.7 & 4.9e+10 & 0.6 & sf + wisepc j133111.02 - 014338.6 & 202.79590 & -1.72740 & 0.084 & 7.39 & 0.096 & 4.7 & 4.9e+10 & 0.7 & sf + wisepc j235208.99 - 095434.4 & 358.03748 & -9.90957 & 0.077 & 7.19 & 0.082 & 4.7 & 4.9e+10 & 0.6 & sf * + wisepc j134033.01 + 015016.0 & 205.13754 & 1.83778 & 0.078 & 7.24 & 0.094 & 4.7 & 4.9e+10 & 0.5 & sf + wisepc j130335.10 - 025417.0 & 195.89626 & -2.90471 & 0.083 & 7.38 & 0.110 & 4.7 & 4.8e+10 & 0.2 & sf + wisepc j130329.04 - 025337.2 & 195.87102 & -2.89366 & 0.079 & 7.27 & 0.092 & 4.7 & 4.8e+10 & 0.3 & sf + wisepc j122902.48 + 114340.3 & 187.26031 & 11.72786 & 0.086 & 7.49 & 0.243 & 4.6 & 4.8e+10 & 2.3 & sf + wisepc j135143.98 + 045833.5 & 207.93326 & 4.97596 & 0.077 & 7.20 & 0.089 & 4.6 & 4.8e+10 & 0.5 & sf + wisepc j170436.61 + 334454.2 & 256.15256 & 33.74839 & 0.090 & 7.60 & 0.099 & 4.6 & 4.8e+10 & 3.0 & sf + wisepc j121734.57 + 033450.9 & 184.39404 & 3.58081 & 0.074 & 7.13 & 0.089 & 4.6 & 4.8e+10 & 1.8 & sf + wisepc j123102.64 + 113929.5 & 187.76102 & 11.65819 & 0.087 & 7.52 & 0.126 & 4.6 & 4.7e+10 & 2.8 & sf + [ table : demi_lirgs ]

the model of ising spins coupled to 2d gravity , in the form of a sum over planar 3 graphs , is studied and in particular the two - point and spin - spin correlation functions are considered . we first solve a toy model in which only a partial summation over spin configurations is performed and , using a modified geodesic distance , various correlation functions are determined . the two - point function has a diverging length scale associated with it . the critical exponents are calculated and it is shown that all the standard scaling relations apply . next the full model is studied , in which all spin configurations are included . many of the considerations for the toy model apply for the full model , which also has a diverging geometric correlation length associated with the transition to a branched polymer phase . using a transfer function we show that the two - point and spin - spin correlation functions decay exponentially with distance . finally , by assuming various scaling relations , we make a prediction for the critical exponents at the transition between the magnetized and branched polymer phases in the full model . epsf by 8 mm 3

hep-th9602028

the work of knizhnik , polyakov and zamolodchikov ( kpz ) @xcite as well as of david , distler and kawai ( ddk ) @xcite made it possible to understand many aspects two - dimensional quantum gravity coupled to conformal field theories . at the same time , it became clear that the models of dynamical triangulations coupled to matter fields provide us with a statistical mechanical realisation of the models described by kpz and ddk , in the same way as the two - dimensional ising model at its critical point can be viewed as a conformal @xmath2 field theory . two aspects of 2d quantum gravity coupled to matter fields remain puzzling : the @xmath3 barrier and the concept of correlation length . the formulae for critical exponents derived by kpz cease to be valid for @xmath4 and the critical exponents are derived by general scaling arguments applied to globally defined operators . at no point is the concept of a divergent correlation length introduced . for ordinary statistical systems the divergence of a correlation length when a critical temperature is approached is believed to be the underlying reason that general scaling arguments work well . the difficulties of defining a local length scale in quantum gravity are well known and have so far prevented a proper treatment of correlation functions by means of continuum methods . working entirely in the context of dynamical triangulations the problems mentioned are not seen directly . statistical models with multiple ising spins living on dynamically triangulated surfaces are perfectly well defined even if @xmath4 and they have a critical point . at least superficially , the only difference compared to @xmath5 is that we can not solve the theory . nevertheless , low temperature expansions and mean field calculations seem reliable for @xmath6 , as is confirmed by the agreement between the theoretical calculations and monte carlo simulations of the systems with large @xmath7 @xcite . the picture which emerges for large @xmath7 from the low temperature expansion @xcite is as follows : for large @xmath8 ( low temperature ) there is a magnetized phase for which @xmath9 separated from a branched polymer phase , where @xmath10 , by a transition at a critical @xmath11 where @xmath12 . likewise it has been possible by the use of dynamical triangulations to address the question of correlation length in two - dimensional quantum gravity . a two - point correlator between `` punctures '' has been calculated in pure gravity as a function of a `` quantum '' geodesic distance and it is found that standard scaling relations , known from the theory of critical phenomena , are satisfied , although with unusual critical exponents @xcite . so far it has not been possible in 2d quantum gravity to calculate correlation functions of matter fields as a function of the geodesic distance . the attempts to measure the spin - spin correlation functions by monte carlo simulations and to define a divergent correlation length as one approaches the critical point , have so far been ambiguous @xcite . the question arises as to whether there is a divergent spin - spin correlation length associated with the phase transition between a magnetized and a non - magnetized phase in the two - dimensional ising model coupled to gravity . rather surprisingly from a continuum point of view , we can , using dynamical triangulations , begin to answer this question in the limit of large @xmath7 , i.e. in the limit where a large number of ising models is coupled to quantum gravity . the statistical model we will define and solve in the following sections is a toy model of 2d quantum gravity coupled to ising spins in the sense that the correct summation over all triangulations is performed , but not all the spin configurations are included . the spin configurations we include are precisely the ones which dominate in the large @xmath7 limit , at large @xmath8 , namely those for which the domains are connected in a tree - like fashion with domain boundaries of minimal size . this model allows us to calculate the two - point functions ( using a certain definition of distance , which is based on the geodesic distance ) and extract the critical exponents . we will verify that these exponents satisfy standard scaling relations and that the geometrical interpretation of some of the exponents is related to the fractal structure of the underlying `` space - time '' , in agreement with general arguments ; that is , we explicitly verify that the exponent @xmath13 is related to the hausdorff dimension by @xmath14 . the model has a third order transition between a tree - like ( i.e. branched polymer ) phase and a magnetized phase . it is found that there is a diverging correlation length in the tree phase associated with the geometric two - point function , but no diverging correlation length associated with the spin - spin correlation function . next we turn our attention to the full model of @xmath0 independent ising spins coupled to 2d gravity , which has a central charge of @xmath15 . following the analysis of the toy model , we show that various two - point functions decay exponentially with distance in the magnetized phase . again there is a diverging correlation length associated with the geometric two - point function . by assuming that all the standard scaling relations still hold for this model and making a few further fairly modest assumptions , we show that the critical exponents for the transition between the magnetized and branched polymer phases , in the full model , are the same as those in the toy model . the rest of this article is organized as follows : in section [ sec : toynomag ] we define the toy model , define a variant of geodesic distance in the model and calculate the two - point function . in sections [ sec : toycrit ] and [ sec : dh ] the critical exponents and hausdorff dimension are calculated for the toy model , whilst section [ sec : spinspin ] discusses the spin - spin correlation function . in section [ sec : toymag ] we verify by direct calculation in an external magnetic field , that the magnetic exponents found by scaling arguments are indeed the correct ones . in section [ sec : full ] we address some of the questions mentioned above for the full model , i.e. in the model where we sum over all the spin configurations . finally , section [ sec : concl ] contains our conclusions . before looking at the full model of @xmath0 spins coupled to 2d gravity ( in the form of a sum over planar 3 graphs ) , we will solve a toy model , which is very similar to the one studied in @xcite . as for that model we will show that there is a magnetized phase for which @xmath16 and a branched polymer phase with @xmath17 . on the boundary @xmath18 , which is the same value as that in @xcite and agrees with the result @xmath19 for @xmath20 , the pure gravity value , which one might expect from the calculation in @xcite . for our model we sum over all possible rooted planar 3 graphs ( let us denote this set of graphs by @xmath21 ) , but we only sum over a subset of the possible spin configurations , namely those for which the domains are connected in a tree - like fashion , with at most one link connecting any two domains . in this paper we shall only consider planar graphs , that is , @xmath22 throughout . each vertex will be weighted with a factor of @xmath23 and has @xmath0 independent spins on it which can take the values @xmath24 ( i.e. vertex @xmath25 has spins @xmath26 on it , with @xmath27 ) . links joining vertices with dissimilar spin configurations will be given a factor of @xmath28 for each spin flavour which differs . thus the grand canonical partition function , t , is [ eqn : gcpf ] t = _ g x^n _ \{s} _ < i j > ( _ = 1^p ( \{s_i^s_j^-1 ) ) , where the first summation is over graphs , with @xmath29 being the number of vertices in the graph . the product is over the nearest neighbour pairs on the graph @xmath30 ( referred to as `` links '' ) and the second summation is over the following set of spin configurations . take the graph @xmath30 and decompose it into a set of one - particle irreducible ( `` 1pi '' ) graphs ( which we shall call `` blobs '' ) connected in a tree - like fashion ( see fig . [ fig : tree ] ) ; note that these blobs are essentially just minimum necked baby universes ( `` minbu ' 's ) . the blobs are fully magnetized , that is , all the vertices within a 1pi blob have the same set of spins , and we will sum over the @xmath31 possible spin configurations for each blob ( except the root blob for which all the spins will be fixed to be @xmath32 ) . ( 100,55)(0,0 ) in the next sections we will solve this model exactly and also calculate various correlation functions , but in order to do this we need first to define how distances are to be measured on the graphs . one possible measure of the distance between two vertices @xmath33 and @xmath34 , is the geodesic distance between them . that is , find the shortest path from @xmath33 to @xmath34 along links , counting a distance of one unit for every link traversed this gives the geodesic distance between the two points . unfortunately , this is quite difficult to deal with analytically , although some results are known for the pure gravity case @xcite . in this paper we use a slightly different definition of distance , which is much simpler to handle . the distance between two vertices will be defined as the shortest path between them along links , counting a distance of one unit only for links that separate two 1pi blobs . since we are interested in correlations between the spins in the root blob , which are held fixed and spins further away , distances will be measured from the root blob . thus all the vertices in the root blob have a distance zero , all the vertices within blobs connected by a single link to the root blob are at distance one and so on ( see fig . [ fig : tree ] ) . it will turn out that the average number of vertices in each blob is very small , so that for many graphs this definition of distance will be quite similar to the geodesic distance , especially in the branched polymer phase for which the partition function is dominated by tree - like graphs . the advantage of measuring distances in this way is that it makes it very easy to define a transfer function , @xmath35 , [ eqn : f ] f(y)= ( x ( 1- y)^- ) + y^2 , where @xmath36 and ( x ) = _ g x^n . this summation is over the set @xmath37 of rooted planar 1pi graphs . this function , @xmath35 , takes a rooted 1pi blob and glues an arbitrary number of trees , each weighted with a factor of @xmath38 , on to the blob . note that each time we glue a tree on to a link in the blob , we pick up a factor of @xmath23 , for the new vertex that is created , a factor of two because we can hang the tree in one of two directions and a factor of @xmath39 to take account of the @xmath31 different ways the spins on the blob and those on the tree s root blob can differ this accounts for the factor of @xmath40 multiplying each @xmath38 . when we sum over all possible rooted 1pi blobs and all possible ways of gluing trees to links this gives the first term in ( [ eqn : f ] ) . more specifically , the blob without any trees attached has a weight of @xmath41 . consider a rooted graph with @xmath29 vertices , it has @xmath42 internal links and we can add an arbitrary number of trees to each link , giving a total contribution of @xmath43 for the graph . when we sum over all 1pi graphs this gives a contribution of _ g x^n ( 1- y)^-l = ( x ( 1- y)^- ) . the second term in ( [ eqn : f ] ) comes from the special case in which the root blob only consists of a single vertex , in which case we get a factor of @xmath23 for the vertex , @xmath44 for the possible ways the spins may differ and @xmath45 for the two trees . ( 100,25)(0,0 ) the partition function , @xmath46 , satisfies @xmath47 and this is represented diagrammatically in fig . [ fig : tft ] ; the shaded circle represents the sum over 1pi graphs , @xmath41 . for fixed @xmath0 and @xmath8 , this equation gives @xmath46 as a function of @xmath23 . by finding the closest singularity to the origin , which occurs at @xmath48 , say , we can determine the free energy for the model . for what follows it will be convenient to define @xmath49 . now , there are essentially two different types of singularity . one occurs when @xmath50 equals @xmath51 , the critical value of @xmath50 for @xmath52 and this will correspond to the magnetized phase . the other occurs when the graphs become tree - like and we shall come back to this case later . consider the first case , for which @xmath54 and hence @xmath55 , then writing @xmath56 as @xmath57 , t_c = _ c + t_c^2 and @xmath58 , where @xmath59 . defining , h ( ) ^= , we get [ eqn : quadhh ] 1- h^2 = ( 1 + e^-2 ) ^p . however , @xmath41 is a known function , from brzin _ et al _ @xcite one can show that [ eqn : model2z ] ( x)= ( 1 - 3 ) , where @xmath60 . this gives @xmath61 , @xmath62 and @xmath63 . hence we can solve ( [ eqn : quadhh ] ) to get @xmath64 with [ eqn : xcmag ] h^2 = , where we have defined @xmath65 . this gives @xmath53 in the magnetized phase we shall prove that it is magnetized later . next we will calculate @xmath67 , the average number of 1pi blobs at distance @xmath68 . define @xmath69 , where the notation means @xmath70 ; this weights each blob at distance @xmath68 with an extra factor of @xmath71 . note that @xmath72 can be regarded as the standard partition function , but with the root blob weighted by an extra factor of @xmath71 . applying the function @xmath73 to this gives @xmath74 , where now each blob at distance one is weighted with the extra factor of @xmath71 . each application of @xmath73 just moves the weights of @xmath71 down the tree , by one unit of distance . thus , _ r = _ v=1 = . _ v=1 , where we have used that @xmath75 . however , @xmath76 and thus @xmath77_{v=1 } \\ & = & \left.\frac{\partial f}{\partial y } \right\vert_{y = t } \nmean_{r-1 } \label{eqn : expo } = \left [ \left.\frac{\partial f}{\partial y}\right\vert_{y = t } \right]^r , \end{aligned}\ ] ] since @xmath78 ( there is only one root blob ) . it will be convenient to define , b(x , , p ) = . _ y = t . using the formula ( [ eqn : f ] ) for @xmath35 , one can calculate @xmath34 , [ eqn : b ] b=. _ t = ( 3 ( x ) -1 ) + 2 x ( 1+e^-2 ) ^2p t , where [ eqn : model2 m ] ( x ) = . note that @xmath79 . evaluating at @xmath53 will give @xmath80 . in the magnetized phase this gives us , [ eqn : bc ] b_c= 1 + ( d -1 ) - . thus at @xmath53 , @xmath81 and the average total number of blobs is = . for large enough @xmath8 , @xmath82 is less than one and positive , that is , the number of blobs decreases exponentially with distance . as @xmath8 is reduced , @xmath82 increases ( since @xmath40 increases and this encourages the tree to branch ) , until @xmath83 at some critical value , @xmath11 ; this corresponds to the boundary of the tree - like region . using @xmath47 , one can show that , = . for @xmath84 , @xmath85 and @xmath86 is finite , whilst for @xmath87 , @xmath88 diverges as @xmath89 . that is , @xmath83 throughout the entire tree - like region and @xmath90 in this region , since @xmath91 . one can easily calculate @xmath66 in the various regions of the phase diagram . in the tree - like region and on the boundary , @xmath92 , hence @xmath93 , where @xmath94 . however , we also have an expression ( [ eqn : b ] ) for @xmath34 and this gives @xmath95 inside the tree phase . hence , @xmath17 here as expected . on the boundary @xmath96 giving @xmath97 . in the magnetized phase @xmath98 , giving @xmath16 as expected . equation ( [ eqn : bc ] ) applies in the magnetized phase and also on the boundary between the two phases . at @xmath11 , @xmath83 and this gives using ( [ eqn : bc ] ) , ^*= - ( 2^ -1 ) . note that this gives an estimate for the location , in the full model , of the transition between the tree and magnetized phases ; compare this with the estimated location of the transition from the tree - like to the unmagnetized ( pure gravity ) phase , given in @xcite . next we will calculate @xmath53 in the tree phase . at @xmath53 we have from @xmath99 , [ eqn : a1 ] t_c = ( x ) + x_c ( 1 + e^-2 ) ^2p t_c^2 and from @xmath83 , using ( [ eqn : b ] ) , [ eqn : b1 ] ( 3 ( x ) -1 ) + 2 x_c ( 1+e^-2 ) ^2p t_c = 1 , where @xmath100 . remembering that @xmath101 and defining @xmath102 , ( [ eqn : b1 ] ) gives ( 3 - 1 ) x x d + 2 d^2 t_c x x^ = 1 and ( [ eqn : a1 ] ) gives [ eqn : c1 ] d^2 t_c^2 x x^ = t_c - x^ . thus , ( 3 - 1 ) x x d + 1 - x^=0 . moreover , t_c = = and thus ( 3 -1 ) x(1-x ) x d + 1-x - 4 x x^2 d = 0 . also ( [ eqn : c1 ] ) gives [ eqn : d1 ] 4x x^2 d = ( 1-x ) . combining these last two equations gives x d + d-1=0 . using ( [ eqn : model2z ] ) and ( [ eqn : model2 m ] ) , x= and substituting into ( [ eqn : d1 ] ) gives = ^-1 and thus @xmath103 is given by [ eqn : xctree ] x_c = d^- in the tree phase . from equations ( [ eqn : xctree ] ) and ( [ eqn : xcmag ] ) one can easily show that there is a third order phase transition with a finite discontinuity , that is , the critical exponent @xmath104 . one can also calculate the average number of vertices at distance @xmath68 . define @xmath105 , where the notation @xmath106 means the function @xmath35 , but with @xmath23 replaced everywhere by @xmath107 . thus @xmath108 is just a tree , but with each vertex in the root blob weighted by an extra factor of @xmath109 . applying @xmath73 , @xmath68 times , just pushes these weights down the tree to a distance @xmath68 , so that in @xmath110 all the vertices at distance @xmath68 have an extra weight of @xmath109 . thus the average number of vertices at distance @xmath68 is given by @xmath111_{z=1 } = \frac{1}{t}\left [ \left . \frac{\partial f}{\partial y } \right\vert_{y = g'_{r-1 } } \frac{\partial g'_{r-1}}{\partial z } \right]_{z=1 } \\ \label{eqn : nmeanr } & = & b^r \nmean_0 \end{aligned}\ ] ] and this is essentially just the ( geometric ) two - point function ( see section [ sec : dh ] ) . the average total number of vertices is = , where @xmath112 is the average number of vertices in the root blob . this is given by _ 0 = b + ( x ) - x d^2 t ( ( x ) + 1 ) . in the magnetized phase this gives , at @xmath53 , _ 0 = + ( d-1 ) + and in the tree phase , _ 0 = . note that @xmath113 , with @xmath114 in the magnetized phase and @xmath115 in the tree phase . the formulae also show that the average number of vertices in a blob is very small , for any values of @xmath8 and @xmath0 ; in fact @xmath116 at @xmath11 . from equation ( [ eqn : expo ] ) we see that there is an exponential decay of the number of blobs with distance . one can define a mass , @xmath117 , in the model through , [ eqn : a2 ] _ r = b^r = e^-m r. the mass is a function of @xmath8 , @xmath0 and @xmath23 . at @xmath53 , as the critical line is approached from the magnetized phase , the mass vanishes ( i.e. there is a correlation length equal to @xmath118 , which diverges ) . note that at @xmath53 , @xmath119 throughout the whole tree phase . let us use this definition of the mass and various scaling relations to calculate the critical exponents . later we will calculate the magnetic exponents directly from the model , for the case @xmath120 , gaining the same results for @xmath121 and @xmath122 ; this will show that our definition of the mass is reasonable and that the scaling relations hold for this toy model . first let us consider the geometric exponents ; note that we have already calculated @xmath66 in section [ sec : blobgam ] . the exponent @xmath13 is defined by @xmath123 for @xmath124 ( where @xmath125 ) , but @xmath126 so that @xmath127 and hence @xmath128 ; in the tree phase @xmath129 and on the boundary @xmath130 . since there are no power law corrections to ( [ eqn : a2 ] ) we have @xmath131 ; note that at small @xmath132 , for @xmath133 , one might expect @xcite that @xmath134 . these sets of exponents satisfy fisher s scaling relation @xmath135 . in reference @xcite it is shown that if the two - point function has associated with it a vanishing mass , then for a suitable definition of the hausdorff dimension , @xmath136 . this gives that @xmath137 in the tree phase , as we might expect for branched polymers and @xmath138 on the critical line . consider now the magnetic exponents , which we will write with a subscript @xmath117 to avoid confusion . evaluating at @xmath53 and letting @xmath139 from the magnetized phase , we have , defining @xmath140 , @xmath141 from ( [ eqn : bc ] ) and hence @xmath142 . if we take @xmath143 ( we shall see later that the spin - spin correlation function has no power law corrections either ) , then using @xmath144 gives @xmath145 . applying the other scaling relations @xmath146 , @xmath147 and @xmath148 , yields @xmath149 , @xmath150 and @xmath151 . the various exponents are listed in table [ tab : exp ] . .critical exponents [ cols="^,^,^,^,^",options="header " , ] before proceeding to check the magnetic exponents explicitly , we shall define @xmath152 and calculate it directly from the model . as before in @xcite we define the hausdorff dimension , @xmath152 , in terms of the two - point function t_2(r ) = _ g x^n w_g , where @xmath153 is the set of planar 3 graphs , with two marked points that are separated by a distance @xmath68 . one of the marked points will be taken to be the vertex , in the root blob , which is connected to the external leg . @xmath154 is just the usual weight for the spin configurations , which appears in ( [ eqn : gcpf ] ) . then @xmath152 is defined , in the continuum limit , by n(r ) ~r^d_h , r , m(x ) r = const . , where n(r ) _ g n x^n w_g = x . that is , by tuning @xmath23 to @xmath53 we are taking a continuum limit ; however this definition of @xmath152 only really makes sense in the tree phase and on the boundary , where the mass vanishes at @xmath53 . now @xmath155 is just t_2(r ) = _ g x^n w_g n_r = t _ r = t b^r _ 0 , where @xmath156 is the number of vertices at distance @xmath68 for graph @xmath30 ; n(r ) = x . as @xmath157 , n(r ) ~r ~r ( - x ) ^- ( x ) ^- . since we are taking the continuum limit with @xmath158 fixed , @xmath159 , n(r ) ~r^2 ( - x ) ^- . in the tree phase @xmath160 giving @xmath137 and on the critical line @xmath161 , so that @xmath162 , as expected . unfortunately , due to the way distances and the hausdorff dimension have been defined in this model , @xmath152 is not well - defined in the magnetized phase . in reference @xcite it is shown , using the geodesic distance , that @xmath163 for pure gravity ( also the geometric exponents are @xmath164 and @xmath165 ) and this may well be true in the whole of the magnetized phase . it is certainly the correct value at @xmath166 , where the toy model reduces to a pure gravity model . thus our definition of distance appears to correctly capture the behaviour of the model in the tree phase and on the critical line , where @xmath90 , but not within the magnetized phase . in this section we will calculate @xmath167 the average total spin at distance @xmath68 ; the summation is over all vertices at distance @xmath68 and all spin flavours on those vertices . since the spins in the root blob are fixed to be @xmath32 , this is essentially a spin - spin correlation function , for spins separated by a distance @xmath68 . to calculate this we will add a magnetic field for vertices at distance @xmath68 . first we will solve the model with @xmath120 and then the general case . let us define f_+(y ) = ( x e^h ( 1 - y e^h ) ^- ) + x ( 1+e^-2 ) ^2 y^2 e^h , with @xmath168 . this is just @xmath35 , but with @xmath23 replaced by @xmath169 . the function takes a number of trees with weights @xmath38 and glues them on to a blob , where each vertex of the blob has a spin of @xmath32 on it and there is a magnetic field @xmath170 applied on these spins . the function @xmath171 is defined in the same way , but with @xmath172 replaced by @xmath173 throughout . let @xmath174 be a tree with positive spins in the root blob and a magnetic field of @xmath170 applied at distance @xmath68 from the root . similarly @xmath175 has spins of @xmath176 in the root blob . then @xmath177 and @xmath178 . define , f(y)= ( x ( 1 - 2xy)^- ) + x y^2 . then @xmath179 that is , to make a graph with spins of @xmath32 on the root blob and a magnetic field at distance @xmath68 , one takes a number of trees with magnetic fields at distance @xmath180 and glues them on to a blob with positive spins , picking up a factor of @xmath28 if the spins on the root of the tree are negative . _ h=0 ; the superscript on @xmath181 denotes the fact that the vertices in the root blob have a single spin each , which is fixed to be @xmath32 . y = t_r-1^+ + e^-2 t_r-1 ^ - _ r^+ = . _ y=(1 + e^-2 ) t but @xmath182 and @xmath183 so that . _ y=(1 + e^-2 ) t = . thus defining @xmath184 , _ r^+ = b t _ r-1^+ = ( b t)^r _ 0^+ = ( bt)^r _ 0 . let us consider the general case of @xmath0 spin flavours for which @xmath185 , @xmath186 e^-4 ( t_r-1^++ + ) + + e^-2 p t_r-1 ^ -- ) _ r^+++ = . however one can easily show that , _ r^ = ( p-2n ) _ r^+++ , where there are @xmath187 plus signs and @xmath188 minus signs in the superscript on the left - hand side . thus , @xmath189 now defining the magnetization at distance @xmath68 by _ r , we have that @xmath190 . it should be noted that if one uses the exponential decay of @xmath191 or @xmath192 to define a spin - spin correlation length , then this quantity will not diverge at the phase transition ; this behaviour should be contrasted with that of the geometric correlation length , which does diverge . if we define the total magnetization @xmath193 , for the case in which all the spins in the root blob are positive , by [ eqn : mgce ] m , where @xmath194 is the average of the total spin , which is a sum over all vertices and all spin flavours , then we have , at @xmath53 , = . note that @xmath195 for @xmath196 , and that @xmath197 throughout the tree phase , showing that this phase is unmagnetized . the other phase has @xmath85 and hence @xmath198 ; this is the magnetized phase as claimed earlier . also at @xmath53 , near @xmath11 , @xmath199 , where @xmath200 , so that @xmath201 and thus @xmath150 , as calculated from the scaling relations . in fact for finite @xmath0 , = + o(()^2 ) , so that the coefficient in front of @xmath202 is non - zero in general ( note that @xmath166 would give @xmath203 ) . in the previous section we defined the magnetization in the grand canonical ensemble by ( [ eqn : mgce ] ) . a different definition is possible namely , _ ce = _ n _ n , where now we are working in the canonical ensemble , that is , we are using the set , @xmath204 of rooted @xmath29-vertex graphs . in this section we will add a magnetic field , @xmath172 , to the toy model . however , we shall only consider the case @xmath120 , as @xmath205 appears to be more difficult to solve . it will be shown that the magnetic exponents @xmath121 and @xmath122 are the same for both definitions of the magnetization and agree with the results from the scaling relations . rather than use the transfer function @xmath35 , which glues trees on to a 1pi blob , we shall use a different transfer function @xmath206 , that glues trees on to a domain ; this is more convenient for the @xmath120 model with a magnetic field . for @xmath207 , it will give a sum over the same set of graphs that we had previously , with the same weights , but we will no longer be able to keep track of distances within the graphs . consider first the case @xmath207 , the function @xmath206 is given by [ eqn : fbar ] ( y)= ( x ( 1 - 2x k ) ^- ) + x k^2 , where k= ( 1 - ) , with @xmath208 and ( x)=_g x^n . the function defined in ( [ eqn : fbar ] ) takes a rooted 3 graph , which will form the core of a domain , and glues on trees weighted with factors of @xmath209 . as before one picks up a factor of @xmath210 for each tree glued on . in this case however we allow trees to be glued on to the root link , which changes the power of the factor in front of @xmath211 compared with that in ( [ eqn : f ] ) . this is necessary to make sure that we correctly sum over all possible domain structures . the factor @xmath209 is the solution of k= e^-2 y + x k^2 . this generates tree graphs whose vertices are weighted with @xmath23 and which have a factor of @xmath212 at the end of each branch . the domain corresponds to the core 3 graph plus all the vertices in @xmath209 . the ends of the branches in @xmath209 , which we have weighted with @xmath212 , correspond to the domain boundaries ; setting @xmath213 will give us a set of domains glued together in a tree - like fashion . thus we have ( y ) = ( x ( 1 - y ) ^- ) + ( 1 - ) ^2 and the grand canonical partition function ( [ eqn : gcpf ] ) is the solution of @xmath214 . this is shown diagrammatically in fig . [ fig : domain ] . ( 100,55)(0,0 ) note that this is just a different way of formulating the same model that we solved previously . one can easily use @xmath215 to calculate @xmath53 in the two phases , gaining the same results as before ( to solve the @xmath0-flavour case one just uses @xmath216 ) . the advantage of this new formulation is that as we use @xmath217 to move down the tree from one domain to the next , the sign of the spins alternates ( at least for @xmath120 ) . this simplifies greatly the task of adding a magnetic field . for the original version of the model it is quite difficult to keep track of the spins on the different blobs . now let us add a magnetic field of @xmath170 , the transfer function which makes a domain with spins of @xmath32 is [ eqn : fbarplus ] _ + ( y ) = ( x e^h ( 1 - y e^h ) ^- ) + ( 1 - ) ^2 . each spin in the domain is given an extra weight of @xmath218 . it will be convenient to define @xmath219 . similarly @xmath220 , which makes a domain with negative spins is ( [ eqn : fbarplus ] ) , but with @xmath172 replaced everywhere by @xmath173 . let @xmath221 be the partition function for graphs with positive spins in the root domain and @xmath222 that for those with a negative root domain . then @xmath223 and @xmath224 . these equations determine @xmath225 . for @xmath226 in the magnetized phase , the critical value of @xmath23 is determined by @xmath227 , where @xmath228 is the critical value for @xmath229 . defining ( ) ^ , we have t^-_c = ( 1 - ^4 ) and t^+_c = ^-1 _ c + ( 1 - ^2 ) ^2 . note that @xmath230 , @xmath231 and @xmath232 from @xcite . however we still have @xmath224 , and using this gives an equation determining @xmath233 . putting x 1 - _ t^+_c = 1 - e^-2 e^-2h ( 1 - ^2 + ^4 ) and @xmath234 , then we have [ eqn : hbar ] e^2 ( 1- ^4 ) = e^2h . this gives @xmath235 and hence @xmath236 . now , ( , h ) = - = 1 - . at @xmath237 we get by differentiating ( [ eqn : hbar ] ) , ( , h 0^+ ) = 1 - , with ^2 = ; note that at @xmath207 , @xmath238 , see equation ( [ eqn : xcmag ] ) . at @xmath8 close to @xmath239 , ( , h 0^+ ) = 2 + o(()^2 ) . hence , @xmath240 as previously , but note that @xmath241 is not equal to @xmath193 now consider @xmath122 , defined by @xmath242 . from ( [ eqn : hbar ] ) , one can show that @xmath241 is a certain function of @xmath233 , @xmath243 , @xmath244 , @xmath50 , @xmath245 , @xmath246 and @xmath247 , where ( x ) . at @xmath11 and small @xmath248 , barring accidental cancellations , then all these functions are a constant plus terms of order @xmath248 ( or smaller ) , except for @xmath246 , ( x ) = _ c + ( - x ) ^+ . since @xmath249 then @xmath250 , giving @xmath251 . note that the value of @xmath122 depends crucially on the fact that @xmath16 in the magnetized phase . suppose that we next try to calculate @xmath252 , defined by ( , h=0 ) ~()^-_m , then we find that @xmath253 and that this contains terms such as @xmath254 , which diverge as we take @xmath172 to zero . thus @xmath252 seems not to be well - defined , even though the scaling relations give @xmath145 . consider now the magnetization in the grand canonical ensemble , @xmath193 , defined by ( [ eqn : mgce ] ) , @xmath121 has already been calculated and this just leaves @xmath122 . we see that @xmath255 at @xmath11 , due to the divergence of @xmath256 , ( ^*,h ) ~. _ ^ * . for graphs with a positive root domain , at @xmath53 , ^+ _ x_c , where @xmath257 . this gives = . thus ( ^ * , h ) ~1- . _ t^-_c . _ t^+_c and again this is a function of various variables all of which are a constant plus @xmath258 , except that @xmath259 , giving @xmath151 . this completes our analysis of the toy model ; we have derived all the critical exponents from the model ( except for @xmath252 ) and shown that all the usual scaling relations hold . now we shall turn our attention to the full model of @xmath0-ising spins coupled to 2d gravity , and will find that many of the considerations in the previous sections apply for the transition between the tree and magnetized phases in this model . consider the full model where we have @xmath0 independent ising spins on each vertex and are summing over all rooted planar 3 graphs and all spin configurations . then the partition function is , with @xmath260 , [ eqn : full ] ( x , ) = _ g x^n ^p , where we have divided out various factors of two and @xmath261 in order to simplify the formulae later on . @xmath29 is the number of vertices in graph @xmath30 . note that there is no vertex and are no spins at the end of the root , so that the product over links does not include the root link . in fact we shall consider a generalization of this model that has @xmath0 coupling constants . expanding out the above equation gives , with the extra coupling constants , [ eqn : general ] ( x,\ { } ) = _ g x^n _ \{s } _ < ij > , @xmath262 where @xmath263 is the flavour @xmath264 spin on vertex @xmath25 , the second summation is over all spin configurations and @xmath265 is the set of coupling constants @xmath266 , with @xmath267 for all @xmath268 . of course by setting @xmath269 one recovers the ordinary model ( [ eqn : full ] ) . alternatively by setting @xmath270 for @xmath271 we recover the o(n ) models studied in @xcite with @xmath272 . to save writing we define @xmath273 . we will show that @xmath46 satisfies @xmath47 where [ eqn : fullf ] f(y)= ( x,\{ } ) + x y^2 , with [ eqn : fullmap ] x=x(1 - 2xy)^- , _ j = _ j ; @xmath274 is defined as in ( [ eqn : general ] ) , but using the set of rooted 1pi graphs , @xmath37 . in a similar fashion to ( [ eqn : f ] ) the function @xmath35 takes a 1pi graph and glues trees on to it , but this time as well as @xmath23 being changed to @xmath50 , the coupling constants are also renormalized . note that a similar renormalization is studied in @xcite . to justify the above equations let us first consider the case @xmath120 . for each link we have a factor such as @xmath275 and when we multiply these factors together , the sum over spins will cause any terms containing odd powers of a given spin to vanish . the only non - vanishing terms correspond to sets of closed loops ; we shall refer to such a non - vanishing term as a loop configuration . thus for each graph @xmath30 , we sum over all ways of drawing sets of closed non - intersecting , non - backtracking loops on the graph ( call this set of loop configurations @xmath276 ) . so , ( x,_1 ) = _ g x^n _ _ 1^l , where @xmath29 is the number of vertices in graph @xmath30 , and @xmath277 is the number of links making up the loops in the loop configuration @xmath278 . as before , we will make the graphs @xmath30 ( where @xmath279 ) by gluing together 1pi blobs ( fig . [ fig : tree ] ) . note that because of the tree - like structure , any given closed loop is wholly contained within a single blob . this means that the partition function for a given graph @xmath30 , factorizes into a product of contributions from the individual blobs . to calculate the transfer function , @xmath35 , we take a rooted 1pi blob , summing over all graphs and loop configurations , as well as over all ways of attaching trees to the blob . if no trees were attached , one would just have @xmath280 , which is ( x,_1 ) = _ g x^n _ \{s } _ < ij > = _ g x^n _ _ 1^l 1^l - l , where @xmath281 is the number of internal links in graph @xmath30 ( i.e. excluding the root link ) ; that is , @xmath282 . attaching a tree weighted with @xmath38 , gives a factor of @xmath283 ( there is an extra vertex and there are two ways of hanging the tree off the link ) . for a given @xmath30 and @xmath278 , each link either contributes @xmath284 if it is part of a closed loop or @xmath285 if it is not . attaching an arbitrary number of trees to an internal link that contributed @xmath285 changes this contribution to @xmath286 ; however for those that contributed @xmath284 we get @xmath287 , since adding a vertex on a link which was part of a closed loop , increases the length of that loop by one and hence gives an extra factor of @xmath284 . thus summing over all ways of attaching trees causes the change : ( 1+_1 s_i s_j ) ( 1 + s_i s_j ) and hence ( x,_1 ) ( x,_1 ) . the extra term in ( [ eqn : fullf ] ) comes from the case in which the blob is just a single vertex . the general case for @xmath205 follows with only minor modifications to the argument . each closed loop is labelled with a spin flavour ( an integer @xmath264 , with @xmath288 ) and loops with different spin flavours can intersect . thus for a given graph @xmath30 and loop configuration @xmath278 , a given link will have @xmath268 spin flavours running through it ( @xmath289 ) , giving a corresponding factor of @xmath290 . adding an arbitrary number of trees to this link changes the contribution to @xmath291 . hence , we get ( [ eqn : fullf ] ) and ( [ eqn : fullmap ] ) for the general case of a rooted 1pi blob with trees , each weighted by @xmath38 , hanging off it . setting @xmath213 , we recover the partition function for graphs with @xmath279 , that is , @xmath47 ; @xmath35 is the transfer function for the full model . as before , we define distance to be the geodesic distance , but count only the links connecting 1pi blobs . again one has @xmath292 , giving the partition function for a model where all blobs at distance @xmath68 are weighted by @xmath71 . the average number of 1pi blobs at distance @xmath68 is _ r = _ v=1 = . since @xmath76 , _ r = . _ y = t ._v=1 = . _ t _ r-1 , so that @xmath293 with @xmath294 . thus we still have an exponential decay of the average number of blobs with distance and [ eqn : fullb ] b= _ y = t , where @xmath295 . one can also show that @xmath296 \label{eqn : dtdx } \\ & \sim & ( x_c -x)^{-\gst},\end{aligned}\ ] ] where ( x,\{ } ) ( ) _ \{}. thus as in the previous case , at @xmath53 , @xmath83 in the tree phase and on its boundary , where @xmath297 , and we have @xmath85 in the magnetized phase . using equations ( [ eqn : fullb ] ) and ( [ eqn : dtdx ] ) one can show , barring accidental cancellations , that @xmath17 in the tree phase and that on the boundary , @xmath298 , which is just durhuus formula @xcite ; where @xmath299 , which is negative , is the value of the string susceptibility for @xmath274 , that is , ( ) _ \{ } ~ ( x_c(\{ } ) - x ) ^-(1 + ( \{ } ) ) . next we consider the spin - spin correlation function . rather than fixing all the spins in the root blob , as we did for the toy model , it is more convenient to add a set of @xmath0 external spins on the root , which are all fixed to be @xmath32 . the partition function is as in ( [ eqn : general ] ) , but with an extra factor for the link connecting the external spins to the root blob , which we shall denote @xmath300 ( we will not add an extra factor of @xmath23 ) . thus , ( x,\ { } ) = _ g x^n _ \{s } , where the weight @xmath301 for link @xmath302 is given by the square bracket in equation ( [ eqn : general ] ) . note that when the sum over spins is performed , the factor @xmath300 just gives a contribution of one , so that the function @xmath211 is unchanged by adding the external spins . the average total spin at distance @xmath68 is _ r^+ = _ g x^n _ \{s } , where the last summation is over all spins , @xmath303 , at distance @xmath68 and all spin flavours , @xmath264 . consider @xmath304 , the only non - zero contributions come from the configurations in which there is a flavour @xmath264 line from the external spins to vertex @xmath303 , in addition to the usual sets of closed loops . define @xmath305 by _ g x^n _ \{s } = p _ r , then @xmath305 is a sum over all rooted graphs and all loop configurations that have a flavour @xmath285 line from the external spins to a vertex @xmath303 at distance @xmath68 ; the location of the vertex @xmath303 is also summed over . however , we can write @xmath305 in terms of @xmath306 . this is shown diagrammatically in fig . [ fig : gr ] , where links included in the flavour @xmath285 line are drawn thicker than those which are not included . the symbol on the left - hand side is used to represent @xmath305 , it has a flavour @xmath285 line of length @xmath68 , note that we are still measuring distances from the root blob ( not from the external spins ) . ( 100,20)(0,0 ) that is , [ eqn : fulld ] _ r= _ r-1 , where @xmath307 , which is drawn as a shaded circle with a thick line passing through it , is the partition function for 1pi graphs with two distinct legs and a flavour @xmath285 line running through ( it does not include factors of @xmath284 on the two legs ) ; @xmath50 and @xmath308 are given by ( [ eqn : fullmap ] ) evaluated at @xmath213 . define @xmath309 to be equal to the square bracket above , then @xmath310 . so that we have an exponential decay with distance , [ eqn : fullmagdecay ] _ r^+ = d^r _ 0 = d^r _ 0^+ . the total average spin is ^+=^+_0 . the magnetization in the grand canonical ensemble is given by [ eqn : fullmag ] m = = ( ) = m_0 ( ) , where @xmath311 is the magnetization of the root blob ( @xmath312 ) . we have used the fact that the average number of vertices at distance @xmath68 is @xmath313 , so that @xmath314 . this is easily shown by defining @xmath315 and following the derivation of ( [ eqn : nmeanr ] ) . note that , at @xmath53 , in the tree phase @xmath83 , so that @xmath197 throughout this phase as one might expect . in the next section it is assumed that @xmath316 , that is , that the spin - spin correlation length does not diverge ; the easiest way to justify this would be to show that @xmath317 , as we already know that @xmath318 . the following calculation will make this assumption plausible . from ( [ eqn : fullmagdecay ] ) one clearly has @xmath319 , ( at least provided that @xmath320 ) , since @xmath321 . to improve this inequality we need an equation for @xmath34 which is simpler than ( [ eqn : fullb ] ) . define @xmath322 to be the partition function for rooted graphs , with a marked vertex at distance @xmath68 it is a sum over all graphs , @xmath323 , all loop configurations and all ways of marking a vertex at distance @xmath68 ; that is , @xmath324 . then following the derivation of ( [ eqn : fulld ] ) we have g_r=g_r-1=b g_r-1 , where @xmath325 is the partition function for 1pi graphs , with two distinct legs it is a sum over all such graphs and all loop configurations . note , @xmath325 differs from @xmath326 in that there is no flavour 1 line running through it . thus , @xmath327 .\end{aligned}\ ] ] now , if one could show that @xmath328 , then this would imply that @xmath329 . consider @xmath330 , the partition function for 1pi graphs with two distinct legs , in which the flavour 1 spins on the two vertices connected to the legs are held fixed at @xmath32 and @xmath176 respectively ; all other spins are summed over . then one can easily see that @xmath331 , since the only non - vanishing terms correspond either to sets of closed loops , or configurations in which there is also a flavour 1 line running between the two fixed spins ; this last set of terms is multiplied by the product of the fixed spins , namely @xmath176 . for small enough values of @xmath308 , @xmath330 is manifestly positive ; the weights @xmath301 for each link will be positive for any spin configuration ( see equation ( [ eqn : general ] ) ) , and hence @xmath332 is just the sum of positive terms . at @xmath333 for all @xmath268 , the system is completely magnetized and hence @xmath334 . one would expect that @xmath335 for other values of @xmath308 . in which case @xmath329 , with @xmath317 in general and @xmath336 only when @xmath337 for all @xmath268 . thus in general @xmath316 and the spin - spin correlation length does not diverge . note that our definition of distance is only good in the tree - like phase and on its boundary , and hence this result does not say anything about the possible divergence of the proper spin - spin correlation length ( defined using the geodesic distance ) at the magnetization transition , between the u and m phases ( see figure [ fig : phase ] ) . figure [ fig : phase ] shows a tentative phase diagram for the standard model defined by ( [ eqn : full ] ) ; t is the ( unmagnetized ) tree - like or branched polymer phase , u the unmagnetized phase and m the magnetized phase . see @xcite for a discussion of various possible alternative phase diagrams . each line of this phase diagram for constant @xmath0 ( where @xmath0 is a non - negative integer ) corresponds to a line through the @xmath0-dimensional phase space of the generalized model . within the u and m phases the model behaves in a similar fashion to the pure gravity case and it seems probable that @xmath16 throughout both these phases . the transition between the magnetized ( m ) and unmagnetized ( u ) phases is caused by @xmath311 vanishing as the coupling constants @xmath265 are varied . from the kpz results , we expect to have @xmath338 along the corresponding critical line , at least for @xmath339 . the tree - like phase has @xmath17 throughout , the generic value for branched polymers . ( 100,77)(0,0 ) now consider the transition from the magnetized to the tree phase , which is caused by @xmath340 . the behaviour of the full model at this transition is very similar to that of the toy model . following the analysis in section [ sec : toycrit ] , we define a mass by @xmath341 . as before there is an exponential decay , so we will take @xmath131 , which gives @xmath342 provided that the scaling relation @xmath343 holds . consider the magnetic exponents , at @xmath53 and fixed integer @xmath0 we vary the @xmath265 in order to approach the phase transition ; the variable @xmath344 will be used to parameterize this , with @xmath345 at the transition . then by definition @xmath346 , but @xmath347 so that @xmath348 . now @xmath121 is defined by @xmath349 , but @xmath350 from ( [ eqn : fullmag ] ) , thus @xmath351 ; note , we have assumed that we are not close to any point at which @xmath311 vanishes ( @xmath352 in fig . [ fig : phase ] is such a point ) . it has also been assumed that @xmath353 at the transition , i.e. that the spin - spin correlation length does not diverge ; this was justified in the previous section . if we assume that @xmath354 with @xmath143 , then we have @xmath355 . assuming the scaling relations @xmath356 and @xmath357 , gives @xmath138 . the calculation in @xcite , which gave that @xmath358 , still applies , so that @xmath359 and hence @xmath18 . this is encouraging as the analysis in @xcite leads one to suppose that @xmath18 provided @xmath16 in the magnetized phase . also , applying the relation @xmath360 gives @xmath151 . now @xmath34 is given by ( [ eqn : fullb ] ) and in the magnetized phase at @xmath53 , @xmath361 . provided that we are away from points at which @xmath311 vanishes , @xmath274 should be analytic and taylor expandable about @xmath362 ( the value of @xmath308 at the m to t phase transition ) . assuming that @xmath363 is expandable in powers of @xmath364 as far as @xmath344 , then barring accidental cancellations @xmath365 and hence @xmath366 and @xmath104 . thus on the assumption of various scaling relations and that our definitions of @xmath117 and @xmath193 are adequate , and barring problems such as accidental cancellations , the set of exponents given in table [ tab : exp ] should also apply for the full model for the m to t transition , away from point @xmath352 . again it is not clear what the value of @xmath152 is , within the m phase ; however one might guess that @xmath163 , the pure gravity value @xcite , throughout the u and m phases ( in which case it is probable that @xmath164 and @xmath165 here ) . note that the mechanism by which the u to t transition occurs is essentially the same and so we would expect to have the same set of exponents , although the magnetic exponents are not relevant since @xmath197 in both the u and t phases . perhaps one should note that a low - temperature expansion matrix model calculation in the limit @xmath367 @xcite gives that @xmath18 at the transition between the tree and magnetized phases , and that the truncated model studied in @xcite , which is supposed to approximate the u to t transition , has @xmath18 and @xmath149 , agreeing with our predictions . in addition the q - state potts model at large @xmath368 @xcite , which is equivalent to the multiple ising model in the limit @xmath369 and may be related to it for finite @xmath368 , has a branched polymer region ( with @xmath17 ) separating two pure gravity regions ( with @xmath16 ) and has @xmath18 and @xmath104 at both transitions . at point @xmath352 , the above argument fails since @xmath370 and hence it is possible to have @xmath371 ; some of the other assumptions may break down here too . thus at this point one might expect to get a different set of exponents . in any case the calculations in this section only apply for non - negative integer @xmath0 and it is not entirely clear whether @xmath372 , the value of @xmath0 at point @xmath352 , is an integer . it is tempting to suppose that @xmath373 , in which case one could understand the breakdown of the kpz formula in terms of the appearance of the branched polymer phase , however monte carlo simulations do not seem to support this hypothesis @xcite . in this article we have studied correlation functions in two - dimensional quantum gravity coupled to ising spins . for the toy model two approximations are used : ( i ) the spin configurations allowed are the ones dominant in the low temperature expansion of multiple ising spins on dynamically triangulated surfaces and ( ii ) the distance between two spins is identified with the number of links separating 1pi subgraphs along a path connecting the spins . in the full model only the second approximation is made . we do not expect this last approximation to be important for values of @xmath8 where @xmath374 , i.e. for values of @xmath8 where the average number of vertices of a generic @xmath1 graph diverges for @xmath375 , since the average number of vertices in an 1pi part of a generic @xmath1 graph in our ensemble is very small . however , if @xmath376 the average number of vertices in a generic @xmath1 graph is itself small , even at the critical point @xmath377 and our definition of distance can not be used to extract in a reliable way the fractal properties of the ensemble of @xmath1 graphs . approximation ( i ) has been shown to be exact in the @xmath378 limit @xcite and numerical simulations @xcite suggest that it is an excellent approximation except for the smallest values of @xmath8 if many ising spins are coupled to two - dimensional gravity . for moderate values of @xmath7 it is difficult to judge if approximation ( i ) is reliable all the way down to @xmath11 . for @xmath2 and @xmath3 ( i ) not a good approximation for @xmath379 $ ] . the toy model allows us to analyze the two - point function ( the puncture - puncture correlator ) and the spin - spin correlator as a function of the distance @xmath68 . for @xmath380 we have @xmath381 and the fractal structure is reliably extracted from the two - point function . we find that @xmath382 for @xmath383 , while it jumps to @xmath162 at @xmath11 . whilst we can not calculate @xmath152 for @xmath384 , one might expect it to equal four , the pure gravity value @xcite . the fact that @xmath152 is different from 4 for @xmath380 is an indication of the very strong interaction between gravity and matter for @xmath385 . a further indication of the strong link between geometry and matter configurations present in the model is found in the scaling relations for the magnetic exponents . they are found from the behaviour of the _ two - point function _ @xmath386 in the region @xmath387 . the exponential decay of @xmath386 _ at _ @xmath388 as a function of the distance @xmath68 defines a length scale ( ) ~ , which diverges for @xmath139 and allows us to define the magnetic exponents using the standard hyper - scaling relations ( note that @xmath389 for @xmath87 ) . the same exponents are found , for the toy model , by a direct calculation in the canonical and grand canonical ensembles . strictly speaking , it seems more natural to define magnetic scaling properties from one of the spin - spin correlators , i.e. either @xmath390 or @xmath192 , however the corresponding correlation length does not diverge at @xmath11 . this is an indication of the geometric nature of the transition . for the full model , a summation over all spin configurations is performed and the only approximation made is in the measure of distance that is used . as before there is an exponential decay of both the two - point function and the spin - spin correlation function , and one can define , in the magnetized phase , a geometric correlation length , which diverges as the tree phase is approached . based on the analysis in section [ sec : fullspin ] we do not expect the spin - spin correlation length , at least as we have defined it , to diverge at the magnetized to tree - like transition . using the geometric length scale and the scaling relations , which have been shown to hold for the toy model , all the critical exponents are calculated for the branched polymer phase and its boundary , on the basis of a small number of assumptions . the exponents are the same as those for the toy model ( see table [ tab : exp ] ) and again @xmath137 within the tree phase and @xmath152 is equal to three on the critical line , showing the strong interaction between the matter fields and the geometry . some interesting questions remain to be answered , such as what happens outside the branched polymer phase , in particular what the values of the geometric exponents and @xmath152 are , and whether there is a diverging length associated with the spin - spin correlation function at the magnetization transition ( between the u and m phases ) . unfortunately our measure of distance is inadequate in this region and it will require further work using probably the full geodesic distance in order to determine this . it would also be useful to locate point @xmath352 on the phase diagram , as it has been suggested @xcite that @xmath391 , in which case there is an intermediate region , for @xmath392 , between the kpz regime and the branched polymer phase .

the physical conditions near the optical continuum peak ( `` hot spot '' ) in the inner narrow line region ( nlr ) of the seyfert 2 galaxy , ngc 1068 , are examined using ultraviolet and optical spectra and photoionization models . the spectra were taken with the _ hubble space telescope_/space telescope imaging spectrograph ( _ hst_/stis ) , through the 01x520 slit , covering the full stis 1200 to 10000 waveband , and are from a region that includes the hot spot , extending 02 , or 14 pc ( for h 75 km sec mpc ) , in the cross - dispersion direction . the spectra show emission - lines from a wide range of ionization states for the most abundant elements , similar to archival faint object spectrograph spectra of the same region . perhaps the most striking feature of these spectra is the presence of strong coronal emission lines , including [ s xii ] which has hitherto only been identified in spectra of the solar corona . there is an apparent correlation between ionization energy and velocity of the emission lines with respect to the systemic velocity of the host galaxy , with the coronal lines blueshifted , most other high excitation lines near systemic , and some of the low ionization lines redshifted . from the results of our modeling , we find that the emission - line gas is photoionized and consists of three principal components : 1 ) one in which most of the strong emission - lines , such as [ o iii ] , [ ne v ] , c iv , arise , 2 ) a more tenuous , highly ionized component , which is the source of the coronal - line emission , and 3 ) a component , which is not co - planar with the other two , in which the low ionization and neutral lines , such as [ n ii ] and [ o i ] , are formed . the first two components are directly ionized by the euv - xray continuum emitted by the central source , while the low ionization gas is ionized by a combination of highly absorbed continuum radiation and a small fraction of unabsorbed continuum scattered by free electrons associated with the hot spot . the combination of covering factor and thomson optical depth of the high ionization components is insufficient to scatter the observed fraction of continuum radiation into our line - of - sight . therefore , the scattering must occur in an additional component of hot plasma , which contributes little or no uv / optical line emission .

astro-ph9910548

ngc 1068 , one of the initial set of emission line galaxies studied by seyfert ( 1943 ) , is the nearest ( z=0.0038 ) and best studied of the seyfert 2 galaxies . based on the widths of their emission lines in optical spectra , seyfert galaxies are generally divided into two types ( khachikian & weedman 1971 ) . seyfert 1s possess broad permitted lines , with full widths at half maximum ( fwhm ) @xmath5 10@xmath6 km s@xmath3 , and narrower forbidden lines , with fwhm @xmath7 500 km s@xmath3 , while seyfert 2s show only narrow emission lines . the optical continua of seyfert 1 galaxies are dominated by nonstellar emission that can be characterized by a power - law ( oke & sargent 1968 ) , whereas this component is much weaker in seyfert 2 galaxies ( koski 1978 ) . spectropolarimetric studies ( miller & antonucci 1983 ; antonucci & miller 1985 ) revealed the presence of strongly polarized continuum and broad permitted line emission within a 30 aperture centered on the optical nucleus of ngc 1068 . these observations were the inspiration for the unified model for seyfert galaxies , according to which the differences between types 1 and 2 are due to viewing angle , with seyfert 2 galaxies characterized by obscuration of their broad - line regions and central engines ( cf . antonucci 1994 ) . the polarization is wavelength independent , which implies that the emission from the hidden continuum source and broad - line region is scattered into our line - of - sight by free electrons within a hot ( @xmath0 3 x 10@xmath8 k ) plasma in the inner narrow - line region ( nlr ) ( miller , goodrich , & mathews 1991 ) . the structure of the nuclear region has been observed extensively with _ hst _ , using the planetary camera ( pre - costar ; lynds et al . 1991 ) , and the faint object camera ( capetti et al . 1995 ) . the peak in the optical continuum is resolved , with a fwhm of approximately 015 , roughly centered within a cloud of dimensions 35 x 17 elongated in the ne - sw direction , the latter consisting primarily of starlight ( lynds et al . 1991 ; crenshaw & kraemer 1999 , hereafter paper i ) . the ne part of the cloud has the physical dimensions and location attributed to the scattering medium modeled by miller , et al . ( 1991 ) , which has been revealed in the uv ( kriss et al . 1993 ; macchetto et al . 1994 ) and which we have discussed in detail in paper i. a bright knot of [ o iii ] @xmath45007 emission , often referred to as cloud b ( evans et al . 1991 ) , overlaps the continuum emission from the hot spot ; its centroid is @xmath0 02 north of that of the hot spot . while cloud b lies close to the apex of emission - line bicone ( evans et al . 1991 ) , there is evidence that the actual hidden nucleus is further to the south and west ( capetti , macchetto , & lattanzi 1997 ) . it has been suggested that a thermal radio source , s1 , 03 south of the continuum peak , is the true nucleus ( evans et al . 1991 ; gallimore et al 1997 ) . the discovery of an h@xmath9o megamaser with a velocity width of 600 km s@xmath3 ( claussen & lo 1986 ) associated with s1 ( cf . greenhill & gwinn 1997 ) , is further evidence that s1 is the nucleus . even so , the optical continuum peak and a portion of cloud b are within 30 pc of the nucleus , and therefore exposed to an intense flux of ionizing radiation . coronal iron lines have been detected in a number of seyfert galaxies ( grandi 1978 ; osterbrock 1981 ; penston et al . 1984 ; osterbrock 1985 ) . the following coronal - lines are known to be present in the spectrum of ngc 1068 : [ fe x ] @xmath46374 ( koski 1976 ) , [ fe xi ] @xmath47892 ( penston et al . 1984 ; osterbrock & fulbright 1996 ) , and [ si x ] 1.430 @xmath10 m ( thompson 1996 ) . it has been recently suggested ( reynolds et al . 1997 ) that the coronal - line emission arises in the outer regions of the x - ray absorber present in many seyfert galaxies ( see reynolds ( 1997 ) and george et al . ( 1998 ) for a discussion of the properties of x - ray absorbers ) . furthermore , krolik & kriss ( 1995 ) have postulated that the scattering medium in seyfert 2 galaxies may be the same gas that produces the x - ray absorption in seyfert 1s . providing better constraints on the physical conditions in the gas in which the coronal lines arise will help in understanding the connection between the emission - line gas and the scattering medium and , possibly , the x - ray absorber . in this paper we will examine the physical conditions near cloud b and the optical hot spot in the nucleus of ngc 1068 . among our results , we report the presence of the highest ionization energy uv / optical lines ever detected in a seyfert 2 galaxy . in section 3 we will discuss the observed relationship between ionization potential and the velocity of the emission - lines with respect to the systemic velocity of the host galaxy . in sections 4 and 5 we will present the details of photoionization models of the emission - line gas . finally we will discuss the relation between the emission - line gas , the scattering of the continuum radiation , and the structure of the inner nlr . we obtained stis long - slit spectra of ngc 1068 over 1150 10,270 on 1998 august 15 . paper i shows the slit position and describes the observations and data reduction . the spectra that are analyzed in this paper are from the central 02 x 01 bin in paper i , which includes the brightest portion of the hot spot in our slit and a portion of knot b south of its centroid . we measured the fluxes of most of the narrow emission lines by direct integration over a local baseline determined by linear interpolation between adjacent continuum regions . for severely blended lines such as h@xmath11 and [ n ii ] @xmath126548 , 6584 , we used the [ o iii ] @xmath45007 profile as a template to deblend the lines ( see crenshaw & peterson 1986 ) . we then determined the reddening of the narrow emission lines from the observed he ii @xmath41640/@xmath44686 ratio , the galactic reddening curve of savage & mathis ( 1979 ) , and an intrinsic he ii ratio of 7.2 , which is expected from recombination ( seaton 1978 ) at the temperatures and densities typical of the nlr ( see also section 4 ) . we determined errors in the dereddened ratios from the sum in quadrature of the errors from three sources : photon noise , different reasonable continuum placements , and reddening . table 1 gives the observed and dereddened narrow - line ratios , relative to h@xmath13 , and errors in the dereddened ratios for each position . at the end of the table , we give the h@xmath13 flux ( ergs s@xmath3 @xmath14 ) in the bin and the reddening value that we determined from the he ii ratios . figure 1 shows the uv and optical spectra of the continuum hot spot in ngc 1068 . emission lines are present from a wide range in ionization state for the most numerous elements , such as strong [ n ii ] @xmath126548 , 6484 and n v @xmath41240 , and emission lines from the first four ionization states of oxygen , as seen in the earlier fos spectra ( kraemer , ruiz , & crenshaw ; hereafter krc ) . the continuum shows no strong evidence of a stellar component , and is clearly the result of scattered continuum radiation from the hidden central source , as we discussed in paper i. also , the strongest permitted lines show broad wings , which is due to reflected broad - line emission . netzer ( 1997 ) attributed the apparent weakness of o iii ] @xmath41663 in the fos spectra of ngc 1068 to an underabundance of oxygen . on the other hand , we used the ratio of n iii ] @xmath41750/o iii ] @xmath41663 as evidence of an overabundance of nitrogen ( krc ) . however , it is clear from these stis data that o iii ] @xmath41663 has been absorbed by galactic al ii @xmath41671 ( see figure 1 ) and , therefore , can not be used to estimate either the o / h or n / o abundance ratios . the most intriguing feature of these spectra , which was not readily apparent in the fos data , is the presence of a number of coronal lines , including those from extremely high ionization states . as shown in table 1 , we have confirmed the presence of [ fe x ] @xmath46374 and [ fe xi ] @xmath47892 , which had been previously detected , as noted in section 1 . furthermore , we have unambiguously detected [ fe xiv ] @xmath45303 ( see figure 1 ) , which previously had only been confirmed to be present in the spectra of the seyfert 1 galaxies iii zw 77 ( osterbrock 1981 ) and mcg -6 - 30 - 15 ( reynolds et al . 1997 ) and , possibly , the seyfert 2 galaxy tololo 0109 -383 ( fosbury & samsom 1983 ; durret & bergeron 1988 ) . perhaps more unexpected is the presence of [ s xii ] @xmath47611 ( see figure 1 ) . s@xmath15 has an ionization energy of 504.7 ev , which makes this highest ionization line ever detected in the spectrum of a seyfert galaxy , outside the x - ray region , and certainly the highest ever seen in the nlr , besting [ si x ] 1.43 @xmath10 m , which has an ionization energy of 401.4 ev ( thompson 1996 ) . since 7611 is within one of the o@xmath9 telluric bands , detection is difficult from the ground . in fact , the only _ confirmed _ detection of [ s xii ] has been in observations of the solar corona during the total eclipse of 30 may 1965 ( jefferies , orrall , & zirker 1971 ) . osterbrock ( 1981 ) detected a line at a measured wavelength of 7613.1 in the spectrum of iii zw 77 , but did not identify it ; we would suggest that this is also [ s xii ] . in addition to the above - mentioned lines , we have identified a number of other coronal lines in these spectra , as indicated by the ionization potentials listed in table 1 ( for the purposes of this paper , we refer to ions with ionization energies greater than 100 ev as `` coronal '' ; this leaves out the [ ne v ] and [ fe vii ] lines , which , although detected in the solar corona , are formed under more typical nlr conditions ) . to our knowledge , these other lines have not been previously identified in the spectra of seyfert galaxies , although we would expect that they should be present , since their ionization energies are lower than that of fe@xmath16 . exceptions , perhaps , are the tentatively identified nickel lines , [ ni xv ] @xmath46702 and [ ni xiii ] @xmath45116 , since the other lines are from more abundant elements . however , halpern & oke ( 1986 ) detected [ ni ii ] @xmath47378 in an off - nuclear spectrum of ngc 1068 , and determined that the nickle abundance may be at least 4 times solar . one of the interesting aspects of the coronal lines is that they are blueshifted with respect to the lower ionization lines , an effect first noted in a number of seyfert galaxies by penston et al . ( 1984 ) . in figure 2 , we plot ionization potential versus recession velocity ( c@xmath17 ) for all the observed emission lines . note that the presence of a blueshifted absorption feature can bias the centroiding of the line redward , as appears to be the case for n v @xmath41240 and c iv @xmath41550 ( krc ) . similarly , the velocity measurement is more difficult for severly blended lines . nevertheless , there is a strong correlation between ionization potential and velocity . a simple explanation is that the lines originate in different regions that are superimposed , such that we are viewing components with different velocities along our line - of - sight . note that most of the lower ionization lines are redshifted with respect to the systemic velocity of the host galaxy . figure 3 shows spatial cross - cut profiles of the [ fe xiv ] @xmath45303 and [ s xii ] @xmath47611 lines , along the slit . clearly , the coronal line emission is concentrated directly on the optical hot spot , in contrast to the suggestion that it extends throughout the nlr in seyfert galaxies ( korista & ferland 1989 ) , but in general agreement with penston et al . ( 1984 ) regarding the origin of the [ fe x ] @xmath46374 and [ fe xi ] @xmath47892 lines , and results from line profile studies ( cf . moore , cohen , & marcy 1996 ) . furthermore , the coronal line emission is clumpier and not as extended as the continuum radiation . of particular interest is the region ne of the hot spot , where the scattered continuum is strongest ( see paper i ) . this does not necessarily mean that the coronal lines can not arise in the same gas that scatters the continuum radiation , but , if they do , conditions in the scatterer must vary such that the lines are often weak . the possible connection between the scatterer and the coronal gas is discussed in section 7.1 . our approach in photoionization modeling of ngc 1068 was described in detail in krc . we will not repeat the details in the current paper , except to point out important differences . as usual , the models are parameterized in terms of the dimensionless ionization parameter , u , which is the number of ionizing photons per hydrogen atom at the illuminated face of the cloud . since the lines of ne@xmath18 and fe@xmath19 are quite strong in in the spectrum of ngc 1068 , we had assumed that this is in part due to an overabundance of these elements relative to solar ( cf . oliva 1997 ) . we still consider this to be true for iron , since _ asca _ data indicate a large fe / o ratio in the x - ray emitting gas ( netzer & turner 1997 ) . however , there is reason to believe that the ne / h ratio is not substantially greater than solar at least in the extreme inner nlr . the strength of the [ ne v ] @xmath43426 relative to hydrogen is sensitive to the optical thickness of the emission - line gas . that is , if a cloud does not extend much beyond the he@xmath20 zone , [ ne v ] @xmath43426/h@xmath13 can be quite large ( @xmath0 5 ) . also , if the ionizing continuum is somewhat harder than we had assumed , there are more photons capable of ionizing ne@xmath21 . in krc , we had made the argument for a supersolar abundance of nitrogen in the inner nlr of ngc 1068 , based on the strength of n v @xmath41240 relative to he ii @xmath41640 and c iv @xmath41550 ( cf . ferland et al . however , the models significantly overpredicted n iv ] @xmath41486 , which might indicate that that we overestimated the nitrogen abundance . the relative strength of n v @xmath41240 , and other resonance lines , can be boosted somewhat by fluoresence and scattering of ultraviolet continuum radiation from the central source ( cf . grandi 1975a , b ) . as we will discuss in section 6.1 , continuum scattering and fluorescence can not fully explain the strength of n v @xmath41240 . however , based on the observed n iv ] @xmath41486/h@xmath13 ratio , there is no reason to believe that the relative abundance of nitrogen is significantly greater than solar . thus , other than iron , we have chosen to assume solar abundances ( cf . grevesse & anders 1989 ) for these models . the numerical abundances , relative to hydrogen , are as follows : he=0.1 , c=3.4x10@xmath22 , o=6.8x10@xmath22 , n=1.2x10@xmath22 , ne=1.1x10@xmath22 , s=1.5x10@xmath23 , si=3.1x10@xmath23 , mg=3.3x10@xmath23 , fe=8.0x10@xmath23 . in krc , we assumed an ionizing continuum similar to that proposed by pier et al . ( 1994 ) , which is a conservative power - law fit from the uv and to the x - ray , using the 2 kev flux and x - ray continuum derived from the bbxrt data ( marshall et al . however , fits to the x - ray continuum by combining _ rosat_/pspc and _ ginga _ , or _ einstein_/ipc and _ exosat _ data ( see pier et al . 1994 , and references therein ) are consistent with a somewhat harder spectral energy distribution ( sed ) . as noted above , the strength of [ ne v ] @xmath43426 depends on the hardness of the ionizing continuum . therefore , we modeled the sed as a broken power - law of the form , f@xmath24 @xmath2 k@xmath25 , as follows : @xmath26 @xmath27 @xmath28 note that we have assumed a harder euv - xray continuum than in krc ( 1.4 , rather than 1.6 ) and have positioned the x - ray break at one - half the energy . we have , however , assumed the same intrinsic luminosity above the lyman limit , q @xmath2 4x10@xmath29 photons sec@xmath3 ( pier et al . 1994 ) , which is typical of seyfert 1 nuclei . as noted in section 3 , these spectra show emission from a wide range of ionization states , which is an indication that we are seeing emission from a range of physical conditions . also , there is an apparent correlation between the redshift of the emission lines and the ionzation potential of the ion from which they arise . thus , it is likely that we are observing emission from several distinct components . an initial guess at temperature and density of much of the emission - line gas ( which we will refer to as the highion component ) can be derived from the ratio of [ o iii ] @xmath125007,4959/[o iii ] @xmath44363 ( osterbrock 1989 ) . the dereddened observed ratio is @xmath0 47 , indicating an electron temperature t@xmath30 @xmath31 20,000k in the low density limit , and little modification by collisional de - excitation of the @xmath32d@xmath9 level . [ fe vii ] lines ratios can also be used to estimate temperature . the ratio [ fe vii ] @xmath43759/[fe vii ] @xmath46087 @xmath7 0.79 , which indicates t@xmath30 @xmath0 25,000k , at electron densities , n@xmath30 , @xmath31 10@xmath33 @xmath34 ( nussbaumer & storey 1982 ) . still , it is possible that much of the [ fe vii ] and [ o iii ] emission arises in different zones within the same gas , particularly since these lines are at the same approximate velocity ( see table 1 ) . if one kinematic component contributes most of the [ o iii ] and [ fe vii ] emission , and presumably other lines within the same range of ionization energy ( see table 1 ) , it is not likely to contribute the highest ionization lines in these spectra . as we discussed in section 3 , such lines as [ fe ix ] @xmath47892 , [ fe xiv ] @xmath45303 , and [ s xii ] @xmath47611 , are blueshifted with respect to the lower ionization lines by several hundred km sec@xmath3 , which suggests that they arise in a different kinematic component ( which we will refer to as coronal ) . it has been previously suggested that a rough estimate of the temperature of the coronal gas can be obtained from the ratio of [ fe xi ] @xmath42649/[fe xi ] @xmath47892 ( penston et al . 1984 ; osterbrock & fulbright 1996 ) . the observed ratio is [ fe xi ] @xmath42649/[fe xi ] @xmath47892 @xmath2 1.4 @xmath35 . unfortunately , based on the most recent calculations for effective collision strengths ( tayal 1999 ) , this ratio is not particularly sensitive to temperature at electron densities @xmath31 10@xmath36 @xmath34 . using these collision strengths and the transition probabilities from mason ( 1975 ) , the dereddened [ fe xi ] ratio indicates that t@xmath30 @xmath37 3 x 10@xmath38 k in the fe@xmath15 zone . since the redshifts of the balmer lines are similar to those of the [ o iii ] lines , we expect coronal will contribute little to the total hydrogen recombination line flux . since the lowest ionization lines appear redshifted compared to the other emission - lines , it is probable that they arise in a separate component , rather than in the more neutral parts of highion . this component must be of fairly low density , since the ratio of [ s ii ] @xmath46716/[s ii ] @xmath46731 @xmath7 0.81 , indicating n@xmath30 @xmath0 2 x 10@xmath6 @xmath14 ( osterbrock 1989 ) . the s@xmath39 lines can arise in partially neutral gas and , thus , we estimate that the atomic hydrogen density for this component is somewhat higher , n@xmath40 @xmath31 a few times 10@xmath38 @xmath14 . the nature of such a component ( which we will refer to as lowion ) was discussed extensively in krc . the fact that the strengths of the low - ionization , collisionally - excited lines must be large relative to the balmer lines in order for them to appear strong in a composite spectrum led us to believe that the gas was screened and , thus , ionized by an absorbed continuum strongly weighted to the x - ray ( krc ) . however , the paucity of ionizing photons in the absorbed continuum and the constraint that this component had the same covering factor as the screening gas led to an underprediction of its contribution to the total emission - line flux . in these spectra it is apparent that much of the high ionization gas is optically thin , since the observed he ii @xmath44686/h@xmath13 ratio is 0.60 , which is unattainable if much of the gas is optically thick near the lyman limit . thus , neither of the other components can provide an effective screen for lowion . kraemer et al . ( 1999a ) have modeled the effect on the nlr of an absorber with a high covering factor within a few parsecs of the continuum source , and this effect is clearly evident in the nlr spectrum of ngc 4151 ( alexander et al . 1999 ; kraemer et al 1999b ) . thus , we suggest the following geometry for the inner nlr of ngc 1068 . the components represented by highion and coronal are essentially co - located , and are ionized by unabsorbed continuum radiation from the hidden nucleus . the low ionization gas lies out of the plane occupied by highion and coronal , and is irradiated by x - rays which penetrate a layer of absorbing gas closer to the nucleus . also a small fraction of unabsorbed continuum , scattered out of the plane by free electrons , is incident upon the low ionization gas . this simple geometry is illustrated in figure 4 . there is no simple way to quantify the fractions of scattered and direct continuum irradiating lowion , but it is thought that at least 1% of the continuum radiation in ngc 1068 is scattered into our line of sight ( miller et al . 1991 ) and there is the example of a large column , x - ray absorber in ngc 4151 ( cf . barr et al . 1977 ) which may vary in optical depth as a function of angle with respect to the ionization cone ( kraemer et al . we have modeled the absorber assuming an ionization parameter , u @xmath2 0.1 , and a column density , n@xmath40 @xmath2 7.4 x 10@xmath41 @xmath14 , where n@xmath40 is the sum of the columns of ionized and neutral hydrogen . these parameters are similar to those describing the x - ray absorber in ngc 4151 ( yaqoob , warwick & pounds 1989 ; weaver et al . given our estimate of the luminosity of the active nucleus , an absorber of density n@xmath40 @xmath2 10@xmath42 @xmath34 would lie at a distance of @xmath0 1 pc from the central engine , closer to the nucleus than the inner part of hot spot . the ionizing continuum incident upon lowion is shown in figure 5 . for the purposes of these models , we assume that highion and lowion are at the same radial distance from the nucleus . based upon the observed physical properties derived from the emission - lines and the simple geometric picture discussed above , we have generated a three - component photoionization model to describe the physical conditions of line emitting gas . most of the emission arises in the component highion , with n@xmath40 @xmath2 6 x 10@xmath38 @xmath14 , u @xmath2 10@xmath43 , and , n@xmath40 @xmath2 10@xmath44 @xmath14 . the ionization parameter was chosen to produce strong [ fe vii ] emission with negligible [ fe xi ] @xmath47892 . the model was truncated at the termination of the he@xmath45 zone , in order to produce the large he ii @xmath44686/h@xmath13 ratio and strong [ ne v ] described above . the density was constrained on the high end by the presence of strong [ ne iv ] @xmath42423 , which has a critical density @xmath0 10@xmath8 @xmath34 , and on the low end by the ionization parameter . lowion is characterized by n@xmath40 @xmath2 3 x 10@xmath38 @xmath14 , u @xmath2 10@xmath46 ( from the combined continuum ) . since the size of the partially neutral zone in x - ray ionized gas can be inordinately large , we chose to truncate the model at the same column density as highion . although not completely radiation bounded , lowion has a considerable extended zone behind the h@xmath39/h@xmath47 boundary , from which [ o i ] @xmath126300 , 6364 , mg ii @xmath42800 , and [ s ii ] @xmath126716 , 6731 arise . for lowion , we used a solar iron abundance , since the enhanced cooling by [ fe ii ] in the extended zone would suppress other lines . the lack of reliable atomic constants for coronal gas at nebular temperatures ( i.e. , t@xmath30 @xmath31 10@xmath8k ) makes it difficult to set the model input parameters for coronal on the basis of emission - line ratios . we have generated this component using cloudy90 ( ferland et al 1998 ) since it includes more a complete model for the coronal - line emission , and includes elements , specifically argon and nickel , which are not included in our code . therefore , we generated a single coronal component to produce the mix of ionization states observed in the blue - shifted emission - lines , assuming the following : n@xmath40 @xmath2 7 x 10@xmath48 @xmath34 , u = 1.7 , n@xmath40 = 4 x 10@xmath41 @xmath14 . in table 1 , the dereddened l@xmath11/h@xmath13 is 30.75 , which is roughly equal to the value from recombination plus collisional excitation in low ionization gas ( osterbrock 1989 ) . thus , there is no evidence for the destruction of trapped l@xmath11 photons by dust . furthermore , dust is not responsible for scattering of continuum radiation by the hot spot ( see paper i ) , which is another indication of the absence of dust in this region . therefore , we have assumed that the emission - line gas is dust - free . in creating a composite model , we first scaled the contributions from highion to provide a rough fit to the high ionization lines such as c iv @xmath41550 , [ ne v ] @xmath43426 , and [ fe vii ] @xmath46087 and lowion to fit lines such as [ n ii ] @xmath46584 , [ o ii ] @xmath43727 , with the result that the contribution of highion to the total h@xmath13 flux is 3 times that of lowion . due to the uncertainties in the atomic data , we have elected not to include the predicted forbidden line strengths from coronal in our scaling . since the balmer lines are associated kinematically with the non - coronal emission lines , it seems reasonable to expect that the contribution from coronal is @xmath31 15% of the total h@xmath13 flux , which is only slightly more than the uncertainty in measurement . in table 2 , we compare the predicted line ratios for highion and lowion and the composite model to the observed / dereddened values . given the simplicity of the model , we have obtained very satisfactory fits for the vast majority of the observed emission lines . there is good agreement over nearly the full ionization sequence , for example c iv @xmath41550 , c iii ] @xmath41909 , and c ii ] @xmath42325 , and o iv ] @xmath41402 , [ o iii ] @xmath45007 , [ o ii ] @xmath43727 , and [ o i ] @xmath46300 . the predicted [ o iii ] @xmath125007,4959/[o iii ] @xmath44363 ratio is 46 , which indicates that t@xmath30 in the o@xmath45 , averaged over the two components , is correct . the [ fe vii ] @xmath43759/[fe vii ] @xmath46087 ratio is 0.55 , lower than the observed value , which indicates a somewhat higher temperature in the fe@xmath19 zone than predicted , which is not surprising since given our model requirement that this zone is co - located with the o@xmath45 zone . the predicted [ s ii ] @xmath46716/@xmath46731 ratio is 1.15 , identical to that observed within the errors , which confirms our assumptions regarding the density and ionization structure of lowion . in general , the model predictions demonstrate that our assumptions regarding the elemental abundances are approximately correct . however , the models do underpredict the strengths of the neon lines somewhat , which may indicate that the neon is supersolar , but probably less than a factor of two . the model prediction for the he ii @xmath44686/h@xmath13 and ratio is only slightly higher than observed . the predicted strengths of the lines formed in the partially neutral envelope of the low ionization gas , such as [ s ii ] @xmath126716 , 6731 and [ o i ] @xmath46300 , are in reasonable agreement with the observations . this indicates that the combined effects of sed and column density are well represented by the models . this result is of particular importance , given the assumption that the components are ionized by different continua . also listed in table 2 are the model predictions for the emitted h@xmath13 flux , the emitting surface area ( the scaled to the reddening - corrected h@xmath13 luminosity divided by the emitted flux ) , and covering factor for each component , assuming a distance of @xmath0 25 pc from the hidden continuum source . at the distance of ngc 1068 ( 14.4 mpc , bland - hawthorne 1997 ) , the 01 slit width corresponds to 7.2 pc , yielding a covering factor for the slit of @xmath0 0.05 . since the covering factor of these components are substantially lower than 0.05 , there is no evidence that we are seeing substantial effects of superimposition of clouds along our line - of - sight . our photoionizaton code does not include pumping of uv resonance lines by scattering of continuum radiation and continuum fluoresence ( cf . ferguson , ferland , & pradhan 1994 ) , which may explain the underpredictions of the strengths of several uv resonance lines , including n v @xmath41240 , c ii @xmath41335 and mg ii @xmath42800 . therefore , we recomputed the lowion model using cloudy90 , assuming a turbulent velocity of 50 km s@xmath3 . although pumping of uv resonance lines is most efficient for gas with large turbulent velocities ( @xmath5 1000 km s@xmath3 ; ferguson et al . 1994 ) , it can still be an important process if the covering factor of the emitting gas is sufficiently large ( cf . hamann & korista 1996 ) and/or if the optical depths of the scattered lines are small ( ferland 1992 ) . we derive a relatively large covering factor for lowion ( @xmath0 20% that of the slit ) , and thus it is not surprising that the model predicts a significant contribution to resonance lines from scattered continuum radiation ( interestingly , this is all due to direct pumping of the uv resonance line , since there is insufficient ionizing radiation incident upon lowion to pump the euv driver lines ) . the cloudy90 predictions for c ii @xmath41335 and mg ii @xmath42800 are listed in table 2 , alongside those from our code ( the predictions for the non - resonance lines were quite similar for the two codes ) , and the agreement with observed flux ratios is quite good . in addition to the large column densities of c@xmath39 ( 3.22 x 10@xmath49 @xmath14 ) and mg@xmath39 ( 1.54 x 10@xmath50 @xmath14 ) , lowion predicts a large column density for o@xmath47 ( 6.02 x 10@xmath49 @xmath14 ) . o i @xmath41302 is present in the far - uv spectrum ( see figure 1 ) and we expect that is also formed by continuum scattering ( unfortunately , it is not included in the code output ) . however , continuum scattering does not appear to have a similarly strong affect in boosting the n v @xmath41240 line , as we discuss below . in table 3 , we list the predictions of the coronal model for the mean ionization fractions ; ions with observed lines are flagged . we elected not to list the predicted emission - line ratios since they may be misleading given the lack of reliable atomic data ( cf . moorwood et al . the model predicts non - negligible populations for each of the observed ionic states , except si@xmath19 and fe@xmath19 , however it is clear that the [ fe vii ] lines arise in lower ionization gas . if we assume that coronal contributes 15% of the observed h@xmath13 flux , the covering factor is 2.2 x 10@xmath51 . assuming isotropic scattering , at small electron scattering optical depths , @xmath52@xmath53 @xmath54 1 , the reflected fraction of continuum radiation , f@xmath55 @xmath7 n@xmath30f@xmath56@xmath57 , where n@xmath30 is the column density of electrons , f@xmath56 is the covering factor , and @xmath57 is the thomson cross - section . for coronal , f@xmath55 @xmath7 7.4 x 10@xmath23 . however , the observed reflected continuum fraction in our spectrum , based on the estimated central source luminosity ( pier et al . 1994 ) , is f@xmath55 @xmath0 1.4 x 10@xmath51 ( the total reflected fraction is f@xmath55 @xmath7 1.5 x 10@xmath58 , consistent with the larger region sampled ; miller et al . thus , the coronal - line emitting gas near the hot spot makes a neglible contribution to the scattered continuum radiation . in krc , we attributed the unusual strength of the n v @xmath41240 line to an overabundance of nitrogen , however , the models significantly overpredicted n iv ] @xmath41486 . even with solar nitrogen abundance , highion overpredicts n iv ] @xmath41486 by a factor of nearly 3 . a check of highion using cloudy90 , assuming a turbulent velocity of 500 km s@xmath3 , produced only a 70% increase in the strength of n v @xmath41240 . the models do not significantly underpredict other lines formed in the same zone as n v @xmath41240 , such as [ ne v ] @xmath43426 and [ fe vii ] @xmath46087 , and the predictions for other uv resonance lines match the data or are well understood ( see above ) . one possible explanation is that n v @xmath41240 and n iv ] @xmath41486 are formed in different regions , with much different nitrogen abundances , but we find no evidence for abundance inhomogeneities for elements other than iron . therefore , the problem with the nitrogen lines remains unresolved . the predicted temperatures for coronal ( t@xmath30 @xmath2 9.29 x 10@xmath38 k at the ionized face , 3.67 x 10@xmath38 k at the point of truncation ) , are within the range expected from the [ fe xi ] @xmath42649/@xmath47892 ratio . however , since the [ fe xi ] ratio is not particularly sensitive to temperature at low density , the estimate of the coronal gas temperature is not well constrained . hopefully , better estimates may be obtained once more accurate collision strengths are calculated for other coronal lines in the temperature range 10@xmath38 10@xmath8 k. in any case , it is clear from our model predictions that the observed ionic states can co - exist in photoionized gas characterized by a single ionization parameter . as we have demonstrated , the ionic states indicated by the coronal emission - lines can co - exist in a large column of highly ionized gas . the ionization parameter , u @xmath2 1.7 , is characteristic of the x - ray absorbers present in many seyfert 1 galaxies ( cf . reynolds 1997 ) . it is tempting to associate the coronal - line emitting gas with the x - ray absorber , as reynolds et al . ( 1997 ) has done for the case of the seyfert 1 galaxy , mcg -6 - 30 - 15 . based on the fraction of seyfert 1s that possess an x - ray absorber , it is likely that the covering factor of the absorber is 0.5 1 ( cf . george et al . 1998 ) . however , we find that the covering factor of the coronal - line gas in ngc 1068 is quite low , and thus different from typical x - ray absorbers . the low covering factor of the coronal gas also makes it unlikely to be the continuum scattering region . furthermore , miller et al . ( 1991 ) estimated a temperature for the scattering medium , t@xmath30 @xmath0 3 x 10@xmath8 k , which is significantly greater than our model predictions . therefore , it is likely that the scattering occurs in a component of gas associated with the hot spot that is more highly ionized than our coronal component . if we use the initial conditions , n@xmath40 @xmath2 200 @xmath34 , u @xmath2 8 , and n@xmath40 @xmath0 10@xmath41 @xmath14 , cloudy90 predicts a mean t@xmath30 @xmath0 4.5 x 10@xmath8 k , which is close to miller et al.s value . if we , again , apply the constraint that this additional component contributes @xmath31 15% of the total h@xmath13 , the covering factor for this component is @xmath31 0.07 and , thus , f@xmath55 @xmath31 1.2 x 10@xmath51 , or approximately 85% that observed . we would expect to see some line emission from this component , but only from the most highly ionized species , such as [ s xii ] @xmath47611 . the covering factor for this component is slightly greater than that constrained by the slit width , which would not be surprising if the scatterer is indeed an x - ray absorber viewed across our line - of - sight , as suggested by krolik & kriss ( 1995 ) , and there was some superposition of clouds . to summarize , we think it is unlikely that the coronal - line gas has a sufficient covering factor to produce the scattered continuum radiation . it is plausible that the scattering occurs in a component of more highly ionized gas , with a high covering factor , which may contribute a fraction of the coronal - line emission . although the physical conditions of both components are within the range observed for x - ray absorbers ( cf . reynolds 1997 ) , we suggest that while the absorber may be associated with the scatterer , neither are associated with the coronal - line gas that we observe . the relative contributions of photoionization and collisional processes ( e.g. , shocks , heating by cosmic rays ) to the physical state of the emission - line gas in ngc 1068 has been a matter of some debate . while kriss et al . ( 1992 ) have attributed the strong c iii @xmath4977 and n iii @xmath4990 seen in _ hut _ spectra to shock heating , ferguson et al . ( 1994 ) suggest that the strength of these lines result from a combination of continuum fluoresence and dielectronic recombination . given the importance of continuum scattering to o i @xmath41302 , c ii @xmath41335 and mg ii @xmath42800 , it is not surprising that the same process enhances c iii @xmath4977 and n iii @xmath4990 . in fact , the recomputed lowion predicts relatively strong c iii @xmath4977 and n iii @xmath4990 ( see table 2 ) . the contribution to both lines from continuum scattering is @xmath0 90% for lowion and @xmath0 50% in highion , primarily from zones near the ionized face of the cloud , where the driver lines are optically thin and pumping is most efficient ( ferland 1992 ) . the models predict c iii ] @xmath41909/c iii @xmath4977 @xmath7 5.2 and n iii ] @xmath41750/n iii @xmath4990 @xmath7 1.1 , compared to 3.15 @xmath59 0.51 and 1.46 @xmath59 0.34 , respectively , from the _ hut _ spectra ( kriss et al . 1992 ) , noting that these ratios are quite sensitive to the atomic parameters used in the code . thus , although we can not rule out additional heating mechanisms , it is clear that continuum pumping can dramatically enhance these lines in photoionized gas . nevertheless , it is interesting to note that turbulent velocities as low as 50 km s@xmath3 can fully account for the resonance scattering . the full width at half maximum of c ii @xmath41335 is @xmath0 1240 km s@xmath3 , corrected for galactic absorption , which indicates that the line is broadened by the summation of different kinematic components . interestingly , the widths of the resolved kinematic components of instrinsic uv absorbers in seyfert 1 galaxies are also typically @xmath0 50 km s@xmath3 ( crenshaw et al . the lack of large scale turbulence implies a lack of violent disruption of the gas . in krc , we attributed the large h@xmath11/h@xmath13 ratio in at least one region to collisional excitation of h@xmath11 by an injection of energetic particles , possibly associated with the radio jet , into the nlr gas this effect was not as apparent in the optical nucleus , and we find no evidence in our stis data for enhancement of h@xmath11 beyond the predictions of the photoinization models . however , there may still be examples of jet / cloud interaction at other locations in the inner nlr of ngc 1068 . this will be addressed in a subsequent paper . one piece of evidence that heating processes other than photoionization are present is that the electron temperatures predicted by the models are somewhat lower than those estimated from the [ fe vii ] . however , the underprediction of the [ fe vii ] ratio is probably due to our assumption that these lines arise in the same gas as the [ o iii ] lines . therefore , while it is possible that collisional effects are important in some of the high ionization gas in the inner nlr , most of the observed properties are consistent with photoionization by the central source . while there is no additional evidence to support our proposed geometry for the inner nlr in ngc 1068 , assuming that the low ionization gas is screened by a large column absorber resolves a problem with the krc model regarding the covering factor of the screened gas . also , it is apparent that conditions in the nlr of ngc 4151 are due to absorption of the ionizing continuum by intervening gas ( alexander et al . 1999 ; kraemer et al . if a column of extremely optically thick gas is present , it may have an important effect in the collimation of the ionizing radiation . however , in order to do so , the absorber must have a large covering factor ( i.e. , 0.5 ) . in addition to ngc 4151 , a large column of x - ray absorbing gas has been detected in the seyfert 1.5 , mrk 6 ( feldmeier et al . 1999 ) , but there are not enough examples to make a statistical determination of the covering factor of the absorber , or its column density as a function of scale - height . nevertheless , if this component has a large covering factor , it will be important , along with the putative molecular torus and any intrinsic anisotropy of the radiation field , in determining the distribution of ionized gas in the nlr . we have analyzed the stis uv and optical spectra of the inner nuclear region of ngc 1068 , near the optical continuum peak or hot spot . we have contructed photoionization models of the narrow - line gas and have been able to successfully match nearly all the observed dereddened emission - line ratios . from our analysis and modeling of these spectra , we can report the following discoveries regarding the physical conditions near the hot spot . \1 . we report the detection of a number of strong coronal emission lines , including [ fe xiv ] @xmath45303 and [ s xii ] @xmath47611 . the latter is the highest ionization uv / optical line ever identified in the nlr of a seyfert galaxy . the coronal lines are blueshifted with respect to both the systemic velocity of the host galaxy and the lower ionization lines in these spectra . this indicates that the coronal lines arise in a distinct component of narrow - line gas . the lower velocities of other emission - lines may be due to contributions from gas at larger radial distances along our line - of - sight , but it is also possible that much of this gas is co - located with the coronal component , which implies that the velocity of the gas is related to its ionization state and/or density . if this gas is in radial outflow from the nucleus , these observations may provide important constraints for models of cloud acceleration ( the kinematics of the inner nlr of ngc 1068 will be the subject of a later paper ) . we have used a three - component model to match the narrow emission - line flux ratios , since the kinematics and range of physical conditions clearly indicate that we are sampling distinct emission - line clouds . most of the high ionization lines , such as c iv @xmath41550 , [ ne v ] @xmath43426 and [ fe vii ] @xmath46087 , arise in a component , highion , which is directly ionized by the hidden continuum source and is optically thin at the lyman limit . low ionization lines , such as mg ii @xmath42800 , [ o ii ] @xmath43727 , and [ n ii ] @xmath46584 , are formed in a component , lowion , which is screened from the central source by an optically thick layer of gas ( n@xmath40 @xmath2 7.4 x 10@xmath41 @xmath14 ) , similar to that seen in ngc 4151 . several of the uv resonance lines formed in the low ionization gas ( o i @xmath41302 , c ii @xmath41335 , and mg ii @xmath42800 ) are enhanced by the scattering of continuum radiation . the turbulent velocities required to produce this effect are low , @xmath0 50 km s@xmath3 . although this is greater than the thermal velocities within the emitting clouds , it is much less than one might expect if the clouds were being disrupted ( e.g. , by fast shocks ) . also , it is an interesting coincidence that the required turbulent velocities are similar to those measured in the resolved kinematic components of the intrinsic uv absorbers in seyfert 1s . unlike our previous study ( krc ) , we find no strong evidence for a large overabundance of neon and nitrogen , although the fe / h ratio in the higher ionization gas appears to be approximately twice solar . interestingly , there can not be significant iron enhancement in the low ionization gas , otherwise the model underpredicts the electron temperature in the partially neutral zone . although iron could be depleted onto dust grains in the neutral gas , we find no evidence of the effects of dust mixed in with the emission - line gas . assuming solar abundances , the models underpredict n v @xmath41240 and overpredict n iv ] @xmath41486 . resonance scattering and continuum fluoresence can not sufficiently pump the n v line to match the observed flux , and , in any case , these processes do not affect the n iv ] strength . although extremely inhomogeneous nitrogen abundances could cause this effect , we think that this is quite unlikely . although the atomic data is not sufficiently reliable to predict coronal - line flux ratios , we have generated a component , coronal , in which all the observed ions are present . hence , it is plausible that these lines form in photoionized gas . we do not believe that this component is responsible for the scattered continuum radiation , since its covering factor and electron temperature are too low . as such , we postulate that a more highly ionized , co - located component is the scatterer . this component has a covering factor and ionization state similar to the x - ray absorbers detected in seyfert 1 galaxies . s.b.k would like to thank swaraj tayal , don osterbrock , anand bhatia , dick fisher , and bruce woodgate for illuminating conversations about coronal emission - lines . s.b.k . would also like to thank gary ferland for useful discussions about resonance scattering and continuum pumping of emission lines . we also thank cherie miskey for help with the figures . s.b.k . and d.m.c . acknowledge support from nasa grant nag 5 - 4103 . kriss , g.a . , 1993 , in proc , 8th iap mtg . , first light in the universe : stars or qso s , eds . b. rocca - volmerange , b. guideroni , m. dennefeld , & j. tran tranh van , ( editions frontieres : paris ) 379 oliva , e. 1997 , in emission lines in active galaxies : new methods and techniques , ed . b.m . peterson , f .- z . cheng , & a.s . wilson ( san francisco : astronomical society of the pacific ) , asp conference series , 113 , 288 lllll l@xmath11 @xmath41216 & 1455 & 13.6 & 11.54 @xmath59 1.04 & 30.75 @xmath59 5.65 + n v @xmath41240 & 1524 & 77.4 & 6.28 @xmath59 0.50 & 16.19 @xmath59 2.86 + o i @xmath41302 & 2028 & & 0.36 @xmath59 0.04 & 0.84 @xmath59 0.14 + c ii @xmath41335 & 2316 & 11.3 & 0.43 @xmath59 0.12 & 0.98 @xmath59 0.19 + o iv ] @xmath41402 @xmath60 ( si iv @xmath41398 ) & 1713 & 77.4 & 2.34 @xmath59 0.29 & 4.99 @xmath59 0.74 + @xmath61si viii ] @xmath41441 & 1251 & 246.5 & 0.25 @xmath59 0.07 & 0.51 @xmath59 0.10 + n iv ] @xmath41486 & 969 & 47.4 & 0.38 @xmath590.08 & 0.76 @xmath59 0.13 + c iv @xmath41550 & 1530 & 47.9 & 10.02 @xmath59 0.70 & 19.83 @xmath59 2.53 + he ii @xmath41640 & 915 & 54.4 & 2.25 @xmath59 0.25 & 4.34 @xmath59 0.57 + c iii ] @xmath41909 @xmath60 ( si iii ] @xmath121883,1892 ) & 771 & 24.4 & 3.52 @xmath59 0.30 & 7.16 @xmath59 0.96 + @xmath61si vii ] @xmath42148 & 783 & 205.1 & 0.29 @xmath59 0.08 & 0.76 @xmath59 0.15 + c ii ] @xmath42326 ( @xmath60 o iii ] @xmath42321 ) & 1509 & 11.3 & 0.22 @xmath59 0.06 & 0.47 @xmath59 0.09 + @xmath61ne iv ] @xmath42423 & 1338 & 63.5 & 0.76 @xmath59 0.11 & 1.44 @xmath59 0.20 + @xmath61fe xi ] @xmath42649 & 531 & 262.1 & 0.49 @xmath59 0.11 & 0.77 @xmath59 0.13 + mg ii @xmath42800 & 1767 & 7.6 & 1.32 @xmath59 0.17 & 1.91 @xmath59 0.21 + o iii @xmath43133 & 1167 & 35.1 & 0.87 @xmath59 0.15 & 1.13 @xmath59 0.16 + he ii @xmath43204 & 1452 & 54.4 & 0.70 @xmath59 0.20 & 0.89 @xmath59 0.21 + @xmath61ne v ] @xmath43346 & 1167 & 97.1 & 1.41 @xmath59 0.15 & 1.74 @xmath59 0.17 + @xmath61ne v ] @xmath43426 & 1377 & 97.1 & 4.05 @xmath59 0.31 & 4.94 @xmath59 0.36 + @xmath61fe vii ] @xmath43588 & 1155 & 100.0 & 0.39 @xmath59 0.07 & 0.46 @xmath59 0.07 + @xmath61o ii ] @xmath43727 & 1110 & 13.6 & 0.48 @xmath59 0.08 & 0.56 @xmath59 0.08 + @xmath61fe vii ] @xmath43760 & 1356 & 100.0 & 0.74 @xmath59 0.07 & 0.85 @xmath59 0.07 + @xmath61ne iii ] @xmath43869 & 1620 & 41.0 & 2.07 @xmath59 0.18 & 2.35 @xmath59 0.19 + @xmath61ne iii ] @xmath43967 & 1542 & 41.0 & 0.77 @xmath59 0.10 & 0.86 @xmath59 0.10 + @xmath61s ii ] @xmath44072 & 1590 & 10.4 & 0.29 @xmath59 0.05 & 0.33 @xmath59 0.05 + h@xmath62 @xmath44101 & 1689 & 13.6 & 0.30 @xmath59 0.05 & 0.33 @xmath59 0.05 + h@xmath63 @xmath44341 & 1209 & 13.6 & 0.62 @xmath59 0.06 & 0.66 @xmath59 0.06 + @xmath61o iii ] @xmath44363 & 1305 & 35.1 & 0.40 @xmath59 0.05 & 0.43 @xmath59 0.05 + he ii @xmath44686 & 1236 & 54.4 & 0.59 @xmath59 0.05 & 0.60 @xmath59 0.05 + h@xmath13 @xmath44861 & 1320 & 13.6 & 1.00 & 1.00 + @xmath61o iii ] @xmath44959 & 1173 & 35.1 & 5.02 @xmath59 0.38 & 4.96 @xmath59 0.38 + @xmath61o iii ] @xmath45007 & 1215 & 35.1 & 15.39 @xmath59 0.98 & 15.12 @xmath59 0.98 + @xmath61ni xiii ] @xmath45116 ? & 705 & 352.0 & 0.06 @xmath59 0.02 & 0.06 @xmath59 0.02 + @xmath61fe vii ] @xmath45159 & 999 & 100.0 & 0.40 @xmath59 0.04 & 0.38 @xmath59 0.04 + @xmath61fe vi ] @xmath45176 & 1755 & 75.5 & 0.22 @xmath59 0.03 & 0.21 @xmath59 0.03 + @xmath61fe xiv ] @xmath45303 & 708 & 361.0 & 0.82 @xmath59 0.06 & 0.78 @xmath59 0.06 + @xmath61ar x ] @xmath45539 ? & 672 & 422.6 & 0.15 @xmath59 0.03 & 0.14 @xmath59 0.03 + @xmath61fe vii ] @xmath45721 & 1092 & 100.0 & 0.92 @xmath59 0.07 & 0.83 @xmath59 0.07 + he i @xmath45876 & 1653 & 24.6 & 0.29 @xmath59 0.12 & 0.25 @xmath59 0.12 + @xmath61fe vii ] @xmath46087 & 1113 & 100.0 & 1.25 @xmath59 0.10 & 1.08 @xmath59 0.10 + @xmath61o i ] @xmath46300 ( @xmath60 [ s iii ] @xmath46312 ) & 720 & & 0.32 @xmath59 0.03 & 0.27 @xmath59 0.03 + @xmath61fe x ] @xmath46374 ( @xmath60 [ o i ] @xmath46364 ) & 720 & 235.0 & 0.95 @xmath59 0.07 & 0.80 @xmath59 0.07 + @xmath61n ii ] @xmath46548 & 1200 & 14.5 & 1.18 @xmath59 0.22 & 0.98 @xmath59 0.22 + h@xmath11 @xmath46563 & 1167 & 13.6 & 3.39 @xmath59 0.51 & 2.81 @xmath59 0.51 + @xmath61n ii ] @xmath46584 & 1197 & 14.5 & 3.54 @xmath59 0.65 & 2.94 @xmath59 0.66 + @xmath61ni xv ] @xmath46702 ? & 618 & 430.0 & 0.07 @xmath59 0.02 & 0.06 @xmath59 0.02 + @xmath61s ii ] @xmath46716 & 1233 & 10.4 & 0.21 @xmath59 0.03 & 0.17 @xmath59 0.03 + @xmath61s ii ] @xmath46731 & 1230 & 10.4 & 0.26 @xmath59 0.03 & 0.21 @xmath59 0.04 + @xmath61ar v ] @xmath47005 ? & 1422 & 59.8 & 0.18 @xmath59 0.03 & 0.15 @xmath59 0.03 + he i @xmath47065 & 1014 & 24.6 & 0.10 @xmath59 0.02 & 0.08 @xmath59 0.02 + @xmath61ar iii ] @xmath47136 & 1455 & 27.6 & 0.37 @xmath59 0.04 & 0.29 @xmath59 0.05 + @xmath61o ii ] @xmath47325 & 1353 & 13.6 & 0.30 @xmath59 0.04 & 0.24 @xmath59 0.04 + @xmath61s xii ] @xmath47611 & 612 & 504.7 & 0.47 @xmath59 0.05 & 0.36 @xmath59 0.05 + @xmath61ar iii ] @xmath47751 & 1590 & 27.6 & 0.09 @xmath59 0.02 & 0.07 @xmath59 0.02 + @xmath61fe xi ] @xmath47892 & 654 & 262.1 & 0.72 @xmath59 0.06 & 0.55 @xmath59 0.06 + @xmath61s iii ] @xmath49069 & 1257 & 23.3 & 0.72 @xmath59 0.09 & 0.51 @xmath59 0.09 + @xmath61s iii ] @xmath49532 & 1116 & 23.3 & 1.82 @xmath59 0.15 & 1.28 @xmath59 0.17 + & & & & + lllll c iii @xmath4977 & ( 1.47 ) & ( 1.08 ) & ( 1.37 ) & + n iii @xmath4990 & ( 0.64 ) & ( 0.78 ) & ( 0.68 ) & + l@xmath11 @xmath41216 & 34.18 & 40.44 & 35.74 & 30.17 @xmath59 5.65 + n v @xmath41240 & 3.33 & 0.00 & 2.50 & 16.19 @xmath59 2.86 + c ii @xmath41335 & 0.03 & 0.09 ( 4.02 ) & 1.01 & 0.98 @xmath59 0.19 + o iv ] @xmath41402 @xmath60 si iv @xmath41398 & 4.74 & 0.00 & 3.56 & 4.99 @xmath59 0.19 + n iv ] @xmath41486 & 2.99 & 0.00 & 2.24 & 0.76 @xmath59 0.13 + c iv @xmath41550 & 32.52 & 0.03 & 24.40 & 19.83 @xmath59 2.53 + he ii @xmath41640 & 6.30 & 1.29 & 5.05 & 4.34 @xmath59 0.57 + o iii @xmath41663 & 1.71 & 0.09 & 1.31 & + n iii @xmath41750 & 0.81 & 0.05 & 0.62 & + c iii ] @xmath41909 @xmath60 si iii ] @xmath121883,1892 & 6.66 & 1.04 & 5.25 & 7.16 @xmath59 0.96 + c ii ] @xmath42326 @xmath60 o iii ] @xmath42321 & 0.24 & 1.96 & 0.67 & 0.47 @xmath59 0.09 + @xmath61ne iv ] @xmath42423 & 1.62 & 0.01 & 1.22 & 1.44 @xmath59 0.20 + @xmath61o ii ] @xmath42470 & 0.00 & 0.77 & 0.19 & + mg ii @xmath42800 & 0.00 & 3.05 ( 8.54 ) & 2.13 & 1.91 @xmath59 0.21 + he ii @xmath43204 & 0.36 & 0.08 & 0.29 & 0.89 @xmath59 0.21 + @xmath61ne v ] @xmath43346 & 1.83 & 0.00 & 1.38 & 1.74 @xmath59 0.17 + @xmath61ne v ] @xmath43426 & 4.99 & 0.00 & 3.74 & 4.94 @xmath59 0.36 + @xmath61fe vii ] @xmath43588 & 0.46 & 0.00 & 0.34 & 0.44 @xmath59 0.07 + @xmath61o ii ] @xmath43727 & 0.00 & 2.81 & 0.70 & 0.56 @xmath59 0.08 + @xmath61fe vii ] @xmath43760 & 0.63 & 0.00 & 0.48 & 0.85 @xmath59 0.07 + @xmath61ne iii ] @xmath43869 & 1.12 & 1.67 & 1.26 & 2.35 @xmath59 0.19 + @xmath61ne iii ] @xmath43967 + h@xmath64 & 0.50 & 0.68 & 0.55 & 0.86 @xmath59 0.10 + @xmath61s ii ] @xmath44072 & 0.00 & 0.87 & 0.22 & 0.33 @xmath59 0.05 + h@xmath62 @xmath44100 & 0.26 & 0.26 & 0.26 & 0.33 @xmath59 0.05 + h@xmath63 @xmath44340 & 0.47 & 0.47 & 0.47 & 0.66 @xmath59 0.06 + @xmath61o iii ] @xmath44363 & 0.63 & 0.07 & 0.49 & 0.43 @xmath59 0.05 + he ii @xmath44686 & 0.87 & 0.19 & 0.70 & 0.60 @xmath59 0.05 + h@xmath13 & 1.00 & 1.00 & 1.00 & 1.00 + @xmath61o iii ] @xmath44959 & 6.60 & 2.85 & 5.66 & 4.96 @xmath59 0.38 + @xmath61o iii ] @xmath45007 & 19.80 & 8.56 & 16.99 & 15.12 @xmath59 0.98 + @xmath61fe vii ] @xmath45721 & 0.79 & 0.00 & 0.60 & 0.83 @xmath59 0.07 + he i @xmath45876 & 0.02 & 0.13 & 0.05 & 0.25 @xmath59 0.12 + @xmath61fe vii ] @xmath46087 & 1.18 & 0.00 & 0.88 & 1.08 @xmath59 0.10 + @xmath61o i ] @xmath46300 @xmath60 [ s iii ] @xmath46312 & 0.00 & 2.29 & 0.57 & 0.27 @xmath59 0.03 + @xmath61o i ] @xmath46364 @xmath60 @xmath61fe x ] @xmath46374 & 1.17 & 0.71 & 1.06 & 0.80 @xmath59 0.07 + @xmath61n ii ] @xmath46548 & 0.00 & 2.65 & 0.66 & 0.98 @xmath59 0.22 + h@xmath11 @xmath46563 & 2.78 & 2.94 & 2.82 & 2.81 @xmath59 0.51 + @xmath61n ii ] @xmath46584 & 0.00 & 7.65 & 1.91 & 2.94 @xmath59 0.66 + @xmath61s ii ] @xmath46716 & 0.00 & 1.03 & 0.26 & 0.17 @xmath59 0.03 + @xmath61s ii ] @xmath46731 & 0.00 & 1.22 & 0.30 & 0.21 @xmath59 0.04 + @xmath61o ii ] @xmath47325 & 0.00 & 0.98 & 0.24 & 0.24 @xmath59 0.04 + @xmath61s iii ] @xmath49069 & 0.00 & 1.17 & 0.30 & 0.51 @xmath59 0.09 + @xmath61s iii ] @xmath49532 & 0.01 & 3.08 & 0.72 & 1.28 @xmath59 0.17 + & & & & + llllllllll si & 0.001@xmath65 & 0.034@xmath65 & 0.191 & 0.339 & 0.288 & 0.099 & 0.046 & 0.002 & + s & & 0.019 & 0.124 & 0.235 & 0.261 & 0.221@xmath65 & 0.107 & 0.025 & 0.007 + ar & 0.001 & 0.012 & 0.101 & 0.179@xmath65 & 0.229 & 0.228 & 0.150 & 0.077 & 0.019 + fe & @xmath65 & & 0.003 & 0.020@xmath65 & 0.076@xmath65 & 0.188 & 0.252 & 0.193@xmath65 & 0.160 + ni & & & 0.001 & 0.022 & 0.106 & 0.180 & 0.181@xmath65 & 0.170 & 0.144@xmath65 +

stochastic reaction networks are dynamical models of biochemical reaction systems and form a particular class of continuous - time markov chains on . here we provide a fundamental characterisation that connects structural properties of a network to its dynamical features . specifically , we define the notion of ` stochastically complex balanced systems ' in terms of the network s stationary distribution and provide a characterisation of stochastically complex balanced systems , parallel to that established in the 70 - 80ies for deterministic reaction networks . additionally , we establish that a network is stochastically complex balanced if and only if an associated deterministic network is complex balanced ( in the deterministic sense ) , thereby proving a strong link between the theory of stochastic and deterministic networks . further , we prove a stochastic version of the ` deficiency zero theorem ' and show that any ( not only complex balanced ) deficiency zero reaction network has a product - form poisson - like stationary distribution on all irreducible components . finally , we provide sufficient conditions for when a product - form poisson - like distribution on a single ( or all ) component(s ) implies the network is complex balanced , and explore the possibility to characterise complex balanced systems in terms of product - form poisson - like stationary distributions .

1507.02195

improved experimental techniques have made it possible to measure molecular fluctuations at a small scale , creating a need for a stochastic description of molecular data @xcite . typically , biochemical reaction networks are modelled as deterministic systems of ordinary differential equations ( odes ) , but these models assume the individual species are in high concentrations and do not allow for stochastic fluctuation . an alternative is stochastic models based on continuous - time markov chains @xcite . as an example of a stochastic reaction system , consider @xmath1[\kappa_2 ] } 2c,\ ] ] where @xmath2 are positive reaction constants . the network consists of three chemical species @xmath3 , @xmath4 and @xmath5 and two reactions . each occurrence of a reaction modifies the species counts , for example , when the reaction @xmath6 takes places , the amount of @xmath3 and @xmath4 molecules are each decreased by one , while two molecules of @xmath5 are created . the species counts are modelled as a continuous - time markov chain , where the transitions are single occurrences of reactions with transition rates @xmath7 and @xmath8 are the species counts @xcite . when modelled deterministically , the concentrations ( rather than the counts ) of the species change according to an ode system . in a classical paper @xcite , kurtz explored the relationship between deterministic and stochastic reaction systems , using a scaling argument large volume limit to link the dynamical behaviour of the two types of systems to each other . other , mainly recent work , also points to close connections between the two types of systems @xcite . in this paper we explore this relationship further . a fundamental link between structural network properties and dynamical features of deterministic reaction networks has been known since the 1970s and 1980s with the work of horn , jackson and feinberg @xcite . specifically , their theory concerns the existence and uniqueness of equilibria in _ complex balanced _ systems , with the ` deficiency zero theorem ' playing a central role in this context . complex balanced systems were called cyclic balanced systems by boltzmann . they have attractive analytical and physical properties ; for example a ( pseudo-)entropy might be defined which increases along all trajectories ( boltzmann s h - theorem ) @xcite . a parallel theory for the stochastic regime is not available , and the very concept of `` complex balanced '' does not currently have a stochastic counterpart . in this paper we develop a theory to fill this gap . we define _ stochastically complex balanced _ systems through properties of the stationary distribution , and we prove results for stochastic reaction networks that are in direct correspondence with the results for deterministic models . in particular , we prove a parallel statement of the deficiency zero theorem and show that all deficiency zero reaction networks have product - form poisson - like stationary distributions , irrespectively whether they are complexed balanced or not . in fact , in the non - complexed balanced case , the network is complex balanced on the boundary of the state space . a second target of our study concerns product - form stationary distributions . such distributions are computationally and analytically tractable and appear in many areas of applied probability , such as , queueing theory @xcite , petri net theory @xcite , and stochastic reaction network theory @xcite . specifically , a complex balanced mass - action network has a product - form poisson - like stationary distribution on every irreducible component @xcite . as an example , the stationary distribution of the complex balanced reaction system is @xmath9 where @xmath10 is an irreducible component of the state space @xmath11 and @xmath12 is a normalising constant . we expand the above result on mass - action systems and give general conditions under which the converse statement is true . in particular , we are interested in providing a structural characterisation of the networks with product - form poisson - like stationary distributions . however , this class of networks is strictly larger than that of complex balanced networks , and a full characterisation seems hard to achieve . we illustrate this with examples . we first introduce the necessary notation and background material ; see @xcite for general references . we assume standard knowledge about continuous - time markov chains . we let @xmath13 , @xmath14 and @xmath15 be the real , the non - negative real and the positive real numbers , respectively . also let @xmath16 be the natural numbers including 0 . for any real number @xmath17 , @xmath18 denotes the absolute value of @xmath19 . moreover , for any vector @xmath20 , we let @xmath21 be the @xmath22th component of @xmath23 , @xmath24 the euclidean norm , and @xmath25 the infinity norm , that is , @xmath26 . for two vectors @xmath27 , we write @xmath28 ( resp.@xmath29 ) and @xmath30 ( resp.@xmath31 ) , if the inequality holds component - wise . further , we define @xmath32 to be one if @xmath30 , and zero otherwise , and similarly for the other inequalities . if @xmath33 then @xmath23 is said to be positive . finally , @xmath34 denotes the index set of the non - zero components . for example , if @xmath35 then @xmath36 . if @xmath37 and @xmath38 , we define @xmath39 with the conventions that @xmath40 and @xmath41 . a reaction network is a triple @xmath42 , where @xmath43 is a set of @xmath44 species , @xmath45 is a set of @xmath46 complexes , and @xmath47 is a set of @xmath48 reactions , such that @xmath49 for all @xmath50 . the complexes are linear combinations of species on @xmath16 , identified as vectors in @xmath51 . a reaction @xmath52 is denoted by @xmath53 . we require that every species is part of at least one complex , and that every complex is part of at least one reaction . in this way , there are no `` superfluous '' species or complexes and @xmath54 is completely determined by the set of reactions @xmath55 , which we allow to be empty . in , there are @xmath56 species ( @xmath57 ) , @xmath58 complexes ( @xmath59 ) , and @xmath60 reactions . given a reaction network @xmath54 , the _ reaction graph _ of @xmath54 is the directed graph with node set @xmath45 and edge set @xmath55 . we let @xmath61 be the number of linkage classes ( connected components ) of the reaction graph . a reaction @xmath62 is _ terminal _ if any directed path that starts with @xmath53 is contained in a closed directed path . we let @xmath63 be the set of terminal reactions . a reaction network @xmath54 is _ weakly reversible _ , if every reaction is terminal . the network in is weakly reversible , since both reactions are terminal . the _ stoichiometric subspace _ of @xmath54 is the linear subspace of @xmath51 given by @xmath64 for @xmath65 , the sets @xmath66 are called the _ stoichiometric compatibility classes _ of @xmath54 ( fig . @xmath67 ) . for the network in , @xmath68 , which is 2-dimensional . we will consider a reaction network @xmath54 either as a deterministic dynamical system on the continuous space @xmath69 , or as a stochastic dynamical system on the discrete space @xmath0 . in the deterministic case , the evolution of the species concentrations @xmath70 at time @xmath71 is modelled as the solution to the ode @xmath72 for some functions @xmath73 and an initial condition @xmath74 . we require that the functions @xmath75 are continuously differentiable , and that @xmath76 if and only if @xmath77 . such functions are called _ rate functions _ , they constitute a _ deterministic kinetics _ @xmath78 for @xmath54 , and the pair @xmath79 is called a _ deterministic reaction system_. if @xmath80 for all reactions , then the constants @xmath81 are referred to as _ rate constants _ and the modelling regime is referred to as _ deterministic mass - action kinetics_. in this case , the pair @xmath82 is called a _ deterministic mass - action system _ , where @xmath83 is the vector of rate constants . in the stochastic setting , the evolution of the species counts @xmath84 at time @xmath71 is modelled as a continuous - time markov chain with state space @xmath0 . at any state @xmath85 , the states that can be reached in one step are @xmath86 for @xmath62 , with transition rates @xmath87 . the functions @xmath88 are called _ rate functions _ , and we require that @xmath89 if and only if @xmath90 . a choice of these functions constitute a _ stochastic kinetics _ @xmath78 for @xmath54 and the pair @xmath79 is called a _ stochastic reaction system_. if the reaction @xmath53 occurs at time @xmath71 , then the new state is @xmath91 where @xmath92 denotes the previous state . if for any reaction @xmath62 @xmath93 then the constants @xmath81 are known as _ rate constants _ , as in the deterministic case , and the modelling regime is referred to as _ stochastic mass - action kinetics_. the pair @xmath82 is , in this case , called a _ stochastic mass - action system_. the evolution of the stochastic as well as the deterministic reaction system is confined to the stoichiometric compatibility classes , @xmath94 in fact , @xmath95 , as @xmath96 takes values in @xmath0 . let @xmath42 be a reaction network . a. a reaction network @xmath97 is a _ subnetwork _ of @xmath54 if @xmath98 . in this case , it follows that @xmath99 and @xmath100 . b. a system @xmath101 , deterministic or stochastic , is a _ subsystem _ of a system @xmath79 if @xmath102 is a subnetwork of @xmath54 and the rate functions agree on the reactions in @xmath103 . c. the subnetwork @xmath104 given by the set of terminal reactions @xmath63 is the _ terminal network _ of @xmath54 . we denote @xmath105 . furthermore , the subsystem @xmath106 of @xmath79 is called the _ terminal system _ of @xmath79 . the connected components of the reaction graph of the terminal network of @xmath54 are called _ terminal strongly connected component _ of @xmath54 . for any complex @xmath107 in @xmath108 , we denote by @xmath109 the subsystem of @xmath54 whose reaction graph is the terminal strongly connected component containing @xmath107 as node . as an example , consider the mass - action system @xmath110[\kappa_2]}2b\cee{<-[\kappa_3]}a\cee{->[\kappa_4]}0\cee{<=>[\kappa_5][\kappa_6]}c.\ ] ] here , there are two terminal strongly connected components , which are @xmath111 and @xmath112 . in particular , @xmath113 is equal to @xmath114 and is given by @xmath110[\kappa_2]}2b.\ ] ] finally , if @xmath82 is a mass - action system , any subsystems @xmath101 is a mass - action systems as well and can be denoted by @xmath115 . in this section we will recapitulate the known characterisation of existence and uniqueness of positive equilibria in complex balanced systems and the connection between complex balanced systems and deficiency zero reaction networks . as we will show in the subsequent section , this characterisation can be fully translated into a similar characterisation for stochastic reaction networks . we start with a definition . [ def : cb ] a deterministic reaction system @xmath79 is said to be _ complex balanced _ if there exists a positive _ complex balanced equilibrium _ , that is , a positive equilibrium point @xmath116 for the system , such that @xmath117 the name ` complex balanced ' refers to the fact that the flow , at equilibrium , entering into the complex @xmath107 equals the flow exiting from the complex . as an example , the mass - action system in is complex balanced for any choice of @xmath118 and @xmath119 is a complex balanced equilibrium . the class of complex balanced systems is an extension of the class of detailed balanced mass - action systems @xcite . for mass - action systems , becomes @xmath120 with the convention that @xmath121 if @xmath122 . in the case of mass - action kinetics , we extend definition [ def : cb ] to the stochastic case , by saying that a stochastic mass - action system @xmath82 is complex balanced if the deterministic mass - action system @xmath82 is complex balanced . we might therefore refer to complex balanced mass - action systems without specifying whether they are stochastically or deterministically modelled . the next theorem is a slight generalization of a classical result @xcite , which provides the backbone for the further characterisation . the generalization includes a property of non - negative equilibria [ thm : complex_balanced ] if a deterministic reaction system @xmath79 is complex balanced , then @xmath54 is weakly reversible . moreover , if @xmath78 is mass - action kinetics , then all equilibria are complex balanced , that is , they fulfil . moreover , there exists exactly one positive equilibrium in each stoichiometric compatibility class , which is locally asymptotically stable . as we are not aware of a proof of this more general formulation , we provide one in appendix [ sec : proofs ] . the _ deficiency _ plays an important role in the study of complex balanced systems . the deficiency of @xmath54 is defined as @xmath123 where @xmath46 is the cardinality of @xmath45 , @xmath61 is the number of linkage classes of the reaction graph of @xmath54 and @xmath124 is the dimension of the stoichiometric subspace @xmath125 @xcite . the definition hides the geometrical interpretation of the deficiency , which we now will explore . let @xmath126 be a basis of @xmath127 . further , define @xmath128 for @xmath62 . let @xmath129 . then @xmath130 @xcite . the space @xmath131 is linearly isomorphic to the stoichiometric subspace @xmath125 if and only if @xmath132 . specifically , consider the homomorphism @xmath133 for @xmath62 , we have @xmath134 and @xmath135 is thus a surjective homomorphism . therefore , @xmath136 which implies that @xmath137 is an isomorphism if and only if @xmath132 . it further follows that the deficiency is a non - negative number . we state here a useful lemma on the deficiency of subnetworks . [ lem : subnet ] let @xmath54 be a reaction network with deficiency @xmath138 . then , the deficiency of any subnetwork of @xmath54 is smaller than or equal to @xmath138 . let @xmath98 and let @xmath102 be the corresponding subnetwork with deficiency @xmath139 . further , let @xmath140 and @xmath141 be the equivalent of @xmath131 and @xmath125 for @xmath102 , respectively . by and since @xmath140 is a subspace of @xmath131 , we have @xmath142 which concludes the proof . we next state two classical results which elucidate the connection between complex balanced systems and deficiency zero systems . a proof of the first and of the second result can be found in @xcite and in @xcite , respectively . the results draw a connection between graphical and dynamical properties of a network . theorem [ thm : equilibrium_boundary ] is given here in a wider formulation than in @xcite ( see appendix [ sec : proofs ] for a proof ) . [ thm : deficiency_zero_iff ] the mass - action system @xmath82 is complex balanced for any choice of @xmath143 if and only if @xmath54 is weakly reversible and its deficiency is zero . [ thm : equilibrium_boundary ] consider a deterministic reaction system @xmath79 , and assume that the deficiency of @xmath54 is zero . if @xmath144 is an equilibrium point and @xmath62 , then @xmath145 only if @xmath53 is terminal moreover , if @xmath78 is mass - action kinetics with rate constants @xmath143 and @xmath145 , then the projection of @xmath146 onto the species space of @xmath147 is a complex balanced equilibrium of @xmath148 . it follows from theorem [ thm : equilibrium_boundary ] that an equilibrium point satisfies for the terminal system , though it is not necessarily a positive equilibrium of @xmath149.the deficiency zero theorem , in the following formulation , is a consequence of the three previous theorems : [ thm : deficiency_zero ] consider a deterministic reaction system @xmath79 for which the deficiency is zero . then the following statements hold : a. if @xmath54 is not weakly reversible , then there exists no positive equilibria ; b. if @xmath54 is weakly reversible and @xmath78 is mass - action kinetics , then there exists within each stoichiometric compatibility class a unique positive equilibrium , which is asymptotically stable . the original formulation is richer than the one presented here @xcite , and @xmath150 . ( a ) the stoichiometric compatibility classes are of the form @xmath151 . ( b ) the two irreducible components on @xmath152 are shown ( black circles and square ) , together with the possible transitions between the states . all states within a component are accessible from each other . the `` square '' component has no active reactions , both reactions are active on the `` black circles '' component . the grey states are transient states which are not in any irreducible component . , scaledwidth=80.0% ] to characterise the stochastic dynamics we introduce the following terminology . let @xmath42 be a reaction network . a. a reaction @xmath62 is _ active _ on @xmath85 if @xmath90 . b. a state @xmath153 is _ accessible _ from a state @xmath85 if there is a sequence of @xmath154 reactions @xmath155 such that 1 . @xmath156 , 2 . @xmath157 is active on @xmath158 for all @xmath159 . let @xmath54 be a reaction network . a non - empty set @xmath160 is an _ irreducible component _ of @xmath54 if for all @xmath161 and all @xmath153 , @xmath162 is accessible from @xmath146 if and only if @xmath163 . a reaction network @xmath54 is _ essential _ if the state space is a union of irreducible components . a reaction network @xmath54 is _ almost essential _ if the state space is a union of irreducible components except for a finite number of states . an essential network is also almost essential . a weakly reversible reaction network is essential @xcite . conditions for being essential can be found in @xcite . any irreducible component is contained in some stoichiometric compatibility class , and a stoichiometric compatibility class may contain several irreducible components ( fig . @xmath164 ) . the stationary distribution @xmath165 on an irreducible component @xmath166 is unique , if it exists . it is characterised by the _ master equation _ @xcite : @xmath167 for all @xmath161 . let @xmath96 denote the stochastic process associated with the system . if @xmath168 follows the law of @xmath165 at time @xmath169 , then the distribution of @xmath96 is @xmath165 for all future times @xmath170 . in this sense , the stationary distribution describes a state of equilibrium of the system . moreover , if @xmath165 exists , then @xmath171 provided that @xmath172 with probability one . as discussed in section [ sec : introduction ] , a connection between mass - action complex balanced systems and their stationary distribution has been made in @xcite : [ thm : anderson ] let @xmath82 be a complex balanced mass - action system . then , there exists a unique stationary distribution on every irreducible component @xmath166 , and it is of the form @xmath173 where @xmath174 is a positive complex balanced equilibrium of @xmath82 and @xmath175 is a normalising constant . in this section we derive stochastic statements corresponding to theorem [ thm : complex_balanced]-[thm : deficiency_zero ] . some of the proofs are deferred to appendix [ sec : proofs ] . we begin with a definition . for an irreducible component @xmath166 , the set @xmath176 of _ active reactions _ on @xmath166 consists of the reactions @xmath62 that are active on some @xmath161 . the subnetwork @xmath177 is called the _ @xmath166-network _ of @xmath54 and the subsystem @xmath178 of @xmath79 is called the _ @xmath166-system _ of @xmath79 . the reactions that are active on @xmath166 determine the dynamics of the stochastic system on @xmath166 . to study the stationary distributions , it is therefore convenient to analyse the @xmath166-systems . note that @xmath176 is empty if and only if @xmath166 consists of a single state . as an example , consider the deficiency zero network , @xmath179 all molecules of @xmath3 and @xmath4 are irreversibly consumed through @xmath180 and @xmath181 , thus the only active reactions on an irreducible component @xmath182 are @xmath183 . the @xmath166-network is therefore @xmath183 , which differs from the terminal system @xmath183 , @xmath111 . the next proposition states that for a deficiency zero reaction network @xmath184 for any irreducible component @xmath166 . note that proposition [ prop : terminal_strong_linkage_classes ] does not hold in general , for example , @xmath185 has @xmath186 for any @xmath187 , while @xmath188 . [ prop : terminal_strong_linkage_classes ] let @xmath54 be a reaction network and @xmath166 an irreducible component such that @xmath189 has deficiency zero . then , @xmath189 is a subnetwork of @xmath104 . in particular , this is true if the deficiency of @xmath54 is zero . see appendix [ sec : proofs ] for a proof . proposition [ prop : terminal_strong_linkage_classes ] can be useful because @xmath176 might be difficult to find , especially if there are many complexes . on the other hand , terminal reactions are easily identified by means of the reaction graph . the next definitions are inspired by definition [ def : cb ] . [ defi : complex_balanced_stationary_distribution ] let @xmath190 be a stochastic reaction system . a stationary distribution @xmath165 on an irreducible component @xmath166 is said to be _ complex balanced _ if @xmath191 for a mass - action system , becomes @xmath192 for any @xmath193 and @xmath161 , with the convention that @xmath121 if @xmath194 . in developing the theory for complex balanced equilibria in the deterministic setting , an important role is played by requiring positivity of the complex balanced equilibrium . our aim is to introduce a similar concept for the stochastic systems . in the deterministic setting , if a state @xmath195 is positive then every rate function calculated on @xmath196 is positive . we find inspiration from this to give the next definition : an irreducible component @xmath166 is _ positive _ if @xmath197 . equivalently , an irreducible component @xmath166 is _ positive _ if all reactions are active on @xmath166 . the next definition follows naturally by analogy with the deterministic setting . [ def : stoch_cb ] a stochastic reaction system @xmath79 is said to be _ stochastically complex balanced _ if there exists a complex balanced stationary distribution on a positive irreducible component . if @xmath166 is positive , then @xmath198 and a complex balanced stationary distribution on @xmath166 satisfies with @xmath199 replaced by @xmath45 . note the similarity between definition [ def : stoch_cb ] and the definition of a complex balance equilibrium ( definition [ def : cb ] ) : the positivity of @xmath166 plays the role of the positivity in definition [ def : cb ] . also note the close similarity between and . [ thm : stochastic_complex_balanced ] let @xmath79 be a stochastic reaction system , and let @xmath166 be an irreducible component . if there exists a complex balanced stationary distribution @xmath165 on @xmath166 then @xmath189 is weakly reversible . moreover , if @xmath78 is mass - action kinetics with rate constants @xmath143 , there exists a complex balanced stationary distribution @xmath165 on @xmath166 if and only if the @xmath166-system of @xmath82 is complex balanced . if this is the case , then @xmath165 has the form @xmath200 where @xmath174 is a positive complex balanced equilibrium of @xmath201 and @xmath175 is a normalising constant . the proof is in appendix [ sec : proofs ] . it is shown in @xcite that the stationary distribution @xmath202 is independent of the choice of complex balanced equilibrium @xmath174 of the @xmath166-system , provided that it is positive . we are now ready to derive stochastic versions of theorem [ thm : complex_balanced]-[thm : deficiency_zero ] . in addition , we will show that a stochastically complexed balanced mass - action system is complex balanced and _ vice versa_. hence , we will show that the deterministic and stochastic systems are intimately connected . the next corollary is an analogue of theorem [ thm : complex_balanced ] . [ cor : stochastic_complex_balanced_positive ] if a stochastic reaction system @xmath79 is stochastically complex balanced then @xmath54 is weakly reversible . moreover , a mass - action system @xmath82 is stochastically complex balanced if and only if it is complex balanced . if this is case , then on every irreducible component @xmath166 there exists a unique stationary distribution @xmath165 . such @xmath165 is a complex balanced stationary distribution and it has the form , where @xmath174 is a positive complex balanced equilibrium of @xmath82 . if @xmath166 is positive , then @xmath203 therefore , by theorem [ thm : stochastic_complex_balanced ] if @xmath79 is stochastically complex balanced then @xmath54 is weakly reversible . moreover , if @xmath78 is mass - action kinetics with rate constants @xmath143 , it follows from theorem [ thm : stochastic_complex_balanced ] that there exists a complex balanced stationary distribution on @xmath166 if and only if @xmath82 is complex balanced . in this case , by theorem [ thm : anderson ] , a stationary distribution exists on every irreducible component and it is of the form . by theorem [ thm : stochastic_complex_balanced ] , it is a complex balanced stationary distribution . corollary [ cor : stochastic_complex_balanced_positive ] might be considered a stochastic version of theorem [ thm : complex_balanced ] , especially if is taken to be equivalent to `` asymptotic stability '' for a deterministic equilibrium . part of the corollary is known @xcite ( see also theorem [ thm : anderson ] ) , and the whole corollary might therefore be considered as an extension of the result in @xcite on mass - action systems . in this sense , theorem [ thm : stochastic_complex_balanced ] provides an even more general version , which deals with complex balanced subsystems of @xmath82 . we now state the parallel versions of theorem [ thm : deficiency_zero_iff]-[thm : deficiency_zero ] for the stochastic setting . [ cor : stoch_deficiency_zero_iff ] the mass - action system @xmath82 is stochastically complex balanced for any choice of @xmath143 if and only if @xmath54 is weakly reversible and its deficiency is zero . the result is an immediate consequence of corollary [ cor : stochastic_complex_balanced_positive ] and theorem [ thm : deficiency_zero_iff ] . [ thm : stationary_boundary ] consider a stochastic reaction system @xmath79 , and assume the deficiency of @xmath54 is zero . let @xmath146 be a state in an irreducible component @xmath166 and let @xmath53 in @xmath55 . then , @xmath204 only if @xmath53 is terminal . moreover , if @xmath78 is mass - action kinetics , then on @xmath166 the stationary distribution has the form @xmath205 where @xmath174 is a positive complex balanced equilibrium for the terminal system , and @xmath206 is a normalising constant . the proof is in appendix [ sec : proofs ] . [ thm : stoch_deficiency_zero ] consider a stochastic reaction system @xmath79 , and assume that the deficiency of @xmath54 is zero . then the following statements hold : a. if @xmath54 is not weakly reversible , then there exist no positive irreducible components ; b. if @xmath54 is weakly reversible , then @xmath54 is essential , and if @xmath78 is mass - action kinetics then there exists a unique stationary distribution on every irreducible component . the proof of the theorem is in appendix [ sec : proofs ] . in case ( i ) , theorem [ thm : stationary_boundary ] provides the form of the stationary distribution . hence we have characterised the stationary distribution for any deficiency zero reaction system , irrespectively whether it is complex balanced or not . consider the two stochastic mass - action systems @xmath207[\kappa_2 ] } b,\quad 10a \cee{<=>[\kappa_3][\kappa_4 ] } 10b\quad\text{and}\quad a \cee{<=>[\kappa_1][\kappa_2 ] } b,\quad 10a \cee{->[\kappa_3 ] } 0.\ ] ] the behaviours of the two corresponding deterministic systems differ substantially , while the behaviours of the stochastic systems are equivalent on the irreducible components @xmath208 with @xmath209 an integer . indeed , in both cases the @xmath210-system is @xmath207[\kappa_2 ] } b,\ ] ] which is complex balanced ( theorem [ thm : deficiency_zero_iff ] ) . it follows from theorem [ thm : stochastic_complex_balanced ] that the stationary distribution on @xmath210 is @xmath211 for a suitable normalizing constant @xmath212 . the stationary distributions are complex balanced , but since @xmath210 is not positive in either of the two networks , we can not conclude that the systems are stochastically complex balanced . indeed , they are not for some choice of rate constants ( corollary [ cor : stoch_deficiency_zero_iff ] ) . incidentally , note that the second network is not almost essential . the above results draw parallels between stochastic and deterministic reaction networks . if a mass - action system is ( stochastically ) complex balanced , then the stationary distribution on every irreducible component is a product - form poisson - like distribution . does the reverse statement hold true too ? if the stationary distribution is a product - form poisson - like distribution on some , or all irreducible components , does it follow that the system is complex balanced ? in the spirit of the first part of the paper we would like to achieve a full characterisation of stochastic systems with product - form poisson - like stationary distributions . however , even though the hypothesis of theorem [ thm : main_thm ] below is rather general , a full characterisation seems hard to achieve . [ thm : main_thm ] let @xmath54 be an almost essential reaction network , @xmath213 a vector of rate constants and @xmath116 a vector with positive entries . the probability distribution @xmath214 $ ] , defined by is a stationary distribution for the stochastic mass - action system @xmath82 for all irreducible components @xmath215 of @xmath54 if and only if @xmath174 is a complex balanced equilibrium for @xmath82 . by theorem [ thm : anderson ] , if @xmath216 is a complex balanced equilibrium for @xmath82 , then the stationary distribution on all irreducible components @xmath160 is of the form . oppositely , assume that is the stationary distribution on @xmath166 for the stochastic mass - action system @xmath82 , for all irreducible components @xmath166 . since @xmath54 is almost essential , there exists a constant @xmath78 such that any states @xmath146 with @xmath217 belongs to an irreducible component @xmath166 . for any @xmath85 , such that @xmath218 we have that @xmath219 and @xmath220 for all @xmath62 . then , since is a stationary distribution and since @xmath146 and @xmath221 are in the same irreducible component for all @xmath62 , we have from @xmath222 for all @xmath161 satisfying . further , using , equation becomes @xmath223 which , by rearranging terms , leads to @xmath224 the equality holds for all @xmath85 satisfying , therefore the polynomials on the two sides of are equal . for any @xmath225 , let @xmath226 be the polynomial @xmath227 the monomial with maximal degree in @xmath228 is @xmath229 , and these differ for all complexes @xmath225 . this implies that @xmath228 , @xmath225 , are linearly independent on @xmath13 , and thus , the polynomials on the two sides of are equal if and only if @xmath230 hence , @xmath174 is a complex balanced equilibrium for @xmath82 and the proof is completed . to infer the existence of positive complex balanced equilibria in theorem [ thm : main_thm ] , the assumptions of the theorem could be weakened . specifically , it is only required that holds for a set of states whose geometry and cardinality allow us to conclude that the polynomials on the two sides of are the same . for to hold , we need @xmath146 to be in a irreducible component and we require @xmath219 and @xmath220 for all reactions @xmath62 , as well as the stationary distribution evaluated in @xmath146 and @xmath231 to be of the form . if a state @xmath146 satisfies this , we call it a _ good state_. a more general condition than being almost essential could be chosen case by case and depends on the monomials appearing in . for example , if the set of complexes coincides with the set of species , then the polynomials in are linear and the existence of @xmath232 _ good states _ in general position implies the existence of a positive complex balanced equilibrium . in general , let @xmath233 be the total degree of the polynomials in . then it is sufficient to have @xmath44 lines in general position with more than @xmath234 good states on each of them . therefore , to conclude that a system is complex balanced it is sufficient to check the behaviour of a finite number of states , lying on a finite number of irreducible components . however , it follows from examples [ ex : not_complex_balanced ] and [ ex : not_complex_balanced_2 ] that the existence of arbitrarily many good states on a few irreducible components does not imply the existence of a positive complex balanced equilibrium in general . finally , in order to postulate that the mass - action system is complex balanced , it is necessary that the vector @xmath174 appearing in theorem [ thm : main_thm ] is the same for every irreducible component , as shown in example [ ex : not_complex_balanced_different_c ] . the following examples are also meant to give an idea of why it is hard to obtain a full characterization of stochastic mass - action systems with a product - form poisson - like stationary distribution on some irreducible component . [ ex : not_complex_balanced ] let @xmath235 and let @xmath236 be an integer . consider the stochastic mass - action system @xmath237{\rho } 2a,\ ] ] where @xmath238 and @xmath239 are the rate constants . the reaction network is almost essential . it is shown in appendix [ sec : calculations ] that the stationary distribution on the irreducible component @xmath208 has the form with @xmath240 , namely @xmath241 where @xmath212 is a normalising constant . however , the mass - action system is not complex balanced as the reaction network is not weakly reversible ( theorem [ thm : complex_balanced ] ) . in particular , by theorem [ thm : main_thm ] , not all irreducible components can have a stationary distribution of the form with @xmath240 . trivially , the absorbing states @xmath242 and @xmath243 have it . additionally , we should point out that there is not an _ equivalent _ system on @xmath166 ( that is , a stochastic mass - action system with the same transition rate matrix on the states of @xmath166 as ) which is complex balanced . consider the case @xmath244 . since the transition from @xmath245 to @xmath246 is possible according to , any equivalent mass - action system must contain the reaction @xmath247 , with rate constant @xmath248 . it can be further shown that any equivalent weakly reversible mass - action system must contain the connected component ( a+b ) at ( 0,0 ) @xmath249 ; ( 2b ) at ( 2,0.6 ) @xmath250 ; ( 2a ) at ( 4,0 ) @xmath251 ; ( 2b ) edge node@xmath248 ( 2a ) ( 2a ) edge node@xmath252 ( a+b ) ( a+b ) edge node@xmath248 ( 2b ) ; this prevents the system from being complex balanced , since there is not a @xmath253 fulfilling for the three complexes @xmath250 , @xmath254 and @xmath249 . let @xmath255 and let @xmath236 be an integer . consider the modification of example [ ex : not_complex_balanced ] given by @xmath256[\rho_2 ] } b\qquad 2b \cee{<=>[\phantom{\theta-}\rho_1+\rho_3\phantom{()1}][\rho_3 ] } 2a,\ ] ] which is weakly reversible . if we let @xmath257 and @xmath258 , then the system reduces to that of example [ ex : not_complex_balanced ] by removing the two reversible reactions . it can be shown that for any parameter choice , is still a stationary distribution on the irreducible component @xmath259 . however , for some choice of parameters the mass - action system is not complex balanced . this can be seen either by direct computation on the system of complex balance equations or by noting that the deficiency of the network is 1 , so there must be a choice of parameters which prevents positive complex balanced equilibria by theorem [ thm : deficiency_zero_iff ] . it can be further shown that irreducible components different from @xmath166 do not possess a product - form poisson - like stationary distribution . [ ex : not_complex_balanced_2 ] consider the stochastic mass - action system with @xmath235 and @xmath260 two positive integers , @xmath261{\rho\theta_1\theta_2 } b & 2b & \xrightarrow{\rho(\theta_1+\theta_2 - 1 ) } 2a\\[-0.3 cm ] 3a & \xrightarrow[\phantom{\rho(\theta_1+\theta_2 - 1)}]{\rho } a+2b\qquad & 2a+b & \xrightarrow[\phantom{\rho(\theta_1+\theta_2 - 1)}]{\rho } 3b . \end{aligned}\ ] ] the reaction network is almost essential . for any @xmath262 , consider the irreducible component @xmath263 . then @xmath264 and @xmath265 , defined as in , are the ( unique ) stationary distributions on the irreducible components @xmath266 and @xmath267 , respectively . for the relevant calculations see appendix [ sec : calculations ] . however , the mass - action system is not complex balanced , since the reaction network is not weakly reversible ( theorem [ thm : complex_balanced ] ) . [ ex : not_complex_balanced_different_c ] theorem [ thm : main_thm ] can be also used to compute the stationary distribution of a stochastic mass - action system which behaves as a complex balanced system on the irreducible components . consider the weakly reversible ( and therefore essential ) stochastic mass - action system @xmath268[\kappa_2 ] } 2a\qquad a+b \cee{<=>[\kappa_3][\kappa_4 ] } 2a+b.\ ] ] on every irreducible component @xmath269 , @xmath262 , the associated continuous time markov chain , which describes the evolution of the counts of @xmath3 , has the same distribution as the process associated with @xmath270[\kappa_2+\kappa_4\theta]}2a,\ ] ] because the transition rates coincide . the latter system is complex balanced for any choice of rate constants . the stationary distribution has the form ( theorem [ thm : main_thm ] ) @xmath271 for some positive constant @xmath212 . the latter gives the stationary distribution of the original system as well . however , the rate of the poisson distribution does depend on @xmath272 , in which case the original system can not be complex balanced ( corollary [ cor : stochastic_complex_balanced_positive ] ) . for the same reason the example does not contradict theorem [ thm : main_thm ] . there are not many means to explicitly calculate the stationary distribution of a stochastic mass - action system . as an example , theorem [ thm : stochastic_complex_balanced ] can be used to determine the stationary distributions of mass - action systems like @xmath273[\kappa_2]}d,\quad 2a\cee{<=>[\kappa_3][\kappa_4]}2b,\quad a\cee{->[\kappa_5]}0.\ ] ] indeed , for any irreducible component @xmath166 different from @xmath274 , the @xmath166-system is given by @xmath273[\kappa_2]}d,\ ] ] which is weakly reversible and has deficiency zero , therefore it is complex balanced . hence , the stationary distribution on @xmath166 has the form @xmath275 where @xmath276 and @xmath277 denote the entries relative to @xmath5 and @xmath131 , respectively . alternatively , since the terminal system is given by @xmath273[\kappa_2]}d,\quad 2a\cee{<=>[\kappa_3][\kappa_4]}2b,\ ] ] theorem [ thm : stationary_boundary ] can be used to compute the stationary distribution . on every irreducible component @xmath166 , it is given by @xmath278 which is equivalent to the previous formula since @xmath279 and @xmath280 are constantly @xmath281 on all irreducible components . if the system does not fulfil the conditions of theorem [ thm : stochastic_complex_balanced ] and neither can be cast as a birth - death process , theorem [ thm : main_thm ] might be useful . the following mass - action system is considered in @xcite : @xmath282 by theorem [ thm : main_thm ] , the stationary distribution can not be poisson . indeed , it is given by the distribution of @xmath283 , where @xmath284 and @xmath285 are two independent poisson random variables with rates @xmath286 and @xmath287 , respectively . hence , @xmath288 in @xcite , the following system is also considered : @xmath289[\kappa_2]}a\qquad 2a\cee{<=>[\kappa_3][\kappa_4]}3a.\ ] ] it has the stationary distribution @xmath290}{i(i-1)(i-2)+\theta_3i}\quad\text{for } x\in{\mathbb{n}},\ ] ] where @xmath291 , @xmath292 , @xmath293 and @xmath294 is a normalising constant . it is interesting that @xmath295 is a poisson distribution if and only if @xmath296 . in fact , and in accordance with our results , the mass - action system is complex balanced if and only if @xmath296 . corollary [ cor : stoch_deficiency_zero_iff ] provides a characterisation of reaction networks that are stochastically complex balanced for any choice of rate constants . it is natural to wonder whether a stationary distribution of the form on some irreducible component @xmath166 for all choices of rate constants implies something specific about the @xmath166-system . if for specific form we intend deficiency zero and weakly reversible , this is not the case , as this is violated in example [ ex : not_complex_balanced_different_c ] . however , in example [ ex : not_complex_balanced_different_c ] the system might be described _ equivalently _ by means of a weakly reversible deficiency zero system for any irreducible component . the question of whether this is always true remains open . we provide here two more examples . [ ex : equivalent_to_complex_balanced_finite_case ] consider the stochastic mass - action system @xmath297}2b\qquad a+3b\cee{->[\kappa_2]}3a+b.\ ] ] the underlying reaction network is considered in figure [ figure ] . on the irreducible component @xmath298 , the markov chain associated with the system has the same distribution as the markov chain associated with @xmath110[3\kappa_2]}2b,\ ] ] since the transition rates coincide . it is interesting to note that the dynamics of the two systems are different when they are deterministically modelled @xcite . due to theorem [ thm : deficiency_zero_iff ] , the latter system is complex balanced for any choice of rate constants . therefore , by theorem [ thm : main_thm ] , the stationary distribution on @xmath166 has the form on both systems for any choice of rate constants . the same argument does not hold , in this case , for the other irreducible components . the same phenomenon as in example [ ex : equivalent_to_complex_balanced_finite_case ] is observed in the stochastic mass - action system @xmath297}3a+b\qquad a+3b\cee{->[\kappa_2]}2b.\ ] ] on the irreducible component @xmath299 , the markov chain associated with the system has the same distribution as the markov chain associated with @xmath110[\kappa_2]}3a+b,\ ] ] since the transition rates coincide , and the latter network is weakly reversible and has deficiency zero . here we state some preliminary results that will be needed in appendix [ sec : proofs ] . [ lem : path ] let @xmath54 be a reaction network . if @xmath300 is a directed path in the reaction graph of @xmath54 , and @xmath301 , then @xmath302 is accessible from @xmath146 . first , note that @xmath303 it is sufficient to note that if @xmath301 , then for any @xmath304 , we have @xmath305 this concludes the proof . [ lem : weakly_reversible ] let @xmath166 be an irreducible component such that @xmath189 has deficiency zero . then , @xmath189 is weakly reversible . in particular , if @xmath54 has deficiency zero , @xmath189 has deficiency zero and is weakly reversible for every irreducible component @xmath166 . if @xmath176 is empty then @xmath189 is weakly reversible and there is nothing to prove . otherwise , if @xmath176 is non - empty , let @xmath306 . by hypothesis , there exists a state @xmath146 in @xmath166 with @xmath301 . this means that @xmath307 is accessible from @xmath146 . moreover , since @xmath146 belongs to an irreducible component @xmath166 , we have that @xmath146 is accessible from @xmath307 as well , which implies that @xmath308 for a certain choice of @xmath309 . in particular , @xmath310 . by the hypothesis of deficiency zero , it follows that @xmath311 , because @xmath312 , defined in , is an isomorphism between the spaces @xmath131 and @xmath125 associated with @xmath189 . therefore , @xmath313 for some integers @xmath314 . since the vectors @xmath315 are linearly independent , @xmath316 for all @xmath317 . hence , each @xmath315 that appears in the sum , must appear at least twice , once with coefficient @xmath318 , once with @xmath319 . consequently , by iteratively reordering the terms @xmath320 , the reactions @xmath321 form a union of directed closed paths in the reaction graph of @xmath54 . in particular , the reaction @xmath322 is contained in a closed directed path of the reaction graph of @xmath189 , and since this is true for every reaction in @xmath176 , @xmath189 is weakly reversible . we conclude the proof by lemma [ lem : subnet ] , since if @xmath54 has deficiency zero , so does every subnetwork of @xmath54 . [ lem : reactions_gamma ] let @xmath54 be a weakly reversible reaction network , and let @xmath166 be an irreducible component . then , for any complex @xmath193 we have @xmath323 one inclusion is trivial , since @xmath324 . for the other inclusion , fix @xmath193 . suppose that there exists @xmath161 with @xmath325 . it follows that any reaction @xmath326 is active on @xmath166 , and therefore is contained in @xmath176 . moreover , since @xmath54 is weakly reversible , for any reaction in @xmath55 of the form @xmath53 , there exists a directed path in the reaction graph of @xmath54 from @xmath327 to @xmath107 . hence , by lemma [ lem : path ] , @xmath221 is accessible from @xmath146 , which implies that @xmath221 is in @xmath166 and that @xmath53 is in @xmath176 , since @xmath328 . therefore , to conclude the proof it suffices to prove that there exists @xmath161 with @xmath325 . if it were no @xmath161 with @xmath325 , then no reaction of the form @xmath329 would be in @xmath176 . since @xmath193 , there exists a reaction of the form @xmath53 . this means that there is @xmath330 , such that @xmath331 . hence , @xmath332 is in @xmath166 with @xmath333 , which concludes the proof . it is proven in @xcite that if a deterministic reaction system @xmath79 is complex balanced , then @xmath54 is weakly reversible . by @xcite , we also know that if @xmath78 is mass - action kinetics , then all positive equilibria are complex balanced , and there exists exactly one positive equilibrium in each stoichiometric compatibility class , which is locally asymptotically stable . therefore , to conclude the proof we only need to prove that in a complex balanced mass - action system @xmath82 , the eventual equilibria on the boundary of @xmath51 are also complex balanced . first of all note that any subsystem @xmath334 of @xmath82 corresponding to a linkage classes @xmath335 of @xmath54 is complex balanced . indeed , the projection of a positive complex balanced equilibrium of @xmath82 onto the space of the species of @xmath335 satisfies for any complex of @xmath336 , hence it is a positive complex balanced equilibrium of @xmath334 . let @xmath174 be an equilibrium point on the boundary . consider a linkage class @xmath335 of @xmath54 , and assume that @xmath337 for any species @xmath125 appearing in the linkage class . then , the projection of @xmath174 onto the species of @xmath335 is a positive equilibrium of @xmath334 , and therefore complex balanced . it follows that @xmath174 satisfies for any complex of @xmath335 . oppositely , assume that there exists a species appearing in the linkage class @xmath335 , such that @xmath338 ( this can only happen on a boundary state ) . remember that by mass - action kinetics , all the rates of reactions whose source complex contains @xmath125 are zero . in particular , all the rates of reactions degrading @xmath125 are zero . consider a complex @xmath107 in @xmath335 that contains @xmath125 . by weakly reversibility , there exists a reaction @xmath329 in @xmath335 . if @xmath327 contains @xmath125 , then @xmath339 . if @xmath327 does not contain @xmath125 , then the reaction @xmath329 produces @xmath125 . since the rate of all reactions degrading @xmath125 is zero at @xmath174 and @xmath174 is an equilibrium , then @xmath340 must be zero as well . by mass - action kinetics , this means that there exists a species @xmath341 such that @xmath141 appears in @xmath327 and @xmath342 . by iteratively applying the same argument with the new species @xmath141 and by weakly reversibility , we obtain that @xmath343 for any reaction @xmath53 in @xmath335 . it follows that @xmath174 satisfies for any complex in @xmath335 , since the equation reduces to @xmath344 . equation is therefore satisfied for any complex of @xmath54 and @xmath174 is a complex balanced equilibrium . this concludes the proof . by ( * theorem 6.1.2 ) , if @xmath144 is an equilibrium point and @xmath62 , then @xmath145 only if @xmath53 is terminal . moreover , if @xmath145 , then @xmath345 for every complex @xmath346 of @xmath347 . now , suppose that @xmath78 is mass - action kinetics with rate constants @xmath143 , and that @xmath145 with @xmath62 ( and therefore @xmath348 ) . consider @xmath349 by the first part of the statement , the reaction graph of the subnetwork @xmath350 is a union of terminal strongly connected components of @xmath54 , and therefore @xmath351 is weakly reversible . moreover , by lemma [ lem : subnet ] , the deficiency of @xmath351 is 0 . it is not hard to see that the canonical projection of @xmath146 onto the space of the species @xmath352 is a positive equilibrium point of @xmath353 , and therefore complex balanced by theorem [ thm : deficiency_zero_iff ] . the proof is concluded by and by noting that , for any complex @xmath354 , @xmath355 if @xmath176 is empty there is nothing to prove . suppose that this is not the case . since @xmath189 has deficiency zero , by lemma [ lem : weakly_reversible ] , it is weakly reversible . for any @xmath356 , by definition there exists @xmath161 such that @xmath90 , which in turn implies @xmath357 . therefore , for any directed path in the reaction graph of @xmath54 that starts with @xmath356 , all the reactions in the path belong to @xmath176 , by definition of @xmath176 . since @xmath189 is weakly reversible , this can only happen if @xmath184 , and this proves the first part of the statement . to conclude the proof , note that if the deficiency of @xmath54 is zero , then by lemma [ lem : subnet ] the deficiency of @xmath189 is zero as well . for the first part of the statement , consider a continuous - time markov chain @xmath358 with state space @xmath359 and transition rate from @xmath360 to @xmath361 given by @xmath87 if @xmath356 , and zero otherwise . the master equation for @xmath358 is @xmath362 with the convention that @xmath363 if @xmath364 . by definition [ defi : complex_balanced_stationary_distribution ] , a stationary distribution for @xmath358 exists and it is of the form @xmath365 , for a suitable normalising constant @xmath366 . since @xmath295 is positive for any @xmath161 ( because it is a stationary distribution on an irreducible component ) , then by standard markov chain theory , we have that for any two states @xmath367 , if @xmath368 is accessible from @xmath369 , then @xmath369 is accessible from @xmath368 . fix @xmath356 and @xmath161 with @xmath90 . then , a directed path from @xmath361 to @xmath360 exists in the graph associated with @xmath358 . the second components of the form @xmath107 of the states in the path , by construction , determine a directed path in the reaction graph of @xmath189 from @xmath327 to @xmath107 . hence , any reaction @xmath356 is contained in a closed directed path , which means that @xmath189 is weakly reversible . assume now that @xmath78 is mass - action kinetics with rate constants @xmath143 and that @xmath174 is a positive complex balanced equilibrium of @xmath82 . then , by theorem [ thm : anderson ] , there exists a ( unique ) stationary distribution on @xmath166 of the form . if a species @xmath370 is not in @xmath371 , then the value of @xmath372 is constant for any @xmath161 , and can be obtained from by modifying the normalising constant . by theorem [ thm : complex_balanced ] and lemma [ lem : reactions_gamma ] , we have that @xmath373 with @xmath374 if @xmath364 . therefore , for any @xmath193 and @xmath161 , @xmath375 which leads to , since @xmath376 is of the form . to prove the converse we first introduce a new stochastic mass - action system @xmath377 , which is given by the reactions of the form @xmath378 where @xmath379 are fictitious species in one to one correspondence with the complexes @xmath199 . the rate constant of the reaction @xmath380 is given by @xmath81 . it is not difficult to see that the sum of the fictitious species is conserved for any possible trajectory . moreover , since any directed path @xmath381 in the reaction graph of @xmath54 corresponds to a directed path @xmath382 in the reaction graph of @xmath383 , we have that @xmath383 is weakly reversible by the first part of the proof . consider the set @xmath384 every state in @xmath385 is of the form @xmath386 , where @xmath161 and @xmath387 is considered as the vector in @xmath388 with entry 1 in the position corresponding to the species @xmath387 and 0 otherwise . since @xmath166 is an irreducible component of @xmath54 and the sum of the fictitious species is conserved , no state outside @xmath385 is accessible from any state in @xmath385 , according to @xmath383 . moreover , the master equation on @xmath385 can be written as @xmath389 if we choose @xmath390 for some positive constant @xmath366 , then the master equation is satisfied due to definition [ defi : complex_balanced_stationary_distribution ] . therefore , if @xmath366 is chosen as a suitable normalising constant , @xmath391 is a stationary distribution on @xmath385 . consider the linear homomorphism @xmath312 as defined in , for the reaction network @xmath383 . let @xmath392 denote the cardinality of a set , and note that @xmath393 . for any vector @xmath315 of the basis of @xmath394 , we have @xmath395 . since the vectors @xmath396 with @xmath317 are linear independent , @xmath312 is an isomorphism and the deficiency of @xmath383 is 0 . since @xmath383 is a deficiency zero weakly reversible reaction network , it follows from theorem [ thm : deficiency_zero_iff ] that the mass - action system @xmath397 is complex balanced . therefore , by theorem [ thm : anderson ] , we have that @xmath398 has the form @xmath399 for a positive complex balanced equilibrium @xmath400 , on any irreducible component @xmath401 contained in @xmath385 . since @xmath390 does not depend on @xmath402 , we have @xmath403 for any @xmath404 . fix a complex @xmath405 . since @xmath189 is weakly reversible , there exists a reaction @xmath329 that is active on @xmath166 . fix @xmath161 such that @xmath325 . then for any @xmath356 we have @xmath220 . if we plug the formula for @xmath406 in for our choice of @xmath146 and @xmath327 , we obtain @xmath407 which leads to @xmath408 the proof is concluded by the fact that the above holds for any fixed @xmath193 , which means that @xmath174 is a positive complex balanced equilibrium of @xmath201 . by lemma [ lem : weakly_reversible ] , @xmath189 is weakly reversible . moreover , for @xmath356 , if @xmath90 then @xmath357 . this implies that for any directed path in the reaction graph of @xmath54 that starts with @xmath356 , all the reactions in the path belong to @xmath176 , by definition of @xmath176 . since @xmath189 is weakly reversible , every directed path in the reaction graph of @xmath54 that starts with @xmath356 is contained in a closed directed path . this implies that @xmath184 , and proves the first part of the statement . now assume that @xmath78 is mass - action kinetics with rate constants @xmath143 . if the deficiency of @xmath54 is zero , then by lemma [ lem : subnet ] the deficiency of the terminal network is zero as well . moreover , @xmath104 is weakly reversible by definition , thus by theorem [ thm : deficiency_zero_iff ] @xmath149 is complex balanced for any choice of rate constants @xmath409 . let @xmath96 be the stochastic process associated with @xmath82 . by the first part of the statement , on @xmath166 only terminal reactions take place and these involve a subset of the species only . without loss of generality , we can assume that @xmath410 is constituted by the first @xmath411 species of @xmath412 . therefore , @xmath166 is of the form @xmath413 , with @xmath414 and @xmath415 . moreover , we have that on @xmath416 , the projection @xmath417 is distributed as the process associated with @xmath149 , for which @xmath416 is an irreducible component . let @xmath174 be a positive complex balanced equilibrium for @xmath149 . hence , by theorem [ thm : anderson ] or corollary [ cor : stochastic_complex_balanced_positive ] , the stationary distribution of the process @xmath418 on @xmath166 is of the form . for the first part , we prove that if an irreducible component @xmath166 is positive , then @xmath54 is weakly reversible . this simply follows from lemma [ lem : weakly_reversible ] : indeed , by the lemma , @xmath189 is weakly reversible and since @xmath166 is positive , @xmath197 . to prove the second part , we have to show that a weakly reversible reaction network is essential , and this is done in @xcite . moreover , a deficiency zero weakly reversible mass - action system is complex balanced , and the proof is concluded by theorem [ thm : anderson ] or corollary [ cor : stochastic_complex_balanced_positive ] . in example [ ex : not_complex_balanced ] , we claim that the stationary distribution on the irreducible component @xmath208 has the form @xmath419 to prove this , it is sufficient to show that @xmath420 satisfies the master equation for every point @xmath421 of @xmath210 . the master equation on @xmath421 is given by @xmath422 by plugging in the formula for @xmath420 and after dividing by @xmath248 and @xmath212 we obtain @xmath423=\frac{1}{x_1!x_2!}[x_1(\theta-1)+x_2(x_2 - 1)].\ ] ] if we multiply by @xmath424 and substitute @xmath425 , it follows that @xmath426 that is @xmath427 which always holds true because the terms cancel each other . in example [ ex : not_complex_balanced_2 ] , we change the notation to @xmath263 . then we claim that the stationary distributions on the irreducible components @xmath266 and @xmath267 are @xmath264 and @xmath265 , respectively , where as before @xmath419 we prove that @xmath264 is the stationary distribution on @xmath266 . the case with @xmath428 is analogue . we prove the result by consider the master equation for @xmath264 on a point @xmath429 , which is as following : @xmath430 as we did for the previous calculations , we plug in the expression for @xmath264 , then divide by @xmath431 , @xmath248 and multiply by @xmath424 . we obtain @xmath432 finally , by substituting @xmath433 with @xmath434 and by performing the calculations , we obtain @xmath344 , which means that the above equation is satisfied . , _ continuous time markov chain models for chemical reaction networks _ , in design and analysis of biomolecular circuits : engineering approaches to systems and synthetic biology , heinz koeppl , douglas densmore , gianluca setti , and mario di bernardo , eds . , springer , 2011 , pp . 342 .

magnetohydrodynamic ( mhd ) simulations have been used to study disk accretion to a rotating magnetized star with an aligned dipole moment . quiescent initial conditions were developed in order to avoid the fast initial evolution seen in earlier studies . a set of simulations was performed for different stellar magnetic moments and rotation rates . simulations have shown that the disk structure is significantly changed inside a radius where magnetic braking is significant . in this region the disk is strongly inhomogeneous . radial accretion of matter slows as it approaches the area of strong magnetic field and a dense ring and funnel flow form at the magnetospheric radius where the magnetic pressure is equal to the total , kinetic plus thermal , pressure of the matter . funnel flows ( ff ) , where the disk matter moves away from the disk plane and flows along the stellar magnetic field , are found to be stable features during many rotations of the disk . the dominant force driving matter into the ff is the pressure gradient force , while gravitational force accelerates it as it approaches the star . the magnetic force is much smaller than the other forces . the funnel flow is found to be strongly sub - alfvnic everywhere . the ff is subsonic close to the disk , but it becomes supersonic well above the disk . matter reaches the star with a velocity close to that of free - fall . angular momentum is transported to the star dominantly by the magnetic field . in the disk the transport of angular momentum is mainly by the matter , but closer to the star the matter transfers its angular momentum to the magnetic field and the magnetic field is dominant in transporting angular momentum to the surface of the star . for slowly rotating stars we observed that magnetic braking leads to the deceleration of the inner regions of the disk and the star spins up . for a rapidly rotating star , the inner regions of the disk rotate with a super - keplerian velocity , and the star spins - down . the average torque is found to be zero when the corotation radius . the evolution of the magnetic field in the corona of the disk depends on the ratio of magnetic to matter energies in the corona and in the disk . most of the simulations were performed in the regime of a relatively dense corona where the matter energy density was larger than the magnetic energy density . in this case the coronal magnetic field gradually opens but the velocity and density of outflowing matter are small . in a test case where a significant part of the corona was in the field dominated regime , more dramatic opening of the magnetic field was observed with the formation of magneto - centrifugally driven outflows . numerical applications of our simulation results are made to t tauri stars . we conclude that our quasi - stationary simulations correspond to the classical t tauri stage of evolution . our results are also relevant to cataclysmic variables and magnetized neutron stars in x - ray binaries . = cmmib10 at 10pt

astro-ph0209426

disk accretion to a rotating magnetized star is important in a number of astrophysical objects , including t tauri stars ( edwards _ et al . _ 1994 ) , cataclysmic variables ( e.g. , warner 1995 ) , and x - ray pulsars ( e.g. , bildsten _ et al . _ 1997 ) . the accretion of matter to a rotating star with a dipole magnetic field is a complicated problem still only partially solved . the important questions which need to be answered include : ( 1 ) what is the structure of the disk near the magnetized star ? ( 2 ) where is the inner radius of the disk ? ( 3 ) what is the nature of the funnel flows ( ff ) ? for example , which force is dominant in lifting matter to the funnel flow ? ( 4 ) how is the accretion rate influenced by the star s magnetic moment @xmath3 and angular velocity @xmath4 ? ( 5 ) what is the mechanism of angular momentum transport between the star and the disk ? what determines whether star spins - up or spins - down ? ( 6 ) what are the necessary conditions for magneto - centrifugally driven outflows from the disk and/or the star ? many of these questions have been investigated analytically , but the conclusions reached by different authors often differ because the simplifying assumptions are different . for example , regarding question ( 2 ) , some authors conclude that the disk should be disrupted in the region where magnetic and matter stresses are comparable ( e.g. , pringle and rees 1972 - hereafter pr72 ; davidson & ostriker 1973 ; lamb , pethick & pines 1973 ; ghosh , lamb , & pethick 1977 ; scharlemann 1978 ; ghosh & lamb 1979 a , b - hereafter gl79a , b ; camenzind 1990 ; knigl 1991 ; shu it et al . hereafter s94 ) . others , argue that the inner radius of the disk should be farther away , at the corotation radius , because the inner regions of the disk are disrupted by magnetic braking ( ostriker & shu 1995 hereafter os95 ; branderburg & campbell 1998 ; elstner & rdiger 2000 ) . question ( 3 ) is investigated in only a few papers which use the bernoulli integral ( lamb , pethick & pines 1977 ; paatz & camenzind 1996 hereafter pc96 ; li & wilson 1999 ; and koldoba _ et al . the authors agree that the flow should become supersonic ( and slow magnetosonic ) just above the disk . on the other hand opinions differ regarding the driving force which pushes matter up into the ff . li and wilson ( 1999 ) ( see also li et al . 1996 ) propose that the twisting of the magnetic field near the base of the ff should be very large , @xmath5 and the magnetic force should be the main one lifting matter to the ff . here , @xmath6 is the toroidal component of the magnetic field and @xmath7 is the poloidal component . other groups ( e.g. , lamb et al . 1977 ; pc96 ; koldoba et al . 2002 ) argue that the ff should be super - alvnic so that the twisting of the field is small so that the magnetic force is also small . in numerical simulations by ms97 , h97 and gbw99 the magnetospheric accretion was reported , but no clear evidence of funnel flows was presented and no analysis of ffs was performed . a significant part of this paper is devoted to the ffs . another important issue which has been discussed over the past @xmath8 years is question ( 5 ) concerning the transport of angular momentum between the disk and the star . what determines the sign of the torque on the star ? in early papers it was supposed that a star can only be spun - up because matter in a keplerian disk brings positive angular momentum to the star ( e.g. , pr72 ) . later , it was recognized that a star can be spun - down due to the part of the star s magnetic flux which passes through the disk outside of the corotation radius ( gl79b , wang 1995 ) . recently , the idea of `` torqueless '' accretion was proposed where mass but not angular momentum is transported to a star ( e.g. s94 ; os95 ; li , wickramasinghe & rdiger 1996 ; li & wickramasinghe 1997 ) . wang ( 1997 ) presented arguments against this idea , but this still remains an open question . question ( 6 ) regarding magneto - centrifugally driven outflows from the disk has been discussed by a number of authors ( e.g. , camenzind 1990 ; knigl 1991 ; s94 ; os95 ; lovelace , romanova & bisnovatyi - kogan 1995 - hereafter lrbk95 ; fendt , camenzind & appl 1995 ; pc96 ; goodson & winglee 1999 ; bardou & heyvaerts 1999 ; agapitou & papaloizou 2000 ) . for example , s94 proposed that poloidal magnetic flux accumulates near the corotation radius and magnetic winds should blow from this point ( x - point ) . lrbk95 proposed that wind may form from the entire region of the disk outside the corotation radius where the magnetic field threading the disk is open . analytical investigations of disk accretion to a magnetized star are of course limited by the different assumptions made . for this reason robust 2d and 3d simulations are essential to further the understanding of the different phenomena . by robust we mean that the result should not depend on initial conditions , boundary conditions , on grid resolution , and other artificial factors . several 2d mhd simulation studies have been made with different initial conditions aimed at disk / star outflows . in an early work hayashi , shibata , and matsumoto ( 1996 ) ( hereafter hsm96 ) investigated the interaction of a non - rotating star with a keplerian accretion disk and observed the opening of magnetic field lines which initially thread both the star and the disk . they found single event outflows and the corresponding inward collapse of the disk on a dynamical time - scale ( less than one period of rotation of the inner radius of the disk ) with the radial velocity of the disk close to free - fall . this fast evolution was the result of the magnetic braking of the disk by the magnetic field linking the disk and non - rotating star through a conducting corona . this explosive behavior may correspond to some episodic accretion events of actual systems . however , it is important to investigate the possible quiescent behavior of the disk - star systems . miller and stone ( 1997 ; hereafter ms97 ) investigated disk - star interaction for different geometries and stellar magnetic fields using the resistive zeus code . ms97 rotated the corona - which occupies all the space between the star and the disk - with the rotation rate of the star . this decreased the initial magnetic braking ( compared to hsm96 ) , so that they were able to perform simulations for several periods of rotation of the inner radius of the disk . in cases with a relatively weak magnetic field , they got results similar to those of hsm96 . they also found the disk collapsing to the star with velocity @xmath9 , the opening of magnetic field lines , and outflows of matter from the disk . however , in the case of a strong magnetic field , particularly in the case which included a uniform homogeneous vertical magnetic field threading the disk , they observed diminished outflows . instead the matter flowed around the magnetosphere to the star . similar results were obtained by hirose et al . ( 1997 ; hereafter h97 ) . goodson , winglee and bhm ( 1997 ) ( hereafter gwb97 ) and goodson , bhm and winglee ( 1999 ) ( hereafter gbw99 ) did much longer simulations in very large simulation regions . they observed quasi - periodic matter outbursts associated with the quasi - periodic opening of magnetic field lines and matter accretion to the star . the density in the corona was chosen to decrease in a special way , @xmath10 , so that the alfvn speed @xmath11 decreases gradually . this is favorable for the opening of magnetic field lines , and for the generation and propagation of outflows . gwb97 and gbw99 do not investigate cases where the density falls off more slowly with distance . fendt and elstner ( 1999 , 2000 hereafter fe99 , fe00 ) investigated disk - star interaction for thousands of rotations of the inner radius of the disk , and observed the opening of magnetic field lines and outflows . however , they treated the disk as a boundary condition so that they could not take into account the back reaction of the disk on the stellar magnetic field . furthermore , the actual outflow of matter from the disk to the corona may be different from that assumed . the above mentioned simulation studies show that outflows appear either in very non - stationary situations ( hsm96 , ms97 ) or for a special distribution of coronal density and very fast rotation of the star ( gwb97 , gbw99 ) . none of the papers give answer to questions ( 1)-(5 ) . also , it is not clear whether or not outflows exist for more quiescent initial conditions , and for cases where the coronal density falls off slowly with distance . in this paper we investigate disk accretion to a rotating magnetized star and the associated funnel flows . we start from initial conditions which give us the possibility to significantly reduce the initial magnetic braking between the disk and the corona . this allows us to investigate the disk - star interaction and funnel flows for long times and to consider questions ( 1)-(5 ) in detail . in 2 we describe the numerical model , including initial and boundary conditions . we also discuss the evolution of the disk without a magnetic field . in 3 we describe in detail the disk - star interaction for the case of slowly rotating stars and in 4 cases of fast rotating stars . in 5 we consider the dependence of disk - star interaction on the magnetic moment @xmath3 . in 6 we analyze the physics of ffs . in 7 we consider the possibility of outflows . in 8 we apply our simulation results to t tauri stars . in 9 give the conclusions from this work . to investigate the disk - star interaction , we numerically solve the mhd equations , @xmath12 @xmath13 here , @xmath14 is entropy , @xmath15 @xmath16 is the stress tensor , @xmath17 is the gravitational acceleration , and @xmath18 is the gravitational potential of the central object . we take the equation of state to be @xmath19 , where @xmath20 for most of this work . we solve equations ( 1)-(4 ) in a spherical coordinate system @xmath21 . in order to model the slow accretion of matter in the disk we modified equations ( 2 ) and ( 4 ) by including a small viscous stress @xmath22 in @xmath23 . because the flow is dominantly in the azimuthal direction the main viscous stress acts in this direction . thus the important viscous stress components are the @xmath24 and @xmath25 terms , @xmath26 , @xmath27 , where @xmath28 is angular velocity of the matter , and @xmath29 is the kinematic viscosity . we adopt the @xmath30 model for the viscosity ( shakura & sunyaev 1973 ) , where @xmath31 , with @xmath32 the isothermal sound speed velocity and @xmath33 is a dimensionless parameter , which was chosen to be @xmath34 or @xmath35 in most of simulations which gave a reasonable speed of matter flow to the central regions . we checked that for the grid sizes used the numerical viscosity is much lower than the considered @xmath36 viscosity ( see 2.5 ) . equations ( 1 ) - ( 4 ) were solved with a godunov type numerical code developed and tested by koldoba _ ( 1992 ) , koldoba and ustyugova ( 1994 ) and ustyugova _ this type of code has also been described by ryu , jones , and frank ( 1995 ) . a modified version of the code incorporating viscosity was tested in a number of hydrodynamic simulations for different values of @xmath36 , from @xmath34 to @xmath37 . for @xmath38 we observed a steady flow of matter to the star with small velocities ( see 2.5 ) corresponding approximately to the shakura - sunyaev accretion model . our godunov type code does not explicitly include magnetic diffusivity ( unlike the codes of ms97 , h97 , & gbw99 ) . however , matter diffuses across the magnetic field owing to the small numerical diffusivity @xmath39 which is of the order of the numerical viscosity @xmath40 , where @xmath41 is the grid spacing . the estimated magnetic reynolds number associated with the radial flow is @xmath42 , where @xmath43 is the characteristic scale . in the region @xmath44 , @xmath45 owing to the small radial velocity @xmath46 . in actual accretion disks , both angular momentum transport and diffusion of the magnetic field , are probably due to turbulence ( e.g. , bisnovatyi - kogan & ruzmaikin 1976 ; balbus & hawley 1991 ) and have similar order of magnitude transport coefficients . in our simulations we observed that both the viscosity and the magnetic diffusivity are small and are of the same order of magnitide . note , that in the @xmath47 direction , the velocity of the flow is high , and the reynolds number is also high , @xmath48 , so that magnetic field lines rotate with the disk in @xmath47 direction , and at the same time matter slowly diffuses in @xmath49direction . this is important for understanding of evolution of magnetic field lines observed in the following sections . a reference value of distance is denoted @xmath50 . the reference value for velocity is taken to be @xmath51 which is the keplerian velocity at @xmath50 . the reference angular rotation rate is @xmath52 , and the corresponding time - scale is @xmath53 . for discussing our results we also use the rotation period at @xmath54 , @xmath55 . for a given magnetic field strength at @xmath50 we can define a reference density as @xmath56 and reference pressure @xmath57 . a reference magnetic moment of the star is then @xmath58 . thus , the calculated variables are : @xmath59 , @xmath60 , @xmath61 , @xmath62 , @xmath63 , @xmath64 . we also use the dimensionless time , @xmath65 . later , we omit the primes but note that any dimensionless variable can readily be converted to its physical value . here , we give parameters for a typical t tauri star . we take the mass and radius of the star to be @xmath66 and @xmath67 . we use as the length scale @xmath50 the initial inner radius of the disk , so that the dimensionless inner radius of the disk is @xmath68 in most of simulations . we place the inner boundary ( a star ) at @xmath69 . then , our reference length is @xmath70 , while the simulation region corresponds to @xmath71 . the reference velocity is @xmath72 , and the corresponding time - scale is @xmath73 . the period of rotation of the inner radius of the disk is @xmath74 . consider a magnetic field strength @xmath75 at the star s surface . then at @xmath76 , the reference magnetic field is @xmath77 and the reference magnetic moment is @xmath78 . the reference density is @xmath79 , or @xmath80 which is typical for t tauri star disks . the reference mass accretion rate is @xmath81 . the reference angular momentum flux is : @xmath82 . we place the inner boundary of the simulation region - the `` star '' - at @xmath69 . the size of external boundary depends on the grid ( see below ) . the spherical grid was inhomogeneous in the @xmath83direction . the inhomogeneity was such that cells at any distance @xmath84 were approximately square with @xmath85 for fixed @xmath86 . this grid gives a high space resolution close to the star which is important in this problem . at the same time simulations in very large regions may be performed . the inhomogeneous grid is a smooth analog of the nested grids used by gbw97 and gbw99 . the angular grid was taken to have @xmath87 in most cases , and @xmath88 in cases with very strong magnetic field ( see 7 ) . the number of points in the @xmath83direction determined the size of the simulation region . the main runs were done with a large radial grid @xmath89 which corresponds to @xmath90 . simulations with a smaller grid @xmath91 corresponding to @xmath92 were done in a number of cases . the large size of the region was chosen in order to minimize the influence of external boundaries on processes in the vicinity of the star . we checked , however , that results of simulations in the larger region almost coincide with those in the smaller region . for the typical grid with @xmath87 , the smallest cell near the star corresponds to @xmath93 which is a resolution about @xmath94 times higher than one in gbw99 . the largest grid is at the external boundary , for @xmath89 and is @xmath95 . these numbers show that a spherical stretched grid is advantageous for simulations of accretion flow to a star with a dipole field . at the inner boundary @xmath96 , we apply free " boundary conditions for the density , pressure , and entropy , @xmath97 , @xmath98 , and @xmath99 . the inner boundary is treated as a rotating perfect conductor @xmath100 , which is analogous to the case of a rotating disk discussed earlier by ustyugova _ ( 1995 , 1999 ) . in the reference frame rotating with the star , the flow velocity @xmath101 is parallel to @xmath102 at @xmath96 ; that is , @xmath103 at @xmath104 . we consider the `` free '' boundary condition for this velocity , @xmath105 . these two conditions determine the direction and value of the velocity vector @xmath101 , that is all three components @xmath106 . the boundary condition at @xmath104 on the magnetic field has @xmath107 , while the poloidal components @xmath108 and @xmath109 are derived from the magnetic flux function @xmath110 , where @xmath111 at the boundary is derived from the frozen - in condition @xmath112 . at the outer boundary @xmath113 , we took fixed boundary conditions for all variables for those cases when the simulation region was very large : @xmath114 . in case of smaller region @xmath115 we took free boundary conditions on all the hydrodynamic variables and the @xmath6 component of the magnetic field , if matter _ outflows _ from the region . for the magnetic field we took @xmath116 and @xmath117 if matter inflows to the region ( @xmath118 ) , then we set this velocity equal to zero . results are very similar at these boundary conditions . we assume reflection symmetry about the equatorial plane . here we present our new method for establishing quasi - equilibrium initial conditions . the star has a magnetic dipole field with the magnetic moment @xmath119 aligned with the rotation axis @xmath120 , with @xmath4 the star s rotation rate . the magnetic field is @xmath121 with components in spherical coordinates @xmath122 and @xmath123 . as initial conditions we set up a low - temperature @xmath124 disk with a high - temperature @xmath125 corona filling the remainder of the simulation region . the disk rotates with angular velocity close to the keplerian value @xmath126 . the disk extends inward to the radius @xmath127 where the ram pressure of the disk is equals to the magnetic pressure of the dipole , @xmath128 . in order to have approximately equilibrium initial conditions it is necessary that the corona initially be rotating . if the corona initially does not rotate ( e.g. , hsm96 ) , or rotates with the angular velocity of the star ( e.g. ms97 ) , while the disk rotates with keplerian velocity , then the coronal magnetic field lines threading the disk and the star lead to magnetic braking of the disk and subsequent fast accretion of the disk with a velocity close to free - fall . furthermore , a strong discontinuity develops in the angular velocity between the disk and the corona ( except at the corotation radius ) . this leads to the generation of a strong @xmath6 component , while the dragging of the magnetic field with the radial inflow of the disk leads to the generation of a @xmath129 component . this twisting of the field is responsible for transient magneto - centrifugally driven outflows ( e.g hsm96 ; ms97 ; hirose et al . such fast evolution may correspond to periods of violent accretion in some strongly non - stationary systems . however , the star - disk systems may exist in more quiescent configurations which can be studied by using appropriate initial conditions . in order to avoid the strong discontinuity of @xmath130 and @xmath6 on the boundary between the disk and the corona , we rotate the initial corona so that its angular velocity is a constant on a given cylindrical radius @xmath131 and equal to the keplerian rotation rate of the disk . thus the corona rotates with different angular velocities on cylinders with different radii @xmath132 . for such initial conditions we observe much slower accretion in the disk with no dramatic outflows . this distribution of @xmath130 in the corona also leads to the twisting of the magnetic field lines . any coronal magnetic field line will be twisted along its length by the differential rotation of nearby layers of corona , but this twist is distributed along the length of the field line . at the boundary between the disk and corona the twist is small . it is much smaller than the twist at the boundary between a rotating disk and non - rotating corona . differentially rotating corona also leads to the magnetic braking of the disk , but it is more gradual and does not lead to a rapid infall of the matter in the disk . earlier , a differentially rotating corona was used as an initial condition in a study of the opening of magnetic loops threading a keplerian disk , but the disk was considered as a boundary condition ( romanova _ et al . _ 1998 ) . initially , we assume that @xmath133 everywhere . also , we assume that the initial poloidal field is that of a dipole and that the plasma is force - free . thus , we search for equilibrium initial conditions for the disk - corona system taking into account gravity , pressure , and rotation . to find the initial equilibrium , we suppose that initially ( @xmath134 ) the matter is barotropic @xmath135 . in this case one have ( 1 ) @xmath136 where @xmath137 ( 2 ) the angular velocity @xmath130 depends only on @xmath131 ; ( 3 ) the centrifugal force has the potential @xmath138 ^ 2r'dr'$ ] ; ( 4 ) the stationary euler equation @xmath139 has the integral : @xmath140 where @xmath18 is the gravitational potential . we suppose that the initial angular velocity of the matter in the equatorial plane is @xmath141 where @xmath142 is the cylindrical radius , and @xmath143 is included to take into account that the disk is slightly non - keplerian ( in our simulations @xmath144 ) . thus , the centrifugal potential in the whole region is @xmath145~ , & r \leq r_d \\ k gm / r~ , & r \geq r_d ~. \end{array } \right . \eqno(7)\ ] ] at the boundary between the disk and corona the initial pressure is taken to be @xmath146 . the dependence of the density on pressure in the corona and in the disk can be expressed as @xmath147 where @xmath148 is the mean particle mass approximately equal to @xmath149 . we consider that @xmath150 . in the corona , at low values of pressure and density @xmath151 the plasma is hot @xmath152 , whereas in the disk , at high pressure and density ( @xmath153 ) the plasma is at a much lower temperature , @xmath154 . initially , the boundary between the disk and corona corresponds to the contact discontinuity , where @xmath155 , @xmath156 in the corona , and @xmath157 in the disk . in this case @xmath158 the initial disk - corona boundary in the equatorial plane is assumed to be at @xmath159 . we assume values of @xmath160 @xmath161 and @xmath124 . at the point @xmath162 , @xmath155 , @xmath163 . thus , we find @xmath164 . then , we can calculate potential @xmath165 at any point @xmath166 and find @xmath167 . now , we can derive pressure @xmath168 at any point of the region , @xmath169 and find the distribution of density taking into account eq . we considered two main types of initial conditions which are described below . in the first type of initial conditions , we place the inner radius of the disk at @xmath170 . we determine the characteristic density and temperature in the corona to be @xmath171 , @xmath172 , and density in the disk @xmath173 . then we determine the pressure at the boundary between the disk and corona , @xmath174 and then derive temperature in the disk , @xmath175 , so that the pressure is continuous across the boundary . then we determine the magnetic field of the star so as to have the magnetic pressure equal to the stagnation pressure of the disk matter at @xmath176 . this gives @xmath177 which is solved for @xmath3 . as matter in the disk flows inward , it forms a funnel flow at @xmath178 . we determine @xmath179 from equation ( 6 ) taking into account that @xmath68 . for @xmath180 matter rotates with an angular velocity corresponding to @xmath181 , while at @xmath182 the rotation is keplerian . the surfaces of constant @xmath130 are cylinders . for such a distribution , there are no strong gradients of @xmath130 in the region . then we derive the pressure and density distributions using equations ( 5)(10 ) . the left panel of figure 1 shows the density distribution in the disk and corona . for this example we took a relatively small region @xmath183 . one can see that the density is high in the disk , and is @xmath184 times smaller in the corona . for these initial conditions the ff starts close to the initial inner radius of the disk . thus the disk provides a reservoir " of matter for the ff . on the other hand there is the following disadvantage . in most cases , we rotate the star with an angular velocity @xmath4 smaller than the initial angular velocity of matter above the star . fortunately , the star is observed to control the rotation of the nearby regions of corona in a few time - steps . we also used initial conditions of type ii , as described below . in a second type of initial conditions we place the inner radius of the disk @xmath127 at the corotation radius , @xmath185 . thus for @xmath180 both the star and the corona rotate with the angular velocity @xmath186 , while at larger distances the rotation is keplerian . for this initial condition , @xmath130 varies smoothly including the variation from the star and to the corona above it . an important aspect of the type ii initial conditions is that the initial inner radius of the disk is far from the magnetosphere . this allows us to check whether or not matter moves inward from the corotation radius . the left panel of figure 1 shows the initial density distribution for type ii initial conditions for @xmath187 . first , we tested our initial conditions with no magnetic field . we observed that for both types of initial conditions the disk exists for more than @xmath188 rotational periods @xmath189 . no high velocities appear in the disk or corona . typical velocities in the disk are @xmath190 . however , our typical time of simulation of magnetic cases is shorter : @xmath191 , so that we show the result of the evolution of the disk without a magnetic field at @xmath192 . we considered three cases , with no viscosity , @xmath193 , and with small viscosities @xmath34 and @xmath35 . figure 2 ( a , b ) shows that at @xmath194 and boundary conditions of type i , the disk s inner edge stays at @xmath195 during the whole period of evolution . the density in the rest of the disk slowly redistributes itself in such way that the external layers move gradually outward . the outer regions of the disk had excess angular momentum compared to a stationary @xmath36-disk ( shakura & sunyaev 1973 ) , and consequently some matter moved outward carrying away extra angular momentum . when we included a small viscosity @xmath34 ( see panel @xmath196 of figure 2 ) , the outer regions of the disk also moved outward , but inside the shown region @xmath197 matter moves inward with velocity @xmath198 for @xmath199 , and with larger velocity @xmath200 for @xmath201 . for even higher viscosity @xmath35 ( panel @xmath202 of figure 2 ) , the velocity of the inflow is @xmath203 in the entire disk , @xmath197 . in both initial conditions we find that the @xmath30viscosity gives a @xmath204 and thus a good starting point for our mhd simulations . the typical accretion rate is @xmath205 for @xmath34 and @xmath206 for @xmath35 . for the type ii initial conditions we observed that for @xmath193 the inner radius of the disk stays at @xmath207 for @xmath208 . this shows that the numerical viscosity is very small ( see panels @xmath209 and @xmath210 of figure 2 ) compared with an @xmath34 viscosity . external regions moved outward with velocity @xmath211 , redistributing density and angular momentum . for @xmath34 and @xmath212 , the disk behaved similarly to one for initial conditions of type i : the equatorial regions accreted slowly inward with velocity @xmath213 ( panels @xmath214 and @xmath215 of figure 2 ) . the accretion rate is @xmath216 for @xmath34 and @xmath205 at @xmath35 . these tests indicate that our hydrodynamic disk and corona are appropriate for subsequent mhd simulations of the disk - star interaction . note , that subsequently we observed that accretion rate to the star in the magnetic case is larger than in hydro case ( see figures 9 and 10 , where accretion rate is @xmath217 at initial conditions of type i , and @xmath218 at initial conditions of type ii , which is connected with small magnetic braking of matter in the disk , because initial conditions give only approximate but not exact initial equilibrium . here , we discuss simulations of accretion to a slowly rotating star with @xmath219 ( @xmath187 ) , which corresponds to a period @xmath220 for a typical t tauri star ( see 2.2 ) . the magnetic moment is @xmath221 , which corresponds to a magnetic field at the surface of the star @xmath222 . we used initial conditions of types i and ii to check for the possible dependence of the evolution on initial conditions . the simulations were performed with the grid @xmath223 which has @xmath114 . first , we describe results of simulations with type i initial conditions . we observed that matter started to move slowly both in the disk and corona with velocities much smaller than free - fall . no dramatic collapse of the disk or fast outflows were observed owing to our quasi - equilibrium initial conditions . figure 3 shows the entire simulation region at different times . the outer part of the disk changes very little during the simulation time @xmath224 . the magnetic field lines twist and become partially open . viewed on this large scale , the disk is relatively flared , but only the much smaller region of the disk @xmath225 is important for disk - star interactions . figure 4 shows the evolution in the smaller region @xmath226 . one can see , that the inner region of the disk is influenced by the magnetic field of the star . for @xmath227 the accretion rate due to the @xmath30 viscosity increased and magnetic flux was accumulated in the central region . we stopped the simulations at this point , because later the accretion rate increased , the inner radius of the disk moved closer to the star , and the case became less interesting for the investigation of the disk star interaction . figure 5 shows the evolution of the density and poloidal magnetic field in the region @xmath228 . the interaction of the disk with the magnetosphere of the star led to the reconstruction of the disk inside the radius which we call the braking " radius @xmath0 . in this case this radius is an approximate analog of the radius of the outer transition zone in the theory of gl79b , which divides areas of an undisturbed viscous disk and inner regions , influenced by a magnetic field . in our case , the disk for @xmath230 is also disturbed by the star s field , but matter still moves to this region due to viscosity . at smaller radii @xmath231 , matter is magnetically braked . by coincidence @xmath232 . the inner regions of the disk become magnetically braked after several rotation periods @xmath189 . in the region of the magnetic braking the density in the disk is @xmath233 times smaller than in the disk without a magnetic field . the inflow speed of the matter increases to @xmath234 . and matter constantly moves in the direction of the star . the flow towards the star stops when the matter stress becomes comparable to the magnetic pressure of the dipole , @xmath235 note that the main term in the matter stress , @xmath236 , is due to the disk s rotation . we introduce the plasma parameter , @xmath237 , which indicates the relative importance of the matter and magnetic stresses . also , we introduce the magnetospheric radius , @xmath238^{1/6}}~ , \eqno(11)\ ] ] which is the radius at @xmath239 where @xmath240 . figure 5 shows that matter accumulates in the region @xmath241 forming a dense ring . the density in the ring is @xmath242 times larger than the initial density at the inner radius of the disk . at @xmath243 , when more matter comes from the disk , the density in the ring becomes @xmath244 times larger . accumulation of matter near @xmath1 and formation of a ring in the inner regions of the disk was also reported by gwb97 , gbw99 , and ms97 , but these simulations were for non - stationary conditions . for @xmath245 matter moves upward , away from the equatorial plane and goes into the funnel flow . the ff is formed after about one period of rotation ( @xmath246 ) and is a stable feature during the length of the run , @xmath208 . figure 5 shows the @xmath240 line ( black line ) . the base of the ff coincides with @xmath247 ; that is ff starts in the area where the magnetic stress is equal to the matter stress . this is observed in many other cases with different parameters ( see 4 & 5 ) . this is in accord with theories developed in 1970 s ( pr72 , gl79a , b ) . we investigate the physics of ffs in 6 . matter from the outer region of the disk @xmath230 tends to accrete inward . often accretion rate is enhanced in the region where magnetic field has larger radial component @xmath129 and hence stronger magnetic braking . the accretion rate of matter from the region of the disk outside the ring varies with time . consequently , the density in the inner ring and in the funnel flows also varies with time ( figure 5 ) . when more matter accumulates in the ring , then the base of the ff is closer to the star , and the magnetospheric radius is smaller . for example , at @xmath248 ( panel b of figure 5 ) , @xmath249 ; at @xmath250 ( panel c ) @xmath251 . subsequently , for @xmath252 , even more matter came from the disk owing to viscosity , and the ff moved even closer to the star . a stationary state for @xmath253 may exist but at smaller values of @xmath36 , @xmath254 . for type ii initial conditions , the inner radius of the disk is taken to be the corotation radius , @xmath187 . the viscosity parameter is @xmath35 , the same as in the above - mentioned simulations . we were able of observe the evolution for more than @xmath255 rotations of the inner radius of the disk . figure 6 shows the evolution of the inner regions of the disk . we observed that the funnel flow did _ not _ form at the corotation radius as predicted in some theories ( e.g. , os95 ) . instead , the disk slowly moves inward towards the star owing to the viscosity ( see figure 6 ) . when the disk reached conditions where @xmath256 , the accreting matter goes up into the funnel flow . the ff formed after @xmath257 rotations . at longer times , the structure of the disk and the ff are similar to those found for type i initial conditions ( figure 5 ) . but one difference is that for type ii initial conditions the inward motion of the inner edge of the disk drags the bases of magnetic loops inward , so that loops of the dipole are inclined to the disk . such a situation is possible when the viscosity is larger than the diffusivity and magnetic flux in the disk accumulates closer to the star ( e.g. , s94 , os95 ) , but the rest of the loop in the corona does nt move so fast , so that the loops are inclined to the disk , and the @xmath129 component is significant . when the magnetic field is inclined away from the axis of the disk , the magnetic force is much stronger than in case with no inclination ( lovelace , berk & contopoulos 1991 , oyed & pudritz 1997 ; fe99 , fe00 ) . we observed opening of the magnetic field lines close to the magnetosphere . the opening was followed by reconnection , and subsequent opening . the process was quasi - periodic with a quasi - period @xmath258 . we also observed quasi - periodic oscillations of the inner radius of the disk between radii : @xmath259 . the nature of the oscillations is the following : ( 1 ) first , the loop is stretched and the magnetic field blocks " the path of matter to the ff . ( 2 ) next , reconnection releases the stress , and matter accretes through the ff . ( 3 ) the new loop starts to stretch , and the process repeats . figure 10 shows the quasi - periodicity of the accretion rate . these oscillations resemble those observed by gwb97 and gbw99 , but on a much smaller scale . h97 and ms97 also observed enhanced accretion after the reconnection of elongated magnetic field lines near the magnetosphere . this work supports the idea of aly & kuijpers ( 1990 ) who proposed this mechanism of quasi - periodic oscillations . we tested the dependence of the results on the viscosity for both types of initial conditions for @xmath34 and @xmath193 . we obtained similar results , but the ff formed at slightly larger distances @xmath132 as a result of smaller accretion rate . simulations with initial conditions of types i and ii have shown examples of possible situations which can be realized in nature . in the remainder of the paper we use the type i initial conditions , because the ff starts earlier in this case . how is angular momentum transported between the disk and the star ? does the star spin - up or spin - down ? in this section we analyze these questions using the run described in @xmath260 . we observed that the magnetic field lines which are responsible for the partial destruction of the disk at radii @xmath261 are also involved in the transport of angular momentum between the star and the disk . the magnetic filed lines connecting the star and the disk decrease the disk s angular velocity in a significant part of the braking region . figure 7 shows the radial distribution of the angular velocities at different times . one can see that the angular velocity of the disk is significantly smaller than the keplerian value in a broad region @xmath262 . we introduce the radius @xmath263 , such that for @xmath264 the magnetic force on the disk is strong enough to cause a significant deviation from keplerian rotation . the radius @xmath263 depends strongly on the accretion rate and varies between @xmath265 at @xmath250 , ( when the inner disk is of low density ) and @xmath266 at @xmath248 ( when the inner disk is dense ) . figure 8 shows the spatial distribution of the angular velocity , @xmath267 . one can see that for @xmath268 the regions of constant angular velocity @xmath269 almost coincide with the regions of constant magnetic flux @xmath111 . thus , the magnetic field line crossing the disk at radius @xmath132 rotates in the corona with angular velocity @xmath179 . this is because most of the corona is matter - dominated , @xmath270 ( see black line in figure 5 ) , and the disk determines the rotation of the magnetic field lines . magnetic field lines which cross the disk in the magnetically - dominated area ( in the ff ) rotate with the angular velocity of the star . an important question is the angular momentum flux to the star . to answer this question , we calculated the flux of angular momentum ( about the @xmath271axis ) carried by the matter @xmath272 and that carried by the ` twist ' of the magnetic field @xmath273 , @xmath274 evaluated at the surface of the star . in addition , we calculated accretion rate to the star , @xmath275 . figure 9 shows the different fluxes as a function of time . most of the angular momentum flux at the star is carried by the magnetic field ; matter carries only @xmath276 of the total flux . the flux @xmath277 is positive so that it acts to _ spin - up _ the star . the flux @xmath277 correlates with @xmath278 because incoming matter from distances @xmath279 is responsible for bringing in positive angular momentum . at distances @xmath280 the angular momentum flux is carried mainly by the matter . these results partially confirm the hypothesis of s94 , os95 , li et al . ( 1996 ) that matter should carry little angular momentum as it approaches the star . however , our results indicate that this does not lead to the torqueless " accretion predicted by these authors , because the angular momentum flux carried by the matter is transferred to a flux carried by the magnetic field with decreasing distance from the star . the small twist of the magnetic field near the star , @xmath281 ( see 6 ) , is sufficient to carry the observed angular momentum transport flux ( see discussion by wang 1997 ) . from our simulations we find that the magnetic field lines responsible for spin - up or spin - down of the star pass through the disk at distances @xmath282 . consequently the spin evolution of the star depends on the location of the corotation radius @xmath283 relative to @xmath263 . a star may spin - up , spin - down , or be in the regime of the torqueless " accretion ( see 4.2 ) . for the case considered in this subsection of a slowly rotating star @xmath187 , the star spins - up . we performed a set of simulations for different angular velocities of the star @xmath4 . simulations were done for type i initial conditions and with viscosity @xmath34 . the grid was @xmath284 corresponding to the smaller region @xmath285 . the fastest rotation considered was such that at @xmath286 , @xmath287 which corresponds to a rotation period @xmath288 for the t tauri parameters of 2.2 . ( the conversion formulae for dimensional period is @xmath289 ) . in this section we discuss the dependence of the spin - up / spin - down rate of the star on @xmath4 . here we discuss the case of a rapidly rotating star , @xmath287 , @xmath286 ( @xmath290 ) , where @xmath291 . we observed that initially the accretion disk moved outward because the magnetic field of the star transferred its angular momentum to the disk . later , for @xmath292 the disk started to move inward towards the star , and finally a funnel flow formed ( see figure 11 ) . in this case the thickness of the inner part of the disk is larger than that in the case of slowly rotating star , and matter accretes preferentially along the top layers of the disk . figure 12 shows the observed fluxes of matter and angular momenta to the star . note , that the flux of an angular momentum is carried mainly by the magnetic field , @xmath277 , which is negative so that the star spins down . thus , we observe that as matter _ accretes _ to the star angular momentum flows out from the star . as in the case of slowly rotating star , we observe a correlation between the matter and angular momentum fluxes . we arranged a set of simulations to test if a particular value of @xmath4 gives `` torqueless '' accretion . according to gl79b , at some angular velocity of the star @xmath293 , positive magnetic torque associated with the region @xmath294 cancels the negative magnetic torque associated with the region @xmath295 and the star can accrete without changing its angular momentum . we performed simulations at different @xmath4 ( corresponding to @xmath296 and @xmath297 ) and observed , that for @xmath298 ( @xmath299 ) the angular momentum flux @xmath300 wanders back and forth around zero ( see figure 13 ) . that is , it is positive for some time ( @xmath301 ) , then becomes negative for a similar length of time , and then become positive again . we conclude that this @xmath4 corresponds to torqueless accretion . the explanation for torqueless accretion was proposed in the @xmath302s ( e.g. , gl79b ) . namely , when more matter is supplied to the disk , the magnetospheric radius @xmath1 moves closer to the star , and as a result there is a positive angular momentum flux to the star . when less matter accretes , the magnetospheric radius @xmath1 moves outward , and the flux of angular momentum is negative . figure 14 shows the evolution of the density and poloidal magnetic field for times corresponding to positive angular momentum flux to the star ( left panels ) , zero angular momentum flux ( middle panels ) and negative angular momentum flux ( right panels ) . the variations of the accretion rate may be connected with accumulation of matter at @xmath303 where in this case @xmath304 and subsequent accretion through instabilities . the ratio of the corotation radius @xmath283 to the magnetospheric radius @xmath1 is of general interest . at the moments when the torque is zero , @xmath305 ( see figure 13 ) , the magnetospheric radius had the following values : at @xmath306 , @xmath307 ; at @xmath308 , @xmath309 ; at @xmath310 , @xmath311 . that is , for @xmath312 the ratio is : @xmath313 . for similar simulations for faster rotation of the star @xmath314 ( @xmath315 ) , the angular momentum flux also wandered around zero value ( see figure 15 ) , but most of time it was negative so that the star spins down . for simulations with more slowly rotating stars or larger corotation radii , @xmath316 , ( @xmath317 ) , we observed that the star spins up . we did simulation runs for different magnetic moments , @xmath318 and @xmath319 , with other parameters fixed , @xmath219 ( @xmath187 ) , and @xmath34 . at @xmath320 the inner radius at of the disk was at @xmath195 in all cases . the grid was similar to the one used in @xmath321 . figure 16 shows result after @xmath248 rotations . one can see that at smaller values of @xmath3 , the inner radius of the disk and ff are settled closer to the star than in the case of larger @xmath3 . in all cases , we observed that the base of the ff approximately coincides with the magnetosphere radius @xmath1 , but at smaller @xmath3 this radius is smaller in accord with equation ( 11 ) . in case of even smaller @xmath3 the direct accretion to the surface of the star is observed , like in simulations by hsm96 and in some simulations by ms97 . the cases with stronger magnetic moment @xmath322 are more appropriate for explanation of t tauri stars , because the gap between the star and the disk is comparable or larger than the radius of the star . at smaller magnetic momenta @xmath323 , the gap is too small . here , we discuss the physics of a typical funnel flow , the case of a slowly rotating star with @xmath187 ( 3 ) after @xmath250 rotations . the specific heat ratio is @xmath324 . figure 17 shows an enlarged picture of the ff . we determined the variation of different parameters as a function of distance ( @xmath325 ) from the surface of the star along a fiducial poloidal field line through the middle of the ff . this field line is shown by the thick black line in figure 17 . figures 18 21 show the variation of different parameters along this field line . figure 18 shows that the density @xmath326 and pressure @xmath327 decrease along the ff , then increase again near the surface of the star . the temperature @xmath328 initially varies slowly , but increases strongly approaching the star as a result of gas compression . figure 19 shows the variation of different velocities along the ff . matter flows along the ff with poloidal velocity @xmath329 . the velocity perpendicular to the poloidal magnetic field , @xmath330 , is very small , @xmath331 . the poloidal velocity @xmath329 increases and becomes larger than the slow magnetosonic velocity @xmath332 at @xmath333 of the way . this point is expected to be closer to the disk in the case of cooler ( and thinner ) accretion disks ( koldoba et al . the flow become supersonic with the mach number reaching @xmath334 at the surface of the star . near the star , the poloidal velocity is close to the free - fall velocity : @xmath335 . the poloidal velocity of the ff is much smaller than the alfvn velocity @xmath336 . the ratio decreases from @xmath337 in the middle of the ff , to @xmath338 at the star . thus , the flow is strongly sub - alfvnic . this validates the analysis of koldoba _ et al . _ ( 2002 ) which proposed that the funnel flows were sub - alfvnic . it is important to understand the forces which drive matter out of the equatorial plane of the disk into the funnel flow , and which forces accelerate the flow towards the star . the force per unit mass @xmath339 along the magnetic field line is obtained by multiplying the euler equation by @xmath340 . this gives @xmath341 ( ustyugova _ et al . _ 1999 ) , where @xmath342 is the angle of the field line to the @xmath271axis . the terms on the right - hand - side represent the centrifugal force , the pressure gradient force , the gravitational force , and the magnetic force . figure 20 shows forces in the outer two - thirds of the poloidal path of the ff . we see that the pressure gradient force is responsible for driving matter up and out of the disk and into the funnel flow , while the gravitational force is responsible for the acceleration of matter in the rest of the ff . note that the magnetic force in equation ( 12 ) is small and acts in the opposite direction . thus , we did not observe the `` magnetic levitation '' force predicted by li & wilson ( 1999 ) . we observed a finite @xmath6 field above the disk in the ff ( see figure 21 ) , but the magnetic force is much smaller than the other forces and it is not sufficient to levitate matter above the disk . figure 21 shows that the angular velocity of the funnel flow decreases gradually from its value at the inner edge of the disk ( @xmath159 ) to a small negative value at the star . the negative value of @xmath130 at the star is due to the fact that the magnetic field has a small inclination in the direction of rotation of the star . for this reason matter moving along the magnetic field line comes to the star with an angular velocity less than @xmath4 ( see also ghosh et al . 1977 ; muzerolle et al . 2001 ) . figure 21 shows the variation of the toroidal magnetic field along the funnel flow . the twisting of the dipolar field is relatively small with the `` twist '' @xmath343 . at the star , this is in accord with estimations by wang ( 1997 ) , who has shown that the twist near the star should be small , but not a zero . away from the funnel flow @xmath345 is of the order of @xmath7 . we never observed conditions where @xmath346 ( discussed , e.g. , by gl79a and agapitou & papaloizou 2000 ) . the reason is that for @xmath347 , a torsional alfvn wave forms and propagates outward into the corona where it dissipates . in reality , matter may accrete through the magnetosphere owing to @xmath348 instabilities , for example , the rayleigh - taylor or kelvin - helmholtz instability ( arons & lea 1976a , b , spruit & taam 1990 ) . realistic investigations of the role of these instabilities is possible with @xmath348 mhd simulations . in our simulations we observed coronal magnetic field lines at large distances from the star opening due to the initial differential rotation . closer to the star the field lines were mostly closed , contrary to the predictions by aly ( 1986 ) , lrbk95 ( see also newman , newman , & lovelace 1992 ; livio & pringle 1992 ; lynden - bell & boily 1994 ; miki & linker 1994 , amari et al . 1996 ; fe99 ; fe00 ; ustyugova et al 2000 ; uzdenskyi , knigl & litwin 2002 ) . the difference in behavior can be explained as follows . the analytical or semi - analytical models suppose that the inertia of the of the matter is very small , @xmath349 , so that the plasma is in the coronal " limit of gold and hoyle ( 1960 ) . however , the inertia of the matter may be important . romanova et al . ( 1998 ) simulated the dynamics of magnetic loops in the corona of a differentially rotating disk where @xmath350 . they observed that in the magnetically dominated regions ( @xmath351 ) the magnetic loops opened whereas in the matter - dominated regions ( @xmath352 ) the loops did not open . the energy - density of the dipole field decreases rapidly with distance ( @xmath353 ) compared with the matter energy - density in our simulations which varies slowly within the computational region . for this reason the magnetically - dominated region is restricted to a relatively small volume , @xmath354 , around the star ( see black line at figure 5 ) . to overcome this feature of the dipole field , gwb97 and gbw99 adopted a rapid fall - off of density in the corona , @xmath355 . for this dependence the region of small @xmath356 is greatly expanded and opening of the magnetic field is favored . in order to investigate possible stronger outflows , we chose parameters so as to increase the region of magnetically - dominated corona . the magnetic moment was increased @xmath357 times , @xmath358 ( which corresponds to a magnetic field @xmath359 at the surface of the star ) . the density in the corona was decreased @xmath360 times , @xmath361 . this led to a higher alfvn speed near the star , so that we increased the inner radius @xmath362 to cover the region of high alfvn velocity , and increased the resolution of the grid to @xmath363 . we have @xmath240 at @xmath170 as in the other runs described in this work . the magnetic field is stronger and the initial density in the disk is about @xmath297 times higher than in other simulations . figure 22 shows the results of the simulations . we observed that magnetic braking led to accretion of the inner part of the disk with velocity @xmath364 . thus matter in the disk concentrated in a dense ring with @xmath365 . the disk angular velocity is significantly smaller than the keplerian value but larger than the angular velocity of the star , @xmath219 , which is a result of the interaction with the strong stellar magnetic field . a significant region above the corona corotates with the disk . analysis of forces , similar to that of 6 shows that in the region @xmath366 , the centrifugal force drives matter along the magnetic field lines , but the force is not strong enough to open the magnetic field lines and matter accretes to the star forming multiple funnel flows ( see figure 22 ) . however , the outer magnetic field lines , which are located at the edge of the magnetically - dominated region , become gradually open owing to combination of centrifugal and magnetic forces . matter accelerates along the region of open field lines up to velocities @xmath367 , where @xmath368 is the escape velocity . we anticipate that if the density in the corona decreases faster than in our model , then matter will outflow from the region . the observed matter and angular momentum fluxes of outflowing matter are small . this test simulation provides evidence of how outflows can be formed from the disk in the quasi - stationary regime . subsequent research is needed in this direction at different distributions of density of matter in the corona . our simulations are well suited for modelling funnel flows to t tauri stars because the entire flow can be well resolved . in the observed systems , the ratio of the inner radius of the disk where the ff begins to the stellar radius is @xmath369 . such conditions are realized in our simulations . the detailed nature of the magnetic fields of t tauri stars is not known , and the fields may differ significantly from a dipole field ( e.g. , safier 1998 ) . however , recent spectral studies of t tauri stars show broad emission lines which give evidence of matter inflow with nearly free - fall velocities ( e.g. , hartmann et al . 1994 ; lamzin , et al . 1996 ; muzerolle et al . 2001 ) . such velocities may be easily explained in the model where matter inflows supersonically in a funnel flow along a dipolar field . in the following we perform numerical estimates for the funnel flow to a typical t tauri star using conversion scheme of 2.1 . we assume the inner radius of the disk is @xmath370 , the star s mass @xmath371 , its radius @xmath372 , and its surface magnetic field @xmath373 . using these values we can derive reference values for the different physical variables . for the flow velocity , @xmath374 . for the period of rotation at @xmath54 , @xmath375 . for the density , @xmath376 . for the accretion rate , @xmath377 . for the temperature of the corona , @xmath378 . in the following we calculate the different physical variables along the middle of the funnel flow shown in figure 17 . from figure 19 the matter reaches a maximum poloidal velocity @xmath379 at the star s surface , which is @xmath380 of the free - fall velocity @xmath381 . the funnel flow will of course give a distribution of velocities along the line of sight to a distant observer . the smallest line of sight velocities are for matter immediately above the disk where the speeds are @xmath382 of @xmath383 . the line of sight velocity distribution is important for the interpretation of observed spectral lines ( e.g. , folha and emerson 2001 ) . the poloidal velocity is @xmath384 the alfvn velocity @xmath336 above the disk ( at @xmath385 ) and decreases to @xmath386 at the star s surface . the velocity across the magnetic field lines @xmath330 is very small ( see figure 19 ) . thus , matter flows along the magnetic field . the flow become supersonic above the disk and the mach number reaches @xmath387 at the surface of the disk . the initial density of matter in the disk at @xmath50 is @xmath388 , and the density of matter in the ff ( see figure 18 ) varies in the range @xmath389 which is typical for t tauri stars . our typical accretion rate to the star through the ff is @xmath390 , which agrees with estimates for t tauri stars ( hartmann _ et al . the initial temperature in the disk is @xmath391 which is several times higher than expected in the disks of t tauri stars at @xmath50 . as mentioned earlier this higher temperature is needed to give adequate resolution in the vertical direction through the disk . we observed that matter in the ff was heated by adiabatic compression to higher temperatures , @xmath392 . the fact that we use `` free '' boundary conditions at the surface of the star means that we can not account for the stand - off shock at the stellar surface , surface heating , and other physical phenomena close to the star . study of these processes can be done with a separate set of simulations . the investigated quasi - stationary funnel flows correspond to classical t tauri stars ( ctts ) ( see hartman 1998 ) where variability is not strong and possible outflows are weak . previous simulations of very fast accretion with subsequent strong outflows were done by hsm96 , ms97 , h97 , gwb97 , and gbw99 , and these may correspond to earlier stages of evolution of young stellar objects ( ysos of classes 0 and i ) which show evidence of episodic non - stationary accretion ( e.g. , fu orionis type stars ) and strong outflows ( e.g. hh30 ) . non - stationary accretion may be initiated by thermal instability ( lightman & eardley 1974 ) or by global magnetic instability of the disk ( lovelace , romanova , & newman 1994 ; lovelace , newman , & romanova 1997 ) , or by tidal interaction with a binary companion ( larson 2002 ) . we have done a wide range of mhd simulations of disk accretion to a rotating aligned dipole in order to understand the different accretion phenomena . the simulations show that funnel flows ( ff ) , where matter flows out of the disk plane and essentially free - falls along the stellar magnetic field lines , are a robust feature of disk accretion to a dipole . specifically , we find that ( 2 ) . the velocity of matter in the ff is much smaller than the alfvn velocity , @xmath393 , so that matter flows along the magnetic field lines . the funnel flow accelerates and become supersonic . the mach number is @xmath394 at the surface of the star . at the star velocity is close to the free - fall velocity : @xmath335 . the angular velocity of the ff gradually varies from its value at the inner edge of the disk to the angular velocity of the star . the ` twist ' of the magnetic field lines in the ff is small , @xmath395 , and it has a maximum approximately in the middle of the ff . the main forces which are responsible for dragging matter to the ff are matter pressure gradient force ( near the disk ) and gravitational force in the rest of the ff . magnetic force is negligibly small . ( 1 ) the magnetic field of the star influences the nearby regions of the disk inside a radius @xmath0 , while viscosity dominates at larger radii . the radius @xmath0 depends on magnetic moment of the star @xmath3 and density in the disk . ( 2 ) . inside the radius @xmath0 the disk is strongly inhomogeneous . the density is @xmath397 times smaller than in the disk without magnetic field . magnetically braked matter accumulates near magnetosphere and forms a dense ring and funnel flow . the star may spin - up or spin - down depending on the ratio of its rotation rate to the rotation rate at the inner radius of the disk . we find that torqueless " accretion is possible when @xmath398 , where @xmath283 is the corotation radius . ( 4 ) . at the star s surface , the angular momentum flux is transported mainly by the ` twist ' of the magnetic field . angular momentum carried by matter in the disk at @xmath279 is transferred almost completely to the magnetic field at the star s surface . the coronal magnetic field is observed to open and close , but strong outflows were not observed for the considered parameters and quasi - equilibrium initial conditions . in the area of the disk where the field is strong , @xmath399 , the magnetic field lines tend to decelerate / accelerate rotation of the disk instead of being opened . besides , opening of magnetic field loops is suppressed by matter dominated corona compared to gbw99 who accepted dependence @xmath355 . strong outflows may be associated with strongly non - stationary accretion in the disk as observed in simulations by hsm96 , ms97 , and h97 or when the disk is very dense and magnetic field lines are inclined to the disk as in simulations by fe99 , fe00 . alternatively the outflows may occur in cases where the density of the corona decreases sufficiently rapidly with distance as in the simulations by gwb97 and gbw99 . sporadic outflows can arise from a global magnetic instability of the disk ( e.g. , lovelace _ et al . _ 1994 , 1997 , 2002 ) , but they may be absent during the quiescent evolution of the disk - star system . this work was supported in part by nasa grant nag5 - 9047 , by nsf grant ast-9986936 . rmm thanks nsf powre grant for partial support . rvel was partially supported by grant nag5 - 9735 , gvu and avk were partially supported by intas grant and by russian program `` astronomy . '' the authors thank the referees for numerous comments and suggestions . bildsten , l. , chakrabarty , d. , chiu , j. , finger , m.h . , koh , d.t . , nelson , r.w . , prince , t.a . , rubin , b.c . , scott , d.m . , stollberg , m. , vaughan , b.a . , wilson , c.a . , wilson , r.b . 1997 , apj supplement , 113 , 367

the evolutionary properties and spatial distribution of i zwicky 18 stellar populations are analyzed by means of hst / acs deep and accurate photometry . the comparison of the resulting colour - magnitude diagrams with stellar evolution models indicates that stars of all ages are present in all the system components , including objects possibly up to 13 gyr old , intermediate age stars and very young ones . the colour - magnitude diagrams show evidence of thermally pulsing asymptotic giant branch and carbon stars . classical and ultra - long period cepheids , as well as long period variables have been measured . about 20 objects could be unresolved star clusters , and are mostly concentrated in the north - west ( nw ) portion of the main body ( mb ) . if interpreted with simple stellar population models , these objects indicate a particularly active star formation over the past hundred myr in izw 18 . the stellar spatial distribution shows that the younger ones are more centrally concentrated , while old and intermediate age stars are distributed homogeneously over the two bodies , although more easily detectable at the system periphery . the oldest stars are best visible in the secondary body ( sb ) and in the south east ( se ) portion of the mb , where crowding is less severe , but are present also in the rest of the mb , although measured with larger uncertainties . the youngest stars are a few myr old , are located predominantly in the mb and mostly concentrated in its nw portion . the se portion of the mb appears to be in a similar , but not as young evolutionary stage as the nw , while the sb stars are older than at least 10 myr . there is then a sequence of decreasing age of the younger stars from the secondary body to the se portion of the mb to the nw portion . all our results suggest that izw18 is not atypical compared to other bcds .

1106.5613

i zwicky 18 ( also mrk 116 or ugca 166 , hereafter izw18 ) is the most intriguing blue compact dwarf ( bcd ) galaxy in the local universe . it became famous right after its discovery , when @xcite measured from its emission - line spectrum an oxygen abundance [ o / h]= 1.14 dex , only 7@xmath0 of the solar value , indicating an almost unprocessed gas content . the first studies on its color and composition @xcite emphasized its almost primordial evolutionary status . izw18 shows very blue colors , u@xmath1b@xmath2 and b@xmath1v@xmath3 @xcite , suggesting the presence of a very young stellar population , with a current star formation rate ( sfr ) much higher than the past mean value . the total mass of izw18 from the rotation curve at a radius of 10@xmath412@xmath5 is estimated to be @xmath6 m@xmath7 ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , where the neutral gas corresponds to @xmath870@xmath0 of the total mass , but only @xmath9 m@xmath7 of hi is associated with the optical part of the galaxy ( e.g. , * ? ? ? * ; * ? ? ? subsequent spectroscopic studies in izw18 @xcite have confirmed its extreme metal deficiency around 1/30@xmath101/50 of z@xmath7 . nowadays izw18 remains the star forming galaxy with the second lowest metallicity and the lowest helium content known @xcite . this makes the system a fundamental point in the derivation of the primordial helium abundance @xcite and in the study of the properties of chemically unevolved galaxies . when discovered by zwicky ( 1966 ) , izw18 was described as `` two galaxies separated by [email protected] and interconnected by a narrow luminous bridge '' , surrounded by two `` very faint flares '' at 24@xmath5 northwest . subsequent ccd ground based images @xcite revealed a more complex structure : the _ two galaxies _ turned out to be two star forming regions of the same galaxy ( usually indicated as nw and se components ) , while the _ two flares _ are just the most prominent of a few nebulosities surrounding izw18 . these minor systems are roughly aligned toward the northwest and were initially considered at the same distance , but subsequent spectroscopic studies showed that only one component ( referred to as component c in * ? ? ? * ) is at the same distance as izw18 and is physically associated with the main body @xcite . the other diffuse objects have been recognized as background galaxies ( see , e.g. , * ? ? ? * ) . in the following we will refer to izw18 and to component c respectively as main body ( mb ) and secondary body ( sb ) of izw18 . both bodies have been resolved into single stars for the first time by hst with the wide field planetary camera 2 ( wfpc2 ) by @xcite and @xcite . the observational features described above triggered the key question over the nature of izw18 : is it a young galaxy which is presently experiencing its first burst of star formation , or is it an old system which has already formed stars in the past ? hst studies by different authors have tried to characterize the evolutionary status of izw18 , with rather discordant results , but with a clear trend intriguingly reminiscent of oscar wilde s dorian gray s picture : the more one looks at it , the older it appears . @xcite , using hst / wfpc2 data , were the first to detect asymptotic giant branch ( agb ) stars in the galaxy , and thus to demonstrate that izw18 is as old as at least several hundred myr , and possibly much older . @xcite failed to detect red giant branch ( rgb ) stars with hst observations made by the advanced camera for surveys ( acs ) , and thus concluded that the galaxy is at most 500 myr old . however , both @xcite and @xcite , from independent re - analyses of the same acs data set , suggested that izw18 should be older than at least 1 - 2 gyr , since it did appear to contain also red giant branch ( rgb ) stars . our own time - series hst / acs photometry ( go 10586 , * ? ? ? * hereafter al07 ) , has allowed us to shed light on the situation , thanks to the combined information from the color - magnitude diagram ( cmd ) and the lightcurves of the cepheids we were able to measure . in fact , al07 not only characterized the rgb and identified its tip ( trgb ) at @xmath11 mag , but , more importantly , detected for the first time a few cepheids whose lightcurves allowed to independently and more reliably pin down the galaxy s distance to the couple mpc accuracy ( fiorentino et al . 2010 , hereafter f10 ; marconi et al . the distance to izw18 was estimated from the cepheids using theoretical @xmath12 wesenheit relations specifically computed for the very low metal abundance of the galaxy , and by theoretically fitting the observed lightcurve of the most reliable classical cepheid with period 8.71 days . the wesenheit relations provide an intrinsic distance modulus for izw18 of @xmath13 ( d=19.0@xmath14 mpc ) adopting a canonical mass - luminosity relation , and of 31.2 @xmath150.2 ( d=@xmath16 mpc ) when an overluminosity of 0.25 dex is assumed for each given mass ( non canonical scenario ) . the theoretical modeling of the 8.71 days cepheid lightcurve provides 31.4 @xmath150.1 ( d=19.0@xmath17 mpc ) . the trgb identified by al07 implies an instrinsic distance modulus of 31.30 @xmath15 0.17 , in excellent agreement with the cepheid - based estimates . this agreement , in turn , provides independent support to our interpretation of the cmd faint red sequence as an rgb . hence , izw18 is actually farther away than originally thought ( for instance * ? ? ? * adopted from the literature a mean value of d=10 mpc ) and is in practice at slightly larger distance than the virgo cluster of galaxies . this made it impossible in the past to resolve its older , hence fainter , stars , and quite challenging even with the outstanding performances of hst s current instrumentation . today izw18 is one of the most distant system ever resolved into stars . in this work , we have used the photometric catalogues of our previous works , which have been focused on determining the age ( al07 ) and the distance through the variable stars ( f10 ) . in this paper we describe the evolutionary properties and the spatial distribution of the resolved stellar populations and candidate star clusters of the main and secondary bodies of izw18 . in section 2 , we briefly summarize the main differences in the photometric procedures adopted by al07 and f10 . the cmds and the spatial distribution of stars in different evolutionary phases are presented in section 3 and 4 , respectively . the results are analysed in section 5 and summarized in section 6 . the data acquisition and reduction have been extensively described by al07 and f10 . here we only summarize their main aspects . time series photometry of izw18 in the f606w ( broad @xmath18 ) and f814w ( @xmath19 ) filters was obtained with the wide field channel ( wfc ) of the acs in 13 different epochs properly spread over a time interval of 96 days ( program i d . 10586 , pi aloisi ) . these proprietary data were complemented by 5 f555w ( @xmath18 ) and 3 f814w archival acs / wfc images of izw18 from hst ( program i d . 9400 , pi thuan ) . izw18 is quite distant , plenty of bright stars hiding and/or contaminating the fainter ones , and plenty of gas , which complicates the photometry of individual stars . in addition , the throughput of the f606w filter is significantly affected by the @xmath20 contribution of ionized gas ( quite abundant in the mb of izw18 ) and makes the star detection and measurement more complicated . thus , depending on the goal , one can find more appropriate to restrict the analysis of the galaxy populations to safer but fewer objects ( al07 ) , or to more uncertain but numerous ones ( f10 ) . in al07 , to produce the galaxy cmd , we built four coadded images by stacking all single epoch images per filter for both archival and proprietary data , separately . then we performed psf photometry of each of these four coadded images using the daophot / allstar package @xcite . the calibration to the johnson - cousins photometric system was obtained using the transformations given by @xcite . since the goal of that paper was to demonstrate that faint red stars actually existed where none had previously been detected , the published cmd was produced following a strategy aimed at minimizing the number of false detections . as a result , it contained only the 2099 sources which were independently detected in all the four deep images ( the @xmath18s and @xmath19s of both datasets ) , with good match in the position ( within one pixel ) and magnitude ( within @xmath21 ) of each source in the different images , and with sharpness @xmath220.5 . these are the most reliable objects measured in each of izw18 s bodies . in the case of f10 , to produce the lightcurves of the candidate variables , we made use of daophotii / allstar / allframe @xcite to get psf photometry on each single epoch ( 15 @xmath19 and 18 @xmath18 images , in total between proprietary and archival data ) . to study the variability we relaxed the al07 selection criteria , since variable stars are more easily distinguishable from spurious objects thanks to their brightness variation in the images at different epochs moreover , to collect a good sample of candidate variables at all magnitudes , we examined objects close to the detection threshold . this photometry resulted in a @xmath23 cmd of izw18 which contains 6859 sources with photometric errors @xmath24 mag and @xmath25 parameter in the range from @xmath26 to @xmath27 , and reaches @xmath28 30.0 , i.e. more than half a magnitude deeper than al07 s photometry . the calibration to the johnson - cousins photometric system was obtained using the transformations given by @xcite , as for the other data set . as shown above , there is a remarkable difference in the number of objects selected with the two approaches . to infer the properties of the stellar populations inhabiting izw18 minimizing the effects of uncertain photometry , we will base our results on the hyper - selected al07 catalogue . nonetheless , we consider it useful to show some features of the larger f10 catalogue both for completeness and for comparison . since our aim is to analyse the evolutionary properties of izw18 s stellar populations , we have put special care in verifying whether the selection differences affect some cmd regions more than others . we have hence divided the cmds of the two sets in 8 zones roughly representative of different evolutionary phases . since the trgb was detected by al07 at @xmath29 , we have adopted this magnitude as the border between bright and faint stars : * zone 1 , stars in the extremely hot portion of the blue plume with @xmath30 and brighter than @xmath29 ; * zone 2 : stars in the extremely hot portion of the blue plume with @xmath30 and fainter than @xmath29 ; * zone 3 : stars in the standard " blue plume , with @xmath31 , and brighter than @xmath29 ; * zone 4 : stars in the standard " blue plume , with @xmath31 , and fainter than @xmath29 ; * zone 5 : objects with @xmath32 and any colour ; * zone 6 : stars with @xmath33 and @xmath34 ; * zone 7 : stars with @xmath35 and @xmath36 ; * zone 8 : all the remaining ones;. the number of objects measured in each zone is reported in table 1 for both bodies and both data sets . lrrrrrrrr zone 1 & 76 & 6 & 649 & 13 & 0 & 0 & 0 & 0 + zone 2 & 84 & 7 & 917 & 20 & 43 & 5 & 108 & 5 + zone 3 & 277 & 23 & 820 & 17 & 93 & 9 & 98 & 4 + zone 4 & 186 & 16 & 707 & 15 & 262 & 29 & 584 & 28 + zone 5 & 34 & 3 & 51 & 1 & 2 & @xmath371 & 2 & @xmath371 + zone 6 & 105 & 8 & 210 & 4 & 60 & 6 & 110 & 5 + zone 7 & 126 & 11 & 349 & 8 & 120 & 13 & 375 & 18 + zone 8 & 305 & 26 & 1060 & 22 & 332 & 36 & 838 & 40 + total & 1187 & 100 & 4744 & 100 & 912 & 100 & 2115 & 100 + the zones are displayed on the al07 cmds in figure [ cmdmap07 ] . figure [ image814 ] shows an f814w image of izw18 with the f10 measured objects colour - coded according to the 8 cmd zones of figure [ cmdmap07 ] . we arbitrarily define as mb objects those with y coordinate in the image smaller than 1200 px and as sb objects those with y@xmath381200 px . in figure [ cmdmap07 ] , and all the following ones , the cmds of the two bodies are splitted following this definition . we have found that the two photometries agree well with each other , once we take the different selection criteria and photometric depth into account . the objects measured by both al07 and f10 fall consistently in the same cmd zone ( i.e. have the same colours and magnitudes ) . the f10 objects not present in the al07 hyper - selected catalogue fall mostly in the fainter portions of the cmds , as expected since they can have larger photometric error . an intriguing difference is that the mb zones 1 and 2 are much more populated in f10 than in al07 ( see table 1 ) , with a ratio n(f10)/n(al07)=8.5 in zone 1 and 10.9 in zone 2 , significantly higher than the total mb ratio 4744/1187=4.0 . figure [ image814 ] reveals that most of the f10 objects of zones 1 , 2 and 4 fall on the gas - rich regions in and around the mb . a possible cause for the discrepancy is therefore gas contamination in the f606w flux , even if in principle all photometric packages should overcome this kind of effects by subtracting the surrounding background flux to each source . as a matter of fact , many of the objects in zone 1 and 2 show a discrepancy in @xmath18-magnitude depending on the used filter , e.g. f606w or f555w . if , we restrict the selection criteria on the f10 catalogue and keep only the objects with @xmath390.1 , most of those in the cmd zone 2 and several of those in zone 1 are rejected . we thus consider as real stars on the gas filaments only the brighter ones with smaller photometric error . finally , we have individually and visually inspected the brighter objects of f10 zone 1 using appropriate routines , and found that some of them do appear as real stars . we consider the vast majority of the blue objects measured by f10 on the gas filaments and not present in the al07 catalogue as likely spurious detection due to gas peaks , but we do believe that some of the brighter zone 1 detections are real stellar objects . to be conservative , we will base all our results on the restricted al07 catalogue . to let the reader easily figure out the involved stellar ages , the left panel of figure [ cmdisotracks ] shows the cmds of the hyper - selected al07 objects with overimposed theoretical isochrones @xcite . the isochrones have metallicity z=0.0004 and are plotted adopting @xmath40=31.4 ( al07 and f10 ) and e(b - v)=0.032 @xcite . the cmds corresponding to the less restrictively selected f10 photometry are displayed in the right panel of figure [ cmdisotracks ] , where stellar evolution tracks are also plotted to let the reader visualize what the involved star masses are . the plotted tracks , again with z=0.0004 , are the classical padova models by @xcite . we note that izw18 is located at high galactic latitude ( @xmath41 ) and therefore its cmds do not suffer from significant contamination from foreground stars belonging to our galaxy , meaning that is not necessary to correct for it our observed diagrams . izw18 is a starforming dwarf and its cmds show all the features typical of these galaxies : a prominent blue plume with mean @xmath42 and a fainter and broad red plume with @xmath33 . once compared with the stellar evolution tracks of figure [ cmdisotracks ] , the blue plume turns out to correspond to high- and intermediate - mass stars in the main sequence ( ms ) phase or at the blue edge of the blue loops covered during central he burning . these two phases are resolved by hst photometry in nearby late - type galaxies , but at the distance of izw18 the photometric error inevitably is large enough to let them merge in just one broad sequence . stars with colours intermediate between the blue and the red plume are either supergiants ( brighter than @xmath43 ) or intermediate - mass stars in the blue loop phase . the brighter ones might also be unresolved star clusters . the red plume is populated by red supergiants ( @xmath44 ) , agb ( @xmath45 ) and rgb stars ( @xmath46 ) . our data allow to cover the brighter 1.5 magnitude portion of the rgb , where @xmath47 is @xmath150.1 mag in @xmath19 and @xmath150.2 in @xmath48 ( al07 ) and the photometric error , estimated from extensive artificial star tests @xcite , is smaller than @xmath150.3 mag in @xmath19 and @xmath150.4 in @xmath48 ( see figure [ cmdmap07 ] ) . the colour extension of the rgb is larger than this error and reflects , as usual , both the spreads in age and in metallicity of the resolved stars . however , given the extremely low metallicity of izw18 , in this case it is mostly the age that contributes to the rgb thickness . in other words , the reddest rgb stars are most likely low mass stars @xmath49@xmath50 gyr old . the redder red plume stars with @xmath51 are presumably carbon stars , not too different from those measured in the magellanic clouds by @xcite . the redder objects at @xmath52 are most probably stars of @xmath53@xmath54 m@xmath7 in their thermal pulsing ( tp ) phase at the end of the agb , since their location in the cmd is perfectly consistent with the predictions of marigo & girardi s ( 2007 ) tp - agb models . these authors show in fact that at very low metallicities ( z @xmath55 ) in ( relatively ) massive intermediate - mass stars efficient nuclear burning at the base of the convective envelope may favour the formation of c - rich models and make their tracks describe pronounced excursions towards lower effective temperatures . these excursions are recognizable in figure [ cmdisotracks ] in the tangled , messy portions of the isochrones above the trgb . 650 px , top panel ) from those in the se region ( y@xmath56650 px , bottom panel ) . there are 478 objects in the nw region and 704 in the se one.,width=321 ] not only does the mb , by definition , host a larger number of stars ; it also contains a larger fraction of young , blue and bright stars than the sb . both the luminosity and the number of the brighter stars in the sb are significantly lower than in the mb . almost no supergiant is found in the sb and no cepheid . moreover , while the blue plume of the mb covers the entire range of blue massive stellar evolution tracks and youngest isochrones , that of the sb turns off the ms around @xmath57 , with apparently no bright star younger than 10 myr being detected there . this is a significant difference , which reflects , as discussed already by @xcite , both a different rate and a different history of star formation , with the mb being more active than the other at recent epochs , and the sb more active earlier than about 100 myr ago , as suggested by the higher density of objects around the isochrones older than this age in the left - bottom panel of figure [ cmdisotracks ] . at the faintest magnitudes , we should also consider that crowding affects the mb more severely , preventing the detection of its faintest stars . this may be particularly the case in the nw part of the mb . the variable stars in izw18 have been identified by f10 exploiting our acs time - series photometry . the intensity - averaged mean magnitudes and colors of the confirmed cepheids are plotted as filled ( red ) triangles in figure [ cmdisotracks ] , while those of the confirmed long period variables ( lpvs ) are plotted as filled ( red ) squares . their position ( see f10 for identifications and nomenclature ) in the cmd indicates that the two bright ultra - long period cepheids ( ulps ) v1 and v15 are supergiants , unless they are hosted by unresolved star clusters making them appear brighter . the classical cepheid v6 with period p=8.71d is on the blue loop and is best fitted by a @xmath86 m@xmath7 star ( see also f10 and * ? ? ? the brighter lpv v4 is a supergiant , while the other confirmed lpv v7 lies at the bright edge of the cmd region of the tp - agb stars and is presumably a mira . for a finer analysis of the evolutionary properties of izw18 components , we have further divided the al07 objects according to their position in the mb . figure [ mb_separate ] shows the cmds of the objects with y coordinate in the image larger ( top panel ) or lower ( bottom panel ) than 650 px ( see figure [ image814 ] ) . this in practice separates objects in the nw ( top ) part of the mb from those in the se ( bottom ) . the two cmds are fairly similar to each other , although the nw one contains more objects in zone 5 and is more dispersed than the other . in particular , the red plume of the se region is much better defined than that of the nw and allows to clearly distinguish the rgb and its tip from the agb . the larger spread of the nw cmd is likely due to the presence of a larger number of bright objects and thick ionized gas filaments that make the measurements of fainter ( and older ) stars more uncertain . it is plausible that the gas ionization is itself due to the massive luminous objects populating the nw region ; so , they may represent the only real difference between the northern and southern parts of the mb . we will further discuss these objects in the next section . in spite of the very different number of objects selected in the two data sets , once the different depths of the resulting cmds ( figure [ cmdisotracks ] ) are taken into account , their stellar evolutionary phases turn out to be populated in similar proportions , both in the mb and in the sb ( see table 1 ) . as discussed in the previous section , the only significant difference is that the mb blue plume is much more populated , bluer and thicker in the f10 cmd . we have analysed the spatial location of the various types of stars populating izw18 and checked if stars of different ages are evenly or unevenly distributed . to this purpose , we have used the cmd zones of figure [ cmdmap07 ] . thanks to the comparison with the evolutionary tracks , we can assign to each zone the following evolutionary properties : * zone 1 corresponds to massive and intermediate mass stars on the ms , younger than @xmath8 1015 myr ; * zone 2 to fainter intermediate mass stars on the ms , younger than@xmath8 70 myr ; * zone 3 to massive and intermediate mass stars on the ms or at the blue edge of the blue loop phase , younger than @xmath8 70 myr ; * zone 4 to fainter intermediate mass stars on the ms or at the blue edge of the blue loop phase , younger than @xmath8 300 myr ; * zone 5 to supergiants , younger than @xmath8 10 myr and , possibly , to unresolved star clusters ; * zone 6 to agb stars , ages from @xmath8 35 to 300 myr ; * zone 7 to rgb stars , ages from @xmath8 1 to 13 gyr ; * zone 8 in most cases corresponds to intermediate - mass stars in the blue loops , but also to some supergiants , ages from @xmath8 15 to 300 myr . the spatial distribution in the x - y map of izw18 of the al07 objects measured in the two bodies and falling in each of the 8 cmd zones is reported in figure [ map07 ] . this map shows that the sb does not contain stars in zone 1 . in other words the sb has had very little , if any , star formation activity in the last few myr . zone 1 stars covering the whole range of magnitudes ( hence with masses from 1520 m@xmath7 up ) are numerous in the mb and show a sort of 8-shaped spatial distribution . if we look at the fainter portion of the extremely blue stars ( zone 2 ) , we find that both bodies contain several of these objects and that in the mb they seem to delineate the same 8-shape as their brighter counterpart . the bright stars in zone 3 have masses above 6 m@xmath7 and are also present in both bodies . their spatial distribution in figure [ map07 ] forms an irregular ellipsoidal clump of objects with significant central concentration ( the mb ) with a comma - shaped companion ( the secondary : zwicky s flare " ) . this is the overall familiar appearance of izw18 , since these are the stars more easily visible , although not resolvable from ground . once again , the sb stars of zone 3 do nt reach the luminosity of the brightest ones hosted in the mb ( see figure [ cmdmap07 ] ) : while the whole mass and luminosity ranges are covered in the latter , only stars less luminous than @xmath58 exist in the former . taking into account that any initial mass function disfavours the most massive stars , we consider it more likely that the brighter blue stars in the sb are 1215 m@xmath7 stars at the blue edge of the blue loop rather than more massive stars . in this case , the star formation activity in the secondary must have been quite low , or absent , in the last 10 myr . fainter ( lower mass ) blue plume stars ( zone 4 ) can be either ms stars with masses between 620 m@xmath7 or stars at the blue edge of the blue loop with mass in the range 35 m@xmath7 ( see the tracks in right panel of figure [ cmdisotracks ] ) . they are present in both bodies , and are more abundant in the sb than in the mb , where they are not found in the central regions and have a spatial distribution complementary to that of their brighter counterpart of zone 3 . this is clearly due to the severe crowding affecting the mb and , particularly , its inner regions , where the many bright stars completely hide fainter ones . in the sb the recent star formation activity has been less intense , crowding is less severe , and faint stars are much more easily detectable . the lower activity of the sb at recent epochs is also confirmed by the objects of zone 5 . as apparent from figure [ map07 ] , we find only 2 such bright objects in the sb ( one of which is quite far away and may actually be a background system ) , and 31 in the mb . the latter are fairly confined in the inner regions , and most often in the nw portion of the mb . they significantly contribute to the severe crowding affecting the central area . if interpreted in terms of individual stars , these objects are supergiants and confirm that the central region of the mb has experienced a strong star forming activity in the last 10 myr or so . admittedly , some of these very bright objects might be not individual stars but unresolved star clusters . we discuss this possibility in section [ cluster_candidates ] . the objects falling in the cmd zone 6 are ( see the tracks in figure [ cmdisotracks ] ) agb stars with mass lower than 9 m@xmath7 , and age ( see the isochrones in the same figure ) in principle between 35 myr and 13 gyr , but in practice peaked at , at least , several hundreds myr , once the stellar lifetimes and phase durations are taken into account . they include the tp - agb and carbon stars with 4 - 5 m@xmath7 ( horizontal finger at @xmath5925.5 in the cmds ) predicted by @xcite for these low metallicities and the carbon stars at @xmath5926.4 discussed by al07 . zone 6 objects are equally present and evenly distributed in both bodies , although missed in the most crowded portions of the mb . the rgb stars are in zone 7 : objects with mass below 1.8 - 2.0 m@xmath7 and age between 1 and @xmath60 gyr . the rgb is notoriously affected by the age - metallicity degeneracy , i.e. the ambiguity that stars with the same brightness and colour can be older and metal poorer or younger and metal richer . hence , without independent spectroscopic metallicity measurements one can not assign the age to rgb stars simply from their position in the cmd . for this reason the photometric identification of the rgb normally puts only a lower limit of 1 - 2 gyr ( i.e. the minimum time required by low - mass stars to reach that phase ) to the age of the examined stellar population . however , in the case of izw18 , its extremely low metallicity somewhat relaxes the effect of the age - metallicity degeneracy , since the redder rgbs can not be metal rich objects and therefore must be old . in these conditions , the colour extension of izw18 rgb , which is broader than the photometric error and covers the whole range of isochrone ages , can be taken as a signature of the presence of stars possibly as old as 13 gyr . rgb stars are faint and therefore their detection is strongly affected by crowding and photometric performances . this is why only with the acs has it been finally possible to resolve and measure them : not even hst was able to reach the rgb , with the less sensitive wfpc2 . the red giants are evenly distributed in the sb and visible only in the outer regions of the mb , due to the excessive crowding of its inner parts . it is interesting to notice that the mb rgbs are located along a semi - arc in the system periphery . we ascribe this asymmetry to the presence of thick gas filaments in the regions where rgbs are not detected : overlapping filaments can clearly hide such faint objects . we thus believe that rgb stars can actually be evenly distributed also in the mb . finally , zone 8 collects all the remaining objects , with a large range of brightness and colour , and different ages and evolutionary phases . those with @xmath6124.5 are stars in the 310 m@xmath7 mass range in the central helium burning phase and with ages between 30 and 300 myr . the brighter ones are 1015 m@xmath7 mass stars in any post - ms evolutionary phase , with age of the order of 1530 myr . these mixed objects are evenly distributed in both bodies , with the only exception of the two most crowded nw and se spots of the mb , where they are missing . interestingly , like the objects of zones 2 , 4 and 7 , they are found in similar numbers in the two bodies . even considering the effect of the different crowding , this suggests that as far as low and intermediate mass stars are concerned , the sb is not significantly less populated than the mb . as for the variable stars , the two ulps v1 and v15 are located in the nw portion of the mb , while the classical cepheid v6 and the two lpvs v4 and v7 are in the se portion ( see f10 figure 5 ) . interestingly , all the confirmed variables belong to the mb : of the 34 good candidate variables examined by f10 , almost half ( 15 ) were actually detected in , or around , the sb , but for none of them satisfactory lightcurves and periods were derived . ground - based observations of the brighter variables are being performed to better characterize their pulsation and evolutionary properties . in particular , further data on the ulps will allow us to verify if they are the extension to higher masses of classical cepheids and , in case , to calibrate them as extremely useful primary indicators of cosmological interest . finally , we emphasize that no star is found in the region between the mb and the sb , either in the al07 or the f10 catalogues . that region looks completely transparent , with only background galaxies being detected there . this shows that the two bodies are completely separated . they are members of a two - body system , but have no star connecting each other . if tidal interactions are at work , they do nt appreciably affect the stellar populations of the two components . at 19 mpc , acs / wfc photometry can not recognize round regularly shaped clusters whose sizes fall within the point spread function . in this section we consider the possibility that zone 5 objects are actually not individual stars but unresolved star clusters , and discuss the implications on izw18 s evolutionary status . versus @xmath48 color - color diagram ( top ) and @xmath18 versus @xmath48 color - magnitude diagram ( bottom ) for the objects in zone 5 ( cluster candidates ) that have been cross - correlated with the wfpc2 @xmath62 photometry presented by @xcite . full symbols denote objects located far away from the mb and sb which could be background galaxies . overplotted are the padova simple stellar populations @xcite for a [email protected] metallicity , ages from 1 myr to @xmath813 gyr , and masses of @xmath64 , @xmath65 , and @xmath66 . small dots indicate the log(age)=6.7 , 7 , 8 , 9 and 10 models . the red arrow represents an @xmath67 extinction vector ( see text ) . , width=302 ] with an intrinsic distance modulus of @xmath40=31.4 , the objects of zone 5 have an absolute bolometric magnitude brighter than -8.4 , certainly consistent with typical cluster luminosities . actually , the brightest objects , with @xmath68 , hence @xmath69 , would be populous " clusters , according to the definitions by @xcite and @xcite . these bigger , more massive clusters are fairly common in starburst dwarfs , such as ngc 1705 @xcite , ngc 1569 @xcite and ngc 4449 @xcite , and tend to be centrally concentrated , just like these brightest objects of izw18 . hence , also if taken as star clusters , the objects of zone 5 would confirm a higher star formation activity in the mb than in the sb . to infer the evolutionary properties of unresolved clusters , two photometric bands are insufficient . fortunately izw18 was observed also in the b band with the wfpc2 by @xcite . we cross - correlated the 31 + 2 objects in zone 5 with the wfpc2 ( f435w , f555w ) photometry presented by @xcite . the cross - correlation provides 23 + 2 objects photometrized in f439w , f555w and f814w . a visual inspection in our images of the objects missed by the cross - correlation shows that they are all located in the most crowded mb region , and thus their position measurement is highly uncertain , also because of blending effects . the wfpc2 magnitudes were transformed into johnson - cousins @xmath70 and @xmath19 magnitudes using the transformations by @xcite . in fig . [ clusters ] we show the @xmath71 versus @xmath48 color - color diagram and the @xmath18 versus @xmath48 color - magnitude diagram for zone 5 objects . the data are compared with the padova simple stellar populations ( ssps ) with metallicity [email protected] , ages from @xmath81 myr to @xmath813 gyr , and masses of @xmath64 , @xmath65 , and @xmath66 @xcite . zone 5 objects have @xmath48 colors in the range @xmath72@xmath73 , and @xmath71 colors in the range @xmath72@xmath74 , and are compatible with ssp ages from a few myr to a hubble time . the two objects with @xmath75 , @xmath760.8 , and @xmath77 , @xmath760.4 , are located in the most crowded region of the main body , and therefore their colors suffer larger uncertanties . the objects concentrated around @xmath78 can be either @xmath79 myr old clusters or clusters in the red loop of the models at ages around @xmath80 myr . we notice that there are some objects with @xmath81 which are redder than the oldest ssps . we can exclude that they are background galaxies because they are all concentrated within izw18 s mb , except for two of them , which are more `` peripheral '' . another possibility is that these objects are highly reddened young embedded clusters ; however , in this case they should have @xmath82 ( see reddening vector in figure [ clusters ] ) , and thus absolute magnitudes brighter than @xmath83 . such high luminosities would imply a super star cluster ( ssc ) classification according to the definition by @xcite . however , it is unlikely that izw18 host so many sscs , considering that ngc 1569 , the dwarf with the highest known current sfr , presents only 4 of them @xcite . several studies have discussed stochastic effects on the integrated colors of clusters ( e.g. * ? ? ? * ; * ? ? ? stochasticity in the imf sampling of a low - mass cluster can heavily change the observed colors with respect to predictions , and in particular , red supergiants make appear the cluster colors redder . @xcite investigated this issue and derived a luminosity limit under which stochastic effects are important . comparing the luminosities of our clusters redder than @xmath84 with their predictions for ages older than 100 myr , we found that our cluster luminosities are well above that limit . therefore , the most likely explanation is that these objects are supergiants with ages younger than @xmath820 myr , implying that the mb has been forming stars at very recent epochs . in fig . [ clusters ] have not been included . , width=302 ] for the zone 5 objects with @xmath85 we have inferred ages and masses using the padova models . this was done minimizing the difference between model and data colors ( b@xmath1v , v@xmath1i , and b@xmath1i ) according to a @xmath86 criterion , varying the age between @xmath81 myr and @xmath813 gyr . we accounted for a galactic reddening of e(b - v)@xmath630.032 , but did not solve for possible internal extinction , because of the small number of bands ( 3 ) and the relatively short wavelength range covered by our data . following the prescriptions of @xcite , the 1 @xmath87 errors were derived as the limits of the 0.68 confidence level region defined by @xmath88 . cluster masses were obtained from the mass - to - light ratios of the best - fitting models in the i band . the results are presented in a mass versus age diagram in fig . [ clusters_agemass ] . cluster ages span the whole hubble time , even if we notice that the majority of clusters have ages younger than a few hundred myr . cluster masses range from @xmath89 to @xmath90 . the observed correlation between cluster ages and masses is an obvious effect of the magnitude limit adopted for zone 5 . we notice that our analysis provides 5 `` old '' ( age@xmath911 gyr ) clusters , with masses between @xmath92 and @xmath90 , and totaling @xmath93 . this implies that @xmath83% of the stellar mass in izw 18 is in the form of old clusters , which sounds unlikely given the fact that this ratio has been found to be considerably lower in other dwarf irregulars with evidence for significant sf at ages @xmath941 gyr ( e.g. , @xmath370.3% in ngc 4449 , annibali et al . this suggests that the majority of the candidate old clusters in izw18 are instead resolved supergiants . concluding , our analysis seems to suggest that the majority of the objects in zone 5 are indeed individual stars , even if we can not exclude that some of them are unresolved stars clusters . nevertheless , even in this scenario , our analysis provides evidence for a particularly active sf over the past hundred myr . the evolutionary properties derived above for the stars measured in izw18 are similar to those of other metal - poor starbursting dwarfs . for instance , once the difference in distance modulus @xmath95 is taken into account , magnitudes and colours of the various types of stars on the red plume of izw18 are in perfect agreement with those of the corresponding stars in sbs1415 + 437 , the other very metal - poor bcd whose stellar populations have been resolved by hst / acs @xcite . indeed , the cmd of sbs1415 + 437 , whose intrinsic distance modulus is 30.66 , shows an unmistakable trgb at @xmath96 and the horizontal finger typical of carbon stars at @xmath97 @xcite . sbs1415 + 437 has a red plume more populated than izw18 , as if it had been more active in previous epochs . in turn , the blue plume of izw18 s mb is more prominent than that of sbs1415 + 437 , suggesting that its current and recent star formation activity is proportionally higher . the results described in the previous sections allow us to draw a qualitative picture of the evolution of the mb and sb . both components contain old and young stars in all evolutionary phases , but it is apparent that the two stellar systems are distinct from each other and have had different histories . the quantitative star formation history ( sfh ) of the two bodies of izw18 will be derived from the cmds using the synthetic cmd method ( see e.g. * ? ? ? * and references therein ) and will be presented in a forthcoming paper @xcite . the spatial distribution of stars in different evolutionary phases displayed in figures [ image814 ] and [ map07 ] suggests that the mb is indeed one single system and not two galaxies separated and interconnected by a narrow luminous bridge " as originally thought by @xcite . we can however understand how he was misled by the 8-shaped distribution of the young blue stars . young massive stars ( zones 1 and 3 of figure [ cmdmap07 ] ) cover the entire mb . zone 3 stars are concentrated on the nw and se star forming regions , following the traditional scheme of starburst dwarfs , where massive young stars usually group together and are concentrated in the inner regions , although not necessarily at the very center of the galaxy . the nw region contains a larger number of bright massive zone 3 objects than the se region . this suggests a more recent / higher star formation activity in the nw clump . zone 1 stars seem ( see figure [ map07 ] ) intriguingly to avoid the areas where zone 3 stars are located . is this because they are different objects ? since they are redder , zone 3 stars might be more evolved or more metal rich than zone 1 stars . we rather believe that either zone 1 stars lie on gas filaments and are slightly blue - shifted by residual gas contamination increasing the f555w flux , or that differential reddening slightly affects zone 3 objects . throughout this paper we have only considered the foreground galactic reddening , @xmath98 , but izw18 does contain a lot of gas and presumably dust and molecules , although in quantities limited by its low metallicity @xcite . it is thus plausible that absorbing interlopers are more abundant in the innermost , star forming regions and that they make the stars of those regions appear redder . in this framework , zone 3 stars are the same kind of objects as zone 1 stars , only affected by internal extinction of the order of @xmath990.1 . lower mass young stars ( zones 2 and 4 ) could help in discriminating between the above interpretations , but they are faint and are much more severely affected by crowding than by any other effect . indeed , although likely to exist also where massive young stars are found , they are only detectable where the latter do nt outshine them . as discussed in the previous sections , in f10 s catalogue many of these objects lie on the gas filaments protruding from the nw spur . while the vast majority of these faint f10 detections on the filaments are likely to be gas peaks misinterpreted as stars , we do believe that some of the brighter ones , which appear as actual stellar sources when individually and visually inspected in the images , are indeed very young stars or unresolved star clusters ( see also * ? ? ? this leads to speculate that star formation is occurring at least in the portions of the filaments closest to the mb . unfortunately , we do nt know whether these filaments are only made of ionized gas or also of neutral one , since the current resolution of hi data ( f. fraternali private communication , lelli et al . in preparation ) does nt allow one to resolve them . neither do we know whether the filaments are infalling , outflowing or comoving with respect to the mb . from van zee s et al . ( 1998 , their figure 6 ) comparison of the @xmath20 and hi distributions , the filaments appear to be a distinct feature of the ionized gas , but the hi resolution is definitely insufficient . it is thus difficult to assess if star formation is physically plausible there ; certainly it is extremely unlikely that it occurs in hot , possibly outflowing bubbles , such as many of izw18 ionized filaments @xcite . also the massive objects of zone 5 are centrally concentrated , as usually found in starburst dwarfs . whether they are resolved supergiants or unresolved star clusters , they are the signature of strong star formation activity . the evidence that they are more abundant in the nw than in the se clump suggests that the activity in the former is higher than in the latter . the nw clump is also known to contain wolf - rayet stars ( see e.g. * ? ? ? * ; * ? ? ? * ) . yet , from emission line spectroscopy , @xcite suggested it to be not as young as the se clump : 10 myr the former and 5 myr the latter ( and 20 myr the sb ) . the cmds of figure [ mb_separate ] do not seem to confirm their suggestion . rather , our cmds suggest that the star formation activity in the nw is more recent than in the se . the circumstance that , of the variable stars reliably characterized by f10 , the two massive ulp cepheids v1 and v15 are located in the nw region , while the lower mass classical cepheid v6 and the fainter lpv v7 are located in the se region , favours the view that the nw on average contains younger stars than the se . all the lower mass evolved stars are likely to be evenly distributed in the mb . rgb stars seem to be located more externally than agbs , but this is most probably a selection effect . we ascribe their apparent confinement in the outer regions to the severe central crowding which prevents their detections . similarly , we interpret their asymmetrical horseshoe distribution as due to the extinction from the thick gas filaments of the northern areas ( see * ? ? ? * for the spatial distribution of dust ) . there is no doubt that faint objects are more easily detected in the se external portions of the mb . the resulting overall scenario of the mb consists of a system with two active , distinct regions of current star formation , one ( the nw one ) stronger and younger than the other . these two active clumps are embedded in a fairly homogeneous body of stars of all ages , from a few myr to several ( possibly 13 ) gyr . the sb is in a rather different situation . as mentioned above , the weaker emission lines of its spectrum already led @xcite to suggest that it is older than the mb emission regions . wfpc2 photometry allowed @xcite to quantitatively assess that its recent sfr is much lower than in the mb . the acs photometry confirms these results . our cmds show that the sb does contain young blue plume stars , but they are more likely stars at least 10 myr old on the blue loop phase . stars more massive than @xmath100 do not seem to be still alive . the sb also appears to contain much less ionized gas than the mb : a further confirmation of its low recent activity . on the other hand , low and intermediate mass stars are at least as numerous in the sb as in the mb , with the objects of zone 4 ( low mass ms stars ) even outnumbering those in the mb ( see table 1 ) . this is a strong indication that at epochs earlier than @xmath8100 myr ago , the sb must have been actively forming stars as , or more than , the mb . if we compare the cmd of the mb se region ( figure [ mb_separate ] ) with that of the sb ( figure [ cmdisotracks ] ) , we see that the two red plumes are similar , both in the rgb and the agb sequences . the blue plume , although containing similar numbers of stars , clearly shows that those in the sb are more evolved than those in the se , that reach brighter magnitudes and bluer colours . the agb and the rgb of the sb are well defined . the colour extension of the latter , combined with the small metallicity range of the system , indicates that stars with the oldest possible ages ( 13 gyr ) are present . we conclude that the sb has started its star formation activity up to @xmath60 gyr ago , and has remained active since then , at rates comparable with those in the mb until relatively recently . however , in the most recent epochs ( i.e. later than 10 myr ago ) its sfr has dropped to values significantly lower than in the mb . we have studied the evolutionary properties and spatial distribution of the stars resolved in izw18 by hst / acs proprietary and archival photometry . to this aim we have analysed both the al07 and the f10 catalogues and found consistent results , independently of the different selection criteria adopted in the two approaches . nonetheless , to be conservative , we have based our conclusions on the al07 most restrictively selected catalogue . the comparison of the cmds resulting from the acs photometry with stellar evolution models ( tracks and isochrones ) indicates that stars of all ages are present in the two bodies . in spite of the impossibility of reaching the oldest ms turn - off in a galaxy 19 mpc away , the extremely low metallicity of izw18 relaxes the age - metallicity degeneracy and makes the red edge of the rgb a signature of the presence of stars up to possibly 13 gyr old . the oldest detectable stars are best visible in the sb and the se portion of the mb , where crowding is less severe , but are present also in the rest of the mb , although measured with larger uncertainties . we have found that the stars are homogeneously distributed over the two bodies , with the younger ones more centrally concentrated , as always found in starburst dwarfs , and the older / fainter ones spread out to the system periphery , also because of selection effects in their detectability . from the maps of the stars spatial distribution ( figure [ map07 ] ) we conclude that old and intermediate age stars are actually distributed homogeneously over the two bodies . * the distance independently estimated from the trgb ( al07 ) and the cepheids ( f10 ) is @xmath101 mpc . * as far as stars are concerned , the sb is completely separated from the mb : no star is detected between the two bodies . if tidal interactions are at work , they do nt appreciably affect the distribution of the stellar populations of the two components . * izw18 contains stars in all the evolutionary phases visible at its distance : massive and intermediate mass stars on the ms , blue and red supergiants , agb and tp - agb stars , carbon stars , rgb stars . hence , it hosts stars of all ages , from a few myr up to possibly 13 gyr . * the mb hosts two bright star forming regions ( nw and se ) , embedded in a common and homogeneous environment of older stars . some stars may have been recently formed on the gas filaments protruding from the nw portion of the mb . * while old and intermediate age stars are present in similar amounts in the mb and in the sb , the youngest stars are only found in the mb and mostly concentrated in its nw region . the se region has also been active recently , but at a lower current rate ; the sb has had definitely less star formation than the mb in the last tens of myr . * of the 34 candidate variable stars studied by f10 , 19 and 15 are in , or around , the mb and the sb respectively . all the confirmed ones are however in the mb : the two ulp cepheids in the nw region ; the classical cepheid and the two lpvs in the se region . * at izw18 s distance , star clusters are not distinguishable from point - like sources . it is thus possible that some of the brightest objects in the main body are actually clusters and not individual stars . comparing their colors with simple stellar population models , we find that the candidate clusters would span ages from a few myr to a hubble time , with the majority of them being younger than @xmath8200 myr . however , the fraction of stellar mass in old ( age@xmath911 gyr ) clusters would be significantly higher than what found in other dwarfs , suggesting that the majority of the candidate old clusters are instead resolved stars . the reddest ( @xmath102 ) brightest objects can not be unresolved clusters and must be supergiants with ages younger than @xmath820 myr . these results show that izw18 is not unique in the sky , but shares the evolutionary properties typical of bcds and , more in general , of late - type dwarfs ( see * ? ? ? * for a review ) . it is not experiencing now its first burst of star formation , but has been forming stars over many gyrs , most likely over the whole hubble time . this implies that its extremely low metallicity can not be explained by a low chemical enrichment due to extremely recent star formation . at best , one may hope that the sfr has been low until relatively recently ( see e.g. * ? ? ? * ) , although the conspicuous presence of stars of all ages does not make this likely . the same problem obviously concerns all the other metal - poor bcds where stars many gyr old have been found , like sbs1415 + 437 @xcite . we recall that no genuinely young galaxy has been found yet by anybody . all the galaxies whose stellar content has been resolved so far contain stars as old as the lookback time allowed by the available instrumentation , whatever their morphological type and metallicity ( see e.g. * ? ? ? * ; * ? ? ? * and references therein ) . hence , since stellar nucleosynthesis has had plenty of time to pollute them , the only ways to explain the low observed metallicity of these galaxies are either to dilute sufficiently their insterstellar medium by accretion of primordial or very metal poor gas , or to remove the stellar nucleosynthesis products through galactic winds triggered by supernova explosions , or both . chemical evolution models of late - type dwarfs and bcds have shown long ago @xcite that infall of metal poor gas alone is not sufficient to reproduce the low metallicity of starburst dwarfs , and that galactic winds must be invoked too . observational evidence for galactic winds has been found for a handful of starburst dwarfs @xcite , and izw18 is indeed one of those where the outflowing ( martin 1996 ) ionized gas associated with the most recent sf activity may indeed have a speed higher than the escape velocity and be able to remove metals from the system . it would be crucial to get detailed information also on the cold gas , see how it is spatially distributed , and what its kinematics is . understanding the chemical and dynamical evolution of izw18 represents a key step to approach the evolution of all dwarf galaxies as well as that of more massive systems which may have formed from the assembly of small objects like it . we thank filippo fraternali and renzo sancisi for interesting conversations on the hi gas properties . financial support for this study was provided by asi through contracts cofis asi - inaf i/016/07/0 and asi - inaf i/009/10/0 . support to the us co - authors was provided by nasa through grants associated with program go-10586 from the space telescope science institute ( stsci ) , which is operated by the association of universities for research in astronomy , inc . , under nasa contract nas5 - 26555 . aloisi , a. , tosi , m. , & greggio , l. 1999 , , 118 , 302 aloisi , a. , van der marel , r. p. , mack , j. , leitherer , c. , sirianni , m. , & tosi , m. 2005 , , 631 , l45 aloisi , a. , et al . 2007 , , 667 , l151 marel , r. p. , mack , j. , leitherer , c. , sirianni , m. , & tosi , m. 2005 , , 631 , l45 annibali , f. , tosi , m. , monelli , m. , sirianni , m. , montegriffo , p. , aloisi , a. , & greggio , l. 2009 , , 138 , 169 annibali , f. , tosi , m. , aloisi , a. & van der marel , r. p. , 2011 , , submitted avni , y. 1976 , , 210 , 642 billett , o. h. , hunter , d. a. , & elmegreen , b. g. 2002 , , 123 , 1454 bomans , d. j. , van eymeren , j. , dettmar , r .- j . , weis , k. , & hopp , u. 2007 , , 51 , 141 cannon , j. m. , skillman , e. d. , garnett , d. r. , & dufour , r. j. 2002 , , 565 , 931 cervio , m. , & luridiana , v. 2004 , , 413 , 145 chandar , r. , leitherer , c. , tremonti , c. a. , calzetti , d. , aloisi , a. , meurer , g. r. , & de mello , d. 2005 , , 628 , 210 cignoni , m. , & tosi , m. 2010 , advances in astronomy , 2010 , davidson , k. , & kinman , t. d. 1985 , , 58 , 321 davidson , k. , kinman , t. d. , & friedman , s. d. 1989 , , 97 , 1591 dufour , r. j. , garnett , d. r. , & shields , g. a. 1988 , , 332 , 752 dufour , r. j. , & hester , j. j. 1990 , , 350 , 149 dufour , r. j. , esteban , c. , & castaneda , h. o. 1996a , , 471 , l87 dufour , r. j. , garnett , d. r. , skillman , e. d. , & shields , g. a. 1996b , from stars to galaxies : the impact of stellar physics on galaxy evolution , 98 , 358 fagotto , f. , bressan , a. , bertelli , g. , & chiosi , c. 1994 , , 104 , 365 fiorentino , g. , et al . 2010 , , 711 , 808 garnett , d. r. , skillman , e. d. , dufour , r. j. , & shields , g. a. 1997 , , 481 , 174 363 , 142 girardi , l. , et al . 2010 , , 724 , 1030 grimes , j. p. , et al . 2009 , , 181 , 272 holtzman , j. a. , burrows , c. j. , casertano , s. , hester , j. j. , trauger , j. t. , watson , a. m. , & worthey , g. 1995 , , 107 , 1065 hunter , d. a. , & thronson , h. a. , jr . 1995 , , 452 , 238 hunter , d. a. , oconnell , r. w. , gallagher , j. s. , & smecker - hane , t. a. 2000 , , 120 , 2383 izotov , y. i. , thuan , t. x. , & lipovetsky , v. a. 1994 , , 435 , 647 izotov , y. i. , & thuan , t. x. 1998 , , 497 , 227 izotov , y. i. , & thuan , t. x. 2004 , , 616 , 768 izotov , y. i. , guseva , n. g. , fricke , k. j. , & papaderos , p. 2009 , , 503 , 61 kunth , d. , lequeux , j. , sargent , w. l. w. , & viallefond , f. 1994 , , 282 , 709 larsen , s. s. , & richtler , t. 2000 , , 354 , 836 legrand , f. 2000 , , 44 , 345 lequeux , j. , peimbert , m. , rayo , j. f. , serrano , a. , & torres - peimbert , s. 1979 , , 80 , 155 lequeux , j. , & viallefond , f. 1980 , , 91 , 269 maz apellniz , j. 2009 , , 699 , 1938 marconi , g. , matteucci , f. , & tosi , m. 1994 , , 270 , 35 marconi , m. , et al . 2010 , , 713 , 615 marigo , p. , & girardi , l. 2007 , , 469 , 239 martin , c. l. 1996 , , 465 , 680 matteucci , f. , & tosi , m. 1985 , , 217 , 391 momany , y. , et al . 2005 , , 439 , 111 olive , k. a. , steigman , g. , & skillman , e. d. 1997 , , 483 , 788 origlia , l. , leitherer , c. , aloisi , a. , greggio , l. , & tosi , m. 2001 , , 122 , 815 pagel , b. e. j. , simonson , e. a. , terlevich , r. j. , & edmunds , m. g. 1992 , , 255 , 325 petrosian , a. r. , boulesteix , j. , comte , g. , kunth , d. , & lecoarer , e. 1997 , , 318 , 390 pilyugin , l. s. 1993 , , 277 , 42 sargent , w. l. w. , & searle , l. 1970 , , 162 , l155 searle , l. , & sargent , w. l. w. 1972 , , 173 , 25 searle , l. , sargent , w. l. w. , & bagnuolo , w. g. 1973 , , 179 , 427 silva - villa , e. , & larsen , s. s. 2011 , , 529 , a25 sirianni , m. , et al . 2005 , , 117 , 1049 skillman , e. d. , & kennicutt , r. c. , jr . 1993 , , 411 , 655 2005 , , 117 , 1049 stasiska , g. , & leitherer , c. 1996 , , 107 , 661 stetson , p. b. 1987 , , 99 , 191 stetson , p. b. 1994 , , 106 , 250 tolstoy , e. , hill , v. , & tosi , m. 2009 , , 47 , 371 tosi , m. , aloisi , a. , mack , j. , & maio , m. 2007 , iau symposium , 235 , 65 tosi , m. 2009 , iau symposium , 258 , 61 van der marel , r. p. , & cioni , m .- r . l. 2001 , , 122 , 1807 van zee , l. , westpfahl , d. , haynes , m. p. , & salzer , j. j. 1998 , , 115 , 1000 zwicky , f. 1966 , , 143 , 192

a quasi 2-dimensional recursive lattice formed by planar elements have been designed to investigate the surface thermodynamics of ising spin glass system with the aim to study the metastability of supercooled liquids and the ideal glass transition . the lattice is constructed as a hybrid of partial husimi lattice representing the bulk and 1d single bonds representing the surface . the recursive properties of the lattices were adopted to achieve exact calculations . the model has an anti - ferromagnetic interaction to give rise to an ordered phase identified as crystal , and a metastable solution representing the amorphous / metastable phase . interactions between particles farther away than the nearest neighbor distance are taken into consideration . free energy and entropy of the ideal crystal and supercooled liquid state of the model on the surface are calculated by the partial partition function . by analyzing the free energies and entropies of the crystal and supercooled liquid state , we are able to identify the melting transition and the second order ideal glass transition on the surface . the results show that due to the coordination number change , the transition temperature on the surface decreases significantly compared to the bulk system . our calculation agrees with experimental and simulation results on the thermodynamics of surfaces and thin films conducted by others .

1502.06562

glass transition on surface / interface / thin film has drawn intensive interests in the last two decades for two reasons . firstly , the importance of surface and thin film in materials science and engineering requires a better understanding of its dynamic and thermodynamic properties . secondly , the confined geometry is a good approach to understand the mysterious glass transition itself , especially to explore the dynamic properties within the thin film geometry . numerous works of both experimental and simulation / calculation approaches have been done [ 1 - 23 ] on glass transition on surface / thin film . keddie and co - workers firstly investigated the supported thin film of ps by ellipsometric measurements @xcite . they prepared several ps films supported by silicon wafers with thickness from @xmath0 to @xmath1 . the measurements indicated the deduction of the @xmath2 with thickness under 400@xmath3 . an empirical relationship between thickness and @xmath2 was given as : @xmath4 , \label{tg_reduction}\ ] ] where @xmath5 is the glass transition temperature of the bulk ps . the @xmath6 and @xmath7 are the parameters fitted to be @xmath8 and @xmath9 respectively , @xmath10 is the thickness of the film . following keddie s work , many researches have been done on different supported thin films of various polymers [ 2 - 3 ] by different characterization methods , such as x - ray reflectivity , positron annihilation , and dielectric @xcite . most results have demonstrated the same phenomenon that for liner polymer the @xmath2 decreases with the thickness of films . however the supported film has a considerable film - substrate interaction , which makes the conclusion controversial . strong attractive interaction between the substrate and thin film may increase the @xmath2 of thin film above the bulk @xmath2 [ 6 ] . van zanten et al . measured @xmath2 of poly-2-vinylpyridine on oxide - coated si substrates , and found it increase by @xmath11k than the bulk @xmath2 , for a @xmath12 film [ 7 ] . forrest and co - workers have done pioneer works in measuring the @xmath2 of free - standing thin films [ 8 - 10 ] . they measured the @xmath2 of free - standing ps films with thickness from @xmath13 to @xmath14 and different molecular weights by brillouin light scattering and transmission ellipsometry . their results showed that the @xmath2 decreases with the thickness of ps thin film with a much larger magnitude : for example , the @xmath2 of @xmath15 film with _ mw _ within @xmath16k to @xmath17k reduces by 70k below the @xmath5 , while this magnitude is around @xmath18k for supported films . the empirical equation ( eq.[tg_reduction ] ) derived for supported film still holds for the low _ mw _ free - standing films . with @xmath19 , the parameter @xmath20 was found to be @xmath21 which is twice of it found for supported films . other than the experimental work , computer simulation and calculation have also been developed to investigate the glass transition on surface / thin film , and most of them are for polymer systems [ 11 - 23 ] . molecular dynamics ( md ) and monte carlo ( mc ) method were usually employed with various modelings , and confirmed the @xmath2 decrease with the thickness reduction for both supported and free standing film , or @xmath2 increase in some particular substrate - film cases . the eq.[tg_reduction ] derived from experiments can also be validated by simulations , nevertheless the explanation for the mechanism of @xmath2 reduction is still a matter of debate . most md simulations verified the experimental observation that for supported film , a strong substrate - film interaction will increase the @xmath2 above the value in the bulk , while the weak substrate - film interaction will lead @xmath2 reduction , and free - standing film shows a much larger @xmath2 reduction than supported film with weak substrate - film interaction [ 12 ] . mattice and co - workers firstly reported the mc simulation in this field [ 13 , 14 ] . de pablo et.al . reported a mc simulation on free - standing films of both linear and cyclic polymeric chains [ 23 ] . basically the @xmath2 reduction with smaller thickness was confirmed in the above investigations . in de pablo and co - workers work , the relation between @xmath2 and @xmath5 can also be fitted to be eq . [ tg_reduction ] with slight difference on the parameters fitting . although the glass transition is not a unique property of polymers , most simulation works were modeled for polymer systems , while the works on small molecule systems are very rare . meanwhile , ising spins glass has also been widely utilized to study the glass transition [ 24 - 37 ] . by different lattices adopted and interactions setup the ising model is capable to describe various systems , such as gas , liquid , crystal and glass , and consequently the phase transitions , like melting and glass transition [ 24 ] . however , very few of the efforts were on surface / thin film glass transition of ising spin glass . this field had been explored by gujrati et al . by applying ising model on a modified bethe lattice to describe the thermodynamics of polymer systems near surface [ 38 - 42 ] . in this paper we follow the similar method to study the glass transition of ising spin glass on the surface / thin film on a specially constructed recursive lattice . except in some rare cases [ 43 - 45 ] , a many - body system with interactions on a regular lattice is difficult to be solved exactly because of the complexity involved with treating the combinatorics generated by the interaction terms in the hamiltonian when summing over all states . the mean - field approximation is commonly adopted to solve this problem , e.g. the flory model of semiflexible polymers [ 46 , 47 ] . on the other hand , recursive lattices enable us to take the explicit treatment of combinatorics on these lattices and no approximation is necessary [ 25 - 28 , 48 ] . the recursive lattice is chosen to have the same coordination number as the regular lattice it is designed to describe . as usual , the coordination number @xmath22 is the number of nearest - neighbor sites of a site . a typical recursive lattices , the husimi lattice , as the analogs of the 2-d lattice is shown in fig . [ fig1 ] . for the bulk system , the recursive lattice calculations have been demonstrated to be highly reliable approximations to regular lattices [ 48 - 50 ] . [ ptb ] in this work we construct a recursive lattice to describe the surface / thin film . this surface recusive lattice ( srl ) is to mimic the 2d case , i.e. the 2d bulk with 1d surfaces . the srl is integrated of square units representing the bulk and single bond representing the surface , the structure is made to have the same coordination numbers with regular 2d square lattice . ising spins will be applied on the lattices to represent a small molecule system . the exact calculation will be drived and the solutions will be discussed . to approximate the regular lattice due to the same coordination number , we firstly need to figure out the coordination number and interactions of a regular lattice of surface / thin film to construct the srl to represent the surface / thin film . the fig . [ fig2 ] shows a regular 2d square lattice of a thin film with thickness equals to 5 and the thick line is the surface . ( in all figures of this paper the thick bond represents surface bond . ) here we label the central layer to be the @xmath23th level , and the layer next to @xmath23th layer is the level @xmath24st . the surface layer is labeled as level @xmath25 . the sites on the surface have coordination number of @xmath26 , while inside the bulk the coordination number is @xmath27 . that is , a hybrid recursive lattice with @xmath28 and @xmath27 is required to describe the surface / thin film . [ ptb ] , scaledwidth=80.0% ] now we take out the basic units inside the bulk and on the surface ( fig . [ fig3 ] ) to construct a recursive lattice . inside the bulk we simply have husimi lattice of @xmath29 . for the surface structure , a single bond representing in the surface unit links on a square unit then we can have sites with @xmath28 . however this unit can not be simply adopted to construct a recursive structure , because the recursive calculation technique ( will be discussed later ) requires an origin point to which the entire tree is symmetrical . for the unit shown in fig . [ fig4]a , wherever we determine the origin is ( point a or b ) , the unit is not symmetrical to that point . we modify the surface unit by replacing one square by an artificial single bond at one lower corner . although this unit differs from the regular lattice we want to approximate , the coordination numbers on the surface square still accord to our design and the calculation in the following sections will show that this approximation is practical . the modification is shown in fig . [ fig4 ] , the modified unit ( fig . [ fig4]b ) is then symmetrical to the point a. [ ptb ] [ ptb ] then we can put the bulk and surface units together to construct a recursive lattice to approximate the regular thin film lattice . the structure of a thickness @xmath30 lattice is shown in fig . [ fig5 ] . [ ptb ] from surface a to b , we have a finite bulk portion with a thickness @xmath30 , and these five layers are labeled as @xmath25 , @xmath24 , and @xmath23 like the regular film lattice in fig . [ fig2 ] . the thick single bond representing surface links two identical bulk trees . in this example , one bulk tree _ _ t__@xmath31 is linked overall with another @xmath32 trees ( only one bulk tree is drawn in above fig . [ fig5 ] ) , while they are all identical and independent with each other except a single surface bond connection . recursively , each of these @xmath32 trees is also linked with another @xmath33 trees by single surface bonds . this lattice is an infinite tree integrated by the finite size bulk portions and infinitely large surfaces . if we look at the surface integrated by thick surface bonds indicated by the arrow at the right lower corner in fig . [ fig5 ] , we can see the surface going through the bulk trees _ _ t__@xmath31 is infinitely large with a coordination number 3 everywhere . a stretched surface is shown in fig . [ fig6 ] to provide a more obvious view of the surface . comparing to the regular lattice surface , the main difference is that one surface also receives thermodynamic contributions from other surface structures , which is caused by the surface unit modification . but this is an approximation we have to take to achieve the recursive calculation . although we have infinite number of bulk trees and surfaces in this lattice , since they are independent and identical , it does not impact the thermodynamic properties of a local region we are going to investigate . [ ptb ] we label each site in a square as shown in fig . [ fig7 ] : [ ptb ] the site in a square unit is labeled as _ _ s__@xmath34 , where @xmath35 is the index of the square unit in the bulk tree , @xmath36 is the index of site . therefore , one site _ _ s__@xmath34 actually has two labels depending on which square unit we refer to . for example _ s__@xmath37 is also _ _ s__@xmath38 in the higher level unit . when we are only focusing on one unit , the site _ _ s__@xmath34 can be denoted as _ _ s__@xmath39 for convenience . in a square unit , the base site is determined to be the site closest to the bulk tree origin ( the @xmath23th square ) and labeled as _ _ s__@xmath40 . therefore , in the upper enlarged circle in fig . [ fig7 ] , the base site is the lowest site , while in the lower enlarged circle , the base site is the left one . the labeling in the origin unit is a special case which will be discussed later . the interactions are also sampled in the enlarged circle : @xmath41 is the interaction between the nearest sites , @xmath42 is the interaction between the second - nearest sites , i.e. the diagonal sites . for surface calculation , we label the surface sites as shown in fig . due to our calculation method , only two labels are necessary in surface labeling . the site close to the bulk site is labeled as @xmath43 , and the site being diagonal to the bulk site is labeled as @xmath44 . [ ptb ] 1.3 ) the interactions and energy in bulk square unit we consider four kinds of interactions in our model : @xmath41 the interaction energy between the nearest sites , @xmath42 the interaction energy between the second nearest sites , @xmath45 the interaction energy of three sites ( triplet ) , and @xmath46 the interaction energy of four sites ( quadruplet ) . we introduce the following variables to count the interactions and magnetic fields of one square unit : @xmath47 where @xmath48 is an ising spin and has the value of @xmath49 or @xmath50 . note that the base site @xmath51 is not included in the magnetic term @xmath52 , because the calculation has a recursive fashion that the energy of one unit is included in the contribution to the next level s unit , and the base site of one unit will be counted as the top site in the next level s energy equation , thus we need to exclude it here to avoid the double counting . then , for a particular cell @xmath20 with a certain spin conformation @xmath53 , the energy of the ising model in one cell is : @xmath54 the total energy of the ising model on srl is the sum of the energy of all cells : @xmath55 in this work we set @xmath41 to be negative to adopt the the anti - ferromagnetic ising model , the neighbor spins preferring different states corresponds to the lowest energy stable state , i.e. the ideal ordered crystal conformation , which have the largest weight comparing to other conformations at the same temperature , while the neighbor spins of the same states are the most unstable conformations with the highest energy , which represent the disordered phase . then the boltzmann weight @xmath56 of state @xmath53 is given by @xmath57 , \label{weights}\ ] ] where @xmath58 @xmath59 is the inverse temperature , and here we set the boltzmann constant @xmath60 to make the temperature to be generalized in the unit of energy . then the partition function of a finite lattice with @xmath61 cells is the sum over products over the boltzmann weight : @xmath62 . \label{pf}\ ] ] in fig . [ fig6 ] , the single bond unit and the square unit alternatively appear on the surface structure . the interactions and energy of the surface square unit is similar to the bulk square unit as we discussed in the previous section . however , here we would like to assign a different nearest neighbor interaction @xmath63 on the surface bond , to distinguish it with @xmath41 in the bulk . this enables us to investigate more complex surface properties . and similarly we may also have a different diagonal interaction ( @xmath64 ) , triplet interaction ( @xmath65 ) , quadruplet interaction ( @xmath66 ) and magnetic field ( @xmath67 ) : @xmath68 on the surface single bond unit , the interaction is much simpler since there is only one nearest neighbor interaction : @xmath69 similar to the role of base site in bulk calculation , here the magnetic field of @xmath70 is not included in the second term because that site will be counted in the next level s energy equation . the ` level ' , in this statement , is indexed by taking the direction indicated by the arrow in fig . [ fig6 ] , and employing an imaginary origin point infinitely far away from the region we concern . since the surface is infinitely large , the selection of ` origin point ' does not affect us to achieve the solutions on surface . by the setup of interaction energy parameters , we can simulate various systems with particular interactions and energy to study their thermodynamic properties and phase transition with the determination of the partition function . the effect of energy parameters will be discussed in section iv . we will generally set @xmath71 to determine the temperature scale for our antiferromagnetic model . the solution based on @xmath71 and all other parameters to be @xmath23 is called the _ reference model_. to discuss the calculation of our model , we first need to introduce the concept of sub - tree contributions . in the finite bulk tree we have an origin at the center of the tree . for each square unit , the base site is the closest site to the origin , and there are three sub - trees coming from the other three sites . the sub - trees could either be three identical portions of the bulk tree , or three surface trees linked by the single surface bond if it is the square unit on the surface . we label the levels in one unit and show the sub - tree contributions in fig . [ fig9 ] . [ ptb ] by this index , the sites on the level @xmath72 is represented as @xmath48 . the sub - tree going to the site on level @xmath72 is marked as _ _ t__@xmath73 . in this way , we can introduce the partial partition function ( ppf ) @xmath74 at the level @xmath75 to represent the contribution of the branch _ _ t__@xmath76 with a certain state of spin @xmath77 as its base site to the lower level @xmath72 , and _ _ t__@xmath73 to the lower level @xmath78 , etc . therefore the ppf @xmath79 at level @xmath72 is a function of the ppfs @xmath74 and @xmath80 at higher level @xmath75 and @xmath81 and the local weight . then we can start from the highest level , count the contributions of sub - trees on each level and recursively go to the lower level unit the origin point , where it gains the contribution of the entire lattice , and the thermodynamics of the entire system can be obtained by the partition function . the ppf of a sub - tree _ _ t__@xmath73 @xmath82 is the sum of the configurations with the spin @xmath83 over the products of the ppfs on higher levels with the local weight @xmath84 . for a square unit , @xmath85 has @xmath86 terms with @xmath87 and @xmath88 is the sum of the other @xmath86 configurations with @xmath89 : @xmath90z_{m+2}(s_{m+2})w(\gamma_{s_{m}=1 } ) , \label{ppf+_recursion}\\ z_{m}(- ) & = \underset{\gamma=9}{\overset{8}{\sum}}[\underset{\{m+1\}}{\overset{2}{{\displaystyle\prod } } } z_{m+1}(s_{m+1})]z_{m+2}(s_{m+2})w(\gamma_{s_{m}=-1 } ) , \label{ppf-_recursion}\ ] ] depending on the spin state ( @xmath91 ) of the origin site , the two sub - trees contributions to the whole system are : @xmath92 where the magnetic term is to count the contribution of origin site which is not contained in eq.[e_alpha ] . the partition function @xmath93 of the entire lattice can be accounted by the contributions to the origin site : @xmath94 the first term is to count the conformation with the origin site spin @xmath95 , the square of ppf represents two sub - trees and the weight of the origin spin itself is counted as @xmath96 , similiarly for the @xmath97 situation in the second term . or , if the origin is defined to be a square unit , the partition function is @xmath98 where @xmath99 is one of the four spins on the origin square , @xmath100 is the ppf of the sub - tree coming into the site @xmath99 , and @xmath101 is the local weight of the origin square . then we can introduce the ratios @xmath102 these two are the ratios of ppfs on the level @xmath72 ; they indicate , although not exactly to be , the probability that the site @xmath48 on level @xmath72 is occupied by a plus spin ( @xmath103 ) or a minus spin ( @xmath104 ) . therefore , these two ratios , which we call the solutions of the lattice , are critical in our model and all the thermodynamic calculations are based on these solutions . define a compact notation for convenience @xmath105{c}x_{m}\text { if } s_{m}=+1\\ y_{m}\text { if } s_{m}=-1 \end{array } \right . , \label{general ratio}\ ] ] by introducing@xmath106 we then have @xmath107 and @xmath108 . substituting these two equations into eq.[ppf+_recursion ] and [ ppf-_recursion ] gives @xmath109z_{m+2}(s_{m+2})w(\gamma),\ ] ] where the sum is over @xmath110 for @xmath87 , and over @xmath111 for @xmath89 . it is clear that the solution ( ratios ) on one level is a function of the solutions on higher levels in a recursive fashion . for convenience , we introduce the polynomials : @xmath112z_{m+2}(s_{m+2})w(\gamma),\label{polynomial q}\\ q_{m-}(y_{m+1},y_{m+2 } ) & = \underset{\gamma=9}{\overset{8}{\sum}}[\underset{\{m+1\}}{\overset{2}{{\displaystyle\prod } } } z_{m+1}(s_{m+1})]z_{m+2}(s_{m+2})w(\gamma),\\ q_{m}(x_{m+1},x_{m+2 } ) & = q_{m+}(x_{m+1},x_{m+2})+q_{m-}(y_{m+1},y_{m+2})=\frac{b_{m}}{b_{m+1}^{2}b_{m+2}}.\end{aligned}\ ] ] then the ratio @xmath103 is just as function of the ratios on higher level @xmath113 and @xmath114 : @xmath115 from the recursive relation shown in eq.[ratios_polynomial_x ] , we may expect a repeating solution , which has the form : @xmath116 and @xmath117 for some value of @xmath118 , so that eq.[ratios_polynomial_x ] holds . such a solution can be called a @xmath119-cycle solution that repeats after the @xmath119-th application of the recursive relation equation . for example , for a 1-cycle solution @xmath120 , we simply have the same ratio @xmath120 on all the levels @xmath121 for a 2-cycle solution @xmath122 and @xmath123 we have @xmath124 that is , the two solutions alternatively appear level by level . in this way , for a particular recursive relation , we can start from a set of initial seeds @xmath122 and @xmath123 to calculate the solutions on lower levels until we reach the recursively repeating solutions , which is called fix - point solution . usually many initial seeds are tried to obtain all the possible solutions for a particular recursive relation . according to our experience , in the lattices discussed in this work , solutions with @xmath125 were never obtained , while solutions with @xmath126 and @xmath127 ( @xmath24-cycle and @xmath128-cycle solutions ) are almost always available . based on the property of @xmath120 , which determines the probability that one site is occupied by the plus spin , a cycle solutions with @xmath125 is hard to imagine . an example 1&2-cycle solutions of the husimi lattice , with @xmath71 and other parameters as @xmath23 on a wide temperature range is shown in fig . [ fig10 ] . [ ptb ] and other parameters as @xmath129,scaledwidth=80.0% ] in fig . [ fig10 ] , at high temperature both @xmath24-cycle and @xmath128-cycle solutions are @xmath130 and every site in the lattice has a @xmath131 probability to be occupied by a + spin . this obviously represents a disordered state . at low temperature , the @xmath24-cycle solution is still @xmath130 , while the @xmath128-cycle solution occurs a transition and gives two branches , one of which goes to @xmath24 and the other goes to @xmath23 with temperature decrease . this indicates , for @xmath128-cycle solution , that two neighbor sites will prefer to have different spin states . in the region close to zero temperature , if one site has @xmath132 ( @xmath133 ) probability to be occupied by @xmath134 spin , then its neighbor sites will have @xmath135 ( @xmath136 ) probability to be occupied by @xmath134 spin ( i.e. @xmath132 probability to be occupied by @xmath137 spin ) . therefore , at zero temperature we will have a plus and minus spins alternative arrangement on the lattice . recalling the anti - ferromagnetic model we employed , this @xmath128-cycle arrangement has the lowest energy ( the most stable state ) and corresponds to the ordered state ( crystal ) . while the @xmath24-cycle solution at the same temperature refers to the metastable state , that is , it is still a stable solution however with a higher energy . the temperature where the @xmath128-cycle solution appears is called the critical temperature , or the melting temperature @xmath138 . at this temperature the amorphous state turns to be the ordered state ( crystallization ) or it can continue as a metastable state , the analog of three states here with liquid , crystal and supercooled liquid implies that it is the melting transition . the thermodynamic details will be discussed in section iii . for an infinite husimi lattice , the calculation to get the fix - point solution is straightforward [ 25 , 37 ] , we can simply start with two artificial initial seeds and recursively calculate the solutions by eq . [ ratios_polynomial_x ] many times until we reach the fix - point solution . while for the srl we developed in this paper , the bulk tree is finite and confined within surface trees , we need to carefully monitor the calculation on each specific site . let us take a srl with thickness of @xmath139 , we label the surface square as @xmath25 , then the square next to @xmath25 is labeled as the @xmath72-th level , then origin square is labeled as @xmath23 . firstly we start from two initial guesses @xmath140 @xmath141 on the surface sites as shown in fig . [ fig11 ] . the calculation through the surface square @xmath25 provides the solution @xmath113 on the top site of level @xmath72-th square . on the other two surface squares labeled as @xmath142 and @xmath143 we use the same initial guess seeds again , however rotate their positions by putting @xmath140 on the top site and @xmath141 on two side sites , then the calculation based on @xmath142 or @xmath143 will give us the solution @xmath144 and @xmath145 on the side sites of level @xmath72-th square . this is because for the anti - ferromagnetic ising model and the properties of @xmath128-cycle solution introduced above , we expect a @xmath128-cycle style solution on our lattice ( and the @xmath24-cycle solution is just a special case when the two @xmath128-cycle solutions are the same ) , thus the rotation avoids us to get the same results on neighbor sites , which makes the @xmath128-cycle solution impossible . [ ptb ] consequently , with the solutions @xmath114 and @xmath113 we obtain @xmath103 on the top site of square level @xmath78 , and on the other branch the rotation of @xmath114 and @xmath113 will give us @xmath146 , then we can continue the calculation to the origin square @xmath23 as shown in fig . [ fig12](a ) . once we reach the solution @xmath122 and @xmath147 at the origin square , we have the states inside the bulk if the solution is the fix - point one . this part of calculation is called downward calculation , which is from the surface to the bulk origin . however , this set of bulk solutions is from the initial guesses on the surface ; it may not be the solution of the real state . we still have to find the fix - point solution on the surface , and the bulk calculation from the surface fix - point solution is then the final result we are looking for . to find the fix - point on the surface , we start from the bulk origin and trace back to the surface ( upward calculation ) , which provides a bulk contribution _ _ t__@xmath148 to the recursive calculation on the surface . [ ptb ] the upward calculation is shown in fig . [ fig12](b ) . at the origin square , from the contributions of three identical trees and the local weight of the origin square , we can calculate the solution on the lower site , which is labeled as @xmath149 . ( index u stands for upward ) . then from @xmath150 , @xmath151 and @xmath152 we can get @xmath153 and so on . the upward calculation will provide the solution @xmath154 at the base site of the surface square unit . this solution will be used as @xmath155 representing the bulk effect to surface in the surface calculation . the situation on surface is more complicated than in the bulk . from the previous introduction on downward and upward calculation , we can see a hint that depending on the direction of calculation , the solutions might be different on the same site , for example the @xmath103 and @xmath156 are different even they are on symmetric sites to the origin . this direction issue does not affect us to explore the solutions describing the bulk , however it is critical in surface calculation . therefore , we firstly classify two directions on the surface with specific labeling , as shown in fig . [ fig13 ] : [ ptb ] in fig . [ fig13 ] , the labeling on the surface is in this way : the solution on the base site is @xmath155 , which is the @xmath154 from upward calculation ; the solution on the sides of the square is label as @xmath140 , and on the top is @xmath141 ; the prime marks the direction that goes out from the square , while the label without prime is the direction going into the square . the scheme a shows a side - to - top direction : the solution going out from the top site , @xmath157 , is calculated from the solution going into the side site @xmath140 and the base site solution @xmath155 . then from @xmath157 we go through a single bond to get the next @xmath140 . the scheme b shows a side&top - to - side direction : the solution going out from the side site , @xmath158 , is calculated from the solution going into the side site @xmath140 , the solution going into the top site @xmath141 , and the base site solution @xmath155 . again from @xmath158 we then go through a single bond to get the next @xmath141 . only after we have done the calculations in both directions , we can get the surface solutions pair , the solution going into the top site @xmath141 , and the solution going into the side site @xmath140 , to do the subsequent bulk calculation . ( note that initially this pair is a guess we made to do the first iteration s bulk calculation ) . the surface calculation scheme is shown in fig . [ fig14 ] . the upward bulk calculation provides us the bulk tree contribution and the solution @xmath155 on the base site of surface square . this solution can be taken as a constant in the surface recursive calculation . in fig . [ fig14 ] , we start from the @xmath155 and the initial seed @xmath140 to do two calculations as mentioned . the recursive calculation in process a will eventually give us a fixed @xmath140 , while the process b will give us a fixed @xmath141 . this new fixed set of @xmath140 and @xmath141 is then the seeds we are going to use for the next iteration s bulk calculation . note one difference in process a and b is that in process a we only need @xmath140 for calculation , which is updated step by step , however in process b after each step we will only have a updated @xmath158 , while @xmath140 is a constant in calculation . thus , in practice we always do the process a first to get a fixed @xmath140 , then use this @xmath140 as a constant in the calculation of process b. [ ptb ] the calculation on the square unit , for example , @xmath159 in process b , is the same as the bulk calculation . the calculation through the single surface bond , for example , @xmath160 in process b , is also similar with the square . the difference is that the ppf only has two terms since there are @xmath27 configurations of a single bond structure : @xmath161 where the sum is over @xmath162 and @xmath128 for @xmath87 , and over @xmath163 and @xmath27 for @xmath89 . then we have the polynomials : @xmath164 the ratio is the function of the ratio on higher level : @xmath165 the calculation scheme to reach the fix - point solutions both on the surface and in the bulk is shown in fig . [ fig15 ] . we do an embedded recursive calculations to get the final solutions on the surface and in the bulk . the @xmath166 iterations give us recursively updated solutions in the bulk until we find the fix - point solution , and a recursive calculation along the surface is embedded in one iteration to take the effect of updated bulk contributions each time . this process is to make sure that we counter the mutual effects from surface to bulk and from bulk to surface . another necessity of this complicated process is that the bulk tree has a finite size thus the fix - point solution is not guaranteed to be reached during one downward calculation . for the structure with thickness @xmath167 , we can always obtain the fix - point solution at bulk origin , although on the first several layers close to surface we can only have numerical calculations instead of exact solutions . ( this ` numerical depth ' depends on the energy parameters setting and it shows always being smaller than @xmath168 according to our experience . ) [ ptb ] since our lattices are infinitely large , it makes no sense to calculate the partition function or free energy for the entire lattice . however an exact treatment called gujrati trick has been well developed to deal with the thermodynamic calculation on recursive lattices in our group s previous work @xcite . by this technique we can approach the thermal properties in a local area ( per site ) by the partial partition functions @xmath169 and solutions @xmath120 we discussed in the previous section . the helmholtz free energy is a function of the temperature and partition function @xmath170 : @xmath171 [ ptb ] here we take a regular husimi lattice as an example to introduce the thermal calculation derived by figuring out the partition functions . recall eq . [ pf_of_ppfs ] , since @xmath93 counts the contribution of the whole system , it must be the sum of contributions of the six sub - trees on level @xmath24 and @xmath128 , and the weight of the two local square units on the origin . if we cut off the two sub - branches on level @xmath128 and joint them together , we will have an identical but smaller lattice . similarly the partition function of this smaller lattice is : @xmath172 we also cut off the 4 sub - branches on level 1 and form two smaller lattices and their partition function will be @xmath173 as an extensive quantity the free energy of the whole system is the sum of the free energies of these three smaller lattices and the two local squares : @xmath174 this yields @xmath175 as the free energy of the two local sqaures , i.e. four sites ( three paris of half - sites and the origin site ) . the free energy per site is : @xmath176 by substituting @xmath177 and @xmath178 we have @xmath179}{b_{1}^{4}[x_{1}^{2}e^{\beta h}+(1-x_{1})^{2}e^{-\beta h}]^{2}b_{2}^{2}[x_{2}^{2}e^{\beta h}+(1-x_{2})^{2}e^{-\beta h } ] } ) \label{freeenergypolynomial / site}\ ] ] recall that for either 1-cycle or 2-cycle fix - point solutions we have @xmath180 , and @xmath181 it follows @xmath182 with the calculation of @xmath183 and fix - point solution @xmath184 and @xmath122 we discussed in the previous section , the free energy can be easily achieved . for a recursive structure with an origin point , we can always do this cut and rearrange trick to obtain the partition function and free energy per site around the origin point . the entropy is the first derivative of the free energy with respect to the temperature : @xmath185 with the free energy and entropy we have the energy per site ( energy density ) as @xmath186 a typical thermodynamic behavior of a husimi square lattice with @xmath71 and all other parameters to be @xmath23 is shown in fig . [ fig17 ] : [ ptb ] and all other parameters to be @xmath129,scaledwidth=80.0% ] in fig . [ fig17 ] , the free energy of two solutions are the same at high temperature . at the melting temperature @xmath138 the 1-cycle s free energy differs from 2-cycle s and is less stable . the continuous entropy at @xmath138 of 1-cycle implies the supercooled liquid state , while the entropy decrease of 2-cycle state implies the crystallization . although here the entropy decrease is not a discontinuous jump as the typical behavior of first order melting transition , we may still treat it as the phases - coexisting point since the 2-cycle state is the most stable state here . the thermodynamics of 1 and 2-cycle solutions agree with our anti - ferromagnetic model . the anti - ferromagnetic ising model prefer to anti - aligned neighbor spins , therefore the 2-cycle state should have the lowest energy , i.e. the crystal , while the 1-cycle solution represents the metastable state with higher energy . with the continuing decrease of temperature we can observe that at @xmath187 the entropy of 1-cycle state ( supercooled liquid ) quickly decreases to zero and becomes negative . this is the kauzmann s paradox and the ideal glass transition . the negative entropy is unphysical thus the metastable state must undergo a transition to be the glass state at @xmath188 , the ideal glass transition temperature . the free energy of the glass state below glass transition temperature is indicated by the last legend in fig . [ fig17 ] . note that this branch is not provided by the calculation , instead it is a theoretical expectation . we can modify our calculation to achieve a stable solution of this glass state branch , but since this part is not our interest we have not done that calculation in this paper . we are following the gujrati trick to solve the free energy and consequent thermal functions of the srl . although the basic principle is the same , the asymmetrical structure of the surface lattice requires further complex tricks to do the calculation . if we take a random site on the surface as the origin , to do the cut and re - joint trick we need to select a local area around the origin , and find matching sub - trees contributing to the local area . here the matchings are more specific than they are in a homogeneous bulk lattice . for example in the husimi lattice discussed in previous section , all the sites are identical , thus if two sites have the same solution @xmath120 on them , the sub - trees cut off from them can be joined together to make a new lattice . however , the sites on the surface of srl have three different situations : on the top of a surface square where the sub - tree s contribution going out , on the side of a surface square where the sub - tree s contribution coming into , and on the bottom of a surface square where the bulk tree s contribution coming into . the matchings must be done on sites with exactly the same situations to make an identical but small lattice , but the asymmetrical structure of srl makes the cutting and matching impossible . wherever we select the origin local area and cut the sub - trees , there will always be two sub - trees left and can not be matched with each another . [ ptb ] fig . [ fig18 ] shows two failure cutting schemes , with 1 square and 2 single bonds area and 2 squares and 4 single bonds area respectively . in the cutting scheme # 1 , by removing the local area surrounded by the discontinuous curves , we have four sub - tree at the point a , a , b and b. the two sub - trees on a and a can be linked together to form a smaller lattice with exactly the same structure . while the sub - trees on b and b do not match with each other , since they are a bulk sub - tree on b and a surface sub - tree contributing the top site of a surface square . in the cutting scheme # 2 , the local area of 2 squares and 4 single bonds is bounded by the solid curves . the sub - trees on c - c and d - d can form smaller lattices . although the new formation of d - d is not identical to the entire lattice , we can still count its partition function and handle the ratio calculation . however the same difficulty , as we encountered in scheme # 1 , presents on the sites e and e. the e site is at the side of a surface square while the e site is on the top . therefore , either we choose odd or even numbers of basic units as the local area around origin , the gujrati trick can not be done . we have to somehow modify the origin structure to avoid the problem of asymmetry . as shown in fig . [ fig13 ] , the single bond and square unit have a head - to - side connection , that is , following the calculation direction , the sub - tree coming from a single bond is always linked to the side site of next level s square , or vice versa . this arrangement is designed to make the structure uniform on the surface , but it also causes the asymmetry . it is necessary to make one head - to - head connection as the origin of the surface , as shown in fig . [ fig19 ] . [ ptb ] in this way , we have two infinite and identical surface trees connected in a head - to - head style , this single bond is then the origin of our entire lattice , and except for this bond , all other surface single bonds are in head - to - side connection . starting from the origin , the closest squares can be labeled as level @xmath24 ; the second closest squares are on level @xmath128 and so on . as shown in fig . [ fig19 ] , by the selection of four squares and six single bonds as the local area , we can form one identical lattice on level @xmath24 at the sites a - a , two identical lattices on level two at the pairs b - b and c - c. the sub - trees cut at d , d , e and e are identical bulk trees so we can pair them to form two bulk lattices . now we can easily derive the free energy of the local area to be : @xmath189 unlike the point origin in husimi lattice , here the origin is a single bond unit . the total partition function thus has four terms due to the four possible states of the single bond : @xmath190+z_{0}^{\text{u}}(-)z_{0}^{\text{l}}(-)\exp[\beta(\overline{j}-2\overline { h } ) ] \label{pf_of_u_and_l_ppfs}\ ] ] to specifically track the spins in the origin area and rematch the sub - trees we define the two sub - trees meeting at the origin bond as upper and lower parts . in equation [ pf_of_u_and_l_ppfs ] , the superscript u or l is to indicate the upper or lower part of the lattice . similar integration can be employed to form the partition functions of smaller lattice we rearranged in eq.[fe_local_surface ] . the ratio of partition functions can be represented by the solutions and polynomials which have been achieved in section ii . however in section ii , we go along the surface and calculate the solution on each site and the polynomial ratios step by step , these detailed solutions are unnecessary here and make it complicated to determine the eq.[fe_local_surface ] . hence we redefine the basic unit as one square and two single bonds linked on its side as shown in fig . [ fig20 ] . [ ptb ] by the selection of this rectangle basic unit , in the upper or lower branch we simply have two units contributing to the sides of next unit , and recursively the whole branch contributing to the origin bond , as indicated by the arrows in the lower part in fig . [ fig20 ] . the contribution of bulk tree can be treated as a constant . the calculation based on this selection will provide only one solution as @xmath191 or @xmath192 on the output site , regardless of it is 1-cycle and 2-cycle solutions on the surface . this single solution is sufficient for us to determine the eq.[fe_local_surface ] . for instance , if we look at the smaller lattices rearranged on points a - a in fig . [ fig20 ] , we will only need the solution @xmath191 and @xmath192 because they are on the output sites a and a. similarly , for either @xmath191 or @xmath192 , recall the eq.[general ratio ] and @xmath107 , @xmath108 with @xmath193 $ ] , then we have @xmath194z_{\text{b}}(s_{\text{b}})\omega(\gamma),\ ] ] where the @xmath195 is the index of the site linked to the bulk tree . with six spins the rectangle unit has @xmath196 possible configurations thus we can introduce the polynomials : @xmath197z_{\text{b}}(s_{\text{b}})\omega(\gamma),\ ] ] @xmath198z_{\text{b}}(s_{\text{b}})\omega(\gamma),\ ] ] @xmath199 then we can obtain the 1-cycle recursive relation : @xmath200 and the fix - point solution is the one cycle solution @xmath191 or @xmath192 , although the actual solution site by site on the surface could be either 1-cycle or 2-cycle . it is important to clarify that the solution @xmath191 or @xmath192 are exactly the same to the solutions we achieved on the surface in section ii . the reason we re - select the basic unit is to obtain a simple polynomial @xmath201 for the calculation in equation ( 3.13 ) and ( 3.14 ) . now we may rewrite eq.[fe_local_surface ] with @xmath202 @xmath203 . \end{split}\ ] ] this local area contains @xmath204 sites , thus the averaged free energy on each site is @xmath205 an reference free energy behaviors of 1-cycle and 2-cycle solution of srl with thickness = 19 , neighbor interaction @xmath71 , surface neighbor interaction @xmath206 and all other parameters to be zero is shown in fig . [ fig21 ] [ ptb ] , @xmath71 , @xmath206 and all other parameters to be zero , scaledwidth=80.0% ] an interesting phenomenon can be observed from fig . the free energy of 2-cycle solution ( crystal state ) is higher than the free energy of metastable state between t = @xmath207 to @xmath208 . this is different from the bulk behavior in fig . [ fig21 ] . the cross point of 1 and 2-cycle s free energy at lower temperature can be determined to be the melting transition . above the melting temperature , our results indicate that the anti - aligned spins arrangement is less stable than the @xmath130 solution . this behavior has not been investigated , yet it is not clear whether it is simply unphysical or it implies a super - heated crystal state . below the melting temperature the behavior is similar to the bulk system . the 2-cycle solution represents the crystal state with a lower free energy and continues to decrease to the free energy of ideal crystal . while the free energy of 1-cycle solution , the metastable state , will reach a minimum point than bind back , this unphysical behavior implies the ideal glass transition . with equations [ entropy ] and [ energy ] we can easily calculate the energy density and entropy from the free energy . the fig . [ fig22 ] shows the entropy derived from the free energy in fig . [ fig21 ] . in fig . [ fig22 ] the black arrow on the entropy show the melting transition at the cross point of free energies ( @xmath209 ) , where the entropy of 1-cycle solution will step to 2-cycle solution s entropy as a first order transition . the ideal glass transition temperature @xmath188 can be clearly observed by the negative entropy of 1-cycle solution . the detailed discussion on surface thermodynamics and the surface effect comparing with bulk system will be presented in the next section . [ ptb ] , @xmath206 and all other parameters to be zero.,scaledwidth=80.0% ] as introduced in previous chapters , the thermodynamic calculations are mainly based on the solutions @xmath120 , i.e. the ratio of partial partition functions on srl . the 1 and 2-cycle solutions of a simple anti - magnetic field husimi case is presented in fig . [ fig10 ] , and we have discussed how the melting transition , crystal state and metastable state can be indicated from the solutions . the reference 2-cycle solutions of srl with @xmath41 and @xmath210 , other parameters as @xmath23 , and thickness = @xmath168 are shown in fig . [ fig23 ] . [ ptb ] the solutions of bulk husimi lattice from fig . [ fig10 ] are also included in fig . [ fig23 ] for comparison . with the temperature increase , the solutions on surface converge to @xmath130 at lower temperature than the husimi bulk solutions . this indicates a lower melting transition temperature on the surface , which is easy to understand with smaller coordination number and less interactions on the surface , the spins on the surface are easier to be anti - aligned with less energy . also , unlike the symmetrical solutions we usually have , the 2-cycle solution on the surface is asymmetric this is because of the hybrid structure on the surface , which defines two different circumstances on the surface square unit and single bond unit respectively . naturally the spins on the surface square unit has the solution closer to the bulk solution , while the spins on the single bond unit has a less stable ( closer to @xmath130 ) solution due to the lower dimension . depending on the initial seeds adopted for surface calculation , we may also have the other set of surface solutions symmetric to the one in [ fig23 ] . the 2-cycle bulk solutions @xmath211 and @xmath212 are the solutions at the bulk origin ( fig . [ fig15 ] ) . since the husimi lattice plays the bulk portion in srl , we can expect the bulk origin solution to be identical to the solutions of husimi lattice if the thickness is large enough to ignore the effects of surface to bulk , and this expectation is confirmed in the region @xmath213 . however the 2-cycle bulk solutions differ from the husimi lattice solutions , and converge to @xmath130 solution quickly above @xmath214 . since the bulk solution comes to be almost steady with thickness larger than @xmath168 , this difference is not really caused by the surface effect . our calculation requires a set of initial seeds for the recursive calculation as the procedure described in fig . [ fig15 ] , the bulk calculation takes the surface solution as its initial seed , in this way , once the surface solution reaches @xmath130 , the initial seeds of @xmath130 will immediately affect the recursive calculation inside the bulk and converges the bulk origin solutions to be @xmath130 . the bulk origin solutions at converging point with different thickness are shown in fig . [ fig24 ] . we can see the convergence occurs with very slight differences no matter how large the thickness is , while theoretically we should have the bulk origin solutions to be identical with husimi bulk solutions . simple to say , because of the property of recursive calculation , the bulk origin solution will be lead away from the exact description by the effect of surface solution in the temperature region between surface melting and husimi bulk melting transition temperatures , which is not really the surface effect in nature . [ ptb ] this is also a reason we are not interested in calculating the thermodynamics of the whole srl . theoretically , the solution on each site of the srl can be approached and therefore we can calculate the thermal properties . nevertheless , there are three facts make this calculation unreliable . firstly , as mentioned above , in the temperature region between surface melting and husimi bulk melting transition , the bulk origin solution is affected by the seeds of @xmath130 solutions on the surface . secondly , the recursive calculation technique requires several steps to reach the fix - point solutions . this implies the calculations on the first few layers closing to the surface are numerical instead of exact calculation . although for large thickness bulk this error can be neglected , the thermal behaviors associated with exact different layers is not useful . thirdly , the recursive structure of srl proportionally generates more surfaces with increasing thickness . unlike the regular lattice , in which the contribution of surface will be neglected with a sufficient large bulk , the srl will have half sites on the surface with infinitely large bulk trees . in this way , the thermodynamics on each site is just the averaged value of surface and bulk values . for a short summary , our approach of srl is good to track the thermal behaviors on the surface with the account of bulk contributions , but not to discover the thermal behaviors change with various thicknesses . in section i we have reviewed the findings that the presence of a free surface dramatically decreases the transition temperatures of bulk system . by comparing the thermodynamics on the surface / thin film we achieved in section ii and iii and in the bulk system , our calculations also clearly indicate the reduction of both melting and ideal glass transitions temperature on the surface . [ fig25 ] show the free energy comparison of husimi bulk system and srl : [ ptb ] in srl the melting and ideal glass transition temperature are dramatically decreased comparing to husimi bulk . in section i we introduced the empirical equation to describe the temperature reduction with the change of the thickness . since the thickness dependence of transition temperature reduction is not available in our methods , and our calculations focus on the transitions right on the surface , we may compare our results to the ratio of glass transition temperatures of bulk and the thinnest free - standing film in others works , either experimental or simulation results . the @xmath188 reduction ratio of srl is @xmath215 . in forrest and co - workers work , the @xmath2 of the thinnest ps film they made is @xmath216k and the @xmath2 of bulk ps is @xmath217k , the reduction ratio is @xmath218 [ 10 ] . in torres and co - workers md simulation , this reduction ratio of a free standing film is @xmath219 [ 12 ] . in de pablo and co - workers mc simulation , this ratio is @xmath220 [ 23 ] . because our results are from the reference case , which is very simple without any particular artificial parameters setup , the fact that our ratios are close to others results may only verify the validity or practice of our method . to particularly describe a real system , the setup of energy parameters is the most critical issue . the effects of energy parameters in our model will be discussed in later section . the fact that similar reduction can be observed in our small moleculars model implies that the lower transition temperature on surface / thin film basically originates from the dimension reduction and less interaction constraints . the chain - structure of polymer system may not be the main reason for transition temperature reduction , although it does play important roles to affect the transition properties . in this section we are going to study the effects of different energy parameters in srl by the variation of one parameter with other parameters fixed . the thickness is always set to be @xmath168 for providing a sufficient tree size and relatively shorter calculation time consumption . in eq.[surface_e_alpha ] we specified the diagonal interaction energy parameter on the surface to be @xmath221 , differed from the diagonal interaction @xmath42 inside the bulk . and similarly we have @xmath63 , ( @xmath65 ) and @xmath66 . this specification is to enable us setup different interaction circumstance on the surface and provide more versatility in our model . however the effects of either @xmath42 or its counterpart on the surface @xmath222 are the same . thus in this section we are going to simplify the case and setup @xmath42 and @xmath221 , @xmath46and @xmath66 to be the same . nevertheless , a variation with fixed @xmath63 will be discussed in details , because the nearest - neighbor interaction @xmath41 and @xmath63 have a much larger weight in the hamiltonian and play more critical roles in the system . the difference setup of @xmath41 and @xmath63 provides the simulation of surface tension which is critical to the surface properties . in the hamiltonian eq.[e_alpha ] , the nearest - neighbor interaction @xmath41 is negative by the definition of anti - ferromagnetic ising model , the system prefers an anti - aligned spins arrangement to obtain a lower energy with a negative first term , then we can also observe that for an anti - aligned spins arrangement , the diagonal spins pairs are in the same state . therefore , unlike the nearest - neighbor interaction @xmath41 , a positive diagonal interaction will make the second term to be negative , which encourages the alternating arrangement and increases the transition temperatures ( i.e. the crystal is more stable to melt ) . on the other hand , a negative @xmath42 will compete with @xmath41 and trend to destroy the ordered state , which decreases the transition temperature . because we have four nearest - neighbor and two diagonal interactions in a square unit , the nearest - neighbor interaction outweigh the diagonal interaction . also , there is no diagonal interaction in the surface single bond unit , which also lowers the contribution of @xmath42 . the variation of @xmath42 will only moderately change the transition temperatures on the surface ; the overall thermal behaviors are similar to the reference case . the thermal behaviors with four different @xmath42 values @xmath223 and @xmath224 are calculated and the melting and ideal glass transition change with @xmath42 variation is summarized in fig . [ fig26 ] and table 2 . it is obvious that positive @xmath42 will make the system more stable and increase the transition temperatures , or vice versa . we also found that @xmath42 can not be larger than @xmath225 otherwise stable solutions can not be reached . [ ptb ] , scaledwidth=80.0% ] [ c]llll + @xmath42 & @xmath138 & @xmath188 & @xmath226 + 0.4 & 1.95 & 1 & 1.950 + 0.2 & 1.82 & 0.95 & 1.916 + 0 & 1.70 & 0.89 & 1.910 + -0.2 & 1.55 & 0.88 & 1.761 + -0.4 & 1.38 & 0.80 & 1.725 the quadruplet interaction @xmath46 is a complicated term . by simply analyzing the fourth term in hamiltonian eq . ( 2.1 ) , it is difficult to give out a clear expectation on the effects of @xmath46 since both system preferred or defective structures can give out either positive or negative values of fours spins product , then consequently either positive or negative values of @xmath46 can against the crystallization . transition temperatures of @xmath227and @xmath228 with other parameters fixed to be 0 are shown in fig . [ fig27 ] . [ ptb ] from the graph above , we can see that the parameter @xmath46 has a similar effect to jp : positive j will make the system more stable and increase both transition temperatures , and vice versa . we also found that , unlike the limited value of @xmath42 , @xmath46 can be assigned with large value such as @xmath228 without destroying the stable solutions . the transition temperatures variation with different @xmath46 are summarized in table 3 . [ c]llll + @xmath46 & @xmath138 & @xmath188 & @xmath226 + 1.0 & 1.89 & 1.00 & 1.89 + 0.8 & 1.88 & 0.98 & 1.92 + 0.6 & 1.84 & 0.96 & 1.92 + 0.4 & 1.78 & 0.94 & 1.89 + 0.2 & 1.76 & 0.92 & 1.91 + 0 & 1.70 & 0.89 & 1.91 + -0.2 & 1.64 & 0.86 & 1.91 + -0.4 & 1.56 & 0.80 & 1.95 + -0.6 & 1.46 & 0.76 & 1.92 + -0.8 & 1.36 & 0.69 & 1.97 + -1.0 & 1.26 & 0.62 & 2.03 one important difference between the effects of @xmath42 and @xmath46 can be observed in table 2 and 3 : although both parameters act similarly in changing the transition temperature , with the @xmath46 variation the ratio of @xmath226 is relatively constant unless @xmath46 is assigned with extraordinary values like @xmath228 , while with the @xmath42 variation the ratio of @xmath226 also changes dramatically . this difference could be useful in modifying the parameters setup to describe the real systems or experimental results . for a real system , factors affecting the thermodynamics can act in different ways , some of them may lift / reduce both transition temperatures but keep the supercooled liquid region constant , or some may change the window size of supercooled liquid state . as mentioned above , for convenience we setup the energy parameters in the bulk and on the surface , such as @xmath42 and @xmath221 , to be the same . nevertheless , our lattice has a fixed bond length ( this length is unit @xmath24 in ising model ) between neighbor sites on the surface and in the bulk , and we also know that the asymmetric interactions and the surface tension are the main reason of unique properties of the surface . therefore with a fixed bong length , we have to somehow modify the surface interaction circumstance to mimic the surface tension or asymmetric interactions . this can only be made possible with different @xmath41 and @xmath63 setup . thus we fixed the bulk nearest - neighbor interaction parameter @xmath41 to be @xmath50 , and observe how the variation of surface bond interaction @xmath63 affect the thermodynamics on the surface . the transition temperatures change with variation of @xmath229 is summarized in fig . [ fig28 ] and table 4 . from fig . [ fig28 ] we can conclude that larger absolute value of negative @xmath63 makes the system more stable and increases both melting and ideal glass transition temperatures . the variation of @xmath63 revealed that the surface is more stable with large surface tension ( @xmath230 ) , or easier to undergo transitions than in the bulk with small surface tension ( @xmath231 ) . positive @xmath63 was also been tested however no stable solution can be reached . this is because the positive @xmath63 will prefer ferromagnetic - aligned spins arrangement which only has @xmath24-cycle solution available . on the other hand it is also hard to imagine a homogeneous system with particles attractive to each other in the bulk but repulsive on the surface . [ ptb ] , scaledwidth=80.0% ] [ c]llll + @xmath63 & @xmath138 & @xmath188 & @xmath226 + -0.5 & 1.12 & 0.55 & 2.04 + -0.7 & 1.38 & 0.70 & 1.97 + -0.9 & 1.60 & 0.82 & 1.95 + -1 & 1.70 & 0.89 & 1.91 + -1.1 & 1.80 & 0.94 & 1.91 + -1.3 & 2.00 & 1.05 & 1.90 + -1.5 & 2.18 & 1.14 & 1.91 it is easy to understand that we have @xmath130 solutions at high temperature , and 2-cycle solutions symmetric to @xmath130 at low temperature , is because that the magnetic field @xmath232 is @xmath23 , the system has no bias on either @xmath134 or @xmath137 spins on a particular site . thus at high temperature one site has @xmath131 probability to be occupied by either @xmath134 or @xmath137 spin . if a non - zero value of magnetic field @xmath232 exists , we can expect the 1-cycle solution off the central @xmath130 line or asymmetric 2-cycle solutions . a positive @xmath232 will prefer more + spins and moves the 1-cycle solution higher than @xmath130 , or vice versa . in our previous research on bulk husimi lattice , the three spins interaction ( triplet ) @xmath45 acts similarly to magnetic field @xmath232 in changing the bias of spins . however , in srl the @xmath45 and magnetic field @xmath232 are not useful , although for a general presentation we still have them in our lattice model and energy equations . the reason is that , the uneven property of the surface on srl determines the 1-cycle solution can only be @xmath130 . if a non - zero magnetic field @xmath232 or triplet interaction @xmath45 presents , the 1-cycle solution on the surface can not be achieved . the single bond unit does not have triple interaction @xmath45 and the effect of @xmath232 is much smaller than it is in the square unit , therefore the anti - aligning property of the single will always prefer to have different spins on it . in another word , if a 1-cycle solution can be achieved on the single surface bond , it must be the @xmath130 solution . even we can still calculate the thermodynamics of 2-cycle solution on the surface , we must have the corresponding 1-cycle solution to determine the melting and ideal glass transitions . therefore , we have to abandon the variation of @xmath232 and @xmath45 , which is a disadvantage of the srl . . we applied recursive lattice technique to investigate the thermodynamics and ideal glass transition on the surface / thin film of ising spin system with a theoretical base . a recursive lattice ( srl ) were constructed to describe a regular finite size square lattice , with a 2d bulk surrounded by 1d surface . the srl utilizes the finite - sized husimi lattice , which is integrated by 2d square units , to be the bulk part , and adopts single bond to connect the finite - sized bulk parts . these single bonds and the outsider bonds of bulk trees assemble the surfaces surrounding the bulk part and slipping through independent bulk parts . the lattice is constructed with particular approximations and compromises and is believed to be reliable approximation to the regular square lattice in the aspect of coordination numbers , i.e. the number of neighbor sites is @xmath27 inside the bulk and @xmath26 on the surface . the ising spins of two possible states @xmath134 or @xmath137 are put on the sites of srl . we assigned anti - ferromagnetic interaction in the model and different spin states in neighbor pairs are preferred to be the stable state ( crystal ) . the partition function of a sub - tree with its base spin state fixed is defined as partial partition function ( ppf ) . based on the recursive properties , the ppf of one sub - tree can be expressed as a function of the ppfs of its sub - sub - trees and a local weight . this recursive relation enables us to derive the ratio of pffs on one site . this ratio is called solution from which the thermodynamics of system can be achieved by gujrati trick . two kinds of stable solutions usually can be achieved by recursive calculation . one is in 2-cycle form and represents the ordered state , the other is in 1-cycle form representing the amorphous / metastable state . the thermal behaviors of 2-cycle and 1-cycle solutions indicate the melting transition by the cross point of free energies of two solutions , and the ideal glass transition by the negative entropy of 1-cycle solution . our results agree with others work that the transition temperatures are dramatically reduced on the surface / thin film comparing which in the bulk . nevertheless our work shows that , regardless of specific properties of particular materials , either polymer or small molecules , this transition temperature reduction can be simply due to the dimension downgrade and lower interaction energy on the surface . the effects of different interaction energy parameters other than nearest - neighbor interaction such as diagonal interaction and four spins ( quadruplet ) interaction were investigated . the variation of energy parameter could either increase or decrease the stability of system and change the transition temperatures according to the hamiltonian . the behaviors of energy parameters can be used to setup a particular model to describe the real system . while our finding agrees well with others experimental and simulation results , comparing to others work , there are two advantages in our approach : 1 ) in this work our model focuses on the small molecules system , it reveals the basic dimension origin of transition temperatures reduction without involving the long chain properties of polymer system ; 2 ) the thermodynamics of systems are derived by exact calculation method , the computation time is much shorter than typical simulation methods , usually the calculation of one set of parameters in the interesting temperature region can be done in less than 100 seconds .

the possibility that the gluino is the next to the lightest supersymmetric particle ( nlsp ) is discussed and it is shown that this situation arises in nonuniversal sugra models within a significant part of the parameter space compatible with all known experimental bounds . it is then shown that the gluino nlsp ( gnlsp ) models lead to a compressed sfermion spectrum with the sleptons often heavier than the squarks at least for the first two generations . the relic density here is governed by gluino coannihilation which is responsible for a relatively small mass splitting between the gluino and the neutralino masses . thus the gnlsp class of models is very predictive first because the susy production cross sections at the lhc are dominated by gluino production and second because the gluino production itself proceeds dominantly through a single channel which allows for a direct determination of the gluino mass and an indirect determination of the neutralino mass due to a linear relation between these two masses which is highly constrained by coannihilation . a detailed analysis of these models shows that the jet production and tagged b - jets from the gluino production can be discriminated from the standard model background with appropriate cuts . it is found that the gnlsp models can be tested with just 10 fb of integrated luminosity and may therefore be checked with low luminosity runs in the first data at the lhc . thus if a gnlsp model is realized , the lhc will turn into a gluino factory through a profuse production of gluinos with typically only a small fraction of total susy events arising from other production modes over the allowed gnlsp model parameter space . nub-3263 + yitp - sb-09 - 10 gluino nlsp , dark matter via gluino coannihilation , and lhc signatures + daniel feldman , zuowei liu , and pran nath + department of physics , northeastern university , boston , ma 02115 , usa , +c . n. yang institute for theoretical physics , + stony brook university , stony brook , ny 11794 , usa .

0905.1148

one of the interesting possibilities that arises within the landscape of possible sparticle mass hierarchies @xcite is that the gluino @xmath5 is the next to the lightest supersymmetric particle ( nlsp ) where neutralino dark matter produces the correct relic abundance of such matter consistent with the wmap observations @xcite . in fact , an analysis in the context of nonuniversal supergravity models ( nusugra ) reveals that the gluino nlsp model ( gnlsp ) arises in a significant part of the parameter space @xcite . amongst the various possible ways that the first four lightest sparticles may stack up in their mass hierarchy , one finds three such hierarchical mass patterns where the gluino is the nlsp @xcite which have been classified as models ( nusp13 , nusp14 , nusp15 ) @xmath6 @xmath7 @xmath6 @xmath8 @xmath6 @xmath9 , where @xmath7 is the nlsp , but the @xmath10 lies close to the @xmath7 mass . ] as given in table([gnlsptable1 ] ) . we will often refer to this subclass of nusugra as the gnlsp class of models . although progress has been made on the parameter space of sparticle masses with coannhilating gluinos @xcite , collider and dark matter detection implications of the gnlsp models have yet to be explored in any great detail . thus in this work we give a dedicated analysis of such a model . since the gluino is a strongly interacting particle , an nlsp gluino will change drastically the typical sparticle analyses . the phenomenology of gnlsp models is very different from that of a model where the gluino is the lsp @xcite which we do not discuss in this paper . we note also that while relatively light gluinos have been studied in detail in reduced @xmath11 gaugino mass models @xcite , the gnlsp situation , which we cover here , was not explored . . hierarchical sparticle mass patterns for the four lightest sparticles , where @xmath12 is the lsp neutralino , and where the gluino is the nlsp that arises in the nusugra models . the labeling of the mass patterns is as given in @xcite . [ cols="^,<,<,<",options="header " , ] @xmath13 & @xmath14 + @xmath15 & 1.72 & @xmath16@xmath17 & 1.60 & @xmath18 @xmath19 & 1.55 & @xmath20 @xmath21 & - & - @xmath22 & @xmath23 & @xmath24 @xmath25 susy - hit@xcite + @xmath26 & ( 20,61,19)% & ( 20,61,19)% @xmath27 & - & 0.03% @xmath28 & 82% & 86% @xmath29 & 100% & 100% and a comparison with msugra . right panel : a display of the lightest squark mass vs the gluino mass in gnlsp@xmath30 and a comparison with msugra . the gluino mass is relatively light in the gnlsp models with an upper limit of a tev under the assumed naturalness assumptions . the lightness of the gluino mass leads to the dominance of the @xmath31 over all other sparticle production processes in the gnlsp models . , title="fig:",width=311,height=245 ] and a comparison with msugra . right panel : a display of the lightest squark mass vs the gluino mass in gnlsp@xmath30 and a comparison with msugra . the gluino mass is relatively light in the gnlsp models with an upper limit of a tev under the assumed naturalness assumptions . the lightness of the gluino mass leads to the dominance of the @xmath31 over all other sparticle production processes in the gnlsp models . , title="fig:",width=302,height=245 ] one observes that the squarks decay back into a gluino with a large branching ratio . the nlo calculation retains the dominance of the @xmath32 at the level of 96% , i.e. , @xmath33 . we note that the model in table([gnlsp1 ] ) has a very small branching fraction @xmath27 ( calculated with susy - hit@xcite and a very similar suppression is seen with isajet @xcite ( @xmath60.1% ) ) . this is to be contrasted with the @xmath34 model ( see eq.([co ] ) ) , which has a @xmath35 of 65 % with susy - hit and 89% with isajet . the gluino decay branching ratios are discussed in further detail later . in the left panel of fig.([mass ] ) we give a comparison of the average mass of the squarks vs the gluino mass in gnlsp and compare it to that for the msugra model . a similar analysis for the lightest squark mass is given on the right panel . in both cases one finds that aside from a very small region , the spectra from these two models do not overlap . quite remarkably for the assumed naturalness conditions as described in sec.([parameter ] ) the gluino mass in the gnlsp models has an upper limit of about a tev , while in msugra this limit extends far beyond . on the other hand , the upper limit on the squark masses is quite large extending to several tev in both cases . it is the relative lightness of the gluino mass in the gnlsp case that enhances the production of the gluinos relative to all other susy production processes such as @xmath36 and @xmath37 at the lhc . this further explains the result of fig.([gluinofactory1 ] ) and of table([gnlsp1 ] ) which show that the @xmath31 production cross section dominates over all others . thus in effect , in the gnlsp case , the lhc will become a gluino factory . nonetheless , the production cross section for the squarks is still significant and their production could be detectable with efficient cuts . for example , the stop is usually the lightest squark in gnlsp models , so its production could be significant , and it can decay via @xmath38 . since the @xmath39 has decays into @xmath40 , the stop will have some leptonic signatures while the gluino is a pure jet signal . an analysis of these leptonic signals requires a further dedicated analysis which is left for a future analysis . we also note that superficially the overlap of the few gnlsp points with the msugra point might be construed as the existence of a degeneracy for these parameter points . however , we need to keep in mind that what is plotted is an average squark mass . further , as already discussed there is a significant splitting between the squark masses and the slepton masses for the msugra model for the first two generations while there is very little splitting in this case for the gnlsp model . thus at the very least the slepton - squark splittings lift any perceived degeneracy indicated in fig.([mass ] ) . we should also point out that although the production cross section of squarks and gluinos can be comparable for gnlsp models and for msugra models , their lhc signatures tend to be significantly different due to the mass spectra of sparticles being rather different and specifically this is so since the gluino is the nlsp in the class of models we consider . * early lhc discovery prospects at low luminosity : * each gnlsp model is subject to the experimental constraints as discussed in section([exp ] ) . we investigate the lhc signatures of 1070 such gnlsp models with pythia coupled to pgs4 @xcite . branching fractions are computed with susy - hit and fed directly into the pythia decay table via the susy les houches accord ( slha ) @xcite interface . for the gnlsp models , this is quite important as one must take into account the radiative decay @xmath35 which can be substantial for this class of models . the lhc detector simulation proceeds with pgs4 with the level 1 ( l1 ) triggers designed to mimic the compact muon solenoid detector ( cms ) specifications @xcite with the lhc detector card . specifically the l1 trigger level cuts that are imposed are as follows@xcite : ( 1 ) inclusive isolated lepton ( @xmath41/@xmath42 ) ( 30 gev ) ; ( 2 ) lepton plus jet ( 20 gev , 100 gev ) ; ( 3 ) isolated di - leptons ( 15 gev ) ; ( 4 ) di - leptons plus jet ( 10 gev , 100 gev ) ; ( 5 ) isolated di - leptons ( 10 gev ) ; ( 6 ) isolated lepton plus isolated @xmath43 ( 15 gev , 45 gev ) ; ( 7 ) isolated di - tau ( 60 gev ) ; ( 8) inclusive isolated photon ( 80 gev ) ; ( 9 ) isolated di - photon ( 25 gev ) ; ( 10 ) inclusive @xmath44 ( 90 gev ) ; ( 11 ) inclusive single - jet ( 400 gev ) ; ( 12 ) jet plus @xmath44 ( 180 gev , 80 gev ) ; ( 13 ) acoplanar jet and @xmath44 with @xmath45 ( 100 gev , 80 gev ) ( jet , @xmath44 ) ; ( 14 ) acoplanar di - jets with ( @xmath46 ) ( 200 gev ) . muon isolation is controlled by employing the cleaning script in pgs4 . sm backgrounds have been generated with qcd multi - jet production due to light quark flavors , heavy flavor jets ( @xmath47 , @xmath48 ) , drell - yan , single @xmath49 production in association with quarks and gluons ( @xmath50/@xmath51 + jets ) , and @xmath52 , @xmath53 , @xmath54 pair production . the standard criteria for the discovery limit are imposed , namely , that the susy signal is taken as discoverable if the number of susy events exceeds @xmath55 or 10 whichever is larger , i.e. , @xmath56 . we implement several classes of post trigger level cuts to analyze the sm background and our event samples . * post trigger level cuts : * as already noted the dominant sparticle production process at the lhc will be @xmath57 , and the dominant decays of the gluino are @xmath58 and depending on the particular part of the parameter space the decay @xmath59@xcite can also be large , and in fact can dominate over the 3 body decay , depending on a confluence of the following : ( a ) the lightness of the gluino , ( b ) the largeness of the squark masses , and ( c ) the neutralino mixing matrix . while a heavy gluino can decay into quarks and squarks and into weak gauge bosons @xmath60@xcite , these decays are either kinematically forbidden , such as into quarks and squarks , or are highly suppressed for the gnlsp models we are discussing here . the above implies that gnlsp models will lead to a preponderance of jet signatures . in our analysis we impose three different sets of post trigger level cuts to optimize the signal and reduce the background from the standard model processes . we classify these as : ( i ) class 1 ( c1 ) , ( ii ) class 2 ( c2 ) , and ( iii ) class 3 ( c3 ) post trigger level cuts . we discuss these in some detail below and discuss the signals that are best detected with these three classes of cuts . before proceeding further , we note that the missing transverse momentum is an important cut both as a trigger level as well as a post trigger level cut allowing one to increase the signal relative to the background . thus susy models with a lsp which is massive tend to produce events at the hadron colliders with a larger missing energy . in order to suppress the standard model background , usually a large missing transverse momentum cut is employed . we note further that since @xmath61 and @xmath62 are produced in the processes @xmath63 as well as with a certain fraction in @xmath57 with subsequent decays of @xmath10 ( this latter case being the most relevant one discussed here ) , one has a significant number of b jets produced in these events . thus b -tagging is a useful instrument in their identification . these features will be seen in the post trigger level cuts we discuss below . _ ( i ) class 1 post trigger level cuts ( c1 ) _ 4 1 . electrons , and muons with @xmath64 gev ( where @xmath65 is the transverse momentum ) and @xmath66 ( where @xmath67 is pseudorapidity ) are selected . jets with @xmath68 gev and @xmath69 are selected . events with @xmath70 gev are selected . 4 since the gnlsp models have a gluino which lies close to the lsp consistent with the relic density constraints , a large portion of events generated by the susy processes have a less energetic jet signal and a relatively smaller missing energy than may be observed , for example , in the case of stau coannihilation ( see @xcite for analysis of such a signature ) . taking the above into account we modified the imposed c1 post trigger level cuts in c2 and c3 which are to be discussed below . as already emphasized , the susy production cross section are dominated by gluino pair production and decays of the gluino . therefore , we only select events that have at least two jets that pass our jet selection condition . we also investigate the effects of putting a softer missing @xmath65 cut and jet @xmath65 cut . as a function of the azimuthal angle @xmath71 between the two hardest jets in the gnlsp model relative to the sm backgrounds with c2 cuts . right panel : @xmath72 distributions of gnlsp models and sm backgrounds with c2 cuts . these distributions act as a guide for implementing the c3 cuts as discussed in the text . , title="fig:",width=283,height=264 ] as a function of the azimuthal angle @xmath71 between the two hardest jets in the gnlsp model relative to the sm backgrounds with c2 cuts . right panel : @xmath72 distributions of gnlsp models and sm backgrounds with c2 cuts . these distributions act as a guide for implementing the c3 cuts as discussed in the text . , title="fig:",width=283,height=264 ] _ ( ii ) class 2 post trigger level cuts ( c2 ) _ 4 1 . electrons , and muons with @xmath64 gev and @xmath66 are selected . 2 . jets with @xmath73 gev and @xmath69 are selected . events with @xmath74 gev are selected . events with at least 2 jets are selected . 4 the dominant sm background for gnlsp models are from qcd , @xmath50/@xmath51 + jets , @xmath47 , and @xmath48 and we focus on these in fig.([distributions ] ) . in the left panel of fig.([distributions ] ) we give an analysis of these backgrounds at the lhc with events / bin / fb@xmath0 as a function of the azimuthal angle @xmath75 between the two leading jets . we also show the distributions for two gnlsp model points with masses of 400 and 500 gev respectively . a similar analysis is presented but as a function of the total jet @xmath65 ( labeled @xmath76 ) in the right panel of fig.([distributions ] ) , where again we also exhibit the distributions for two gnlsp model points . the analysis provides a good starting point for charting out the post trigger level cuts needed to reduce the background and enhance the signal . thus the analysis suggests that one may cut out the events that have a large @xmath75 as this cut will suppress the qcd background due to light quark flavors , @xmath47 and @xmath77 . a veto on isolated electrons or muons was applied to reject the background events containing @xmath51 or @xmath50 leptonic decays . a simple counting of events after applying c2 cuts reveals that the gnlsp models are nearly lepton free , as is the case for the @xmath78 and di - jets background . however , the standard model backgrounds due to drell - yan process and ( @xmath50/@xmath51 + jets ) have ( @xmath79 ) of the total events that contain electrons or muons , while for @xmath48 background it is about @xmath80 . the @xmath52 , @xmath53 , @xmath54 pair production resulting in multi - leptonic backgrounds have an even larger percentage of leptonic events . therefore , the @xmath81 veto significantly enhances the gnlsp signals over the standard model background . this leads us to investigate cut class c3 . + + _ ( iii ) class 3 post trigger level cuts ( c3 ) _ 4 1 . apply cut set c2 . electron or muon veto is imposed . 3 . @xmath82 gev . 4 . the azimuthal angle @xmath83 between jet1 ( the hardest jet ) and jet2 ( the second hardest jet ) is chosen so that @xmath84 . 5 . the azimuthal angle @xmath85 between jet1 ( the hardest jet ) and @xmath44 is chosen so that @xmath86 . 6 . the azimuthal angle @xmath87 between jet2 ( the second hardest jet ) and @xmath44 is chosen so that @xmath88 . as is indicated from the preceding discussion , jets with and without tagged b jets are important signals for the discovery of the gnlsp models . another important signature is @xmath89 , which is the average magnitude of the missing transverse momentum , where the average extends over all events passing the cuts . we discuss now several signatures for the gnlsp models as given in fig.([c3cuts ] ) . the analysis of fig.([c3cuts ] ) is given at @xmath90 under post trigger level cuts c3 with an integrated luminosity of 10 fb@xmath0 . the left panel of fig.([c3cuts ] ) exhibits the average @xmath44 as a function of the gluino mass . here one finds that essentially all of the parameter points of the gnlsp models have an average @xmath44 which is larger , and often significantly larger , than for the standard model case for which the average @xmath44 is found to be @xmath91 257 gev under the same set of cuts . the right panel illustrates the ratio @xmath92 vs the gluino mass where @xmath93 is the total number of susy events and @xmath94 is the standard model background , again with the imposition of post trigger level cuts c3 . here one finds that the ratio @xmath92 lies above 5@xmath95 discovery limits for gluino masses up to 800 gev making this ratio an important channel for the discovery of the gnlsp models . the left panel of fig.([c3cutschannelandcompare ] ) demonstrates that the events containing a single tagged b - jet leads to a discoverable signal over a wide range of gluino masses . specifically this panel gives the ratio @xmath96 vs the gluino mass where @xmath97 is the number of susy events with 1 tagged b - jet and @xmath98 is the number of standard model events with 1 tagged b - jet under the imposition of class c3 cuts . this channel provides a @xmath99 discovery for gluino masses up to about 600 gev . we briefly comment on the ability to tag b - jets in this case . naively one would expect that for cases in which the dominant decays are @xmath100 , with sizeable branchings in b quarks , that the b - jet events would be a gold plated signal . however , here the b s come out rather soft since the mass splitting of the @xmath101 is rather small , typically around 50 gev ( the phenomenon responsible for the satisfaction of the relic density is precisely this mass split ) . while the ability to tag the b s is possible , it is indeed more difficult than the canonical situation seen on the hyperbolic branch of rewsb @xcite ( for recent work on b - tagging analyses see @xcite ) . nonetheless , as discussed above the signal for singly tagged b - jets is strong enough that it can be a useful one . the right panel of fig.([c3cutschannelandcompare ] ) we compare the 4-jet discovery limits under c1 and c3 cuts . here it is found that the sm background drops by an appreciable amount as one goes from c1 cuts to c3 cuts , allowing one to extend the discovery limit in 4-jet channel by over 100 gev with c3 cuts relative to imposing the c1 cuts in this channel . amongst the classes of n - jet events , we find this channel to be the most enhanced when passing from c1 to c3 . combining all the channels analyzed in fig.([c3cuts],[c3cutschannelandcompare ] ) one finds that gluino masses up to 800 gev are discoverable by this technique for a gnlsp model with just 10 fb@xmath0 of integrated luminosity . thus the validity of the gnlsp models can be tested with first data from the lhc . many of the gnlsp model points have an lsp neutralino which is bino like . however , there is also a significant set of models that have an lsp with large higgsino components ( many of which are at high @xmath102 but this is not exclusively so ) . the large @xmath103 parameter points are easily spotted by examining the @xmath102 vs @xmath104 plot in fig.([a - tanbeta ] ) . the gnlsp models have important implications for the direct detection of neutralino dark matter . an analysis of the spin independent and spin dependent cross sections in dark matter experiments is given in fig.([dark2 ] ) implemented with micromegas . included are published limits from the zeplin - iii experiment@xcite , the first five - tower cdms data@xcite and the xenon 10 results @xcite . projected limits ( indicated by * ) from cdms and lux are also shown @xcite . the @xmath105 vs @xmath106 analysis shows some interesting results . first one finds that there are a class of gnlsp models which are beginning to be constrained by the direct detection dark matter experiments . these models would also produce large @xmath107 production cross sections and will be easily visible at the lhc . however , interestingly there is another class of low mass neutralino ( and low mass gluino ) models which have rather small spin independent cross sections which are outside the reach of the direct detection experiments in the foreseeable future . nonetheless these models would still lead to rather large @xmath107 production cross sections and hence will be visible at the lhc . thus the lhc can detect many of the gnlsp models which are most likely inaccessible to the direct detection dark matter experiments . on the flip side , for the large set of low mass gnlsp models one still has the possibility of a very light neutralino ( gluino ) with a sizeable higgsino component and consequently a large spin independent cross section . thus if a light gluino is indeed indicated early on at the lhc , it may also provide a hint of the size of the dark matter signal in direct detection of dark matter . we note in passing that a plot of @xmath105 vs @xmath108 looks very similar to the left panel of fig.([dark2 ] ) as the gluino and neutralino mass are related at the low scale via @xmath109 . we note that in the gnlsp class of models , the direct annihilations of @xmath110 into electron positron pairs is helicity suppressed . for example , we obtain @xmath111 ( at @xmath112 ) for the model of table([gnlsp1 ] ) while the self annihilation into @xmath113 are equally small . annihilations into @xmath114 are found to be the largest ( this particular model has @xmath115 . thus a significant boost will be needed to explain the recent cosmic ray excess@xcite . it is useful to give benchmarks for the three gnlsp models a , b , and c discussed in sec.([gaugino ] ) . as mentioned above , we go beyond the perturbative calculation , and include the sommerfeld enhancement of the cross section in these benchmarks . in table([benchain ] ) we give benchmarks for model gnlsp@xmath116 . the benchmarks are chosen to exhibit a significant diversity in the input values . some of the low lying spectrum as well as the spin independent cross section @xmath117 and the spin dependent cross section @xmath118 corresponding to table([benchain ] ) are exhibited in table([benchaout ] ) . the analysis of table([benchaout ] ) shows a variation over two orders of magnitude for @xmath117 . as discussed already the variation arises due to changes in the higgsino vs the bino component of the lsp . similar benchmarks for model gnlsp@xmath119 are given in table([benchbin ] ) and the corresponding light sparticle spectrum and the corresponding spin independent cross section @xmath117 and the spin dependent cross section @xmath118 are exhibited in table([benchbout ] ) . finally , a similar analysis for the model gnlsp@xmath30 is given in table([benchgin ] ) and table([benchgout ] ) . in the above we have given an analysis of a class of models with nonuniversalities which lead to a gluino as the nlsp . several important observations emerge from this analysis which have bearing on the observation of sparticles at the lhc . perhaps the most important of these is that if the gluino is the nlsp , then the @xmath57 cross section at the lhc dominates over all others in the gnlsp models . the dominance of the gluino production and the fact that the @xmath31 production cross sections are large implies that the observation of supersymmetry via the gluino production can occur with the first data from the lhc . it is found that the dominant signal of the gluino nlsp model are multi - jets , tagged b - jets and missing energy and it is possible to devise post trigger level cuts which discriminate these models above the standard model backgrounds . such cuts which reduce the background and enhance the signal to the background ratio were devised and implemented in this paper . we note also that the inverse of the lhc process , namely @xmath120 , is largely responsible for the satisfaction of the relic density when the neutralino and the gluino coannihilate . further , the analysis of gnlsp models reveals that there exists a significant region of the parameter space in these models where the neutralino has a large higgsino content , and the neutralino - proton spin independent scattering cross section is sizeable and can be probed with the current experimental sensitivity , and sensitivities that would be achievable in future experiments . however , there are also other regions of the parameter space where the neutralino is mostly bino like and in this case the spin independent cross sections can fall well below the current experimental sensitivity , and well below the sensitivity that would be achievable in the near future experiments . interestingly , the bino cases , though difficult to discover in dark matter experiments , can be accessible at the lhc since the gluino mass in these models lies within the reach of the lhc even at low luminosities . another aspect of the gnlsp models was also discussed which relates to the compressed nature of the sfermion mass spectrum relative to the case of universal gaugino masses . here the sleptons in the first two generations could be almost degenerate in mass and often even heavier than their squark counterparts . several benchmarks for the gnlsp models were also given to facilitate further work . it was also pointed out that a test of the gnlsp models can be done with just 10 fb@xmath0 of integrated luminosity and thus it is one of the models that can be checked with the early data at the lhc . finally , we emphasize once again that if the gluino is an nlsp then the production of gluinos will dominate all other sparticle production making the lhc effectively a gluino factory . _ acknowledgments _ : this research is supported in part by nsf grants phy-0653342 ( stony brook ) and phy-0757959 ( northeastern ) . the work of df was also supported in part by office of the vice - provost at northeastern university . each of the mass ratios listed in table([nonunitable1 ] ) gives two gaugino mass relations at the gut scale . of these one is ` unstable ' to the inclusion of a singlet @xmath121 term breaking while the other one is ` stable ' and remains valid when one includes a singlet @xmath121 term breaking along with the non - singlet breaking . below we list only the ` stable ' mass relations . the mass relations are labeled numerically ( 1)-(23 ) as in table([nonunitable1 ] ) . they are , @xmath122 and @xmath123 in each of the mass relations in eq.([sum ] ) vanishes . ] @xmath124 in addition to the above , models ( 6 ) , ( 8) , ( 11 ) , ( 12 ) , ( 15 ) , ( 17 ) , ( 22 ) in table([nonunitable1 ] ) satisfy the relation @xmath125 , while model ( 19 ) in table([nonunitable1 ] ) satisfies the relation @xmath126 , and model ( 21 ) in table([nonunitable1 ] ) satisfies the relation @xmath127 . these mass relations would be appropriately modified at low scales by the renormalization group evolution . thus at the one loop level the mass relations at the electroweak scale are ( r-1 ) m_1(q ) + m_2(q ) -r m_3(q)=0 , where @xmath128 is the ratio as given by table([nonunitable1 ] ) . assuming one can determine with accuracy the gaugino masses , these mass relations can be a useful indicator of the specific @xmath121 type breaking and hence of the type of nonuniversal sugra model one has at the gut scale . here we give a one loop analysis of how the quasi degeneracy of the lsp and of the gnlsp comes about and then discuss a similar phenomenon for the squarks and the sleptons for the first two generations . we begin with the gaugino masses which at the gut scale obey the nonuniversality condition of eq.([non1 ] ) which we restate in more compact notation as follows : for model @xmath129 we have m_1^(i)= _ 1^(i ) , m_2^(i)= _ 2^(i ) , m_3^(i)= _ 3^(i ) , [ non2 ] where _ 1^(i)= ( 1+a_i _ i ) , _ 2^(i ) = ( 1+b_i _ i ) , _ 3^(i)= ( 1+c_i _ i ) . [ non3 ] with the above boundary conditions , the gaugino masses at the electroweak scale @xmath130 , where @xmath131 is the gut scale and @xmath132 is the electroweak scale , are given by m_a^(i)(t)= ^(i)_a _ a(t ) , a=1,2,3 . here correspond to the @xmath133 , @xmath134 and @xmath11 gauge groups , and @xmath135 are the corresponding fine structure constants for these groups at the electroweak scale . for the msugra case , @xmath136 , and one simply has that at one loop m_1(t ) : m_2(t ) : m_3(t)= _ 1(t):_2(t):_3(t ) . using the experimental values of the gauge coupling constants at the electroweak scale one finds the three gaugino masses roughly in the ratio @xmath137 . in this case the gaugino masses are split in a very significant way . however , in the presence of nonuniversalities the ratios will be modified in a very different way . thus with the inclusion of the modifications @xmath138 , the three gaugino masses at the electroweak scale will be roughly in the ratio @xmath139 . it is clear then that the choice @xmath140 will make masses of gaugino 1 and of gaugino 3 almost degenerate . of course , more realistically there would be mixings between the gauginos and the higgsinos and the mass eigenstates will be admixtures of these . thus the mass relation of the lightest neutralino and of the gluino will be more complicated . still the above approximation may roughly hold when the neutralino is mostly a bino . + nonuniversalities also enter in the masses for the squarks and for the sleptons . for the first two generation down squarks one finds m^2_d_il(t)=m_0 ^ 2+m_di^2 + _ g [ _ 3 + _ 2 + _ 1 ] ^2 + ( -+^2_w ) m_z^2 ( 2 ) , + m^2_d_ir(t)=m_0 ^ 2+m_di^2 + _ g [ _ 3+_1 ] ^2 -^2_w m_z^2 ( 2 ) , where f_a= _ a^2 f_a , f_a(t)= ( 1- ) , where @xmath141 , @xmath142 , and @xmath143 for the gauge groups @xmath133 , @xmath134 and @xmath11 . for the case of the first two generations of charged leptons one has m^2_e_il(t)=m_0 ^ 2+m_ei^2 + _ g [ _ 2 + _ 1 ] ^2 + ( -+^2_w ) m_z^2 ( 2 ) , + m^2_e_ir(t)=m_0 ^ 2+m_ei^2 + _ g f_1 ^2 -^2_w m_z^2 ( 2 ) . using the above we find that = m_di^2 - m_ei^2 + _ g [ _ 3 ^ 2 f_3- _ 1 ^ 2 f_1 ] . [ diff ] it is now easily seen that the last brace in eq.([diff ] ) can vanish or even turn negative by appropriate choice of @xmath144 vs @xmath145 . the above situation leads to a near equality of the squark and of the slepton masses discussed in the text of the paper . the @xmath41 parameter also has a strong dependence on nonuniversalities . this is seen by examining the relation that determines @xmath146 , i.e. , ^2=(m_h_1 ^ 2-m_h_2 ^ 2 ^ 2)(^2 - 1)^-1 -m_z^2 + ^2 , where @xmath147 is the loop correction . thus @xmath148 , and @xmath149 are sensitive to nonuniversalities since m_h_1 ^ 2=m_0 ^ 2+_g ( _ 2(t ) + _ 1(t ) ) ^2 , where the nonuniversalities enter via the @xmath150 functions . similarly @xmath149 is given by m_h_2 ^ 2=^2e(t ) + a_0m_0 ( t)+m_0 ^ 2(h(t)-k(t)a_0 ^ 2 ) , where the tilde functions @xmath151 and @xmath152 are modified due to nonuniversality while the functions @xmath153 and @xmath154 are not unaffected ( for definitions of these functions see the first paper of @xcite and the references therein which also gives a more detailed discussion of this topic ) . the dependence on nonuniversalties is more complicated in this case because of the coupling with the top quark . specifically , one can derive the following relation which gives the explicit dependence on nonuniversalities = ( t^2 - 1)^-1(^2g_a-t^2(^2e_a+a_0m_0f_a ) ) + , where @xmath155 , @xmath156 , @xmath157 , and @xmath158 . one can make a semi - quantitative estimate of the dependence of @xmath146 on nonuniversalities from above . we note , however , that @xmath41 does not enter sensitively in the sum rule for the first two generation of squarks and sleptons and thus an estimate of the compression of the sfermion spectrum in the first two generations can be made without estimate of the @xmath41 parameter . we provide here sample model points for the gnlsp class of models . each model point obeys experimental constraints as discussed in the text . we use here suspect 2.41 coupled to micromegas ( mo ) 2.2.cpc along with an independant code which agrees with mo at the pertuabtive level but accounts for the non - perturbative effects discussed in the text . similar model points may be obtained with other spectrum calculators coupled to mo [ see , for example , table([gnlspcompare ] ) where one such comparison is given , which , however , is only at the perturbative level ] . ( gev ) & @xmath159 ( gev ) & @xmath160 & @xmath161 & @xmath162 + @xmath163 & 2949 & 692 & 3658 & 35 & 0.566 & -0.847 @xmath164 & 2706 & 783 & 4408 & 37 & 0.560 & -0.839 @xmath165 & 2529 & 946 & 3873 & 41 & 0.560 & -0.837 @xmath166 & 2967 & 910 & 5114 & 27 & 0.557 & -0.834 @xmath167 & 2574 & 1058 & 4197 & 42 & 0.557 & -0.833 @xmath168 & 2821 & 1019 & 5050 & 20 & 0.554 & -0.830 @xmath169 & 3008 & 1252 & -3241 & 27 & 0.558 & -0.836 @xmath170 & 2746 & 1265 & 5186 & 17 & 0.551 & -0.824 @xmath171 & @xmath172 & @xmath173 & @xmath174 & @xmath175 & @xmath176 + model & ( gev ) & ( gev ) & ( gev ) & ( gev ) & ( gev ) & @xmath177 & ( pb ) & ( pb ) + @xmath163 & 117 & 285 & 343 & 343 & 1560 & 2130 @xmath178 0.241 & 3.0@xmath179 & 8.0@xmath180 @xmath164 & 117 & 337 & 387 & 869 & 1225 & 1952 @xmath178 0.004 & 2.7@xmath181 & 2.8@xmath182 @xmath165 & 116 & 398 & 456 & 480 & 1190 & 1540 @xmath178 0.135 & 2.3@xmath179 & 2.4@xmath180@xmath166 & 117 & 399 & 454 & 1080 & 1276 & 2724 @xmath178 0.003 & 1.3@xmath181 & 1.1@xmath182 @xmath167 & 116 & 447 & 510 & 531 & 1161 & 1514 @xmath178 0.131 & 2.2@xmath179 & 1.9@xmath180 @xmath168 & 117 & 448 & 507 & 1064 & 1149 & 2886 @xmath178 0.003 & 2.1@xmath181 & 1.4@xmath182 @xmath169 & 120 & 551 & 618 & 645 & 1332 & 2718 @xmath178 0.108 & 1.3@xmath179 & 1.0@xmath180 @xmath170 & 116 & 557 & 624 & 970 & 1032 & 2960 @xmath178 0.007 & 6.7@xmath181 & 3.3@xmath182 ( gev ) & @xmath159 ( gev ) & @xmath160 & @xmath161 & @xmath162 + @xmath183 & 2421 & 736 & 3414 & 51 & 0.000 & -0.841 @xmath184 & 3406 & 734 & 4655 & 35 & 0.000 & -0.848 @xmath185 & 2890 & 945 & -2977 & 47 & 0.000 & -0.844 @xmath186 & 3772 & 988 & 5894 & 46 & 0.000 & -0.837 @xmath187 & 2857 & 1158 & -2631 & 24 & 0.000 & -0.842 @xmath188 & 2943 & 1142 & 5006 & 12 & 0.000 & -0.831 @xmath189 & 3188 & 1376 & -2479 & 7 & 0.000 & -0.837 @xmath190 & 2659 & 1380 & 5028 & 37 & 0.000 & -0.825 @xmath171 & @xmath172 & @xmath173 & @xmath174 & @xmath175 & @xmath176 + model & ( gev ) & ( gev ) & ( gev ) & ( gev ) & ( gev ) & @xmath177 & ( pb ) & ( pb ) + @xmath183 & 116 & 313 & 364 & 557 & 1223 & 409 @xmath178 0.009 & 2.4@xmath179 & 1.2@xmath191 @xmath184 & 119 & 316 & 364 & 576 & 1734 & 2520 @xmath178 0.003 & 1.4@xmath181 & 2.0@xmath182 @xmath185 & 118 & 417 & 484 & 582 & 1507 & 665 @xmath178 0.035 & 2.0@xmath179 & 5.2@xmath191@xmath186 & 119 & 428 & 488 & 779 & 1818 & 1756 @xmath178 0.002 & 1.0@xmath181 & 5.6@xmath179 @xmath187 & 119 & 502 & 566 & 583 & 1475 & 2610 @xmath178 0.140 & 1.8@xmath179 & 1.7@xmath180 @xmath188 & 117 & 502 & 565 & 948 & 1299 & 3177 @xmath178 0.002 & 1.7@xmath181 & 6.8@xmath179 @xmath189 & 117 & 601 & 672 & 701 & 1676 & 3382 @xmath178 0.113 & 1.8@xmath179 & 7.4@xmath191 @xmath190 & 116 & 598 & 669 & 1051 & 1121 & 2031 @xmath178 0.004 & 4.3@xmath181 & 1.6@xmath182 ( gev ) & @xmath159 ( gev ) & @xmath160 & @xmath161 & @xmath162 + @xmath192 & 1604 & 450 & 2035 & 49 & -0.317 & -0.852 @xmath193 & 2119 & 696 & 2860 & 44 & 0.291 & -0.845 @xmath194 & 2443 & 948 & 2823 & 11 & -0.227 & -0.842 @xmath195 & 3850 & 1111 & 4388 & 9 & 0.209 & -0.840 @xmath196 & 2599 & 1270 & 3656 & 14 & -0.009 & -0.833 @xmath197 & 2458 & 1479 & 5045 & 35 & 0.462 & -0.823 @xmath198 & 2087 & 453 & 2359 & 21 & 0.292 & -0.862 @xmath199 & 1958 & 674 & 2950 & 22 & 0.299 & -0.843 @xmath200 & 3874 & 1098 & 4455 & 8 & -0.018 & -0.839 @xmath201 & 2543 & 1337 & 4188 & 49 & -0.147 & -0.826 @xmath202 & 3288 & 1431 & 5879 & 15 & 0.703 & -0.822 @xmath203 & 3942 & 1755 & 5995 & 36 & 0.080 & -0.825 @xmath171 & @xmath172 & @xmath173 & @xmath174 & @xmath175 & @xmath176 + model & ( gev ) & ( gev ) & ( gev ) & ( gev ) & ( gev ) & @xmath177 & ( pb ) & ( pb ) + @xmath192 & 113 & 185 & 227 & 228 & 830 & 516 @xmath178 0.022 & 3.6@xmath179 & 9.0@xmath191@xmath193 & 115 & 291 & 338 & 423 & 1073 & 1065 @xmath178 0.056 & 1.4@xmath179 & 1.6@xmath180@xmath194 & 115 & 404 & 461 & 508 & 1315 & 2488 @xmath178 0.075 & 1.3@xmath179 & 1.2@xmath180@xmath195 & 117 & 481 & 546 & 587 & 2078 & 3939 @xmath178 0.098 & 1.4@xmath179 & 1.0@xmath180 @xmath196 & 116 & 542 & 610 & 609 & 1320 & 2677 @xmath178 0.200 & 2.9@xmath179 & 1.9@xmath180 @xmath197 & 115 & 632 & 704 & 716 & 830 & 2042 @xmath178 0.130 & 2.0@xmath179 & 9.7@xmath191 @xmath198 & 115 & 189 & 223 & 333 & 1114 & 1938 @xmath178 0.047 & 5.1@xmath204 & 2.2@xmath180 @xmath199 & 114 & 289 & 333 & 581 & 924 & 1887 @xmath178 0.012 & 1.2@xmath204 & 1.8@xmath191@xmath200 & 117 & 483 & 546 & 816 & 2088 & 3996 @xmath178 0.010 & 1.1@xmath204 & 6.0@xmath182 @xmath201 & 116 & 576 & 648 & 797 & 1242 & 956 @xmath178 0.018 & 5.4@xmath204 & 1.3@xmath191 @xmath202 & 117 & 634 & 708 & 909 & 1310 & 3558 @xmath178 0.015 & 1.7@xmath204 & 8.0@xmath182 @xmath203 & 119 & 769 & 849 & 948 & 1965 & 2909 @xmath178 0.031 & 3.7@xmath204 & 1.6@xmath191 d. feldman , z. liu and p. nath , phys . rev . lett . * 99 * , 251802 ( 2007 ) . d. n. spergel _ et al . _ [ wmap collaboration ] , astrophys . j. suppl . * 170 * , 377 ( 2007 ) ; e. komatsu _ et al . _ [ wmap collaboration ] , astrophys . j. suppl . * 180 * , 330 ( 2009 ) . d. feldman , z. liu and p. nath , phys . b * 662 * , 190 ( 2008 ) . d. feldman , z. liu and p. nath , jhep * 0804 * , 054 ( 2008 ) . s. profumo and c. e. yaguna , phys . d * 69 * , 115009 ( 2004 ) ; phys . d * 70 * , 095004 ( 2004 ) s. profumo , phys . d * 72 * , 103521 ( 2005 ) . n. bernal , a. djouadi and p. slavich , jhep * 0707 * , 016 ( 2007 ) . i. gogoladze , r. khalid and q. shafi , arxiv:0903.5204 [ hep - ph ] . h. baer , k. m. cheung and j. f. gunion , phys . d * 59 * , 075002 ( 1999 ) ; s. raby and k. tobe , nucl . b * 539 * , 3 ( 1999 ) . h. baer , a. mustafayev , e. k. park , s. profumo and x. tata , jhep * 0604 * , 041 ( 2006 ) . nath and r. arnowitt , phys . d * 56 * , 2820 ( 1997 ) ; j. r. ellis , k. a. olive and y. santoso , phys . b * 539 * , 107 ( 2002 ) [ arxiv : hep - ph/0204192 ] ; h. baer , a. mustafayev , s. profumo , a. belyaev and x. tata , jhep * 0507 * , 065 ( 2005 ) [ arxiv : hep - ph/0504001 ] ; u. chattopadhyay and d. das , arxiv:0809.4065 [ hep - ph ] . see @xcite for further references specifically on the nonuniversality of the gaugino masses . a. h. chamseddine , r. arnowitt and p. nath , phys . * 49 * ( 1982 ) 970 ; p. nath , r. l. arnowitt and a. h. chamseddine , nucl . b * 227 * , 121 ( 1983 ) . l. hall , j. lykken and s. weinberg , phys . rev . * d27 * , 2359 ( 1983 ) ; for historical reviews see : h. p. nilles , phys . rept . * 110 * , 1 ( 1984 ) ; p. nath , arxiv : hep - ph/0307123 . a. arvanitaki , c. davis , p. w. graham , a. pierce and j. g. wacker , phys . d * 72 * , 075011 ( 2005 ) ; a. arvanitaki , s. dimopoulos , a. pierce , s. rajendran and j. g. wacker , phys . d * 76 * , 055007 ( 2007 ) [ arxiv : hep - ph/0506242 ] ; a. alves , o. eboli and t. plehn , phys . d * 74 * , 095010 ( 2006 ) ; j. alwall , m. p. le , m. lisanti and j. g. wacker , phys . b * 666 * , 34 ( 2008 ) ; phys . rev . d * 79 * , 015005 ( 2009 ) ; j. alwall , s. de visscher and f. maltoni , jhep * 0902 * , 017 ( 2009 ) ; g. l. kane , a. a. petrov , j. shao and l. t. wang , arxiv:0805.1397 [ hep - ph ] . b. s. acharya , p. grajek , g. l. kane , e. kuflik , k. suruliz and l. t. wang , arxiv:0901.3367 [ hep - ph ] . ellis , k. enqvist , d.v . nanopoulos and k. tamvakis , phys . b * 155 * , 381 ( 1985 ) ; m. drees , phys . b * 158 * , 409 ( 1985 ) ; g. anderson , c.h . chen , j.f . gunion , j.d . lykken , t. moroi and y. yamada , [ hep - ph/9609457 ] . a. corsetti and p. nath , phys . d * 64 * , 125010 ( 2001 ) ; u. chattopadhyay and p. nath , phys . d * 65 * , 075009 ( 2002 ) ; a. birkedal - hansen and b. d. nelson , phys . d * 67 * , 095006 ( 2003 ) ; u. chattopadhyay and d. p. roy , phys . d * 68 * , 033010 ( 2003 ) ; d. g. cerdeno and c. munoz , jhep * 0410 * , 015 ( 2004 ) ; g. belanger , f. boudjema , a. cottrant , a. pukhov and a. semenov , nucl . b * 706 * , 411 ( 2005 ) ; h. baer , a. mustafayev , e. k. park , s. profumo and x. tata , jhep * 0604 * , 041 ( 2006 ) ; k. choi and h. p. nilles jhep * 0704 * ( 2007 ) 006 ; i. gogoladze , r. khalid , n. okada and q. shafi , arxiv:0811.1187 [ hep - ph ] ; s. bhattacharya , a. datta and b. mukhopadhyaya , phys . d * 78 * , 115018 ( 2008 ) ; m. e. gomez , s. lola , p. naranjo and j. rodriguez - quintero , arxiv:0901.4013 [ hep - ph ] ; b. altunkaynak , p. grajek , m. holmes , g. kane and b. d. nelson , arxiv:0901.1145 [ hep - ph ] ; u. chattopadhyay , d. das and d. p. roy , arxiv:0902.4568 [ hep - ph ] ; s. bhattacharya and j. chakrabortty , arxiv:0903.4196 [ hep - ph ] . martin , arxiv:0903.3568 [ hep - ph ] . a. djouadi , j. l. kneur and g. moultaka , comput . commun . * 176 * , 426 ( 2007 ) ; present version 2.41 ( 2008 ) . b. c. allanach , comput . commun . * 143 * , 305 ( 2002 ) [ arxiv : hep - ph/0104145 ] ; present version 3.02 ( 2009 ) . w. porod , comput . commun . * 153 * , 275 ( 2003 ) ; present version 2.23 ( 2005 ) . g. belanger , f. boudjema , a. pukhov and a. semenov , comput . commun . * 176 * , 367 ( 2007 ) ; g. belanger , f. boudjema , a. pukhov and a. semenov , comput . commun . * 174 * , 577 ( 2006 ) ; g. belanger , f. boudjema , a. pukhov and a. semenov , comput . commun . * 149 * ( 2002 ) 103 ; present version 2.2 cpc . a. pukhov , arxiv : hep - ph/0412191 . e. komatsu _ et al . _ [ wmap collaboration ] , astrophys . j. suppl . * 180 * , 330 ( 2009 ) . g. degrassi , p. gambino and g. f. giudice , jhep * 0012 * ( 2000 ) 009 ; d. a. demir and k. a. olive , phys . d * 65 * , 034007 ( 2002 ) [ arxiv : hep - ph/0107329 ] ; a. j. buras et.al . b * 659 * ( 2003 ) 3 ; m. e. gomez , t. ibrahim , p. nath and s. skadhauge , phys . d * 74 * ( 2006 ) 015015 ; g. degrassi , p. gambino and p. slavich , phys . b * 635 * ( 2006 ) 335 . e. barberio _ et al . _ [ heavy flavor averaging group ] , arxiv:0808.1297 [ hep - ex ] . m. misiak _ et al . _ , phys . * 98 * ( 2007 ) 022002 . s. r. choudhury and n. gaur , phys . b * 451 * , 86 ( 1999 ) ; k. s. babu and c. f. kolda , phys . rev . lett . * 84 * , 228 ( 2000 ) [ arxiv : hep - ph/9909476 ] ; c. bobeth , t. ewerth , f. kruger and j. urban , phys . d * 64 * , 074014 ( 2001 ) ; r. l. arnowitt , b. dutta , t. kamon and m. tanaka , phys . b * 538 * , 121 ( 2002 ) [ arxiv : hep - ph/0203069 ] ; a. dedes , h. k. dreiner , u. nierste and p. richardson , arxiv : hep - ph/0207026 . j. k. mizukoshi , x. tata and y. wang , phys . d * 66 * , 115003 ( 2002 ) [ arxiv : hep - ph/0208078 ] . t. ibrahim and p. nath , phys . d * 67 * , 016005 ( 2003 ) [ arxiv : hep - ph/0208142 ] . t. aaltonen _ et al . _ [ cdf collaboration ] , phys . lett . * 100 * , 101802 ( 2008 ) [ arxiv:0712.1708 [ hep - ex ] ] . a. djouadi , m. drees and j. l. kneur , jhep * 0603 * , 033 ( 2006 ) . g. abbiendi _ et al . _ [ opal collaboration ] , eur . j. c * 35 * , 1 ( 2004 ) . r. barate _ [ lep working group for higgs boson searches ] , phys . b * 565 * , 61 ( 2003 ) ; the aleph , delphi , l3 and opal collaborations , lhwg - note 2005 - 01 . v. m. abazov _ et al . _ [ d0 collaboration ] , phys . b * 660 * , 449 ( 2008 ) . t. aaltonen _ et al . _ [ cdf collaboration ] , arxiv:0811.2512 [ hep - ex ] . a. belyaev , s. dar , i. gogoladze , a. mustafayev and q. shafi , arxiv:0712.1049 [ hep - ph ] . r. l. arnowitt , b. dutta , a. gurrola , t. kamon , a. krislock and d. toback , phys . * 100 * ( 2008 ) 231802 ; r. l. arnowitt _ et al . _ , b * 649 * , 73 ( 2007 ) . d. feldman , z. liu and p. nath , aip conf . proc . * 1078 * , 116 ( 2009 ) . n. bhattacharyya , a. datta and s. poddar , phys . d * 78 * , 075030 ( 2008 ) . d. feldman , z. liu and p. nath , phys . d * 78 * , 083523 ( 2008 ) . s. biswas and b. mukhopadhyaya , arxiv:0902.4349 [ hep - ph ] . k. griest and d. seckel , phys . d * 43 * , 3191 ( 1991 ) . k. l. chan , u. chattopadhyay and p. nath , phys . d * 58 * , 096004 ( 1998 ) ; j. l. feng , k. t. matchev and t. moroi , phys . rev . lett . * 84 * , 2322 ( 2000 ) ; u. chattopadhyay , a. corsetti and p. nath , phys . d * 68 * , 035005 ( 2003 ) ; h. baer , c. balazs , a. belyaev , t. krupovnickas and x. tata , jhep * 0306 * , 054 ( 2003 ) . s. p. martin and p. ramond , phys . d * 48 * , 5365 ( 1993 ) [ arxiv : hep - ph/9306314 ] . h. baer , v. barger , g. shaughnessy , h. summy and l. t. wang , phys . d * 75 * , 095010 ( 2007 ) . s. p. martin , phys . d * 75 * , 115005 ( 2007 ) . g. l. kane and j. p. leveille , phys . b * 112 * , 227 ( 1982 ) ; p. r. harrison and c. h. llewellyn smith , nucl . b * 213 * , 223 ( 1983 ) ; e. reya and d. p. roy , phys . b * 141 * , 442 ( 1984 ) ; s. dawson , e. eichten and c. quigg , phys . d * 31 * , 1581 ( 1985 ) . w. beenakker , r. hopker , m. spira and p. m. zerwas , nucl . b * 492 * , 51 ( 1997 ) . . a. djouadi , m. m. muhlleitner and m. spira , acta phys . b * 38 * , 635 ( 2007 ) ; m. muhlleitner , a. djouadi and y. mambrini , comput . commun . * 168 * , 46 ( 2005 ) ; a. djouadi , j. kalinowski and m. spira , comput . commun . * 108 * , 56 ( 1998 ) . w. beenakker , m. klasen , m. kramer , t. plehn , m. spira and p. m. zerwas , phys . lett . * 83 * , 3780 ( 1999 ) ; w. beenakker , r. hopker and m. spira , arxiv : hep - ph/9611232 . skands _ et al . _ , jhep * 0407 * , 036 ( 2004 ) . pgs-4 ; j. conway et . al , cdf : this code simulates lhc detector effects , with a tracking system , em and hadronic calorimetry , and a muon system , taking events simulated with pythia @xcite and reconstructs photons , electrons / muons , hadronically decaying taus , and hadronic jets . g. l. bayatian _ et al . _ [ cms collaboration ] , j. phys . g * 34 * , 995 ( 2007 ) . h. e. haber and g. l. kane , nucl . b * 232 * , 333 ( 1984 ) ; e. ma and g. g. wong , mod . lett . a * 3 * , 1561 ( 1988 ) ; h. baer , r. m. barnett , m. drees , j. f. gunion , h. e. haber , d. l. karatas and x. r. tata , int . j. mod . phys . a * 2 * , 1131 ( 1987 ) ; h. baer , x. tata and j. woodside , phys . d * 42 * , 1568 ( 1990 ) . m. toharia and j. d. wells , jhep * 0602 * , 015 ( 2006 ) . h. baer , r. m. barnett , m. drees , j. f. gunion , h. e. haber , d. l. karatas and x. r. tata , int . j. mod . phys . a * 2 * , 1131 ( 1987 ) . r. h. k. kadala , p. g. mercadante , j. k. mizukoshi and x. tata , eur . j. c * 56 * , 511 ( 2008 ) . v. n. lebedenko _ et al . _ , arxiv:0812.1150 [ astro - ph ] . z. ahmed _ et al . _ [ cdms collaboration ] , phys . lett . * 102 * , 011301 ( 2009 ) . j. angle _ et al . _ [ xenon collaboration ] , phys . lett . * 100 * , 021303 ( 2008 ) . see r. gaitskell , j. filippini , et al . http://dendera.berkeley.edu/plotter/entryform.html based on : supercdms ( projected ) 2-st@soudan , supercdms proposal , supercdms ( projected ) 25 kg ( 7-st@snolab),supercdms proposal , lux proposal , 300 kg lxe projection .

we discuss and extend recent conjectures relating partial null limits of correlation functions of local gauge invariant operators and the expectation value of null polygonal wilson loops and local gauge invariant operators . we point out that a particular partial null limit provides a strategy for the calculation of the anomalous dimension of short twist - two operators at weak and strong coupling . + 0.6 cm _ _ department of physics , the pennsylvania state university , + university park , pa 16802 , usa

1110.0758

increasingly efficient perturbative computational techniques for higher - order calculations and increasingly efficient use of integrability of the worldsheet theory in @xmath0 have brought about , in the last few years , remarkable progress in our understanding of @xmath1 super - yang - mills ( sym ) theory and have exposed unexpected and fascinating relations between _ a priori _ unrelated quantities . such a connection , proposed in @xcite , links gluon scattering amplitudes with the expectation value of certain null polygonal wilson loops in @xmath1 sym theory . initially suggested at strong coupling based on the ads / cft correspondence , this relation was generalized and successfully tested at weak coupling @xcite as well . correlation functions of local gauge invariant operators are natural observables in a conformal field theory such as @xmath2 sym theory . two - point functions are determined by the dimension of the operators ; since the latter are known , at least in principle , through use of integrability , so are the two - point functions . apart from the dimension and @xmath3 charges of operators , three point functions are determined by additional coupling - constant dependent `` structure functions '' whose evaluation is less clear . it is possible to argue that , if some of the operators carry large quantum numbers , the calculation can be carried out in a semiclassical expansion @xcite . even though the axioms of conformal field theories guarantee that higher - point correlation functions are determined by the two- and three - point functions , an explicit evaluation along this line is not straightforward . it is therefore interesting to devise methods to directly evaluate them either for generic position of operators or in special limits . it was recently suggested @xcite that the duality between wilson loops and scattering amplitudes can be extended to include certain special classes of correlation functions . more precisely , for operators in the stress tensor multiplet , the following relation should hold in the planar limit : _ x_i , i+1 ^ 20 = = [ allks ] where @xmath4 is a null polygonal wilson loop with corners at positions @xmath5 with @xmath6 . initially proposed as a relation between correlation functions of bosonic operators , bosonic wilson loops and mhv amplitudes , this triality conjecture was extended to supersymmetric correlators , supersymmetric wilson loops and generic superamplitudes @xcite . this relation was proven at the level of the unregularized integrand in @xcite and shown to hold in explicit examples @xcite in the presence of a dimensional regulator . it was moreover suggested and explicitly demonstrated for four - point correlation functions though two - loop order @xcite that the relation ( [ allks ] ) is not restricted to operators in the tensor multiplet , but rather holds for more general 1/2-bps operators , including operators with large quantum numbers . another class of interesting observables is provided by the correlation function of wilson loops and local operators . such quantities are interesting for several reasons . for example , they characterize the expansion of a wilson loop in local operators = _ ic_i ( ) o_i(x ) ; the coefficients @xmath7 may be found in the obvious way in terms of the correlation function @xmath8 and the two - point function @xmath9 . moreover , they may be used to factorize the expectation value of a product of two wilson loops or , if the coefficients @xmath7 are known , to simply evaluate this expectation value : & = & _ i , jc_i()c_j( ) _ i(x)o_j(y)&=&_ic_i ( ) _ i(x ) w()&=&_i , j_ij(x - y)w()o_i(x ) _ j(y ) w( ) , [ factorization ] where @xmath10 are the relevant conformal blocks . a second motivation for analyzing correlation functions of wilson loops and local operators is that , for a special choice of operator ( given by the chiral lagrangian ) , they contain information about the higher - loop corrections to the expectation value of the wilson loop . this is akin to the lagrangian insertion formalism @xcite used in @xcite to evaluate loop corrections to correlation functions of local operators in the null separation limit . a generalization of the relation between correlation functions of operators and null polygonal wilson loops was recently proposed in @xcite . more precisely , starting with an @xmath11-point correlation function and taking the limit in which @xmath12 points are sequentially null - separated one should find that _ x_ii+1 ^ 20 ~ [ abt ] where the expectation values on the right - hand side are taken in the fundamental representation . non - vanishing values for both correlators , @xmath13 and @xmath14 , require that the total r - charges of the product of operators in the two correlators are zero . it therefore follows that the operator @xmath15 must have vanishing r - charges.@xmath16 , the correlation function acquires the interpretation of form factor at zero momentum . indeed , it was argued in @xcite that form factors may be interpreted in terms of the expectation value of certain zig - zag wilson loops . if the position of the operator is integrated over , _ i.e. _ if the momentum inflow though it vanishes , the zigzag wilson line becomes closed . ] generalizations of ( [ abt ] ) are fairly straightforward to formulate . for example , _ x_ii+1 ^ 20 ~ [ ww ] where the upper index denotes a restriction to the connected part of the correlation functions and the points @xmath17 with @xmath18 are generic . one may moreover consider a second limit in which the points in this set become sequentially null separated while remaining at generic positions with respect to the points @xmath5 , @xmath6 . in this limit the original correlation function equals the correlation function of two null polygonal wilson loops . ratios of the type ( [ abt ] ) , ( [ ww ] ) are good observables . indeed , as discussed in detail in @xcite , in the null limit correlators develop the same type of singularities as null polygonal wilson loops . since these divergences are located around the wilson loop cusps or , alternatively , around the operators that are sequentially null separated , in the ratios ( [ abt ] ) and ( [ ww ] ) these divergences cancel out leaving behind a finite quantity , which should exhibit the symmetries of the theory . in particular , the conformal symmetry of the correlation functions should be realized . it would be interesting to understand whether the additional operator insertions in eqs . ( [ abt ] ) and ( [ ww ] ) have any interpretation in terms of scattering amplitudes when the operators position is not integrated over . here we will prove , to all orders in a weak coupling expansion and in the regularized theory with @xmath19 and take the null separation limit for generic @xmath20 . since we will keep the complete @xmath20 dependence , our arguments hold in all dimensions . ] , the first equality in equation ( [ allks ] ) as well as equations ( [ abt ] ) and ( [ ww ] ) for all twist-2 operators and for a finite rank of an @xmath21 gauge group . is replaced with @xmath22 , as mentioned in @xcite . ] we will also discuss two - field operators outside this class , containing fermions and field strengths . while not identical in details , our strategy will be similar in spirit with the eikonal line arguments used in @xcite ; we will separate the feynman diagrams contributing to correlation functions into classes depending on whether or not there exists r - charge flow between operators and show that certain sequences of propagators between null - separated points are equivalent to null wilson lines . we will also show that diagrams in which there is no r - charge exchange between charged fields at null - separated points have softer singularities than if r - charge flow is present . we will then extend these arguments to larger classes of operators and also to larger classes of correlation functions , in which not all points are sequentially null - separated ; we will refer to them as `` partial null limits '' . we will also see that the relation between correlation functions in the null separation limit and the expectation value of null polygonal wilson loops is not restricted to four - dimensional gauge theories but rather holds in all dimensions . the arguments in this note point to a generalizations of ( [ abt ] ) and ( [ ww ] ) to limits in which any operator is null - separated from at least one and at most two other operators . in this limit a correlation function reduces to the correlation function of certain non - local operators built out of fundamental fields and open wilson lines . let us consider the correlator of two @xmath23 $ ] and two @xmath24 $ ] operators at positions @xmath25 in the limit @xmath26 and all other distances being nonzero . denoting by w(x , y)=pe^_x^y a , it is not difficult to see that _ ( x_1)o(x_2)o(x_3)o(x_4 ) . [ corropenwl ] clearly , the open wilson lines are null . the proportionality coefficient depends on the regularization scheme and on the order of limits . this generalization provides a direct link between four - point correlators of bps operators and two - point functions of non - bps operators . indeed , each non - local operators @xmath27 $ ] may be expanded in twist - two operators thus reducing the four - point correlator reduces to a superposition of two - point functions . making explicit this decomposition should allow one to read off the anomalous dimensions of twist - two operators . this approach may be particularly efficient for the lowest twist - two operators @xmath28 $ ] which is a member of the konishi multiplet and may offer an alternative approach to the calculation of anomalous dimensions of short operators at strong coupling . the rest of this note is organized as follows . in [ generalities ] we discuss general features of correlation functions , define the regularization scheme and the null limit , and outline the proof of relations ( [ allks ] ) , ( [ abt ] ) and ( [ ww ] ) . later sections contain some of the details completing this proof . we proceed in [ no_insertions ] to discuss the correlation function of 2-field scalar operators in the stress tensor multiplet in the null separation limit and identify the relevant feynman diagrams that contribute in this limit . we then proceed in [ w_insertion ] to extend the discussion in [ no_insertions ] to the case of @xmath11-point correlation function with @xmath12 null - separated points . we will finish this section with comments on the more general correlators of the type mentioned in equation ( [ ww ] ) . in [ large_charge ] we discuss twist-2 operators with higher spin and extend the results described in the previous sections to their correlation functions . in [ comments ] we comment on other weak and strong coupling features of the correlator / wilson loop relations . appendix [ opotherfields ] details the null and partial - null separation limit of correlation functions of two - field operators constructed from fermions and gauge fields . let us consider the correlation function of some number of operators of length 2 in a gauge theory with an @xmath21 gauge group ; we will keep @xmath29 arbitrary and discuss the large @xmath29 limit at the end . such a correlator is symmetric under the permutation of positions of identical operators ; for example , the correlation function of @xmath12 operators in the @xmath30 representation of @xmath31 will be symmetric under the permutation of positions of all operators . taking the limit in which the operators are null - separated requires choosing a specific sequence of positions , _ e.g. _ @xmath32 and setting @xmath33 . this choice breaks their permutation symmetry to one of its cyclic subgroups ; different choices of sequences will lead , in this limit , to different dominant terms which , through the correlator / amplitude relation @xcite , are related to ( squares of ) different color - ordered amplitudes . to take the limit in which ( some ) operators are null - separated it is useful to start with the momentum space correlation function and fourier - transform it to position space . in momentum space each operator @xmath34 carries nontrivial momentum + @xmath35 which is split between the fields composing it , @xmath36 , as shown in fig . [ general ] . one may arbitrarily choose a sequence of propagators connecting adjacent points ; denoting this sequence by @xmath37 with @xmath38 the momenta of the lines attaching this sequence to the rest of the diagram and @xmath39 the corresponding indices , and by @xmath40 the green s function obtained by removing all @xmath41 , it is easy to see that _ 1(q_1)_n(q_n)= _ i=1^n_j=1 ^ 2d^dp_ij[d^dk]^d(q_i - p_i1-p_i2 ) l(p_ij , p_i+1,j , k)_a_i g(k)_a_1 a_n , where the measure factor @xmath42 $ ] stands for integration over the momenta of all external lines of @xmath43 . the factors @xmath44 may contain momentum factors arising from derivatives present in operators . fourier - transforming back to position space implies that _ 1(x_1)_n(x_n)= _ i=1^n _ j=1 ^ 2d^d p_ij _ x_il(x_i , x_i+1,k)_a_i_x_i+1 g(k)_a_1 a_n , [ general_form ] where @xmath45 are differential operators which are present if @xmath34 contain derivatives and ( x_1 , x_2,k)_a_i = d^dp_1 d^dp_2 e^ip_1x_1 + ip_2x_2 l(p_1 , p_2,k)_a_1 . the presentation ( [ general_form ] ) of correlation functions is quite general and does not assume any specific structure for the operators @xmath34 apart from their two - field structure . in the following we will mainly restrict to operators carrying nontrivial r - charge . by considering all possible choices of @xmath46 we will identify the one that is dominant _ i.e. _ it has the strongest singularity in the limit @xmath47 . we will moreover see that this @xmath46 may be interpreted in terms of a null wilson line with fields attached to it . , height=170 ] while this appears to be a classical computation , it may be promoted to an all - loop one in a regularization scheme in which the ( partial ) null - separation limit can be decoupled from the integrals over the internal momenta @xmath38 . such a scheme indeed exists : it suffices to take the limit @xmath33 such that in the null separation limit , where @xmath48 is the mass scale introduced by dimensional regularization . ] |x_i , i+1|^2 k^2 0 [ def_scheme ] for all possible internal momenta @xmath38 . with this assumption , the null separation limit can be taken at the level of the ( regularized ) integrand . in the following sections we will show that , in this case , a sequence of scalar propagators with gluons attached to in though three - point vertices reduces in the null separation limit to a null wilson line with the same number of attached gluons ; this yields the desired results , to all orders in weak coupling perturbation theory . as discussed in @xcite , in other schemes this is only a proportionality relation . as outlined here , the arguments in the following sections focus on only two operators at a time and justify the appearance of a wilson line between their insertion points in the limit in which they are null - separated . there arguments will also support the relation between correlation functions in the limit in which @xmath49 but @xmath50 for some subset of operators and the correlation function of appropriately capped open null zig - zag wilson lines , as illustrated in eq . ( [ corropenwl ] ) . correlation functions of operators carrying r - charge vanish unless the total r - charge vanishes . if the r - charge of subsets of operators adds up to zero , then one may organize the feynman diagrams into classes labeled by the number of operators among which there exists nontrivial r - charge flow . diagrammatically , between any operators connected by an r - charge flow there exists a sequence of propagators of fields charged under r - symmetry . for example , the correlator ( x_1)(x_2)(x_3)(x_4)will have two such classes . in one of them any one of the four operators is connected to two others by some sequence of scalar or fermion lines , while in the second class any one of the four operators is connected to only one other operator by two distinct sequences of scalar of fermion lines . with the strategy outlined in the previous section , let us now proceed to prove the first equality ( [ allks ] ) for charged 2-field bps operators to all orders in perturbation theory and in the presence of a dimensional regulator . we will begin by showing that , in the limit of null separation , the diagrams exhibiting a connected r - charge flow dominate over the diagrams with disconnected flows and that the former reduce to the expectation value of a wilson loop in the adjoint representation . moreover , it will turn out that the dominant diagrams will contain a continuous sequence of scalar propagators . let us consider a sequence of @xmath12 one - gluon vertices connected by scalar propagators , as shown in fig . [ manyssg ] . the momenta going to the two endpoints are denoted @xmath51 and @xmath52 while the momenta carried by gluons are denoted by @xmath53 and the momenta of the scalars between vertices are denoted by @xmath54 . throughout we will not write explicitly the propagator of the gluons attached to the vertices between the points @xmath55 and @xmath56 . the starting expression is thus : & & l(x_1 , x_2 , k_1 , , k_n)__1_n = d^dp_1d^dp_2e^ip_1x_1+ip_2x_2d^dq_1d^dq_n-1 + & & [ many_gss_ini ] + & & ^(d)(k_1+q_1+p_1)^(d)(k_2-q_1+q_2)^(d)(k_n-1-q_n-2+q_n-1)^(d)(k_n - q_n-1+p_2 ) , height=71 ] by performing all the integrals except the one over @xmath51 , schwinger - parametrizing the propagators and representing the factors of @xmath51 in the numerator as derivatives with respect to @xmath55 it follows that & & l(x_1 , x_2 , k_1 , , k_n)__1_n = d^dp_1 + & & = ( -i)^(n+1)_i=1^n ( -2i+2_j=1^i-1k_j_i+k_i_i ) ( _ i=1^n+1_0^d_i)e^-i_ik_ix_2 + & & d^dp_1e^ip_1(x_1-x_2)+i_i=1^n+1_i(p_1+_j=1^i-1k_j)^2 - 0_i . a change similar to the one done in the 1-vertex example allows us to further simplifies this expression . more precisely , we change the integration variables such that @xmath57 with @xmath58 and @xmath59 , while @xmath60 . the jacobian for this change of variables is @xmath61 . the change of variables @xmath57 with @xmath58 , @xmath59 and @xmath60 further simplifies this expression . moreover , writing @xmath62 for @xmath6 and with @xmath63 , the unit sum constraint on the @xmath64 variables becomes just @xmath65 . together with the evaluation of the @xmath51 integral , these transformations lead to : [ many_gss ] & & l(x_1 , x_2 , k_1 , , k_n)__1_n = ( -i)^(n+1)_i=1^n ( -2i+2_j=1^i-1k_j_i+k_i_i ) + & & ( _ i=1^n_t_i-1 ^ 1 dt_i)e^-i_ik_i(x_2t_i - x_1(1-t_i ) ) _ 0^d^ne^i(t_i , k_i)-i-0 - 0/ , with some functions @xmath66 whose expressions are not important . the integral over @xmath67 is of the general type [ allints ] i_m(z , f)=d ^m-2+e^if - i-0 - 0/ with some choice of @xmath68 and with some function @xmath69 depending on the momenta @xmath53 and the affine parameters @xmath70 ; such integrals will also appear in later sections in diagrams involving other types of fields . the rescaling @xmath71 together with the null - separation limit implies that _ z^20 i_m(z , f ) = ( -1)^(1-m-)z^2(m+-1 ) [ generic_int ] for @xmath72 . for @xmath73 the integral does not vanish , but reduces to an @xmath69-dependent ( and @xmath74-independent ) constant . the leading term in the light - like limit arises from the highest power of @xmath75 brought down by the differentiation with respect to @xmath55 . it is easy to see that it is @xmath76 ; this factor cancels the @xmath61 measure factor arising form the change of variables and leaves behind an @xmath12-independent @xmath67 factor which generates , upon integration , a factor of the position space scalar propagator @xmath77 . is introduced here as a schwinger parameter . ] the remaining integral represents @xmath12 correctly ordered wilson line vertices together with the exponential factors needed to fourier - transform the gluon propagator : & & l(x_1 , x_2 , k_1 , , k_n)__1_n&&=_4 - 2(x_1-x_2 ) _ t_i-1 ^ 1 dt_ie^-i_ik_i(x_2t_i - x_1(1-t_i ) ) , [ many_gss_final ] where , as mentioned before , @xmath63 . we have therefore recovered , in the scheme used here , the results of the eikonal approximation : a continuous sequence of scalar propagators connecting two operators that become light - like separated is equivalent to a light - like wilson line between the operators positions . factor which makes the right - hand side of eq . ( [ many_gss_final ] ) vanish in the null separation limit . ] the derivation described above points to a simple rule for identifying the diagrams that survive in the light - like limit : once all propagators are written with standard @xmath78 denominators , one simply counts the number of momentum factors of lines connecting the two insertion points . if this number is equal to the number of propagators minus one , then the corresponding diagram yields a sufficiently singular contribution . otherwise it drops out in the null separation limit . ) the measure factor @xmath61 counts the number of propagators minus one and the derivative factors count the number of momenta along the path connecting the two insertion points . each derivative generates a @xmath75 factor ; to have a sufficiently singular contribution it is necessary , according to ( [ generic_int ] ) , that these two factors cancel out or yield some negative power of @xmath67 . ] for this reason the four - point gluon - scalar interaction can not contribute to the leading term in the null - separation limit : since such a vertex has no derivatives , replacing a three - point vertex in eq.([many_gss_ini ] ) with one such vertex would lead to an eq . ( [ many_gss ] ) with one fewer derivative factor and hence to an extra factor of @xmath67 in the integral . it will therefore not be sufficiently singular to contribute in the limit in which r - charge flows between light - like separated points . apart from a continuous sequence of scalar propagators , r - charge flow between two operators may be realized by a sequence of scalars and fermions . we illustrate this possibility in fig . [ many_and_sff ] with two scalars and any number of fermions ; the sequence of fermion propagators may also be interrupted by further scalar lines , in which case more of the open - ended lines become scalars . , height=68 ] the relation between the singularity in the null separation limit and the structure of the momentum dependence of the integrand of the fourier transform of such a sequence suggests that , for a fixed number of propagators , the most singular terms arise from the terms with the highest number of momentum factors in the numerator . this illustrates the importance of the spin of the exchanged particle in the null - separation limit . since no derivatives appear in the interaction of fermions and scalars and of fermions and gluons , the only source of numerator momenta in fig . [ many_and_sff ] is the numerator of fermion propagators . consequently , the diagrams with more than the two scalar propagators will have a softer singularity than the diagram shown in this figure . let us then discuss the situation of the highest number of fermion propagators @xmath79 for a total of @xmath12 lines . the relevant fourier - transform is & & l(x_1,x_2 , k_1 , , k_n)__1_n-2&=&d^d p_1 d^d p_2 e^ip_1x_1+ip_2x_2 d^d q_1 d^d q_n-1 + & & ^(d)(k_1+q_1+p_1)^(d)(k_2-q_1+q_2)^(d)(k_n-1-q_n-2+q_n-1)^(d)(k_n - q_n-1+p_2 ) . carrying out the @xmath80 integrals , schwinger - parametrizing the resulting propagators and changing variables as discussed in the previous section leads to an expression of the type & & l(x_1,x_2 , k_1 , , k_n)__1_n-2&~&(_i=1^n_t_i-1 ^ 1 dt_i)e^-i_ik_i(x_2t_i - x_1(1-t_i ) ) + & & t^_n-1_n-2 _1_1_0^d^ne^i(t_i , k_i)-i-0 - 0/ , where @xmath81 is a product of dirac matrices . it is easy to see that , among the integrals defined in ( [ allints ] ) , only @xmath82 with @xmath73 can appear from the action of derivatives . as expected , it follows from ( [ generic_int ] ) that all such diagrams are not sufficiently singular in the limit of null separation and thus can be ignored . it is clear that a similar argument goes through if some of the fermion lines are replaced with scalar lines ; such contributions can also be ignored . the typical diagram that does not exhibit r - charge flow between two charged operators involves at least one gluon propagating along any possible path between them see fig . [ noflow ] . the sequence of gluon propagators may be interrupted by scalar or fermion lines . since the fermion - gluon interaction has no derivatives , the discussion in the previous section implies that such diagrams are not sufficiently singular in the null - separation limit and we can ignore them . we will therefore focus on gluons and scalars and , for the same reason as above , also ignore the four - point interactions . , height=71 ] the momentum dependence of the gluon - scalar three - point vertices guarantees that feynman diagrams of the type shown in fig . [ noflow ] have the correct momentum dependence to potentially yield leading order contributions in the limit of null separation of insertion points ; this would be the case if , upon fourier transform , each vertex contributes a factor of @xmath83 . such a relation also appeared in a closer inspection reveals however that these potentially dangerous diagrams do not contribute . to see this let us assume that there are @xmath12 vertices and @xmath11 propagators between insertion points . of the external lines not attached to the insertion points , at least two are scalars , leaving only at most @xmath79 outgoing gluon lines and consequently at most @xmath79 free lorentz indices . since there are @xmath12 factors of @xmath83 , one for each vertex , it immediately follows that at least two such factors are necessarily contracted ( otherwise there would be more than @xmath84 free indices ) and at least one more @xmath85 numerator factor is generated . , _ i.e. _ additional propagator - like factors . ] since the feynman diagrams contributing to fig . [ noflow ] depend on a number of additional vectors _ e.g. _ the momenta of external particles one may wonder whether two of the indices arising from the @xmath12 vertices may be contracted with such vectors rather than produce an @xmath85 . such contributions can be ruled out by noticing , on the one hand dimensional analysis requires that the number of numerator factors be fixed by a factor of @xmath38 . ] and on the other factors of external momenta lower the singularity through the absence of @xmath75 factors . it therefore follows that , regardless of the mechanism that is responsible for the reduction of the number of free vector indices either contraction with some @xmath53 or the appearance of an explicit @xmath85 factor the null separation limit @xmath86 contains no singularities . this completes the proof that , if a connected r - charge flow can exist between scalar operators , then only feynman diagrams leading to this flow contribute in the null separation limit . we have therefore shown that , in the presence of a dimensional regulator and in the scheme described in [ generalities ] , the correlation function of length-2 scalar operators in the null separation limit is proportional to the expectation value of a null polygonal wilson loop in the adjoint representation with cusps at the positions of the original operators : ( x_1)(x_n)= w_n _ _ i=1^n(x_i - x_i+1 ) . in general , as explained in @xcite , a scheme - dependent coefficient function will appear on the right - hand side of this equation and will capture the differences between different possible light - like limits and their inter - relation with the regulator . the arguments presented here hold for finite values of the rank of the gauge group and do not rely on the existence of supersymmetry . our proof extends the arguments of @xcite to the regularized theory . we will comment in the appendix on correlation functions of operators built from other fields . let us now proceed to discuss more general limits of correlation functions , in which only a subset of operators participate in the null separation limit . let us first consider the correlation functions of @xmath87 operators , @xmath12 of which participate in the null separation limit . the additional operator @xmath15 is located at a generic position , @xmath88 for all @xmath89 . as discussed in the introduction , such an operator should have vanishing r - charge , otherwise a nonvanishing correlation function @xmath90 implies that @xmath91 . the arguments in the previous section apply equally well if the additional operator does not affect the r - charge flow between two light - like separated operators . indeed , if @xmath15 is connected to the scalar and/or fermion lines as in fig . [ extra1](a ) and [ extra1](b ) , then it acts similarly to any other multi - point interaction arising from the lagrangian : the continuous sequence of scalar propagators will continue to dominate in the presence of an additional operator insertion . a similar conclusion that the presence of the additional operator does not affect the conclusion of the previous section can be reached if the operators is inserted in the diagrams in fig . ( [ noflow ] ) while being attached to some of the fields denoted by open lines see fig . [ extra1](c ) . dimensional arguments also imply that diagrams that are subleading in the absence of the additional operator can not acquire a stronger singularity . , height=94 ] the other possible way of inserting the additional operator is by interrupting the sequence of propagators explicitly shown in figs . [ manyssg ] , [ many_and_sff ] and [ noflow ] , as illustrated in fig . [ extra2 ] for the feynman diagrams dominant in the null separation limit . one may intuitively expect that , since the insertion point @xmath92 is not null - separated from @xmath55 and @xmath56 , the limit @xmath93 is non - singular were integrated over . ] . a short calculation shows that this is indeed the case ; we shall illustrate it with the example of an operator @xmath15 with arbitrary number of derivatives inserted in a continuous sequence of scalar propagators . is non - singular . [ extra2],height=79 ] assuming that the operator is inserted at position @xmath92 , that the fields that are wick - contracted with the scalars in the original sequence of propagators carry @xmath94 and @xmath95 derivatives and using @xmath96 in eq ( [ many_gss ] ) it is easy to see that fig . [ extra2 ] will contribute the following to the correlation function of the @xmath87 operators : & & l(x_1 , a , x_2 , k_1 , , k_n , o)__1_n&=&(_a)^n_1 l(x_1,a , k_1 , , k_r)__1_r ( _ a)^n_2 l(a , x_2 , k_r+1 , , k_n)__r+1_n , where @xmath97 and @xmath97 stand for products of derivatives with potentially suppressed free indices and @xmath98 contains the fields of @xmath15 that are not contracted with the scalar line . neither one of the two @xmath99 factors is singular as @xmath86 . it therefore follows that these terms , as well as the others obtained by replacing some or all of the scalars with gluons and/or fermions give subleading contributions compared to the feynman diagrams in which @xmath100 is not interrupting the sequence of scalar propagators connecting two light - like separated points . thus , in the limit in which @xmath5 with @xmath6 are sequentially null separated , the correlator @xmath101 becomes proportional to the correlation function of the null wilson loop in the adjoint representation with corners at positions @xmath5 and the additional operator @xmath15 . it is not difficult to generalize the arguments in the previous section to the insertion of several operators which are placed at generic positions relative to the points @xmath5 with @xmath102 . in such case the only constraint stemming from r - charge conservation is that the total charge of the insertions vanishes : @xmath103 . if none of the operators interrupts the r - charge flow between two null - separated operators they act , as in the case of a single operator insertion , as a lagrangian interaction vertex . if a sequence of scalar propagators is interrupted by an additional operator then it becomes non - singular in the limit in which its beginning and end points are null separated . it therefore follows that [ partialnull ] _ x_ii+1 ^ 20 = . as before , the wilson loop is in the adjoint representation and the upper index denotes the fact that on both sides of this equation one should restrict to the connected part of the correlation functions . relaxing this constraint will replace the right - hand side numerator with a sum over connected components , each of which being the product of the correlation function of @xmath4 and some subset of @xmath104 and the correlation function of the remaining operators . for sufficiently many additional operator insertions one may take a further limit , in which @xmath105 but at generic positions compared to @xmath106 . dividing by @xmath107 it then follows that _ = 1 + . [ relww ] as before , the upper index on the right - hand side denotes a restriction to the connected part of that correlator ; the constant term is the contribution of one of the disconnected components of @xmath108 . the other disconnected components have a different singularity structure and their contribution to the ratio ( [ relww ] ) vanishes in the null separation limit . one may consider the factorization of the correlation function of wilson loops , as outlined in eq . ( [ factorization ] ) . from the perspective of eq . ( [ relww ] ) , it corresponds to a particular factorization of the correlation function of @xmath109 operators into the sum of products of @xmath87 and @xmath110-point correlation functions . the arguments described above are independent of the rank of the @xmath21 gauge group . to make contact with a strong coupling analysis it is necessary to consider the large-@xmath29 limit . in the absence of operator insertions this limit is standard _ = w_n_^2 . [ fact_adj ] diagrammatically , this relation arises from the fact that the feynman diagrams of the type shown in fig . ( [ largen])(a ) are subleading in the large @xmath29 limit compared to diagrams in figs . ( [ largen])(b ) . the graph in fig . [ largen](c ) may or may not contribute depending on the rest of the diagram . that is , the only feynman diagrams that survive in this limit are those in which the inner and outer index lines of the wilson loop are connected to themselves by some webs of vertices and propagators from the lagrangian . moreover , away from the wilson loop , the feynman diagram is planar . the double - lines denote wilson lines and the wavy line represent gluons . [ largen],height=113 ] it is not difficult to generalize this picture to the correlation function of a wilson loop and an operator , @xmath111 . the feynman diagrams that survive the large-@xmath29 limit are those in which the operator @xmath15 is connected only to one index line of the wilson loop . n w_no(a ) _ = 2w_n_w_no(a ) _ . the factor of @xmath112 arises because the operator may be connected either to the inner of the outer index line ( _ e.g. _ it may be inserted on the gluon line in figs . ( [ largen])(b ) and ( c ) ) . this factor is in agreement with the discussion in @xcite on the insertion of the ( integrated ) dilaton vertex operator . following a similar reasoning it is easy to take the planar limit for the correlator of a wilson loop and several operators . the feynman graphs can be organized into sets of diagrams in which some operators are connected to one wilson loop index line and the other operators are connected to the other wilson loop index in a planar way , _ i.e. _ [ opsninfty ] & & _ n w_no(a_1)(a_m ) _ + & = & _ k=0^m_\{j_1 , , j_k}\{1 , , m } w_no(a_j_1)(a_j_k ) _ w_no(a_j_k+1)(a_j_m ) _ , where the second sum runs over all subsets of @xmath113 with @xmath38 elements . if the total r - charge of the operators appearing in an expectation value is nonzero , @xmath114 , the corresponding term vanishes identically . for example , if @xmath115 the r - charges of the two operators must be equal in absolute value ; if their common value is nonvanishing , then all correlators involving one wilson loop and one operator vanish identically and , in the sum above , only the terms with @xmath116 and @xmath117 survive . last , the factorization of the connected part of the correlation function of two wilson loops follows a similar pattern : _ = 2w_n _ w_m_w_nw_m^ _ . [ wwlim ] this term arises form the @xmath116 and @xmath118 terms in eq . ( [ opsninfty ] ) . the other terms disappear in the null limit because there is no r - charge flow between operators ( in fact there are no fields that propagate between the operators in one set and the operators in the other ) and thus are subleading compared to eq . ( [ wwlim ] ) . as reviewed in [ intro ] , it was suggested in @xcite that the relation between correlation functions and wilson loops is not restricted to the operators in the stress tensor multiplet . if such a generalization is indeed correct , it is tempting to expect that it would hold in the presence of additional insertions as well . such a generalization , both in the absence and in the presence of additional operator insertions , is potentially very interesting . a relation between correlation functions and wilson loops suggests that , at least in the null separation limit and at string coupling , correlation functions should have a semiclassical description . while such an approach is difficult to justify for operators of small dimension ( even in the null separation limit ) , a justification is readily available @xcite for operators whose dimension scales like @xmath119 . the discussion in the previous sections can be extended without much difficulty to the case when the operators inserted at positions @xmath120 are twist-2 operators ^s=_n=0^s c_n ( ) , [ ops ] where the coefficients @xmath121 are determined by requiring that these operators have definite anomalous dimensions . at one loop , they are given @xcite in terms of the jacobi polynomials ^s = i^s(nd_1+nd_2)^sp_s^(0,0 ) ( ) |__1=_2 , where @xmath122 are covariant derivatives in the adjoint representation in the light - like direction specified by the vector @xmath12 . our regularization scheme implies however that the following arguments hold even if @xmath123 does not have definite anomalous dimension . as we saw in the previous sections , in the null separation limit , the feynman diagrams that dominate the correlation function of 2-field scalar operators generate r - charge flow between insertion points and contribute the scalar propagator @xmath124 multiplied by a wilson line stretched between @xmath55 and @xmath56 , see eq . ( [ many_gss ] ) . the additional derivatives present in the operators ( [ ops ] ) act either on the scalar propagator prefactors @xmath125 or on the wilson line factor in the usual way : ( nd)^pz(x_1)(nd)^mz(x_2)(nd_x_1)^p(nd_x_2)^m z(x_1)z(x_2 ) . the derivative action on @xmath125 yields terms that are more singular than @xmath125 . since the action of derivatives on the wilson line does not generate any @xmath126 singularities , it follows that the most singular term in the null separation limit arises solely from all derivatives acting on the scalar propagator prefactors : _ x_i , i+10^s_n(x_n)^s_1(x_1)&=&w_n_m_1=0^s_1_m_n=0^s_nc_m_1c_m_n _ i=1^n_x_i_x_i ( x_i - x_i+1 ) + & & + ( ) [ corr_tw2 ] where @xmath127 and the left derivative with respect to @xmath128 in the first term acts on the @xmath129 factor . compared to the contribution of a scalar line with the same number of external gluons @xmath130.[corner],height=94 ] the contribution of the gauge fields present in the covariant derivatives can be shown to also be subleading . to understand this it is useful to interpret the @xmath131 $ ] term in the covariant derivative as a standard @xmath132 vertex from which one strips off the derivative . thus , for the purpose of understanding the @xmath133 dependence of the gauge field contribution , we may simply analyze the expression in eq . ( [ many_gss_ini ] ) without the @xmath134 propagator and without the momentum contribution of the first gluon - scalar vertex see fig . [ corner](b ) . it is useful to recall that the final @xmath135 dependence is governed by the power of @xmath67 in equation ( [ many_gss ] ) ; there , @xmath61 in the measure arose from the propagators and each derivative generates a factor of @xmath75 . thus , the two diagrams in fig . [ corner ] will have identical singularities in the @xmath136 limit . this singularity however is softer than that of the diagram in which the derivative acts on the scalar field ; consequently , we may ignore the contribution of the gauge fields in the covariant derivatives . thus , eq . ( [ corr_tw2 ] ) reduces to _ x_i , i+10^s_1(x_1)^s_n(x_n)=w_n_m_1=0^s_1_m_n=0^s_nc_m_1c_m_n _ i=1^n_x_i_x_i ( x_i - x_i+1 ) . it is straightforward to extend the arguments above , along the lines of [ w_insertion ] , to the correlation function of @xmath87 operators , @xmath12 of which are taken to be null separated . we find & & _ x_i , i+10^s_1(x_1)^s_n(x_n ) o^s_a(a)= w_no^s_a(a ) _ + & & _ m_1=0^s_1_m_n=0^s_nc_m_1c_m_n _ i=1^n _ x_i _ x_i(x_i - x_i+1 ) . normalizing this expression to the correlation function of the @xmath12 null - separated operators it follows immediately that _ x_i , i+1 ^ 20= . for several operator insertions at generic positions one finds for the connected part of the correlation function that & & _ x_i , i+10 + & & = , in close analogy with the correlation function of two - field bps operators . let us comment here on correlation functions of higher - twist operators . in @xcite it was suggested and demonstrated through two loops for four - point correlators , that the relation between correlation functions and null polygonal wilson loops is not restricted to two - field bps operators or , more generally , operators in the stress tensor multiplet . to see this in the regularization scheme we employed here let us consider operators connected by two or more non - overlapping sequences of propagators allowing for r - charge flow . following the discussion in [ no_insertions ] , in the limit in which the two operators are null - separated , each such sequence is equivalent to a wilson line multiplied by a position space scalar . moreover , we have also seen that if fields other than gluons are attached to the wilson line , the singularity of the scalar propagator is softened . the same holds if two wilson lines between the same two null - separated points are connected by a gluon . indeed , since the interaction vertex between a wilson line and a gluon is proportional to @xmath137 , a gluon of this type necessarily yields a factor of @xmath138 which softens the singularity of the scalar propagator . we therefore conclude that , if two operators are connected by more than one r - charge carrying sequence of propagators , the leading contribution to the null separation limit comes from feynman graphs in which no field connects two such sequences . this in turn implies that , in the planar limit , _ x_i , i+1 ^ 20 = [ allksj ] where @xmath139 stands for a scalar operator with @xmath140 fields . in general , an additional scheme - dependent factor may appear on the right - hand side above . unlike the two - field scalar operators however , the argument above breaks down for a finite rank gauge group . this argument goes through unmodified if additional operators , placed at arbitrary positions , are present in the correlation function . indeed , as discussed in previous sections , if any of the additional operators affect the r - charge flow between the null - separated ones , the resulting feynman diagram will have a subleading singularity . in the diagrams in which the additional operators act as regular interaction vertices , the null separation limit proceeds as if they were absent , implying that & & _ + & & = _ k=0^m_\{j_1 , , j_k}\{1 , , m } w_no(a_j_1)(a_j_k ) _ w_no(a_j_k+1)(a_j_m ) _ . similarly to the partial null - separation limit of correlation functions of two - field operators ( [ partialnull ] ) , we expect that the scheme - dependent coefficient functions arising both in the numerator and denominator of the left - hand side of this equation cancel each other out . while the weak coupling arguments for the relation between correlation functions and null polygonal wilson loops are relatively straightforward , a detailed derivation of this relation remains mysterious from a strong coupling perspective . moreover , as already mentioned in @xcite for two additional operators , if the charges and dimensions of the local operators are not large enough to invalidate the semiclassical expansion , the correlation function between a wilson loop and some number of local operators such as the one on the right - hand side of eq . ( [ ww ] ) factorizes as w_no_1(a_1) o_m(a_m ) _ _ j=1^mw_no_j(a_j ) _ . if the operators @xmath141 are the gauge theory dual of the integrated dilaton vertex operators ( _ i.e. _ they are the @xmath1 action ) , a relation of this type can be understood at weak coupling @xcite by simply representing @xcite @xmath141 as @xmath142 . one may also expect that a similar relation may hold if the integrals over the positions of the operators are omitted . it is however not immediately clear how to understand such a relation if the operators @xmath141 are other bps operators ; it would be interesting to explore the origin and limitations of such a factorization at weak coupling . since the arguments discussed in [ no_insertions ] build the closed null polygonal wilson loops one side at a time , they can be used to also prove the relation ( [ corropenwl ] ) and its generalization to higher - point correlation functions and correlation function of open null zig - zag wilson lines with arbitrary number of cusps . such relations may provide a semiclassical approach complementary to that of @xcite to the calculation of anomalous dimensions of short operators , in particular of members of the konishi multiplet . in this note we discussed in detail the relation between the null and partial null limits of correlation functions and the expectation value of null polygonal wilson loops as well as the correlation functions of such wilson loops and local gauge invariant operators . twistor space techniques were used in @xcite to construct a recursion relation for the expectation value of null polygonal wilson loops ; this recursion relation mirrors that of the unregularized integrand of scattering amplitudes proposed in @xcite ; see also ref . the two structurally different terms arise from the self - intersections of the wilson loop due to a bcfw - type deformation and from a quantum term in the loop equation , respectively ( see @xcite for details ) . through the construction in @xcite a similar recursion relation holds for correlation functions of gauge invariant operators in the null - separation limit . it would be interesting to investigate the existence of similar recursion relations for partial null - limits of correlation functions . it is not difficult to see that the bcfw - type deformation will lead to additional terms arising form the intersection of the deformed wilson loop and the line(s ) representing the insertion points of the additional operators . when such an intersection occurs , one of the corners of the deformed wilson loop becomes null - separated from one of the additional operators . it may be possible that such contributions can also be expressed in terms of null wilson loops . it may , in fact , be possible to construct bcfw - type recursion relations for correlation functions away from null or partial null limits . indeed , it has been argued in @xcite that strong coupling correlation functions in the supergravity approximation obey such a relation . while the supergravity approximation does not yield the desired null limit , the fact that string amplitudes have a bcfw presentation @xcite suggests that it may be possible to construct such a recursion relation also for strings in @xmath0 . we are grateful to m. campiglia , g. korchemsky , d. skinner and a. tseytlin for discussions and to g. korchemsky , p.h . damgaard and a. tseytlin for comments on a preliminary draft . this work is supported in part by the us department of energy under contract de - fg02 - 201390er40577 ( oji ) and the a.p . sloan foundation . the arguments in [ no_insertions ] and [ w_insertion ] may be extended to correlation functions of 2-field operators containing fermions or gauge fields ; one may interpret this as a step towards the proof of the supersymmetric generalization of the correlation function / wilson loop relation . on dimensional grounds one expects that , since the engineering dimensions of such fields are larger than that of scalars , the correlators will be more singular in the null separation limit than the correlation function of scalar operators . the position space fermion propagator is obtained by fourier - transforming the momentum space one in the standard way . s(x_1-x_2)&=&i/_1^_0d & = & -(1-)^2 - 2 ^ 3 - 2(1- ) . not unexpectedly , this propagator is more singular in the limit in which @xmath55 and @xmath56 are null - separated than the scalar propagator . let us consider next a sequence of fermion propagators with gluons attached to it . the important property of the fermion vertices is that they do not contain any derivatives . however the momentum factor in the numerator of the fermion propagator will play a role similar to the derivative in the scalar - gluon vertex . the contribution of such a sequence to the expectation value of operators is the factor & & l(x_1,x_2,k_1 , , k_n)__1 _n + & = & d^dp_1d^dp_2e^ip_1x_1+ip_2x_2d^dq_1d^dq_n-1 + & & ^(d)(k_1+q_1+p_1)^(d)(k_2-q_1+q_2)^(d)(k_n-1-q_n-2+q_n-1)^(d)(k_n - q_n-1+p_2 ) . following the same steps as in the case of scalar operators we find that [ procfermi ] & & l(x_1,x_2,k_1 , , k_n)__1 _n + & = & ( -i)^(n+1)(-1)^n^(-i ) _ i=1^n^ _ i^_i(-i+_j=1^ik_j_i)(_i=1^n _ t_i-1 ^ 1 dt_i ) + & & e^-i_ik_i(x_2t_i - x_1(1-t_i ) ) _ 0^d^ne^i(t_i , k_i)-i-0 - 0/ . the @xmath67 integral is exactly of the same type as discussed before ; therefore , the leading term in the @xmath143 limit arises from the highest power of @xmath75 generated by the derivatives with respect to @xmath144 . dirac matrix algebra may be used to simplify the numerator ; discarding terms proportional to @xmath85 we find that [ finalfermi ] & & l(x_1,x_2,k_1 , , k_n)__1 _n + & & = i^n+1s(x_1-x_2)(_i=1^n(x_1-x_2)^_i ) ( _ i=1^n_t_i-1 ^ 1 dt_i)e^-i_ik_i(x_2t_i - x_1(1-t_i ) ) , a position space fermion propagator multiplied by a wilson line with @xmath12 gluons attached to it . this is the same structure as in the case of a sequence of scalar propagators . there are two classes of diagrams which could potentially yield the same singularity as the sequence of fermion propagators . one of them involves the fermion - scalar vertex . having such a vertex does not lead to a smaller number of numerator momenta . however , since this vertex does not bring a vector index being proportional to only @xmath145 , its contribution at the level of eq . ( [ procfermi ] ) will contain a factor of ( x/_1-x/_2)_5(x/_1-x/_2)=-(x_1-x_2)^2_5 , and this will soften the singularity of @xmath99 compared to eq . ( [ finalfermi ] ) . the other potential contributions arise from the gluon self - interactions ; an example is shown in fig . ( [ fermions_3 g ] ) . it is not difficult to see that , in a general renormalizable gauge , their momentum dependence leads , after fourier transform , to factors of @xmath138 in the numerator which will soften the singularity in the null separation limit and render these diagrams subleading . the arguments in [ w_insertion ] carry over unmodified to the case when one or several additional operators are added to such correlation functions and kept at generic positions relative to the operators which become null - separated ; we will not repeat them here . similarly to the case of scalar operators , we find that the @xmath12-partial null limit , @xmath147 with @xmath6 and @xmath148 , of the connected part of the correlation function of @xmath109 operators is given by _ x_ii+1 ^ 20 = . the feynman diagrams contributing to correlation functions of two - field operators containing gauge field strengths do not follow the classification we have been using ; this is because there are at least two operators between which there is no direct r - charge flow ( _ i.e. _ if a flow exists at all it must pass through other operators ) . we will see , however , that unlike previous situation when absence of a flow led to absence of a sufficiently strong singularity in the null - separation limit , here a singularity does occur . the main difference compared to previous discussions relates to the existence of a derivative in the field strength acting along the sequence of propagators not carrying r - charge . as we will see , this will lead to the desired singularity in the limit in which a sequence of gluon propagators and three - point vertices stretch between two null - separated points . l(x_1 , x_2)_^ = ( -i_+i_)^(-i_+i_)d^dp_1 . the @xmath51 integral is exactly the same as that of a scalar propagator . introducing a schwinger - parameter and carrying out the momentum integral leads again to the position space scalar propagator , except that now there are two additional derivatives acting on it . each derivative yields a factor of @xmath75 , leading to l(x_1 , x_2)&= & ( ( x_1-x_2)__-(x_1-x_2)__)^ ( ( x_2-x_1)__-(x_2-x_1 ) _ _ ) + & & _ 0^ e^-i-0/ . [ gluon_prop ] this is again an integral of the type ( [ allints ] ) and , following ( [ generic_int ] ) , it will be proportional to @xmath149 . the singularity , stronger than that of a propagator between two scalar operators , is a reflection of the dimensional analysis . it is not difficult to deduce the properties of a sequence of gluon propagators and three - gluon vertices . as in the case of scalar propagators , each gluon propagator brings one positive power of @xmath67 . to compensate for it and end up with a singularity as strong as ( [ gluon_prop ] ) the null limit , each derivative in the 3-gluon vertex must yield a factor of @xmath75 , in close analogy with the derivative in the scalar - gluon vertex see [ sgv ] . each such derivative also generates a factor of @xmath150 ; due to the antisymmetry of the field strength contributions ( cf . ( [ gluon_prop ] ) ) these factors must carry the lorentz indices of the out - going gluons . the details are essentially identical to those in sec . [ sgv ] and we will not repeat them here . the conclusion is that , up to a factor of @xmath149 , the sequence of gluon propagators and three - point vertices between two null - separated points reduces in the null limit to a wilson line with as many gluon vertices as the original number of vertices : l(x_1,x_2,k)_^__1 _n & & ( ( x_1-x_2)__-(x_1-x_2 ) _ _ ) ^ ( ( x_2-x_1)__-(x_2-x_1 ) _ _ ) + & & _ i=1^ni(x_1-x_2)__i _ t_i-1 ^ 1 dt_ie^-i_ik_i(x_2t_i - x_1(1-t_i ) ) , where the field strengths in the two operators at positions @xmath55 and @xmath56 carry indices @xmath151 and @xmath152 , respectively . other diagrams , in which the sequence of gluons is interrupted by scalars , fermions or ghosts will have subleading contributions in the null limit . similarly , four - point lagrangian vertices as well as the commutator terms in the field strength will drop out for the same reason . thus , similarly to the correlation function of operators constructed out of scalars and fermions , correlation functions of operators containing field strengths reduce to the expectation value of a null polygonal wilson loop with cusps at the positions of the original operators . as in the case of correlation functions of operators without gauge fields , inclusion of one or more operators at generic positions relative to the null separation points is straightforward with the same result quoted in eqs . ( [ abt ] ) and ( [ ww ] ) . g. p. korchemsky , j. m. drummond , e. sokatchev , `` conformal properties of four - gluon planar amplitudes and wilson loops , '' nucl . * b795 * , 385 - 408 ( 2008 ) . [ arxiv:0707.0243 [ hep - th ] ] . a. brandhuber , p. heslop , g. travaglini , `` mhv amplitudes in n=4 super yang - mills and wilson loops , '' nucl . * b794 * , 231 - 243 ( 2008 ) . [ arxiv:0707.1153 [ hep - th ] ] . e. i. buchbinder , a. a. tseytlin , `` on semiclassical approximation for correlators of closed string vertex operators in ads / cft , '' jhep * 1008 * , 057 ( 2010 ) . [ arxiv:1005.4516 [ hep - th ] ] . r. a. janik , p. surowka , a. wereszczynski , `` on correlation functions of operators dual to classical spinning string states , '' jhep * 1005 * , 030 ( 2010 ) . [ arxiv:1002.4613 [ hep - th ] ] . d. e. berenstein , r. corrado , w. fischler and j. m. maldacena , `` the operator product expansion for wilson loops and surfaces in the large n limit , '' phys . d * 59 * , 105023 ( 1999 ) [ arxiv : hep - th/9809188 ] . k. zarembo , `` open string fluctuations in @xmath0 and operators with large r charge , '' phys . d * 66 * , 105021 ( 2002 ) [ arxiv : hep - th/0209095 ] . v. pestun and k. zarembo , `` comparing strings in @xmath0 to planar diagrams : an example , '' phys . d * 67 * , 086007 ( 2003 ) [ arxiv : hep - th/0212296 ] . j. gomis , s. matsuura , t. okuda and d. trancanelli , `` wilson loop correlators at strong coupling : from matrices to bubbling geometries , '' jhep * 0808 * , 068 ( 2008 ) [ arxiv:0807.3330 ] . l. f. alday and a. a. tseytlin , `` on strong - coupling correlation functions of circular wilson loops and local operators , '' arxiv:1105.1537 . a. miwa and t. yoneya , `` holography of wilson - loop expectation values with local operator insertions , '' jhep * 0612 * , 060 ( 2006 ) [ hep - th/0609007 ] . m. sakaguchi and k. yoshida , `` a semiclassical string description of wilson loop with local operators , '' nucl . b * 798 * , 72 ( 2008 ) [ arxiv:0709.4187 ] . a. miwa , `` bmn operators from wilson loop , '' jhep * 0506 * , 050 ( 2005 ) . [ hep - th/0504039 ] . g. w. semenoff , d. young , `` exact 1/4 bps loop : chiral primary correlator , '' phys . lett . * b643 * , 195 - 204 ( 2006 ) . [ hep - th/0609158 ] .

we prove that the cardinality of the intersection of two compact exact lagrangian submanifolds in a cotangent bundle is bounded from below by the dimension of the hom space of the guillermou s sheaf quantizations of the lagrangians in tamarkin s category . this gives a purely sheaf - theoretic new proof of a result of nadler and fukaya - seidel - smith , which asserts that the cardinality is at least the sum of the betti numbers of the base space .

1701.02057

an important problem in symplectic geometry is to study exact lagrangian submanifolds in cotangent bundles . arnold s nearby lagrangian conjecture asserts that any compact exact lagrangian submanifold of a cotangent bundle is hamiltonian isotopic to the zero - section . in symplectic geometry , non - displaceability of lagrangian submanifolds is also a central problem , which is also originally due to arnold . recall that a diffeomorphism @xmath0 of a symplectic manifold @xmath1 is said to be a hamiltonian diffeomorphism if there exists a hamiltonian isotopy with compact support @xmath2 } \colon x \times [ 0,1 ] \to x$ ] such that @xmath3 and @xmath4 . the problem is the existence or non - existence of a hamiltonian diffeomorphism @xmath5 satisfying @xmath6 for two compact lagrangians @xmath7 and @xmath8 and the estimation of the cardinality of the intersection . for a cotangent bundle and its zero - section , laudenbach - sikorav @xcite proved the following . let @xmath9 be a compact @xmath10-manifold without boundary and @xmath11 its cotangent bundle equipped with the canonical exact symplectic structure . then for a hamiltonian diffeomorphism @xmath5 of @xmath11 such that the intersection of the zero - section @xmath9 and its image @xmath12 is transversal , one has an inequality @xmath13 note that this inequality had been proved by chaperon @xcite in the case @xmath9 is a torus . although arnold s nearby lagrangian conjecture has been proved only for some special cases , combined with the inequality , it implies the following statement for any two compact exact lagrangians . in fact , this was already proved by nadler @xcite and fukaya - seidel - smith @xcite with the use of the floer theory . [ thm : intronfss ] for any two compact connected exact lagrangian submanifolds @xmath7 and @xmath8 of @xmath11 intersecting transversally , one has @xmath14 in this paper , we study intersection of two compact exact lagrangian submanifolds in a cotangent bundle using a different method based on the microlocal sheaf theory , especially guillermou s sheaf quantization and the category considered by tamarkin . in particular , we give a purely sheaf - theoretic new proof of . the microlocal sheaf theory was introduced and systematically developed by kashiwara and schapira @xcite . one of the key ingredients of the theory is the notion of microsupports of sheaves . let us recall it briefly . in the sequel , let @xmath15 be a field . for a @xmath10-manifold @xmath1 , we denote by @xmath16 the bounded derived category of sheaves of @xmath15-vector spaces . with an object @xmath17 , one can associate its microsupport @xmath18 , a closed subset of the cotangent bundle @xmath19 of @xmath1 . roughly speaking , @xmath18 describes directions in which the cohomology of @xmath20 does not propagate . the microsupport is conic , that is , invariant by the action of @xmath21 on @xmath19 . tamarkin @xcite gave a new approach to non - displaceability problems based on the microlocal sheaf theory . roughly speaking , he constructed a sheaf whose microsupport coincides with a given lagrangian submanifold of a cotangent bundle outside the zero - section and deduced a non - displaceability result from it . such a sheaf is called a _ sheaf quantization _ of the lagrangian . however , the problem is that microsupports of sheaves are conic and there is no such sheaf for a non - conic lagrangian . tamarkin s idea to treat non - conic lagrangians and not necessarily homogeneous hamiltonian isotopies is to add a variable and reduce the problem to the exact cases . one can consider a sheaf quantization for a non - conic lagrangian by adding a variable and conifying " it . after tamarkin s work , guillermou - kashiwara - schapira @xcite and guillermou @xcite proved the existence of sheaf quantization of hamiltonian isotopies and compact exact lagrangian submanifolds in cotangent bundles . below we review their results more precisely , following the formulation of guillermou - schapira @xcite for the result of tamarkin @xcite . see for more details . note that sheaf - theoretic approaches to symplectic geometry also appeared in @xcite . in what follows , again let @xmath9 be a non - empty compact connected @xmath10-manifold without boundary and regard its cotangent bundle @xmath11 as an exact symplectic manifold . tamarkin @xcite considered a category @xmath22 defined by a localization of @xmath23 by objects whose microsupports are contained in @xmath24 , where @xmath25 is the homogeneous symplectic coordinate of @xmath26 . for a compact subset @xmath27 of @xmath11 , one defines a subcategory @xmath28 by the subcategory consisting of @xmath20 with @xmath29 , where @xmath30 . then one can prove that if @xmath31 , @xmath32 for any @xmath33 and @xmath34 . moreover we define a category @xmath35 as a quotient of @xmath22 by what are called torsion objects . the category @xmath35 has a hamiltonian invariance property . namely , for a hamiltonian diffeomorphism @xmath36 , one can construct a functor @xmath37 such that @xmath38 in @xmath35 for any @xmath33 . the fact that hom spaces in @xmath35 can be described as inductive limits of those in @xmath22 implies the following non - displaceability theorem : if @xmath27 and @xmath39 are compact subsets of @xmath11 and if there exist objects @xmath33 and @xmath40 such that @xmath41 , then @xmath27 and @xmath39 are mutually non - displaceable , that is , @xmath42 for any hamiltonian diffeomorphism @xmath5 of @xmath11 . after the work of tamarkin , guillermou - kashiwara - schapira @xcite constructed a sheaf quantization of a hamiltonian isotopy of @xmath11 and proved the non - displaceability of the zero - section of a cotangent bundle by using the quantization . let @xmath43 be an open interval of @xmath44 containing @xmath45 and @xmath46 a hamiltonian isotopy such that @xmath47 . adding a variable , one can canonically associate a homogeneous hamiltonian isotopy @xmath48 , where @xmath49 is the complement of the zero - section . denote by @xmath50 the conic lagrangian submanifold associated with @xmath51 . the authors proved the existence and uniqueness of a sheaf quantization @xmath52 of @xmath53 whose restriction to @xmath54 is the constant sheaf of the diagonal . using the quantization , one can construct a family of sheaves @xmath55 such that the microsupport of @xmath56 is related with the intersection @xmath57 of the zero - section and its image by @xmath58 and all @xmath56 have the same cohomology . the authors recovered the result of laudenbach - sikorav @xcite by applying the microlocal morse lemma and the morse inequality for sheaves to @xmath56 . note also that the existence of the functor @xmath59 mentioned above is now a consequence of the existence of @xmath60 . after these works , guillermou @xcite constructed a sheaf quantization of a compact exact lagrangian submanifold of a cotangent bundle and studied topological properties of the lagrangian by using the quantization . let @xmath61 be a compact connected exact lagrangian submanifold of @xmath11 . choosing a primitive of the liouville 1-form , one can define a conification @xmath62 , which is a conic lagrangian submanifold of @xmath63 . guillermou proved the existence and uniqueness of a sheaf quantization @xmath64 of @xmath62 such that @xmath65 is 0 and @xmath66 is isomorphic to the constant sheaf on @xmath9 for a sufficiently large @xmath67 . the object @xmath68 is called the canonical sheaf quantization . using the quantization , he recovered results of abouzaid and kragh @xcite , saying that the maslov class of @xmath61 is zero , and fukaya - seidel - smith @xcite and abouzaid @xcite , saying that the projection @xmath69 is a homotopy equivalence . in this paper , we study the relation between intersection of two compact exact lagrangian submanifolds of a cotangent bundle @xmath11 and hom spaces of guillermou s sheaf quantizations in the category @xmath35 . let @xmath7 and @xmath8 be compact connected exact lagrangian submanifolds of @xmath11 and chose primitives @xmath70 of the liouville 1-form . let us denote by @xmath71 the canonical sheaf quantization associated with @xmath72 . [ prp : introcohlagr ] one has an isomorphism @xmath73 ) \simeq h^l(m;k_m ) \quad \text{for any $ l \in \bz$}.\ ] ] in particular , this proposition implies that @xmath74 for any compact connected exact lagrangian submanifolds @xmath7 and @xmath8 by tamarkin s non - displaceability theorem . note that for the special case where one of the lagrangian is the zero - section , the non - emptiness was proved by gromov @xcite . moreover we prove that the hom space can be used not only for the non - displaceability but also for the estimate of the cardinality of the intersection . namely , we prove that the cardinality is bounded from below by the dimension of the hom space . [ thm : intromain ] assume that @xmath7 and @xmath8 intersect transversally . then one has an inequality @xmath75).\ ] ] this theorem is proved by a combination of techniques of the above three works . by results of tamarkin @xcite , one can construct a sheaf @xmath76 on @xmath77 whose @xmath78-th cohomology ( supported by @xmath79 ) gives @xmath80)$ ] . then we apply the morse inequality for sheaves to @xmath76 and the function @xmath81 as in guillermou - kashiwara - schapira @xcite and take an inductive limit . the key proposition is that the dimension of the stalk of the local cohomology sheaf @xmath82 is equal to the cardinality of @xmath83 , where @xmath84 is the projection . in order to prove this , we use microlocal inverse images and some techniques developed by kashiwara - schapira @xcite . combining with , we can recover as a corollary . in fact , using sections of @xmath76 , we can obtain more precise estimates including the difference of the values of the primitives @xmath85 ( see for more details ) . by the result of tamarkin @xcite , guillermou @xcite , and ours , the category @xmath35 has the following three properties : ( i ) hamiltonian invariance ( @xcite ) , ( ii ) @xmath86 ) \simeq h^l(l;k ) \ ( l \in \bz)$ ] for a compact exact lagrangian @xmath61 ( a particular case of and by results in @xcite ) , and ( iii ) the dimension of the hom space is a lower bound of the cardinality of the intersection ( ) . the floer cohomology @xmath87 has similar properties to ( i)(iii ) , though the approach is totally different . in the floer theory , one defines the floer cohomology so that its rank gives a lower bound of the intersection , but the proof of hamiltonian invariance is difficult . in contrast , in the sheaf - theoretic approach , one constructs the category @xmath35 so that it satisfies the property ( i ) , but ( iii ) is unclear from the definition . although our approach is purely sheaf - theoretic and does not use floer cohomology or fukaya categories , it seems to be closely related to approaches by nadler @xcite and fukaya - seidel - smith @xcite . in order to discuss the relation , let us review their results briefly . nadler @xcite proved the following after the work of nadler - zaslow @xcite . let @xmath88 be a compact real analytic manifold without boundary . then the infinitesimal fukaya category @xmath89 is equivalent to the bounded derived category of constructible sheaves @xmath90 . moreover , assuming that @xmath88 is simply - connected , he proved any relatively spin compact connected exact lagrangian submanifold @xmath61 with vanishing maslov class of @xmath91 corresponds to ( a shift of ) the constant sheaf of rank @xmath92 on @xmath88 . in particular , these imply that @xmath61 is isomorphic to ( a shift of ) the zero - section @xmath88 in @xmath89 and for any pair of such lagrangian submanifolds @xmath93 , one has an isomorphism ( up to shift ) @xmath94 by the definition of the floer cohomology , the isomorphism implies . fukaya - seidel - smith @xcite independently proved the isomorphism in @xmath89 and for a @xmath10-manifold @xmath88 ( not necessarily simply - connected , see ( * ? ? ? * theorem 1.3 ) ) . note that their assumptions of relatively spin and vanishing maslov class can be removed , thanks to results of abouzaid @xcite , and abouzaid and kragh @xcite , respectively . also in our approach , we can prove an isomorphism between any compact exact lagrangian and the zero - section in terms of sheaf quantization . in fact , as pointed out by t. kuwagaki , for any compact connected exact lagrangian submanifold @xmath61 , the canonical sheaf quantization @xmath68 is isomorphic to the canonical sheaf quantization @xmath95 associated with the zero - section @xmath9 of @xmath11 in the category @xmath35 ( see ) . in our ongoing work with him , we extend the notion of sheaf quantization to non - compact lagrangians and explore a relation between the nadler - zaslow equivalence and guillermou s sheaf quantization . during the preparation of this paper , the author also learned that c. viterbo announced that he had proved the relation between the section of @xmath76 and the floer cochain complex @xmath96 . this paper is organized as follows . in , we recall the microlocal sheaf theory due to kashiwara and schapira @xcite . in , we review the results of @xcite about tamarkin s non - displaceability theorem and sheaf quantization of hamiltonian isotopies and compact exact lagrangian submanifolds in cotangent bundles . in , we prove and . in appendix , we prove the functoriality " of guillermou s canonical sheaf quantization with respect to hamiltonian diffeomorphisms , which seems to be useful for further developments . note that results in appendix are independent and not used in the previous sections . the author wishes to express his sincere gratitude to stphane guillermou for many helpful discussions . he is also very grateful to pierre schapira for many enlightening discussions and helpful advice . he also thanks tomohiro asano , vincent humilire , and tatsuki kuwagaki for many stimulating discussions . he expresses his gratitude to imj - prg and equipe analyse algbrique " for hospitality during the preparation of this paper . this work was supported by a grant - in - aid for jsps fellows 15j07993 and the program for leading graduate schools , mext , japan . in this paper , all manifolds are assumed to be real manifolds of class @xmath10 . throughout this paper , let @xmath15 be a field . in this section , we recall some definitions and results from @xcite . we mainly follow the notation in @xcite . let @xmath1 be a @xmath10-manifold of dimension @xmath97 . for a locally closed subset @xmath27 of @xmath1 , we denote by @xmath98 its closure . one also denotes by @xmath99 or simply @xmath100 the diagonal of @xmath101 . we denote by @xmath102 the tangent bundle of @xmath1 and by @xmath103 the cotangent bundle of @xmath1 . if there is no risk of confusion , we simply write @xmath104 instead of @xmath105 . if @xmath1 is a submanifold of a manifold @xmath106 , one denotes by @xmath107 the conormal bundle to @xmath1 in @xmath106 . in particular , @xmath108 denotes the zero - section of @xmath19 . we set @xmath109 . let @xmath110 be a morphism of manifolds . with @xmath111 we associate morphisms and a commutative diagram @xmath112 ^ -{\pi_x } & x \times_y t^*y \ar[d]^-\pi \ar[l]_-{f_d } \ar[r]^-{f_\pi } & t^*y \ar[d]^-{\pi_y } \\ x \ar@{=}[r ] & x \ar[r]_-f & y , } \end{split}\ ] ] where @xmath113 is the projection and @xmath114 is induced by the transpose of the tangent map @xmath115 . we also set @xmath116 if @xmath111 is a closed embedding , the above notation coincides with the previous one . we denote by @xmath117 a local homogeneous coordinate system of @xmath19 . the cotangent bundle @xmath19 is an exact symplectic manifold with the liouville form @xmath118 . the antipodal map @xmath119 is defined by @xmath120 . for a subset @xmath27 of @xmath19 , we denote by @xmath121 its image by the map @xmath122 . we denote by @xmath123 the constant sheaf with stalk @xmath15 and by @xmath124 the bounded derived category of sheaves of @xmath15-vector spaces on @xmath1 . for a morphism of manifolds @xmath110 , one can define grothendieck s six operations @xmath125 between derived categories of sheaves . since we work on the field @xmath15 , we simply write @xmath126 instead of @xmath127 . for an inclusion @xmath128 of a locally closed subset @xmath88 of @xmath1 and @xmath17 , we define @xmath129 we simply write @xmath130 for @xmath131 . one denotes by @xmath132 the dualizing complex on @xmath1 , that is , @xmath133 , where @xmath134 is the natural morphism . note that @xmath135 is isomorphic to @xmath136 $ ] , where @xmath137 is the orientation sheaf on @xmath1 . more generally , for a morphism of manifolds @xmath110 , we denote by @xmath138 the relative dualizing complex . recall the definition of the microsupport @xmath18 of @xmath17 . [ def : microsupport ] let @xmath17 and @xmath139 . one says that @xmath140 if there is a neighborhood @xmath141 of @xmath142 in @xmath19 such that for any @xmath143 and any @xmath10-function @xmath144 on @xmath1 ( defined on a neighborhood of @xmath143 ) with @xmath145 , one has @xmath146 . the following properties can be checked from the definition of microsupports . 1 . the microsupport is a conic ( i.e. , invariant by the action of @xmath21 on @xmath19 ) closed subset of @xmath19 . 2 . @xmath147 . the microsupports satisfy the triangular inequality : if @xmath148 is a distinguished triangle in @xmath16 , @xmath149 for @xmath150 . we also use the notation @xmath151 . we denote by @xmath152 the ( unbounded ) derived category of sheaves of @xmath15-vector spaces on @xmath1 . an object @xmath153 is said to be locally bounded if for any relatively compact open subset @xmath141 of @xmath1 , one has @xmath154 . we denote by @xmath155 the full subcategory of @xmath152 consisting of locally bounded objects . a microsupport of an object of @xmath155 can be defined in a totally same way as in , since it is a local notion . \(i ) if @xmath20 is a non - zero local system on a connected manifold @xmath1 , then @xmath156 . conversely , if @xmath157 then the cohomology sheaves @xmath158 are local systems for all @xmath159 . \(ii ) if @xmath1 is a smooth submanifold of a manifold @xmath106 , then @xmath160 . \(iii ) let @xmath144 be a @xmath10-function and assume that @xmath161 whenever @xmath162 . let @xmath163 and @xmath164 . then @xmath165 next we recall the microlocal morse theory and morse inequality for sheaves . let @xmath166 be a @xmath10-function . we set @xmath167 the following proposition is called ( a particular case of ) the microlocal morse lemma . see ( * ? ? ? * proposition 5.4.17 and corollary 5.4.19 ) for more details . the classical theory corresponds to the case @xmath20 is the constant sheaf @xmath123 . [ prp : microlocalmorse ] let @xmath17 and @xmath166 be a @xmath10-function . assume that @xmath168)$ ] is compact for any @xmath169 . let @xmath170 in @xmath44 or @xmath171 and assume that @xmath172 for any @xmath173 . then the canonical morphism @xmath174 is an isomorphism . the morse inequality can also be extended to sheaves as follows . for a bounded complex @xmath175 of @xmath15-vector spaces with finite - dimensional cohomology , we set @xmath176 let @xmath17 and @xmath166 a @xmath10-function . we consider the following assumptions 1 . @xmath168)$ ] is compact for any @xmath169 , 2 . the set @xmath177 is finite : @xmath178 , 3 . if one sets @xmath179 has finite - dimensional cohomology for all @xmath180 . [ thm : morseineq ] assume that _ ( 1)(3 ) _ are satisfied . then 1 . @xmath181 has finite - dimensional cohomology , 2 . one has @xmath182 for any @xmath183 . note that implies @xmath184 for any @xmath183 . we consider bounds for microsupports of proper direct images , non - characteristic inverse images , and @xmath185 . let @xmath110 be a morphism of manifolds and @xmath186 a closed conic subset . the morphism @xmath111 is said to be _ non - characteristic _ for @xmath27 if @xmath187 see for the notations @xmath113 and @xmath114 . in particular , if @xmath111 is a submersion , @xmath111 is non - characteristic for any closed conic subset in @xmath188 . note that submersions are called smooth morphisms in @xcite . one can show that if @xmath111 is non - characteristic for @xmath27 , then @xmath189 is a conic closed subset in @xmath19 . [ thm : operations ] let @xmath110 be a morphism of manifolds , @xmath17 , and @xmath190 . 1 . assume that @xmath111 is proper on @xmath191 . then @xmath192 . 2 . assume that @xmath111 is non - characteristic for @xmath193 . then the canonical morphism @xmath194 is an isomorphism and @xmath195 . [ prp : ssprod ] let @xmath196 be manifolds and denote by @xmath197 the projection @xmath198 . let @xmath199 . then one has @xmath200 using and ( ii ) for the diagonal embedding @xmath201 , one can prove the following . [ prp : sshom ] let @xmath202 and assume that @xmath203 . then @xmath204 where @xmath205 is the fiberwise sum . we consider estimates of microsupports of non - proper direct images for special cases . let @xmath206 and @xmath207 be vector spaces and @xmath208 a constant linear map , that is , the map of the form @xmath209 with @xmath210 is a linear map of vector spaces . the map @xmath211 induces the maps @xmath212_-{u_d } \ar[rd]^-{u_\pi } & \\ t^*x \times v_1 \times v_1^ * \ar[rd]_-{v_\pi } & & t^*x \times v_2 \times v_2^*. \ar[ld]^-{v_d } \\ & t^*x \times v_2 \times v_1^*. & } \end{split}\ ] ] note that for a subset @xmath27 of @xmath213 we have @xmath214 . let @xmath211 be as above and @xmath215 a closed subset . one sets @xmath216 [ prp : nonproper ] let @xmath211 be as above and @xmath217 . then @xmath218 let @xmath219 be manifolds . we write @xmath220 and @xmath221 for short . we use the same symbol @xmath197 for the projections @xmath222 and @xmath223 . we also denote by @xmath224 the projection @xmath225 . similarly , we denote by @xmath226 the projection @xmath227 . one denotes by @xmath228 the composite of @xmath229 and the antipodal map on @xmath230 . let @xmath231 and @xmath232 . we set @xmath233 we define the operations of composition of kernels as follows : @xmath234 let @xmath235 and assume that 1 . @xmath236 is proper on @xmath237 , 2 . @xmath238 . then by and estimates of microsupports of tensor products ( see ( * ? ? ? * proposition 5.4.14 ) ) , we have @xmath239 if there is no risk of confusion , we simply write @xmath240 instead of @xmath241 . let @xmath242 be a subset and set @xmath243 . we denote by @xmath244 the subcategory of @xmath16 consisting of sheaves whose microsupports are contained in @xmath27 . by the triangular inequality , the subcategory @xmath244 is a triangulated subcategory . we set @xmath245 the localization of @xmath16 by @xmath246 . a morphism @xmath247 in @xmath16 becomes an isomorphism in @xmath248 if @xmath211 is embedded in a distinguished triangle @xmath249 with @xmath250 . for a closed subset @xmath39 of @xmath251 , @xmath252 denotes the full triangulated subcategory of @xmath248 consisting of @xmath20 with @xmath253 . in the case @xmath254 for @xmath139 , we simply write @xmath255 instead of @xmath256 . note that our notation is the same as in @xcite and slightly differs from that of @xcite . by using the following refined microlocal cut - off lemma " , we can construct an isomorphic object in the localized category with an estimate of the microsupport . recall that a cone is said to be proper if it contains no lines . [ prp : cutoff ] let @xmath257 , @xmath60 a proper closed cone of @xmath258 , and @xmath259 an open cone . let @xmath17 and @xmath175 a conic neighborhood of @xmath260 . then there exist @xmath261 and a morphism @xmath262 satisfying the following conditions : 1 . @xmath211 is an isomorphism in @xmath263 , 2 . @xmath264 . next we recall the notion of microlocal inverse images . let @xmath110 be a morphism of manifolds , @xmath265 . we set @xmath266 . let @xmath267 . the microlocal inverse image of @xmath268 at @xmath142 is defined by the pro - object of @xmath269 @xmath270 where @xmath271 ranges over morphisms in @xmath272 which are isomorphisms in @xmath273 . in @xcite , the authors simply wrote @xmath274 , but in this paper we use the notation @xmath275 in order to emphasize the dependence on the point @xmath142 . in this paper , we only consider the cases where the microlocal inverse images belong to @xmath269 . the next proposition is a slight generalization of ( * ? ? ? * proposition 6.1.9 ) ( though it is almost written in loc . cit . ) . [ prp : microlocalinverse ] 1 . let @xmath267 . if @xmath276 in a neighborhood of @xmath142 , then @xmath277 belongs to @xmath269 . moreover for any neighborhood @xmath175 of @xmath142 , one has @xmath278 2 . let @xmath190 and @xmath279 . assume the following conditions : 1 . @xmath111 is non - characteristic for @xmath193 , 2 . there exist finitely many points @xmath280 such that @xmath281 + then one has an isomorphism @xmath282 \(i ) is the same as ( * ? ? ? * proposition 6.1.9 ( i ) ) . \(ii ) we shall use a similar argument to the proof of ( * ? ? ? * proposition 6.1.9 ) . set @xmath283 , @xmath284 . take proper closed convex cones @xmath285 and open convex cones @xmath286 such that @xmath287 , @xmath288 , @xmath289 , and @xmath290 . by the refined microlocal cut - off lemma ( ) , there exist morphisms @xmath291 such that @xmath292 is an isomorphism in @xmath293 and @xmath294 satisfies the conditions 1 . @xmath111 is non - characteristic for @xmath295 , 2 . @xmath296 for any @xmath297 . embedding the morphism @xmath298 into a distinguished triangle @xmath299 , we find that @xmath111 is non - characteristic for @xmath300 by the triangular inequality . moreover we get @xmath301 . hence by ( ii ) , @xmath302 , which shows @xmath303 is an isomorphism in @xmath269 . it remains to apply ( * ? ? ? * proposition 6.1.9 ( ii ) ) to each @xmath294 . let @xmath304 be a locally closed lagrangian submanifold and @xmath305 . pure and simple sheaves along @xmath306 at @xmath142 are defined in ( * ? ? ? * definition 7.5.4 ) . in this subsection , we briefly recall them . let @xmath305 and @xmath166 be a @xmath10-function on @xmath1 . the function @xmath144 is said to be transversal to @xmath306 at @xmath142 if @xmath307 and the manifolds @xmath306 and @xmath308 intersect transversally at @xmath142 . for @xmath309 , we define the following lagrangian subspaces in @xmath310 : @xmath311 recall the definition of the inertia index ( see @xcite and ( * ? ? ? * a.3 ) ) . let @xmath312 be a symplectic vector space and let @xmath313 three lagrangian subspaces of @xmath314 . we define a quadratic form @xmath315 on @xmath316 by @xmath317 . then @xmath318 is defined as the signature of @xmath315 . one sets @xmath319 [ prp : maslovshft ] let @xmath320 be @xmath10-functions transversal to @xmath306 at @xmath142 . let @xmath17 and assume that @xmath321 in a neighborhood of @xmath142 . then @xmath322.\ ] ] in the situation of , @xmath20 has microlocal type @xmath323 with shift @xmath324 at @xmath142 if @xmath325\ ] ] for some ( hence for any ) @xmath10-function @xmath144 transversal to @xmath306 at @xmath142 . if @xmath326 , @xmath20 is said to be _ pure _ along @xmath306 at @xmath142 . if moreover @xmath327 , @xmath20 is said to be _ simple _ along @xmath306 at @xmath142 . if @xmath20 is pure ( resp . simple ) at all points of @xmath306 , one says that @xmath20 is pure ( resp . simple ) along @xmath306 . when @xmath306 is a conormal bundle to a submanifold @xmath328 in a neighborhood of @xmath142 , that is , @xmath329 has constant rank , then @xmath17 is simple along @xmath306 at @xmath142 if @xmath330 $ ] in @xmath255 for some @xmath331 . [ prp : homsimple ] let @xmath196 be manifolds , @xmath332 a lagrangian submanifold , and @xmath333 . denote by @xmath197 the projection @xmath198 . let @xmath199 and assume that @xmath334 has microlocal type @xmath335 with shift @xmath336 at @xmath337 along @xmath338 . then @xmath339 has microlocal type @xmath340 with shift @xmath341 at @xmath342 along @xmath343 . the microlocal inverse images preserve purity and simplicity under some assumptions . [ prp : inversesimple ] let @xmath110 be a morphism of manifolds , @xmath265 and set @xmath266 . let @xmath344 be a lagrangian submanifold of @xmath188 such that @xmath113 is transversal to @xmath344 at @xmath345 . let @xmath190 and assume that @xmath346 in a neighborhood of @xmath345 and @xmath20 has microlocal type @xmath61 with shift @xmath347 along @xmath344 at @xmath345 . 1 . for a sufficiently small open neighborhood @xmath175 of @xmath142 , @xmath348 is a lagrangian submanifold of @xmath19 , isomorphic to @xmath349 by @xmath114 . @xmath350 has microlocal type @xmath61 with shift @xmath347 at @xmath351 . next we consider the method to calculate stalks of local cohomology sheaves of direct images . [ prp : localcohdirect ] let @xmath17 and @xmath352 . set @xmath353 . assume that 1 . @xmath111 is proper on @xmath191 , 2 . there exist finitely many points @xmath354 such that @xmath355 3 . if one sets @xmath356 , for any @xmath180 , there exist smooth lagrangian submanifolds @xmath357 containing @xmath358 and @xmath359 such that @xmath360 has microlocal type @xmath361 with shift @xmath362 along @xmath357 at @xmath358 and @xmath363 4 . for any @xmath364 and any @xmath365 , @xmath114 is transversal to @xmath357 at @xmath358 . then one has the following : 1 . for any @xmath180 , there exists a sufficiently small open neighborhood @xmath366 of @xmath337 such that @xmath367 is a lagrangian submanifold of @xmath188 for any @xmath368 . 2 . let @xmath144 be a @xmath10-function on @xmath106 transversal to @xmath369 at @xmath370 for all @xmath371 and @xmath372 . then one has @xmath373,\ ] ] where the shifts @xmath374 are defined as follows : @xmath375 where @xmath376 means @xmath377 . \(i ) is the same as ( * ? ? ? * corollary 7.5.12 ( i ) ) . \(ii ) the proof is similar to that of ( * ? ? ? * theorem 7.3.1 ) . set @xmath378 . by the assumption ( 4 ) and the transversality of @xmath144 , @xmath379 is transversal to @xmath357 at @xmath358 . hence by ( 3 ) we have @xmath380\ ] ] for some shift @xmath381 . the assumption ( 2 ) implies @xmath382 combining this with the assumption ( 1 ) , we obtain @xmath383 for the shifts , we refer to ( * ? ? ? * theorem 7.3.1 ) and omit the details . in what follows , until the end of the paper , let @xmath9 be a non - empty compact connected manifold without boundary . in this section , we review tamarkin s approach to non - displaceability problems in symplectic geometry based on the microlocal sheaf theory . we also review sheaf quantization of hamiltonian isotopies and compact exact lagrangian submanifolds in a cotangent bundle . guillermou - kashiwara - schapira @xcite constructed a sheaf quantization of a hamiltonian isotopy . since microsupports of sheaves are conic subset of a cotangent bundle , the microlocal sheaf theory is related to the exact ( homogeneous ) symplectic structure rather than the symplectic structure of @xmath11 . in order to treat non - homogeneous hamiltonian isotopies and non - conic lagrangian submanifolds , an important trick is to add a variable and conify " them . we set @xmath384 and define the map @xmath385 & t^*m \\ ( x , t;\xi,\tau ) \ar@{|->}[r ] \ar@{}[u]|-{\rotatebox{90}{$\in$ } } & ( x;\xi/\tau ) . \ar@{}[u]|-{\rotatebox{90}{$\in$ } } } \end{split}\ ] ] let @xmath43 be an open interval in @xmath44 containing @xmath45 and @xmath46 a hamiltonian isotopy with compact support . note that @xmath386 is the identity for @xmath54 : @xmath47 . then one can construct a homogeneous hamiltonian isotopy @xmath48 such that the following diagram commutes : @xmath387 ^ -{\wh{\phi } } \ar[d]_-{\rho \times \id } & t^*_{\tau>0}(m \times \br ) \ar[d]^-{\rho } \\ t^*m \times i \ar[r]_-{\phi } & t^*m . } \end{split}\ ] ] here @xmath51 is called a homogeneous hamiltonian isotopy if it is a hamiltonian isotopy whose hamiltonian function @xmath388 is homogeneous of degree @xmath92 : @xmath389 for any @xmath390 . see ( * ? ? ? * subsection a.3 ) for more details . for simplicity , we set @xmath391 and consider a homogeneous hamiltonian isotopy @xmath392 and the associated homogeneous hamiltonian @xmath393 . we define a conic lagrangian submanifold @xmath394 by @xmath395 note that @xmath396 for any @xmath397 ( see for the definition of @xmath398 ) . the set @xmath399 is called the twisted graph of @xmath400 . for a homogeneous hamiltonian isotopy @xmath401 , there exists a unique object @xmath402 satisfying the following conditions : 1 . @xmath403 , 2 . @xmath404 , where @xmath405 is the diagonal of @xmath406 . moreover @xmath60 is simple along @xmath53 and both projections @xmath407 are proper . the object @xmath60 is called the _ quantization _ of @xmath51 . note also that @xmath408 . tamarkin @xcite ( see also guillermou - schapira @xcite ) considered the following categories , from which he deduced non - displaceability of lagrangian submanifolds . denote by @xmath117 a local homogeneous coordinate system on @xmath11 and by @xmath25 the homogeneous coordinate system on @xmath26 . define the maps @xmath409 we also use the notation @xmath410 for @xmath411 , one sets @xmath412 note that the functor @xmath413 is a left adjoint to @xmath414 . the functor @xmath415 defines a projector on the left orthogonal @xmath416 . by using this projector , tamarkin proved that the localized category @xmath417 is equivalent to the left orthogonal @xmath416 : @xmath418 we denote by @xmath22 the quotient category @xmath417 for short . for an object of @xmath22 , we always take the canonical representative in @xmath416 via the projector . [ prp : mord ] for @xmath419 , one has an isomorphism @xmath420 tamarkin @xcite proved the non - displaceability theorem by using the category @xmath22 and the notion of torsion objects . moreover guillermou - schapira @xcite proved that torsion objects form a triangulated subcategory and introduced the quotient category @xmath35 , which we will explain below . for a compact subset @xmath27 of @xmath11 , we define a full subcategory @xmath28 of @xmath22 by @xmath421 the following separation theorem is due to tamarkin @xcite . [ thm : separate ] let @xmath27 and @xmath39 be compact subsets of @xmath11 and assume that @xmath31 . then for @xmath33 and @xmath40 , one has @xmath32 . we give the outline of the proof of the theorem due to guillermou - schapira @xcite ( a new proof of tamarkin s result ) . denote by @xmath422 the function @xmath423 . recall the notation @xmath424 . then one can show that @xmath425 hence by and the microlocal morse lemma ( ) , we have the conclusion . for non - displaceability problems , we need a hamiltonian invariance property . one can get such invariance by taking the quotient by torsion objects . define the translation map @xmath426 for @xmath427 , there exists a canonical morphism @xmath428 an object @xmath427 is said to be a _ torsion object _ if @xmath429 for some @xmath430 . denote by @xmath431 the subcategory of torsion objects in @xmath432 . let @xmath427 and assume that @xmath433 $ ] for some compact interval @xmath434 $ ] of @xmath44 . then @xmath20 is a torsion object . the subcategory @xmath431 is a full triangulated subcategory . the triangulated category @xmath35 is the quotient category of @xmath22 by @xmath431 : @xmath435 . hom spaces in @xmath35 can be described as inductive limits of those in @xmath22 . [ prp : mort ] for @xmath419 , one has an isomorphism @xmath436 recall that for a symplectic manifold @xmath1 , a diffeomorphism @xmath0 is said to be a hamiltonian diffeomorphism if there exists a hamiltonian isotopy with compact support @xmath437 \to x$ ] such that @xmath3 and @xmath4 . let @xmath36 be a hamiltonian diffeomorphism and @xmath438 a hamiltonian isotopy satisfying the above conditions , where @xmath43 is an open interval containing @xmath439 $ ] . let @xmath48 be the associated homogeneous hamiltonian isotopy and @xmath440 the quantization of @xmath51 . then the composition with @xmath441 does not satisfy ( * ? ? ? * ( 3.3 ) ) in general , @xmath442 is bounded for any relatively compact subinterval @xmath443 of @xmath43 . the author learned the detailed proof from s. guillermou . one can prove it using the properness of @xmath444 and the fact that @xmath445 , where @xmath446 and @xmath447 . ] defines a functor @xmath448 which induces a functor @xmath449 ( see ( * ? ? ? * proposition 3.29 ) ) . moreover for a compact subset @xmath27 of @xmath11 and @xmath33 , the commutative diagram implies that @xmath450 hence the functor also induces @xmath37 . the following is a hamiltonian invariance theorem due to tamarkin @xcite . [ thm : invham ] let @xmath33 and @xmath37 be as above . then one has an isomorphism @xmath451 combining with and , we can deduce the non - displaceability theorem . two compact subsets @xmath27 and @xmath39 of @xmath11 are said to be _ mutually non - displaceable _ if for any hamiltonian diffeomorphism @xmath36 , @xmath42 . [ thm : nondisp ] let @xmath27 and @xmath39 be compact subsets of @xmath11 . assume that there exist @xmath33 and @xmath40 such that @xmath41 . then @xmath27 and @xmath39 are mutually non - displaceable . guillermou @xcite proved the existence of a sheaf quantization of a compact exact lagrangian submanifold of @xmath11 . let @xmath61 be a compact connected exact lagrangian submanifold of @xmath11 and choose a primitive of the liouville form @xmath452 satisfying @xmath453 with the pair @xmath454 , we associate the conification @xmath455 of @xmath61 with respect to @xmath111 by @xmath456 if there is no risk of confusion , we simply write @xmath62 instead of @xmath457 . consider the category @xmath458 consisting of sheaves microsupported in @xmath459 . by the compactness of @xmath61 , there is @xmath460 such that @xmath461 hence for any @xmath462 , the restrictions @xmath463 and @xmath464 are locally constant . for an object @xmath462 , one defines @xmath465 by @xmath466 for any @xmath467 ( independent of @xmath468 ) . we denote by @xmath469 the full subcategory of @xmath458 consisting of @xmath20 such that @xmath470 . guillermou @xcite proved the following existence and uniqueness of a canonical sheaf quantization of a compact exact lagrangian submanifold in a cotangent bundle . [ thm : guiquan ] let @xmath61 and @xmath471 be as above . 1 . there exists an object @xmath472 satisfying @xmath473 . 2 . moreover @xmath68 in _ ( i ) _ is unique up to a unique isomorphism and simple along @xmath62 . we call @xmath68 the _ canonical quantization _ of @xmath62 . in this section , we study intersection of two compact exact lagrangians in cotangent bundles using the results presented in the previous sections . let @xmath474 be compact connected exact lagrangian submanifolds of @xmath11 and choose primitives @xmath70 satisfying . for the pair @xmath72 we define the conification @xmath475 of @xmath335 with respect to @xmath476 as in . by theorem [ thm : guiquan ] , there exist the canonical quantizations @xmath477 satisfying the conditions ( i ) and ( ii ) . first we prove @xmath478 ) \simeq h^l(m;k_m)$ ] for any @xmath183 . let @xmath62 be a conic lagrangian submanifold of @xmath63 obtained as a conification of some compact connected exact lagrangian submanifold @xmath61 of @xmath11 . [ lem : orthogonal ] one has an inclusion @xmath479 there exists @xmath480 such that @xmath481 . let @xmath482 . since @xmath483 , by we have @xmath484 applying the microlocal morse lemma ( ) to @xmath485 and the function @xmath81 , we obtain the vanishing @xmath486 by the inclusion @xmath487 . [ prp : cohlagr ] there exists @xmath488 such that @xmath489)$ ] is isomorphic to @xmath490 for any @xmath491 and @xmath183 . in particular , one has an isomorphism @xmath492 ) \simeq h^l(m;k_m ) \quad \text{for any $ l \in \bz$}.\ ] ] by , for any @xmath183 we have @xmath493 ) = \hom_{\db(m \times \br)}(f_1,{t_c}_*f_2[l]).\ ] ] by the compactness of @xmath494 and @xmath495 , there exist @xmath496 satisfying @xmath497 and @xmath498 . take a sufficiently large @xmath488 such that @xmath499 . since @xmath500 and @xmath501 , we get @xmath502 for any @xmath491 . since @xmath503 , we can apply the microlocal morse lemma ( ) and obtain @xmath504 the second assertion follows from . for the special case @xmath61 is the zero - section @xmath505 of @xmath11 , the isomorphism @xmath86 ) \simeq h^l(m;k_m)$ ] was already proved by guillermou - schapira @xcite . in this case , @xmath506 is the canonical quantization associated with the zero - section @xmath505 . guillermou - schapira @xcite proved that the functor @xmath507 is fully faithful ( see ( * ? ? ? * corollary 5.8 ) ) . we thus obtain @xmath508 ) \simeq \hom_{\db(m)}(k_m , k_m[l ] ) \simeq h^l(m;k_m).\ ] ] in fact , we can prove for general @xmath7 and @xmath8 from the above special case . the following was pointed out to the author by tatsuki kuwagaki . [ prp : isomt ] for any compact connected exact lagrangian submanifold @xmath61 of @xmath11 , one has an isomorphism @xmath509 by the compactness of @xmath61 , we can take a sufficiently large @xmath460 such that @xmath510 . since @xmath511 , there exists a canonical morphism @xmath512 the cone of this morphism is supported in @xmath513 $ ] and hence a torsion object . therefore the morphism is an isomorphism in @xmath35 . a similar argument shows that the morphism @xmath514 is an isomorphism in @xmath35 . we thus obtain the result . hence follows from the isomorphism . [ cor : nondislagr ] let @xmath7 and @xmath8 be compact connected exact lagrangian submanifolds of @xmath11 . then @xmath7 and @xmath8 are mutually non - displaceable . it follows from and . next we prove that the cardinality of the intersection @xmath515 is bounded from below by the dimension of the hom space between the canonical sheaf quantizations of @xmath7 and @xmath8 in the category @xmath35 . hereafter we assume that @xmath516 recall the isomorphism @xmath517 since @xmath518 and @xmath519 are submersions , by ( ii ) we have @xmath520 and @xmath521 hence @xmath522 and by , we obtain @xmath523 set @xmath524 . [ prp : localcohrr ] let @xmath525 . let moreover @xmath526 _ _ ( _ _ hence @xmath527 . _ ) _ 1 . for any @xmath297 , there is a sufficiently small neighborhood @xmath528 of @xmath529 in @xmath11 such that @xmath530 is a smooth lagrangian submanifold of @xmath531 . 2 . there exist @xmath532 simple along @xmath533 at @xmath534 @xmath535 such that @xmath536 consider the projections @xmath537 and the diagonal embedding @xmath538 then we have an isomorphism @xmath539 by , we get @xmath540 here the involution @xmath541 is defined by @xmath542 moreover @xmath543 is simple along @xmath544 by . furthermore , by we find that @xmath545 is non - characteristic for @xmath546 . hence we obtain an isomorphism @xmath547 where @xmath548 is the relative dualizing complex . by the assumption that @xmath7 and @xmath8 intersect transversally , @xmath549 is transversal to @xmath544 at @xmath550 for any @xmath372 . thus ( i ) follows ( i ) , and ( ii ) follows from ( ii ) and ( ii ) . [ lem : conicclosure ] one has @xmath551 in other words , @xmath552 _ _ ( _ _ see subsection @xmath553 for the notation @xmath554 , and @xmath555 associated with the constant linear map @xmath556 . _ ) _ set @xmath557 then @xmath558 is equal to @xmath559 . for the fist equality , it suffices to check the equality for @xmath560 . by the compactness of @xmath7 and @xmath8 , there exists @xmath561 such that @xmath562 for any @xmath563 . therefore the same inequality holds on the closure @xmath564 of @xmath306 . hence if @xmath560 then @xmath565 . by , , and , the set @xmath566 can be estimated as @xmath567 hence we obtain @xmath568 [ prp : localcohr ] let @xmath569 . then one has @xmath570 take a relatively compact contractible open neighborhood @xmath141 of @xmath143 and a sufficiently small @xmath571 , and consider an open neighborhood @xmath572 of @xmath573 in @xmath77 . then there exists a sufficiently large @xmath460 such that @xmath574 is constant on @xmath575 and @xmath576 is constant on @xmath577 . thus we have isomorphisms @xmath578},k_{m \times \br \times \br}[1])|_{s^{-1}(u_0 ) } \\ \simeq \ , & \rg_{s^{-1}(u_0 ) \cap ( m \times \br \times ( -\infty ,- a])}(k_{s^{-1}(u_0)})[1 ] . \end{split}\ ] ] here the first isomorphism follows from ( ii ) , since the fibers of the submersion @xmath518 are one - dimensional and @xmath579 is contractible . therefore we obtain @xmath580 by the distinguished triangle @xmath581 with @xmath582 is supported in some compact subset , we find that @xmath583 and @xmath584 is proper on @xmath585 . replacing @xmath574 by @xmath582 , we may assume that @xmath584 is proper on the support from the beginning . the inclusion shows that @xmath586 where @xmath587 satisfies @xmath588 at each point @xmath589 , by there exist smooth lagrangian submanifolds @xmath590 of the forms as in and @xmath591 simple along @xmath590 at @xmath592 @xmath593 such that @xmath594 and @xmath595 . moreover , by the transversality of @xmath7 and @xmath8 , @xmath596 is transversal to @xmath590 at @xmath592 for any @xmath371 and @xmath372 . by ( i ) , for any @xmath180 , there exists a sufficiently small neighborhood @xmath366 of @xmath337 such that @xmath597 are smooth lagrangian submanifolds of @xmath63 . furthermore , each @xmath598 is of the form @xmath599 where @xmath175 is a sufficiently small neighborhood of some point @xmath600 satisfying @xmath601 . hence the transversality of @xmath7 and @xmath8 implies that the function @xmath81 is transversal to @xmath598 for any @xmath371 and @xmath372 . thus the result follows from ( ii ) . [ prp : homstarineq ] let @xmath170 in @xmath44 or @xmath171 . assume that @xmath602 then one has an inequality @xmath603 by and , we have @xmath604 . hence by we obtain @xmath605 set @xmath606 . we shall apply the morse inequality for sheaves ( ) to @xmath607 and the function @xmath608 . combining with , we get @xmath609 and the intersection is finite . moreover by , the inclusion is in fact an equality . let @xmath610 . we also set @xmath611 then by , we get @xmath612 by , we obtain @xmath613 which shows @xmath614 combining with , we have the conclusion . c. viterbo announced the relation between the sections of @xmath76 on @xmath615 and the floer cochain complex @xmath616 filtered by @xmath617 . motivated by his work , we use sections not only on @xmath77 but also on @xmath618 in . [ thm : main ] one has an inequality @xmath619).\ ] ] first recall the isomorphism in @xmath73 ) \simeq \varinjlim_{c \to + \infty } \hom_{\cd(m)}(f_1,{t_c}_*f_2[l]).\ ] ] by , there exists a sufficiently large @xmath488 such that @xmath620 ) \simeq \hom_{\cd(m)}(f_1,{t_c}_*f_2[l])\ ] ] for any @xmath491 and @xmath183 . moreover , replacing @xmath621 by a larger constant if necessary , we may assume that @xmath622 by the uniqueness of the canonical quantization ( ) , we have an isomorphism @xmath623 . hence , replacing @xmath624 by @xmath625 , we may assume that holds for any @xmath626 and @xmath627 from the beginning . thus it suffices to show that @xmath628).\ ] ] recall also the isomorphism in @xmath629 ) \simeq h^l_{m \times [ 0,+\infty)}(m \times \br ; \chom^\star(f_1,f_2)).\ ] ] hence the result follows from for the case @xmath630 and @xmath631 . let @xmath7 and @xmath8 be compact connected exact lagrangian submanifolds of @xmath11 intersecting transversally . then one has an inequality @xmath632 it follows from and . in this section , we prove the `` functoriality '' of guillermou s canonical sheaf quantization with respect to hamiltonian diffeomorphisms . we remark that results in this section are independent of the results in the previous sections and not used for the proofs of them . let @xmath61 be a compact connected exact lagrangian submanifold of @xmath11 and @xmath111 a primitive of the liouville form @xmath633 . we define the conification @xmath457 of @xmath61 with respect to @xmath111 as in . let @xmath5 be a hamiltonian diffeomorphism of @xmath11 and @xmath438 be a hamiltonian isotopy , where @xmath43 is an open interval containing @xmath439 $ ] , such that @xmath3 and @xmath47 . we denote by @xmath634 the associated hamiltonian and by @xmath635 the associated hamiltonian vector field on @xmath11 . the homogeneous lift @xmath51 of @xmath386 is described as follows ( see ( * ? ? ? * proposition a.6 ) ) : @xmath636 where @xmath637 and @xmath638 hence we get @xmath639 on the other hand , we have equalities @xmath640 here , for a vector field @xmath1 , @xmath641 denotes the lie derivative with respect to @xmath1 , and the third equality follows from cartan s formula . moreover , the fourth equality follows from the definition of the hamiltonian vector field : @xmath642 . hence setting @xmath643 , we get @xmath644 thus we find that @xmath645 is a primitive of @xmath633 on @xmath646 and obtain the following . let @xmath648 be the canonical sheaf quantization of @xmath457 . let @xmath36 be a hamiltonian diffeomorphism and @xmath649 the associated functor _ _ ( _ _ see _ ) _ . then one has an isomorphism @xmath650 where @xmath643 . by , we have @xmath651 by the uniqueness of the canonical quantization ( ) , it remains to show that @xmath652 let @xmath48 be the associated homogeneous hamiltonian isotopy and @xmath440 its sheaf quantization . consider the composite @xmath653 . by the compactness of @xmath62 , there exists @xmath460 satisfying @xmath654 } \wh{\phi}_s(\wh{l } ) \subset t^*(m \times ( -a , a)).\ ] ] if necessary , shrinking @xmath43 , we may assume that @xmath655 from the beginning . set @xmath656 . we shall show that @xmath657 first by , we have @xmath658 by the definition of @xmath53 ( see ) , we obtain @xmath659 denote by @xmath660 the closed embedding for any @xmath397 . then by the definition of @xmath53 , we also have @xmath661 moreover by , we get @xmath662 for any @xmath397 . hence the inclusion follows from the above estimates , , and . since @xmath43 is contractible , we have an isomorphism @xmath663 where @xmath664 is the projection . in particular , we get @xmath665 and @xmath666 . a similar argument shows that @xmath667 . k. fukaya , p. seidel , and i. smith , the symplectic geometry of cotangent bundles from a categorical viewpoint , in _ homological mirror symmetry _ , vol . * 757 * of _ lecture notes in phys . _ , 126 , springer , berlin , 2009 . s. guillermou and p. schapira , microlocal theory of sheaves and tamarkin s non displaceability theorem , in _ homological mirror symmetry and tropical geometry _ , 15 * of _ lect . notes unione mat . _ , 4385 , springer , cham , 2014 .

ion coulomb crystals ( icc ) , formed by atomic ions at low temperatures in radiofrequency and penning ion traps , are structures that have remarkable properties and many applications . images of coulomb crystals are striking and reveal the crystal structure , which arises from a balance between the trapping forces acting on the ions and their mutual coulomb repulsion . applications of these structures range from frequency standards and quantum simulation through to measurement of the cross sections of chemical reactions of ions .

1411.4945

many solids exist in the form of a crystal , where atoms are arranged in a regular pattern and have fixed positions relative to each other . such crystals are normally three - dimensional but atoms can also , in some cases , form two - dimensional structures such as graphene . the equilibrium distances between atoms are essentially determined by the overlap of the atomic wavefunctions , resulting in typical inter - atomic distances of 0.3 nm and a density of the order of 3@xmath0 m@xmath1 . when it is heated , a crystal will often undergo a transition to a liquid , in which the density remains similar to that of the crystal but the atoms no longer have fixed positions relative to each other . atomic ions can also form crystal structures when they are held in ion traps . ion traps are devices that use a combination of electric and magnetic fields to confine the motion of charged particles ( in this case , atomic ions ) to a small region of space . when the ions have low energies , they accumulate at the centre of the trap , but since they repel each other due to the coulomb force , there is a limit to how close they can approach each other . in equilibrium , and at low enough temperatures , the ions will then form a regular crystal - like structure with the inter - particle spacing determined by the balance between the trapping fields and the coulomb repulsion between the ions . for typical values of the trapping fields ( as discussed below ) this results in a spacing of the order of 10@xmath2 m and hence a density of the order of @xmath3 m@xmath1 . this is many orders of magnitude lower than the density of conventional crystals and even of the air around us . in fact , it is comparable to the density of residual gas molecules in an ultra - high vacuum system ( such as would be used to enclose the ion trap ) . these novel crystal structures therefore have many interesting properties that contrast with those of conventional crystals . on the other hand , there are also some similarities of behaviour , even though the densities differ by many orders of magnitude . this review discusses the formation , structure and applications of ion coulomb crystals ( icc ) . it is arranged as follows . section [ sec : traps ] presents the basic physics of ion traps and cooling techniques ; section [ sec:1dicc ] discusses the properties of one - dimensional crystals ( i.e. strings or chains of ions ) while section [ sec:3dicc ] deals with the three - dimensional case . section [ sec : applications ] discusses applications including experiments to study phase transitions and quantum mechanical effects in ion coulomb crystals . finally , section [ sec : conclusions ] presents some conclusions . ion traps were first developed in the late 1950s and the three - dimensional radiofrequency trap , often called the paul trap , was developed in bonn by the group of wolfgang paul , who was awarded a nobel prize in 1989 for his work in this area @xcite . there are many reviews of ion traps available which explain in detail how the radiofrequency trap works and discuss the many applications that have been developed : see for example @xcite . ) is typically of the order of 1 mm ( see text).,width=188 ] in this review , we will concentrate mainly on a two - dimensional version of the paul trap , called the linear radiofrequency ( rf ) trap . briefly , the linear rf trap works by using a set of four electrodes ( usually made from circular rods : see figure [ fig : linear ] ) to generate a quadrupole potential in a region of space under ultra - high vacuum conditions . the rods are aligned parallel to the @xmath4-axis and the radiofrequency potential is required in order to confine ions in the @xmath5-plane , i.e. to push them towards the central @xmath4-axis . it is not possible to achieve this radial confinement by applying a static ( dc ) potential . to see this , consider the effect of applying a positive potential to two rods opposite each other , with the other two electrodes grounded . this would tend to push a positively - charged ion away from the positively charged rods and towards the @xmath4-axis , but the ion would also be attracted to the two other rods . this would lead to the loss of the ion . so with a static potential it would be possible to trap in ( say ) the @xmath6-direction but only with the motion in the orthogonal @xmath7-direction being unstable . reversing the sign of the applied voltage would have the opposite effect , giving stable motion along @xmath7 and unstable motion along @xmath6 . however , it turns out that if the applied potential is made to oscillate between positive and negative values , such that the stable and unstable directions are constantly reversing , the final time - averaged motion of the ion can be stable in _ both _ directions simultaneously , so long as the oscillation frequency is set to be within a certain range @xcite . the average applied potential in this situation is zero , so it is not intuitively obvious that stable trapping can be achieved , but this can be demonstrated from a careful mathematical analysis of the system @xcite . this analysis shows that the final motion consists of an oscillation of the ion in the plane perpendicular to the @xmath4-axis of the trap as if it were sitting in an effective potential that had a two - dimensional minimum along this axis . this is called the _ pseudopotential _ and this motion is referred to as the _ secular motion_. superimposed on this motion is a faster and smaller - amplitude oscillation at the frequency of the applied field , called the _ micromotion_. the pseudopotential arises because this driven micromotion takes place in a nonuniform field such that although the electric field at any point averages to zero , the average force acting on the ion is non - zero because it is moving around inside the trap . the motion is properly described by a mathieu equation and solution of this equation gives an expression for the oscillation frequencies but where we give a numerical value for the frequency we will quote the value of @xmath8 in hz . ] : @xmath9 where @xmath10 is the angular frequency of the applied rf potential ( with amplitude @xmath11 ) and @xmath12 is given by @xmath13 here @xmath14 is the internal radius of the trap ; @xmath15 and @xmath16 are the charge and mass of the ion respectively . equation [ eq : omega ] is valid for @xmath17 . equation [ eq : q ] is exact if the electrodes have a hyperbolic rather than a circular cross - section . circular electrodes do not produce a pure quadrupole potential like hyperbolic ones do , but they are easier to manufacture . the additional higher order terms they produce in the potential are not important close to the trap axis . it is also necessary to confine the particle s motion in the @xmath4-direction , i.e. along the axis of the trap , and this is achieved by applying an additional static potential at the two ends of the trap . this is sometimes done by segmenting the rf rods and adding a dc potential to the end segments , and sometimes it is done by including extra _ electrodes at the ends of the rods . in either case the result is a harmonic potential along the axis that pushes ions back towards the central part of the trap . this potential is usually relatively weak , with the confinement in the radial direction ( from the rf pseudopotential ) much stronger . this in turn means that the ion oscillation frequencies are generally higher for the transverse ( @xmath5 ) motion and lower for the axial ( @xmath4 ) motion . linear rf traps can vary in size but for the experiments to be described here , typical dimensions involve a separation of the rod electrodes on the order of 1 mm with the trapping region a few mm in length . the applied frequency would be typically 1 to 10mhz and the resulting secular oscillation frequencies would be in the region of a few hundred khz . however , some traps may be several mm across and others , manufactured using microfabrication techniques , are as small as 100@xmath2 m and typically have much higher oscillation frequencies @xcite . we have described the trapping mechanism for a single ion but the same principles also apply when many particles are being trapped . however , it is clear that although all ions will feel a force towards the centre of the trap , eventually there must come a point at which the coulomb repulsion between the ions balances the trapping potential . this leads to a maximum density of particles that can be accommodated in the trap . the effect can be described in terms of the space charge potential that builds up as the particles accumulate . it can be shown that space charge effects lead to a uniform density of particles in the trap at low temperatures , which has important consequences for large ion coulomb crystals . an important feature of all radiofrequency ion traps is the presence of micromotion . this is the small - amplitude motion of ions which is driven directly by the applied radiofrequency field . since the pseudopotential arises as a result of this motion , it is an unavoidable feature of rf traps . it is important because it can lead to heating up of an ion cloud ( in effect , it can couple the energy stored in the rf field into the secular motion of the ions ) . it will also lead to doppler broadening effects in the spectra of ions in traps . for these reasons , it is often desirable to minimise or eliminate the micromotion . in the linear trap , the rf field is confined to a plane perpendicular to the @xmath4-axis . since the field approximates to a quadrupole field ( remember it is generated by four electrodes ) , the amplitude of the field ( and the resulting micromotion ) rises linearly with the distance from the axis of the trap . this means that along the axis of the trap ( i.e. along the line @xmath18 ) there is no micromotion . this is a point to which we will return later . the addition of a uniform magnetic field @xmath19 along the trap axis ensures three - dimensional trapping . the separation of the endcaps is @xmath20 and the internal diameter of the ring is @xmath21 , which is typically of the order of 1 cm ( see text).,width=302 ] the penning trap offers an alternative way of confining the motion of ions . in this type of trap , three - dimensional confinement is achieved by using three electrodes ( a ring and two endcaps ) having the shape of hyperboloids of revolution about the @xmath4-axis ( see figure [ fig : trap ] ) . the hyperbolic shape of the electrodes results in a quadrupole electrostatic saddle potential given by @xmath22 where @xmath23 is the applied potential , @xmath20 is the separation of the endcaps and @xmath21 is the internal diameter of the ring electrode . a positive potential @xmath23 between the endcaps and the ring confines the axial motion of a positively - charged ion because the ion is repelled from both endcaps . the quadrupole potential well therefore gives rise to an axial oscillation of the ion at a frequency @xmath24 . the radial motion is unstable because the ion will be attracted towards the ring electrode , which encircles the @xmath4-axis . however , the addition of a strong axial magnetic field @xmath19 stops the ion moving towards the ring and instead forces the motion into a combination of circular orbits in the radial plane ( a slow magnetron motion at @xmath25 combined with a faster modified cyclotron motion at @xmath26 ) . in this way , three - dimensional confinement is achieved @xcite . the three trap oscillation frequencies are given by @xmath27 @xmath28 @xmath29 where @xmath30 , the pure cyclotron frequency , is equal to @xmath31 . one advantage of the penning trap is that there is no micromotion ; however , the energy associated with the magnetron motion is negative , due to the negative radial potential energy described by equation [ eq : saddle ] , and this results in complications for the stability of ions in the trap and the effectiveness of laser cooling , as will be discussed below . the internal diameter of a penning trap ranges from a few mm to a few cm , depending on the application . as a very rough guide , a magnetic field of a few tesla ( usually from a superconducting magnet ) gives typical oscillation frequencies in the range 100khz to 1mhz for atomic ions of interest . penning traps are often constructed from a stack of open cylindrical electrodes rather than hyperbolic electrodes as shown in figure [ fig : trap ] . this is because cylindrical electrodes are much easier to manufacture , and in addition , good optical access is generally easier to achieve . it is possible to design such electrodes to eliminate higher - order terms in the potential they produce close to the centre of the trap @xcite . in order to observe crystals in ion traps , it is necessary to cool the ions , i.e. to reduce their kinetic energy @xcite . since the rf micromotion amplitude in a linear trap depends on the amplitude of the secular motion , it is sufficient to remove energy from the secular motion . although there are several other methods available for cooling of particles in traps , including resistive cooling @xcite and buffer gas cooling @xcite , the method that is most important for ion coulomb crystals is laser cooling @xcite . this is because it is the only cooling method that is able to reduce the temperature of ions to low enough values for crystallisation to be observed . laser cooling is based on the exchange of momentum between photons from a laser and the ions . each time an ion in the trap absorbs a photon from a laser beam tuned close in frequency to a strongly allowed transition from the ground state to an excited state of the ion , the ion absorbs the momentum carried by the photon ( equal to @xmath32 ) . in order to make sure that the effect of this is to slow the ion down , it is necessary for absorption to take place only when the ion is moving towards the laser . this can be done by making use of the doppler effect , which is the shift in the frequency of the light as seen by the ion , as a result of its motion . by setting the laser frequency slightly below resonance , the ion will only see the light as resonant if it is moving towards the laser ; if it is moving away from the laser , the doppler shift takes it further out of resonance and no light is absorbed . this process , when repeated many thousands of times , is able to slow ions down from the high energies ( typically a few electron volt ) which they have when they are created inside the trap or when they enter the trap from elsewhere in the apparatus . however , there is a limit to how low a temperature can be reached in this way . this limit comes from the fact that the light must be re - radiated from the ion before it can absorb another photon ( this takes place on average after one lifetime of the excited atomic state : for resonance transitions of ions this lifetime is usually of the order of a few ns ) . the emission of a photon is in a random direction , so although the recoil of the ion averages to zero , it gives rise to a random walk in momentum space which results in an average kinetic energy that is determined by the balance between cooling from the laser and heating from the recoil . this generally corresponds to an equilibrium temperature of around 1 mk for typical ions of interest @xcite . this technique is called _ doppler cooling _ as it depends on the doppler effect ; there are other types of laser cooling that are able to reach lower temperatures . these are not generally used for work with large ion crystals , but techniques such as raman cooling @xcite and optical sideband cooling @xcite can be applied to small ion strings in a linear rf trap . this sub - doppler cooling is essential for much of the research in quantum information processing carried out using ion traps . laser cooling can be applied only to a limited number of different ion species . this is because an ion must have a suitable resonance transition available at a wavelength that can be reached by continuous - wave lasers , and the energy level structure of the ion must be such that the ion can cycle rapidly between its ground state and the excited state reached by absorption of laser light . this is only possible directly with mg@xmath33 , be@xmath33 and hg@xmath33 ions . however , if an ion has a metastable ( long - lived ) state to which it can decay from its excited state , it is sometimes possible to use a second laser to bring the ion back up to the excited state , so that it can resume the cooling process . this allows a number of other ions to be used for laser cooling experiments , including ca@xmath33 , sr@xmath33 , yb@xmath33 and ba@xmath33 . the total number of singly - charged ion species that can be laser cooled is about ten @xcite . there are no suitable doubly - charged ions because the laser wavelengths required are too far into the ultraviolet where no tuneable continuous - wave lasers are available . from the point of view of the physics of ion coulomb crystals , the species of ion being used to create the crystals is unimportant ( so long as it can be laser cooled effectively ) . this is because the physics of the crystal depends only on the temperature of the ions ( see the next section ) and on the coulomb repulsion between the ions . at typical distances of several @xmath2 m , the detailed electronic structure of the ions is completely irrelevant as they behave like point charges . the simplest type of ion coulomb crystal is a linear string of ions along the axis of an ion trap . this arises most naturally in the case of a linear rf trap , where the confinement along the axis of the trap is typically much weaker than the radial confinement . if several cold ions are present in a trap , they therefore tend to line up along the trap axis it is this configuration that has the lowest potential energy . for @xmath34 ions along the trap axis , the total potential energy of the system is given by @xmath35 the equilibrium positions of the ions can be found by minimising this energy as a function of the @xmath36 . the resulting positions have been tabulated by james @xcite and have been confirmed in many experiments . the length scale of the linear crystals formed in this way can be found by evaluating the equilibrium separation of two ions , @xmath37 . this is the separation at which the inward force due to the trapping potential is exactly balanced by the outward repulsion of the ions , i.e. @xmath38 giving @xmath39 m for calcium ions having an axial oscillation frequency @xmath24 of @xmath40500khz . as the number of ions increases , the separation between adjacent ions at the centre of the string reduces approximately as @xmath41 @xcite . ( figure courtesy of t. mehlstubler , ptb).,width=302 ] strings of ions in a trap were first reported in 1992 @xcite . a typical recent image of a string of ions in shown in figure [ fig : strings](a ) @xcite . the ions are illuminated by laser light that is also used for laser cooling and is therefore resonantly scattered by the ions . very long strings can also be formed ( see for example the images in @xcite ) . in these long strings it can be seen that the separation of adjacent ions is a smooth function of position , with its minimum value at the centre of the string . sometimes an ion of a different species may be incorporated into one of these strings , but does not emit any fluorescence because it does not have a resonant response to the laser frequency . in this case the dark ion occupies a site that would otherwise be occupied by a bright ion . the dark ion is said to be sympathetically cooled by the other ( laser cooled ) ions , and this happens through the coulomb interaction , which couples the motions of all the particles @xcite . the string of ions can be considered as a set of coupled oscillators and as such the small - amplitude oscillations of the system are best treated by finding the normal modes of oscillation of the string . for oscillations along the @xmath4 direction , the normal modes can be found starting from the total energy given in equation [ eq : energy ] above @xcite . for all values of @xmath34 , the first ( lowest frequency ) mode is the centre of mass mode at @xmath24 ( where all the ions move together without changing their separations ) and the second mode ( called the _ breathing mode _ , where the whole string expands and shrinks symmetrically with respect to the central point ) is always at a frequency of @xmath42 . the remaining axial modes are at higher frequencies and involve more complicated motions . the oscillation modes of small axial strings of ions are important for quantum information processing applications , as discussed in section [ sec : qip ] . if the ion string is excited by an additional , weak , force ( e.g. from an oscillating potential applied to the electrodes ) , the string will respond resonantly when the excitation frequency matches the frequency of one of its normal modes . by then taking images at different phases of the exciting field , it is possible to build up a sequence of images showing the evolution of the normal modes of the string @xcite . the string also has transverse oscillations , where the motions of the ions are perpendicular to the trap axis . these oscillations include the radial centre of mass mode , which has the single - ion radial oscillation frequency @xmath43 . the other radial modes are all at lower frequencies . the lowest frequency mode , the so - called zigzag mode , has adjacent ions moving in opposite directions . if the radial confinement is relaxed , the frequency of this zigzag mode approaches zero and at that point the ion string becomes unstable and acquires a permanent zigzag displacement @xcite ( see section [ sec : phase ] ) . this is illustrated in figure [ fig : strings](b ) . m. * ( b ) * collage of linear crystals of 19 ions . the applied voltage was kept constant for all of these experimental images . each pixel is equivalent to @xmath44@xmath2 m in the centre of the trap . the circles around the ions are the calculated positions , from ref . @xcite , and not from a fit to the data . the bars at the bottom of the images represent a length of 50@xmath2 m . reprinted by permission from macmillan publishers ltd : nature communications 4 , 2571 , copyright 2013.,width=377 ] axial strings can also be observed in penning traps , as has recently been demonstrated by the imperial group @xcite . images of strings containing one to nine ions are shown in figure [ fig : penningstrings ] , along with a longer string containing 29 ions . the physics of this arrangement is identical to that of the linear rf trap as the axial confinement is the same in both cases . however , the traps differ in the way they achieve radial confinement , and this affects the formation of three - dimensional crystals ( see section [ sec:3dicc ] ) . a third type of trap in which long strings of ions can be observed is the ring trap introduced by walther s group at mpq in garching @xcite . this is in effect a very long linear rf trap which is curved round to make a complete ring . it therefore does not have any axial confinement so ions are free to move around the ring . this trap can accommodate an extremely large number of ions in a string with a uniform spacing . although in principle the ions would move freely around the trap if it were perfect , in reality there will always be places where additional potentials ( e.g. from deposits on the surface of the electrodes ) act to give a potential barrier that stops the ions moving past that point . this means that the ions can be laser cooled in a similar manner to the case of the linear rf trap . the mpq group found that there was a critical linear density up to which the linear string was stable , and for densities greater than this the string began to kink and eventually to form cylindrical shells , as shown in figure [ fig : birkl ] @xcite . we return to this point in section [ sec : applications ] . in a radiofrequency ring trap @xcite . the structure of the crystal changes as the linear density of ions increases . in ( a ) the ions form a simple string ; in ( b ) they have a zigzag configuration ; in ( c ) the ions form a structure consisting of two interwoven helices and in ( d ) there are three interwoven helices . visualisations of the structures in ( c ) and ( d ) are given below . reprinted by permission from macmillan publishers ltd : nature 357 , 310 , copyright 1992.,width=453 ] finally , even longer structures can be observed in an ion beam moving in a storage ring . this was first demonstrated using the pallas table - top storage ring @xcite , similar in design to the ring trap described above @xcite . the sign that crystallisation of the circulating beam had taken place was the sudden narrowing of the transverse distribution of the ions in the beam . a comprehensive review of this topic has been given by schramm and habs @xcite . ions confined in an ion trap constitute an example of a plasma . however , unlike a normal plasma where there are equal numbers of positively and negatively charged particles , a cloud of ions has only one sign of charge . in effect the presence of the confining potential is equivalent to the presence of a uniform density of particles of the opposite charge to the confined ions . this type of plasma is therefore referred to as a _ single - component plasma _ @xcite . it is well known from classical physics considerations that charged particles in a plasma will form a crystal structure when cold enough . this has been studied extensively ( see , for example , @xcite ) . the behaviour of the plasma is best described in terms of the so - called _ coulomb coupling parameter _ , @xmath45 , which is given by @xmath46 where @xmath47 is called the _ wigner - seitz radius _ , and the other symbols have their usual meanings . the wigner - seitz radius is a measure of the average distance between particles and it is related to the number density @xmath48 by @xmath49 . the coupling parameter is therefore essentially the ratio of the nearest - neighbour coulomb energy to the average thermal energy of the particles . simulations show that plasmas behave like a gas when @xmath45 is below 1 ( called the _ weak coupling regime _ ) and like a liquid when @xmath50 ( called the _ strong coupling regime _ ) . at higher values of @xmath45 the plasma forms a regular crystal - like structure . for an infinite plasma this phase transition occurs at @xmath51 @xcite . a simple calculation allows us to estimate the parameters required for a coulomb crystal to form in an ion trap : assuming a typical doppler - cooled temperature of 1 mk we find that the inter - ion distance should be less than about 10@xmath2 m and the density should be at least 2.5@xmath52 m@xmath1 ( or 2.5@xmath53 cm@xmath1 ) . this is achievable fairly easily with linear rf traps having dimensions of the order of mm and rf drive voltages of a few hundred volt at several mhz . large icc generally have a spheroidal shape with the aspect ratio determined by the ratio of the strength of confinement ( i.e. the trap oscillation frequencies ) in the radial and axial directions . they have a uniform density which depends on the trapping strength . icc containing a small number of ions behave in a similar manner but the exact configuration of the ions needs to be found by calculation or simulation . extensive numerical simulations have been performed to study the formation of ion coulomb crystals . these have shown that for infinite systems the minimum energy state of the crystal has a body - centred cubic crystal structure . however , for finite systems other structures can appear such as spherical @xcite or cylindrical @xcite shells . crystallisation occurs at a similar value of @xmath45 to that for the infinite case , but the process tends to start at lower values of @xmath45 . in these cases there may be parameters for which the central part of the cloud of ions is crystallised but not the outer part . the group of schiller have carried out simulations of these systems that are able to reproduce the observed configuration of the ions in a three - dimensional crystal very accurately . the simulations are based on a detailed calculation of the forces acting on ions , including the confining potential , the coulomb repulsion , stochastic forces to represent heating effects such as collisions with residual gas molecules , and the interaction with the cooling laser @xcite . micromotion can be included explicitly if necessary but this lengthens the time required for the simulation to run as the time step then needs to be reduced in order to follow the fast motion at the applied radiofrequency . however , its effects are not too severe when the cloud is crystallised and it can often be ignored . the simulations reproduce well the process of crystallisation of the ions and show how the crystal ends up in its final configuration . the results of these calculations are simulated images of the ions which can then be compared to experimentally - obtained images ( see section [ sec : rfcrystals ] ) . by including the effect of heating and cooling processes , simulated images corresponding to different ion temperatures can be made . similar work has been carried out by other groups using different approaches to model the effect of a finite temperature @xcite . in the simulations it is also possible to include ions of other species , in order to see their effect on the observed image , which of course does not show any fluorescence from other species as they will not be resonant with the laser . typically , ions of different mass - to - charge ratios will form concentric shells , with the heavier ions further away from the trap axis . the presence of a shell of high mass - to - charge ratio ions can be seen in the shape of the inner fluorescing part of the crystal , which is now flattened into a cylinder towards the centre rather than a complete spheroid @xcite . however , for ions with very close mass - to - charge ratios ( e.g. different isotopes of the same element ) , this separation does not take place but the fluorescing ions are pushed towards one end of the crystal due to the light pressure from the axial laser beam . this is illustrated in figure [ fig : schiller ] where different ionic species separate into different regions of the crystal @xcite . the behaviour of clouds of ions containing different species is an example of sympathetic cooling , which is the process by which ions of one laser - cooled species are able to cool ions of another species in the same trap through the coulomb interaction . ba@xmath33 ions ( blue ) , 300 sympathetically cooled @xmath55ba@xmath33 ions ( red ) , 100 co@xmath56 ions ( pink ) and 200 organic molecular ions with a mass of 470 amu ( green ) . the simulation in ( b ) has a higher laser cooling rate than in ( a ) , leading to a higher equilibrium temperature for the sympathetically cooled species ( which can be seen in the blurring of the green outer layer . reprinted figure with permission from c. b. zhang _ et al . _ , physical review a 76 , 012719 , 2007 . copyright 2007 by the american physical society.,width=453 ] there are some differences in the way that ion coulomb crystals form in a penning trap compared to an rf trap , due to the fact that the whole crystal rotates . this rotation frequency can take a range of different values and the exact configuration of the crystal ( in effect , its aspect ratio ) adjusts as the rotation frequency changes . this is because the effective radial potential in the frame rotating with the crystal depends on its rotation frequency . simulations of crystals in a penning trap must take this rotation into account . in ref @xcite the many different structures observed for a 15-ion crystal could be reproduced in simulations , and this allowed the rotation frequency of the crystal to be estimated ( see section [ sec : penningcrystals ] ) . in this case the simulations do not attempt to follow the process of crystallisation , but simply find iteratively the lowest energy configuration of the system under the influence of the trapping fields , the crystal rotation and the coulomb repulsion . other physical systems can also show this sort of crystallisation . the idea was first suggested by wigner in the 1930s @xcite in a discussion about the states of electrons in metals . realisation of a wigner crystal for electrons in three dimensions is difficult due to quantum mechanical effects . however , two - dimensional crystallisation of electrons on the surface of a superfluid has been observed @xcite . for a full discussion about the theory and a review of experimental evidence for wigner crystallisation of electrons , see the review by tsidilkovski @xcite . many groups have observed and studied the formation of ion coulomb crystals in radiofrequency ion traps . the first observations were made in a three - dimensional paul trap , where the strength of the confinement is similar in all three dimensions and the micromotion is zero only at the exact centre of the trap rather than along a line as in the linear rf trap . images of small numbers of laser - cooled ions in different configurations were obtained in this way @xcite . much larger numbers of ions can be crystallised in linear radiofrequency ion traps due to the lower level of micromotion in these traps . images have been obtained for more than 10@xmath57 ions in these traps . given the linear symmetry of the trap itself , the ions tend to arrange themselves in open or closed cylindrical shells @xcite . since the trapping potential felt by an ion depends on the mass to charge ratio for that ion , different species experience different forces and this results in a separation of the different species , as discussed in section [ sec : sims ] . it is notable that even though laser cooling can only be applied to one species , a cloud of ions consisting of more than one species can be maintained at a temperature that is low enough for the whole system to remain crystallised by means of sympathetic cooling . for example , in one experiment a cloud of 15 ions was kept crystallised by a single laser - cooled ion in this way @xcite . ca@xmath58/@xmath59mg@xmath58 bi - crystal at three different end cap voltages in a linear ion trap @xcite . the crystal is symmetric under rotations about the trap axis , @xmath4 , and contains approximately 300 @xmath59mg@xmath58 ions ( red ) and 3000 @xmath60ca@xmath58 ions ( blue ) . the aspect ratios of the crystals are determined by the ratio of axial to radial trapping frequencies for the two species . this parameter is always higher for @xmath60ca@xmath58 than for @xmath59mg@xmath58 and it increases ( for both species ) in the sequence ( a ) to ( c ) . reprinted figure with permission from m. drewsen _ et al . _ , physical review letters 81 , 2878 , 1998 . copyright 1998 by the american physical society.,width=453 ] figure [ fig : drewsen ] shows images of an ion crystal containing two species ( @xmath60ca@xmath58 , blue , and @xmath59mg@xmath58 , red ) . the images are obtained by superimposing light from the two different species in two separate exposures . the strength of the confinement for mg@xmath33 is always stronger than for ca@xmath33 due to its lighter mass , so it is always positioned along the trap axis with the ca@xmath33 around it in a ring . note that the overall shape of the crystal is always roughly ellipsoidal even though it consists of two different species @xcite . in the experiments of the schiller group in dusseldorf it was shown that is possible to determine the exact number of ions present by careful observation of the detailed shape of the crystal and comparison with the results of computer simulations @xcite . in the same manner , comparison of simulated images of ion crystals that include thermal effects with experimentally obtained images allows the ion crystal temperature to be estimated . because the number of ions in the trap can be determined so accurately , this system can be used to study the rates of chemical reactions between trapped ions and neutral gas in the chamber ( see section [ sec : reactions ] ) . crystals of ions in linear rf traps tend to be prolate , that is , long and thin like a cigar . this is because the strength of the radial confinement is generally much greater than that of the axial confinement . in the extreme case , the crystal consists of a long string of ions on the axis ( see section [ sec:1dicc ] ) . on the other hand , the relative strengths of the axial and radial confinement in a penning trap can be adjusted over a wider range , and in particular it is possible to adjust the parameters of the trap to change the configuration of a small crystal from a linear string , through different three - dimensional structures , to a planar crystal . the bollinger group at nist has been most active in this area and has worked routinely with crystals that consist of a single two - dimensional plane containing hundreds of ions @xcite . there are two main problems with the creation and observation of ion coulomb crystals in the penning trap . the first problem is that laser cooling is not very effective for the magnetron motion , which for a single ion is a slow orbit around the centre of the trap . because the total energy of this motion is negative , energy has to be supplied to the ion to reduce the amplitude of the magnetron motion . one way to do this is to offset the laser cooling beam radially from the centre of the trap @xcite . this has the effect of applying a torque to the ion as well as providing damping . for large clouds of ions the equivalent of the magnetron motion is a global rotation of the whole cloud , which now behaves like a plasma . note that in the penning trap the crystal always rotates , due to the presence of the magnetic field . this rotation frequency , @xmath43 , is linked to the number density , @xmath48 , through the relation @xcite @xmath61 the density of the cloud is therefore maximum when the rotation frequency is @xmath62 , and this condition is termed _ brillouin flow_. it is often possible to reach this condition using just a radially offset laser beam . in the case of large clouds it is more effective to apply the torque in a different way , and the nist group has developed the _ rotating wall technique _ for this purpose @xcite . here a radial linear or quadrupole electric field , rotating at some frequency @xmath63 , perturbs the ion plasma and forces it to rotate at the same frequency . since this rotation frequency is linked to the density of the ions through equation [ eq : density ] , it has the effect of controlling the ion density and also the shape of the plasma @xcite . laser cooling can then be provided with an axial beam ; if the cooling is strong enough , the plasma will crystallise in the same way as in an rf trap . the rotating wall therefore allows the density and shape of the ion coulomb crystal to be controlled . ions in a penning trap @xcite . the crystal is rotating but the image is reconstructed using timing information from a position - sensitive photomultiplier . reprinted by permission from macmillan publishers ltd : nature 484 , 489 , copyright 2012.,width=377 ] the second problem with the penning trap is also a consequence of the fact that the crystal is always rotating . the rotation means that the crystal can not easily be imaged without blurring , so in order to obtain clear images it becomes necessary to gate the camera so that the ions are always imaged at the same point in their rotation . since the rotation is locked to the applied rotating wall frequency ( @xmath63 ) , this is relatively straightforward provided the camera is capable of fast gating . as an alternative , a position - sensitive photomultiplier can be used for imaging . in this way the bollinger group have obtained spectacular images of rotating crystals consisting of a single plane of ions ( see figure [ fig : hexagonal ] ) @xcite . a strong advantage of the penning trap for this type of work is the absence of micromotion , which would be significant in an rf trap for the ions far removed from the axis of the trap . ion crystals in the penning trap can form in a number of different configurations . in particular , for a two - dimensional crystal consisting of a single layer of ions , a triangular lattice is normally formed . as the rotation frequency of the crystal ( @xmath43 ) is increased ( thus effectively increasing the strength of the radial confinement ) , the crystal splits into two planes , but also changes to a square lattice @xcite . as @xmath43 is further increased , the square lattice becomes rhombic and eventually triangular again before undergoing another split into three planes . this process then repeats . the large single - layer planar crystals have interesting transverse modes of vibration and behave rather like vibrating drum - skins @xcite . the frequencies of these modes can be calculated and compared to experimentally measured values . in their experiments , the bollinger group at nist worked with a planar crystal containing several hundred ions in a single layer . with an inter - ion spacing of around 20@xmath2 m , the diameter of the crystal is a few hundred @xmath2 m . the modes were excited using an optical dipole force from a pair of laser beams inclined at a small angle to the plane of the crystal and detuned from the optical resonance of the be@xmath33 ions at 313 nm . the dipole force is spin - state dependent so by initially placing all ions in a superposition of their spin states , any mode can be excited by choosing a suitable frequency difference between the two beams . in this way the effective temperature of each mode could be determined . the measured temperatures were consistent with the expected doppler cooling limit of 0.4mk , except for the centre of mass mode , which had a higher temperature of a few mk . this arises because the pulses of light used for laser cooling were turned on and off in a very short time . the resulting radiation pressure , which was uniform across the whole crystal , excited the centre of mass mode , increasing its temperature above the doppler limit . in recent work the nist group has carried out quantum simulation experiments using this system ( see section [ sec : qip ] ) . conventional crystals of atoms can be investigated by means of bragg scattering of x - rays . this is a well - established technique for determining crystal structures : from the angles of scattered rays information can be obtained on the spacing of the atoms and the type of crystal lattice . in order for the technique to be effective , the wavelength of the radiation used must be comparable to the typical distances between atoms in the crystal ( i.e. less than 1 nm ) hence the use of x - rays . bragg scattering can also be used to study ion coulomb crystals and in this case , since the typical distance between ions is of the order of a few @xmath2 m , optical radiation has a suitable wavelength , giving scattering angles of a few degrees . this means that radiation at the ion s resonance wavelength , which is used for laser cooling , can also act as the incident radiation for bragg scattering . the nist group has used this technique with a three - dimensional crystal of be@xmath33 ions in a penning trap , to demonstrate different configurations of the crystal . bragg scattering from ions in rf traps is not possible due to the rf micromotion and the lack of long - range order in the crystals . in penning traps , medium - sized crystals with roughly spherical symmetry tend to form in concentric shells . long - range order starts to set in at a diameter of around 37@xmath47 but the predicted body - centred cubic ( bcc ) bulk configuration does not form unless the crystal has a diameter of around 65@xmath47 , meaning that the crystal contains at least 300 000 ions . there is clear evidence that under these circumstances the configuration is a bcc crystal @xcite . this can be deduced from the bragg scattering pattern , which is a series of concentric rings ( due to the rotation of the crystal ) with characteristic angular spacings . for smaller numbers of ions , a number of other crystal configurations were observed , because in this case the presence of the surface affects the arrangement of ions . these included crystals with 5-fold symmetry , observed through both the bragg pattern and direct imaging of the crystal @xcite . at the other extreme , the imperial group has used a penning trap to prepare and image small numbers of ions ( up to 20 ions ) in different configurations as the trapping parameters are varied . for low axial potential the expected configuration is an axial string , as explained above . as the axial potential is increased , first the string of ions kinks into a zigzag shape , and then it forms several three - dimensional structures ( depending sensitively on the number of ions ) until eventually it becomes a planar crystal . these different configurations can also be obtained from simulations that attempt to find the lowest energy configuration of the crystal for given trapping parameters , and a good agreement between the simulations and observations has been obtained ( see figure [ fig : zoo ] ) . in these experiments the rotation frequency of the ion crystal in the laboratory frame is not known , because there is no rotating wall applied . it can take a range of values , depending on the size , position and frequency of the laser beam , and this can be checked for consistency with the observed structures @xcite . note that for a small number of ions it is not strictly appropriate to describe the crystal shape as a spheroid . however , the overall aspect ratio and density of the ions are always very close to what would be expected for a larger number of ions , once surface effects are accounted for . m scale bar in the bottom right - hand pane which applies to all the images . reprinted by permission from macmillan publishers ltd : nature communications 4 , 2571 , copyright 2013.,width=642 ] there is nothing in the equations of motion of particles in rf traps that restricts the nature of the trapped particles to be atomic ions . at around the same time that paul was developing the first three - dimensional trap for atomic ions , a different research group was also working on a trap for charged dust particles with dimensions of the order of micrometres @xcite . the physics of this trap is identical to that of the now familiar rf trap but the parameters are rather different : because the mass to charge ratio of the particles is much higher , the frequency of the applied potential is typically around 100hz and the required amplitude is a few hundred volts . damping is provided by the air . figure [ fig : wuerker ] shows a photograph from this paper , demonstrating that the structures formed are very similar to those formed by atomic ions . note that in this photograph each particle s image becomes a line because of the relatively large micromotion in this system . as has been mentioned above , ion coulomb crystals undergo what might loosely be called phase transitions between different configurations as the trapping parameters are changed . strictly , this term should only be used in a system containing large numbers of particles but it is also convenient to use it here . consider first a long string of ions in a linear rf trap ( as in figure [ fig : strings](a ) ) . as the strength of the axial confinement of the ions is increased , there is a well - defined point at which the string changes into a zigzag shape ( see figure [ fig : strings](b ) ) . this would typically be referred to as a second - order phase transition . initially the degree of kinking into the second dimension is very small but rises rapidly as the axial confinement is further strengthened . in the zigzag configuration ( provided that the radial potential does not have perfect radial symmetry ) , the string has two stable states which are mirror images of each other ; these two states are degenerate in energy . further , more complicated , phase transitions follow as the crystal becomes more of a solid three - dimensional structure ( as shown in figure [ fig : strings](c ) where the ions form a helical structure ) . figure [ fig : birkl ] shows a similar process of evolution between different crystal structures for large numbers of ions in a radiofrequency ring trap . this linear to zigzag transition and the phase diagram of the system have been extensively studied theoretically ( see , for example , @xcite ) . unfortunately the spacing of the ions is not uniform throughout the string , but it is possible to make the approximation that close to the centre of the string the linear density of ions is roughly constant , and this allows useful calculations to be performed . recently , these classical calculations of ion string dynamics have been extended into the quantum regime ( see , for example , @xcite ) . the structural transition between the linear string and the zigzag structure can be used for a completely different sort of study , illustrated in figure [ fig : strings](d ) and ( e ) . if the confining potential of a linear string is _ rapidly _ changed across the transition point , different regions of the chain will evolve into a zigzag at the same time , but will not be able to communicate if the quench is fast enough . therefore there will be defects in the crystal where regions with opposite zigzag displacements meet . figure [ fig : strings](d ) shows a localised defect where two such regions meet ( referred to as an _ odd defect _ ) , and figure [ fig : strings](e ) shows an _ extended defect _ where the orientation of the zigzag changes slowly . the formation of these defects is closely related to the formation of defects in the early universe through the process of spontaneous symmetry breaking ( the so - called kibble - zurek mechanism ; see @xcite ) . the density of defects created depends on the speed of the quench from the linear to zigzag configurations . recent experimental studies @xcite have shown that this system reproduces the features expected theoretically for the kibble - zurek mechanism applied to ion chains @xcite and have confirmed the calculated scaling with quench time . the simplest possible example of this system is a string of three ions , and this is amenable to calculations @xcite . in a potential corresponding to an axial frequency @xmath24 and with a transverse frequency @xmath64 , it can be shown that the critical point , where the crystal kinks , is given by @xmath65 = 2.4 @xcite . if one considers the potential energy of the whole system , it can be seen that at low values of @xmath24 there is a potential well corresponding to transverse vibrations of the string in a _ kink mode _ ( where the central ion moves in the opposite direction to the other two ions ) . this is the so - called _ soft mode _ and is the lowest frequency of vibration of the system . as the critical point is approached , this vibration frequency drops and the potential becomes flatter . at the critical point , it is no longer quadratic at the centre , but quartic . beyond this point , the potential has a double well and this means that there are two degenerate configurations , which are mirror images of each other and have the central ion displaced from the trap axis in the opposite direction to the other two ions . this offers the intriguing possibility to observe quantum mechanical tunnelling between these two degenerate configurations of the three - ion system . however , this is technically very difficult as the spacing of the double well potential has to be comparable to the spread of the ionic wavefunction in order for the tunnelling rate to be observable @xcite . ions in a coulomb crystal constitute a system with unusually attractive properties for many experiments : in particular , they are nearly stationary , they are well isolated from each other and from the environment , and their internal electronic states can be manipulated with a high degree of control . this makes them ideal for use in experiments in the area of cavity quantum electrodynamics ( cqed ) , which deals with the interaction between atoms and light in an optical cavity , at the level of single of single photons . the first experiment in this area , carried out by drewsen s group at aarhus , demonstrated that by incorporating a high - finesse optical cavity into an ion trap structure , ions could be located precisely along the axis of the cavity so they interact strongly with the cavity light @xcite . since the ions act cooperatively , it was possible to reach the collective strong coupling regime , where the coherent exchange of energy between the ions and the optical cavity was faster than the decay of light in the cavity . for this , a crystal containing at least 500 ions was required . at any time , the cavity contains at most one photon . figure [ fig : cavity ] shows the change in the cavity reflectivity around a cavity resonance with and without the presence of ions . this plot is a clear demonstration that the strong coupling regime has been reached . this system has a number of applications apart from a demonstration of the physics of cqed , including a technique for studying the vibrational modes of large coulomb crystals in a sensitive and non - invasive manner @xcite . it has also been used to observe electromagnetically induced transparency ( eit ) , where a separate optical control field can switch the atomic absorption on and off in a controlled manner @xcite . a classical computer uses an array of two - state systems ( classical _ bits _ ) to carry out calculations using binary arithmetic . a quantum computer is a device that uses a system of simple two - state quantum mechanical systems ( e.g. spin1/2 particles ) as _ qubits _ for a similar purpose . there are several reasons why the construction of a quantum computer would be very desirable , and one of these is that some computational problems are effectively impossible for a classical computer ( e.g. the factorisation of very large numbers that are the product of two prime numbers ) , whereas they are believed to be tractable for a modest quantum computer using specially designed algorithms that take advantage of the way that quantum mechanical systems behave . just like classical computers work using classical gate operations ( such as and , or and not ) on classical bits , so quantum computers use quantum gates acting on qubits @xcite . quantum computing ( or more generally _ quantum information processing _ ) has now become a huge field of study , with various physical systems being investigated as realisations of qubits ( see for example @xcite ) , but we do not attempt to review it in detail here . many demonstrations of the essential operations necessary for building a quantum computer have been carried out using trapped ions @xcite . this field of application of trapped ions was initiated by cirac and zoller in 1995 @xcite . they proposed the realisation of a cnot gate , one of the basic elements of a quantum computer , using two ions in an rf trap . the internal electronic state of each ion forms one qubit , with the ion s ground state representing a logical 0 and a long - lived excited state ( an excited electronic , hyperfine or zeeman state ) representing the logical 1 . an essential component of their proposal was the use of the common vibrational mode of the two - ion system as a means of communicating between the ion qubits . it was therefore necessary to be able to cool the system to its lowest vibrational quantum state , for which optical sideband cooling would be essential . the first direct realisation of the cirac - zoller scheme was carried out by blatt s group at innsbruck @xcite . since then , ion coulomb crystals have been used in realisations of increasingly complex demonstrations of quantum computing algorithms . for a detailed review of this aspect of the use of icc , see the reviews by roos @xcite and ozeri @xcite . one of the requirements for a practical quantum computer is for it to be scalable to large numbers of qubits , which in this case means large numbers of ions . however , as the number of ions in a string rises , the number of possible modes of oscillation also increases . since each mode has a different frequency , it is difficult to separate out individual sidebands , which makes sideband cooling more difficult . this puts a practical limit on the number of qubits that can be manipulated in a single ion trap . up to 14 ions in a string have now successfully been put into an entangled state @xcite ; however , this number is approaching the limit of what is practically possible . for larger numbers of ions it will be necessary to move to a system where ions are held in separate traps or in separate trapping regions within the same trap ( see , for example , @xcite ) . they would then be moved between different trapping regions in order to carry out gate operations between specific pairs of ions . transport of individual ions inside a trap structure without significant heating of the motional state has been demonstrated in several experiments ( e.g. @xcite ) . _ quantum simulation _ refers to the use of one well - controlled quantum system to simulate the behaviour of another quantum system that may be harder to control or measure . of course classical computers may be used for the purpose of simulating quantum mechanical systems ( e.g. the behaviour of a set of spin1/2 particles ) , but as the size of the system increases , simulation using a classical computer becomes computationally difficult and for large enough systems it is impossible to carry out meaningful simulations using current technology . the difficulty comes from having to keep track of all the coherences between individual elements of a quantum system . it was feynman who pointed out in 1982 that the best way to simulate such a system would be to use another quantum mechanical system @xcite . it turns out that trapped ions are ideal for this type of application . this is because the particles are very stable and well isolated from the environment , so the lifetimes of delicate quantum states such as entangled states and coherent states can be very long . furthermore , the particles have very well - controlled interactions with each other and with external fields such as magnetic fields , lasers and microwaves , and these interactions can be controlled to match the physics of different systems . in addition , it is possible to measure the final internal ( electronic ) and external ( motional ) states of trapped ions to high accuracy using well - developed techniques , so detailed information on how the trapped - ion system is behaving can be obtained . recent reviews of the field of quantum simulation with trapped ions have been given by schneider _ _ @xcite and johanning _ et al . _ @xcite . coulomb crystals are very important in this area because in general if trapped ions are to be used in some sort of quantum simulator , they need to be in a stable configuration with fixed distances between them , so an icc is ideal for this . as an example , many experiments make use of a linear string of ions in a linear rf trap , where each ion qubit represents a spin1/2 particle in a magnetic field . in this case the ground state of the ion represents , for example , the spin down state @xmath66 and the internal excited state of the ion represents the spin up state @xmath67 . in order to simulate the well - known ising model of interacting spins , we need to be able to simulate a magnetic field acting on the spins , and to engineer interactions between them . the effect of the magnetic field is represented by irradiating the ions with coherent radiation ( a laser or radiofrequency field , depending on the type of qubits being employed ) . this causes each qubit to oscillate continuously between the @xmath66 and @xmath68 states ( an example of rabi flopping ) , in the same way that a spin exposed to a perpendicular magnetic field would oscillate between its spin - down and spin - up states as it precessed around the magnetic field direction . spin interaction for two ions can be engineered using a standing wave light field which provides a force on each ion that depends on which state it is in . in this way the force on one ion also leads to an effect on the other ion due to their mutual coulomb interaction : if they are both in the same state , they move together but if they are in different states , they move in opposite directions and their coulomb interaction energy changes . this represents a spin - spin interaction because the energy of each ion now depends on the state of the other ion . alternatively , an effective spin spin interaction can be engineered using a spatially varying magnetic field which causes ions to move slightly when their spin state changes , leading to an interaction between the ions mediated through the coulomb interaction @xcite . experiments based on these ideas have been used to study the properties of a quantum magnet represented by two trapped ions @xcite . by varying the parameters of the system , it was possible to observe both ferromagnetic and paramagnetic behaviour in this system . by extending the system to three ions in a chain ( the three - spin ising model ) , additional effects can be observed . for this , the interactions are engineered such that adjacent spins have a lower energy when they point in opposite directions ( i.e. anti - ferromagnet interactions ) . clearly not all three spins can arrange themselves in this way , so the system is said to exhibit _ spin frustration_. this is closely linked to quantum mechanical entanglement in the system , and the well - controlled nature of ion trap techniques allows all the dynamics and properties of the system to be studied in detail @xcite . in a penning trap it is possible to prepare a two - dimensional triangular array of ions , as discussed in section [ sec : penningcrystals ] . this is ideal for simulation of a two - dimensional array of spins in a magnetic field @xcite . in this experiment , performed on a planar crystal containing several hundred be@xmath33 ions ( see figure [ fig : hexagonal ] ) , the ion qubit comprises the two zeeman sub - levels of the ground state of the ion , separated by 124ghz in the magnetic field of the trap . an optical dipole force that is dependent on the qubit state is generated using a pair of laser beams close in frequency to the laser cooling transition at 313 nm , as described in section [ sec : penningcrystals ] . this creates an effective interaction @xmath69 between any two qubits @xmath70 and @xmath71 , which is mediated via the spatial properties of the transverse vibrational modes of the crystal @xcite . by tuning the difference frequency of the laser beams , different vibrational modes can be selected , and this changes the way the effective spin spin interaction depends on the separation of the qubits . in general , it is found that @xmath72 where @xmath73 is a constant and @xmath74 is the separation of qubits @xmath70 and @xmath71 . the spatial dependence of the interaction can be varied simply by changing the detuning of the laser beams , and this allows a range of interactions to be studied , from a constant force ( @xmath75 ) through a coulomb - like force ( @xmath76 ) to a dipole - dipole force ( @xmath77 ) . ions in coulomb crystals can be used for investigation of cross - sections of various types of reaction @xcite . for instance , if a buffer gas is introduced into the ion trap chamber , any reactions between the buffer gas and the ions will result in a change in mass to charge ratio of the trapped ions ; if they are in a coulomb crystal , they will then separate from the original species , as described in section [ sec : rfcrystals ] . this will result in a change in the shape of the crystal , which can be modelled through simulations . observation of the time - behaviour of the crystal shape , including the presence or absence of non - fluorescing species , will therefore give information on reaction cross sections . the same principle was also used in the measurement of multi photon ionisation cross - sections of mg and mg@xmath33 . a variation on this is to use a cloud of cold atoms rather than a buffer gas . in this case charge exchange cross sections or even the formation of cold molecular ions can be studied @xcite . for a review of this topic see @xcite . experiments have shown that it is possible to create and trap simple molecular ions ( e.g. h@xmath56 , hd@xmath33 ) using electron bombardment of low pressure gas in an rf trap . if the trap also contains laser - cooled be@xmath33 ions , the molecular ions are sympathetically cooled to temperatures in the mk range @xcite . one important potential application in this area is the use of coulomb crystals of sympathetically cooled molecular ions of hydrogen or deuterium for very precise measurement of ro - vibrational transition frequencies . the high precision and insensitivity to external perturbations could be exploited for frequency standards applications and also for the study of any possible variation of the fundamental constants with time @xcite . it is perhaps questionable whether it is strictly correct to describe two trapped ions as a coulomb crystal @xcite , but we include it here in order to demonstrate a particularly important application for future frequency standards and ultra - high precision spectroscopy . in order to create an optical frequency standard , ideally we require an ion that has an optical transition that is suitable for laser cooling , has a second transition ( the _ clock transition _ ) that has very narrow linewidth , and whose electronic state can be detected efficiently by the presence or absence of fluorescence . some atomic ions that are good candidates for frequency standards applications ( e.g. @xmath78al@xmath33 ) have suitable weak electronic transitions that are insensitive to external fields but do not have transitions that are suitable for laser cooling and for detection of the electronic state of the ion . quantum logic spectroscopy can couple such an ion to a second ion that can be laser cooled and whose electronic state can be detected with high efficiency ( e.g. @xmath79be@xmath33 ) . introduced first by the wineland group at nist @xcite , quantum logic spectroscopy uses the excitation of the vibrational mode of a two - ion coulomb crystal to transfer information about the electronic state of one ion ( the _ spectroscopy ion _ ) to the other ion ( the _ logic ion _ ) . in this way , laser cooling and state detection are carried out on the logic ion , but the ultra - high resolution spectroscopy is carried out on the spectroscopy ion . this allows a wider range of potential frequency standards candidates to be investigated . the technique has been used by the wineland group to demonstrate a frequency standard with a fractional frequency inaccuracy below 10@xmath80 @xcite . it is also being exploited in other groups for the construction of ultra - high stability frequency standards and for the measurement of possible variations of the fundamental constants . in this review we have described the physics underlying the formation of ion coulomb crystals , and we have discussed the methods used in various experiments to create , study and apply them in different areas of physics . these novel and unique structures have properties that make them attractive for demonstration of a variety of physical phenomena , some of which are familiar , but less accessible , in other situations . their applications are growing as scientists further develop the techniques required to create them , to manipulate and measure their properties and to determine the quantum state of the particles of which they are composed . ion coulomb crystals provide a very rich physical system for study , combining both classical physics in their formation and structure , and quantum physics in many of their applications . the comparison with conventional crystals reveals both similarities and fundamental differences . the range of applications , spanning across many areas of physics including thermodynamics , plasma physics , chemical reactions , fundamental physics , quantum optics and quantum information processing , demonstrates the power of ion traps and laser cooling . many of the studies that are performed with ion coulomb crystals simply could not be carried out using alternative techniques , because the degree of precise control and delicate measurement which is possible with trapped ions can not be achieved in other systems . more applications are expected in the future for these uniquely beautiful and versatile objects . this work was supported in part by the european commission strep picc ( fp7 2007 - 2013 grant number 249958 ) . we also gratefully acknowledge financial support towards networking activities from cost action mp 1001 - ion traps for tomorrows applications . richard thompson gained his dphil at the university of oxford in the area of atomic spectroscopy . following postdoctoral work in germany and the uk , he worked at the national physical laboratory on the first experiments with laser cooled trapped ions in the uk . in 1986 he moved to imperial college london where he is now a professor of experimental physics . his research has focussed on the dynamics of laser - cooled ions in penning traps , including the creation and manipulation of ion coulomb crystals in this system .

within a dense environment (atoms/ ) at ultracold temperatures ( ) , a single atom excited to a rydberg state acts as a reaction center for surrounding neutral atoms . at these temperatures almost all neutral atoms within the rydberg orbit are bound to the rydberg core and interact with the rydberg atom . we have studied the reaction rate and products for _ ns _rb rydberg states and we mainly observe a state change of the rydberg electron to a high orbital angular momentum _ l _ , with the released energy being converted into kinetic energy of the rydberg atom . unexpectedly , the measurements show a threshold behavior at for the inelastic collision time leading to increased lifetimes of the rydberg state independent of the densities investigated . even at very high densities ( ) , the lifetime of a rydberg atom exceeds at compared to at . in addition , a second observed reaction mechanism , namely rb molecule formation , was studied . both reaction products are equally probable for but the fraction of rb created drops to below for .

1605.04883

ultracold rydberg atoms are studied in increasingly dense environments as they give rise to collective phenomena in the strongly blockaded regime @xcite , foerster - type energy transfers @xcite , rydberg - dressed ensembles @xcite or to achieve larger optical depths for quantum optical applications @xcite . at increasingly higher densities it becomes more likely that ground state atoms reside within the orbit of the rydberg - electron and ultralong - range rydberg molecules @xcite can be created with exotic butterfly @xcite and trilobite @xcite shapes . the largest density achievable at ultracold temperatures are in quantum degenerate gases , where the backaction of a rydberg atom on the superfluid can be studied @xcite . the generation of phonons by immersing a single rydberg atom may even be used to image the wavefunction of the rydberg electron @xcite . for such advanced schemes , it is necessary that the rydberg atom stays in its original state sufficiently long to have a measurable impact on the surrounding quantum gas . in a previous study it was already shown that the lifetime of rydberg atoms inside a dense cloud of atoms is reduced @xcite but no quantitative study on the origins of the reduced lifetimes have been performed so far . prior to the present work , inelastic collisions between rydberg atoms and neutral particles have been investigated but in regimes of energies and densities that are completely different from those of the present work . chemi - ionization reactions creating rb@xmath8 were examined in the existing literature @xcite leading to the ionization of the rydberg electron as a consequence of the short - range interaction between the ionic core and the incoming neutral atom @xcite . inelastic collisions of highly excited rydberg atoms in a dense background gas at room temperature were observed in the late 1970 s @xcite and early 1980 s @xcite , and simultaneously the theoretical framework associated with such a process was developed by adopting a semiclassical description of the dynamics , in combination with a constant rydberg electron - neutral atom scattering length @xcite in applications of the fermi pseudopotential . this approach describes satisfactorily the overall behavior of the @xmath11-mixing cross section as a function of the principal quantum number @xmath12 , in a regime where the collision energy between the neutral and the rydberg atom , @xmath13 , exceeds the relevant energy scale of the rydberg transitions @xmath14 , i.e. , @xmath15 , thus justifying the semiclassical treatment . however , for a rydberg excitation ( @xmath16 ) in an ultracold ( @xmath17 ) and a dense background gas ( @xmath18@xmath19 ) , more than 1000 neutral atoms are within the rydberg orbit . this requires a new theoretical treatment for inelastic rydberg - neutral collisions to be developed . the two observed reaction channels for a single rydberg atom in an ultracold and dense environment . the red sphere depicts the rydberg ionic core and the rydberg electron is shown in blue , whereas the neutral atoms are shown in green . ( i ) the rydberg atom changes its angular momentum and both collision partners share the released energy as kinetic energy leading to an l - mixing reaction channel . ( ii ) a deeply bound rb@xmath8 molecular ion is formed through chemi - ionization by an associative ionization reaction.,scaledwidth=45.0% ] ( 0,0 ) ( -203,100)i ( -35,100)ii ( -104,90)rb ( -175,90)rb*(@xmath20 ) ( -215,10)rb*(@xmath21 ) ( -58,10)rb@xmath8 in this work , we study the role of the ionic rydberg core , the rydberg electron and the neutral ground - state atom collisions across a large range of principal rydberg quantum numbers in limiting the collisional lifetime of rb rydberg atoms in a dense , ultracold atom cloud . we identify two reaction products , shown in fig . [ fig : reaction ] , which are explained using a new theoretical framework . a quantum mechanical treatment of the reaction pathways is implemented along with a semiclassical description of the interactions at short internuclear distances , with an explicit treatment of the neutral - rydberg energy mediated by the rydberg electron - neutral collisions . finally , the reaction dynamics and the branching ratio of these ultracold chemical reactions are analyzed . ( a ) spectrum of a 90__s _ _ rydberg atom in a bec . one of the excitation lasers is focused to the center of the bec with a focal waist of . therefore , primarily the peak density @xmath22@xmath19 of the bec is addressed with a detuning of approximately . each data point is taken with 10 becs , and each bec was probed 50 times . during the subsequent experiments within one bec , the peak density drops from @xmath23@xmath19 to @xmath24@xmath19 and approximately @xmath25 atoms of the initial @xmath26 of the bec are in the thermal cloud at the end of the experiment . the thermal atoms surrounding the bec contribute to the signal at very small red detuning . the laser detuning is related to the local density where the rydberg atom is excited @xcite . [ fig : spectrum90s](b ) nearest neighbor distribution of the distance between the rydberg core ( @xmath27 ) and a neutral atom for two different densities of the bec . ( c ) pair potential of a 90__s__+5__s _ _ state including the rydberg electron - neutral scattering interaction and the rydberg core - neutral polarization potential @xmath28 . the state crossing visible at is the butterfly state @xcite crossing the @xmath29 state . the highest energy available in the system comes from the photon recoil , @xmath30 and therefore most of the particles inside the rydberg orbit are classically trapped.,title="fig : " ] + ( a ) spectrum of a 90__s _ _ rydberg atom in a bec . one of the excitation lasers is focused to the center of the bec with a focal waist of . therefore , primarily the peak density @xmath22@xmath19 of the bec is addressed with a detuning of approximately . each data point is taken with 10 becs , and each bec was probed 50 times . during the subsequent experiments within one bec , the peak density drops from @xmath23@xmath19 to @xmath24@xmath19 and approximately @xmath25 atoms of the initial @xmath26 of the bec are in the thermal cloud at the end of the experiment . the thermal atoms surrounding the bec contribute to the signal at very small red detuning . the laser detuning is related to the local density where the rydberg atom is excited @xcite . [ fig : spectrum90s](b ) nearest neighbor distribution of the distance between the rydberg core ( @xmath27 ) and a neutral atom for two different densities of the bec . ( c ) pair potential of a 90__s__+5__s _ _ state including the rydberg electron - neutral scattering interaction and the rydberg core - neutral polarization potential @xmath28 . the state crossing visible at is the butterfly state @xcite crossing the @xmath29 state . the highest energy available in the system comes from the photon recoil , @xmath30 and therefore most of the particles inside the rydberg orbit are classically trapped.,title="fig : " ] the reaction dynamics are studied in a nearly pure bec of approximately @xmath31 @xmath32 atoms in the magnetically trapped spin polarized ground - state @xmath33 produced in a quic trap @xcite . the trapping frequencies are @xmath34 in the radial and @xmath35 in the axial direction corresponding to thomas - fermi radii @xcite of by . the atom number and trap frequencies give rise to a peak density of @xmath23atoms/@xmath1 in the bec . the rydberg excitation is created by a two - photon excitation scheme , which couples the ground state to the @xmath36 state for principal quantum numbers _ n _ from 40 to 149 , off - resonant via the intermediate state 6__p__@xmath37 . both excitation lasers are pulsed simultaneously with a repetition rate of , enabling up to 400 experiments within a single bec at a fixed laser frequency . the excitation probability per shot is kept well below one . the possible single rydberg atom in the bec is subsequently ionized by an electric field pulse , with a rise time and a variable delay time . the atom number and density of the bec drops during the repeated measurements but the extracted values were tested to be independent of this atom loss . further details of the experiment are described in @xcite . under these conditions , a rydberg atom is excited with , on average , one ( @xmath38 ) or up to tens of thousands ( @xmath39 ) of neutral atom perturbers inside the rydberg orbit , depending on the principal quantum number @xmath12 and the density . the neutral atoms inside the rydberg orbit cause a frequency shift for the excitation of a rydberg atom and can be used to modify the resonant density region in the bec @xcite : in a first approximation , the resonance frequency is shifted in relation to the local density , @xmath40 , in which the rydberg atom is excited , due to the fermi pseudopotential as @xmath41 with the reduced planck constant as @xmath42 , the electron mass as @xmath43 , and the @xmath44-wave scattering length as @xmath45 . this leads for triplet scattering in rubidium ( @xmath46 @xcite ) to a spectral line shift of for @xmath47atoms/@xmath48 independent of @xmath12 . the situation is depicted also in fig . [ fig : spectrum90s ] , showing a 90__s _ _ spectrum of a single rydberg atom in a bec for which a laser detuning of approximately addresses the peak density @xmath22@xmath19 . the temperature of the ultracold atoms is approximately . the largest energy available in the system comes from the recoil energy resulting from the two - photon excitation , and is @xmath49 ( @xmath50 ) , where @xmath51 is the planck constant and @xmath52 the boltzmann constant . at these low temperatures , many neutral atoms are trapped inside the rydberg orbit , as shown in fig . , by the rydberg electron interaction potential and the polarization potential from the ionic core of the rydberg atom . the atoms inside the rydberg orbit approach the rydberg core after the rydberg atom is excited , leading to inelastic and reactive processes , namely @xmath11-mixing collisions and chemi - ionization as depicted in fig . [ fig : reaction ] . since the neutral atoms start in an interacting frame , the usual concepts of scattering theory are not well - defined in this case as there are no asymptotic initial states . the figure of merit is therefore the collision time observed instead of a collision rate . this time is much shorter than the radiative lifetimes of rydberg atoms , making a dense atom cloud an ideal test bed for studying ultracold rydberg - neutral atom collisions . ( a ) released energy during the @xmath11-changing collision . the black dots indicate the most likely released energy during a collision , whereas the bars indicate the full width at half maximum of the fitted probability distribution of the released energy . the energies observed correspond to a state change to the next lower hydrogenic manifold ( @xmath53 ) . for @xmath54 the most likely kinetic energy gained is not shown anymore because the released energy is too low to extract a well - defined maximum out of the energy distribution , which is depicted by the arrow . the colors of the released energies are according to the level schema shown on the right , in which hy denotes the hydrogenic manifold . ( b ) state selective ionization analysis of the detected ions for a @xmath55 state . with an immediate ( blue curve ) linear ionization field ramp going to within applied after the excitation pulse , almost all atoms arrive between 1 and corresponding to the initially excited @xmath55 state . with a delayed ionization , the rydberg atom undergoes an inelastic collision ( red : delay of ) and the required ionization voltage changes by a factor of 2 to 4 corresponding to a high @xmath11-state @xcite.,title="fig : " ] + ( a ) released energy during the @xmath11-changing collision . the black dots indicate the most likely released energy during a collision , whereas the bars indicate the full width at half maximum of the fitted probability distribution of the released energy . the energies observed correspond to a state change to the next lower hydrogenic manifold ( @xmath53 ) . for @xmath54 the most likely kinetic energy gained is not shown anymore because the released energy is too low to extract a well - defined maximum out of the energy distribution , which is depicted by the arrow . the colors of the released energies are according to the level schema shown on the right , in which hy denotes the hydrogenic manifold . ( b ) state selective ionization analysis of the detected ions for a @xmath55 state . with an immediate ( blue curve ) linear ionization field ramp going to within applied after the excitation pulse , almost all atoms arrive between 1 and corresponding to the initially excited @xmath55 state . with a delayed ionization , the rydberg atom undergoes an inelastic collision ( red : delay of ) and the required ionization voltage changes by a factor of 2 to 4 corresponding to a high @xmath11-state @xcite.,title="fig : " ] the state - changing collision of the studied @xmath3rb rydberg atom with the neutral atom reagent is an exoergic process , whereby a state - change occurs from the excited _ ns _ state to the lower - lying hydrogenic manifold . in the case of rubidium , with a quantum defect of the _ s _ state ( @xmath56 @xcite ) , this is the @xmath57 hydrogenic manifold . we can measure the change in kinetic energy of the rydberg atoms undergoing this collision process by monitoring the arrival time of the ionized rydberg atoms on the microchannel plate detector . for this measurement the rydberg atoms are ionized after a delay time , e.g. @xmath58 , which is significantly longer than the collisional lifetime @xmath59 of the rydberg states . after the @xmath11-changing collision , the rydberg atom and the neutral atom fly apart in opposite directions , due to momentum conservation and their equal mass , with respect to the excitation location . once the ionization pulse is applied at time @xmath60 , the rydberg atom accelerates towards the microchannel plate detector . the varying spatial positions of the rydberg atom at the time of ionization are detected as a spread in ion arrival time . using ion trajectory simulations , we calculate the expected ion arrival times based on the distance _ d _ from the center where the rydberg atoms are excited . this mapping of initial position to ion arrival time determines the average kinetic energy of the detected rydberg atoms . the total energy @xmath61 released in this reaction is twice the energy of the detected rydberg atom @xmath62 , because the energy released during the collision must be shared between both collision partners : @xmath63 further details for this method are given in appendix [ a : energyrelease ] . fig . plots the released energy , gained during the inelastic collision , versus the principal quantum number . the released energy @xmath61 corresponds for low principal quantum numbers well to the change in potential energy from the excited @xmath64 rydberg state to the hydrogenic manifold below , which is the @xmath57 manifold because of the quantum defect of rb with @xmath56 . the energy @xmath61 can be calculated according to the rydberg formula @xcite @xmath65 with the effective principal quantum number @xmath66 taking into account the quantum defect , and the modified rydberg constant @xmath67 for @xmath3rb . due to the decreasing energy level spacing , the released energy @xmath61 decreases with higher principal quantum numbers . furthermore , the change in kinetic energy as confirmed by the measurement shown in fig . strongly indicates that we do not couple for any investigated @xmath12 to the ( @xmath68 ) hydrogenic manifold or any levels with larger energy separations . the threshold ionization electric field of the rydberg atoms can be exploited to examine the angular momentum change of the rydberg atom . at a slew rate of the ionization electric field of @xmath69 , rydberg atoms with @xmath70 ionize diabatically at up to four times the adiabatic ( classical ) ionization threshold field ( @xmath71 ) @xcite . therefore , the higher electric field required to ionize the rydberg atom , which is visible in fig . , shows that the final state must be a high @xmath11-state . based on the measured released energy and the ionization threshold change combined , the most likely populated final state after the inelastic collision of this reaction channel is the @xmath57 hydrogenic manifold below the excited @xmath64 state . the energy release restricts the possible state change to one effective principal quantum number below the initial state . for low @xmath12 , the most probable released energy is in accordance to a state change to the next lower lying manifold . for high principal quantum numbers , for which the energy resolution is not precise enough to determine the target state , the state selective ionization measurements show that a high @xmath11-state is populated . ( a ) potential energy landscape of the @xmath72 state including the interaction between the rydberg electron and a neutral atom including the polarization potential @xmath28 of the rydberg ionic core . at @xmath73 , the @xmath72 and the neighboring butterfly state couples and causes an avoided crossing . from a classical viewpoint , neutral atoms initially at shorter distances than the crossing will follow the @xmath74 potential until the ion neutral interaction takes over ( violet ) . at larger distances , the probability to adiabatically follow the butterfly state at the crossing of the @xmath64 potential with the butterfly state is almost one , as the neutral atom approaches the rydberg ionic core . in both cases a short range coupling at @xmath75 can lead to a state change to the trilobite state ( green ) , which for large internuclear separations turns into a high @xmath11-state of the next lower lying hydrogen manifold ( hy ) . ( b ) theoretical adiabatic landau - zener probabilities at the butterfly crossing with the _ d _ and _ p _ states , respectively , as functions of the principal quantum number @xmath12 . the error bars have been calculated as it is explained in the appendix [ a : lz].,title="fig : " ] + ( a ) potential energy landscape of the @xmath72 state including the interaction between the rydberg electron and a neutral atom including the polarization potential @xmath28 of the rydberg ionic core . at @xmath73 , the @xmath72 and the neighboring butterfly state couples and causes an avoided crossing . from a classical viewpoint , neutral atoms initially at shorter distances than the crossing will follow the @xmath74 potential until the ion neutral interaction takes over ( violet ) . at larger distances , the probability to adiabatically follow the butterfly state at the crossing of the @xmath64 potential with the butterfly state is almost one , as the neutral atom approaches the rydberg ionic core . in both cases a short range coupling at @xmath75 can lead to a state change to the trilobite state ( green ) , which for large internuclear separations turns into a high @xmath11-state of the next lower lying hydrogen manifold ( hy ) . ( b ) theoretical adiabatic landau - zener probabilities at the butterfly crossing with the _ d _ and _ p _ states , respectively , as functions of the principal quantum number @xmath12 . the error bars have been calculated as it is explained in the appendix [ a : lz].,title="fig : " ] theoretical results for the different pathways . probability to find the rydberg atom in different states after an ion - neutral collision : @xmath74 state ( blue ) , @xmath76 state ( red ) , @xmath77 state ( orange ) and in the @xmath57 trilobite state ( green ) . the rb@xmath8 reaction product will be discussed in the section [ s : rb2 ] . ] the rydberg - neutral collision is determined by the molecular potential energy landscape . in particular for distances below @xmath78 , it is known that the electron - neutral @xmath79-wave shape resonance at significantly affects this landscape @xcite . we thus calculate the molecular potentials by means of green s function approach @xcite including energy - dependent @xmath44-wave and @xmath79-wave triplet scattering phase - shifts for e@xmath80-rb collisions . as an example , the potential energy curve ( pec ) associated with the 90s state is shown in panel ( a ) of fig . [ fig : pec ] . to study the collision dynamics , we follow the trajectory of an initial pair state @xmath81 when the two collision partners approach each other . at each avoided crossing in the pecs , we determine the probability for the initial state to follow a specific path by applying the landau - zener formula @xcite including the relative velocity of the two partners at each crossing . to begin with , the probabilities of ending up in either the @xmath82 or the @xmath83 state , is denoted as @xmath84 and @xmath85 , respectively are shown in panel ( b ) of fig . [ fig : pec ] . landau - zener probabilities strongly depend on the kinetic energy at the crossing point , which in the present approach is dominated by the energy difference between the initial _ ns _ state and the final state reached via the butterfly state pathway . at our very low temperatures , the rydberg level splitting by far dominates over the initial kinetic energy of the atoms , which was not the case for experiments at room temperature @xcite . in panel ( b ) of fig . [ fig : pec ] , aside from the noticeable variance in principal quantum number , one observes an overall trend : at low principal quantum numbers @xmath84 and @xmath85 are systematically larger in comparison with @xmath86 . this translates into a larger population of the butterfly state at high principal quantum numbers . the varying probability in @xmath85 and @xmath84 is related with the derivative of the @xmath83 and @xmath82 rydberg wave function , respectively at the @xmath79-wave crossing point . a more detailed explanation of the calculation of the landau - zener probabilities is given in appendix [ a : lz ] . for rydberg - neutral distance @xmath87 the butterfly and @xmath83 states are energetically close to the trilobite state coming from the @xmath57 hydrogenic manifold , as shown in fig . , leading to a coupling between these states . the non - adiabatic couplings of the involved electronic states have been quantified by means of the @xmath88-matrix @xmath89 @xcite , where the adiabatic states @xmath90 and @xmath91 have been obtained by diagonalizing the hamiltonian by means of a truncation of the hilbert space @xcite instead of the green s function method . as it is shown in appendix [ a : pmatrix ] , the @xmath82 state has a negligible coupling with respect to the trilobite state , and hence every atom ending up in the @xmath82 state after passing through the butterfly region will reach the short - range region dominated by chemical forces . however , atoms in the @xmath83 state or butterfly state after the butterfly region experience a considerable coupling to the trilobite state correlated with the @xmath57 hydrogenic manifold . even though atoms will have a chance to reach the short - range interaction region in the trilobite state , rydberg atoms in high angular momentum states , such as those forming trilobite states , show an extremely narrow autoionization resonance width @xcite . this translates into a very small probability of chemi - ionization reactions . figure [ coupling ] displays the probability to find the rydberg atom in a given state after a collision with the perturber . these probabilities have been calculated by analyzing all the available reaction pathways as well as accounting for the coupling among them . at short rydberg - neutral distances the ion - neutral polarizability attraction dominates the electron - neutral interaction . this leads to associative ionization , an autoionization process , to be elucidated in the next section . however , the characterization of all the relevant couplings associated with the adiabatic rydberg - neutral states allows us to calculate the probability to find the rydberg atom in a particular state after a short - range collision , i.e. , when the neutral leaves the butterfly region on its way back to the long - range region . due to the fast timescales of the short - range physics we can only resolve the final states in the long - range region , in our experimental setup . the results for the relevant pathway probabilities are shown in fig . [ coupling ] , where the associative ionization probability of reaction given by eq . ( [ eq : ai ] ) has been included . the associative ionization probability leads to the second reaction channel , namely the rb@xmath8 formation . even though the model predicts the right fraction for the molecule formation , which will be discussed in the next section , there is a discrepancy for the @xmath11-changing collision at @xmath92 . the theory predicts a probability of to end up in the @xmath76 state whereas the measurement ( fig . ) shows a smaller fraction ( @xmath93 ) for the atoms which ionize at the lower electric field ( @xmath94 ) . for the theoretical results shown in fig . [ coupling ] , the contribution of neutral atoms at distances closer to the rydberg atom than the distance of the butterfly state crossing the @xmath64 state has been neglected . certainly , these atoms will increase the probability to have associative ionization reactions since @xmath74 states at small internuclear distances are easily ionized . moreover , @xmath74 states do not show any coupling with trilobite states . potential energy curves relevant to the formation of rb@xmath8 for rydberg states from @xmath95 to the ionization threshold . for each rydberg state there are two pec attached , namely @xmath96 showing a primarily repulsive nature whereas the other labeled @xmath97 displays a far more attractive nature that supports many bound states . the dark blue line represents the potential curve @xmath96 which correlates with the @xmath98 state , and the crossing point of this potential with the @xmath99 pec correlates with the onset where autoionization can begin into the channel rb@xmath100 ( red dotted line ) . this crossing point @xmath101 represents outer boundary of the associative ionization region . the inner boundary of this autoionization region ends at the inner turning point @xmath102 , where the collision energy ( solid orange line ) is equal to the potential . ] a highly excited atom colliding with a ground state atom can experience a chemi - ionization process @xcite , which is the ionization of the excited atom as a consequence of the short - range interaction between the ionic rydberg core of the excited atom and the neutral atom . this leads to associative ionization ( ai ) , which in the case at hand is identified as @xmath103 the previous sections demonstrate that a neutral atom interacting with a rydberg atom transits through a multi - pathway rydberg energy landscape ( see panel ( a ) of fig . [ fig : pec ] ) . at very short internuclear distances @xmath104 , however , the electronic cloud of the rydberg ionic core starts to overlap with the electronic cloud of the valence electron of the neutral atom , leading to a stronger interaction between the neutral ground state atom and the ionic core . this short - range interaction translates into two pecs at short range , which correlate at long range with the rb@xmath105+rb atomic asymptote . the ground state pec of the molecular ion is a @xmath106 electronic state , labeled as @xmath107 for simplicity , which supports a large number of bound states , i.e. , rb@xmath108 , whereas the second pec is a mostly repulsive electronic state @xmath109 ( weakly attractive between and ) , labeled as @xmath110 . let us assume that our initial rydberg state , an @xmath64 state , follows the butterfly state ( see fig . ) until the next state crossing and adiabatically changes into the @xmath111 state , which is the state it enters the short range region as depicted in fig . [ fig : rb2pot ] . attached to this rydberg state are the @xmath112 and @xmath110 electronic states that are degenerate in energy for large internuclear distances . the probability to transfer into either of the two available ( @xmath113 and @xmath114 ) states is approximately the same ( @xmath115 ) as @xmath116 decreases through the region and exchange interaction becomes important . further , we notice that the @xmath110 state attached to the @xmath111 state shows a crossing point at @xmath101 with the @xmath112 potential that correlates with the rb@xmath105 + rb asymptote . thus , for @xmath117 , @xmath110 ( thick blue curve in fig . [ fig : rb2pot ] ) can be interpreted as the lower boundary for an ionization continuum of potential energy curves @xmath112 that correlate with the state rb@xmath108+e@xmath80 . on the other hand , @xmath101 also denotes the internuclear distance where the energy splitting between @xmath118 and @xmath119 is equal to the rydberg binding energy in the ungerade bound channel , i.e. , @xmath120 , and hence for @xmath117 it is possible for the rydberg electron to autoionize , which then forms rb@xmath108 in the state @xmath118 , leading to the product rovibrational state of the ai reaction . the reaction probability for this reaction is given by @xcite @xmath121,\ ] ] where @xmath122 is the radial velocity for the rydberg - neutral collision for an @xmath44-wave atom - atom collisions . @xmath13 denotes the collision kinetic energy and @xmath123 is the reduced mass for the colliding partners . @xmath102 stands for the inner classical turning point associated with the collision energy @xmath13 and @xmath124 represents the width of the autoionization resonances which are in general proportional to @xmath125 . the probability @xmath126 to create deeply bound rb@xmath8 molecules is shown , calculated from the measured fraction of ionization events that produce rb@xmath8 @xmath127 . the delay time of the ionization in this experiment is chosen such that almost all atoms experienced an inelastic collision . the blue error bars show the experimental uncertainty due to the finite electric field within the science chamber that drags away the molecular ions created . as a comparison , two data sets ( @xmath74 @xcite and @xmath128 @xcite ) from measurements of a thermal cloud ( @xmath129 ) with a peak density of only @xmath130@xmath19 are added . the black dots represent the theoretical values for the probability of reaction by means of eq . ( [ eq : ai ] ) and using the corresponding landau - zener probabilities . the error bars have been calculated from the errors shown in panel ( b ) of fig . [ fig : pec ] ] experimentally , the state selective ionization measurement , plotted in panel ( b ) of fig . [ fig : nlchange ] , shows that most of the excited rydberg atoms undergo an inelastic collision if the delay time before the ionization is long enough . therefore , the branching ratio between the two reaction channels can be calculated by measuring the fraction of rb@xmath8 molecules following an inelastic collision compared to the total signal arriving at the detector . figure [ fig : rb2fraction ] shows the resulting reaction probability associated with the formation of rb@xmath8 . the same figure also shows our theoretical predictions based on eq . ( [ eq : ai ] ) , including the landau - zener probabilities shown in panel ( b ) of fig . [ fig : pec ] . this calculation utilizes the appropriate collision energy , which turns out to be dominated by the energy difference between adjacent rydberg states through the butterfly state , as it was indicated in the preceding section . for the purposes of this estimate , the width function has been approximated by the formula @xmath131 in eq . ( [ eq : ai ] ) , which is of the same order as the theoretical limit of @xmath132 predicted by two - channel quantum defect theory with one closed rydberg channel of autoionizing resonances , and this evidently yields overall agreement with our experimental data . the branching ratio of the reaction products , as shown in fig . [ fig : rb2fraction ] , strongly depends on the principal quantum number of the rydberg atom , but not on the excitation laser detuning , which is related to the atom density , in which the rydberg atom is excited . for @xmath133 , the rb@xmath8 formation is as likely as an @xmath11-changing collision . the fraction of deeply bound rb@xmath8 molecules drops for higher principal quantum numbers to below above @xmath7 for the @xmath74 state . the theoretical results following eq . ( [ eq : ai ] ) show the same trend as the experimental data in this range of principal quantum numbers , however the theoretical predictions are systematically below the observed experimental data at high principal quantum numbers . the reason for this discrepancy might conceivably be associated with our relatively crude utilization of the landau - zener approximation for the avoided crossings involving highly excited rydberg states , since in reality the highly oscillating nature of the rydberg state wave functions can result in overlapping avoided crossings . the branching ratio of the low density measurements was extracted out of spectra measured in a thermal cloud at a temperature of @xmath129 @xcite . the excitation laser pulse duration was and the ionization was applied directly after the excitation . the data points for the @xmath74 state agree with the high density bec data , which means that most of the rydberg atoms went through the chemical reaction during the long excitation time , under the assumption that the branching ratio is independent of the density . finally , the molecular ions formed are deeply bound , which was tested by comparing the number of molecules detected for the @xmath134 state with an electric field ionization of and , which are respectively 7 times and 32 times the classical ionization threshold of the @xmath134 state . the number of molecules measured was not lower for higher electric fields , which means that the molecular bond was not disturbed with application of high ionization fields . the same behavior was observed for the @xmath72 state ( classical ionization field of ) with an applied ionization field of and . in addition , it was tested that the signal with a field ionization well below the ionization threshold stays the same , which means that the molecules have autoionized during the process . it was found that the effect of the photoionization from the laser is negligible . measurements in which the laser was kept on after the excitation pulse showed that the number of molecules detected was only approximately higher compared to the normal case in which both lasers are turned off simultaneously . the present experimental approach allows us to measure the amount of products formed as a result of ultracold chemical reactions , as it has been shown . however , the key advantage of the density regime utilized in this study , is the possibility to observe chemical reaction in the time domain , allowing a very near characterization of the two reactions at hand : @xmath11-mixing collisions and chemi - ionization . the time scale on which this state - changing collision occurs could in principle be examined with the state selective ionization measurement shown in fig . by evaluating the time evolution of the measured signal in the initial state and final state . to achieve a better time resolution , however , a two - step ionization was performed with rise times for the electric field of . the electric field of the first ionization pulse was chosen such that it ionizes mainly the initial state , but not the final state ( e.g. for @xmath55 in fig . ) . a subsequent second ionization pulse , at least four times higher than the classical ionization field , was applied to ionize the remaining rydberg atoms . all detected atoms of the second ionization pulse must have undergone an @xmath11-changing collision before the first ionization pulse . this fraction is extracted from the data and its evolution with delay time is used to determine the lifetime of the initial rydberg state , shown in fig . [ fig : taus ] , under the influence of the inelastic collisions . an example of how the collision time is extracted is shown in appendix [ a : tau ] . extracted inelastic collision times for the @xmath11-changing collision ( top ) and the formation of rb@xmath8 ( bottom ) . the inelastic collision time @xmath135 of the reactions was extracted out of the measured signal of the reaction products @xmath136 by fitting @xmath137 with the error bars indicating the confidence interval of the fit taking into account multiple measurements at each data point . unexpectedly , at around a principal quantum number of @xmath138 , a threshold is reached leading to significantly higher lifetimes of the rydberg state at higher principal quantum numbers independent of laser detuning and thus the density . the rb@xmath8 formation time could not be measured for high @xmath12 because the total number of molecules measured was too low at these principal quantum numbers . ] quite unexpectedly we observe a threshold behavior in the collisional lifetime in the region of @xmath138 . the overall shape of the curves does not depend on the laser detuning and therefore the density , which only causes an overall offset factor to the collision time . the unexpected long lifetime for @xmath139 is therefore solely a state - dependent effect . the developed theoretical framework to explain the two observed reactions is applicable once the neutral atom crosses the butterfly region . the neutral atom is accelerated due to the butterfly state , leading to collisional times @xmath140 for both reactions . therefore , we conclude that the observed collisional times are related with the physics beyond the butterfly region , which is fully dominated by the rydberg electron colliding with neutral atoms . on the other hand , taking into account the large number of perturbers within the rydberg orbit under the present experimental conditions , a molecular dynamic simulation assuming a microcanonical ensemble and including only the ion - neutral interaction was performed , as shown in fig . [ fig : tausim ] . to simplify the potential , only the attractive force due to the polarizability of the neutral atoms is taken into account in the simulation : the rydberg core can be treated as an ion for the interaction with the neutral atoms within the rydberg orbit , because the rydberg electron does not shield the charge of the core within this region . the rydberg core and the polarizability @xmath141 @xcite of the neutral rubidium atoms then lead to the polarization potential @xmath142 at an internuclear distance @xmath116 , with the vacuum permittivity @xmath143 and the elementary charge @xmath144 . comparison between the inelastic collision time ( see also fig . [ fig : taus ] ) and the collision time of a molecular dynamics simulation of an ion and a neutral atom with the polarization potential @xmath28 only . the density for the simulation was chosen such that the density is according to fermi s density shift ( eq . [ eq : fermishift ] ) with @xmath145@xmath19 ) . the classical simulation with point - like particles reproduces the order of magnitude for the inelastic collision time at low principal quantum numbers . on the one hand , the full quantum treatment including the electron must accelerate this process for lower @xmath12 , which is very likely due to the butterfly state crossing the @xmath64 state . on the other hand , it must slow down the inelastic collision above the threshold region @xmath146 so that collisional lifetimes above are possible even at the peak density of the bec . ] the simulated collision time of this system is too slow to explain the observed reaction times for low quantum numbers as shown in fig . [ fig : tausim ] . in addition , this kind of interaction does not depend on the principal quantum number and therefore can not reproduce the observed threshold behavior . this is a strong indication that the rydberg electron plays a key role in the collision dynamics , as one expected based on investigations of the reaction pathways , which was also found to be the case increasing the scattering cross section at similar temperatures @xcite . the rydberg electron interacting with the neutral atom can accelerate the collision process of the inelastic collision especially due to the anti - crossing of the butterfly state with the excited @xmath64 state , as shown in panel ( a ) of fig . [ fig : pec ] , which is the case for lower quantum numbers @xmath147 with collision times of a few . for higher principal quantum numbers , the collision time exceeds for @xmath148 . here we suspect that a quantum mechanical effect sets in which prevents the neutral atoms for colliding inelastically with the rydberg atom . a possibility for this could be quantum reflection at the steep slope of the butterfly crossing the initial @xmath64 state as depicted in fig . , which is already known as a mechanism to form bound states in rubidium @xcite and depends on the principal quantum number . on the other hand , very recently a similar scenario has been explored finding deviations from the expected additive nature of the many - body landscape for the interaction between rydberg atoms and neutral atoms @xcite , which may be also relevant to understand the measured collisional times . in fig . [ fig : tausim ] the collision time of the two different reaction channels , shown in fig . [ fig : taus ] , are compared . the rb@xmath8 formation time can be up to two times the @xmath11-changing collision time . it is unlikely that the rb@xmath8 molecules are created as a subsequent collision after an @xmath11-changing collision , because for the kinetic energy gained during the collision , the rydberg atom leaves the dense bec region within a microsecond and the scattering cross section is reduced , due to the higher velocity ( e.g. for @xmath134 ) . this work explores the fundamental limit of the lifetime of a rydberg atom excited in a ultracold , dense gas , which is limited by inelastic collisions of the rydberg atom with a neutral ground state atom . two reaction products are observed within the system . either deeply bound rb@xmath8 molecules are created or the rydberg atoms change their angular momentum . these two reaction product states have been explained from a new theoretical quantum mechanical framework based on the analysis of the reaction pathways including the role of the rydberg electron . both channels are equally probable for rydberg @xmath74 states at @xmath9 . for higher principal quantum numbers , the probability of molecular rb@xmath8 production diminishes ( for @xmath10 ) . the rydberg electron plays a key role for the interaction , especially because of the @xmath79-wave scattering resonance leading to the butterfly state in rubidium . it accelerates the inelastic collision compared to the simulated collision time of the ion - neutral system for low principal quantum numbers ( @xmath147 ) for which the collision time is on the order of a few microseconds at the bec densities ( @xmath47@xmath19 ) . for high principal quantum numbers ( @xmath149 ) , the inelastic collision time surprisingly exceeds even in the very dense environment due to the observed threshold behavior for the inelastic collision time between @xmath138 . hence , rydberg states at principal quantum numbers above 110 are better suited , e.g. , for applications such as quantum optics , for which a high density can benefit the experiment . similarly , our findings have deep implications for rydberg - rydberg interaction - based quantum information processing @xcite , in particular by identifying collisions as decoherence sources , which strongly depend on the principal quantum number chosen . an implication of these observed long - lived states for high principal quantum numbers in dense media is the opportunity to image the wave function of a rydberg atom in - situ @xcite . the findings of this work can be used to understand the decoherence limits of utilizing rydberg atoms in cold , dense atom samples . most interesting will be future studies of the collisional lifetimes of rydberg atoms using atomic species , which do not have a _ p_-wave shape resonance for electron - neutral collisions , for example strontium . based on the collision mechanisms described in this work , the collisional lifetime depends on the initial temperature of the neutral atom reagent . these studies could be extended by measuring the collisional lifetime of rydberg atoms in thermal clouds with varying densities and temperatures . finally , the ultimate understanding of the lifetime of rydberg states in high density media will definitively help to understand the complex intimate interplay between few - body physics and many - body physics , which usually seems to be the most applied approach to these systems . we thank andreas khn and david peter for helpful discussions . c. h. g. acknowledge fruitful discussion with francis robicheaux and matt eieles . this work was supported by the deutsche forschungsgemeinschaft ( dfg ) within the sfb / trr21 and the project pf 381/13 - 1 . parts of this work was also founded by erc under contract number 267100 . s.h . acknowledges support from dfg through the project ho 4787/1 - 1 , m.s . acknowledges support from the carl zeiss foundation . this work was supported by the department of energy , office of science , under award no . de - sc0010545 . ( 1,0.46923913 ) ( 0,0 ) ion arrival time spread due to the @xmath11-changing collision . the top panel shows schematically the positions of the rydberg atoms at the moment of the ionization . the rydberg atoms are located on a sphere with the radius @xmath150 , which depends on the released energy @xmath151 . the bottom panel shows the relative ion arrival time of the ionized rydberg atoms which went through an @xmath11-changing collision . the highest principal quantum number ( dark blue ) shows the lowest spread , which means that the energy release during the collision was low . the lowest principal quantum number ( dark red ) shows the widest spread because of the higher energy release during the inelastic collision . all rydberg atoms are ionized after a delay time with an electric field of .,title="fig : " ] ( 0.96977349,0.1615649 ) ( 0,0 ) ion arrival time spread due to the @xmath11-changing collision . the top panel shows schematically the positions of the rydberg atoms at the moment of the ionization . the rydberg atoms are located on a sphere with the radius @xmath150 , which depends on the released energy @xmath151 . the bottom panel shows the relative ion arrival time of the ionized rydberg atoms which went through an @xmath11-changing collision . the highest principal quantum number ( dark blue ) shows the lowest spread , which means that the energy release during the collision was low . the lowest principal quantum number ( dark red ) shows the widest spread because of the higher energy release during the inelastic collision . all rydberg atoms are ionized after a delay time with an electric field of .,title="fig : " ] ( 0.58788768,0.36271499)(0,0)[lb ] ( 0,0 ) ion arrival time spread due to the @xmath11-changing collision . the top panel shows schematically the positions of the rydberg atoms at the moment of the ionization . the rydberg atoms are located on a sphere with the radius @xmath150 , which depends on the released energy @xmath151 . the bottom panel shows the relative ion arrival time of the ionized rydberg atoms which went through an @xmath11-changing collision . the highest principal quantum number ( dark blue ) shows the lowest spread , which means that the energy release during the collision was low . the lowest principal quantum number ( dark red ) shows the widest spread because of the higher energy release during the inelastic collision . all rydberg atoms are ionized after a delay time with an electric field of .,title="fig : " ] ( 0,0 ) ion arrival time spread due to the @xmath11-changing collision . the top panel shows schematically the positions of the rydberg atoms at the moment of the ionization . the rydberg atoms are located on a sphere with the radius @xmath150 , which depends on the released energy @xmath151 . the bottom panel shows the relative ion arrival time of the ionized rydberg atoms which went through an @xmath11-changing collision . the highest principal quantum number ( dark blue ) shows the lowest spread , which means that the energy release during the collision was low . the lowest principal quantum number ( dark red ) shows the widest spread because of the higher energy release during the inelastic collision . all rydberg atoms are ionized after a delay time with an electric field of .,title="fig : " ] ( 0.99919676,0.22019241)(0,0)[lb ] ( 0,0.36271475)(0,0)[lb ] + ion arrival time spread due to the @xmath11-changing collision . the top panel shows schematically the positions of the rydberg atoms at the moment of the ionization . the rydberg atoms are located on a sphere with the radius @xmath150 , which depends on the released energy @xmath151 . the bottom panel shows the relative ion arrival time of the ionized rydberg atoms which went through an @xmath11-changing collision . the highest principal quantum number ( dark blue ) shows the lowest spread , which means that the energy release during the collision was low . the lowest principal quantum number ( dark red ) shows the widest spread because of the higher energy release during the inelastic collision . all rydberg atoms are ionized after a delay time with an electric field of .,title="fig : " ] the released energy during an @xmath11-changing collision , as shown in fig . , can be extracted out of the arrival time spread of the ionized rydberg atoms on the detector . the method will be explained by the means of fig . [ fig : ionspread ] , which shows schematically ion flight trajectories leading to the ion arrival time spread . in addition , a measurement series of the ion arrival time spread for different principal quantum number is shown in this figure . all rydberg atoms were ionized with the same ionization electric field ( ) and a constant delay time between the excitation and the ionization ( ) , so that the spread of the ion arrival time for different principal quantum numbers can be directly compared . the full width at half maximum of the ion arrival time spread increases from for the highest state shown , @xmath152 , to for @xmath153 . the ion arrival time distribution @xmath154 can be simulated as a parameter of the sphere radius @xmath155 , which is the distance the rydberg atoms have traveled before the ionization with respect to the excitation center @xmath156 . due to the electric field gradients inside the experiment chamber and the asymmetric distance distribution from the ion detector to the rydberg atoms , the ion arrival time is asymmetric . the ion arrival time distribution of fig . [ fig : ionspread ] is reconstructed by fitting a linear combination of the simulated arrival time distributions @xmath154 to the measured arrival time distribution @xmath157 : @xmath158 the fit coefficients @xmath159 reflect the probability , @xmath160 , that a rydberg atom has traveled the distance @xmath155 in the time between the @xmath11-changing collision and the ionization . this spatial distribution can be converted into a distribution of the released energy @xmath161 with eq . ( [ eq : energyrelease ] ) . the maximum and the full width at half maximum of the energy distribution @xmath161 are finally plotted in fig . and the most probable total energy release corresponds for lower principal quantum numbers , for which the energy release is high , to the energy difference of the initially excited state to the next lower lying manifold . the ion arrival time spread was measured not only for but also for higher delay times ( up to ) to increase the sensitivity especially for high principal quantum numbers , for which the released energy is low . the method to simulate and extract information out of the ion arrival time was also used recently in another context by faoro and coworkers to measure in real space the effect of the van der waals forces between individual rydberg atoms @xcite . analysis of the @xmath11-changing collision time for a 100s state at a laser detuning of . the rydberg atoms were excited for and the first ionization pulse of mainly ionizes the rydberg atoms ( orange ) which are in the initial state . the subsequent higher ionization pulse ( ) ionizes the remaining rydberg atoms which went through the @xmath11-changing collision ( red ) . the sum of both pulses ( blue ) stays constant within the experimental uncertainty . the mean collision time is extracted out of the red fit curve @xmath162 as and the error given is coming from the confidence interval of the fit . the error bar of every data point is showing the standard error from the mean from 32 averages . ] the time of an inelastic collision is determined by measuring the reaction products as a function of the delay time @xmath163 between the excitation and ionization . an exemplary analysis is shown for the 100s state at a laser detuning of in fig . [ fig : taurb100s ] . after a variable delay time @xmath163 , a first ionization pulse ( ) is applied to ionize the 100s rydberg atoms , but not the rydberg atoms which went through an @xmath11-changing . the second ionization pulse ( ) ionizes the remaining rydberg atoms , which must be as a consequence in a high @xmath11 state . the measured signal is not zero at @xmath164 because rydberg atoms can already collide inelastically during the time of the excitation ( ) . the time constant @xmath135 of the collision is fitted with @xmath165 which results for the examined state in a collision time of @xmath166 . the sum of the ions detected in both ionization pulses is constant within the experimental uncertainty , which shows that all excited rydberg atoms are detected independent from the delay time before the ionization . avoided crossing between the butterfly state ( blue ) and the 88@xmath77 state ( black ) . this avoided crossing is studied by means of a landau - zener approach with eq . ( [ eq : lz ] ) , where @xmath167 denotes the energy gap at the crossing point and the orange line depicts the slope at the crossing point . ] in the landau - zener framework the non - adiabatic ( na ) transition probability is given by @xmath168 where @xmath169 denotes the velocity at the crossing point , @xmath170 is the difference between slopes of the diabatic potentials associated with the states under consideration @xcite , and @xmath167 denotes the energy difference of the two involved pec at the crossing point ( see fig . [ zoom ] ) . in fig . [ zoom ] the physics behind the landau - zener approach is illustrated for the crossing between the butterfly states and the 88@xmath77 state , as it was already shown in fig . [ fig : pec ] . the probability to end up in the 88@xmath77 state is given by eq . ( [ eq : lz ] ) , where the velocity at the crossing point in atomic units is given by @xmath171 for each landau - zener crossing an error bar has been estimated assuming a deviation @xmath172 20 @xmath173 in @xmath170 and 10 @xmath173 in @xmath167 , and the results are shown in panel ( b ) of fig . [ fig : pec ] of the main text . @xmath88-matrix elements associated with the butterfly - trilobite coupling ( blue ) , @xmath83-trilobite ( red ) and @xmath82-trilobite crossing ( orange ) for two different principal quantum numbers , @xmath174 ( dashed lines ) and @xmath175 ( solid lines ) . ( b ) probability for non - adiabatic transitions associated with the @xmath88-matrix displayed in panel ( a ) , see text for details . ] the coupling between the trilobite state and the rest of the involved rydberg states has been calculated by means of the nonadiabatic coupling @xmath88-matrix defined @xmath176 where @xmath177 and @xmath178 represent the adiabatic states involved . the @xmath88-matrix associated with the couplings between the trilobite state and the relevant states for @xmath11-mixing collisions are shown for two different quantum numbers @xmath12 = 40 and @xmath12 = 90 in panel ( a ) of fig . [ fig : pmatrix ] . observe that the @xmath82 state shows a negligible coupling with respect to the trilobite state , and hence every atom ending up in the @xmath82 state after passing through the butterfly region will reach the short - range region dominated by chemical forces . however , atoms in the @xmath83 state or butterfly state after the butterfly region will experience a considerable coupling with the trilobite state correlated with the @xmath57 hydrogenic manifold . the probability for non - adiabatic transitions can be calculated in atomic units as @xcite @xmath179 where @xmath180 @xcite , and @xmath169 denotes the velocity at the crossing point , @xmath167 stands for the energy difference between the involved adiabatic states , and @xmath181 denotes the value of the @xmath88-matrix at the crossing point , which corresponds to its maximum value . as it is noticed in panel ( a ) of fig . [ fig : pmatrix ] , some of the observed crossings ( regions where the @xmath88-matrix shows a local maximum ) are close to each other . therefore , eq . ( [ d2 ] ) has been extended to all distances , i.e. @xmath182 @xcite , where @xmath88 denotes the @xmath88-matrix associated with the cited adiabatic states . in panel ( b ) of fig . [ fig : pmatrix ] , the non - adiabatic transition probability for the crossings in panel ( a ) fig . [ fig : pmatrix ] are shown , where it is clearly observed that the trilobite state is strongly coupled to the butterfly state ( blue line ) . a large coupling between the trilobite state and the @xmath76 state is also observed ( orange line ) . these couplings are used in the theoretical description of the observed @xmath11-mixing collisions as discussed in the body of this paper . 49ifxundefined [ 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.99.163601 [ * * , ( ) ] link:\doibase 10.1038/nphys2413 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.112.043602 [ * * , ( ) ] link:\doibase 10.1103/physrevx.5.031015 [ * * , ( ) ] link:\doibase 10.1126/science.1244843 [ * * , ( ) ] http://arxiv.org/abs/1602.06313 [ ( ) ] , link:\doibase 10.1103/physrevlett.112.073901 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.113.053601 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.113.053602 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.85.2458 [ * * , ( ) ] link:\doibase 10.1038/nature07945 [ * * , ( ) ] http://arxiv.org/abs/1602.08400 [ ( ) ] , link:\doibase 10.1126/science.1211255 [ * * , ( ) ] link:\doibase 10.1038/nature12592 [ * * , ( ) ] link:\doibase 10.1088/1367 - 2630/17/5/053046 [ * * , ( ) ] link:\doibase 10.1088/0022 - 3700/20/6/011 [ * * , ( ) ] link:\doibase 10.1016/0370 - 1573(95)00074-q [ * * , ( ) ] link:\doibase 10.1016/s0166 - 1280(99)00134 - 7 [ * * , ( ) ] link:\doibase 10.1007/s10876 - 011 - 0438 - 7 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.115.013003 [ * * , ( ) ] link:\doibase 10.1063/1.1673523 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.35.644 [ * * , ( ) ] link:\doibase 10.1103/physreva.15.1945 [ * * , ( ) ] link:\doibase 10.1103/physreva.17.904 [ * * , ( ) ] link:\doibase 10.1088/0022 - 3700/15/15/016 [ * * , ( ) ] link:\doibase 10.1103/physreva.14.1354 [ * * , ( ) ] link:\doibase 10.1103/physreva.15.631 [ * * , ( ) ] link:\doibase 10.1103/physreva.18.1339 [ * * , ( ) ] @noop ( ) link:\doibase 10.1103/physreva.58.r2664 [ * * , ( ) ] https://books.google.de/books/about/bose_einstein_condensation_in_dilute_gas.html?id=ibk0g3_5iiqc&pgis=1[__ ] , ed . ( , ) link:\doibase 10.1103/physrevlett.116.053001 [ * * , ( ) ] link:\doibase 10.1103/physreva.93.032512 [ * * , ( ) ] link:\doibase 10.1017/cbo9780511524530 [ _ _ ] ( , , ) link:\doibase 10.1103/physreva.83.052515 [ * * , ( ) ] link:\doibase 10.1002/andp.19083300403 [ * * , ( ) ] link:\doibase 10.1103/physreva.69.063405 [ * * , ( ) ] link:\doibase 10.1016/j.physleta.2004.02.062 [ * * , ( ) ] link:\doibase 10.1088/0953 - 4075/35/10/101 [ * * , ( ) ] link:\doibase 10.1088/0953 - 4075/35/10/102 [ * * , ( ) ] link:\doibase 10.1016/b978 - 0 - 08 - 020940 - 1.50001 - 3 [ _ _ ] ( , ) link:\doibase 10.1016/0375 - 9601(79)90127 - 0 [ * * , ( ) ] link:\doibase 10.1038/ncomms5546 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.112.143008 [ * * , ( ) ] link:\doibase 10.1103/physreva.81.053607 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.105.163201 [ * * , ( ) ] http://arxiv.org/abs/1601.06881 [ ( ) ] , link:\doibase 10.1103/revmodphys.82.2313 [ * * , ( ) ] link:\doibase 10.1103/physreva.93.030701 [ * * , ( ) ]

recent work has indicated that ghost imaging may have applications in standoff sensing . however , most theoretical work has addressed transmission - based ghost imaging . to be a viable remote - sensing system , the ghost imager needs to image rough - surfaced targets in reflection through long , turbulent optical paths . we develop , within a gaussian - state framework , expressions for the spatial resolution , image contrast , and signal - to - noise ratio of such a system . we consider rough - surfaced targets that create fully developed speckle in their returns , and kolmogorov - spectrum turbulence that is uniformly distributed along all propagation paths . we address both classical and nonclassical optical sources , as well as a computational ghost imager .

1110.0845

ghost imaging is a transverse active - imaging technique that exploits the correlation between two light beams to image a target without spatially - resolving measurements of the light beam that has undergone target interaction . the two beams are the signal , which interacts with the target and then is measured by a single - pixel bucket detector , and the reference , which is directly measured by a high spatial - resolution detector . an intensity cross - correlation between the signal beam that encounters the target and the reference beam that impinges on the high spatial - resolution detector imparts target information the ghost image to the cross correlation between the photocurrents obtained from the two detectors , information that is unavailable from either photocurrent alone . most ghost - imaging experiments and related theory deal with the transmissive case , in which the bucket detector is placed behind the target , and we image the target s intensity - transmission profile . in the first ghost imaging experiment , the signal and reference beams were the quantum - mechanically entangled outputs from a spontaneous parametric downconverter ( spdc ) @xcite , which afforded them a phase - sensitive cross correlation stronger than permitted by classical physics @xcite . the system was run at low flux , in which the spdc s post - selected output state , within the photodetectors response time , could be taken to be a biphoton ; the ghost image was then formed by counting coincidences between the bucket and high spatial - resolution detections instead of a photocurrent cross correlation . subsequently , ghost imaging was done with a high - flux pseudothermal source , i.e. , a classical - state source realized by sending a laser beam through a rotating ground - glass diffuser and a 50 - 50 beam splitter to create identical , spatially - incoherent signal and reference beams with a phase - insensitive cross correlation . the ghost image was then obtained by continuous - time cross correlation of the bucket and high spatial - resolution photocurrents @xcite . later , the theories of spdc and pseudothermal ghost imaging were united in a gaussian - state treatment that showed , in both cases , that ghost - image formation by photocurrent correlation arises from classical coherence propagation , but the nonclassical spdc source offers much higher image contrast and a modest spatial - resolution advantage in near - field operation @xcite . recently , a computational ghost - imaging method was introduced , in which the need for a reference beam is removed @xcite . conventional ghost imaging relies on the intensity cross - correlation between the signal beam on the target and the reference beam on the high spatial - resolution detector . in spdc and pseudothermal ghost imaging , these beam patterns are random . if , however , a spatial light modulator is used to impart a sequence of known spatial patterns to a laser beam prior to propagation to the target , then diffraction theory can be used to calculate noiseless versions of the associated reference - beam intensity patterns . these computed reference - beam results can be employed , for cross - correlation ghost - image formation , as if they came from photodetection of a physical reference beam . more importantly , computational ghost imaging opens the door for more advanced computational image reconstruction techniques , notably compressive sensing @xcite . all of the preceding discussion has been framed for ghost imaging of transmissive targets . recent experiments have shown the feasibility of an alternate ghost - imaging configuration , in which the bucket detector views the target in reflection , rather than in transmission @xcite . in this reflective setup , the source and detector can , in principle , be moved arbitrarily far away from the target , suggesting that ghost imaging could be a viable remote - sensing technique . atmospheric turbulence will surely be a relevant concern in any such standoff - sensing application , just as it is for astronomical imaging and laser radar . to date , there has been theoretical study of the impact of turbulence on ghost imaging in transmission @xcite , but there has only been an initial turbulence - free theory development for ghost imaging in reflection @xcite , and that work was limited to pseudothermal sources . in this paper we will extend the analysis from @xcite to include spdc ghost imaging in reflection and , for both spdc and pseudothermal operation , the presence of atmospheric turbulence in the propagation paths . the setup we shall consider is the lensless ghost - imaging configuration shown schematically in fig . [ fig : setup ] @xcite . pseudothermal ghost imaging can be described quantitatively using semiclassical photodetection theory in which the fields are described classically and photodetection incurs the shot noise associated with the discreteness of the electron charge whereas quantum photodetection theory is required for the spdc case , owing to the nonclassical nature of its output state @xcite . nevertheless , to provide a unified development , we will employ quantum photodetection to characterize both pseudothermal and spdc ghost imaging . we begin , in sec . [ sec : setup ] , with the theoretical framework for our analysis . here we will describe our models for the pseudothermal and spdc sources , reflection from the target , atmospheric propagation , quantum photodetection , and image formation . in sec . [ sec : rescon ] we will derive expressions for the spatial resolution and image contrast for pseudothermal and spdc ghost images , and in sec . [ sec : snr ] we will derive their respective signal - to - noise ratios ( snrs ) . then , in sec . [ sec : comp ] , we shall consider a computational ghost imager , in which the reference beam is calculated instead of measured @xcite . because the reference beam is not an optical field , we find it convenient in sec . [ sec : comp ] to employ the semiclassical theory . this section will also include a quantitative performance comparison between pseudothermal , sdpc , and computational ghost images formed in reflection . we conclude , in sec . [ sec : discussion ] , with a discussion of our findings , and an assessment of the situations in which reflective ghost - imaging might have practical value . m to the ccd camera , while the signal beam travels @xmath0 m to the target , and the reflected light travels @xmath0 m to the bucket detector . the image is then formed by cross correlation between the photocurrents obtained from the bucket detector and the high spatial - resolution ( ccd camera ) detector.,width=336 ] we are interested in ghost imaging with classical pseudothermal light , and the nonclassical output of an spdc . in both cases our signal and reference beams will be represented quantum - mechanically as @xmath1-units , @xmath2-going , positive - frequency , field operators@xmath3 and @xmath4 , as functions of transverse - coordinate vector @xmath5 and time at the output of the fig . [ fig : setup ] beam splitter . their baseband field operators have the canonical commutation relations @xmath6 = 0 , \\[.05 in ] \label{eq : comm2}&[{\hat{e}}_j({\bm\rho}_1,t_1),{\hat{e}}^\dagger_k({\bm\rho}_2,t_2 ) ] = \delta_{jk}\delta({\bm\rho}_1-{\bm\rho}_2)\delta(t_1-t_2),\end{aligned}\ ] ] for @xmath7 and @xmath8 , where @xmath9 is the kronecker delta , and @xmath10 is the unit impulse . following @xcite , we will take both the pseudothermal and spdc sources to produce signal and reference beams that are in zero - mean , jointly gaussian states , which are then completely characterized by their nonzero autocorrelation and cross - correlation functions . all the autocorrelations will be assumed to have the same gaussian - schell model form , @xmath11 & = \frac{2p}{\pi a_0 ^ 2}e^{-\frac{|{\bm\rho}_1|^2+|{\bm\rho}_2|^2}{a_0 ^ 2 } } e^{-\frac{|{\bm\rho}_1-{\bm\rho}_2|^2}{2\rho_0 ^ 2}}e^{-\frac{(t_1-t_2)^2}{2 t_0 ^ 2}},\end{aligned}\ ] ] for @xmath8 , where @xmath12 is the photon flux , @xmath13 is the source s @xmath14 intensity radius , @xmath15 is its coherence length , and @xmath16 is its coherence time . the distinction between the gaussian - state models for the pseudothermal and spdc sources lies in their signal - reference cross correlations , as we will now explain . the pseudothermal signal and reference beams start as continuous - wave ( cw ) laser light , that is passed through a rotating ground - glass diffuser to make it spatially incoherent . signal and reference beams are then obtained by 50 - 50 beam splitting . the signal and reference thus have no phase - sensitive cross correlation , and the maximum phase - insensitive cross correlation allowed by classical physics given their autocorrelation functions , i.e. , @xmath17 \begin{split}\label{eq : ccorrpis } { \langle { \hat{e}}^\dagger_s&({\bm\rho}_1,t_1){\hat{e}}_r({\bm\rho}_2,t_2 ) \rangle}_c \\[.05 in ] & = \frac{2p}{\pi a_0 ^ 2}e^{-\frac{|{\bm\rho}_1|^2+|{\bm\rho}_2|^2}{a_0 ^ 2 } } e^{-\frac{|{\bm\rho}_1-{\bm\rho}_2|^2}{2\rho_0 ^ 2}}e^{-\frac{(t_1-t_2)^2}{2 t_0 ^ 2 } } , \end{split}\end{aligned}\ ] ] where the subscript @xmath18 indicates that these are classical - state cross correlations . the coherence time is directly related to how fast the ground - glass rotates , the coherence length is typically on the order of a few wavelengths , and the intensity radius is set by the radius of the initial laser beam . a type - ii phase - matched , collinear , cw spdc source emits orthogonally - polarized , co - propagating signal and idler beams , which are separated by a polarizing beam splitter to become the signal and reference beams for ghost imaging . they are in a maximally - entangled , jointly - gaussian state with no phase - insensitive cross correlation , and the maximum phase - sensitive cross correlation permitted by quantum theory given their autocorrelation functions , namely @xcite @xmath19 \begin{split}\label{eq : qcorrps } { \langle { \hat{e}}_s({\bm\rho}_1,t_1)&{\hat{e}}_r({\bm\rho}_2,t_2 ) \rangle}_q = \frac{2p}{\pi a_0 ^ 2 } e^{-\frac{|{\bm\rho}_1|^2+|{\bm\rho}_2|^2}{a_0 ^ 2 } } \!\ ! \left[e^{-\frac{|{\bm\rho}_1-{\bm\rho}_2|^2}{2\rho_0 ^ 2}}e^{-\frac{(t_1-t_2)^2}{2t_0 ^ 2 } } + i \left(\frac{2}{\pi}\right)^\frac{1}{4 } \!\sqrt{\frac{a_0 ^ 2}{p t_0 \rho_0 ^ 2}}e^{-\frac{|{\bm\rho}_1-{\bm\rho}_2|^2}{\rho_0 ^ 2}}e^{-\frac{(t_1-t_2)^2}{t_0 ^ 2 } } \right ] , \end{split}\end{aligned}\ ] ] where the subscript @xmath20 indicates that these are quantum ( nonclassical ) cross correlations . the signal and reference beams in fig . [ fig : setup ] first propagate @xmath0 m over spatially - separated paths from their shared source location @xcite . the reference beam is then measured by a high - spatial resolution ccd camera , while the signal field reflects off the target and travels @xmath0 m to the bucket detector . we will assume statistically - independent turbulence on all three paths : the reference path ( @xmath21 ) from the source to the ccd ; the signal path ( @xmath22 ) from the source to the target ; and the target - return path ( @xmath23 ) from the target to the bucket detector . to account for the kolmogorov - spectrum turbulence that is uniformly distributed along the three propagation paths , we make use of the quantum version of the extended huygens - fresnel principle @xcite @xmath24 where @xmath25 , and @xmath26 is a complex - valued random process that encapsulates the effects of turbulence on path @xmath27 from @xmath5 to @xmath28 . the real and imaginary parts of @xmath29 , which we will denote @xmath30 and @xmath31 , are the logamplitude and phase fluctuations imposed on the field received at transverse coordinate @xmath28 over path @xmath27 from a point source at transverse coordinate @xmath5 . because all the path lengths in fig . [ fig : setup ] are equal , we have suppressed the @xmath32 line - of - sight propagation delay in eq . ( [ eq : prop ] ) . we have also suppressed any time - dependence of the turbulence . physically , this amounts to assuming that the integration time , @xmath33 , employed in the fig . [ fig : setup ] correlator is shorter than the @xmath34ms coherence time of the turbulence . however , that assumption is _ not necessary for deriving the spatial resolution and image contrast achieved in the presence of turbulence , although it _ will be relevant to our signal - to - noise ratio analysis . _ _ the mutual coherence function of @xmath35 is taken to be @xcite @xmath36 & = e^{-(|{\bm\rho}'_1-{\bm\rho}'_2|^2+({\bm\rho}'_1-{\bm\rho}'_2)\cdot({\bm\rho}_1-{\bm\rho}_2 ) + |{\bm\rho}_1-{\bm\rho}_2|^2)/2\rho_m^2},\end{aligned}\ ] ] which is the square - law approximation to the rigorous @xmath37-law behavior for uniformly - distributed turbulence @xcite . in eq . ( [ eq : turbcorr ] ) , @xmath38 is the turbulence coherence length on path @xmath27 , given by @xmath39 in terms of the constant turbulence strength - parameter , @xmath40 , along path @xmath27 @xcite . we will assume that the path length @xmath41 satisfies the far - field coherence - propagation conditions for pseudothermal and spdc light discussed in @xcite , viz . , @xmath42 , and @xmath43 , respectively . for convenience , we also define @xmath44 and @xmath45 . these are the far - field , turbulence - free , on - target coherence length and intensity radius produced by the gaussian - schell autocorrelation function from eq . ( [ eq : autocorr ] ) . most real - world targets have surfaces that are sufficiently rough on the scale of an optical wavelength that reflections from them present fully - developed speckle patterns at appreciable distances from the surface @xcite . we will model such a target as a planar 2-d object with random , microscopic height variations having depths on the order of several wavelengths , and transverse variations on the order a wavelength and an average intensity - reflection coefficient @xmath46 . thus , following laser - radar theory @xcite , we shall assume that this target has a random field - reflection coefficient , @xmath47 , that is a zero - mean , complex - valued gaussian random process characterized by the autocorrelation function @xmath48 where the deterministic pattern @xmath46 is what we are tying to image . target reflection is a passive process , so we require @xmath49 , which is in conflict with the gaussian statistics and the delta - function term in eq . ( [ eq : scatter ] ) . that delta function , however , leads to quasi - lambertian reflection , implying that at standoff distances a realistic bucket detector will only capture a very small fraction of the reflected light . hence using the reflection statistics from the preceding paragraph is a reasonable approximation for ghost imaging of rough - surfaced targets . however , because some light may be absorbed by or transmitted through the target , the quantum model for target reflection needs to be @xmath50 here @xmath51 is a vacuum - state field operator , needed to preserve the free - field commutator relations given , for the source fields , in eqs . and . in what follows we will assume that the target s intensity - reflection coefficient , @xmath46 , is space - limited to the region illuminated by the signal beam , i.e. , @xmath52 for @xmath53 this condition will give the ensemble - average pseudothermal and sdpc ghost images convolution - integral relationships with @xmath46 from which to determine their spatial resolutions . see @xcite for information about ghost imaging field - of - view when the preceding restriction on @xmath46 is not satisfied . loss also occurs in the photodetection process , because the detectors in fig . [ fig : setup ] will , in general , have sub - unity quantum efficiencies . therefore , the field operator @xmath54 driving the photocurrent , @xmath55 , from ccd pixel @xmath56 , and the field operator @xmath57 driving the photocurrent , @xmath58 , from the bucket detector satisfy @xcite @xmath59 for @xmath60 or @xmath61 . here , @xmath62 is the quantum efficiency , which is taken to be the same for both the bucket detector and the ccd camera , and @xmath63 is a vacuum - state field operator whose inclusion models the loss incurred by having @xmath64 , cf . eq . ( [ eq : targetloss ] ) . we will assume ghost - image formation via continuous - time , pixel - wise correlation of the photocurrents from the ccd and the bucket detector , formalized as @xmath65 where @xmath66 is the center coordinate of the ccd s @xmath56th pixel , and @xmath33 is the correlation integration time . from quantum photodetection theory @xcite , we have that @xmath67 where @xmath68 is the effective ( sub - unity quantum efficiency ) photon flux on the active region @xmath69 for detector @xmath27 , and @xmath70 is the impulse response of the detectors output circuit , which we have taken to be the same for both the bucket detector and the ccd camera . to enable the performance evaluations in the sections to follow , we make the following assumptions . first , each pixel is small enough that the the average photon flux impinging on it obeys @xmath71 , where @xmath72 is the area of the @xmath73 . second , we assume that the detector has a gaussian impulse response , @xmath74 whose bandwidth , @xmath75 , greatly exceeds that of the incident light for the pseudothermal source ( @xmath76 ) , and is much smaller than that of the incident light for the spdc source ( @xmath77 ) . finally , we assume that the correlation integration time is sufficiently long to capture a ghost image , so @xmath78 for the pseudothermal source , and @xmath79 for the spdc source . the spatial resolution and contrast of the ghost image are properties of the average cross correlation , @xmath80 . to determine this ensemble - average cross correlation we : ( 1 ) write @xmath58 and @xmath55 in terms of the field operators that drive them , @xmath57 and @xmath54 ; ( 2 ) apply the canonical commutator relations to normally - order the resulting photocurrent - product expressions ; and ( 3 ) use eq . to write the detected fields in terms of @xmath81 and @xmath82 . from the linearity of expectation , we can bring the ensemble - average operation inside the time - domain integration to get @xmath83 & \times { \langle { \hat{e}}'^\dagger_r({\bm\rho}_p,\tau_1 ) { \hat{e}}'^\dagger_t({\bm\rho},\tau_2 ) { \hat{e}}'_r({\bm\rho}_p,\tau_1 ) { \hat{e}}'_t({\bm\rho},\tau_2 ) \rangle}.\end{aligned}\ ] ] we then use eqs . and to back - propagate the target - return field operator to the target , and find @xmath84 & = & \int\ ! d{\bm\rho}_2 \!\int\ ! d{\bm\rho}_3\ , { \langle t^*({\bm\rho}_2)t({\bm\rho}_3 ) \rangle}{\langle e^{\psi^*_t({\bm\rho},{\bm\rho}_2)}e^{\psi_t({\bm\rho},{\bm\rho}_3 ) } \rangle}\nonumber \\[.05 in ] & \times & { \langle { \hat{e}}'^\dagger_r({\bm\rho}_p,\tau_1 ) { \hat{e}}'^\dagger_s({\bm\rho}_2,\tau_2 ) { \hat{e}}'_r({\bm\rho}_p,\tau_1 ) { \hat{e}}'_s({\bm\rho}_3,\tau_2 ) \rangle } \nonumber \\[.05 in ] & \times & \frac{e^{-ik_0(|{\bm\rho}_2|^2-|{\bm\rho}_3|^2)/2l + ik_0{\bm\rho}\cdot({\bm\rho}_2-{\bm\rho}_3)/l}}{(\lambda_0l)^2},\end{aligned}\ ] ] where we have used the statistical independence of the target s reflection coefficient , the signal and reference fields at the source , and the turbulence on all three propagation paths . substituting for @xmath85 from eq . and performing the @xmath86 integral eliminates the turbulence term from the preceding @xmath80 expression , which shows that turbulence on the the target - to - bucket path has no impact on the ghost image s spatial resolution or image contrast . now , by back - propagating the field operators to the source by means of eq . with @xmath87we have that @xmath88 & = & \frac{1}{\lambda_0 ^ 4l^6}\int\ ! d{\bm\rho}_2 \ , { \mathcal{t}}({\bm\rho}_2 ) \int\ ! d{\bm\rho}'_1 \int\!d{\bm\rho}'_2 \int\!d{\bm\rho}''_1 \int \ ! d{\bm\rho}''_2 \nonumber \\[.05 in ] & \times & { \langle { \hat{e}}^\dagger_r({\bm\rho}'_1,\tau_1 ) { \hat{e}}^\dagger_s({\bm\rho}'_2,\tau_2 ) { \hat{e}}_r({\bm\rho}''_1,\tau_1 ) { \hat{e}}_s({\bm\rho}''_2,\tau_2 ) \rangle } \nonumber \\[.05 in ] & \times & { \langle e^{\psi^*_r({\bm\rho}_p,{\bm\rho}'_1)}e^{\psi_r({\bm\rho}_p,{\bm\rho}''_1 ) } \rangle}{\langle e^{\psi^*_s({\bm\rho}_2,{\bm\rho}'_2)}e^{\psi_s({\bm\rho}_2,{\bm\rho}''_2 ) } \rangle } \nonumber \\[.05 in ] & \times & e^{ik_0[{\bm\rho}_p\cdot({\bm\rho}'_1-{\bm\rho}''_1 ) + { \bm\rho}_2\cdot({\bm\rho}'_2 - { \bm\rho}''_2)]/l},\end{aligned}\ ] ] where we have again invoked the statistical independence of the signal and reference fields at the source and the turbulence on the source - to - ccd , and source - to - target paths . we can now directly evaluate all remaining moments . the turbulence moments are available from eq . , and the fourth - order field moment can be expressed in terms of second - order moments with the gaussian - state moment - factoring theorem @xcite : @xmath89 until this point , our analysis applies equally well to the pseudothermal and spdc ghost imagers . that equivalence disappears , however , when we evaluate the second - order moments in eq . using eqs . for the pseudothermal ghost imager , and using eqs . , , and for the spdc ghost imager . the results we obtain are @xmath90 , \label{eq : cavgmean}\end{aligned}\ ] ] for the pseudothermal ( classical ) imager , and @xmath91 & \times & \left [ 1 + \frac{\omega_b t_0}{4}\frac{e^{-|{\bm\rho}-{\bm\rho}_p|^2/\alpha\rho_l^2}}{\alpha}\left(1+\frac{1}{4\sqrt{\pi}\mathcal{i}}\right)\right ] . \label{eq : qavgmean}\end{aligned}\ ] ] for the spdc ( quantum ) ghost imager , where @xmath92 is the area of @xmath93 , @xmath94 is the source brightness in photons per spatiotemporal mode , and @xmath95 is a turbulence - induced performance - degradation factor . we see that both average correlation functions are the sum of the same featureless background term @xmath96 and different image - bearing terms : @xmath97 for the classical - state ( pseudothermal ) source , and @xmath98 & \times & \frac{\omega_b t_0}{4}\int\!d{\bm\rho}\ , { \mathcal{t}}(-{\bm\rho } ) e^{-\frac{|{\bm\rho}-{\bm\rho}_p|^2}{\alpha\rho_l^2}},\end{aligned}\ ] ] for the nonclassical - state ( quantum ) source . the average pseudothermal ghost image is erect , and consists of a scaled version of the target s intensity - reflection coefficient @xmath46 convolved with a gaussian point - spread function whose @xmath99 radius is @xmath100 . the average spdc ghost image is inverted , with a different scaling but the same point - spread function . thus , both the pseudothermal and spdc ghost images have the same spatial resolution , @xmath101 } = \label{eq : res } } \\[.05 in ] & & \hspace*{-.1 in } \frac{\lambda_0l } { \pi a_0}\sqrt{1+\frac{a_0 ^ 2}{2 } ( 1.09 k_0 ^ 2 l)^\frac{6}{5}\left[(c_{n , s}^2)^\frac{6}{5 } + ( c_{n , r}^2)^\frac{6}{5}\right]},\end{aligned}\ ] ] where we have utilized eq . to write the turbulence coherence lengths on each path in terms of their respective strength parameters . equation shows that turbulence - limited resolution prevails for the far - field lensless ghost image when either the source - to - target coherence length ( @xmath102 ) or the source - to - reference coherence length ( @xmath103 ) becomes smaller than the source size ( @xmath13 ) . when @xmath104 and @xmath105 , far - field lensless ghost imaging in reflection yields the same @xmath106 spatial resolution previously reported for far - field lensless ghost imaging in transmission @xcite . however , even when turbulence does not degrade ghost - image spatial resolution , there is still a significant difference between the transmissive and reflective cases , namely the @xmath107 factor that is present in our results but is absent from those in @xcite . this factor , which will obey @xmath108 in a standoff - sensing scenario , is the bucket detector s angular subtense at the target , and represents the fraction of the quasi - lambertian reflected light that this detector collects . no such factor appeared in @xcite , because that work assumed the bucket detector collected all the light that was transmitted through the target . the featureless background terms in our pseudothermal and spdc ghost images affect the images contrast . to assess these contrasts , we will adopt the definition and approach presented in @xcite . in particular , the image contrast @xmath109 is the ratio of the difference between the brightest and darkest pixels in the image to the featureless background : @xmath110-\min_{\mathcal{r}}[\hat{c}({\bm\rho}_p)]}{c_0},\end{aligned}\ ] ] where @xmath111 is the region within which @xmath46 is nonzero . we will assume that the ghost imager completely resolves the target , so that we can use the approximation @xmath112 defining @xmath113 as the target s reflective cross - section , and taking @xmath114 = 1 $ ] , we get the following image contrast expressions for our two ghost imagers , @xmath115 and @xmath116 if @xmath117 where it is nonzero , then @xmath118 is the area of the target . moreover , @xmath119 is the spatial - resolution area in the absence of turbulence . it follows that the pseudothermal ghost - image contrast is approximately @xmath120 , where @xmath121 is number of no - turbulence resolution cells in the image . note that our analysis to this point has assumed a dc - coupled correlator in fig . [ fig : setup ] . the contrast of a pseudothermal ghost image can be improved by suppressing the background term through ac - coupling , i.e. , measuring the cross covariance , rather than the cross correlation , between the bucket and high spatial - resolution photodetector s outputs , as was done in the pseudothermal experiments of scarcelli _ et al . _ the pseudothermal imager s contrast does not depend on its source brightness , @xmath122 , but that for the spdc does . at high source brightness , @xmath123 , the spdc image contrast reduces to the result we found for the pseudothermal case . the normal operating regime for a cw spdc , however , is one of low brightness , @xmath124 , for which eq . reduces to @xmath125 the number of photon pairs emitted by an spdc in one detector time - constant is low , so @xmath126 , and our assumption that the target is contained within the region illuminated by the signal beam implies that @xmath127 if @xmath117 where it is nonzero . consequently , unlike the pseudothermal case , @xmath128 is the norm for sdpc ghost imaging in dc - coupled operation . in keeping with the analysis done in @xcite , we will be evaluating the signal - to - noise ratio ( snr ) pixel - wise as the ratio of the square of the expected value to the variance for the photocurrent cross - correlation function , @xmath129 } = \frac{{\langle \hat{c}({\bm\rho}_p ) \rangle}^2}{{\langle \hat{c}^2({\bm\rho}_p ) \rangle}-{\langle \hat{c}({\bm\rho}_p ) \rangle}^2}.\end{aligned}\ ] ] the main complication in evaluating eq . is the second moment , @xmath130 after using eq . to write the currents in terms of the fields , eqs . and to normally order the fields , and eqs . and to propagate the fields back to the source , we are left needing to evaluate an eighth - order moment of the fields , two sixth - order moments of the fields , a fourth - order moment of the fields , a fourth - order moment of the target , three fourth - order moments of the turbulence , and propagation of these terms over 12 unique paths . to make the analysis more tractable , we use several simplifying assumptions . first , we ac - couple both of the detector outputs by inserting a narrowband notch centered at the origin in frequency space , resulting in the frequency response @xmath131 where @xmath132 . we also assume that the ghost imager can resolve all structure on the target , allowing us to use the approximation in eq . . to evaluate the fourth - order turbulence moments , we can no longer use the square - law approximation in eq . that models the turbulence s mutual - coherence function as gaussian , because this approximation is not valid for fourth - order moments @xcite . to evaluate these terms we assume that the turbulence coherence length on each path is much larger than the on - target coherence length @xmath133 . we also assume the turbulence coherence lengths are all much larger than @xmath13 , implying that the turbulence functions are independent of coordinates at the source and detection planes . although these assumptions are somewhat limiting , they still allow weak - to - medium strength turbulence for a typical situation . for instance , if the target is standing off @xmath134 km , the source aperture has radius @xmath135 cm , and we are at a wavelength @xmath136 m , this assumption is satisfied for @xmath137 . moreover , this situation has no spatial - averaging of the turbulence at the target , and thus represents a worst - case scenario for the snr . for @xmath138 and @xmath139 being coordinates at the ccd or bucket detectors , and @xmath140 being coordinates at the target , the preceding coherence - length assumptions imply that @xmath141 + 2{\rm re}[\psi_m(\bm 0 , { \bm\rho}_2 ) ] } \rangle}.\end{aligned}\ ] ] decomposing the turbulence function into its log - amplitude and phase components as @xmath142 , and taking the logamplitude to be gaussian distributed with mean equal to minus its variance @xcite , then gives @xmath143 \approx e^{4 k_{\chi_m } ( { \bm\rho}_1-{\bm\rho}_2)},\end{aligned}\ ] ] where @xmath144 for @xmath145 , is the log - amplitude covariance function . the logamplitude variance will be taken to be @xcite @xmath146 where we have used the rytov - approximation expression , whose validity is ensured by our assumption of weak - to - medium strength turbulence . even with these simplifications , the snr evaluation is quite tedious , if somewhat straightforward . the higher - order field moments and the fourth - order target moment can be reduced to a sum of the products of second - order moments by application of the gaussian moment - factoring theorem . the ac - coupling removes several of these terms , and we are left with integrals and fourier transforms of the product of gaussian terms , the turbulence correlation functions , and the target s intensity - reflection coefficient , @xmath46 . to simplify our final snr expressions , we define two new terms , @xmath147 where @xmath148 is the two - circle overlap function for circles of diameter @xmath149 , @xmath150 & & \hspace*{-.2in}\left\{\begin{array}{ll } \frac{d^2}{2}\left [ \cos^{-1 } \left ( \frac{|\bm\zeta|}{d}\right ) -\frac{|\bm\zeta|}{d}\sqrt{1-\frac{|\bm\zeta|^2}{d^2 } } \right ] , & \mbox{for $ |\bm\zeta|\le d$ } \\[.05 in ] 0 , & \text{elsewhere , } \end{array}\right.\end{aligned}\ ] ] @xmath151 , and we have utilized the dimensionless difference coordinates @xmath152 , where @xmath28 and @xmath153 are coordinates at the bucket detector . both @xmath154 and @xmath155 have significant physical interpretations . similar to @xmath118 , @xmath154 is another measure of the reflective area of the target , and thus directly related to the number of on - target resolution cells , and the subsequent amount of time it takes to form an image . @xmath155 is a metric for the spatial averaging of the target - induced speckle over the bucket detector . for very small detectors @xmath156 , while for large detectors it is inversely proportional to the area of the bucket detector , viz . @xmath157 . to further simplify our snr expressions , we make two final assumptions : first that @xmath158 , which is equivalent to saying that the ghost image consists of at least @xmath159 resolution cells ; and second , that the bucket detector is at least as large as the area of the source beam , or @xmath160 . applying these conditions we arrive at the following snr expressions for pseudothermal and spdc reflective ghost imaging through atmospheric turbulence , @xmath161 ^ 2}{\displaystyle \frac{4}{\omega_bt_i}\delta^2\text{source } + \delta^2\text{path}\left[1+\frac{1}{4\sqrt{\pi}\mathcal{i}}\right]^2 + \frac{t_0 4\sqrt{2}}{t_i}\delta^2\text{detect}\left[1+\frac{1}{4\sqrt{\pi}\mathcal{i}}\right]+ \frac{t_0}{t_i}\delta^2\text{mix}\left[\frac{2}{\sqrt{3}}+\frac{1}{4\sqrt{\pi}\mathcal{i}}\right]}.\end{gathered}\ ] ] where , for ease of comparison with the pseudothermal imager , we have inverted the image coordinates in the spdc case . the terms that appear in the noise denominators of eqs . ( [ eq : csnr ] ) and ( [ eq : qsnr ] ) are as follows : @xmath162 , \\ & \delta^2\text{detect } = \frac{{\mathcal{t}}({\bm\rho}_p ) \rho_l^2 \sqrt{\pi}}{16 \sqrt{2 } a_p \eta^2 \mathcal{i}^2}\frac{l^2}{a_b } , \\ & \delta^2\text{mix } = \frac{{\mathcal{t}}({\bm\rho}_p)}{\eta\mathcal{i}}\frac { l^2}{a_b}e^{4\sigma_r^2}\nonumber\\ & \quad\quad+\frac{\pi\rho_l^2{\mathcal{t}}^2({\bm\rho}_p)}{a_p\eta\mathcal{i}}\left[\frac{4}{3}+\beta^{-1}\right]e^{4(\sigma_s^2+\sigma_t^2)},\end{aligned}\ ] ] where @xmath163 for @xmath164 denotes the turbulence - induced logamplitude variance on the reference ( @xmath21 ) , signal ( @xmath22 ) , and target ( @xmath23 ) paths . each of these noise terms has an important physical interpretation , as we explain below . the first noise term , @xmath165 , captures the fluctuations associated with decorrelation of transverse locations in the image arising from the source s time evolution . thus , it is inversely proportional to the number of on - target resolution cells , @xmath166 , and the number of different source patterns measured . for the pseudothermal imager , for which @xmath167 holds , this is the number of field - coherence times in the integration time , so we find @xmath165 scaled by @xmath168 . for the spdc imager we have @xmath169 , so it is the detectors response time that limits the measurements correlation behavior , hence we find @xmath165 scaled by @xmath170 in that imager s snr denominator . the @xmath171 term is due to the optical - path interactions that degrade the measurement , i.e. , the turbulent atmosphere , and the scattering off the rough target . the noise contribution from target scattering can be mitigated by aperture averaging , which , as noted above , is quantified by @xmath155 . the effect of the turbulence is somewhat more complicated , as it contributes a scaling factor @xmath172 to all field - variation terms on path @xmath173 . in addition , the logamplitude fluctuations also cause a deviation that results in the squared - mean term in the denominator of eq . no longer being canceled . these effects are particularly devastating because @xmath171 is the only noise term whose strength is independent of the detector integration time . this is because we have assumed time - independent target behavior and @xmath33 less than the @xmath34ms turbulence coherence time . consequently , regardless of the source brightness and integration time , @xmath171 sets an ultimate upper limit on the snr . the noise term @xmath174 results from the statistics of photodetection at both the reference and bucket detectors , and as such is inversely proportional to the photon fluxes impinging on their active regions . thus , @xmath174 is inversely proportional to the size of each pixel , @xmath72 , and the solid - angle subtense factor @xmath107 . detection noise is mitigated by temporal averaging , but exacerbated by having a fast detector , so the pseudothermal @xmath174 is scaled by @xmath175 , while for the spdc imager it is scaled only by @xmath168 . the spdc case is also scaled by by @xmath176 , a result of that source s stronger - than classical correlation . the @xmath177 term contains noise contributed by mixing between source plus path noise on the reference arm with detection noise on the signal arm , and vice versa . for both imagers this term has a temporal scaling of @xmath168 , and for the spdc imager the quantum correlation imparts an additional factor of @xmath178 . to get a better understanding of the snr behavior for pseudothermal and spdc ghost imagers , we will consider three limiting cases . the first is the long integration - time limit in which @xmath171 dominates the noise denominators for both the pseudothermal and spdc ghost imagers , leading to the _ identical maximum ( saturation ) snr value @xmath179-e^{-4(\sigma_r^2+\sigma_s^2+\sigma_t^2)}},\end{aligned}\ ] ] regardless of the strength of the illumination . _ for our next two cases , we will assume the integration time is sufficiently short that snr saturation is not approached . then , in the high - brightness ( @xmath180 ) case , for which detection noise can be neglected , performance is dominated by source fluctuations , @xmath165 , yielding @xmath181 for pseudothermal ( @xmath18 ) and spdc ( @xmath20 ) imagers , respectively . with all parameters other than source coherence - time being equal , the spdc high - brightness snr exceeds that of the pseudothermal source by a factor @xmath167 , where @xmath16 is the pseudothermal source s coherence time . this is because the much faster fluctuations in the spdc output fields lead to faster decorrelation , and hence higher snr . that said , however , the reader is reminded that cw spdc operation is ordinarily in the low - brightness regime , hence we turn our attention now to that special case . in the low - brightness , @xmath182 , regime , snr behavior is dominated by the beat between the detection noises , @xmath174 . here we find that @xmath183 for the pseudothermal and spdc cases . the spdc snr is linear in source brightness , whereas the pseudothermal snr is quadratic in that parameter . this behavior is to be expected , because the spdc low - brightness output approaches the biphoton state , which is known to have lower detection noise than classical - state light . furthermore , @xmath16 for the pseudothermal source will greatly exceed @xmath16 for the spdc , and , as noted in the preceding paragraph , @xmath167 holds for the pseudothermal source . it follows that the low - brightness spdc snr is much higher than that of the pseudothermal imager . the reader is cautioned to remember , however , that pseudothermal ghost imaging is typically performed in the high - brightness regime . it is also instructive to compare our snr results which apply to reflective ghost imaging through atmospheric turbulence to the transmissive ghost imaging , no - turbulence results from @xcite . the principal differences between the snr behaviors in these two cases is as follows . our snr expressions have the average intensity reflection coefficient @xmath184 appearing in lieu of the magnitude - squared of the field reflection coefficient @xmath185 that is seen in the transmissive case . moreover , the transmissive - case snr from @xcite has neither target - induced speckle nor turbulence contributions , so it has neither @xmath171 term nor any turbulence - induced noise - magnification factors . the reflective case s @xmath165 term has a factor of @xmath186 and its @xmath177 term has a factor of @xmath187 , but both @xmath188 contributions are absent from the transmissive case . these additional reflective - case contributions are direct results of the rough - surface scattering . in both transmissive and reflective operation there is averaging of the source randomness at the target , but in the reflective case the fields scatter incoherently off the target , and their fourier components are averaged at the bucket detector . this second averaging depends on the ratio of the source size to the bucket size , i.e. , @xmath189 . as @xmath189 increases , the bucket detector captures more light , this term becomes less significant , and we approach the transmissive behavior for these terms . finally , the reflective - snr terms associated with the bucket detector measurement in @xmath174 and @xmath177 are scaled by @xmath190 , relative to the corresponding terms in the transmissive case . this scale factor is the inverse of the solid angle subtended at the target by the bucket detector . its presence in the reflective case is a consequence of the quasi - lambertian scattering from the rough surface it follows from the snr differences highlighted above that the low - brightness snr asymptotes for transmissive and reflective operation only differ by the latter s including the solid - angle subtense factor @xmath107 . their high - brightness snr asymptotes are also similar , with the only difference being the reflective case s factor of @xmath186 . the most profound difference , however , is due to the @xmath171 term that is only present for reflective operation . this term implies that the snr of reflective ghost imaging has a maximum value , snr@xmath191 , whereas no such saturation occurs in transmissive ghost imaging , i.e. , its snr can grow without bound as @xmath192 increases . furthermore , in reflective imaging , this limiting effect can be quite severe : even with no turbulence , for @xmath193 we find @xmath194 , and for @xmath195 we have @xmath196 . so , for realistic standoff sensing , the snr will be limited to single - digit values if no further measures are taken to suppress the fluctuations arising from target - induced speckle and atmospheric turbulence . we have developed the performance of reflective ghost imaging from classical and nonclassical sources in a quantum gaussian - state framework . next we consider a computational ghost imager , a variant of the pseudothermal case in which the `` chaotic '' signal field is created by deterministically modulating the wavefront of a laser with a spatial light modulator ( slm ) , and no reference - arm light beam is required . instead , knowledge of the slm phase pattern is used to calculate the signal - beam intensity at the target for use in lieu of a reference - beam measurement @xcite . if an independent pseudorandom phase process is applied to each slm pixel , then far - field propagation can be argued to yield gaussian field statistics per the central limit theorem . thus it is appropriate for us to model the computational ghost imager s far - field statistics with a gaussian - schell model , taking the source s coherence length to be approximately the width of an slm pixel , and source s intensity diameter to be the diameter of the slm . the laser that illuminates the slm will be taken to emit coherent - state light , hence the slm s output beam is in classical state and semiclassical theory will give quantitatively identical results for its far - field photodetection statistics as the full quantum treatment @xcite . also , because the reference field at the target is computed , rather than measured , it is most naturally modeled as a complex - valued gaussian random process . consequently , and without loss of generality , we choose to use semiclassical theory for our analysis of computational ghost imaging because it lets us deal with only classical random processes , rather than a combination of quantum field operators and classical random processes . we thus use @xmath197 to represent the complex envelope of the target - plane signal field , and @xmath198 to represent the computed complex - envelope of the field that would have been present on the ccd camera . these two complex - valued , zero - mean , gaussian random processes are completely characterized by their nonzero autocorrelation and cross - correlation functions , which we take to be as given in eqs . for their field - operator pseudothermal counterparts . that these correlation functions are consistent with classical random - process theory follows from the pseudothermal source s producing classical - state light . the principal distinction between computational and pseudothermal ghost imaging which is also its main advantage is that there is neither detection noise nor turbulence on its computed reference - arm intensity , which is found via @xmath199 with @xmath200 . the bucket detector photocurrent , in semiclassical theory , is given by @xmath201 h(t-\tau ) , \end{aligned}\ ] ] where @xmath202 is the impinging photon flux , and conditioned on that photon flux the shot noise @xmath203 is a zero - mean random process with autocorrelation function @xmath204 . the computational ghost image is then obtained from the correlation function @xmath205 because the shot - noise is zero mean , given @xmath206 , it does not contribute to @xmath207 , hence the computational ghost imager s spatial resolution and image contrast analysis is identical to the sec . [ sec : rescon ] treatment of the pseudothermal ghost imager once we substitute the computational system s classical random processes for the pseudothermal imager s field operators , and then let @xmath208 , and set @xmath209 to account for the absence of turbulence on the computed reference . using this approach we immediately find the spatial resolution to be @xmath210 where @xmath211 and the image contrast is identical to that given for the pseudothermal imager in eq . . in the latter regard we note that we could replace @xmath212 in eq . ( [ cgi ] ) with the zero - mean process , @xmath213 , thereby achieving the same high - contrast operation as is realized with ac - coupling in pseudothermal ghost imaging . turning now to the computational ghost imager s signal - to - noise ratio , we find more significant differences between its analysis and that of the pseudothermal ghost imager . in our sec . [ sec : snr ] snr analysis , the second moment of the photocurrent cross - correlation function involved an eighth - order field moment , two sixth - order field moments , a fourth - order field - moment , three fourth - order moments of the turbulence , and a fourth - order moment of the target s field - reflection coefficient . for the computational ghost imager s snr analysis we only have to evaluate an eighth - order field moment , a sixth - order field moment , two fourth - order moments of the turbulence , and the fourth - order moment of the target s field - reflection coefficient . this simplification is because the other terms involved the path and detection noises on the reference arm , which are absent from the computational configuration . the computational snr is then given by @xmath214-e^{-4(\sigma_s^2+\sigma_t^2 ) } ) + \frac{{\mathcal{t}}({\bm\rho}_p)}{\eta\mathcal{i}}\frac { l^2}{a_b}}.\end{aligned}\ ] ] it follows that computational ghost imaging shares the same saturation snr as pseudothermal and spdc ghost imaging , from eq . ( [ eq : snrsat ] ) , with @xmath215 set to zero because there is no turbulence in the computational imager s reference arm . likewise , computational ghost imaging has the same high - brightness snr asymptote as pseudothermal ghost imaging , from eq . ( [ eq : snrc_h ] ) , with @xmath216 . on the other hand , computational ghost imaging has a significantly improved low - brightness snr asymptote , given by @xmath217 in comparison with _ both pseudothermal and spdc ghost imaging . _ for quantitative comparison between the snrs of the pseudothermal , spdc , and computational systems we will first have to put them on equal footing . the dimensionless brightness quantity @xmath122 that appears in our snr formulas is the photon flux per spatiotemporal mode of the source . however , because of the dramatically different coherence times of the spdc ( @xmath169 ) and the pseudothermal and slm sources ( @xmath167 ) , comparisons based on equal @xmath122 values will be for sources with dramatically different photon fluxes , whereas we would prefer to compare snrs at equal photon fluxes . to do so we introduce the dimensionless quantity @xmath218 , which is the photon flux per source spatial mode and detector temporal mode . because all three systems are taken to have detectors with identical characteristics , identical values for @xmath219 will yield identical photon fluxes . making this substitution we can easily compare the low - brightness snrs from eqs . , , and . in the extreme low - brightness regime , the computational imager has the best snr , followed by the spdc and pseudothermal imagers . we can also compare the high - brightness limits eqs . and , where the shorter coherence time of the spdc source gives it a clear advantage . because the different sources exit their low- and high - brightness snr regimes and enter into snr saturation for different photon fluxes , assessing that behavior requires us to explicitly choose some operating parameters . first , we assume the operating wavelength is @xmath220 m , the source coherence length is @xmath221 mm , the source intensity radius is @xmath222 cm , and that the bucket detector matches the source size , i.e. , @xmath223 . these give us the on - target parameters of @xmath224 m , and @xmath225 m . the photodetectors are taken to have @xmath226 quantum efficiency , with the ccd pixel area being @xmath227 . we assume the target is at @xmath228 km range , that it has an effective area of @xmath229 , and @xmath230 at the point for which we will perform our snr evaluations . all paths are assumed to have moderate - strength ( @xmath231 ) turbulence . finally , we take the pseudothermal and computational sources to have temporal coherence times satisfying @xmath232hz , while for the spdc source @xmath233hz . we consider two scenarios : first , when the source is weak enough ( @xmath234 ) that the spdc output is still a stream of biphotons , but we are not deep into the low - brightness regime ; and second , when the source is sufficiently strong ( @xmath235 ) to make the spdc output appear classical @xcite . the resulting snr curves for these two cases are shown in figs . [ fig : snr_l ] and [ fig : snr_h ] , respectively . from fig . [ fig : snr_l ] we see that the spdc imager is the worst performer when it is in its biphoton regime , while the computational ghost imager is to the top performer there . from fig . [ fig : snr_h ] , however , we see that when the spdc output appears to be classical , its much shorter coherence time allows it to reach the saturation - snr limit well before the other two systems , whose snr curves are nearly identical . . the source brightness is @xmath234 , so that the spdc output is a stream of biphotons and all three systems are well below their respective high - brightness snr asymptotes . other parameters used in computing these curves are given in the text.,width=336 ] . the source brightness is @xmath235 , so that the spdc output appears to be classical and the pseudothermal and computational ghost imagers are in their high - brightness limits wherein they only differ by a factor of @xmath236 . other parameters used in computing these curves are given in the text.,width=336 ] we have derived the key performance characteristics of reflective ghost imaging in the context of realistic standoff sensing applications . this was done within a gaussian - state framework that allowed for analysis of classical pseudothermal sources and the nonclassical output of an spdc . our results for the spatial resolution , image contrast , and snr were analogous to those calculated for transmissive ghost imaging in the absence of turbulence @xcite , with a few significant differences that result from the roughness of the target surface , turbulence on the optical paths , and the standoff nature of the measurement . first we shall discuss the impact of rough - surface scattering in the absence of turbulence , then we shall discuss the additional effects incurred when turbulence is present . rough - surfaced targets scatter their impinging illumination in a random manner , casting random speckle patterns in their far fields that , on average , correspond to a quasi - lambertian distribution . consequently , the ghost - imager s bucket detector will , on average , collect only a fraction , @xmath107 , of the target - scattered light . this angular subtense factor which is absent in transmissive ghost imaging appears in the average photocurrent cross - correlation functions in eqs . and , and the noise terms associated with the bucket detector , i.e. , @xmath174 and @xmath177 . the surface roughness has another major effect on the snr : its time - independent speckle pattern implies the existence of a finite saturation - snr value in the limit of long integration time . the target - speckle contribution that leads to a saturation snr appears in @xmath171 , and is a function of @xmath189 , the ratio of the source size to the bucket detector size . here we may incur a trade - off between spatial resolution and snr . specifically , by enlarging the bucket detector we can increase the averaging of the target speckle , and thus the saturation snr , without affecting the imager s spatial resolution . ultimately , however , there will be a practical limit beyond which the receiving aperture size can no longer be increased . beyond that point @xmath189 , and thus the saturation snr , can only be increased by decreasing the source size @xmath13 . however , as seen in eq . , spatial resolution is inversely proportional to @xmath13 , so increasing the saturation snr in this manner will degrade spatial resolution . the effects of turbulence on the image contrast and spatial resolution of reflective ghost imaging are similar those previously found for transmissive ghost imaging @xcite . in particular : turbulence does not change the image contrast ; turbulence in the target - to - bucket path has no effect on spatial resolution ; and turbulence on the signal and reference paths degrades spatial resolution in the same manner , i.e. , degradation occurs when they become smaller than the source size . in both spdc and pseudothermal ghost imaging it is therefore advantageous to propagate the reference field through a controlled ( turbulence - free ) environment . turbulence also degrades ghost - image snr in two ways : first , its logamplitude fluctuations magnify several existing noise terms ; and second , these fluctuations also contribute to the time - independent noise term @xmath171 , decreasing the saturation snr from its speckle - only value . in computational ghost imaging , the reference field is computed , removing the turbulence on that path , and thus improving both the spatial resolution and the snr . the computational case also has significantly reduced noise for low - brightness illumination , as there is no longer detection noise on the reference arm , making its low - brightness snr behavior superior to that of both pseudothermal and spdc ghost imaging . more generally , figs . [ fig : snr_l ] and [ fig : snr_h ] show that the computational ghost imager has a higher snr than the pseudothermal system in both low - brightness and high - brightness operation , and than the spdc system in its low - brightness ( biphoton ) regime . however , in high - brightness operation , the spdc ghost imager s significantly shorter coherence time enables it to reach its saturation snr much quicker than either the computational or the pseudothermal ghost imagers . although a cw spdc will not be able to reach that high - brightness regime , pulsed systems are capable of doing so @xcite . insofar as standoff sensing is concerned , the essential conclusion to be drawn from the preceding summary of our work is simple : the computational approach is the ghost - imaging configuration that shows the most promise for this application . it has the best spatial resolution , the highest saturation - snr , and general snr performance second only to high - brightness spdc operation . in addition , the computational approach obviates the need for a physical reference path , which makes it naturally amenable to obtaining 3d ghost images . in particular , pseudothermal and spdc ghost imaging require that reference - arm measurements be made on an intensity pattern corresponding to the one that is projected onto the target . as shown theoretically in @xcite and experimentally in @xcite , computational ghost imaging allows reference intensity patterns to be computed at a variety of target ranges so that ghost images can be formed for these target ranges from the same bucket - detector data . this range sectioning is something that _ can not be done , without separate measurements for each possible target range , in pseudothermal or spdc ghost imaging . _ we have reserved our final comment for our choice of image reconstruction via photocurrent cross - correlation . it has already been demonstrated @xcite that ghost imaging reconstruction can be performed in transmission through more advanced methods , such as compressive sensing . a preliminary no - turbulence performance comparison between a simple cw - laser radar and ghost imaging in reflection via cross correlation has shown their similarity in spatial resolution and snr @xcite . if advanced image - reconstruction techniques can be successfully applied to computational ghost imaging , the resulting system could have notable advantages over current laser radars . this work was supported by the u.s . army research office multidisciplinary university research initiative grant no . w911nf-05 - 1 - 0197 , the w. m. keck foundation center for extreme quantum information theory , the darpa information in a photon program under u.s . army research office grant no . w911nf-10 - 1 - 0404 , and the darpa quantum sensors program under afrl contract no . fa8750 - 09-c-0194 , for the sake of generality , we have allowed there to be turbulence in the reference path , although a lens could be used to cast a minified reference - beam pattern onto a ccd camera within a protected environment near the source , hence avoiding turbulence on that path . that no additional quantum - noise term is needed to ensure that this expression for @xmath237 preserves free - field commutator brackets follows from the normal - mode decomposition for propagation through turbulence [ j. h. shapiro , appl * 13 , * 2614 ( 1974 ) ] . a similar square - law approximation is available for cases in which the turbulence strength is not uniformly distributed along each propagation path , but we have opted to limit our attention to uniform distributions . in quantum photodetection , the currents at the outputs of the bucket and ccd detectors are classical stochastic processes , @xmath238 and @xmath239 , whose statistics are equivalent to those of the quantum measurements @xmath58 and @xmath55 given in the next paragraph . we say `` appear classical '' because the phase - sensitive cross - correlation function from eq . ( [ eq : qcorrps ] ) _ always exceeds the classical limit set by the autocorrelation functions from eq . ( [ eq : autocorr ] ) . however , for @xmath123 , that cross - correlation function is so close in value to the classical limit , given in eq . ( [ eq : ccorrpis ] ) , to render this measure of nonclassicality exceedingly small . _

we present 12co ( 21 ) data of three virgo spirals ngc 4330 , ngc 4402 and ngc 4522 obtained using the submillimeter array . these three galaxies show clear evidence of ram pressure stripping due to the cluster medium as found in previous hi imaging studies . using high - resolution co data , we investigate how the properties of the inner molecular gas disc change while a galaxy is undergoing hi stripping in the cluster . at given sensitivity limits , we do not find any clear signs of molecular gas stripping . however , both its morphology and kinematics appear to be quite disturbed as those of hi . morphological peculiarities present in the molecular and atomic gas are closely related with each other , suggesting that molecular gas can be also affected by strong icm pressure even if it is not stripped . co is found to be modestly enhanced along the upstream sides in these galaxies , which may change the local star formation activity in the disc . indeed , the distribution of h emission , a tracer of recent star formation , well coincides with that of the molecular gas , revealing enhancements near the local co peak or along the co compression . fuv and h share some properties in common , but fuv is always more extended than co / h in the three galaxies , implying that the star - forming disc is rapidly shrinking as the molecular gas properties have changed . we discuss how icm pressure affects dense molecular gas and hence star formation properties while diffuse atomic gas is being removed from a galaxy . [ firstpage ] ism : molecules galaxies : clusters : intracluster medium galaxies : evolution galaxies : ism galaxies : spiral .

1701.02750

since @xcite suggested that a galaxy might lose its interstellar medium ( ism ) by interacting with the intracluster medium ( icm ) in the cluster environment , much evidence for ram pressure stripping has been found to date . in early days , @xcite have shown that virgo galaxies are generally more deficient in hi compared to field galaxies based on their single - dish observations . in addition , @xcite have found that the hi content is well correlated with the location of galaxies in a sense that hi is more deficient at smaller distances from the cluster centre . then the hi synthesis imaging studies such as @xcite and @xcite have shown that hi - deficient galaxies near the cluster centre have small hi - to - optical extents , truncated within the stellar disc in many cases . in a more recent high - resolution hi imaging study of @xmath150 selected virgo galaxies by @xcite , a number of galaxies have been found with various scales of extraplanar hi gas or long hi tails , indicating that ram pressure stripping is indeed acting in the cluster environment . as a galaxy loses its ism , the star formation rate is expected to be suppressed , which agrees well with the observations . as a good example , @xcite find that the massive star formation rate of virgo spirals is lower than their field counterpart by a factor of 2.5 on average . @xcite also show that many virgo spirals have a smaller h@xmath0 extent compared to their stellar disc , reflecting that the process truncating hi discs may also be acting on the star - forming disc and h@xmath0 emission . on the other hand , molecular gas is unlikely to be as easily stripped as atomic hydrogen , since it is more tightly bound to the galactic centre and the density is higher . in fact , most previous studies find that the molecular gas mass of the cluster population is not significantly different from that of field galaxies @xcite . in addition , more recent studies are finding clumpy dust features in the upstream side of hi gas stripped galaxies , which are likely to be surviving dense clouds that are unveiled after diffuse atomic gas is removed ( @xcite ; kenney , abramson & bravo - alfaro 2015 ) . on the other hand , however , the opposite results also have been reported . @xcite find that h@xmath2 mass normalized by the dynamical mass of galaxies tends to get larger with increasing clustercentric distance , which supports that molecular gas can be also deficient in the cluster environment . more recently , @xcite show that hi - deficient galaxies in the cluster environment tend to be also modestly deficient in molecular gas . however , the molecular gas fraction to the optical luminosity of spiral galaxies measured using 12co ( 1@xmath30 ) ranges quite widely for all environments @xcite . this implies that the molecular gas fraction alone may not be a good tracer of molecular gas deficiency . due to the scatter in this relation and opposing observational results , it is still arguable whether molecular gas can be affected by the icm pressure in a similar way to hi gas , and it is then still puzzling why star formation appears to be quenched in hi stripped galaxies if molecular gas , which is the more direct ingredient for star formation , is not deficient . of the sample in the _ b _ band . ngc 4569 that we have not observed using the sma but is included in our discussion as a good representative of a galaxy at post - peak pressure , is shown in green . the sample galaxies , located at 0.41 mpc from m87 , adopting a virgo distance of 16 mpc @xcite , make a nice sequence of ram pressure stripping from early stage , close to peak pressure and post - peak pressure in their orbit . ] no clear evidence for molecular gas stripping yet low star formation activities in cluster galaxies may imply distinct molecular gas properties in high - density environments as supported by some previous observations . for example , @xcite have found a highly asymmetric co morphology in an hi deficient virgo cluster galaxy . @xcite also have found very peculiar co distributions in some of virgo spirals undergoing active hi stripping . these studies indicate that molecular gas can be potentially disturbed by strong icm pressure whether it is stripped or not . particularly along the side experiencing icm pressure , it has been suggested that interstellar gas including the molecular phase can be pushed up against the centre of a galaxy . this leads to ism compression and increasing h@xmath2 formation ; hence , it locally triggers intensive star formation @xcite . indeed , ism compression with high molecular fraction in the upstream side is seen in a number of galaxies experiencing icm pressure ( @xcite ; nehlig , vollmer & braine 2016 ) , which are often accompanied by an enhancement of star formation . molecular gas enhancement among the virgo cluster members has been also recently reported by @xcite . these observations clearly show that molecular gas properties and hence star formation activity within a stellar disc can be affected by ram pressure . therefore , in order to get a deeper understanding on how galaxies become passive after hi stripping , not only the possibility of stripping but also the detailed properties of molecular gas inside a galaxy under strong icm pressure need to be probed . hence , using the submillimeter array ( sma ) , we have taken high - resolution 12co ( 2@xmath31 ) imaging data of a subsample of virgo galaxies that have lost atomic gas significantly by icm pressure . some preliminary results are published in @xcite , and in this work , we present more complete analysis on both morphology and kinematics of three virgo galaxies based on the sma data . this paper is organized as follows . we introduce the sample in section [ sec2 ] . details of observations and data reduction procedure are provided in section [ sec3 ] . in section [ sec4 ] , we present the sma data , describing the co morphology and kinematics . in section [ sec5 ] , we compare the co and other wavelength data to discuss how the molecular gas in these galaxies has been affected by the icm pressure . in section [ sec6 ] , we summarize the results and conclude . a distance of 16 mpc ( 1 arcsec @xmath178 pc ) to virgo cluster is adopted in this work ( yasuda , fukugita & okamura 1997 ) . [ cols="<,^,^,^,^",options="header " , ] [ tab3 ] _ note . _ @xmath4the co - to - h@xmath5 conversion factor of 3.2 @xmath6 pc@xmath7 ( k km s@xmath8)@xmath8 is adopted from @xcite . + in this section , we present the results of our sma observations . the 12co ( 2@xmath31 ) flux and its uncertainty are measured in jy km s@xmath8 as follows , @xmath9 where @xmath10 is the total flux of co emission in each channel , @xmath11 is the channel separation of the final cube ( 5 km s@xmath8 ) , and @xmath12 is the rms of each channel outside co emission . continuous features above @xmath13 are considered as real signal in individual channels . co linewidths are calculated using the velocities , where the flux density corresponds to 20 per cent and 50 per cent of the peaks in the receding and the approaching side , adopting the definition of @xcite : @xmath14 @xmath15 the co velocity is determined following @xcite : @xmath16 following the recipe from @xcite , 12co ( 2@xmath31 ) line luminosity is measured as follows : @xmath17 in @xmath18 , where @xmath19 is an integrated total 12co ( 2@xmath31 ) flux in jy km s@xmath8 , @xmath20 is the luminosity distance in mpc , @xmath21 is the observation frequency in ghz , and _ z _ is the redshift . adopting a typical 12co ( 2@xmath31)/(1@xmath30 ) ratio of nearby galaxies ( @[email protected] ; @xcite ) and using the conversion relation of normal spirals ( @xmath24 = 3.2 @xmath6 pc@xmath7 ( k km s@xmath8)@xmath8 ; @xcite ) , the molecular gas mass can be calculated by @xmath25 where @xmath26(2@xmath31 ) is the luminosity of 12co ( 2@xmath31 ) transition . the global co properties of the sample are summarized in table [ tab3 ] . on the left of figs [ n4330][n4522 ] , the co intensity map , velocity field and position - velocity diagram ( pvd ) with global and radial profile are presented in ( a ) , ( b ) and ( c ) for each galaxy . the overlays with other wavelength data shown on the right - hand side of figs [ n4330][n4522 ] are discussed in section [ sec5 ] . as seen in figs [ n4330](a)(c ) , the co morphology of ngc 4330 is highly asymmetric , with the south - west extent in the downstream side being about 75 per cent ( 40 arcsec versus 53 arcsec ) of the northeast but only 68 per cent when the outermost 4 arcsec of the south - west disc , which is bent down to the south ( indicated by arrow in fig . [ n4330]a ) is excluded ( 36 arcsec versus 53 arcsec ) . the central part , within 30 arcsec in diameter ( @xmath12 kpc at the distance of virgo ) , is also asymmetric due to the strong local peak in the north - east from the stellar disc centre . the distinct extents and surface densities are also clearly seen in radial profiles . in the end of the southwest disc of the downstream side , co is found to be slightly bent ( fig . [ n4330]a ) . this bending is also found in many other wavelengths such as uv , h@xmath0 and hi , but all are different in scale and angle from one another as further discussed in section [ sec5 ] . in the case of co , this bending part is clumpy , almost identified as an independent blob or clump ( see the arrow in fig . [ n4330]c ) . this co clump corresponds to 2 per cent of the total in flux ( @xmath11.9@xmath2710@xmath28 @xmath6 , comparable to that of a large molecular cloud ) . the centre of the clump is off from the mid - plane of the main disc toward the south by @xmath14 arcsec ( @xmath1312 pc ) . the co kinematics also shows peculiar structures , especially along the south - west concentration and the end of the tail as shown in fig . [ n4330](b ) . the velocity gradient of the south - west is quite steep , while it is more slowly rising on the other side within the small radii from the centre , reaching 72 km s@xmath8 on the approaching side but only 43 km s@xmath8 on the receding at 15 arcsec radii . in the south - west clump , the velocity gradient is inverted . the kinematical complexity is also clearly seen in the pvd ( fig . [ n4330]c ) . a distinct component of a steep velocity gradient in the central @xmath113 arcsec ( @xmath11 kpc ) is quite noticeable . this may indicate a molecular ring or bar . in addition , the co peak is off from the stellar disc centre by @xmath1270 pc ( fig . [ n4330]a ) , while the co kinematic centre is more or less consistent with the stellar disc centre ( fig . [ n4330]c ) . as shown in fig . [ n4402](a ) , the co disc of ngc 4402 is slightly more extended in the west ( 62 arcsec in the west versus 58 arcsec in the east ) , but the difference is subtle and not as significant as in hi . what makes this case look highly asymmetric is the north - west quarter of the downstream side ( the west side of the disc ) . in this region , a modest co bump ( indicated by arrow in fig . [ n4402]a ) is found as in hi ( fig . [ n4402]d ) . meanwhile , the southern part of the co disc looks quite compressed along the enhanced far - ultraviolet ( fuv ) emission ( fig . [ n4402]e ) , which will be discussed more in detail in section [ sec5 ] . the northern part above the major axis is measured to be 91 per cent of that of the south in flux . although the co peak agrees well with the stellar disc centre , the inner co disc also appears to be slightly more stretched toward the west , reflecting the outer co disc . the co kinematics reveals the evidence for non - circular motions as shown in fig . [ n4402](b ) . in the inner region , the iso - velocity curves are highly skewed and not perfectly parallel with the minor axis . the skewness in velocity must be the result of two distinct disc components in the inner @xmath115 arcsec ( @xmath11 kpc ) being projected on the sky as seen in its pvd ( fig . [ n4402]c ) . the velocity gradient in the outer part is quite different in the two sides of the disc along the major axis . in the north - west quarter ( in the downstream side ) , the kinematics of modest co bump is slightly deviated from the main disc . it also has a distinct co component with steep velocity gradient in the inner @xmath115 arcsec . this likely indicates a nuclear bar or ring , although we favor a nuclear bar as this would also explain the skewed isovelocity contours . in ngc 4402 , @xcite also find a nuclear molecular disc traced by 12co ( 1@xmath30 ) in the inner @xmath110 arcsec . we do not find any direct relevance of these nuclear structures with ram pressure stripping . while the overall size of 12co ( 2@xmath31 ) disc in the sma image is consistent with the extent measured by a single dish for ngc 4330 @xcite and ngc 4402 ( lee et al . in preparation ) , our sma data of ngc 4522 reveal only the inner @xmath157 per cent of its single - dish map in size @xcite . while @xcite have detected co along the extraplanar h@xmath0 and hi in both ends of the disc , we did not detect any such features in the south - west in our sma data due to the lack of sensitivity and/or potentially due to the diffuse nature of gas in the outer molecular disc . the north - east end was not covered in the sma observations . therefore , we limit our discussion to only the inner co disc for this particular case in this section and we cite the iram single - dish data when it is needed for comparisons with other wavelength data in section [ sec5 ] . as shown in fig . [ n4522 ] , the co extent is comparable in both sides ( 27.4 arcsec in the north - east versus 27.6 arcsec in the south - west ) . the inner co disc of ngc 4522 is found to be curved in the opposite way to the outer co and hi disc , i.e. to the south - east as the inner h@xmath0 disc . we find two features sticking out from the inner part , one from the end of north - east disc and another almost to the same direction but connected from the centre of the main disc , more to the south . the north - east blob coincides with the morphology of the inner co disc in the iram image ( fig . [ multi]c ; @xcite ) , which is smoothly connected to the disc farther extended in the north - east , coinciding with a dust loop in this region . while the downward bending coincides with the inner spiral structure including the dust feature ( see fig . [ hst ] in appendix a ) . the co velocity structure within the main disc generally shows that of a regularly rotating disc as seen in figs [ n4522](b ) and ( c ) . the velocity keeps rising up to @xmath110 arcsec on both sides , then almost flattens out , and toward the end of the disc , the velocity rises again . the velocity gradient along the north - east branch is overall consistent with that of the main disc on the same side . however , the gradient is smaller and the velocity rises more slowly compared to the other side . in this section , we compare 12co ( 2@xmath31 ) properties with other wavelength data including hi @xcite , optical ( dss2 red ) , fuv @xcite , and h@xmath0 ( ngc 4330 and ngc 4569 from @xcite , ngc 4402 from @xcite , ngc 4522 from koopmann , kenney , & young 2001 ) . co traces star - forming gas , and h@xmath0 and fuv emission are good indicators of star formation with time - scales of @xmath120 and 100 myr , respectively @xcite . we also compare the radial gas surface density between atomic and molecular hydrogen . the hi column density is calculated using , @xmath29 in cm@xmath7 , where @xmath31 is the integrated intensity ( brightness temperature ) in @xmath32 @xcite . the h@xmath2 surface density is estimated by the following relation : @xmath33 in @xmath6 pc@xmath7 , where @xmath34 is 12co ( 2@xmath31 ) integrated intensity , and @xmath35 is the same as in equation ( 6 ) . considering the high inclination of all three galaxies in our sample , we utilize the strip integral method instead of ellipse fitting to derive radial surface density because ellipse fitting is usually inappropriate for edge - on galaxies when the spatial resolution is limited @xcite . this method calculates _ strip integrals _ by integrating hi column density perpendicularly to the major axis , using lucy s ( 1974 ) iterative deconvolution method . then the strip integrals of approaching and receding sides are deprojected separately to infer the face - on hi surface density @xcite . the surface density of both atomic and molecular hydrogen is measured in @xmath6 pc@xmath7 . the comparison between atomic and molecular gas radial distributions is found in the bottom on the left - hand column of figs [ n4330][n4522 ] . overlays between co and other multiwavelength data are presented on the right - hand column of figs [ n4330][n4522 ] and fig . [ multi ] . for the reference , the iram 12co ( 2@xmath31 ) maps are also overlaid in the composite map . the sensitivity of the iram cubes is @xmath110 mk per 5 km s@xmath8 channel ( the same channel width as ours ) , which is comparable to the sma cubes or slightly better in the cases of ngc 4330 and ngc 4522 when the different spatial resolutions are taken into account ( @xmath111 arcsec versus @xmath15 arcsec for iram and sma , respectively ) . for ngc 4402 , the sensitivity of sma data is somewhat better than the iram data not only because of the sufficient integration time with the sma but also because of the poor weather condition when the iram data were taken . for ngc 4330 and ngc 4402 , the iram and sma maps are generally in good agreement in morphology , considering the beam sizes , while we are missing most of the extraplanar molecular of ngc 4522 in the sma data due to the lack in coverage , sensitivity , and/or potentially due to the diffuse nature of molecular gas in the extragalactic space as further discussed in section 5.1.3 . asymmetry is ubiquitous in a range of wavelengths data except in the red optical image ( dss2 red ) . in fig . [ multi](a ) , we see that the hi is truncated within the stellar disc in the north - east , where h@xmath0 and fuv emission reveals an upturn feature @xcite . a hint of upturn is also found in the north - east end of the upstream of radio continuum at 6 cm and 20 cm @xcite . the single - dish 12co ( 2@xmath31 ) data @xcite appear to be similar to the radio morphology , also with some hint of upturn , although the extent of the molecular gas is much less than that of hi . in our sma data , we are missing some outermost features including the tip of the nose in the north - east . instead , the sma data clearly reveal very detailed structures of the inner co disc such as the difference in scaleheight between the north and the south along the major axis ( up to @xmath10.36 kpc , fig . [ n4330]a ) which is not as clear as this in the previous single - dish data . the south - west side , i.e. the downstream of the galaxy where an hi gas tail is present , is bent toward the south . on this side , we also find fuv , h@xmath0 and radio continuum tail but the extents and bending angles are all different in various wavelengths ( fig [ multi]a ) . the distribution becomes patchy in the outer co disc as seen in both iram single - dish data @xcite as well as our sma data . intriguingly , the co blob toward the end is located in almost the same galactic radius as one of the distinct h@xmath0 blobs and off from the mid - plane in the same direction as the tails of the other wavelengths ( fig [ multi]a ) . as clearly seen , the molecular disc , which extends to only half the stellar disc or less , shows many properties in common with the morphologies in the other wavelengths , reflecting the impact of the icm pressure . although there are subtle differences , this suggests that the molecular gas in this galaxy has been also affected in similar ways by the same mechanism that is responsible for the peculiarities found in the other wavelengths . this indicates that icm pressure can change molecular gas properties well inside the stellar disc . the co emission shows a number of properties in common with the hi gas ( fig . [ n4402]d ) . it is somewhat extended in the west , i.e. the same side where the hi is more extended . both phases show a bump in the west ( indicated by arrow in fig . [ n4402]a ) , although the scale is distinct because the hi bump is visible in extraplanar gas . hi is truncated within the mid - plane of the stellar disc on both sides , with a sharp cut - off in the east to south along the upstream . meanwhile , the northern side is more extended with a short tail pointing the north - west in the downstream side . the extent of the molecular gas is not significantly different in the receding and the approaching side , yet its distribution is quite distinct . these differences between the two sides of the co disc can be also clearly seen in the radial velocity distributions ( fig . [ n4402]c ) . note that the hi extent appears to be almost twice as long as the co in the west due to the tail ( figs [ n4402]c and d ) . the molecular gas disc is thicker in the east ( especially third contour level from the outermost contour ) , reflecting the change in fuv disc thickness across the disc . although the overall morphology of fuv is quite different from that of co , both the fuv enhancement and co compression are present in the east and south - east ( fig . [ n4402]e ) . h@xmath0 is tightly correlated with co in general , with a similar distribution and extent . also , strong h@xmath0 knots are found near or within the local co peaks ( fig [ n4402]f ) . as with ngc 4330 , the co morphology is surprisingly similar to that of hi , strongly suggesting that the impact of the icm has reached the inner interstellar gas disc . with given sensitivity limit , we are missing the outer part of the molecular gas disc in the sma data for this case . as shown by @xcite , however , the overall morphology of co is quite similar to that of hi , revealing a significant fraction of molecular gas outside the stellar disc ( fig [ multi]c ) . this galaxy must be one of the best examples found to date where extraplanar molecular gas is present . in addition to the sensitivity , the molecular gas could be more diffuse in nature compared to the gas within the disc , which could make the detection more difficult . in fact , we recently have detected some extragalactic 13co ( 1@xmath30 ) in ngc 4522 using the atacama large millimeter array ( alma ) - cycle 3 . the nature of this extraplanar co gas will be further analyzed using 12/13co ratio , and will be presented in a separate work . the inner co disc is slightly bent down to the south as h@xmath0 emission . the strongest h@xmath0 blobs near the galactic centre are located within the inner co disc ( fig [ n4522]f ) , but there is an offset between the inner fuv enhancement and the inner co disc ( fig [ n4522]e ) . however , this offset in the inner region might be simply due to the dust extinction effect . on the other hand , the offset between co ( the iram data)/h@xmath0 and fuv at large radii is real since there is almost no dust beyond the co extent . the extent of fuv is larger than the co and h@xmath0 extents within the stellar disc by a factor of 1.5 ( fig . [ multi]c ) . however , the fuv enhancement is roughly consistent with the co compression in the downstream side as in ngc 4402 . this implies that there is significant change in the recent star formation activity and the distribution of molecular gas in a very short time as further discussed in section [ sec5].3 . most previous studies using single - dish radio telescopes did not find a significant difference in co luminosities between field galaxies and cluster members @xcite . however , some individual galaxies in the cluster environment have been reported to show very distinct co distributions from those with normal hi content in the field ( e.g. @xcite ) . more recently , @xcite have shown that the molecular gas surface density of galaxies that are hi - deficient tends to be low . these suggest that the molecular gas of cluster galaxies can be affected in some ways by ram pressure even if it is not stripped . indeed , our high - resolution co data clearly show that molecular gas has been affected in a similar manner as diffuse gas by ram pressure , and the impact of icm winds reaches as deeply as a few kpc from the galactic centre , as we further elaborate below . firstly , the co morphology in our sample is all found to be highly asymmetric and disturbed . the inner molecular gas disc of many field galaxies does show some peculiarities to a certain degree , including asymmetry ( e.g. bima song survey ; @xcite ) . however , the overall asymmetry found in the co morphology of all three galaxies is similar to that of the hi morphology . in figs [ multi](a)(c ) , the composite maps of cool ism contents ( co and hi ) , star formation indicators ( fuv and h@xmath0 ) and the old stellar disc ( dss2 red ) of our sma sample are presented . a rough direction of the icm wind based on the hi morphology for each case is shown with thick arrows . secondly , the detailed co structure of the sample shows good connections with the hi features which have been pushed and/or stripped by the icm pressure . in ngc 4330 , the co clump in the south - west end seems to have been pushed down to the south from the mid - plane of the stellar disc , similarly to its hi tail ( fig . [ n4330]d ) . in addition , similar kinematical structures are found in this side , i.e. in the downstream in both hi and co ( fig . [ pvd ] ) . for example , the co gas reveals an inverted velocity gradient in the end of south - west like hi which shows an anomalous velocity component in the same side @xcite . this implies that dense molecular phase gas can be pushed by ram pressure in addition to the diffuse atomic phase gas . although the morphology is suggestive of gas being pushed away , one problem with this possibility is that both the hi tail and the co clump in the downstream are found with higher velocities than the main disc , i.e. the opposite direction of the system velocity of cluster centre . this is inconsistent with what is generally expected in the case of the gas pushed by the icm as found in other virgo spirals ( e.g. ngc 4522 ; kenney , van gorkom & vollmer 2004 ) . however , this is still feasible if the gas has been losing angular momentum in extraplanar space as suggested by @xcite . alternatively , the co clump might be an example of a dense cloud unveiled after the stripping of surrounding diffuse gas , which hence appears to stand alone , decoupled from the rest of the disc , similar to the ones seen as narrow dusty plumes found in virgo galaxies such as ngc 4402 and ngc 4522 @xcite or ngc 4921 , a coma spiral @xcite . if this is an example of those decoupled clouds , it is the first time it has been seen clearly in co. however , one aspect that makes this case particularly interesting is its gas kinematics . rather than being smoothly connected from the inner disc , the velocity gradient at the location of the co clump turns back . this velocity structure strongly suggests that this clump is the molecular gas that has been pushed off from the main disc . in ngc 4402 , both hi and co are compressed from the south - east to south along the upstream . meanwhile in the north - west , a modest co bump is found to be pointing to the same direction as the hi bump ( fig . [ n4402]d ) . the compressed co in the upstream side reveals a steeper velocity gradient than the other side . meanwhile , the co bump in the north - west downstream side ( fig . [ n4402]b ) is found to have larger velocities than the main disc ( fig . [ n4402]c ) , i.e. closer to the cluster centre . this is consistent with our expectations if the gas is accelerated by ram pressure due to the icm . in ngc 4522 , the overall co morphology from the iram data @xcite is very similar to the hi morphology , with the extraplanar gas in the downstream side and compression in the upstream side ( fig . [ multi]c ) . this strongly suggests that the molecular gas has been influenced by the same mechanism that affects the atomic gas , i.e. ram pressure due to the icm wind . in the upstream side , the interstellar gas can also get compressed by the icm pressure , which potentially results in the formation of molecules as inferred by the excess of co in this side ( see also the example of ngc 4654 ; @xcite ) . as shown in fig . [ pvd ] , the co kinematics also suggests that the icm pressure can disturb the molecular gas , deep inside a galactic potential well . in ngc 4330 and ngc 4402 , the co velocity gradients are quite distinct in the receding and the approaching side , an asymmetry also observed in hi . in the outer part of ngc 4330 and ngc 4402 ( the co clump and the modest co bump , respectively ) where the molecular gas is off from the mid - plane , we also find the velocity gradient deviates from the overall gas flow of the gas disc . the comparisons with the hi pvd is also intriguing ( the bottom row of fig . [ pvd ] ) . in the upstream of ngc 4330 and ngc 4402 , the velocity gradients of the molecular gas and atomic gas are similar to each other , but they deviate in the downstream . in ngc 4522 , the deviation starts at @xmath10.8 kpc from the centre on both sides . this may indicate a delay in stripping of gas in distinct phases , i.e. more diffuse gas dragged prior to relatively dense gas . our results clearly show that the molecular gas in galaxies experiencing active hi stripping can become highly asymmetric , as demonstrated by simulations ( e.g. @xcite ) . in addition , our high - resolution data indicate that the influence of strong icm pressure reaches quite deep inside of the galactic disc . although molecular gas may be pushed off from the stellar disc together with atomic gas as in ngc 4522 @xcite , active ram pressure does not seem to always cause molecular gas stripping ( ngc 4330 and ngc 4402 ) . indeed , ngc 4569 , which is thought to have lost more than 90 per cent of its hi gas while crossing the cluster core a few hundred myr ago ( hi deficiency : 1.47 , a projected distance of 1.7 degree ; @xmath10.5 mpc to m87 , the bottom right of fig . [ multi ] ) , has a comparable molecular gas fraction to optical luminosity of our sma sample ( ngc 4330 , ngc 4402 and ngc 4522 ) . adopting the h@xmath2 deficiency of @xcite , single - dish measured co fluxes and the mw conversion factor result in @xmath36 of 0.48 , 1.25 , 0.97 and 1.19 for ngc 4330 , ngc 4402 , ngc 4522 and ngc 4569 , respectively . considering that the sample of @xcite does not include galaxies as faint as ngc 4330 or ngc 4522 , the expected h@xmath2 mass could have been overestimated for these two galaxies . this supports that molecular gas mass does not necessarily change significantly during and after severe hi stripping , while its detailed molecular gas properties may change . these changes to the co gas properties are consistent with those from @xcite , who found enhanced levels of warm molecular gas which resulted from additional ism - heating due to ram pressure driven shocks in ngc 4330 , ngc 4402 and ngc 4522 . all three galaxies have lost hi gas significantly , containing only @xmath37 @xmath120 per cent of hi mass compared to field galaxies of similar size . the star formation quenching time - scale , @xmath38 , measured based on the youngest stellar age at the location , where hi gas is truncated ( i.e. one way to estimate how long ago star formation was suppressed in a galaxy ) , ranges from a few tens myr to a few hundreds myr for our sample @xcite . it is however comparable among the sample within the uncertainties , and they share several properties in common in star formation activity in spite of subtle differences due to the icm wind angle and the projection on the sky as shown in fig . [ multi ] . the spatial distributions of co and h@xmath0 ( a tracer of recent star formation with a timescale of @xmath3920 myr ; @xcite ) are remarkably well correlated in the inner few kpc radius ( fig . [ multi ] ) . the locations of local co peaks are generally coincident with the h@xmath0 knots as shown by high - resolution data . the extents and overall morphology are also similar to each other as seen in the comparison with single - dish data , although the h@xmath0 blobs outside the old stellar disc do not always have a co counterpart ( e.g. in both ends of ngc 4330 s fuv disc ; fig . [ multi]a ) . meanwhile , fuv ( a star formation tracer of a time - scale of @xmath1100 myr ; @xcite ) is somewhat distinct from co and h@xmath0 in extent and morphology , at least on one side , with fuv being more extended than co and h@xmath0 . although the fuv disc of many field galaxies are larger than their h@xmath0 disc ( e.g. @xcite ) , these three galaxies are distinct from the field population in a sense that both fuv and h@xmath0 are highly truncated within the stellar disc . these observations indicate that star - forming discs have been shrunk considerably ( by a factor of up to 1.5 ) while undergoing strong ram pressure . then the question is what causes the quenching of star formation , whether it is molecular gas stripping , the change of molecular gas properties , or both . as seen by the morphological / kinematical peculiarities of molecular gas in our sample , the icm pressure is strong enough to strip the hi gas in the outer disc or push the hi disc to one side , and seems to be able to also affect the molecular gas at a few kpc radii of the stellar disc . this can cause a number of consequences . first , in the outer disc where stars must have been formed mostly from low density interstellar gas , star formation activity will gradually decrease as the atomic gas is stripped . even when quenching starts , some young stars from the last star formation event may be still observable , and indeed , the fuv emission outside co and h@xmath0 discs yet inside the old stellar disc could be good examples of this process . our sma data suggest that these galaxies are close to or approaching peak pressure @xcite , which is also supported by the results of simulations in the cases of ngc 4330 and ngc 4522 @xcite . therefore , they will experience strong(er ) ram pressure for a while , losing more hi . as long as hi is supplied into the disc , the galaxy should be capable of forming molecular gas , keeping up with star formation . hence , within one galaxy where star formation can be regulated , the fuv morphology should be more or less in good agreement with those of h@xmath0 , which traces more recent star formation on a time - scale of @xmath37 20 myr @xcite . this is indeed true for isolated galaxies such as the sample of @xcite . in addition , the fuv extent in isolated galaxies is usually smaller than the hi extent as shown in the comparison of radial profiles @xcite . in our galaxies , however , the h@xmath0 looks much smaller than the fuv disc , and quite distinct in morphology . this implies that star - forming disc in these galaxies has been shrunken at least in the last 100 myr . while the outer disc undergoes star formation truncation , the inner disc is also less likely to form stars in the same manner as field galaxies , despite enough molecular gas present . in the upstream side , molecular gas is found to be compressed as if it is pushed by the icm pressure . this can be the result of direct molecular gas compression but also molecular gas excess due to formation from compressed diffuse gas @xcite . consequently , the star formation rate is likely to be locally increased as @xcite suggested ( see also @xcite ) . indeed , we find strong fuv enhancement on the same side where the outer co disc is found to be compressed . alternatively , this might be the result of the removal of dust due to the icm pressure and hence the upstream side being unveiled . however , the fuv disc also shows morphological peculiarities and it is quite distinct from the old stellar disc ( e.g. a tail in ngc 4330 , different extents and scale heights of ngc 4402 , extraplanar fuv emission in ngc 4522 ) , which can not be explained by dust stripping alone . in addition , more evidence for increased star formation due to gas compression by the icm pressure has been found in other galaxies ( e.g. @xcite ) , which also seems to be the case in our sample . local fuv enhancements found near co compression within the old stellar disc ( such as the north - east of ngc 4330 , the east of ngc 4402 and the south - east of ngc 4522 ) can be good examples of induced star formation by ram pressure ( figs [ n4330]e[n4522]e and figs [ multi]a c ) . some interstellar gas removed from the main stellar disc yet lingering around in a galactic halo may be re - accreted onto the disc after core crossing . ngc 4569 shown in fig . [ multi](d ) , is known to be such case . the single - arm structure in hi agrees well in morphology and kinematics with a model of stripped gas falling back to the galaxy , which assumes that the galaxy is currently moving away from the cluster centre after crossing the core @xmath1a few hundreds myr ago @xcite . its star formation quenching time - scale estimated by @xcite is also somewhat longer than our sma sample by @xmath1100 myr or so , indicating that this galaxy has experienced strong icm pressure relatively longer ago than the other three galaxies . what is intriguing in the case of ngc 4569 is that the extent of the fuv disc is comparable to that of the h@xmath0 and co ( fig . [ multi]d ) . in this case , the star formation is limited in the central region of the galaxy as most of the hi gas at larger galactic radii is stripped . some gas can be re - obtained but not enough gas is left to replenish the entire disc . therefore , the fuv extent can decrease to a similar size as the h@xmath0 extent because fuv emission may fade as time goes by , if molecular gas can not be replenished because the atomic gas has been stripped . at this stage , it is similar to isolated galaxies in a sense that fuv is overall well correlated with h@xmath0/co , but now the star - forming disc is somewhat shrunken in comparison to what it used to be during active hi stripping or before . the three galaxies that we observed using the sma are expected to go through a similar evolutionary sequence and will be found with a truncated fuv disc of a comparable size to the h@xmath0 and co disc within their old stellar disc . judging from the star formation quenching timescales by @xcite and simulations by @xcite , the time it will take for our sma sample to become a galaxy like ngc 4569 is expected to be only a few hundreds myr . lccc & ngc 4330 & ngc 4402 & ngc 4522 + & + @xmath40)@xmath4 & 2.1 & 1.4 & 3.3 + @xmath41@xmath42 & 8.2 & 63.8 & 20.4 + @xmath43@xmath44 & 12551775 & 13381892 & 11381609 + + & + @xmath45 ( kpc)@xmath46 & 0.24.2 & 0.24.9 & 0.22.1 + @xmath47@xmath48 & 1.4109 & 6.6568 & 1.5187 + @xmath49@xmath50 & 140 & 145 & 122 + [ tab4 ] @xmath51projected distance from m87 . + @xmath52icm density . + @xmath53range of possible orbital velocities . + @xmath54radius from the galactic centre ( minimum and maximum values from either side ) . + @xmath55molecular gas surface density ( minimum and maximum values from either side ) . + @xmath56rotational velocity , @xmath57 , @xmath58 : the inclination angle , @xmath59 : the hi linewidth @xcite . + while a galaxy is undergoing very strong ram pressure , some molecular gas might be pushed away from the stellar disc as clearly seen in ngc 4522 . in order to test whether dense molecular gas can get stripped from a galaxy or not , we compare the ram pressure acting on our sample due to the icm wind with the restoring force ( per unit area ) by the galactic potential as a function of galactic radii ( i.e. @xmath60 versus @xmath61 ) in fig . [ restore ] . the icm density and the orbital velocity of each galaxy are adopted from the model of the virgo cluster @xcite . the shaded region indicates the range of possible orbital velocities at a given projected clustercentric distance . to estimate the restoring force at given galactic radii , we take the mean molecular gas and the hi gas surface densities along the major axis from the radial profiles ( i.e. figs [ n4330]c[n4522]c ) using our sma co data and the hi data ( viva ; @xcite ) . hence , in this calculation we do not include the iram data of ngc 4522 , and therefore the restoring force of this case in fig . [ restore ] is only the lower limit of the lowest possible value . the parameters used to estimate the restoring force and ram pressure are summarized in table [ tab4 ] . our rough estimate suggests that it is not impossible to remove the molecular gas in the very outer region in some circumstances . however , by adopting the rotational velocity of the co disc , which in all cases has a radius less than 5 kpc , we are focused on a region whose potential is dominated by the stellar component , not the dark matter halo . therefore , in practice , the actual stripping of molecular gas must be more difficult than our estimation . also , depending on the encounter angle between the icm wind and the galactic disc , the inner molecular gas can be better shielded , which will make the stripping even harder . lastly , considering that the ism is one entity after all and the total gas surface density of the two phases is more appropriate for more realistic estimation , the chance for the molecular gas to get stripped can be even less . in addition , even if molecular gas is pushed off from the galactic disc , whether the extraplanar molecular gas leaves the galactic halo for good or partially comes back is still debatable . considering that cluster galaxies are not significantly different from their field counterpart in co luminosity @xcite , the molecular gas completely stripped out of a galaxy is inferred to be a very small fraction . our estimate is , however , only the prediction for how easy or hard it is to remove the molecular gas from a galaxy completely , and the molecular gas can be still disturbed without stripping , as our data clearly show . in fact , the recent result by @xcite that severely hi - deficient galaxies tend to be found with lower molecular gas surface density may simply reflect the fact that the molecular gas of galaxies under strong icm pressure is perturbed , not necessarily implying molecular gas being deficient . extraplanar molecular gas may also form new stars ( e.g. eso 137 - 001 with a prominent rps tail found in a massive cluster , abell 3627 ; @xcite ) . the fuv emission with several h@xmath0 knots along the hi tail of ngc 4330 and the extraplanar fuv / h@xmath0 emission in ngc 4522 might be such a case . outside the old stellar disc , stars may form in two ways , from stripped molecular gas , or alternatively from newly formed molecular gas . indeed , @xcite in their simulations show that stripped diffuse gas can form dense molecular gas by cooling process , which can possibly form new stars . however , in extraplanar space , it is questionable how efficiently stripped atomic hydrogen can be pressurized , turning into molecular form @xcite . therefore , star formation taking place outside the main disc is more likely to be utilizing the molecular gas that is pushed from the main disc but has not left the galactic halo . as in ngc 4522 , however , there are several extraplanar h@xmath0 regions with no detectable co. this suggests that the h@xmath2 gas depletion time - scale ( @xmath62 ) in extraplanar space can be much shorter than what it is on the disc ( of the order of a few gyr ; @xcite ) . meanwhile , @xmath62 at the location of h@xmath0/co blob is measured to be comparable to what is normally found in the galactic disc ( see appendix b ) . therefore , there might be some extraplanar co even in ngc 4330 and ngc 4402 , but with mass smaller than our detectability or that of the iram @xcite . together with this , there is also an important difference between ngc 4330 and ngc 4522 , which is the presence of co in the extraplanar star formation region . in the case of ngc 4330 , more rapidly increasing ram pressure compared to ngc 4522 might be responsible for faster dissociation of molecular cloud , and hence absence of co along the fuv tail . alternatively , some fraction of stripped molecular gas might be destroyed by high - energy photon emitted from massive young stars @xcite . therefore extraplanar or stripped co may not be always observable even if it is pushed off from the stellar disc . in this work , we have presented high - resolution 12co ( 2@xmath31 ) data obtained using the sma of three virgo spirals , ngc 4330 , ngc 4402 and ngc 4522 , which are undergoing active hi stripping due to ram pressure . by comparing with single - dish 12co ( 2@xmath31 ) , hi , fuv , h@xmath0 and dss2 red data , we have discussed how icm pressure changes molecular gas and star formation properties . most important results from our study can be summarized as follows : 1 . the overall co morphology and kinematics are quite asymmetric and disturbed . morphological peculiarities found in co are closely related to hi morphology in these galaxies , indicating that molecular gas is also affected by the ram pressure . we find the co clump in the south - west of ngc 4330 ( figs [ n4330]a c ) and the modest co bump in the north - west of ngc 4402 ( figs [ n4402]a c ) , with distinct velocity structure from the overall velocity gradient of the molecular disc . this also supports the idea that molecular gas can be displaced by the external pressure inside a stellar disc . 3 . both fuv and h@xmath0 are enhanced where hi and co are compressed , supporting that ram pressure can also trigger star formation temporarily ( e.g. figs [ n4402]d f of ngc 4402 ) . in these galaxies , however , fuv shows distinct morphology and extent from those of co , while h@xmath0 and co are overall in good agreement ( fig . [ multi ] ) . the large offset between fuv and h@xmath0 must indicate the star formation has been recently quenched , likely over the last 100 myr as the molecular gas properties have changed . 4 . to compare with ngc 4569 ( fig . [ multi]d ) , which is thought to have already crossed the cluster core , the offset between fuv and h@xmath0/co is expected to smooth out over time ( @xmath1a few hundreds myr ) after the core crossing as the galaxy further loses hi , i.e. the capability of forming molecular gas , and star formation is completely shut down at large galactic radii . in summary , our results suggest that a strong icm pressure not only strips diffuse atomic gas but also changes the properties of dense molecular gas in the inner few kpc of a galaxy . molecular gas can be pushed from the stellar disc in the outer part as reported in previous studies , yet we find that molecular gas is less likely to be completely stripped from a galaxy . on the side where the icm pressure acts , molecular gas can be enhanced by compression and molecular formation from atomic gas , locally increasing star formation . as the gas is rapidly used up , the star - forming disc will decrease in size as no sufficient atomic gas is left to form more molecular gas . on the opposite side , molecular gas can be perturbed in a way that reflects the atomic gas , which may prevent a galaxy from forming stars as before , i.e. injecting turbulence into molecular gas is likely to cause star formation rates to be inefficient @xcite . the combination of these two effects must be an important driver that makes galaxies evolve to become passive in the cluster environment . in order to investigate the molecular gas properties and its conditions for star formation under strong icm pressure more quantitatively , we are currently probing the density and temperature of molecular gas using multitransitions and isotopes of co ( lee et al . in preparation ) . our future work is expected to provide further insights on the evolution of molecular gas and thus star formation efficiency in galaxies undergoing ram pressure . support for this work was provided by the national research foundation of korea to the center for galaxy evolution research ( no . 2010 - 0027910 ) and science fellowship of posco tj park foundation . this work has been also supported by nrf grant no . 2015r1d1a1a01060516 . the submillimeter array is a joint project between the smithsonian astrophysical observatory and the academia sinica institute of astronomy and astrophysics and is funded by the smithsonian institution and the academia sinica . this work made use of heracles , ` the hera co - line extragalactic survey ' @xcite . we acknowledge the usage of the hyperleda database ( http://leda.univ-lyon1.fr ) . this research has made use of the goldmine database @xcite , and it made use of aplpy , an open - source plotting package for python hosted at http://aplpy.github.com . st was supported by the alvin e. nashman fellowship in theoretical astrophysics . bl is grateful to e .- j . chung for the technical help with gipsy . 99 abramson a. , kenney j. d. p. , crowl h. h. , chung a. , van gorkom j. h. , vollmer b. , schiminovich d. , 2011 , aj , 141 , 164 abramson a. , kenney j. d. p. , 2014 , aj , 147 , 63 alatalo k. , et al . , 2015 , apj , 812 , 117 barmby p. , et al . , 2006 , apj , 650 , l45 bekki k. , 2014 , mnras , 438 , 444 bigiel f. , leroy a. , walter f. , brinks e. , de blok w. j. g. , madore b. , thornley m. d. , 2008 , aj , 136 , 2846 bigiel f. , et al . , 2011 , apj , 730 , l13 blitz l. , rosolowsky e. , 2006 , apj , 650 , 933 bhringer h. , briel u. g. , schwarz r. a. , voges w. , hartner g. , trmper j. , 1994 , nature , 368 , 828 boselli a. , cortese l. , boquien m. , boissier s. , catinella b. , gavazzi g. , lagos c. , saintonge a. , 2014 , a&a , 564 , a67 cayatte v. , van gorkom j. h. , balkowski c. , kotanyi c. , 1990 , aj , 100 , 604 chung a. , van gorkom j. h. , kenney j. d. p. , vollmer b. , 2007 , apj , 659 , l115 chung a. , van gorkom j. h. , kenney j. d. p. , crowl h. , vollmer b. , 2009 , aj , 138 , 1741 chung e. j. , 2012 , phd thesis , yonsei university chung e. j. , kim s. , 2014 , pasj , 66 , 11 crowl h. h. , kenney j. d. p. , van gorkom j. h. , vollmer b. , 2005 , aj , 130 , 65 crowl h. h. , kenney j. d. p. , 2008 , aj , 136 , 1623 davies r. d. , lewis b. m. , 1973 , mnras , 165 , 231 drimmel r. , spergel d. n. , 2001 , apj , 556 , 181 fujita y. , nagashima m. , 1999 , apj , 516 , 619 fumagalli m. , krumholz m. r. , prochaska j. x. , gavazzi g. , boselli a. , 2009 , apj , 697 , 1811 gavazzi g. , boselli a. , donati a. , franzetti p. , scodeggio m. , 2003 , a&a , 400 , 451 gil de paz a. , et al . , 2007 , apjs , 173 , 185 giovanelli r. , haynes m. p. , 1985 , apj , 292 , 404 gunn j. e. , gott j. r. , iii , 1972 , apj , 176 , 1 haynes m. p. , giovanelli r. , 1984 , aj , 89 , 758 helfer t. t. , thornley , m. d. , regan , m. w. , wong t. , sheth k. , vogel s. n. , blitz l. , bock d. c .- j . , 2003 , apjs , 145 , 259 henderson b. , bekki k. , 2016 , apj , 822 , l33 hidaka m. , sofue y. , 2002 , pasj , 54 , 33 jchym p. , combes f. , cortese l. , sun m. , kenney j. d. p. , 2014 , apj , 792 , 11 kenney j. d. p. , young j. s. , 1989 , apj , 344 , 171 kenney j. d. p. , young j. s. , hasegawa t. , nakai n. , 1990 , apj , 353 , 460 kenney j. d. p. , van gorkom j. h. , vollmer b. , 2004 , aj , 127 , 3361 kenney j. d. p. , abramson a. , bravo - alfaro h. , 2015 , aj , 150 , 59 kennicutt r. c. , jr . , 1998 , ara&a , 36 , 189 koopmann r. a. , kenney j. d. p. , young j. , 2001 , apjs , 135 , 125 koopmann r. a. , kenney j. d. p. , 2004a , apj , 613 , 851 koopmann r. a. , kenney j. d. p. , 2004b , apj , 613 , 866 kregel m. , van der kruit p. c. , de blok w. j. g. , 2004 , mnras , 352 , 768 kronberger t. , kapferer w. , ferrari c. , unterguggenberger s. , schindler s. , 2008 , a&a , 481 , 337 lee b. , chung a. , 2015 , pkas , 30 , 491 leroy a. k. , walter f. , brinks e. , bigiel f. , de blok w. j. g. , madore b. , thornley m. d. , 2008 , aj , 136 , 2782 leroy a. k. , et al . , 2009 , aj , 137 , 4670 lucy l. b. , 1974 , aj , 79 , 745 mei s. , et al . , 2007 , apj , 655 , 144 merluzzi p. , et al . , 2013 , mnras , 429 , 1747 mok a. , et al . , 2016 , mnras , 456 , 4384 nehlig f. , vollmer b. , braine j. , 2016 , a&a , 587 , a108 paturel g. , petit c. , prugniel p. , theureau g. , rousseau j. , brouty m. , dubois p. , cambrsy l. , 2003 , a&a , 412 , 45 qi c. , 2012 , available at : http://www.cfa.harvard.edu/@xmath1cqi/mircook.html rengarajan t. n. , iyengar k. v. k. , 1992 , mnras , 259 , 559 rhee m .- h . , van albada t. s. , 1996 , a&as , 115 , 407 salo h. , et al . , 2015 , apjs , 219 , 4 skrutskie m. f. , et al . , 2006 , aj , 131 , 1163 sofue y. , koda j. , nakanishi h. , onodera s. , kohno k. , tomita a. , okumura s. k. , 2003a , pasj , 55 , 17 sofue y. , koda j. , nakanishi h. , onodera s. , 2003b , pasj , 55 , 59 sofue y. , koda j. , nakanishi h. , hidaka m. , 2003c , pasj , 55 , 75 solomon p. m. , downes d. , radford s. j. e. , barrett j. w. , 1997 , apj , 478 , 144 stark a. a. , knapp g. r. , bally j. , wilson r. w. , penzias a. a. , rowe h. e. , 1986 , apj , 310 , 660 strong a. w. , mattox j. r. , 1996 , a&a , 308 , l21 sun m. , donahue m. , voit g. m. , 2007 , apj , 671 , 190 sun m. , donahue m. , roediger e. , nulsen p. e. j. , voit g. m. , sarazin c. , forman w. , jones c. , 2010 , apj , 708 , 946 swaters r. a. , 1999 , phd thesis , university of groningen tonnesen s. , bryan g. l. , 2012 , mnras , 422 , 1609 verheijen m. a. w. , 1997 , phd thesis , university of groningen vollmer b. , cayatte v. , balkowski c. , duschl w. j. , 2001 , apj , 561 , 708 vollmer b. , balkowski c. , cayatte v. , van driel w. , huchtmeier w. , 2004 , a&a , 419 , 35 vollmer b. , soida m. , otmianowska - mazur k. , kenney j. d. p. , van gorkom j. h. , beck r. , 2006 , a&a , 453 , 883 vollmer b. , braine j. , pappalardo c. , hily - blant p. , 2008 , a&a , 491 , 455 vollmer b. , 2009 , a&a , 502 , 427 vollmer b. , et al . , 2012a , a&a , 537 , a143 vollmer b. , wong o. i. , braine j. , chung a. , kenney j. d. p. , 2012b , a&a , 543 , a33 walter f. , brinks e. , de blok w. j. g. , bigiel f. , kennicutt r. c. , jr . , thornley m. d. , leroy a. , 2008 , aj , 136 , 2563 warmels r. h. , 1988a , a&as , 72 , 19 warmels r. h. , 1988b , a&as , 72 , 427 wong o. i. , kenney j. d. p. , murphy e. j. , helou g. , 2014 , apj , 783 , 109 yasuda n. , fukugita m. , okamura s. , 1997 , apjs , 108 , 417 yoon h. , 2013 , ms thesis , yonsei university we estimate the molecular gas depletion time - scale ( the ratio of molecular gas mass and star formation rate ) in extraplanar region of ngc 4522 following the prescription below . we first calculate molecular gas mass in two regions ( figure [ deple ] , red squares ) using iram 12co ( 2@xmath31 ) data of @xcite . using h@xmath0 image @xcite , the star formation rate in each region ( fig . [ deple ] , red squares ) is estimated as follows : @xmath63 in @xmath6 yr@xmath8 where @xmath64 is the h@xmath0 luminosity in erg s@xmath8 @xcite . measured depletion time - scales are 1.1 gyr ( region 1 ) and 0.8 gyr ( region 2 ) , respectively .

the fluctuating accretion model of and its extension by , seeks to explain the spectral - timing properties of the x - ray variability of accreting black holes in terms of inward - propagating mass accretion fluctuations produced at a broad range of radii . the fluctuations modulate the x - ray emitting region as they move inwards and can produce temporal - frequency - dependent lags between energy bands , and energy - dependent power spectral densities ( psds ) as a result of the different emissivity profiles , which may be expected at different x - ray energies . here we use a simple numerical implementation to investigate in detail the x - ray spectral - timing properties of the model and their relation to several physically interesting parameters , namely the emissivity profile in different energy bands , the geometrical thickness and viscosity parameter of the accretion flow , the strength of damping on the fluctuations and the temporal coherence ( measured by the ` quality - factor ' , ) of the fluctuations introduced at each radius . we find that a geometrically thick flow with large viscosity parameter is favoured , and confirm that the predicted lags are quite robust to changes in the emissivity profile , and physical parameters of the accretion flow , which may help to explain the similarity of the lag spectra in the low / hard and high / soft states of cyg x-1 . we also demonstrate the model regime where the light curves in different energy bands are highly spectrally coherent . we compare model predictions directly to x - ray data from the narrow line seyfert 1 galaxy and the bhxrb cyg x-1 in its high / soft state and show that this general scheme can reproduce simultaneously the time lags and energy - dependence of the psd . [ firstpage ] accretion : accretion discs galaxies : active

astro-ph0512394

a common characteristic in active galactic nuclei ( agn ) and x - ray binary systems ( xrbs ) is their strongly variable x - ray emission . most of the variability power is in the form of aperiodic fluctuations that span several orders of magnitude in temporal frequency ( e.g. * ? ? ? * ; * ? ? ? rapid , large amplitude x - ray variations , on time - scales down to milliseconds in xrbs and minutes in agn , are commonly observed ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , supporting the expectation from accretion theory that the bulk of the emission must originate close to the central black hole . however , in many sources ( in both agn and xrbs ) , large - amplitude x - ray variations , of tens of per cent fractional rms , are observed over several decades of time - scales ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , including values orders of magnitude longer than the viscous time - scale of the inner disc , suggesting that these fluctuations must originate at large radii . this discrepancy motivated the appearance of propagating - fluctuation models , as proposed by @xcite , who noted that while the _ emission _ might be produced only in the innermost regions , the _ variability _ can originate throughout the accretion flow . in this model , accretion rate fluctuations arise over a wide range of radii and , correspondingly , a wide range of time - scales , and propagate inwards to modulate the central x - ray emission . as noted by @xcite the model of @xcite is consistent with the observed linear relation between rms - variability amplitude and mean count rate in xrb and agn x - ray light curves , which implies a coupling between fluctuations on different time - scales , ruling out models where the variability arises from strictly independent flares or active regions @xcite . fluctuating accretion models , on the other hand , are consistent with this relation : the fluctuations couple together as they propagate down to the centre , so low frequency fluctuations produced at large radii modulate higher frequency fluctuations produced further in . a linear rms - flux relation is also observed in neutron star xrb systems , where the emission mechanism is most likely different to the one operating in black - hole systems @xcite . this result provides additional evidence for fluctuating - accretion scenarios because it suggests that the variability originates in the accretion flow and not just locally via the mechanism producing the emission @xcite . besides explaining the broad range of variability time - scales and the rms - flux relation , the fluctuating accretion model can also explain the spectral - timing properties of the variability , as noted by @xcite . for example , it has long been known ( e.g. * ? ? * ; * ? ? ? * ) that x - ray variations in black hole x - ray binaries ( bhxrbs ) often show hard lags , i.e. a delay in hard x - ray variations with respect to soft x - rays , which is larger for variations on longer time - scales , and at higher energies . the magnitudes of the lags are typically of order one per cent of the variability time - scale . similar time - scale - dependent , hard lags have recently been discovered in agn , albeit on much longer time - scales , commensurate with their higher black hole masses ( e.g. * ? ? ? ? * ; * ? ? ? in their analytical extension to lyubarskii s model , @xcite explain these lags by invoking a radially extended x - ray emitting region with an energy - dependent profile . in this scenario , the response of the emission to a fluctuation in the accretion flow is a function of the inward propagation time - scale , and hence radius of origin , of the fluctuation , combined with the emissivity profile . hence , if the emissivity profile is more centrally concentrated at higher energies , hard - band lags are produced such that the lags are larger for longer time - scale variations . @xcite show that the same basic picture can also explain the energy dependence of the psd of bhxrbs and agn , where there is relatively more high - frequency power observed at higher energies than at lower energies ( e.g. * ? ? ? * ; * ? ? ? this is because the emitting region acts as a low - pass filter and , as the emitting region of the soft band is more extended , the higher frequency variations are filtered more strongly in the soft band . due to its success in explaining many aspects of the x - ray variability data , the fluctuating accretion model of @xcite and @xcite warrants a deeper investigation , which is the aim of this paper . as shown by the brief analytical treatment of @xcite , in principle , the model produces time - scale - dependent lags of approximately the same amplitude as observed in the data , and energy - dependent psds . however it is not clear exactly how the lag spectrum ( lag as a function of fourier frequency ) and energy dependence of the psd varies as a function of emissivity profile , or the parameters of the accretion flow ( assuming the standard disc model of @xcite e.g. the viscosity parameter @xmath1 or the scale - height of the flow ) , or due to the effects of radial damping of variations in the accretion flow . furthermore , since @xcite analytically determined spectral - timing properties by making the simplifying assumption that the perturbation introduced into the accretion flow at each radius is a delta - function in time and radius , it is important to determine the effects on spectral - timing properties of a more realistic model , where stochastic variations over a broader range of time - scales are introduced at each radius . finally , the fluctuating accretion model should , in principle , produce light curves which are highly correlated in different bands ( i.e. spectrally coherent , see @xcite and appendix b of this paper ) , but this aspect of the model has not yet been studied . to study these various effects in more detail , in section [ model ] we introduce a computational toy model for a fluctuating accretion flow , based on the work of @xcite and @xcite , and explore the dependence of the psd , time lags and coherence on the model parameters in section [ properties ] . we show how the model predictions fit x - ray data from agn and the bhxrb cyg x-1 in sections [ agn ] and [ xrb ] . specifically , we will concentrate on explaining the spectral - timing properties of the bhxrb cyg x-1 in its high / soft state , and agn which show similar variability properties , since these states show rather simple @xmath2-type psd shapes , without complex quasi - periodic oscillations ( qpos ) . we discuss the implications of our results in section [ discussion ] and summarise our results in section [ conclusions ] . the model is based on the scenario proposed by @xcite , where fluctuations propagate inward through the accretion flow , modulating the emission of the inner regions . the fluctuations are produced on time - scales related to the viscous time - scale at the radius of origin and are uncorrelated for different radial scales . @xcite noted that , in this class of models , the accretion rate fluctuations can be carried efficiently , and be produced up to high fourier frequencies , by a geometrically thick accretion flow , identifying it with an extended corona over the thin disc . we use standard accretion disc @xcite considerations to relate the fluctuation time - scales and propagation speeds to the radial position in the disc , since the same relations are applicable to geometrically thick and optically thin accretion flows , as well as the standard geometrically thin disc . following @xcite , we assume that each independent annulus produces a pattern of fluctuations @xmath3 , having most of the variability power at the local viscous frequency @xmath4 ( e.g * ? ? ? * ) , where @xmath5 is the disc scale height to radius ratio , @xmath1 is the viscosity parameter , the radial position @xmath6 is in units of gravitational radii @xmath7 and the frequency is given in terms of @xmath8 . the local accretion rate at any radius in the disc is allowed to fluctuate around a value @xmath9 , as @xmath10 , where @xmath11 . as this new value of the accretion rate propagates inward , it serves as the @xmath12 for the fluctuations produced further in , modulating the higher frequency fluctuations . the propagation speed is assumed to equal the local radial drift velocity @xmath13 . a key feature of this model is that the absolute amplitude of accretion rate variations is proportional to the local accretion rate , @xmath9 , at every radius . considering that the characteristic frequencies of the input fluctuation @xmath14 decrease with radius , this implies that the amplitude of high frequency fluctuations is proportional to the actual value of the accretion rate , given by fluctuations on longer time - scales , and this produces the rms - flux relation . to generate light curves , following @xcite , we assume that the x - rays are emitted by a radially ( but not necessarily vertically ) extended region . this emitting region might correspond to an accreting corona , possibly sandwiching an optically thick , geometrically thin disc , or be another type of optically thin accretion flow ( e.g. adaf , @xcite ) . having no pre - defined emission mechanism we simply assume that the emitted flux per unit area is proportional to the local accretion rate , which introduces the variability , and to a stable radial emissivity profile . the emissivity profile , @xmath15 , is taken to follow the radial rate of gravitational energy loss in the accretion disc , @xmath16 , where @xmath17 is the inner radius of the disc ( and of the emission ) , which we will fix at @xmath18 in our simulations . this profile describes the _ total _ energy loss . however , as the emitted spectrum might be radially dependent , a given energy band can have a different emissivity profile and so we will use @xmath19 , where the emissivity index @xmath20 is a model parameter . for the calculation of the variability patterns , we discretize the models of @xcite and @xcite by assuming that a finite number of annuli produce independent fluctuations . these annuli are spaced keeping a constant ratio between consecutive radii , to mimic the behaviour of a continuous accretion flow that produces equal variability power at each radial scale . this geometric spacing of the radii introduces equal power per decade in time - scale , reproducing the @xmath2 psd shape , as predicted by @xcite . all these independent annuli are assumed to produce a pattern of small accretion rate fluctuations , @xmath3 . to meet the requirement of producing most of the variability power at the viscous frequency , these signals are modelled as random fluctuations with a lorentzian - shaped psd of variable width , with peak frequency @xmath21 . as @xmath22 ( e.g. see @xcite ) . set to equal @xmath23 the width of the lorentzian is governed by the quality factor @xmath0 , equal to the ratio of lorentzian peak frequency to the full width at half maximum . we used broad input psds , with @xmath24 , a value similar to the four - lorentzian decomposition of the psd of cyg x-1 in the low / hard state @xcite and also much narrower psds with @xmath25 , to study the influence of this parameters on the final psds and cross spectra . the normalisation of these signal psds was chosen to produce a final fractional variability of the output light curves @xmath26 , by assigning each of the signals an rms @xmath27 , where @xmath28 is the number of independent annuli . we used the method described by @xcite to generate the signal time series @xmath3 , i.e. by generating periodograms with the underlying signal psd and the correct statistical properties of a noise process and obtaining @xmath29 through their inverse fourier transform . the resulting @xmath29 have zero mean and rms amplitude @xmath30 . the accretion rate at a given annulus , @xmath31 , is then calculated iteratively as the product of the accretion rate at that position , including its contribution to the fluctuation , @xmath32 , by the accretion rate at the annulus lying directly outside , which includes the fluctuations from all outer annuli , @xmath33 so that each @xmath34 is a flicker noise time series with power law psd of slope @xmath35 up to @xmath36 , and shows a linear rms - flux relation . the effect of the extended emitting region is accounted for when calculating the _ light curves_. first , an emissivity index @xmath20 is assumed for each energy band . then , the light curves are constructed by adding the variability patterns @xmath31 from all annuli , which are weighted by the flux originating from each annulus ( i.e. @xmath37 , where @xmath38 is the width of the annulus ) , and time - shifted by the propagation time of fluctuations from the outermost annulus to the present annulus . the disc parameters are either kept constant or allowed to vary radially as @xmath39 , where c is a scaling constant and @xmath40 is an index which governs the radial variation ( e.g. constant @xmath41 and @xmath1 correspond to @xmath42 ) . we will normally use @xmath43 , to produce thick disc parameters @xmath44 and @xmath45 at the innermost radius . radially - varying disc parameters change the radial separation of independent annuli but preserve the separation of independent characteristic frequencies . therefore , we fix the geometrically spaced frequencies and use @xmath46 to calculate the corresponding radii and propagation speeds . fig . [ fig0 ] shows a light curve realisation with emissivity index @xmath47 ( a ) , together with its psd , plotted as frequency @xmath48 power ( b ) , and its rms variability amplitude plotted vs the average count rate of short sections of the light curve ( c ) . finally , damping of the fluctuations as they propagate through the accretion flow affects the power spectra of the transmitted signals . the magnitude of the damping effect depends on the nature of the fluctuations and their propagation mechanism , which are not specified in our phenomenological model . in their analytical model , @xcite include this damping effect by using a green function for diffusion in the geometrically thin disc approximation . here , for simplicity and generality , we use a simple prescription for damping by assuming that fluctuations of frequency @xmath49 are significantly damped after travelling a distance @xmath50 , so that fluctuations on frequencies much higher than the local viscous frequency are strongly suppressed ( see e.g. * ? ? ? * but notice that in that case damping was specifically derived for diffusive propagation of a delta - function shaped impulse in mass accretion rate ) . the damping prescription is implemented by first taking the fourier transform of @xmath51 at each radial step @xmath52 , then multiplying the real and imaginary coefficients of the fourier transform at each frequency @xmath49 by a frequency - dependent factor @xmath53 ( where the damping coefficient d is a free parameter ) and inverse fourier - transforming the result to get the new , damped @xmath54 . throughout this paper we will use @xmath7 as units of distance and @xmath55 as units of time . for a quick check of frequencies and time - scales note that @xmath56 where @xmath57 , so the frequencies quoted would be in units of hz for a @xmath58 black hole and units of @xmath59 hz for a @xmath60 black hole . the extended emitting region acts as a low - pass filter on the psd . this effect can be understood by considering that the signals at low temporal frequencies are imprinted in the flux from most of the emitting region , while higher frequencies are only imprinted in the emission from smaller , inner , areas . as the light curve has additive components from all these regions , this produces comparatively less variability power at high frequencies , creating a bend in the psd . the ` filtered ' psd can be approximated by multiplying the original psd(@xmath49 ) by the squared fraction of the total flux that is produced within the characteristic radius @xmath61 of each frequency @xmath49 as @xmath62 the filter factor tends to 1 as @xmath61 tends to @xmath63 , i.e. for low temporal frequencies the original variability power is preserved , and decreases monotonically as @xmath61 reaches @xmath17 , reducing the variability power at the high frequency end . obviously , as the emissivity profiles @xmath64 used are steep , it makes little difference to the filtered psd if the emission region is truncated at a large radius rather than extending to infinity . the filtering effect due to the origin of the signals within the emitting region could be easily calculated once the emissivity profile is defined , and its effect is simply to reduce the overall _ normalisation _ of the contribution to the psd from each signal ( in a frequency - dependent way ) , without distorting the signal psd shape . however , the finite travel time of the fluctuations through the emitting region distorts the psd shape of the input signals , by smoothing out high frequency fluctuations , thus reducing the variability amplitude further , while damping can also affect the low - frequency power . therefore , the total effective filtering of the underlying fluctuations is quite complex , so we will primarily use numerical simulations to demonstrate its effects in the following section . the frequency - resolved time lags can be approximated by defining , for each signal @xmath65 ( produced at radius @xmath66 ) an ` average travel time ' , @xmath67 , by weighting the travel time of the fluctuations from @xmath66 to the radius @xmath6 , @xmath68 by the emissivity profile @xmath64 , @xmath69 if the signals are simple fluctuations on the viscous time - scale only ( i.e. defined by a delta - function and not a broad lorentzian psd ) , then @xmath66 can be expressed in terms of frequency and this formula gives the mean travel time as a function of @xmath49 . otherwise , if the signals are characterised by broad lorentzian psds , a given temporal frequency will have important contributions from a range in radii @xmath66 . in this case , the mean travel time at a given frequency @xmath49 is a sum of @xmath70 for each contributing ( i.e. overlapping ) lorentzian signal psd , weighted by the variability power at @xmath49 contributed by that signal . time lags between two energy bands appear when they are characterised by different emissivity indices , since this produces different mean travel times for each band . we will use the difference of @xmath71 calculated with the corresponding @xmath20 indices as an approximation to the time lags between energy bands . an explicit form for the filter factor of the psd and for the time lags is given in appendix a. the model was constructed to produce power law - psd time series that also reproduce the non - linear properties of agn and bhxrb light curves . the exact psd shape and spectral - timing properties between different energy bands , depend on a few key model parameters : emissivity indices , disc structure parameters , psd of the input signals and damping coefficient . here we will explore how each of these parameters affect the observable properties . an overview of the spectral - timing measurements we use here is provided in appendix b. the loss of gravitational energy in the accretion disc requires an energy release per unit area proportional to @xmath72 @xcite . a given energy band will have the same radial emissivity profile only if it represents a constant fraction of the emitted flux at each radius . however , following @xcite , we assume the emitted spectrum hardens towards the centre , so the actual radial emissivity profile will be steeper for harder bands . we will refer to time series constructed with emissivity index @xmath47 as the ` soft ' light curve and use higher values of @xmath20 for the ` hard ' bands . in this subsection , we assume the product of disc structure parameters @xmath73 , and that there is no damping ( @xmath74 ) . the psd of the final variability pattern at @xmath17 for a realisation with input signal @xmath24 , is shown by the solid line in fig.[1a_per ] . this psd would correspond to an infinitely concentrated emitting region located at @xmath17 , or equivalently , a compact emitting region located entirely _ within _ it keeps a @xmath2 behaviour ( corresponding to a flat top in this plot of power multiplied by frequency ) up to @xmath75 , as no annuli contribute with significant variability power above this frequency , the psd bends downward . the dotted and dashed lines show the psds of soft and hard light curves from a realisation with the same underlying accretion fluctuations , with @xmath76 and @xmath77 , respectively . the figure highlights the filtering effect of the extended emission on these two light curves , which makes them lose variability power at high frequencies , shifting the bend to lower frequencies . on longer time - scales , all the psds flatten to their original slope of -1 . the resulting psds bend continuously to steeper slopes at high frequencies but , for a limited frequency range , they can be approximated fairly well by a singly bending power law model , as has been used to fit real data . following e.g. @xcite , we will use the following bending power law model to fit the psd , @xmath78 table [ t1 ] gives the values for the low and high frequency slopes , @xmath79 and @xmath80 respectively , and the break frequencies @xmath81 that result from fitting equation [ eq1 ] to psds with emissivity indices @xmath20= 3 , 4 and 5 , for broad ( @xmath24 ) and narrow ( @xmath25 ) input signal psds . these values depend slightly on the frequency range used for fitting , but reveal the general trend of more high - frequency power for harder energy bands , given by flatter @xmath82 , and higher @xmath81 and/or flatter @xmath83 . note that the underlying psds bend continuously to steeper slopes at higher frequencies , so limiting the frequency range used for fitting , excluding higher frequencies , produces flatter high - frequency slopes . [ t1 ] .bending power - law model fits to the psds of simulated light curves , with emissivity indices @xmath20= 3 , 4 and 5 , for broad ( @xmath24 ) and narrow ( @xmath25 ) input signal psds . higher @xmath20 values , corresponding to higher energy bands , retain more high - frequency power , evident in the flatter low - frequency slope @xmath82 and in and higher @xmath81 and/or flatter @xmath80 . the fitting range used was @xmath84 . note that the underlying psds bend continuously to steeper slopes at higher frequencies , so limiting the frequency range used for fitting , excluding higher frequencies , produces flatter slopes above the break . [ cols="<,^,^,^,^",options="header " , ] damping can also reduce the coherence , especially in the case of broad input signal psds . the dot - dashed line in fig . [ input_coh ] , shows the coherence for light curves of emissivity indices @xmath85 and @xmath86 , @xmath24 and damping coefficient @xmath87 . these light curves only differ from those corresponding to the dashed line on the same plot by the effect of damping , which increases the loss of coherence at the highest frequency bin from 2 to 4 per cent . again , a larger difference in emissivity indices makes this drop more pronounced , and damping effects increase these losses accordingly : for @xmath88 and @xmath89 the coherence at the highest frequency bin drops by @xmath90 per cent and for @xmath88 and @xmath91 , by 60 per cent . finally , damping can produce slightly steeper lag spectra , reducing the lags at high frequencies . this effect is small : damping coefficients of @xmath74 and @xmath87 produce lags differing by a factor of @xmath92 at frequencies close to the measured break - frequency @xmath93 and less at lower frequencies . agn x - ray light curves from long monitoring campaigns typically show power law psds with a slope @xmath35 at low frequencies , bending or breaking at high frequencies to slopes @xmath94 or steeper ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? the high - frequency psd is energy dependent , with higher energy psds showing a higher break frequency and/or flatter slopes above the break , depending on model assumptions ( e.g. @xcite ) . significant frequency - dependent time lags have been measured in a few agn [ ngc 7469 @xcite ; @xcite ; ngc 4051 @xcite ; ngc 3783 @xcite ] . in all cases , these lags increase with time - scale and energy separation of the bands and show magnitudes of up to a few percent of the variability time - scale . the measured coherence is usually high ( @xmath95 ) up to about the break frequency , sometimes dropping by tens of per cent above the break @xcite . this behaviour is qualitatively consistent with the predictions of the fluctuating accretion model outlined in section 3 . now we will compare the model predictions directly to agn data to check this consistency quantitatively . the time lags in reported by @xcite are among the largest measured in agn , with values of @xmath96 0.25 , 2.5 and 4.5 per cent of the variability time - scale , when calculated between the 0.1 - 0.5 kev band and the 0.5 - 2 , 2 - 5 and 5 - 10 kev bands , respectively . the slopes of the lag spectra are not well constrained but are consistent with -1 . to reproduce these large lags with the model , we chose a fairly extended soft emitting region , with emissivity index @xmath97 , assumed @xmath98 at @xmath99 , used a value of @xmath100 to produce a lag - spectrum slope @xmath35 and used @xmath74 . using eq . [ eq1 ] , we calculated the lag spectra for these model parameters , choosing emissivity indices for the remaining energy bands to match the observed lags . we used @xmath101 for the calculation , although other @xmath0 values produce equally good fits as the width of the input signal psd has a very small effect in the lag spectrum over the frequency range used here ( see sec . [ input ] ) . the resulting lag spectra for @xmath102=3.3 , 5.5 and @xmath63 are shown in fig . [ lags_ngc ] , in dotted , solid and dashed lines , respectively . to convert light - crossing times to seconds , we assumed a mass of @xmath103 @xcite . overlaid in this figure , we show the lags measured from data for , where the observed light curves were constructed as described in @xcite . although the errors on the data are large , it is clear that the chosen model parameters are consistent with the amplitude , slope and energy dependence of the measured lags . was monitored with _ rxte _ and on long and short time - scales respectively by @xcite , producing a well sampled psd over 6 decades in frequency . the high frequency psd in shows strong energy dependence , varying from a slope @xmath104 in the 0.1 - 0.5 kev band to @xmath105 in the 2 - 5 kev band . the difference in emissivity indices for these energy bands , suggested by the lag spectra , can also explain the energy dependence of the high frequency psd slope . to measure the energy dependence we produced model light curves with broad input signal psds , @xmath24 , and model parameters as described above and fitted the final psd with a bending power law model . to make a fair comparison with the data , we followed the procedure used by @xcite by fixing the low - frequency slope @xmath82 to -1.1 and the bend frequency @xmath81 to the best fitting value for an intermediate emissivity index , and fitted the simulated psds over a similar range in frequencies . using again @xmath97 for the 0.1 - 0.5 kev and @xmath106 for the 2 - 5 kev band , we obtained psds that differ by @xmath107 , spanning a similar range to that measured for the real data , @xmath108 . the resulting psds and bending - power law fits are shown in fig . [ psd_surr_ngc ] , for emissivity indices @xmath109 2.5 , 3.3 and 5.5 . notice that in this figure the psd normalisations have been rescaled to match at the low frequency end , to highlight the different changes in slope . the filled circles in this figure correspond to @xmath110 , which would represent the 0.5 - 2 kev band , as seen in section 4.1 . correspondingly , the difference in high frequency psd slope between the @xmath111 and @xmath112 light curves is @xmath113 , placing the @xmath112 light curve between the 0.5 - 1.0 kev ( @xmath114 ) and 1.0 - 2.0 kev ( @xmath115 ) bands . to check the consistency of the time - scales , we compared the bend frequency @xmath116 measured by @xcite , using the combined _ rxte _ and light curves , to model light curves covering a similar range in frequencies . as the data used for this fit corresponded to the 410 kev band , we used the ` hard ' simulated light curves with @xmath20 between 5.5 and @xmath63 . the best fit bend frequencies for @xmath117 and @xmath118 , respectively , are @xmath119 and @xmath120 . equating these values to @xmath121 hz yields a mass of @xmath122 , consistent with the reverberation mapping mass of @xmath123 @xcite . a similar energy dependence of the psd has been found in other agn . as an example , the psd of the seyfert 1 galaxy can also be fit with a bending power law model obtaining flatter high - frequency slopes for higher energy light curves @xcite . the energy dependence in this case is less pronounced than in , possibly suggesting that the energy bands used have closer emissivity indices . incidentally , emissivity indices that are closer in value would also produce smaller time lags , as observed in . however , more parameters contribute to define the observed lags , for example , the mass of the central object changes the frequency range probed in terms of the light crossing times . a comparison of the fractional lags at , for example , the psd bend frequency in each case gives a mass - independent measure , so an extrapolation of the lag spectra in published by @xcite down to the break time - scale would suggest that the lags in this case are indeed systematically larger than in . we now apply our simple model to data from the well - studied , persistent black hole xrb cyg x-1 , in the high / soft state . our model produces light curves with a @xmath124 psd shape down to low frequencies . this is because we assume that the input signals originate at many different radii throughout the accretion disc , are equally separated in the logarithm of frequency and have identical rms amplitudes of variability . psds of this shape are seen for several agn ( see previous section ) and cyg x-1 in its high / soft state . we will use data from a short ( 2.3 ks ) _ rxte _ observation of cyg x-1 in the high / soft state , observation i d 10512 - 01 - 09 - 01 , obtained on 1996 june 18 . using proportional counter array ( pca ) light curves with @xmath125 s resolution , we measured the lag spectrum and psd in two bands , 25.1 kev ( soft band ) and 813 kev ( hard band ) . we subtracted poisson noise , including the appropriate deadtime correction , e.g. @xcite , before calculating the ratio of the psds . to keep the fitting simple , we consider only the ratio of the psds , and not the energy - dependent psd shapes themselves , so we do not attempt to match the psd shape by assuming any particular input signal . this approach allows us to estimate the lags and filtering effect in a simple analytical way , by assuming that each frequency contributing to the lag spectrum and psd ratio corresponds to a single radius , so input signal psds do not overlap in frequency , with the lags and psd filtering determined using the analytical expressions given in appendix a. we note that @xcite do not claim any evidence of energy dependence of the psd in their fitting of the same data used here , but we do find a significant energy dependence using the psd ratio ( see below ) . the discrepancy may result from the fact that the measured psd ratio is , in fact , more sensitive than independent fits to the psd , because the light curves in different bands are correlated , so that statistical scatter in the psd due to the stochastic nature of the light curves is in the same direction in both bands , and its effect on the psd ratio is therefore mitigated . we first stress that due to the complexity of fitting even our relatively simple model to the data , and the difficulty of quantifying certain measurement uncertainties ( see below ) , we will only test the broad consistency of our model with the data . therefore , we will not quote statistical errors on parameters derived here , which should be treated as only indicative of the underlying physical parameters . since our model contains many parameters with complex degeneracies between them , we only fit the model with a few parameters left free , to demonstrate whether the model can reproduce the data . for all our fits , we assumed a black hole mass of 10 m@xmath126 . the mass provides the scaling factor between the units of light - crossing time - scale used in our method , and the observed units of seconds . the effect of changing mass is therefore equivalent to that of a simple linear change in the quantity @xmath127 ( see section 3.2 ) at all radii , and is rather weak except at high frequencies . consistent with our earlier simulations , we initially keep @xmath73 at all radii . we assume a fixed outer disc radius @xmath128 @xmath129 , but allow the inner disc radius @xmath17 to vary , down to a minimum value of 1.23 @xmath129 ( i.e. the minimum stable orbit for a kerr black hole ) . for simplicity , the soft emissivity index was initially fixed at @xmath88 , while @xmath102 was allowed to vary . we also introduced an additional normalising factor for the lags , so that all the lags predicted by the model can be increased by a single factor , which we leave free to vary in the fits . this renormalisation represents the possibility that the fluctuations propagate at some speed proportional to the theoretically predicted local drift velocity ( see section 2.1 ) , but not necessary equal to it . note that the psds in different energy bands can have different overall normalisations for a variety of reasons . for example , if a strong constant thermal emission component is present , as in the high / soft state ( see @xcite ) , the variability is diluted in the soft band and the psd normalisation reduced . alternatively , spectral pivoting at high energies ( e.g. due to compton cooling of the corona ) , which we do not model here , can lead to a larger normalisation in the soft band compared to the hard band ( e.g. see @xcite for evidence of this effect in the agn ngc 3227 ) . for these reasons , when fitting our model we allow the normalisation of the psd ratio to be free . the observed lag and soft / hard psd ratio spectra are shown in fig . [ cyghilagratio ] , together with the best - fitting model resulting from a joint fit of the lag and filtering equations to the data , i.e. we assume the same parameters for fitting both the lag and ratio plot . the reduced chi - squared , @xmath130 , indicates an excellent fit , for a renormalisation factor of 1.86 ( i.e. fluctuations propagate at @xmath131 times lower drift velocity than given by standard disc theory ) , @xmath132 , and @xmath133 . we caution however that the goodness - of - fit is almost certainly over - estimated , since the random errors in the soft and hard psds are correlated at low frequencies due to the intrinsic correlation between the light curves in these bands ( the correlation is much weaker at high frequencies due to poisson noise effects ) , and hence the error bars at these frequencies are almost certainly over - estimated . furthermore , there is likely to be some additional uncertainty in psd ratio at high frequencies ( @xmath134 hz ) , since an assumed poisson noise level had to be subtracted from each psd before the ratio was taken , and the uncertainty in the assumed noise level is not accounted for in the fit . the small inner disc radius preferred by the data is consistent with a rapidly rotating black hole . however , this interpretation is strongly governed by our assumptions about the emissivity profile , and our simplifying assumption that each frequency corresponds to a single input signal . since the travel time of material from 6 @xmath129 to 1.23 @xmath129 is very short , the value of @xmath17 is not driven by lags at lower frequencies , but is mainly due to a combination of the relatively large high - frequency lags , and the weak drop in psd ratio at high frequencies . indeed , the frequencies corresponding to signals at 6 @xmath129 and 1.23 @xmath129 are 65 hz and 700 hz respectively , so that if the edge of the emitting region were observed at 6 @xmath129 a very pronounced drop in both the lags and the psd ratio would be observed approaching this frequency , hence the smaller radius is preferred . however , if the input signals overlap substantially , the larger lags from signals at larger radii can contribute even at high frequencies , so a strong drop in lag and psd ratio would not be expected ( see section [ input ] ) . there may also be substantial emission at radii smaller than @xmath17 , which is not considered in our model . this would also increase the lags at high frequencies and reduce the effects of filtering which cause the drop in psd ratio at high frequencies . thus , the small value of @xmath17 inferred from our fit is quite model - dependent and should be treated with caution . the fact that the model can explain the soft / hard psd ratio also implies that it can explain the results of fourier frequency - resolved spectral analysis of the high / soft state continuum , as presented by @xcite . the fourier - resolved spectra become harder at higher temporal frequencies , with the change in photon index consistent with the frequency - dependent change in psd ratio which we measure here . finally , for completeness we note here that , unlike the smooth @xmath2 shape seen in the high / soft state , cyg x-1 in the low / hard state is characterised by a band - limited psd with a few humps , which can be modelled as four broad lorentzian components @xcite . the lag spectrum also reveals a stepped structure , but it still follows approximately a @xmath135 power law slope . although the model can fit reasonably well the general trend seen in the psd ratio and lag spectra , it can not reproduce these detailed features , indicating that some of the model assumptions are not appropriate in this case . in particular , our simple implementation of the model uses power law emissivity indices and equal variability power per contributing annulus , so these smooth functions can not produce additional structure at a few given frequencies . the general scheme of propagating fluctuations can still be correct but additional assumptions need to be made , for example , assuming that a few annuli have much enhanced variability power will certainly produce a bumpy psd and might reproduce the steps in the lag spectra , as suggested by @xcite and also @xcite . the extended emitting region in our model is responsible for the high frequency bend in the psd , and the associated radial emissivity profiles produce the energy dependence of the psd and the time lags . in this section we will summarise the spectral - timing properties produced by the model , consider the implications of the model for the size of the emitting region , and discuss some possible improvements . keeping the local variability time - scales tied to the propagation time - scale produces time lag spectra of power law slope @xmath136 or flatter , and lags of @xmath137 of the variability time - scale , for a wide range of model parameters . as discussed in sections [ agn ] and [ xrb ] , these simple assumptions produce lag spectra that match the data well in time - scale dependence and amplitude . we note once again that these lags arise solely due to the difference in emissivity profiles of the x - ray energy bands and do not involve any other spectral evolution of the emitting region . the amplitude of the lags depends mostly on the emissivity indices of the energy bands , increasing rapidly with their difference , up to @xmath138 , above which the lag values tend to saturate . significant lags can appear between energy bands characterised by similar emissivity profiles and , correspondingly , similar psd shapes . in particular , the psd of cyg x-1 in the high / soft state shows weak energy dependence ( see section 5.1 ) , but the psd ratio and lags can still be reproduced simultaneously with close emissivity indices for each energy band ( @xmath139 ) , provided that the propagation time - scale is slightly longer than the local fluctuation time - scale , and assuming that the inner disc radius is small . the power - law shape of the lag spectra is quite robust , its slope depends only weakly on disc structure parameters @xmath140 . incidentally , the stability of the lag spectra might explain the behaviour seen in , e.g. cyg x-1 , in different spectral states . this object shows very different psd and energy spectra in the high / soft and low / hard states , indicative of different disc configurations , but surprisingly similar lags @xcite . a comparison with agn and bhxrb data in the high / soft state shows that the filtering effect of the extended emitting region , acting on a simple @xmath2 underlying psd shape , can broadly reproduce their psd shape and energy dependence . cyg x-1 in the low / hard state , however , requires a more complex _ intrinsic _ psd . in either case , the extended emitting region introduces a bend in the psd in addition to any intrinsic curvature , and produces the energy dependence of the filtered psds . note that the bending power - law , used to fit agn data , is only an approximation to the actual psds produced by the fluctuating - accretion model . the filtered psd bends down continuously at high frequencies and has no well - defined high - frequency power law slope . the difference between the bending power - law model and the filtered psds can be appreciated in fig . [ psd_surr_ngc ] , where the single - bend psd fit overestimates the power in the highest frequency bins . however , as this region of the psds from real data is often heavily affected by poisson noise , the associated error bars and scatter are large and it is not possible to appreciate any deviations from a simple power law slope . therefore , even better agn data would be needed to discern if the model can replicate the exact psd shape or if additional variability components are needed . we also note here that , although @xcite fit the high / soft state psd from combined data from 1996 june 4 - 18 with a simple power law ( index -2.1 ) up to @xmath141 hz , the signal - to - noise at these frequencies is still fairly low , and the psd during that time is known vary in shape between observations @xcite , which makes interpretation of the underlying shape even more difficult . @xcite have recently demonstrated that the high / soft state psd can be fitted with weak lorentzians _ in addition _ to a @xmath142 power - law component with an exponential cut - off . this last component is reminiscent of our simulated psds . therefore , our model may be considered as representative of the times when the lorentzians in the psd are very weak or absent . in our model , thick disc parameters , @xmath143 , in the inner regions of the accretion flow and inner radius @xmath144 put the break frequency around @xmath145 ( i.e. @xmath962 - 20 hz for a @xmath146 black hole , @xmath147 for a @xmath148 ) . these values are in general agreement with the average break frequencies of agn and bhxrbs , indicating that fluctuations on the viscous time - scales of a geometrically thick accretion flow are appropriate to explain the variability . a geometrically thick disc is necessary to produce the observed high frequency fluctuations as _ viscous _ frequencies . if the fluctuations were instead produced on dynamical time - scales , or a magnetic time - scale related to this , as in e.g. @xcite , then a thinner flow might be allowed . however , the fluctuations need not only be produced but also propagated , which poses a difficulty for much thinner accretion flows , as fluctuations on time - scales much shorter than the propagation time - scale are easily damped @xcite . this , of course , does not rule out an additional thin disc , possibly underlying the thick flow , that might contribute to the flux but not to the variability . in our implementation of the variability model , the x - ray emitting region extends out to large radii . the bend in the psd is produced by the radial distribution of variability time - scales and emissivity profiles alone and is not related to a characteristic time - scale at the maximum radius of emission . therefore , an outer edge to the emitting region might be allowed but is not required . the steep emissivity profiles concentrate most of the emitted flux in the centre so it would make little difference to the psd if the emitting region were truncated at some radius or if it extended out to infinity . however , the extent of the emitting region has a major impact on the lag spectra . if , for example , the x - ray emission is confined to the central few @xmath149 and responds to the variability pattern produced outside this radius , then time lags would still appear due to the radial segregation of the energy bands but , to first order , the lags would have the same value at _ all _ fourier frequencies ( e.g. * ? ? ? * ) . in our case , frequency - dependent lags appear because the fluctuations are produced _ within _ the emitting region . this consideration gives further support to the idea that @xmath81 does not correspond to the characteristic frequency of an outer edge of the emitting region , as noted above , as these lags are observed far below the break frequency . this , in turn , implies that the x - ray emitting region should extend at least as far as the radius corresponding to the longest time - scales where frequency - dependent lags are observed . we have made a number of simplifying assumptions that seem to reproduce the broad spectral timing properties of agn and bhxrbs . the detailed predictions of psds and lag spectra can be improved by including some additional effects . as a first step , the assumption that all radii in the disc produce the same amplitude of fluctuations can be relaxed , allowing for strongly enhanced variability at a few discrete annuli . this scenario might represent cyg x-1 in the low / hard state where the psd appears to be composed of a few broad lorentzians , so it would be interesting to check if the effect of the extended emitting region would reproduce well the steps observed in the psd ratio and in the lag spectra . to simplify the calculations , we have only allowed for power - law radial emissivity profiles . though this appears to be a reasonable assumption given the profile of total energy loss expected for an accretion disc , it is not clear that each energy band should follow a simple power - law profile . for example , the energy spectrum reveals other components in addition to the direct continuum that might be expected from an optically thin x - ray emitting corona . in bhxrbs in the high / soft state , black body disc emission is also seen , although it does not vary significantly on short time - scales @xcite , so may not contribute to the observed psd and lag spectra . however , in several agn ( e.g. ngc 4051 , @xcite ) , strongly variable soft excess emission is seen below @xmath150 kev , which may be associated with comptonised disc emission and hence could contribute a different component to the emissivity profile in the soft band . other complex profiles might be expected . for example , by fitting the profile of the broad iron line in with a relativistic disc - line model , @xcite have inferred a broken power - law emissivity profile , which steepens from @xmath111 ( i.e. similar to the value we infer for ngc 4051 ) to @xmath151 within 6 @xmath149 of the black hole . thus , more complex emissivity profiles , perhaps with breaks or additional components , and taking account of general relativistic effects , could be considered . additional variable components , arising for example through reflection or reprocessing of the primary variable emission , can also affect the coherence and time lags and should be included in more detailed variability models . in particular , in high quality data from cyg x-1 , @xcite show that a reflection component is necessary to reproduce the detailed _ energy dependence _ of the lags . the simple fluctuating - accretion model presented here agrees generally well with the observational variability measurements , but still needs to be set on physical grounds to be fully validated . a physical treatment must consider two things , the source of the variability ( i.e. source of accretion rate fluctuations , for example ) , and the source of emission . work on the evolution of a physical ( if simplified ) accretion flow , implemented with numerical simulations ( e.g * ? ? ? * ) , indicate that the necessary accretion rate fluctuations and wave - like propagation on viscous time - scale can be realised physically . finally , a model for the emission process , and its coupling to the accretion rate fluctuations , will be necessary to provide a physical motivation for the radial emissivity profiles of each energy band and , with it , a meaningful interpretation of the emissivity indices . we have used a computational fluctuating - accretion model to reproduce the spectral - timing properties of x - ray light curves from agn and bhxrbs . in the model , the variability is produced throughout the accretion flow as perturbations of the accretion rate on the local viscous time - scale , produced by many signals which are geometrically spaced in temporal frequency . the fluctuations propagate inward at the radial drift velocity and modulate the fluctuations produced further in . this general scheme , introduced by @xcite , produces light curves which possess @xmath2 type psds over a broad range of time - scales and follow a linear rms - flux relation . following @xcite , x - ray emission from a radially extended region is assumed to track the local accretion rate fluctuations and to show a power law radial emissivity profile , which steepens for higher energy bands . the different emissivity profiles and the inward propagation produce differences between the psds of each energy band and introduce time lags . our implementation of the model is successful in reproducing the observed variability properties of agn and bhxrbs , as follows : 1 . we reproduce the expected @xmath124 frequency - dependence of time lags between energy bands , and lag amplitudes of a few per cent of the variability time - scales , predicted by @xcite and observed in bhxrb and agn variability data . we demonstrate that the amplitudes and slopes of the lag spectra are only weakly dependent on changes in the emissivity indices , the radial dependence of the product of disc parameters @xmath152 , and the strength of damping of accretion fluctuations . the robust nature of the lag spectra may help to explain the similarity of lag spectra in the low / hard and high / soft states in cyg x-1 @xcite , despite the very different psd and energy - spectral shapes in these states . the extended emitting region suppresses short time - scale fluctuations , producing a gradual bend in the psd at high frequencies . the ` filtered ' psd shapes resemble those of agn and bhxrbs in the high / soft state ( e.g. @xcite ) and can be approximately fitted with a bending power - law model . the bend frequencies obtained match the observed frequencies if the fluctuations are produced on the viscous time - scales of a geometrically thick accretion flow . values of @xmath153 gave adequate bend frequencies compared to data from seyfert galaxy and the bhxrb cyg x-1 . 4 . the fact that all energy bands are produced by the same emitting region can maintain a high coherence ( @xmath154 ) between light curves of different energy bands , as observed by e.g. @xcite , @xcite . however , the coherence can be substantially reduced at high frequencies if input signals at each radius have fairly broad psds and the difference between emissivity profiles in different bands is large . coherence is also reduced if damping of the inward - propagating fluctuations is significant . these results may provide an explanation for the drops in coherence observed at high frequencies in various bhxrbs and agn @xcite . the model can simultaneously reproduce both the observed energy - dependence of psd shape and the lag spectrum for the agn ngc 4051 and the bhxrb cyg x-1 in the high / soft state , although the inferred propagation times in cyg x-1 are slightly slower than the drift velocity expected for a standard disc . in the low / hard state of cyg x-1 , our simple model can reproduce the general slope of the lag spectrum but can not account for the stepped shape of the lag spectrum . modelling the detailed features of this spectrum would require additional assumptions about the mechanism producing the variability . direct comparison with spectral - timing data from agn and bhxrbs shows good general agreement with the model . however , we note that if the model is correct , the empirical bending or broken power - law models used to fit the psds of agn and bhxrbs in the high / soft state must break down for high signal - to - noise data , because they only approximate the gradual bending shape predicted by our model . also , we note that the model implies that the psd bend time - scale measured for agn and bhxrbs is a function of both the underlying psd of accretion fluctuations ( which can contain intrinsic breaks ) _ and _ the emissivity profile of the observed energy band ( which introduces a high - frequency bend ) . therefore , the psd bend time - scale is not necessarily associated with the outer radius of the emitting region , and this region can , in principle , be extremely large , provided the emissivity profile is relatively steep . this research has made use of data obtained from the high energy astrophysics science archive research center ( heasarc ) , provided by nasa s goddard space flight center . we wish to thank the two anonymous expert referees for their valuable comments . pa acknowledges support from the international max - planck research school in astrophysics ( imprs ) , travel support from grants associated with the nasa _ rxte _ guest observer program , and the hospitality of nasa s goddard space flight center . pu acknowledges a research associateship from the national research council . 99 axelsson , m. , borgonovo , l. , larsson , s. , 2005 , a&a , 438 , 999 churazov , e. , gilfanov , m. , revnivtsev , m. 2001 , mnras , 321 , 759 cui , w. , zhang , s. n. , focke , w. , swank , j. h. , 1997 , apj , 484 , 383 fabian , a.c . , vaughan , s. , nandra , k. , et al . 2002 , mnras , 335 , l1 gierliski m. , zdziarski a. 2003 , mnras , 343 , l84 gilfanov , m. , churazov , e. , revnivtsev , m. , 2000 , mnras , 316 , 923 kato , s. , fukue , j. , mineshige , s. 1998 , black - hole accretion disks , kyoto ( kyoto university press ) king , a. r. , pringle , j. e. , west , r. g. , livio , m. 2004 , mnras , 348 , 111 kotov , o. , churazov , e. , gilfanov , m. 2001 , mnras , 327 , 799 lyubarskii , y. e. 1997 , mnras , 292 , 679 markowitz , a. , et al . 2003 , apj , 593 , 96 markowitz , a. 2005 , ( astro - ph/0508521 ) m@xmath155hardy , i. m. , papadakis , i. e. , uttley , p. , page , m. j. , mason , k. o. 2004 , mnras , 348 , 783 m@xmath155hardy , i. m. , gunn , k. f. , uttley , p. , goad , m. r. 2005 , mnras , 359 , 1469 miyamoto s. , kitamoto s. , 1989 , nature , 342 , 773 narayan , r. , yi , i. , 1994 , apj , l13 nowak , m. a. , 2000 , mnras , 318 , 361 nowak , m. a. , wilms , j. , dove , j. b. , 1999a , apj , 517 , 355 nowak , m. a. , vaughan , b. a. , wilms , j. , dove , j. b. , begelman , m. c. , 1999b , apj , 510 , 874 papadakis , i. e. , nandra , k. , kazanas , d. 2001 , apj , 554,133 pottschmidt , k. , wilms , j. , nowak , m. a. , heindl , w. a. , smith , d. m. , staubert , r. 2000 , a&a , 357 , 17 pottschmidt , k. , et al . , 2003 , a&a , 407 , 1039 press , w.h . , teukolsky , s.a . , vetterling , w.t . , flannery , b.p . , 1992 , _ numerical recipes _ , cambridge university press ( cambridge ) reig p. , papadakis i. , kylafis n. d. , 2002 , a&a , 382 , 202 revnivtsev , m. , gilfanov , m. , churazov , e. 2000 , a&a , 363 , 1013 shakura , n.i . , sunyaev , r. 1973 a&a , 24 , 337 shemmer , p. , uttley , p. , netzer , h. , m@xmath155hardy , i. m. 2003 , mnras,343,1341 timmer , j. , knig , m. 1995 , a&a , 300 , 707 uttley , p. 2004 , mnras , 347 , 61 uttley , p. , m@xmath155hardy , i. m. 2001 , mnras , 323 , 26 uttley , p. , m@xmath155hardy , i. m. , papadakis , i. e. , 2002 , mnras , 332 , 231 uttley p. , taylor , r. d. , m@xmath155hardy , i. m. , page , m. j. , mason , k. o. , lamer , g. , fruscione , a. , 2004 , mnras , 347 , 1345 uttley , p. , m@xmath155hardy , i. m. , vaughan , s. , 2005 , mnras , 359 , 345 uttley , p. , m@xmath155hardy , i. m. , 2005 , mnras , 363 , 586 van der klis , m. 1995 , lewin , van paradijs , van den heuvel , eds . x - ray binaries . cambridge university press , campridge , p.252 vaughan , b. , nowak , m. 1997 , apj , 474 , l43 vaughan , s. , fabian , a.c . , & nandra , k. 2003 , mnras , 339 , 1237 [ lastpage ] as mentioned in sec . [ model ] , the extended emitting region acts as a low - pass filter on the psd . this filter factor , @xmath156 , can be approximated as : @xmath157 taking the emissivity profile @xmath158 , the filter factor takes the form ( for @xmath159 ) : @xmath160 in our implementation , the fluctuations are produced on the viscous time - scale at the radius of origin , so @xmath161 . substituting this value in the equation above gives the psd filtering factor as a function of frequency @xmath49 , where @xmath49 is in units of @xmath162 . this approximation for the filtering effect is appropriate for narrow - input - psd signals , where most of the variability power produced by each annulus is produced on the corresponding viscous time - scale . finite travel time effects contribute to cancel out variability produced at higher frequencies than @xmath163 , so the filtered psd in this case is distorted further . for the calculation of the time lags , the average travel time of a signal produced at @xmath66 can be approximated as : @xmath164 if the fluctuations propagate with viscous velocity , @xmath165 , then @xmath166 and the average travel time can be calculated using the emissivity profile given above . for example , for the case where @xmath167 , the average travel time is : @xmath168 \nonumber\\ / \left [ \frac{r_s^{2-\gamma } -r_{\rm min}^{2-\gamma } } { 2-\gamma } -r_{\rm min}^{0.5}\frac{r_s^{1.5-\gamma } -r_{\rm min}^{1.5-\gamma } } { 1.5-\gamma } \right]\end{aligned}\ ] ] substitution of the exponentials with logarithms in the appropriate terms give the equation for the rest of the cases . for signals produced on the viscous time - scale of the radius of origin , @xmath66 can be substituted by @xmath61 as above , producing average travel times as a function of temporal frequency . finally , the difference between @xmath169 calculated for two different emissivity indices @xmath20 gives the time lag spectrum corresponding to this @xmath20 pair . we estimate the psd using the periodogram , where the variability power , @xmath170 , is calculated for discretely sampled frequencies @xmath171 as : @xmath172 where @xmath173 and @xmath174 are the real and imaginary parts of the discrete fourier transform of the time series @xmath175 , @xmath176 is the average count rate , @xmath177 is the sampling time interval and @xmath178 is the number of sampled points @xcite . the cross spectrum between two time series , e.g. simultaneous soft and hard light curves , @xmath175 and @xmath179 , is defined as @xmath180 , where @xmath181 and @xmath182 are the fourier transforms of the respective light curves . the cross spectrum is a complex - valued function , from where the coherence and phase lags can be extracted . the coherence @xmath183 for discretely sampled time series is calculated as : where @xmath185 and @xmath186 are the real and imaginary parts of the cross spectrum @xmath187 and angle brackets represent averaging over independent measurements ( i.e. the numerator in the equation is the modulus - squared of the averaged cross - spectrum ) . note that here , the averaging of independent measurements of the cross - spectrum and psd is carried out by binning up in frequency , as well as by averaging measurements of the cross - spectrum measured at the same frequency from separate light curve segments . for a more detailed discussion on the measurement and interpretation of coherence and time lags , and the determination of errors on these measurements , see @xcite , @xcite and , in the context of agn , @xcite .

detailed stellar evolution calculations have been performed to quantify the influence of the main sequence mass and the metallicity z of the progenitor on the structure of the exploding wd which are thought to be the progenitors of sne ia . in particular , we study the effects of progenitors on the brightness decline relation which is a corner stone for the use of sne ia as cosmological yard - stick . both the typical and z can be expected to change as we go back in time . we consider the entire range of potential progenitors with 1.5 to 7 and metallicities between z=0.02 to . our study is based on the delayed detonation scenario with specific parameters which give a good account of typical light curves and spectra . based on the structures for the wd , detailed model calculations have been performed for the hydrodynamical explosion , nucleosynthesis and light curves . the main sequence mass has been identified as the decisive factor to change the energetics of the explosion and , consequently , dominates the variations in the rise - time to decline relation of light curves . has little effect on the color index b - v . for similar decline rates , the flux at maximum brightness relative to the flux on the radioactive tail decreases systematically with by about . this change goes along with a reduction of the photospheric expansion velocity by about 2000 km / sec . a change in the central density of the exploding wd has similar effects but produces the opposite dependency between the brightness to tail ratio and and , therefore , can be separated . the metallicity alters the isotopic composition of the outer layers of the ejecta . selective line blanketing at short wavelengths decreases with z and changes systematically the intrinsic color index b - v by up to , and it alters the fluxes in the u band and the uv . the change in b - v is critical if extinction corrections are applied . the offset in the calibration of is not monotonic in z and , in general , remains . we use our results and recent observations to constrain the progenitors , and to discuss evolutionary effects of sne ia with redshift . the narrow spread in the fiducial rise - time to decline relation in local sne ia restricts the range of main sequence masses to a factor of 2 . the upper limit of 1 day for the difference between the local and distance sample support the need for a positive cosmological constant . the size of evolutionary effects are small ( ) but are absolutely critical for the reconstruction of the cosmological equation of state .

astro-ph0104257

the last decade has witnessed an explosive growth of high - quality data for supernovae both from the space and ground observatories with spectacular results , and new perspectives for the use of sne ia as cosmological yard sticks and for constraining the physics of supernovae . one of the most important new developments in observational supernova research was to establish the long - suspected correlation between the peak brightness of sne ia and their rate of decline , @xmath1 , by means of modern ccd photometry ( phillips 1993 ) . sne ia have provided new estimates for the value of the hubble constant ( @xmath10 ) based on a purely empirical procedure ( hamuy et al . 1996ab , riess , press & kirshner , 1996 ) , and on a comparison of detailed theoretical models with observations ( hflich & khokhlov 1996 , hereafter hk96 ; nugent et al . the values obtained are in good agreement with one another . more recently , the routine successful detection of supernovae at large redshifts , z ( e.g. perlmutter et al . 1995 , 1997 ; riess et al . 1998 ; garnavich et al . 1998 ) , has provided an exciting new tool to probe cosmology . this work has provided results that are consistent with a low matter density in the universe and , most intriguing of all , yielded hints for a positive cosmological constant @xmath11 of @xmath12 . it is worth noting that the differences in the maximum magnitude between @xmath11=0 and 0.7 is @xmath13 for redshifts between 0.5 to 0.8 . these results prompted the quest for the nature of the the dark energy , i.e. cosmological equation of state . current candidates include a network of topological defects such as strings , evolving scalar fields ( i.e. quintessence ) , or the classical cosmological constant . for a recent review , see ostriker & steinhardt ( 2001 ) , and perlmutter , turner & white ( 1999b ) . to separate between the candidates by sne ia , the required accuracy has to be better than @xmath14 to @xmath15 ( albrecht & weller 2000 ) . the results on @xmath16 and future projects to measure the cosmological equation of state depend on the empirical @xmath1 which is calibrated locally . this leaves systematic effects as the main source of concern . indeed , there is already some evidence that sne ia undergo evolution . it has been argued , that the local sn ia sample covers all the possible variations that may come from different progenitors , different explosion mechanisms and environments , etc .. for local sne ia , the observational and statistical characteristics depend on their environment . they occur less often in ellipticals than in spirals , and the mean peak brightness is dimmer in ellipticals ( branch et al . 1996 ; wang , hflich & wheeler 1997 ; hamuy et al . 2000 ) . in the outer part of spirals the brightness is similar to ellipticals while , in more central regions , both intrinsically brighter and dimmer sne ia occur ( wang et al . these dependencies show us that sne ia likely depend on the underlying population and may undergo evolution . if the evolution realizes then we have to know it and take it into account going back in time . otherwise we can not _ safely _ use a local calibration . in principle , more distant sample could come from younger and more metal - poor progenitors , or the dominant explosion scenario may change . there is general agreement that sne ia result from some process of combustion of a degenerate white dwarf ( hoyle & fowler 1960 ) . within this general picture , three classes of models have been considered : ( 1 ) an explosion of a co - wd , with mass close to the chandrasekhar mass , which accretes mass through roche - lobe overflow from an evolved companion star ( whelan & iben 1973 ) . the explosion is then triggered by compressional heating near the wd center . ( 2 ) an explosion of a rotating configuration formed from the merging of two low - mass wds , caused by the loss of angular momentum due to gravitational radiation ( webbink 1984 , iben & tutukov 1984 , paczyski 1985 ) . ( 3 ) explosion of a low mass co - wd triggered by the detonation of a helium layer ( nomoto 1980 , woosley et al . 1980 , woosley & weaver 1986 ) . only the first two models appear to be viable . the third , the sub - chandrasekhar wd model , has been ruled out on the basis of predicted light curves and spectra ( hflich et al . 1996b , nugent et al . 1997 ) . delayed detonation ( dd ) models ( khokhlov 1991 , woosley & weaver 1994 , yamaoka et al . 1992 ) have been found to reproduce the optical and infrared light curves and spectra of typical sne ia reasonably well ( hflich 1995 , hereafter h95 ; hflich , khokhlov & wheeler 1995 , hereafter hkw95 ; hk96 ; fisher et al . 1998 ; nugent et al . 1997 ; wheeler et al . 1998 ; hflich et al . 2000 ; lentz et al . 2001 ; gerardy et al . this model assumes that burning starts as subsonic deflagration and then turns to a supersonic , detonative mode of burning . due to the one - dimensional nature of the model , the speed of the subsonic deflagration and the moment of the transition to a detonation are free parameters . the moment of deflagration - to - detonation transition ( ddt ) is conveniently parameterized by introducing the transition density , @xmath17 , at which ddt happens . the amount of @xmath18ni , @xmath19 , depends primarily on @xmath20 ( h95 ; hkw95 ; umeda et al . 1999 ) , and to a much lesser extent on the assumed value of the deflagration speed , initial central density of the wd , and initial chemical composition ( ratio of carbon to oxygen ) . models with smaller transition density give less nickel and hence both lower peak luminosity and lower temperatures ( hkw95 , umeda et al . in dds , almost the entire wd is burned , i.e. the total production of nuclear energy is almost constant . this and the dominance of @xmath20 for the @xmath21 production are the basis of why , to first approximation , sne ia appear to be a one - parameter family . the observed @xmath1 can be well understood as a opacity effect ( hflich et al . 1996b ) , namely , as a consequence of the rapidly dropping opacity at low temperatures ( hflich , khokhlov & mller 1993 , khokhlov , mller & hflich 1993 ) . less ni means lower temperature and , consequently , reduced mean opacities because the emissivity is shifted from the uv towards longer wavelengths with less line blocking . a more rapidly decreasing photosphere causes a faster release of the stored energy and , as a consequence , steeper declining lcs with decreasing brightness . the dd models thus give a natural and physically well - motivated origin of the @xmath1 relation of sne ia within the paradigm of thermonuclear combustion of chandrasekhar - mass co white dwarfs . nonetheless , variations of the other parameters lead to some deviation from the @xmath1 . e.g. a change of the central density results in an increased binding energy of the wd and a higher fraction of electron capture close to the center which reduce the @xmath21 production ( hflich et al . 1996b ) . because dd - models allow to reproduce the observations , we use this scenario to test the influence of the underlying stellar population on the explosion . we note that detailed analyses of observed spectra and light curves indicate that mergers and deflagration models such as w7 may contribute to the supernovae population ( hflich & khokhlov 1996 , hatano et al . in particular , the classical `` deflagration '' model w7 with its structure similar to dd models has been successfully applied to reproduce optical light curves and spectra ( e.g harkness , 1987 ) . the evidence against pure deflagration models for the majority of sneia includes ir - spectra which show signs of explosive carbon burning at high expansion velocities ( e.g. wheeler et al . 1998 ) and recent calculations for 3-d deflagration fronts by khokhlov ( 2001 ) which predict the presence of unburned and partial burned material down to the central regions . currently , pure deflagration models may be disfavored for the majority of sneia but , clearly , they can not be ruled out either . previously , hflich , wheeler & thielemann ( 1998 , hereafter hwt98 ) studied evolutionary effects induced by the progenitor . they calculate differences in the lc and nlte - spectra as a function of parameterized values of the integrated c / o ratio @xmath22 and metallicity of the exploding wd . this study showed that a change of @xmath22 alters the energetics of the explosion which results in an off - set of the brightness - decline relation . most prominently , this effect can be identified by a change in the fiducial rise - time to decline relation @xmath23 . the offset in @xmath1 is given by @xmath24 where @xmath25 is the dispersion in the rise time of the fiducial light curve . aldering et al . ( 2000 ) showed that @xmath23 are identical within @xmath26 for the local and distant sample lending strong support for the notion that we need a positive @xmath16 . a change in the metallicity z causes a change in the burning conditions at the outer layers of the wd and it alters the importance of the line blanketing in the blue to the uv . based on detailed calculations , effects of similar order have been found for both the delayed dd and the deflagration scenario ( hwt98 , lentz et al . recent studies showed the additional effect that z will influence the final structure of the progenitor and the resulting lcs ( umeda et al . 1999 , domnguez et al . 2000 , hflich et al . however , the former two studies were restricted to the progenitor evolution whereas the latter included the connection between the progenitor and the lc but it was restricted to a progenitor of @xmath27 and two metallicities , z=0.02 and 0.004 . a more comprehensive study may be useful to eliminate potential problems due to evolution of the progenitors for the determination of the cosmological equation of state , and it may provide a direct link to the progenitors of sn ia . in this work we connect @xmath0 and and the initial metallicity of the wd to the light curves and spectral properties of sne ia for the entire range of potential progenitors . in section 2 we discuss the evolutionary properties of our models . in section 3 , the results are presented for the explosion , nucleosynthesis , the light curves and spectral properties . in the final , concluding section , our model calculations are related to observations , and we discuss constraints for the progenitors and implications for the cosmological equation of state . co white dwarfs are the remnants of the evolution of low and intermediate mass stars ( becker & iben 1980 ) . their progenitors are stars less massive than @xmath28 , which is the lower stellar mass for which a degenerate carbon ignition occurs after the central helium exhaustion . the precise value of @xmath28 depends on the chemical composition ( see domnguez et al . , 1999 , for a recent evaluation of this mass limit ) . it ranges between 6.5 and 8 @xmath2 . on the base of updated theoretical stellar models of intermediate mass stars , domnguez et al ( 1999 ) found final co core masses in the range 0.55 - 1.04 @xmath2 , in good agreement with semi - empirical evaluations of the wd masses ( see e.g. weidemann 1987 ) . in this paper , we use co wd structures obtained by evolving models with main sequence masses @xmath0 between 1.5 and 7 @xmath2 and metallicities z between @xmath29 and 0.02 . in the following a label identifies a particular progenitor model , namely @xmath30 for a progenitor with a ms mass of @xmath31 and z= @xmath32 . ] these models have been obtained by means of the frascati raphson - newton evolutionary code ( franec ) , which solves the full set of equations describing both the physical and chemical evolution of a star by assuming hydrostatic and thermal equilibrium and a spherical geometry ( chieffi & straniero 1989 , chieffi , limongi & straniero , 1998 ) . for a detailed description of the adopted input physics see straniero , chieffi & limongi ( 1997 ) and domnguez et al ( 1999 ) . because we are interested in the final chemical structure of a co wd , let us recall the main properties and the major uncertainties of the evolutionary phases during which the co core forms . in table 1 we show some properties of our models . in columns 1 to 9 we give : ( 1 ) the initial composition ( z and y ) , ( 2 ) the model name , ( 3 ) the main sequence mass ( in @xmath2 ) , ( 4 ) the mass of the co core at the beginning of the tp phase ( in @xmath2 ) , ( 5 ) the c abundance ( mass fraction ) at the center , ( 6 ) the mass ( in @xmath2 ) of the homogeneous carbon - depleted central region , ( 7 ) the averaged c / o ratio within the final , @xmath331.37 @xmath2 , co white dwarf after accretion , and ( 8) the @xmath18ni mass ( in in @xmath2 ) synthesized during the explosion . the c and o profiles ( mass fraction ) of the co core for selected thermally pulsing models are shown in figs . [ z22 ] to [ corate ] . in particular , fig . [ z22 ] shows the changes induced by a different initial mass , while figs . [ m3 ] and [ m5 ] illustrate the effect of the metallicity . the internal c and o profiles of a wd are generated in three different evolutionary phases of the progenitor , namely : i ) the central he burning , ii ) shell he - burning during the early asymptotic - giant - branch ( agb ) phase , and iii ) shell he - burning during the thermally pulsing agb phase . as illustrated in the figures , they produce three distinct layers . the central he - burning produces the innermost homogeneous layer . this phase is initially dominated by the carbon production via the 3@xmath34 reaction occurring in the center of a convective core . once sufficient @xmath35c is synthesized , the @xmath36 reactions becomes competitive with 3@xmath34 . carbon is partially burned into @xmath37o . since the opacity of a c - o mixture is larger than that of a he mixture , the extension ( in mass ) of the convective core increases with time . when the he mass fraction in the convective core is reduced down to @xmath330.1 , the he - burning is mainly controlled by @xmath36 , and most of the oxygen in the convective core is synthesized during the late he - burning . the final abundances in the innermost region of a wd is strongly dependent on the duration of the last @xmath38 of the entire he - burning lifetime . in column 7 of table 1 , we report the size ( in solar masses ) of this innermost homogeneous region , which corresponds to the maximum extension of the convective core . the intermediate region of the final c / o structure is characterized by a rising carbon abundance . it is produced during the early - agb when the he - burning shell advances in mass until it approaches the h - rich envelope . the amount of carbon ( oxygen ) left behind increases ( decreases ) due to the progressive growth of the temperature in the shell which favors the 3@xmath34 reactions with respect to the @xmath34 capture on @xmath35c . in addition , the short lifetime does not allow a substantial conversion of carbon into oxygen . finally , a thin external layer is built up during the thermally pulsing agb . at the beginning of a thermal pulse , the large energy flux is locally produced by the 3@xmath34 reactions . it induces the formation of a convective shell that rapidly overlaps the whole inter - shell region . owing to the large he reservoir , a huge amount of carbon is produced at the base of the convective shell . after few years ( 10 - 100 yr depending on the core mass ) the convective shell disappears and a quiescent he - burning takes place . it is during this longer phase that the @xmath36 reactions convert a certain part of the carbon produced during the pulse into oxygen . the c / o ratio left below the he - rich layer by the he burning - shell depends on the rate of the @xmath34 captures on carbon . note that the outer blip in the carbon profile is the result of the last thermal pulse where the he has not yet fully depleted . the size , in mass , of this third layer depends on the duration of the thermally pulsing agb phase . although it is influenced by the assumed mass loss rate , it is generally believed that for m@xmath393 m@xmath40 the co core can not increase more than 0.1 - 0.2 m@xmath40 during the agb . an even smaller increase of the co core is expected for larger @xmath0 . the subsequent phase has been calculated by accreting h / he rich material on the resulting co wd . note that we have assumed that the progenitor ejects its h / he - rich envelope prior to the onset of the accretion epoch . the accreted matter has a final c / o ratio of @xmath331 . when the star reaches a mass close to 1.37 m@xmath40 , ignition occurs close to the center . @xmath41 has been adjusted to enforce that the thermonuclear runaway occurs at the same central density @xmath42 in all models . in fig . [ z22 ] , the chemical structures of our models are shown as a function of @xmath0 ( for z=0.02 ) . for low stellar masses , the core he - burning happens under lower central temperatures ( see e.g. domnguez et al . this favors the @xmath34 captures on @xmath35c , which are in competition with the 3@xmath34 resulting in a slightly smaller central c / o ratio for low @xmath0 . however , the size of the region of central he - burning , @xmath43 , is increasing with @xmath0 . in a star with 7 m@xmath40 , the maximum size of the convective core is about 0.7 m@xmath40 while , in the 1.5 m@xmath40 model , it is only 0.25 m@xmath40 . this is the dominating factor for changes in the mean c / o ratio which , in general , produces the monotonic relation that @xmath22 decreases with increasing @xmath0 ( table 1 ) . in figs . [ m3 ] and [ m5 ] , the chemical structures are given for various z for stars with 3 and 5 @xmath2 , respectively . the sizes of the innermost homogeneous region is not a monotonic function of z. this is due to the peculiarity of very low metallicity intermediate mass stars ( see chieffi et al . 2001 ) . in these stars ( z@xmath44 ) , the central hydrogen burning proceeds via the pp - chain instead of the cno - cycles , as it happen for larger metallicity . it results in a smaller he - core ( ezer & cameron 1971 , tornamb & chieffi 1986 ) . for solar metallicity , the higher opacities and the steeper temperature gradients produce a smaller he core ( becker & iben 1979 , 1980 ) . in all cases the dependence on z of the final averaged c / o ratio ( column 7 of table 1 ) is small compared to the effect of main sequence mass . the variation with z ranges between 5 and 10@xmath45 . for the final chemical structure , the most important uncertainty is due to the ambiguity in nuclear reaction rate of @xmath36 . the innermost region is also sensitive to the treatment of turbulent convection which may affect the duration of the late central he - burning lifetime and the size of the convective core . the rate of the @xmath36 at astrophysical energies is not well established ( see e.g. buchmman 1996 , 1997 ) . the cross section around the gamow peak is dominated by ground state transitions through four different processes : the e1 amplitudes due to the low - energy tail of the @xmath46 resonance at @xmath47 and to the subthreshold resonance at @xmath48 , and the e2 amplitudes due to the @xmath49 subthreshold resonance at @xmath50 and to the direct capture to the @xmath51 ground state , both with the corresponding interference terms . besides ground state transitions , also cascades , mainly through the e2 direct capture to @xmath52 @xmath53 and @xmath54 @xmath55 states , have to be considered . obviously a higher rate ( @xmath56 factor of 2 ) of this reaction reduces the carbon abundance left by both the core and the shell he - burning . some indications in favor of an high value for this reaction rate comes from the rise times to maximum light in sne ia ( hwt98 ) , and from recent studies of pulsating wds ( metcalfe , winget & charboneeau 2001 ) . concerning turbulent convective mixing , it only affects the region of the wd structure produced during the core he - burning . the two major uncertainties are related to the possible existence of breathing pulses ( see castellani et al . 1985 , caputo et al . 1986 ) and the possible existence of a sizeable convective core overshoot . as recently pointed out by imbriani et al . ( 2001 ) , since breathing pulses increase the late core he - burning lifetime , they significantly reduce the central c / o . in this regard , they mimic the effects of a high @xmath36 rate . on the contrary , convective core overshoot does not affect the central c / o , but it may enlarge the size of the convective core and , in turn , may produce a larger central , c - depleted region . in the models presented in this paper , we have neglected breathing pulses and convective core overshoot . all the models , except the 3p0z13lr , have been obtained by means of a high rate for the @xmath36 reaction ( as given by caughlan et al . , 1985 ) . the 3p0z13lr model has been obtained by adopting the alternative low rate presented by caughlan & fowler ( 1988 ) . a comparison between the two models with different @xmath36 reaction rate is shown in fig . [ corate ] . changing the @xmath36 rate from the high to the low value drastically alters the chemical profiles of the progenitor . the carbon abundance increases by about a factor two . at the time of the explosion , the average composition of the wd changes from oxygen - rich ( @xmath57 ) to carbon - rich ( @xmath58 ) ! the consequences for the lcs are discussed below . although a variation in the assumed convective mixing scheme may change the quantitative result of our analysis , the overall conclusions and tendencies can not be significantly altered because its influence is limited to the innermost part of the pre - explosive structure . spherical dynamical explosions and corresponding light curves are calculated . we consider delayed detonation ( dd ) models , because these have been found to reproduce the optical and infrared light curves and spectra of sne ia reasonably well ( hflich 1995b ; hkw95 ; hk96 ; nugent et al . 1997 ; wheeler et al . 1998 ; lentz et al . 2000 ; hflich et al . 2000 ; gerardy et al . model parameters have been chosen which allow to reproduce light curves and spectra of typical type ia supernovae . for our set of models , the differences can be attributed to changes in the progenitor structure of the co - wd . as reference , we use the explosion of a progenitor with 5 @xmath2 at the main sequence and solar metallicity ( model 5p0z22 ) . at the time of the explosion of the wd , its central density is 2.0@xmath59 g/@xmath60 and its mass is close to 1.37@xmath2 . the transition density @xmath20 from deflagration to detonation is chosen to be 2.3@xmath61 g/@xmath60 . explosion models are calculated using a one - dimensional radiation - hydro code ( hk96 ) that solves the hydrodynamical equations explicitly by the piecewise parabolic method ( colella & woodward 1984 ) . nuclear burning is taken into account using an extended network of 606 isotopes from n , p to @xmath62 ( thielemann , nomoto & hashimoto 1996 and references therein ) . the propagation of the nuclear burning front is given by the velocity of sound behind the burning front in the case of a detonation wave and in a parameterized form during the deflagration phase calibrated by detailed 3-d calculations ( e.g. khokhlov 1995 , 2000 ; niemeyer & hillebrandt 1995 ) . we use the parameterization as described in domnguez & hflich ( 2000 ) . for a deflagration front at distance @xmath63 from the center , we assume that the burning velocity is given by @xmath64 , where @xmath65 and @xmath66 are the laminar and turbulent velocities with @xmath67}\ ] ] here @xmath68 is the atwood number , @xmath69 is the characteristic length scale , and @xmath70 and @xmath71 are the densities in front of and behind the burning front , respectively . the quantities @xmath34 and @xmath69 are directly taken from the hydro at the location of the burning front and we take @xmath72 . the transition density is treated as a free parameter . the description of the deflagration front does not significantly influence the final structure of the explosion ( domnguez & hflich 2000 ) . the total @xmath18ni production is governed by the pre - expansion of the wd and , consequently , is determined by the transition density @xmath20 , at which the burning front switches from the deflagration to the detonation mode ( h95 ) . from the physical point of view , @xmath20 should be regarded as a convenient way to adjust the amount of material burned during the deflagration phase . the value @xmath20 can be adjusted to produce a given amount of @xmath21 . this code includes the solution of the frequency - averaged radiation transport implicitly via moment equations , expansion opacities , and a detailed equation of state ( see sect . as expected from previous studies ( see introduction ) , the overall density , velocity and chemical structures are found to be rather insensitive to the progenitor , including the production of elements . although explosions and light curves have been calculated for the entire set of stellar cores , we will concentrate our detailed discussion on the extreme cases and the reference model . results for intermediate models can be understood accordingly and interpolated using the quantities given in table 1 . the final density , velocity and chemical structures and detailed production of isotopes are shown in figs . [ rho ] , [ chem ] and [ isotope ] for the extreme cases in metallicity ( z=0.02 and z=10@xmath73 , 5 m@xmath40 ) and the extremes in @xmath0 ( 1.5 m@xmath40 to 7 m@xmath40 , z=0.02 ) . between 0.511 to 0.589 @xmath74 of @xmath21 are produced ( table 1 ) . the production of individual isotopes varies only by about 10 % ( fig . [ isotope ] ) . for the reference model @xmath75 , the final element abundances are given in table 2 . variations in the final density and velocity structure are correspondingly small ( fig . [ rho ] ) . in delayed detonations , almost the entire wd is burned . the total release of nuclear energy depends mainly on the fuel , i.e. on the integrated c / o ratio @xmath22 ( hwt98 ) . however , as usual for delayed detonation models , the deflagration phase is key for our understanding of the final results . during the deflagration phase , about @xmath76 of fuel are burned in our models ( figs . [ rho ] & [ chem ] ) . in all explosions but the progenitors with @xmath77 with z=0.02 and @xmath78 with z=0.001 , the deflagration front will propagate in the carbon - depleted layers . the amount of total energy produced during the deflagration phase and the binding energy of the progenitor determines the pre - expansion of the outer layers and , consequently , the overall chemical structure . the binding energy of the wd is dominated by the central density @xmath42 at the time of the explosion . note that the c / o ratio has little influence on the structure of the wd because the pressure is dominated by degenerate electrons and the electron to nucleon ratio @xmath79 is identical for @xmath80 and @xmath81 . the total energy production during the deflagration phase is governed by @xmath20 and by the nuclear energy release per gram , i.e. the composition . both @xmath82 and @xmath20 have been kept the same in all models . variations can be understood by the change of the mean c / o ratio and the mass @xmath83 of the central , carbon depleted region . the latter influences the temperature and , consequently , the laminar speed and the atwood number ( eq . 1 ) . at the central layers , all the material is burned up to iron - group elements ( fig . [ chem ] ) . some additional variation in the total @xmath21-mass is caused by the c / o ratio of the matter burned during the detonation phase . for all models , the transition between @xmath21- and si - rich layers is between 0.58 and 0.98 @xmath2 . for @xmath84 and , to a much lesser extend , for @xmath85 , this transition region overlaps with layers of lower c - depleted ( fig . 1 ) . from the nuclear physics , the transition between complete and incomplete si burning occurs in a narrow temperature range around @xmath86 . a locally lower c / o ratio results in lower burning temperatures . consequently , the @xmath87/si boundary is shifted inwards and @xmath19 is reduced . overall , differences between the models remain small because the nuclear energy production for c- and oxygen burning differs only by @xmath88 . an increasing @xmath0 for a given metallicity leads to changes in @xmath89 and @xmath22 by 33% and -25% , respectively ( table 1 ) . increasing @xmath0 from 1.5 to 7.0 @xmath2 alters the expansion velocity of a given mass element ( fig . [ rho ] , right panel ) and it results in a shift of the chemical interfaces between complete , incomplete si and explosive c burning by @xmath90 ( fig . [ chem ] ) . if we decrease the metallicity from solar ( @xmath91 ) to @xmath92 for stars with @xmath93 , @xmath89 and @xmath22 change by as little as + 10 % and -7 % , respectively . this means that the density and velocity structure of the chemical structures are virtually indistinguishable ( fig . [ rho ] ) . for stars with @xmath0 @xmath94 , variations in @xmath89 increase up to 30 % but , still , variations in @xmath22 remain at a level of 10 % . the overall energetics , density and velocity structure remain mostly unchanged but the pre - expansion and , consequently , the chemical interfaces between different regimes of burning change by @xmath95 . for all @xmath0 , the most noticeable difference is the increasing @xmath96 production with z in the layers of incomplete si burning which changes the spectra in the blue , and in the uv ( see hwt98 and below ) . as mentioned in sect . 2.3 , there is some indirect evidence for a high cross section of this key reaction but a low rate can not be ruled out either . the consequences of a low rate are strong . e.g. , for a progenitor with @xmath98 and @xmath99 , the low rate suggested by caughlan & fowler ( 1988 ) will increase @xmath89 and @xmath22 from 0.26 to 0.51 and 0.74 to 1.22 when compared to our favorite rate ( caughlan et al . the explosion becomes more energetic by about 20 % and the deflagration front propagates faster . the result is an increase of the @xmath100 production by about 10 % and a shift in the chemical interfaces by about + 2500 km / sec . based on the explosion models , the subsequent expansion , bolometric and monochromatic light curves are calculated ( hflich et al . 1998 , and references therein ) . the lc - code is the same used for the explosion , except that @xmath101 ray transport is included via a monte carlo scheme and nuclear burning is neglected . in order to allow a more consistent treatment of the expansion , we solve the time dependent , frequency averaged radiation moment equations . the frequency - averaged variable eddington factors and mean opacities are calculated by solving the frequency - dependent transport equations . about one thousand frequencies ( in one hundred frequency groups ) and about nine hundred depth points are used . at each time step , we use t(r ) to determine the eddington factors and mean opacities by solving the frequency dependent radiation transport equation in a comoving frame and integrate to obtain the frequency averaged quantities . the averaged opacities have been calculated assuming local thermodynamics equilibrium ( lte ) . both , the monochromatic and mean opacities are calculated in the narrow line limit . scattering , photon redistribution and thermalization terms used in the light curve opacity calculations are taken into account . in previous works , the photon redistribution and thermalization terms have been calibrated for a sample of spectra using the formalism of the equivalent two level approach ( h95 ) . here , for increased consistency , we use the same equations and atomic models but solve the rate equations simultaneously with the light curves calculation at about every 100@xmath102 time step , on the expense of some simplifications in the nlte - part compared to h95 . for the opacities , we use the narrow line limit and for the radiation fields , we use the solution of the monochromatic radiation transport using @xmath103 frequency groups . both the old and new approach are about equivalent in accuracy with consistent results . most noticeable , now , b - v is bluer by about @xmath104 . in the following discussion is based on the same set of models used in the previous section . in figs . [ lc ] and [ dep ] , we show the b and v light curves and some quantities at maximum light . overall the different phases of light curves can be understood in the usual way including the bump at about day 35 which can be attributed to the change in the opacities between the layers of complete and incomplete burning ( domnguez 1991 , 1994 ) . for the reference model 5p0z22 , a maximum brightness @xmath105 of @xmath106 is reached at about 18.25 days after the explosion . the color index @xmath107 is @xmath108 . as discussed in the introduction , the amount of @xmath18ni , its distribution and the expansion rate of the envelope are the dominant factors which determine the absolute magnitude at maximum and the light curve shape . with all model parameters fixed but the progenitor mass and the metallicity , the differences of the light curves can be understood based on the previous discussion of the explosion models . by increasing @xmath0 from 1.5 to 7.0 @xmath2 for z=0.02 , both @xmath109 and @xmath105 decrease by @xmath110 consistent with a change in the @xmath21 mass by 14 % ( fig . [ lc ] , upper panel , and fig . [ dep ] ) . the similarity in the density and velocity structures produces almost identical conditions at the photosphere . b - v is insensitive to a change in @xmath0 ( @xmath111 \leq 0.01^m$ ] ) . relative to the reference model , the rise times vary between -0.5 d ( 1p5z22 ) to + 1.2d ( 7p0z22 ) . the decline rate @xmath4 is hardly affected . a change in @xmath0 will result in an offset / dispersion in @xmath1 by up to @xmath112 . interestingly , the fluxes on the radioactive tail are much more similar than could be expected from the spread in the @xmath21 masses by 14 % ( fig . [ lc ] ) . the change in @xmath19 is almost compensated by the differences in the energy deposition of @xmath113-rays from the radioactive decay . in fig . [ gam ] , the escape probability for hard radiation is shown as a function of time . a significant fraction of @xmath113 photons is thermalized up to about 150 to 200 days . the actual value of thermalization depends on the expansion rate which is decreasing with mass ( see above ) . e.g. the fraction of thermalized @xmath113 photons for the models 1p5z22 , 5p0z22 and 7p0z22 are 24.2 % , 25% and 27 % , respectively . the increase in the efficiency for the thermalization amounts to 11 % over the mass range and almost compensate the decrease in the @xmath21 mass . note that the ratio between maximum and tail brightness is decreasing with @xmath105 . this effect is opposite to the observed @xmath1 relation ( e.g. hamuy et al . if realized in nature , a wide range in @xmath0 would increase the dispersion in @xmath114 by about @xmath112 . the presence of this effect would reveal itself by an additional change in the expansion velocity measured by the doppler shift of lines . e.g. at maximum light , weak lines would indicate an expansion velocity at the photosphere which is smaller by @xmath115 if we compare model 7p0z22 vs. 1p5z22 . the discussion above applies to all metallicities because @xmath89 and @xmath22 vary in a similar range . for progenitors with @xmath116 , a change in the metallicity has little effects on the energetics and , consequently , on the light curves ( fig . [ lc ] , middle panel ) . the most important effect is the systematic decline of b - v by @xmath117 when z is changed from 0.02 to @xmath29 . this effect can be attributed to a change of the line blending by fe in the decoupling region of photons , i.e. the atmosphere ( hwt98 , lentz et al . 2000 ) . in b and v at maximum light , opacities are dominated by electron scattering ( hkm93 ) but iron lines are more important in b compared to v. consequently , a lower metallicity results in increase of the flux ratio between b and v. in the u - band and the uv , the opacities are dominated by lines . a change in the line blending will cause both a change of the radius of the flux formation and the specific flux . therefore , a decrease in z may result in either an in- or decrease of the monochromatic flux depending on the density structure . these findings apply to all ms - masses in our sample . to some extend , the exception are the models with @xmath118 for which the central carbon concentration varies with metallicity and produces a change in @xmath19 by 3 % and a corresponding change in @xmath109 and @xmath105 . a low nuclear rate @xmath119 increases the explosion energy compared to our standard rate ( 3p0z13lr vs. 3p0z13 , see fig . [ lc ] , lower panel ) . the rise time in v is reduced by 2.7 days ( 15.3d for 3p0z13lr vs. 18.0d for 3p0z13 ) . the enhanced escape probability for @xmath113-rays ( fig . [ gam ] ) explains the remaining differences including the increased maximum brightness to tail ratio and the moderate increase of @xmath105 and @xmath109 . these results are consistent with previous findings which identified the importance of the c / o ratio for the change in the rise time of typical sne ia ( hwt98 ) . using a delayed detonation model and realistic structures for the exploding white dwarf , we have studied the influence of the progenitor star on the light curves and spectral properties of type ia supernovae . * stellar models : * we considered stars between 1.5 to 7 @xmath2 and metallicities between @xmath91 ( solar ) to @xmath120 which covers the full range of potential progenitors . the progenitor structures are based on detailed calculations for the stellar evolution starting at the pre - main sequence up to the thermal pulses when most of the stellar envelope is ejected and a white dwarf is formed with a mass between 0.5 and 1.0 @xmath74 . its size increases with @xmath0 and , to a lesser extend , changes with the metallicity . the subsequent accretion and burning at the surface of the wd let it grow to @xmath121 . as a final chemical structure , the wd shows a central region of reduced c abundance between 0.21 to 0.32 originating from the convective he - burning , a layer of increased c abundance from the he - shell burning , and a layer originating for the accretion phase . the mean c / o ratio decreases by about 30 % over the entire mass range . the sensitivity on the metallicity is much weaker ( @xmath122 ) , and not monotonic . * supernovae : * our study of sne ia is based on delayed detonation models because they have been found to reproduce the monochromatic light curves and and spectra of sne ia reasonably well including the brightness decline relation @xmath1 . deviation from a perfect relation are due to variations in the central density , properties of the deflagration front , and the progenitor structure . all parameters but the progenitors have been fixed to produce lcs and spectra typical for normal sne ia . in this work , rise times to maximum light are between 17.7 to 19.4 days , @xmath123 to @xmath124 , and @xmath125 to @xmath126 . differences between the models and light curves remain small because the nuclear energy production by burning carbon and oxygen to iron - group elements differs by as little as @xmath88 . the change of @xmath0 is the decisive factor to change the energetics . the @xmath87 production varies by about 14 % and the velocities of the various chemical layers differ by up to 1500 km / sec . a change in the metallicity hardly affects the overall structure of the progenitor . as already discussed in detail in hwt98 , its main effect is a change in the production of @xmath127 in the outer layers of incomplete si burning . @xmath0 alters the @xmath1 relation which may be offset by up to @xmath5 . in addition , @xmath0 changes the flux ratio between maximum light and the radioactive tail , and it alters the photospheric expansion velocities @xmath128 measured by the doppler shift of lines . e.g. a change in @xmath129 by @xmath5 is coupled to a decrease in @xmath130 at maximum light by @xmath131 . note that a change in the central density @xmath42 of the wd has a similar effect on @xmath129 but with the opposite sign for @xmath132 ( hflich , 2001 ) . in principle , this allows to decide whether differences in @xmath129 between sn with similar @xmath1 are related to a change in the progenitor or the central density at the thermonuclear runaway which is sensitive to the accretion rate . in contrast to @xmath0 , the metallicity z hardly changes the light curve shapes ( @xmath133 ) . it alters the line blocking by iron group elements at the photosphere mainly in the uv , u and b but hardly in v ( hwt98 ) . in the models presented here , b - v becomes systematically bluer with decreasing z ( up to @xmath134 ) . because b - v is the basic color index used to correct for interstellar extinction , the metallicity effect can systematically alter the estimates for the absolute brightness by up to @xmath5 . * @xmath135 : * at the example of a progenitor with @xmath136 and z=0.001 , we have tested the influence of the low nuclear rate @xmath119 on the outcome . using the lower rate suggested by caughlan & fowler ( 1988 ) instead caughlan et al . ( 1985 ) results in more energetic explosions because @xmath22 increases by a factor of @xmath137 . the rise times to maximum light are 15.3d instead of 18.0d . from detailed observations of nearby supernovae , riess et al . ( 1999 ) find the following relation between the rise time @xmath138 and the maximum brightness @xmath139}\ ] ] the theoretical lcs peak at @xmath140 and @xmath141 for @xmath142 and @xmath143 , respectively . from the empirical fit of riess , we would expect a rise time between 17.6 and 18.4 days which favors the high rate @xmath119 of caughlan et al . note that the uncertainties in absolute values for the rise times are @xmath56 1 to 2 days ( hwt98 ) . * constraints on the progenitor : * empirically , the @xmath1 has been well established with a rather small statistical error @xmath144 ( @xmath145 : riess et al . 1996 ; @xmath146 : schmidt et al . ( 1998 ) ; @xmath147 phillips 1999 ; @xmath148 riess et al . 1999 ; @xmath149 perlmutter et al . 1999a ) . from theoretical models , a spread of 0.3 to @xmath150 can be expected ( hflich et al . 1996 ) . this may imply a correlation between free model parameters , namely the properties of the progenitors , the central density or the transition density @xmath20 . in this study , we find a spread in @xmath1 of about @xmath5 for progenitors with @xmath0 between @xmath151 to @xmath152 . this may suggest a more narrow range in @xmath0 for realistic progenitors . from the fiducial rise time , progenitors with @xmath0 @xmath153 are favored . this number should be regarded as a hint because uncertainties in the lc models . another constraint can be obtained from the observed spread in the rise - time to decline relation . riess et al . ( 1999 ) find a spread in @xmath138 of @xmath154 days whereas our models show a spread of @xmath56 1.7 days for @xmath155 . to be consistent with the observations , the range of main sequence masses has to be reduced by a factor of two . this range is an upper limit because additional variations in the population of sneia such as explosion scenarios or the central density of the wd at the time of ignition will likely result in lower correlations . * the cosmological equation of state : * in the following , we want to discuss our results in context of sne ia as probes for cosmology and for the determination of the cosmological equation of state . we limit the discussion to the effects due to different progenitors . for a discussion of other systematic effects such as grey dust , gravitational lensing , the influence of a change in the importance of different possible scenarios ( e.g merger vs. @xmath121 models ) etc . we want to refer to the growing literature in this field ( e.g. schmidt et al . 1998 , hwt98 , perlmutter et al . 1999a ) . recently , there has been strong evidence for a positive cosmological constant ( e.g. perlmutter et al . 1999a , riess et al . this evidence is based on observations that sne ia in the redshift range between 0.5 to 1.2 appear to be dimmer by about @xmath156 for redshifts between 0.5 to 0.8 which is comparable to the variations produced by different progenitors . however , both the internal spread in @xmath1 ( see above ) and the similarity in @xmath1 between the local sne ia and the high - z sample ( @xmath157 , aldering et al . 2000 ) limit the likely range of models and , consequently , evolutionary effects to @xmath158 up to redshifts of @xmath159 . in addition , we do not expect a drastic change in the metallicity between local and supernovae at @xmath160 . taking the linear dependence of b - v on the metallicity ( fig . [ dep ] ) and realistic ranges for z , also reddening of non - grey dust will not change the conclusion on the need for a positive cosmological constant . as discussed in the introduction , the quest for the nature of the dark energy is one of the central questions to be addressed in future ( e.g. white 1998 , perlmutter et al . 1999b , albrecht & weller 2000 , ostriker & steinhardt 2001 ) . for @xmath161 , we can expect both very low metallicities and a significant change of the typical @xmath0 . from this study , systematic effects due to different progenitors are limited to @xmath162 . therefore , without further corrections for the progenitor evolution , some of the alternatives for the nature of the dark energy may be distinguished without correction for evolution . however , for a more detailed analysis , an accuracy of about @xmath163 ( weller & albrecht 2001 ) is required . in this paper , we have shown how a combination of spectral and lc data or different characteristics of the lc can help to achieve this goal . * limitations : * finally , we also want to mention the limitations of this study . qualitatively , our results on the z dependence agree with a previous study ( hflich et al . 2000 ) which was based on a progenitor with @xmath164 calculated by nomoto s group ( umeda et al . the relation between @xmath22 and the offset in the @xmath1 relation has been confirmed . however , the influence of z on @xmath22 was found to be about twice as large , and the central c concentrations are systematically larger . the differences point towards a general problem . the details of the central structure and evolution in the convective he burning core depend sensitively on the treatment of convection , semi - convection , overshooting , and breathing pulses ( lattanzio 1991 , schaller et al . 1991 , bressan et al . 1993 , vassiliadis & wood 1993 ) . for a detailed discussion , see domnguez et al . in particular , the central c concentration may vary between 0.1 and 0.5 . we want to note , that our value is consistent with direct measurements of the central c / o ratio found by the analysis of pulsational modes of wds ( metcalfe et al . these uncertainties will affect the efficiency to separate the contribution of @xmath0 and z , respectively , i.e. the reason for @xmath22 but not the observable relations for lcs and spectra . we provide limits on the size of evolutionary effects due to @xmath0 and metallicity of the progenitor ( @xmath9 ) . other evolutionary effects may be due to a systematic change in the dominant progenitor scenario ( e.g. mergers vs. single degenerate ) or the typical separation in binary systems which contribute to the sne ia at a given time . the latter may alter the accretion rates and , consequently , the central densities of the wd at the time of the thermonuclear runaway . we did not consider the effect of the progenitor and of its pre - conditioning just prior to the explosion on the propagation of the burning front and , in particular , on the deflagration to detonation transition or , alternatively , the phase of transition from a slow to a very fast deflagration ( hillebrandt , 1999 private communication ) . although the model parameters have been chosen to allow for a representation of typical " sne ia , more comprehensive studies and detailed fitting of actual observations are needed e.g. to detangle effects due to a change in the ignition density vs. the progentor . in particular , observations of local sne ia have to be employed to narrow down and test for the proposed range of flux ratio between maximum and tail , and its relation to the expansion velocities . * acknowledgments : * this work has been supported by nasa grant nag5 - 7937 , the murst italian grant cofin2000 , by the mec spanish grant pb96 - 1428 , by the andalusian grant fqm-108 and it is part of the italy - spain integrated action ( murst - mec agreement ) hi1998 . the calculations for the explosion and light curves were done on a beowulf - cluster financed by the john w. cox - fund of the department of astronomy at the university of texas . albrecht a. , weller j. 2000 , american astronomical society , 197 , 61 aldering g. , knop , p. , nugent p. 2000 , 119 , 2110 becker s.a . , iben i. jr . 1979 , 232 , 831 becker s.a . , iben i. jr . 1980 , 273 , 111 branch d. , romanishing w. , baron e. 1996 , 465 , 73 & e467 , 473 bressan a. , fagotto f. , bertelli g. , chiosi c. 1993 , a&as 100 , 647 buchmann l. 1996 , 468 , l127 buchmann l. 1997 , 479 , l153 caputo f. , castellani , v. , chieffi a. , pulone l. , tornamb a. 1989 , 340 , 241 caughlan g.r , fowler a.m. , harris m.j . , zimmerman b.a . 1985 , at . data nucl . dat a tables 32 , 197 caughlan g.r , fowler a.m. 1988 , atomic data nucl . table , 40 , 283 chieffi a. , dominguez i. , limongi m. , straniero o. 2001 , apj , in press chieffi a. , straniero o. 1989 , 71 , 47 chieffi a. , limongi m. , straniero o. 1998 , 502 , 737 collela , p. , & woodward , p.r . 1984 , j.comp.phys . , 54 , 174 domnguez i. , chieffi a. , limongi m. , straniero o. 1999 , 524 , 226 domnguez i. , & hflich p. 2000 , apj 528 , 854 domnguez i. , hflich p. , straniero o. , wheeler c.j . 2000 , in : _ nuclei in the cosmos _ , eds . n. pranzos & s. harissulos , edition frontiers , paris , p. 259 domnguez i. 1994 , in : abstracts of the conferences on thermonuclear supernovae , eds . p. ruiz - lapuente , r. canal & j. isern . domnguez i. 1991 , phd thesis , university of barcelona , isbn 84.7875.736.8 ezer d. , cameron a. 1972 , apss 14 , 399 fisher a. , branch d. , hflich p. , khokhlov a. 1998 , 494 , 47 garnavich p.m. et al . 1998 , apj 509 , 74 gerardy , c.l , hflich p. , fesen r. , & wheeler j.c . 2001 , apj in preparation hamuy m. , phillips m.m , schommer r.a . suntzeff n.b , maza j. , aviles a. 1996a , 112 , 2391 hamuy m. , phillips m.m . , suntzeff n.b . , schommer r.a . , maza j. , aviles a. 1996b , 112 , 2398 hamuy m. et al . 2000 , 120 , 1479 harkness r. 1987 , in : 13th texas symposium on relativistic astrophysics , world scientific publishing co. , p. 413 hatano k. , branch d. , lentz e.j . , baron e. , filippenko a.v . , garnavich p. 2000 , apj 543l , 49 hillebrandt w. 1999 , private communication at the aspen workshop on sne ia hflich p. 2001 , , in preparation hflich p. , nomoto , k. , umeda h. , wheeler j.c . 2000 , 528 , 590 hflich p. , wheeler j.c . , thielemann f.k 1998 , 495 , 617 [ hwt98 ] hflich p. , khokhlov a. 1996 , 457 , 500 [ hk96 ] hflich , p. , khokhlov a. , wheeler j.c . , phillips m.m . , suntzeff n.b . , hamuy m. 1996a , apj 472l , 81 hflich , p. ; dominik , c. ; khokhlov , a. ; mller , e. ; wheeler , j.c . 1996b , @xmath165 texas symposium on relativistic astrophysics , annals of the new york academy of science 759 , 348 hflich , p. 1995 , 443 , 89 [ h95 ] hflich p. , khokhlov a. , wheeler j.c . 1995 , 444 , 211 [ hkw95 ] hflich p. , khokhlov a. , wheeler j.c . 1993 , a&a , 270 , 223 hoyle p. , fowler w.a . 1960 , 132 , 565 iben i.j , tutukov a.v . 1984 , apjs 54 , 335 iben i.j 1972 , 178 , 433 ivanov v.d . , hamuy , m. , pinto , p.a . , 2000 , 542 , 588 iwamoto , k. , brachwitz , f. , nomoto , k. , kishimoto , n. , umeda , h. , hix w.r . , thielemann f.k . 1999 , apjs 125 , 439 khokhlov a. 2001 , , in press & astro - ph/0008463 khokhlov a. 1995 , 449 , 695 khokhlov , a. , mller , e. , hflich p. 1993 , a&a 270 , 223 khokhlov a. 1991 , 245 , 114 lattanzio j. 1991 , apjs 76 , 215 lentz e.j . , baron e. , branch d. , hauschildt p. , nugent p.e . 2000 , apj 530 , 966l metcalfe t.s . , winget d.e . , charbonneau p. 2001 , , submitted niemeyer , j. c. , & hillebrandt , w. 1995 , apj 452 , 779 nomoto , k. 1980 , apj 248 , 798 nomoto k. , thielemann f .- k . , yokoi k. 1984 , 286 , 644 nugent , p. , baron , e. , hauschildt , p. , & branch . , d. 1997a , 485 , 812 ostriker p. , steinhardt p.j . 2001 , scientific american , jan . 2001 , p. 47 paczyski b. 1985 , in _ cataclysmic variables and low - mass x - ray binaries _ , eds . lamb & j. patterson , ( dordrecht : reidel ) p. 1 perlmutter , s. , et al . 1999a , apj 517 , 565 perlmutter s. , turner m.s . , white m. 1999b , physical review letters 83 , 670 perlmutter c. et al . 1997 , 483 , 565 perlmutter c. et al . 1995 , apj 440 , l95 phillips , m.m . 1993 , 413 , 105 phillips , m.m . , lira p. , sunzeff n.b . , schommer r.a . , hamuy m. , maza j. 1999 , 118 , 1766 renzini a. , fusi pecci f. 1988 , 26 199 riess a.g . , press w.h . , kirshner r.p . 1995 , 438 , l17 riess a.g . , press w.h . , kirshner r.p . 1996 , 473 , 88 riess a.g . , et al . 1998 , 116 , 1009 riess a.g . , et al . 1999 , 118 , 2675 saha a. et al . 1997 , apj 486 , 1 schaller g. , shaerer d. , meynet g. , maeder a. 1992 , a&as 96 , 269 schmidt , b. p. , et al . 1998 , apj 507 , 46 straniero , o. chieffi , a. and limongi m. 1997 , 490 , 425 thielemann f.k . , nomoto k. , hashimoto m. 1996 , 460 , 408 tornamb a. , chieffi . a. 1986 , , 220 , 529 umeda , h. , nomoto , k. 1999 , 513 , 861 vassiliadis e. , wood p.r . 1993 , 413 , 641 wang l. , hflich p. , wheeler j.c . 1997 , 483 , 29 webbink , r. f. 1984 , apj 277 , 355 weller j. , albrecht a. 2001 , _ opportunities for future supernova studies of cosmic acceleration _ , astro - ph/0008314 whelan j. , iben i.jr . 1973 , apj 186 , 1007 wheeler j. c. , hflich p. , harkness r. p. , spyromilio j. 1998 , apj 496 , 908 white m. 1998 , apj 506 , 495 weidemann v. 1987 , a&a 188 , 74 woosley s.e . , weaver t.a . 1994 , 423 , 371 woosley s.e . , weaver t.a . 1986 , 24 , 205 woosley s. e. , weaver t.a . , taam r.e . 1980 , in : type i supernovae , ed . j.c.wheeler , austin , u.texas , p.96 yamaoka h. , nomoto k. , shigeyama t. , & thielemann f. 1992 , apj 393 , 55 llcccccc z=0.02 & 1p5z22 & 1.5 & 0.55 & 0.21 & 0.27 & 0.75 & 0.589 y=0.28 & 3p0z22 & 3.0 & 0.57 & 0.21 & 0.28 & 0.76 & 0.584 & 5p0z22 & 5.0 & 0.87 & 0.29 & 0.46 & 0.72 & 0.561 & 7p0z22 & 7.0 & 0.99 & 0.28 & 0.70 & 0.60 & 0.516 & & & & & & & z=10@xmath166 & 1p5z13 & 1.5 & 0.59 & 0.24 & 0.31 & 0.76 & 0.587 y=0.23 & 3p0z13 & 3.0 & 0.77 & 0.26 & 0.39 & 0.74 & 0.567 y=0.23 & 5p0z13 & 5.0 & 0.90 & 0.29 & 0.58 & 0.66 & 0.541 & 6p0z13 & 6.0 & 0.98 & 0.29 & 0.71 & 0.60 & 0.522 & & & & & & & & low rate & & & & & & & 3p0z13lr & 3.0 & 0.76 & 0.51 & 0.38 & 1.22 & 0.620 & & & & & & & z=10@xmath167 & 3p0z14 & 3.0 & 0.80 & 0.27 & 0.41 & 0.73 & 0.568y=0.23 & 5p0z14 & 5.0 & 0.90 & 0.29 & 0.58 & 0.65 & 0.541 & 6p0z14 & 6.0 & 0.99 & 0.28 & 0.72 & 0.59 & 0.511 & & & & & & & z=10@xmath73 & 5p0z00 & 5.0 & 0.89 & 0.32 & 0.49 & 0.70 & 0.549 y=0.23 & 7p0z00 & 7.0 & 0.99 & 0.31 & 0.59 & 0.62 & 0.525 ccccccc he & c & o & ne & na & mg & si 6.62e-04 & 1.19e-02 & 9.18e-02 & 5.34e-03 & 4.91e-05 & 1.82e-02 & 2.61e-01 & & & & & & p & s & cl & ar & k & ca & sc 2.51e-05 & 1.59e-01 & 4.00e-06 & 3.28e-02 & 1.99e-06 & 3.44e-02 & 1.02e-08 & & & & & & ti & v & cr & mn & fe & co & ni 1.61e-05 & 9.07e-04 & 6.56e-04 & 2.68e-02 & 6.57e-01 & 6.15e-03 & 6.47e-02

we calculate the dynamical tides raised on a non rotating solar type star by a close stellar or planetary companion . dissipation arising from a turbulent viscosity operating in the convection zone and radiative damping in the radiative core are considered . we compute the torque exerted on the star by a companion in circular orbit , and determine the potentially observable magnitude of the tidally induced velocity at the stellar photosphere . these calculations are compared with the results obtained by assuming that a very small frequency limit can be taken in order to calculate the tidal response ( _ equilibrium tide _ ) . for a standard solar model , the latter is found to give a relatively poor approximation at the periods of interest of several days , even when the system is far from resonance with a normal mode . this behavior is due to the small value of the brunt - visl frequency in the interior regions of the convection zone . it is shown that although the companion may go through a succession of resonances as it spirals in under the action of the tides , for a fixed spectrum of normal modes its migration is controlled essentially by the non resonant interaction . we find that the turbulent viscosity that is required to provide the observed circularization rates of main sequence solar type binaries is about fifty times larger than that simply estimated from mixing length theory for non rotating stars . we discuss the means by which this enhanced viscosity might be realized . these calculations are applied to 51 pegasi . we show that the perturbed velocity induced by the tides at the stellar surface is too small to be observed . this result is insensitive to the magnitude of the turbulent viscosity assumed and is not affected by the possibility of resonance . for this system , the stellar rotation and the orbital motion are expected to be synchronized if the mass of the companion is as much as one tenth of a solar mass .

astro-ph9801280

theoretical analyses of the tidal interaction between close binaries can be classified according to whether an equilibrium tide is assumed or the dynamical tide is taken into account . the theory of the equilibrium tide is based on the assumption that a star subject to the tidal disturbance of a companion instantly adjusts to hydrostatic equilibrium ( darwin 1879 ) . a calculation including the dynamical tide takes into account the fact that gravity or g modes can be excited in the convectively stable layers of the star and that resonances between the tidal disturbance and the normal modes of the star can occur ( cowling 1941 ) . so far , dynamical tides have been studied only in massive close binaries , which have a convective core and a radiative envelope ( zahn 1975 , 1977 ; savonije & papaloizou 1983 , 1984 , 1997 ; papaloizou & savonije 1985 , 1997 ; savonije , papaloizou & alberts 1995 ) . in this paper , we examine the effect of dynamical tides excited by a companion on a solar type star , in which a radiative core is surrounded by a convective envelope . this is of particular interest in connection with circularization of solar type binaries . it has been proposed that circularization occurs through the action of turbulent viscosity , originating in the convective envelope , on the tide . however , according to claret & cunha ( 1997 ) ( see also goodman & oh 1997 ) , who have used the equilibrium tide formalism of zahn ( 1989 ) , the circularization rate resulting from this mechanism is too small by about two orders of magnitude to account for the circularization timescales required on the main sequence . the tidal response calculation undertaken here is also of interest in connection with the newly discovered planets , some of which are found to orbit around solar type stars with a period comparable to that of the high order g modes of the star . one such example is 51 pegasi ( mayor & queloz 1995 ; marcy & butler 1995 ) . in these binaries , g mode oscillations are excited by the companion in the radiative region beneath the convective envelope . they become evanescent in the convection zone where they are damped by their interaction with the convective eddies . this dissipation leads to an exchange of angular momentum between the star and the orbit if the stellar rotation and the orbital motion are not synchronized . here we assume that the orbital frequency is initially larger than the rotational frequency of the star . then tidal interaction results in the decay of the orbit and the spin up of the star . if the mass of the secondary companion is considerably smaller than that of the primary , the timescale for orbital decay is smaller than the stellar spin up timescale , and the companion eventually plunges into the primary . but if the mass of the companion is large enough , synchronization may occur before the binary has merged , stopping further orbital decay . estimates based on the theory of the equilibrium tide ( rasio _ et al . _ 1996 ; marcy _ et al . _ 1997 ) suggest that the orbital decay timescale and the stellar spin up timescale for a system like 51 pegasi are longer than the inferred age of the primary if the companion is a jovian like planet . in this paper we examine the effect of resonances on these timescales , and determine the potentially observable magnitude of dynamical tides at the photosphere of a solar type star . we also compare the dynamical tide calculations with the results of an asymptotic analysis we carry out in the limit of small frequencies which should correspond to the adiabatic equilibrium tide theory . the paper is organized as follows : in [ sec : analysis ] , we study the tidal response of the star to the perturbation by a companion in a circular orbit with a period in the range 413 days . in [ sec : linearized ] , we first consider the linear adiabatic response and then , away from resonance with a g mode , extend the analysis using first order perturbation theory to calculate the torque due to dissipation in the convective envelope . this mechanism is then found to be more important than non adiabaticity arising from heat transport in the radiative interior ( i.e. radiative damping ) . however , this is not the case in the vicinity of a g mode resonance . there we also calculate the torque due to non adiabaticity in the radiative core using a wkb treatment of the non adiabatic terms . we find that the torque at effective resonances is mainly determined by radiative damping . an analysis valid for very low frequencies ( _ equilibrium tide _ ) is given in [ sec : asy ] . the orbital circularization timescale for systems initially in non circular orbits can be derived from the response calculations for companions in circular orbits . this is done in [ sec : times ] . we then discuss how this might be used to calibrate the magnitude of the turbulent viscosity required to fit the observations in [ sec : cal ] . numerical calculations are presented in [ sec : results ] . the results assuming the equilibrium tide are given in [ sec : eqres ] . in [ sec : dyres ] we present the results of the dynamical tide calculations . we describe the tidal response of the star to a companion in circular orbit , give the induced velocity at its surface and the tidal torque . we describe the resonances and show that , for the periods of interest of several days , they are not expected to affect the orbital evolution of the binary . in [ sec : comp ] we compare the calculations based on the dynamical and equilibrium tides . we find that , for the standard solar model , at the orbital periods of interest , because of the long timescale associated with convection , the equilibrium tide calculations give a relatively poor approximation to the results of the dynamical tide calculations . we find that the viscosity that is required to provide the observed circularization rates is about 50 times larger than that simply estimated from mixing length theory and discuss the means by which this viscosity might be enhanced in [ sec : calibration ] . however , we note that _ the strength of the resonances for orbital periods larger than @xmath0 days and the perturbed velocity at the surface of the star are insensitive to the magnitude of the turbulent viscosity assumed_. only for periods @xmath1 days is the strength of the resonances decreased by a factor @xmath1 . the observable width of the resonances as well is reduced when the viscosity is increased . we also give the relation between the orbital evolution , circularization and spin up timescales and the orbital frequency in [ sec : fitting ] . finally in [ sec : discussion ] we discuss and summarize our results , applying them to 51 pegasi in [ sec : peg ] . the calculations presented in this section will be applied to close binary systems where the primary is a solar type star and the secondary a stellar or planetary companion . the orbital periods of interest lie in the range @xmath2@xmath3 days . the rotational angular velocity of the primary is assumed to be small compared to the orbital frequency , so that it can be neglected . quadrupolar tidal forcing thus occurs through potential perturbations with periods in the range @xmath4@xmath5 days . when calculating the tidal response well away from a condition of resonance with a g mode , we firstly calculate the tidal response assuming it to be adiabatic throughout the star . first order perturbation theory is then used to calculate the dissipation occurring in the convective envelope . the idea here ( as is borne out by the numerical results ) is that although short wavelength g modes are excited in the radiative core , when they are away from resonance they do not play an important role in comparison to the global component of the tidal response . also the variations in the convective envelope occur on a comparatively long length scale , making the adiabatic approximation a reasonable one . when there is a resonance with a high order g mode , the response becomes one with a very short length scale such that non adiabaticity in the radiative core can not be neglected . however , the modes are of high order such that a wkb treatment of the non adiabatic effects is possible and this is used close to resonance where the normal mode dominates the response . such non adiabatic effects turn out to be more important than the action of turbulent viscosity in the convective envelope , with the torque at significant resonances being determined mainly by non adiabatic effects . the linearized momentum , mass , and energy equations governing the adiabatic response of the non rotating star to the perturbing potential @xmath6 may be written ( see , for example , unno _ et al . _ 1989 ) @xmath7 where @xmath8 is the pressure , @xmath9 is the density , @xmath10 is the entropy , @xmath11 is the lagrangian displacement vector , @xmath12 denotes the lagrangian perturbation and the primed quantities are eulerian perturbations . we use the cowling ( 1941 ) approximation , applicable to stars with high central condensation , which neglects the perturbation to the stellar gravitational potential . we also have the thermodynamic relation @xmath13 where @xmath14 is the temperature and @xmath15 and @xmath16 are the adiabatic exponents of chandrasekhar . this relation together with equation ( [ energy ] ) leads to @xmath17 where @xmath18 with @xmath19 being the brunt - visl frequency and @xmath20 the acceleration due to gravity . following cowling ( 1941 ) , only the dominant quadrupole term is considered in the perturbing potential due to the companion . for a binary system with a circular orbit , this is given in spherical polar coordinates @xmath21 by the real part of @xmath22 where the spherical harmonic @xmath23 with @xmath24 , @xmath25 being the associated legendre polynomial with indices @xmath26 and @xmath27 . here @xmath28 is the orbital angular velocity , @xmath29 where @xmath30 is the orbital separation , and @xmath31 is the mass of the companion . we adopt the same angular and time dependence for the perturbations , so that @xmath32 @xmath33 and @xmath34 are proportional to @xmath35 . the corresponding expression for the lagrangian displacement is @xmath36 y_{n , m}(\theta,\varphi ) e^{- i m \omega t } . \label{xi}\ ] ] the factor @xmath37 will be henceforth taken as read , so that hereafter the perturbations will be taken to depend only on @xmath38 . physical perturbations are then found by taking real parts after inserting this factor . the horizontal displacement @xmath39 is given by the non radial equation of motion ( [ momentum ] ) : @xmath40 this relation together with equation ( [ rhop ] ) allow @xmath41 and @xmath33 to be eliminated from the system ( [ momentum])([energy ] ) . the radial equation of motion ( [ momentum ] ) and the continuity equation ( [ mass ] ) can then be written as a pair of ordinary differential equations for @xmath42 and @xmath39 : @xmath43 \xi_h + \frac{f r^2 \rho}{\gamma_1 p } , \label{ode1 } \\ \frac{d \xi_h}{dr } & = & \frac{1}{r } \left ( 1 - \frac{ap}{m^2 \omega^2 \rho } \frac{dlnp}{dr } \right ) \xi_r - \left ( a + \frac{1}{r } \right ) \xi_h + \frac{afr}{m^2 \omega^2 } , \label{ode2}\end{aligned}\ ] ] the solution of which requires two boundary conditions . at the surface of the star we take a free boundary : @xmath44 , i.e. @xmath45 the boundary condition at @xmath46 , where equations ( [ ode1 ] ) and ( [ ode2 ] ) have a regular singularity , is that the solutions be regular . since at @xmath47 @xmath48 , @xmath49 and @xmath50 , where @xmath51 and @xmath52 are respectively the central pressure and density , this leads to @xmath53 . the interaction between convective motions and the tidal flow is expected to lead to the dissipation of tidally excited waves ( e.g. zahn 1977 ) . we model this effect as arising from a turbulent viscosity . to do this we suppose there is an additional dissipative force per unit mass acting in the convection zone given in spherical coordinates by : @xmath54 where @xmath55 is the flow velocity and @xmath56 is the turbulent viscosity . here , we assume that variation in the radial direction is the most significant and note that the viscous force is defined in such a way as to lead to a positive definite energy dissipation rate . for the turbulent viscosity coefficient , we take ( see , for example , xiong , cheng & deng 1997 ) @xmath57 where @xmath58 and @xmath59 are two constants , @xmath58 being on the order of unity , @xmath60 is the orbital period , @xmath61 is the mixing length , and @xmath62 is the convective timescale . the viscosity is then @xmath63 for small forcing frequency @xmath64 . the factor @xmath65 , where we shall use @xmath66 , allows for a reduction of efficiency of the damping of high frequency oscillations to which the convection can not adjust . a similar prescription with @xmath67 has been proposed by zahn ( 1966 ) and used by zahn ( 1989 ) , whereas @xmath66 has been considered by goldreich & keeley ( 1977 ) and used by campbell & papaloizou ( 1983 ) and goldman & mazeh ( 1991 ) . goodman & oh ( 1997 ) have also recently put forward some arguments in favor of @xmath66 . for the mixing length , we shall take the standard relation @xmath68 and set @xmath69 . in principle , equations ( [ momentum])([energy ] ) should be solved with @xmath70 added on the right hand side of equation ( [ momentum ] ) in the convective envelope . however , this would increase the order of the differential system to be solved and make the numerical calculations much more complicated . instead , we have found it adequate to solve first the adiabatic problem and then to treat the dissipative effect using a first order perturbation theory . this is valid everywhere , except very close to resonances , because dissipation is weak enough so that the imaginary parts of @xmath42 and @xmath39 are much smaller in magnitude than their real parts . thus we solve equations ( [ momentum])([energy ] ) without dissipative terms , and use these ( real ) solutions to calculate the rate of energy dissipation @xmath71 due to convection which is given by : @xmath72 where the integration is over the volume @xmath73 of the convective envelope and the angular dependence of @xmath70 and @xmath55 has to be taken into account . using the relation @xmath74 and the expression ( [ xi ] ) for @xmath11 , we get @xmath75 dr , \label{pt1}\ ] ] where @xmath76 and @xmath77 are respectively the inner and outer radius of the convective envelope . noting that the ratio of the rate of exchange of energy and angular momentum with the orbit is given by the pattern speed of the tidal disturbance , @xmath78 the torque exerted by the companion on the star is given by @xmath79 this torque is positive because the star is non rotating . when the frequency of the tidal wave is equal to that of some adiabatic normal mode frequency of the star , we can no longer use first order perturbation theory because it would give an infinite torque . however , when the frequency @xmath28 is very close to a resonant frequency @xmath80 , the torque will be given by an expression of the form @xmath81 where @xmath82 is an amplitude and @xmath83 is the damping rate for the mode . first order perturbation theory assumes dissipative effects are small in the response calculation . therefore it is valid only for frequencies such that @xmath84 . however , the damping rate , if small , can be found from first order perturbation theory applied , as described above , very close to resonance where the mode dominates the response . then it is given by ( see , for example , goldstein 1980 ) @xmath85 very close to resonance , where @xmath86 is the kinetic energy of the mode : @xmath87 ^2 dv = \frac{24 \pi}{5 } m^2 \omega^2 \int_0^{r_{\sun } } \rho r^2 \left ( \xi_r^2 + 6 \xi_h^2 \right ) dr,\ ] ] the integration being over the volume @xmath88 of the star ( in the first integral , the angular dependence of @xmath55 has to be taken into account ) . in ( [ gamma ] ) , the total energy of the mode is @xmath89 because at resonance there is equipartition between kinetic and potential energy . we calculate @xmath83 using ( [ gamma ] ) as outlined above making sure that @xmath28 is close enough to @xmath80 by checking that @xmath83 remains approximately constant when @xmath28 is slightly changed . to get @xmath90 we fit the torque obtained from first order perturbation theory to the expression ( [ restor ] ) in the region approaching the resonance where still @xmath91 . this procedure works satisfactorily when @xmath83 is small with consequent strong resonances . this appears to be the situation when turbulent viscosity alone is assumed to act . however , radiative damping can not be neglected for resonances at low forcing frequency and this is discussed below . non adiabatic effects become important when the radiative diffusion time across the length scale associated with the tidal response shortens to become comparable to the wave propagation time across the star . in principle , these effects should be taken into account both in the radiative core and above the convection zone . however , since g modes , which are excited in the radiative core , are evanescent in the convective envelope , we do not expect non adiabaticity to play an important role above the convection zone . to take into account non adiabatic effects , equation ( [ energy ] ) has to be modified in the radiative core to : @xmath92 where @xmath93 is the perturbed radiative flux . the radiative flux is given by the radiative diffusion equation @xmath94 where @xmath14 is the temperature and @xmath95 is the radiative conductivity , with @xmath96 being the stefan boltzmann radiation constant , @xmath97 the velocity of light and @xmath98 the opacity . therefore @xmath99 - \mbox{\boldmath $ \nabla$}_h^2 \left ( kt ' \right ) , \label{grflux}\ ] ] where @xmath100 is the non radial component of the operator @xmath101 . we now suppose that close to resonance , the response behaves exactly like a free g mode with very large radial wavenumber @xmath102 so that wkb theory can be used together with the local dispersion relation to evaluate @xmath103 . the dominant term in the right hand side of ( [ grflux ] ) is then @xmath104 , and all the other terms can be neglected . for a high order free g mode , the local dispersion relation gives ( see , for example , unno _ et al . _ 1989 ) : @xmath105 this expression of @xmath102 is derived under the adiabatic approximation . however , since we want to incorporate non adiabatic effects to the lowest order , we do not need to take them into account in evaluating @xmath102 . we have @xmath106 in the radiative core because @xmath107 there . the perturbed temperature @xmath108 can be expressed as a function of @xmath41 and @xmath33 using the thermodynamic relation @xmath109 where @xmath110 and @xmath111 . the equation of state in the radiative core of a solar type star is primarly that of a perfect gas ( we neglect here the radiation pressure which is very small compared to the gas pressure ) . we then have @xmath112 and @xmath113 . using the above , the fact that @xmath114 , @xmath115 and the relation ( [ thermo ] ) , we can recast ( [ nad ] ) under the form @xmath116 where @xmath117 for high order free g modes , we have ( see , for example , unno _ et al . _ 1989 ) @xmath118 which means that @xmath119 in the radiative core of a solar type star at the frequencies of interest . the non adiabatic correction of the term associated with pressure in equation ( [ nad1 ] ) is then very small compared to the non adiabatic correction of the term involving @xmath42 . therefore equation ( [ nad1 ] ) can be approximated by @xmath120 where we have defined @xmath121 . we note that because we have identified the tidal response with a normal mode , this calculation is valid only close to resonances . equation ( [ rhopnad ] ) is similar to ( [ rhop ] ) but with @xmath122 being replaced by @xmath123 . the system of differential equations we have to solve to get the non adiabatic response of the star is then the same as ( [ ode1])([ode2 ] ) but with the following modification in the radiative core . in equations ( [ ode1 ] ) and ( [ ode2 ] ) , @xmath124 where it appears as a coefficient of @xmath42 , has to be replaced by @xmath123 . the system of differential equations so obtained is complex . in general , for the periods we consider , @xmath125 is small . we then calculate both the real and imaginary parts of the response , so that the torque can be computed directly from : @xmath126 where the integral is over the volume @xmath88 of the star . the angular dependence of @xmath6 and @xmath33 has to be taken into account in this expression . equation ( [ torgen ] ) can be recast under the form ( savonije & papaloizou 1983 ) : @xmath127 as in [ sec : torque ] , we can calculate the damping rate @xmath128 in resonances due to non adiabatic effects using ( [ gamma ] ) . here , the rate of energy dissipation is calculated from the torque ( see [ det ] ) . in the limit of small @xmath129 , the following relations may be obtained for the _ adiabatic equilibrium tide _ : @xmath130 and , if @xmath131 : @xmath132 where the ` _ eq _ ' subscript denotes the equilibrium value . using these displacements , the torque may be calculated using equations ( [ pt1 ] ) and ( [ det ] ) . however , we comment that equation ( [ xieq ] ) , which states that the lagrangian perturbation to the pressure is zero , can only be derived in the adiabatic low @xmath129 limit if the brunt - visl frequency is not zero ( in practice , one also requires that the forcing period be short compared to the thermal timescale of the star , but the latter is so long that it can be assumed to be infinite in this context ) . equations ( [ xieq])([weq ] ) do not apply in a finite region where strictly @xmath133 in that case the fluid is locally barotropic , and the displacement can be written as the gradient of a potential : @xmath134.\ ] ] the continuity equation then gives for low frequencies @xmath135 equation ( [ crp ] ) gives a second order differential equation for @xmath136 this applies inside the region where @xmath133 it is possible , using the two available boundary conditions for ( [ crp ] ) , to match @xmath137 given by ( [ xieq ] ) at the boundaries of such a region , but not in general @xmath138 given by ( [ weq ] ) . this means that there will tend to be a discontinuity in the tangential displacement at the boundaries for low frequencies . when @xmath139 is not zero but very small , in particular small compared to @xmath140 which corresponds physically to the convective timescale being much longer than the forcing period , the tidal response more closely matches that given by ( [ crp ] ) than that given by equations ( [ xieq])([weq ] ) . this feature causes a very slow convergence towards the low frequency limiting solution ( _ equilibrium tide _ ) found here , as well as near discontinuous behavior near the inner convective envelope boundary . this is borne out by our numerical results ( see [ sec : velocity ] and [ sec : comp ] ) . the torque , @xmath141 gives the rate at which angular momentum is transferred from the orbit to the star . we can then calculate a tidal evolution ( decay ) timescale of the circular orbit : @xmath142 where @xmath143 is the reduced mass . in principle , the variation of the torque with @xmath28 has to be taken into account for the total decay time to be calculated . however , since the torque increases as the companion spirals in , @xmath144 is mainly determined by the initial position of the companion , and a good estimate can be obtained using the above formula . this exchange of angular momentum also results in the spin up of the star , the timescale of which is given by @xmath145 ( savonije & papaloizou 1983 ) , with @xmath146 being the stellar moment of inertia . out of resonance , angular momentum deposition _ initially _ occurs mainly in the convective envelope where the turbulent viscosity acts ( goldreich & nicholson 1989 ) . it is then of interest to calculate the spin up timescale for the convective envelope alone , which is @xmath147 , where @xmath148 is the moment of inertia of the convection zone . we note however that on the long timescales associated with tidal evolution , angular momentum may be redistributed throughout the star . in practice , we find that the torque arising from a companion in a circular orbit varies with frequency approximately @xmath149 this result can be used to relate the orbital circularization timescale to the initial orbital decay timescale provided that the eccentricity is not too large . in practice , both these timescales can be significantly longer than the spin up timescale of the star due to its relatively small moment of inertia . we should then consider that the star is synchronized with the orbit . the ratio between the orbital decay timescale and the circularization timescale @xmath150 is found to be about 6 for the calculated frequency dependence of the circular orbit torque ( see , for example , the expressions given in savonije & papaloizou 1983 , 1984 ) . this appears to be independent of whether the star is assumed to be synchronously rotating or non rotating in that the circularization rate calculated assuming no rotation , as we do here , gives a reasonable estimate in the synchronous case also . in addition , to evaluate @xmath150 for an equal mass system , we have to take into account the reciprocal torque exerted by the primary on the companion . to do this , we take @xmath150 to be proportional to a factor which is @xmath151 when @xmath152 and 1 when @xmath153 . to a reasonable approximation , we then get : @xmath154 one of the purposes of this paper is to calculate the tidally induced velocities on the star . in order to do this , the processes responsible for dissipating the disturbance should be included as accurately as possible . an important dissipation process is that associated with turbulent friction in the convection zone . as this process has been suggested as being responsible for circularizing the orbits of main sequence binaries of sufficiently short period ( see mathieu 1994 ) , we investigate whether reasonable assumptions about the behavior of turbulent viscosity can lead to the observed circularization rates . zahn & bouchet ( 1989 ) have investigated the pre main sequence evolution of late type binaries in which the stars are fully convective . the main conclusion of their work was that circularization occurs at the very beginning of the hayashi phase , with hardly any decrease of the eccentricity on the main sequence . the cutoff period they predict is between 7.3 and 8.5 days . according to them , observations show a cutoff period around 8 days independent on the age of the star and are then in agreement with their results . however , we note that for this agreement to be reached , some observations had to be discarded . those by mayor & mermilliod ( 1984 ) , which indicated that the cutoff period of late type binaries was at most 5.7 days , and those by mathieu & mazeh ( 1988 ) , which showed that the cutoff period in the 4 @xmath155 old cluster m 67 was more than 10 days . in a review article , mathieu ( 1994 ) ( see also mathieu 1992 and mathieu _ et al . _ 1992 ) has confirmed that `` cutoff periods increase with age , consistent with active main sequence tidal circularization '' . main sequence cutoff period is very likely to be 4.3 @xmath156 ( an upper limit being 6.4 @xmath156 ) , and cutoff periods for solar mass binaries are 7.05 @xmath156 in the pleiades ( 0.1 @xmath155 ) , 8.5 @xmath156 in the hyades ( 0.8 @xmath155 ) , 12.4 @xmath156 in m 67 ( 4 @xmath155 ) and 18.7 @xmath156 in the halo field ( 16 @xmath155 ) . we therefore conclude that active circularization does take place for main sequence binaries and that it is less efficient than proposed by zahn & bouchet ( 1989 ) on the pre main sequence . recently , claret & cunha ( 1997 ) have applied the formalism developed by zahn ( 1989 ) to different stellar models . they have computed the parameters which enter into the expression for the circularization timescale , which is based on treatment of the equilibrium tide , for a wide grid of stellar models as a function of mass and time . their conclusion is that turbulent dissipation is too low by a factor 100200 during the main sequence to fit the observed cutoff periods . the calculations presented in the following section are applied to the standard solar model described by christensen et al . _ ( 1996 ) . the torque associated with the equilibrium tide was calculated as indicated in [ sec : asy ] . as mentioned there , this calculation is only expected to apply at very low frequencies . for periods between 4.23 and 12.4 @xmath156 , and @xmath157 , the calculated torque can be interpolated by the following power law : @xmath158 for the calculation of the dynamical tide , we solve numerically the differential equations ( [ ode1 ] ) and ( [ ode2 ] ) using a shooting method to an intermediate fitting point ( press _ et al . _ 1986 ) . to evaluate non adiabatic effects in the radiative core close to resonances , we modify these equations in the way described in [ sec : torquenad ] . we define the dimensionless quantity @xmath159 where @xmath77 is the outer radius of the convective envelope . with this notation , the equations are integrated from @xmath160 to @xmath161 the radiative core extends from @xmath162 to @xmath163 . the results presented below have been obtained with the values @xmath157 and @xmath69 ( @xmath164 being the ratio of the mixing length @xmath61 to the pressure scale height ) . we discuss the effect of changing these parameters . for illustration purposes , we plot the horizontal and radial displacements induced in the star at orbital periods of @xmath165 @xmath156 and 8.46 @xmath156 away from resonance in the adiabatic approximation . the first period is that inferred for the system 51 pegasi . the spatial distribution of the real parts of @xmath166 and @xmath167 are shown in figure [ fig1 ] . these represent typical values of the radial and horizontal velocities , the maximum values being three and six times larger respectively . since these quantities depend on the perturbing mass through the ratio @xmath168 , they have been represented in units of this factor . as expected , the stellar response shows oscillations between turning points near the center and the inner radius of the convection zone , where @xmath169 otherwise it is evanescent . the horizontal displacement varies rapidly in the photosphere because the temperature drops to zero rapidly there . we see from figure [ fig1 ] that , when the perturbing mass @xmath152 , the maximum radial velocity at the surface of the star is about 6 and 1 @xmath170 for @xmath165 @xmath156 and 8.46 @xmath156 respectively . these numbers drop to @xmath171 and @xmath172 respectively when the perturbing mass is one jupiter mass ( @xmath173 ) . the radial displacement and the perturbed pressure at the surface of the star are well approximated by the equilibrium values ( [ xieq ] ) and ( [ peq ] ) respectively . these quantities are found to be insensitive to the existence of resonances with the consequence that the radial velocity at the surface of the star never differs much from the values given above . for the smallest periods considered , the ratio @xmath174 at the surface of the star can vary by up to one order of magnitude on passage through resonance . this is due to the fact that @xmath39 is proportional to @xmath175 ( see [ xih ] ) . even though this ratio remains small , it can vary by up to an order of magnitude as a resonance is passed through . the numerical results indicate that both the amplitude and the wavelength of the response increase with the orbital frequency , in agreement with the theoretical expectation of a smaller radial order for higher frequencies ( christensen dalsgaard & berthomieu 1991 and references therein ) . finally , as expected ( see [ sec : asy ] ) , the plots shown in figure [ fig1 ] ( see also figure [ fig3 ] ) indicate that at the boundary of the radiative core and the convection zone there is a near discontinuity in the mean value of @xmath39 obtained after averaging out the interior oscillations . _ i ) resonances _ : figure [ fig2 ] shows @xmath176 versus @xmath28 in a log log representation for three different small frequency intervals in the vicinity of @xmath165 , 8.5 and 12.4 @xmath156 . on each plot , the dotted line gives the values obtained from the theory of the equilibrium tide as given by ( [ asyint ] ) . since the torque depends on the perturbing mass through the factor @xmath177 , it has been represented in units of this . these plots show several resonances , which occur when the frequency of the tidal wave is equal to the frequency of some normal mode of the star . the left panels show the torque arising from convective dissipation , through turbulent viscosity , alone ( see [ sec : torque ] ) . these plots have been displayed for comparison with the right panels , for which radiative damping has been taken into account in the resonances ( see [ sec : torquenad ] ) . as indicated by the table [ table1 ] , the normal mode damping rate due to radiative damping ( @xmath128 ) is much larger than that due to convective dissipation ( @xmath83 ) . ccccccccccccccccccccccccccccccccc @xmath178 & @xmath28 & @xmath83 & @xmath128 + ( days ) & ( @xmath179 ) & ( @xmath179 ) & ( @xmath179 ) + 4.23 & @xmath180 & @xmath181 & @xmath182 + 8.5 & @xmath183 & @xmath184 & @xmath185 + 12.4 & @xmath186 & @xmath187 & @xmath188 + [ table1 ] thus the torque in the center of resonances , where they are significant , is predominantly determined by radiative damping . for the frequencies we consider , this contribution to the torque becomes much smaller than that due to turbulent viscosity in the tail of the resonances . this means that non adiabatic effects in the radiative core are negligible away from resonances . therefore , in the right panels of figure [ fig2 ] , we have plotted the torque resulting from radiative damping acting alone in the center of resonances , and that resulting from convective dissipation acting alone away from resonances . a comparison between the strength of the resonances shown in the right and left panels indicates the importance of non adiabatic effects in the radiative core . as expected , the resonances are weakened and broadened , this effect being marginally important for @xmath189 @xmath156 . we now discuss the properties and the effect of resonances on the tidal torque . from now on , when resonances are discussed , we shall refer to the calculations which take into account radiative damping . in the neighborhood of @xmath165 , 8.5 and 12.4 @xmath190 the relative separation , @xmath191 between successive resonances is respectively @xmath192 @xmath172 and @xmath193 the relative width , @xmath194 of the resonances is respectively @xmath195 @xmath196 and @xmath197 here we have arbitrarily defined the width of a resonance as being the frequency interval over which @xmath176 is at least 3 times larger than the minimum torque obtained just out of this resonance . to calculate the probability of the companion being in a resonance , we have to take into account the fact that it drifts away from the resonances much more rapidly than elsewhere . the relevant quantity for calculating the tidal evolution timescale is @xmath198 ( see expression [ torb1 ] ) . for a fixed oscillation spectrum , we can approximate this probability by @xmath199 times the ratio of the mean value of @xmath198 over a resonance to the mean value of @xmath198 between two resonances , where the mean value is defined by @xmath200 with the integrals being taken over the relevant frequency interval . this gives a probability of being in a resonance which is close to 0.7% for @xmath165 and 8.5 @xmath156 and 2% for @xmath201 @xmath156 . the fact that the probability of being in a resonance increases with @xmath178 is not significant , because resonances get weaker when the period increases ( see figure [ fig2 ] ) . we note that this discussion applies only if the _ a priori _ probability of being in any frequency interval of a given width is independent of the frequency as might be expected to be a reasonable assumption if the normal mode spectrum is fixed . however , different circumstances may apply if the combined effect of orbital and stellar evolution were to lock the companion in a resonance with changing location . but we shall not consider the possibility of this process here . as the companion spirals inwards , it goes through a succession of resonances . however , for a fixed normal mode spectrum , the above calculation tells us that its migration is controlled essentially by the non - resonant interaction . this can be seen by comparing @xmath202 evaluated over a large frequency range , both taking into account and neglecting the resonances . such a comparison shows that neglecting resonances changes @xmath202 by at most a few per cent . _ ii ) relation between the mean torque , @xmath28 and the circularization timescale _ : here we interpolate the numerical results to express the torque as a power of the frequency . to begin with , we consider the three frequency intervals described above . we take the appropriate torque to be @xmath203 where the mean values are taken over the frequency intervals displayed in figure [ fig2 ] . the results can be interpolated with the following relation : @xmath204 we have checked that the above formula gives a good estimate of the torque at other non resonant frequencies between 4.23 and 12.4 @xmath156 . since the index of the power law ( [ dynint ] ) is close to 4 , the circularization timescale @xmath150 is given by ( [ tcircb ] ) . at @xmath201 @xmath156 , @xmath150 is found from the above formula to be 56 times larger than the timescale of 4 @xmath155 that is indicated by the observations . we note that both the dynamical and equilibrium tide calculations give a power law with an index close to 4 , which results in the circularization timescale being proportional to the binary period @xmath178 to the 13/3 . for comparison , zahn ( 1977 ) , zahn ( 1989 ) and goldman & mazeh ( 1991 ) , using equilibrium tide calculations , found @xmath150 to be proportional to @xmath178 raised to the power 16/3 , 13/3 and 10/3 respectively . the difference between these results can be related to a different choice of @xmath205 in the expression ( [ nu ] ) for @xmath206 these authors used an expression similar to ( [ nu ] ) with respectively @xmath207 , 1 and 2 . the fact that we obtain an index close to 13/3 by setting @xmath66 or even @xmath67 ( see below ) , in contrast to the results above , is at least partially due to the effectively smaller value of @xmath59 we used ( see below ) . we comment further that tassoul ( 1988 ) found @xmath208 for his postulated alternative hydrodynamical mechanism for tidal circularization . the results presented above show that the torque corresponding to the dynamical tide is smaller than that given by the adiabatic equilibrium tide for all the frequencies we have computed . however , the difference tends to decrease as the frequency gets smaller . from expressions ( [ asyint ] ) and ( [ dynint ] ) , we calculate that the ratio of these torques is indeed about 6.0 and 4.8 for @xmath165 and 12.4 @xmath156 respectively . figure [ fig3 ] shows @xmath209 , @xmath210 and @xmath211 in units @xmath168 versus @xmath212 in the range @xmath213 for @xmath165 and 12.4 @xmath156 . it is clear from these plots that @xmath42 departs from the asymptotic value in the convective envelope . the difference is not large , being about 17% , but the derivatives of @xmath42 and @xmath39 from which the torque is calculated ( see [ pt1 ] ) depart more from their asymptotic values . in the limit where the magnitude of the brunt - visl frequency is everywhere large compared to the tidal forcing frequency , calculations based on the dynamical tide should converge towards those based on the equilibrium tide . this is because the convective timescale is small enough that the convective motions adjust essentially instantaneously to the tidal forcing . we have checked this expectation by artificially increasing the magnitude of the brunt - visl frequency in the convection zone . except in the part of the convective envelope just below its outer radius , wherever @xmath214 , @xmath215 being an arbitrary constant , we make the replacement @xmath216 . in figure [ fig4 ] we plot @xmath217 versus @xmath212 in the range @xmath218 for @xmath201 @xmath156 and for @xmath219 100 and 400 . for comparison we also plot the case corresponding to the original solar model . as expected , @xmath217 converges towards 1 when the magnitude of the brunt - visl frequency is increased . we note that @xmath220 , which corresponds to @xmath221 , gives @xmath222 smaller than 5% . this confirms that the asymptotic limit is reached when the magnitude of the brunt - visl frequency is very large compared to the frequency of the tidal wave . we note that the very slow convergence towards the equilibrium tide was predicted from the arguments presented in [ sec : asy ] . also expected was the discontinuity in the mean value of @xmath223 obtained after averaging out the interior oscillations , at the boundary of the radiative core and the convection zone that is observed in figure [ fig3 ] ( see also figure [ fig1 ] ) . we shall limit the comparison of our results with observations of main sequence binaries because our calculations do not apply to pre main sequence stars , which have a much larger convective envelope than the sun . as mentioned above , the observed circularization timescale we have to fit is then 4 @xmath155 for @xmath201 @xmath156 ( mathieu 1994 ) . as indicated above , when using a simple estimate of the turbulent viscosity based on mixing length theory for non rotating stars , the circularization timescale we get from our calculations for this period is 56 times larger than 4 @xmath155 . this indicates that either + ( i ) solar type binaries are not circularized through turbulent viscosity acting on tidal perturbations ( but see tassoul 1988 and kumar & goodman 1996 for other suggested tidal mechanisms ) , or + ( ii ) dissipation in the convective envelope of solar like stars is significantly more efficient than is currently estimated . tassoul ( 1995 ) postulates that efficient tidal dissipation occurs in a very thin eckman layer close to the surface of a tidally deformed star and that this process greatly increases the efficiency of tidal interactions . but a refutation of the notion that the free surface boundary condition appropriate to the tidally deformed star , rather than the more common rigid boundary condition , leads to such an effective boundary layer , has been given by rieutord & zahn ( 1997 ) . further tassoul & tassoul ( 1997 ) state that their mechanism is inapplicable to extreme mass ratio cases such as 51 pegasi that we consider later in the paper . we now consider briefly here the mechanism proposed by kumar & goodman ( 1996 ) , namely enhanced dissipation associated with high order oscillation modes excited through parametric instability . the growth rate for the most rapidly growing modes is expected to be @xmath224 , where @xmath42 is the radial displacement in the primary oscillation , which we shall assume to be the equilibrium tide , evaluated at @xmath225 , and @xmath226 is its frequency . if we assume that the non linear development of the parametric instability and subsequent dissipation of the excited modes leads to an effective viscosity , and frictional dissipation rate @xmath227 which is big enough to suppress the linear instability , then we expect from the classical theory of darwin ( 1879 ) that there will be a phase lag @xmath228 associated with the tide given by @xmath229 for a binary of unit mass ratio and period @xmath230 @xmath156 , synchronization occurs on a timescale very much shorter than that required for circularization , so that we assume that the stellar rotation is synchronized with the orbit and @xmath231 in the calculation of @xmath232 for small eccentricity , the circularization timescale @xmath150 is approximately given by @xmath233 , where the apsidal motion frequency is @xmath234 , with @xmath235 being the apsidal motion constant ( cowling 1938 ) . we use the equilibrium tide value ( [ xieq ] ) to estimate @xmath236 at @xmath225 as @xmath237 for a mass ratio of unity . we then get @xmath238 for @xmath228 given by ( [ darwin ] ) . for and orbital period of 12.4 @xmath156 , this gives @xmath239 @xmath240 since @xmath241 , it is not very likely that this mechanism will be able to explain the observed circularization rates . however , we stress that this has not been shown from a full non linear calculation of the development of parametric instability . in general , a large increase in the simply estimated turbulent viscosity coefficient is needed in order to explain the observed circularization rate . we now investigate what is needed to achieve this and give the numerical results of tests we have carried out . in all cases , we have checked that _ the velocity at the surface of the star is not sensitive to the magnitude of the turbulent viscosity assumed_. in addition , as indicated above , the resonances are essentially controlled by radiative damping in the radiative core as long as @xmath242 . when the turbulent viscosity is enhanced , @xmath83 is increased . however , in the tests we present below , for orbital periods larger than @xmath0 days , @xmath128 stays large enough compared to @xmath83 so that , although the non resonant torques increase , the central structure and strength of the resonances is determined by radiative damping . for orbital periods on the order of @xmath1 days , @xmath83 can become comparable to @xmath128 . then the damping factor in [ restor ] has to be replaced by @xmath243 , and the strength of the resonance is reduced by a factor @xmath1 . we first consider the effect on the circularization timescale of _ varying the parameters _ @xmath58 , @xmath59 , @xmath205 and @xmath61 in the expression ( [ nu ] ) for @xmath56 . note that the denominator we used in this expression is @xmath244 rather than @xmath245 using the latter with @xmath246 is equivalent to setting @xmath247 in the former . at present , it seems that our knowledge of convection does not allow discrimination between these possibilities ( see , for example , the discussion in zahn 1989 ) . however , we note that goodman & oh ( 1997 ) have recently put forward some arguments in favor of @xmath247 . using @xmath248 in ( [ nu ] ) results in a circularization time for @xmath201 @xmath156 that is 10 times larger than that obtained with @xmath246 . if we set @xmath249 , @xmath150 is decreased by only a factor 2 . this is because when @xmath246 the factor @xmath250 is already smaller than , or even very small compared to , unity in a large part of the convective envelope . thus it seems that adjusting the way in which turbulent viscosity responds to short period forcing can not produce the required enhancement in this case . in table [ table2 ] we summarize the results obtained for different values of @xmath205 and @xmath59 , and in table [ table3 ] we indicate the corresponding index of the power law in expression ( [ dynint ] ) . we note that @xmath246 corresponds to the denominator in ( [ nu ] ) being @xmath251 , whereas @xmath252 corresponds to @xmath253 . ccccccccccccccccccccccccccccccccc @xmath178 & @xmath59 & @xmath205 & @xmath150 + ( days ) & & & ( @xmath155 ) + 4.23 & 1 & 2 & 2.46 + & & 1 & 1.37 + 12.4 & & 2 & 220 + & & 1 & 211 + & & & + 4.23 & @xmath254 & 2 & 39.4 + & @xmath255 & 1 & 5.98 + 12.4 & @xmath254 & 2 & 2010 + & @xmath255 & 1 & 685 + [ table2 ] ccccccccccccccccccccccccccccccccc @xmath59 & @xmath205 & index of the power law + 1 & 2 & 3.85 + & 1 & 4.3 + & & + @xmath254 & 2 & 3.3 + @xmath255 & 1 & 4.1 + [ table3 ] we note that setting @xmath256 with @xmath66 decreases the index of the power law in ( [ dynint ] ) down to @xmath257 this gives @xmath150 proportional to the orbital period to the @xmath258 , which is similar to the value found by goldman & mazeh ( 1991 ) . zahn & bouchet ( 1989 ) have argued that the prescription @xmath66 suggested by goldreich & keeley ( 1977 ) and used later by campbell & papaloizou ( 1983 ) and goldman & mazeh ( 1991 ) ( see also goodman & oh 1997 ) would lead to cutoff periods in clear conflict with the observational data which they claim require @xmath259 the results presented for the model adopted here do not support this statement . if @xmath246 , @xmath150 hardly changes when @xmath205 is changed from 2 to 1 . if @xmath252 , the difference between @xmath67 and @xmath66 is not dramatic for @xmath260 @xmath156 . within the uncertainties associated with convection , our results do not allow a distinction to be made between @xmath67 and @xmath66 . finally , taking the mixing length to be the distance to the top boundary of the convective envelope rather than 3 times the pressure scale height does not affect significantly the circularization timescale . we note that the torque is directly proportional to @xmath58 and @xmath261 . however , @xmath58 is expected to be on the order of unity , and @xmath164 is usually taken to be between 1 and 4 . the magnitude of the turbulent viscosity given by ( [ nu ] ) would be increased if the magnitude of the brunt - visl frequency was larger in the convection zone . this is because the convective timescale decreases . we comment that a reduction in the convective timescale , while maintaining the same length scale , implies larger convective velocities which would have to occur without increasing the heat flux . this is the essential feature of the modification . to illustrate the effect of increasing the brunt - visl frequency we consider the following dimensionless number : @xmath262 which is the superadiabatic temperature gradient when radiation pressure and variations of the mean molecular weight are neglected . the accuracy with which @xmath263 is known from helioseismic observations is not better than @xmath264 ( gough 1984 ) . however , in most of the convective envelope , this parameter , estimated from mixing length theory applied to a non rotating stars , is much smaller than @xmath265 we have made a numerical investigation in which we increased @xmath266 in the convection zone by replacing @xmath263 by min@xmath267 @xmath268 being an arbitrary constant , wherever @xmath269 ( except just below the outer radius of the convective envelope ) . we have considered @xmath270 and @xmath271 . it is doubtful that consideration of present helioseismic data could preclude such an increase of @xmath266 ( thompson 1997 ) . figure [ fig5 ] shows @xmath263 in the convective envelope versus @xmath212 for the original solar model and for @xmath270 and @xmath271 . we also display the factor by which @xmath266 has been increased in each case . the circularization timescales we find with @xmath246 and @xmath66 for @xmath201 @xmath156 when @xmath270 and @xmath271 are respectively 13 and 7.7 @xmath155 , which are now larger than the observed value by factors of 3 and 2 respectively . small discrepancies of this magnitude could be dealt with by adjustments to the mixing length or @xmath272 when either @xmath270 or @xmath271 , the circular orbit torque is found to be proportional to @xmath273 . if , keeping @xmath66 , we adopt @xmath256 , @xmath150 is increased only by a factor 1.3 compared to the case @xmath246 for @xmath271 . this is because the factor @xmath250 is smaller than , or even very small compared to , unity in almost all the convection zone whatever @xmath59 between 1 and @xmath254 . although an increase in @xmath274 of the magnitude we consider might be thought to be unrealistic , we note that such an increase in the deep layers of the convection zone has also been considered by dsilva ( 1995 ) as a means of explaining the dynamics of sunspots without invoking too strong a magnetic field . in his model , which applies strictly to a star rotating at the same rate as the sun , @xmath274 has to be larger than @xmath275 @xmath276 , which means that it has to be multiplied on average by a factor @xmath277 in the solar model we use , we need to multiply @xmath274 by a factor between 100 and 400 for @xmath278 , between 10 and 100 for @xmath279 and between 1 and 10 for @xmath280 in order to get such a minimum value . if we do this , the circularization timescale we get for @xmath201 @xmath156 is only 6 times larger than the observed one . in this context , it is possible that if the proposed increase in @xmath274 is related to the stellar rotation , this may be even greater for the more rapidly synchronously rotating star that is expected in the equal mass binary case . we note that numerical simulations of turbulent convection in the presence of rotation show an increase of @xmath274 with the effect of rotation ( brummell , hurlburt & toomre 1996 ) . this is because , as pointed out by brummel _ et al . _ ( 1996 ) , rotation influences the thermodynamic mixing properties of the convection in such a way that it leads to a decrease in correlation between temperature fluctuations and vertical velocities . the efficiency of the vertical convective transport is then weakened , with a subsequent enhanced superadiabatic mean stratification in the interior of the fluid ( see their figure 8.a ) . + this suggests that the magnitude of the brunt - visl frequency in the convective envelope of rotating stars is actually larger than the values given by the solar model we have been using here . however , it seems questionable that the extremely large increase required to account for the observed circularization rates can be achieved . another means of increasing the total amount of dissipation is to assume that turbulent viscosity acts down to some depth below the inner boundary of the convective envelope . this might be expected if convective overshooting takes place . however , this might not be a very effective process because of the slow convective motions expected and the rapid increase in @xmath274 that occurs as the radiative zone is entered . estimates based on the observed solar oscillation frequencies give an upper limit between 0.05 ( basu 1997 ) and @xmath281 ( christensen dalsgaard , monteiro & thompson 1995 ) times the pressure scale height on the extent of overshoot below the convection zone . here we consider a simple illustrative situation in which convective blobs or some other turbulent motions are able to penetrate into the stratified radiative core over some fraction @xmath282 of the pressure scale height @xmath283 producing a turbulent viscosity . we model this by setting @xmath56 to be constant from a distance @xmath284 above the inner radius of the convective envelope down to the same distance below this radius , equal to its value at the top of this zone . because such a viscosity is able to act on the short wavelength part of the tidal response associated with g modes it has a dramatic effect . for @xmath285 and @xmath157 , the circularization timescale for @xmath201 @xmath156 is decreased by a factor about 400 , being now 8 times smaller than the observed timescale . if we set @xmath256 ( see discussion above ) , @xmath150 is increased by a factor 15 , being about 4 times larger than the observed timescale . for @xmath286 and @xmath157 , we get a circularization timescale for @xmath201 @xmath156 about 30 times larger than the observed one . for the calculated timescale to be in agreement with the observations , we need @xmath282 between 0.4 and 0.5 with @xmath157 . in the model we have adopted , the effect is of less importance for shorter periods because the number of oscillations of the response in the region of the radiative core where turbulent dissipation is introduced decreases with forcing frequency . therefore , the torque does not vary with @xmath28 as a simple power law with an index close to 4 . however , in reality , the effectiveness of the turbulent viscosity should be reduced for short wavelength disturbances giving a compensating effect to make it relatively less effective at low frequencies . the above calculations show that overshooting is not likely to be efficient enough to decrease the circularization timescales by a factor of about 50 . to get such an effect , we indeed require an extent of overshoot below the convection zone at least 5 or 10 times larger than that deduced from the observations . in addition , we have not taken into account the fact that overshooting leads to an increase of the g modes length scale through a decrease in the buoyancy or the magnitude of @xmath287 this in turn would decrease the amount of turbulent dissipation associated with overshooting . as we have already mentioned above , _ calculations based on both the dynamical and equilibrium tide theories give a torque proportional to the orbital frequency raised to a power @xmath1 _ ( see [ asyint ] and [ dynint ] ) . if circularization of solar type binaries does occur through the action of turbulent viscosity on the tides , then its magnitude has to be calibrated so as to account for the observed timescale . we have discussed in the previous section , somewhat speculatively , how the required enhancement of the magnitude of the viscosity above that obtained from simple estimates might be envisaged to occur . since the enhancement might depend on forcing frequency , it is not clear that the resulting torque will still be proportional to the frequency to the @xmath1 . however , the increase to @xmath288 described above gave torques that approximately preserved this power law so that in the absence of additional information , we shall suppose it holds . then the calibration acts only to adjust the coefficient of the power law . we note that the observations do not rule out any exponent between 3 and 5 ( see below ) . since our calculations can only strictly be applied to solar type stars , we calibrate our results using @xmath289 @xmath155 for @xmath201 @xmath156 . this gives ( in cgs ) : @xmath290 or , equivalently : @xmath291 the corresponding formul for the orbital and spin up timescales ( in giga years ) are : @xmath292 @xmath293 and @xmath294 ( for the solar model we use , @xmath295 @xmath296 and @xmath297 @xmath296 ) . since the torque is proportional to @xmath298 we can use the relation ( [ tcircb ] ) for @xmath150 , so that the circularization time is given by @xmath299 even though our calculations can only be applied to solar type stars , it is of interest to compare the circularization timescales we get from ( [ tcirc ] ) with the observed ones . for @xmath300 , 7.05 , 8.5 and 18.7 @xmath156 , ( [ tcirc ] ) gives respectively @xmath301 , 0.3 , 0.8 and 24 @xmath155 , to be compared with the observed timescales 0.003 , 0.1 , 0.8 and 16 @xmath155 respectively . the agreement for @xmath302 @xmath156 is within a factor 1.5 . for smaller periods , circularization is expected to occur when the convective envelopes of the stars are larger , making turbulent dissipation more efficient . we note that mathieu _ et al . _ ( 1992 ) have already pointed out that a power law @xmath303 provides a close fit to the slope of the observed cutoff periods . however , the observations are equally well fitted with an index of 10/3 and an index of 16/3 can not be ruled out ( mathieu _ et al . _ 1992 ) . if the simple estimate of the turbulent viscosity based on mixing length theory for non rotating stars is used , the coefficients in the formul for the torque have to be divided by a factor @xmath304 whereas those in the formul for the timescales have to be multiplied by the same factor . in figure [ fig6 ] , we have plotted @xmath144 in units @xmath305 and @xmath150 in units @xmath306 versus @xmath28 and @xmath178 in a log log representation . it is of interest to apply these results to the system 51 pegasi , for which the orbital period ( assuming the observed oscillations are due to a companion ) is @xmath165 @xmath307 if the companion is a jupiter mass planet ( @xmath308 ) , then the tidal orbital evolution timescale given by ( [ torb ] ) is @xmath309 @xmath310 the star spin up timescale ( [ tsp ] ) is @xmath311 @xmath155 and the spin up timescale of the convective envelope is @xmath312 @xmath155 . all of these timescales are long compared with the inferred age of 51 pegasi ( edvardsson _ et al . _ if the companion is a low - mass star of @xmath313 , as has been recently suggested , @xmath144 is 100 times smaller while @xmath314 and @xmath315 are @xmath316 times smaller . we then expect the primary star to be synchronized with the orbit , in which case exchange of angular momentum is no longer taking place . synchronization is actually expected if the mass of the companion is larger than about 10 jupiter masses . the orbital decay timescale is also smaller than the age of the system , but since @xmath317 , tidal interaction stops before the companion has plunged into the central star . if the simple estimate of the turbulent viscosity based on mixing length theory for non rotating stars is used , all these timescales have to multiplied by @xmath318 . in that case , synchronization is expected if the mass of the companion is larger than about 70 jupiter masses . the planetary companion interpretation has been questioned recently by the reported 4.23 @xmath156 modulation in the line profile of 51 pegasi ( gray 1997 ) , and the possibility that this modulation may be due to g mode oscillations has been considered ( gray & hatzes 1997 ) . we note that , according to our results , such a modulation could not be due to g mode oscillations tidally driven by a companion . for the oscillation to have a period of @xmath319 @xmath156 , the orbital period would have to be 8.46 @xmath156 . the maximum perturbed radial velocity at the surface of the star induced by the companion would then be between @xmath172 and 1 @xmath170 for a perturbing mass between @xmath320 and 1 solar mass . _ these numbers do not depend on the magnitude of the turbulent viscosity assumed , and are not expected to be affected by the possibility of resonance_. these velocities are at least about 50 times smaller than the observed ones . in this paper , we have studied the dynamical response of the star to the tidal perturbation of a companion . we have computed the torque due to dissipation in the convective envelope using first order perturbation theory . in the vicinity of resonance , we have also calculated the torque due to non adiabaticity in the radiative core using a wkb treatment . we have found that the torque at effective resonances is mainly determined by radiative damping . we have carried out an analysis based on the adiabatic equilibrium tide , and showed that agreement with the dynamical tide calculations can be rather poor . for the unmodified stellar model and the periods of interest of several days , the torque derived using the equilibrium tide is 4 to 6 times larger than that corresponding to the dynamical tide . we have found that the presence of fixed resonances do not affect the long term orbital evolution of the binary , so that the different timescales ( orbital evolution , circularization and spin up ) are mainly determined by the non resonant interaction . our calculations show that the viscosity that is required to provide the observed circularization rates of solar type binaries is @xmath318 times larger than that simply estimated from mixing length theory for non rotating stars . we have explored some means by which this viscosity might be enhanced . we have found that it could become large enough if the magnitude of the brunt - visl frequency in the deep convective envelope were increased sufficiently . such an increase is expected to be produced by the effect of rotation on convection , but it is questionable whether it can be of sufficient magnitude . we note that _ the strength of the resonances for orbital periods larger than @xmath0 days and the perturbed velocity at the surface of the star are insensitive to the magnitude of the turbulent viscosity assumed_. only for periods @xmath1 days is the strength of the resonances decreased by a factor @xmath1 . the effective widths of the resonances affecting the tidal torques as well are reduced when the viscosity is increased . we have applied our results to 51 pegasi , and showed that the oscillations which have been observed at the surface of this star can not be a tidally driven non radial g mode . also we have found that the stellar rotation and the orbital motion of this system are expected to be synchronized if the mass of the companion exceeds 0.1 solar mass . we are grateful to m.j . thompson for supplying us a solar model , and thank d.o . gough , r.d . mathieu and m.j . thompson for helpful discussions . this work is supported by pparc through grant gr / h/09454 and by nsf and nasa through grants ast 9315578 and nag 54277 . acknowledge support by the center for star formation studies at nasa / ames research center and the university of california at berkeley and santa - cruz . mathieu , r. d. , duquennoy , a. , latham , d. w. , mayor , m. , mazeh , t. , & mermilliod , j.c . 1992 , in binaries as tracers of stellar formation , ed . a. duquennoy , & m. mayor ( cambridge : cambridge university press ) , 278

continuing on our previous work [ a. donev , a. nonaka , y. sun , t. g. fai , a. l. garcia and j. b. bell , comm . app . math . and comp . sci . , 9 - 1:47 - 105 , 2014 ] , we develop semi - implicit numerical methods for solving low mach number fluctuating hydrodynamic equations appropriate for modeling diffusive mixing in isothermal mixtures of fluids with different densities and transport coefficients . we treat viscous dissipation implicitly using a recently - developed variable - coefficient stokes solver [ m. cai , a. j. nonaka , j. b. bell , b. e. griffith and a. donev , commun . comput . phys . , 16(5):1263 - 1297 , 2014 ] . this allows us to increase the time step size significantly for low reynolds number flows with large schmidt numbers compared to our earlier explicit temporal integrator . also , unlike most existing deterministic methods for low mach number equations , our methods do not use a fractional time - step approach in the spirit of projection methods , thus avoiding splitting errors and giving full second - order deterministic accuracy even in the presence of boundaries for a broad range of reynolds numbers including steady stokes flow . we incorporate the stokes solver into two time - advancement schemes , where the first is suitable for inertial flows and the second is suitable for the overdamped limit ( viscous - dominated flows ) , in which inertia vanishes and the fluid motion can be described by a steady stokes equation . we also describe how to incorporate advanced higher - order godunov advection schemes in the numerical method , allowing for the treatment of ( very ) large peclet number flows with a vanishing mass diffusion coefficient . we incorporate thermal fluctuations in the description in both the inertial and overdamped regimes . we validate our algorithm with a series of stochastic and deterministic tests . finally , we apply our algorithms to model the development of giant concentration fluctuations during the diffusive mixing of water and glycerol , and compare numerical results with experimental measurements . we find good agreement between the two , and observe propagative ( non - diffusive ) modes at small wavenumbers ( large spatial scales ) , not reported in published experimental measurements of concentration fluctuations in fluid mixtures . our work forms the foundation for developing low mach number fluctuating hydrodynamics methods for miscible multi - species mixtures of chemically reacting fluids . # 1 # 1 # 1 # 1#1 # 1#1 # 1#1 # 1|#1| # 1#1 # 1#1

1410.2300

flows of realistic mixtures of miscible fluids exhibit several features that make them more difficult to simulate numerically than flows of simple fluids . firstly , the physical properties of the mixture depend on the concentration of the different species composing the mixture . this includes both the density of the mixture at constant pressure , and transport coefficients such as viscosity and mass diffusion coefficients . common simplifying assumptions such as the boussinesq approximation , which assumes a constant density and thus incompressible flow , or assuming constant transport coefficients , are uncontrolled and not appropriate for certain mixtures of very dissimilar fluids . secondly , for liquid mixtures there is a large separation of time scales between the various dissipative processes , notably , mass diffusion is much slower than momentum diffusion . the large schmidt numbers @xmath0 typical of liquid mixtures lead to extreme stiffness and make direct temporal integration of the hydrodynamic equations infeasible . lastly , flows of mixtures exhibit all of the numerical difficulties found in single component flows , for example , well - known difficulties caused by advection in the absence of sufficiently strong dissipation ( diffusion of momentum or mass ) , and challenges in incorporating thermal fluctuations in the description . here we develop a low mach number approach to isothermal binary fluid mixtures that resolves many of the above difficulties , and paves the way for incorporating additional physics such as the presence of more than two species @xcite , chemical reactions @xcite , multiple phases and surface tension @xcite , and others . stochastic fluctuations are intrinsic to fluid dynamics because fluids are composed of molecules whose positions and velocities are random . thermal fluctuations affect flows from microscopic to macroscopic scales @xcite and need to be consistently included in all levels of description . fluctuating hydrodynamics ( fhd ) incorporates thermal fluctuations into the usual navier - stokes - fourier laws in the form of stochastic contributions to the dissipative momentum , heat , and mass fluxes @xcite . fhd has proven to be a very useful tool in understanding complex fluid flows far from equilibrium @xcite ; however , theoretical calculations are often only feasible after making many uncontrolled approximations @xcite , and numerical schemes used for fluctuating hydrodynamics are usually far behind state - of - the - art deterministic computational fluid dynamics ( cfd ) solvers . in this work , we consider binary mixtures and restrict our attention to isothermal flows . we consider a specific equation of state ( eos ) suitable for mixtures of incompressible liquids or ideal gases , but otherwise account for advective and diffusive mass and momentum transport in full generality . recently , some of us developed finite - volume methods for the incompressible equations @xcite . we have also developed low mach number isothermal fluctuating equations @xcite , which eliminate the stiffness arising from the separation of scales between acoustic and vortical modes @xcite . the low mach number equations account for the fact that for mixtures of fluids with different densities , diffusive and stochastic mass fluxes create local expansion and contraction of the fluid . in these equations the incompressibility constraint should be replaced by a `` quasi - incompressibility '' constraint @xcite , which introduces some difficulties in constructing conservative finite - volume techniques @xcite . in section [ sec : equations ] we review the low mach number equations of fluctuating hydrodynamics for a binary mixture of miscible fluids , as first proposed in ref . @xcite . the numerical method developed in ref . @xcite uses an explicit temporal integrator . this requires using a small time step and is infeasible for liquid mixtures due to the stiffness caused by the separation of time scales between fast momentum diffusion and slow mass diffusion . in recent work @xcite , some of us developed temporal integrators for the equations of fluctuating hydrodynamics that have several important advantages . notably , these integrators are semi - implicit , allowing one to treat fast momentum diffusion ( viscous dissipation ) implicitly , and other transport processes explicitly . furthermore , these temporal integrators are constructed to be second - order accurate for the equations of linearized fluctuating hydrodynamics ( lfhd ) , which are suitable for describing thermal fluctuations around stable macroscopic flows over a broad range of length and time scales @xcite . importantly , the linearization of the fluctuating equations is carried out _ automatically _ by the code , making the numerical methods very similar to standard deterministic cfd schemes . finally , specific integrators are proposed in ref . @xcite to handle the extreme separation of scales between the fast velocity and the slow concentration by taking an _ overdamped _ limit of the inertial equations . in this work , we extend the semi - implicit temporal integrators proposed in ref . @xcite for incompressible flows to account for the quasi - incompressible nature of low mach number flows . we apply these temporal integrators to the staggered - grid conservative finite - volume spatial discretization developed in ref . @xcite , and additionally generalize the treatment of advection to allow for the use of monotonicity - preserving higher - order godunov schemes @xcite . our work relies heavily on several prior works , which we will only briefly summarize in the present paper . the spatial discretization we describe in more detail in section [ sub : advection ] is identical to that proposed by donev _ et al _ @xcite , which itself relies heavily on the treatment of thermal fluctuations developed in refs . @xcite . a key development that makes the algorithm presented here feasible for large - scale problems is recent work by some of us @xcite on efficient multigrid - based iterative methods for solving unsteady and steady variable - coefficient stokes problems on staggered grids . our high - order godunov method for mass advection is based on the work of bell _ et al . _ @xcite . the temporal integrators developed in section [ sub : temporaldiscretization ] are a novel approach to low mach number hydrodynamics even in the deterministic context . in high - resolution finite - volume methods , the dominant paradigm has been to use a splitting ( fractional - step ) or projection method @xcite to separate the pressure and velocity updates @xcite . we followed such a projection approach to construct an explicit temporal integrator for the low mach number equations @xcite . when viscosity is treated implicitly , however , the splitting introduces a commutator error that leads to the appearance of spurious or `` parasitic '' modes in the presence of physical boundaries @xcite . there are several techniques to reduce ( but not eliminate ) these artificial boundary layers @xcite , and for sufficiently large reynolds number flows the time step size dictated by advective stability constraints makes the splitting error relatively small in practice . at small reynolds numbers , however , the splitting error becomes larger as viscous effects become more dominant , and projection methods do not apply in the steady stokes regime for problems with physical boundary conditions . methods that do not split the velocity and pressure updates but rather solve a combined stokes system for velocity and pressure have been used in the finite - element literature for some time , and have more recently been used in the finite - volume context for incompressible flow @xcite . here we demonstrate how the same approach can be effectively applied to the low mach number equations for a binary fluid mixture @xcite , to construct a method that is second - order accurate up to boundaries , for a broad range of reynolds numbers including steady stokes flow . we test our ability to accurately capture the static structure factor for equilibrium fluctuation calculations . then , we test our methods deterministically on two variable density and variable viscosity low mach number flows . first , we confirm second - order deterministic accuracy in both space and time for a lid - driven cavity problem in the presence of a bubble of a denser miscible fluid . next , we simulate the development of a kevin - helmholtz instability as a lighter less viscous fluid streams over a denser more viscous fluid . these tests confirm the robustness and accuracy of the methods in the presence of large contrasts , sharp gradients , and boundaries . next we focus on the use of fluctuating low mach number equations to study giant concentration fluctuations . in section [ sec : giantfluct ] we apply our methods to study the development of giant fluctuations @xcite during free diffusive mixing of water and glycerol . we compare simulation results to experimental measurements of the time - correlation function of concentration fluctuations during the diffusive mixing of water and glycerol @xcite . the relaxation times show signatures of the rich deterministic dynamics , and a transition from purely diffusive relaxation of concentration fluctuations at large wavenumbers , to more complex buoyancy - driven dynamics at smaller wavenumbers . we find reasonably - good agreement given the large experimental uncertainties , and observe the appearance of propagative modes at small wavenumbers , which we suggest could be observed in experiments as well . at mesoscopic scales , in typical liquids , sound waves are very low amplitude and much faster than momentum diffusion ; hence , they can usually be eliminated from the fluid dynamics description . formally , this corresponds to taking the zero mach number singular limit @xmath1 of the well - known compressible fluctuating hydrodynamics equations system @xcite . in the compressible equations , the coupling between momentum and mass transport is captured by the equation of state ( eos ) for the pressure @xmath2 as a local function of the density @xmath3 and mass concentration @xmath4 at a specified temperature @xmath5 , assumed to be time - independent in our isothermal model . the low mach number equations can be obtained by making the ansatz that the thermodynamic behavior of the system is captured by a reference pressure @xmath6 , with the additional pressure contribution @xmath7 capturing the mechanical behavior while not affecting the thermodynamics . we will restrict consideration to cases where stratification due to gravity causes negligible changes in the thermodynamic state across the domain . in this case , the reference pressure is spatially constant and constrains the system so that the evolution of @xmath8 and @xmath9 remains consistent with the thermodynamic eos @xmath10 physically this means that any change in concentration must be accompanied by a corresponding change in density , as would be observed in a system at thermodynamic equilibrium held at the fixed reference pressure and temperature . the eos defines density @xmath11 as an implicit function of concentration in a binary liquid mixture . the eos constraint ( [ eq : general_eos ] ) can be re - written as a constraint on the divergence of the fluid velocity @xmath12 , @xmath13 where @xmath14 is the total diffusive mass flux defined in ( [ eq : f_def ] ) , and the solutal expansion coefficient @xmath15 is determined by the specific form of the eos . in this work we consider a specific _ linear _ eos , @xmath16 where @xmath17 and @xmath18 are the densities of the pure component fluids ( @xmath19 and @xmath20 , respectively ) , giving @xmath21 it is important that for this specific form of the eos @xmath22 is a material constant independent of the concentration ; this allows us to write the eos constraint ( [ eq : div_v_constraint ] ) in conservative form @xmath23 and take the reference pressure @xmath24 to be independent of time . the specific form of the density dependence ( [ eq : beta_simple ] ) on concentration arises if one assumes that two incompressible fluids do not change volume upon mixing , which is a reasonable assumption for liquids that are not too dissimilar at the molecular level . surprisingly the eos ( [ eq : eos_quasi_incomp ] ) is also valid for a mixture of ideal gases . if the specific eos ( [ eq : eos_quasi_incomp ] ) is not a very good approximation over the entire range of concentration @xmath25 , ( [ eq : eos_quasi_incomp ] ) may be a very good approximation over the range of concentrations of interest if @xmath17 and @xmath18 are adjusted accordingly . our choice of the specific form of the eos will aid significantly in the construction of simple conservative spatial discretizations that strictly maintain the eos without requiring complicated nonlinear iterative corrections . in fluctuating hydrodynamics , stochastic contributions to the momentum and mass fluxes that are formally modeled as @xcite @xmath26 where @xmath27 is boltzmann s constant , @xmath28 is the shear viscosity , @xmath29 is the diffusion coefficient , @xmath30 is the chemical potential of the mixture with @xmath31 , and @xmath32 and @xmath33 are standard zero mean , unit variance random gaussian tensor and vector fields , respectively , with uncorrelated components , @xmath34 and similarly for @xmath35 . a standard asymptotic low mach analysis @xcite , formally treating the stochastic forcing as smooth , leads to the _ isothermal low mach number _ equations for a binary mixture of fluids in conservation form @xcite , @xmath36 where the deterministic and stochastic diffusive mass fluxes are denoted by @xmath37 here @xmath38 is a symmetric gradient , @xmath39 is the density of the first component , @xmath40 is the density of the second component , and @xmath41 is the gravitational acceleration . the gradient of the non - thermodynamic component of the pressure @xmath42 ( lagrange multiplier ) appears in the momentum equation as a driving force that ensures the eos constraint ( [ eq : div_v_constraint ] ) is obeyed . we note that the bulk viscosity term gives a gradient term that can be absorbed in @xmath42 and therefore does not explicitly need to appear in the equations . temperature dynamics and fluctuations are neglected in these equations ; however , this type of approach can be extended to include thermal effects . the shear viscosity @xmath43 and the mass diffusion coefficient @xmath44 in general depend on the concentration . note that the two density equations ( [ eq : rho1_eq],[eq : rho2_eq ] ) can be combined to obtain the usual continuity equation for the total density , @xmath45 and the primitive ( non - conservation law ) form of the concentration equation , @xmath46 our conservative numerical scheme is based on eqs . ( [ eq : momentum_eq],[eq : rho1_eq],[eq : div_v_constraint],[eq : rho_eq ] ) . in ref . @xcite we discussed the effect of the low mach constraint on the thermal fluctuations , suitable boundary conditions for the low mach equations , and presented a gauge formulation of the equations that formally eliminates pressure in a manner similar to the projection operator formulation for incompressible flows . importantly , the gauge formulation demonstrates that although the low mach equations have the appearance of a constrained system , one can write them in an unconstrained form by introducing a gauge degree of freedom for the pressure . for the purposes of time integration , one can therefore treat these equations as standard initial - value problems @xcite and use the temporal integrators developed in ref . @xcite . it is important to note that the equations of fluctuating hydrodynamics should be interpreted as a mesoscopic coarse - grained representation of the mass , momentum and energy transport in fluids @xcite . as such , these equations implicitly contain a mesoscopic coarse - graining length and time scale that is larger than molecular scales @xcite and can only formally be written as stochastic partial differential equations ( spdes ) . a coarse - graining scale can explicitly be included in the spdes @xcite ; such a coarse - graining scale explicitly enters in our finite - volume spatio - temporal discretization through the grid spacing ( equivalently , the volume of a grid cell , or more precisely , the number of molecules per grid cell ) . additional difficulties are posed by the fact that in general the noise in the nonlinear equations is multiplicative , requiring a careful stochastic interpretation ; the mori - zwanzig projection formalism @xcite suggests the correct stochastic interpretation is the kinetic one @xcite . for compressible and incompressible flows , the spdes of _ linearized _ fluctuating hydrodynamics ( lfhd ) @xcite can be given a precise continuum meaning @xcite . in these linearized equations one splits each variable into a deterministic component and small fluctuations around the deterministic solution , e.g. , @xmath47 , where @xmath48 is a solution of the deterministic equations ( [ eq : momentum_eq],[eq : rho1_eq],[eq : div_v_constraint],[eq : rho_eq ] ) with @xmath49 and @xmath50 . here @xmath51 is the solution of a _ linear _ additive - noise equation obtained by linearizing ( [ eq : c_t_primitive ] ) to first order in the fluctuations and evaluating the noise amplitude at the deterministic solution ; more precisely , lfhd is an expression of the central limit theorem in the limit of weak noise . in this work , in the stochastic setting we restrict our attention to lfhd equations . as discussed in ref . @xcite , we do not need to write down the ( rather tedious ) complete form of the linearized low mach number equations ( for an illustration , see section [ sub : overdamped ] ) since the numerical method will perform this linearization automatically . namely , the complete nonlinear equations are essentially equivalent to the lfhd equations when the noise is sufficiently weak , i.e. , when the hydrodynamic cells contain many molecules . the low mach number equations pose additional difficulties because they represent a coarse - graining of the dynamics not just in space but also in time . as such , even the linearized equations can not directly be interpreted as describing a standard diffusion process . this is because the stochastic mass flux @xmath52 in the eos constraint ( [ eq : div_v_constraint ] ) makes the velocity formally white - in - time @xcite . we note , however , that the analysis in ref . @xcite shows that there is a close connection between mass diffusion and advection by the thermally - fluctuating velocity field , and thus between @xmath52 and velocity fluctuations . this suggests that a precise interpretation of the low mach constraint in the presence of stochastic mass fluxes requires a very delicate mathematical analysis . in this work we rely on the implicit coarse - graining in time provided by the finite time step size in the temporal integration schemes to regularize the low mach equations @xcite . furthermore , for the applications we study here , we can neglect stochastic mass fluxes and assume @xmath53 , in which case the difficulties related to a white - in - time velocity disappear . at small scales , flows in liquids are viscous - dominated and the inertial momentum flux @xmath54 can often be neglected in a zero reynolds number approximation . in addition , in liquids , there is a large separation of time scales between the fast momentum diffusion and slow mass diffusion , i.e. , the schmidt number @xmath55 is large . this makes the relaxation times of velocity modes at sufficiently large wavenumbers much smaller than those of the concentration modes . formally treating the stochastic force terms as smooth for the moment , the separation of time scales implies that we can replace the _ inertial _ momentum equation ( [ eq : momentum_eq ] ) with the _ overdamped _ steady - stokes equation @xmath56 the above equations can be used to eliminate velocity as a variable , leaving only the concentration equation ( [ eq : c_t_primitive ] ) . note that the density equation ( [ eq : rho_eq ] ) simply defines density as a function of concentration and thus is not considered an independent equation . the solution of the stokes system @xmath57 where @xmath58 and @xmath59 are applied forcing terms , can be expressed in terms of a generalized inverse stokes linear operator ) is an affine rather than a linear operator . ] @xmath60 $ ] that is a _ functional _ of the viscosity ( and thus the concentration ) , @xmath61\left(\v f,\ , h\right).\ ] ] in the linearized fluctuating equations , one must linearize around the ( time dependent ) solution of the deterministic nonlinear equation @xmath62 where we have used the shorthand notation @xmath63 , @xmath64 , @xmath65 . here the velocity is an implicit function of concentration defined via @xmath66 which we can write in shorthand notation as @xmath67\left(\bar{\rho}\v g,\;\left(\bar{\rho}_{2}^{-1}-\bar{\rho}_{1}^{-1}\right)\grad\cdot\left(\bar{\chi}\bar{\rho}\grad\bar{c}\right)\right).\label{eq : v_bar_overdamped}\ ] ] here we develop second - order integrators for the deterministic overdamped low mach equation ( [ eq : c_bar_overdamped],[eq : v_bar_overdamped ] ) . in the stochastic setting , the solution of ( [ eq : steady_stokes ] ) is white in time because the stochastic mass and momentum fluxes are white in time . this means that the advective term @xmath68 requires a specific stochastic interpretation , in addition to the usual regularization ( smoothing ) in space required to interpet all nonlinear terms appearing in formal fluctuating hydrodynamics spdes . by performing a precise ( albeit formal ) adiabatic mode elimination of the fast velocity variable under the assumption of infinite separation of time scales , donev _ et al . _ arrive at a stratonovich interpretation of the random advection term @xmath68 ( see appendix a of ref . this analysis does not , however , directly extend to the low mach number equations since it relies in key ways on the incompressibility of the fluid . generalizing this sort of analysis to the case of variable fluid density is nontrivial , likely requiring the use of the gauge formulation of the low mach equations , and appears to be beyond the scope of existing techniques . variable ( i.e. , concentration - dependent ) viscosity and mass diffusion coefficient can be handled using existing techniques although there are subtle nonlinear stochastic effects arising from the fact that the noise in the velocity equation is multiplicative and the invariant measure ( equilibrium distribution ) of the fast velocity depends on the slow concentration . in the linearized setting , however , the difficulties associated with the interpretation of stochastic integrals and multiplicative noise disappear . the complete form of the linearized equations contains many terms and is rather tedious . since we will never need to explicitly write this form let us illustrate the procedure by assuming @xmath29 and @xmath28 to be constant . for the concentration , we obtain the linearized equation @xmath69 where @xmath70 relates concentration fluctuations to density fluctuations via the eos . here we split @xmath71 into a component @xmath72 that is continuous in time and a component @xmath73 that is white in time , @xmath74\left(\bar{\rho}^{\prime}\v g\d c,\;\left(\bar{\rho}_{2}^{-1}-\bar{\rho}_{1}^{-1}\right)\grad\cdot\left(\bar{\rho}\chi\grad\left(\d c\right)\right)\right),\\ \d{\v v}_{f } & = & \m{\mathcal{l}}^{-1}\left[\bar{\eta}\right]\left(\grad\cdot\m{\sigma},\;\grad\cdot\m{\psi}\right).\end{aligned}\ ] ] the term @xmath75 in ( [ eq : c_t_lin ] ) is interpreted as an additive noise term with a rather complicated and potentially time - dependent ( via @xmath76 ) spatial correlation structure . in this work we develop numerical methods that solve the overdamped linearized equation ( [ eq : c_t_lin ] ) to second - order weakly @xcite . our baseline spatio - temporal discretization of the low mach equations is based on the method of lines approach where we first discretize the ( s)pdes in space to obtain a system of ( s)odes , which we then solve using a single - step multi - stage temporal integrator . the conservative finite - volume spatial discretization that we employ here is essentially identical to that developed in our previous works @xcite and @xcite . in summary , scalar fields such as concentration and densities are cell - centered , while velocity is face - centered . in order to ensure conservation , the conserved momentum @xmath77 and mass densities @xmath78 and @xmath8 are evolved rather than the primitive variables @xmath79 and @xmath9 . diffusion of mass and momentum is discretized using standard centered differences , leading to compact stencils similar to the standard laplacian . stochastic mass fluxes are associated with the faces of the regular grid , while for stochastic momentum fluxes we associate the diagonal elements with the cell centers and the off - diagonal elements with the nodes ( in 2d ) or edges ( in 3d ) of a regular grid with grid spacing @xmath80 . here we focus our discussion on three new aspects of our spatio - temporal discretization . after summarizing the dimensionless numbers that control the appropriate choice of advection method and temporal integrator , in section [ sub : advection ] we describe our implementation of two advection schemes and a discussion of the advantages of each . in section [ sub : gmres ] we describe our implicit treatment of viscous dissipation using a gmres solver for the coupled velocity - pressure stokes system . in section [ sub : temporaldiscretization ] we describe our overall temporal discretization strategies for the inertial and overdamped regimes . the suitability of a particular temporal integrator or advection scheme depends on the following dimensionless numbers : @xmath81 where @xmath82 is the kinematic viscosity . observe that the first two depend on the spatial resolution and the typical flow speed @xmath83 , while the schmidt number is an intrinsic material property of the mixture . also note that the physically relevant reynolds @xmath84 and peclet @xmath85 numbers would be defined with a length scale much larger than @xmath80 , such as the system size , and thus would be much larger than the discretization - scale numbers above . in this work , we are primarily interested in small - scale flows with @xmath86 and large @xmath87 ( liquid mixtures ) . the choice of advection scheme for concentration ( partial densities ) is dictated by @xmath88 . if @xmath89 , centered advection schemes will generate non - physical oscillations , and one must use the godunov advection scheme described below . however , it is important to note that in this case the spectrum of fluctuations will not be correctly preserved by the advection scheme ; if fluctuations need to be resolved it is advisable to instead reduce the grid spacing and thus reduce the cell peclet number to @xmath90 and use centered advection . the choice of the temporal integrator , on the other hand , is determined by the importance of inertia and the time scale of interest . if @xmath91 is not sufficiently small , then there is no alternative to resolving the inertial dynamics of the velocity . now let us assume that @xmath92 , i.e. , viscosity is dominant . if the time scale of interest is the advective timescale @xmath93 , where @xmath94 is the system size , then one should use the inertial equations . however , the inertial temporal integrator described in section [ sub : temporaldiscretization ] will be rather inefficient if the time scale of interest is the diffusive time scale @xmath95 , as is the case in the study of diffusive mixing presented in section [ sec : giantfluct ] . this is because the crank - nicolson ( implicit midpoint ) scheme used to treat viscosity in our methods is only a - stable , and , therefore , if the viscous courant number @xmath96 is too large , unphysical oscillations in the solution will appear ( note that this problem is much more serious for fluctuating hydrodynamics due to the presence of fluctuations at _ all _ scales ) . in order to be able to use a time step size on the diffusive time scale , one must construct a _ stiffly accurate _ temporal integrator . this requires using an l - stable scheme to treat viscosity , such as the backward euler scheme , which is however only first - order accurate . constructing a second - order stiffly accurate implicit - explicit integrator in the context of variable density low mach flows is rather nontrivial . furthermore , using an l - stable scheme leads to a damping of the velocity fluctuations at large wavenumbers and is inferior to the implicit midpoint scheme in the context of fluctuating hydrodynamics @xcite . therefore , in this work we choose to consider separately the overdamped limit @xmath97 and @xmath98 ( note that the value of @xmath85 is arbitrary ) . in this limit we analytically eliminate the velocity as an independent variable , leaving only the concentration equation , which evolves on the diffusive time scale . we must emphasize , however , that the overdamped equations should be used with caution , especially in the presence of fluctuations . notably , the validity of the overdamped approximation requires that the separation of time scales between the fast velocity and slow concentration be uniformly large over _ all _ wavenumbers , since fluctuations are present at _ all _ lengthscales . in the study of giant fluctuations we present in section [ sec : giantfluct ] , buoyancy effects speed up the dynamics of large - scale concentration fluctuations and using the overdamped limit would produce physically incorrect results at small wavenumbers . in microgravity , however , the overdamped limit is valid and we have used it to study giant fluctuations over very long time scales in a number of separate works @xcite . we have implemented two advection schemes for cell - centered scalar fields , and describe under what conditions each is more suitable . the first is a simpler non - dissipative centered advection discretization described in our previous work @xcite . this scheme preserves the skew - adjoint nature of advection and thus maintains fluctuation - dissipation balance in the stochastic context . however , when sharp gradients are present , centered advection schemes require a sufficient amount of dissipation ( diffusion ) in order to avoid the appearance of gibbs - phenomenon instabilities . higher - order godunov schemes have been used successfully with cell - centered finite volume schemes for some time @xcite . in these semi - lagrangian advection schemes , a construction based on characteristics is used to estimate the average value of the advected quantity passing through each cell face during a time step . these averages are then used to evaluate the advective fluxes . our second scheme for advection is the higher - order godunov approach of bell , dawson , and schubin ( bds ) @xcite . additional details of this approach are provided in section [ sub : bds ] . the bds scheme can only be used to advect cell - centered scalar fields such as densities . this is because the scheme operatse on control volumes , and therefore applying it to staggered field requires the use of disjoint control volumes , thereby greatly complicating the advection procedure for non - cell - centered data . we therefore limit ourselves to using the skew - adjoint centered advection scheme described in refs . @xcite to advect momentum . although some godunov schemes for advecting a staggered momentum field have been developed @xcite , they are not at the same level of sophistication as those for cell - centered scalar fields . for example , in ref . @xcite a piecewise constant reconstruction is used , and in ref . @xcite only extrapolation in space is performed but not in time . in our target applications , there is sufficient viscous dissipation to stabilize centered advection of momentum ( note that the mass diffusion coefficient is several orders of magnitude smaller than the kinematic viscosity in typical liquids ) . the bds advection scheme is not skew - adjoint and thus adds some dissipation in regions of sharp gradients that are not resolved by the underlying grid . thus , unlike the case of using centered advection , the spatially discrete ( but still continuous in time ) fluctuating equations do not obey a strict discrete fluctuation - dissipation principle @xcite . nevertheless , in high - resolution schemes such as bds artificial dissipation is added locally in regions where centered advection would have failed completely due to insufficient spatial resolution . furthermore , the bds scheme offers many advantages in the deterministic context and allows us to simulate high peclet number flows with little to no mass diffusion . for well - resolved flows with sufficient dissipation there is little difference between bds and centered advection . note that both advection schemes are spatially second - order accurate for smooth flows . simple advection schemes , such as the centered scheme described in our previous work @xcite , directly computes the divergence of the advective flux @xmath99 evaluated at a specific point in time , where @xmath100 is a cell - centered quantity such as density , and @xmath79 is a specified face - centered velocity . by contrast , the bds scheme uses the multidimensional characteristic geometry of the advection equation @xmath101 to estimate time - averaged fluxes through cell faces over a time interval @xmath102 , given @xmath103 , as well as a face - centered velocity field @xmath104 and a cell - centered source @xmath105 that are assumed _ constant _ over the time interval . in actual temporal discretizations @xmath106 is a midpoint ( second - order ) approximation of the velocity over the time step . similarly , @xmath107 will be a centered approximation of the divergence of the diffusive and stochastic fluxes over the time step . in the description of our temporal integrators , we will use the shorthand notation @xmath108 to denote the approximation to the advective fluxes used in the bds scheme for solving ( [ eq : adv_diff ] ) , @xmath109 bds is a conservative scheme based on computing time - averaged advective fluxes through every face of the computational grid , for example , in two dimensions , @xmath110 where @xmath111 is the given normal velocity at the face , and @xmath112 represents the space - time average of @xmath100 passing through face-@xmath113 in the time interal @xmath114 . the extrapolated face - centered states @xmath112 are computed by first reconstructing a piecewise continuous profile of @xmath115 in every cell that can , optionally , be limited based on monotonicity considerations . the multidimensional characteristic geometry of the flow in space - time is then used to estimate the time - averaged flux ; see the original papers @xcite for a detailed description . in the original advection bds schemes in two dimensions @xcite and three dimensions @xcite , a piecewise - bilinear ( in two dimensions ) or trilinear ( in three dimensions ) reconstruction of @xmath100 was used . subsequently , the schemes were extended to a quadratic reconstruction in two dimensions @xcite . note that handling boundary conditions in bds properly requires additional investigations , and the construction of specialized one - sided reconstruction stencils near boundaries . in our implementation we rely on cubic extrapolation based on interior cells and the specified boundary condition ( dirichlet or neumann ) to fill ghost cell values behind physical boundaries , and then apply the bds procedure to the interior cells using the extrapolated ghost cell values . bds advection , as described in @xcite , does not strictly preserve the eos constraint , unlike centered advection . the characteristic extrapolation of densities to space - time midpoint values on the faces of the grid , @xmath116 and @xmath117 , are not necessarily consistent with the eos , unlike centered advection where they are simple averages of values from neighboring cells , and thus guaranteed to obey the eos by linearity . a simple fix that makes bds preserve the eos , without affecting its formal order of accuracy , is to enforce the eos on each face by projecting the extrapolated values @xmath116 and @xmath117 onto the eos . in the @xmath118 sense , such a projection consists of the update @xmath119 and similarly for @xmath120 , or equivalently , @xmath121 . note that this projection is done on each face only for the purposes of computing advective fluxes and is distinct from any projection onto the eos performed globally . the temporal discretization described in our previous work @xcite was fully explicit , whereas the discretization we employ here is implicit in the viscous dissipation . the implicit treatment of viscosity is traditionally handled by time - splitting approaches , in which a velocity system is solved first , without strictly enforcing the constraint . the solution is then projected onto the space of vector fields satisfying the constraint @xcite . this type of time - splitting introduces several artifacts , especially for viscous - dominated flows ; here we avoid time - splitting by solving a combined velocity - pressure stokes linear system , as discussed in detail in ref . @xcite . the implicit treatment of viscosity in the temporal integrators described in section [ sub : temporaldiscretization ] requires solving discretized unsteady stokes equations for a velocity @xmath79 and a pressure @xmath42 , @xmath122 for given spatially - varying density @xmath8 and viscosity @xmath28 , right - hand sides @xmath123 and @xmath124 , and a coefficient @xmath125 . we solve these linear systems using a gmres krylov solver preconditioned by the multigrid - based preconditioners described in detail in ref . this approach requires only standard velocity ( helmholtz ) and pressure ( poisson ) multigrid solvers , and requires about two - three times more multigrid iterations than solving an uncoupled pair of velocity and pressure subproblems ( as required in projection - based splitting methods ) . there are two issues that arise with the stokes solver in the context of temporal integration that need special care . in fluctuating hydrodynamics , typically the average flow @xmath126 changes slowly and is much larger in magnitude than the fluctuations around the flow @xmath127 . in the predictor stages of our temporal integrators the convergence criterion in the gmres solver is based on relative tolerance . because the right - hand side of the linear system and the residual are dominated by the deterministic flow , it is hard to determine when the fluctuating component of the flow has converged to the desired relative accuracy . in the corrector stage of our predictor - corrector schemes , we use the predicted state as a reference , and switch to using absolute error as the convergence criterion in gmres , using the same residual error tolerance as was used in the predictor stage . this ensures that the corrector stage gmres converges quickly if the predicted state is already a sufficiently accurate solution of the stokes system . another issue that has to be handled carefully is the imposition of inhomogeneous boundary conditions , which leads to a linear system of the form @xmath128 where @xmath129 comes from non - homogeneous boundary conditions . both of these problems are solved by using a residual correction technique to convert the stokes linear system into one for the _ change _ in the velocity and pressure @xmath130 relative to an initial guess or _ reference _ state @xmath131 , which is typically the last known velocity and pressure , _ except _ that the desired inhomogeneous boundary conditions are imposed ; this ensures that boundary terms vanish and the stokes problem for @xmath132 is in homogeneous form . note that any dirichlet boundary conditions for the normal component of velocity should be consistent with @xmath124 , and any dirchlet boundary conditions for the tangential component of the velocity should be evaluated at the same point in time ( e.g. , beginning , midpoint , or endpoint of the time step ) as @xmath124 . in this section we construct temporal integrators for the spatially - discretized low mach number equations , in which we treat viscosity semi - implicitly . for our target applications , the reynolds number is sufficiently small and the schmidt number is sufficiently large that an explicit viscosity treatment would lead to an overall viscous time step restriction , @xmath133 we present temporal integrators in which we avoid fractional time stepping and ensure _ strict _ ( to within solver and roundoff tolerances ) conservation and preservation of the eos constraint . the key feature of the algorithms developed here is the implicit treatment of viscous dissipation , without , however , using splitting between the velocity and pressure updates , as discussed at more length in the introduction . the feasibility of this approach relies an efficient solver for stokes systems on a staggered grid @xcite ; see also section [ sub : gmres ] for additional details . in the temporal integrators developed here , we treat advection explicitly , which limits the advective courant number to @xmath134 mass diffusion is also treated explicitly since it is typically much slower than momentum diffusion and in many examples also slower than advection . explicit treatment of mass diffusion leads to an additional stability limit on the time step since the diffusive courant number must be sufficiently small , @xmath135 where @xmath136 is the number of spatial dimensions and @xmath80 is the grid spacing . implicit treatment of mass diffusion is straightforward for incompressible flows , see algorithm 2 in ref @xcite , but is much harder for the low mach number equations due to the need to maintain the eos constraint ( [ eq : eos_quasi_incomp ] ) via the constraint ( [ eq : div_v_constraint ] ) . even with explicit mass diffusion , provided that the reynolds is sufficiently small and the schmidt number is sufficiently large , a semi - implicit viscosity treatment results in a much larger allowable time step . in algorithm [ alg : lminertial ] we give the steps involved in advancing the solution from time level @xmath137 by a time interval @xmath102 to time level @xmath138 , using a semi - implicit trapezoidal temporal integrator @xcite for the inertial fluctuating low mach number equations ( [ eq : momentum_eq],[eq : rho1_eq],[eq : div_v_constraint],[eq : rho_eq ] ) . in algorithm [ alg : lmoverdamped ] we give an explicit midpoint temporal integrator @xcite for the overdamped low mach number equations ( [ eq : rho1_eq],[eq : rho_eq],[eq : steady_stokes ] ) . in order to ensure strict conservation of mass and momentum , we evolve the momentum density @xmath139 and the mass densities @xmath78 and @xmath8 ( an equally valid choice is to evolve @xmath78 and @xmath120 ) . whenever required , the primitive variables @xmath140 and @xmath141 are computed from the conserved quantities . unlike the incompressible equations , the low mach number equations require the enforcement of the eos constraint ( [ eq : eos_quasi_incomp ] ) at every update of the mass densities @xmath78 and @xmath8 , notably , both in the predictor and the corrector stages . this requires that the right - hand side of the velocity constraint ( [ eq : div_v_constraint ] ) be consistent with the corresponding diffusive fluxes used to update @xmath78 . in order to preserve the eos and also maintain strict conservation , algorithms [ alg : lminertial ] and [ alg : lmoverdamped ] use a splitting approach , in which we first update the mass densities and then we update the velocity using the updated values for the density @xmath8 and the diffusive fluxes that will be used to update @xmath78 . note , however , that after many time steps the small errors in enforcing the eos due to roundoff and solver tolerances can accumulate and lead to a systematic drift from the eos . this can be corrected by periodically projecting the solution back onto the eos using an @xmath118 projection , see section iii.c in ref . @xcite . in our presentation of the temporal integrators , we use superscripts to denote where a given quantity is evaluated , for example , @xmath142 . even though we use continuum notation for the divergence , gradient and laplacian operators , it is implicitly understood that the equations have been discretized in space . the white - noise random tensor fields @xmath32 and @xmath33 are represented via one or two collections of i.i.d . uncorrelated normal random variables @xmath143 and @xmath144 , generated independently at each time step , as indicated by superscripts and subscripts @xcite . spatial discretization adds an additional factor of @xmath145 due to the delta function correlation of white - noise , where @xmath146 is the volume of a grid cell @xcite . for simplicity of notation we denote @xmath147 . 1 . compute the diffusive / stochastic fluxes for the predictor . note that these can be obtained from step [ enu : inertialcorrector ] of the previous time step , @xmath148 2 . take a predictor forward euler step for @xmath78 , and similarly for @xmath120 , or , equivalently , for @xmath8 , @xmath149 3 . compute @xmath150 and calculate corrector diffusive fluxes and stochastic fluxes , @xmath151 4 . take a predictor crank - nicolson step for the velocity , using @xmath152 as a reference state for the residual correction form of the stokes system , @xmath153+\grad\cdot\left(\sqrt{\frac{\eta^{n}k_{b}t}{\d t\,\d v}}\,\overline{\m w}^{n}\right),\\ \grad\cdot\v v^{\star , n+1 } & = & -\grad\cdot\left(\beta\rho^{-1}\v f^{\star , n+1}\right).\end{aligned}\ ] ] [ enu : corradv]take a corrector step for @xmath78 , and similarly for @xmath120 , or , equivalently , for @xmath8 , @xmath154 where @xmath155 and @xmath156 . [ enu : inertialcorrector]compute @xmath157 and compute @xmath158 6 . take a corrector step for velocity by solving the stokes system , using @xmath159 as a reference state , @xmath160\\ \grad\cdot\v v^{n+1 } & = & -\grad\cdot\left(\beta\rho^{-1}\v f^{n+1}\right).\end{aligned}\ ] ] . several variants of the inertial algorithm [ alg : lminertial ] preserve deterministic second - order accuracy . for example , in the corrector stage for @xmath78 , for centered advection we use a trapezoidal approximation to the advective flux , @xmath161 but we could have also used a midpoint approximation @xmath162 without affecting the second - order weak accuracy @xcite . note that bds advection by construction requires a midpoint approximation to the advective velocity ; no analysis of the order of stochastic accuracy is available for bds advection at present . in the corrector step for velocity , in algorithm [ alg : lminertial ] we use corrected values for the viscosity , but one can also use the values from the predictor @xmath163 . 1 . calculate predictor diffusive fluxes and generate stochastic fluxes for a half step to the midpoint , @xmath164 2 . generate a random advection velocity by solving the steady stokes equation with random forcing , @xmath165 3 . take a predictor midpoint euler step for @xmath78 , and similarly for @xmath120 , or , equivalently , for @xmath8 , @xmath166 and compute @xmath167 . 4 . calculate corrector diffusive fluxes and generate stochastic fluxes , @xmath168 where @xmath169 is a collection of random numbers generated independently of @xmath170 . solve the corrected steady stokes equation @xmath171+\rho^{\star , n+\myhalf}\v g\\ \grad\cdot\v v^{\star , n+\myhalf } & = & -\grad\cdot\left(\beta\rho^{-1}\v f^{\star , n+\myhalf}\right).\end{aligned}\ ] ] 6 . correct @xmath78 , and similarly for @xmath120 , or , equivalently , for @xmath8 , @xmath172 and set @xmath157 . for explicit temporal integrators , we relied on a gauge formulation to write the low mach equations in the form of a standard unconstrained initial - value problem , thus allowing us to use standard integrators for odes @xcite . in the semi - implicit case , however , we do not use a gauge formulation because the stokes solver we use works directly with the pressure and velocity . this makes proving second - order temporal accuracy nontrivial even in the deterministic context ; we therefore rely on empirical convergence testing to confirm the second - order deterministic accuracy . in the stochastic context , there is presently no available theoretical analysis when bds advection is employed ; existing analysis @xcite assumes a method of lines ( mol ) discretization in which space is discretized first to obtain a system of sodes . for centered advection , which does lead to an mol discretization , the algorithms used here are based on the second - order weak temporal integrators developed in ref . in particular , for the case of the inertial equations ( [ eq : momentum_eq],[eq : rho1_eq],[eq : div_v_constraint],[eq : rho_eq ] ) we base our temporal integrator on an implicit trapezoidal method . it should be emphasized however that the analysis in ref . @xcite applies to unconstrained langevin systems , while the low mach equations are constrained by the eos . nevertheless , the deterministic accuracy of the method is crucial even when fluctuations of primary interest because in linearized fluctuating hydrodynamics the fluctuations are linearized around the solution of the determinisitc equations , which must itself be computed numerically @xcite accurately in order to have any chance to compute the fluctuations accurately . for the case of the overdamped equations ( [ eq : rho1_eq],[eq : rho_eq],[eq : steady_stokes ] ) , we base our temporal integrator on an implicit midpoint method . in this case the analysis presented in ref . @xcite does apply since the velocity is not a variable in the overdamped equations and the limiting equation for concentration is unconstrained . this analysis indicates that the overdamped temporal integrator below is second - order weakly accurate for the _ linearized _ overdamped low mach number equations . in this section we apply the inertial and overdamped low mach algorithms described in section [ sec : spatiotemporal ] in a stochastic and several deterministic contexts . first , we demonstrate our ability to accurately model equilibrium fluctuations by analyzing the static spectrum of the fluctuations . next , we confirm the second - order deterministic order of accuracy of our methods on a low mach number lid - driven cavity test . next , we confirm that the bds advection scheme enables robust simulation in cases when there is little or no mass diffusion ( i.e. , nearly infinite peclet number ) . lastly , we use the inertial algorithm to simulate the development of a kelvin - helmholtz instability when a lighter less viscous fluid is impulsively set in motion on top of a heavier more viscous fluid . one of the key quantities used to characterize the intensity of _ equilibrium _ thermal fluctuations is the static structure factor or static spectrum of the fluctuations at thermodynamic equilibrium . we examine the static structure factors in both the inertial and overdamped regimes . we use arbitrary units with @xmath173 , @xmath174 , molecular masses @xmath175 @xmath176 , and pure component densities @xmath177 , @xmath178 . we initialize the domain with @xmath179 which gives @xmath180 . the diffusion coefficient was constant@xmath181 , whereas the viscosity varies linearly from @xmath182 to @xmath183 ( for the inertial tests ) , and from @xmath182 to @xmath184 ( for the overdamped tests ) as @xmath9 varies from @xmath185 to @xmath186 , but note that at equilibrium the concentration fluctuations are small so the viscosity varies little over the domain . we assume an ideal mixture , giving chemical potential @xmath187 $ ] @xcite . at these conditions , the equilibrium density variance is @xmath188 , where @xmath189 is the volume of a grid cell ( see appendix a1 in @xcite ) . we use a periodic system with @xmath190 grid cells with @xmath191 , with the thickness in the third direction set to give a large @xmath192 and thus small fluctuations , ensuring consistency with linearized fluctuating hydrodynamics . a total of @xmath193 time steps are skipped in the beginning to allow equilibration of the system , and statistics are then collected for an additional @xmath194 steps . we run both the inertial and overdamped algorithms using three different time steps , @xmath195 @xmath196 and @xmath197 , the largest of which corresponds to 40% of the maximum allowable time step by the explicit mass diffusion cfl condition . .[tab : srho ] equilibrium static structure factor @xmath198 averaged over all wavevectors for the inertial and overdamped algorithms using three different time steps . the exact solution from theory is @xmath199 , allowing us to estimate an order of accuracy from the average error over all wavenumbers . note that there are significant statistical errors present , especially at small wavenumbers , and these make it difficult to reliably estimate the asymptotic order of accuracy empirically when the error is very small ( as for the overdamped integrator ) . [ cols="^,^,^,^,^",options="header " , ] in order to extract a relaxation time , we fit the numerical results for the normalized time - correlation function to an analytical formula . for the first four wavenumbers @xmath200 , clear oscillations ( propagative modes ) were observed , as illustrated in the left panel of fig . [ fig : watergly_tau ] . for these wavenumbers we used the fit @xmath201 where the relaxation time @xmath202 , the coefficient @xmath203 and the period of oscillation @xmath204 are the fitting parameters . for the remaining wavenumbers , we used a double - exponential decay for the fitting , @xmath205 where @xmath206 , @xmath207 and @xmath208 are the fitting parameters . this leads to good fits for @xmath209 ; for a few transitional wavenumbers such as @xmath210 the fit is not as good , as illustrated in the left panel of fig . [ fig : watergly_tau ] . from the fit ( [ double ] ) we obtain the relaxation time @xmath202 as the point at which the amplitude decays by @xmath211 . a similar procedure was also used to obtain the relaxation time from the experimental data of croccolo et al . @xcite for all wavenumbers . the experimental data shows monotonically decaying correlation functions @xmath212 for _ all _ measured wavenumbers , not consistent with the oscillatory correlation function observed for the four smallest wavenumbers in our simulations @xcite , see the left panel of fig . [ fig : watergly_tau ] . we believe that this mismatch is due to the way measurements for different wavenumbers of similar modulus are averaged in the experimental calculations . in our two - dimensional simulations , we do not perform any averaging over wavenumbers . we believe that the experimentally measured time correlation functions capture the _ real _ part of the decay times _ only _ and thus have the form of a sum of exponentials . due to the lower time resolution and the fact that the static structure factor is not known , for the experimental data we used a single exponential fit and added an offset to account for the background noise , @xmath213 dynamics of concentration fluctuations during free - diffusive mixing of water and glycerol . ( left ) numerical results for the time correlation functions for several selected wavenumbers about @xmath214s from the beginning of the experiment . symbols indicate results from the simulations and lines of the same color indicate the fit to ( [ trig ] ) for the first ( @xmath215 ) and fourth ( @xmath216 ) wavenumbers , or to ( [ double ] ) for the remaining wavenumbers . note that the statistical errors due to the finite averaging increase with time and the tails of the correlation functions are not reliably estimated . ( right ) relaxation or decay times as a function of wavenumber at several points in time . empty symbols show results from computer simulations , and filled from experimental measurements @xcite.,title="fig:",scaledwidth=49.0%]dynamics of concentration fluctuations during free - diffusive mixing of water and glycerol . ( left ) numerical results for the time correlation functions for several selected wavenumbers about @xmath214s from the beginning of the experiment . symbols indicate results from the simulations and lines of the same color indicate the fit to ( [ trig ] ) for the first ( @xmath215 ) and fourth ( @xmath216 ) wavenumbers , or to ( [ double ] ) for the remaining wavenumbers . note that the statistical errors due to the finite averaging increase with time and the tails of the correlation functions are not reliably estimated . ( right ) relaxation or decay times as a function of wavenumber at several points in time . empty symbols show results from computer simulations , and filled from experimental measurements @xcite.,title="fig:",scaledwidth=49.0% ] in the right panel of fig . [ fig : watergly_tau ] we compare simulation and experimental results for the real part of the decay or relaxation time @xmath217 , at several points in time measured from the beginning of the experiment . good agreement is observed between the two with the same qualitative trends : a diffusive relaxation time @xmath218 for large wavenumbers , with a maximum at @xmath219 , and then another minimum at @xmath220 . note that decay times are not reported by croccolo _ et al _ @xcite for wavenumbers @xmath221 since that work focuses on the effect of gravity for @xmath222 . in our analysis of the experimental data we included all measured wavenumbers , including those for which propagative modes are observed . here , the diffusion coefficient varies with concentration according to ( [ eq : chi_c ] ) ; very similar results for the relaxation times were obtained by keeping @xmath223 constant , as suggested by croccolo _ et al _ @xcite . this indicates that the dynamic correlations are not very sensitive to the concentration dependence @xmath224 . in future work we will perform a more careful comparison to experiments . extracting the static structure factor from experimental measurements is complicated by several factors , including the presence of optical prefactors such as the transfer function of the instrument , and the appearance of additional contributions to the scattered light intensity such as shot noise , contributions due to giant temperature fluctuations @xcite , and capillary waves @xcite . we therefore study the evolution of the static structure factor using simulations only . in the left panel of fig . [ fig : watergly_s_k ] we show numerical results for the static structure factor @xmath225 of the discrete concentration field averaged along the @xmath226-axes , at a series of times @xmath227 chosen to match those of the experimental measurements . instead of ensemble averaging , here we performed a temporal average of the spectrum of the vertically - averaged concentration over a period of @xmath228s , _ ending _ at the time indicated in the legend of the figure . the characteristic @xmath229 power law decay at large wavenumbers and the plateau at small wavenumbers predicted by ( [ eq : s_c_c ] ) are clearly observed in fig . [ fig : watergly_s_k ] , consistent with a value of @xmath124 decreasing with time . a quantitative difference is seen between the results for variable and constant diffusion coefficients , consistent with a different value of the imposed concentration gradient @xmath124 due to the somewhat different evolution of @xmath230 . evolution of the static structure factor during free - diffusive mixing of water and glycerol . ( left ) with earth gravity . solid lines show results for simulations in which @xmath224 depends on concentration according to ( [ eq : chi_c ] ) , while dashed lines of the same color correspond to simulations in which @xmath223 is constant . fluctuations at large wavenumbers follow a @xmath229 power law but are damped by gravity at small wavenumbers . ( right ) without gravity ; observe the large difference in the vertical axes showing `` giant '' fluctuations in the microgravity case . note that these are results from a single simulation , mimicking a single experimeent , and therefore there are large statistical uncertainties at small wavenumbers ( large decorrelation times).,title="fig:",scaledwidth=49.0%]evolution of the static structure factor during free - diffusive mixing of water and glycerol . ( left ) with earth gravity . solid lines show results for simulations in which @xmath224 depends on concentration according to ( [ eq : chi_c ] ) , while dashed lines of the same color correspond to simulations in which @xmath223 is constant . fluctuations at large wavenumbers follow a @xmath229 power law but are damped by gravity at small wavenumbers . ( right ) without gravity ; observe the large difference in the vertical axes showing `` giant '' fluctuations in the microgravity case . note that these are results from a single simulation , mimicking a single experimeent , and therefore there are large statistical uncertainties at small wavenumbers ( large decorrelation times).,title="fig:",scaledwidth=49.0% ] in the right panel of fig . [ fig : watergly_s_k ] we show numerical results for the static structure factor that would be obtained had the experiment been performed in microgravity ( @xmath231 ) . in this case , we use the overdamped algorithm [ alg : lmoverdamped ] since there is a persistent large separation of time scales between the slow concentration and fast velocity . we see clear development of a @xmath229 power law as predicted by ( [ eq : s_c_c ] ) for @xmath231 . note that here the concentration gradient is established instantaneously , in fact , it is the largest in the initial configuration and then decays on the diffusive time scale ; this is different from simulations of the development of giant fluctuations in microgravity during the gradflex experiment reported in @xcite , in which the concentration gradient is slowly established on a diffusive scale . the results in fig . [ fig : watergly_s_k ] show that it takes some time for the giant fluctuations at smallest wavenumbers to develop ; the diffusive relaxation time corresponding to the smallest wavenumber studied , @xmath232 is @xmath233s . after a time @xmath234 , the amplitude of the fluctuations @xmath235 decays slowly due to the diffusive mixing , and eventually the system will fully mix and reach thermodynamic equilibrium . we have developed a low mach number algorithm for diffusively - mixing mixtures of two liquids with potentially different density and transport coefficients . in the low mach number setting , the incompressible constraint is replaced by a quasi - compressible constraint that dictates that stochastic and diffusive mass fluxes must create local expansion and contraction of the fluid to maintain a constant thermodynamic ( base ) pressure . we employed a uniform - grid staggered - grid spatial discretization @xcite . following prior work in the incompressible simple - liquid case @xcite , we treated viscosity implicitly without splitting the pressure update , relying on a recently - developed variable - coefficient stokes solver @xcite for efficiency . this approach works well for any reynolds number , including the viscous - dominated overdamped ( zero reynolds number ) limit , even in the presence of nontrivial boundary conditions . furthermore , by using a high - resolution bds scheme @xcite to advect the concentration we robustly handled the case of no mass diffusion ( no dissipation in the concentration equation ) . in our spatial discretization we strictly preserved mass and momentum conservation , as well as the equation of state ( eos ) constraint , by using a finite - volume ( flux - based ) discretization of advective fluxes in which fluxes are computing using extrapolated values of concentration and density that obey the eos . our temporal discretization used a predictor - corrector integrator that treats all terms except momentum diffusion ( viscosity ) explicitly @xcite . we empirically verified second - order spatio - temporal accuracy in the deterministic method . in the stochastic context , establishing the weak order of accuracy is nontrivial in the general low mach number setting . for centered advection our temporal integration schemes can be shown to be second - order accurate for the special case of a boussinesq constant - density ( incompressible ) approximation , or in the overdamped ( inertia - free ) limit . existing stochastic analysis does not apply to the case of bds advection because godunov schemes do not fit a method - of - lines approach , but rather , employ a space - time construction of the fluxes . the presence of nontrivial density differences between the pure fluid components and nonzero mass diffusion coefficient , complicates the analysis even for centered advection , due to the presence of a nontrivial eos constraint on the fluid dynamics . it is a challenge for future work to develop improved numerical analysis of our schemes in both the deterministic and the stochastic setting . in future work , we will demonstrate how to extend the algorithms proposed here to multispecies mixtures of liquids using a generalization of the low mach number constraint . the nontrivial multispecies formulation of the diffusive and stochastic mass fluxes has already been developed by some of us in the compressible setting @xcite . it is also possible to include thermal effects in our formulation , by treating the temperature in a manner similar to the way we treated concentration here . two key difficulties are constructing a spatial discretization that ensures preservation of an appropriately generalized eos , as well as developing temporal integrators that can handle the moderate separation of time scales between the ( typically ) slower heat diffusion and ( typically ) faster momentum diffusion . in particular , it seems desirable to also treat temperature implicitly . such implicit treatment of mass or heat diffusion is nontrivial because it would require solving coupled ( via the eos constraint ) velocity - temperature or velocity - concentration linear systems , and requires further investigation . in the staggered - grid based discretization developed here , we can only employ existing higher - order godunov advection schemes for the cell - centered scalar fields such as concentration and density . it is a challenge for future work to develop comparable methods to handle advection of the staggered momentum field . this would enable simulations of large reynolds number flows . it should be noted , however , that our unsplit approach is most advantageous at small reynolds numbers . a challenge for future work on low mach number fluctuating hydrodynamics is to account for the effects of surface tension in mixtures of immiscible or partially miscible liquids . this can be most straightforwardly accomplished by using a diffuse - interface model , as some of us recently did in the compressible setting for a single - fluid multi - phase system @xcite . one of the key challenges is handling the fourth - order derivative term in the concentration equation in a way that ensures stability of the temporal integrator , as well as developing a consistent discretization of the korteweg stresses on a staggered grid @xcite . the semi - implicit temporal integrators we described here can deal well with a broad range of reynolds or schmidt numbers in the deterministic ( smooth ) setting . in the context of fluctuating hydrodynamics , however , all modes are thermally excited and treatment of viscosity based on a crank - nicolson method ( implicit midpoint rule ) are bound to fail for sufficiently large schmidt numbers ( or sufficiently low reynolds numbers ) . in this work we solved this problem for the case of infinite schmidt , zero reynolds number flows by taking an overdamped limit of the original inertial equations before temporal discretization . it is a notable challenge for the future to develop uniformly accurate temporal integrators that work over a broad range of reynolds or schmidt numbers , including the asymptotic overdamped limit , in the presence of thermal fluctuations . we would like to thank fabrizio croccolo and alberto vailati for sharing their experimental data on water - glycerol mixing , as well as numerous informative discussions . this material is based upon work supported by the u.s . department of energy office of science , office of advanced scientific computing research , applied mathematics program under award number de - sc0008271 and under contract no . de - ac02 - 05ch11231 . additional support for a. donev was provided by the national science foundation under grant dms-1115341 .

we show that if an interlacing particle system in a two - dimensional lattice is a determinantal point process , and the correlation kernel can be expressed as a double integral with certain technical assumptions , then the moments of the fluctuations of the height function converge to that of the gaussian free field . in particular , we examine a specific particle system that was previously studied in and compute the corresponding green s function .

1109.4444

the gaussian free field is widely considered to be a universal object describing the flucutations of heights of random surfaces . previous work has rigorously shown this to be the case in specific models ( @xcite,@xcite ) . in this paper , we show that if an interlacing particle system in two dimensions can be described as a determinantal point process , and mild assumptions are made about the correlation kernel , then the covariances of the fluctuations of the height function are governed by a particular green s function . a general formula for the green s function is given . in particular , we will use this general theorem to show that in an interlacing particle system that arises from the representation theory of the orthogonal groups , the green s function is @xmath0 note that @xmath1 is the green s function for the laplace operator with dirichlet boundary conditions on the set @xmath2 . it turns out that there is a map @xmath3 from the surface to @xmath4 . we will show that the fluctuations of the height function converge to a gaussian process whose covariance is given by the pullback of @xmath1 under @xmath3 . * particle system . * now let us describe this particle system , which was the initial motivation for this paper . introduce coordinates on the plane as shown in figure [ configur ] . denote the horizontal coordinates of all particles with vertical coordinate @xmath5 by @xmath6 , where @xmath7 . there is a wall on the left side , which forces @xmath8 for @xmath5 odd and @xmath9 for @xmath5 even . the particles must also satisfy the interlacing conditions @xmath10 for all meaningful values of @xmath11 and @xmath5 . [ configur ] by visually observing figure [ configur ] , one can see that the particle system can be interpreted as a stepped surface . we thus define the height function at a point to be the number of particles to the right of that point . define a continuous time markov chain as follows . the initial condition is a single particle configuration where all the particles are as much to the left as possible , i.e. @xmath12 for all @xmath13 . this is illustrated in the left - most iamge in figure [ jumps ] . now let us describe the evolution . we say that a particle @xmath14 is blocked on the right if @xmath15 , and it is blocked on the left if @xmath16 ( if the corresponding particle @xmath17 or @xmath18 does not exist , then @xmath14 is not blocked ) . each particle has two exponential clocks of rate @xmath19 ; all clocks are independent . one clock is responsible for the right jumps , while the other is responsible for the left jumps . when the clock rings , the particle tries to jump by 1 in the corresponding direction . if the particle is blocked , then it stays still . if the particle is against the wall ( i.e. @xmath20}=0 $ ] ) and the left jump clock rings , the particle is reflected , and it tries to jump to the right instead . when @xmath14 tries to jump to the right ( and is not blocked on the right ) , we find the largest @xmath21 such that @xmath22 for @xmath23 , and the jump consists of all particles @xmath24 moving to the right by 1 . similarly , when @xmath14 tries to jump to the left ( and is not blocked on the left ) , we find the largest @xmath25 such that @xmath26 for @xmath27 , and the jump consists of all particles @xmath28 moving to the left by 1 . in other words , the particles with smaller upper indices can be thought of as heavier than those with larger upper indices , and the heavier particles block and push the lighter ones so that the interlacing conditions are preserved . [ jumps ] figure [ jumps ] depicts three possible first jumps : left clock of @xmath29 rings first ( it gets reflected by the wall ) , then right clock of @xmath30 rings , and then left clock of @xmath29 again . in terms of the underlying stepped surface , the evolution can be described by saying that we add possible `` sticks '' with base @xmath31 and arbitrary length of a fixed orientation with rate 1/2 , remove possible `` sticks '' with base @xmath31 and a different orientation with rate 1/2 , and the rate of removing sticks that touch the left border is doubled . cube . if one uses the dual convention that this is a cube - shaped hole then the orientations of the sticks to be added and removed have to be interchanged , and the tiling representations of the sticks change as well . ] a computer simulation of this dynamics can be found at + http://www.math.caltech.edu/papers/orth_planch.html[@xmath32 . this particle system falls in the universality class of the anisotropic kardar - parisi - zhang ( akpz ) equation with a wall . the kpz equation was first introduced in @xcite and is of interest to physicists , see @xcite . similar markov chains have been previously studied in @xcite without the wall , and in @xcite with a different ( `` symplectic '' ) interaction with the wall . * more general particle systems . * by following the proof , the author realized that a more general statement could be proved . if a point process in a two - dimensional lattice is determinantal , and the correlation kernel can be expressed as a double integral with certain technical assumptions ( see definition [ normal ] below ) , then the moments of the fluctuations of the height function can be governed by a green s function . the exact statement is theorem [ maintheorem ] . we then can use this theorem to determine the green s function for the specific point process described above . * motivations . * there are three reasons for proving the results in the more general case . the first reason is that the proofs are not much more difficult . the second reason is that it is easier to check the conditions for the general case than it is to repeat the full calculations . the third reason is that the general result tells us that the formula for the green s function only depends on @xmath33 , and therefore should only depend on the horizontal movement of the particles . for instance , in the model of @xcite , the expression for @xmath1 was explicitly computed . in the model of @xcite , the particles had two different jump rates , and the expression for @xmath1 was the same . our general results should imply , for example , that @xmath1 will be the same for any number of different jump rates . in order for a model to have a different expression for @xmath1 , there should be some change along the horizontal direction . for example , in the model considered in this paper , a reflecting wall is added to the model of @xcite , and we see that @xmath1 is changed . the correlation kernels in previous problems ( @xcite,@xcite ) should satisfy the conditions of the more general case , although the author has not rigorously checked the details . * * there are several statements which should be true , but are not pursued in this paper . one of them concerns the one - point fluctuations . namely , after a scaling of @xmath34 , the fluctuations of the height function at a single point should converge in moments to a standard gaussian . the proof would consist mainly in proving that the variance is proportional to @xmath35 , which by @xcite immediately implies the convergence to a gaussian . the appropriate scaling was first demonstrated in @xcite using a renormalization group argument . using this convergence , one can then show that the convergence to the gaussian free field occurs in distribution , not just in moments . analogous ( rigorous ) statements are found as theorem 1.2 and proposition 5.6 in @xcite . it should also be possible to prove similar results when @xmath36 , but this was also not pursued . * organization of paper . * in section [ statement of the main theorem ] , the theorem for more general particle systems ( theorem [ maintheorem ] ) is stated . in sections [ algebrageneral ] and [ analysisgeneral ] , the algebraic and analytic steps of the proof are given . in section [ specific results ] , it is shown that the particle system described above satisfies the conditions of theorem [ maintheorem ] . section [ asymptotics ] deals with the necessary asymptotic analysis . * acknowledgments . * the author would like to thank alexei borodin and maurice duits for insightful comments . the author was supported by the national science foundation s graduate research fellowship program . suppose we have a family of point processes on @xmath37 which runs over time @xmath38 . ( note that these are different co - ordinates from the introduction ) . in other words , at any time @xmath39 , the system selects a random subset @xmath40 . if @xmath41 , then we say that there is a particle at @xmath42 . for any @xmath43 and @xmath44 , let @xmath45 $ ] be defined by @xmath46 assume that there is a map @xmath47 on @xmath48 such that @xmath49_{1\leq i , j\leq k}.\ ] ] the maps @xmath50 and @xmath47 are called the _ kth correlation function _ and the _ correlation kernel _ , respectively . a function @xmath51 on @xmath52 is called a _ conjugating factor _ if there exists another function @xmath53 on @xmath54 such that @xmath55 note that if @xmath51 is a conjugating factor , then @xmath56_{1\leq i , j\leq k } = \det[c(x_i , n_i , x_j , n_j)k(x_i , n_i , x_j , n_j , t)]_{1\leq i , j\leq k}.\ ] ] two kernels @xmath47 and @xmath57 are called _ conjugate _ if @xmath58 for some conjugating factor @xmath51 . if a correlation kernel exists , the point process is called _ determinantal_. on a discrete space , a point process is determined uniquely by its correlation functions ( see e.g. @xcite ) . therefore , if we are given two determinantal point process on a discrete space with conjugate kernels , they must have the same law . the set @xmath59 is called the _ nth level_. given a subset @xmath40 , let @xmath60 be the cardinality of the set @xmath61 . assume that the numbers @xmath60 take constant finite values which are independent of the time parameter @xmath39 . in words , this means that the number of particles on the @xmath62th level is always @xmath60 . further assume that @xmath63 for all @xmath62 . let @xmath64 denote the elements of @xmath61 . a subset @xmath65 is called _ interlacing _ if @xmath66 @xmath67 assume that at any time @xmath39 , the system almost surely selects an interlacing subset . let @xmath68 equal @xmath69 . define the random _ height function _ by @xmath70 @xmath71 in words , @xmath72 counts the number of particles to the right of @xmath42 at time @xmath39 . we wish to study the large - time asymptotics of this particle system . let @xmath73 , n=[n\eta ] , t = n\tau$ ] , where @xmath74 is a large parameter . define @xmath75 to be @xmath76 let @xmath77 be defined by @xmath78 @xmath79 in words , @xmath77 is the fluctuation of the height function around its expectation . before stating the theorem , we need to state some more assumptions on the kernel . suppose the kernel @xmath47 is conjugate to a kernel @xmath57 such that @xmath57 satisfies the following property : there is a number @xmath80 such that whenever @xmath81 , @xmath82 @xmath83 further suppose that for @xmath84 , @xmath85 suppose @xmath86 is a complex - valued function on @xmath87 . to save space , we will sometimes write @xmath88 . expressions such as @xmath89 will be shorthand for @xmath90 , @xmath91 and @xmath92 , respectively . assume @xmath93.also suppose there exists a differentiable map @xmath94 from @xmath95 to the upper half - plane @xmath96 such that @xmath94 is a critical point of @xmath97 . in other words , @xmath98 note that @xmath3 need not be onto . for any @xmath99 , if the set @xmath100 is nonempty , let @xmath101 denote its supremum . [ normal ] with the notation above , a determinantal point process on @xmath102 is _ normal _ if all of the following hold : * for all @xmath103 , the limit @xmath104 exists and is a positive real number . * for all @xmath103 , as @xmath105 approaches @xmath101 from the left , @xmath106 . * @xmath47 is conjugate to some @xmath57 such that - hold for some integer @xmath80 . * set @xmath107 , @xmath108 $ ] and @xmath109 $ ] for @xmath110 , where @xmath111 . let @xmath112 denote @xmath113 and let @xmath114 denote @xmath115 . for finite integers @xmath116 and @xmath117 , as @xmath118 @xmath119 where @xmath120 and @xmath121 are steepest descent paths , @xmath122 , and @xmath123 are complex - valued meromorphic functions satisfying the identity @xmath124 here , we have written @xmath125 for @xmath126 . * for any @xmath127 , the following indefinite integral satisfies @xmath128 where the sum is taken over all @xmath129-cycles in @xmath130 and the indices are taken cyclically . * for any finite interval @xmath131 $ ] , @xmath132 $ ] and the lesbesgue measure of the set @xmath133 : i'(x)\in 2\pi{\mathbb{z}}+[-\delta,\delta]\}$ ] is @xmath134 for some positive @xmath135 . the following remarks will help explain the definition . [ normalexplanation ] ( 1 ) the assumption that @xmath136 occurs naturally . one often finds that for @xmath137 , @xmath138+k_1,[n\eta],[n\nu]+k_2,[n\eta],n\tau)=\frac{1}{2\pi i}\int_{\bar{{\omega}}}^{{\omega } } \frac{dz}{z^{k_1-k_2 + 1}}=\frac{\im({\omega}^{k_2-k_1})}{\pi(k_2-k_1)},\ ] ] where the contour crosses the positive real line . by setting @xmath139 , we see that the right hand side reduces to the ubiquitous sine kernel . when @xmath140 , we see that @xmath141,[n\eta])=\frac{1}{2\pi}(\log{\omega}-\log\bar{{\omega}})=\frac{\arg{\omega}(\nu,\eta,\tau)}{\pi}.\ ] ] since the left hand side equals zero , we expect @xmath142 . \(2 ) since @xmath93 , this means that @xmath97 has a critical point at both @xmath3 and @xmath143 . as @xmath3 approaches the real line , the two critical points coalesce into a triple zero , so @xmath144 converges to @xmath145 as @xmath39 approaches @xmath101 . we need a control for how quickly this convergence to @xmath145 occurs , in order to order to control the behavior near the boundary of @xmath95 . more specifically , it controls the bound in proposition [ edgeasymptotics ] . there is a heuristic understanding for why ( 2 ) should hold . the function @xmath97 has two critical points which coalesce into a triple zero . the simplest example of such a function is @xmath146 as @xmath39 approaches @xmath145 . in this case , the solution to @xmath147 is @xmath148 . then @xmath149 . \(3 ) assumptions will be elucidated when we interpret the particles as lozenges . in particular , see remark [ lozengeremark ] . \(4 ) it is common for the kernel to be expressed in this form ; previous examples are @xcite and @xcite . if the kernel has a different expression with the same asymptotics as in propositions [ prop6.9 ] and [ edgeasymptotics ] , the results still hold . \(5 ) in particular , holds if there always exist @xmath150-substitutions and an expression @xmath151 such that @xmath152 where @xmath153 and @xmath154 . this is because of lemma 7.3 in @xcite , which refers back to @xcite , which says that @xmath155 \(6 ) this is a technical lemma which allows lemma [ zerolemma ] to be applied . we can now state the main theorem . [ maintheorem ] suppose we are given a normal determinantal point process . for @xmath156 , let @xmath157 be distinct points in @xmath95 , and let @xmath158 . define the function @xmath1 on the upper half - plane to be @xmath159 then @xmath160 where @xmath161 is the set of all involutions in @xmath162 without fixed points . we note that these are the moments of a linear family of gaussian random variables : see appendix a. using the results of @xcite , it should be possible to show that @xmath163 converges to a gaussian , but this was not pursued . the most natural way to view @xmath54 is as a square lattice . however , it turns out that a hexagonal lattice is more useful . to obtain the hexagonal lattice , take the @xmath62th level and shift it to the right by @xmath164 . see figure [ particles ] . figure [ particles ] also shows that the particle system can be interpreted as lozenges . each lozenge is a pair of adjacent equilateral triangles . see figure [ lozenges ] . by setting the location of each triangle to be the midpoint of its horizontal side , each lozenge can be viewed as a pair @xmath165 , where the black triangle is located at @xmath42 and the white triangle is located at @xmath166 . for example , in figure [ particles ] there are lozenges @xmath167 and @xmath168 . the three types of lozenges can be described as follows . for lozenges of type i , @xmath169 . for lozenges of type ii , @xmath170 . for lozenges of type iii , @xmath171 . note that a lozenge of type i is just a particle . we say that @xmath172 is _ viable _ if @xmath173 , or @xmath174 . a sequence @xmath175 of viable elements is _ non - overlapping _ if @xmath176 are all distinct from each other and @xmath177 are also all distinct from each other . we do , however , allow the possibility of @xmath178 . equal @xmath179 . the black line on the left represents the points where @xmath180 . examples of @xmath181 are @xmath182.,height=192 ] [ particles ] [ lozenges ] the statement and proof of the next proposition are similar to theorem 5.1 of . [ lozengekernel ] suppose the kernel @xmath47 is conjugate to some @xmath57 such that - hold for some @xmath80 . if @xmath183 and @xmath184 @xmath185 is a sequence of non - overlapping viable elements of @xmath52 , then @xmath186_{1\leq i , j\leq k}.\end{gathered}\ ] ] [ lozengeremark ] the equations can now be intuitively understood . equation says that each black triangle is located in exactly one of the three lozenges around it , and equation makes an identical statement for white triangles . equations and say that lozenges of type ii almost surely do not occur far to the right of the particles , with controlling the off - diagonal entries in the determinant and controllling the diagonal entries . similarly , equations and says that lozenges of type iii almost surely do occur far to the right of the particles . this intuition will be exploited in the proof of thereom [ lozengekernel ] . we proceed by induction on the number of lozenges that are not of type i. when this number is zero , the statement reduces to and . for any set @xmath187 of non - overlapping , viable elements , let @xmath188 and @xmath189 denote the left and right hand sides of , respectively . first , as a preliminary statement , we prove that if @xmath190 for @xmath191 , then @xmath192 note that if @xmath193 is replaced by @xmath194 in , the statement is immediate , since the black triangle at @xmath195 must be contained in exactly one lozenge . expand the determinants into a sum over permutations @xmath196 . now split the sum into two parts , one part where the @xmath197 fixes @xmath198 and one part where @xmath197 does not fix @xmath198 . the first part equals @xmath199,\end{gathered}\ ] ] which , by , equals @xmath189 . the second part equals @xmath200.\end{gathered}\ ] ] since @xmath201 by assumption , implies that this expression equals @xmath145 . therefore holds . in a similar manner , if @xmath202 for @xmath191 , then implies that @xmath203 again , the statement holds if @xmath193 is replaced by @xmath194 . now let us prove that @xmath193 and @xmath194 still agree if we add a lozenge of type ii to @xmath204 . suppose that @xmath205 is viable and that @xmath206 is non - overlapping . then equation is equivalent to @xmath207 and the same holds for @xmath194 instead of @xmath193 . by the induction hypothesis , @xmath208 thus , implies @xmath209 assume for now that @xmath210 for @xmath211k . then we cam apply equation , which implies that @xmath212 and the same statement holds for @xmath194 . thus , @xmath213 if @xmath214 is non - overlappinng , then is again applicable . we repeatedly apply and as often as possible . first , suppose that this can be done indefinitely . then @xmath215 since lozenges of type ii almost surely do not appear when we look far to the right of the particles , @xmath216 by expanding the determinant into a sum over @xmath217 , and imply that @xmath218 now suppose that and can only be applied finitely many times . this means that @xmath219 equals either @xmath220 or @xmath221 in the first case , @xmath222 is non non - overlapping . this implies @xmath223 ( because two of the rows are idential ) and @xmath224 ( because a triangle can not be in two different lozenges at the same time ) . thus , @xmath193 and @xmath194 agree . a similar argument holds in the second case . thus , @xmath193 and @xmath194 agree whenever a lozenge of type ii is added to @xmath204 . an identical holds for type iii lozenges , except that we use and instead of and . we have been describing a lozenge as a pair @xmath165 . it can also be described as @xmath225 , where @xmath166 is the location of the white triangle and @xmath226 is the type of the loznege . thus the proposition can be restated as the following statement . [ alternative ] for any non - overlappign @xmath227 , @xmath228_{1\leq i , j\leq k},\end{gathered}\ ] ] where @xmath229 this is a result of the correspondences @xmath230 there are two different formulas for the height function . one formula is @xmath231 it is possible to only use to complete the proof . however , when there are multiple points on one level , i.e. not all @xmath232 are distinct , the computation becomes much more complicated . this is because lozenges of type i will appear in multiple sums of the form . we can avoid this difficulty by introducing another formula for the height function : @xmath233 where , for @xmath234 , @xmath235 therefore , the expression @xmath236\right)\ ] ] can be expressed as a sum of terms of the form @xmath237\prod_{l = k'+1}^k [ h_{n_l , n_l'}(x_l)-\mathbb{e}(h_{n_l , n_l'}(x_l))]\right).\ ] ] assume that the following sets are disjoint : @xmath238 then @xmath239,\ ] ] where the matrix blocks are : @xmath240_{1\leq i , j\leq k ' } \\ a_{12 } & = [ \tilde{k}(s_i , n_i , x_j , p_j , t)]_{1\leq i\leq k',\ k'+1\leq j\leq k } \\ a_{21 } & = [ -\tilde{k}(x_i-\delta_{p_i-1},p_i-1,s_j , n_j , t)]_{k'+1\leq i\leq k,\ 1\leq j\leq k'}\\ a_{22 } & = [ -(1-\delta_{ij})\tilde{k}(x_i-\delta_{p_i-1},p_i-1,x_j , p_j , t)]_{k'+1\leq i , j\leq k}\end{aligned}\ ] ] by applying corollary [ alternative ] to and , we see that @xmath241 equals the right hand side of with the @xmath242 terms removed . it is well - known that subtracting the expectation corresponds to putting zeroes on the diagonal . for example , this is noticed in the proof of theorem 7.2 of @xcite . write the determinant in as a sum over permutations @xmath197 in @xmath162 . if the cycle decomposition of @xmath197 contains the cycle @xmath243 of length @xmath244 and @xmath245 denotes the matrix in the right hand side of , then the contribution from @xmath197 is @xmath246 where @xmath247 correspond to other cycles of @xmath197 . let @xmath248 denote @xmath249 if @xmath250 , and @xmath251 if @xmath252 . since the sum over @xmath248 only affects the matrix terms @xmath253 and @xmath254 , the contribution from @xmath197 is @xmath255 where @xmath256 denote other cycles . in other words , the contribution from @xmath197 can be expressed as a product over the cycles in the cycle decomposition of @xmath197 . note that if @xmath197 fixes any points , then the correponding contribution is zero because all the diagonal entries are zero . in , set @xmath108 , n_l=[n\eta_l],$ ] and @xmath107 . our goal is to find the limit of as @xmath257 . expanding the determinant into a sum over @xmath258 , we just saw that the contribution from a fixed @xmath197 is of the form . first note that if any of the @xmath259 denotes @xmath260 , then @xmath261 this is because each @xmath262 is proportional to @xmath263 ( by proposition [ prop6.9 ] , so @xmath264 is proportional to @xmath265 , but the sum is only taken over @xmath266 terms . therefore , can be expressed as a single term of the form in , and in this term @xmath267 . now we will prove ( stated as theorem [ goalinanalysissection ] below ) that @xmath268 once this is proven , implies that the total contribution from @xmath269 equals zero . when @xmath270 , then the right hand side is just @xmath271 , completing the proof of theorem [ maintheorem ] . recall the definitions of @xmath97 and @xmath3 from section [ statement of the main theorem ] . set @xmath272 to be @xmath273 [ productofkernels ] for @xmath274 , let @xmath275 , @xmath276 $ ] , @xmath277 $ ] and @xmath107 . for @xmath278 , let @xmath279 denote @xmath280 , let @xmath281 denote @xmath282 and let @xmath283 denote @xmath284 . let @xmath285 and @xmath286 . let @xmath287 . then @xmath288 } k(x_1,n_1,y , n_2,t)k(y , n_2,x_3,n_3,t ) \\ = o\left(\frac{1}{n}\right)+\frac{e^{n\re((g_1({\omega}_1)-g_3({\omega}_3)))}}{2\pi n\sqrt{\vert g_1''({\omega}_1)\vert}\sqrt{\vert g_3''({\omega}_3)\vert}}\int_{\gamma_+\cup\gamma_-}\frac{dz}{2\pi g'_{\nu_1}(z)}\\ \times \big [ f({\omega}_1,z)f(z,{\omega}_3)\frac{e^{in\im(g_1({\omega}_1))-i{\theta}_1}}{e^{in\im(g_3({\omega}_3))+i{\theta}_3 } } + f({\bar{\omega}}_1,z)f(z,{\omega}_3)\frac{e^{-in\im(g_1({\omega}_1))+i{\theta}_1}}{e^{in\im(g_3({\omega}_3))+i{\theta}_3}}\\ + f({\omega}_1,z)f(z,{\bar{\omega}}_3)\frac{e^{in\im(g_1({\omega}_1))-i{\theta}_1}}{e^{-in\im(g_3({\omega}_3))-i{\theta}_3 } } + f({\bar{\omega}}_1,z)f(z,{\bar{\omega}}_3)\frac{e^{-in\im(g_1({\omega}_1))+i{\theta}_1}}{e^{-in\im(g_3({\omega}_3))-i{\theta}_3}}\big].\end{gathered}\ ] ] let @xmath289 denote @xmath290,\eta_2,\tau , z)$ ] , let @xmath291 denote @xmath292,\eta_2,\tau)$ ] and @xmath293 denote @xmath294 . fix some @xmath295 and split up the sum into two parts : the first part is from @xmath296 to @xmath297 , while the second sum is from @xmath297 to @xmath298 . since there are no particles to the right of @xmath299 in the limit @xmath257 , the sum from @xmath299 to @xmath300 can be ignored . it is common to refer to the first sum as the bulk and the second sum as the edge . first examine the bulk . by proposition [ prop6.9 ] , @xmath301\\ + \mathcal{o}(g_2''({\omega}_2)^{-4}n^{-3 } ) + \mathcal{o}(g_2''({\omega}_2)^{-7}n^{-4}),\end{gathered}\ ] ] where @xmath302 denotes the other fifteen terms that occur in the sum . first let us examine the error term in the bulk . by ( 2 ) of definition [ normal ] , each term in the error is bounded by @xmath303 and @xmath304 , respectively . there are @xmath305 terms , and since @xmath306 , we must have @xmath307 and @xmath308 . therefore the sum is @xmath309 . now let us return to the main term in the bulk . for eight of the sixteen terms in @xmath302 , the expression @xmath310 cancels in the numerator and the denominator . by propositions [ oneproposition ] , these eight terms are @xmath309 . by [ riemannapproximation ] , the other eight terms equal @xmath311 d\nu+o\left(\frac{1}{n}\right),\end{gathered}\ ] ] where @xmath312 represent the other seven terms . of the eight total terms , four have @xmath313 and four have @xmath314 . for the four terms with the expression @xmath293 , make the substitution @xmath315 . the new integration path is @xmath316 . by taking the partial of with respect to @xmath105 and using the chain rule , @xmath317 which implies @xmath318 for the four terms with @xmath319 , make the substitution @xmath320 . the path of integration is @xmath321 . finally , the integral becomes @xmath322.\end{gathered}\ ] ] now we sum over the edge . by proposition [ edgeasymptotics ] and ( 2 ) of definition [ normal ] , the sum is bounded above by @xmath323 as long as @xmath324 , the sum over the edge is also @xmath309 . [ goalinanalysissection ] for @xmath325 , let @xmath275 and set @xmath276,n_i = n\eta_i$ ] . for @xmath325 , let @xmath279 denote @xmath326 , let @xmath281 denote @xmath282 and let @xmath283 denote @xmath284 . let @xmath327 and @xmath328 . then @xmath329 } \cdots \displaystyle\sum_{y_l>[n\nu_l ] } \prod_{i=1}^l k(y_i , x_i , y_{i+1},x_{i+1},t ) \\ \rightarrow \left(\frac{1}{2\pi}\right)^l \int_{\gamma_1^{+}\cup\gamma_1^{- } } dz_1 \cdots \int_{\gamma^{+}_l\cup\gamma^{-}_l } dz_l \frac{f(z_1,z_2)}{g'_{\nu}(z_1 ) } \ldots \frac{f(z_l , z_1)}{g'_{\nu}(z_l)}.\end{gathered}\ ] ] the indices are taken cyclically . by proposition [ prop6.9 ] , the product has @xmath330 terms . each application of proposition [ productofkernels ] decreases the number of terms by a factor of 4 , so repeated applications of proposition [ productofkernels ] yields the result . we now return to the particle system with a reflecting wall described in the introduction . for notational reasons , it is more convenient to use different co - ordinates . instead of labeling the levels as @xmath331 , it is more convenient to label them as @xmath332 . if the @xmath333 is at least as high as the @xmath334 level , then this will be denoted as @xmath335 . this happens if and only if @xmath336 . using the notation of section [ algebrageneral ] , @xmath337 and @xmath338 . along the horizontal direction , we will use a square lattice , so that the particles live on @xmath339 instead of @xmath340 or @xmath341 . let @xmath342 be defined by @xmath343 let @xmath344 denote the ( normalized ) jacobi polynomial with parameters @xmath345 . the normalization is set so that for any nonzero complex number @xmath346 , @xmath344 satisfies @xmath347 let @xmath348 be defined for nonnegative integers @xmath349 by @xmath350 note that for @xmath351 , @xmath352 by theorem 4.1 of @xcite , the correlation functions are determinantal with kernel @xmath353 @xmath354 where the @xmath346-contour is the unit circle and the @xmath355-contour is a circle centered at the origin with radius bigger than @xmath356 . [ isnormal ] the determinantal point process is normal . the green s function is given by @xmath0 once we prove the point process is normal , the expression for the green s function follows from theorem [ maintheorem ] with @xmath357 @xmath358 in section [ specificalgebra ] , we show that the third condition in definition [ normal ] is satisfied . in section [ specificanalysis ] , we show that the fourth and second conditions are satisfied . since these are conditions are the hardest to prove , we will focus mainly on their proofs . the fifth conditions follows from the substitution @xmath359 and ( 5 ) of remark [ normalexplanation ] . let @xmath360 equal @xmath361 and @xmath362 . then @xmath363 satisfies for @xmath364 . using and the orthogonality relation , it is straightforward to check that and hold . what happens is that in the left hand side of or , one obtains six terms , three of which come from and three of which coe from . the three terms from always sum to @xmath145 , while the three terms from sum to @xmath145 or @xmath356 . now we will prove - when @xmath365 . the term equals zero , so we only need to look at . explicitly , the expression is @xmath366 and we want the asymptotic result when @xmath367 in such a way that @xmath368 is @xmath145 or @xmath356 . expand the paranthetical expression @xmath369 to get two terms , each of which is a double integral . since @xmath370 , the term with @xmath371 goes to zero . for the remaining term , expand @xmath372 to get two terms . for the term with @xmath373 , make the substitution @xmath374 . what remains is @xmath375 now deform the @xmath346-contour to the circle @xmath376 and the @xmath355-contour to the circle @xmath377 , where @xmath378 . with these deformations , @xmath379 , so the double integral goes to zero . however , residues are picked up when the contours pass through each other . these residues equal @xmath380 there is a residue at @xmath381 which equals @xmath382 , and a residue at @xmath383 which equals @xmath145 for @xmath384 and @xmath385 for @xmath386 . since @xmath387 , this proves and when @xmath365 . the case when @xmath388 is similar . it remains to show and . when considering the product of two kernels , we obtain a quadruple integral . after the substitutions @xmath389 and @xmath390 , the part of the integrand that depends on @xmath349 is just @xmath391 . therefore , deforming contours so that @xmath392 gives and . for this section , we need a slightly different expression for the kernel . by ( 40)(42 ) of @xcite , the kernel equals @xmath393 @xmath394 @xmath395 where @xmath396 is any real number , and the arc from @xmath397 to @xmath398 is outside the unit circle and does not cross @xmath399 $ ] . set @xmath400 by proposition 5.1.1 of @xcite , we can take @xmath95 to be @xmath401 where @xmath402 [ inverseomega ] let @xmath403 denote @xmath404 . then @xmath405 in general , @xmath406 where @xmath407 and @xmath244 are @xmath408 let @xmath409 denote the polynomial @xmath410 corresponding to @xmath411 . note that @xmath412 . by definition , @xmath403 is the zero of @xmath413 that is in the upper half - plane . therefore , @xmath414 . now let us return to the proof of the fourth condition in definition [ normal ] . start by examining . expanding the parantheses , we obtain four terms corresponding to @xmath415 , and @xmath416 . for the two terms with @xmath417 , make the substitution @xmath418 . what remains are two terms , corresponding to @xmath416 and @xmath419 . therefore , equals @xmath420 we now need to deform the contours in to steepest descent paths . in other words , we need @xmath421 for all @xmath346 on the @xmath346-contour and @xmath422 for all @xmath355 on the @xmath355-contour . by lemma [ inverseomega ] and the definition of @xmath97 , we see that @xmath423 . if @xmath424 , then @xmath425 . since the steepest descent paths can go completely outside the unit circle ( see proposition 5.1.2 of @xcite ) , follows from . , and on the right is @xmath426 . white regions indicate @xmath427 and shaded regions indicate @xmath428 . the double zero occurs at @xmath113 . the arc @xmath355 goes from @xmath397 to @xmath398 . the unit circle has been drawn on the right.,title="fig : " ] [ contours ] if we deform the contours to the steepest descent paths @xmath120 and @xmath121 in figure [ contours ] , we get that asymptotically becomes @xmath429 @xmath430 plus possibly the residues at @xmath431 . since @xmath121 goes outside the unit circle and the critical point of @xmath432 lies inside the unit circle , the second double integral is negligible . now we need to compute the possible residues at @xmath431 . if the contours pass through each other , then the residues at @xmath431 equal @xmath433 @xmath434 where @xmath435 is any complex number satisfying and . see figure [ residuecontours ] . if the contours do not pass through each other , then there is no contribution from the residues . for notational convenience , set @xmath436 and @xmath355 contours from figure [ contours ] . they intersect at @xmath435.,title="fig : " ] [ residuecontours ] it is important to note that @xmath437 is arbitrarily selected . the only requirement on @xmath435 is that it satisfies the inequalities and , and the only requirement on @xmath398 is that @xmath438 . so there exists @xmath378 such that if @xmath439 , then @xmath440 also satisfies those inequalities . now we need to compute and . expanding the parantheses , we get four terms corresponding to @xmath441 . for the terms corresponding @xmath442 and @xmath443 , make the substitution @xmath418 . therefore , the sum of , , , equals @xmath444 where the contour crosses @xmath445 if @xmath446 , and it crosses @xmath447 if @xmath448 . for each integral , deform the contour to a circular arc of constant radius . then the absolute value of the integrand is maximized at the endpoints . to see this , note that @xmath449 is constant on the arc , so it suffices to analyze @xmath450 now note that @xmath451 therefore , @xmath452 increases as we move counterclockwise along the arc in the upper half - plane , and @xmath452 decreases as we move counterclockwise along the arc in the lower half - plane . if @xmath446 , then @xmath453 is maximized when @xmath452 is maximized , which occurs at the endpoints since the contour crosses @xmath445 . likewise , if @xmath448 , then @xmath452 is minimized at the endpoints . thus , in both cases , @xmath453 is maximized at the endpoints . using a standard asymptotic analysis ( see e.g. chapter 3 of @xcite ) , we get that the asymptotic expansion of is @xmath454 for some constants @xmath455 . to complete the proof , notice that if @xmath456 for some selection of @xmath457 , then the asymptotic expansion of the kernel would depend on @xmath437 . but @xmath437 was arbitrarily selected , so this is impossible . now that the fourth condition has been proved , it remains to show that the second condition in definition [ normal ] holds . recall that @xmath458 is the root of @xmath459 that lies in the upper half - plane , where @xmath407 is the polynomial from lemma [ inverseomega ] . we thus need to solve @xmath460 since @xmath461 is a double zero of @xmath462 , we thus have to solve @xmath463 which implies that @xmath464 . in other words , as @xmath105 approahces @xmath101 , @xmath465 . plugging this into the expression for @xmath466 gives the result . [ zerolemma ] suppose that @xmath467 $ ] and @xmath468 $ ] . suppose that as @xmath469 , the lesbesgue measure of the set @xmath133 : i'(x)\in 2\pi{\mathbb{z}}+[-\delta,\delta]\}$ ] is @xmath134 for some positive @xmath135 . let @xmath470 $ ] depend on @xmath74 . then @xmath471 let @xmath472 denote @xmath473 . note that @xmath474 . fix some @xmath475 and consider @xmath476 we bound this sum in two cases . * case 1*. assume @xmath477 $ ] . for @xmath478 , taylor s theorem says that @xmath479+[\frac{1}{2}i''(c_k)(t_k - t_s)^2]=:[i_1(t_k)]+[i_2(t_k)]\ ] ] for some @xmath480 between @xmath481 and @xmath472 . we will prove that @xmath482 using the inequality @xmath483 we have that @xmath484 furthermore , for @xmath485 , @xmath486 also , @xmath487 using , the definition of @xmath488 and @xmath489 , and respectively , @xmath490 * case 2*. assume that @xmath491 . in this case , only a simple estimate is needed : @xmath492 since the estimate in case 1 is much better than the estimate in case 2 , we need an upper bound on how frequently case 2 can occur . in other words , we need an upper bound on the measure of the set @xmath133 : i'(x)\in 2\pi\mathbb{z}+(\delta,\delta)\}$ ] . we assumed that @xmath493 : i'(x)\in 2\pi\mathbb{z}+[\delta,\delta]\}\vert = { \mathcal{o}}(\delta^a).\ ] ] now we need to sum over all @xmath349 in the set @xmath494 . there are @xmath495 terms for which case 2 applies . therefore , @xmath496 and setting @xmath497 yields the result . [ oneproposition ] suppose @xmath498 is a function such that for each @xmath499 , @xmath500 where @xmath501 and @xmath502 satisfy the same assumptions as in lemma [ zerolemma ] . further suppose that the error term @xmath503 is uniform , i.e. @xmath504.\ ] ] then as @xmath257 , @xmath505 let @xmath378 . by assumption , there exists @xmath506 such that for @xmath507 and @xmath508 $ ] @xmath509 by summing over @xmath510 , we see that @xmath511 the proposition now follows from lemma [ zerolemma ] . [ bvlemma ] suppose @xmath501 is a function on @xmath131 $ ] with bounded variation . then as @xmath257 , @xmath512 let @xmath378 . by the definition of a riemann sum , there exists @xmath245 such that @xmath513 furthermore , for @xmath514 , @xmath515 finally , for @xmath516 , @xmath517 [ riemannapproximation ] suppose @xmath518 is a function such that for each @xmath499 , @xmath519 where @xmath501 is a function on @xmath131 $ ] of bounded variation . further suppose that the error term @xmath503 is uniform , i.e. @xmath520.\ ] ] then @xmath521 let @xmath378 . by assumption , there exists @xmath506 such that for @xmath507 and @xmath508 $ ] @xmath522 we can see this by noting that when @xmath523 , @xmath524 . furthermore , increasing @xmath510 by @xmath356 causes @xmath39 to increase by @xmath263 , which causes @xmath525 to increase by @xmath356 . by summing over @xmath510 , we see that @xmath526 the proposition now follows from lemma [ bvlemma ] . [ prop6.9 ] for @xmath110 , let @xmath111 , @xmath112 denote @xmath113 , @xmath114 denote @xmath115 , and @xmath527 denote @xmath528 . with the assumptions from section [ statement of the main theorem ] , @xmath529.\end{gathered}\ ] ] first , we show that the main term is correct . by assumption , we can deform @xmath120 and @xmath121 so that @xmath530 passes through @xmath531 for @xmath110 . the contributions to the integral away from @xmath531 are exponentially small , so we can replace @xmath530 with @xmath532 , where @xmath533 and @xmath534 are steepest descent paths near @xmath112 and @xmath535 , respectively . the integration over @xmath536 expands into four integrations corresponding to @xmath537 . we explicitly do the calculation for @xmath538 . the other three calculations are essentially identical . make the substitutions @xmath539 and @xmath540 . in the neighborhood of @xmath541 and @xmath542 , we have @xmath543 which imply @xmath544 then we get @xmath545 , \end{aligned}\ ] ] where the last equality follows from @xmath546 . the @xmath547 appears because the maps @xmath548 and @xmath549 are both two - to - one . it still remains to show that the error term is correct . the remainder of this section is devoted to proving this . the idea is to reduce the double integral to progressively simpler forms . first , by a reparametrization , the integral over two arcs in @xmath550 can be written as a integral in @xmath551 . second , by using a taylor approximation , the integral in @xmath551 can be written as a product of two integrals in @xmath552 , each of which is of the form @xmath553 , where @xmath554 has a maximum @xmath555 in the interval of integration . third , by using the implicit function theorem , this integral reduces to the form @xmath556 , where the interval of integration is a small neighbourhood @xmath555 . fourth , this last integral is a slight generalization of @xmath557 , which is dealt with by the well - known watson s lemma ( lemma [ watsonslemma ] below ) . since the first two steps have been done before ( see chapters 3 and 4 of @xcite ) , we will focus mostly on the third and fourth steps . [ watsonslemma ] suppose @xmath558 and @xmath559 is a complex valued , absolutely integrable function on @xmath560 $ ] : @xmath561 also suppose that @xmath562 , where @xmath563 and @xmath501 is continuously differentiable in some neighbourhood of @xmath564 . then for any @xmath565 and @xmath566 $ ] , @xmath567 see proposition 2.1 of @xcite . suppose @xmath558 and @xmath559 is a complex valued , absolutely integrable function on @xmath560 $ ] : @xmath561 also suppose that @xmath568 is twice continuously differentiable in some neighbourhood of @xmath564 . then for any @xmath565 and @xmath569 $ ] , @xmath570 where @xmath571 and @xmath572 see pages 5860 of @xcite . [ 4.8 ] suppose that @xmath452 and @xmath568 are infinitely continuously differentiable in some neighbourhood of @xmath573 . also suppose that @xmath555 is a local maximum of @xmath452 and @xmath574 . then for any @xmath565 and @xmath569 $ ] , @xmath575 where @xmath576 @xmath577 where @xmath578 is an infinitely differentiable function solving @xmath579 before continuing , a few estimates on @xmath578 are needed . [ estimateonv ] let @xmath578 be as in . \(a ) @xmath580 @xmath581 \(b ) set @xmath582 . then @xmath583 for @xmath584 in particular , @xmath585 and @xmath586 . \(c ) let @xmath587 then @xmath588 @xmath589 \(d ) with the same bounds on @xmath590 , @xmath591 \(a ) the proof comes from page 69 of @xcite . it follows immediately from using implicit differentiation of and setting @xmath592 . \(b ) first notice that if @xmath593 with @xmath594 and @xmath595 are the solutions to @xmath596 , then @xmath597 . we will use @xmath598 therefore , we obtain bounds on @xmath578 by solving @xmath596 , which is equivalent to solving @xmath599 in other words @xmath600 . we will use the intermediate value theorem to estimate roots of @xmath601 . for @xmath602 and @xmath603 where @xmath604 is any real number , we have @xmath605 where the @xmath606 are polynomials which satisfy the following inequalities when @xmath607 @xmath608 now setting @xmath609 where @xmath610 , @xmath611 since @xmath612 this implies that @xmath613 for @xmath614 . by an identical argument , for @xmath615 and @xmath603 where @xmath607 , @xmath616 and setting @xmath609 where @xmath617 , @xmath618 for @xmath619 and @xmath620 , @xmath621 therefore @xmath622 for @xmath619 . for @xmath623 and @xmath624 , @xmath625 so @xmath626 . thus the lower bound holds in both cases . the last statement follows because @xmath627 \(c ) differentiating yields @xmath628 or equivalently @xmath629 to estimate @xmath630 , let us first estimate @xmath631 . by a taylor expansion , @xmath632 by the triangle inequality and part ( b ) , @xmath633 which , by and , implies @xmath634 to estimate the inverse of @xmath631 , use @xmath635 which , by setting @xmath636 and @xmath637 , implies that @xmath638 multiplying by @xmath639 finishes the proof of ( c ) . \(d ) differentiating twice yields @xmath640 from part ( c ) and a taylor approximation for @xmath641 , @xmath642 since @xmath643 , @xmath644 by and the estimates on @xmath590 , @xmath645 combining , and , @xmath646 and using @xmath647 on the second term gives the result . [ eh ] suppose that @xmath452 and @xmath568 are infinitely continuously differentiable in some neighbourhood of @xmath573 . also suppose that @xmath555 is a local maximum of @xmath452 and @xmath574 . let @xmath648 and @xmath649 be positive numbers such that @xmath650 and assume @xmath651 equals the right - hand side of . let @xmath652 @xmath653 then for any @xmath565 , @xmath654 use lemma [ 4.8 ] . by part ( a ) of lemma [ estimateonv ] , @xmath655 by parts ( c ) and ( d ) of lemma [ estimateonv ] , @xmath656 when @xmath657 , @xmath658 @xmath659 @xmath660 @xmath661 when @xmath662 , @xmath663 @xmath664 @xmath665 @xmath666 thus @xmath667 , and @xmath668 we can finally wrap up the proof of proposition [ prop6.9 ] . since @xmath349 is not too small ( polynomial in @xmath669 ) , the exponential terms are small enough to be ignored . therefore the error term is @xmath670 . the main term is of order @xmath671 . thus , when multiplying two integrals of the form in corollary [ eh ] , we get @xmath672 as needed . in proposition [ prop6.9 ] , the error term blows up at the edge . therefore a better bound is needed . to get this bound , we simply use the first term in watson s lemma , as opposed to using two terms . since the method of the proof is identical as before and the details are simpler , the proof will be omitted . the exact statement is the following . [ edgeasymptotics ] for @xmath110 , let @xmath111 , @xmath112 denote @xmath113 , @xmath114 denote @xmath115 , and @xmath527 denote @xmath528 . with the assumptions in section [ statement of the main theorem ] , @xmath673 \end{gathered}\ ] ] this section is adapted from @xcite . given a domain @xmath674 , let @xmath675 be the space of smooth , real - valued functions on @xmath676 that are supported on a compact subset of @xmath193 . let @xmath677 be the dirichlet inner product on @xmath675 . let @xmath678 be the completion of @xmath675 under the dirichlet inner product . note that @xmath678 is a hilbert space . a * gaussian free field * is any gaussian hilbert space @xmath679 of random variables denoted by @xmath680 one random variable for each @xmath681 such that @xmath682=(a , b)_{\nabla}.\ ] ] in other words , the map from @xmath678 to @xmath679 defined by sending @xmath125 to the random variable @xmath680 preserves the inner product . let @xmath685 be the eigenfunctions of the dirichlet laplacian on @xmath193 which form an orthonormal basis for @xmath686 , and let @xmath687 denote the corresponding eigenvalues . by the spectral theorem for compact self - adjoint operators , all the @xmath688 are negative . suppose the ordering satisfies @xmath689 . define the operator on @xmath686 : @xmath690 define @xmath691 to be the set of all @xmath692 such that @xmath693 . we may view @xmath694 has a hilbert space with inner product @xmath695 . note that when @xmath696 , the sum @xmath697 may not converge in @xmath686 , but it converges in the space of distributions . note that by integration by parts , @xmath698 . therefore , instead of considering @xmath678 as the hilbert space , we can also consider @xmath699 . then @xmath700=(f , g)_{1/2}=((-\delta)^{-1/2 } f,(-\delta)^{-1/2 } g)=((-\delta)^{-1}f , g).\ ] ] if @xmath703 are mean zero random variables such that @xmath704 is gaussian for all @xmath705 , then @xmath706= \begin{cases } \displaystyle\sum_{\sigma}\prod_{j=1}^{k/2 } \mathbb{e}[x_{\sigma(j)}x_{\sigma(j+1 ) } ] , \ \ k\ \text{even}\\ 0 , \ \ k \ \text{odd } , \end{cases}\ ] ] where the sum is over all involutions on @xmath707 with no fixed point . \99 ) c. boutillier , modles de dimres : comportements limites , ph.d . thesis , universit e de paris - sud , 2005 . a. borodin and p. ferrari , _ anisotropic growth of random surfaces in 2 + 1 dimensions _ ( 2008 ) , http://arxiv.org/abs/0804.3035[arxiv:0804.3035v1 ] a. borodin and j. kuan , _ random surface growth with a wall and plancherel measures for _ @xmath708 , comm . math , volume 63 , issue 7 , pages 831 - 894 , july 2010 . http://arxiv.org/abs/0904.2607[arxiv:0904.2607 ] a. borodin and j. kuan , _ asymptotics of plancherel measures for the infinite - dimensional unitary group _ , adv . in math , * 219 * ( 2008 ) , 894931 . http://arxiv.org/abs/0712.1848[arxiv:0712.1848v1 ] a.l . barabsi and h.e . stanley , _ fractal concepts in surface growth _ , cambridge university press , cambridge , 1995 . m. duits , _ the gaussian free field in an interlacing particle system with two jump rates _ , http://arxiv.org/abs/1105.4656[arxiv:1105.4656v1 ] . r. kenyon , _ height fluctuations in the honeycomb dimer model _ , comm . * 281 * ( 2008 ) , 675709 . m. kardar , g. parisi and y. zhang , _ dynamic scaling of growing interfaces _ , phys . rev . lett , volume 56 , issue 9 ( 1986 ) , 889892 . a. lenard , _ correlation functions and the uniqueness of the state in classical statistical mechanics _ , , volume 30 , number 1 , 3544 . miller , peter . _ applied asymptotic analysis_. american mathematical society : providence , rhode island , 2006 . s. sheffield , _ gaussian free field for mathematicians _ , probab . theory related fields 139 ( 2007 ) , no . 3 - 4 , 521541 . http://arxiv.org/abs/0812.0022[arxiv:0812.0022v1 ] . soshnikov , gaussian fluctuation for the number of particles in airy , bessel , sine and other determinantal random point fields , j. stat . phys . * 100 * ( 2004 ) , 491522 . http://arxiv.org/math-ph/9907012v2[arxiv:math-ph/9907012v2 ] j. warren and p. windridge , _ some examples of dynamics for gelfand tsetlin patterns _ , electronic j. of probability * 14 * ( 2009 ) , no . 59 , 17451769 . http://arxiv.org/abs/0812.0022[arxiv:0812.0022v1 ] . wolf , kinetic roughening of vicinal surfaces , phys . * 67 * ( 1991 ) , 1783-1786 .

movements of molecular motors on cytoskeletal filaments are described by directed walks on a line . detachment from this line is allowed to occur with a small probability . motion in the surrounding fluid is described by symmetric random walks . effects of detachment and reattachment are calculated by an analytical solution of the master equation in two and three dimensions . results are obtained for the fraction of bound motors , their average velocity and displacement . the diffusion coefficient parallel to the filament becomes anomalously large since detachment and subsequent reattachment , in the presence of directed motion of the bound motors , leads to a broadening of the density distribution . the occurrence of protofilaments on a microtubule is modeled by internal states of the binding sites . after a transient time all protofilaments become equally populated .

cond-mat0312422

molecular motors are proteins that convert the free energy released from chemical reactions into directed movements @xcite . here we focus on linear cytoskeletal motors which move along cytoskeletal filaments . many of these motors are processive in the sense that a single motor molecule can move a cargo over a large distance . the most prominent examples are ( conventional ) kinesin and certain types of myosins , which move along microtubules and actin filaments , respectively . in the cell , these motors are involved in transport processes , reorganization of the cytoskeleton and cell division @xcite . however , experiments on the movements of molecular motors can also be done in vitro , which has lead to the development of various single molecule assays . in these experiments , one can measure the velocities , step sizes , walking distances and forces for single motor molecules , see , e.g. @xcite . in additions , they have stimulated a lot of theoretical work devoted to the walks of molecular motors along filaments , see , e.g. @xcite . in the experiments , one observes that even processive motors unbind from their filamentous tracks after a certain walking distance , which is typically of the order of 1 @xmath0 m . for a kinesin molecule , this means that it makes about 100 steps of 8 nm before unbinding @xcite . myosin v motors have a comparable walking distance , but a larger step size of 36 nm , so that they detach after about 3050 steps @xcite . unbound motors then diffuse in the surrounding fluid until they rebind to the same or another filament and continue their directed walk . on larger scales , the motors thus perform complex random walks , which consist of alternating sequences of directed movements along filaments and non - directed diffusion in the surrounding fluid , as shown schematically in fig . [ fig : randwalk ] . these random walks have been discussed by ajdari using scaling arguments @xcite . recently , we have introduced lattice models to describe the random walks of the motors as random walks on a lattice , where certain lines of lattice sites represent the filaments @xcite . when bound to these lines , the motors perform directed random walks . detachment from these lines is allowed to occur with a small , but non - zero probability . diffusive motion in the surrounding fluid is described by symmetric random walks . these models are designed to study generic properties of motor movements , but they can also be used to describe specific motor molecules , since all model parameters can be determined from the measured transport properties , see ref . @xcite . in addition , motor motor interactions can be easily included into these models , e.g. , mutual exclusion of motors from the binding sites of the filaments , which leads to self - organized density profiles in closed systems @xcite and boundary - induced phase transitions in open tube systems @xcite . for the random walks of single motors or , equivalently , for an ensemble of non - interacting motors , we have obtained a number of exact results for the cases of a single filament embedded in two - dimensional or three - dimensional space as reported in @xcite . in particular , these random walks exhibit anomalous drift behavior ; the average position of the motor advances slower than linearly with time . the same drift behavior is found for the movement along a single filament immobilized in open compartments with the same dimensionality , which are more easily accessible to experiments @xcite . in the present article , we present a detailed derivation of the analytical results of ref . @xcite for movements in two and three dimensions without confining boundaries . analytical results are obtained by the following method , which is a variant of the method of fourier laplace transforms for random walks in homogeneous space , see , e.g. , @xcite : by using fourier laplace transforms of the probability distributions , the master equations of the random walk can be transformed into a set of algebraic equations , one of which , however , requires the evaluation of a non - trivial integral . solving these algebraic equations , solutions for the fourier laplace transformed probability distributions and their moments are obtained and closed expressions in terms of integrals can be derived for the time - dependent probability distributions and moments . these can , on the one hand , be evaluated numerically to obtain results for all times ; on the other hand , asymptotic results for small and large times can be obtained fully analytically by using the tauberian theorems , which we summarize in the appendix . in this way , we derive expressions for the fraction of bound motors , the average displacement and dispersion , and the effective velocities and diffusion coefficients . the analytical results are compared to data from monte carlo simulation and are found to be in very good agreement . in addition to the anomalous drift behavior , the random walks of molecular motors also exhibit strongly enhanced diffusion in the direction parallel to the filament . our article is organized as follows : we start with the two - dimensional case in section [ sec:2d ] and discuss the three - dimensional case in section [ sec:3d ] . in both cases , we derive probability distributions and their moments for both the bound and unbound motor . the fact that filaments may consist of several protofilaments is taken into account in the final subsections of sections [ sec:2d ] and [ sec:3d ] , where these are modeled by several internal states of the bound motors . in section [ sec : sticking_prob ] , we extend the discussion to include a variable sticking probability for motors arriving at the filament . at the end , we include a short summary of our results . consider a discrete time random walk on a two dimensional square lattice with lattice sites labeled by integer coordinates @xmath1 . at each step a particle has probability 1/4 to jump into any of the four directions . for modeling the motion of a motor on a microtubule , we choose different behavior on the line with @xmath2 . here the probability to jump from @xmath3 to @xmath4 equals @xmath5 , while jumps to @xmath6 have probability @xmath7 ; the probability to jump to @xmath8 is @xmath9 and the probability to make no jump is @xmath10 , see fig . [ fig : randwalk_2d ] . the latter parameter is needed for modeling realistic situations , where the diffusion coefficient in the fluid is much larger than on the filament @xcite . the ordinary random walk in two dimensions has @xmath11 , @xmath12 , and @xmath13 . we shall assume that the escape probability @xmath14 is small . for @xmath15 the problem amounts to a directed random walk on the line with @xmath2 . the average speed of a motor particle on the filament line is @xmath16 . per step there is a probability @xmath17 to unbind . the probability that the motor is still bound after @xmath18 steps is @xmath19 the master equation for this dynamics reads @xmath20 as initial condition we take an ensemble of particles at @xmath21 , so @xmath22 let us now define the fourier laplace transforms of the probability distribution along the filament @xmath23 and the full distribution @xmath24 , @xmath25 and @xmath26 then the master equations are reduced to an algebraic equation relating @xmath27 and @xmath28 as given by @xmath29-\frac{1}{4}\right)e^{ir}+\left(\frac{\delta}{2}-\frac{1}{4}\right)e^{-ir } + \gamma-\frac{1-\epsilon}{2}\cos q\right]\ , p_{{\rm b}}(r , s ) . \end{aligned}\ ] ] here the first line is what one would get in the case of a symmetric random walk in two dimensions , and the second line corrects those terms that are changed by the presence of the filament . this equation has the obvious solution @xmath30p_{{\rm b}}(r , s ) } { s+1-{\frac{1}{2}\,}\cos q-{\frac{1}{2}\,}\cos r}. \end{aligned}\ ] ] by integrating this result over @xmath31 we also obtain @xmath27 on the left hand side . it thus satisfies a linear equation , that can be easily solved . introducing the variable @xmath0 via @xmath32 we may use the equalities @xmath33 after some computation , we then end up with the probability distribution @xmath34 for the motors bound to the filaments . the probability distribution for all , bound and unbound motors follows via eq . ( [ p2= ] ) . it is easy to check that this is correct for @xmath15 ( random walk in one dimension ) and for @xmath11 , @xmath35 , and @xmath13 ( non - biased random walk in 2d ) . in the following , we extract the transport properties of the motor s random walks from the solution ( [ sold=2 ] ) . the value at @xmath36 gives us the laplace transform @xmath37 of the probability @xmath38 that the motor particle is bound to the filament line with @xmath2 : @xmath39 the inverse is @xmath40 } = \int_0^{{\,\epsilon\,}^{-2}}\frac{{{\rm d}}y}{\pi}\frac{(1-{\,\epsilon\,}^2 y)^{t+1/2}}{\sqrt{y}\ , [ 1+(1 - 2{\,\epsilon\,})y]}.\ ] ] this expression is exact . it holds for all @xmath18 and for all @xmath14 , @xmath41 and can be evaluated numerically . values for @xmath42 obtained in this way are shown fig . [ fig : pb_d2_eps.03.05.08 ] ( lines ) for three different values of the detachment rate @xmath14 . comparison to results of monte carlo simulations ( data points ) shows that the agreement is very good . let us now derive the asymptotic behavior for small and large times , respectively . if @xmath43 is small and @xmath44 , we have @xmath45 then the inverse behaves as @xmath46 the second integral is an obvious limit of the last integral in eq . ( [ nnult= ] ) . for deriving the short - time behavior we may start from the series of equalities as given by @xmath47 the integrals over the exponential terms are most easily carried out using dimensional regularization . to show how that works , let us consider the last term . we need to consider the expression @xmath48 in the limit of small @xmath49 , where we used that , in dimensional regularization , integrals of powers are put equal to zero . the limit @xmath50 can now be taken . using the same procedure for the other terms , we obtain @xmath51 for small @xmath18 , which represents a series in powers of @xmath52 . for @xmath53 , this is somewhat surprising : although the motors detach at times @xmath54 , the recurrent behavior of the random walk brings them mostly back to the filament . for @xmath55 , this just says that the motor did not have enough time to escape from the line . the half - integer powers are related to the long range diffusion away from the filament . note that , for the short time limit , we have assumed that @xmath56 , i.e. , that @xmath18 is large compared to the time required for one step of the random walk . in that interval the approximation ( [ n0s2 ] ) holds with the term @xmath57 being due to diffusion . for smaller times , the random walk exhibits discrete steps . our short time result would hold for arbitrarily small times in the limit in which the walk on the line becomes a continuous time random walk . a continuous time random walk with exponential waiting time distribution is obtained approximately , if @xmath58 is small , which is the case for realistic applications of our model to molecular motor setups @xcite . the same remark holds for all the short time results discussed in the following . another way to derive ( [ n0t<= ] ) is to expand ( [ n0s= ] ) in powers of @xmath59 and to use the tauberian theorem which states that the inverse laplace transform of @xmath60 is given by @xmath61 see appendix [ sec : app : taubthms ] and , e.g. , ref . this theorem holds both for positive and negative values of @xmath62 . it also shows that positive integer powers of @xmath63 of @xmath37 in the limit of small @xmath63 do not contribute to long time tails . for large times , @xmath64 , the expression ( [ n0t= ] ) for @xmath42 can be evaluated using the expansion @xmath65 for small @xmath66 which leads to @xmath67 alternatively , one can expand ( [ n0s= ] ) in powers of @xmath68 and use again the tauberian theorem . it follows from ( [ n0t>= ] ) that for large @xmath18 the probability to be bound to the filament decays as @xmath69 in agreement with scaling arguments @xcite . the expression ( [ sold=2 ] ) for the fourier laplace transformed probability distribution @xmath27 contains much more information . at linear order in @xmath70 we get the laplace transform of the average position of motor particles along the filament line , @xmath71 we obtain @xmath72 ^ 2}.\ ] ] in the limit of small @xmath14 and large @xmath18 , this implies @xmath73 ^ 2}.\ ] ] we invert the laplace transform by taking the @xmath74 term apart . we then get for the average position the asymptotic behavior @xmath75 for large @xmath18 . the exact expression is @xmath76 } { \sqrt{y}\ , [ 1+y(1 - 2{\,\epsilon\,})]^2 } , \ ] ] which deviates from ( [ n1t = i ] ) for times of order unity . the full expression ( [ n1t = i_exact ] ) is evaluated numerically and plotted in fig . [ fig : displ_d2_b_ub_all ] . the same figure contains data points as obtained from monte carlo simulations which confirm the analytical result ( [ n1t = i_exact ] ) . for short times we proceed as above . we expand @xmath77{\nonumber \\ } & = & { v_{\rm b}}t\,(1-\frac{8}{3}\,\frac{\epsilon \sqrt t}{\sqrt\pi } ) \end{aligned}\ ] ] this leads to @xmath78 . thus , the average position @xmath79 and speed @xmath80 of the motors bound to the filament are given by @xmath81 where @xmath82 is the average speed if the particles did not leave the line . for large times , the asymptotic expression ( [ n1t = i ] ) leads to @xmath83 therefore , for large @xmath18 , the average speed of bound motors behaves as @xmath84 the last relation confirms the scaling @xmath85 , which has been used in the scaling approach @xcite . the effective motor velocity is reduced by a factor @xmath86 , i.e. , by the probability that a motor is in the bound state . the relation @xmath87 also applies to a simple two - state random walk , where motion is directed in one of the states only . in contrast to the simple two - state random walk , however , the probability @xmath88 is time - dependent here . the factor @xmath89 in ( [ venhd=2 ] ) is solely due to the fact that only the bound motors are considered . we will show below , that this factor is absent , if all , bound and unbound , motors are considered . by expanding the expression ( [ sold=2 ] ) for the fourier laplace transformed probability @xmath27 up to second order in @xmath70 , the second moment of the distribution of bound motors is found to behave as @xmath90 the laplace transform is again inverted by complex integration . in the contour integral for @xmath91 , we may replace @xmath92 by @xmath93 , since the subtracted integral vanishes , as can be seen by closing the contour of integration in the right half plane . closing the contour along the negative real axis , we obtain @xmath94\sqrt{1-{\,\epsilon\,}^2x}}{[1+(1 - 2{\,\epsilon\ , } ) x]^3\sqrt{x } } - \frac{2{v_{\rm b}}^2}{\pi\epsilon^4}\int_0^{{\,\epsilon\,}^{-2}}{{\rm d}}x \ , \frac{[(1-x\epsilon^2)^t-1](1-{\,\epsilon\,}^2x)^{3/2}}{[1+(1 - 2{\,\epsilon\ , } ) x]^3x^{3/2 } } \nonumber\\ & & { } + \frac{2(1-\gamma-{\,\epsilon\,})}{\pi{\,\epsilon\,}^2}\int_0^{{\,\epsilon\,}^{-2}}{{\rm d}}x\ , \frac{[1-(1-x\epsilon^2)^t]\sqrt{1-{\,\epsilon\,}^2x}}{[1+(1 - 2{\,\epsilon\ , } ) x]^2 \sqrt{x } } \nonumber\\ & & { } - \frac{1}{4\pi{\,\epsilon\,}^2}\int_0^{{\,\epsilon\,}^{-2}}{{\rm d}}x\,\frac{(1 - 2{\,\epsilon\,}^2x ) [ ( 1-x\epsilon^2)^t-1+{\,\epsilon\,}^2xt][1-(1 - 2{\,\epsilon\,})x]}{[1+(1 - 2{\,\epsilon\ , } ) x]^2 x^{3/2 } \sqrt{1-{\,\epsilon\,}^2x } } \nonumber\\ & \approx & \frac{2{v_{\rm b}}^2}{\pi\epsilon^4 } \int_0^\infty{{\rm d}}x \ , \frac{(1-e^{-x\epsilon^2 t})(1 - 3x)}{(1+x)^3x^{3/2 } } + \frac{2(1-\gamma)}{{\,\epsilon\,}^2}\int_0^\infty\frac{{{\rm d}}x}{\pi}\ , \frac{1-e^{-x{\,\epsilon\,}^2 t } } { ( 1+x)^2\sqrt{x}}\nonumber\\ & & { } + \frac{1}{4{\,\epsilon\,}^2}\int_0^\infty\frac{{{\rm d}}x}{\pi}\ , \frac{(1-e^{-x{\,\epsilon\,}^2 t})(1-x ) } { ( 1+x)^2x\sqrt{x}}. \end{aligned}\ ] ] we have evaluated this exact expression numerically and compared it with simulation data . as shown in fig . [ fig : deltanb2_d23 ] , the agreement is again very good . for short times we proceed as above . @xmath95{\nonumber \\}&=&{v_{\rm b}}^2t^2 ( 1-\frac{16}{5}\,\frac{\epsilon \sqrt t}{\sqrt\pi})+ ( 1-\gamma)t+\frac{8\gamma-7}{3}\,\frac{\epsilon t\sqrt t}{\sqrt\pi}. \end{aligned}\ ] ] this implies the normalized second moment @xmath96 and the dispersion @xmath97 this result holds again in the limit of continuous time , i.e. for @xmath10 close to one . the leading term taking into account the time discretization is @xmath98 , i.e. the dispersion at short times , which arises from the walks along the filament , is smaller in discrete than in continuous time . for the large time result , which we will derive next , the choice of continuous or discrete time makes no difference . the relative dispersion @xmath99 is small since @xmath53 . using the dispersion @xmath100 , we may also calculate the time - dependent diffusion coefficient @xmath101 in the scaling regime @xmath102 , @xmath103 is much larger than its limiting value @xmath104 for @xmath15 . this enhanced diffusion arises from the fact that each motor may detach from the microtubule with probability @xmath105 for any value of @xmath106 , according to eq . ( [ punb= ] ) . this leads to a considerable broadening of the bound motor distribution . for large @xmath18 one can make a change of variables in the integral expression ( [ 2ndmomentbound_lf ] ) and use @xmath107 as the new integration variable . this leads to the asymptotic behavior @xmath108 and @xmath109 the diffusion coefficient behaves as @xmath110 for large @xmath18 . the limiting value of the diffusion coefficient , @xmath111 is large compared to the diffusion coefficient of the one - dimensional random walk along the filament , @xmath112 . this broadening of the distribution occurs since the unbound motors lag behind the bound ones , which implies that also the rebinding motors lag behind those that have been bound for some time . at large times @xmath113 scales as @xmath114 and one may look for a scaling form of the density . for small @xmath63 and small @xmath70 , the fourier laplace transformed probability distribution @xmath27 as in ( [ sold=2 ] behaves as @xmath115 the inverse fourier transform follows from the continuum limit of eq . ( [ p0rz ] ) . it then follows that , in this limit , @xmath116 for @xmath117 , while for positive @xmath49 , one finds @xmath118 the inverse laplace transform of this expression now leads to @xmath119 for @xmath120 which implies that the overall motion is slower than ballistic . for @xmath121 , we obtain @xmath122 after a change of variables from @xmath63 to @xmath123 , the @xmath124-integral can be taken over the whole real axis ( at the expense of a factor @xmath125 ) . evaluation of this gaussian integral leads to @xmath126 for @xmath127 . this expression vanishes linearly as @xmath128 for small @xmath49 and even exponentially fast as @xmath49 approaches @xmath129 from below . the density profile @xmath130 as given by ( [ pn0t= ] ) is plotted in fig . [ fig : bound_profile_d2 ] for several values of the time @xmath18 . comparison with the results of mc simulations shows that the asymptotic expression as given by ( [ pn0t= ] ) is very good for times which excess about 8000 time steps . at smaller times , the asymptotic expression overestimates the maximum of @xmath131 and underestimates the tails of @xmath131 for large @xmath49 , see fig . [ fig : bound_profile_d2 ] . simulation data are obtained by averaging over @xmath132 realizations of the random walk . it is somewhat tedious to show that the moments @xmath88 , @xmath133 and @xmath134 as obtained from ( [ pn0t= ] ) agree with the previously derived expressions . to verify this , one may , in an intermediate step , use the substitution @xmath135 which implies @xmath136 eventually every motor will unbind and diffuse in the surrounding fluid . we now discuss the effects of the filament on the behavior of the unbound motors . the fourier laplace transform @xmath28 of the probability distribution @xmath24 of bound and unbound motors as given by eq . ( [ p2= ] ) can be rewritten as @xmath137 the first part is @xmath27 as given by eq . ( [ p0app ] ) , describing the bound motors . the second part describes the corresponding probability distribution @xmath138p_{{\rm b}}(r , s ) . \end{aligned}\ ] ] of the unbound motors . for small @xmath63 , small @xmath70 , and small @xmath31 , taking into account that @xmath139 and @xmath140 , the expression ( [ p2decompose ] ) leads to @xmath141 expanding ( [ p_ub_2d= ] ) or , for large times , ( [ p2ass= ] ) in powers of @xmath70 and @xmath31 yields the moments of the distribution of unbound motors . the total fraction of unbound motors as obtained from ( [ p2decompose ] ) for @xmath142 is , of course , @xmath143 expressing motor conservation . the average longitudinal position of the unbound motors is given by @xmath144 as follows from ( [ p_ub_2d= ] ) and ( [ p2ass= ] ) . the last equality holds for small @xmath43 and small @xmath63 , i.e. for large @xmath18 . for small @xmath18 , one has from the exact expression in ( [ n_1ub_s_d=2 ] ) @xmath145 which leads to @xmath146 for the average position and speed of the unbound motors . for large @xmath18 , the asymptotic expression given in ( [ n_1ub_s_d=2 ] ) leads to @xmath147 which implies @xmath148 whereas each individual motor has zero average velocity in the fluid , the statistical velocity @xmath149 is non - zero , since it is driven by unbinding from the cloud of motors moving on the filament : the cloud of unbound motors advances , because motors rebind to the filament and others detach from it , and those detaching have propagated a certain distance compared to those rebinding . since for large times all motors are detached from the filament most of the time , equation ( [ large_t_displ ] ) gives the asymptotic displacement of a motor , if averages are taken over all bound and unbound motors , i.e. , @xmath150 at large times as follows from ( [ n0t>= ] ) , ( [ venhd=2 ] ) , ( [ large_t_displ ] ) , and @xmath151 . the effective ( time - dependent ) velocity at large times is @xmath152 , as predicted by scaling arguments @xcite . this can also be obtained by inspection of eq . ( [ n_1ub_s_d=2 ] ) : the displacement of all motors is given by @xmath153 the normalization factor for all motors is unity . the effective velocity is obtained from the inverse : @xmath154 fig . [ fig : displ_d2_b_ub_all ] shows the displacement of the bound and unbound motors , obtained by numerical evaluation of the exact integrals ( dotted lines ) and corresponding simulation data ( circles for the bound motors and diamonds for the unbound ones ) . the displacement averaged over all motors , @xmath155 , is also shown in fig . [ fig : displ_d2_b_ub_all ] ( solid line and squares ) . for small times it is equal to the displacement of the bound motors , for intermediate times it interpolates to the curve for the unbound motors . since almost all motors are detached for large times , the displacement of all motors is then equal to the displacement of the unbound ones . the longitudinal position has the second moment @xmath156 at short times , this means @xmath157 and thus @xmath158 the last relation implies that , in the longitudinal direction , the released motors spread in a broad cloud , not narrowly centered around its average . at long times , one obtains @xmath159 this implies the dispersion and the diffusion coefficient as given by @xmath160 the order of magnitude of the diffusion coefficient is again @xmath161 . the prefactor @xmath162 is slightly smaller than the prefactor @xmath163 of eq . ( [ dparline ] ) . the order of magnitude @xmath164 tells us that , like on the line , longitudinal diffusion is strongly enhanced by the unbinding from and binding back to the line . the diffusion behavior perpendicular to the filament is determined by ( [ p2ass= ] ) up to quadratic order in @xmath31 . the average transverse position vanishes and the dispersion of the transverse position is given by @xmath165 and @xmath166 this implies @xmath167 at short times . it is small because the motors started in a state bound to the line . at large times one has @xmath168 the transverse diffusion coefficient @xmath169\ ] ] approaches the free value @xmath170 . so the transverse diffusion starts out very small , and finally reaches its value in free space . the quantity @xmath171 is plotted in fig . [ fig : m^2_d2 ] . there are small deviations in comparison to the simulations at small times arising from the time discretization . finally , we derive the density profile of the unbound motors . after inversion of the fourier laplace transforms , the profile ( [ p2ass= ] ) becomes in real space @xmath172 comparing with ( [ pn0s ] ) we have a shifted value for @xmath173 and an extra factor @xmath43 . the temporal form thus becomes @xmath174 for @xmath175 and @xmath127 . this again vanishes for @xmath176 . for @xmath50 it is finite when @xmath177 is finite , while it is zero for negative @xmath49 , in the present approximation . though detachment followed by diffusion to negative @xmath49 may occur , the attainable values are of order unity , and vanish to leading order in @xmath178 . microtubules , the filament tracks of kinesin and dynein motors , consist of 13 parallel protofilaments @xcite , each of which provides a possible track for these motors . to incorporate this in our model , we assume that the pinning line at @xmath2 has @xmath179 internal states where @xmath179 may be equal to @xmath180 . the average occupation of each of these states is denoted by @xmath181 with @xmath182 . there is a small probability , @xmath183 , that a motor goes from track @xmath184 to @xmath185 within one time step , and similarly for going to track @xmath186 . to take into account the cylindrical structure of the microtubule , we identify @xmath187 with @xmath188 . we assume that after detaching , the motor may randomly go to the right or to the left of the tubule ; likewise , when attaching to the tubule either from right or left , the motor has equal probability to attach to any of the tracks . in this model the total fraction of motors at position @xmath49 along the tubule is @xmath189 the dynamics given by eqs . ( [ me1 ] ) , ( [ me2 ] ) , ( [ me3 ] ) , and ( [ me4 ] ) remains valid and leads to the same solution , as it is insensitive to the internal distribution over the tracks . the motion on the individual tracks is described by a master equation analogous to eq . ( [ me2 ] ) and given by @xmath190 when summed over @xmath184 this indeed leads back to eq . ( [ me2 ] ) . to solve eq . ( [ me2j ] ) , another fourier transform is needed which is defined by @xmath191 with @xmath192 and @xmath193 . we assume that , at @xmath194 , all motors are located on the track @xmath188 which corresponds to the initial condition @xmath195 going to the fourier laplace transform , we find @xmath196 the surviving fractions on the individual tracks are found from the expressions for @xmath36 . in the time representation , the surviving fractions behave as @xmath197 the first term describes the symmetric distribution of the motors over the @xmath179 tracks ; in eq . ( [ n0t= ] ) we showed that it decays algebraically . the second term represents the asymmetry arising from the initial condition that only one track is populated . there are two decay mechanisms of this asymmetry . the term @xmath198 expresses that detachment and subsequent reattachment to randomly chosen tracks restores the symmetry . this happens in particular for motors that quickly return to a randomly chosen track , implying ordinary exponential decay . the factor @xmath199 expresses that hopping to neighboring tracks also restores the symmetry . for large @xmath179 , the smallest of these rates is @xmath200 . so the exchange process dominates when this value is larger than @xmath201 . therefore , the asymmetry between the average occupation of the various tracks disappears after the discussed transient time . now , let us consider the same kind of random walk on a three - dimensional cubic lattice , in which the line @xmath202 represents the filament which attracts and binds the motors . away from the filament the jump probabilities are equal to @xmath203 , while on the filament they are given by @xmath204 and @xmath205 in the forward and backward direction , respectively , equal to @xmath206 for each of the four sideward directions , and equal to @xmath10 to make no step , see fig . [ fig : randwalk_3d ] . the average short - time velocity is @xmath207 while the sticking probability is @xmath208 $ ] . we denote the transverse coordinate as @xmath209 . the master equation for this dynamics has the form @xmath210 in these equations , @xmath211 and @xmath212 denote the four transverse nearest neighbor vectors which connect a filament site to the four adjacent non - filament sites , @xmath213 . the summations over @xmath214 in ( [ me31 ] ) and ( [ me32 ] ) or @xmath215 run over the four possible values . ( [ me33 ] ) holds for any of the four values of @xmath214 with the sum over @xmath215 running over the other three vectors . we can follow the same steps as in the two - dimensional case using again fourier laplace transforms . the fourier transformation in the perpendicular directions leads to a transverse fourier vector @xmath216 . the equivalent of eq . ( [ p2= ] ) becomes @xmath217\ , p_{\rm b}(r , s)}{3 + 3s-\cos r-\cos q_1-\cos q_2}.\ ] ] by doing the integrals over @xmath218 and @xmath219 , we derive the expression for @xmath220 . the necessary integral is @xmath221 where @xmath222 is the complete elliptic integral of the first kind . we also use the relation @xmath223 where @xmath224 is given by ( [ integral_d3 ] ) . we then get @xmath225 \,i(r , s)}.\ ] ] for large @xmath63 , one may use @xmath226 , to verify that @xmath227 , which is required by our initial condition that all motors started at @xmath228 . for small @xmath14 and small @xmath70 and @xmath63 , we may approximate ( [ pnulb_d3 ] ) as @xmath229 which is analogous to ( [ p0app ] ) . this implies in real space that @xmath230 for @xmath127 . decomposing @xmath231 , the inverse laplace transform leads to @xmath232 for large times , the variable @xmath63 will be small . since @xmath233 $ ] for @xmath234 @xcite , we may conclude that @xmath235 diverges as @xmath236\qquad { \rm with}\qquad { { \tau_0}}=\frac{3}{16}\ ] ] for small @xmath63 and @xmath70 . this implies in particular that @xmath237 ^ 2+\pi^2},{\nonumber \\}{\rm and}\qquad j_2(s ) & \approx & \frac{2\pi\epsilon}{3}\ , \frac{\pi}{[\ln{{\tau_0}}s]^2+\pi^2}. \label{j12ass}\end{aligned}\ ] ] the laplace transformed survival fraction @xmath37 is equal to @xmath238 . it then follows from ( [ p03d= ] ) that @xmath239 ^ 2+j^2_2(s)}.\ ] ] for @xmath240 , using ( [ j12ass ] ) and neglecting @xmath63 with respect to @xmath241 , we get for the number of particles on the line the simple result @xmath242 which confirms the scaling @xmath243 predicted by the scaling approach @xcite . we can also derive the latter result from @xmath244\ln({{\tau_0}}s)$ ] using the tauberian transforms summarized in appendix [ sec : app : taubthms ] . the survival fraction is shown in fig . [ fig : pb_d3_eps.03.05 ] for two values of @xmath14 . again the exact integral ( [ n0_d=3_exact ] ) is evaluated numerically ( lines ) and compared to simulation data ( data points ) . the agreement is good . computation of the first moment @xmath245 with @xmath246 as given by ( [ p03d= ] ) leads to @xmath247j_2(s ) } { \left([j_1(s)-s]^2+j_2 ^ 2(s)\right)^2}.\ ] ] for small times , this behaves again as @xmath248 corresponding to the velocity @xmath207 . for large times , we find @xmath249 to leading order in @xmath250 where @xmath251 is euler s constant . this implies the average position and velocity @xmath252 the position of bound motors as a function of time is shown in fig . [ fig : n0_d3_eps.03.05 ] . the agreement between the analytical result and the simulations is again quite good . for the second moment , we obtain @xmath253 ^ 3}+\frac{1-\gamma-{\,\epsilon\,}}{{v_{\rm b}}}n_1(s)+\frac{\frac{1}{2\pi{\,\epsilon\,}s } j^2(s)}{[s+j(s)]^2},\ ] ] where we have taken into account the quadratic correction term to @xmath254 , and @xmath255}{[(-s+j_1(s))^2+j_2 ^ 2(s)]^3 } + \frac{2(1-\gamma-{\,\epsilon\,})}{\pi}\int_0^\infty{{\rm d}}s\,\frac{e^{-st}[(j_1(s)-s)j_2(s)]}{[(-s+j_1(s))^2+j_2 ^ 2(s)]^2}\nonumber\\ & & { } - \frac{1}{\pi^2{\,\epsilon\,}}\int_0^\infty{{\rm d}}s\,\frac{e^{-st}[j_2 ^ 3(s)-sj_1(s)j_2(s)+j_1 ^ 2(s)j_2(s)]}{[(-s+j_1(s))^2+j_2 ^ 2(s)]^2 } + \frac{1}{2\pi{\,\epsilon\ , } } . \end{aligned}\ ] ] the last term emerges from the singularity at @xmath256 in the third term of @xmath257 and represents diffusion in the unbound state . for large times , this expression leads to the asymptotic relations @xmath258+\frac{1}{2\pi{\,\epsilon\ , } } , \\ \overline{n^2}(t ) & \approx & 2{v_{\rm b}}^2\left(\frac{3}{2\pi{\,\epsilon\,}}\right)^2 \,[3 ( \ln \frac{t}{{{\tau_0}}}+\gamma_e)^2-\frac{\pi^2}{3}]+\frac{1}{3}t,\end{aligned}\ ] ] and @xmath259+\frac{1}{3}t . \end{aligned}\ ] ] with @xmath260 as in ( [ int+def_taun ] ) . the longitudinal diffusion coefficient is @xmath261.\ ] ] notice that , at typical times @xmath262 , this is still of order @xmath263 , and , thus , much larger than the value without the unbinding mechanism . in contrast to the two - dimensional case discussed above , the leading term in three dimensions is given by the usual diffusion in the unbound state , but there are large logarithmic corrections of order @xmath264 . this can be seen in fig . [ fig : deltanb2_d23 ] , where @xmath265 is shown for both cases . as in the two - dimensional case , the spatial distribution of the motors bound to the line can be derived in a somewhat explicit form . for @xmath127 one has @xmath266}-st.\ ] ] for large @xmath49 and @xmath18 , we may use the saddle point approximation . the condition @xmath267 yields @xmath268 for very small @xmath43 and @xmath269 this means that @xmath63 is indeed small , @xmath270 the saddle point values are @xmath271 the second derivative has a negative sign , which allows us to choose the contour from @xmath272 to @xmath273 , where @xmath274 is the saddle point value given in ( [ ssp= ] ) . the integration over gaussian fluctuations yields @xmath275 as a function of @xmath49 , this curve starts at @xmath276 , has a maximum and goes to zero at @xmath277 . the apparent divergence near @xmath277 of the last expression is an artifact of the saddle point approximation . we can check the normalization : @xmath278 it differs from the exact value by a factor of order @xmath279 , which is anyhow only logarithmic . this occurs because the small @xmath49-behavior has not been treated properly by the saddle point approximation . we have compared the profiles of the bound motors obtained from the saddle point approximation with simulation data , evaluating eq . ( [ profile_3d ] ) with @xmath63 taken from the numerical solution of eq . ( [ saddle_point_cond ] ) and normalizing the saddle point profile . as expected , agreement with the simulation data is only good for large @xmath49 . therefore we have also taken the inverse laplace transform of ( [ profile3d_bound ] ) numerically . the result is shown in fig . [ fig : bound_profile_d3 ] for @xmath280 and @xmath281 in comparison with simulation results . to obtain the simulated profiles , simulations were performed again with @xmath132 motors particles , most of which are , however , detached from the filament at the times the density profile was measured . while agreement is good for large @xmath49 , there are deviations in the region around the maximum , which are probably due to the approximation used in eq . ( [ p03d= ] ) . also for the motors that detached from the pinning line , we can follow the steps of previous section . at large times , the transport properties obtained for the unbound motors dominate also the results that are obtained if averages are taken over all , bound and unbound , motors , because at large times the motors spend most of the time detached from the filament from eq . ( [ p3= ] ) , we now find the fourier laplace transform of the distribution of unbound motors to be given by @xmath282}\ ] ] for small @xmath70 , @xmath283 , and @xmath63 . the term linear in @xmath70 implies @xmath284 ^ 2}{\nonumber \\}{\rm and}\qquad \overline{n}_{{\rm ub } } & \approx & \frac{3{v_{\rm b}}}{2\pi{\,\epsilon\,}}(\ln\frac{t}{{{\tau_0}}}+\gamma_e ) . \end{aligned}\ ] ] the last result is just half of the value for the bound particles ( [ nb3d= ] ) . as in the two - dimensional case , the latter results gives also the average position of all , bound and unbound , motors at large times . the logarithmic growth of the average position and the corresponding time - dependent velocity confirm the predictions of scaling arguments , which give @xmath285 @xcite . the time - dependent velocity is again given by @xmath286 at large times . for the second moment we get @xmath287 for small @xmath63 or large @xmath18 , this expression leads to the asymptotic relations @xmath288 ^ 2},{\nonumber \\}\overline { n^2}_{{\rm ub } } & \approx & \frac{t}{3}+\frac{9{v_{\rm b}}^2}{2\pi^2{\,\epsilon\,}^2 } [ ( \ln\frac{t}{{{\tau_0}}}+\gamma_e)^2 - 1 ] , { \nonumber \\}\delta n^2_{{\rm ub } } & \approx & \frac{t}{3}+\frac{9{v_{\rm b}}^2}{4\pi^2{\,\epsilon\,}^2 } [ ( \ln\frac{t}{{{\tau_0}}}+\gamma_e)^2 - 2],\qquad { \rm and}{\nonumber \\}d_\parallel & \approx & \frac{1}{6}+\frac{9{v_{\rm b}}^2}{4\pi^2{\,\epsilon\,}^2\,t } [ \ln\frac{t}{{{\tau_0}}}+\gamma_e ] , \end{aligned}\ ] ] where the logarithmic correction to the free diffusion coefficient is half of the correction to the bound diffusion coefficient ( [ largetimediff_d=3 ] ) . that such leading singularities have different numerical prefactors in different quantities could have been anticipated , since the cloud of random walkers is smeared , with spread as large as the average . in @xmath289 the effect was stronger , since already the leading terms had different numerical prefactors , cf . ( [ dparline ] ) for @xmath290 with eq . ( [ dparall ] ) for @xmath291 . for transversal transport the situation is much simpler @xmath292}.\ ] ] for large @xmath18 one has small @xmath63 , so @xmath293 . this implies @xmath294 and the transverse diffusion coefficient is given by @xmath295.\ ] ] the limiting values are just those without the pinning line . they are reached for @xmath296 . finally , we derive an expression for the probability distribution ( or , equivalently , the density profile in the case of many non - interacting motors ) away from the filament using again the saddle point approximation . from ( [ pubd=3 ] ) we deduce @xmath297 for @xmath298 , where @xmath299 is a modified bessel function . this leads to the profile @xmath300/{v_{\rm b}}}.\ ] ] for closing the contour around the negative real axis we need @xmath301.\ ] ] we get a result of the type @xmath302/{v_{\rm b } } } } \nonumber\\ & & \times [ \cos \frac{nj_2(s)}{{v_{\rm b}}}\,j_0(m\sqrt{6s})- \sin \frac{nj_2(s)}{{v_{\rm b}}}\,y_0(m\sqrt{6s})].\end{aligned}\ ] ] the saddle point approximation can be done for large enough @xmath177 , where @xmath303 we then get for the distribution away from the filament ( @xmath304 ) @xmath305 where the saddle point values are @xmath306 { \nonumber \\}&=&-\frac{2\pi{\,\epsilon\,}n}{3{v_{\rm b}}\ln{{\tau_0}}s}(1+\frac{1}{\ln{{\tau_0}}s } ) { \nonumber \\ } & & { } + \frac{2\pi{\,\epsilon\,}n}{3\ln^2{{\tau_0}}s}\ , \frac{m{v_{\rm b}}\sqrt{3/(2s)}}{{v_{\rm b}}t - n-{v_{\rm b}}m\sqrt{3/(2s ) } } \ ] ] and @xmath307 with the saddle point value of @xmath63 as given by @xmath308}.\ ] ] as before we can assume that the line has @xmath179 internal states , which for @xmath309 model the protofilaments on the microtubule . the motion on the individual tracks is described by a master equation analogous to eq . ( [ me2 ] ) , @xmath310 this can be analyzed as in the @xmath289 case . one finds for @xmath311 @xmath312 the surviving fractions on the individual protofilaments are found by inserting @xmath36 . in the time representation it reads @xmath313 the asymmetry decays as @xmath314 $ ] . as in the two - dimensional situation , the asymmetry has no influence on the total occupation @xmath315 of site @xmath49 along the microtubule , that was the subject of study in previous subsections . we now want to incorporate the possibility that a motor need not bind to the line when it collides with it . we consider two approaches . let us now consider a tubule located above the line @xmath2 of a two - dimensional plane . we consider the following jump rates : detaching : line@xmath316plane @xmath17 ; attaching plane@xmath316 line @xmath317 ; reduced jumps in the plane away from position below the line : @xmath318 . the probability to jump from the tubule to the line with @xmath2 is equal to @xmath319 and the probability to jump from the line with @xmath2 to the tubule is given by @xmath317 . in addition , the motor particle jumps with probability @xmath318 from the line with @xmath2 to a neighboring line with @xmath320 . so for @xmath321 no reattachment occurs . for @xmath322 the other extreme occurs : one can not jump from the line @xmath2 to the rest of the plane ; if initially all walkers were on the tubule , they will go no further than below it , but not wander in the plane . thus we consider the master equations @xmath323 as initial condition we choose all motors on the location @xmath324 of the tubule , @xmath325 the fourier laplace transforms yield @xmath326p(q , r ) } \nonumber\\ & & = { \frac{1}{2}\,}{\,\epsilon\,}p(r)- { \frac{1}{2}\,}\eta\cos q p_0(r ) \end{aligned}\ ] ] @xmath327p(r ) & \nonumber\\ & = 1+{\frac{1}{2}\,}\eta p_0(r ) . & \end{aligned}\ ] ] we can integrate @xmath328 over @xmath31 @xmath329 . \end{aligned}\ ] ] with @xmath330 it follows that @xmath331 eliminating @xmath332 now yields @xmath333-i{v_{\rm b}}\sin r\big)^{-1 } \label{soletad=2}. \end{aligned}\ ] ] for @xmath334 the @xmath335-dependent term becomes @xmath336 , so then the previous situation is recovered . for @xmath322 one can check that @xmath337 , showing that no motors reach the fluid ( total sticking ) . for small parameters one has @xmath338 so the only effect is the effective detaching probability @xmath339 for the probability to be in the fluid no subtraction as in ( [ p2ass= ] ) is needed . we have immediately @xmath340 we can now look for the enhancement of the speed on the tubule ; it was factor @xmath89 in eq . ( [ venhd=2 ] ) . let us assume that @xmath335 is small , so that there is a large time domain , where we may neglect it . let us thus set @xmath341 then we get @xmath342 there are two domains : \1 ) @xmath343 , @xmath344 . here @xmath345 , @xmath346 . thus @xmath347 and @xmath348 . this is still a sharp profile . although particles have detached , the remaining ones go firmly with the bare speed @xmath349 . \2 ) @xmath350 , @xmath351 . this we discussed already in ( [ venhd=2 ] ) . the relation @xmath352 just says that @xmath353 . let us now include a variable sticking probability in our model : if a motor reaches the filament , it rebinds to it with a probability @xmath354 , while it is reflected from the filament with probability @xmath355 . such a behavior can be due to steric constraints , if e.g. a motor with attached bead diffuses close to the filament , but with the bead between the motor and the filament . in the long time regime the introduction of this additional parameter is expected to reduce to the probability that a motor is bound to the filament and thus the effective time - dependent velocity by a factor @xmath354 . this has been confirmed by simulations for the case of random walks in open compartments @xcite . in this section we show analytically that this is indeed the case . let us begin with the simpler case @xmath289 . to include the sticking probability , the master equations for @xmath356 have to be modified : on the lines @xmath320 the rate for hopping to @xmath2 , i.e. to the filament is @xmath357 , while there is a rate @xmath358 not to jump , equivalently a motor on these lines attempts to hop to the filament with the usual rate @xmath359 , but the jump is rejected with probability @xmath355 . the modified master equations are @xmath360 @xmath361 the equivalent of eq . ( [ p1a= ] ) now contains also a term with @xmath362 , the fourier laplace transform of the probability distribution @xmath363 along the lines with @xmath320 adjacent to the filament line : @xmath364\ , p_{{\rm b}}(r , s ) . \end{aligned}\ ] ] @xmath27 and @xmath362 are related via the fourier laplace transform of eq . ( [ master_eq_d2_piad_m0 ] ) : @xmath365p_{{\rm b}}(r , s)-1\right).\ ] ] using this expression for @xmath362 , we can proceed in the same way as above and obtain @xmath366[1+\frac{1-\pi_{\rm ad}}{\pi_{\rm ad}}(1-e^{-\mu})]-\frac{\epsilon}{2}e^{-\mu}}.\ ] ] for small @xmath70 and @xmath63 , this leads to the asymptotic relation @xmath367 which has exactly the form of eq . ( [ p0app ] ) , but with an effective detachment rate @xmath368 . doing the analogous calculations for the unbound motors we find @xmath369 which corresponds to eq . ( [ p2ass= ] ) again with the effective detachment rate @xmath368 . hence in the long time regime the only effect of the sticking probability @xmath354 is a rescaling of the detachment rate . thus the probability for a motor to be bound to the filament for large times decays as @xmath370 and the average displacement grows as @xmath371 , i.e. both quantities are reduced by a factor @xmath354 as expected from the scaling approach @xcite . for @xmath372 the calculation is completely analogous . the sticking probability is introduced in the same way as above : a motor at a neighboring site of the filament attempts to jump to the filament with rate @xmath373 as usual , but the attempt is only successful with probability @xmath354 , so that the motor remains at its site with probability @xmath374 . then the probability distributions @xmath375 , @xmath362 and @xmath27 are related via @xmath376p_{{\rm b}}(r , s)-1\right)\ ] ] and @xmath377\ , p_{\rm b}(r , s)}{3 + 3s-\cos r-\cos q_1-\cos q_2}\nonumber\\ & & + \frac{(1-\pi_{\rm ad})[2-\cos q_1-\cos q_2]p_1(r , s)}{3 + 3s-\cos r-\cos q_1-\cos q_2}. \end{aligned}\ ] ] from these two equations we obtain a rather complicated expression for @xmath27 , which , in the limit of small @xmath63 and @xmath70 , can be reduced to @xmath378 where @xmath224 is the integral ( [ integral_d3 ] ) . for the unbound motors we find @xmath379}.\ ] ] both equations differ from those without the parameter @xmath354 , i.e. from the case @xmath380 , only by a rescaling of the detachment rate , just as in the two - dimensional case discussed above . therefore in this case too , the long time displacement and the probability to be bound to the filament are reduced by a factor @xmath354 , as these quantities are proportional to @xmath381 . in summary , we have calculated various transport properties arising from the random walks of molecular motors . over large length scales ( @xmath382 m ) , molecular motors perform random walks which consist of alternating sequences of directed movements along filaments and non - directed brownian motion in the surrounding fluid . here , we have described these walks as random walks on a cubic lattice and derived analytical solutions for the cases of a single filament in two or three dimensions using fourier laplace transforms . we have obtained closed expressions for the probability distributions of bound and unbound motors and their moments , which can be evaluated numerically for all times . the asymptotic behavior at small and large times was obtained fully analytically . in this way , we derived the fraction of bound motors , the average position and dispersion , as well as effective velocities and diffusion coefficients . all these results were found to be in excellent agreement with results from monte carlo simulations . the random walks of molecular motors exhibit anomalous drift behavior . in two dimensions the average position of both the bound and unbound motors grows as @xmath383 at large times @xmath18 , while in three dimensions the displacements grows only logarithmically . in addition , diffusion parallel to the filament is strongly enhanced . in the two - dimensional case , the diffusion coefficient has an anomalously high value , which is of the order @xmath384 where @xmath14 denoted the small detachment probability , see ( [ dparline ] ) . in the three - dimensional case , there are large logarithmic corrections to the usual diffusion behavior , again of the order @xmath384 , see ( [ largetimediff_d=3 ] ) . finally let us emphasize that similar behavior is also obtained for random walks in confined geometries which have effectively the same dimensionality @xcite . these geometries are accessible to _ in vitro _ experiments . in addition , unbinding of motors from filaments and rebinding to them might also be important for the design of nanotechnological devices using molecular motors as transport systems , which has been proposed by several groups @xcite . . expresses his gratitude for hospitality at the max planck institute in golm , where a major part of his work was done . the tauberian theorems allow one to obtain the asymptotic behavior of a function @xmath385 at large times @xmath18 from the small @xmath63 behavior of its laplace transform @xmath386 $ ] , see , e.g. , @xcite . the following inverse laplace transforms @xmath387 $ ] are used for the random walks of molecular motors : @xmath388 & \approx & \frac{{\alpha}t^{{\alpha}-1}}{\gamma(1+{\alpha})}{\nonumber \\}{\cal l}^{-1}[-s^{-\alpha}\ln s]&\approx & \frac{t^{{\alpha}-1}}{\gamma(1+{\alpha})}[1+{\alpha}\ln t-{\alpha}\psi(1+{\alpha})]{\nonumber \\}{\cal l}^{-1}[s^{-\alpha}\ln^2 s]&\approx & \frac { t^{{\alpha}-1}}{\gamma(1+{\alpha } ) } [ { \alpha}\ln^2 t+2\ln t-2{\alpha}\ln t \psi(1+{\alpha})-2\psi(1+{\alpha})+{\alpha}\psi^2(1+{\alpha})]{\nonumber \\}{\cal l}^{-1}[-\ln^3 s]&\approx & \frac{1}{t}\ , [ 3 ( \ln t+\gamma_e)^2-\frac{\pi^2}{3 } ] . \end{aligned}\ ] ] for each of the four neighbor sites , at filament sites , a motor steps forward with probability @xmath394 and backward with probability @xmath395 ; jumps to each of the adjacent non - filament sites which lead to unbinding occur with probability @xmath396 . the dwell probability is @xmath10 . ] to each of the six neighbor sites , at a filament sites , a forward step has probability @xmath397 , a backward step @xmath395 , and a jump to each of the six adjacent non - filament sites has probability @xmath398 ; as in two dimensions , a motor does not step at all with probability @xmath10 . as in fig . [ fig : randwalk_2d ] , the line of black lattice sites represents the filament . the shaded ares with white lattice sites indicate the planes spanned by the @xmath49- and @xmath399 axes of the lattice perpendicular to the paper plane . ]

_ swift _ discovered grb 050713a and slewed promptly to begin observing with its narrow field instruments 72.6 seconds after the burst onset , while the prompt gamma - ray emission was still detectable in the bat . simultaneous emission from two flares is detected in the bat and xrt . this burst marks just the second time that the bat and xrt have simultaneously detected emission from a burst and the first time that both instruments have produced a well sampled , simultaneous dataset covering multiple x - ray flares . the temporal rise and decay parameters of the flares are consistent with the internal shock mechanism . in addition to the _ swift _ coverage of grb 050713a , we report on the konus - wind ( k - w ) detection of the prompt emission , an upper limiting gev measurement of the prompt emission made by the magic imaging atmospheric cherenkov telescope and _ xmm - newton _ observations of the afterglow . simultaneous observations between _ swift _ xrt and _ xmm - newton _ produce consistent results , showing a break in the lightcurve at t+ ks . together , these four observatories provide unusually broad spectral coverage of the prompt emission and detailed x - ray follow - up of the afterglow for two weeks after the burst trigger . simultaneous spectral fits of k - w with bat and bat with xrt data indicate that an absorbed broken powerlaw is often a better fit to grb flares than a simple absorbed powerlaw . these spectral results together with the rapid temporal rise and decay of the flares suggest that flares are produced in internal shocks due to late time central engine activity .

astro-ph0602490

the _ swift _ gamma - ray burst explorer @xcite has been returning unprecedented data about gamma ray bursts ( grbs ) for the past 18 months . of particular interest in the bursts followed by _ swift _ have been the early time lightcurves of the afterglows which have shown much greater structure and different temporal decay properties than expected , leading to much discussion in the literature regarding the nature of the transition between the prompt emission , thought to be due to synchrotron radiation from internal collisions @xcite , and the afterglow , also thought to be due to synchrotron radiation , though it remains somewhat unclear whether the emission arises in internal or external shocks . _ swift _ observations have shown that grb lightcurves can be described by a canonical 3-segment shape . this shape consists of 1 - an early steep decay phase ( f@xmath2 where 3@xmath3 @xmath4 @xmath3 5 ; t @xmath5 500 s ) 2 - a very shallow decay phase ( 0.5 @xmath3 @xmath6 @xmath3 1.0 ; 500 s @xmath5 t @xmath5 @xmath7 s ) and 3 - a normal decay phase ( 1 @xmath3 @xmath8 @xmath3 1.5 ; @xmath7 s @xmath5 t ) @xcite . superimposed on this smooth decay profile , _ swift _ has shown that bright x - ray flares , 2 to 100 times as bright as the underlying afterglow , are common during the early ( t @xmath3 10 ks ) stages of grbs @xcite . the observation of grb050117 @xcite marked the first time that _ swift _ slewed to and settled on a grb while the prompt gamma ray emission was still in progress , arriving 192 seconds after the burst alert telescope ( bat ) triggered on the burst which had a t@xmath9 of 168 seconds . due to irregularities in the observing mode of the xrt and the proximity of the south atlantic anomaly to _ swift _ during the observation of grb050117 , however , only very sparse data was collected by the xrt , totaling 11.4 seconds in the first orbit and 946 seconds overall . this left large gaps in the lightcurve coverage and severely limited the quality of the spectral analysis that could be performed . we report here on the observation of grb 050713a @xcite , a burst of t@xmath9 = 70 seconds to which _ swift _ slewed and began collecting data with the narrow field instruments ( nfis ) in just 72.6 seconds , while the prompt gamma ray emission was still detectable by the bat . this burst marks just the second time that the bat and xrt have collected simultaneous data on a burst and it marks the first time that both instruments have produced a well sampled , simultaneous dataset covering multiple flares in the prompt emission . in addition to the _ swift _ coverage of grb 050713a , we report also on prompt and followup observations carried out by konus - wind , magic , _ xmm - newton _ and ground based optical observatories . in section 2 we describe the observations and data analysis from all instruments including ground follow - up . in section 3 we discuss the implications of the observations in light of the new theoretical understanding emerging from _ swift _ observations of grbs . in section 4 we summarize and present our conclusions . quoted uncertainties are at the 90% confidence level for one interesting parameter ( i.e. , @xmath10=2.71 ) unless otherwise noted . many different observatories and instruments have observed grb 050713a . we devote the following section to a description of the observations and analysis carried out by each instrument team . all spectral fits were performed using xspec v11.3 . the _ swift _ bat @xcite triggered on grb 050713a at 04:29:02.39 ut , measuring a peak 1-second flux of @xmath11 photons @xmath12 s@xmath13 . t@xmath9 measured in the 15350 kev energy range is @xmath14 s @xcite . the onset of the burst as defined by the bat trigger is preceded by a weak , hard ( photon index = 1.26 ) precursor at t@xmath060 s. bat data were processed using the bat ground software build 11 and bat calibration database files build 11 . at the time of the bat trigger , the flux rose rapidly and remained elevated during a 12 s long , multipeaked burst ( fig 1 ) . at t@xmath0 + 12 s , the bat flux rapidly decayed with a powerlaw decay rate of @xmath15 for 5 seconds before breaking to a more shallow decay of @xmath16 at t@xmath0 + 17 s. this decay continued until t@xmath0 + 40 s at which point the bat flux had decayed to near background levels . at t@xmath0 + 50 s , a flare is seen with peak flux @xmath17 ergs @xmath12 s@xmath13 , extrapolated into the xrt 0.210.0 kev bandpass , followed by a flare with peak flux @xmath18 ergs @xmath12 s@xmath13 at t@xmath0 + 65 s , another at t@xmath0 + 105 s with peak flux @xmath19 ergs @xmath12 s@xmath13 and some hint of further emission at the onset of a flare seen in the xrt at t@xmath0 + 160 s. a weak but statistically significant precursor is seen at t@xmath070 s to t@xmath050 s followed by a period of no significant emission from t@xmath050 s to the burst trigger . the spectrum of the entire bat dataset is well fit by a power - law spectrum with photon index = @xmath20 , though there is evidence for a slightly harder index of 1.45 during the plateau and a softening to @xmath21 during the rapid decay , and further softening to @xmath22 during the weak flares . using the global fit of @xmath23 , the fluence is @xmath24 ergs @xmath12 in the 15350 kev energy range . grb 050713a triggered konus - wind ( k - w ) @xcite at t@xmath0(k - w)= 04:29:01.745 ut . it was detected by the s2 detector , which observes the north ecliptic hemisphere , with an incident angle of 18@xmath251 . the k - w lightcurve in 3 bands is shown in figure 2 . the propagation delay from wind to _ swift _ is 1.387 s for this grb . correcting for this factor , one sees that the k - w trigger time corresponds to t@xmath0 + 0.742 s. prior to t@xmath0(k - w)-0.512 s data were collected by k - w in a survey mode with lower time resolution of 2.944 s and only 3 broad spectral channels , 1870 kev , 70300 kev and 3001160 kev . from t@xmath0(k - w ) to t@xmath0(k - w)+491.776 s , 64 spectra in 101 channels were accumulated on time scales varying from 64 ms near the trigger to 8.19 s by the time the signal became undetectable . the multichannel spectra cover the 18 kev14 mev energy range but no statistically significant emission is seen above 2 mev . data were processed using standard konus - wind analysis tools . joint spectral analysis was carried out using the bat data between 15 and 150 kev and the konus data from 20 to 2000 kev . the spectra were fit by a power law model with an exponential cut off : dn / de @xmath26 e@xmath27 e@xmath28 where e@xmath29 is the peak energy of the @xmath30 f@xmath31 spectrum and @xmath32 is the photon index . the spectrum of the main pulse is well fit ( fig 3 ) with photon index @xmath33 and e@xmath34 kev ( @xmath35=138/119 dof ) . joint fits between bat and konus were also made for other time intervals , including one which shows the faint precursor detected by both instruments at t@xmath3660 s , and will be addressed in greater detail in section 3.3 . the main pulse fluence in the 20 kev to 2 mev range is @xmath37 erg @xmath12 . the 256-ms peak flux measured from t@xmath0 + 1.2 s in the 20 kev to 2 mev band is @xmath38 erg @xmath12 s@xmath13 and the t@xmath9 durations of the burst in the g1 , g2 , g3 energy bands are @xmath39 s , @xmath40 s and @xmath41 s , respectively . the xrt @xcite performs an automated sequence of observations @xcite after _ swift _ slews to a grb detected by the bat . when the spacecraft first settles on the target , a short image ( 0.1 s followed by a longer 2.5 s image if a position is not determined in 0.1 s ) is taken to determine an accurate position . following the image , the xrt switches into either windowed timing ( wt ) mode ( a high timing accuracy mode with 1 dimensional position information ) if the source count rate is above 2 counts s@xmath13 , or photon counting ( pc ) mode ( the more traditional operating mode of x - ray ccds in which full 2 dimensional position information is retained but with only 2.5 s timing resolution ) if the count rate is below 2 counts s@xmath13 . xrt collected a 0.1 s image mode frame upon settling on grb 050713a 73 seconds after the bat trigger , which yields a count rate of 314 counts s@xmath13 . following the image mode frame , xrt cascaded down through its automated mode sequence and collected its first wt frame 4.5 seconds later . at the onset of the wt data , the xrt count rate was about 100 counts s@xmath13 and decaying as a powerlaw . this initial powerlaw decay in the xrt wt data together with the image mode data point measured at a flux level @xmath13 times higher just 4.6 s earlier clearly indicates that the xrt settled and began taking data during the latter portion of the flare detected in the bat at t@xmath0 + 65 seconds ( see fig 4 ) . xrt remained in wt mode throughout the entire first orbit of data collection on grb 050713a , also observing the flare detected by the bat at t@xmath0 + 105 and a lower level flare not clearly detected by the bat at t@xmath0 + 155 s. following a 65 minute period of occultation by the earth , xrt began observations again at t@xmath0 + 4300 s , now observing in pc mode since the countrate of the source had decayed below 2 counts s@xmath13 . a small flare at t@xmath0 + 10 ks and the indication of another flare at t@xmath0 + 45 ks are seen in the late time xrt lightcurve data , superimposed on an otherwise steady powerlaw decay . xrt observations continue to monitor the source until t@xmath0 + 1.8@xmath42 s , a total exposure time of 178 ks , at which time the source had decayed below the xrt detection threshold . xrt data are processed using the _ xrtpipeline _ ground software version 0.9.9 , the redistribution response matrices swxwt0to2_20010101v007.rmf ( wt data ) and swxpc0to12_20010101v007.rmf ( pc data ) , and ancillary response files generated with the _ xrtpipeline _ task _ xrtmkarf_. the x - ray afterglow position determined from ground processing of the data is with an uncertainty of 3.2 arcseconds . this is 10.5 arcseconds from the reported bat position , 0.5 arcseconds from the optical counterpart reported by malesani et al . , @xcite and 1.5 arcseconds from the initial xrt position calculated onboard the satellite and automatically distributed via the gcn network @xcite . an x - ray image compiled from the first segment of xrt pc data is shown as figure 5 with the bat , xrt and optical counterpart error circles displayed . a faint background source is detected 30 arcseconds due south of the grb afterglow at a constant flux level of @xmath43 counts s@xmath13 . the contribution of this steady source has been removed from the calculation of the afterglow lightcurve . a timeline of the xrt ( as well as other ) observations of grb 050713a is shown in table 1 . the lightcurve will be broadly treated in two parts . the first part is the initial orbit of data , during which the lightcurve is characterized by bright flares which are simultaneously observed by the bat as well as the k - w instrument at higher energies . due to the extreme variability in this portion of the lightcurve , a global decay index can not be determined from the xrt data . the second part is the remainder of the xrt data from the second orbit onward , which is characterized primarily by a broken powerlaw decay , though at least one small flare is seen superimposed atop this global decay . _ swift _ finished slewing to grb 050713a at t@xmath0 + 73 s , during the flare which began at t@xmath0 + 65 s. the xrt short image frame is collected just after the peak of this flare , at a flux of @xmath44ergs @xmath12 s@xmath13 , and the first 20 frames of wt data record the decay of the flare . fitting a simple powerlaw to this decay from t@xmath0 + 79 s to t@xmath0 + 100 s setting t@xmath0 to be the bat burst trigger time , we find a powerlaw index of @xmath45 ( 1 @xmath46 ) . at t@xmath0 + 105 s a new flare begins , which rises with a powerlaw index of @xmath47 for 5 - 10 s , flattens at the peak of @xmath48 ergs @xmath12 s@xmath13 for 5 - 10 s , then decays with a more shallow powerlaw index of @xmath49 for about 30 s. at t@xmath0 + 165 s a third flare is detected , which rises with a powerlaw slope of @xmath50 for 5 - 10 s , flattens at the peak of @xmath51 ergs @xmath12 s@xmath13 for 5 - 10 s , then decays with a slope of @xmath52 for 70 s before the end of the observing window due to earth occultation . the second orbit of data in the xrt is the only single orbit of data in which the afterglow is characterized by a well sampled ( greater than 100 events total ) lightcurve devoid of any obvious flaring activity . during the 1600 seconds of data in this orbit , from t@xmath0 + 4360 s to t@xmath0 + 5952 s , the lightcurve decays steadily as a powerlaw with decay index of about 1.0 . the third orbit of data is characterized by another flare , beginning at t@xmath0 + 10 ks , lasting throughout the entire orbit ( about 2 ks ) and reaching a peak flux of @xmath53 ergs @xmath12 s@xmath13 . a powerlaw fit to the rising portion of the flare yields a slope of @xmath54 while the decaying portion yields a slope of @xmath55 . this flare seems to be superposed atop the underlying afterglow decay of decay index @xmath56 . observations were interrupted after 150 s during the fourth orbit due to the occurrence of grb 050713b , and observations of grb 050713a remained suspended until t@xmath0 + 40 ks . some suggestion of another flare is seen in the orbit of data beginning at t@xmath0 + 45 ks , though the statistics are poor . while afterglow data from the xrt alone do not clearly require a break in the afterglow powerlaw , _ xmm - newton _ data ( see @xmath572.4 ) from t@xmath0 + 21 ks to t@xmath0 + 50 ks provide an accurate measure of the late - time decay slope ( @xmath58 ) which can not fit the xrt data from orbits 2 and 3 without a break in the powerlaw . the joint xrt-_xmm - newton _ lightcurve will be further discussed in @xmath572.4 . table 2 summarizes the flares and their temporal fits . the xrt spectral analysis of this burst is somewhat complicated by the high degree of flaring activity seen . in all cases , spectra are binned to a minimum of 20 counts per bin in order to use @xmath35 statistics . fitting the entire first orbit of data , the spectrum is well fit by a highly absorbed powerlaw with photon index @xmath59 and @xmath60 @xmath12 , which is significantly above the galactic column ( @xmath61 @xmath12 ) in the direction of grb 050713a . we are also able , due to the large number of counts in each of the early flares in the dataset , to fit a spectrum to both the rising and decaying portions of the flares . in doing so we see the typical hard to soft evolution of the flares @xcite . the second orbit of data shows a significantly different spectrum from the first , with a harder spectrum with photon index @xmath62 and a lower value of @xmath63 , possibly indicating a period of energy injection @xcite . the third orbit is well fit by a softer powerlaw similar to that which fit the first orbit with photon index @xmath64 and @xmath65 . during the period of overlapping coverage between _ swift _ and _ xmm - newton _ , xrt has 3.5 ks of exposure time at a mean countrate of 0.04 counts s@xmath13 for a total of about 150 events during the simultaneous observing period . fitting a spectrum to this overlapping coverage yields a photon index @xmath66 and @xmath67 . the corresponding mean unabsorbed 0.210.0 kev flux during the overlap period as measured by xrt is @xmath68 ergs @xmath12 s@xmath13 . the data collected after the third orbit ( i.e. , after the temporal break in the lightcurve at t@xmath0+@xmath69 ks ) are too sparse to justify fitting with higher order models , but a simple absorbed powerlaw fit yields a spectrum of photon index @xmath70 with @xmath71 . this is consistent with the x - ray photon index found in orbits 1 and 3 and is marginally softer than the photon index found during orbit 2 which , as noted above , suggests a period of energy injection . _ xmm - newton _ follow - up observations of grb 050713a commenced t@xmath0 + 23.6 ks ( for the epic - pn ) and t@xmath0 + 20.9 ks ( for the two epic - mos cameras ) . the _ xmm - newton _ data were processed with the epproc and emproc pipeline scripts , using the _ xmm - newton _ sas analysis package , version 6.5 . a bright rapidly decaying source is detected near the aimpoint of all three epic detectors , localized at . the net exposures after screening and deadtime correction are 24.1 ks ( pn ) and 27.0 ks ( mos ) . all three epic cameras ( pn and 2 mos ) were used in full window mode with the medium filter in place . source spectra and lightcurves for all 3 epic cameras were extracted from circular regions of 20 arcseconds radius centered on the afterglow . background data were taken from a 60 arcseconds circle on the same chip as the afterglow , but free of any x - ray sources . fitting the afterglow lightcurve with a simple power - law decay results in a decay index of @xmath72 . several flares are present in the background lightcurve , so as a conservative check , we also excluded times where the background rate is @xmath73countss@xmath13 . the afterglow decay rate is then @xmath74 , consistent with the above value . the decay rate from the mos lightcurve ( for the two detectors combined ) is also consistent at @xmath75 . afterglow and background spectra were extracted with the same regions used for the lightcurves , while ancillary and redistribution response files were generated with the sas tasks arfgen and rmfgen respectively . as with xrt data , source spectra were binned to a minimum of 20 counts per bin in order to use @xmath76 statistics . the pn and mos spectra were fitted jointly , allowing only the cross normalization to vary between the detectors , which is consistent within @xmath77 . the two mos spectra and responses were combined to maximize the signal to noise , after first checking that they were consistent with each other . the average net source count rates obtained over the whole observation are @xmath78countss@xmath13 for the pn and @xmath79countss@xmath13 per mos module . allowing the absorption column to vary in the spectral fit results in a formally acceptable fit ( @xmath80 ) . the @xmath81 obtained is @xmath82@xmath12 , while the continuum photon index @xmath83 . the time - averaged , unabsorbed , 0.210.0 kev flux obtained for the afterglow is @xmath84erg@xmath12s@xmath13 . these values are consistent with the _ swift _ xrt measurement obtained at the time of the _ xmm - newton _ observation . the _ xmm - newton _ afterglow spectra were also sliced into three segments of approximately 8 ks in length , in order to search for any spectral evolution within the _ xmm - newton _ observation . no change in the continuum parameters was found , all three spectral segments being consistent with photon index @xmath85 and @xmath86@xmath12 . the spectrum obtained from the pn detector and residuals to an absorbed power - law model ( with @xmath87 and @xmath88@xmath12 ) are shown in figure 6 . the power - law decay index obtained from the _ xmm - newton _ observation ( @xmath89 ) appears to be steeper than that obtained from the _ swift _ xrt in orbit 2 ( @xmath90 ) . in order to compare between the _ xmm - newton _ and _ swift _ afterglow lightcurves , a combined lightcurve from the _ xmm - newton _ and _ swift _ observations was produced , scaling to the absorbed continuum fluxes measured in the 0.510 kev band . the joint _ swift _ and _ xmm - newton _ lightcurve is shown in figure 7 , zoomed to better display the region at which the lightcurve break occurs . a single power - law decay slope of @xmath91 is an extremely poor fit to the lightcurve in this region , with a fit statistic of @xmath92 . indeed the lightcurve from t@xmath0 + 4 ks until t@xmath0 + 1000 ks can be better fitted with a broken power - law . there is a flat decay index of @xmath93 at early times and a steeper decay index of @xmath94 at later times , with the break in the decay occurring at t@xmath95 ks . the fit statistic is then @xmath96 . the remaining contribution towards the @xmath76 originates from two small possible flares present near t@xmath97 ks and t@xmath98 ks . the magic telescope @xcite was able to observe part of the prompt emission phase of grb 050713a as a response to the alert provided by _ swift_. the observation , at energies above @xmath99 , started at t@xmath0 + 40 s , 20 s after reception of the alert . it overlapped with the prompt emission phase measured by _ swift _ and k - w , and lasted for 37 min , until twilight . the observation window covered by magic did not , however , contain the burst onset peak detected at kev - mev energies , where the _ swift _ and k - w spectra were taken . the same region of the sky was observed 48 hours after the burst onset , collecting an additional 49 minutes of data , which was used to determine the background contamination . the magic ( _ major atmospheric gamma imaging cherenkov _ ) telescope is currently the largest single - dish imaging air cherenkov telescope ( iact ) in operation , with the lowest energy threshold ( @xmath100 at zenith , increasing with zenith angle ) . in its fast slewing mode , the telescope can be repositioned within @xmath130 s. in case of an alert by gcn , an automated procedure takes only a few seconds to terminate any pending observation , validate the incoming signal and start slewing toward the grb position , as was the case for grb 050713a . using the standard analysis , no significant excess of @xmath101-like air showers from the position of grb 050713a above @xmath99 was detected @xcite . this holds both for the prompt emission and during the subsequent observation periods . figure 8 shows the number of excess events during the first 37 minutes after the burst , in intervals of 20 s. for comparison , the number of expected background events in the signal region , stable and compatible with statistical fluctuations , is shown . upper limits to the gamma - ray flux are given in table 3 . this is the first observation of the grb prompt emission phase performed by an iact . optical followup observations of grb 050713a performed by the uvot and by ground based observatories are summarized in table 4 . the earliest optical afterglow measurement comes from the raptor - s robotic telescope at the los alamos national laboratory in los alamos , new mexico at r=18.4 @xmath102 0.18 in a coadded series of 8x10 second images with a midpoint observation time of t@xmath0 + 99.3 s @xcite . a nearly simultaneous measurement was made by the robotic liverpool telescope in a coadded series of 3 @xmath12 minute exposures in the r@xmath103 band with a midpoint of observation of t@xmath0 + 3 minutes @xcite . later detections below the digitized sky survey limits were reported within the first 60 minutes after the burst trigger in the r band by the nordic optical telescope ( t@xmath0 + 47 m ) , in the i band by the galileo italian national telescope in the canary islands and in the infrared j , h , and k bands by the astronomical research consortium telescope at apache point observatory ( t@xmath0 + 53 m ) . due to the bright ( v=6.56 ) star hd204408 which is located just 68 arcseconds from the position of the burst , the uvot background level at the position of the afterglow is significantly higher than usual , resulting in abnormally poor sensitivity of the instrument in detecting the afterglow of grb 050713a . considering this high background , the non - detection of the afterglow by the uvot is not surprising . all other reported optical observations of the afterglow position have yielded only upper limits . most of the upper limits are near in time to the actual detections but at brighter magnitudes and thus do not produce strong constraints on the decay rate of the optical afterglow . the r - band measurement made at t@xmath0 + 10.3 hours by the lulin telescope in taiwan , however , is at a sufficiently late epoch to place a useful constraint on the optical decay rate . fitting a simple powerlaw to the two well defined measurements at t@xmath0 + 99.3 s and t@xmath0 + 180 s and the upper limit at t@xmath0 + 10.3 hours yields an upper limit on the power law decay slope of @xmath104 , as is shown in figure 4 . a radio followup observation made with the vla reports no detection at t@xmath0 + 4.3 days . the k - w light curve in the 181160 kev energy range is similar to the _ swift_-bat light curve ( fig 1 ) . the small precursor peak detected by bat at t@xmath070 to t@xmath050 s is seen by k - w at statistically significant levels in all three broad , pre - trigger bands : g1 ( 1870 kev ) , g2 ( 70300 ) , and g3 ( 3001160 kev ) . the other smaller peaks detected by the bat after the burst trigger are not seen at statistically significant levels in the k - w data , despite the fact that the peaks at t@xmath0 + 50 s and at t@xmath0 + 65 s are more intense in the bat energy range than the precursor is . the detection by k - w of the precursor but not the later flares is indicative of the harder spectral index seen in the precursor as compared to the later flares ( see section 3.3 for discussion of separate spectral fits to individual flares ) . the xrt lightcurve with bat data overplotted is shown in figure 4 . both the x - ray and gamma - ray data in the first orbit are dominated by flaring activity , making it difficult to draw a conclusion regarding the underlying powerlaw decay index from this orbit alone . the xrt data beginning at t@xmath0 + 4 ks ( orbit 2 ) and extending until t@xmath0 + 40 ks show a significantly flatter powerlaw decay slope of @xmath105 , implying that a break in the powerlaw decay has occurred near the end of the first orbit of xrt coverage at t@xmath0+@xmath1300 s and that a period of energy injection occurs from t@xmath0+@xmath1300 s to t@xmath0+@xmath115 ks . another break in the lightcurve then occurs near t@xmath0 + 25 ks to a steeper , `` normal '' , pre - jetbreak decay slope , as shown by the _ xmm - newton _ data ( @xmath106 ) . support for this notion of the presence of an energy injection phase may be drawn from the harder x - ray spectral slope of the second orbit of xrt data ( photon index = 1.9 @xmath102 0.13 ) compared to the first orbit ( photon index = 2.28 @xmath102 0.04 ) , the third orbit ( photon index = 2.25 @xmath102 0.23 ) , and the later data ( photon index = 2.8 @xmath102 0.6 ) ( table 5 ) . _ xmm - newton _ data coverage nicely fills much of the data gap in the xrt coverage between t@xmath0 + 15 ks and t@xmath0 + 40 ks and provides high signal to noise data in this regime , producing a confident determination of the lightcurve break . the global picture of the lightcurve of this burst is one in which the early data ( prior to t@xmath0 + 12 s ) shows a bright plateau in the 15 kev to 1 mev energy range , consisting of multiple overlapping peaks . at t@xmath0 + 12 s the emission drops rapidly , consistent with a curvature radiation falloff @xcite until subsequent flaring activity begins to be seen in the 0.3150 kev region with some indication of flux at higher energies from k - w . due to the rapid rise and decay of the flares , internal shocks from continued central engine activity appears to be the most likely explanation for these flares @xcite . the earliest ground based optical detections are reported at this time also , suggesting that the flares may also be optically bright . the lack of higher resolution timing information in the optical data , though , admits the possibility that the optical emission may be unassociated with the emission mechanism responsible for the x - ray flares . it is possible that the optical emission is due to synchrotron emission from the reverse shock ( rs ) , though the much higher flux level of the x - ray flare peaks compared to the optical measurements suggests that the x - ray flares themselves are not due to inverse compton scattering of the optical synchrotron emission of the rs @xcite . following this prompt emission phase , an energy injection phase begins which dominates the lightcurve until at least t@xmath0 + 16 ks . during the energy injection phase , continued activity of the central engine adds energy to the afterglow of the burst , either through additional ejection events or through the realization of energy contained in previously ejected outward moving relativistic shells which only collide at later times , producing late time internal shock emission which is then added to the overall decay @xcite . it may be expected , if the energy injection phase is due to continued central engine activity , that flaring behavior would continue to be observed during this period and , indeed , some evidence for small scale flaring activity during both the second and third orbit of xrt data can be seen , though at a much reduced significance in comparison to the flaring activity of the first orbit . near t@xmath0 + 25 ks , the energy injection phase ends , giving way to a steeper decay slope similar to what is often seen in grbs after the prompt emission phase and prior to the possible onset of a traditional jet - break @xcite . many flares superimposed on top of the overall decay of grb 050713a show the typical properties seen in other bursts : that @xmath107t / t @xmath1 0.1 and that the peak flux level is negatively correlated with the time of the flare @xcite . these two properties of flares seen in _ swift _ grb afterglows have been cited as evidence for flares being produced through accretion processes onto the central compact object @xcite , but we offer here that the constancy of the @xmath107t / t value of flares may partly be a by - product of the overall decay of the afterglow in so much as the sensitivity of the xrt to flares is naturally degraded as the overall flux level of the afterglow decays , thus _ requiring _ flares at later times ( and hence , lower flux levels ) to be longer in duration for enough counts to be collected to produce a significant flare seen above the background . such a case can be seen in comparing the early time flares in the first orbit of grb 050713a to the flare seen in the the third orbit . during the first orbit , the underlying flux level beneath the flares is poorly determined , but can be assumed to be 10 - 100 counts s@xmath13 . we are dominated in this portion of the lightcurve by the poissonian error in the flux , which in a 10 second integration will be 10 - 32 counts , or 3 - 10@xmath108 . thus , for a flare to appear at the 6 sigma level above the background during this portion of the lightcurve , at most a 60@xmath108 increase in fluence above the normal powerlaw decay is needed , which can be acquired in a few seconds by the introduction of a flare with twice the flux of the underlying afterglow . during the third orbit , however , the underlying afterglow flux level has dropped to @xmath1 0.1 counts s@xmath13 . during a 10 second integration at 0.1 counts s@xmath13 the poisson error alone is 1 count , so for a flare to be detectable at 6 sigma above background at these count levels , the total fluence must be 6 counts , implying an increase in the rate from 0.1 counts s@xmath13 to 0.6 counts s@xmath13 during the 10 s interval , a 6 fold increase , which has been seen only in the brighter flares . in order to be sensitive to the same 60@xmath108 increase in flux level as during the first orbit , the flare which occurs at a flux level of 0.1 counts s@xmath13 needs to have a poission error which is 1/6 of the total counts in the observation , i.e. , 36 counts must be collected , which implies an exposure time of at least 180 s if produced by the introduction of a flare with twice the flux of the underlying afterglow . in other words , because the afterglow flux level decays as t@xmath27 , the exposure time needed to acquire the same fluence level increases as t@xmath109 . thus , we see that in moving from the first orbit at t@xmath0 + 100 s to the third orbit at t@xmath0 + 10000 s , assuming a typical underlying powerlaw decay of the afterglow of @xmath110 1 , we have greatly decreased the temporal resolution of xrt to detect flares ( from a few seconds to a few hundred seconds ) . this is not to imply that there is not another more physical cause for the constancy of the @xmath107t / t ratio seen in flares , but rather to note that the typical grb seen by the _ swift _ xrt does not provide sufficient flux at times typically greater than a few ks to detect the shorter timescale flares that are so often seen during the first orbit . in grb 050713a , a hint of emission above the afterglow powerlaw decay appears in the xrt data at t@xmath0 + 45 ks , though the statistics are , predictably , poor . this time is overlapped by _ xmm - newton _ data , though , so we can look for evidence of a short flare in the _ xmm - newton _ data at this time . in figure 9 we show the _ xmm - newton _ lightcurve , plotted linearly and zoomed near t@xmath0 + 45 ks . though a 1 - 2 sigma deviation above the background decay is seen at t@xmath0 + 45 ks , the _ xmm - newton _ data appear consistent with a statistical fluctuation rather than a true flare similar to those seen earlier during the burst . the presence of multiple flares in grb050713a argues against `` one - shot '' emission mechanisms such as synchrotron self - compton emission in a reverse shock or deceleration of the blastwave @xcite and it argues in favor of a mechanism which can produce repeated flares , such as late time central engine activity . while it may remain possible that one of the several flares in grb050713a is due to the rs or the onset of the afterglow due to external shocks , the steep temporal decays of all the temporally fitted flares coupled with the photon indices of the flares ( 1.25 @xmath1 2.5 ; table 6 ) do not satisfy the closure relations of sari et al . , ( 1998 ) , chevalier & li ( 1999 ) and sari et al . , ( 1999 ) for propagation of the blast wave into either a wind or constant density ism . together these points seem to argue in favor of an internal shock origin for the flares seen in this burst . due to the relatively narrow spectral response function of the bat ( 15150 kev for mask - tagged events ) and the xrt ( 0.310 kev ) , a spectral fit to data from only one of the two high energy instruments on _ swift _ is usually not able to discriminate between higher order spectral models . analysts and authors are usually limited to choosing between a power - law or band function . in grb 050713a we have a rare case of data coverage overlap between bat and xrt ( 0.3150 kev ) and also between bat and k - w ( 15 kev14 mev ) . taking advantage of this where appropriate , considering the relative flux levels in the three instruments , we have jointly fitted spectral datasets between the two pairs of instruments . during the precursor and from t@xmath0 + 0 to t@xmath0 + 16.5 s , we perform joint fitting between bat and k - w data . from t@xmath0 + 16.5 to t@xmath0 + 78 s we have only bat data . from t@xmath0 + 78 to t@xmath0 + 116 s and during the onset of the flare at t@xmath0 + 160 we perform joint fitting between xrt and bat . we have grouped the data into segments ( as shown in table 6 ) in order to temporally separate data which we expect may show significantly different spectral parameters . segments 1 - 4 contain bat and k - w data and are segmented to separate the precursor from the prompt emission and the prompt emission from the rapid decay phase . segment 5 contains bat data only and segments 6 - 10 contain xrt and bat data . these are segmented to distinguish the 3 flares which have overlapping data and also to separate the rise of each flare from the decay of each flare . we attempt fits to each of these segments using 4 different spectral models : 1 ) an absorbed powerlaw 2 ) an absorbed cutoff or broken powerlaw ( cutoff for data extending beyond 150 kev , broken otherwise ) 3 ) an absorbed band function and 4 ) an absorbed blackbody plus powerlaw . the precursor is the most poorly sampled of all the regions . despite the low number of counts in the region , a cutoff powerlaw is favored over a single powerlaw at @xmath111 confidence according to the f - test . of all the segments fit , the precursor has the hardest photon index , regardless of the model which is used to perform the fits . the plateau of the prompt emission is best fit by an exponentially cutoff powerlaw model with photon index @xmath112 and e@xmath113 kev . next to the precursor , the prompt plateau has the second hardest photon index of all segments fit , regardless of the model used . as with the other data segments which contain k - w data , the rapid decay segment is poorly fit by a simple powerlaw and is best fit by a cutoff powerlaw or band function . the photon index of the cutoff powerlaw in segment 3 is quite similar to that in the prompt plateau , but the cutoff energy is somewhat lower ( 312 kev compared to 421 kev in the plateau ) , suggesting that the highest energy flux is `` shutting off '' during the rapid decay phase . this segment is an extension of the prompt segment to slightly later times , encompassing slightly more data . the cutoff powerlaw or band function is the best fit , with photon indices similar to segment 2 and e@xmath114 between that in segments 2 and 3 . this segment contains only bat data and is included for completeness , though the narrowness of the bat spectral response limits the ability to distinguish between models . a simple powerlaw is a good fit with photon index of 1.83 . is unconstrained . the powerlaw plus blackbody model produces a good fit to this segment but only with a very minimal blackbody component , effectively reproducing the fit of the simple absorbed powerlaw . therefore we consider the powerlaw plus blackbody model inapplicable to this segment . only in this segment , the data time ranges are mismatched between xrt and bat ( due to xrt observations beginning towards the end of the flare decay ) . rather than ignore this flare or consider only the later part of the flare decay where xrt and bat data coverage overlap , we have chosen to fit the entire bat time range from t@xmath0 + 68 to t@xmath0 + 95 s together with the t@xmath0 + 79 to t@xmath0 + 95 s xrt data ( note that the image mode data taken at t@xmath0 + 73 s are highly piled up and can not be used spectrally ) for consistency with our treatment of the other flares . a simple powerlaw is a good fit to this segment , yielding @xmath115@xmath12 and a photon index of 2.47 , significantly softer than the rise of the flare , as expected . in the rise of the brightest flare seen in xrt , both an absorbed powerlaw plus blackbody model and an absorbed broken powerlaw model are significantly better fits ( f - test probability @xmath116 ) than a simple absorbed powerlaw . the powerlaw plus blackbody indicates @xmath117@xmath12 and a relatively soft photon index of 2.0 with a blackbody temperature of kt@xmath118 kev . we note that this value of kt is below the xrt energy band and may therefore indicate a non - physical spectral solution . the absorbed broken powerlaw indicates @xmath119@xmath12 and photon indices of @xmath120 and @xmath121 , broken at 3.4 kev . these two models are somewhat degenerate in this dataset , with both models producing a roll over in flux at low ( below 0.5 kev ) and high ( above 50 kev ) energies . the decay portion of this flare is well fit by a simple absorbed powerlaw with @xmath122@xmath12 and photon index of 2.68 . we note , however , that both an absorbed broken powerlaw and absorbed powerlaw plus blackbody are equally good fits to the data . the rise of the last flare with overlapping data is well fit by a simple absorbed powerlaw with @xmath123@xmath12 and photon index @xmath124 , however the absorbed powerlaw plus blackbody is , strictly , a better fit according to the f - test , though only at about the @xmath125 confidence level ( f - test probability = 0.219 ) , with @xmath126@xmath12 , kt=200 kev and photon index of 2.43 . we note that this value of kt is above the xrt - bat energy band and may therefore indicate a non - physical spectral solution . the decay of this flare is well fit by an absorbed powerlaw with @xmath127@xmath12 and photon index of 2.55 , though as with segment 9 , the absorbed powerlaw plus blackbody is also an acceptable fit with @xmath128@xmath12 , kt=5.2 kev and photon index of 2.83 . it should be noted that the bat flux is very near the noise level in this segment and really provides only an upper limit on the spectral fitting process in the higher energy region . we have produced the broadband sed ( spectral energy distribution ) of the afterglow of grb 050713a over the time range from t@xmath0 + 20 s to t@xmath0 + 1000 s ( fig 10 ) . this timerange includes detections of the burst afterglow in the optical from the raptor - s and liverpool telescopes ( corrected for the galactic extinction in this direction of @xmath129=1.04 @xcite ) and in the x - ray from _ swift _ bat and xrt . it also includes upper limits in the gamma - ray energy range from k - w ( whose detectable emission ends at t@xmath0+@xmath110 s ) and in the gev energy range from magic . a similar sed has been addressed by the magic collaboration in their paper regarding the magic flux upper limit @xcite in which they note that the sed composed of data from _ swift _ and magic ( 0.2 kev to 400gev ) is fit by a band function at low energy and that the magic data are consistent with a single unbroken powerlaw extending from e@xmath114 ( at @xmath1 400 kev ) to the magic limits up to 500 gev . we confirm this result , citing a best fit photon index for a single powerlaw fit from 400 kev to 500 gev of @xmath130 = 2.1 @xmath102 0.1 and a reduced @xmath131 = 1.66 for 63 dof . we further note that in performing our fit to the magic data , we have treated the magic upper limits as data points during our fit , thus our photon index or 2.1 is only a lower limit on the true photon index of a powerlaw which would fit the true flux level at gev energies . our results here are , therefore , consistent with the analysis of the albert et . , in which they show that their data are consistent with a powerlaw photon index of 2.5 from 400 kev to 500 gev . we add that a band function fit is not , however , consistent with the data when we also consider the contemporaneous optical detections . the relative faintness of the optical detections compared to the x - ray detections combined with the upper limits from k - w and magic requires an absorbed broken powerlaw fit . figure 10 shows the best fit to the entire sed using an absorbed powerlaw ( dotted ) , absorbed broken powerlaw ( solid ) and absorbed band function ( dashed ) . the spectral parameters and fit characteristics for each of these fits are shown in table 7 . we have not corrected for the attenuation of flux above 10gev due to photon - photon interactions with the infrared background @xcite , however , our spectral fit results will remain valid independent of this effect due to the constraints placed by the k - w limiting flux measurement from 20 kev to 14 mev . grb 050713a is one of the rare bursts observed simultaneously in soft x - rays ( xrt ) , hard x - rays ( bat ) and gamma - rays ( k - w ) . the broad spectral coverage of these simultaneous measurements has allowed us to fit the early prompt emission , rapid decay , and several flares in the early emission with several different spectral models . in general we find a cutoff powerlaw model to be a good fit to segments with data extending into the mev range , thus able to constrain the high energy component of the model . for data segments with 0.3150 kev coverage ( bat and xrt data ) we find that a simple absorbed powerlaw is often an adequate fit to the data , though an absorbed powerlaw plus blackbody or absorbed broken powerlaw model seems to sometimes be a marginally better fit during periods of flaring activity . the lightcurve structure of grb 050713a is quite typical of many grbs that have been observed by _ it has an early section showing steep decay slopes of @xmath132 and bright flares extending until t@xmath0+@xmath11 ks , followed by a break to a flatter section with decay slope @xmath133 lasting until t@xmath0+@xmath125 ks , followed by a break to a steeper slope of @xmath58 . we have temporally separated the early , flaring portion of the burst into 10 segments and attempted to fit each segment using 4 different spectral models : 1 ) an absorbed powerlaw 2 ) an absorbed cutoff or broken powerlaw ( cutoff for data extending beyond 150 kev , broken otherwise ) 3 ) an absorbed band function and 4 ) an absorbed blackbody plus powerlaw . in all segments where at least two instruments provide significant , simultaneous levels of emission , and hence the spectral data span more than 2 decades in energy , we find that at least one of the higher order spectral models is acceptable and , in several cases , is a better fit to the data than a simple absorbed powerlaw . this suggests that the spectral shape of grb flares , while consistent with a simple absorbed powerlaw when viewed through any particular narrow spectral window , is intrinsically fit in the broadband by a model with attenuated flux above ( and possibly below ) some threshold energy . it has long been known that grb prompt emission is better fit by spectral models with a high ( and sometimes low ) energy cutoff than by a simple absorbed powerlaw @xcite , and thus the indication that grb flares are fit by a similar spectral model suggests that similar emission mechanisms may be responsible for the production of flares and for the prompt emission itself , namely internal shocks produced as a result of central engine activity . since the discovery of x - ray flares in grbs by _ swift _ , relatively few of the flares have been observed simultaneously across a broad enough energy range to determine whether such higher order models are necessary to fit their spectra , making the multi - instrument observations of grb 050713a unique and valuable . we have also examined the temporal properties of the flares seen in grb 050713a as early as t@xmath0 + 80 s and as late as t@xmath0 + 10 ks . in all cases we find the flares to have steep powerlaw rise and decay slopes and 0.1 @xmath5 @xmath107t / t @xmath5 1 , which also suggests internal rather than external shocks as the production mechanism for the flares @xcite . we have noted that the presence of multiple flares and the failure of those flares to fit the closure relations associated with the external shock in a wind or constant density ism further argues in favor of the internal shock origin for the flares . we have furthermore discussed the difficulty that _ swift _ xrt will face in detecting late time flaring activity . we have noted that the xrt will have difficulty resolving late , short - timescale flares due to the low xrt count rates typically involved . data from higher throughput instruments such as _ xmm - newton _ _ epic _ will be important for constraining flares at these times . grb 050713a has simultaneous coverage at moderately late times with _ xmm - newton _ _ epic _ , but no conclusive evidence of flaring in the _ xmm - newton _ data has been found in this case . finally , we have created a broadband sed of the flaring region of grb 050713a from 0.002 kev to 500gev at times from t@xmath0 + 20 s to t@xmath0 + 1000 s. we find that the sed is inconsistent with a single absorbed powerlaw or an absorbed band function and is best fit by an absorbed broken powerlaw . this overall sed again implies that grb flares are best fit by a spectral model similar to that of the prompt emission itself and thus suggests a common mechanism for the emission from the prompt phase and from flares . this work is supported at penn state by nasa contract nas5 - 00136 . we gratefully acknowledge the contributions of dozens of members of the _ swift _ team at psu , university of leicester , oab , gsfc , asdc , mpe and our subcontractors , who helped make this observatory possible and to the flight operations team for their support above and beyond the call of duty . this work contains observations obtained with _ xmm - newton _ , an esa science mission with instruments and contributions directly funded by esa member states and the usa ( nasa ) . the k - w experiment is supported by russian space agency contract and rfbr grant 06 - 02 - 16070 . albert , j. , et al . , 2006 , apj , 641 , l9-l12 aptekar , r.l . , et al . , 1995 , , 71 , 265 band , d. , et al . , 1993 , , 413 , 231 barthelmy , s. , et al . , 2005 , , 120 , 143 barthelmy , s. , et al . , 2005 , nature 438 , 994 burrows , d.n . , et al . , 2005a , , 120 , 165 burrows , d.n . , et al , 2005b , science , 309 , 1833 chevalier , r.a . , and li , z.y . , 1999 , , 520 , l29 dickey , j.m . and lockman , f.j . 28 , p215 falcone , a.d . , et al . , 2005 , gcn 3581 falcone , a.d . , 2006 , , 641 , 1010 gallant , y.a . , et al . , 1999 , a&as , 138 , 549 g gehrels , n. , et al . , 2004 , , 611 , 1005 gendre , b. , et al . , 2005 , a&as , accepted , astro - ph/0603431 hill , j.e . , et al . , 2004 , , vol . 5165 , 217 hill , j.e . , et al . , 2005 , , 639 , 303 ioka , k. , et al , 2005 , apj , 631 , 429 de jager , o.c . and stecker , f.w . 2002 , apj , 566 , 738 kneiske , t. m. , et al . , 2004 , a&as , 413 , 807 kobayashi , , s. , et al . , 2005 , astro - ph/0506157 kraft , r.p . , burrows , d.n . , nousek , j.a . , 1991 , , 344 , 355 kouveliotou , c. , et al . , 1993 , , 413:l101 , l104 malesani , d. , et al . , 2005 , gcn 3582 mirzoyan , r. et al . , 2005 , ip procs 29@xmath134 icrc nousek , j.a . , et al . , 2005 , , 642 , 389 palmer , d.m . , et al . , 2005 , gcn 3597 perna , r. , armitage , p.j . , zhang , b. , 2006 , apj , 636 , 29 piro , l. , 2005 , , 623 , 314 piro , l. , 2005 , astro - ph/0506020 primack , j.r . , bullock , j.s . , somerville , r.s . 2005 , aip conf . proceedings # 745 , p23 romano , p. , et al . , 2006 , a&as , 450 , 59 roming , p. , et al . , 2005 , , 120 , 95 ryde , f. , 2005 , , 625 , 95 sari , r. , et al . , 1998 , , 497 , l17 sari , r. , and piran , t. , 1999 , , 520 , 641 schlegel , d.j . , finkbeiner , d.p . , and davies , m. , 1998 , , 500 , 525 wren , j. , et al . , 2005 , gcn 3604 zhang , b. and meszaros , p. , 2004 , ijomp a , v19 , 15 , 2385 zhang , b. , et al . , 2006 , , 642 , 354 cllcll swift - bat & 05 - 07 - 13 - 04:29:02.4 & 05 - 07 - 13 - 04:32:00 & 178 & 0 + konus - wind & 05 - 07 - 13 - 04:29:03.1 & 05 - 07 - 13 - 04:37:14.8 & 491.8 & 0.7 + magic(limit ) & 05 - 07 - 13 - 04:29:42 & 05 - 07 - 13 - 05:06:45 & 2223 & 40 + swift - xrt & 05 - 07 - 13 - 04:30:14 & 05 - 08 - 01 - 04:37:02 & 167740 & 72 + xmm - newton & 05 - 07 - 13 - 10:17:00 & 05 - 07 - 13 - 18:22:00 & 20900 & 21000 + cccccc 79 & 101 & 22 & na & 5.6 @xmath102 1.8 & @xmath135 ( from bat ) + 101 & 161 & 60 & 23.3 @xmath102 5 & 8.4 @xmath102 1.8 & @xmath136 + 161 & 304 & 143 & 8.9 @xmath102 3 & 6.1 @xmath102 1.2 & @xmath137 + 9751 & 11840 & 2089 & 5.76 @xmath102 1.8 & 11.0 @xmath102 2.4 & @xmath53 + cccc mcdonald obs , tex & t@xmath0 + 22.4 s & unfilt & 17.7 ( lim ) + raptor - s , lanl & t@xmath0 + 99.3 s & r & 18.4 @xmath102 0.18 + liverpool robotic telescope , canary islands & t@xmath0 + 180 s & r & 19.2 + _ swift _ & t@xmath0 + 252 s & v & 17.98 + _ swift _ & t@xmath0 + 309 s & u & 17.81 + _ swift _ & t@xmath0 + 311 s & uvm2 & 17.13 + _ swift _ & t@xmath0 + 325 s & uvw1 & 16.85 + _ swift _ & t@xmath0 + 326 s & uvw2 & 17.08 + _ swift _ & t@xmath0 + 351 s & b & 18.08 + red buttes obs , wy & t@xmath0 + 27 m & r & 19.4 ( lim ) + red buttes obs , wy & t@xmath0 + 31 m & i & 18.2 ( lim ) + nordic optical tel & t@xmath0 + 47 m & r & @xmath5 dss limit + galileo national telescope , canary islands & t@xmath0 + 48 m & i & @xmath5 dss limit + arc telescope , apache point obs & t@xmath0 + 53 m & j , h , k & detected + red buttes obs , wy & t@xmath0 + 93 m & r & 19.4 ( lim ) + red buttes obs , wy & t@xmath0 + 98 m & i & 18.7 ( lim ) + lulin telescope , taiwan & t@xmath0 + 10.3h & r & 22.4 ( lim ) + vla , nrao & t@xmath0 + 4.3d & 8.5 ghz & 96 microjan + cccc _ swift _ orbit 1 & 2.28 @xmath102 0.04 & @xmath149 & pre- energy injection phase + _ swift _ orbit 2 & 1.90 @xmath102 0.13 & @xmath150 & energy injection phase + _ swift _ orbit 3 & 2.25 @xmath102 0.23 & @xmath151 & flare during energy injection + _ swift _ after orbit 3 & 2.8 @xmath102 0.6 & @xmath152 & post - break + _ xmm - newton _ & 2.1 @xmath102 0.05 & @xmath153 & post - break + + instr & bat & and & k - w & & bat & & xrt & and & bat & + pl : & na & na & na & na & na & 0.52 & 0.59 & 0.60 & 0.54 & 0.38 + pl : phind & 1.26 & 1.44 & 1.61 & 1.47 & 1.83 & 2.47 & 1.72 & 2.68 & 2.52 & 2.55 + pl:@xmath154 & 1.54 & 1.96 & 1.39 & 1.95 & 0.98 & 1.20 & 1.21 & 1.07 & 1.04 & 1.00 + pl : dof & 12 & 101 & 91 & 98 & 26 & 74 & 98 & 125 & 22 & 37 + cutoffpl : phind & 0.913 & 1.26 & 1.31 & 1.25 & na & na & na & na & na & na + cutoffpl : e@xmath114 & 270 & 421 & 312 & 400 & na & na & na & na & na & na + cutoffpl:@xmath35 & 1.20 & 1.60 & 1.02 & 1.29 & na & na & na & na & na & na + cutoffpl : dof & 11 & 110 & 90 & 100 & na & na & na & na & na & na + bknpl : & na & na & na & na & na & 0.58 & 0.43 & 0.45 & na & na + bknpl : phind1 & na & na & na & na & 0.95 & 2.69 & 1.26 & 1.93 & na & na + bknpl : e@xmath155 & na & na & na & na & 5.8 & 4.1 & 3.4 & 1.8 & na & na + bknpl : phind2 & na & na & na & na & 2.1 & 1.5 & 2.0 & 2.8 & na & na + bknpl:@xmath154 & na & na & na & na & 1.06 & 1.31 & 1.03 & 1.04 & na & na + bknpl : dof & na & na & na & na & 23 & 73 & 97 & 124 & 21 & 36 + band : & na & na & na & na & na & 0.33 & 0.49 & 0.54 & 0.41 & 0.33 + band:@xmath32 & -1.13 & -1.27 & -1.27 & -1.29 & -2.01 & -1.69 & -1.46 & -2.32 & -1.98 & -2.17 + band:@xmath156 & -1.26 & -9.36 & -2.39 & -9.29 & -9.07 & -1.88 & -9.36 & -9.32 & -9.04 & -8.78 + band : e@xmath114 & 101 & 761 & 244 & 636 & 994 & 10.6 & 30.7 & 10.6 & 10.6 & 10.6 + band:@xmath154 & 1.85 & 1.61 & 1.03 & 1.28 & 1.06 & 2.11 & 1.11 & 1.13 & 1.26 & 1.12 + band : dof & 10 & 109 & 89 & 101 & 23 & 73 & 97 & 124 & 21 & 36 + pl+bb : & na & na & na & na & na & 0.59 & 1.17 & 0.63 & 0.50 & 0.44 + pl+bb : kt & 3.59 & 59.4 & 26.4 & 44.97 & na & 7.1 & 0.1 & 4.4 & 200 & 5.2 + pl+bb : bb@xmath157 & 0.068 & 2.74 & 0.46 & 1.47 & na & 0.07 & 2.48 & 0.01 & 2.90 & 0.01 + pl+bb : phind & 1.00 & 1.64 & 1.76 & 1.63 & na & 2.75 & 2.00 & 2.77 & 2.43 & 2.83 + pl+bb : pl@xmath157 & 0.053 & 13.97 & 7.69 & 11.65 & na & 1.53 & 2.32 & 2.03 & 0.51 & 0.44 + pl+bb:@xmath154 & 1.15 & 1.79 & 1.03 & 1.60 & na & 1.33 & 1.04 & 1.07 & 0.90 & 0.98 + pl+bb : dof & 10 & 109 & 89 & 101 & na & 73 & 97 & 124 & 21 & 36 + lc pl : & @xmath158 @xmath12 + pl : phind & 2.14 + pl:@xmath154 & 10 ( 65 dof ) + bknpl : & @xmath159 @xmath12 + bknpl : phind1 & 1.1 @xmath102 0.1 + bknpl : e@xmath155 & 1.3 @xmath102 0.2 kev + bknpl : phind2 & 2.2 @xmath102 0.1 + bknpl:@xmath154 & 1.20 ( 63 dof ) + band : & @xmath160 @xmath12 + band:@xmath32 & 1.3 + band:@xmath156 & 2.2 + band : e@xmath114 & 10.6 kev + band:@xmath154 & 2.97 ( 63 dof ) + magic observations . filled circles : number of excess events for 20 s intervals , in the 37 min window after the burst onset . open circles : number of background events in the signal region . no significant source signal is detected above the background . ]

we study the implications of a noncommutative geometry of the minisuperspace variables for the frw universe with a conformally coupled scalar field . the investigation is carried out by means of a comparative study of the universe evolution in four different scenarios : classical commutative , classical noncommutative , quantum commutative , and quantum noncommutative , the last two employing the bohmian formalism of quantum trajectories . the role of noncommutativity is discussed by drawing a parallel between its realizations in two possible frameworks for physical interpretation : the nc - frame , where it is manifest in the universe degrees of freedom , and in the c - frame , where it is manifest through-dependent terms in the hamiltonian . as a result of our comparative analysis , we find that noncommutative geometry can remove singularities in the classical context for sufficiently large values of . moreover , under special conditions , the classical noncommutative model can admit bouncing solutions characteristic of the commutative quantum frw universe . in the quantum context , we find non - singular universe solutions containing bounces or being periodic in the quantum commutative model . when noncommutativity effects are turned on in the quantum scenario , they can introduce significant modifications that change the singular behavior of the universe solutions or that render them dynamical whenever they are static in the commutative case . the effects of noncommutativity are completely specified only when one of the frames for its realization is adopted as the physical one . non - singular solutions in the nc - frame can be mapped into singular ones in the c - frame .

hep-th0408071

over the latest years a great deal of work and effort has been done in the direction of understanding canonical noncommutative field theories and quantum mechanics ( see @xcite and references therein ) . the recent interest in these theories is motivated by works that establish a connection between noncommutative geometry and string theory @xcite . intensive research is carried out to investigate their interesting properties , such as the ir - uv mixing and nonlocality @xcite , lorentz violation @xcite , new physics at very short distances @xcite , and the equivalence between translations in the noncommutative directions and gauge transformations ( see , e.g. @xcite ) . several investigations have been pursued to verify the possible role of noncommutativity in a great deal of cosmological scenarios . among them we quote newtonian cosmology @xcite , cosmological perturbation theory and inflationary cosmology @xcite , noncommutative gravity @xcite , and quantum cosmology @xcite . in a previous work @xcite , an investigation into the influence of noncommutativity of the minisuperspace variables in the early universe scenario was carried out for the kantowski - sachs universe . although noncommutativity effects proved to be relevant to the universe history at intermediate times , they were shown not to be capable of removing the future and past cosmological singularities of that model in the classical context . in the quantum context , on the other hand , non - singular universe solutions were shown to be present . however , since they exist for the commutative quantum kantowski - sachs universe , their presence in the ensemble of solutions of the noncommutative quantum model can not be attributed to the noncommutativity effects . although the investigation carried out in @xcite was restricted to a particular model , one expected that some of the results obtained there could be of general validity . indeed , in this work we shall show that noncommutativity can appreciably modify the evolution of the friedman - robertson - walker ( frw ) universe with a conformally coupled scalar field @xcite . as in reference @xcite , our investigation is carried out by means of a comparative study of the universe evolution in four different scenarios : classical commutative , classical noncommutative , quantum commutative and quantum noncommutative . the main motivation for the choice of the conformally coupled scalar field is that it admits exact solutions in the simpler cases discussed along this work and it is rich enough to be useful as a probe for the significant modifications noncommutative geometry introduces in classical and quantum cosmologies . the analytical treatment renders easy the study of the singular behavior of the model in its four versions . as we shall show later , even in the classical context noncommutative geometry can remove singularities . moreover , depending on the value of the noncommutative parameter , noncommutative classical models can mimic quantum effects . as an interpretation for quantum theory , we are adopting the bohmian one , which we briefly review in section 4 . originally proposed by bohm in 1952 @xcite , and further developed by him in a collaboration with hiley @xcite , such an interpretation of quantum theory has acquired an increasing number of adepts along the years @xcite . decisive contributions for the development of the bohmian quantum physics were made by bell @xcite , who was its arduous defender for three decades , by holland @xcite , and by d. drr , s. goldstein and n. zangh ( see , e.g. , @xcite-@xcite and ref . therein ) . due to the capability it has to reproduce the experimental results of quantum mechanics and quantum field theory providing an intuitive interpretation of the underlying dynamics , bohmian quantum physics is presently an issue of interest for broad community ( see , e.g. , @xcite ) . in this work the main reason that compels us to adopt the bohmian interpretation in quantum cosmology is the absence of external observers in the primordial quantum universe , which renders the standard copenhagen interpretation inapplicable in its description . as an alternative , the bohmian interpretation has been employed in several works of quantum cosmology and quantum gravity ( see , e.g. , @xcite ) . other interesting aspect of the bohmian approach to quantum theory , which motivates us for its adoption , is the efficient framework it provides for comparison between the classical and quantum counterparts of a physical model in the common language of trajectories . we shall benefit from this facility in our comparative study of the four versions of the frw universe . since our preference for bohmian interpretation is justified for technical , rather than philosophical reasons , we shall not focus our discussion on fundamental questions regarding the interpretation of quantum theory ( for references on this subject see @xcite ) . instead , whenever possible , we shall give preference for the interpretation - independent aspects . this work is organized as follows . sections 2 and 3 are devoted to a comparative study of the classical frw with a conformally coupled scalar field and its noncommutative counterpart . in section 4 we present the commutative quantum version of the model , and analyze it by using the bohmian formalism of quantum trajectories . a similar study is carried out in sections 5 and 6 , which are concerned with the noncommutative quantum version of the model . in section 7 we end up with a general discussion and summary of the main results . as a reference for the identification of the noncommutative effects later , it is interesting to consider first the commutative classical frw universe , which we describe as follows . we shall restrict our considerations to the case of constant positive curvature of the spatial sections . the action for the conformally coupled scalar field model in this case is @xcite @xmath1 , \label{9}\ ] ] where @xmath2 is the four - metric , @xmath3 its determinant , @xmath4 is the scalar curvature , and @xmath5 is the scalar field . units are chosen such that @xmath6 and @xmath7 , where @xmath8 is the planck length . for the frw model with an homogeneous scalar field the following _ ansatz _ of minisuperspace can be adopted @xmath9{c}ds^{2}=-n^{2}(t)dt^{2}+a^{2}(t)\left [ \frac{dr^{2}}{1-r^{2}}+r^{2}\left ( d\theta^{2}+\sin^{2}\theta d\varphi^{2}\right ) \right ] , \\ \\ \varphi=\varphi(t).\hspace{7.2cm}\end{array } \right . \label{10}\ ] ] by substituting ( [ 10 ] ) in ( [ 9 ] ) and rescaling the scalar field as @xmath10 we have the following minisuperspace action @xmath11 the corresponding hamiltonian is @xmath12 = n\mathcal{h } , \label{12}\ ] ] where the canonical momenta are@xmath13 as the poisson brackets for the classical phase space variables we have @xmath14 the equations of motion for the metric and matter field variables @xmath15 , @xmath16 @xmath17 and @xmath18 that follow from ( [ 12 ] ) and ( [ 13 ] ) are @xmath9{c}\dot{a}=\left\ { a , h\right\ } = -np_{a}/2a,\\ \\ \dot{p}_{a}=\left\ { p_{a},h\right\ } = 2n,\hspace{1cm}\\ \\ \dot{\chi}=\left\ { \chi , h\right\ } = np_{\chi}/2a,\hspace{0.2cm}\\ \\ \dot{p}_{\chi}=\left\ { p_{\chi},h\right\ } = -2n\chi / a . \end{array } \right . \label{14}\ ] ] from now on we shall adopt conformal time gauge @xmath19 the general solution of ( [ 14 ] ) for @xmath15 and @xmath17 in this gauge is @xmath20@xmath21 where the super - hamiltonian constraint @xmath22 imposes the relation @xmath23 as it can be seen , the classical commutative solutions are necessarily singular in the past and in the future @xmath24 and @xmath25 present plots of the solution for @xmath26 in the dashed curves for given values of @xmath27 and @xmath28 [ @xmath29 is fixed by ( [ 18 ] ) ] . let us introduce a noncommutative classical geometry in our universe model by keeping the hamiltonian with the same functional form as ( [ 12 ] ) , but now valued on noncommutative variables , @xmath30 , \label{19}\ ] ] where @xmath31 , @xmath32 , @xmath33 and @xmath34 satisfy the deformed poisson brackets @xmath35 by making the substitution @xmath36 the theory defined by ( [ 19 ] ) and ( [ 20 ] ) can be mapped into a theory where the metric and matter variables satisfy the poisson brackets @xmath37 written in terms of @xmath38 , @xmath39 , @xmath40 and @xmath41 the hamiltonian ( [ 19 ] ) exhibits the noncommutative content of the theory through @xmath0-dependent terms . two distinct physical theories , one considering @xmath38 and @xmath39 , and the other considering @xmath31 and @xmath32 as the physical scale factor and matter field can be assumed to emerge from ( [ 19 ] ) , ( [ 20 ] ) , ( [ 21 ] ) and ( [ 22 ] ) . in the case where @xmath38 and @xmath39 are assumed as the preferred variables for physical interpretation , the theory can be interpreted as a commutative one with a modified interaction . we shall refer to this theory as being realized in the c - frame . to the other possible theory , which assumes @xmath31 and @xmath32 as the constituents of the physical metric and matter field , we shall refer as realized in the nc - frame . there are works that privilege the c - frame approach ( e.g. @xcite ) , and others the nc - frame ( e.g. @xcite ) . some works rely on the assumption that the difference between c- and nc - variables is negligible ( e.g. @xcite ) . however , as showed in @xcite , even in simple models the difference in behavior between these two types of variables can be appreciable . in what follows we will show that in the cosmological scenario the assumption of the nc- or c - frame point of view as preferential for physical interpretation leads to dramatic differences in the analysis of the universe history . a parallel between the theories in both frames realizations is draw in the classical context here , and in the quantum context in section 6 . methodologically , the route we shall follow in the computation of the configuration variables in this classical context is the same as that adopted in the noncommutative quantum case discussed later on . we shall depart from the c - frame and calculate @xmath42 , @xmath43 , @xmath44 and @xmath45 . after that , we shall use ( [ 21 ] ) to obtain the corresponding @xmath46 and @xmath47 in the nc - frame . the computation of the physical quantities is rendered simpler by the gauge choice @xmath48 , which from now on will be assumed as the gauge employed in all calculations . the equations of motion for the variables @xmath42 , @xmath43 , @xmath49 and @xmath45 are @xmath9{c}\dot{a}_{c}=\left\ { a_{c},h\right\ } = -\frac{1}{2}\left ( 1-\theta ^{2}\right ) p_{a_{c}}+\theta\chi_{c},\\ \\ \dot{p}_{a_{c}}=\left\ { p_{a_{c}},h\right\ } = 2a_{c}-\theta p_{\chi_{c}},\hspace{1.7cm}\\ \\ \dot{\chi}_{c}=\left\ { \chi_{c},h\right\ } = \frac{1}{2}\left ( 1-\theta ^{2}\right ) p_{\chi_{c}}+\theta a_{c},\hspace{0.3cm}\\ \\ \dot{p}_{\chi_{c}}=\left\ { p_{\chi_{c}},h\right\ } = -2\chi_{c}-\theta p_{a_{c}}.\hspace{1.3cm}\end{array } \right . \label{23}\ ] ] according to the values of @xmath0 , the system ( [ 23 ] ) allows three types of solutions , which we describe below . the general solutions for @xmath42 , @xmath43 , @xmath49 and @xmath45 in the @xmath51 case are @xmath52{c}a_{c}(t)=e^{\theta t}\left [ a\sqrt{1-\theta^{2}}\cos\left ( \sqrt { 1-\theta^{2}}t\right ) + b\sqrt{1-\theta^{2}}\sin\left ( \sqrt{1-\theta^{2}}t\right ) \right ] \\ \\ + e^{-\theta t}\left [ c\sqrt{1-\theta^{2}}\cos\left ( \sqrt{1-\theta^{2}}t\right ) + d\sqrt{1-\theta^{2}}\sin\left ( \sqrt{1-\theta^{2}}t\right ) \right ] , \end{array } \label{24}\ ] ] @xmath52{c}\chi_{c}(t)=e^{\theta t}\left [ a\sqrt{1-\theta^{2}}\cos\left ( \sqrt { 1-\theta^{2}}t\right ) + b\sqrt{1-\theta^{2}}\sin\left ( \sqrt{1-\theta^{2}}t\right ) \right ] \\ \\ t}\left [ c\sqrt{1-\theta^{2}}\cos\left ( \sqrt{1-\theta^{2}}t\right ) + d\sqrt{1-\theta^{2}}\sin\left ( \sqrt{1-\theta^{2}}t\right ) \right ] , \end{array } \label{25}\ ] ] @xmath53 + 2e^{-\theta t}\left [ -d\cos\left ( \sqrt{1-\theta^{2}}t\right ) + c\sin\left ( \sqrt{1-\theta^{2}}t\right ) \right ] , \label{25.5}\ ] ] @xmath54 + 2e^{-\theta t}\left [ -d\cos\left ( \sqrt{1-\theta^{2}}t\right ) + c\sin\left ( \sqrt{1-\theta^{2}}t\right ) \right ] . \label{26}\ ] ] from ( [ 21 ] ) , ( [ 24])-([26 ] ) we can calculate @xmath46 and @xmath47 as @xmath52{c}a_{nc}(t)=e^{\theta t}\left\ { \left [ a\sqrt{1-\theta^{2}}-b\theta\right ] \cos\left ( \sqrt{1-\theta^{2}}t\right ) + \left [ b\sqrt{1-\theta^{2}}+a\theta\right ] \sin\left ( \sqrt{1-\theta^{2}}t\right ) \right\ } \\ \\ + e^{-\theta t}\left\ { \left [ c\sqrt{1-\theta^{2}}+d\theta\right ] \cos\left ( \sqrt{1-\theta^{2}}t\right ) + \left [ d\sqrt{1-\theta^{2}}-c\theta\right ] \sin\left ( \sqrt{1-\theta^{2}}t\right ) \right\ } , \end{array } \label{27}\ ] ] @xmath52{c}\chi_{nc}(t)=e^{\theta t}\left\ { \left [ a\sqrt{1-\theta^{2}}-b\theta\right ] \cos\left ( \sqrt{1-\theta^{2}}t\right ) + \left [ b\sqrt{1-\theta^{2}}+a\theta\right ] \sin\left ( \sqrt{1-\theta^{2}}t\right ) \right\ } \\ \\ -e^{-\theta t}\left\ { \left [ c\sqrt{1-\theta^{2}}+d\theta\right ] \cos\left ( \sqrt{1-\theta^{2}}t\right ) + \left [ d\sqrt{1-\theta^{2}}-c\theta\right ] \sin\left ( \sqrt{1-\theta^{2}}t\right ) \right\ } . \end{array } \label{28}\ ] ] the constraint @xmath22 in the present case can be written as @xmath55 in the limit where @xmath56 , equation ( [ 28.5 ] ) is reduced to the equation ( [ 18 ] ) . in the same limit , the nc- and c - frame solutions , given by ( [ 27])-([28 ] ) and ( [ 24])-([25 ] ) , respectively , coincide and match with the commutative solutions ( [ 16 ] ) and ( [ 17 ] ) . from ( [ 24 ] ) , ( [ 27 ] ) and ( [ 28.5 ] ) it can be seen that , as in the commutative case , @xmath46 and @xmath42 are unavoidably singular in the past and in the future . the exponential multiplying factors in both solutions can model their shape giving rise to bounces at intermediate times , as is depicted in the thin and thick solid lines in fig . @xmath57 for representative values of @xmath58 and @xmath28 [ @xmath29 is fixed by ( [ 28.5 ] ) ] . [ ptb ] fc1.eps when @xmath60 the solutions for @xmath42 , @xmath43 , @xmath49 and @xmath45 are @xmath61 @xmath62@xmath63@xmath64 as the corresponding @xmath46 and @xmath65 we have @xmath66@xmath67 the constraint @xmath22 in the present case can be written as @xmath68 as it can be seen , the universe solutions in the case @xmath59 are characterized by a qualitative behavior that differs from the one corresponding to the case@xmath69 . there exist non - singular bouncing solutions for both @xmath46 and @xmath42 , as depicted in fig . . however , the correspondence between the nc- and c - frames can be broken for some values of the integration constants . non - singular solutions in the nc - frame can correspond to singular solutions in the c - frame . an interesting example is the case where @xmath71 and @xmath72 . this corresponds to a bouncing universe in the nc - frame that has no counterpart in the c - frame , where the universe is singular at all times . the general solutions for @xmath42 , @xmath43 , @xmath49 and @xmath45 when @xmath74are @xmath52{c}a_{c}(t)=e^{\theta t}\left [ a\sqrt{\theta^{2}-1}\cosh\left ( \sqrt{\theta ^{2}-1}t\right ) + b\sqrt{\theta^{2}-1}\sinh\left ( \sqrt{\theta^{2}-1}t\right ) \right ] \\ \\ + e^{-\theta t}\left [ c\sqrt{\theta^{2}-1}\cosh\left ( \sqrt{1-\theta^{2}}t\right ) + d\sqrt{\theta^{2}-1}\sinh\left ( \sqrt{\theta^{2}-1}t\right ) \right ] , \end{array } \label{35}\ ] ] @xmath52{c}\chi_{c}(t)=e^{\theta t}\left [ a\sqrt{\theta^{2}-1}\cosh\left ( \sqrt { \theta^{2}-1}t\right ) + b\sqrt{\theta^{2}-1}\sinh\left ( \sqrt{\theta^{2}-1}t\right ) \right ] \\ \\ -e^{-\theta t}\left [ c\sqrt{\theta^{2}-1}\cosh\left ( \sqrt{\theta^{2}-1}t\right ) + d\sqrt{\theta^{2}-1}\sinh\left ( \sqrt{\theta^{2}-1}t\right ) \right ] , \end{array } \label{36}\ ] ] @xmath75 + 2e^{-\theta t}\left [ d\cosh\left ( \sqrt{\theta^{2}-1}t\right ) + c\sin\left ( \sqrt{\theta^{2}-1}t\right ) \right ] , \label{37}\ ] ] @xmath76 + 2e^{-\theta t}\left [ d\cos\left ( \sqrt{\theta^{2}-1}t\right ) + c\sinh\left ( \sqrt{\theta^{2}-1}t\right ) \right ] . \label{37.5}\ ] ] from ( [ 21 ] ) , ( [ 24])-([26 ] ) we can calculate the corresponding @xmath46 and @xmath47 as @xmath52{c}a_{nc}(t)=e^{\theta t}\left\ { \left [ a\sqrt{\theta^{2}-1}+b\theta\right ] \cosh\left ( \sqrt{\theta^{2}-1}t\right ) + \left [ b\sqrt{\theta^{2}-1}+a\theta\right ] \sinh\left ( \sqrt{\theta^{2}-1}t\right ) \right\ } \\ \\ + e^{-\theta t}\left\ { \left [ c\sqrt{\theta^{2}-1}-d\theta\right ] \cosh\left ( \sqrt{\theta^{2}-1}t\right ) + \left [ d\sqrt{\theta^{2}-1}-c\theta\right ] \sinh\left ( \sqrt{\theta^{2}-1}t\right ) \right\ } , \end{array } \label{38}\ ] ] @xmath52{c}\chi_{nc}(t)=e^{\theta t}\left\ { \left [ a\sqrt{\theta^{2}-1}+b\theta\right ] \cosh\left ( \sqrt{\theta^{2}-1}t\right ) + \left [ b\sqrt{\theta^{2}-1}+a\theta\right ] \sinh\left ( \sqrt{\theta^{2}-1}t\right ) \right\ } \\ \\ -e^{-\theta t}\left\ { \left [ c\sqrt{\theta^{2}-1}-d\theta\right ] \cosh\left ( \sqrt{\theta^{2}-1}t\right ) + \left [ d\sqrt{\theta^{2}-1}-c\theta\right ] \sinh\left ( \sqrt{\theta^{2}-1}t\right ) \right\ } . \end{array } \label{39}\ ] ] when @xmath77 the constraint @xmath22 is reduced to @xmath78 as in the case where @xmath59 , there are present non - singular bouncing solutions in both nc- and c - frames . @xmath79 depicts one example . again we find that , depending on the values of the integration constants , non - singular universes in the nc - frame can correspond to singular universes in the c - frame [ fig . @xmath80 . it is now clear that the @xmath59 case establishes a division between qualitatively different ensembles of noncommutative universe solutions , since only for @xmath81 non - singular universes can exist . but this is not the only interesting property that appears when @xmath82 as it will be shown later , the solutions in this case have a particular behavior that reveals the capability noncommutativity has to mimic quantum effects under special conditions . here we present the quantum version of the commutative universe model discussed in section 2 . the frw universe with conformally coupled scalar field has already been investigated in the basis of the wheeler - dewitt equation in @xcite and @xcite , the latter using bohmian trajectories . however , in reference @xcite there was a restriction to the regime of small scale parameter , and the wavefunctions considered were different from the ones studied in this work . the quantization of minisuperspace model is here carried out by employing the dirac formalism ( for details see @xcite ) . by making the canonical replacement @xmath83 and @xmath84 in ( [ 12 ] ) we obtain , applying the dirac quantization procedure , @xmath85 \psi(a,\chi)=0 , \label{45}\ ] ] which is the wheeler - dewitt equation for the conformally coupled scalar field model . this equation can be solved by separating the @xmath15 and @xmath17 variables , as it has been done in the literature ( see @xcite and references therein ) . however , by making a change of variables there is another route to tackle the problem that is interesting by the ensemble of solutions it generates . we shall present it here , and show , later on , that such a route it is particularly suitable for application in the noncommutative quantum case . by making the coordinate change @xmath86 we can rewrite ( [ 45 ] ) as @xmath87 \psi(\xi,\eta)=0.\label{47}\ ] ] by plugging in the _ ansatz _ @xmath88 in ( [ 47 ] ) we obtain , after simplification , @xmath89 a solution to ( [ 49 ] ) is @xmath90 where @xmath91 and @xmath92 are bessel functions of the second kind , @xmath93 and @xmath94 are constants and @xmath95 is a real number . the solution of the wheeler - dewitt equation ( [ 47 ] ) is therefore @xmath96 such a kind of wavefunction also appears , e.g. , in the study quantum wormholes @xcite and in quantum cosmology of the kantowski - sachs universe @xcite . the contribution corresponding to @xmath92 is usually discarded because it leads to a solution that is divergent in the classically forbidden region of the potential . from the point of view of the quantum trajectories , as it will be clear in the next subsection , there is no fundamental reason for this solution be discarded . however , since in this work our main interest is in the influence of noncommutativity in cosmology , rather than in the foundations of quantum theory , we shall give preference for wavefunctions that are also admissible in interpretations other than the bohmian one . we shall therefore discard the @xmath92 contribution and write the solution of ( [ 47 ] ) as is a continuous index , in the most general case the summation can be replaced by an integral . ] @xmath97 in order establish a framework where all versions of the universe model can be compared , it is interesting to appeal to a common language . this is provided by the bohmian quantum trajectory formalism , which we briefly describe in this section . for a more detailed account of the subject , see the references given in the introduction . in the formulation presented here , we shall benefit from ideas proposed in @xcite . the wavefunction will be assumed as not a constituent of the physical system , as originally assumed by bohm @xcite , but as a representation of it . quantum information theory tells us that the wavefunction has a non - physical character @xcite . bohmian quantum physics should in some way be in accordance with this fact . actually , the comprehension of the meaning of the wavefunction as a representation of a quantum system is crucial to achieving the understanding of quantum mechanics from any perspective @xcite . the same point of view seems to be suitable for adoption in quantum cosmology . we shall briefly comment this aspect when applying the bohmian formalism in the description of the noncommutative version of our universe model . when dealing with a quantum model one must have a clear picture of the elements of ontology of the theory . by the elements of ontology , we mean what the theory is essentially about . the orthodox quantum theory based on the copenhagen interpretation , e.g. , is about observers that realize measurements . in the bohmian interpretation , on the other hand , quantum theory is concerned with the physical systems , which can be particles , waves , strings , etc . in this work the object of attention is the primordial quantum universe , characterized , in the minisuperspace formalism , by the configuration variables @xmath15 and @xmath17 . having fixed the objects of ontology of the theory , we must determine how they evolve in time . this is done with the aid of the wavefunction , whose role is to provide us the evolution law . the procedure is best illustrated in the context of non - relativistic quantum mechanics . bohmian non - relativistic quantum mechanics is concerned with the behavior of point particles that move in space describing quantum trajectories . an evolution law is ascribed to them according to the rule @xmath98 } { \psi^{\ast}\psi}\right\ } = \frac{\nabla s}{m } , \label{60}\ ] ] where @xmath99 is the wavefunction and @xmath100 is obtained from the polar decomposition @xmath101 . as in the orthodox interpretation , the wavefunction satisfies the schrdinger equation @xmath102 equations ( [ 60 ] ) and ( [ 61 ] ) specify completely the theory . without any other axiom , all phenomena governed by nonrelativistic quantum mechanics , from spectral lines and quantum interference phenomena to scattering theory , superconductivity and quantum computation follow from the analysis of the dynamical system defined by ( [ 60 ] ) and ( [ 61 ] ) @xcite . the expectation value of a physical quantity associated with a hermitian operator @xmath103 in the standard formalism can computed in the bohmian formulation by ensemble averaging the corresponding `` beable '' @xmath104 } { \psi^{\ast}\psi}\right\ } = a\left ( x^{i},t\right ) , \label{61.5}\ ] ] which represents the same quantity when seen from the bohmian perspective . ) as taking the local expectation value of the observable @xmath105 . such a nomenclature is not adopted here because it is unsuitable to be used in theories where the object of ontology is an individual system , as in quantum cosmology . ] in the context of non - relativistic quantum mechanics , it can be shown from first principles that an ensemble of particles obeying the evolution law ( [ 60 ] ) the associated probability density in the configuration space must be given by @xmath106 @xcite . this is why computing the ensemble average of @xmath107@xmath108 we arrive at the same results of the standard operatorial formalism . notice that the law of motion ( [ 60 ] ) can itself be obtained from ( [ 61.5 ] ) by associating @xmath109 with the beable corresponding to the velocity operator @xmath110\right ) = \frac{\nabla s}{m}.\label{62.5}\ ] ] as a consequence of being an objective theory of point particles describing trajectories in space , bohmian quantum mechanics does not give to probability a privileged role . instead , as discussed in @xcite , such a formulation probability is a derived concept , a decurrent of the law of motion of the point particles . the bohmian formulation is thus eminently suitable for the study of individual systems , as the primordial quantum universe . in the remaining of this section , we will be concerned with the application of the theory to the commutative quantum universe discussed in the beginning of this section . in the next sections a similar study will be carried out for the noncommutative quantum case . in this subsection we shall apply the bohmian formalism to obtain information about the evolution of the quantum frw universe with a conformally coupled scalar field . in the description of quantum cosmology employing quantum trajectories we shall extend the evolution law ( [ 62.5 ] ) to the minisuperspace variables . in the commutative case the resulting bohmian minisuperspace formalism matches with the minisuperspace version of the bohmian quantum gravity proposed in @xcite , and employed to study the conformally coupled scalar model in @xcite . from ( [ 62.5 ] ) we find , in the gauge @xmath111@xmath112 } { \psi^{\ast}\psi}\right\ } = -\frac { 1}{2}\frac{\partial s}{\partial a } , \label{69}\]]@xmath113 } { \psi^{\ast}\psi}\right\ } = \frac { 1}{2}\frac{\partial s}{\partial\chi}. \label{69.5}\ ] ] by changing ( [ 69 ] ) and ( [ 69.5 ] ) into the@xmath114 coordinates defined by ( [ 46 ] ) we obtain @xmath115 in what follows@xmath116 we shall solve the system ( [ 70 ] ) in two examples where the a universe is characterized by a wavefunction of the type ( [ 51 ] ) . let us consider first the example where there is a single bessel function in ( [ 51 ] ) . in this case the wavefunction is given by @xmath117 where @xmath93 is a constant . since the bessel function @xmath118 is real for @xmath119 real and @xmath120 , the phase can be read directly from the exponential in ( [ 74 ] ) : @xmath121 . the equations of motion ( [ 70 ] ) in this state are therefore reduced to @xmath122 whose solutions are @xmath123 the corresponding @xmath26 and @xmath124 obtained from ( [ 46 ] ) are given by @xmath125 quantum effects can therefore remove the cosmological singularity , giving rise to bouncing universes . under suitable conditions , solution ( [ 78 ] ) can represent a quantum universe that is indistinguishable from a noncommutative classical universe . this is seen by noting that it can be mapped into the solution for @xmath42 ( eq . [ 29 ] ) in the case where @xmath59 by identifying @xmath126 , @xmath127 and @xmath128 , or to the solution for @xmath46 ( eq . [ 33 ] ) by identifying @xmath129 @xmath130 and @xmath131 . in the case where @xmath132 equation ( [ 78 ] ) describes a static universe with arbitrary large scale factor , a highly nonclassical behavior . let us now consider a wavefunction that is a superposition of two bessel functions in ( [ 51 ] ) , that is @xmath133 as the corresponding phase we find @xmath134 , \label{81}\ ] ] where the @xmath135 and @xmath136 were chosen as real coefficients . the equations of motion ( [ 70 ] ) for this state are @xmath137 \xi\sin\left [ \left ( \mu-\nu\right ) \eta\right ] } { a_{1}^{2}k_{i\mu/2}^{2}+a_{2}^{2}k_{i\nu/2}^{2}+2a_{1}a_{2}k_{i\mu/2}k_{i\nu/2}\cos\left [ \left ( \mu-\nu\right ) \eta\right ] } , \hspace{1.8 cm } \label{82}\]]@xmath138 } { a_{\mu}^{2}k_{i\mu /2}^{2}+a_{\nu}^{2}k_{i\nu/2}^{2}+2a_{1}a_{2}k_{i\mu/2}k_{i\nu/2}\cos\left [ \left ( \mu-\nu\right ) \eta\right ] } , \label{83}\ ] ] where prime means derivative with respect to the argument . the system ( [ 82])-([83 ] ) is set of nonlinear coupled differential equations . analytical solutions are difficult to find . numerical solutions , on the other hand , can be easily computed for @xmath139 and @xmath140 . once the solutions for @xmath139 and @xmath140 are found , the corresponding @xmath26 and @xmath124 are determined from ( [ 46 ] ) . the qualitative properties of the solutions of autonomous system such as ( [ 82])-([83 ] ) can be determined by analyzing the associated field of velocities . from the right hand side ( rhs0 of ( [ 82])-([83 ] ) we can see that the field of velocities has its direction inverted under the substitution @xmath141 @xmath142 . therefore , to have a qualitative picture of the associated flow one must be concerned only with the relative sign of @xmath143 and @xmath119 . for simplicity , let us fix @xmath144 , and , without loss of generality , consider @xmath145 the direction fields corresponding to the two possible combinations of relative sign between @xmath143 and @xmath119 are depicted in figs . and@xmath147 for representative values of these constants , where the orbits of some solutions are also plotted . as it can be seen , the case with positive @xmath143 and negative @xmath119 favors the formation of closed orbits , which correspond to non - singular cyclic universes . the case with positive @xmath143 and @xmath119 , on the other hand , tends to privilege the open orbits . the presence of the trigonometric functions in the rhs of ( [ 82])-([83 ] ) is the reason for the repetitive pattern of the direction fields observed along the @xmath148 direction with period @xmath149 in figs . @xmath146 and@xmath147 . the systematic appearance of closed orbits along the @xmath148 direction in fig . @xmath146 , and the possibility of varying their amplitudes with an appropriate choice of the initial conditions and the constants @xmath143 and @xmath119 , show us that the cyclic universe solutions can present a variable @xmath150 another quantity that is variable is the number of @xmath151-folds between its maximum and minimum size configurations . this information can be read directly from the logarithmic plot of @xmath26 in fig . @xmath152 , where the solution depicted corresponds to one of the , closed orbits of fig . @xmath146 . another interesting non - singular solution type is depicted in fig . this corresponds to one of the open orbits of fig . @xmath154 , and is an example of an universe that passes by sequence of infinite bounces starting in the infinite past and never ending . the sequence is itself enveloped by a larger bounce@xmath155 a different non - singular solution type is present in the case where @xmath156 . since the phase of such a kind of state is @xmath157 , the corresponding universe is necessarily static . [ ptb ] fc2.eps [ ptb ] fc3.eps after having studied the individual manifestation of noncommutative and quantum effects in the conformally coupled scalar field universe , we are now in time to study the combination of them in a unique model . this is achieved by considering the quantum version of equation ( [ 20 ] ) : @xmath158=i\theta\text { , \ } [ \widehat{a},\widehat { p}_{a}]=i\text { , \ } [ \widehat{\chi},\widehat{p}_{\chi}]=i\text{,\ \ } [ \widehat{p}_{a},\widehat{p}_{\chi}]=0 . \label{100}\ ] ] according to the weyl quantization procedure @xcite , the commutation relation above between the observables @xmath159 and @xmath160 can be realized in terms of commutative functions by making use of the moyal star product defined as below @xmath161 the commutative coordinates @xmath38 and @xmath39 are called weyl symbols of the operators @xmath159 and @xmath160 , respectively . a wheeler - dewitt equation the can be adopted for the noncommutative scalar field universe is @xmath162 \psi ( a_{c},\chi_{c})+4\left ( a_{c}^{2}-\chi_{c}^{2}\right ) \star\psi(a_{c},\chi_{c})=0 , \label{102}\ ] ] which is obtained by moyal deforming ( [ 45 ] ) . by using the properties of the moyal product , it is possible to write the ( [ 102 ] ) as @xmath163 \psi(a_{c},\chi _ { c})+4\left ( \widehat{a}^{2}-\widehat{\chi}^{2}\right ) \psi(a_{c},\chi _ { c})=0 , \label{103}\ ] ] where @xmath164 equation ( [ 104 ] ) is nothing but the operatorial version of equation ( [ 21 ] ) . the notation @xmath38 and @xmath39 is now justified . these symbols match exactly with the canonical variables defined by ( [ 21 ] ) . here we are in face with the same situation as in the classical case . two consistent cosmologies are possible . one considering @xmath165 and @xmath166 as the operators associated with the physical metric , and the other considering @xmath167 and @xmath168 in the application problems we shall consider the two possibilities . in order to draw a parallel between the noncommutative quantum universe and the other three universe types , it is necessary to have a prescription of how to compute bohmian trajectories in noncommutative quantum cosmology . the simplest way to do this is by extending the bohmian formulation discussed in section 4 along the same lines proposed in @xcite . the procedure consists in departing from the c - frame and using the beable mapping ( [ 61.5 ] ) to ascribe an evolution law to the canonical variables . in our time gauge for the noncommutative cosmology , @xmath48 ( see section 3 ) , the hamiltonian ( [ 19 ] ) reduces simply to@xmath169 .\label{104.5}\ ] ] we can therefore use @xmath170 to generate time displacements and obtain the bohmian equations of motion for @xmath171 and @xmath172 as @xmath173\right ) = -\frac{1}{2}\left ( 1-\theta^{2}\right ) \frac{\partial s}{\partial a_{c}}+\theta\chi_{c},\label{105}\]]@xmath174\right ) = \frac{1}{2}\left ( 1-\theta^{2}\right ) \frac{\partial s}{\partial\chi_{c}}+\theta a_{c}.\label{106}\ ] ] the connection between the c- and nc - frame variables is established by applying the beable mapping to the operatorial equations ( [ 104 ] ) , that is , by defining @xmath175 and @xmath176 . once the trajectories are determined in the c - frame , one can find their counterparts in the nc - frame by evaluating the variables @xmath15 and @xmath17 along the c - frame trajectories , @xmath177 , \label{109}\]]@xmath178 .\label{110}\ ] ] in what follows we shall illustrate the application of the formalism in noncommutative quantum cosmology . by using the representations @xmath179 and @xmath180 , we can write the noncommutative wheeler - dewitt equation ( [ 103 ] ) as @xmath181 \psi(a_{c},\chi _ { c})=0 , \label{111}\ ] ] where @xmath182 . the separation of variables can be made by changing to the new set of coordinates @xmath183 which allow us to rewrite ( [ 111 ] ) as @xmath184 \psi(\xi,\eta)=0 . \label{113}\ ] ] the computation of the bohmian trajectories is rendered easy by expressing the equations of motion ( [ 105 ] ) and ( [ 106 ] ) in the same hyperbolic coordinates as the wavefunction . after the change of variables , the bohmian equations of motion can be written as @xmath185 the equations ( [ 109 ] ) and ( [ 110 ] ) , responsible by the nc - c - frame correspondence , can be written in the new set of coordinates as@xmath186 -\frac{\theta}{2}\xi^{-1}\cosh\eta\partial_{\eta } s\left [ \xi(t),\eta(t)\right ] , \label{113.7}\]]@xmath187 -\frac{\theta}{2}\xi^{-1}\sinh\eta \partial_{\eta}s\left [ \xi(t),\eta(t)\right ] . \label{113.8}\ ] ] before starting our comparative study by computing bohmian trajectories corresponding to specific solutions of ( [ 113 ] ) , let us discuss the case of real wavefunctions . while in the commutative bohmian quantum cosmology real wavefunctions always represent static universes , in the noncommutative bohmian quantum cosmology they can represent dynamical universes , a property pointed out in @xcite for the kantowski - sachs model . in the frw with a conformally coupled scalar field under consideration , equations ( [ 113.2 ] ) tell us that to real wavefunctions correspond always a nontrivial and identical dynamics . moreover , from ( [ 113.7 ] ) and ( [ 113.8 ] ) we can see that when @xmath157 the nc- and c - frame realizations are indistinguishable , representing the same universe . this universe is determined by solving equations ( [ 113.2 ] ) and substituting the solutions in ( [ 112 ] ) . as a result , we find @xmath188 real wavefunctions therefore always represent non - singular bouncing universes . complex wavefunctions , on the other hand , can give rise to a great variety of dynamics , where the distinction between the frames of physical realization can be crucial . in the same way as in the classical analog , we can distinguish three cases : @xmath189 and @xmath190 in this case , equation ( [ 113 ] ) can be solved by using the _ ansatz _ @xmath192 as a result , we find @xmath193 whose solution is @xmath194 where @xmath195 and @xmath196 are whittaker functions , @xmath197 and @xmath198 are constants and @xmath95 is a real number . the piece corresponding to the @xmath196 contribution leads to a divergent wavefunction in the classically forbidden region . we shall therefore discard its contribution in a similar way as in the commutative model discussed before . we can thus write the solution of ( [ 113 ] ) as @xmath199 . \label{117}\ ] ] in the limit were @xmath200 the whittaker functions @xmath201 are reduced to the bessel functions@xmath202 through the relation , @xmath203 and therefore the wavefunction ( [ 117 ] ) matches with the commutative wavefunction ( [ 51 ] ) . in the sequel we present two examples of application of the bohmian formalism in the investigation of the properties of the wavefunctions . the wavefunction is of the type @xmath204 where @xmath93 is a constant . since the whittaker function @xmath201 is real for @xmath205and@xmath206 real and @xmath120 , the phase can be read directly from the exponential : @xmath207 the equations of motion for @xmath208 and @xmath148 in this state are@xmath209 as the solutions for @xmath42 and @xmath210 we have @xmath211 where @xmath212 @xmath213 and@xmath214 was absorbed by redefining the origin of time . from ( [ 113.7 ] ) and ( [ 113.8 ] ) we can write @xmath215 where @xmath216 again we found bouncing solutions that can be mapped into the classical solution of the same type with a suitable identification of the integration constants . but this is not the only interesting property exhibited in this case . from ( [ 137 ] ) we can see that to each universe in the nc frame there corresponds at least one universe in the c - frame , as it is shown in fig . @xmath217 . for @xmath218 positive , the correspondence is one to one and exists only for @xmath219 . smaller values of @xmath220 would imply a singular universe in the nc - frame . for negative values of @xmath218 , on the other hand , the correspondence is defined for all values of @xmath220 . to each universe in the nc - frame there correspond two universes in the c - frame . an exception occurs for @xmath221 , where the curve @xmath222 achieves its minimum and the correspondence is one - to - one . this value of @xmath220 marks the division between the two regimes that govern the nc - c - frame correspondence : large @xmath223 and small @xmath220 and of large @xmath223 and @xmath220 . a similar behavior was previously found in @xcite in non - relativistic bohmian quantum mechanics when studying the harmonic oscillator . the capability of the noncommutativity effects to promote the interplay between large and small scale distances was interpreted in that reference as manifestation of a sort of ur - uv mixing in the oscillator orbits . [ ptb ] fc4.eps the wavefunction function ( [ 117 ] ) is a sum of two functions , @xmath224 the corresponding phase is @xmath225 , \label{152}\ ] ] where , as in the commutative case , @xmath135 and @xmath136 were chosen as real coefficients . the equations of motion ( [ 113.2 ] ) in this state are @xmath226 \sin\left [ \left ( \mu-\nu\right ) \eta\right ] } { a_{1}^{2}w_{\mu\theta/2\sqrt{\beta},i\mu/2}^{2}+a_{2}^{2}w_{\nu\theta/2\sqrt{\beta } , i\nu/2}^{2}+2a_{1}a_{2}w_{\mu\theta/2\sqrt{\beta},i\mu/2}w_{\nu\theta /2\sqrt{\beta},i\nu/2}\cos\left [ \left ( \mu-\nu\right ) \eta\right ] } , \label{153}\]]@xmath227 } { a_{1}^{2}w_{\mu\theta/2\sqrt{\beta},i\mu/2}^{2}+a_{2}^{2}w_{\nu\theta/2\sqrt{\beta},i\nu/2}^{2}+2a_{1}a_{2}w_{\mu\theta/2\sqrt{\beta } , i\mu/2}w_{\nu\theta/2\sqrt{\beta},i\nu/2}\cos\left [ \left ( \mu-\nu\right ) \eta\right ] } + \theta , \label{154}\ ] ] where prime means derivative with respect to the argument . as in the commutative counterpart , we have an autonomous set of non - linear coupled differential equations to solve . in order to render easy the comparison with that case , let us fix @xmath144 . again its is possible to find bouncing solutions , as it is shown in fig @xmath228 , where the splitting between nc- and c - frame evolutions is quantitatively irrelevant . the effect of noncommutativity in that case is manifest through the suppression of the sequence of bounces that appears in the commutative counterpart and by shifting the time where the universe achieves its minimum size [ see fig . @xmath153 ] . the case where @xmath230 , which in the commutative counterpart corresponds to a static universe , here has a non - trivial dynamics . one example is presented in fig . @xmath231 , where a cyclic solution similar to that of fig . @xmath152 is depicted . the splitting between nc- and c - frame evolutions in this case is again quantitatively irrelevant . as in the classical counterpart , the case @xmath59 is marked by a peculiar behavior . equation ( [ 113 ] ) in this case is reduced to the first order partial differential equation @xmath232 \psi(\xi,\eta)=0 , \label{160}\ ] ] whose general solution is @xmath233 where @xmath234 is any differentiable function of @xmath235 the equations of motion ( [ 113.2 ] ) in this case are reduced to @xmath236 whose solutions are @xmath237 as the solutions for @xmath42 and @xmath210 we have @xmath238 while the corresponding @xmath26 and @xmath124 are @xmath239@xmath240 in both nc- and c - frame realizations of noncommutativity we find bounce solutions whenever @xmath241 , while for @xmath242 the universe is necessarily singular [ fig . @xmath243 the novelty here is that these universe solutions are the most general ones available . no matter what is the wavefunction , the fate of the universe is determined uniquely by @xmath244 . at first sight it could seem strange that the wavefunction can not have any influence on the fate of the universe , independent of its functional form . however , if we realize that the information provided by the wavefunction is about the universe evolution law , we find that the wavefunction is playing its hole providing us equations ( [ 162 ] ) in the same way as in all the other cases previously discussed . since the kinetic term is quenched by noncommutativity effects , what we found in this case is exactly what one would expect to find : a poor and highly constrained dynamics , similar to the one that appears when a magnetic field projects a system onto its lowest landau level ( see @xcite and ref . therein ) . in this last case the most general wavefunction that satisfies ( [ 111 ] ) can be written as @xmath245 . \label{180}\ ] ] contrary to the previous cases , the contribution corresponding to @xmath246 is not divergent in the classically forbidden region . moreover , each of the whittaker functions @xmath247 and @xmath196 in this case is complex , and therefore can give rise to a dynamics that differs from the ones of the examples previously discussed . for simplicity , we shall consider only the example where @xmath248 for this wavefunction , the equations of motion ( [ 113.2 ] ) for @xmath139 and @xmath140 can be shown to be @xmath249 , \label{182}\]]@xmath250 the equations ( [ 182 ] ) and ( [ 183 ] ) can be solved analytically in the limit of large @xmath208 , where the contribution coming from the term containing the whittaker functions in the right hand side of ( [ 182 ] ) can be approximated by @xmath251 in this regime , ( [ 182 ] ) can be simplified to @xmath252 the solutions of ( [ 183 ] ) and ( [ 184 ] ) are @xmath253@xmath254 as the expressions for @xmath42 and @xmath43 we have @xmath255 , \label{187}\]]@xmath256 , \label{188}\ ] ] where @xmath257 was absorbed by redefining the origin of time . the corresponding @xmath46 and @xmath47 obtained from ( [ 113.7 ] ) and ( [ 113.8 ] ) are @xmath258 \nonumber\\ & \label{190}\\ & -\frac{\theta a_{c_{0}}}{\sqrt{\left\vert \beta\right\vert } } e^{-\sqrt{\left\vert \beta\right\vert } t}\sinh\left [ \frac{\alpha \sqrt{\left\vert \beta\right\vert } } { 4a_{c_{0}}^{2}}\left ( 1-e^{2\sqrt { \left\vert \beta\right\vert } t}\right ) + \theta t\right ] , \nonumber\end{aligned}\]]@xmath259 \nonumber\\ & \label{191}\\ & -\frac{\theta a_{c_{0}}}{\sqrt{\left\vert \beta\right\vert } } e^{-\sqrt{\left\vert \beta\right\vert } t}\cosh\left [ \frac{\alpha \sqrt{\left\vert \beta\right\vert } } { 4a_{c_{0}}^{2}}\left ( 1-e^{2\sqrt { \left\vert \beta\right\vert } t}\right ) + \theta t\right ] .\nonumber\end{aligned}\ ] ] from ( [ 185 ] ) we can see that the physical meaning of the approximation assumed is that of early times . figure @xmath260 depicts the scale factors @xmath46 and @xmath42 in an interval where the approximation proposed is accurate . depending on the values of @xmath261 and @xmath220 , the deviation in the behavior of them can be very large . the situation here is more or less similar to that of case 1 of section 6 . we shall therefore not enter the details . let us consider , for example , the case where @xmath262 for any instant of time @xmath263 where the approximations ( [ 187])-([191 ] ) are valid , we can see that to each universe in the nc - frame there correspond two universes in the c - frame : one with @xmath220 large and the other with @xmath220 small . the graph of @xmath264/2a_{c_{0}}$ ] [ which in ( [ 190 ] ) has a role similar to that of @xmath223 in case 1 , section 6 ] is identical in shape to that of fig . [ tbh ] fc5.eps in this work , we carried out an investigation into the role of noncommutative geometry in the cosmological scenario by introducing a noncommutative deformation in the algebra of the minisuperspace variables along the same lines proposed in @xcite and followed in @xcite . as a cosmological model to carry out such an investigation , we chose a friedman - robertson - walker universe with conformally coupled scalar field . a parallel was drawn between the realizations of noncommutativity in two possible frameworks for physical interpretation : the c - frame , where it manifest through @xmath0-dependent terms the hamiltonian , and in the nc - frame , where it is manifest directly in the universe degrees of freedom . the influence of noncommutativity in the universe evolution and its capability to remove cosmological singularities was investigated by means of a comparative study of the frw model in four different versions : classical commutative , classical noncommutative , quantum commutative and quantum noncommutative . the confrontation between the classical and quantum versions was rendered easy by the bohmian interpretation of quantum theory , which provided a common language for comparison through the quantum trajectory formalism . an extension of the bohmian formulation to comprise noncommutative effects was previously proposed for the kantowski - sachs model in @xcite . in our comparative study we have dealt with the noncommutative quantum model along the same lines . the beable mapping commonly employed in bohmian quantum mechanics was extended to noncommutative quantum cosmology . in the commutative context , our formulation is reduced to the one proposed by holland @xcite in the minisuperspace approximation . in the classical context , the main result of our investigation is that , contrary to the noncommutative kantowski - sachs model , for @xmath0 sufficiently large the noncommutative frw can be non - singular . when @xmath265 , noncommutativity can give rise to bouncing universes in the nc- and c - frame realizations . the @xmath59 case is of particular interest since it reveals the capability noncommutativity has to mimic quantum effects under special conditions . the bouncing solutions that appear in both nc- and c - frame realizations [ eqs . ( [ 29 ] ) and ( [ 30 ] ) ] can be mapped into the commutative quantum solution ( [ 78 ] ) with an appropriate identification of the integration constants . therefore noncommutative classical quantum universe can be indistinguishable from a commutative quantum universe . similar correspondences involving the noncommutative classical universe and other bouncing solutions of noncommutative quantum cosmology were described in section 6 . while in the classical context non - singular universe solutions can exist only in the noncommutative universe model and for @xmath266 , in the quantum context one may find non - singular universe solutions even in the commutative case . the qualitative behavior of the universe solutions in noncommutative quantum cosmology was discussed in section 4 , where examples were presented that contain non - singular periodic solutions , non - singular solutions presenting one bounce , as well as non - singular solutions containing an infinite sequence of bounces enveloped by a larger bounce . when noncommutativity effects are turned on in the quantum scenario , they give rise to dynamical universes in situations where bohmian quantum cosmology admits only static universes . one example is that of real wavefunctions , which in the noncommutative frw model with a conformally coupled scalar field represent always non - singular bouncing universes . noncommutativity effects can also induce a dynamical behavior yielding non - singular periodic universes in cases where the commutative counterpart is static and the noncommutative wavefunction is complex . an example was presented in section 6 . the investigation into the nc - c - frame correspondence revealed that , even in the classical context , the description of the universe evolution provided by these two possible scenarios for the realization of noncommutativity can differ radically . in section 3 we showed that for some values of the integration constants the universe can be non - singular in the nc - frame , while its c - frame counterpart is singular . an example of the drastic difference that can occur between the nc- and c - frame realizations was exhibited in that section for the case where @xmath59 . it consists of a universe that is non - singular in the nc - frame and that has no correspondent in the c - frame , where it is at all times singular . in the quantum context the distinction between the nc- and c - frame descriptions was shown to be as relevant as it is in the classical one . an example was worked out that discusses how a universe with large @xmath267 in the nc - frame can correspond to two universes in the c - frame , one with large @xmath268 @xmath267 , and other with very small @xmath269 . such an interplay between small and large scale distances was previously reported in @xcite , where it was interpreted as a sort of ir - uv mixing , in analogy with noncommutative field theory . other examples involving the nc - c - frame correspondence were presented and solved numerically for the case @xmath191 and analytically for the case @xmath73 . again the case where @xmath59 provided an interesting example . when @xmath59 noncommutativity effects act to drop the kinetic term from the wheeler - dewitt equation . this justifies the poor and highly constrained dynamics found in this case . no matter what is the wavefunction in this case , if the initial conditions are non - singular the universe is non - singular and experiments a single bounce in both nc- and c - frame descriptions ( fig . @xmath270 ) . in case noncommutativity of the minisuperspace variables has in fact played a role in the evolution of the primordial universe , as proposed in @xcite , the study carried out in this work renders evident the need of an ontology in order to have a clear picture of the essential features of the noncommutative universe models . the correspondence between degrees of freedom in two different frames of realization is not sufficient to define the theory completely , which is only fixed by assuming one of them as the physical frame . this necessity seems not to be an exclusive feature of the cosmological model considered here , where the dramatic difference in the universe evolution can be attributed , in part , to the fact that the noncommutativity in question is that of the system s degrees of freedom - the minisuperspace variables- . in the models where noncommutativity does not involve directly the system s degrees of freedom , as the canonical noncommutative field theories that come from string theory @xcite , the study of the correspondence between the nc- and c - frame descriptions is also a relevant subject . in the context of gauge theories , where the connection between the nc- and c- frames is via the seiberg - witten map , an investigation into the properties of the theory that have resemblance with gravity was carried out , e.g. , in @xcite , where the equivalence between spacetime translations and gauge transformations is shown to occur in the nc - frame . in the c - frame , on the other hand , where such an equivalence seems to be lost , noncommutative fields can be interpreted as ordinary theories immersed in a gravitational background generated by the gauge field , as shown in the interesting work by rivelles @xcite , and further in @xcite . the author acknowledges nelson pinto neto and jos abdalla helayel - neto for corrections in earlier versions of this manuscript d. drr , s. goldstein , n. zangh , bohmian mechanics and the meaning of the wave function , experimental metaphysics - quantum mechanical studies in honor of abner shimony , ed . r.s.cohen , m. horne , and j. stachel , boston studies in the philosophy of science ( kluwer , 1996 ) [ quant - ph/9512031 ] . d. drr , s. goldstein and n. zangh , bohmian mechanics as the foundation of quantum mechanics , contribution to bohmian mechanics and quantum theory : an appraisal , edited by j.t . cushing , a. fine , s. goldstein , kluwer academic press .

randomly branching polymers with _ annealed _ connectivity are model systems for ring polymers and chromosomes . in this context , the branched structure represents transient folding induced by topological constraints . here we present computer simulations of melts of annealed randomly branching polymers of segments in and dimensions . in all cases , we perform a detailed analysis of the observed tree connectivities and spatial conformations . our results are in excellent agreement with an asymptotic scaling of the average tree size of , suggesting that the trees behave as compact , _ territorial _ fractals . the observed swelling relative to the size of ideal trees , , demonstrates that excluded volume interactions are only partially screened in melts of annealed trees . overall , our results are in good qualitative agreement with the predictions of flory theory . in particular , we find that the trees swell by the combination of modified branching and path stretching . however , the former effect is subdominant and difficult to detect in dimensions .

1610.03100

randomly branched polymers or trees display surprisingly rich physics . in statistical mechanics , lattice trees are believed to fall into the same universality class as lattice animals @xcite and their critical exponents are related to those of magnetic systems @xcite . in polymer chemistry , the deliberate ( or accidental @xcite ) incorporation of monomers with higher functionality into the polymerisation processes modifies materials properties @xcite . in this context , one has to distinguish the environmental conditions under which chains are studied from those under which they are synthesised and where their connectivity is said to be _ quenched_. here we are interested in randomly branched polymers with _ annealed _ connectivity , whose structure is meant to represent the transient folding of topologically constrained ring polymers @xcite ( fig . [ fig : rubinsteinelk ] ) and chromosomes @xcite . at the light of the recent results by lang @xcite and smrek and grosberg @xcite who analysed the threadable fraction of the minimal area encircled by non - concatenated ring polymers in melt , it is a non - trivial and still open question , if @xcite or to which extent @xcite this analogy @xcite holds also for these systems . however , having shown that it provides at least an excellent approximation @xcite , we now proceed to analyse in some detail the statistical properties of melts of annealed trees . behaviour of a two - dimensional ring ( orange ) in an array of impenetrable fixed obstacles ( schematically represented by the black dots ) . to maximize its conformational entropy , the ring adopts a transiently branched shape ( a ) , which can then be mapped to a lattice tree ( b ) with _ annealed _ connectivity . the figure is adapted from @xcite . , scaledwidth=22.0% ] as customary in polymer physics @xcite , we are primarily interested in exponents describing how expectation values for observables scale with the weight , @xmath5 , of the trees : @xmath6 or the path distance @xmath7 between tree nodes : @xmath8 here , @xmath9 denotes the average branch weight ; @xmath10 the average contour distance or length of paths on the tree ; @xmath11 the mean - square gyration radius of the trees ; and @xmath12 and @xmath13 the mean - square spatial distance and contact probability of nodes as a function of their contour distance , @xmath7 . for ideal , non - interacting trees @xcite @xmath14 , @xmath15 and @xmath16 . for interacting systems , the only exactly known exponent is @xmath17 for self - avoiding trees in @xmath2 dimensions @xcite . flory theory provides a simple and insightful description of a wide range of interacting tree systems @xcite , but the results are obtained through uncontrolled approximations and rely on the cancellation of large errors @xcite . for the present problem , flory theory predicts that trees in a melt ought to behave as compact fractals with @xmath18 and that trees should swell by a combination of path swelling and modified branching . as plausible as this prediction might be , it needs to be corroborated by more rigorous approaches . here we present the , to our knowledge , first computational study of the properties of melts of annealed trees in @xmath1 and @xmath2 dimensions . the article is part of a series @xcite , where we use a combination of computer simulations , flory theory and scaling arguments to investigate the connectivity and conformational statistics of randomly branched polymers with excluded volume interactions . we employ the same notation , definitions and numerical methodologies introduced in our previous work @xcite on single self - avoiding trees in good solvent , which we briefly summarise in sections [ sec : theory ] and [ sec : modmethods ] . results for trees connectivity and spatial conformations are outlined in sec . [ sec : results ] and discussed in detail in sec . [ sec : discussion ] . finally , conclusions are sketched in sec . [ sec : concls ] . we are interested in randomly branched polymers with annealed connectivity and repulsive , short - range interactions between monomers . [ sec : modelunits ] and [ sec : observables ] summarize our choices of the employed lattice model , units and notation and the definition of observables . [ sec : flory ] reviews predictions of flory theory for interacting trees . finally , secs . [ sec : blobsize1 ] and [ sec : blobsize2 ] provide a justification for the model parameters and the sizes of simulated trees by employing the concept of `` blob size '' . all our numerical results are obtained for trees embedded in @xmath1 and @xmath2 dimensions , even though many theoretical expressions are conveniently expressed for general @xmath19 . we study lattice trees on the @xmath20- and @xmath21-cubic lattice . the functionality of the nodes is restricted to the values @xmath22 ( a leaf or branch tip ) , @xmath23 ( linear chain section ) , and @xmath24 ( branch point ) . connected nodes occupy adjacent lattice sites . a tree conformation , @xmath25 , can be described by the set of node positions , @xmath26 , in the embedding space and a suitable representation of its connectivity graph , @xmath27 . we employ a data structure in the form of a linked list , which retains for each node , @xmath28 , its position , @xmath29 , functionality , @xmath30 , and the indices @xmath31 of the nodes to which it is connected . since our models do not include a bending energy , the lattice constant equals the kuhn length , @xmath32 , of linear paths across ideal trees . we measure energy in units of @xmath33 , length in units of the lattice constant or kuhn length , @xmath32 , and mass in units of the number of kuhn segments . similarly , we specify the density by the kuhn segment number density , @xmath34 . we use the letters @xmath5 and @xmath35 to denote the mass of a tree or a branch , respectively . with @xmath5 kuhn segments connecting the nodes of a tree , there are @xmath36 nodes in a tree . the symbols @xmath37 and @xmath7 are reserved for contour lengths of _ linear _ paths on the tree , while @xmath38 and @xmath39 denote contour distances from a fixed point , typically the tree center . spatial distances are denoted by the letters @xmath40 and @xmath41 . examples are the tree gyration radius , @xmath42 , spatial distances between nodes , @xmath43 , and the spatial distances , @xmath44 , of a node from the tree center of mass . for ideal trees , nodes do not interact and their asymptotic branching probability , @xmath45 , is controlled via a chemical potential for branch points , @xmath46 where @xmath47 is the total number of 3-functional nodes in the tree . all our results are obtained @xcite for a value of @xmath48 . interactions between nodes are accounted for via @xmath49 where @xmath50 is the total number of kuhn segments inside the elementary cell centered at the lattice site @xmath28 . in all cases we employ the same , large free energy penalty of @xmath51 for overlapping pairs of kuhn segments , eq . ( [ eq : hvolumeinteractions ] ) . the pair repulsion is so strong , that single trees are effectively self - avoiding , while the local occupancy fluctuations in our melts is @xmath52 _ for all densities_. a tree is a branched structure free of loops . its connectivity can be characterized in a number of ways . locally , nodes connecting kuhn segments differ according to their functionality , @xmath53 , with branch points having functionality @xmath54 and branch tips @xmath55 . given @xmath56 the total number of tree nodes with functionality @xmath53 , they satisfy the relations : @xmath57 for our choice of parameters , ideal trees are characterized by an asymptotic branching probability @xcite @xmath58 the large scale structure of a tree can be analyzed in terms of the ensemble of sub - trees generated by cutting bonds . removal of a bond splits a tree of weight @xmath5 into two trees of weight @xmath35 and @xmath59 . defining @xmath60 as the branch weight , the corresponding ensemble average grows as a characteristic power of the tree weight , @xmath61 ( eq . ( [ eq : epsilon ] ) ) . alternatively , the tree connectivity can be analyzed in terms of : ( 1 ) the statistics of minimal distances , @xmath62 , of two nodes @xmath63 along linear paths on the tree , and its corresponding ensemble average @xmath64 ; ( 2 ) the average distance , @xmath65 of nodes from the central node ; ( 3 ) the average length @xmath66 of the longest distance from the central node . for ensemble averages one expects ( eq . ( [ eq : rho ] ) ) : @xmath67 with @xcite @xmath68 . similarly , we characterize the statistics of branches by measuring the average branch weight , @xmath69 , as a function of the _ longest _ contour distance of nodes from the branch root , @xmath70 . finally , we consider the average weight of the `` core '' of tree , @xmath71 , made of segments whose distance from the central node does not exceed @xmath72 . the overall spatial extension of the tree is best described through the mean - square gyration radius @xmath73 which is the average square distance of a node from the tree center of mass . asymptotically , it is expected to scale as @xmath74 ( eq . ( [ eq : nu ] ) ) . the spatial conformations of linear paths of length @xmath7 on trees of total mass @xmath5 can be characterized using standard observables for linear polymers , namely : ( 1 ) the mean - square end - to - end distance @xmath75 ; ( 2 ) for a given contact distance @xmath76 , the corresponding end - to - end closure probability @xmath77 . they are expected to scale , respectively , as ( eqs . ( [ eq : nu_path ] ) and ( [ eq : theta_path ] ) ) : @xmath78 and @xmath79 and to be asymptotically independent of tree weight . by construction @xcite , @xmath80 . flory theories @xcite are formulated as a balance of an entropic elastic term and an interaction energy : @xmath81 the central element of the flory theory of interacting trees is the elastic free energy of ideal annealed trees @xcite , @xmath82 ensembles of quenched trees are characterised by fixed values of @xmath83 and the flory energy needs to be minimized over @xmath40 . linear chains represent a special case , where @xmath84 . in annealed trees , interactions can modify the branching statistics as well as the spatial conformations . with larger contour distances leading to larger spatial distances between repelling monomers , the flory energy needs to be simultaneously minimized over both variables , @xmath40 and @xmath37 . optimising @xmath37 for a _ given _ size , @xmath85 , yields @xmath86 and @xmath87 thus independently of the physical origin of the effect , swollen trees with @xmath88 are predicted to display both , modified connectivities with @xmath89 and path swelling with @xmath90 . in melts , volume interactions are screened @xcite and dominated by high - order collisions of dense systems . inspecting all terms of order @xmath91 in a ( standard ) virial - type expansion of the interaction energy term in eq . ( [ eq : fflory ] ) @xcite : @xmath92 shows that interactions are estimated to be irrelevant , if @xmath93 . even without swelling , this is the case in @xmath94 dimensions , where @xmath95 suggests ideal tree behavior . in @xmath96 dimensions , for @xmath97 the series is dominated by the @xmath98 limit . minimising the sum of eqs . ( [ eq : fgutin ] ) and ( [ eq : inttermmelt ] ) in this limit with respect to @xmath37 and @xmath40 yields : @xmath99 eqs . ( [ eq : nu_of_d_melt ] ) to ( [ eq : rho_of_d_melt ] ) predict that in the melt state annealed trees are compact fractals . interestingly , @xmath100 has the same value as the critical exponent @xmath101 for linear self - avoiding walks , suggesting a deeper analogy between the two problems @xcite . assuming that entropic effects are too small to induce density inhomogeneities , we can estimate the asymptotic behaviour from the assumption @xcite that the asymptotic segment self density at the tree center of mass converges to the melt segment density @xmath34 . more precisely @xcite , @xmath102 , where @xmath103 is the gyration or shape tensor , @xmath104 and @xmath105 is the spatial position of the tree centre of mass . neglecting asphericity , @xmath106 while our simulation results ( fig . [ fig : rg2aspectratios]b ) suggest in @xmath2 with @xmath107 a slightly larger value . by using the latter quantity , the blob size @xcite @xmath108 where we expect the crossover from the ideal to the asymptotic regime to occur is implicitly defined via @xmath109 where @xcite @xmath110 . corresponding arguments in @xmath1 dimensions yield @xmath111 with a blob size @xmath112 our original simulation of tree melts in @xmath2 dimensions reported in ref . @xcite were carried out at a segment density of @xmath113 , which was imposed by the mapping to the corresponding ring polymer problem . in this case , the estimated blob size of @xmath114 equals the size , @xmath115 , of the largest trees we were able to simulate . below we report results for measured gyration radii _ etc . _ for these systems , but we exclude them from the estimation of asymptotic exponents . instead , we rely in @xmath2 dimensions on data obtained for a smaller density of @xmath116 . with @xmath117 we safely expect to have reached the asymptotic regime . similarly , in @xmath1 dimensions for @xmath118 , we get @xmath119 . to simulate melts of annealed lattice trees we have used a variant of the `` amoeba '' monte carlo algorithm by seitz and klein @xcite ( sec . [ sec : amoebaalgo ] ) . the quantitative analysis of tree connectivities , tree spatial conformations and the estimation of critical exponents defined in eqs . ( [ eq : epsilon])-([eq : nu_path ] ) has been carried out by applying the `` burning algorithm '' @xcite ( sec . [ sec : treesanalysis ] ) and ordinary fitting procedures ( sec . [ sec : extractexps ] ) . tabulated values and other details on the derivation of critical exponents for single - tree statistics are also reported in the supplemental material @xcite . parametric plots of the mc - time ( @xmath120 ) evolution of the ensemble - average square gyration radius , @xmath121 , _ vs. _ the mean - square displacement of the tree center of mass , @xmath122 . non - equilibrated ( resp . , equilibrated ) values of the plots correspond to regions above ( resp . , below ) the black solid line @xmath123 . different colors correspond to different tree masses ranging from ( bottom symbols ) @xmath124 to ( top symbols ) @xmath125 ( melts for @xmath126 ) and @xmath127 ( ideal trees and melts for @xmath128),scaledwidth=50.0% ] . amoeba trial moves simultaneously modify the tree connectivity , @xmath27 , and the tree conformation , @xmath129 . they are constructed by randomly cutting a leaf ( or node with functionality @xmath22 ) from the tree and placing it on a randomly chosen site adjacent to a randomly chosen node with functionality @xmath130 , to which the leave is then connected . trial moves , @xmath131 , are accepted with probability : @xmath132 where @xmath133 is the total number of 1-functional nodes in the initial / final state and @xmath134 , eqs . ( [ eq : hideal ] ) and ( [ eq : hvolumeinteractions ] ) . it should be noted , that our version of the amoeba algorithm is slightly modified with respect to the original one of ref . @xcite as we impose @xcite node functionalities @xmath135 . similar algorithms displacing entire branches are more efficient for single trees @xcite , but are likely to encounter difficulties when generalized to the dense systems we are mostly interested in . in contrast , the small non - local mass transport of the amoeba algorithm is not obstructed by the volume interactions , since it falls into the range of the natural occupancy fluctuations in our tree melts . our simulations start from linearly connected random walks as initial states . the total computational effort for equilibrating the systems as a function of the corresponding systems sizes is summarised in table si @xcite . as illustrated by fig . [ fig : mc_equilibration ] , the tree gyration radii equilibrate over a time scale during which the tree centers of mass diffuse over the corresponding distance . quite curiously , the performance of the `` amoeba '' algorithm as a function of the monte carlo time steps is non - monotonic in the system density @xmath34 , see fig . s1 @xcite . in fact , while @xmath21 self - avoiding trees ( @xmath136 , studied in our previous work @xcite ) and @xmath21 melts for @xmath137 take roughly the same time to reach equilibrium , @xmath20 and @xmath21 tree melts at density @xmath138 require simulations which are @xmath139 times longer . in particular , this is the reason why we have only been able to reach tree weights up to @xmath140 compared to @xmath115 in ref . @xcite . at the moment we have no explanation accounting for this , apparently counterintuitive , behavior . as in our previous work @xcite , we have analysed tree connectivities using a variant of the `` burning '' algorithm for percolation clusters @xcite . the algorithm is very simple , and consists of two parts . in the initial inward ( or burning ) pass branch tips are iteratively `` burned '' until the tree center is reached . in the subsequent outward pass one advances from the center towards the periphery . the inward pass provides information about the mass and shape of branches . the outward pass allows to reconstruct the distance of nodes from the tree center . by employing a data structure in the form of a linked list , which retains for each node , @xmath28 , its position , @xmath29 , functionality , @xmath30 , and the indices @xmath31 of the nodes to which it is connected , each step of the burning algorithm consists in removing from the list all sites with functionality @xmath141 ( tips ) and updating the functionalities and the indices of the remaining nodes accordingly . the algorithm stops when only one node remains in the list . in order to find the minimal path length @xmath7 between any pair of tree nodes @xmath28 and @xmath142 , we have modified the algorithm by requiring that sites @xmath28 and @xmath142 are not removed from the list . accordingly , the algorithm stops when nodes @xmath28 and @xmath142 are the only tips left of the `` burned '' tree . by using the remaining linked list it is then trivial to find the corresponding path length @xmath7 . in order to get reliable estimates of critical exponents `` @xmath143 '' in the large-@xmath5 limit and of the corresponding errors , we stay close to the procedure developed by janse van rensburg and madras @xcite and employed in our previous work @xcite , and combine the results obtained from fitting the @xmath5-dependent data to two functional forms . for the specific example of the exponent @xmath101 , they are given by the following expressions : 1 . a simple power - law behavior with @xmath144 ( @xmath145 , @xmath101 ) fit parameters : @xmath146 and 2 . a power - law behavior with a correction - to - scaling term ( @xmath147 ) with @xmath148 ( @xmath145 , @xmath149 , @xmath150 , @xmath101 ) fit parameters : @xmath151 here , we have carried out a one - dimensional search for the value of @xmath152 for which the fit yields a vanishing @xmath153 term . for the other exponents , we have employed analogous expressions . ( [ eq : fitrg22 ] ) has been used on data with @xmath154 and @xmath155 for @xmath156 ( @xmath157 ) and @xmath128 , respectively . ( [ eq : fitrg21lin2 ] ) has been employed on the whole range with @xmath158 . best fits are obtained by minimizing @xcite @xmath159 ^ 2 $ ] , where @xmath160 is the number of data points used in the fit procedure . quality of the fit is estimated by the normalized @xmath161 , where @xmath53 is the number of fit parameters . when @xmath162 the fit is deemed to be reliable @xcite . the corresponding @xmath163-values provide a quantitative indicator for the likelihood that @xmath164 should exceed the observed value , if the model were correct @xcite . all fit results are reported together with the corresponding errors , @xmath165 and @xmath166 values . final estimates of critical exponents are calculated as averages of all independent measurements . corresponding uncertainties are given in the form @xmath167(statistical error)@xmath167(systematic error ) , where the `` statistical error '' is the largest value obtained from the different fits @xcite while the `` systematic error '' is the spread between the single estimates , respectively . in those cases where eq . ( [ eq : fitrg21lin2 ] ) fails producing trustable results we have retained only the @xmath144-parameter fit , eq . ( [ eq : fitrg22 ] ) , and a separate analysis of uncertainties was required , see the caption of table svii @xcite for details . error bars reported in table [ tab : expsummary ] are given by @xmath168 . in the following sections , we discuss the structure of trees in @xmath20 and @xmath21 melts by considering the scaling behaviors of the observables defined in sec . [ sec : observables ] and their corresponding critical exponents . similarly to the general outline of our previous work @xcite on single self - avoiding trees in good solvent , it is particularly instructive to compare the properties of interacting trees to the ones for ideal trees . for this reason , but also for keeping this work autonomous and self - consistent in its own , all figures and tables of this work , including the supplementary ones @xcite , contain the same data for ideal trees originally reported in @xcite . branching statistics . average fraction of @xmath169-functional nodes , @xmath170 , as a function of the total number of tree nodes , @xmath36 . the asymptotic branching probability @xcite @xmath45 is @xmath171 . , scaledwidth=50.0% ] our results for the average number of branch points , @xmath172 , as a function of @xmath5 are listed in tables sii - iii @xcite . figure [ fig : branchnumber ] shows that the ratios of 3-functional nodes , @xmath173 , reach their asymptotic value already for moderate tree weights . interestingly , our results for ideal trees as well as interacting trees perfectly agree to each other with asymptotic branching probability @xcite @xmath174 . in fact , due to the multiple occupation of lattice sites in the melt , the branching probabilities remain virtually unchanged in spite of the interactions . corresponding distributions @xmath175 are well described by gaussian statistics with corresponding variances increasing linearly with @xmath5 , see fig . s2 @xcite . path length statistics . mean contour distance between pairs of nodes , @xmath10 . straight lines correspond to the large-@xmath5 behaviour @xmath176 with critical exponents @xmath177 given by the best estimates summarised in table [ tab : expsummary ] . , scaledwidth=50.0% ] our results for ( a ) the mean contour distance between pairs of nodes , @xmath10 , ( b ) the mean contour distance of nodes from the central node , @xmath65 , and ( c ) the mean _ longest _ contour distance of nodes from the central node , @xmath66 are summarized in tables sii - iii @xcite and plotted in fig . [ fig : pathlengthstat_l ] and fig . s3 @xcite . as discussed in sec . [ sec : observables ] , the three quantities are expected to scale with the total tree weight @xmath5 as @xmath178 . extracted single values for @xmath177 s including more details on their statistical significance ( @xmath164 and @xmath166-values ) are summarized in table siv @xcite . our final best estimates for @xmath177 s ( straight lines in fig . [ fig : pathlengthstat_l ] and fig . s3 @xcite , and tables [ tab : expsummary ] and siv @xcite ) are obtained by combining the corresponding results for @xmath65 , @xmath66 and @xmath10 . the relation between branch weight and path length can also be explored on the level of branches . we have analyzed the scaling behavior of : ( 1 ) the average branch weight , @xmath179 , as a function of the longest contour distance to the branch root , @xmath180 , and ( 2 ) the average branch ( or tree core ) weight , @xmath181 , inside a contour distance @xmath72 from the central node of the tree . corresponding results are shown in fig . [ fig : avmass_vs_lbranch ] . both data sets show universal behavior at intermediate @xmath180 and @xmath72 , and saturate to the corresponding expected limiting values @xmath182 and @xmath183 . for large @xmath180 ( respectively , @xmath72 ) the relation @xmath184 ( resp . , @xmath185 ) is expected to hold . for @xmath186 ( and with an analogous expression for @xmath187 , we have estimated @xmath177 as @xmath188 . numerical results are reported in the corresponding insets of fig . [ fig : avmass_vs_lbranch ] , the large - scale behaviour agreeing well with the best estimates for @xmath177 s ( horizontal lines ) summarized in table [ tab : expsummary ] . branch weight statistics . average branch weight , @xmath189 as a function of the total tree mass , @xmath5 . straight lines correspond to the large-@xmath5 behaviour @xmath190 with critical exponents @xmath191 given by the best estimates summarised in table [ tab : expsummary ] . , scaledwidth=50.0% ] the scaling behavior of the average branch weight , @xmath190 , defines the critical exponent @xmath191 . single values of @xmath189 for each @xmath5 ( see tables sii - iii @xcite ) are plotted in fig . [ fig : nbranch ] , where the straight lines have slopes corresponding to our best estimates for @xmath191 s ( table [ tab : expsummary ] ) , see also table siv @xcite for details . we notice , in particular , that the scaling relation @xmath192 holds within error bars . in order to extract the critical exponent @xmath100 which defines the scaling behavior @xmath193 , we have selected paths of length close to the average length ( @xmath194 ) and to the trees maximal length ( @xmath195 ) and calculated corresponding mean - square end - to - end distances @xmath196 and @xmath197 ( see fig . [ fig : r2maxlmax ] and corresponding tabulated values in tables sv - vi @xcite . combination of the two ( table svii @xcite ) led to our best estimates for @xmath100 in the different ensembles summarized in table [ tab : expsummary ] . not surprisingly , these values agree well with the differential exponents @xmath198 reported in the l.h.s insets of fig . [ fig : pathr2_contactfreq ] . then , we have calculated the mean closure probabilities , @xmath199 ( fig . [ fig : pathr2_contactfreq ] , right - hand panels ) , normalised to the corresponding `` mean - field '' expectation values @xmath200 . as in the case of single self - avoiding trees in good solvent @xcite , @xmath199 for interacting trees markedly deviate from the mean - field prediction , which defines a _ novel _ critical exponent @xmath201 , @xmath202 . estimated values for @xmath201 s are reported in table [ tab : expsummary ] . conformational statistics of trees . ( a ) mean square gyration radius , @xmath203 . straight lines correspond to the large-@xmath5 behaviour @xmath204 with critical exponents @xmath101 given by the best estimates summarised in table [ tab : expsummary ] . ( b ) tree average aspect ratios , @xmath205 and @xmath206 . , scaledwidth=50.0% ] finally , we have measured the mean - square gyration radius , @xmath11 ( tabulated values in tables sv - vi @xcite ) , and the average shape of trees as a function of tree weight , @xmath5 ( fig . [ fig : rg2aspectratios ] , panels a and b respectively ) . estimated values of critical exponents @xmath101 ( straight lines in fig . [ fig : rg2aspectratios]a ) are summarized in table [ tab : expsummary ] , while details about their derivation are given in table svii @xcite . with + & & & & & & & annealed connectivity + & relation to & theory & simul . & theory & simul . , @xmath207 & theory & simul . , @xmath208 & & & simul . , @xmath137 + & other exponents & & & & & & & & & + & & & & & & & & & & + @xmath177 & & @xmath209 & @xmath210 & @xmath211 & @xmath212 & @xmath213 & @xmath214 & & & @xmath215 + & & & & & & & & & & + @xmath191 & @xmath216 & @xmath209 & @xmath217 & @xmath211 & @xmath218 & @xmath213 & @xmath219 & & & @xmath220 + & & & & & & & & & & + @xmath100 & & @xmath209 & @xmath221 & @xmath222 & @xmath223 & @xmath224 & @xmath225 & & & @xmath226 + & & & & & & & & & & + @xmath101 & @xmath227 & @xmath228 & @xmath229 & @xmath209 & @xmath230 & @xmath231 & @xmath232 & & & @xmath233 + & & & & & & & & & & + @xmath201 & & & @xmath234 & & @xmath235 & & @xmath236 & & & @xmath237 + & & & & & & & & & & + we have analyzed the behavior of interacting trees with annealed connectivity in @xmath20 and @xmath21 melts in terms of a small set of exponents defined in the introduction and in sec . [ sec : observables ] . with weights of up to @xmath127 segments , our tree sizes are comparable to those of linear chains in similar studies @xcite . however , for many purposes the average contour distance , @xmath238 , between monomers provides a more suitable comparison . the reader should thus bear in mind , that extracted exponents are either ( i ) effective ( crossover ) exponents valid for the particular systems and system sizes we have studied or ( ii ) _ estimates _ of true , asymptotic exponents , which suffer from uncertainties related to the extrapolation to the asymptotic limit . these effects are particularly pronounced for the high density @xmath21 melts we studied in ref . for example , we initially reported @xcite the results of a simple power law fit including all data for tree sizes @xmath239 suggesting @xmath240 . in contrast , the present more refined analysis yields @xmath241 . if this result is acceptable as an effective exponent for tree sizes @xmath242 , it mostly illustrates that one can not extract the asymptotic behaviour assuming a _ single _ , small correction to scaling , if the available tree sizes barely reach the crossover ( sec . [ sec : blobsize2 ] ) . the discussion below focuses on results for lower density @xmath20 and @xmath21 melts , where the studied tree sizes are significantly larger than the estimated blob size . our results are summarised in table [ tab : expsummary ] . most of them are in quantitative or at least good qualitative agreement with the predictions from flory theory @xcite . first , this implies that the reported exponents @xmath100 for melts of trees fit to the predicted @xcite value @xmath243 which , intriguingly , corresponds to the flory @xcite exponent @xmath101 for self - avoiding _ linear _ chains in good solvent conditions . second , the results for the exponent @xmath101 are in agreement with the prediction @xmath244 and confirm that trees in melts behave as `` territorial '' polymers . this result is relatively simple to understand as @xmath244 represents the minimal amount of swelling compatible with steric packing requirements . thus not too much stock should be put in the fact that the flory value for @xmath101 turns out to be exact in the present case . more interestingly , the theory makes two testable , non - trivial predictions for the contributions of connectivity changes and path stretching to the overall swelling @xcite ( eqs . ( [ eq : rho_of_nu ] ) and ( [ eq : nupath_of_nu ] ) ) : firstly , both effects should only depend on the magnitude of the overall swelling , but not its physical origin and secondly path stretching is expected to be dominant as @xcite @xmath245 for weakly swollen trees . the first prediction is well borne out by the comparison of @xmath21 self avoiding trees and @xmath20 melt trees . in both cases , the trees are expected to swell to @xmath17 . interestingly , the observed values for @xmath177 and @xmath100 turn out to be almost identical and close to the predicted values . the second prediction is confirmed by noticing that trees in @xmath20 and @xmath21 melts swell almost exclusively at the path level ( fig . [ fig : r2maxlmax ] ) , while the observed modifications of the connectivities ( figs . [ fig : pathlengthstat_l ] and [ fig : nbranch ] ) , are very weak . in absolute terms for our largest trees with @xmath125 ( see tables sii - iii @xcite and sv - vi @xcite ) : @xmath246 , @xmath247 and @xmath248 while @xmath249 , @xmath250 and @xmath251 . for other quantities , flory theory is even qualitatively wrong . a particularly interesting case are the average contact probability @xmath13 between nodes at path distance @xmath7 . as shown in fig . [ fig : pathr2_contactfreq ] and table [ tab : expsummary ] , @xmath13 s for interacting trees deviate consistently from the nave mean - field estimate of @xmath252 . this is yet another illustration of the subtle cancellation of errors in flory arguments , which are built on the mean - field estimates of contact probabilities @xcite . motivated by the close analogy between non - concatenated ring polymers and annealed lattice trees @xcite , we have studied the statistical properties of tree melts in @xmath1 and @xmath2 dimensions . we have used the same methodology as in our recent work on self - avoiding trees @xcite , _ i.e. _ variants of the amoeba @xcite and burning @xcite algorithms for the monte carlo simulation and the connectivity analysis ( sec . [ sec : modmethods ] ) . table [ tab : expsummary ] summarises our estimates of the asymptotic values of the exponents describing the scaling behavior of the average branch weight , @xmath190 , the average path length , @xmath176 , the mean - square path extension , @xmath253 , and the tree and branch gyration radii , @xmath254 ( sec . [ sec : results ] ) . our results are in excellent agreement with an asymptotic scaling of the average tree size of @xmath3 , suggesting that the trees behave as compact , _ territorial _ @xcite fractals ( fig . [ fig : rg2aspectratios ] ) . moreover , we find that the trees swell by the combination of modified branching and path stretching . however , the former effect is subdominant and difficult to detect in @xmath2 dimensions . our results for dense systems contribute to the evidence suggesting that flory theory @xcite provides a useful framework for discussing the behavior of interacting trees . that flory theory should work for trees is not a foregone conclusion . in the case of linear chains , the approach is notorious ( and appreciated ) for the nearly perfect cancellation of large errors in the estimation of _ both _ terms in eq . ( [ eq : fflory ] ) @xcite . this delicate balance might well have been destroyed for trees , where the flory energy needs to be simultaneously minimized with respect to @xmath37 and @xmath40 . in two forthcoming publications , we will generalise flory theory to trees of finite size and extensibility @xcite and we will analyse the distribution functions for the quantities , whose _ mean _ behaviour we have explored above and in ref . @xcite in an attempt to go beyond flory theory @xcite . _ acknowledgements _ ar acknowledges grant prin 2010hxaw77 ( ministry of education , italy ) . re is grateful for the hospitality of the kavli institute for theoretical physics ( santa barbara , usa ) and support through the national science foundation under grant no . nsf phy11 - 25915 during his visit in 2011 , which motivated the present work . in particular , we have benefitted from long and stimulating exchanges with m. rubinstein and a. yu . grosberg on the theoretical background . this work was only possible thanks to generous grants of computer time by psmn ( ens - lyon ) and p2chpd ( ucb lyon 1 ) , in part through the equip@meso facilities of the flmsn . 10 j. isaacson and t. c. lubensky . flory exponents for generalized polymer problems . , 41:l469l471 , 1980 . w. a. seitz and d. j. klein . excluded volume effects for branched polymers . , 75:51905193 , 1981 . j. a. m. s. duarte and h. j. ruskin . the branching of real lattice trees as dilute polymers . 42:15851590 , 1981 . g. parisi and n. sourlas . critical behavior of branched polymers and the lee - yang edge singularity . , 46:871874 , 1981 . m. e. fisher . yang - lee edge singularity and @xmath255 field - theory . , 40:16101613 , 1978 . d. a. kurtze and m. e. fisher . yang - lee edge singularities at high - temperatures . , 20:27852796 , 1979 . a. bovier , j. frhlich , and u. glaus . . north - holland , amsterdam , k. osterwalder and r. stora ( eds . ) , 1984 . p. bacova , l. g. d. hawke , d. j. read , and a. j. moreno . dynamics of branched polymers : a combined study by molecular dynamics simulations and tube theory . , 46:46334650 , 2013 . m. rubinstein and r. h. colby . . oxford university press , new york , 2003 . w. burchard . solution properties of branched macromolecules . , 143:113 , 1999 . a. r. khokhlov and s. k. nechaev . polymer chain in an array of obstacles . , 112a:156160 , 1985 . m. rubinstein . dynamics of ring polymers in the presence of fixed obstacles . , 57:30233026 , 1986 . s. p. obukhov , m. rubinstein , and t. duke . dynamics of a ring polymer in a gel . , 73:12631266 , 1994 . m. kapnistos et al . unexpected power - law stress relaxation of entangled ring polymers . , 7:997 , 2008 . a. yu . grosberg . annealed lattice animal model and flory theory for the melt of non - concatenated rings : towards the physics of crumpling . , 10:560565 , 2014 . a. rosa and r. everaers . ring polymers in the melt state : the physics of crumpling . , 112:118302 , 2014 . a. rosa and r. everaers . computer simulations of randomly branching polymers : annealed vs. quenched branching structures . , 49:345001 , 2016 . a. grosberg , y. rabin , s. havlin , and a. neer . crumpled globule model of the three - dimensional structure of dna . , 23:373378 , 1993 . a. rosa and r. everaers . structure and dynamics of interphase chromosomes . , 4:e1000153 , 2008 . t. vettorel , a. y. grosberg , and k. kremer . statistics of polymer rings in the melt : a numerical simulation study . , 6:025013 , 2009 . leonid a. mirny . the fractal globule as a model of chromatin architecture in the cell . , 19:3751 , 2011 . m. lang . ring conformations in bidisperse blends of ring polymers . , 46:11581166 , 2013 . j. smrek and a. yu . minimal surfaces on unconcatenated polymer rings in melt . , 5:750754 , 2016 . m. e. cates and j. m. deutsch . conjectures on the statistics of ring polymers . , 47:21212128 , 1986 . t. ge , s. panyukov , and m. rubinstein . self - similar conformations and dynamics in entangled melts and solutions of nonconcatenated ring polymers . , 49:708722 , 2016 . g . de gennes . . cornell university press , ithaca , 1979 . m. doi and s. f. edwards . . oxford university press , new york , 1986 . a. yu . grosberg and a. r. khokhlov . . aip press , new york , 1994 . b. h. zimm and w. h. stockmayer . the dimensions of chain molecules containing branches and rings . , 17:13011314 , 1949 . m. daoud and j. f. joanny . conformation of branched polymers . , 42:13591371 , 1981 . a. m. gutin , a. yu . grosberg , and e. i. shakhnovich . polymers with annealed and quenched branchings belong to different universality classes . , 26:12931295 , 1993 . j. des cloizeaux and g. jannink . . oxford university press , oxford , 1989 . r. everaers , a. yu . grosberg , m. rubinstein , and a. rosa . flory theory of randomly branching polymers i : asymptotic behaviour . , 2016 . r. everaers . flory theory of randomly branching polymers ii : interacting trees of finite size and finite extensibility . , 2016 . a. rosa and r. everaers . beyond flory theory : distribution functions for interacting lattice trees . , 2016 . e. j. janse van rensburg and n. madras . a nonlocal monte carlo algorithm for lattice trees . , 25:303333 , 1992 . p. j. flory . . cornell university press , ithaca ( ny ) , 1953 . a. yu . grosberg and s. k. nechaev . from statistics of regular tree - like graphs to distribution function and gyration radius of branched polymers . , 48:345003 , 2015 . the result is consistent with the `` traditional '' picture @xcite that polymers in semi - dilute solutions can be represented as chains of blobs , where each blob contains @xmath256 monomers and has a linear size @xmath257 . since blobs are densily packed @xcite , the monomer density inside each blob equals the overall density or , neglecting prefactors , @xmath258 . for standard semi - dilute solutions of _ linear _ chains , @xmath259 and @xmath260 _ decreases _ with density @xcite . for our melts of trees , @xmath15 ( see fig . [ fig : rg2aspectratios]a ) and @xmath261 ( equivalent within numerical prefactors to eq . ( [ eq : blob3d ] ) ) which _ increases _ with density . this picture ought to break when density fluctuations ( see discussion at the end of sec . [ sec : modelunits ] ) become comparable to @xmath34 , or @xmath262 . h. j. heermann , d. c. hong , and h. e. stanley . backbone and elastic backbone of percolation clusters obtained by the new method of ` burning ' . , 17:l261l266 , 1984 . d. stauffer and a. aharony . . taylor & francis inc . , 1994 . see supplemental material at [ url ] for complementary figures and tables cited in the article . w. h. press , s. a. teukolsky , w. t. vetterling , and b. f. flannery . cambridge university press , cambridge , 2 edition , 1992 . h. meyer , j. p. wittmer , t. kreer , a. johner , and j. baschnagel . . , 132:184904 , 2010 . j. p. wittmer et al . scale - free static and dynamical correlations in melts of monodisperse and flory - distributed homopolymers . , 145:10171126 , 2011 . n. schulmann , h. meyer , j. p. wittmer , a. johner , and j. baschnagel . . , 45:16461651 , 2012 . p. j. flory . . interscience , new york , 1969 . the result is trivially obtained by expanding eqs . ( [ eq : rho_of_nu ] ) and ( [ eq : nupath_of_nu ] ) around the ideal values @xmath263 and @xmath264 . t. vettorel , a. y. grosberg , and k. kremer . territorial polymers . , 62:72 , 2009 . * supplemental material [ cols="^,^,^,^,^,^,^,^ " , ] & & & + & @xmath265 & @xmath40 & @xmath266 & @xmath267 & @xmath265 & @xmath40 & @xmath266 & @xmath267 & @xmath265 & @xmath40 & @xmath266 & @xmath267 + & @xmath268 & @xmath269 & @xmath270 & @xmath271 & @xmath272 & @xmath269 & @xmath270 & @xmath273 & @xmath274 & @xmath269 & @xmath275 & @xmath276 + 5 & @xmath277 & @xmath269 & @xmath270 & @xmath278 & @xmath272 & @xmath269 & @xmath270 & @xmath279 & @xmath280 & @xmath269 & @xmath275 & @xmath278 + 10 & @xmath281 & @xmath269 & @xmath270 & @xmath139 & @xmath282 & @xmath269 & @xmath270 & @xmath283 & @xmath284 & @xmath269 & @xmath275 & @xmath285 + 20 & @xmath281 & @xmath269 & @xmath270 & @xmath286 & @xmath287 & @xmath269 & @xmath270 & @xmath286 & @xmath288 & @xmath269 & @xmath275 & @xmath289 + 30 & @xmath268 & @xmath269 & @xmath290 & @xmath291 & @xmath292 & @xmath269 & @xmath290 & @xmath293 & @xmath272 & @xmath269 & @xmath270 & @xmath289 + 45 & @xmath294 & @xmath269 & @xmath295 & @xmath286 & @xmath272 & @xmath269 & @xmath295 & @xmath293 & @xmath272 & @xmath269 & @xmath296 & @xmath289 + 75 & @xmath297 & @xmath269 & @xmath298 & @xmath293 & @xmath272 & @xmath269 & @xmath298 & @xmath139 & @xmath272 & @xmath269 & @xmath299 & @xmath289 + 150 & @xmath300 & @xmath269 & @xmath301 & @xmath291 & @xmath287 & @xmath269 & @xmath301 & @xmath302 & @xmath272 & @xmath269 & @xmath303 & @xmath304 + 230 & @xmath287 & @xmath269 & @xmath305 & @xmath286 & @xmath306 & @xmath269 & @xmath305 & @xmath302 & @xmath272 & @xmath269 & @xmath307 & @xmath304 + 450 & @xmath294 & @xmath269 & @xmath308 & @xmath286 & @xmath272 & @xmath269 & @xmath309 & @xmath293 & @xmath272 & @xmath269 & @xmath310 & @xmath304 + 900 & @xmath311 & @xmath269 & @xmath312 & @xmath286 & @xmath313 & @xmath269 & @xmath314 & @xmath293 & @xmath268 & @xmath269 & @xmath315 & @xmath304 + 1800 & & & & & & & & & @xmath280 & @xmath269 & @xmath316 & @xmath304 + @xmath317 & @xmath318 & @xmath319 & @xmath320 + + + & @xmath321 & @xmath322 & @xmath323 & @xmath324 & @xmath325 + 5 & @xmath326 & @xmath327 & @xmath328 & @xmath329 & @xmath330 + 10 & @xmath331 & @xmath332 & @xmath333 & @xmath334 & @xmath335 + 20 & @xmath336 & @xmath337 & @xmath338 & @xmath339 & @xmath340 + 30 & @xmath341 & @xmath342 & @xmath343 & @xmath344 & @xmath345 + 45 & @xmath346 & @xmath347 & @xmath348 & @xmath349 & @xmath350 + 75 & @xmath351 & @xmath352 & @xmath353 & @xmath354 & @xmath355 + 150 & @xmath356 & @xmath357 & @xmath358 & @xmath359 & @xmath360 + 230 & @xmath361 & @xmath362 & @xmath363 & @xmath364 & @xmath365 + 450 & @xmath366 & @xmath367 & @xmath368 & @xmath369 & @xmath370 + 900 & @xmath371 & @xmath372 & @xmath373 & @xmath374 & @xmath375 + 1800 & @xmath376 & @xmath377 & @xmath378 & @xmath379 & @xmath380 + + + & @xmath381 & @xmath382 & @xmath383 & @xmath384 & @xmath385 + 5 & @xmath386 & @xmath327 & @xmath387 & @xmath329 & @xmath330 + 10 & @xmath388 & @xmath389 & @xmath390 & @xmath391 & @xmath392 + 20 & @xmath393 & @xmath394 & @xmath395 & @xmath396 & @xmath397 + 30 & @xmath398 & @xmath399 & @xmath400 & @xmath401 & @xmath402 + 45 & @xmath403 & @xmath404 & @xmath405 & @xmath406 & @xmath407 + 75 & @xmath408 & @xmath409 & @xmath410 & @xmath411 & @xmath412 + 150 & @xmath413 & @xmath414 & @xmath415 & @xmath416 & @xmath417 + 230 & @xmath418 & @xmath419 & @xmath420 & @xmath421 & @xmath422 + 450 & @xmath423 & @xmath424 & @xmath425 & @xmath426 & @xmath427 + 900 & @xmath428 & @xmath429 & @xmath430 & @xmath431 & @xmath432 + 1800 & @xmath433 & @xmath434 & @xmath435 & @xmath436 & @xmath437 + @xmath317 & @xmath318 & @xmath319 & @xmath320 + + + & @xmath438 & @xmath439 & @xmath440 & @xmath441 & @xmath442 + 5 & @xmath443 & @xmath327 & @xmath444 & @xmath445 & @xmath330 + 10 & @xmath446 & @xmath447 & @xmath448 & @xmath449 & @xmath450 + 20 & @xmath451 & @xmath452 & @xmath453 & @xmath454 & @xmath455 + 30 & @xmath456 & @xmath457 & @xmath458 & @xmath459 & @xmath460 + 45 & @xmath461 & @xmath462 & @xmath463 & @xmath464 & @xmath465 + 75 & @xmath466 & @xmath467 & @xmath468 & @xmath469 & @xmath470 + 150 & @xmath471 & @xmath472 & @xmath473 & @xmath474 & @xmath475 + 230 & @xmath476 & @xmath477 & @xmath478 & @xmath479 & @xmath480 + 450 & @xmath481 & @xmath482 & @xmath483 & @xmath484 & @xmath485 + 900 & @xmath486 & @xmath487 & @xmath488 & @xmath489 & @xmath490 + + + & @xmath491 & @xmath492 & @xmath493 & @xmath494 & @xmath495 + 5 & @xmath496 & @xmath327 & @xmath497 & @xmath445 & @xmath330 + 10 & @xmath498 & @xmath499 & @xmath500 & @xmath501 & @xmath502 + 20 & @xmath503 & @xmath504 & @xmath505 & @xmath506 & @xmath507 + 30 & @xmath508 & @xmath509 & @xmath510 & @xmath511 & @xmath512 + 45 & @xmath513 & @xmath514 & @xmath515 & @xmath516 & @xmath517 + 75 & @xmath518 & @xmath519 & @xmath520 & @xmath521 & @xmath522 + 150 & @xmath523 & @xmath524 & @xmath525 & @xmath526 & @xmath527 + 230 & @xmath528 & @xmath529 & @xmath530 & @xmath531 & @xmath532 + 450 & @xmath533 & @xmath534 & @xmath535 & @xmath536 & @xmath537 + 900 & @xmath538 & @xmath539 & @xmath540 & @xmath541 & @xmath542 + + + & @xmath543 & @xmath492 & @xmath544 & @xmath494 & @xmath495 + 5 & @xmath545 & @xmath546 & @xmath547 & @xmath548 & @xmath549 + 10 & @xmath550 & @xmath551 & @xmath552 & @xmath553 & @xmath554 + 20 & @xmath555 & @xmath556 & @xmath557 & @xmath558 & @xmath559 + 30 & @xmath560 & @xmath561 & @xmath562 & @xmath563 & @xmath564 + 45 & @xmath565 & @xmath566 & @xmath567 & @xmath568 & @xmath569 + 75 & @xmath570 & @xmath571 & @xmath572 & @xmath573 & @xmath574 + 150 & @xmath575 & @xmath576 & @xmath577 & @xmath578 & @xmath579 + 230 & @xmath580 & @xmath581 & @xmath582 & @xmath583 & @xmath584 + 450 & @xmath585 & @xmath586 & @xmath587 & @xmath588 & @xmath589 + 900 & @xmath590 & @xmath591 & @xmath592 & @xmath593 & @xmath594 + 1800 & @xmath595 & @xmath596 & @xmath597 & @xmath598 & @xmath599 + @xmath600 & @xmath601 & @xmath602 & @xmath603 + + + @xmath150 & @xmath604 & @xmath605 & @xmath606 & @xmath607 + @xmath165 & @xmath608 & @xmath609 & @xmath610 & @xmath611 + @xmath612 & @xmath613 & @xmath614 & @xmath615 & @xmath616 + & @xmath617 & @xmath618 & @xmath619 & @xmath620 + @xmath150 & @xmath621 & @xmath621 & @xmath621 & @xmath621 + @xmath165 & @xmath622 & @xmath623 & @xmath624 & @xmath625 + @xmath612 & @xmath626 & @xmath627 & @xmath627 & @xmath627 + & @xmath628 & @xmath629 & @xmath630 & @xmath631 + & & @xmath632 + + + @xmath150 & @xmath633 & @xmath634 & @xmath635 & @xmath636 + @xmath165 & @xmath637 & @xmath638 & @xmath639 & @xmath640 + @xmath612 & @xmath641 & @xmath642 & @xmath643 & @xmath644 + & @xmath645 & @xmath646 & @xmath647 & @xmath648 + @xmath150 & @xmath621 & @xmath621 & @xmath621 & @xmath621 + @xmath165 & @xmath642 & @xmath649 & @xmath650 & @xmath651 + @xmath612 & @xmath652 & @xmath653 & @xmath654 & @xmath627 + & @xmath655 & @xmath656 & @xmath655 & @xmath657 + & & @xmath658 + + + @xmath150 & @xmath659 & @xmath660 & @xmath661 & @xmath662 + @xmath165 & @xmath663 & @xmath664 & @xmath665 & @xmath666 + @xmath612 & @xmath667 & @xmath668 & @xmath626 & @xmath627 + & @xmath669 & @xmath670 & @xmath671 & @xmath672 + @xmath150 & @xmath621 & @xmath621 & @xmath621 & @xmath621 + @xmath165 & @xmath673 & @xmath674 & @xmath675 & @xmath676 + @xmath612 & @xmath677 & @xmath678 & @xmath679 & @xmath627 + & @xmath680 & @xmath681 & @xmath682 & @xmath683 + & & @xmath684 + + + @xmath150 & @xmath685 & @xmath686 & @xmath687 & @xmath688 + @xmath165 & @xmath689 & @xmath690 & @xmath691 & @xmath692 + @xmath612 & @xmath693 & @xmath694 & @xmath695 & @xmath696 + & @xmath697 & @xmath698 & @xmath699 & @xmath700 + @xmath150 & @xmath621 & @xmath621 & @xmath621 & @xmath621 + @xmath165 & @xmath701 & @xmath702 & @xmath703 & @xmath704 + @xmath612 & @xmath705 & @xmath706 & @xmath707 & @xmath707 + & @xmath708 & @xmath709 & @xmath710 & @xmath711 + & & @xmath712 + @xmath713 & @xmath714 & @xmath715 & @xmath716 + + + & @xmath717 & @xmath718 & @xmath719 & @xmath720 + 5 & @xmath721 & @xmath722 & @xmath723 & @xmath724 + 10 & @xmath725 & @xmath726 & @xmath727 & @xmath728 + 20 & @xmath729 & @xmath730 & @xmath731 & @xmath732 + 30 & @xmath733 & @xmath734 & @xmath735 & @xmath736 + 45 & @xmath737 & @xmath738 & @xmath739 & @xmath740 + 75 & @xmath741 & @xmath742 & @xmath743 & @xmath744 + 150 & @xmath745 & @xmath746 & @xmath747 & @xmath748 + 230 & @xmath749 & @xmath750 & @xmath751 & @xmath752 + 450 & @xmath753 & @xmath754 & @xmath755 & @xmath756 + 900 & @xmath757 & @xmath758 & @xmath759 & @xmath760 + 1800 & @xmath761 & @xmath762 & @xmath763 & @xmath764 + + + & @xmath717 & @xmath765 & @xmath766 & @xmath767 + 5 & @xmath768 & @xmath769 & @xmath770 & @xmath771 + 10 & @xmath772 & @xmath773 & @xmath774 & @xmath775 + 20 & @xmath776 & @xmath777 & @xmath778 & @xmath779 + 30 & @xmath780 & @xmath781 & @xmath782 & @xmath783 + 45 & @xmath784 & @xmath785 & @xmath786 & @xmath787 + 75 & @xmath788 & @xmath789 & @xmath790 & @xmath791 + 150 & @xmath792 & @xmath793 & @xmath794 & @xmath795 + 230 & @xmath796 & @xmath797 & @xmath798 & @xmath799 + 450 & @xmath800 & @xmath801 & @xmath802 & @xmath803 + 900 & @xmath804 & @xmath805 & @xmath806 & @xmath807 + 1800 & @xmath808 & @xmath809 & @xmath810 & @xmath811 + @xmath713 & @xmath714 & @xmath715 & @xmath716 + + + & @xmath717 & @xmath812 & @xmath813 & @xmath647 + 5 & @xmath814 & @xmath815 & @xmath816 & @xmath817 + 10 & @xmath818 & @xmath819 & @xmath820 & @xmath821 + 20 & @xmath822 & @xmath823 & @xmath824 & @xmath825 + 30 & @xmath826 & @xmath827 & @xmath828 & @xmath829 + 45 & @xmath830 & @xmath831 & @xmath832 & @xmath833 + 75 & @xmath834 & @xmath835 & @xmath836 & @xmath837 + 150 & @xmath838 & @xmath839 & @xmath840 & @xmath841 + 230 & @xmath842 & @xmath843 & @xmath844 & @xmath845 + 450 & @xmath846 & @xmath847 & @xmath848 & @xmath849 + 900 & @xmath850 & @xmath851 & @xmath852 & @xmath853 + + + & @xmath717 & @xmath854 & @xmath855 & @xmath856 + 5 & @xmath857 & @xmath858 & @xmath859 & @xmath860 + 10 & @xmath861 & @xmath862 & @xmath863 & @xmath864 + 20 & @xmath865 & @xmath866 & @xmath867 & @xmath868 + 30 & @xmath869 & @xmath870 & @xmath871 & @xmath872 + 45 & @xmath873 & @xmath874 & @xmath875 & @xmath876 + 75 & @xmath877 & @xmath878 & @xmath879 & @xmath880 + 150 & @xmath881 & @xmath882 & @xmath883 & @xmath884 + 230 & @xmath885 & @xmath886 & @xmath887 & @xmath888 + 450 & @xmath889 & @xmath890 & @xmath891 & @xmath892 + 900 & @xmath893 & @xmath894 & @xmath895 & @xmath896 + + + & @xmath717 & @xmath897 & @xmath898 & @xmath899 + 5 & @xmath900 & @xmath901 & @xmath902 & @xmath903 + 10 & @xmath904 & @xmath905 & @xmath906 & @xmath907 + 20 & @xmath908 & @xmath909 & @xmath910 & @xmath911 + 30 & @xmath912 & @xmath913 & @xmath914 & @xmath915 + 45 & @xmath916 & @xmath917 & @xmath918 & @xmath919 + 75 & @xmath920 & @xmath921 & @xmath922 & @xmath923 + 150 & @xmath924 & @xmath925 & @xmath926 & @xmath927 + 230 & @xmath928 & @xmath929 & @xmath930 & @xmath931 + 450 & @xmath932 & @xmath933 & @xmath934 & @xmath935 + 900 & @xmath936 & @xmath937 & @xmath938 & @xmath939 + 1800 & @xmath940 & @xmath941 & @xmath942 & @xmath943 + @xmath944 & @xmath945 & @xmath946 + + + @xmath150 & & & @xmath947 + @xmath165 & & & @xmath948 + @xmath612 & & & @xmath949 + & & & @xmath950 + @xmath150 & @xmath621 & @xmath621 & @xmath621 + @xmath165 & @xmath951 & @xmath952 & @xmath953 + @xmath612 & @xmath954 & @xmath955 & @xmath956 + & @xmath957 & @xmath958 & @xmath959 + & & @xmath960 + + + @xmath150 & & & + @xmath165 & & & + @xmath612 & & & + & & & + @xmath150 & @xmath621 & @xmath621 & @xmath621 + @xmath165 & @xmath961 & @xmath962 & @xmath963 + @xmath612 & @xmath964 & @xmath965 & @xmath966 + & @xmath967 & @xmath968 & @xmath969 + & & @xmath970 + + + @xmath150 & & & @xmath971 + @xmath165 & & & @xmath972 + @xmath612 & & & @xmath973 + & & & @xmath974 + @xmath150 & @xmath621 & @xmath621 & @xmath621 + @xmath165 & @xmath975 & @xmath976 & @xmath977 + @xmath612 & @xmath978 & @xmath694 & @xmath627 + & @xmath979 & @xmath980 & @xmath981 + & & @xmath982 + + + @xmath150 & & & @xmath983 + @xmath165 & & & @xmath984 + @xmath612 & & & @xmath985 + & & & @xmath986 + @xmath150 & @xmath621 & @xmath621 & @xmath621 + @xmath165 & @xmath987 & @xmath988 & @xmath989 + @xmath612 & @xmath990 & @xmath991 & @xmath992 + & @xmath993 & @xmath994 & @xmath995 + & & @xmath996 +

if a dynamical system is long - lived and non - resonant ( that is , if there is a set of tracers that have evolved independently through many orbital times ) , and if the system is observed at any non - special time , it is possible to infer the dynamical properties of the system ( such as the gravitational force or acceleration law ) from a snapshot of the positions and velocities of the tracer population at a single moment in time . in this paper we describe a general inference technique that solves this problem while allowing ( 1 ) the unknown distribution function of the tracer population to be simultaneously inferred and marginalized over , and ( 2 ) prior information about the gravitational field and distribution function to be taken into account . as an example , we consider the simplest problem of this kind : we infer the force law in the solar system using only an instantaneous kinematic snapshot ( valid at 2009 april 1.0 ) for the eight major planets . we consider purely radial acceleration laws of the form^{-\alpha}$ ] , where is the distance from the sun . using a probabilistic inference technique , we infer ( 95 percent interval ) , largely independent of any assumptions about the distribution of energies and eccentricities in the system beyond the assumption that the system is phase - mixed . generalizations of the methods used here will permit , among other things , inference of milky way dynamics from _ gaia_-like observations .

0903.5308

the _ gaia _ satellite @xcite will measure positions and velocities for millions to billions of stars at varying precision . one of the principal goals of this mission is to provide the data necessary to infer the dynamical state of the milky way . however , there are issues in principle with inference of dynamics from a _ snapshot _ or instantaneous set of configuration and velocity measurements : the instantaneous positions and velocities have no _ necessary _ relationship with the gravitational potential or accelerations . indeed , despite a considerable literature ( for example , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) there is no methodology for performing this inference that naturally handles all of the issues , including finite and non - trivial observational uncertainties or noise , missing data , non - steady aspects of the mass distribution , and the ( incredibly likely ) possibility that the potential is not ( simply ) integrable . robust inference may not even be _ possible _ if the milky way has significant time - dependence or is strongly chaotic or is far from showing any simple symmetries ( such as axisymmetry ) . certainly there is no hope for dynamical inference on the massive scale required for the _ gaia _ data set if we can not perform it on much simpler , much more symmetrical , much older ( in a dynamical sense ) , and much smaller ( in a data sense ) systems . in what follows , we take one of the simplest possible systems the solar system and the smallest possible data set the positions and velocities of the major planets at a moment in time and perform a complete dynamical inference . for a test system we could also have chosen the black hole at the galactic center , where similar considerations apply . however , this system has additional issues with `` missing data '' because not all six phase - space coordinates are directly measurable for all stars . the solar system truly is the simplest problem in this class . our inferential starting point is orbital roulette @xcite . in this previous work , it was assumed that the orbital angles ( in the action angle formalism ) are uniformly distributed . thus , dynamical model parameters that correspond to orbital angles that suspiciously lack diversity are rejected . more specifically , the method considers a large range of possible dynamical model parameters , computes orbital angles at every value of the parameters , and a distribution statistic on those orbital angles . if the distribution statistic is designed to monotonically increase with the diversity of the angles or the flatness of the angle distribution , the `` best fit '' dynamical model parameters are those that optimize the distribution statistic . this kind of approach is inherently frequentist : on many applications of these procedures to independent datasets a confidence interval captures the true dynamical parameters on a guaranteed fraction of trials . however , for any given dataset the confidence interval produced might not represent a credible set of parameters . in what follows we cast the dynamical parameters estimation as a probabilistic inference problem ( a `` bayesian '' approach ) . we adopt the same core assumption as the roulette problem : we assume that the orbital angles are uniformly distributed . however , in this framework , we must also construct prior probability distribution functions for the dynamical model parameters and conditional probabilities of the data given the model to encode the assumptions of a long - lived and bound solar system . these prior and conditional probability distributions and the data create posterior probability distribution functions for all the parameters . we use this new method to infer the gravitational force law ( radial dependence and amplitude ) . what is new here , in the context of solar system dynamics , is that we perform this inference with only a snapshot of the kinematic state , that is , with only the positions and velocities of the planets at a _ single instant of time . _ of course the kinematic snapshot we employ is , in fact , a set of initial conditions for a solar system integration @xcite . these initial conditions were determined not by a single measurement at a single epoch , but are in fact the result of an optimization of a solar system integration to observed planetary positions over many decades . in the context of this paper a demonstration of a method it is best to think of these `` data '' as `` simulated data '' useful for testing the method . they just happen to be data that have been simulated by the analog computer we know as the solar system . what is new here , in the context of dynamical inference , is that our method is fully probabilistic or bayesian . this is important for future problems , such as the _ gaia _ problem , or for inferring the mass of the black hole at the center of the milky way , because in these real data analysis problems , the data points come with non - trivial and highly correlated observational uncertainties , and because entire dimensions of phase space are missing or unobserved . at the galactic center , we do not know the radial distance to anything accurately , and in the _ gaia _ data set many of the radial velocities will not be measured . the bayesian framework handles these real - world data issues very naturally , although in fact they are _ not _ important in the test problem we solve here . aside from these issues of principle , it is also the case , as we will show , that the bayesian method performs extremely well . of course , a lot is known about the gravitational force law in the solar system , so we do not expect , at the outset , to be surprised by our results . the first force - law inference in the solar system ( * ? ? ? * and also work by contemporaries , particularly hooke , who may have priority ) made use of full orbit shape determination @xcite . in this sense , newton s problem find the force law ( from among a small , discrete group of possible force laws ) that leads to elliptical orbits with the sun at one focus was much easier than the problem we have set for ourselves . of course , along the way , newton had to develop for the first time the general principles of kinematics and dynamics ! we are going to assume spherical symmetry of the solar system s force law and gravitational potential , although nothing in the general inference formalism that follows will require this . consider a radial force law ( really acceleration law ) of the form @xmath3^{-\alpha}\,{{\boldsymbol{\hat{r}}}}\quad,\ ] ] where @xmath4 is an amplitude , @xmath1 is the distance from the sun , @xmath5 is a distance scale ( in this case we will use @xmath6 so that @xmath4 can be thought of as the acceleration at earth s orbit ) , @xmath7 parameterizes the radial dependence , and @xmath8 is the radial direction . in this model , the list of free parameters is @xmath9 where we have taken the logarithm of @xmath4 because in inference problems , dimensioned parameters are usually best handled in the log @xcite . the potential @xmath10 ( potential energy per unit planet mass ) , radial effective potential @xmath11 , and binding energy per unit mass @xmath12 are @xmath13^{1-\alpha } \quad,\ ] ] @xmath14 @xmath15 where @xmath16 is the square of the magnitude of the planet s angular momentum per unit mass ( or @xmath17 ) , and @xmath18 is the radial component of the velocity ( the component of @xmath19 parallel to @xmath20 ) . the perihelion and aphelion distances @xmath21 and @xmath22 are both found by setting @xmath23 . with these , we can define a radial asymmetry @xmath24 as @xmath25 where we have called this `` @xmath24 '' because in the kepler newton @xmath26 case it is the orbital eccentricity . one way of thinking of this radial asymmetry is that at any point in the space made up of the dynamical parameters @xmath27 and the binding energy @xmath28 , the radial asymmetry @xmath24 is a dimensionless description of the angular momentum magnitude . importantly for what follows , we can define a `` radial angle '' @xmath29 that increases linearly with time from perihelion passage through next perihelion passage . any planet at radius @xmath1 on an orbit with perihelion distance @xmath21 and aphelion distance @xmath22 can be assigned this angle @xmath29 by @xmath30 \displaystyle \pi+\pi\,\frac{t(r)-t({r_{\mathrm{ap}}})}{t({r_{\mathrm{peri}}})-t({r_{\mathrm{ap } } } ) } & \mbox{for $ v_r < 0 $ } \end{array}\right . \quad,\ ] ] where the first numerator is the time it takes to go from @xmath21 to @xmath1 outbound , the first denominator is the time it takes to go from @xmath21 to @xmath22 outbound , the second numerator and denominator are the times inbound , and all time differences between two radii can be computed numerically for general values of @xmath7 by integrating the inverse of the radial velocity between these radii . the first - order form of this integral has an integrable singularity at the perihelion and aphelion , which can be handled by an appropriate change of variables ( e.g. , * ? ? ? a planet observed at a set of random times spanning many orbits will be observed to have radial angles @xmath29 drawn from a flat distribution in the range @xmath31 . this radial angle is one of the angles in the action angle formulation of the system , which is integrable for the simple reason that it is spherically symmetric . in what follows , we are going to use and compare several methods for inferring the force - law parameters @xmath27 ( the amplitude @xmath32 and radial exponent @xmath7 of the spherical force law ) from an instantaneous snapshot of the positions and velocities of the eight major planets . this snapshot was taken from jpl s horizons system which provides highly accurate ephemerides for solar system objects @xcite . it is an extrapolation ( at the time of writing ) to 2009 april 1.0 , approximately 400 years after the important publication of kepler ( 1609 ) . this kinematic snapshot is given in table [ table : eph ] . since this snapshot is obtained by integrating the positions and velocities of solar system bodies , the accuracy is limited by ( i ) the correctness of the dynamical model used , ( ii ) the numerical integration of the equations of motion , and ( iii ) the accuracy to which the initial conditions are known . it is generally believed that the dynamical model used is correct and complete , and that the numerical integration is sufficiently accurate . the main uncertainty in the ephemerides is then that due to the uncertainty in the initial conditions . the current set of initial conditions ( de405 ; * ? ? ? * ) is a fit to a set of optical , radar , and vlbi observations as well as to a set of spacecraft range and doppler points from various space missions . the uncertainties are the largest for the outer planets , since the data for these are almost entirely from optical observations ( with the exception of jupiter ) , and because neptune has not been observed over a full orbit since the start of precise measurements . a comparison between the de405 ephemerides and more recent observations shows that the positions of the inner planets are known to a fractional accuracy of approximately @xmath33 , while those of the outer planets are known to a fractional accuracy of @xmath34 to @xmath35 @xcite . uncertainties in the velocities are at the same fractional magnitude . this kinematic snapshot is not , of course , a fair data set with which to perform the inference below , for the main reason that the `` measured '' kinematic state of the solar system is in fact the output of fitting observations with a dynamical model that _ assumes _ @xmath26 . for this reason , the data should be thought of as `` idealized '' or `` simulated data '' and the work must be considered a test of the method rather than a definitive inference . the virial theorem relates the time averages @xmath36 and @xmath37 of a test particle s kinetic and potential energies through @xmath38 where @xmath7 is the exponent in the radial force law . given that a planet s potential energy is a function of the dynamical parameters @xmath27 , while its kinetic energy is not , the virial relation for each planet becomes a one - dimensional locus in the @xmath27 space . using kinetic and potential energies computed from the observations as a proxy for the time - averaged energies , these loci are shown in s [ fig : virial_main ] and [ fig : virial_zoom ] . the fact that all eight lines cross near a single point in the space is encouraging that the system is virialized ( as we expect ) and that the inference will work . that the eight lines so nearly intersect in a single point is because , in fact , many of the planets are on circular orbits ; the probabilistic method developed in this paper does not assume this , but it does retain the precision that is possible because of this situation . also shown in s [ fig : virial_main ] and [ fig : virial_zoom ] is the region of parameter space in which one or more of the planets is unbound because @xmath39 or , equivalently , @xmath40 . in what follows , we will assign vanishing probability to regions of parameter space in which one or more planets is unbound . however , as we will see , this does not affect any of our conclusions . also shown in [ fig : virial_main ] is the region of parameter space in which one or more of the planets has @xmath41 , where @xmath42 is the radius of the sun . this part of parameter space ought also to be excluded , although in practice this is not necessary for any of what follows . in preparation for what follows , we pre - compute all of the planet radial angles @xmath43given their positions @xmath44 and velocities @xmath45as a function of the dynamical parameters . these angles are shown in [ fig : anglesplanets ] . in orbital roulette @xcite , the idea is to compute , for each point @xmath27 in parameter space , the @xmath46 radial angles @xmath43 and analyze them statistically for being well mixed . in practice , this means applying a distribution test or multiple distribution tests to the angles and preferring parameters for which these tests are more consistent with a random or flat distribution of angles in the range @xmath31 . because each test provides one constraint , and in the case described here the parameter space is two - dimensional , at least two qualitatively different tests are required to make localized constraints in @xmath27 . in addition to a test for a flat angle distribution we also apply a test for an angle energy correlation . about the simplest consistency test for the calculated angles is a test of the mean of the angles : is the mean consistent with the expected mean for a uniformly distributed set of @xmath46 planets ? for this to perform well one must fold the angles of the inbound planets onto the interval @xmath47 $ ] , that is , disregard the information in the _ sign _ of the radial velocity ; then the expected mean of the angles is equal to @xmath48 ( for details on how to test this assumption , see @xcite ) . the fact that we have to perform this mapping indicates that the procedure is _ ad hoc_. indeed , for a uniform distribution on the circle there is no specific meaning to the perihelion and the aphelion , or to any two points , such that no real meaning can be attached to the mean of the angles between two arbitrary points on the circle . better , one can test the consistency of the full distribution function of the angles with a uniform distribution . this could again be done with the full @xmath49 $ ] distributed angles or with the folded angles ; the results will depend on this choice . testing an observed distribution for consistency with an expected distribution often involves comparing cumulative distribution functions @xcite . the kolmogorov smirnov ( k s ) test is the simplest in practice , since the distribution of the test statistic the maximum difference between the cumulative distributions can be approximated by an analytic function @xcite . s test is by construction most sensitive to deviations near the median value ; this rules out dynamical parameters at which the planets bunch up at perihelion or at aphelion , the situation in which about half are at perihelion and half at aphelion can easily dupe the test . a statistic can be chosen that is sensitive to deviations at all values , such as the anderson darling statistic @xcite . however , no approximate analytic description of the distribution of this statistic exists and in practice this distribution has to be obtained by monte carlo sampling ( e.g. , * ? ? ? * ) . a statistic more appropriate to the problem at hand ( although we are not primarily interested in a careful examination of the differences between different frequentist procedures ) is kuiper s statistic @xcite . this statistic the sum of the maximum distance of the observed cumulative distribution above and below the expected cumulative distribution is invariant under periodic shifts of the angle and was specifically designed to test uniform distributions on the circle . the advantage over anderson darling is that the asymptotic distribution of the kuiper statistic is known ( e.g. , * ? ? ? all of these tests for the uniformity of the distribution of the angles are shown in [ fig : freq ] . these tests can fail when for a certain combination of dynamical parameters some planets are near aphelion while other planets are near perihelion . this situation appears here in [ fig : anglesplanets ] , at @xmath7 far from @xmath50 , where there are large regions in which the inner planets , especially mercury and venus , are near perihelion , while the outer planets , especially uranus and neptune , are near aphelion and _ vice versa_. this prevents the mean angle and k s tests from excluding those regions . the kuiper test performs better . a second constraint ( for the two - dimensional parameter space ) comes from a second test . in regions of parameter space in which the inner planets are all near perihelion and the outer planets are all near aphelion , a significant correlation between the angles and the energies exists . this correlation is unphysical if the system is not being observed at any special time . a non - parametric test for the correlation is preferred here as the angle energy correlation will not in general be linear . we perform a test of the angle energy correlation using kendall s @xmath51 @xcite . this is a rank test ; it only considers the relative ordering of the angles and energies of different planets ( for details on this test see @xcite ) . that this test is in a sense orthogonal to the tests of the uniformity of the angle distribution can be seen in [ fig : freq ] . all of the frequentist tests permit acceptance or rejection of a dynamical model at a certain confidence level . 95 and 99 percent confidence intervals for all of the frequentist tests are shown in [ fig : freq ] . also shown is the combination of the tests of the uniformity of the angle distribution with the test of the angle energy correlation . the frequentist procedures perform reasonably well in this simple problem because the number of dynamical parameters is small and the data have vanishing errors and no missing components . as the number of dynamical parameters increases the frequentist must find larger numbers of tests in order to break degeneracies . while data uncertainties can be included by sampling the error distribution of the data and combining the results ( e.g. , * ? ? ? * ) , this will perform badly in the limit of low signal - to - noise or missing data . these difficulties are related to the fact that these procedures use only a very crude model of the data , that is , that the angles are distributed uniformly and that angle energy correlations should be absent , which allows no room for discovery of structure in phase space . a fully bayesian treatment of this problem can treat the phase - space distribution function as an unknown function to be inferred from the data . modeling the full phase - space distribution function permits simultaneous inference of missing data and properly marginalized probability distributions for dynamical parameters . imagine that we have the three - vector positions @xmath44 and three - vector velocities @xmath45 at some time @xmath52 for @xmath46 planets @xmath53 and a parameterized model for the gravitational acceleration law ( force law per unit mass ) @xmath54 , a function of position @xmath20 and a list of parameters @xmath27 . we wish to obtain an estimate of the posterior probability distribution @xmath55 for the parameters , where @xmath56 is the set of all planet positions and velocities . we employ bayes s theorem as follows : @xmath57 where , as usual , @xmath58 is the likelihood or the probability distribution function for the data given the model parameters , evaluated at the observed values of the data , @xmath59 is the prior probability distribution function for the parameters , and the denominator is ( for our purposes here ) a normalization constant . for our chosen parameterization of the dynamical parameters , @xmath27 , a broad flat or uniform prior in the space represents a reasonable description of our ( assumed ) prior knowledge . the much more challenging problem is to specify the likelihood of the dynamical parameters , the conditional probability distribution function @xmath58 . without detailed knowledge of how the solar system formed , this probability distribution is also a representation of our prior beliefs . ideally our ability to learn from the observed data will not be too sensitive to these beliefs . it is easier to express beliefs about the angle @xmath60 , radial asymmetry @xmath24 , and binding energy @xmath28 of each planet than its position and velocity . this is because , given our assumption that the planets constitute an angle - mixed population , jeans s theorem @xcite tells us that the distribution function , which is proportional to the probability of observing a planet at a certain locus in phase space , is only a function of the integrals of the motion . we wish to assign zero probability to dynamical parameters that lead to any of the computed binding energies @xmath61 being negative ( because we defined @xmath62 to be bound ) . we would also like to express our prior information or assumption that the system is long - lived ( in units of the dynamical time ) , that it is non - resonant , and that we are not seeing the system at any special time , or that the radial angles @xmath29 will be randomly distributed between @xmath63 and @xmath64 . in the absence of any better information , we will try to be as agnostic as possible about the _ actions _ ( conserved quantities ) of the planets but extremely confident that all radial angles @xmath31 are equally likely . in the simple spherical or radial situation under consideration here , in which there are no missing data , we can rewrite the likelihood as a function of the planets radial coordinates , as the orientation of the orbit does not depend on the dynamical parameters : @xmath65 because dimensioned parameters are usually best handled in the log , and @xmath66 is the jacobian matrix of all the partial derivatives of @xmath67 with respect to @xmath68 . for spherical potentials , this jacobian is given by @xmath69 is the radial period , which depends on the dynamical parameters and the integrals of the motion . for general @xmath7 this radial period can be computed numerically ( see [ sec : model ] ) . the derivative of the radial asymmetry with respect to the specific angular momentum squared can be written in terms of the perihelion and aphelion distance as @xmath70^{-1 } \left[{j^2}-a r_0 ^ 3\left(\frac{{r_{\mathrm{peri}}}}{r_0}\right)^{3-\alpha}\right]^{-1}\,.\ ] ] in the special case @xmath71 this jacobian is @xmath72 however , a priori we do _ not _ know the distribution function from which radial asymmetries and binding energies were drawn . fixing @xmath73 to a broad distribution would be making a strong assumption : regardless of what the data say we would continue to dogmatically believe that the distribution function was broad , which could result in poor inferences about the model as a whole . we can assume that the planets properties were drawn independently and identically distributed , without making strong assumptions about what that distribution was . this is achieved by introducing auxiliary parameters @xmath74 that if known would specify the distribution function . since we do not know these nuisance parameters , introducing a prior @xmath75 and marginalizing over them , @xmath76 is then part of the inference task . with all of this in place , we apply equation ( [ eq : bayes ] ) to get the posterior distribution over the dynamical parameters @xmath77 \ , \int{\mathrm{d}}{{\boldsymbol{{\theta}}}}\;p({{\boldsymbol{{\theta}}}},{{\boldsymbol{{\omega } } } } ) \,\prod_{i=1}^n \,p(e_i,{\ln{\epsilon}}_i|{{\boldsymbol{{\theta}}}},{{\boldsymbol{{\omega}}}})\quad , \label{eqn : posterior}\ ] ] where each planet s value of @xmath78 is a function of phase - space position @xmath79 and dynamical parameters @xmath27 , and the jacobian is evaluated at each planet s value of @xmath78 . for situations in which we have missing data or noisy observations , further unknown quantities will be added to the model and marginalized over . as the model becomes more complicated it is more likely that markov chain monte carlo ( e.g. , * ? ? ? * ) or other approximate computational methods will be required . to keep things as simple as possible , at first we model the distribution function @xmath80 as a product of a top - hat function in @xmath81 from @xmath82 to @xmath83 with a top - hat function in @xmath24 from @xmath84 to @xmath85 . in this context , the phase - space parameter list is @xmath86 in what follows , we only consider values @xmath87 , where @xmath4 and @xmath88 provide very distant ( uninformative ) limits ( below , we will take the limit ) , and @xmath89 . these enforce our assumption or prior information that the system is bound . for our initial model we also set the prior @xmath90 flat or uniform in all the parameters . now the marginalization required to compute the posterior ( equation [ eqn : posterior ] ) , an integral over all four of the phase - space distribution parameters in @xmath91 , can be performed analytically ; it leaves @xmath92^{2-n } \left[1-[1-e_l]^{2-n}-[e_m]^{2-n } + [ e_m - e_l]^{2-n}\right]\\ & \quad\times\prod_{i=1}^n |j({\ln{\epsilon}},e,{\phi_r};r_i , v_{r , i},{j_i^2})|\quad , \end{split}\ ] ] where @xmath93 is the lowest planetary binding energy at this point @xmath27 in dynamical parameter space , @xmath94 is the highest , @xmath95 is the lowest planetary radial asymmetry at this point @xmath27 , @xmath96 is the highest , and we have taken the limit in which the range of the parameters @xmath97 goes to infinity , or @xmath98 , @xmath99 . this posterior probability distribution for @xmath27 represents our posterior beliefs marginalized over all possible values of @xmath100 of the tophat model . the results of any inference depend both on the data and on modeling assumptions . thus , it is always sensible to question any assumptions and investigate how sensitive the results are to them . the marginalization in our basic method is tractable at the expense of believability : it seems unlikely that the true distribution function over the planet properties is uniform in the chosen parameterization . if the true distribution function is far from this assumption , the inference performed here could be in trouble because the zero prior probability assigned to all other reasonable forms of the distribution function makes it impossible for the model to learn this . this family of distribution functions also has no special status : a distribution which is uniform in one parameterization will not be uniform in another . instead of modeling the distribution in @xmath101 , we could have chosen to model the distribution function in terms of @xmath102 . this set of parameters is , in a sense , more natural as it is closer to action angle coordinates , and , thus , the jacobian @xmath103 is more close to constant . indeed , from equation ( [ eq : jacobian ] ) it is clear that , in the case of @xmath71 , this jacobian does not depend on the eccentricity of the orbit any longer and [ fig : ejacobiansplanets ] shows that this jacobian is close to constant for all values of @xmath32 and @xmath7 . however , it is not obvious a priori whether using this parameterization will give better results . rather than guessing , we can assume that the distribution over radial asymmetries is uniform in an unknown parameterization , and infer from the data what this parameterization should be . we tried assuming that the radial asymmetry was uniform over some range of @xmath24 , @xmath104 or @xmath105 , with equal prior probability over each choice . this method corresponds to extending @xmath91 to parameterize a richer class of distributions . as we did with nuisance parameters before , we can marginalize over this choice . in general we could marginalize over a very flexible class of distributions for both the binding energy and radial asymmetry . a generic way of representing distributions is with a histogram , where any distribution can be represented within an arbitrary tolerance for sufficiently narrow bins . some care is required in putting a prior @xmath106 over the heights of the bins . it is tempting to use a dirichlet distribution , which allows analytical marginalization . however , this assumes that the heights of the bins are unrelated except through their overall normalization . an arbitrarily spiky distribution function is not only unphysical , but we would be unable to infer its shape from point observations . we attempted to construct a sensible prior over distributions using histograms with many bins by coupling the heights of the bins so that the densities appear smooth . since the focus on this paper is on the principles of dynamical inference and not on the specific result on the solar system s force law , we only provide a brief description of how we implemented this approach . we refer the interested reader to the references provided in the following for more details on the specifics . a multivariate gaussian prior can be put over the log bin heights , approximating a logistic gaussian process @xcite . we used 100 equal - width bins coupled with a gaussian covariance function @xcite . we put uniform priors over the start and end points of the histogram and over physically reasonable ranges of the covariance function s parameters . the posterior over the dynamical parameters was estimated by markov chain monte carlo simulation of all the unknowns , gibbs sampling was used to update the dynamical parameters over a grid of values for which we had precomputed the jacobian term . slice sampling @xcite was used to update all nuisance parameters , except the bin heights which were updated with a metropolis hastings method ( * ? ? ? * eq . 15 ) . in our basic method we evaluate the jacobian @xmath107 for each planet at the observed radius , radial velocity , and specific angular momentum , for each value of the dynamical parameters @xmath27 , and multiply the magnitudes of the determinants of these jacobians together . following equation ( [ eq : basic_posterior ] ) we multiply this product with the value of the marginalized product of the distribution function and finally multiply this with the ( uniform ) prior on the dynamical parameters in order to obtain the posterior probability distribution for the dynamical parameters . it is instructive to look at the magnitude of the determinant of the jacobian for each planet as a function of the dynamical parameters in [ fig : jacobiansplanets ] . these factors seem to show a clear preference for the dynamical parameters of each planet to lie on its virial locus . why is this ? for nearly circular orbits the last two factors in equation ( [ eq : dedj2 ] ) are equal and are unbounded for perfectly circular orbits ; this results in a natural preference for nearly circular orbits for all of the planets , i.e. , for each planet to lie somewhere along its virial locus . at the same time the measured non - zero radial velocity limits the extent to which a planet s orbit can be circularized by a suitable choice of the dynamical parameters and , thus , how close to the circular singularity an individual planet can be brought . this naturally leads to a weighting scheme in which planets that are closer to circular orbits ( as measured by the observed angle between the planet s velocity and the tangent to the circular orbit going through its present location ) receive a higher weight in the analysis than planets on less circular orbits . this weighting scheme is good in that planets on more circular orbits are indeed more informative about the potential than planets on less circular orbits . therefore , it may seem that the decision to model the radial asymmetries as uniform in @xmath24 was fortuitous , but in fact , as we will discuss below in [ sec : sensitivity ] , the results obtained using the basic method are robust in that more flexible models learn that the tophat in @xmath24 leads to a good interpretation of the data . the posterior probability distribution is shown in [ fig : bayes2d ] . a strong peak is apparent near @xmath108 and the newtonian solar value for @xmath32 . the width of the probability distribution is indicated by the 95 and 99 percent posterior contours . these contours are defined to enclose the smallest area that holds 95 and 99 percent , respectively , of the posterior probability distribution . in order to infer the exponent @xmath7 of the force power - law we perform a second marginalization of the posterior probability distribution , this time over the ( for our purposes ) uninteresting parameter @xmath32 . this gives a posterior probability distribution for the parameter @xmath7 @xmath109 this probability distribution is shown in [ fig : bayes1d ] . the 95 and 99 percent posterior intervals in this figure are defined to exclude 2.5 and 0.5 percent , respectively , of the distribution on either side of the central region . the result of the inference is not a value for the parameters but a posterior probability distribution . a `` best fit '' value for @xmath7 could be obtained according to various criteria . for example , the posterior mean minimizes the expected square difference from the true parameter . however , since the posterior distribution is multi - modal , a point - estimate does not capture our uncertainty . the posterior is better summarized by a credible interval , containing a given fraction of the probability mass . the 95 percent posterior interval is @xmath110 . this compares favorably with the results obtained by @xcite who inferred @xmath71 from a much richer data set ( though a less rich model set ) . modern tests constrain the value of the exponent @xmath7 to a fractional accuracy of 10@xmath11110@xmath112 on solar system scales @xcite using lunar laser ranging tests @xcite and keplerian tests comparing @xmath113 and the rate of precession for different planets @xcite . our method appears to work well : the true dynamical parameters @xmath4 and @xmath7 are plausible under a fairly tight posterior distribution , see [ fig : bayes2d ] . the confidence intervals from the frequentist tests , [ fig : freq ] , are somewhat broader than the posterior intervals , but are not directly comparable . frequentist confidence intervals do not represent beliefs about the parameters for this dataset , simply a region that satisfies some sampling properties . as a result it is unsurprising that the frequentist confidence intervals appear less good if interpreted as beliefs given the data . it is difficult to use the frequentist confidence intervals to make statements about @xmath7 alone : without the assumptions of the bayesian method we can not marginalize over the parameter @xmath4 . one approach that allows frequentist guarantees is to set @xmath4 adversarially , but this pessimistic view would give unreasonably large error bars for @xmath7 . that the posterior distribution @xmath114 turns out to be multi - modal is no surprise in the light of the virial considerations from s [ fig : virial_main ] and [ fig : virial_zoom ] . the two peaks in @xmath114 correspond to the two main regions in parameter space in which the different virial loci cross . these virial considerations , and also the frequentist techniques , are very similar to the probabilistic approach in that they all prefer each planet to be in a non - special region of radial angle space , that is , between perihelion and aphelion ; this can only happen simultaneously near the points in parameter space at which virial loci cross . the advantage of the probabilistic approach is that it explains and quantifies this reasoning , and uses it to set formal limits on the dynamical parameters . the strongly peaked jacobian factors shown in [ fig : jacobiansplanets ] depend strongly on the choice of parameterization . if the model was expressed using the radial asymmetries squared rather than the radial asymmetries , the jacobians , some of which are shown in [ fig : ejacobiansplanets ] , are much flatter . if the model is maintained by also transforming the prior over radial asymmetry distributions to the new parameterization , the posterior distribution over dynamical parameters will be unchanged . setting all the radial asymmetries close to one is penalized elsewhere in the mathematical expression . however , setting the prior over radial asymmetry distributions to consider only uniform distributions was a strong assumption . if we had only considered distributions uniform in the radial asymmetry squared , this combined with the flat jacobians in [ fig : ejacobiansplanets ] would not have given as tight a posterior distribution . fortunately we can infer a suitable parameterization , or equivalently the distribution function from a richer family of possibilities , from the data . [ fig : altbayes1d ] shows the posterior over @xmath7 from using the two more flexible priors outlined in [ sec : altmethods ] . assuming that the radial asymmetry is uniform in one of @xmath105 , @xmath24 , or @xmath104 gave a 95% credible interval of @xmath115 . a variety of different flexible priors for the radial asymmetry distribution gave very similar results . allowing the @xmath32 constant in the force law to come from a flexible family of distributions changed the posterior very little . the 95% credible interval from the flexible non - parametric prior on both distributions gave a 95% credible interval of @xmath116 . the fine details of the posterior distribution , specifically the relative mass in the two modes , are sensitive to prior assumptions . this is unsurprising with only eight data points . however , the position of the bulk of the posterior mass , which gives the confidence interval is surprisingly robust across a wide range of modeling choices . as a warning , it is possible to obtain poor results through bad modeling choices . the reasonable posterior from our basic method , [ fig : bayes1d ] , was to some extent due to chance . the family of possible radial asymmetry distributions is inflexible as it forces this distribution to be flat over some range in radial asymmetry . using the basic method it is not possible to learn the true distribution , which is not flat in radial asymmetry , from data , because the truth is assigned zero prior probability . we were fortunate to choose a parameterization in which the inflexible family was not too unreasonable . the non - parametric approach is much more flexible in that its weak smoothness constraints assign non - zero prior probabilities to all reasonable distribution functions such that the true one could be picked out with enough data . however , care must also be taken with non - parametric modeling . an unsmooth dirichlet prior over bin heights in a non - parametric prior over distributions gave a broad and biased distribution over @xmath7 . the necessity of a proper probabilistic approach as laid out in this paper has been recognized before in the context of inferring the mass of the galaxy from a small set of highly informative tracers : distant satellite galaxies @xcite and high velocity stars @xcite . the limitations of those works are that they choose restrictive forms of the distribution function in which the distribution of angular momentum is fixed as either radial or isotropic , or given as an inflexible parametric function in between these two extrema @xcite . the precision of their results for the fundamental parameters of the galaxy is limited by the systematic uncertainty arising from this assumption . in the case of estimating the local escape velocity from a sample of high velocity stars , a power law model for the energy part of the distribution function was used ; some progress has been made marginalizing over the exponent , a nuisance parameter , with a prior coming from cosmological simulations @xcite . in the future , using the technique introduced in this paper there is no need to make strong assumptions about the degree of anisotropy of the system at hand or the form of the distribution function beyond the assumption of complete angle mixing . the results of this paper show that by introducing flexible models for the distribution function that can learn from the data , robust constraints on the parameters of a dynamical system can be obtained . the approach developed here can be applied to other ( perhaps more pressing ) dynamical inference problems in which test particles can be relied upon to be well - mixed in angle space . one such problem is the dynamics in the region surrounding the black hole at the galactic center . often in these problems complications arise because of large observational uncertainties ( often highly correlated ) , the absence of some of the six - dimensional phase - space coordinates , and selection effects , all of which are absent in the simple problem considered here . it will be necessary for these problems to model the full six - dimensional phase - space distribution function . this will complicate the marginalization over the phase - space parameters , which was trivial here in the basic method , but at the same time it will permit the discovery of structure in the phase - space distribution , which can aid with the presence of missing data and large observational uncertainties . that this approach has much potential has been shown before in the case of the galactic center ; the assumption that a set of stars at the galactic center is part of a disk - like population has been successful in reconstructing missing data @xcite . extension of the approach developed in this paper will permit incorporation of much more of the available data on the dynamics in the central region of the galaxy . this , in turn , will lead to a better determination of the mass of the black hole and the surrounding density profile . on larger scales , approaches like that developed in this paper will prove to be essential for the analysis of the large data sets of upcoming surveys such as _ gaia_. as the duration of the _ gaia _ mission is vanishingly smaller than any dynamical timescale , the problem posed is essentially the same as the problem posed here : infer the dynamics from a snapshot of the kinematics . our approach can not simply be applied to this larger problem because the system is almost certainly _ not _ ( trivially ) integrable , and the assumption of mixed angles and lack of resonances is invalidated by the clear abundance of substructure in the halo ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) and the disk @xcite ; indeed , jeans himself , in the paper in which he first wrote down his eponymous theorem @xcite , argued _ against _ assuming a steady state for the galaxy . modeling the details of the phase - space distribution function will be even more important in this context . however , we expect that a large fraction of the galaxy , even a large fraction of the stellar halo , is well - mixed , such that mixed - angle approaches are expected to lead to valuable inferences . combining the mixed - angle and unmixed - substructure techniques into a general inference about the dynamics of the galaxy requires a fully probabilistic method and the approach developed in this paper is a baby step toward this ambitious goal . first of all we thank scott tremaine for many inspiring discussions that helped shape this paper . it is also a pleasure to thank kyle cranmer , dustin lang , yuri levin , phil marshall , bill press , and hans - walter rix for very helpful discussions and the anonymous referee for valuable comments . this research was partially supported by nasa ( adp grant nnx08aj48 g ) , the nsf ( grant ast-0908357 ) , and a research fellowship of the alexander von humboldt foundation . this project made use of the horizons system provided by the solar system dynamics group of the jet propulsion laboratory , the nasa astrophysics data system , and the open - source python modules scipy , numpy , and matplotlib . adelberger , e. g. , heckel , b. r. , & nelson , a. e. , 2003 , annu . nucl . part . , 53 , 77 anderson , t. w. & darling , d. a. , 1952 , ann.math.stat . , 23 , 193 beloborodov , a. m. & levin , y. , 2004 , , 613 , 224 beloborodov , a. m , levin , y. , eisenhauer , f. , genzel , r. , paumard , t. , gillessen , s. , & ott , t. , 2004 , , 648 , 405 belokurov , v. , et al . , 2006 , , 642 , l137 belokurov , v. , et al . , 2007 , , 654 , 897 binney , j. & tremaine , s. , 2008 , galactic dynamics : second edition ( princeton university press ) bovy , j. , hogg , d. w. , & roweis , s. t. , 2009 , , 700 , 1794 dehnen , w. , 1998 , , 115 , 2384 fischbach , e. & talmadge , c. l. , 1999 the search for non - newtonian gravity , ( berlin : springer ) giorgini , j. d. , yeomans , d. k. , chamberlin , a. b. , chodas , p. w. , jacobson , r. a. , keesey , m. s. , lieske , j. h. , ostro , s. j. , standish , e. m. , & wimberly , r. n. , 1996 , , 28 , 1158 jeans , j. h. , 1915 , , 76 , 70 jeffreys , h. , 1939 , theory of probability ( oxford : clarendon press ) johnston , k. v. , zhao , h. , spergel , d. n. , & hernquist , l. , 1999 , , 512 , l109 kaasalainen , m. & binney , j. , 1994 , phys.rev.lett . , 73 , 2377 kendall , m. , 1938 , biometrika , 30 , 81 kepler , j. , 1609 , astronomia nova , trans . w. h. donahue , 1992 ( cambridge university press ) kochanek , c. s. , 1996 , , 457 , 228 kolmogorov , a. n. , 1941 , ann . , 12 , 461 koposov , s. , et al . , 2008 , , 686 , 279 kuiper , n. h. , 1962 , proc . koninklijke nederlandse akademie van wetenschappen a , 63 , 38 leonard . , t. , 1978 , j.roy.stat.soc.b , 40 , 113 leonard , p. j. t. & tremaine , s. , 1990 , , 353 , 486 little , b. & tremaine , s. , 1987 , , 320 , 493 neal . , r. m. , 1993 , probabilistic inference using markov chain monte carlo methods . technical report crg - tr-93 - 1 , department of computer science , university of toronto neal . , r. m. , 1999 , in bayesian statistics 6 , ed.j . m. bernardo et al .. ( oxford university press ) , 475 neal . , r. m. , 2003 , ann.stat . , 31 , 705 newton , i. , 1687 , philosophiae naturalis principia mathematica , trans . f. cajori , 1934 ( university of california press ) oort , j. h. , 1932 , bull.astron.inst.netherlands , 6 , 249 perryman , m. a. c. , et al . , 2001 , , 369 , 339 press , w. h. , teukolsky , s. a , vetterling , w. t. , & flannery , b. p. , 2007 , numerical recipes : the art of scientific computing , 3rd edition ( cambridge university press ) rasmussen , c. e. & williams . , c. k. i. , 2006 gaussian processes for machine learning ( mit press ) schwarzschild , m. , 1979 , 232 , 236 sivia , d. s. & skilling , j. , 2006 , data analysis : a bayesian tutorial ( oxford university press ) smith , m. c. , et al . , 2007 , , 379 , 755 standish , e. m. , 1998 , jpl planetary and lunar ephemerides , de405/le405 , jpl iom 312 , f-98 - 048 ( pasadena , ca : jpl ) standish , e. m. , 2004 , , 417 , 1165 stephens , m. a. , 1970 , j. r. stat . soc . b , 32 , 115 talmadge , c. , berthias , j .- p . , hellings , r. w. , & standish , e. m. , 1988 , phys . lett . , 61 , 1159 williams , j. g. , turyshev , s. g. , & boggs , d. h. , 2004 , phys . lett . , 93 , 261101 willman , b. , et al . , 2005 , , 626 , l85 [email protected]@[email protected]@[email protected]@.l mercury & 0 & 324190175 & 0 & 090955208 & -0 & 022920510 & -4 & 627851589 & 10 & 390063716 & 1 & 273504997 + venus & -0 & 701534590 & -0 & 168809218 & 0 & 037947785 & 1 & 725066954 & -7 & 205747212 & -0 & 198268558 + earth & -0 & 982564148 & -0 & 191145980 & -0 & 000014724 & 1 & 126784520 & -6 & 187988860 & 0 & 000330572 + mars & 1 & 104185888 & -0 & 826097003 & -0 & 044595990 & 3 & 260215854 & 4 & 524583075 & 0 & 014760239 + jupiter & 3 & 266443877 & -3 & 888055863 & -0 & 057015321 & 2 & 076140727 & 1 & 904040630 & -0 & 054374153 + saturn & -9 & 218802228 & 1 & 788299816 & 0 & 335737817 & -0 & 496457364 & -2 & 005021061 & 0 & 054667082 + uranus & 19 & 930781147 & -2 & 555241579 & -0 & 267710968 & 0 & 172224285 & 1 & 357933443 & 0 & 002836325 + neptune & 24 & 323085642 & -17 & 606227355 & -0 & 197974999 & 0 & 664855006 & 0 & 935497207 & -0 & 034716967 + currentlabel ] ) for each of the eight planets in the solar system . for the combinations of dynamical parameters in the light gray region in the lower right at least one planet becomes unbound . when the dynamical parameters are in the light gray region in the upper left at least one planet has @xmath41 . the light gray regions overlap in the dark gray region . in the units used in this , the `` true '' value of @xmath4 lies at @xmath117 . ] for each of the eight planets as a function of the dynamical parameters . the gray triangular region in the bottom - right corner is the region excluded by the condition that all the planets are bound . each planet has an angle range of @xmath118 if it has radial velocity @xmath119 ( outgoing from perihelion ) or @xmath120 if it has @xmath121 ( incoming from aphelion).,title="fig : " ] for each of the eight planets as a function of the dynamical parameters . the gray triangular region in the bottom - right corner is the region excluded by the condition that all the planets are bound . each planet has an angle range of @xmath118 if it has radial velocity @xmath119 ( outgoing from perihelion ) or @xmath120 if it has @xmath121 ( incoming from aphelion).,title="fig : " ] + for each of the eight planets as a function of the dynamical parameters . the gray triangular region in the bottom - right corner is the region excluded by the condition that all the planets are bound . each planet has an angle range of @xmath118 if it has radial velocity @xmath119 ( outgoing from perihelion ) or @xmath120 if it has @xmath121 ( incoming from aphelion).,title="fig : " ] for each of the eight planets as a function of the dynamical parameters . the gray triangular region in the bottom - right corner is the region excluded by the condition that all the planets are bound . each planet has an angle range of @xmath118 if it has radial velocity @xmath119 ( outgoing from perihelion ) or @xmath120 if it has @xmath121 ( incoming from aphelion).,title="fig : " ] + for each of the eight planets as a function of the dynamical parameters . the gray triangular region in the bottom - right corner is the region excluded by the condition that all the planets are bound . each planet has an angle range of @xmath118 if it has radial velocity @xmath119 ( outgoing from perihelion ) or @xmath120 if it has @xmath121 ( incoming from aphelion).,title="fig : " ] for each of the eight planets as a function of the dynamical parameters . the gray triangular region in the bottom - right corner is the region excluded by the condition that all the planets are bound . each planet has an angle range of @xmath118 if it has radial velocity @xmath119 ( outgoing from perihelion ) or @xmath120 if it has @xmath121 ( incoming from aphelion).,title="fig : " ] + for each of the eight planets as a function of the dynamical parameters . the gray triangular region in the bottom - right corner is the region excluded by the condition that all the planets are bound . each planet has an angle range of @xmath118 if it has radial velocity @xmath119 ( outgoing from perihelion ) or @xmath120 if it has @xmath121 ( incoming from aphelion).,title="fig : " ] for each of the eight planets as a function of the dynamical parameters . the gray triangular region in the bottom - right corner is the region excluded by the condition that all the planets are bound . each planet has an angle range of @xmath118 if it has radial velocity @xmath119 ( outgoing from perihelion ) or @xmath120 if it has @xmath121 ( incoming from aphelion).,title="fig : " ] test for the absence of angle energy correlations ; combined confidence intervals from the kuiper test and the kendall test ; combined confidence intervals from the k s test and the kendall test . in the plots with a single statistic the darkest region corresponds to the 95 percent confidence region , the lighter region to the 99 percent confidence region . the same is true for the plots with combinations of statistics , except that a bonferroni correction has been applied to the significances of the individual statistics that make up the combination . in each plot the lightest region is excluded because at least one planet becomes unbound for those parameter values.,title="fig : " ] test for the absence of angle energy correlations ; combined confidence intervals from the kuiper test and the kendall test ; combined confidence intervals from the k s test and the kendall test . in the plots with a single statistic the darkest region corresponds to the 95 percent confidence region , the lighter region to the 99 percent confidence region . the same is true for the plots with combinations of statistics , except that a bonferroni correction has been applied to the significances of the individual statistics that make up the combination . in each plot the lightest region is excluded because at least one planet becomes unbound for those parameter values.,title="fig : " ] + test for the absence of angle energy correlations ; combined confidence intervals from the kuiper test and the kendall test ; combined confidence intervals from the k s test and the kendall test . in the plots with a single statistic the darkest region corresponds to the 95 percent confidence region , the lighter region to the 99 percent confidence region . the same is true for the plots with combinations of statistics , except that a bonferroni correction has been applied to the significances of the individual statistics that make up the combination . in each plot the lightest region is excluded because at least one planet becomes unbound for those parameter values.,title="fig : " ] test for the absence of angle energy correlations ; combined confidence intervals from the kuiper test and the kendall test ; combined confidence intervals from the k s test and the kendall test . in the plots with a single statistic the darkest region corresponds to the 95 percent confidence region , the lighter region to the 99 percent confidence region . the same is true for the plots with combinations of statistics , except that a bonferroni correction has been applied to the significances of the individual statistics that make up the combination . in each plot the lightest region is excluded because at least one planet becomes unbound for those parameter values.,title="fig : " ] + test for the absence of angle energy correlations ; combined confidence intervals from the kuiper test and the kendall test ; combined confidence intervals from the k s test and the kendall test . in the plots with a single statistic the darkest region corresponds to the 95 percent confidence region , the lighter region to the 99 percent confidence region . the same is true for the plots with combinations of statistics , except that a bonferroni correction has been applied to the significances of the individual statistics that make up the combination . in each plot the lightest region is excluded because at least one planet becomes unbound for those parameter values.,title="fig : " ] test for the absence of angle energy correlations ; combined confidence intervals from the kuiper test and the kendall test ; combined confidence intervals from the k s test and the kendall test . in the plots with a single statistic the darkest region corresponds to the 95 percent confidence region , the lighter region to the 99 percent confidence region . the same is true for the plots with combinations of statistics , except that a bonferroni correction has been applied to the significances of the individual statistics that make up the combination . in each plot the lightest region is excluded because at least one planet becomes unbound for those parameter values.,title="fig : " ] of the jacobian of the transformation from the energy @xmath28 , radial asymmetry @xmath24 , and radial angle @xmath29 coordinates to the relevant positional and kinematical observables , evaluated at the observed positions and velocities of the planets , as a function of the dynamical parameters . gray scales are linear with darker shades for larger values.,title="fig : " ] of the jacobian of the transformation from the energy @xmath28 , radial asymmetry @xmath24 , and radial angle @xmath29 coordinates to the relevant positional and kinematical observables , evaluated at the observed positions and velocities of the planets , as a function of the dynamical parameters . gray scales are linear with darker shades for larger values.,title="fig : " ] + of the jacobian of the transformation from the energy @xmath28 , radial asymmetry @xmath24 , and radial angle @xmath29 coordinates to the relevant positional and kinematical observables , evaluated at the observed positions and velocities of the planets , as a function of the dynamical parameters . gray scales are linear with darker shades for larger values.,title="fig : " ] of the jacobian of the transformation from the energy @xmath28 , radial asymmetry @xmath24 , and radial angle @xmath29 coordinates to the relevant positional and kinematical observables , evaluated at the observed positions and velocities of the planets , as a function of the dynamical parameters . gray scales are linear with darker shades for larger values.,title="fig : " ] + of the jacobian of the transformation from the energy @xmath28 , radial asymmetry @xmath24 , and radial angle @xmath29 coordinates to the relevant positional and kinematical observables , evaluated at the observed positions and velocities of the planets , as a function of the dynamical parameters . gray scales are linear with darker shades for larger values.,title="fig : " ] of the jacobian of the transformation from the energy @xmath28 , radial asymmetry @xmath24 , and radial angle @xmath29 coordinates to the relevant positional and kinematical observables , evaluated at the observed positions and velocities of the planets , as a function of the dynamical parameters . gray scales are linear with darker shades for larger values.,title="fig : " ] + of the jacobian of the transformation from the energy @xmath28 , radial asymmetry @xmath24 , and radial angle @xmath29 coordinates to the relevant positional and kinematical observables , evaluated at the observed positions and velocities of the planets , as a function of the dynamical parameters . gray scales are linear with darker shades for larger values.,title="fig : " ] of the jacobian of the transformation from the energy @xmath28 , radial asymmetry @xmath24 , and radial angle @xmath29 coordinates to the relevant positional and kinematical observables , evaluated at the observed positions and velocities of the planets , as a function of the dynamical parameters . gray scales are linear with darker shades for larger values.,title="fig : " ] of the jacobian of the transformation from the energy @xmath28 , radial asymmetry squared @xmath104 , and radial angle @xmath29 coordinates to the relevant positional and kinematical observables , evaluated at the observed positions and velocities of the planets , as a function of the dynamical parameters . grayscales are linear with darker shades for larger values . only the jacobians for mercury and venus are shown here ; the corresponding jacobians for the other planets are very similar to these.,title="fig : " ] + of the jacobian of the transformation from the energy @xmath28 , radial asymmetry squared @xmath104 , and radial angle @xmath29 coordinates to the relevant positional and kinematical observables , evaluated at the observed positions and velocities of the planets , as a function of the dynamical parameters . grayscales are linear with darker shades for larger values . only the jacobians for mercury and venus are shown here ; the corresponding jacobians for the other planets are very similar to these.,title="fig : " ] with 95 and 99 percent posterior intervals . top : distribution over radial asymmetry is assumed to be uniform in one of @xmath105 , @xmath24 or @xmath104 . bottom : results from a non - parametric prior over the distributions of both @xmath24 and @xmath32.,title="fig : " ] + with 95 and 99 percent posterior intervals . top : distribution over radial asymmetry is assumed to be uniform in one of @xmath105 , @xmath24 or @xmath104 . bottom : results from a non - parametric prior over the distributions of both @xmath24 and @xmath32.,title="fig : " ]

we study multi - soliton solutions of the four - dimensional su(n ) skyrme model by combining the hedgehog ansatz for su(n ) based on the harmonic maps of into and a geometrical trick which allows to analyze explicitly finite - volume effects without breaking the relevant symmetries of the ansatz . the geometric set - up allows to introduce a parameter which is related to the t hooft coupling of a suitable large limit , in which and the curvature of the background metric approaches zero , in such a way that their product is constant . the relevance of such a parameter to the physics of the system is pointed out . in particular , we discuss how the discrete symmetries of the configurations depend on it .

1508.07055

one of the most intriguing theoretical results in quantum field theory ( qft henceforth ) has been the realization that fermions can emerge out of purely bosonic lagrangians as solitonic excitations ( for a detailed review see @xcite ) . the clearest demonstration that the importance of this result goes far beyond pure theoretical physics is given by the skyrme theory @xcite which is one of the most important models of nuclear and particle physics . the skyrme term @xcite allows the existence of static soliton solutions with finite energy , called _ skyrmions _ ( see @xcite ) describing fermionic degrees of freedom ( see @xcite and references therein ) . the wide range of applications of the theory in other areas ( such as astrophysics , bose - einstein condensates , nematic liquids , multi - ferric materials , chiral magnets and condensed matter physics in general @xcite ) is well recognized by now . it is also worth to emphasize that the skyrme model appears in a very natural way in the context of the ads / cft correspondence @xcite . from the point of view of nuclear physics , it is very important to have analytic tools allowing to analyse the skyrme model when interacting skyrmions are present within bounded regions . this case is relevant whenever one wants to take into account the effects of the finite size on the topological properties of the skyrmions themselves . it is commonly believed that many - nucleons systems ( such as the ones occurring in nuclear pasta @xcite ; for a review see @xcite ) are completely out of reach of the analytical techniques provided by soliton theory already in the @xmath4 case ( while , at a first glance , the @xmath5 case is even worse ) . in particular , the task to compute physical parameters of such multi - nucleon systems with analytic multi - skyrmionic configurations is believed to be completely hopeless . recently , the generalized hedgehog ansatz in the @xmath4 case , introduced in @xcite , allowed the construction of the first multi - skyrmions at finite volume : namely , the first exact solutions of the skyrme model representing interacting elementary skyrmions with a non - trivial winding number , in which finite - volume effects can be explicitly taken into account ( arriving at a good prediction for the compression modulus ) , was obtained @xcite . the way to do this is to write the system in a modified cylinder - like " metric whose curvature is parametrized by a length @xmath6 . then multi - skyrmionic configurations look like necklaces of elementary skyrmions interacting in bounded tube - shaped regions . the ground state of such multi - skyrmions has the remarkable property that , although the bps bound in terms of the winding can not be saturated , a new topological charge exists which leads to a different bps bound , which can instead be saturated . in this paper , we consider the @xmath5 case . a very powerful technique to construct multi - skyrmionic configurations in unbounded regions is given by an ansatz for @xmath5 skyrmions introduced in @xcite and @xcite , based on harmonic maps of @xmath0 into @xmath1 , which allowed the construction of many interesting numerical multi - skyrmionic configurations . here we will exploit the fact that the metric we use is spherically symmetric , just like flat space , so the harmonic map ansatz can be used also in the present case without modifications . we shall see that in the new metric the equations simplify with respect to those studied in @xcite and @xcite ; in particular they become autonomous , thus techniques from dynamical systems theory become available . thanks to the choice of the background geometry , a novel type of large @xmath2 limit becomes possible in which @xmath2 is large and the curvature is small in such a way that their product is constant . in this way one can see that both the effects of the curvature become negligible and the field equations remain autonomous ( so that , in such a limit , one can kill the curvature of the metric but keeping all the advantages of the technique of @xcite ) . it is also worth to emphasize that it is precisely the large @xmath2 limit which discloses in the clearest possible way the role of the skyrme model as low energy limit of qcd . a non constructive proof of the existence of such nontrivial solutions is also provided and a few numerical solutions are exhibited as well . this paper is organized as follows : in the second section , the @xmath4 case treated in @xcite will be briefly reviewed in order to set the stage . in the third section , the ansatz for the skyrmions will be described . in the fourth section , the general equations of motion are written down and the flat - large @xmath2 limit is discussed . in the fifth section , we study the system using techniques from the theory of dynamical systems ; in particular , the stability of the fixed points in the @xmath7 case is analyzed , and their dependence on the geometric t hooft parameter is discussed . in the sixth section , some nontrivial numerical solution of the equations of motion in the @xmath8 case will be studied . in the seventh section , some conclusions will be drawn . in the appendix , it is proved on general grounds that the equations of motion do have nontrivial solutions which can be interpreted as genuine @xmath5 skyrmions . in this section , the construction of analytic multi - skyrmionic configurations in the @xmath4 case @xcite will be shortly reviewed . the action of the four dimensional skyrme model is given by @xmath9 \ , \nonumber\end{aligned}\ ] ] where the planck constant and the speed of light have been set to @xmath10 , and @xmath11 and @xmath12 are the coupling constants . the @xmath13 s are the generators of the flavor group , which in this section is @xmath4 . notice that we allowed for curved metrics . the coupling constants @xmath11 and @xmath12 are related to the couplings @xmath14 and @xmath15 used in @xcite . ] by @xmath16 the non - linear sigma model term of the skyrme action is necessary to take into account pions . the second term is the only covariant term leading to a well - defined hamiltonian formalism in time which supports the existence of skyrmions . the field equations following from the above action are @xmath17=0 . \label{nonlinearsigma1}\ ] ] the following standard parametrization of the @xmath4-valued scalar @xmath18 will be adopted@xmath19 where @xmath20 is the @xmath21 identity matrix ; to describe a spherically symmetric field configuration we use the hedgehog ansatz @xcite:@xmath22@xmath23 in order to mimic finite - volume effects without loosing the nice properties of the hedgehog ansatz we will consider the following curved background , but this has already been considered in @xcite . the geometry in eq . ( [ metric ] ) was considered in @xcite but with a motivation different from the analysis of finite - volume effects . consequently , the main results obtained in @xcite ( the derivation of both a novel bps bound which can be saturated and analytic multi - skyrmions at finite volume and an explicit formula for the compression modulus in good agreement with experiments ) are indeed novel . ] @xmath24@xmath25 where @xmath26 is the length of the @xmath27interval . the total volume of space is @xmath28 . this geometry describes three - dimensional cylinders whose sections are @xmath0 spheres , so that parameter @xmath6 plays the role of the ( finite ) diameter of the transverse sections of the tube . the fact that this parameter replaces the radial variable @xmath29 in the metric also leads , as we will see , to considerable simplification of the equations of motion , even allowing to find exact solutions in the @xmath4 case . moreover , the curvature of this metric is proportional to @xmath30 . as it will be explained in the next sections , the explicit presence of this parameter in eq . ( [ metric ] ) together with the @xmath2 of @xmath5 allows to define a smooth flat limit in which @xmath31 and so all the effects of the curvature disappear ( however , the global topology of space remains cylindrical even in the flat limit and so it differs from the trivial @xmath32 topology of flat static _ unbounded _ skyrmions ) . thus , in a sense , the above metric is introduced just as a regulator " whose local effects can be removed at the end . in the present context , flat limit " really means @xmath33 so that , from the practical point of view , already when @xmath6 is around @xmath34 @xmath35 all the effects of the curvature are negligible and , consequently , even in the flat limit in eq . ( [ flatl2 ] ) finite volume effects will not disappear . it is also worth to emphasize that the well known result that elementary skyrmions should be quantized as fermions ( which originally was derived on flat spaces ) has been extended to space - times with compact orientable three - dimensional spatial sections in @xcite ( and the metric in eq . ( [ metric ] ) belongs to this class ) . since , at the end , we will be interested in the flat limit in eq . ( [ flatl2 ] ) , one may wonder whether it would be possible to start from the very beginning with a flat metric . in fact , as it has been shown in @xcite , the background metric in eq . ( [ metric ] ) is a very suitable tool to take into account finite volume effects ( since the total spatial volume is finite ) without breaking relevant symmetries of the hedgehog ansatz , with the additional advantage of simplifying the field equations . a further relevant advantage of the above background metric is that , unlike what happens in the usual unbounded case , it allows to define in a very transparent way a smooth large @xmath2 limit of the @xmath5 skyrmions . therefore , it is much more convenient to analyze the skyrme theory first within the background metric in eq . ( [ metric ] ) , and take the flat limit only later . the effectiveness of such a choice for the metric is also shown by the results in @xcite and @xcite in which it has been shown that , unlike what happens in flat space , the equations for the yang - mills - higgs system ( in the sector with non - vanishing non - abelian electric and magnetic charges ) possess analytic solutions even in the case in which the higgs coupling is non - zero . with the above ansatz the skyrme field equations reduce in the static case to the following scalar differential equation for the skyrmion profile @xmath36 @xcite : @xmath37 \right ) = 0\ ] ] the winding number @xmath38 for such a configuration reads:@xmath39 in the present case , the natural boundary conditions correspond to the choice:@xmath40 and with these boundary conditions the winding number takes the integer value @xmath41 . these boundary conditions are unique in that they ensure @xmath42 , which correspond to bosonic and fermionic states for even and odd @xmath41 , respectively . smooth solutions exist for any @xmath41 satisfying the above boundary conditions for a finite range @xmath43 . in particular multi - soliton solutions exist , which represent skyrmions with winding number @xmath41 living in a finite spatial volume @xmath28 . it is worth to remark that the large @xmath41 limit in the present context is quite natural in order to consider thermodynamical properties of the multi - skyrmions system and @xmath41 is the baryon number : obviously , a thermodynamical analysis only makes sense in the cases in which the number of particles is very large . now we switch to the @xmath5 case with generic @xmath2 . our analysis will be based on the techniques introduced in @xcite , which we shall now briefly describe . the hedgehog ansatz for @xmath5 spherically symmetric skyrmions living in flat minkowski metric @xmath44 is based on a suitable family of projectors from @xmath0 into @xmath1 ( see @xcite ) . such projectors @xmath45 can be written as @xmath46 where @xmath47 is an @xmath2-component complex vector of two complex variables @xmath48 and @xmath49 which locally parametrize @xmath0 , where @xmath50 , and @xmath51 . the first @xmath52 is constructed with a holomorphic @xmath53 while the gram - schmidt procedure gives rise to the others . indeed , @xmath54 can be found by its action on any vector @xmath55 @xcite as @xmath56 consequently , the other vectors @xmath57 are determined inductively : @xmath58 . thus the operator @xmath59 corresponding to the family of vectors @xmath60 ( for @xmath61 ) reads @xmath62 where , due to the orthogonality of the projectors , we have @xmath63 . due to the holomorphy of @xmath64 , the following identities of the above defined vectors can be proved @xcite : @xmath65 & \partial_{\bar{\xi}}|v_{k}\rangle= -|v_{k-1}\rangle\frac{|v_{k}|^{2}}{|v_{k-1}|^{2}},\hspace{5mm}\hspace{5mm}\partial_{\xi}\left ( \frac { |v_{k-1}\rangle}{|v_{k-1}|^{2}}\right ) = \frac{|v_{k}\rangle}{|v_{k-1}|^{2}}. \label{aaa}\ ] ] for @xmath5 the components of @xmath66 , up to an irrelevant overall factor which cancels in the projector , are functions of only @xmath49 . the @xmath5 hedgehog ansatz defined in @xcite reads then@xmath67 where we defined @xmath68 . such an ansatz involves the introduction of @xmath69 projectors and of @xmath69 profile functions @xmath70 , @xmath71 ; . note that the projector @xmath72 is not included in the above formula since it is a linear combination of the others . one of the main results in @xcite and @xcite has been to show that the above ansatz in eq . ( [ ansatzsu(n ) ] ) , when inserted into the full skyrme field equations in the case in which the background metric is the standard flat metric in spherical coordinates in eq . ( [ flat ] ) , gives rise to a consistent set of @xmath69 coupled non - linear differential equations for the @xmath69 profiles @xmath73 . moreover , such field equations can also be derived as stationary equations for the energy functional with respect to variations of the profiles . a close inspection of the computations in @xcite shows that the main requirement in order for the ansatz to work is the @xmath74 invariance of the background metric . this then suggests that the above ansatz may also work in the finite - volume metric in eq . ( [ metric ] ) adopted in @xcite . this is what we show in the next section . in this section we switch back to the finite - volume @xmath74-invariant metric in eq . ( [ metric ] ) . a direct computation shows that the ansatz in eq . ( [ ansatzsu(n ) ] ) , when inserted into the full skyrme field equations in this metric gives rise to a consistent system of @xmath69 coupled _ autonomous _ non - linear differential equations for the @xmath69 profiles @xmath73 which is simpler than the flat _ non - autonomous _ system analyzed in @xcite as it will be now discussed . furthermore , also in the present case the field equations can be derived as stationary equations for the energy functional with respect to variations of the @xmath75 s . it is worth emphasizing here that the explicit presence of the parameters @xmath2 and @xmath76 allows to consider a flat large @xmath2 limit in which the curvature is negligible and the product @xmath77 remains constant . hence , the present formalism is also relevant for people only interested in the skyrme model on flat space - times . it is important to notice that in such a flat limit , while the effects of the curvature of the metric in eq.([metric ] ) disappear , _ the effects of the cylindrical topology do not_. from now on we shall use the following energy @xmath78 $ ] and length @xmath79 $ ] units:@xmath80 & = \frac{f_{\pi}}{4e}\approx1\ gev\ , \label{conv1}\\ \left [ l\right ] & = \frac{2}{ef_{\pi}}\approx0.6\ fm\ \rightarrow \label{conv2}\]]@xmath81 following the same steps of @xcite in the new metric , one arrives at the following expressions for the winding number , the total energy of the @xmath5 hedgehog and the field equations respectively : @xmath82 \right\vert _ { x =- l/2}^{x = l/2}\ , \label{winding(n)}\]]@xmath83 \ , \label{enersu(n)}\]]@xmath84 \ \ \mathbf { , } \label{equ(n)}\ ] ] where we introduced the quantities @xmath85 @xmath86 , and we defined @xmath87 notice that the winding number gets _ two _ different contributions ( while in the minkowski case it just gets one ) , as a consequence of the different topology of the cylinder - like metric we are using . another important point is that when the parameter @xmath6 is very large the field equations ( [ equ(n ) ] ) tend to linear equations , on the other hand when @xmath6 is small the nonlinear terms become very important . thus , this parameter ( more precisely , the quantity @xmath88 ) controls the nonlinearity of the theory , and thus it plays the rle of an additional coupling constant . all the analyses we shall perform in the following will confirm the fact that the physics of the model crucially depends on it . more precisely , both the numerical analysis of sect . [ numsol ] and the existence theorem of the appendix tell us that the global properties of the solutions of the system depend on the dimensionless parameter @xmath89 , which controls the shape of the cylinder . by comparing the present total energy in eq . ( [ enersu(n ) ] ) and the field equations in eq . ( [ equ(n ) ] ) with the corresponding expressions in @xcite , the first of the advantages mentioned above in working within the metric in eq . ( [ metric ] ) is apparent . namely , the field equations become an autonomous system which can , as such , be analyzed with the powerful tools of dynamical systems theory . in the special case in which all the profiles are equal , i.e. @xmath90 all the equations of the above system become proportional and equivalent to : @xmath91 = 0\ ] ] upon setting @xmath92 , this is precisely eq . , valid in the @xmath4 case . this is to be expected since setting all the profiles to be equal corresponds to looking for solutions which are embeddings of the @xmath4 ones in @xmath5 . in the next sections we shall see numerically in some cases that the above system of equations does admit nontrivial solutions , i.e. solutions whose profiles are not equal nor proportional , while in the appendix we shall prove analytically the existence of such solutions . such solutions have winding numbers which in general are nonvanishing , therefore can be interpreted as genuine @xmath5 multi - skyrmion configurations . as said , the formalism developed so far strongly relies on the usage of a spacetime whose spatial sections have a finite radius @xmath93 , and whose curvature goes to zero as @xmath94 . in this section we will show that there is a natural way to accommodate this flat limit , which requires to take the large @xmath2 limit as well . the latter is a common tool in quantum field theory , see e.g. @xcite for a review . the field equations in eq . ( [ equ(n ) ] ) can be rewritten as follows@xmath95 \ \ \mathbf { , } \label{largen1}\ ] ] where we introduced the effective radius @xmath96 the nonlinear part of the field equations ( [ largen1 ] ) behaves smoothly in the large @xmath2 limit provided also @xmath6 is large , so that @xmath97 is kept constant . more precisely : @xmath98 hence , the proper way to consider the large @xmath2 limit is to simultaneously consider the _ flat _ limit @xmath99 in such a way that the parameter @xmath100 in eq . ( [ largen2 ] ) remains finite . since in the large @xmath2 limit in eq . ( [ largen4 ] ) the nonlinear part of the field equations eq . ( [ largen1 ] ) does not depend on @xmath2 and @xmath6 separately but only on the effective radius @xmath97 defined in eq . ( [ largen2 ] ) , the quantity @xmath101 plays the rle of a _ geometric _ ( it defines a length scale ) t hooft coupling . physically , the parameter @xmath100 in eq . ( [ largen2 ] ) represents the effective area " available for each skyrmion within each section of the tube . in conclusion , one can see the metric defined in eq . ( [ metric ] ) just as a technical device in order to analyze multi - skyrmionic configurations since , at the end , one can turn off all the effects of the curvature ( keeping the cylindrical topology ) . in this limit , the solutions of the field equations represent multi - skyrmionic configuration living in a _ flat _ tube - shaped region whose sections have a radius much bigger than the scale @xmath97 . when n is large the equations read : @xmath102 \ \ \mathbf { , } \label{largen10}\ ] ] the first two terms in the above equation ( involving summations ) describe the free part of the theory , while the rest ( involving @xmath97 ) is responsible for interactions , therefore defines nontrivial dynamics . therefore we find our prescription a reasonable way to implement flat ( large volume ) limit within our formalism . in the next section we will demonstrate ( for @xmath103 ) , that the dynamics _ does _ depend on @xmath97 . one of the strongest advantages of our formalism lies in the fact that we have an _ autonomous _ system of differential equations . the latter can be qualitatively described by its phase portrait , an important characteristic of which is the set of critical points . in particular if under some change of parameters the set of critical points remains the same ( their number and their character ) , then one does not expect that dynamics of the system changes dramatically under the same change . on the contrary , if the number and/or character of the critical points changes , one may definitely say that the dynamics of the system is qualitatively different . in this section we are going to study our system by using the tools of dynamical system theory ( we will follow @xcite ) . in particular , we will analyze how changing the value of @xmath100 the number of the critical points of the dynamical system , as well as their linear stability properties , change . for simplicity , we shall perform the analysis in the @xmath7 case . in this case we have two profiles , @xmath104 and @xmath105 , and the equations of motion are given by : @xmath106 & = 0\ , \nonumber \\ \overset{{\cdot\cdot}}{f_{1}}\left ( \frac{2}{3}+\frac{1}{2r^{2}}\left ( 1-\text{cos}f_{1}\right ) \right ) + \frac{1}{3}\overset{{\cdot\cdot}}{f_{0}}+\frac{\text{sin}f_{1}}{4r^{2}}\left [ \overset{{\cdot}}{f_{1}}^{2}-4-\frac{2\left ( 1-\text{cos}f_{1}\right ) } { r^{2}}+\frac{\left ( 1-\text{cos}f_{0}\right ) } { r^{2}}\right ] & = 0\ . \label{eqmn3}\end{aligned}\ ] ] notice that the two equations go one into the other by exchanging @xmath104 and @xmath105 . we are going to exploit this symmetry . in order to study critical points of the system ( [ eqmn3 ] ) , first of all we rewrite it as a system of first order equations as follows : @xmath107 where @xmath108 , \quad \vec{g}\left(\vec{z}\right ) = \left[\begin{array}{c } -\hat a^{-1}\left[\begin{array}{c } q(f_0,f_1,\delta ) \\ q(f_1,f_0,e ) \end{array}\right ] \\ \delta \\e \end{array}\right],\ ] ] and we introduced the following notations for brevity @xmath109 , \quad q(\xi , \eta,\theta ) \equiv \frac{\text{sin } \xi } { 4r^{2}}\left [ \theta^2 -4-\frac{2\left ( 1-\text{cos } \xi \right ) } { r^{2}}+\frac{\left ( 1-\text{cos}\eta\right ) } { r^{2}}\right].\ ] ] stationary points @xmath110 are defined by vanishing of the vector field @xmath111 , which defines the flow of the autonomous equation in phase space : @xmath112 one can easily check that @xmath113 , therefore @xmath110 can be written as follows : @xmath114\ ] ] where @xmath115 and @xmath116 are solutions of the following algebraic equations @xmath117 equations imply one of the three possibilities ( the fourth possibility does not have real solutions ) : 1 . @xmath118 ; @xmath119 ; 2 . @xmath119 ; @xmath120 ; 3 . @xmath118 ; @xmath121 . linearizing in the vicinity of the critical point @xmath110 we obtain @xmath122 where @xmath123 is the jacobian matrix of the diffeomorphism @xmath124 taken at the point @xmath110 @xmath125 in order to analyze stability of the critical point @xmath110 i.e. in order to understand whether phase trajectories starting near this point run away " exponentially or not , one has to check whether eigenvalues of @xmath123 have real part or they are purely imaginary . below we present a complete analysis of the critical points . since in the algebraic equations the unknowns @xmath126 and @xmath127 enter via sine and cosine , it is sufficient to consider critical points modulo @xmath128 . we identify the following four independent subcases : @xmath129 @xmath130 ; @xmath129 @xmath131 ; @xmath129 @xmath132 ; @xmath129 @xmath133 . in the ( equal profile ) case @xmath134 jacobian matrix is @xmath135\ ] ] whose four eigenvalues are : @xmath136 therefore this fixed point is unstable ; this property does not depend on @xmath97 . also in this case we have equal profiles @xmath137 . in this case the jacobian matrix is @xmath138\ ] ] and the eigenvalues @xmath139 are purely imaginary , therefore this point is stable for all @xmath97 . now we consider @xmath140 , @xmath141 . in this case the jacobian matrix @xmath142\ ] ] has the eigenvalues @xmath143 unlike the previous cases , now the properties of this critical point change at a critical value of the parameter @xmath97 . indeed , while the eigenvalues @xmath144 and @xmath145 are always purely imaginary , @xmath146 and @xmath147 are purely imaginary only when @xmath148 . in this regime this critical point is stable . for @xmath149 instead the first two eigenvalues become real , therefore the fixed point becomes unstable . this is a nice example of how the stability properties change at a critical value of the parameter . interestingly , the value of @xmath97 at which the transition happens is the same below which new fixed points appear ( i.e. those occurring in cases 2 and 3 described above ) . in this case the equations are exactly the same of case 3 , only with @xmath152 and @xmath153 exchanged . the eigenvalues are therefore exactly the same so that the same discussion applies . we now consider cases 2 and 3 in the above list . since they are related by the exchange @xmath154 , they will have the same eigenvalues and hence the same stability properties . let us then consider case 2 : @xmath155 one may immediately check that a real solution only exists for @xmath156 , and it is given by @xmath157 modulo @xmath128 . for @xmath158 this critical point disappears . the jacobian matrix reads @xmath159\ ] ] and its eigenvalues are : @xmath160 in the whole range @xmath161 , the first two eigenvalues are purely imaginary , while @xmath144 and @xmath145 are real , therefore this critical point is unstable . at @xmath162 , @xmath144 and @xmath145 vanish . the analysis we just performed discloses the crucial dependence of the dynamics of our system on the parameter @xmath97 . in fact we saw that there is a critical value @xmath162 at which the nature and the number itself of the critical points change . for @xmath163 and for @xmath158 we have two _ qualitatively _ different phase portraits , thereby different dynamics . these are genuine finite volume effects ( since the parameter @xmath97 represents the available surface per baryon within the tube - shaped region ) which can be disclosed only within the present framework . in fig.[f1 ] we present location of critical points for @xmath164 and @xmath158 . as one can easily see , the two plots are qualitatively different : while the latter grid is invariant under the discrete group of reflections @xmath165 about the axes @xmath126 and @xmath127 , the former is not , that is the symmetry is broken . moreover the former is sensitive to the particular value of @xmath97 , while the latter is always the same for all @xmath158 . finally , the concentration of different critical points changes as well , since some of them changed color ( i.e. type ) when @xmath97 passed through the critical value , while new ( green ) fixed points appeared . these metamorphoses of the phase portrait hint at the presence of some sort of phase transition occurring at strong coupling ( recall that @xmath166 plays the rle of a coupling constant ) . it is interesting to notice that , in our units , this critical value is of the order of the @xmath35 , so although irrelevant from a flat limit perspective , it may have important consequences in view of applications to nuclear physics , where the dimensions of the tube - shaped region should be precisely of that order of magnitude in order to model an atomic nucleus @xcite . on the left and for @xmath167 on the right . in the latter case the grid is independent on a particular choice of @xmath97 , while for the former case we chose @xmath168 . blue points represent stable fixed points , red points represent absolutely " unstable fixed points ( i.e. all eigenvalues of the corresponding jacobi matrix are real ) , while green points represent softer " unstable fixed points ( i.e two eigenvalues values are real and two imaginary ) . [ f1 ] , title="fig : " ] on the left and for @xmath167 on the right . in the latter case the grid is independent on a particular choice of @xmath97 , while for the former case we chose @xmath168 . blue points represent stable fixed points , red points represent absolutely " unstable fixed points ( i.e. all eigenvalues of the corresponding jacobi matrix are real ) , while green points represent softer " unstable fixed points ( i.e two eigenvalues values are real and two imaginary ) . [ f1 ] , title="fig : " ] in this section we perform a numerical study of the system in the @xmath169 case . in this case we have three profiles , @xmath170 , and the equations of motion reduce to : @xmath171 = 0,\\ & \overset{{\cdot\cdot}}{f_{1}}\left ( 1+\frac{2}{r_{0}^{2}}\left ( 1-\text{cos}f_{1}\right ) \right ) + \frac{\overset{{\cdot\cdot}}{f_{0}}+\overset{{\cdot\cdot}}{f_{2}}}{2 } + \\ & + \frac{\text{sin}f_{1}}{r_{0}^{2}}\left [ \overset{{\cdot}}{f_{1}}^{2}-4- \frac{8\left ( 1-\text{cos}f_{1}\right ) } { r_{0}^{2}}+ \frac{3\left ( 1-\text{cos}f_{0}\right ) } { r_{0}^{2}}+ \frac{3\left ( 1-\text{cos}f_{2}\right ) } { r_{0}^{2}}\right ] = 0,\\ & \overset{{\cdot\cdot}}{f_{2}}\left ( 1+\frac{2}{r_{0}^{2}}\left ( 1-\text{cos}f_{2}\right ) \right ) + \frac{\overset{{\cdot\cdot}}{f_{0}}+2\overset{{\cdot\cdot}}{f_{1}}}{3 } + \frac{\text{sin}f_{2}}{r_{0}^{2}}\left [ \overset{{\cdot}}{f_{2}}^{2}-4- \frac{6\left ( 1-\text{cos}f_{2}\right ) } { r_{0}^{2}}+ \frac{4\left ( 1-\text{cos}f_{1}\right ) } { r_{0}^{2}}\right ] = 0.\end{aligned}\ ] ] the first and the second go one onto the other upon exchanging @xmath172 . we can exploit this symmetry to look for solutions which have @xmath173 ( another possibility allowed by the symmetry is to set @xmath174 and at the same time @xmath175 , however this would give vanishing winding number , so we do not consider it ) . let us also call @xmath176 . in this case we get the two independent equations : @xmath177 = 0,\\ \overset{{\cdot\cdot}}{g}\left ( 1+\frac{2}{r_{0}^{2}}\left ( 1-\text{cos}g\right ) \right ) + \overset{{\cdot\cdot}}{f}+ \frac{\text{sin}g}{r_{0}^{2}}\left [ \overset{{\cdot}}{g}^{2}-4- \frac{8\left ( 1-\text{cos}g\right ) } { r_{0}^{2}}+ \frac{6\left ( 1-\text{cos}f\right ) } { r_{0}^{2}}\right ] = 0,\end{aligned}\ ] ] and the energy density ( which is defined by @xmath178 ) is given by : @xmath179 + \nonumber\\ & + \frac{1}{r_{0}^{4}}\left [ 9\left ( 1-\text{cos}f\right ) ^{2 } -12 \left ( 1-\text{cos}f\right ) \left ( 1-\text{cos}g\right ) + 8\left ( 1-\text{cos}g\right ) ^{2}\right ] .\end{aligned}\ ] ] before solving the system we must choose a set of boundary conditions to impose on @xmath180 and @xmath181 in such a way to get solutions with a nontrivial winding number . recall that we have to impose ( anti-)periodic boundary conditions on the field @xmath182 : @xmath183 for simplicity we shall limit ourselves to the periodic case . according to the ansatz we are using the field can be written as : @xmath184 it is convenient to use the second form . we notice that the matrix @xmath182 is diagonal in the basis @xmath185 , therefore the above boundary condition is equivalent to its diagonal matrix elements . the diagonal matrix elements of @xmath182 are @xmath186 thus the boundary conditions are satisfied if @xmath187 with these conditions the winding number is given by @xmath188 we also need to impose conditions on the first derivative of @xmath182 , i.e. of @xmath180 and @xmath181 . we choose the simplest possibility which is compatible with the periodicity of the first derivative of @xmath182 , i.e. @xmath189 we have solved numerically the system ( [ su(4)eqs ] ) with the boundary conditions ( [ su(4)bcs1 ] ) and ( [ su(4)bcs2 ] ) , for some values of the integers @xmath190 and @xmath41 , taking @xmath191 , and we computed the energies of the solutions . the results are summarized in the tables below and in the one in sect . [ eqprofsubsect ] . we have allowed @xmath190 and @xmath41 to vary in a small range , beyond which our numerical procedure is not very accurate . [ c]p1cmp1cmp1cmp1cm|p1cmp1cmp1cmp1cm|p1cmp1cmp1cmp1cm@xmath190 & @xmath41 & @xmath38 & @xmath193 & @xmath190 & @xmath41 & @xmath38 & @xmath193 & @xmath190 & @xmath41 & @xmath38 & @xmath193 + -1 & 1 & 2 & 26.68 & 1 & 0 & 6 & 11.43 & 1 & 1 & 14 & 43.87 + -2 & 2 & 4 & 100.65 & 2 & 0 & 12 & 42.41 & 2 & 2 & 28 & 168.31 + -3 & 3 & 6 & 228.42 & 3 & 0 & 18 & 92.08 & 3 & 3 & 42 & 376.07 + & & & & 4 & 0 & 24 & 161.61 & & & & + the energies are exactly the same if we change sign simultaneously to @xmath190 and @xmath41 . there are sectors in which , to a very good approximation , the energy grows like the square of the modulus of the winding number , with different coefficients in different sectors . however , this is not true in all sectors , for example : [ c]p1cmp1cmp1cmp1cm@xmath190 & @xmath41 & @xmath38 & @xmath193 + 0 & 1 & 8 & 24.48 + 0 & 2 & 16 & 93.20 + we notice that the solutions with the smallest winding numbers are not energetically favored , since to achieve small winding numbers both profiles must grow going along @xmath194 in opposite directions , while the smallest energy solutions are those where only one of the profile grows while the other oscillates and stays small , i.e. _ either _ @xmath190 _ or _ @xmath41 are different from zero , but not both . notice also that increasing @xmath41 costs much more than increasing @xmath190 , despite their contributions to the winding number being not so different . in figs.1 - 6 the plots of some solutions are reported . [ m-1n1 ] [ m-2n2 ] [ m1n0 ] [ m2n0 ] [ m1n1 ] [ m2n2 ] if we impose that the profiles are equal , i.e @xmath195 , the equations of motion reduce of course to ( [ eqprofeq ] ) . in this case the same procedure followed above tells us that the boundary conditions to achieve a periodic @xmath182 are @xmath196 in which case the winding number is simply @xmath197 in order to compare the equal profile situation with the non equal profile one we have to consider cases with the same winding number . this is achieved if in the latter case we have @xmath198 . we did the computation in the two simplest cases , with @xmath191 , and the results are : [ c]p1cmp1cmp1cm|p1cmp1cmp1cmp1@xmath199 & @xmath200 & @xmath201 & @xmath190 & @xmath41 & @xmath38 & @xmath193 + 1 & 20 & 82.67 & 2 & 1 & 20 & 82.67 + 2 & 40 & 325.70 & 4 & 2 & 40 & 325.57 + we see that the energies are the same ( in the second case the discrepancy is due to small numerical inaccuracies : the numerical computation for the @xmath202 , @xmath203 case gave the winding number of the solutions to be 40.13 , thus signalling such inaccuracies ) . therefore , in topological sectors whose winding number is compatible with equal profile configurations , with the chosen values of the parameters , equal - profile solutions and non - equal profile ones are energetically equivalent . besides the above analysis , which was performed at fixed @xmath6 , we also studied what happens when this parameter is varied , considering for definiteness the lowest winding number case , i.e @xmath204 . we checked that when @xmath6 is increased the solutions tend to linear functions , in agreement with triviality of the dynamics for large radius . instead , for values of @xmath6 ( or @xmath89 if we allow also @xmath26 to vary ) substantially smaller than @xmath10 the numerical procedure becomes unstable and the results unreliable . this is of course to be expected since lowering @xmath6 the non - linearity of the problem increases . this is in agreement with the bounds ( [ ban10.5]-[ban11.7 ] ) , which are more difficult to satisfy when @xmath6 is small , and trivial when @xmath6 is large . in the small @xmath6 case a more sophisticated numerical analysis would therefore be needed . in particular it would be interesting to see whether at strong coupling non - equal profile solutions are energetically favored with respect to equal profile ones with the same winding number , as suggested by the fixed point analysis of section [ fpoints ] . another interesting issue requiring a more accurate numerical work is the identification of a range of the parameters in which the dependence of the energy on the winding number is linear rather than quadratic . we hope to come back on these interesting issues in a future investigation . in this paper the @xmath5 skyrme model is studied in a tube - shaped geometry which allows to consider finite volume situations . the use of such a geometry allows to introduce a parameter which regulates the dynamics as a new ( inverse ) coupling constant . these simplifications , which make the system autonomous , allow us to perform a detailed study of the system using several techniques , i.e. dynamical systems , numerical solutions and rigorous existence theorems . all these analyses confirm the relevance of the parameter @xmath6 introduced by the metric on the dynamics of the system . as a further bonus , when combined with @xmath2 as @xmath205 , this parameter defines an effective t hooft coupling . in the large @xmath2 limit in which such coupling is kept fixed , flat space - time is recovered . this effective t hooft parameter determines the discrete symmetries of the skyrmions configurations . these results clearly show that the present framework is a very promising tool to study interacting skyrmions systems . a very interesting problem which we hope to analyse in a future publication is to extend the present analysis at finite temperature and/or chemical potential . the authors would like to thank j. zanelli for enlightening discussions and comments . this work has been funded by the fondecyt grant 1120352 . the centro de estudios cientficos ( cecs ) is funded by the chilean government through the centers of excellence base financing program of conicyt . m. k. acknowledges partial support from unina and compagnia di san paolo in the framework of the program star 2013 . in this appendix we prove analytically that non - trivial solutions which are not trivial embeddings of @xmath4 into @xmath5 ( in which the profiles are not proportional ) do indeed exist . we will focus for simplicity on the @xmath7 case but the same argument can be easily extended to the general case . the basic mathematical tool of this section is a well - known result of nonlinear functional analysis , the _ schauder theorem _ ( see for a detailed pedagogical review on mathematical tools to deal with non - linear partial differential equations @xcite ) . the statement of the schauder theorem ( @xcite ) is the following . let @xmath206 be a banach space between any pair of elements of the space is defined , and it is called complete if ( with respect to the chosen metric ) from every cauchy sequence one can extract a convergent subsequence ( see , for instance , @xcite ) . ] . let @xmath207 a bounded closed convex set in @xmath206 , and let @xmath208 be a compact operator from a banach space @xmath206 into itself ( see , for a detailed discussion , @xcite ) is called _ compact _ if and only if , for any bounded sequence @xmath209 , the sequence @xmath210 has a convergent subsequence . ] from the banach space @xmath206 into itself such that @xmath208 maps @xmath207 into itself:@xmath211 : c\rightarrow c\ . \label{ban2}\ ] ] then the map @xmath212 $ ] has ( at least ) one fixed point in @xmath207 . in other words , under the above hypotheses , there exists a solution to the equation @xmath213 = x\ . \label{ban3}\ ] ] let us recall the equations of motion for the @xmath7 case : @xmath106 & = 0\ , \label{su(3)eq1}\\ \overset{{\cdot\cdot}}{f_{1}}\left ( \frac{2}{3}+\frac{1}{2r^{2}}\left ( 1-\text{cos}f_{1}\right ) \right ) + \frac{1}{3}\overset{{\cdot\cdot}}{f_{0}}+\frac{\text{sin}f_{1}}{4r^{2}}\left [ \overset{{\cdot}}{f_{1}}^{2}-4-\frac{2\left ( 1-\text{cos}f_{1}\right ) } { r^{2}}+\frac{\left ( 1-\text{cos}f_{0}\right ) } { r^{2}}\right ] & = 0\ . \label{su(3)eq2}\ ] ] we want to prove that the system in eqs . ( [ su(3)eq1 ] ) and ( [ su(3)eq2 ] ) admits non - trivial solutions in which the profiles @xmath104 and @xmath105 are not proportional . in order to achieve this goal , let us rewrite it as coupled integral equations ( for notational simplicity in this section we shall consider @xmath214 $ ] instead of @xmath215 $ ] , since everything just depends on the length @xmath26 of the tube ) : where @xmath218 \ , \ \label{int1.35}\\ h_{1}\left ( \rho\right ) & = \left [ \overset{{\cdot}}{f_{1}}^{2}-4-\frac{2\left ( 1-\text{cos}f_{1}\left ( \rho\right ) \right ) } { r^{2}}+\frac{\left ( 1-\text{cos}f_{0}\left ( \rho\right ) \right ) } { r^{2}}\right ] , \label{int2.35}\\ a_{i } & = \left ( \frac{2}{3 } + \frac{1}{2\,r^{2}}\left ( 1-\cos\,f_{i}\right ) \right ) ^{-1 } , \quad i=0,1\end{aligned}\ ] ] where @xmath219 , @xmath220 , and @xmath221 and @xmath222 represent the initial data for the two profiles @xmath104 and @xmath105 and their derivatives at @xmath223 . it is a trivial computation to show that the above system of integral equation is equivalent to the system in eqs . ( [ su(3)eq1 ] ) and ( [ su(3)eq2 ] ) . the system in eqs . ( [ int1 ] ) and ( [ int2 ] ) can be written as a fixed point condition for the following vectorial operator @xmath224 acting component - wise on pairs of @xmath225 functions @xmath226:@xmath227 \times c^{2}\left [ 0,l\right ] \rightarrow c^{2}\left [ 0,l\right ] \times c^{2}\left [ 0,l\right ] \ , \\ \overrightarrow{f}\left ( x\right ) & = \left ( f_{0}\left ( x\right ) , f_{1}\left ( x\right ) \right ) \in c^{2}\left [ 0,l\right ] \times c^{2}\left [ 0,l\right ] \ , \end{aligned}\]]@xmath228 = \overrightarrow{t}\left [ \ \overrightarrow{f}\left ( x\right ) \right ] = \left ( t_{0}\left ( x\right ) , t_{1}\left ( x\right ) \right ) \ , \label{int2.5}\ ] ] with @xmath229 where the functions @xmath230 are defined in eqs . ( [ int1.25 ] ) , ( [ int2.25 ] ) , ( [ int1.35 ] ) and ( [ int2.35 ] ) . it is then obvious that the system in eqs . ( [ int1 ] ) and ( [ int2 ] ) can be written as the following fixed - point condition@xmath231 \ , \label{fixedpoint}\ ] ] where the operator @xmath224 has been defined in eqs . ( [ int2.5 ] ) , ( [ int3 ] ) and ( [ int4 ] ) . hence , now the task is to define @xmath224 as a compact operator from a bounded closed convex sub - set of a banach space into itself . in order to achieve this goal , first of all let us define the following metric into the space @xmath232 \times c^{2}\left [ 0,l\right ] $ ] : @xmath233 } { \sup}\left\vert f_{0}(x)-g_{0}(x)\right\vert + \underset{x\in\left [ 0,l\right ] } { \sup}\left\vert \frac{df_{0}}{dx}-\frac{dg_{0}}{dx}\right\vert + \underset{x\in\left [ 0,l\right ] } { \sup } \left\vert \frac{d^{2}f_{0}}{dx^{2}}-\frac{d^{2}g_{0}}{dx^{2}}\right\vert + \nonumber\\ & + \underset{x\in\left [ 0,l\right ] } { \sup}\left\vert f_{1}(x)-g_{1}(x)\right\vert + \underset{x\in\left [ 0,l\right ] } { \sup}\left\vert \frac{df_{1}}{dx}-\frac{dg_{1}}{dx}\right\vert + \underset{x\in\left [ 0,l\right ] } { \sup}\left\vert \frac{d^{2}f_{1}}{dx^{2}}-\frac{d^{2}g_{1}}{dx^{2}}\right\vert \ , \label{ban4.01}\\ \overrightarrow{f}\left ( x\right ) & = \left ( f_{0}\left ( x\right ) , f_{1}\left ( x\right ) \right ) \ , \ \overrightarrow{g}\left ( x\right ) = \left ( g_{0}\left ( x\right ) , g_{1}\left ( x\right ) \right ) \ \in c^{2}\left [ 0,l\right ] \times c^{2}\left [ 0,l\right ] \ .\nonumber\end{aligned}\ ] ] with respect to this metric , which is induced by a norm , the space @xmath232 \times c^{2}\left [ 0,l\right ] $ ] is a banach space which we will call @xmath206 . _ the next task _ to apply the schauder theorem is to define a bounded closed convex sub - set @xmath207 of the banach space defined above ( using the metric in eq . ( [ ban4.01 ] ) ) such that @xmath224 maps @xmath207 into itself . let us define @xmath207 as@xmath234 \ \right . \ \left\vert f_{0}(x)-a_{0}\right\vert \leq b\ , \ \left\vert f_{1}(x)-a_{1}\right\vert \leq b\ \ , \label{ban5}\\ & \left . \ \left\vert \frac{df_{0}}{dx}\right\vert \leq b\ , \ \ \left\vert \frac{d^{2}f_{0}}{dx^{2}}\right\vert \leq b\ , \ \ \left\vert \frac{df_{1}}{dx}\right\vert \leq b\ , \ \left\vert \frac{d^{2}f_{1}}{dx^{2}}\right\vert \leq b\right\ } \ , \quad\ \ \ b\in\mathbb{r } _ { + } \ , \nonumber\end{aligned}\ ] ] where @xmath221 are the initial data appearing in eqs . ( [ int1 ] ) and ( [ int2 ] ) so that @xmath207 is closed by definition . it is easy to see that @xmath207 is bounded as well , since from the definition of @xmath207 it follows that@xmath235 which clearly implies that there exists @xmath236 such that @xmath237 , given any two functions @xmath238 . it remains to prove that @xmath207 is convex , i.e. to check that if @xmath239 and @xmath240 both belong to @xmath207 then also @xmath241 belongs to @xmath207 , @xmath242 $ ] . this can be seen as follows:@xmath243 while the conditions on the derivatives are trivially satisfied . from the definition of @xmath207 in eq . ( [ ban5 ] ) we deduce the inequalities @xmath245 } { \sup}\left\vert h_{0}\left ( x\right ) \right\vert & \leq l\left\ { \frac{b}{2 } + \frac{3}{8r^{2}}\left [ b^{2}+4+\frac{6}{r^{2}}\right ] \right\ } \ , \label{useful1}\\ \underset{x\in\left [ 0,l\right ] } { \sup}\left\vert h_{1}\left ( x\right ) \right\vert & \leq l\left\ { \frac{b}{2 } + \frac{3}{8r^{2}}\left [ b^{2}+4+\frac{6}{r^{2}}\right ] \right\ } \ , \label{useful2}\ ] ] @xmath245 } { \sup}\left\vert \frac{dh_{0}}{dx}\right\vert & \leq\frac{b}{2 } + \frac{3}{8r^{2}}\left [ b^{2}+4+\frac { 6}{r^{2}}\right ] \ , \label{useful3}\\ \underset{x\in\left [ 0,l\right ] } { \sup}\left\vert \frac{dh_{1}}{dx}\right\vert & \leq\frac{b}{2 } + \frac{3}{8r^{2}}\left [ b^{2}+4+\frac { 6}{r^{2}}\right ] \ , \label{useful4}\ ] ] ( where we used the fact that @xmath246 ) , which will be needed later . _ the next task _ to apply the schauder theorem is to prove that @xmath224 is a compact operator . to do this we advocate the _ ascoli - arzel theorem _ , which states that if a sequence of functions ( defined on a compact metric space ) @xmath247 is uniformly bounded and equicontinuous , then a convergent subsequence can be extracted from it is uniformly bounded if @xmath248 , where @xmath236 does not depend on @xmath41 ; @xmath249 is said to be equicontinuous if , given @xmath250 , @xmath251 such that @xmath252 whenever @xmath253 _ and , moreover , _ @xmath152 _ does not depend on _ @xmath41 ( otherwise the sequence would be continuous but not equicontinuous : see @xcite ) . ] . thus compactness of @xmath224 is equivalent to the statement that if @xmath254 is a sequence in @xmath207 , then the sequence @xmath255 = \left ( t_{0}\left ( n;x\right ) , t_{1}\left ( n;x\right ) \right ) $ ] is uniformly bounded and equicontinuous . uniform boundedness can be proved as follows:@xmath256a_0+b_0l+lh_0 ( n;x ) . and therefore , from eq . ( [ useful1 ] ) , one gets:@xmath257 \right\ } \ , \label{ban10}\ ] ] and , similarly , for @xmath258 : @xmath259 \right\ } \ . \label{ban11}\ ] ] now , the sequence of images @xmath260 $ ] has to belong to @xmath207 as well . as it always happens ( see @xcite and @xcite ) this will give some constraints on the range of the initial data as well as on the parameters @xmath261 , @xmath26 and @xmath97 since , from the definition of @xmath207 , one has to require that for any @xmath41 @xmath262 consequently , as it can be easily seen comparing eqs . ( [ cons1 ] ) , ( [ cons2 ] ) , ( [ ban5 ] ) , ( [ ban6 ] ) and ( [ ban7 ] ) with eqs . ( [ useful1 ] ) , ( [ useful2 ] ) , ( [ useful3 ] ) , ( [ useful4 ] ) , ( [ ban10 ] ) and ( [ ban11 ] ) , the following constraints arise:@xmath263 \right\ } & \leq b\ , \label{ban10.5}\\ \left\vert b_{1}l\right\vert + l^{2}\left\ { \frac{b}{2 } + \frac{3}{8r^{2}}\left [ b^{2}+4+\frac{6}{r^{2}}\right ] \right\ } & \leq b\ , \label{ban11.5}\]]@xmath264 \right\ } & \leq b\ , \label{ban10.6}\\ \left\vert b_{1}\right\vert + l\left\ { \frac{b}{2 } + \frac{3}{8r^{2}}\left [ b^{2}+4+\frac{6}{r^{2}}\right ] \right\ } & \leq b\ , \label{ban11.6}\]]@xmath265 & \leq b\ \ , \label{ban10.7}\\ \frac{b}{2 } + \frac{3}{8r^{2}}\left [ b^{2}+4+\frac{6}{r^{2}}\right ] & \leq b\ . \label{ban11.7}\ ] ] therefore , in order for this theorem to work , the length @xmath26 of the tube - shaped region in which these multi - skyrmions are living can not exceed the bounds defined in eqs . ( [ ban10.5 ] ) and ( [ ban11.5 ] ) ( indeed , if @xmath26 is too large , the bounds will be violated at a certain point can not be arbitrarily big for the theorem to hold can be also seen by observing that in the limit @xmath266 the domain of definition of the functions would not be compact any more , thus invalidating the ascoli - arzel theorem ] . ) . it is also to be noticed that one can not obtain a very large value for the allowed @xmath26 by increasing @xmath261 since the left hand sides of eqs . ( [ ban10.5 ] ) and ( [ ban11.5 ] ) increase faster than the right hand sides . moreover , the situation gets worse if @xmath100 is very small ( as all the above inequalities are violated at a certain point ) . however , if @xmath97 is large ( namely , in the flat limit ) , eqs . ( [ ban10.7 ] ) and ( [ ban11.7 ] ) are always satisfied and eqs . ( [ ban10.6 ] ) and ( [ ban11.6 ] ) become mild constraints on the initial data . on the other hand , it is trivial to see that it is always possible to choose the initial data and @xmath261 , @xmath26 and @xmath97 in such a way that all the above inequalities are fulfilled . _ the second step _ to prove that @xmath208 is compact is to show that if @xmath267 is a sequence in @xmath207 then the sequence @xmath268 $ ] is _ equicontinuous_. to show this , we must evaluate , _ for a generic _ @xmath41 , the absolute values of following differences:@xmath269 where @xmath270 . after trivial manipulations ( which use the fact that all the functions @xmath271 belong to @xmath207 and consequently eqs.([ban6 ] ) and ( [ ban7 ] ) are satisfied ) one arrives at @xmath272 \right\ } \right ] \ , \label{ban14}\\ \left\vert t_{1}\left ( n;x\right ) -t_{1}\left ( n;y\right ) \right\vert & \leq\left\vert x - y\right\vert \left [ \left\vert b_{1}\right\vert + l\left\ { \frac{1}{2}b+ \frac{3}{8r^{2}}\left [ b^{2}+4+\frac{6}{r^{2}}\right ] \right\ } \right ] \ , \label{ban15}\ ] ] in summary , eqs . ( [ ban10 ] ) , ( [ ban11 ] ) , ( [ ban10.5 ] ) , ( [ ban11.5 ] ) , ( [ ban10.6 ] ) , ( [ ban11.6 ] ) , ( [ ban10.7 ] ) and ( [ ban11.7 ] ) show that , if @xmath275 is any sequence in @xmath207 , then the sequence @xmath276 $ ] is uniformly bounded in @xmath207 . secondly , eqs . ( [ ban14 ] ) , ( [ ban15 ] ) , ( [ ban16 ] ) and ( [ ban18 ] ) show that , if @xmath267 is any sequence in @xmath207 , then the sequence @xmath277 $ ] is _ equicontinuous_. consequently , by virtue of the ascoli - arzel theorem , from any sequence @xmath278 $ ] one can extract a convergent subsequence : this , together with the bounds ( [ ban10.5])-([ban11.7 ] ) , implies that the operator @xmath224 is a compact operator from a bounded closed convex set into itself . thus , it is possible to apply the schauder theorem , which ensures that eq . ( [ fixedpoint ] ) ( which is equivalent to our original system ) has at least one solution . moreover , it is always possible to choose appropriately the inital data @xmath221 and @xmath222 in such a way that the two profiles are not proportional . the conclusion is that not only one can construct numerical solutions ( which are very interesting by themselves ) , as we do in the main text , but also one can prove analytically that non - trivial multi - skyrmions in which the profiles are not proportional do indeed exist . besides the intrinsic mathematical elegance of the fixed - point schauder - type argument , the present procedure also discloses the presence of the bounds in eqs.([ban10.5])-([ban11.7 ] ) on the radius @xmath97 and on the length @xmath26 of the tube - shaped region in which these multi - skyrmions are living ( which constrain its shape ) , as well as on the other parameters of the model . at the present stage of the analysis , it is not possible yet to say whether such a bound is just a limitation of the method or it signals some deeper physical limitation on the volume of the regions in which one constrains these skyrmions to live . however , to understand whether or not @xmath5 multi - skyrmions can fit into very large tube - shaped regions is certainly a very interesting and deep question on which we hope to come back in a future investigation . f. jonietz , s. mhlbauer , c. pfleiderer , a. neubauer , w. mnzer , a. bauer , t. adams , r. georgii , p. bni , r. a. duine , k. everschor , m. garst , a. rosch , _ science _ * 330 * ( 2010 ) , 1648 - 1651 s. seki , x. z. yu , s. ishiwata , y. tokura , _ science _ * 336 * ( 2012 ) , 198 - 201 .

motivated by recent experimental results for the step sizes of dynein motor proteins , we develope a cellular automata model for intra - cellular traffic of dynein motors incorporating special features of the hindrance - dependent step size of the individual motors . we begin by investigating the properties of the aggressive driving model ( adm ) , a simple cellular automata - based model of vehicular traffic , a unique feature of which is that it allows a natural extension to capture the essential features of dynein motor traffic . we first calculate several collective properties of the adm , under both periodic and open boundary conditions , analytically using two different mean - field approaches as well as by carrying out computer simulations . then we extend the adm by incorporating the possibilities of attachment and detachment of motors on the track which is a common feature of a large class of motor proteins that are collectively referred to as cytoskeletal motors . the interplay of the boundary and bulk dynamics of attachment and detachment of the motors to the track gives rise a phase where high and low density phases separated by a stable domain wall coexist . we also compare and contrast our results with the model of parmeggiani et . al . ( phys . rev . lett . * 90 * , 086601 ( 2003 ) ) which can be regarded as a minimal model for traffic of a closely related family of motor proteins called kinesin . finally , we compare the transportation efficiencies of dynein and kinesin motors over a range of values of the model parameters .

cond-mat0607322

molecular motors are protein molecules that drive a wide range of intra - cellular activities including transport of molecular cargo @xcite . there are many similarities between collective molecular motor transport and vehicular traffic @xcite . in recent years non - equilibrium statistical mechanics has found unusual application in research on traffic flow of various different types of objects starting from objects as small as molecular motors to macroscopic objects like vehicles @xcite . analytical as well as numerical techniques of the statistical physics are being used to understand rich variety of physical phenomena exhibited by traffic systems . some of these phenomena , observed under different circumstances , include phase transitions , criticality and self - organized criticality , metastability and hysteresis , phase - segregation , etc . a common modeling strategy is to represent the motile objects ( e.g. , a vehicle or a molecular motor ) by a self - propelled particle , ignoring its structural details , and then treating the traffic as a system of interacting particles driven far from equilibrium . these models belong to a class of non - equilibrium systems called _ driven - diffusive lattice gases _ @xcite . in most of these traffic models the dynamics of the particles is formulated using the language of _ cellular automata _ ( ca ) @xcite . to our knowledge , the first model for molecular motor traffic was formulated in 1968 in the context of collective movement of ribosomes on messenger rna track @xcite . in recent years several groups have independently developed a class of minimal generic models for traffic of molecular motors which move on tracks that are filamentary proteins . all these models are essentially extensions of the totally asymmetric simple exclusion process ( tasep ) @xcite which is one of the simplest models of driven diffusive lattice gas systems . in these models @xcite the molecular motors are represented by particles whereas the sites for the binding of the motors with the tracks are represented by a one - dimensional discrete lattice . just as in tasep , the motors are allowed to hop forward , with probability @xmath0 , provided the site in front is empty . however , unlike tasep , the particles can also get `` attached '' to an empty lattice site , with probability @xmath1 , and `` detached '' from an occupied site , with probability @xmath2 from any site except the end points . parmeggiani et al . @xcite demonstrated a novel phase where low and high density regimes , separated from each other by domain walls , coexist . they interpreted this spatial organization as traffic jam of molecular motors . none of the models of molecular motor traffic mentioned above distinguish between kinesins and dyneins which form the two superfamilies of motor proteins that move on the same type of tracks , namely , microtubules . on the other hand , detailed experiments over the last two years have established that , in contrast to kinesins , dyneins can take steps of four different sizes depending on the opposing force or hindrance . one of the aims of this paper is to introduce a minimal model that distinguishes between these two features of kinesin and dynein motors . in this paper we begin by investigating the aggressive driving model ( adm ) , a stochastic ca model for traffic flow that is closely related to the nagel - schreckenberg ( nasch ) model @xcite . one of the reasons for studying this model is that it allows natural extensions so as to capture the essential features of dynein motor traffic including the unique features of dynein stepping ( which we shall explain in section [ dyneinexperiment ] ) . besides , the adm model is an interesting model of vehicular traffic in its own right and is also related to the fukui - ishibashi ( fi ) model @xcite . however , in contrast to the fi model , it still shows spontaneous jam formation . we investigate the properties of the adm by approximate analytical calculations as well as by computer simulations . then , we use an extended version , which we refer to as the dynein traffic model ( dtm ) , for a quantitative desciption of intra - cellular traffic of dynein motors . the paper is organized as follows . in the next section we describe the adm and the method of simulation . in section [ sec3 ] we investigate the properties of the adm with periodic boundary conditions and we describe the analytical theories for calculating its flow properties . we present a comparison of the adm with nasch model at the end of section [ sec3 ] . in section [ sec4 ] we investigate the density profiles and phase diagram of the adm with open boundary conditions . in section [ sec5 ] we describe the experimentally observed hindrance - dependence of the step sizes of dynein motors and introduce the dynein traffic model ( dtm ) . we present the results for the dtm with periodic boundary conditions in section [ sec6 ] and those under open boundary conditions in section [ sec7 ] . finally we summarize the main results and the conclusions in section [ sec8 ] . in the cellular automata model of aggressive driving a lane is represented by a one - dimensional lattice . the boundary conditions may be periodic or open . each of the lattice sites represents a cell that can be either empty or occupied by at most one vehicle at a time . the speed @xmath3 of each vehicle can take one of the allowed integer values @xmath4 . let @xmath5 and @xmath6 be the position and speed , respectively , of the @xmath7th vehicle . then we define the ( distance ) headway of the @xmath7th vehicle at time @xmath8 by @xmath9 , i.e. as the number of empty cells in front of this car . at each time step @xmath10 the state of all vehicles on this 1-d lattice is updated in _ parallel _ according to the following rules : + @xmath11 : _ acceleration : _ if @xmath12 then @xmath13 and if @xmath14 then @xmath15 , that is , @xmath16 + @xmath17:_randomization : _ if @xmath18 , the speed of the @xmath7th vehicle is decreased randomly by one with probability @xmath19 ; that is , @xmath20 with probability @xmath19 + @xmath21:_vehicle movement : _ each vehicle is moved forward so that @xmath22 . + step @xmath11 reflects the tendency of drivers to drive the vehicle as fast possible , without exceeding the maximum speed of the vehicle , and avoiding accidents between vehicles at the same time . thus , if there is enough gap in front , vehicles in this model can accelerate to the maximum allowed velocity within one single timestep.this captures at least one type of aggressive driving and hence the name . the randomization in the step @xmath17 takes into account the different behavioral patterns of the individual drivers , especially non - deterministic acceleration and over - reaction while slowing down . as usual , the _ flux _ is defined to be the number of vehicles crossing a detector site per unit time . in the context of vehicular traffic , the most important quantity of interest is the so - called _ fundamental diagram _ which depicts the dependence of flux on the density of vehicles . the number of empty sites in between a pair of vehicles is usually taken as a measure of the corresponding _ distance - headway_. the _ time - headway _ is defined as the time interval between the passage of two successive vehicles recorded by a detector placed at a fixed position on the highway . we have calculated all these characteristic quantities for the adm and will present these results in the following sections . before presenting the results for the adm , we would like to compare and contrast it with a few other well known models of vehicular traffic . in the nasch model , the calculation of the speed of a vehicle at the next time step ( @xmath23 ) during the acceleration stage requires the knowledge of its speed at previous time step @xmath8 and its speed after the deceleration stage depends on the available headway in front of it , whereas , in the aggressive driving model the calculation of the speed of a vehicle at next time step does not require any knowledge of its velocity at previous time step and depends only on the available headway in front of the vehicle . in contrast to the nasch model it therefore has no velocity memory . from now onwards , we shall refer to this model as aggressive driving model ( adm ) . this adm differs from the fukui - ishibashi ( fi ) model @xcite at step @xmath17 of the updating procedure . in the fi model the randomization is applied only to those vehicles whose final velocities become @xmath24 after the acceleration stage and , therefore , the fi model is unrealistic for normal traffic . consequently , the fi model fails to capture overreactions at braking which are responsible for spontaneous jam formation ( see e.g. @xcite ) . in the special case @xmath25 , the adm reduces to nasch model @xcite with @xmath25 . in this limit the fundamental diagram is given by exact expression @xcite @xmath26.\ ] ] the symmetry about @xmath27 in this fundamental diagram breaks down for all @xmath28 . [ fig-2 ] shows the fundamental diagram of the adm for different values of @xmath24 for fixed @xmath29 and @xmath30 . [ fig-3](a ) and fig . [ fig-3](b ) , show variation of flux and average speed , respectively , with @xmath31 for different values of the braking probability @xmath19 for fixed @xmath32 . \(a ) ( b ) \(a ) ( b ) for @xmath25 , the fundamental diagram of adm has a perfect particle - hole symmetry with a flow maximum at @xmath33 . however , as in the nasch model , this particle - hole symmetry breaks down for all @xmath34 and the maximum shifts to lower densities with increasing @xmath35 . the system remains in the _ free - flow _ regime for densities on the left side of this maximum where the flux increases with increasing density . the densities on the right hand side of this maximum correspond to _ congested flow _ regime where flux starts decreasing with increasing density and finally vanishes at @xmath36 . for a given @xmath24 , the maximum value of the flux starts decreasing with increasing braking probability @xmath19 . the fundamental diagram of the adm shows unusual behavior in the deterministic limit @xmath37 where the flux vanishes at @xmath33 for all @xmath28 . the reason for this unusual behavior will be explained in the following sections . the _ distance headway _ is usually defined as the distance from a selected point on a vehicle to the same point on the corresponding lead vehicle ( i.e. , the next vehicle downstream ) . since in our model all vehicles have the same length we can use the number @xmath38 of empty cells in front of vehicle @xmath7 as a measure of the headway . in fig . [ fig-5](a ) we have shown the distribution @xmath39 of the distance headway in adm obtained from simulations . at low densities the distance headway distribution shows a broad peak near @xmath40 . this corresponds to the free - flow regime where the cars are distributed almost homogeneously . in contrast , at higher densities the peak in the distribution occurs at a smaller distance headway . in fact , the most probable distance headway decreases with increasing density . finally , at sufficiently high densities , the maximum of the probability distribution occurs only at @xmath41 . thus , with increase of vehicle density , the compact cluster of jammed vehicles becomes larger while large headways are strongly surpressed . \(a ) ( b ) the time headway distribution is determined by ( a ) the time interval between the departure from one site and arrival at the next site and ( b ) the waiting time at a given site ; the latter depends not only on the hindrance from the vehicle in front but also on the randomization parameter @xmath19 . equivalently , the time - headway depends not only on the spatial distance - headway but also on the velocity of the vehicles . a few typical time headway distributions @xmath42 in adm are shown in fig . [ fig-5](b ) for a few different densities of the vehicles . at sufficiently low densities it shows a peak at @xmath43 as , because of the _ parallel _ updating scheme , minimum two time steps must elapse between the arrival of a vehicle at two successive sites even when it moves totally unhindred by any other vehicle . since mean time headway is the inverse of the flux , it is expected to exhibit a minimum when plotted against the density . the trend of variation of the most probable time headway with increasing density is also similar , as can be seen also in fig . [ fig-5 ] . at low densities the peak is rather sharp and it becomes much broader at higher densities . compared to the corresponding results for the nasch model @xcite , large headways are surpressed in the adm ; this is caused by the possibility of large acceleration whereas in the nasch only allowed acceleration is unity . the broader distribution at higher densities arises from the longer waiting times at each site which is caused by the hindrance from the vehicle immediately in front . this stochastic model becomes deterministic in the limit @xmath44 . in this special case , the deterministic update rules of the model can be written as @xmath45 which leads to two types of steady states depending on density of vehicles @xcite . at low densities , the system can self - organize so that @xmath46 for all @xmath7 and , therefore every vehicle can move with @xmath24 , giving rise to the corresponding flux @xmath47 . this steady state is , however , possible only if enough empty cells are available in front of every vehicle , i.e. , for @xmath48 and the corresponding maximum flux is @xmath49 . on the other hand , for @xmath50 , @xmath51 and , therefore , the relevant steady states are characterized by @xmath52 , i.e. flow is limited by density of holes . since the average distance headway is @xmath53 , the fundamental diagram of the model in the deterministic limit @xmath44 is given by _ exact _ expression @xmath54.\ ] ] this is identical to the fundamental diagram of the nasch model in the deterministic limit , despite the slightly different dynamics . as we discussed earlier in this paper that in the special case @xmath55 the adm reduces to nasch model with @xmath56 and hence in the deterministic limit @xmath37 , @xmath57 for all densities @xmath31 as expected . however , for @xmath28 , the properties of the adm with maximum allowed speed @xmath24 in the deterministic limit @xmath37 are not exactly identical to those of the same model with maximum allowed speed @xmath58 and @xmath44 . if @xmath59 , then , for @xmath60 , all initial states lead to @xmath57 because in the steady state system self - organizes itself in such a way that there is a maximum headway of one lattice site in front of each vehicle and hence speed of all vehicles becomes zero immediately after the randomization step ( step @xmath17 in update rules ) . however , for @xmath61 and @xmath37 , @xmath62 for all @xmath63 . the maximal attainable velocity for every vehicle in this limit becomes @xmath64 . the fundamental diagram of adm for @xmath65 in the deterministic limit @xmath37 is given by _ expression + @xmath66 & $ \text{\ \ for\ } c \le 1/2 $ \\ 0 & $ \text{\ \ for\ } c \ge 1/2$. } % \end{cases}\ ] ] this unusual behavior of the adm is different from the corresponding behaviour in the nasch model . in the deterministic limit @xmath37 of the nasch model , irrespective of @xmath24 and @xmath31 , all random initial states lead to @xmath57 @xcite , because a car which has velocity @xmath67 will never move again . there are several possible ways of extrapolating to this limit since only finite systems can be treated in computer simulations . we here investigate the case @xmath69 . the fundamental diagram of the model is plotted in fig . [ fig-6 ] for different values of @xmath19 in this limit . this fundamental diagram has a form quite different from that in the case of finite @xmath24 . the flow does not vanish in the limit @xmath70 since already one single vehicle produces a finite value of flow , @xmath71 . @xmath72 is a monotonically decreasing function of @xmath31 . another characteristic feature of this fundamental diagram is the absence of the characteristic plateau which is exhibited by the nasch model with @xmath73 @xcite . in this section we will present the site - oriented mean - field ( somf ) and car - oriented mean - field ( comf ) approaches for calculating the fundamental diagram of the adm with periodic boundary conditions following the methods of @xcite after a brief review of the earlier works done in this regard . a somf theory was developed earlier for the fi model @xcite . starting with a microscopic relation for the updating rule , which describes the occupancy of each site on the lattice , a macroscopic time - evolution relation is obtained for the average speed of the vehicles by carrying out statistical averages . mean field equations are obtained as the asymptotic limit of the evolution relation . this gives average vehicle speed in the long time limit as a function of the vehicle density . a comf theory for the fi model was developed in @xcite starting with the basic equations which describe the time evolution of the headway in front of each car . by introducing the concept of inter - car spacing longer and shorter than the maximum attainable velocity @xmath24 , the average speed of the vehicles has been obtained analytically as a function of car density in the asymptotic limit which corresponds to the steady state . in the somf @xcite approach , @xmath74 denotes the probability that there is a vehicle with speed @xmath75 at site @xmath76 at time step @xmath8 . then , obviously , @xmath77 is the probability that the site @xmath76 is occupied by a vehicle at the time step @xmath8 ( irrespective of its speed ) and @xmath78 is the corresponding probability that the site @xmath76 is empty . using the definition + @xmath79 for the flux @xmath80 one can determine the mean - field fundamental diagram for the given @xmath19 , provided one can determine @xmath81 in the mean - field approximation . according to the update rules of the adm , the time evolutions of the probabilities @xmath74 are given by the following equations : + _ step i : _ acceleration(@xmath82 ) @xmath83 _ step ii : _ randomization ( @xmath84 ) @xmath85 _ step iii : _ movement of vehicles ( @xmath86 ) @xmath87 recall that adm with @xmath56 is identical to the nasch model with @xmath56 . therefore , for nontrivial results of the adm one must consider @xmath88 . for @xmath89 , the full somf equations read ( with @xmath90 ) + + @xmath91 in the steady state , i.e. for @xmath92 , the @xmath74 are independent of @xmath8 . for periodic boundary conditions the system becomes homogeneous in the steady state and hence the @xmath76-dependence of @xmath93 also drops out . ( [ somf20])-([somf22 ] ) then give @xmath81 explicitly as a function of the density @xmath31 . the steady state flux for @xmath89 is then given by + @xmath94 the results obtained from the somf theory are plotted in fig . [ fig-7](left ) for a few values of @xmath19 along with the corresponding numerical data from computer simulation . the agreement between the fundamental diagrams obtained from this simple somf theory and those obtained from computer simulations is quite poor because the important correlations between neighboring sites are neglected in this approach . interestingly , in contrast to the nasch model @xcite , the fundamental diagram shows an inflection point at intermediate densities . this non - convexity of the flow - density relation becomes more pronounced for large values of the randomization @xmath19 . asymptotically , for large densities @xmath95 , the flow in the adm will be identical to that in the corresponding nasch model , i.e.@xmath96 . however , in the nasch model , and in many other traffic models , the flow at _ any _ density @xmath31 can never exceed @xmath97 , the flow on the jammed branch . but , in contrast , because of the possibility of large accelaration of the vehicles in the adm , the flow can far excced the value @xmath97 at intermediate densities . then , for obvious mathematical reasons , any smooth function with the asymptotic behaviour @xmath97 has to exhibit an inflection point . in the nasch model somf always systematically underestimates the true flux because of the effective particle - hole attraction @xcite . surprisingly , the same is not always true in the adm ( see e.g. @xmath30 in fig . [ fig-7](left ) ) . this indicates that the correlations in the adm at intermediate densities are somewhat different from those in the nasch case . since now somf _ overestimates _ the flow over a range of density , this indicates the presence of effective particle - particle attraction , instead of particle - hole attraction in that regime . this is a consequence of the large accelerations of the vehicles which lead to a tendency towards particle - particle aggregation this tendency becomes stronger at large values of the randomization @xmath19 , where fluctuations that reduce the velocity of a car temporarily become more likely . in the next section we describe an improved mean field theory , namely car - oriented mean field theory , which takes into account certain correlations between the sites . here , we present the car - oriented mean - field ( comf ) theory @xcite of adm with @xmath98 . the central quantity in comf theory is the probability @xmath99 to find at time @xmath8 ( exactly ) @xmath7 empty sites in front of a vehicle , i.e. the spatial headway distribution . this approach is also known as _ empty interval method _ or _ interparticle distribution function method_. for a nice introduction and list of related references we refer to @xcite . the time evolution of the probabilities @xmath99 can conveniently be expressed through the probability @xmath100 @xmath101 that a car moves @xmath102 sites in the next time step . in order to find the time evolution of the @xmath99 we first determine from which configurations at time @xmath8 a given state a time @xmath23 could have evolved . take for instance a car called second car in the following which has @xmath103 free sites in front , i.e. its distance to the next car ahead ( called first car in the following ) is @xmath104 sites . since the velocity difference of the two cars is at most 2 , a headway of @xmath7 sites at time @xmath23 must have evolved from a headway of length @xmath105 , @xmath7 , @xmath104 or @xmath106 in the previous time step . a headway of @xmath105 sites evolves into a headway of @xmath7 sites only if the first car moves ( with probability @xmath107 ) and the second car brakes in the randomization step ( with probability @xmath19 ) , i.e. the total probability for this process is @xmath108 . the headway will remain constant if the first car moves with probability @xmath109 and second car brakes with probability @xmath19 ( total probability for this process is @xmath110 ) or the first car moves with probability @xmath111 and second car moves with probability @xmath0 ( total probability for this process is @xmath112 ) . similarly , a headway of @xmath104 sites evolves into a headway of @xmath7 sites if the first car does not move ( probability @xmath113)and second car brakes with probability @xmath19 ( total probability being @xmath114 ) or the first car moves with probability @xmath109 and second car moves with probability @xmath0 ( total probability for this process is @xmath115 ) . finally , a headway of @xmath106 evolves into a headway of @xmath7 only if the second car moves with probability @xmath0 ( total probability for this @xmath116 ) . the special cases @xmath117 and @xmath118 can be treated in an analogous fashion . in this way one obtains the time evolution of the probabilities as @xmath119 , \\ % \fl \qquad{p_1}(t+1 ) & = & { g_1}(t){p_0}(t ) + ( p{g_0}(t)+q{g_1}(t ) ) [ { p_1}(t)+{p_2}(t ) ] + q{g_0}(t){p_3}(t),\quad \\ % \fl \qquad{p_2}(t+1 ) & = & { g_2}(t){p_0}(t ) + ( p{g_1}(t)+q{g_2}(t ) ) [ { p_1}(t)+{p_2}(t ) ] \nonumber \\ \fl \qquad & + & ( p{g_0}(t)+ q{g_1}(t)){p_3}(t ) + q{g_0}{p_4}(t ) , \\ % \fl \qquad { p_3}(t+1 ) & = & p{g_2}(t)[{p_1}(t)+{p_2}(t ) ] + ( p{g_1}(t)+q{g_2}(t)){p_3}(t ) \nonumber \\ \qquad & + & ( p{g_0}(t)+q{g_1}(t)){p_4}(t ) + q{g_0}(t){p_5}(t ) , \\ % \fl \qquad { p_n}(t+1 ) & = & p{g_2}(t){p_{n-1}}(t ) + ( p{g_1}(t)+q{g_2}(t ) ) { p_n}(t ) \nonumber \\ \qquad & + & ( p{g_0}(t)+q{g_1}(t)){p_{n+1}}(t ) + q{g_0}(t){p_{n+2}}(t ) \ \ \ % \hspace{2.0 in } ( n \ge 4 ) \label{last}\end{aligned}\ ] ] a car will not move in next time step if there is no empty site in front of it ( probability @xmath120 ) or if there is exactly one empty site in front of it and it decelerates in the randomization step 2 ( probability @xmath121 ) . it will move one site if either there is exactly one empty site ahead and it does not decelerate ( probability @xmath122 ) or there are at least two empty sites in front , but the car decelerates in step 2 ( probability @xmath123 ) . in all other cases it will move two sites . therefore the probability @xmath124 that a car moves @xmath102 sites in the next time step is given by @xmath125 \nonumber \end{aligned}\ ] ] where we have used the normalization condition @xmath126 to rewrite the probabilities @xmath124 in terms of @xmath120 , @xmath127 , @xmath128 and @xmath129 only . the probabilities can also be related to the density @xmath130 of cars . since each car which has the headway @xmath7 to the next car one in front of it occupies @xmath104 sites we have following relation : @xmath131 here we are mainly interested in the stationary state ( t @xmath132 ) with @xmath133 @xmath134 . in order to determine the probabilities in the stationary state we introduce the generating function + @xmath135 after multiplying corresponding equation in [ [ first]-[last ] ] by @xmath136 and summing over all equations one finds an explicit expression for the generating function , @xmath137 with @xmath138 note that @xmath139 . the denominator of @xmath140 has two zeros located at @xmath141 with @xmath142 and @xmath143 . the normalization condition ( [ second ] ) is equivalent to @xmath144 and is already satisfied by ( [ generating ] ) . the density relation implies @xmath145 , where @xmath146 denotes the derivative of @xmath140 . in order to have @xmath147 and @xmath148 the generating function must be analytic in the unit disc @xmath149 . therefore the zero @xmath150 of the denominator has to be cancelled by a corresponding zero of the numerator . the equation @xmath151 yields a relation between the variable @xmath152 and @xmath153 so that @xmath140 only depends on one free parameter , e.g. @xmath152 . this parameter , in turn , is a function of the only physically relevant parameter , the density c , via @xmath145 . to obtain the fundamental diagram we have to calculate the flux . it is given by @xmath154 . \label{eq-29}\ ] ] in order to calculate the flux @xmath80 for a given set of @xmath31 and @xmath19 one has to solve the following two equations numerically . @xmath155 and @xmath156 where @xmath146 denotes the derivative of @xmath140 . ( [ pdashone ] ) can be written as @xmath157 eqs . ( [ appenda1 ] ) and ( [ appenda3 ] ) were solved numerically . values of @xmath152 and @xmath153 thus obtained for a given set of @xmath31 and @xmath19 are used to calculate the values of @xmath158 , @xmath159 and @xmath160 . finally flux @xmath161 is calculated using equation ( [ eq-29 ] ) . the results obtained from comf are plotted in fig . [ fig-7](right ) for a few values of @xmath19 along with the corresponding numerical data from computer simulation . fundamental diagrams obtained from comf show an excellent agreement with the numerical data in the limit @xmath162 . thus comf can capture the important correlations much better than somf . especially it is able to reproduce the occurance of an inflection point at larger values of @xmath19 . the small deviations are due to the fact that comf neglects the correlations between the headways in front of successive vehicles . \(a ) ( b ) in fig . [ fig-9 ] we have plotted the fundamental diagram of the adm with @xmath163 and the average speed of vehicles against their density along with that of the nasch model with @xmath163 for a few values of @xmath19 . in the absence of randomization , i.e. for @xmath44 , this model and the nasch model give identical fundamental diagram and variation of average speed with density . in the presence of randomization , i.e. @xmath164 , the flow in the adm is always larger than that of the corresponding nasch model due to the faster acceleration . this difference is most pronounced at densities slighty beyond the maximum flow . cells . vehicles move from left to right , and are represented by dark circles . the left boundary consists of mini system of @xmath24 cells . this left mini system is occupied by at most one vehicle with probability @xmath165 . similarly the right boundary consists of a mini system of @xmath24 cells and particles are extracted from it with probability @xmath166 . we shall represent each cell by a site on the lattice formed by these cells . ] in this section we consider the adm with open boundary conditions where vehicles move deterministically , i.e. with randomization probability @xmath44 . a schematic representation of the analyzed system is shown in fig . [ fig-10 ] . our main system consist of @xmath167 cells . this main system is connected to two mini systems of length @xmath24 on each side @xcite . this is done to provide a proper insertion and extraction strategy allowing us to investigate the whole spectrum of the possible states . the state of the mini system of the left boundary has to be updated every time step before the vehicles of whole system . the update procedure consists of two steps . if any cell of the left mini system is occupied it has to be emptied first . then a vehicle is inserted in the system with probability @xmath165 . the position of the inserted vehicle has to satisfy the following conditions : ( i ) the headway between the inserted vehicles in the mini system and the first vehicle in the main system is equal to @xmath35 , and ( ii ) the distance to the main system has to be minimum i.e. if there is no vehicle present in the main system within first @xmath24 cell then the rightmost cell of the left boundary is occupied . the right boundary consists of @xmath24 cells and vehicle are removed from these cells with probability @xmath166 . these boundary conditions are capable of generating all flows observed in the case of periodic boundary conditions , including the maximal flow . from now onwards , we shall represent the cells by the sites of a lattice formed by the cells . the above insertion and extraction scheme generates the maximum flow of the corresponding aggressive driving model with periodic boundary conditions for @xmath168 , i.e. @xmath169 for small @xmath165 and large @xmath166 , the system is found in the free flow regime . in fig . [ fig-12 ] we have shown density profiles over a spatial region located in the middle of the bulk of the system for @xmath170 and @xmath171 in the free flow regime . the density profile shows a periodic structure with a period of oscillation @xmath172 . for any arbitrary @xmath24 , we find that the period of this oscillating pattern is @xmath173 . \(a ) ( b ) in order to understand this periodicity we first consider the density profile for very low injection rates and maximum extraction rate ( @xmath174 ) ( see fig . [ fig-12 ] ) . for @xmath175 the probability of inserting a vehicle at the rightmost site of the left mini - system in two successive time step is very small and , therefore , the vehicles at the beginning of the system do not feel influence of each other . this means that a vehicle which is inserted at the rightmost site of the left mini system moves to @xmath176 at the next time step and can be found at site @xmath177 after @xmath8 times steps ( @xmath178 ) . the density on these sites is @xmath179 ( @xmath180 ) . for increasing injection rate @xmath165 , the probability of inserting a vehicle in two successive time steps increases which results in the increase in the hindrance that a vehicle feels from the front vehicle at the beginning of the system . this can be explained as follows @xcite : suppose we insert a vehicle a in the mini - system at time step @xmath181 and a vehicle at time step @xmath182 . considering the system at time step @xmath182 , we see that vehicle a is on site @xmath176 and will move with velocity @xmath183 where as vehicle b will occupy position @xmath184 in the next time step @xmath185 because the vehicles are inserted in left mini - system in such a way that the headway between the inserted vehicle and the next vehicle downstream is is @xmath183 . at time @xmath186 , vehicle a is on @xmath177 and vehicle b is on @xmath187 . in other words , we can say that the hindrance due to left boundary condition leads to a shift of the position of the vehicles within the system . this shift is reflected in the oscillation in the density profile of [ fig-12 ] . the probability of finding a vehicle at @xmath188 is smaller than at @xmath177 and it is much smaller for @xmath189 and even much smaller for @xmath190 . as we move from the free flowing regime to congested flow regime ( keeping @xmath165 fixed and decreasing @xmath166 ) something interesting happens : the oscillations start vanishing and envelope of density profile rises ( see fig . [ fig-13 ] ) . for low values of @xmath166 , the density profile is just a constant whose value increases with decreasing @xmath166 . this phenomenon is due to the hindrance that vehicles feel at the right boundary with decreasing probability @xmath166 . consequently a jam develops at the right boundary which expands to the left with decreasing @xmath166 . in order to identify the regions of free - flow and congested flow in the phase diagram of the adm we measure the bulk density and the flux in the middle of the open system by varying the boundary rates . density - flux pairs falling on the free flow branch of the periodic system are identified as belonging to the free flow phase , those falling on the jammed branch as congested flow . since density profile shows a periodic structure in the free flow regime , in order to compute the bulk density in the middle of the system , we average over the densities of @xmath24 lattice sites ( i.e. one period of oscillation ) for a given @xmath24 . the phase diagram of the adm with open boundary conditions for @xmath170 and @xmath191 is shown in fig . [ fig-11 ] . the system will be found either in free - flow or congested - flow regime depending on the values of @xmath165 and @xmath166 . here , the @xmath165 = @xmath166 line does not separate the free flowing and congested flow regime . instead , the free flow regime is larger than the congested flow regime . the span of the free - flow regime increases with increasing @xmath24 . in the special case @xmath25 the above insertion and extraction scheme leads us to the phase diagram of tasep with open boundary and where the line @xmath165 = @xmath166 separates free flow and congested flow regime . and @xmath191 . ] in this section we first mention the main experimentally observed features of the steppings of dynein motors . then , by extending the adm , we develope a simple theoretical model that captures the essential features of dynein stepping . in order to understand the mechanism of a single dynein motor , mallik et al . @xcite extracted the step size of single dynein motor from their experimental data . in their experiment , hindrance against forward movemnent of dyneins was caused by an opposing force . in principle , this hindrance could also be created by other motors . the smallest possible step size would be @xmath192 nm as the equispaced binding sites on the microtubule form a lattice with lattice constant of @xmath192 nm . mallik et al . @xcite observed that in the absence of hindrance the step sizes of dyneins were mostly @xmath193 nm , i.e. , four times the lattice constant . moreover , the step size decreased with increasing hindrance : under weak hindrance the step size was approximately @xmath194 nm , under intermediate hindrance step size was about @xmath195 nm , whereas under strong hindrance dynein takes steps of @xmath196 nm . on the basis of these observations , mallik et al . @xcite suggested a _ molecular gear mechanism _ for dynein motors . in their generic model of molecular motor traffic , parmeggiani et al . @xcite implicitly assumed a hindrance - independent step size of the motors . therefore , in the light of the experimental observations on dynein steppings @xcite , one may interpret the model developed in @xcite to be a minimal model for the traffic of kinesin motors which are known to take steps of @xmath192 nm irrespective of the hindrance . therefore , to model the traffic of dynein motors , which can take steps of upto @xmath197 nm ( i.e. , four times the lattice spacing in between the successive binding sites ) at one go in the absence of hindrance , we need a more sophisticated model . in the next subsection we describe the model which we propose for the traffic of dynein motors . our dynein traffic model ( dtm ) has been obtained by extending the adm which we have discussed extensively in the preceeding sections . the lattice sites in this case represent the dynein binding site on the microtubule track and the lattice constant is @xmath192 nm . in order to capture the fact that in the absence of hindrance a single dynein motor can take a single step of @xmath197 nm , we set @xmath198 in the adm . moreover , to capture the attachment and detachment of the motors from the microtubule track , we also allow the detachment of a motor from an occupied site with rate @xmath2 and attachment of a motor to an empty site with rate @xmath199 . the state of the system is updated in a random sequential manner . in this dtm , a single dynein motor can move forward by four lattice sites ( i.e. , @xmath197 nm ) in one single step if the available gap is greater than or equal to @xmath197 nm ; otherwise , the step size will be equal to the available gap as the mutual exclusion between the motors hinders the motion of the following dynein . in the dtm with @xmath200 for different values of the binding constant @xmath201 and with ( a ) random - sequential updating , ( b ) parallel updating.,title="fig : " ] in the dtm with @xmath200 for different values of the binding constant @xmath201 and with ( a ) random - sequential updating , ( b ) parallel updating.,title="fig : " ] \(a ) ( b ) in the case of periodic boundary condition we allow attachment and detachment of motors from all lattice sites with probability @xmath199 and @xmath2 respectively . this prescription is very similar to that followed by parmeggiani et al.@xcite in their generic model of molcular motor traffic . in order to compare and contrast our dtm with the model of parmeggiani et al.@xcite , all our computer simulations of the dtm have been carried out by random sequential updating which was adopted in ref.@xcite . we have carried out limited investigation of the dtm also by implementing parallel updating . but , unless explicitly stated otherwise , by computer simulation of the dtm we shall mean simulation using random sequential updating . fig . [ dyperiodic ] shows the values of flux @xmath202 as a function of the binding constant @xmath203 for different values of @xmath2 obtained from the computer simulations implementing random - sequential updating ( fig . [ dyperiodic](a ) ) and parallel updating ( fig . [ dyperiodic](b ) ) . in both the cases the flux initially increase with increasing @xmath201 . however , beyond a threshold value of @xmath201 the flux starts decreasing with the further increase of @xmath201 . comparing the figures [ dyperiodic](a ) and [ dyperiodic](b ) we find that , for the same set of values of the parameters , higher flux is obtained with parallel updating ; this effect becomes more pronounced at higher @xmath2 . this is consistent with the well known result that flux obtained with parallel updating is higher than that obtained with random - sequential updating even in the simpler situation of tasep . we have carried out a site - oriented mean field ( somf ) calculation for the dtm following the same procedure as we followed earlier for the adm . for @xmath204 , we get @xmath205 \\ \label{eq - dynpar2 } { c_1}(i , t+1 ) & = & qc(i+1,t)d(i , t)c(i-1,t)\\ \label{eq - dynpar3 } { c_2}(i , t+1 ) & = & qc(i+1,t)d(i-1,t)d(i , t)c(i-2,t)\\ \label{eq - dynpar4 } { c_3}(i , t+1 ) & = & qc(i+1,t)d(i-2,t)d(i-1,t)d(i , t)c(i-3,t)\\ \label{eq - dynpar5 } { c_4}(i , t+1 ) & = & qd(i-3,t)d(i-2,t)d(i-1,t)d(i , t)c(i-4,t)\end{aligned}\ ] ] in the steady state , i.e. @xmath92 , @xmath74 are independent of @xmath8 . for periodic boundary conditions the system becomes homogeneous in the steady state and hence the @xmath76-dependence of @xmath93 also drops out . therefore the steady state flux for @xmath204 is given by @xmath206\ ] ] where @xmath0 is the probability that the motor does , indeed , hop , instead of getting detached from the track . finally , substituting @xmath207 , the steady flux in the dtm , under the somf approximation , is given by @xmath208 where @xmath31 is given by the well - know ratio @xmath209 of langmuir equilibrium density and @xmath210 . in figs . [ dyperiodic](a ) and ( b ) we have shown the curves obtained from eq . ( [ eqn - flux ] ) to compare the predictions of the somf theory with the corresponding simulation data . although the agreement is not very good , we find that the somf provides better estimates of flux at higher values of the detachment probability @xmath2 . one interesting feature of the somf result is that it is an overestimate of the flux for random - sequential updating whereas it is an underestimate of that corresponding to parallel updating . moreover , the somf estimates are closer to the simulation data from random - sequential updating than those from parallel updating . this behaviour indicates that the underlying correlations are rather subtle . and parallel updating under periodic boundary conditions . ] the agreement between somf theory and computer simulations is much better in the dtm than in the adm . is it an artefact of the different values of @xmath211 used in fig.[fig-7](a ) and fig . [ dyperiodic](b ) ? in order to investigate this possibility , we have studied a hypothetical dtm with @xmath212 by somf as well as by computer simulation implementing parallel updating . the somf estimate for the flux in this hypothetical dtm with @xmath213 is given by @xmath214 where @xmath215 . this somf estimate is compared in fig.[hypdtm ] with the numerical data obtained from the simulation of the same model with parallel updating . the agreement is as good as that in fig . [ dyperiodic ] for the actual dtm with @xmath200 . thus , the reason for better success of somf in dtm than in adm remains a challenging open problem for future investigation . a schematic representation of the analysed system with open boundary conditions is given in fig . [ dyneinmodel ] . to capture the attachment and detachment of the motors from the microtubule we allow the detachment of a motor from an occupied site with rate @xmath2 and attachment of a motor to an empty site with rate @xmath199 in the bulk i.e. from all site other except those who belong to the reservoirs at the left and right boundary . our proposed model with open boundary conditions reduced to the model of parmeggiani et al.@xcite , which is a minimal model for the intra - cellular traffic of kinesin motors , if one sets @xmath216 . in the dtm obtained from monte carlo simulations and plotted against the rescaled space variable @xmath217 for different values of @xmath218 . the common parameter values are @xmath219 . , title="fig : " ] in the dtm obtained from monte carlo simulations and plotted against the rescaled space variable @xmath217 for different values of @xmath218 . the common parameter values are @xmath219 . , title="fig : " ] a competition between bulk dynamics ( langmuir kinetics ) and boundary induced non - equilibrium effects ( tasep - like dynamics ) is expected only if the particles injected either at boundary or somewhere in bulk visits a finite fraction of the total system size . in that case particles will spend enough time on lattice to feel mutual influence and eventually would produce collective effects . study of competition between bulk and boundary dynamics for large systems ( @xmath220 ) requires that the kinetic rates @xmath199 and @xmath2 decrease simultaneously with increasing system size @xmath167 . this can be illustrated by considering the following heuristic argument given in ref . the average time @xmath221 spent by a particle before detachment is roughly of the order of @xmath222 @xmath223 . during this time @xmath221 the number of sites @xmath7 visited by a given particle is of the order of @xmath224 . therefore the fraction @xmath225 ( @xmath226 ) of the lattice site visited by a given particle during this time would tend to zero for fixed @xmath2 as @xmath227 . in order that a given particles explores a finite fraction of the total sites before detaching for system size @xmath220 , one has to define the total " detachment rate @xmath228 such that @xmath218 remain constant as @xmath227 . a similar argument shows that a vacancy visits a finite fraction lattice sites until it is filled by the attachment of a particle if @xmath199 scales to zero as @xmath229 with fixed total " detachment rate @xmath230 @xcite . therefore we define total detachment rate @xmath231 and total attachment rate @xmath232 such that @xmath218 and @xmath230 remain constant as @xmath227 . note that the binding constant @xmath233 remains unchanged as @xmath227 . and @xmath234 . ] in fig . [ densityprofile ] ( a ) we have plotted the typical average density profiles @xmath235 obtained from monte carlo simulations for this model in rescaled space variable @xmath217 ( where @xmath236 denotes occupation of nth lattice site in the bulk i.e. @xmath237 ) by choosing a path in the parameter space along the curves with fixed @xmath238 and @xmath201 while increasing @xmath239 . for low values of kinetic rates @xmath240 the system remain in low - density phase where average density in the bulk remains constant . for high values of kinetic rates @xmath218 and @xmath230 the system goes to high density phase where average density in the bulk again constant . the bulk density in this case is determined by the well know ratio @xmath209 of langmuir equilibrium density . for intermediate values of the rates , for example @xmath241 , the density profile in the bulk exhibits unusual feature where regions of high density and low density are connected by a steep rise . in fig . [ densityprofile ] ( b ) we have shown the density profiles obtained by parmeggiani et al . @xcite for their model where @xmath242 . comparison of figs . [ densityprofile ] ( a ) and [ densityprofile ] ( b ) shows that the average bulk density in the low density phase in fig . [ densityprofile ] ( a ) is smaller than in fig . [ densityprofile ] ( b ) . this observed decrease in the bulk density in the low density phase for identical values of the parameters @xmath243 and @xmath218 is due to the fact that particles in low density phase move with higher velocity which leads to higher current and lower density . [ densityprofile ] ( a ) also shows oscillations in the density profile at the beginning of the system in the low density phase . these oscillations result from the hindrance that particles have at the beginning of the system from each other @xcite . these oscillations die out for higher systems sites . comparison of fig . [ densityprofile ] ( a ) and fig . [ densityprofile ] ( b ) also shows that in the case of dynein motors the domain wall is found inside the the system for slightly higher value of @xmath218 . figure [ systemsize ] shows the average density profile @xmath244 computed from monte carlo simulation in rescaled space variable @xmath245 for different system sizes . the width of the transition region decreases with increasing system size . the data obtained from our simulations suggest a sharp discontinuity of the density profile in terms of the rescaled space variable @xmath217 in the limit @xmath246 . therefore the low and high density phases separated by a sharp domain wall coexist in our model over an intermediate range of parameter values where boundary and bulk kinetic rates compete against each other . this discontinuity in the density profile is stable and the position of the domain wall is determined by the values of the kinetic rates as shown in fig . [ dwposition ] . this coexistence of high and low density phase separated by a domain wall can be regarded as a _ traffic jam _ for molecular motors . . the common parameter values are @xmath247 and @xmath248 . ] ( b ) @xmath249 ( c ) @xmath250 and ( d ) @xmath251 ( e ) @xmath252 ( f ) @xmath253 ( g ) @xmath254 and ( h ) @xmath255 . other common parameter values are @xmath256 . ] in order to identify the regions of coexistence , we have obtained the phase diagram of our model for intra - cellular traffic of dynein motors by varying the boundary rates @xmath165 and @xmath166 for fixed values of @xmath201 and @xmath218 . [ phasediag](a ) shows the phase diagram for @xmath257 . in this case model reduces to the aggressive driving model with random sequential updating . for @xmath257 , by varying boundary rates @xmath165 and @xmath166 one gets three kind of phases namely low density phase ( ld ) , high density phase ( hd ) and maximal current phase ( mc ) . for very small values of @xmath218 ( @xmath258 ) the boundary rates @xmath165 and @xmath166 dominate and the structure of the phase diagram is determined only by the boundary rates @xmath165 and @xmath166 . for @xmath259 one obtained a phase digram similar to fig . [ phasediag](a ) . on increasing the value of @xmath218 the boundary and bulk rates start competing with each other and in this case one gets a phase diagram where mc phase disappears and one can identify four distinct regions in the phase diagram namely , low density phase ( ld ) , low density high density coexistence region ( ld - hd ) , high density phase ( hd ) and meissner " ( m ) phase . the meissner phase @xcite has some interesting features that are genuinely distinct from the high density phase . the density profile in the bulk is independent of the boundary rates @xmath165 and @xmath166 and is determined only by the bulk . on further increase of @xmath218 the low density phase also disappears from the phase diagram as shown in fig . [ phasediag](c ) and in this case one gets hd phase , m phase and coexistence region . for large values of @xmath218 the phase diagram is spanned only by the m phase as shown in fig . [ phasediag ] ( d ) . in fig . [ phasediag](e)-(h ) we have shown phase diagrams of the model of parmeggiani et al . @xcite for the identical set of parameters . comparison of the phase diagrams shown in figs . [ phasediag](a)-(d ) and [ phasediag](e)-(h ) shows that the regions of coexistence of low and high density phases are slightly different in the @xmath260 plane for identical values of the parameters @xmath201 and @xmath218 . comparison of @xmath197 nm step of dynein at in the absence of hindrance and @xmath192 nm step of kinesin implies that as a cargo transporter the dynein is four times more fuel - efficient than kinesin as both require one molecule of atp as fuel . to study the transportation efficiency of dynein motors and kinesin motors as a function of the parameters @xmath201 and @xmath218 we define the transportation efficiency by the relation @xmath261 and ( b ) @xmath218 . the common parameter values are @xmath262 and @xmath263,title="fig : " ] and ( b ) @xmath218 . the common parameter values are @xmath262 and @xmath263,title="fig : " ] the above relation has been defined in such a way that if a dynein motor takes 4 steps of 8 nm in each attempt ( i.e. for each atp hydrolysis ) then its efficiency will be 100 % similarly if a kinesin motor takes 1 step of 8 nm in each attempt then its efficiency will be equal to 25 % . the transportation efficiency of cytoplasmic dynein motors ( @xmath198 ) and kinesin motors ( @xmath242 ) are plotted in fig . [ efficiency ] ( a ) and fig . [ efficiency ] ( b ) for different values of @xmath218 and @xmath201 . there is practically no difference between the efficiency of dynein and kinesin motors for very large values of @xmath218 and @xmath201 ( @xmath264 and @xmath265 ) as for very large values of @xmath218 and @xmath201 system is found in high density phase ( hd ) . in this paper we have first investigated the properties of the aggressive driving model ( adm ) which is a simple cellular automata model for vehicular traffic . one of the motivations for considering this model is that the rule for aggressive driving can be naturally extended to capture the special features of step sizes of dynein motors and , therefore , the adm is ideally suited for extending so as to study intra - cellular molecular motor traffic by dynein motors . the adm shows different behavior for @xmath56 and @xmath28 . for @xmath56 the model is identical to the nasch model with @xmath56 which has perfect particle - hole symmetry . this symmetry is broken for @xmath28 . the fundamental diagram of this model in the special limit @xmath73 has a form which is quite different from that of the nasch model in the limit @xmath73 . we have also shown few distance headway and time - headway distributions . we have calculated the fundamental diagram using two different mean - field approaches , namely , site - oriented mean - field approach ( somf ) and car - oriented mean - field approach ( comf ) . a simple somf theory shows a poor agreement with the simulation data . however , an improved mean field theory , namely comf , shows good agreement with the numerical data obtained from computer simulations . we compare our adm with the nagel - schreckenberg model which captures essential features of normal driving . we have also investigated the density profiles and phase diagrams of this model replacing the periodic boundary conditions by open boundary conditions . the density profile of this model with open boundary conditions shows periodic structures in the free - flowing regime whose period of oscillation depends on the maximum attainable velocity @xmath24 . we have extended the adm to develope a dynein traffic model ( dtm ) which is a model of intra - cellular molecular motor traffic from cell periphery towards the nucleus of the cell . we have investigated the properties of this model with periodic and open boundary conditions . under open boundary conditions , dtm shows an unusual feature where low and high density phases separated by a static domain wall coexist over a range of parameter values which can be interpreted as a traffic jam of molecular motors . this is in sharp contrast to the phase diagram of the adm which does not exhibit such coexistence of congested and free - flowing regions . the occurrence of the phase is , thus intimately related to the competition between the hopping and the kinetics of attachment / detachment of the motors on the track . finally , we have compared the efficiencies of dynein and kinesin motors for different values of parameters . for very large values of the parameters @xmath218 and @xmath201 system is found in the high density phase and , in that case , one observes practically no difference between the efficiencies of kinesin and dynein motors . to our knoweledge , our dtm is the first model of traffic - like collective transport of _ dynein _ motors on filamentary microtubule tracks . a model that incorporates both the species of dyneins and kinesin motors , which move in opposite directions along the same track , may provide deep insight into experimentally observed bidirectional traffic @xcite . * acknowledgements * : this work has been supported ( through dc ) , in part , by the council of scientific and industrial research ( csir ) of the government of india . dc also thanks max - planck institute for physics of complex systems , dresden , and prof . frank jlicher for hospitality during a short visit when a part of this manuscript was prepared . unsrt m. schliwa ( ed . ) , _ molecular motors _ , ( wiley - vch , 2003 ) . j. howard , _ mechanics of motor proteins and the cytoskeleton _ ( sinauer associates , massachusetts , 2001 ) . d. chowdhury , a. schadschneider , k. nishinari , phys . * 2 * , 318 ( 2005 ) see the special issue of physica a ( 2006 ) for the close similarities and crucial differences between biological traffic and vehicular traffic . physica scripta ( _ royal swedish academy of sciences _ ) , * t106 * , 13 ( 2003 ) . d. chowdhury , l. santen , and a. schadschneider , phys . rep . * 329 * , 199 ( 2000 ) . b. schmittmann and r.k.p . zia , in : _ phase transition and critical phenomena _ , vol . 17 , eds . c. domb and j. l. lebowitz ( academic press , 1995 ) . g. schtz , in : _ phase transition and critical phenomena _ , vol . 19 , eds . c. domb and j. l. lebowitz ( academic press , 2000 ) . v. privman ( ed . ) , _ nonequilibrium statistical mechanics in one dimension _ ( cambridge university press , 1997 ) j. marro and r. dickman , _ nonequilibrium phase transitions in lattice models _ , ( cambridge university press , 1999 ) s. wolfram , _ theory and applications of cellular automata _ ( world sci . , 1986 ) ; _ a new kind of science _ ( wolfram research inc . , 2002 ) c. macdonald , j. gibbs and a. pipkin , biopolymers , * 6 * , 1 ( 1968 ) . c. macdonald and j. gibbs , biopolymers , * 7 * , 707 ( 1969 ) . b. derrida , phys . rep . * 301 * , 65 ( 1998 ) . a. parmeggiani , t. franosch and e. frey , phys . lett . * 90 * , 086601 ( 2003 ) ; phys . e * 70 * , 046101 ( 2004 ) . e. frey , a. parmeggiani and t. franosch , genome inf . * 15 * , 46 ( 2004 ) and references therein . r. lipowsky , s. klumpp and t.m . nieuwenhuizen , phys . lett . * 87 * , 108101 ( 2001 ) . r. lipowsky and s. klumpp , physica a * 352 * 53 ( 2005 ) and references therein ; r. lipowsky , y. chai , s. klumpp , s. liepelt and m. j. i. mller , in @xcite and references therein . m.r . evans , r. juhasz and l. santen , phys . e * 68 * , 026117 ( 2003 ) . v. popkov , a. rakos , r.d . williams , a.b . kolomeisky a nd g.m . schtz , phys . e * 67 * , 066117 ( 2003 ) . a. schadschneider , in : _ traffic and granular flow 97 _ , ed . by m. schreckenberg and d.e . wolf ( springer , 1998 ) k. nagel and m. schreckenberg , j. physique i * 2 * , 2221 ( 1992 ) m. schreckenberg , a. schadschneider , k. nagel and n. ito , phys . rev . e * 51 * , 2939 ( 1995 ) ; a. schadschneider and m. schreckenberg , j. phys . a * 26 * , l679 ( 1993 ) m. fukui , y. ishibashi , j. phys . , * 65 * , 1868 ( 1996 ) a. schadschneider , physica a * 313 * , 153 ( 2002 ) k. ghosh , a. majumdar , d. chowdhury , phys . rev . e * 58 * , 4012 ( 1998 ) k. nagel and h. j. herrmann , physica a * 199 * , 254 ( 1993 ) m. sasvri and j. kertsz , phys . e * 56 * , 4104 ( 1997 ) sascha grabolus , diplom thesis , university of cologne ( 2001 ) b. h. wang , y. r. kwong and p. m. hui , phys . e * 57 * , 2568 ( 1998 ) ; physica a * 254 * , 122 ( 1998 ) b. h. wang , l. wang , p.m. hui and b. hu , phys . e * 58 * , 2876 ( 1998 ) a. schadschneider and m. schreckenberg , j. phys . a * 30 * , l69 ( 1997 ) d. ben - avraham : in _ nonequilibirum statistical mechanics in one dimension _ , edited by v. privman , cambridge university press ( 1997 ) . r. barlovic , t. huisinga , a. schadschneider , and m. schreckenberg , phys . e * 66 * , 046113 ( 2002 ) s. cheybani , j. kertsz and m. schreckenberg , phys . rev e * 63 * , 016107 ( 2000 ) r. mallik , b. c. carter , s. a. lex , s. j. king and s. p. gross , nature * 427 * , 649 ( 2004 ) . s. cheybani , j. kertsz and m. schreckenberg , phys . rev e * 63 * , 016108 ( 2000 ) . gross , phys . * 1 * , r1 ( 2004 ) .

for the generalized statistical mechanics based on the tsallis entropy , a variational perturbation approximation method with the principle of minimal sensitivity is developed by calculating the generalized free energy up to the third order in variational perturbation expansion . the approximation up to the first order amounts to a variational approach which covers the variational method developed in phys . rev . lett . 80 , 218 ( 1998 ) by lenzi , and the approximations up to higher orders can systematically improve variational result . as an illustrated example , the generalized free energy for a classical harmonic oscillator ( considered in the lenzi s joint work ) are calculated up to the third order , and the resultant approximations up to the first , second , and third orders are numerically compared with the exact result .

cond-mat0107553

tsallis non - extensive statistical physics ( tnesp ) is one of theories for generalizing the boltzmann - gibbs statistical mechanics and thermodynamics @xcite . its formalism is based on the tsallis entropy with a parameter @xmath1 , the index of non - extensiveness for a system . it takes the conventional statistical mechanics as its special case of @xmath2 . a distinct feature of it is the power - law distributions ( @xmath3 ) instead of the exponential law in the conventional statistical mechanics . since tsallis pioneered work in 1988 @xcite , the tnesp has greatly developed , and successfully been used for investigating many systems with a long - range interaction , long - time memory , or fractal structured space - time @xcite , in which the non - extensive effect exists @xcite . for a new physical theory , to develop basic and analytic approximation tools for the tnesp should be a fundamental task for developing and employing it . to finish this task is particularly necessary for the tnesp because to perform calculation with the tsallis statistics is usually difficult due to the presence of the power - law distributions . since the tnesp with @xmath4 is reduced to the conventional statistical mechanics , it will be natural to generalize variational method and perturbational theory , the two basic approximation methods in the conventional statistical mechanics @xcite , to the tnesp . in fact , some progress has been made in this direction . early in 1993 , employing concavity properties of tsallis entropy , plastino and tsallis established a generalized bogoliubov inequality and accordingly developed a scheme of variational approximation for the free energy in the generalized canonical ensemble @xcite . later in 1998 , lenzi @xmath5 generalized the perturbation method by deriving approximate expressions of the generalized free energy ( gfe ) up to the second order and obtained , through analyzing the approximate expression up to the second order , another generalized bogoliubov inequality which affords a different variational method @xcite . recently , mendes @xmath6 gave a comparative study on the aforementioned variational methods @xcite , and illustrated that the variational method in ref . @xcite gives a better approximation for @xmath7 and is easier to perform than the variational method in ref . @xcite , albeit the latter can lead to a better approximation than the former for the case of @xmath8 . nevertheless , the usefulness of both variational and perturbation methods is limited . the variational approximate scheme in ref . @xcite is valid only for the case of @xmath9 , and the variational approximate scheme in ref . @xcite works only for the case of @xmath10 . as for the perturbation method , the @xmath11th - order perturbation approximation can be used only when @xmath12 . moreover , as is well known in the conventional statistical mechanics , to control the approximate accuracy of variational methods is not straightforward , and the perturbation method is valid only for an exactly soluble system with a really small perturbation . hence , for systematically improving variational method and extending the valid range of @xmath1 , it is worthwhile developing new approximate approaches . in the conventional statistical mechanics and some other branches of physics , a variational perturbation idea @xcite which collects merits and overcomes drawbacks of both the perturbation and the variational methods has developed and now becomes a powerful tool called variational perturbation theory @xcite ( sometimes nominated other names . ) . the author or with his collaborators also developed some schemes for the variational perturbation theory @xcite(refs . @xcite have briefly introduced references on the schemes for variational perturbation theory ) . in the present paper , we intend to generalize the variational perturbation theory to the tnesp . in the generalized canonical ensemble , we will consider an expansion for the gfe analogous to that in ref . @xcite but with artificially introduced some auxiliary parameter @xmath13 and an expansion index @xmath14 , and derive the expressions of the first four terms in the expansion series which will lead to approximations for the gfe up to the third order . from the truncated expressions of the gfe at the first , second and third orders ( with taking @xmath15 ) , the auxiliary parameter @xmath13 will be determined according to a principle of minimal sensitivity ( pms ) @xcite at the first , second and third orders , respectively . the value of @xmath13 determined at some order entering the truncated expression of the gfe at the same order produces the approximation for the gfe up to the same order . one will see that the adoption of the pms is crucial to the variational perturbation approximation ( vpa ) method . it is the pms that guarantees the meaningfulness , non - perturbational nature and effectiveness of the vpa results . to illustrate the method , we will consider a classical harmonic oscillator , which was used in ref . @xcite , and calculate the gfe up to the third order in the variational perturbation expansion . numerical comparisons between the first- , second- and third - order and exact results are made and indicate that the vpa method provides a better approximation than the variational method in ref . @xcite . nowadays , owing to the so - called normalization " problem of tsallis non - extensive thermostatistics , there have existed four versions for the tnesp @xcite . the first version with the usual constraint on the internal energy in ref . @xcite was used , for some special systems , only a couple of times in the past , and the second version with the non - normalized constraint on the internal energy in ref . @xcite has been intensively studied and used , and furthermore the aforementioned variational and perturbation methods were developed in this version . the third version with the normalized constraint on the internal energy in ref . @xcite is a satisfactory version , but is very complicated for performing because there exist implied relations between relevant quantities , and so the fourth version with the non - normalized constraint on centered operators appears to unentangle the problem on the implied relations in the third version @xcite . happily , the four versions can be easily derived from just any one of them @xcite . moreover , we notice that the second version has been applied to many systems with providing satisfactory theoretical and/or experimental results @xcite . hence , here we will use the second version worked out in ref . @xcite to perform the vpa method . next , the tsallis statistical mechanics with the canonical ensemble and the perturbation and variational methods in ref . @xcite will briefly be introduced for the convenience of our later investigations . in sect . iii , the vpa method will be stated and truncated expressions of the expansion series of the gfe in the second version in ref . @xcite for a system will be derived up to the third order . iv will contribute to investigating the classical harmonic oscillator , and conclusions will be made in sect . v. for a system with a non - extensive index @xmath1 , the tsallis entropy is defined as @xcite @xmath16 where @xmath17 is the boltzmann constant and @xmath18 is the total number of microscopic possibilities @xmath19 of the system . note that @xmath1 is a real number and characterizes the degree of non - extensiveness . when @xmath4 , eq.(1 ) leads to the usual entropy @xmath20 . the three cases of @xmath7 , @xmath4 and @xmath21 characterize superextensivness , extensiveness and subextensiveness of systems , respectively . based on the tsallis entropy , a generalized equilibrium thermostatistics can be established by making @xmath22 extremal with appropriate constraints present . for the canonical ensemble , four possible candidates exist for the constraints and hence lead to the aforementioned four versions for tsallis statistical mechanics . this section will use the following non - normalized constraint on the internal energy @xmath23 @xcite : @xmath24 where @xmath25 is the @xmath26th eigenvalue of the hamiltonian of the system . ( the corresponding constraint used in the first version of tsallis statistical mechanics is @xmath27 , and the one in the third version is the normalized one @xmath28 . ) thus , making @xmath22 extremal with present the constraint eq.(2 ) and the normalization property of the probability yields the following power - law distribution @xcite @xmath29^{1/(1-q)}}{z_q(\beta^*)}}\ ] ] with the generalized partition function @xmath30^{1/(1-q)},\ ] ] where @xmath31 is the lagrange multiplier associated with the internal - energy constraint , eq.(2 ) . in the case of @xmath7 , the summation in the distribution eq.(2 ) will be cut off for those energy eigenvalues higher enough to give negative probabilities . it is easy to verify that when @xmath2 , the power - law distribution eq.(2 ) tends to the conventional exponential - law distribution . from eqs.(2 ) , ( 3 ) and ( 4 ) , one can easily read @xmath32 where the @xmath1-logarithm function @xmath33 is a generalization of the usual logarithm function . through legendre transform on @xmath34 , which depends on @xmath31 , one can find the relation @xmath35 . note that the relation between the generalized internal energy and tsallis entropy takes the same form as in the conventional statistical mechanics . in fact , introducing @xmath36 , and defining generalized thermodynamic function in the same way as that in the conventional thermodynamics , one can find the same legendre structure as in the conventional thermodynamics . for example , one can have the gfe @xmath37 using the basic limit formula @xmath38 , one can verify that the above generalized thermodynamic function and partition function are reduced to the usual ones in the limit of @xmath2 . from the above , one can see that although the usual thermodynamical legendre structure remains valid in tsallis statistical mechanics , it is evident that the original calculation techniques can not be directly borrowed into the generalized theory . one has to design calculation techniques for the generalized theory . as was stated in the introduction , a united presentation of the perturbation and variational methods for the tnesp has been given in ref . next , for convenience of later contrast , we give a brief introduction on them in the notations here . assume that the hamiltonian can be written as @xmath39 where , @xmath40 is the hamiltonian of a soluble model , @xmath41 is small enough so that it can be considered as a perturbation on @xmath40 , and @xmath42 is the perturbation parameter . the gfe of the system with @xmath43 , @xmath44 , is a function of @xmath42 and can be expanded as @xmath45 with @xmath46 , @xmath47 , @xmath48 . eq.(8 ) is a perturbation expansion series of @xmath44 in powers of @xmath42 , and to truncate it at some order , for example @xmath11th order , can provide an approximation to @xmath44 up to the @xmath11th order . in ref . @xcite , approximate expressions for @xmath44 was derived up to the second order . through analyzing the approximation expression of @xmath44 up to the second order , ref . @xcite obtained the generalized bogoliubov inequality @xmath49 where the symbol @xmath50 is the operator corresponding to the physical quantity @xmath51 and @xmath52 ( and @xmath53 ) is the @xmath26th eigenstate for @xmath54 . this inequality stems from the approximation expression of @xmath44 up to the first order in the perturbation expansion , and affords a variational method . according to ref . @xcite , the approximate expression of @xmath44 up to the @xmath11th order is valid only for the case of @xmath55 because the derivation of the approximate expressions requires the interchangeability between the sum over the quantum numbers and the derivative with respect to @xmath42 . as was stated below eq.(4 ) , when @xmath7 , the sum in eq.(4 ) and , accordingly , eq.(6 ) has a cutoff and so its upper limit depends upon @xmath42 , leading to the above limitation on the resultant expressions in ref . therefore , the variational method in ref . @xcite can be used only for @xmath10 . we stop here for introducing the perturbation and variational methods in ref . @xcite . next , we turn to our investigation in the present paper . for simplicity of the notation , @xmath56 will be taken from now on . to develop the vpa method , we consider calculating the gfe @xmath57 ( of course , the method can be used to calculate other generalized thermodynamic potentials ) . we begin with modifying the hamiltonian @xmath43 . firstly , differently from eq.(7 ) in the perturbation method , we add the zero term @xmath58 to @xmath43 and write it as @xmath59 with @xmath60 and @xmath13 an auxiliary parameter . in eq.(10 ) , @xmath61 is the hamiltonian of some exactly soluble model which is originally not included in @xmath43 , and the exact solubility of the model is not affected by values of the auxiliary parameter @xmath13 . that is to say , for a value of @xmath13 , the energy eigenvalues @xmath62 ( non - degenerate , for simplicity ) , eigenstates @xmath63 , the probability @xmath64 , the generalized partition function @xmath65 and the gfe @xmath66 for the model with @xmath61 are exactly known . secondly , an artificial parameter @xmath14 is inserted as a factor before @xmath67 in eq.(10 ) , that is , @xmath43 in eq.(10 ) is now modified as @xmath68 thus , we have a new system with the hamiltonian @xmath69 which becomes @xmath61 and @xmath43 in the cases of @xmath70 and @xmath15 , respectively . note that different from @xmath43 in eq.(10 ) , @xmath69 depends on @xmath13 for @xmath71 . the above modification provides a possibility that we calculate @xmath57 for the original system through considering the new system . for the system with @xmath69 , suppose that the set of energy eigenvalues are the spectra @xmath72 , and the corresponding eigenfunctions is @xmath73 . when the system is in thermal contact with a reservoir , @xmath74 when we adopt a generalized canonical ensemble , we have the probability distributions @xmath75 , and can calculate the gfe @xmath76 in the presence of the constraint eq.(2 ) according to the definition and expression eq.(6 ) . in analogy to eq.(8 ) , one can expand @xmath76 as the following series @xmath77 with @xmath78 and @xmath79 . this is only a formal expansion , and the parameter @xmath14 acts as an expansion order index and need nt be small . taking @xmath15 in the last equation , we obtain an expansion series of @xmath57 for the system with @xmath43 which is independent of @xmath13 if the resultant series is not truncated . now we have a simple discussion on the expansion eq.(12 ) . of course , if @xmath61 is naively one part of the original @xmath43 and @xmath80 can be regarded as a perturbation on @xmath61 ( in this case , @xmath13 is one of system parameters in @xmath43 ) , then eq.(12 ) with @xmath15 is just the perturbation expansion series of @xmath57 , eq.(8 ) . if @xmath61 is not a naive part of @xmath43 , as we assumed in eq.(10 ) , the truncated expression at the first order in @xmath14 can lead to a variational method , and substituting the value @xmath13 from the variational method into eq.(12 ) with @xmath15 will produce an expansion series of @xmath57 around the variational result . in this case , if the resultant variational method can provide a good approximation result for @xmath57 , then it can be presumed that the expansion of @xmath57 around the variational result maybe afford a good non - perturbational approximation which will improve variational results . this is the original version of the variational perturbation theory , but the expansion series is possibly divergent @xcite . if the variational method can not provide a good approximation result for @xmath57 , the expansion of @xmath57 around the variational result will make no senses . that is to say , in general , the above expansion series eq.(12 ) seems to be useless . however , as was mentioned in sect.i , once the pms is employed to single out an appropriate value of @xmath81 , eq.(12 ) can afford a presumably convergent , systematic non - perturbation approximation method , a modern version of the variational perturbation theory @xcite . in eq.(12 ) , truncating the expansion series at some order in @xmath14 and then letting @xmath15 , the resultant truncated expression is a function of the parameter @xmath13 , albeit eq.(12 ) with @xmath15 is independent of @xmath13 . in principle , the value of the truncated expression can possibly take any value with variance of @xmath13 , including the exact value of @xmath57 . since @xmath57 is a constant in the space of @xmath13 ( the exact @xmath57 is independent of @xmath13 ) , it should be believable that a requirement of the truncated result varying most slowly with the parameter @xmath13 can likely make the truncated result provide a most reliable approximate result for the exact @xmath57 . employing this requirement to determine the parameter @xmath13 is the main spirit of the so called pms @xcite . generally , the curve of a function varies more slowly near its extremum than in its slope part , and so a simple realization of the pms is to require the first derivative of the truncated expression with respect to @xmath13 to be zero . thus , substituting the appropriate value of @xmath13 determined at the truncated order into the corresponding truncated expression of @xmath57 will provide a reasonable and reliable approximation for @xmath57 up to the same order . such a truncated result at some order is the approximate result of @xmath57 up to the same order in the vpa method . in brief , the vpa method consists in only two crucial steps : one is to formally expand the quantity @xmath57 in consideration with an adjustable parameter @xmath13 entered , and the other is to determine the value of @xmath13 from the truncated expression of @xmath57 according to the pms . now , we continue to perform the first step , derive the first four terms in eq.(12 ) ( the derivations here are similar to those in ref . @xcite ) , and simultaneously indicate the vpa procedures up to the first , second and third orders , respectively . the first term in eq.(12 ) is straightforwardly @xmath82 with @xmath83^{1/(1-q)}.\ ] ] for deriving the other three terms , eigenstates of @xmath84 are needed , and so we first consider the eigenequation , @xmath85 . by mimicing rayleigh - schrdinger perturbation theory , after substituting the expansions @xmath86 and @xmath87 into the eigenequation , one can formally have @xmath88 @xmath89|m{\stackrel{(0)}{>}}\nonumber \\ & & \hspace*{0.2 cm } -{\frac { 1}{2}}\sum_{m\not = n } { \frac { ( h_{i,\ ; mn})^2}{(e_n^{(0 ) } -e_m^{(0)})^2}}|n{\stackrel{(0)}{>}}\end{aligned}\ ] ] and so on . here , @xmath90 . by the way , the expansion expressions , eqs.(15 ) and ( 16 ) , together with the pms can lead to a vpa method to solve schrdinger equation @xcite(prd)@xcite . for the second term of eq.(12 ) , @xmath91 can be calculated as @xmath92\biggl |_{\epsilon=0}=z_{q,\ ; \epsilon}^{-q } \sum_n [ 1-(1-q)\beta^ * e_n]^{\frac { q } { 1-q } } { \frac { \partial e_{n,\ ; \epsilon}}{\partial \epsilon}}\biggl|_{\epsilon=0}\ ] ] with @xmath93 the generalized partition function for the system with @xmath69 . from hellmann - feynman theorem @xmath94 @xcite , we have @xmath95 ^{\frac { q } { 1-q } } < \epsilon , n|\hat{h}_i|n,\epsilon>\biggr ] \biggl|_{\epsilon=0 } \nonumber \\ & = & \sum_n p_q^q(e_n^0)h_{i,\ ; nn } = < h_i>_q^0 , \end{aligned}\ ] ] which takes the same form as eq.(5 ) in ref . @xcite . thus , the truncated expression for @xmath57 at the first order , @xmath96 , is @xmath97 the right hand side of last equation has the same form as that in the right hand side of the inequality , eq.(9 ) . @xmath96 is a function of @xmath13 and a different value of @xmath13 produces a different approximation for @xmath57 which maybe have nothing to do with the exact value of @xmath57 . however , if the value of @xmath13 in eq.(21 ) is chosen from roots of the following condition @xmath98 according to the pms , then @xmath96 in eq.(21 ) would produce a most reliable approximation for @xmath57 up to the first order . since @xmath57 has the same form as the right hand side of the inequality , eq.(9 ) , the vpa procedure up to the first order for the case of @xmath99 is consistent with the variational method based on the bogoliubov inequality in ref . @xcite , and so the appropriate root of eq.(22 ) for @xmath13 makes @xmath96 minimal . for the case of @xmath100 , an example in next section shows that the appropriate root of eq.(22 ) for @xmath13 also makes @xmath96 minimal . this suggests that the vpa procedure up to the first order is generally a variational method which provides a reliable upper limit for the exact @xmath57 . in order to give the truncated expression for the gfe at the second order @xmath101 , we calculate @xmath102 as follows . a straightforward differentiation gives @xmath103+\sum_n ( p_q(e_n^{(0)}))^q { \frac { \partial<\epsilon , n|h_i|n,\epsilon>}{\partial \epsilon } } \biggl|_{\epsilon=0 } .\end{aligned}\ ] ] from eq.(15 ) , we have @xmath104 . consequently , eq.(23 ) becomes @xmath105 \nonumber \\ & & + 2\sum_n ( p_q(e_n^{(0)}))^q \sum_{m\not = n } { \frac { |h_{i,\ ; nm}|^2}{e_n^{(0)}-e_m^{(0 ) } } } .\end{aligned}\ ] ] formally , the right hand side of eq.(24 ) can be rewritten as eq.(6 ) in ref . @xcite . according to the pms , using the reasonable root of the condition @xmath106 as the value of @xmath13 , one can give the approximation result for @xmath57 up to the second order from @xmath107 . the approximation result up to the second order would improve the approximation result up to the first order . similarly , one can calculate @xmath108 . in the calculation , we need the additional relation @xmath109 , which can be easily obtained from eqs.(15 ) , ( 17 ) and ( 18 ) . thus , @xmath108 can be written as @xmath110 \sum_{m\not = n}{\frac { |h_{i , nm}|^2}{e_n^{(0)}-e_m^{(0 ) } } } \nonumber \\ & & + 6\sum_n ( p_q(e_n^{(0)}))^{q } \biggl [ \sum_{l\not = n}\sum_{m\not = n}{\frac { h_{i , nl}h_{i , lm}h_{i , nm } } { ( e_n^{(0)}-e_m^{(0)})(e_n^{(0)}-e_l^{(0)})}}- \sum_{m\not = n } { \frac { |h_{i , nm}|^2 h_{i,\ ; nn } } { ( e_n^{(0)}-e_m^{(0)})^2 } } \biggr ] .\end{aligned}\ ] ] then one can have the truncated expression for @xmath57 at the third order , @xmath111 . furthermore , using the reasonable root of the condition @xmath112 as the value of @xmath13 , one can give the approximation result for @xmath57 up to the third order from @xmath113 . in the same way , one can consider vpa results for @xmath57 up to higher orders . obviously , the vpa results up to various orders constitute a sequence of the approximation results of @xmath57 . owing to the employment of the pms , the values of @xmath13 determined depend upon the truncated orders , and the value of @xmath13 at a given order is different from values of @xmath13 at other orders . presumably , it is the dependence of @xmath13 upon the truncated order that makes the sequence of the vpa results of @xmath57 up to various orders converge to the exact @xmath57 . although we can not verify this convergent property of the variational perturbation theory , investigations in other fields have provided such a few examples @xcite . additionally , in the example of next section , generally , the vpa result up to the third order approaches more closely the exact @xmath57 than the approximations up to the lower orders . the second crucial step plays a vital role in the vpa method , and one have to perform it very carefully . sometimes , eqs.(22 ) , ( 25 ) and ( 27 ) can not provide appropriate roots for @xmath13 , and in those cases one should render the second derivative of the truncated expressions for @xmath57 with respect to @xmath13 zero to determine @xmath13 . this can be understandable because the curve of a function varies more slowly near its knee than in its slope part ( as a matter of fact , when there exist both an extremum and a knee , one should analyze behaviors of the truncated expressions as a function of @xmath13 near the extremum and knee to determine @xmath13 according to the pms ) . in case the second derivative condition can not provide an appropriate root for @xmath13 , too , then the vpa method can not provide an approximation result for @xmath57 up to the truncated order in consideration , and one should further consider vpa result up to the next order . furthermore , when there exist multi - roots of the vanishing derivatives for @xmath13 , one should choose to adopt the root near which the truncated result varies most slowly with @xmath13 . next section , we will show how to determine @xmath13 according to the pms in a concrete example . in concluding this section , we intend to emphasize one point . in the above calculations , the interchange between the differentiation with respect to @xmath14 and the sum over eigenstates was involved . as was mentioned in last section , the interchangeability between the differentiation and the sum leads to a limitation for the variational and perturbation methods in ref . nevertheless , for the vpa method developed here , the relevant calculations are only formal calculations and the resultant expressions are independent of @xmath14 , and presumably , we suggest that the vpa method need nt suffer those limitations on the methods in ref . @xcite . at least , this point yields no problem in the example of next section . for a one - dimensional classical harmonic oscillator with mass @xmath114 and an angular frequency @xmath115 , the hamiltonian is @xmath116 with @xmath117 the coordinate and @xmath118 the momentum . in a manner analogous to the one used for the quantum oscillator , we can associate with the oscillator a continuous energy spectra @xmath119 , where @xmath120 is an arbitrary positive constant with the dimension of energy and @xmath11 any positive real number including @xmath121 @xcite . then , from eq.(4 ) , the generalized partition function can be written and calculated as @xmath122^{1/(1-q ) } dn={\frac { 1}{(2-q)\beta^*\delta_o}}\ ] ] with @xmath123 for the case of @xmath21 and @xmath124 $ ] for the case of @xmath7 ( in this section , we suppose that @xmath125 . ) . the right hand side of eq.(28 ) is valid only for @xmath9 and @xmath126 is divergent for @xmath127 . so , eq.(6 ) leads to the gfe for the classical harmonic oscillator @xmath128.\ ] ] last equation with @xmath129 is just eq.(14 ) in ref . @xcite . on the other hand , for a particle with mass @xmath114 which moves in a one - dimensional box , the hamiltonian @xmath130 with @xmath131 for @xmath132 and @xmath133 for @xmath134 . in analogy to what we did for the oscillator , we can associate with the particle a continuous energy spectra @xmath135 , where @xmath136 is an arbitrary positive constant with the dimension of action and @xmath11 any positive real number including @xmath121 . so the generalized partition function for the particle in the box can be written and calculated as @xmath137^{1/(1-q ) } dn=\biggl \ { \begin{array}{ll } { \frac { l}{2\delta_b}}\sqrt{{\frac { 2 m}{(1-q)\beta^ * } } } b({\frac { 1}{2}},{\frac { 2-q}{1-q } } ) , & q<1 \\ { \frac { l}{2\delta_b}}\sqrt{{\frac { 2 m}{(q-1)\beta^ * } } } b({\frac { 1}{2}},{\frac { 3-q}{2(q-1 ) } } ) , & q>1 \end{array}\end{aligned}\ ] ] with @xmath123 for the case of @xmath21 and @xmath138 for the case of @xmath7 . in eq.(30 ) , @xmath139 is the beta function . note that @xmath65 is convergent only for the case of @xmath140 . in the calculation of the above equation , the formulae 8.380(1 ) and 8.380(3 ) in ref . @xcite were employed . owing to @xmath141 with @xmath142 the gamma function ( 8.384(1 ) in ref . @xcite ) , eq.(30 ) with @xmath143 is nothing but eq.(11 ) in ref . @xcite . from eqs.(6 ) and ( 30 ) , the gfe for the classical particle in the box is easily calculated as @xmath144 with @xmath145 since the above two systems are exactly solved , ref . @xcite employed them to illustrate the variational method there . in the present paper , we will use them to illustrate the vpa method . that is , by regarding the hamiltonian of the classical particle in the box as @xmath61 in eq.(10 ) and using the vpa scheme in last section , we will calculate the gfe for the classical harmonic oscillator up to the third order and then make comparisons among the results up to the various orders and the exact result for the oscillator . next , the present section will be divided into three subsections . in subsection a , the truncated variational perturbation expressions for the classical harmonic oscillator will be given up to the third order . the values of @xmath13 in the truncated expressions will be determined according to the pms in subsection b , and the approximated values of the gfe will be calculated and compared with the exact result in subsection c. take the width @xmath146 of the box as the adjustable parameter @xmath13 in our vpa scheme . then , @xmath147 in last section is now @xmath148 for the classical particle in the box , the speed @xmath149 corresponds to the particle energy @xmath62 . in a state with the energy @xmath62 , the particle moves in the box back and forth at the constant speed @xmath150 , and so the probability of finding the particle in one of the directions near @xmath117 ( @xmath151 ) is @xmath152 with @xmath153 ( except for @xmath154 ) , but zero out of the box . thus , in the present case , the matrix elements @xmath155 appeared in eqs.(20 ) , ( 24 ) and ( 26 ) can be straightforwardly calculated as @xmath156 where @xmath157 for @xmath158 and @xmath159 . now we are at the position to calculate the truncated expressions for @xmath57 of the classical harmonic oscillator up to the third order from eqs.(13),(20),(24 ) and ( 26 ) . the zeroth - order expression is eq.(31 ) . at the first order , one can have , from eq.(20 ) , @xmath160^{q/(1-q ) } dn \nonumber \\ & = & { \frac { m\omega^2l^2}{24}}(z_{q,\ ; 0})^{-q}\biggl \ { \begin{array}{ll } { \frac { l}{2\delta_b}}\sqrt{{\frac { 2 m}{(1-q)\beta^ * } } } b({\frac { 1}{2}},{\frac { 1}{1-q } } ) , & q<1 \\ { \frac { l}{2\delta_b}}\sqrt{{\frac { 2 m}{(q-1)\beta^ * } } } b({\frac { 1}{2}},{\frac { q+1}{2(q-1 ) } } ) , & q>1 \end{array } \nonumber \\ & = & { \frac { m \omega^2 l^2}{24}}{\frac { 3-q}{2 } } c_q \;.\end{aligned}\ ] ] the right hand side of last equation is identical to eq.(12 ) in ref . @xcite . at the second order in @xmath14 , eq.(24 ) leads to @xmath105 \nonumber \\ & = & \beta^ * q z_{q,\ ; 0}^{q-1}\biggl [ ( f_q^{(1)})^2-{\frac { m^2\omega^4l^4}{(24)^2}}(z_{q,\ ; 0})^{1 - 2q}\int_0^n[1-(1-q)\beta^ * { \frac { n^2\delta_b^2}{2 m l^2}}]^{(2q-1)/(1-q ) } dn\biggr ] \nonumber \\ & = & \beta^ * q z_{q,\ ; 0}^{q-1}(f_q^{(1)})^2 - \beta^ * q z_{q,\ ; 0}^{q-1}{\frac { m^2\omega^4 l^4}{(24)^2}}(z_{q,\ ; 0})^{1 - 2q}\biggl \ { \begin{array}{ll } { \frac { l}{2\delta_b}}\sqrt{{\frac { 2 m}{(1-q)\beta^ * } } } b({\frac { 1}{2}},{\frac { q}{1-q } } ) , & q<1 \\ { \frac { l}{2\delta_b}}\sqrt{{\frac { 2 m}{(q-1)\beta^ * } } } b({\frac { 1}{2}},{\frac { 3q-1}{2(q-1 ) } } ) , & q>1 \end{array } \nonumber \\ & = & \beta^ * { \frac { m^2\omega^4 l^4}{(24)^2 } } { \frac { ( q-3)(1-q)^2}{4 } } c_q \;.\end{aligned}\ ] ] finally , we consider the third order . eq.(26 ) , in the present case , is reduced to @xmath161^{{\frac { 2q-1}{1-q } } } dn \nonumber \\ & & + { \beta^*}^2 q(2q-1)z_{q,\ ; 0}^{-q}{\frac { m^3\omega^6l^6}{(24)^3 } } \int_0^n[1-(1-q)\beta^ * { \frac { n^2\delta_b^2}{2 m l^2}}]^{{\frac { 3q-2}{1-q } } } dn \nonumber \\ & = & { \beta^*}^2 q(q+1 ) z_{q,\ ; 0}^{2q-2}(f_q^{(1)})^3 - 3{\beta^*}^2q^2 z_{q,\ ; 0}^{-1}{\frac { m^2\omega^4l^4}{(24)^2 } } f_q^{(1 ) } \biggl \ { \begin{array}{ll } { \frac { l}{2\delta_b}}\sqrt{{\frac { 2 m}{(1-q)\beta^ * } } } b({\frac { 1}{2}},{\frac { q}{1-q } } ) , & q<1 \\ { \frac { l}{2\delta_b}}\sqrt{{\frac { 2 m}{(q-1)\beta^ * } } } b({\frac { 1}{2}},{\frac { 3q-1}{2(q-1 ) } } ) , & q>1 \end{array } \nonumber \\ & & + { \beta^*}^2 q(2q-1)z_{q,\ ; 0}^{-q}{\frac { m^3\omega^6l^6}{(24)^3}}\biggl \ { \begin{array}{ll } { \frac { l}{2\delta_b}}\sqrt{{\frac { 2 m}{(1-q)\beta^ * } } } b({\frac { 1}{2}},{\frac { 2q-1}{1-q } } ) , & q<1 \\ { \frac { l}{2\delta_b}}\sqrt{{\frac { 2 m}{(q-1)\beta^ * } } } b({\frac { 1}{2}},{\frac { 5q-3}{2(q-1 ) } } ) , & q>1 \end{array } \nonumber \\ & = & { \beta^*}^2{\frac { m^3\omega^6l^6}{(24)^3}}{\frac { ( 1+q)(3-q)(q-1)^3}{8}}c_q \;.\end{aligned}\ ] ] thus , collecting eqs.(31),(34),(35 ) and ( 36 ) , one can get @xmath162 , @xmath163 and @xmath164 , the truncated expressions for @xmath57 of the classical oscillator at the first , second and third orders . when @xmath7 , the truncated expression for @xmath57 at the third order is @xmath165 \;,\end{aligned}\ ] ] and when @xmath166 , it is @xmath167 \;,\end{aligned}\ ] ] for the case of @xmath168 , @xmath169 in eq.(38 ) is divergent . in eqs.(37 ) and ( 38 ) , all the terms in which the powers in @xmath170 are lower than the second and the third powers give the truncated expressions for @xmath57 at the first and second orders , respectively . next subsection , we will determine @xmath146 in them according to the pms to get approximations for @xmath57 . in the following subsections , when a numerical calculation is performed , we will take @xmath171 , and we will consider only the case of @xmath9 which is the convergent range of @xmath57 for the oscillator . ( @xmath172 corresponds to the case @xmath173 in ref . @xcite . ) @xmath146 must be positively real . up to the first order in @xmath14 , for the cases of both superextensiveness and subextensiveness , eq.(22 ) leads to the equation @xmath174 to see whether @xmath162 is most insensitive to @xmath146 near the positive root of eq.(39 ) or not , one can analyze varying property of @xmath162 as a function of @xmath146 . for illustration , the function @xmath162 is depicted for @xmath175 and some values of @xmath1 in fig.1 . the dependence of @xmath162 upon @xmath146 for some values of @xmath1 and at @xmath175 . the solid , dashed , dot - dashed , dotted and long - dashed curves from right to left correspond to @xmath176 and @xmath177 , respectively . every curve in this figure has a minimum at a positive root for @xmath146 . ] fig.1 indicates that the positive root of eq.(39 ) for @xmath146 makes @xmath162 reach a minimum and simultaneously be most insensitive to @xmath146 . so , we should choose this positive root as the value of @xmath146 @xmath178 which is identical to eq.(13 ) in ref . the minimum @xmath179 provides an upper limit for @xmath57 . thus , the vpa up to the first order is really a variational method and it , in the case of @xmath10 , is consistent with the variational method in ref . @xcite which is based on the generalized bogoliubov inequality , eq.(9 ) in ref . @xcite . up to the second order , for the cases of both superextensiveness and subextensiveness , eq.(25 ) leads to the equation @xmath180 last equation produces four roots for @xmath146 . the roots are real only when @xmath181 and @xmath182 , and two of them are positive as well as the other roots are negative . to choose an appropriate root as @xmath146 , @xmath163 , as a function of @xmath146 , is representatively illustrated for @xmath175 and some values of @xmath1 in fig.2 . the dependence of @xmath163 upon @xmath146 for some values of @xmath1 and at @xmath175 . the dot - dashed and dotted curves are for @xmath183 and @xmath184 , respectively , and the solid and dashed curves in the image are for @xmath185 and @xmath186 , respectively . horizontal lines represent exact values of @xmath57 , and every pair of curve and horizontal line with an identical line - type is for the same value of @xmath1 . in the figure , every curve has a minimum and a maximum . ] in fig.2 , horizontal lines represent exact values of @xmath57 , and every pair of curve and horizontal line with an identical line type is for the same value of @xmath1 . from curves in fig.2 , one can see that the smaller positive root makes @xmath163 minimal whereas the larger positive root makes @xmath163 maximal . hence , @xmath163 is more insensitive to @xmath146 near the smaller positive root than near the other positive root , and so the smaller root should be chosen as the value of @xmath146 , @xmath187 , for @xmath181 . for the other values of @xmath1 , eq.(25 ) can not produce reasonable values for @xmath146 . in this case , the vanishing requirement for the second derivative of @xmath163 with respect to @xmath146 should be considered and yields the following equation for @xmath146 @xmath188 which is valid for the cases of both superextensiveness and subextensiveness . eq.(42 ) produces four roots for @xmath146 . there exist real roots only when @xmath189 and @xmath190 , and for @xmath191 , two of them are positive as well as the other roots are negative , and for @xmath10 , only one positive root exists . to choose an appropriate root as @xmath146 for the case of @xmath192 , @xmath163 , as a function of @xmath146 , is typically illustrated for @xmath175 and @xmath193 in fig.3 . for completeness , we also depict @xmath163 for @xmath194 ( which is involved in the range of @xmath195 ) in fig.3 as an image . similar to fig.2 , and the solid curve is for @xmath193 and the dashed curves in the image is for @xmath194 . the solid curve has two knees and the dashed line in the image has only one knee . ] in this figure , the solid curve for @xmath193 has two knees which correspond to the two positive roots of eq.(42 ) for @xmath146 : @xmath196 and @xmath197 . obviously , @xmath163 is more insensitive to @xmath146 near the larger positive root than near the other positive root . the same situation occurs for other negative values of @xmath1 ( @xmath192 ) , and so the larger positive root of eq.(42 ) should be chosen as the value of @xmath146 , @xmath187 , for @xmath192 . furthermore , as is illustrated in the image of fig.3 , for the case of @xmath198 , @xmath163 is most insensitive to @xmath146 near the positive root of eq.(42 ) , and so the positive root of eq.(42 ) should be chosen as the value of @xmath146 , @xmath187 , for @xmath195 . by the way , for the case of @xmath199 , there exists not any appropriate root of eq.(41 ) or eq.(42 ) for @xmath146 and so the vpa method can not provide an approximation for @xmath57 up to the second order in this case . finally , we determine @xmath146 so as to approximate @xmath57 up to the third order . for the cases of both superextensiveness and subextensiveness , eq.(27 ) leads to the equation @xmath200 last equation produces six roots for @xmath146 . among the six roots , there exist three positive roots for @xmath201 , two positive roots for @xmath202 and only one positive root for @xmath203 , @xmath204 and @xmath205 . in fig.4 , @xmath164 , as a function of @xmath146 , is typically depicted for some values of @xmath1 and @xmath31 . the dependence of @xmath164 upon @xmath146 for some values of @xmath1 and @xmath31 . the solid , short - dashed , dot - dashed , dotted and long - dashed curves correspond to @xmath206 and @xmath207 , respectively . the solid and the long - dashed curves have a minimum , the short - dashed and the dot - dashed curves have a minimum and a maximum , and the dotted curve has two minima and one maximum . ] in fig.4 , the dotted curve is for @xmath208 , and has two minima and a maximum , which corresponds to the case of having three positive roots . in this case , it is evident , from fig.4 , that @xmath164 is most insensitive to @xmath146 near the smallest positive root of eq.(43 ) ( the curvature of the curve at the smallest positive root is smaller than that at the largest positive root ) , and so the smallest positive root should be chosen as the value of @xmath146 , @xmath209 , for @xmath201 . the short - dashed and the dot - dashed curves are for @xmath210 and @xmath211 and have a minimum at the smaller positive root and a maximum at the other positive root . so , for the case of having two positive roots , the smaller positive root is appropriate for @xmath146 when @xmath202 . as for the cases of @xmath203 and @xmath8 , the solid and the long - dashed curves indicate that @xmath164 reaches an extremum , the minimum at the positive root , and so it can be taken as the value of @xmath146 , @xmath209 . according to the pms , we have determined @xmath146 for all cases which we are interested in . in the above , analytical and numerical discussions are done with mathematica , and we do not list all expressions of roots of eqs.(41 ) , ( 42 ) and ( 43 ) because they are too lengthy . note that although @xmath146 is determined as @xmath212,@xmath187 and @xmath209 for approximating @xmath57 up to the first , second and third orders , respectively , it is not meant that @xmath146 is being expanded as a series . from analysis and results in last subsection , employing mathematica package , one can get , for various ranges of @xmath1 , the expressions of @xmath213 , @xmath214 and @xmath215 , the approximations of the gfe @xmath57 up to the first , second and third orders . regarding them as functions of @xmath216 ( @xmath217 ) and the non - extensiveness index @xmath1 , respectively , we can numerically calculate and compare them with the exact @xmath57 , eq.(29 ) . the exact free energy , @xmath57 in eq.(29 ) , reaches a maximum at @xmath218 for any given value of @xmath9 . as a function of @xmath1 , @xmath57 in eq.(29 ) has a maximum for any @xmath219 or so , and , otherwise , reaches first a maximum and then a minimum when @xmath1 increases up to @xmath220 . in ref . @xcite , the approximate gfe from variational method there was considered for @xmath221 , and when @xmath216 increases or when @xmath1 decreases , the discrepancies between the approximate values and the exact values become more and more evident . here , we consider the vpa gfe up to the first order , @xmath213 , for all values of @xmath9 , which with @xmath221 is identical to that in ref . basically , @xmath213 , as a function of @xmath216 or @xmath1 , mimics the feature of the exact free energy , and for a given value of @xmath216 , when @xmath1 approaches @xmath222 or is sufficiently negative , the error @xmath223 is very small whereas @xmath224 is not small for intermediate values of @xmath1 . interestingly , the dependent feature of @xmath224 upon @xmath1 is similar to that of curvature of @xmath162 at @xmath225 , @xmath226 , upon @xmath1 ( for subextensiveness case , when @xmath216 is small , there exists not such a similarity . ) . for an illustration of this similarity , in fig.5 , we depict the dependence of @xmath224 ( dashed curve ) and @xmath227 ( solid curve ) upon @xmath1 at @xmath228 . the dependence of @xmath224 ( dashed curve ) and @xmath227 ( solid curve ) upon @xmath1 at @xmath228 . ] in fig.5 , the image is drawn for the range of @xmath229 , and evidently indicates that while the solid curve reaches the maximum , the dashed curve also gets to its maximum . this similarity can be understood from the pms . smaller the curvature at @xmath225 is , more slowly @xmath162 varies near @xmath225 , and so , closer @xmath213 approaches @xmath57 according to the spirit of the pms , yielding the similarity . the approximation of @xmath57 up to the second order , @xmath214 , improves the approximation of @xmath57 up to the first order , @xmath213 , the variational result for the range of @xmath230 where @xmath214 makes senses , and the approximation of @xmath57 up to the third order , @xmath215 , generally further improves the variational result . for a comparison and an illustration , the exact and various approximate results are shown at @xmath231 in fig.6 . the dependence of @xmath57 and its approximations up to various orders upon @xmath1 at @xmath231 . the solid , short - dashed , dot - dashed and dotted curves are the exact @xmath57 and its approximations @xmath213 , @xmath214 and @xmath215 , respectively . the dot - dashed curve interrupts at @xmath232 or so , and almost coincides with the dotted curve when @xmath233 or so . the long - dashed curve is @xmath234 , and goes up quickly when @xmath235 or so . ] in fig.6 , the solid , short - dashed , dot - dashed and dotted curves are @xmath57 , @xmath213 , @xmath214 and @xmath215 , respectively ( note that the dot - dashed curve , @xmath214 , interrupts at @xmath232 or so , and almost coincides with the dotted curve , @xmath215 , when @xmath233 or so ) . additionally , the long - dashed curve in fig.6 is @xmath234 , and obviously it is a bad approximation for @xmath57 when @xmath235 or so , albeit @xmath236 can produce as a good approximation for @xmath57 as @xmath214 does . fig.6 indicates that the improvement of @xmath214 to the variational result is substantial when @xmath237 . furthermore , the approximation of @xmath57 up to the third order mimics the exact @xmath57 better than @xmath214 , and substantially improves the variational result for the range of @xmath238 where @xmath214 is invalid . for the cases of both @xmath239 and @xmath10 , @xmath214 and/or @xmath215 only slightly improve the variational result.(by the way , because @xmath57 and its various approximations here vary slowly with @xmath1 near their maxima , the humps of those curves in fig.6 are not evident . ) for further and more clear illustration and comparison , we also consider the dependence of the various approximations here upon @xmath216 . in figs.7,8,9 and 10 , @xmath57 , @xmath213 , @xmath214 and @xmath215 are depicted for some typical values of @xmath1 as the solid , short - dashed , dot - dashed and dotted curves , respectively , and , for the sake of clearness , we redraw them for the range of @xmath240 in the image . the dependence of @xmath57 and its approximations up to various orders upon @xmath216 at @xmath241 . the solid , short - dashed and dotted curves are the exact @xmath57 and its approximations @xmath213 and @xmath215 , respectively . the long - dashed curve is @xmath234 , going up when @xmath242 or so . they are redrawn for the range of @xmath240 in the image . ] similar to fig.7 , but @xmath243 , and the dot - dashed is @xmath214 . ] similar to fig.7 , but @xmath244 ] similar to fig.7 , but @xmath245 fig.7 is drawn for @xmath241 , and so there is not a dot - dashed curve owing to the invalidness of @xmath214 for this case . in this figure , the long - dashed curve is @xmath234 which goes up when @xmath242 or so , and is not given in fig.8 , 9 and 10 because it behaves too badly . fig.7 indicates that although @xmath214 does not exist , @xmath215 substantially improves the variational result and provide a better approximation for @xmath57 . fig.8 is depicted for @xmath243 . in this figure , although the dot - dashed and dotted curves are lower than the solid curve , they indicate that in this case , the approximations up to the first , second and third orders are distinct from each other , and evidently the second- and third - order approximations more closely approach the exact value than the variational result . fig.9 and 10 are drawn for @xmath244 and @xmath246 , respectively , which was considered in fig.1 in ref . @xcite . for these cases , the short - dashed , dot - dashed and dotted curves almost coincide and suggests that the approximations up to the second and third orders provide only a very small corrections to the variational results . by considering the gfe for a system , this paper proposed a vpa scheme for the generalized statistical mechanics based on the tsallis entropy . for approximating the gfe , we derived the truncated expressions for @xmath57 up to the third order in the variational perturbation expansion , and the classical harmonic oscillator was considered in detail for an illustration . the model investigation , albeit being a little academic @xcite , illustrates that the approximation up to the first order amounts to a variational method and covers the variational method in ref . @xcite , and the approximations up to the second and third orders improve the variational result and tend to approach the exact result . frankly , the variational perturbation expansion technique is formally similar to the perturbation expansion in ref . @xcite , the work in the present paper is to introduce the variational perturbation idea into the perturbation expansion and use the pms for determining the auxiliary parameter in the vpa scheme . it is these revisions that make the variational perturbational approximation method be non - perturbational , take the variational result as the first - order approximation and systematically improve the variational result . we believe that the investigation in the present paper is useful , at least , for calculating the generalized thermodynamical functions based on the tsallis statistics . finally , we intend to point out that since path - integral formalism has been developed @xcite , it is worth while developing vpa method within the formalism of the generalized thermodynamics . 99 s. abe and y. okamoto , eds . , _ nonextensive statistical mechanics and its applications _ , lectures notes in physics , vol . * 560 * , ( springer , berlin , 2001 ) . a full and timely list of references can be found in the following webpage : http://tsallis.cat.cbpf.br/biblio.htm . c. tsallis , j. stat . phys . * 52 * , 479 ( 1988 ) . r. p. feynman , _ statistical mechanics : a set of lectures _ , ( addison - wesley publishing company , inc . , reading , massachusetts , 1972 ) . a. plastino and c. tsallis , j. phys . a * 26 * , l893 ( 1993 ) . e. k. lenzi , l. c. malacarne and r. s. mendes , phys . lett . * 80 * , 218 ( 1998 ) . r. s. mendes , c. a. lopes , e. k. lenzi and l. c. malacarne , physica a , * 344 * , 562 ( 2004 ) . e. hylleraas , z. phys . * 65 * , 209 ( 1930 ) ; l. cohen , proc . soc . a 68 , 419 ( 1955 ) ; a 68 , 425 ( 1955 ) . h. kleinert and v. schulte - frohlinde , _ critical properties of @xmath247-theories _ , world scientific , singapore , 2001 , chapter 19 ; h. kleinert , _ path integrals in quantum mechanics , statistics , polymer physics , and financial markets _ , 3rd ed . , world scientific , singapore , 2004 . w. f. lu , c. k. kim , j. h. yee , k. nham , phys . d * 64 * , 025006 ( 2001 ) ; w. f. lu , s. k. you , jino bak , c. k. kim and k. nham , j. phys . a * 35 * , 21 ( 2002 ) ; w. f. lu , c. k. kim , jae hyung yee , k. nham , phys . lett . b * 540 * , 309 ( 2002 ) ; w. f. lu , c. k. kim , k. nham , phys . b * 546 * , 177 ( 2002 ) ; w. f. lu , phys . lett . b * 602 * , 261 ( 2004 ) . p. m. stevenson , phys . d * 23 * , 2916 ( 1981 ) ; phys . b * 100 * ( 1981 ) 61 ; phys . rev . d * 24 * , 1622 ( 1981 ) ; s. k. kaufmann and s. m. perez , j. phys . a * 17 * , 2027 ( 1984 ) ; p. m. stevenson , nucl . b * 231 * , 65 ( 1984 ) . e. m. f. curado and c. tsallis , j. phys . a * 24 * , l69 ( 1991 ) ; corrigendum , j. phys . a * 24 * , 3187 ( 1991 ) ; corrigendum , j. phys . a * 25 * , 1019 ( 1992 ) . c. tsallis , r. s. mendes and a. r. plastino , physica a * 261 * , 534 ( 1998 ) . s. martnez , f. nicols , f. pennini and a. r. plastino , physica a * 286 * , 489 ( 2000 ) ; g. l. ferri , s. martnez and a. r. plastino , physica a * 347 * , 205 ( 2005 ) . g. l. ferri , s. martnez and a. r. plastino , j. stat . p04009 ( 2005 ) . p. m. stevenson , phys . d * 30 * , 1712 ( 1984 ) . r. p. feynman , phys . rev . * 56 * , 340 ( 1939 ) ; d. b. lichtenberg , phys . d * 40 * , 4196 ( 1989 ) ; w. namgung , j. kr . . soc . * 32 * , 649 ( 1998 ) . a. duncan and h. f. jones , phys . rev . d 47 , 2560 ( 1993 ) ; c. m. bender , a. duncan , and h. f. jones , phys d * 49 * , 4219 ( 1994 ) ; r. guida , k. konishi , and h. suzuki , ann . ( n.y . ) * 241 * , 152 ( 1995 ) ; r. guida and k. konishi , ann . phys . * 249 * , 109 ( 1996 ) ; e. braaten and e. radescu , phys . 89 * , 271602 ( 2002 ) ; j .- kneur , m. b. pinto and r. o. ramos , phys . * 89 * , 210403 ( 2002 ) . i. s. gradshteyn and i. m. ryzhik , _ table of integrals , series , and products _ ( academic press , new york , 1980 ) .

halos , filaments , sheets and voids in the cosmic web can be defined in terms of the eigenvalues of the smoothed shear tensor and a threshold . using analytic methods , we construct mean maps centered on these types of structures for gaussian random fields corresponding to cosmological initial conditions . each map also requires a choice of shear at the origin ; we consider three possibilities . we find characteristic sizes , shapes and other properties of the central objects in these mean maps and explore how these properties change with varying the threshold and smoothing scale , i.e. varying the separation of the cosmic web into different kinds of components . the mean maps become increasingly complex as the threshold decreases to zero . we also describe scatter around these mean maps , subtleties which can arise in their construction , and some comparisons between halos in the maps and collapsed halos at final times . -1 cm

1505.07887

the web of cosmological large scale structure is both strikingly evident in observations and simulations and difficult to characterize . its existence has been known for several decades @xcite , and a huge number of different approaches have been used to study it since . recent overviews of many of these directions can be found in the proceedings of the @xcite . these cosmological structures evolved via gravitational instability from initial conditions which have been characterized by measurements of the cosmic microwave background ( e.g. , @xcite ) . these early time initial conditions are well described by gaussian random fields , which can also be classified into cosmic web structures . an assortment of different approximate methods can be used to estimate properties of the late time counterparts of the structures found in the initial conditions . properties include the fraction of mass which will collapse into a certain type of cosmological structure , the number of such structures , etc . classic works include @xcite ; some textbook descriptions are found in @xcite and @xcite . for instance , peaks in the initial conditions can be associated with collapsed halos at late times on average @xcite , and properties of the peaks can be connected to final time properties of the halos ( for instance by tracing back particles in simulations to their early time configurations such as in , e.g. , @xcite ) . halos are the most studied cosmological structures , in part due to their close identification to galaxies @xcite . this note extends studies of cosmic structures within the specific cosmic web classification considered in @xcite . ( their detailed particular definitions are in [ sec : background ] below . ) in that work , they compared analytic predictions for initial random gaussian fields ( equivalently , their values linearly extrapolated to final time ) and measurements of fully evolved dark matter simulations at late times ( redshift @xmath1 ) . for some values of their parameters , including the primary one which we focus upon in this work , the volume fractions of the structures for gaussian random fields and structures for fully evolved dark matter simulations counterparts at late times were comparable . the analytic mass fractions followed the same trends as found in the final time simulations , but were somewhat too high for sheets and too low for filaments . ( halos , which have the largest overdensities and thus might be expected to have the largest differences in the two methods , contribute a very small fraction to both the mass and volume fractions in this structure classification . ) this analytic classification of cosmological gaussian initial conditions has further consequences , as the power spectrum also determines mean spatial properties around a given central point . we analyze these spatial properties in this note . we find mean properties of regions around different choices of origin , classified as halos , filaments , or sheets , regions we call mean halos , mean filaments and mean sheets . ( the boundary around a `` mean void '' is more challenging to define as for many parameter choices the mean background is also a void , so that a boundary is not evident ) . the central shear values of these mean objects are an additional parameter which must be chosen . we explore three natural choices for these mean object central shear values , and measure sizes , shapes , shear profiles , and mean linearly extrapolated overdensities of the resulting mean halos , filaments and sheets . we also discuss scatter around these mean configurations , subtleties in their construction , and some possible relations between mean halos and final time collapsed halos . we consider structures in dark matter only . background is given in [ sec : background ] : the particular shear - based web classification we use , analytic estimates for average shear in different objects , the construction of mean maps and the measurement of some of their properties . in [ sec : examp ] we consider three choices of central shear , showing mean map examples for different objects and trends in their properties when changing the parameters defining the cosmic web . we also discuss scatter around the mean . in [ sec : collapse ] we turn to possible connections of the mean halos to collapsed halos at late times . in [ sec : conclude ] we summarize and conclude . the three appendices give details of the calculations ( shear , zeldovich displacement and scatter ) , subtleties which arise in some interpretations of one of our construction methods and properties of mean maps corresponding to slightly different constraints at the origin used by @xcite . the power spectrum used is for a @xmath2cdm universe with @xmath3 @xcite , calculated with the @xcite transfer function . the cosmic web can be separated into halos ( sometimes called nodes ) , filaments , sheets and voids . we will use the term halo and node interchangeably , as both terms are commonly used for these structures . there are many different web classifications based upon density , morphology , shear and beyond , some of which concentrate on just finding one particular type of structure ( e.g. halos or voids ) and others which classify an entire map . we use a classification of the latter sort , assigning a type of cosmic structure to every region based upon its shear . we begin with details of our shear definition and classification . we follow @xcite for definitions and for much of the analysis methodology below . the shear is defined on the initial gaussian perturbations as @xmath4 our potential @xmath5 has the same spatial dependence as the newtonian potential @xmath6 at early times but is time independent : @xmath7 . here @xmath8 is newton s constant , @xmath9 is the mean background density of the universe , and @xmath10 is the linear growth factor describing how linear structure grows as the universe expands with scale factor @xmath11 . the derivatives @xmath12 are with respect to the comoving coordinates @xmath13 . the potential @xmath14 is related to the initial density perturbations evolved linearly to the present time , @xmath15 , via the poisson equation : @xmath16 at times earlier than the present time the linearly evolved overdensity obeys @xmath17 the linearly evolved overdensity at the present time , @xmath18 at position @xmath13 is defined in terms of ( the linearly evolved to the present time ) density via @xmath19 . the factor @xmath20 is the square root of the integral of the power spectrum @xmath21 associated with @xmath15 , smoothed on scale @xmath22 @xmath23 i.e. , the root mean square fluctuations of the density contrast @xmath15 . we use a gaussian window function , @xmath24 in part to connect more closely with @xcite , as we use their parameters as a starting point below . one of the parameters of the classification is the smoothing scale @xmath22 . we will show results from @xmath25 , studied in depth by @xcite , but note trends we saw for other cases we explored , @xmath26 . again , to make the notation clear , because @xmath27 , @xmath15 equals the linearly evolved density perturbation today and @xmath20 corresponds to its commonly used definition as the square root of the integral of the power spectrum linearly evolved to the present day . some other approaches include @xmath10 in the overdensity , and thus in @xmath28 . as the fluctuations @xmath20 are associated with the overdensity , in our definition they have a factor of @xmath10 as well , and so @xmath29 remains time independent . this is one reason for introducing @xmath20 into our definition of shear . other definitions of shear also sometimes take out the trace , again not the case here . when the shear in numerical simulations is considered , eq . [ eq : sheardef ] is often used as well , but with the the fully evolved nonlinear potential ( determined by the evolved densities in the simulation ) , rather than its linearly evolved counterpart used in the analytic work in this note . the potential @xmath28 can also be used to find the zeldovich approximation to the displacement of initial positions at a later time . in the lagrangian approach to structure formation , the final position @xmath30 of an initial mass component with initial position @xmath31 is related via the displacement @xmath32 : @xmath33 in the zeldovich approximation , @xmath34 for the initial configuration , with coordinate @xmath31 , multiplying the linear final time density by @xmath10 and taking @xmath11 small gives the initial density ( which is linear because it is small , @xmath35 as @xmath36 and @xmath37 ) . the shear can thus be thought of either as a property of the linearly evolved final density perturbations , or as a property of positions at early times ( which then change , using eq . [ eq : zeldef ] ) . we will discuss properties in both interpretations . the @xmath38 shear matrix @xmath29 has three eigenvalues , traditionally ordered as @xmath39 . their sum , the trace of the shear , is proportional to the density by eq . [ eq : delphi ] , @xmath40 halos , filaments , sheets and voids can then be defined as regions where ( 3,2,1,0 ) respectively of the @xmath41 are above some threshold @xmath42 . in one way of estimating later structure formation , an initial region is expected to collapse at late times if it linearly evolves to cross some overdensity threshold @xmath43 . as the linearly evolved overdensity @xmath44 , and regions classified here as halos might be expected to collapse at late times ( this is an assumption ) , this relation suggests some minimum @xmath41 , i.e. nonzero @xmath0 , for initial conditions . note that this condition also means that the mean background , @xmath45 , is also classified as a void . there are thus two scales in this classification , the smoothing @xmath22 and the threshold @xmath0 . many different choices have been used in the literature . the same eigenvalue classification is sometimes used for dark matter simulations , but ( as mentioned above ) using the shear associated with the fully evolved nonlinear potential , see , e.g. , @xcite and references therein . for example , in dark matter nonlinear simulations evolved to redshift zero , @xcite found @xmath46 and @xmath47 separates structure into these different classes in a way that matches visual estimates , while @xcite considered @xmath48 and @xmath49 and found improvements for @xmath50 based upon volume fractions and mass fractions . one comparison between the above analytic and simulation approaches is by @xcite . in both approaches , they measured the mass fraction and volume fraction in the four cosmic structures as a function of @xmath0 , and for two values of @xmath22 . they found reasonable correspondence of volume and mass fractions for filaments , sheets and voids between the analytically calculated linearly evolved perturbations and the numerically evolved fully nonlinear simulations . given this , we looked for further characteristic properties of the cosmic web in the analytic construction and report these properties here . we focus for the most part on the parameters used in @xcite . by eq . [ eq : delphi ] , where each type of structure corresponds to @xmath15 crossing some particular threshold @xmath51 @xcite . the distributions of halos using @xmath0 vs @xmath51 was done in @xcite where it was found that the @xmath0 constraint worked well . in analytic work inspiring the work here , @xcite considered a mixed constraint , using both @xmath0 and @xmath51 to define halos and voids , and compared density anisotropies at large distances between the analytic ( linearly evolved to final time ) and simulated ( nonlinearly evolved to final time ) calculations . the mixed constraint makes the identification of filaments and sheets less direct . we give properties using the @xcite constraint in appendix [ sec : papsheex ] . structures classified using eigenvalues of the hessian of the density ( e.g. @xcite ) and the hessian of the velocities ( e.g. @xcite ) have also been considered in great detail by many authors . ] we will use these definitions of halos , filaments , sheets and voids to analytically calculate mean configurations of each , for initial random gaussian field configurations with different choices of @xmath22 and @xmath0 . unlike press - schechter theory where one changes the smoothing scale to consider different mass halos , here the one smoothing scale produces a separation of a full configuration into halos , filaments , sheets and voids , i.e. one smoothing scale completely classifies an entire map . our mean objects correspond to one such specific smoothing scale and threshold . the analytic definition of halos , filaments , sheets and voids in terms of @xmath0 allows the calculation of volume and mass fractions in each type of object and the average value of each of the eigenvalues @xmath52 in the gaussian random field initial conditions . the probability distribution for the ordered eigenvalues @xmath53 of a gaussian random field is @xcite @xmath54 where @xmath55 and @xmath56 . note that our @xmath41 differ by a factor of @xmath20 from that in @xcite because we include @xmath20 in our definition of shear , eq . [ eq : sheardef ] . expressions for pairs of eigenvalues or individual eigenvalues are found in @xcite and another interesting rewriting is in @xcite . the probability of the eigenvalues restricted by some condition @xmath57 , given for example in terms of @xmath0 , is given by @xmath58 one case of interest for us is where the shear in the region @xmath57 obeys the constraints for a halo , filament , sheet or void in terms of @xmath0 . the mean eigenvalues for this particular constraint @xmath57 , which we will denote by @xmath59 , are @xmath60 in this example , for each structure the corresponding normalization @xmath61 , eq . [ eq : pccalc ] , has the same integration limits as the the corresponding integral over @xmath62 ( it changes for different structures ) . for example , for a halo , @xmath63 we use the notation @xmath59 to indicate these constrained averages . it depends both upon @xmath0 and the type of structure ( halo , filament , sheet or void ) . in practice , for any given @xmath0 we integrated these expressions numerically using mathematica . another constraint goes beyond restricting eigenvalues in terms @xmath0 for each structure , requiring the @xmath28 from which they come to also correspond to extremal values of the potential ( @xmath64 ) . in this case the shear obeys conditions to be a halo , filament , sheet or void in terms of @xmath0 , and its potential also obeys @xmath65 . the measure for the volume average of points obeying this constraint is found in @xcite , starting with the probability distribution for the random gaussian fields @xmath66 times the constraint on @xmath67 . they were interested in peaks and minima in particular , as their main interest was for halos and voids ( they also had a restriction on @xmath28 , which we do not impose here , and thus it drops out of the expressions ) . their initial measure for integrating the probability density over volume , shear and @xmath67 becomes an integral over the eigenvalues similar to that in eq . [ eq : dorav ] , except with an additional measure factor @xmath68 ( from changing variables in the constraint @xmath69 from @xmath67 to position and integrating over position ) . @xmath70 here the integral again is over the values of @xmath41 obeying the constraint to be a halo , filament , sheet or void . we will call the corresponding average shear values @xmath71 . so , for example , for a halo , one has : @xmath72 these two kinds of average shear , @xmath59 and @xmath73 , are shown in fig . [ fig : lamcens ] below for the 4 kinds of objects , as the dashed and solid lines , and are quite similar . we will consider one other shear average below , that of the mean objects we construct . the eigenvalues of the average shear within these mean objects will be called @xmath74 . one can go beyond the analytic volume fractions , mass fractions and the constrained shear eigenvalue averages for each structure type by using the power spectrum . the power spectrum can be used to calculate mean spatial properties of the shear around any given central shear configuration . we make maps of these mean shear values around a given central shear value , and study properties of regions within them , in particular those regions sharing the same cosmic web classification as the central point . we use the method of @xcite , following @xcite , to find the mean values of @xmath75 ( and thus @xmath76 ) everywhere around a given configuration @xmath77 at the origin . their calculation , for our case of interest , is summarized in appendix [ sec : shearcalc ] . denoting the constraint at the origin as @xmath78 , we thus get a map @xmath79 , a mean shear at every point . this shear map depends not only on @xmath78 as explicitly indicated , but also upon the linear power spectrum and smoothing . the shear eigenvalues at the origin corresponding to the origin constraint @xmath78 are denoted as @xmath80 . , second along @xmath81 and smallest along @xmath82 . ] once the mean shear @xmath83 is known everywhere around a given central configuration , a choice of @xmath42 separates the mean map into halos , filaments , sheets or voids . if the origin constraint @xmath78 corresponds to a halo , the nearest points are usually also halo points , similarly for filaments , sheets and voids . we consider the central objects in the mean maps in the most detail . the central object in a mean map is the region sharing the same cosmic web classification ( e.g. filament ) as the origin and which are continuously connected to the origin via regions with this same classification . far away from the origin , the correlations and thus the mean shear go to zero and the mean configuration becomes the mean background ( again , this is a void region for @xmath50 ) . in between the origin and mean background , depending upon parameters , several different types of structures may occur . for instance mean halos tend to have filaments and then sheets around them before reaching the mean background . similarly , filaments tend to have sheets around them ( and sometimes halos within , towards their ends ) . voids do not always have a clear boundary because for @xmath84 they also can include the mean background . several quantities can be associated with these resulting mean map objects , such as volume , shape , and average values of shear @xmath74 . all of these depend upon @xmath78 , @xmath0 and @xmath22 . we report the volume of these mean objects in terms of the lagrangian volume for a halo associated with the same window function and smoothing scale @xmath22 , @xmath85 . this lagrangian volume is found by by integrating the background average density at early times ( as @xmath86 ) against the window function to give @xmath87 and then converting to @xmath88 . for our gaussian window function , @xmath89 . the correspondence between shear and initial gaussian perturbations means that one can view the volumes of objects in our maps as lagrangian volumes as well , as their volume also directly gives their mass at early times . for one choice of shear at the origin , @xmath90 , boundary conditions for the zeldovich displacement at the origin are also specified , as @xmath64 . in this case we also calculate the mean linearly extrapolated lagrangian displacement at late times given by eq . [ eq : zel ] . the mean displacement of the edges of the mean objects gives one way of estimating ( using linear evolution ) their final size . an additional measurement we make is of the average of the linearly extrapolated eulerian overdensity at final time in these mean objects , given by @xmath91 . in simulations , this has been measured by running the clock backwards , starting with collapsed halo regions in nonlinear simulations . particles in the final halo are tagged and then traced back in time , and the average of the shears of these particles in the initial conditions is used to calculate the linearly extrapolated eulerian overdensity at final times . @xcite found for lower final halo masses that on average @xmath92 , and @xcite found similar trends for higher mass halos . for more discussion of relations between collapsed halos and halos in this shear based classification . ] we quote overdensities in terms of this reference @xmath93 below . to summarize , for central halo , filament and sheet regions in mean maps we will measure volume , shape ( axes lengths ) , average shear eigenvalues within , @xmath74 , linearly extrapolated average overdensity@xmath94 , and in some cases mean zeldovich linearly extrapolated displacement @xmath95 . most of our focus is on the objects which include the origin ( objects not connected to the origin will be discussed briefly in appendix [ sec : subtle ] ) . we show examples below of cross sections of these regions along the 3 pairs of spatial axes and of the three shear eigenvalues and linearly extrapolated overdensity along the 3 spatial axes . the mean maps for a given cosmology are determined by the choice of central shear @xmath96 , the threshold defining the separation into different structures , @xmath0 , the smoothing @xmath22 , the window function and the power spectrum . we hold the window function ( gaussian ) and power spectrum fixed and show results for @xmath97 ( commenting on changes for other values ) . the threshold @xmath0 and central shears @xmath78 vary as indicated . to construct a mean map , a choice of shear at the origin , @xmath78 , is required . mean maps can be constructed for any choice of shear at the origin @xmath78 , but for a given type of structure and @xmath0 , we found three choices of particular interest , all motivated by the analytic calculations of average shear given some constraint . two are associated with @xmath59 , found via eq . [ eq : dorav ] , and one with @xmath71 , which includes eq . [ eq : extrfactor ] in the calculation of the average shear eigenvalues . they all yield a value for @xmath98 , the shear at the origin as follows ( implicitly for choice 1 , explicitly for choices 2,3 ) . 1 . : : @xmath99 : + origin shear choice 1 fixes @xmath98 implicitly , by requiring that the average @xmath41 within a particular mean object in the map , @xmath74 , corresponds to @xmath59 for the corresponding type of structure . that is , the mean object with shear @xmath98 at the origin has average shear corresponding to the average analytic shear , eq . [ eq : dorav ] for that kind of structure . for example , for a halo in the mean map , @xmath100 , where the left hand side is the average over the values in the mean map , and similarly for the corresponding filament ( using @xmath101 , etc . ) . in practice , we change the central value of shear until the mean shears within the halo region centered at the origin ( which changes in size and shape as the central shear changes ) , obeys eq . [ eq : haloca ] . one possible interpretation is that this gives an `` average '' such object ( e.g. halo ) , in a smoothed simulation . practically , we could not always find solutions @xmath98 satisfying the constraint @xmath102 for all @xmath0 and all structures , in addition , other complications in this implicit definition can arise ( appendix [ sec : subtle ] ) . our examples below are in the regime where the associated corrections from these possible complications seem small enough to be ignored and where we could find solutions . + 2 . : : @xmath103 : + origin shear choice 2 sets the central shear @xmath98 equal to the average shear @xmath59 in eq . [ eq : dorav ] ( similar to @xcite , who used it for long distance properties at final times and whose work inspired this study ) . this means putting the average shear for a certain structure ( for instance a halo ) at the origin and then measuring properties of the object made up of the enclosing region of points sharing the same cosmic web classification . in more detail , for a halo , the origin would have shear eigenvalues @xmath104 . although slightly more complicated to interpret , these maps are much easier to construct , as the central shear is explicitly and immediately calculable . mean properties around such a point might be interpreted as mean properties around an average point in the corresponding structures in a large map , rather than properties around the central point of a structure ( i.e. the peak of a halo does not have average halo properties ) . can be read off from fig . [ fig : profiles ] . ] + 3 . : : @xmath105 : + origin shear choice 3 is the volume average of shears obeying some particular cosmic web constraint where in addition @xmath28 is at an extrema , i.e. where @xmath65 for all @xmath106 . this is reasonable to expect for a center of a halo or void ( minimum or peak ) and we can also consider this condition for the center of a filament or sheet . these @xmath107 are calculated with the extra measure factor derived from eq . [ eq : extrfactor ] . intuitively , the mean objects corresponding to these central shear values , at least for halos and voids , might correspond to stacking objects in a map , centered on their extremal points ( which is often the point used to define the center of such objects ) . in fig . [ fig : lamcens ] we show the central values for halos , filaments , sheets and voids for these three constructions , as a function of @xmath0 . as central value 1 required implicit solving of an equation , only a few values are shown , as points ( and again , we could not find solutions for all @xmath0 in this case , and the construction is ill - defined for voids ) . the solid line is for case 2 , @xmath108 , the integral in eq . [ eq : dorav ] , the dashed line is for case 3 , the volume average over extremal points , @xmath109 . the central values are very similar for the three constructions . with these central values we can calculate the mean shear around the origin , and classify each region as halo , filament , sheet , or void . in fig . [ fig : aep ] , we show example cross sections of regions obeying constraint 1 for the central object , i.e. where @xmath74 within the central object is approximately equal to the corresponding @xmath59 , along the three pairs of axes . we use colors to distinguish between structures : red for halos , dark blue for filaments , cyan for sheets and yellow for voids . , and threshold @xmath110 . they chose @xmath0 by trying to get similar analytic volumes in each of the different types of structures . for the sheet shown we changed @xmath0 to 1.25 to avoid subtleties discussed in appendix [ sec : subtle ] . ] the mean background ( @xmath45 ) is also a void ( yellow ) for any @xmath50 . ( as voids comprise the eventual mean background into which these structures are found , and the majority of the volume , as noted earlier it was not clear how to impose @xmath111 for their case . thus no void example is shown . ) as anticipated , the mean central halo ( at top left , fig . [ fig : aep ] ) is surrounded by a filament , then a sheet , and then eventually reaches mean density ( which is a void ) , similarly the mean filament ( at top right , fig . [ fig : aep ] ) is surrounded by a sheet and then void and mean density , and so on . the mean filament does not extend between halos for this value of @xmath0 although as @xmath0 is lowered halos do eventually appear in the mean filament endpoints . the mean halo has comparable size in all three directions , the filament is extended in one direction compared to the other two , and the sheet has one direction much smaller than the other two , as expected . the larger directions are a few times larger than the smoothing scale @xmath22 . [ fig : aep ] also lists the mean extrapolated overdensity and volume for each central object . ( by construction , for choice of origin 1 , @xmath112 . ) for the @xmath0 and @xmath22 here ( 0.762 and @xmath113 respectively ) , the halo volume is slightly smaller than expected for the lagrangian region associated with a halo of mass @xmath87 , and the extrapolated density is higher , @xmath114 . explored in detail by @xcite , @xmath115 , the trend reverses , @xmath116 . ] the change in these quantities using a different choice of @xmath98 can be seen in fig . [ fig : bbks ] , where we show mean halo , filament and sheet maps for the same @xmath22 and @xmath42 values , but choice 3 for shear at the origin , @xmath117 . they are similar to their counterparts in fig . [ fig : aep ] , but slightly smaller , as is expected with their smaller values of @xmath98 ( see fig [ fig : lamcens ] ) . choice 3 of shear at the origin also fixes @xmath118 at the origin , implying values for the mean zeldovich displacement for points in these structures . in fig . [ fig : bbks ] , black lines along each axis connect the position of the initial edges , after they are shifted by their mean zeldovich displacements ( @xmath119 ) . ( no line is shown if the displacement crosses the origin . a displacement crossing the origin means that the estimates of the final edge values are well outside of the range of their validity . ) under these displacements , the mean halo contracts and the direction with the largest shear at early times becomes the most contracted at late times , as expected and seen as well in @xcite , see also @xcite . in addition , the mean filament and sheet both contract ( only slightly in some directions ) , while the void , not shown , expands in all directions . as mentioned earlier , the final time estimate for the linear displacement of the initial region edges provides one ( very rough ) estimate of the mean region s final size in the lagrangian picture . more generally , the shapes of the mean objects in these maps as the threshold @xmath0 changes can be seen in fig . [ fig : varlam ] , for @xmath97 and our 3 choices of shear at the origin . ( note that as these are the maximum of each coordinate that the actual filament length is , for instance twice the maximum @xmath82 value . ) solid points give the extent along the 3 axes for values of @xmath0 where we solved choice 1 ( @xmath111 ) , solid lines denote choice 2 , mean maps centered on @xmath120 , and the lighter dashed lines are for choice 3 , @xmath121 . we only sampled central values for condition 1 because it required solving an implicit equation for each @xmath0 and each different type of structure , and again , we could not always find a solution . for a fixed smoothing scale ( here @xmath22 ) , raising the threshold @xmath0 in general raises the average shears within objects , @xmath74 . as can be seen , this simultaneously decreases the volume of the associated mean region for any object beside a void . the shapes of halos , i.e. ratios of extent along different axes , change strongly as the threshold @xmath0 varies . , not shown here , we found that the sizes of the structures along the different axes ( in units of @xmath22 and relative to those seen in fig . [ fig : varlam ] ) increase for lower @xmath122 and decrease for larger @xmath123 . for example , at low @xmath0 , the size along a given coordinate axis ( @xmath124 ) for a given @xmath0 is about twice as large ( 2/3 as large ) , in units of @xmath22 , for @xmath22 = 0.5 ( 19.0 ) @xmath125 . the ratio of largest sheet edge to largest filament edge is close to constant as @xmath22 changes , for @xmath84 , but the filament / halo largest edge ratio changes ( drops ) with increasing @xmath22 until @xmath126 . ] for choice 3 of shear at the origin , the trends of the mean zeldovich displacements of the edges of the mean objects could also be studied . we found that the halo edges contract for all @xmath42 and that as @xmath0 increases the displacements go through the origin and become negative , indicating the approximation is well beyond its range of validity . for filaments and sheets , although 2 or 1 of the edges are / is displaced quickly towards the origin even for small @xmath0 , for the other edge or edges , as @xmath0 increases , the displacement either is small at first ( although causing the displaced edge to move towards the origin ) , or causes the object to first expand ( the two larger sheet axes ) before again having the displaced edge move towards the origin . the mean zeldovich displaced position of the edges divided by the original position of the edges takes roughly the same form for all choices of smoothing we studied . for many values of @xmath0 and @xmath22 , the mean objects around average points resemble those in figs . [ fig : aep],[fig : bbks ] above , albeit with different side lengths as just noted . however , as @xmath0 tends to zero , not only do central structures get much larger as seen in fig . [ fig : varlam ] , but also qualitatively different behavior appears . cross sections for the case @xmath127 and case 2 , @xmath128 , are shown in fig . [ fig : lam0 ] . the mean maps are shown for a much larger range of scales , because unlike the two earlier cases shown in figs . [ fig : aep],[fig : bbks ] , where outside the central regions the shear quickly tends to a void , for this low @xmath0 non - void structures persist out to large distances . ( we consider @xmath129 rather than exactly zero as the functions used to calculate the shears have limited accuracy . ) for this threshold , in part because voids and halos only comprise 8% of the total volume each , with sheets and filaments taking 42% each , the structures which are not the mean background ( not voids ) are extremely large . in addition , unlike fig . [ fig : aep ] , mean filaments have halos within their endpoints , mean sheets have filaments within their outskirts , mean voids have sheets outside of them , and so on , as can be seen . ( the presence of halos within the ends of filaments is characteristic of individual filaments in simulations , which are often found by looking for bridges between halos , e.g. @xcite . ) around mean halos , the filaments fan out and do not end within the scales ( @xmath130 ) we consider in 2 directions . increasing @xmath50 , these extra embedded structures decrease in size and eventually disappear , smoothly transforming into objects similar to those shown in fig . [ fig : aep ] . for @xmath97 this transition happens around @xmath131 , for @xmath132 this happens at an only slightly larger @xmath133 . ( as might be expected , this rich structure complicates using choice 1 to calculate @xmath98 , and is discussed in appendix [ sec : subtle ] . ) the scatter around this mean configuration is significant as well , as discussed in [ sec : scatter ] below . beyond shape , volume , extrapolated overdensity and mean zeldovich displacement of edges of each type of object , one can look at more details of the mean shear itself , for instance the shear values along any given axis . for the mean objects in figs . [ fig : aep],[fig : bbks ] , values along 3 axes for the 3 shear eigenvalues ( and the linearly extrapolated overdensity @xmath134 ) are shown in fig . [ fig : profiles ] . the dotted straight black line is the threshold @xmath0 , crossing this means changing structure classificiation ( e.g. from a halo to a filament ) . the profiles for voids with @xmath98 corresponding to central shear choices 2 and 3 ( @xmath110 ) are also shown , as these are explicitly calculable . central shear choices 1 , 2 and 3 correspond to line types dot - dashed , solid and dashed respectively . the different choices of shear at the origin give similar profiles for halos , but the filaments , sheets and voids show some large differences . the void profile , when calculable ( i.e. choices 2 and 3 at the origin ) , has mass compensation at a distance away from the origin in the linearly extrapolated over density as expected . note that the mean shears take a very long time to approach their eventual value of zero even though they fall below the threshold @xmath42 fairly quickly . one other scale in the system is the scatter , the variance of the shear around the mean values ( calculated in appendix [ sec : shearcalc ] , eq . [ eq : scatterexp ] ) . the variance depends only on @xmath22 , the window function , and the position relative to the constrained origin , independent of the central value of the mean configuration and the threshold @xmath0 . it is simplest to interpret along the 3 axes , where it starts at zero at the origin ( where the shear constraint is imposed ) and then increases to a constant @xmath1350.2 at a distance @xmath136 for one eigenvalue and at a larger distance , @xmath137 , for the other two . ( calling the matrix in eq . [ eq : scatterexp ] @xmath138 , where @xmath139 are each pairs of indices , the scatter shown is @xmath140 along @xmath106=11 , 22 or 33 respectively . ) the effect of this scatter on the classification of a given region depends upon the configuration at the origin and @xmath0 . we show in fig . [ fig : scatprof ] the shear eigenvalue profiles along each of the 3 axes both for the filament case seen in in fig . [ fig : profiles ] ) , and for the filament configuration with @xmath141 , but now also show the envelope of the variance . in both examples , far from the origin the mean value of shear approaches zero , while the scatter approaches its asymptotic value . at large distances , if @xmath0 is high , the configurations close to the mean do not scatter enough to change their classification from a void to anything else . as @xmath0 drops , a one - sigma fluctuation will mix many different kinds of structures even at large distances from the origin . thus , for instance , in fig . [ fig : lam0 ] of the cross sections for low @xmath0 , beyond about 20 @xmath125 , the mean configurations are within one - sigma of all of the structures ( e.g. , the mean halo is within one - sigma of the mean filament , sheet and void ) . thus the different ways of constructing the mean maps , via different constraints @xmath78 at the origin , provide , for the same @xmath0 , similar objects in size and shape . the mean central objects increase in size and change shape as the threshold @xmath0 is decreased , becoming very complicated as the threshold goes to zero . for very small @xmath0 , the scatter strongly mixes the different structures even far from the origin . the scatter about the mean configurations is independent of the central value and approaches its asymptotic limit at larger distances from the origin as @xmath22 increases . in general , any smoothing @xmath22 , and any threshold @xmath0 can be used to define halos , filaments , sheets and voids . one can also ask how objects in the smoothed map , such as halos , are related to other ways of defining such objects ( e.g. via gravitational collapse at later times ) . as halos are the most well studied collapsed objects , we investigated correspondences between our mean halos defined using @xmath102 and final collapsed halos such as are identified in numerical dark matter simulations . initial conditions for halos and their final properties have been extensively studied in @xcite . final collapsed halos can be characterized by their mass and the measured linearly extrapolated overdensity of their initial regions . we checked whether we could find , for each @xmath22 , a @xmath0 such that our mean halo map overdensities matched @xmath142 from @xcite and @xmath143 . , and there are different mass definitions for different quantities , with large conversion factors @xcite . for the former , the average mass of all halos with @xmath144 and their average @xmath145 from integrating over the mass function , e.g. , @xcite , gives @xmath146 until @xmath22 increases to @xmath147 . however @xmath148 for the smoothings @xmath149 which we considered . ] we considered all three origin choices in [ sec : choices].they gave similar results : for smaller @xmath22 , fixing @xmath150 to the value matching @xmath22 gave a @xmath0 driven close to zero , or even below , and the lagrangian volumes were much larger than that associated with @xmath87 , and for larger @xmath22 the opposite happened . for @xmath151 , for cases 1,2,3 respectively we found that @xmath152 and @xmath153 . ( this can be compared to the average mass for @xmath22 and above at this scale , @xmath154 . ) given the approximations involved and the difficulty in taking into account the higher mass halos , this might be a close enough agreement . in summary , only for one smoothing scale @xmath22 did it seem possible to find a @xmath0 where the mean halo had its volume close to the lagrangian volume @xmath155 , for our 3 choices of mean object centers . initial condition halos do not always give final halos , and not all halos at final times are tagged as halos in the initial conditions , for reasons which have been studied in detail in e.g. @xcite . for instance , external tidal fields might destroy a initial local peak , or a peak might form at a later time due to a merger . if we take this choice of threshold as a preferred one for smoothing @xmath97 , because it gives a better connection to collapsed halos ( of one fixed mass ) at late times , then we also can consider the characteristic sizes for other objects for this smoothing , from fig . [ fig : varlam ] . however , this lack of direct connection to collapsed halos ( of one fixed mass ) is not necessarily surprising . the mean objects we find for each kind of structure should be more characteristic of the smoothed map than , e.g. , for halos , the collapsed objects which can form after full nonlinear evolution . for modified constraints on both @xmath0 and @xmath156 and taking @xmath108 , @xcite compared the analytically predicted density anisotropy at large , i.e. close to linear , distances to that produced by a stack of numerically simulated halos , finding rough agreement . properties of the mean objects for the origin constraints considered by @xcite are given in the appendix [ sec : papsheex ] . a spatial distribution of gaussian random fields can be classified into halos , filaments , sheets and voids using a smoothing scale @xmath22 and threshold @xmath0 . these can be associated with linearly evolved initial conditions or with the initial conditions themselves for structure formation . many smoothing scales and thresholds have been chosen for use in different contexts . the comparison of fractions of volume and mass in the different structures has shown some agreement between analytic calculations of initial gaussian random fields ( specifically their linearly extrapolated values at the present time ) and numerical simulations of the full nonlinear evolution ( e.g. , @xcite ) . we explored further properties of cosmic structures in the gaussian random field configurations , for the same classification and parameters used by @xcite . we used the properties of gaussian random fields , combined with the power spectrum , to calculate properties of mean halos , filaments , sheets and voids with three choices of central shears @xmath98 . these central shears are related either to the mean shears for each type of object , or to the volume average of the extremal shear values for each object . for low @xmath0 the mean structures are very complicated , while at larger @xmath0 the mean structures centered at the origin smoothly transition to the mean background through the sequence halo to filament to sheet to void . the mean halo has approximately similar extent in all three directions , the mean filament has one larger direction and the mean sheet is small along one direction and larger along the other two . we showed examples of mean structures at large and small @xmath0 and discussed trends of their shapes with changes in threshold and in smoothing . one method of choosing the central shear was an implicit condition , which we could not always solve . when the central point is taken to be an extremal point ( @xmath157 ) , there is a natural value for the zeldovich displacement at the origin and thus one can also calculate the mean zeldovich displacement of the edge points of the mean objects , giving a final linear approximation to the size of each of these objects . for some @xmath0 , the estimated zeldovich change in the mean filaments and sheets is small , while the mean filament and sheets are large relative to the smoothing scale ( and sometimes the nonlinear scale ) , suggesting that these characteristic sizes might persist in the nonlinear maps in some form as well . we also characterized the scatter around the mean configurations , which tends to a constant at a small multiple of @xmath22 , the smoothing . the effect of scatter on the classification of structures around the mean background depends in part on @xmath0 , we showed as an example the scatter for mean filaments for two values of @xmath0 . there is not a clear identification of the mean halos with collapsed halos or collapsed halos above some mass . matching the expected average linearly extrapolated overdensity for a given smoothing @xmath22 implies a threshold @xmath0 , and leads to a lagrangian volume ( final estimated mass ) which is too small ( large ) relative to @xmath87 for large ( small ) @xmath22 , although there might be a sweet spot near @xmath158 . these estimates were only approximate as different mass definitions are used in the quantities being compared . as these mean objects are an additional characterization of initial gaussian random field configurations and smoothed maps , it would be very interesting to compare them to mean objects in simulations with the same smoothing and threshold , both at initial and at late ( fully evolved ) times . this comparison is nontrivial . although the division of halo , filament , sheet and void regions is direct in simulations , as are their volume and mass fractions , the separation into specific distinct objects , e.g. different filaments , is not immediate . splitting halo , filament , sheet and void regions into objects so that their stack gives our mean objects might be a natural way of defining such individual objects in the full map , but presumably additional assumptions will be required . the stacking might be most naturally associated with doing a volume average over extremal points as in our choice 3 of origin shear . this work is in a different direction than the extremely interesting research on finding the probability distribution of initial conditions which are consistent with a given final measured nonlinear configuration ( e.g. the line of study pursued in @xcite ) although both kinds of studies involve cosmic web classifications of initial conditions and their inferred counterparts at late times . several free parameters appeared in the construction , including the threshold @xmath0 , the smoothing @xmath22 and the central values @xmath98 . we chose a few different @xmath98 , however others are possible . connections to other natural scales in the initial conditions would be interesting to further pursue as well , such as those discussed in @xcite and @xcite . jdc thanks j. peacock , a. pontzen and j. whitehouse for suggestions , and especially m. white for numerous discussions . we thank n. dalal , v. desjacques , f. leclercq , m. neyrinck and the anonymous referee for helpful comments on the draft . alonso , d. , eardley , e. , peacock , j.a . , 2014 , arxiv:1406.4159 bardeen , j.m . , bond , j.r . , kaiser , n. , szalay , a.s . , 1986 , apj , 304 , 15 blumenthal , g. r. , faber , s. m. , primack , j. r. , rees , m. j. , 1984 , nature , 311 , 517 bond j. r. , kofman l. , pogosyan d. , 1996 , nature , 380 , 603 borzyszkowski , m. , ludlow , a.d . , porciani , c. , 2014 , mnras , 445 , 4124 colberg , j. m. , 2007 , mnras , 375 , 337 dalal , n. , white , m. , bond , j.r . , shirokov , a. , 2008 , apj , 687 , 12 desjacques , v. , 2008 , mnras , 388 , 638 desjacques , v. , smith , r.e . , 2008 , prd78 , 023527 despali , g. , tormen , g. , sheth , r.k . , 2013 , mnras , 431 , 1143 doroshkevich , a.g . , 1970 , astrophysics , 6 , 320 einasto , j. , klypin , a.a . , saar , e. , shandarin , s.f . , 1984 , mnras , 206 , 529 eisenstein , d. , hu , w. , 1999 , apj , 511 , 5 elia , a. , ludlow , a.d . , porciani , c. , 2012 , mnras , 421 , 3472 forero - romero , j.e . , hoffman , y. , gottlober , s. , klypin , a. , yepes , g. , 2009 , mnras , 396 , 1815 hahn , o. , porciani , c. , carollo , c. m. , dekel , a. , 2007 , mnras , 375 , 489 hoffman , y. , metuki , o. , yepes , g. , gottlober , s. , forero - romero , j.e . , libeskind , n.i . , knebe , a. , 2012 , mnras , 425 , 2049 jasche , j. , leclercq , f. , wandelt , b.d . , 2015 , jcap , 1 , 36 kaiser , n. , 1984 , apjl , 284 , 9 lam , t.y . , sheth , r.k . , desjacques , v. , 2009 , mnras , 399 , 1482 leclercq , f. , jasche , j. , wandelt , b. , 2015a , jcap , 6 , 15 leclercq , f. , jasche , j. , wandelt , b. , 2015b , jcap , 5 , 26 leclercq , f. , jasche , j. , wandelt , b. , 2015c , a&a letters , 576 , 17 lee , j. , shadarin , s.f . , 1998 , apj , 500 , 14 ludlow , a.d . , borzyszkowski , m . , porciani , c. , 2014 , mnras , 445 , 4110 ludlow , a.d . , porciani , c. , 2011 , mnras,413,1961 mo , h. , van den bosch , f. , white , s.d.m . , 2010 , `` galaxy formation and evolution '' , cambridge university press , cambridge papai , p. , sheth , r.k . , 2013 , mnras , 429 , 1133 peacock , j. a. , 1998 , `` cosmological physics '' , cambridge university press , cambridge pillepich , a. , porciani , c. , hahn , o. , 2010 , mnras , 402 , 191 planck collaboration , 2013 , a & a , 571 , 16 press , w.h . , schechter , p. , 1974 , apj , 187 , 425 pogosyan , d. , pichon , c. , gay , c. , prunet , s. , cardoso , j.f . , sousbie , t. , colombi , s. , 2009 , mnras , 396 , 635 robertson , b.e . , kravtsov , a.v . , tinker , j. , zentner , a.r . , 2009 , apj 696 , 636 robles , s. , dominguez - tenreiro , r. , onorbe , j. , martinez - serrano , f.j . , 2015 , arxiv : 1504.06297 shandarin , s.f . , zeldovich , ia . b. , 1983 , comments on astrophysics , 10 , 33 shandarin , s. , 2009 , jcap , 02 , 031 shen , j. , abel , t. , mo , h.j . , sheth , r.k . , 2006 , apj , 645 , 783 `` proceedings of iaus 308 : the zeldovich universe : genesis and growth of the cosmic web '' , to appear . tinker , j. , kravtsov , a.v . , klypin , a. , abazajian , k. , warren , m. , yepes , g. , gottlober , s. , holz , d.e . , 2008 , apj , 688 , 709 white m. , 2001 , a&a , 367 , 27 white , m. , 2014 , mnras , 439 , 3630 zeldovich , ia.b . , einasto , j. , shandarin , s.f . , 1982 , nature , 300 , 407 the mean shear @xmath159 can be calculated as described in @xcite ( based in part upon @xcite ) . their focus was on @xmath160 , for only one of the boundary conditions @xmath78 which we consider , so we give more details for completeness . a more general angle - averaged probability distribution @xmath161 was found by @xcite . defining @xmath162 , + @xmath163 , and @xmath164 , @xcite use the gaussianity of @xmath165 to get @xmath166 which for conditions @xmath78 at the origin then give @xmath167 the zeldovich displacement mean value , shown for central shear choice 3 in fig . [ fig : bbks ] can also be calculated analogously to that for @xmath168 when the displacement at the origin is fixed to be zero ( as it is for choice 3 of central shear ) : @xmath169 the variance of @xmath170 is @xcite : @xmath171 which is zero at the origin ( where there is a fixed constraint ) and then increases , asymptotically reaching @xmath172 . these give the mean value of the shear , zeldovich displacement when @xmath173 at the origin , and scatter at position @xmath13 when the shear at the origin is set by the boundary condition @xmath78 , which we explore in the body of this paper . the three central shear values @xmath98 which we consider are in [ sec : choices ] . to evaluate eq . [ eq : fullexp ] , eq . [ eq : psiexp ] and eq . [ eq : scatterexp ] , we follow @xcite in choosing the shear at the origin to be diagonal @xmath174 , and substitute @xmath175 : @xmath176 the expression for @xmath177 can be found by replacing @xmath75 by @xmath178 . for the scatter , eq . [ eq : scatterexp ] , @xmath179 this is independent of the value at the origin , and the threshold @xmath0 . the correlations @xmath180 are @xmath181 for @xmath182 , and @xmath183 the basis functions @xmath184 depend on the primordial matter power spectrum @xmath185 . we use a power spectrum for a @xmath2cdm universe with @xmath3 @xcite , calculated with the @xcite transfer function . for the @xmath184 associated with @xmath29 ( appendix of @xcite ) , \psi_5(r)&= \frac{1}{\sigma^2}\int_0^\infty d \ln k \delta_\delta^2(k)w_r(k)^2 [ \frac{1}{15 } j_0(kr ) \\ & + \frac{2}{21}j_2(kr ) + \frac{1}{35 } j_4(kr ) ] \ ; . \\ \end{array } \label{eq : threepsi}\ ] ] these are shown in fig . [ fig : threepsis ] for smoothing @xmath187 , the case considered in most detail in this note . there are features at scales a few times @xmath22 . for @xmath177 s correlations , the associated basis functions are ( e.g. , @xcite ) : @xmath188 these are used to calculate the mean displaced edges of the cosmological structures shown as solid black lines in fig . [ fig : bbks ] . method 1 of assigning @xmath98 ( in [ sec : choices ] ) constrains its value implicitly by requiring the average shear in a mean object of a certain kind , centered at the origin , to be equal to the mean value of shear for that kind of object in eq . [ eq : dorav ] . in addition to not always being able to solve this constraint by systematic trial and error , there are also conceptual subtleties with this approach , although there is an intuitive appeal to requiring the average of the shear within the central mean object in our construction to be the analytically calculated average of the shear within that type of object , @xmath59 . one complication was already noted for voids , as constructing a mean void volume is challenging when there is no boundary between a mean `` void '' configuration and the mean background , i.e. no mean void `` size '' . in practice , restricting to halos , filaments and sheets , we took objects centered on each kind of structure and calculated the average shear within these objects for a given @xmath0 . we then varied the central shear until the mean shear within each object was the one given by eq . [ eq : dorav ] for that object . however , the mean maps often have more than just the central type of object and the void structure in them ; all four structures can be present in a mean map centered on any one type of structure . so when one calculates the mean shear in the mean objects of a certain structure , one possible way is to include the contributions from all the different maps where that structure appears . that is all regions with that given structure , even if not part of the central object , should be included as well . if the average in the mean maps is chosen to include the contributions from these other regions , the calculation of mean shear becomes more complicated . we give an example by showing how adding these regions to calculate the mean shear can affect the calculation of mean shear for filaments . generally , a mean halo at the origin also has filament regions around it . so in addition to the mean filament region around the origin , when the origin is a filament , the mean of filament shears including other regions would include the shear contribution from filament regions around the mean central halo , calculated for the same threshold . the relative contributions of these filaments in the different maps , filaments around mean halos , around central mean filaments , ( and sheets and voids in some cases ) , can be estimated if one breaks down the volume fraction for each kind of object into contributions from each configuration : i.e. filaments when there is a mean halo at center , when there is a mean filament at center and when there is a mean sheet at center . this is @xmath189 here the volume fractions @xmath190 of each structure are available from eq . [ eq : dorav ] , and @xmath191 is the filament volume associated with each mean halo , etc . , found by constructing the mean halo , etc . and measuring the volume . these volumes depends upon the central shears which one is trying to find . this is especially challenging if contributions to the mean filamentary shear are large from any mean map besides that centered on the mean filament , and gets even more self - referential once filaments appear in the outskirts of sheets or halos appear in the outskirts of filaments , i.e. the last two terms @xmath192 . for some @xmath0 these additional terms @xmath193 seem small . then the calculation ignoring these other terms , that is , averaging the filament shear only over the mean object centered on the origin which is a filament , seems justified even in this more complicated interpretation ( and similarly for sheets and halos in their respective cases ) . for filaments these other terms can be estimated for size by the following method . for @xmath0 large enough , the mean halo structure only appears when it is the central object , so @xmath98 can be found for the mean halo by imposing @xmath194 for a mean halo centered at the origin . also , again for @xmath0 large enough , @xmath195 . then filaments only appear around the central mean halo and as the central mean filament . so we have : @xmath196 and the filamentary volume around the mean halo , @xmath197 is determined by the mean halo , already in hand . as we also know @xmath198 , we can constrain @xmath199 this can be used to see if the fraction of mean filament shear due to filaments around halos is large relative to that due to a filament centered at the origin . for @xmath97 halos and @xmath200 , @xmath201 and @xmath202 , giving a fraction of filament volume due to filaments around halos of about 1/35 , which we thus neglected in calculating the filament central value shown in fig . [ fig : aep ] . repeating the argument for sheets , for this threshold and smoothing , the relatively large sheet to filament volume fraction @xmath203 , combined with the large @xmath204 , makes @xmath204 too large to reasonably neglect in this interpretation . ( by trial and error ) to meet the constraint @xmath102 by considering the volume centered on the sheet alone . more generally , configurations where any @xmath205 were often quite hard to find via trial and error . this happens , for @xmath97 , at @xmath206 for filaments and @xmath207 for sheets . ] in fig . [ fig : aep ] we thus considered instead @xmath208 to impose @xmath209 in a sheet configuration . ( for the other scale studied in detail by @xcite , @xmath210 , filament contributions around halos and the filament to halo fraction ( 1/12 ) were both too large for their @xmath211 to permit neglect of @xmath191 . ) one could include these subleading contributions directly instead of going to regimes where they can be neglected , as well , in some cases , when halos are not found in filament edges , and filaments are not found in sheet edges . it would be interesting to further explore this interpretation , although the issues with voids might make it difficult to push it very far . this mean halo / filament / sheet interpretation and argument also gives a characteristic mean halo , filament and sheet volume and an expression for @xmath212 in terms of the number of halos , the volume fractions of filaments and halos and the mean volumes of filaments and halos . for the simplest case , for instance , @xmath213 , and analogously for sheets . [ cols="<,^,^,^,^,^,^",options="header " , ] this method for calculating mean shear was used by @xcite to find the mean shear due to a given configuration at the origin . they took the origin to correspond to 1 ) the sum of all halos with mass above some smoothing scale and 2 ) all voids associated with a smoothing scale , at large distances and final times . to find the central values of shear , they imposed conditions both on the eigenvalues of the shear individually and on their sum ( the density ) . we give here properties of the mean configuration maps for their boundary conditions at the origin . note that we consider the mean smoothed values given their boundary conditions ( that is we smooth the shear both at the origin and at the position it is being measured ) . they chose @xmath214 , where the latter was found by integrating eq . [ eq : dorav ] with either their halo or void constraints but with an additional constraint in the measure , a step function in the density . for halos , @xmath215 and @xmath216 , for voids , @xmath217 and @xmath218 . this is analogous to our choice of central shear 2 , except for the additional density constraint . in addition , voids and halos have different thresholds @xmath0 . for the comparison with simulations , as @xmath22 increases , @xmath219 decreases and @xmath0 increases , so halos with a larger smoothing ( higher mass ) will also obey the constraints for a lower @xmath22 ( this was used in this note in [ sec : collapse ] ) . thus for halos , @xcite interpret the resulting shear at the origin @xmath98 as due to the sum of halos with mass associated with @xmath22 and above . they take simulated halos with masses corresponding to @xmath22 and above in a numerical simulation , stack them , and compare measured and analytically predicted anisotropies in density at large distances , finding reasonable agreement . they consider masses @xmath220 , from which we find @xmath22 . in table 1 , we show halo and void configurations they used , with the corresponding @xmath221 , volume ( @xmath222 ) , shape , average @xmath223 ( eulerian ) overdensity , and @xmath98 , @xmath74 . we use a gaussian window function . interestingly , for these constraints , the central value of shear and the average value in the halo or void volume are very close . the density cut is a stringent constraint on the shear and the average density in the regions is close to the minimum ( maximum for voids ) imposed density . the associated volumes for the halos correspond to smaller masses than the lagrangian volumes of the halos associated with @xmath22 .

the energy relaxation channels of hot electrons far from thermal equilibrium in a degenerate two - dimensional electron system are investigated in transport experiments in a mesoscopic three - terminal device . we observe a transition from two dimensions at zero magnetic field to quasi one - dimensional scattering of the hot electrons in a strong magnetic field . in the two - dimensional case electron - electron scattering is the dominant relaxation mechanism , while the emission of optical phonons becomes more and more important as the magnetic field is increased . the observation of up to 11 optical phonons emitted per hot electron allows us to determine the onset energy of lo phonons in gaas at cryogenic temperatures with a high precision ,mev . numerical calculations of electron - electron scattering and the emission of optical phonons underline our interpretation in terms of a transition to one - dimensional dynamics .

1104.1645

non - equilibrium phenomena on the nanoscale increasingly gain interest as more and more devices are based on nanoscale electronics . the ongoing miniaturization trend results in state - of - the - art transistors , used for information processing , in which only a small number of electrons is moving through a conducting channel at a given time . semiconductor - based quantum information processing relies on the coherent dynamics in nanostructures . for both classical and , in particular , quantum circuits the detection of information relies on electronic signals strong enough to be measurable . an inevitable consequence are interactions in nonequilibrium giving rise to quantum noise and back - action . our experiments aim at understanding the underlying physics of interacting nanoscale circuits where only few electrons far from thermal equilibrium carry the information . we study in transport experiments at low temperatures the simplest case of individual nonequilibrium electrons . after their injection these `` hot '' electrons move at first ballistically with well - defined , high kinetic energy before they relax in an otherwise degenerate fermi liquid interested in the case of a high mobility 2des since here electrons near the fermi edge have a momentum mean - free path @xmath1 of several micrometers and therefore move ballistically through the mesoscopic device . the scattering length of the hot electrons , however , at first strongly decreases as a function of kinetic energy as the phase space for scattering processes grows . for larger kinetic energies the electron - electron scattering length increases again because of the high velocity and short interaction times . in a magnetic field perpendicular to the plane of the 2des , the lorentz force tends to guide hot electrons to move along the edges of the conducting mesa of the 2des . in this article we focus on the transition between two - dimensional scattering in a low magnetic field to quasi one - dimensional scattering in the quantum limit of edge channel transport of the hot electrons . one - dimensional '' refers to a situation where the lateral width of the effective transport channel still exceeds the magnetic length . relevant energy - loss mechanisms of hot electrons in a 2des are the emission of plasmons , acoustic or optical phonons , or scattering with `` cold '' equilibrium electrons in the degenerate fermi sea . the radiation of photons is strongly suppressed because of the momentum missmatch . the emission of plasmons by hot electrons is particularly hard to capture in pure transport experiments so we will address plasmons only briefly from the theoretical point of view in this article . the interaction between hot electrons and acoustic phonons constitutes a minor contribution to the relaxation of hot electrons in our nonequilibrium experiments . it therefore wo nt be a focus here , but detailed investigations in zero magnetic field have already been conducted on comparable samples by using a novel phonon spectroscopy technique.@xcite it is worth mentioning , however , that acoustic phonons play a major role in interactions between _ electrically separated _ nanostructures in nonequilibrium without@xcite and with@xcite a perpendicular magnetic field applied . the emission of optical phonons has been studied in zero magnetic field in various types of experiments since the 1960 s . it usually shows up as a very weak oscillatory signal as a function of the kinetic energy of hot electrons , with an oscillation frequency equal to the phonon energy . it has been observed in several materials including gaas in photoconductivity,@xcite in raman scattering experiments@xcite as well as in pure transport experiments in which electrons tunnel vertically between layers of a heterostructure.@xcite later the emission of optical phonons has also been observed in lateral devices defined in gaas - based heterostructures.@xcite the scattering of hot electrons in zero magnetic field with a cold 2des has been experimentally investigated in lateral three - terminal devices where the three regions have been separated by electrostatic barriers.@xcite it has been demonstrated that the electron - electron interaction in such mesoscopic three - terminal devices gives rise to effects such as a `` negative resistance''@xcite and can be used for avalanche amplification of a current of hot injected electrons.@xcite building on refs . and here we find that the transition from two - dimensional to one - dimensional scattering , as a perpendicular magnetic field is increased , goes along with a change in the importance of electron - electron scattering versus the emission of optical phonons . these two processes dominate the scattering dynamics of hot electrons in a degenerate two- or one - dimensional electron system in the whole range of available magnetic fields . at low fields hot electrons relax mainly via electron - electron scattering while the emission of optical phonons becomes more and more important at high fields . still , amplification of the injected electron current based on electron - electron scattering occurs even in a large perpendicular magnetic field . our measurements indicate that the scattering time between hot electrons and optical phonons strongly decreases as a function of the perpendicular magnetic field while the electron - electron scattering time increases . our experimental results are backed up by numerical calculations of the electron - electron scattering time and the emission rate of optical phonons as a function of the perpendicular magnetic field . finally , changing the magnetic field direction reveals a contribution of hopping transport via high energy localized bulk states of the 2des where a strong perpendicular magnetic field hinders the screening of disorder . our sample is based on a gaas / algaas heterostructure with a 2des below the surface . the charge carrier density is @xmath2 which corresponds to a fermi energy of @xmath3 ( determined at a temperature of @xmath4 ) . its electron mobility ( measured at the temperature @xmath5 ) is @xmath6 . the resulting equilibrium momentum mean - free path is @xmath7 m , an order of magnitude longer than the relevant distances in our sample . a hall - bar like mesa , visible as elevated area in the scanning electron micrograph in fig . [ sample](a ) , ( color online ) ( a ) atomic force micrograph of the sample . the elevated areas still contain a 2des below the surface , which is removed by wet - etching elsewhere . in addition , metal electrons are visible on top of the hall bar marked as emitter ( be ) and collector ( bc ) . ( b ) model that explains the avalanche amplification effect which is based on scattering of hot electrons with the degenerate fermi sea of the 2des at low temperatures . ] was created by electron - beam lithography and wet etching . the device has several ohmic contacts , three of which are used in the experiments as the emitter `` e '' , the collector `` c '' and an additional `` side '' contact . seven metal electrodes [ not all of them are visible in fig . [ sample](a ) ] have been fabricated by electron - beam lithography and evaporation , and are used to create electrostatic barriers by applying gate voltages . the two barriers used here are named emitter barrier be and collector barrier bc . the latter has been designed as a broad ( 300 nm gate width ) classical barrier whereas the former is actually a quantum point contact . note that the exact nature of the emitter is not important in the experiments shown here ( similar measurements with a broad barrier as emitter yield comparable results ) . the sample was measured in a @xmath8he cryostat at an electron temperature of @xmath9mk . unless stated otherwise a negative dc voltage is applied to the emitter contact while the dc currents @xmath10 and @xmath11 flowing into the grounded collector and side contacts are measured with low noise current amplifiers . the electric potentials on all other ohmic contacts of the nanostructure are left floating . great care was taken to tune the voltage offsets at the inputs of both current amplifiers to zero and hence avoid a superimposed current between side contact and collector . the emitter current @xmath12 is derived via kirchhoff s current law , @xmath13 , where we define the three currents to be positive if electrons flow into the sample from the emitter and leave the sample at the side contact and the collector [ compare arrows in fig . [ sample](a ) ] . this scenario would be expected in the limit of diffusive transport for which the sample can be described as a network of ohmic resistors . electron - electron scattering can cause major deviations from this ohmic case , as we have already discussed in detail in refs . and . the following paragraph gives a brief overview of their main results . we keep the emitter barrier be nearly pinched off and hence almost the entire voltage @xmath14 applied to the emitter drops across be . all electrons injected via be into the central region of the sample then have an excess kinetic energy @xmath15 close to @xmath16 because of the barrier s transmission probability that depends exponentially on energy . these _ hot _ electrons then excite via electron - electron scattering _ cold _ electrons which leave behind unoccupied states in the otherwise degenerate fermi sea . if we name these unoccupied states in the conduction band `` holes '' ( not to be confused with valence band holes ) the scattering process far from thermal equilibrium , schematically depicted in fig.[sample](b ) , can be interpreted as excitation of electron - hole pairs . if bc is configured such that excited electrons can pass but most holes are reflected , the positively charged holes will reside between be and bc until they will eventually be neutralized by electrons drawn in from the side contact . in its extreme , this can cause @xmath17 and consequently @xmath18 , where more electrons leave the device at the collector compared to the number of injected electrons . this is the case of avalanche amplification of the injected current . [ current_color_zerofield](a ) and [ current_color_zerofield](b ) ( color online ) measurements at constant @xmath19 ( @xmath20 ; see ref . for details on determining the barrier height in units of energy ) . ( a , b ) @xmath10 and @xmath11 as a function of @xmath21 and @xmath14 . ( c ) @xmath10 as in ( a ) , but plotted as a function of the injected current @xmath12 instead of @xmath21 . an extended region at the bottom ( yellow ) has @xmath22 ( see ref . for details ) . contour lines are spaced by ; white for @xmath23 , black for @xmath24 and dashed for @xmath25 . ( d ) collector current , same type of plot as in ( c ) . a very small area of @xmath26 is marked in the bottom left corner . ( e ) current transfer ratio @xmath27 , same axes . contour lines at @xmath28 ; the extended range of constant current at the bottom corresponds to @xmath29 . ] depict @xmath10 and @xmath11 measured as the gate voltage @xmath21 ( x - axis ) , defining the emitter barrier be , and the bias voltage @xmath14 ( y - axis ) have been varied . data are only taken in the regime of be nearly pinched off , namely within the roughly diagonal stripe of the graph with variable color . the upper left regions of the two plots ( gray ) are characterized by @xmath30 and have therefore not been mapped out in detail . to prevent excessive heating of the sample the overall dissipated power has been limited to @xmath31nw which results in no data for the lower right regions ( also gray ) of figs . [ current_color_zerofield](a ) and [ current_color_zerofield](b ) . the gate voltage @xmath21 controls the emitter current @xmath12 while the injected electrons have an energy close to @xmath32 . within the measured area of fig.[current_color_zerofield](a ) , a narrow stripe of @xmath33 is visible in which amplification of the injected current occurs @xcite . [ current_color_zerofield](c ) shows the amplification effect in a more instructive way . the raw data from fig.[current_color_zerofield](a ) is plotted as a function of the injected current @xmath12 and the bias voltage @xmath14 . the area of negative side current is enclosed by the dashed contour line marking @xmath25 . other contour lines of constant current are white at @xmath23 and black at @xmath24 . @xmath10 strongly depends on the energy of the injected electrons . this behavior is caused by the energy dependence of the electron - electron scattering length @xmath34 . near its absolute minimum at @xmath35mev , @xmath36 is actually smaller than the distance @xmath37 between be and bc.@xcite in this regime multiple scattering processes lead to the excitation of a large number of electron - hole pairs between be and bc . if only excited electrons with @xmath38 can escape via bc ( tuned to a barrier height near the fermi edge ) a positive charge can built up between be and bc which results in @xmath23 as the side contact is grounded . at a slightly larger energy of the injected electrons our data suggest @xmath39@xcite , meaning that scattering tends to happen just beyond the collector barrier . in this regime , electrons can be backscattered from behind bc , and then leave the sample at the grounded side contact . this manifests itself in a positive side current and a small in some cases even negative collector current . the latter can be seen in fig . [ current_color_zerofield](b ) or better [ current_color_zerofield](d ) , which shows @xmath11 as a function of @xmath12 and @xmath14 in the same fashion as for @xmath10 in fig . [ current_color_zerofield](c ) . in figs . [ current_color_zerofield](c ) and [ current_color_zerofield](d ) the positive @xmath10 and rather small @xmath11are clearly visible as ridge - like structures at @xmath40 ( see contour lines ) . for @xmath41mv we find @xmath22 [ fig . [ current_color_zerofield](a ) ] . this can be interpreted in terms of @xmath36 exceeding the sample dimensions so that the injected electrons pass the device without electron - electron scattering . most injected electrons then move ballistically into the collector contact.@xcite fig . [ current_color_zerofield](e ) shows the current transfer ratio defined as @xmath42 for the same set of data as in the other subfigures . the maximal value observed in the data presented here is @xmath43 , although we have already reached higher amplification factors of @xmath44 in a different sample . @xmath22 in the high - energy region in fig . [ current_color_zerofield](a ) corresponds to @xmath45 and @xmath46 in figs . [ current_color_zerofield](e ) and [ current_color_zerofield](d ) , respectively . note that the highest excess kinetic energies studied here up to @xmath47 are large compared to the fermi energy of @xmath48 and also exceed the energy of optical phonons ( @xmath49mev in gaas ) by far . in the data presented in fig . [ current_color_zerofield](c ) the emission of optical phonons is faintly visible as `` wiggles '' in the contour lines with extrema at @xmath50mv and @xmath51mv . optical phonons will be discussed in detail in sec . [ sec - phonon ] in context of a magnetic field applied perpendicularly to the plane of the 2des . it should be mentioned that the energy range used for the hot electrons in our experiments also exceeds the intersubband energy of the 2des which is in the order of @xmath52mev . nevertheless we do not observe any signatures of intersubband scattering which , therefore , seems to be inefficient compared to intrasubband scattering . most probably the majority of hot electrons still occupy the lowest subband which they used to populate in the emitter contact before being injected across be . moreover , the maximum energy of @xmath53mev is even larger than the vertical confinement energy of the 2des , which could cause scattering of hot electrons into three - dimensional bulk states . we have not observed any signatures of transport through the bulk of the heterostructure in our experiments , though . in a magnetic field @xmath54 perpendicular to the plane of the 2des the lorentz force @xmath55 acts on electrons perpendicularly to their momentary velocity @xmath56 and forces them to move along the edges of the conducting mesa . in a simple classical picture ballistic electrons are again and again reflected at the edge and move along well - defined skipping orbits with the cyclotron radius @xmath57 and the cyclotron frequency @xmath58 , where @xmath59 is the effective mass of the electrons . this classical limit has indeed often been observed in magnetic focusing experiments close to equilibrium and in moderate magnetic fields ( an example including nonequilibrium scattering is ref . ) . in a quantum - mechanical description the angular momentum quantization only allows cyclotron radii with @xmath60 , which for @xmath61 is equal to the so - called magnetic length @xmath62 , and where @xmath63 is the landau level index . for the case of @xmath64 and strong quantization ( small @xmath65 ) the landauer - bttiker description of the quantum hall effect has often been used . this model assumes a nondissipative one - dimensional motion of current - carrying electrons within edge channels.@xcite while in the past electrons close to thermal equilibrium have been studied in a perpendicular magnetic field , here we are interested in hot electrons far from thermal equilibrium with @xmath66 . in the following the landau level @xmath65 therefore pertains to the energy of the hot electrons rather than the number of filled landau levels in a degenerate 2des . in this nonequilibrium situation we aim at observing the transition from a two - dimensional electron system ( large @xmath67 and large @xmath65 ) to the case of one - dimensional motion ( small @xmath67 and small @xmath65 ) in a strong perpendicular magnetic field . since @xmath68 we expect to observe one - dimensional behavior to occur at sizable magnetic fields and not too large kinetic energies of the hot electrons . to achieve this we have extended the previous results@xcite summarized above by applying a perpendicular magnetic field . [ current_color](a ) ( color online ) ( a ) side current as a function of injected current @xmath12 and bias voltage @xmath14 in a perpendicular magnetic field of @xmath69 t ; contour lines are spaced by and colored as in fig.[current_color_zerofield](c ) . the inset shows the raw data as a function of @xmath14 and @xmath70 [ as for @xmath71 in fig.[current_color_zerofield](a ) ] . ( b ) @xmath10 and @xmath11 as a function of the excess energy @xmath72 of the injected electrons for a constant @xmath73na [ horizontal solid line , vertical line in fig . [ current_color](a ) ] . ( c ) energy dependence of the currents as in ( b ) but for @xmath71 ( vertical trace of fig . [ current_color_zerofield](c ) for @xmath73na ) . vertical lines in ( b ) and ( c ) are spaced by @xmath74mev . ] shows a measurement analogous to fig . [ current_color_zerofield](c ) , but in a magnetic field of @xmath69 t which corresponds to a filling factor of @xmath75 ( @xmath76 ) for equilibrium electrons at the fermi energy . the raw data is shown as inset analogously to fig . [ current_color_zerofield](a ) . we define the sign of @xmath54 to be positive for @xmath54 directed upwards , i.e. , such that electrons injected via be into the sample are guided to the left , away from the side contact [ fig . [ sample](a ) ] . the influence of the field direction will be discussed in detail in sec.[sec - edges ] . compared to the case of no magnetic field applied , at @xmath69 t the emitter current is slightly reduced ( by about @xmath77 ) . this is related to the conductance quantization and corresponding increase of the resistance of a 2des in a strong perpendicular magnetic field . however , the absolute value of the minimum of @xmath78 and the maximum of @xmath79 ( the current transfer ratio ) are much more reduced , roughly by factors of four and two , respectively [ compare figs . [ current_color_zerofield](c ) and [ current_color](a ) ] . a notable difference compared to zero magnetic field is that for @xmath69 t @xmath10 is strongly modulated with a period close to @xmath80mev . the oscillations of @xmath10 are also mapped onto @xmath11 as can best be seen in fig.[current_color](b ) which plots both currents as a function of @xmath72 for a fixed @xmath73na [ marked by a vertical line in fig.[current_color](a ) and a horizontal line in fig . [ current_color](b ) ] . these oscillations are caused by the emission of optical phonons with an energy @xmath81mev and will be discussed in detail in sec . [ sec - phonon ] . the analogous measurement for @xmath82 shown in fig . [ current_color](c ) allows a quantitative comparison near maximum amplification . at zero magnetic field amplification clearly is much stronger ( deeper minimum of @xmath10 ) while phonon - induced oscillations are only seen at relatively small energies of @xmath83mev . these observations imply that in our sample the electron - electron scattering rate between be and bc decreases in a perpendicular magnetic field while at the same time the emission of optical phonons becomes more significant . furthermore , the shape and current profile of the region of @xmath23 [ framed by dashed contour lines in figs.[current_color_zerofield](c ) and [ current_color](a ) ] corresponding to @xmath84 changes drastically if a perpendicular magnetic field is applied . for relatively small injected currents @xmath12 the region of @xmath23 extends towards larger @xmath72 in a perpendicular magnetic field . at the same time the absolute minimum of @xmath10(@xmath14 ) , which is relatively broad at @xmath71 , shifts towards smaller energies @xmath85mev and narrows in a magnetic field . this first ( and deepest ) minimum of @xmath10due to optical phonons at @xmath69 t still extends to relatively large currents up to @xmath86a . with increasing @xmath12 the main minimum shifts towards larger @xmath72 while the phonon - induced oscillations are constant in @xmath72 . we conclude that in a large perpendicular magnetic field the dependence of the amplification effect on the energy @xmath87 is strongly altered by the emission of optical phonons while its dependence on the dissipated power , @xmath88 , is still dominated by electron - electron scattering . could the observed change in the amplification effect then be related to a transition of scattering of hot electrons in two dimensions to one - dimensional scattering at finite @xmath54 ? a first answer to this question can be given by a comparison of the screening length @xmath89 nm of the 2des at @xmath71 with the width of the current - carrying channel at @xmath69 t , approximately given by the cyclotron radius in the quantum hall regime . at @xmath69 t we find @xmath90 nm , which is in the same order of magnitude as @xmath91 . the main minimum of @xmath10 in fig . [ current_color](a ) at relatively large currents where the phonon - induced oscillations are weak occurs at energies in the order of @xmath92 . this corresponds to @xmath93 nm ( and @xmath94 ) which is considerably larger than the two - dimensional screening length , @xmath89 nm . hence we do not expect our data to resolve the complete transition from two - dimensional to fully one - dimensional scattering of hot electrons . the comparison of the length scales suggest that our data can be interpreted as a signature of quasi one - dimensional scattering . we start a more detailed discussion by highlighting some numbers relevant for the scattering dynamics of the hot electrons . in the next two sections we will then present theoretical considerations pertaining to scattering of hot electrons with a degenerate 2des in a perpendicular magnetic field . the area occupied by 2des between be and bc measures slightly more than a square micrometer and therefore contais about 3000 electrons . the potential in this area is increased in a regime of avalanche amplification by up to @xmath95mv ( see ref . for details on the setup used to obtain this number ) which corresponds to a reduction of the resident electrons by roughly 10% . in the following we restrict the discussion to effects at considerably larger energy scales . for simplicity we now assume that the injected electrons move ballistically from be to bc across the distance of @xmath96 nm . then the average number of hot electrons traveling simultaneously between be and bc is @xmath97 . we find a maximum of @xmath98 for our measurements at @xmath99a and @xmath100mev . however , amplification ( @xmath23 ) occurs at @xmath101 and most of the following discussions apply to this case for which we can neglect direct coulomb interaction between hot electrons . scattering of hot electrons with a three - dimensional degenerate electron system is discussed in standard textbooks on landau fermi - liquid theory.@xcite in three dimensions an energy - dependent quasiparticle relaxation is usually considered that can be divided into two regimes . on the one hand , the relaxation of excited electrons at rather low energies with momenta @xmath102 is dominated by particle - hole excitations . on the other hand , at rather high energies ( in the order of several electron volts in typical metals ) beyond the plasma frequency the emission of plasmons becomes important . in two dimensions this clear separation into two regimes breaks down at relatively high electron densities for which the fermi energy @xmath103 by far exceeds the mean mutual coulomb energy @xmath104 between conduction band electrons ( and the thomas - fermi wavelength exceeds the inter - particle distance ) : @xmath105 fulfills @xmath106.@xcite our sample resides with @xmath107 in an intermediate regime and we expect that the relaxation of hot electrons with sizable excess energies via the emission of plasmons should play a role at zero magnetic field . nevertheless , our transport measurements have proven to be relatively insensitive to the emission of plasmons and we have not been able to identify traces of plasmon excitations in our present data . if , in addition , a perpendicular magnetic field @xmath54 is applied , the two - dimensional bulk electrons become rigid against perturbations at energies low compared to the landau - level separation @xmath108 . for hydrodynamic plasmons , i.e. , in the limit of very small momenta @xmath109 , the only difference between the two cases of @xmath110 and @xmath71 is this finite gap @xmath111 . @xcite however , at sufficiently high magnetic field , the transport properties of a disordered 2des are determined by the one - dimensional edge channels , hence , the bulk magneto - plasmons would not be visible in a transport measurement . in a one - dimensional system , interactions are strong and the electron system can be described as a luttinger liquid rather than a landau fermi liquid . in the low - energy regime the relaxation is dominated by dephasing @xcite . for technical reasons , investigations of the electron - plasmon interaction in the luttinger liquid are limited to intermediate energies corresponding to momenta @xmath112.@xcite one would expect that the scattering of a single hot electron with the degenerate electron systems displays a transition from landau - fermi to luttinger liquid behaviour as the external magnetic field is increased . this feature is absent in the present experiment , which can be understood by considering the width of the one - dimensional channel as described in the following . here we calculate the electron - electron scattering length of hot electrons moving in a quasi one - dimensional channel of an otherwise degenerate 2des . the width of the quasi one - dimensional edge channel , produced by a perpendicular magnetic field , is approximately given by the cyclotron radius @xmath113 . we specifically consider the realistic scenario of weak two - dimensional scattering in which the width of the channel exceeds the screening length of the 2des by far , @xmath114 . edge channels form in a perpendicular magnetic field because the physical edges of the hall bar represent boundaries which bend the landau levels upwards in energy . formally , the degeneracy of the electrons in each landau level is lifted near the edges according to @xmath115 , where @xmath116 denotes the momentum component of a hot electron parallel to the edge of the 2des . a good approximation for etched edges is a hard - wall confinement of the 2des , which allows an exact calculation of the dispersion@xcite @xmath117 the resulting single - electron wavefunctions resemble those of the free 2des , @xmath118 , with @xmath119 being the @xmath65-th hermite function . the electrons within the quasi one - dimensional channel interact via the two - dimensional coulomb potential @xmath120 where @xmath121 creates an electron with momentum @xmath122 in the @xmath65-th landau level , @xmath123 is the ordinary two - dimensional coulomb matrix element with the exchange momentum @xmath124 , and @xmath125 describes the interaction between electrons in the landau levels @xmath65 and @xmath126 . further details have been discussed in ref . . in the limit of hydrodynamic transitions , i.e. @xmath127 , eq . ( 2 ) reduces to the ordinary plane - wave coulomb interaction . et al . _ calculated the electron - electron scattering rate at @xmath71 @xcite . it is straightforward to write down the analog in presence of a perpendicular magnetic field @xmath128^{''}&\\ & \times(1-f_n(p_{y}+q_{y}))\,,&\end{aligned}\ ] ] by summing over those exchange momenta @xmath124 which are related to electron - electron interaction within the quasi one - dimensional channel ( @xmath129 ) . here the dielectric susceptibility @xmath130 is given by @xmath131 and momentum sums are cut off at @xmath132 , where @xmath133 is the width of the hall bar ( see figure 1 ) ; @xmath134 is the fermi distribution in the @xmath65-th landau level . for our numerical evaluation we use the quadratic dispersion in eq . ( [ eq : dispersion ] ) which is the exact solution for a hard - wall confinement . since a clear physical separation of the edge channels at high energies seems unlikely , we assume that each in - situ injected electron with energy @xmath135 occupies any one of the landau - levels fulfilling @xmath136 with equal probability and average the inverse quasi - particle lifetime over these channels . in fig . [ lambda_ee ] ( color online ) calculated scattering length for scattering between one hot electron with kinetic energy @xmath137 and cold electrons of the degenerate 2des as a function of excess kinetic energy for different filling fractions with fully occupied landau - levels @xmath138 and @xmath139 ( corresponding to @xmath71 ) . to compare the values of @xmath65 with experimental data , corresponding values of @xmath54 are given as well , calculated for the charge carrier density of our sample . ] we show the electron - electron scattering length @xmath140 calculated with ( 4 ) for the @xmath138 landau levels fully occupied by the degenerate 2des ( integer filling factors @xmath141 ) as well as the @xmath71 result ( corresponding to @xmath139 ) obtained in a previous calculation.@xcite the overall shape of all curves is similar , exhibiting a rapid decrease of @xmath36 for @xmath142 followed by a gradual increase of @xmath143 in the limit @xmath144 . this general behavior has already been discussed in detail for the case of @xmath71 in ref . . for @xmath145 the scattering length never drops below the distance between be and bc ( @xmath146 nm ) . this explains the overall weaker amplification ( smaller @xmath147 ) which we observe at @xmath69 t compared to the case of @xmath71 [ compare figs.[current_color_zerofield](c ) and [ current_color](a ) ] . it can be qualitatively understood since the width of the quasi one - dimensional channel shrinks with increasing magnetic field and the phase space for electron - electron scattering is hence reduced . the @xmath148 curve apparently converges to the zero field curve at large @xmath137 while at low @xmath137 large differences remain . this behavior expresses the fact that at high energies forward scattering dominates which resembles scattering in a one - dimensional channel . our calculations suggest that the differences in the electron - electron scattering dynamics between the one - dimensional and the two - dimensional cases disappear at high @xmath137 and @xmath149 . in fig . [ lambda_ee ] at low energies the scattering length at small filling fractions ( @xmath138 ) differs from its 2des counterpart for @xmath71 ( @xmath139 ) . the origin of this discrepancy is the altered dispersion , yielding a different asymptotic behavior of the appropriate susceptibilities @xmath150 within the particle - hole continuum . hence this technical discrepancy at small energies should not be mistaken for a realistic prediction . in summary , the calculated scattering length @xmath36 plotted in fig . [ lambda_ee ] explains several aspects of the measured magnetic field dependence expressed in our data in figs . [ current_color_zerofield](c ) and [ current_color](a ) . the increase of @xmath36 with increasing magnetic field causes the measured overall weaker amplification at finite @xmath54 . the similarity of the calculated @xmath151 curves for different values of @xmath54 explains the overall similar behavior ( except for phonon - induced effects ) observed at different perpendicular magnetic fields . the calculations in particular predict that amplification should occur even at high perpendicular magnetic fields ( up to @xmath152 ) while the strongest amplification is expected for @xmath153 @xmath154 where we find a broad minimum of @xmath155 . the calculations suggest a transition towards one - dimensional scattering to happen for @xmath156 . the increase of @xmath36 with increasing magnetic field in this regime especially at large energies suggests that the region of amplification should be restricted to lower energies . this corresponds to the narrowing of the main minimum of @xmath10 towards lower @xmath72 in a finite perpendicular magnetic field . the ballistic regime visible in figs . [ current_color_zerofield](c ) and [ current_color](a ) as extended areas of @xmath157 in the limit of large @xmath87 is expected for @xmath158 . it can therefore be qualitatively explained by the calculated monotonic increase of @xmath36 at large @xmath159 ( fig . [ lambda_ee ] ) . we have succeeded in qualitatively explaining some of the differences between the data at zero versus finite perpendicular magnetic field by the comparison with our calculations of @xmath36 . however , other features can only be understood by taking the relaxation of hot electrons via emission of optical phonons into account . in our experiment the emission of optical phonons can be clearly seen as periodic oscillations of the measured currents ( see fig . 3 ) , which is the subject of the next section . electrons carrying enough excess energy @xmath160 can lose part of their energy by emitting an optical phonon . the electronic momenta in our experiments cover only about 10% of the first brillouin zone . in this range of small wave vectors optical phonons have an almost constant dispersion relation , with an energy of @xmath161mev in gaas.@xcite we therefore expect the hot electrons to emit phonons with a fixed energy @xmath81mev and a wide range of momenta . in fact , the momentum transfer is only limited by the quadratic dispersion relation of the electrons and the scattering cross - section for the emission of optical phonons is accordingly large . since their dispersion relation is flat , the emitted optical phonons have almost zero velocity . their main decay channels are reabsorption by electrons or , more efficient , the emission of two acoustic phonons . the momenta of the resulting acoustic phonons are not confined to the plane of the 2des ; consequently , they tend to disappear into the three - dimensional crystal and have a negligible chance to be reabsorbed by the 2des . the oscillations in @xmath10 and @xmath11 as a function of @xmath14 in fig . [ current_color](b ) are clearly caused by the emission of optical phonons , as their period [ see vertical lines in fig . [ current_color](b ) ] is close to the energy of the lo phonons in gaas @xmath81mev . the oscillations occur because the number of optical phonons that can be emitted per hot electron increases one by one at integer multiples of @xmath162 if @xmath72 is increased . at large magnetic fields we have observed up to eleven of these oscillations which implies the sequential emission of eleven optical phonons by the individual hot electrons . by averaging over several periods of the oscillations we have determined the _ energy of the lo phonons in gaas _ at a cryogenic temperature @xmath9mk with a high precision to be @xmath163 . this result implies that in our experiments the interaction between electrons and phonons is dominated by the bulk lo phonons . if other decay channels , e.g. , interface optical phonons which have different energies,@xcite were also important the observed oscillations of @xmath10 would appear more irregular . the coupling between the two - dimensional electron system in a gaas / algaas heterostructure and interface phonon modes is expected to be reduced compared to bulk modes because of the small probability function of the electrons at the interface . @xmath11 has local maxima and @xmath10 local minima at integer multiples of @xmath164 [ vertical lines in fig.[current_color](b ) ] which corresponds to a rigid relative phase of @xmath165 between the oscillations of the two currents . however , in fig . [ current_color](b ) this behavior is caused by the way the data are depicted , namely for a fixed @xmath12 . in fact , @xmath166 determines the phase between @xmath10 and @xmath12 . to overcome this limitation we therefore replot the raw data [ see inset of fig.[current_color](a ) ] as a function of @xmath167 in fig . [ phase](a ) . ( color online ) ( a ) same data as in the inset of fig . [ current_color](a ) but as a function of reduced gate voltage @xmath168 instead of @xmath12 ( see main text for details ) . ( b ) reproduction of fig . [ current_color](a ) to allow a direct comparison . ( c ) vertical traces of ( a ) at @xmath169 as a function of @xmath72 for all three currents . integer multiples of @xmath170mev are marked by vertical lines ; ( d ) traces analogous to ( c ) , but for @xmath82 ( fig . [ current_color_zerofield ] ) and @xmath171 . ] for comparison fig . [ phase](b ) reproduces the plot for constant @xmath12 already shown in fig.[current_color](a ) . @xmath172 is defined as the gate voltage at which the emitter opens for a given @xmath14 , and current starts to flow ( @xmath173 ) . compared to the choice of a constant @xmath12 this processing technique does not require a rigid phase relation between @xmath10 and @xmath11 . the disadvantage is a small uncertainty regarding an offset of @xmath14 which , however , does not effect the relation between @xmath10 and @xmath11 . fig . [ phase](c ) shows a vertical trace of fig . [ phase](a ) at @xmath174 and fig . [ phase](d ) an analoguous graph for @xmath71 . interestingly , the rigid phase difference of @xmath165 between the phonon - induced oscillations in @xmath10 and @xmath11 remains ( independent of @xmath54 ) . this expresses a direct linkage between @xmath10 and @xmath11 caused by electron - electron scattering which in itself is correlated to the emission of optical phonons . regardless of the value of @xmath54 the emitter current , which is no longer fixed , steeply grows at first as the emitter voltage @xmath14 is increased . at @xmath71 @xmath12 then develops a broad maximum at @xmath175mev [ fig . [ phase](d ) ] . the gradual decrease of @xmath12 at @xmath176 is caused by an increase of the overall resistance . this is linked to an increase of the electron - phonon scattering rate with growing excess energy . at @xmath69 t @xmath12 is further reduced and , interestingly , stays roughly constant in the regime of strong phonon - induced oscillations of @xmath11 and @xmath10 . a closer look reveals that @xmath12 is slightly modulated and decreases stepwise at integer multiples of @xmath162 [ vertical lines in fig.[phase](c ) ] . this weak modulation can be explained as an incremental increase of the overall device resistance whenever additional optical phonons contribute to electron scattering . it is a relative small effect since the device resistance is dominated by the almost pinched - off emitter . the oscillations of @xmath11 and @xmath10 ( at finite @xmath54 ) are much stronger and can not be explained in terms of the weak modulation of the overall resistance . instead they express the correlated dynamics of electron - phonon and electron - electron scattering . at relatively small excess energies @xmath177mev the collector current @xmath11 steadily increases and exceeds the emitter current @xmath18 while the side current decreases accordingly , becoming negative . this avalanche amplification effect has been discussed in detail in sec . [ sec - review ] and [ sec - transition ] . in short , the increase in amplification is caused by the decrease of @xmath36 as @xmath72 is increased . at @xmath69 t the amplification is reduced compared to @xmath71 while the emission of optical phonons has become an important scattering process [ oscillations in fig . [ phase](c ) ] . in the extreme case , the hot electrons will first emit as many optical phonons as possible and therefore end up with an energy less than @xmath164 before most of the electron - electron scattering occurs . as expected for such a scenario the negative side current @xmath10(@xmath72 ) reaches its absolute minimum at @xmath178mev , just before the first optical phonon can be emitted . for higher energies @xmath72 the side current then oscillates with local minima at integer multiples of @xmath164 where yet another optical phonon can be emitted . our calculations presented in the previous section suggest that @xmath36 increases with increasing perpendicular magnetic field as electron - electron scattering becomes weaker due to the transition from two - dimensional to quasi one - dimensional scattering . we will see that the emission rate of optical phonons increases under the transition from two to one dimensions.@xcite our observation that the scattering with optical phonons takes over from electron - electron scattering as the perpendicular magnetic field is increased can therefore be interpreted as indication for a transition from two - dimensional to one - dimensional scattering . in the following we estimate the electron - phonon scattering time @xmath179 by counting phonon - induced oscillations of @xmath10 as a function of @xmath72 . until now we have assumed that if the electron - phonon scattering length is much smaller than the electron - electron scattering length ( @xmath180 ) , the maximum possible number of optical phonons @xmath181 will be emitted first . afterwards hot electrons with the reduced excess energy @xmath182 will scatter with the 2des . while this assumption seems reasonable for a rough picture , it disregards the influence of the sample s geometry on the correlation between the emission of lo phonons and electron - electron scattering , namely that only scattering processes happening in the region between emitter be and collector barrier bc contribute to the observed amplification effect . the apparent decrease of the amplitude of the phonon - induced oscillations at large energies @xmath72 [ fig . [ phase](c ) ] is related to this geometrical restriction . consider the length of the path @xmath183 that an electron subjected to a perpendicular magnetic field travels along the mesa edge from be to bc . assuming that backscattering of the hot electrons is strongly suppressed in a large perpendicular magnetic field ( the extreme case are landauer - bttiker edge channels which only allow forward scattering ) , we can make an accurate statement . phonon - induced oscillations with undiminished amplitude will be observed if ( 1 ) @xmath184 ( as is the case in our sample ) and ( 2 ) @xmath185 . here @xmath186 is the number of phonons already emitted when the electron reaches bc . in order to roughly estimate the electron - phonon scattering time @xmath179 from our measurements in different magnetic fields we now completely neglect all other energy relaxation processes including electron - electron scattering . we can then express the velocity of a hot electron with initial kinetic energy @xmath187 after emission of @xmath188 phonons as @xmath189 the mean distance a hot electron travels until it has emitted @xmath190 optical phonons is @xmath191 in the last step we assumed for simplicity that @xmath192 is independent of the energy of the electron ( as long as @xmath193 ) , which is justified by previous calculations suggesting only a weak energy dependence.@xcite the expectation that on average @xmath190 phonons have already been emitted after the hot electron has traveled the distance @xmath194 is expressed in the inequality @xmath195 which leads to the approximation @xmath196 \nonumber\\ & = & { \ensuremath{\tau_{\text{e - ph}}}}\left ( \sum_{k=0}^{{\ensuremath{n_{\text{ph}}}}-1 } v_k + \frac{1}{2 } v_{{\ensuremath{n_{\text{ph } } } } } \right)\,.\end{aligned}\ ] ] inserting eq . [ eq : vk ] and solving for @xmath179 results in @xmath197 we estimate the distance the electrons travel to reach the collector barrier ( by going along the edge ) to be about @xmath198 m [ compare fig . [ sample](a ) ] for the magnetic field values considered here ; for @xmath82 no transport along the edge is expected and @xmath199 is used instead . [ differentfields ] ( color online ) @xmath10 as a function of @xmath72 at constant @xmath200 as in fig.[current_color](b ) and [ current_color](c ) for various perpendicular magnetic fields . the vertical lines in ( b ) and ( c ) mark the value of @xmath72 above which the amplitude of the phonon - induced oscillations starts to decrease considerably . for ( a ) and ( c ) these values of @xmath72 are outside of the plots . ( e ) scattering times displayed in table [ table ] as a function of magnetic field ( determined from ( a)(d ) and additional measurements ) . the leftmost ( rightmost ) data point represents a lower ( upper ) limit as indicated by arrows ( see table [ table ] ) . the solid lines show theory curves obtained by heuristic application of bockelmann s theory ( numerical calculations , no fit parameters involved ) . the three branches belong to transitions between electronic subbands ( @xmath201 , @xmath202 , @xmath203 ) in @xmath204 direction via the emission of an optical phonon . ] shows the measured @xmath10 as a function of @xmath14 for several magnetic fields at an injected current @xmath205 . assuming that the emission of optical phonons is indeed by far the most efficient scattering process for hot electrons with @xmath206 we expect a constant oscillation amplitude of @xmath10 as long as @xmath207 . this means that all optical phonons that are energetically allowed are emitted before the electron reaches the collector barrier . accordingly we can interpret an onset of a decrease of the oscillation amplitude of @xmath10 as a function of @xmath87 as the electron energy at which @xmath208 . this energy is marked in figs . [ differentfields](b ) and [ differentfields](c ) by vertical lines and corresponds to @xmath209 . inserting this relation into eq . [ eq : taufirst ] then yields @xmath210 which can be used to calculate @xmath179 from the data for different magnetic fields . for @xmath82 [ fig . [ differentfields](a ) ] the phonon - induced oscillations quickly weaken as the excess energy @xmath72 is increased . we interpret the absence of a constant oscillation amplitude as a scattering time that exceeds the calculated value for @xmath211 . in addition , for @xmath71 our assumption of forward scattering only does not hold , which tends to increase the effective @xmath194 . hence , our model tends to underestimate @xmath179 for @xmath71 . for @xmath212 [ fig.[differentfields](d ) ] , the oscillation amplitude is constant over the whole energy range measured , showing 9 oscillations , so we conclude @xmath213 . the values for @xmath186 and @xmath214 obtained from the data for different magnetic fields shown in figs . [ differentfields](a ) [ differentfields](d ) and similar measurements are summarized in table [ table ] . .[table]estimated number of optical phonons @xmath186 emitted within the distance @xmath194 that hot electrons travel between emitter and collector ( for the algorithm of determining @xmath186 see fig . [ differentfields ] ) . the electron - phonon scattering times @xmath179 are then calculated with eq . ( [ eq : tausecond ] ) . [ cols="^,^,^,^,^,^,^,^,^",options="header " , ] in fig . [ differentfields](e ) @xmath179 ( logarithmic scale ) is plotted as a function of the magnetic field . the errorbars express the uncertainty in determining @xmath186 taken to be @xmath215 . for the two extremal values , @xmath216 and @xmath217 , the plotted values of @xmath179 represent a lower ( @xmath82 ) and upper ( @xmath212 ) limit , respectively . in summary we observe a dramatic decrease of @xmath179 by more than an order of magnitude as the perpendicular magnetic field is increased from zero to a filling fraction beyond @xmath152 . the interaction between hot electrons and optical phonons in nanostructures has been investigated theoretically in detail with and without applied perpendicular magnetic field.@xcite extending the calculations of bockelmann _ _ et al.__@xcite telang _ _ et al.__@xcite investigated the effect of a magnetic field on the electron - phonon scattering rate in quasi one - dimensional systems rigorously by solving the schrdinger equation with boundaries . they found that the scattering rate only weakly depends on the electron energy in a quasi one - dimensional channel . relying on this weak energy dependence , for convenience here we numerically integrate bockelmann s formula for a fixed initial energy of @xmath218mev of the injected electron . this energy value has been chosen to be similar to the one used in ref . where the electron - phonon scattering time in a vertical tunneling structure is calculated for zero magnetic field ( their result of @xmath219ps is not applicable to our case , though , since we have a lateral nanostructure in a 2des ) . we use the edge channel diameter @xmath113 as the lateral confinement width . the @xmath204 component of the phonon momenta ( along the direction of electron confinement ) can be treated as a continuum since @xmath220 . we assume the electrons to occupy the lowest perpendicular subband @xmath221 ( compare last paragraph of sec . [ sec - review ] ) , hence phonon emission mainly occurs within the plane of the 2des . the solid lines in fig . [ differentfields](e ) express our numerical results for @xmath222 . they are in excellent agreement with experimental data even though they contain no free parameters . these theory curves correspond to three branches of intersubband scattering ( subbands in @xmath204 direction ) under emission of an optical phonon . most experimental data points lie just below the lowest branch which is expected if the overall scattering rate is dominated by the fastest process [ lowest branch fig . [ differentfields](e ) ] with a slight contribution of slower processes ( higher branches ) . the transitions between the three branches result from the according discontinuities in the electronic one - dimensional density of states ( see bockelmann s results@xcite ) . these jumps have been observed and discussed in detail in numerical calculations by telang _ _ et al.__@xcite . until now , only perpendicular magnetic fields that guide the injected electrons away from the side contact have been discussed ( @xmath110 ) . in this section , the discussion is extended to the opposite field direction ( @xmath223 ) that tends to direct the injected electrons into the side contact [ compare figs . [ cyclotron](d ) and [ cyclotron](e ) ] . in a nave model one would expect @xmath224 ( and @xmath225 ) for @xmath226 as long as @xmath227 is large enough to confine the current to edge channels , and for a perfect side contact without reflection . [ cyclotron](a ) and [ cyclotron](b ) ( color online ) ( a , b ) @xmath10 as a function of @xmath12 and @xmath14 as in fig . [ current_color ] , for @xmath228 @xmath229 ) at perpendicular magnetic fields of @xmath230 t and @xmath231 t . contour lines are spaced by . ( c ) vertical traces ( @xmath232 is plotted as a function of @xmath72 ) through the absolute minimum of @xmath10 for three different magnetic fields : @xmath230 t at @xmath233na ( a ) , @xmath231 t at @xmath234na ( b ) and @xmath71 at @xmath235na ( from data similar to fig . [ current_color](a ) but for @xmath236 as in [ cyclotron](a ) and ( b ) ) . ( d f ) sketches of the classically expected skipping orbit motion of hot electrons with energies of @xmath237mev ( d , e ) and @xmath238mev ( f ) corresponding to the absolute minima of @xmath10 at @xmath239 t . a zone of depletion extending an estimated from the etched edges into the hall bar is marked in blue . ] show measurements of @xmath10 as a function of @xmath14 and ( relatively small ) @xmath12 for opposite field directions @xmath240 . all other parameters are kept equal ( the color scale has been adapted , though ) . as in the previous figures , white contour lines denote @xmath23 while for black lines @xmath24 . vertical traces ( for constant @xmath12 ) including the respective absolute minimum of @xmath10 are plotted in fig . [ cyclotron](c ) for both magnetic field directions @xmath240 as well as for @xmath71 . the nave expectation @xmath241 for @xmath223 based on edge channel transport is clearly not fulfilled . instead we observe both oscillations of @xmath10 induced by the emission of optical phonons and amplification ( @xmath23 ) . nevertheless , the differences for opposite magnetic field directions are striking . for @xmath223 the phonon - induced oscillations decrease much quicker as @xmath72 is increased and the region of amplification ( @xmath23 ) is much smaller . in addition the maximum amplification effect [ absolute mimimum of @xmath10 marked by @xmath242 in fig . [ cyclotron](c ) ] occurs at a much larger energy @xmath243mev for @xmath223 compared to @xmath244mev for @xmath110 . the absolute minimum for @xmath82 occurs between these two values at @xmath245mev . for even more negative magnetic fields , @xmath10stays positive for all @xmath72 , thus no amplification effect can be observed at all . this can be seen in fig.[otherfields ] ( color online ) @xmath232 versus @xmath72 for constant @xmath246na . included is a measurement for @xmath231 t as already shown in fig . [ cyclotron](c ) and two additional curves for @xmath247 t and @xmath248 t measured at @xmath249 @xmath250 . the slightly different collector barrier height for @xmath231 t of @xmath251 is not important for the qualitative behavior of interest here . ] which shows the @xmath252 trace from fig . [ cyclotron](c ) as well as measurements for @xmath253 and . the latter field is large enough to prevent amplification ( @xmath24 ) . the landauer - bttiker edge channel picture is limited to linear - response transport for @xmath254 , and is therefore not appropriate to describe our data at @xmath255 . in this non - linear regime scattering of the hot electrons , incomplete screening of the lorentz force acting on hot electrons in the bulk of the 2des , and the energy - dependent cyclotron radius have to be accounted for . in the following , we develop a semiclassical approach to model the magnetic field dependence far from equilibrium . as before we assume that the cyclotron radius @xmath256 roughly defines the width of the channel in which most of the hot electrons move along the edge of the 2des . this is assured by the large lorentz force acting on the hot electrons which forces them back to the edge even after scattering . the cyclotron radius at the energy @xmath244mev of the absolute minimum of @xmath10 for @xmath257 t is @xmath258 nm , still relatively small . hence the electrons move along the edge in a narrow channel as sketched in fig.[cyclotron](d ) . for the reversed magnetic field @xmath231 t electrons are guided towards the side contact as shown in fig.[cyclotron](e ) . this should lead to @xmath241 , and indeed @xmath10 becomes positive for relatively small @xmath83mev , where we observe the maximum amplification for @xmath110 [ fig . [ cyclotron](c ) ] . fig.[cyclotron](f ) sketches the situation at an energy of @xmath243mev , corresponding to @xmath259 nm at @xmath260 . at this energy , @xmath261 is already comparable to the width of the side contact and energy exchange between the two edge channels with opposite current direction can occur . as a result excited electrons can reach the collector barrier and amplification can occur similarly to @xmath110 . furthermore , within the region of backscattering ( at the `` mouth '' of the side contact ) electron - electron scattering is enhanced as the two `` edge channels '' of opposite direction tend to merge into a region without a preferred direction . for large energies @xmath72 the side current therefore becomes more negative for @xmath223 compared to @xmath110 . our semiclassical model based on edge transport of hot electrons in a strong perpendicular magnetic field fails to explain one crucial observation , though . in the region of relatively small energies @xmath83mev where @xmath24 for @xmath223 we observe phonon - induced oscillations while our model does _ not _ propose such an effect since it assumes that all current - carrying electrons are guided towards the side contact . in this case the emission of optical phonons should not influence the number of hot electrons reaching this contact . even backscattering at a non - perfect ohmic contact which is in our case one millimeter away would not allow such a behavior since after having traveled such a distance all electron - hole pairs would have recombined within the edge channel , and no effects caused by phonon emission should be visible . in addition we would expect to see @xmath241 , but fig . [ cyclotron](c ) shows that this is not the case . the observation of phonon - induced oscillations in this regime points towards leakage of energy into bulk states related to the emission of optical phonons by hot electrons moving along the edge . however , free hot electrons scattered into the bulk of the 2des tend to be directed back to the edge of the 2des by the lorentz force acting on them . this contradiction can be resolved by taking localized bulk states into account which are a result of disorder combined with reduced screening in a magnetic field . a hot electron can travel from be to bc by a combination of energy relaxation and hopping between these localized states . our experimental observations are consistent with the assumption that in a large magnetic field a small part of the overall current is carried by such a hopping transport mechanism . finally , we observe that the amplitude of the phonon - induced oscillations decays faster as the energy @xmath72 is increased for @xmath223 compared to @xmath110 . this behavior might be related to the transport via localized states but will not be discussed in detail here . in conclusion we have studied the energy relaxation of hot electrons injected at an energy @xmath66 into an otherwise degenerate 2des . the transport measurements have been performed in a mesoscopic three - terminal device in which two of the contacts ( emitter and collector ) are separated by tunable electrostatic barriers . the emitter current and the energy of the injected electrons are fully controlled while the currents into the other two contacts are measured . our main observation is that as a function of a perpendicular magnetic field the electron - electron scattering rate decreases while the emission of optical phonons increases drastically . quantitatively , the corresponding electron - phonon relaxation time declines from over 2ps at zero magnetic field to below 0.2ps at strong perpendicular magnetic fields . numerical calculations within the theoretical standard approaches quantitatively confirm our data and support the following interpretation of our experiments . we have observed the transition from two - dimensional scattering of hot electrons at zero magnetic field towards one - dimensional dynamics at large perpendicular magnetic fields . in the quasi one - dimensional limit the interaction of hot electrons with optical phonons becomes so strong that we clearly observe the emission of more than 10 optical phonons by individual hot electrons in the current signals . this allows us to determine the onset energy of lo phonons in gaas at @xmath262mk with high precision to @xmath0mev . finally , measurements as a function of the direction of a strong perpendicular magnetic field suggest that the expected flow of the hot electrons along the mesa edges is accompanied by a second transport contribution which we interpret as hot electrons hopping between localized bulk states of the 2des . we thank j.p . kotthaus and a. govorov for helpful discussions . financial support by the german science foundation via sfb 631 , lu 819/4 - 1 , and the german israel program dip , the german excellence initiative via the `` nanosystems initiative munich ( nim ) '' , and lmuinnovativ ( funs ) is gratefully acknowledged . the main focus of this article is to study @xmath10 as a function of the kinetic excess energy @xmath72 of the injected electrons . fig . [ current_line ] ( color online ) @xmath10 as a function of @xmath12 for several values of @xmath263 ( given in the plot ) ; @xmath249 @xmath264 . @xmath71 in ( a , b ) and @xmath265 t in ( c , d ) . curves for relative low excess energies @xmath266mev [ equal spacings of 10mev in ( a ) and energies as marked in ( c ) ] are plotted in ( a , c ) while the plots in ( b , d ) concentrate on @xmath267mev ( energies marked ) . ] shows the side current @xmath10 as a function of @xmath12 for @xmath71 ( left ) as well as for @xmath265 t ( right ) . the @xmath71 data have already been discussed in ref . . data for @xmath268mev are plotted in fig.[current_line](a ) . in this low - energy regime , @xmath10 first rapidly decreases as a function of @xmath12 before it grows again for larger currents . strikingly , for @xmath269 nm all curves measured at energies below @xmath270mev follow the same straight line which is actually parallel to the line expected for ohmic behavior [ dashed in fig.[current_line](a)].@xcite . this curve shape can be interpreted as an ohmic contribution to @xmath10 ( which is proportional to @xmath12 and is determined by the ohmic resistances of the three - terminal device ) plus a negative contribution that saturates at @xmath271a . the saturation of the negative contribution to @xmath10 is explained by taking into account the neutralization of electron - hole pairs created by electron - electron scattering . the generation rate of the electron - hole pairs should be roughly proportional to the number of injected electrons , so neutralization becomes more efficient at larger @xmath12 . this is related to the positive charge building up between be and bc due to the amplification effect which hinders the escape of hot electrons via bc . since electrons that can not escape add to the neutralization of holes a steady state is reached in which the negative contibution to @xmath10saturates.@xcite the ohmic contribution of @xmath10 proportional to @xmath12 is straightforward to explain . at energies at which the electron - electron scattering length @xmath36 is small , multiple scattering processes result in many electrons with small kinetic energies and almost arbitrary direction of their momentum . these electrons mimic a diffusive motion , the prerequisite for an ohmic behavior.@xcite the regime of high energies @xmath267mev is depicted in fig . [ current_line](b ) . here we observe the transition to @xmath36being longer than the device dimensions , which | in its extremes | results in @xmath25 ( observed at @xmath272mev ) because the hot electrons move ballistically through the sample.@xcite in a strong perpendicular magnetic field the amplification effect is reduced , hence fig . [ current_line](c ) has a differently scaled y - axis . the preferred directional motion along the edges of the hall bar at @xmath265 t prevents ohmic contributions even at large currents . this explains part of the more complex behavior seen for low energies @xmath268mev in fig.[current_line](c ) at @xmath265 t , namely that the bunching of curves @xmath10(@xmath12 ) for several energies at high @xmath12 is missing . in addition , in a large positive magnetic field the minimum of @xmath10 strongly shifts as a function of energy and @xmath23 persists to larger @xmath12 and larger energies @xmath72 compared to @xmath71 [ compare figs . [ current_line](c ) and [ current_line](d ) ] . a detailed discussion of these effects which reflect the more complex situation in a strong perpendicular magnetic field including the enhanced emission of optical phonons would be very difficult . in a positive magnetic field the high - energy limit @xmath25 is reached at a larger @xmath72 as for @xmath71 [ see also figs . [ current_color_zerofield](c ) and [ current_color](a ) ] . this tendency might be understood in terms of the stronger interaction with optical phonons which shifts the regime of ballistic motion throughout the hall bar towards higher velocities of the injected electrons .

we introduce various quantitative and mathematical definitions for price momentum of financial instruments . the price momentum is quantified with velocity and mass concepts originated from the momentum in physics . by using the physical momentum of price as a selection criterion , the weekly contrarian strategies are implemented in south korea kospi 200 and u.s . s&p 500 universes . the alternative strategies constructed by the physical momentum achieve the better expected returns and reward - risk measures than those of the traditional contrarian strategy in weekly scale . the portfolio performance is not understood by the fama - french three - factor model . price momentum , momentum / contrarian strategies , alternative stock selection rule

1208.2775

searching for the existence of arbitrage is an important task in finance . in the case of the statistical arbitrages , regardless of their origins such as market microstructure , firm - specific news / events , and macroeconomic factors , it is possible to exploit arbitrage opportunities via trading strategies in order to take consistent profits . among such kinds of the statistical arbitrage chances , they are also called as market anomalies , if origins of them are not well - explained nor understood quantitatively and qualitatively @xcite . to academic researchers in finance , it is very useful to testing the robustness of the efficient market hypothesis @xcite and no - arbitrage theorem . although they had played the keystone roles in asset pricing theory and general finance , their statuses recently have been changed as alternative theories that intrinsically allow the pricing anomalies have appeared in financial markets , as instances , the adaptive market hypothesis @xcite and behavioral finance @xcite . hunting for the systematic arbitrage opportunities is also crucial to market practitioners such as traders and portfolio managers on wall street because it is the core of money - making process that is the most important role of them . among these market anomalies , price momentum has been the most well - known example to both groups . since jegadeesh and titman s seminal paper @xcite , it has been reported that the prices of financial instruments exhibit the momentum effect that the future price movement tends to keep the same direction along which it has moved during a given past period . it is also realized that the momentum strategy , a long - short portfolio based on the momentum effect , has been a profitable trading strategy in the stock markets of numerous developed and emerging countries during a few decades even after its discovery @xcite . in addition to the existence in equity markets , the momentum effect large enough to implement as the trading strategy is also found in other asset classes such as foreign currency exchange @xcite , bond @xcite , futures @xcite , and commodities markets @xcite . in spite of its success in profitability over diverse asset classes and markets , its origin has not been fully understood in the frame of traditional mainstream finance . this is why the momentum effect is one of the most famous market anomalies . attempts to explain the momentum effect with factor analysis have failed @xcite and the reason why the momentum effect has persisted over decades still remains mysterious . the fama - french three factor model is able to explain small portions of the momentum return @xcite . the lead - lag effect or auto-/cross - sectional correlation between equities are one of the possible answers to the momentum effect @xcite . the sector momentum is another partial interpretation on the anomaly @xcite . additionally , the behavioral aspects of investors such as collective response to financial news and events have broadened the landscape of understanding on the momentum effect @xcite . transaction cost is also considered a factor caused the momentum effect @xcite . unfortunately , none of these explanations is capable of providing the entire framework for explaining why the momentum of price dynamics exists in many financial markets . not only demystification on the origins of the price momentum , pursuit on the profitability and implementability of the momentum effect in financial markets also has been interesting to academics and practitioners . for example , although several studies @xcite found that the momentum strategies in some asian markets such as japanese stock market are not profitable , asness et al . @xcite discovered that the momentum strategy in japan becomes lucrative , when it is combined with other negatively correlated strategies such as value investment . not only in several stock markets , the hybrid portfolio of value and momentum also outperforms each of the value and momentum portfolios across the asset classes . their study paid attention to the implementation of the momentum strategy combining with fundamental value investment factors such as book - market ( bm ) ratio which also has been used to unveil the origins of the momentum effect in fama - french three factor analysis . in other words , their work can be understood as the construction of the hybrid portfolio to increase the profitability and stability of the portfolios based on the momentum strategy . moreover , the selection criteria for the hybrid portfolio are considered as the multiple factors related to the momentum returns whether they are positively correlated or negatively correlated . academically , this observation imposes the important meaning in the sense that these multiple filters can explain their contributions to the momentum returns . in practical viewpoint , it is obviously the procedure for generating trading profits in the markets . another method for improving the profitability of the momentum strategy is introducing various selection rules to the construction of the momentum portfolio . first of all , simple variations in the original momentum selection rule can be made . moskowitz et al . @xcite suggested new trading strategies based on time series momentum which constructs the momentum portfolios by time series regression theory . it is not simply from cumulative return during a lookback period as a sorting variable but from an autoregressive model of order one which can forecast the future returns under given conditions such as the past returns and volatilities . the predicted return is used as the selection criterion for the time series momentum strategy . the time series momentum portfolio performs very well even during market crisis . it also shares the common component which drives the momentum return with the cross - sectional momentum strategy across many asset classes . this fact imposes that the momentum strategy is improved by the modified cumulative return criterion and there is a possibility to find the better momentum strategies in performance and risk . besides only considering the cumulative return , the introduction of alternative proxies for the portfolio selection rules has been also worth getting attention . george and hwang @xcite used 52-week high price as a selection criterion and the momentum portfolio based on the 52-week high price generated stronger returns . additionally , the tests on the momentum portfolios , which are doubly - sorted by the cumulative return or sector momentum and the 52-week high price , exhibit the superiority of the 52-week high price criterion . the factor analysis also shows that the return from the 52-week high price factor is not only stronger than the traditional and sector momentum factors but also statistically more significant and important in the momentum return modeling . the dominance of the 52-week high momentum criterion is also observed in the various international stock markets @xcite . reward - risk measures are also able to serve as the ranking criteria . rachev et al . @xcite used the reward - risk measures as the sorting criteria for their momentum portfolios instead of the cumulative return over the estimation period . in their work , value - at - risk ( var ) , sharpe ratio , r - ratio , and starr were used as alternative ranking rules . in the s&p 500 universe from 1996 to 2003 , their momentum portfolios constructed by the reward - risk measures provided the better risk - adjusted returns than the traditional momentum strategy . in addition to that , the new momentum portfolios had lower tail indexes for winner and loser baskets . in other words , these momentum strategies based on the reward - risk measures obtained the better risk - adjusted returns with acceptance of the lower tail risk . back to physics , the momentum in price dynamics of a financial instrument is also an intriguing phenomenon because the persistent price dynamics and its reversion can be understood in terms of inertia and force . the selection rules of the momentum strategy is directly related to the ways of how to define and measure `` physical '' momentum in price dynamics of the instrument . when the instrument is considered as a particle in a one - dimensional space , the price momentum is also calculated if mass and velocity are defined . since the momentum effect exists , it can be concluded that price of an equity has an inertia that makes the price keep their direction of movements until external forces are exerted . in this analogy , the external force corresponds to any exogenous market events and information such as good / bad news , changes in psychology and macroeconomic situation , and imbalance in supply and demand . this idea is also able to explain why the cumulative return based momentum strategy generates the positive expected returns . however , it has been not much attractive to physicists yet . most of the econophysics community does nt have been interested in trading strategy and portfolio management so far . recently , choi @xcite suggested that a trading strategy can be considered as being in the spontaneous symmetry breaking phase of arbitrage dynamics . in his work , the return dynamics had a parity in the return which can be broken by choosing the ground state . when a control parameter is smaller than a critical value , the strategy is in the arbitrage phase and we expect the non - zero expected return which is not permitted in the efficient market hypothesis . random fluctuation around the non - zero value makes variance of the strategy return and the risk of loss still exists . the important caveats were not only that the arbitrage strategy can theoretically generate the non - zero expected returns emergent from the symmetry breaking but also that the idea is empirically meaningful when it is applied to real trading strategies . for the simple back - testing , the control parameter , which triggers phase transition , was estimated from an autocorrelation coefficient of the strategy return time series . if the strategy is expected to be in the arbitrage phase , the strategy is exploited and if not , the execution is stopped . as an empirical test , the weekly contrarian strategy was executed based on the scheme using the symmetry breaking arbitrage . the contrarian strategy with the scheme had the better expected return and sharpe ratio than the traditional contrarian strategy does . in this paper , we introduce various definitions for the physical momentum of equity price . based on those definitions , the equity price momentum can be quantified from real historical data in the south korean kospi 200 and s&500 universes . after computing the physical momentum , the implementation of the contrarian strategies based on the candidates for the price momentum increases the validity of our approach to measuring the physical momentum in equity price . empirically , these new candidates for the selection criteria originated from the physical momentum idea provide the better returns and sharpe ratios than the original criterion , i.e. the cumulative return . the structure of this paper is the following . in the next chapter , the definition of velocity in equity price space and possible candidates for financial mass are introduced and then the price momentum is defined with the financial velocity and mass . in section [ sec_phys_data ] , we specify the datasets used for our analysis . in section [ sec_phys_result ] , results for the physical momentum strategies are given . the fama - french three - factor analysis is given in section [ sec_phys_factor_analysis ] . in section [ sec_phys_conclusion ] , we conclude the paper . when a one - dimensional space for price of a financial instrument is introduced , it is possible to consider that the price is in motion on the positive half - line . although the negative price is conceptually proposed by sornette @xcite , the negative price of the instrument is not allowed in practice . the price dynamics of the financial instruments are now changed to a one - dimensional particle problem in physics . to extend the space to the entire line , the log price is mapped to the position @xmath0 in the space by @xmath1 where @xmath2 is the price of the instrument . this transformation is not new to physicists because baaquie @xcite already introduced the same transformation to derive path integral approach to option pricing theory . it was used in order to find the relation between the black - schole equation and schrdinger equation . with this re - parametrization , an option pricing problem was transformed to a one - dimensional potential wall problem in quantum mechanics . however , it was not for introducing the physical momentum concept mentioned above . with the log return , @xmath0 covers the whole line from the negative to positive infinity . in addition to the physical intuition , the log price has some advantages in finance . first of all , it is much simpler to calculate the log return from the log price because the difference of two log prices is the log return . contrasting to the log return , the raw return is more complicated to compute than the log return . secondly , one of the basic assumptions in mathematical finance is that the returns of financial instruments are log - normally distributed and we can handle normally - distributed log returns . with the advantages of the log price described above , it is natural to introduce a concept of velocity into the one - dimensional log price space . in the case of the log price , the log return @xmath3 per time scale is expressed in @xmath0 by @xmath4 in the limit of infinitesimal time interval ( @xmath5 ) , the log return becomes @xmath6 where @xmath7 is the velocity of the instrument in the log price space , @xmath0 . when the mapping between the log price and position in the one - dimensional space is introduced , it imposes the relation between the log return and velocity . although this relation works only in the limit of @xmath5 , it can be used as the approximation in the discrete time limit if the length of the whole time series is long enough to make the time interval relatively shorter . the cumulative return @xmath8 is expressed in @xmath7 by @xmath9 since the log return is usually small such as @xmath10 in real data , higher - order terms in @xmath7 can be treated as higher - order corrections on @xmath8 and it is possible to ignore the higher - order corrections if @xmath10 . in this sense , the cumulative return can be approximated to @xmath7 . however this relation is broken in the cases of heavy tail risks caused by financial crisis or firm - specific events such as bankruptcy , merger and acquisition , and good / bad earning reports of the company . since @xmath11 in these events can be comparable to one or greater than one , the higher - order perturbations should be considered . based on the correspondence , the concept of the price momentum can be quantified by using the classical momentum in physics with @xmath12 where @xmath13 has the same role to physical mass . in particular , when velocity is given in the log return , the contribution of the mass to the price momentum can be expressed in the following way , @xmath14 the financial mass @xmath13 plays a role of amplifying the price change as the mass becomes larger . this amplification is understood as filtering of market information on price . consideration on the volume can catch more market information . for example , large transaction volumes at the peak or trough could impose the change in the trend from the viewpoint of technical analysis . some instruments are heavily influenced by the investors psychology and other market factors but other are nt . in this sense , the mass can act a role of the filter which is unique to each instrument and encodes the instrument - specific characteristics . this interpretation is also well - matched to the physical analogy that mass is a physical constant which is unique to each particle . the original ranking criterion in the traditional momentum strategy is a special case of this physical momentum definition . in the cumulative return momentum strategy , it is assumed that each of equities has the identical mass , @xmath15 . however , the identical mass assumption seems not to be reasonable because each equity has distinct properties and shows inherent price evolutions . in order to capture these heterogeneities between the characteristics of each equity , the departure from the identical financial mass for all equities is more natural and the introduction of the financial mass concept to the momentum strategy looks plausible . although the physical momentum concept can be applied to other asset classes , we focus only on the equities in this paper . as described in the previous paragraph , the financial mass can convey the instrument - specific information . however , it is obvious that all kinds of information can not work as the candidates for the mass because it should be well - matched to intrinsic properties of physical mass . in this sense , liquidity is a good candidate for the financial mass . its importance in finance is already revealed in many financial literatures in terms of bid - ask spread , volume , and fractional turnover rate @xcite . in particular , datar et al.@xcite reported that past turnover rate is negatively correlated to future return . with the same size of the momentum , the larger turnover rate brings the poorer future return i.e. illiquid stocks exhibit higher average returns . even after controlling other factors such as firm - size , beta , and bm ratio , the past turnover rate has the significant negative correlation with the future return . it is possible to understand that the trading volume incorporates integrated opinions of investors and makes the price approach to the equilibrium asymptotically . in the viewpoint of information , trading can be understood as the exchange of information between investors with inhomogeneous information . more transactions occur , more information is widely disseminated over the whole market and the price change becomes more meaningful . lee and swaminathan @xcite also provided the similar result that stocks with low past trading volumes tend to have high future returns . additionally , the study found that the momentum strategy among high volume stocks is more profitable . the similar result is obtained in the south korean market @xcite . the possible mass candidates , which are also well - matched to the analogy of physical mass , are volume , total transaction value , and inverse of volatility . if the trading volume is larger , the price movement can be considered the more meaningful signal because the higher volume increases the market efficiency . the amount of the volume is proportional to mass @xmath13 . as mentioned in the previous paragraph , the relation between the trading volume and asset return is already studied in finance @xcite . instead of the raw volume , we need to normalize the daily volume with the total number of outstanding shares and this normalized value is also known as a turnover rate . the reason of this normalization is that some equities intrinsically have the larger trading volumes than others because the total number of shares enlisted in the markets are much larger than other equities or because they get more investors attentions which cause more frequent trades between investors . the share turnover rate , i.e. trading volume over total outstanding shares is expressed in @xmath16 in the paper . similar to the volume , the daily transaction value in cash can be used as the financial mass . if an equity on a certain day has the larger transaction value , investors trade the equity frequently and the price change has more significant meanings . additionally , the transaction value contains more information than the volume . for examples , even though two equities record the same daily volume and daily return on a given day , the higher priced equity exhibits the larger trading value if two prices are different . the more important meaning is that even though market information such as close price , volume , return , and price band are identical , the trading value in cash can be different . as an instance , when one equity is traded more near the lowest price of the daily band but the other is traded mainly around the daily highest price region , the total transaction values of two equities are obviously different . it also needs to be normalized because each equity price is different . the normalization of dividing total transaction value by market capitalization is expressed in @xmath17 in the paper . the return volatility @xmath18 is inversely proportional to the financial mass @xmath13 . if the volatility of a certain equity in a given period is larger , the equity price is easy to fluctuate much severely than other equities with the smaller volatilities . this correspond to the situation in physics that a lighter object can move more easily than a heavy object under the same force . so if it is heavy in the sense of the volatility , the asset price with larger mass is under the smaller volatility . this definition of the financial mass is also matched with the analogy used in baaquie s works @xcite . in his works , the black - scholes equation was transformed into hamiltonian of a particle under the potential which specifies the option . the mass of a particle in the hamiltonian was exactly same to the inverse of the return volatility . since the volatility is also interesting to economists and financiers , there are series of the literature covering the link between volatility and return @xcite . with the fractional volume and fractional transaction value as the proxies for the mass , it is possible to define two categories of the physical momentum , @xmath19 and @xmath20 over the period of the window size @xmath21 . the latter one is the reminiscent of the center - of - mass momentum in physics and the similar concept is used as the embedded capital gain in grinblatt and han @xcite . since two different categories for the momentum calculation , two for return , and two for mass are available , there are eight different momentum definitions for an equity . it is easily found that the cumulative return can be expressed in @xmath22 by @xmath23 an this shows that the traditional momentum in finance is a special case of the physical momentum . in this sense , let s call @xmath24 . in addition to that , since exponential function and log function are strictly increasing functions , the mapping between @xmath25 and @xmath26 is one - to - one . since the return volatility over the period contains more practical meanings than the sum of daily volatilities during the period , the third class of the physical momentum is defined by @xmath27 where @xmath28 is the average velocity at time @xmath29 during the past @xmath21 periods . there are also two different definitions for @xmath30 computed from the normal return and log return . this is closely related to the sharpe ratio , @xmath31 , @xmath32 where @xmath33 is the risk - free rate . if the risk - free rate is small and ignorable , @xmath30 approaches to the sharpe ratio . the momentum strategy with this ranking criterion is the reminiscent of the sharpe ratio based momentum strategy by rachev et al . similar to the sharpe ratio , @xmath30 can be related to the information ratio that uses excessive returns over the benchmark instead of the risk - free rate in the definition . however , we do nt consider the risk - free rate nor the benchmark return as a reference point of the portfolio returns in this paper . with @xmath34 , @xmath35 , and @xmath30 , total eleven different definitions of physical momentum including the traditional cumulative return are the possible candidates for the physical equity momentum . each of them is originated from the physical and financial foundations . additionally , they are relatively easier to quantify than other risk measures used in rachev s work @xcite . although it is possible to consider more complicated functions of other market data for the price momentum , it is beyond the scope of this paper . the market data and component - change log of the kospi 200 universe are downloaded from korea exchange . the covered time horizon starts january 2003 and ends december 2012 . the daily market information and roaster for s&p 500 components are collected from bloomberg . the time window is identical to the kospi 200 . the comparison of the performances between the traditional momentum strategy and physical momentum strategy is able to test the validity of the physical momentum definition . instead of the traditional momentum strategy that uses the raw return during the lookback period as a ranking criterion , we can construct the momentum portfolio ranked by the various definitions of the physical momentum . after finding the performance , each of the momentum strategies from the various momentum criteria are compared with others in order to measure the validity of a given momentum definition . details on the momentum strategy will follow . the most important variables of the momentum strategy are the length of the lookback ( or estimation ) period @xmath36 , the length of the holding period @xmath37 , and the sorting criterion @xmath38 . the traditional momentum strategy uses the cumulative return during the lookback period as a ranking criterion , i.e. a triplet of the traditional momentum strategy is @xmath39 @xcite . on the reference day ( @xmath40 ) , the cumulative returns of all instruments in the market universe during the periods from @xmath41 to @xmath42 are calculated . after sorting the instruments in the order of the ascending criterion , numbers of ranking groups are constructed and each of the ranking groups has the same number of the instruments . as an instance , if there are 200 equities and we consider 10 groups , each of sorted ranking groups has 20 equities as group constituents . following the convention of jegadeesh and titman @xcite , the loser group who has the worst performers in the market is named as r1 and the winner group with the best performers is the last one , r10 . and then the momentum portfolio is constructed by buying the winners and short - selling the losers with the same size of positions in cash in order to make the composite portfolio dollar - neutral . for the winner and loser portfolios , each group member is equally weighted in the group in which it is . the constructed momentum portfolio is held until the end of the holding period ( @xmath43 ) . on the last day of the holding period ( @xmath43 ) , the momentum portfolio is liquidated by selling the winner group off and buying the loser group back . on the first day of each unit period , the momentum portfolio is constructed as explained in the previous paragraph . for examples , a weekly momentum portfolio is selected in every monday unless it is not a holiday . monthly portfolios are formulated on the first business day in every month . for multiple - period holding strategies , there exists overlapping period between two strategies constructed by the same criterion at the different reference dates . the reasons of this construction are followings . first of all , the momentum return from this construction is not dependent with the starting point of the strategy formation . for example , when we implement the 12-month lookback and 12-month holding momentum strategy , construction of the portfolio occurs at the beginning of each year . since the return results are always interfered by the seasonal effects such as january effect or others related to business cycle and taxation , it is difficult to discern the momentum effect from the seasonal effects . second , the portfolios from overlapped periods can generate the larger numbers of return samples to fortify the statistical significance . since the dataset here only has twelve years of historical data comparing with other studies which uses much longer time periods as datasets , its statistical significance could be lowered by the smaller size of our samples if we use the non - overlapped portfolios . third , jegadeesh and titman @xcite reported that there were not big differences between the returns by the overlapped and non - overlapped portfolios . finally , the portfolio construction here can be considered as diversification which helps to mitigate the large fluctuation of returns in the momentum portfolio . for example , in the case of 12-month holding strategies , we posses 12 different portfolios at a given moment and it is definitely diversification of the portfolio . based on these reasons , it is more sensible that the overlapping portfolios are used in our case . when we buy the winner and loser portfolios which provide expected returns for those groups of @xmath44 and @xmath45 respectively , the return by the momentum portfolio @xmath46 is simply @xmath47 because we short - sell the losers in the portfolio . when we implement the trading strategy in the real financial markets , a transaction cost including brokerage commission and tax is always important because they actually erode the trading profits . the implemented momentum return or transaction - cost - adjusted return @xmath48 is @xmath49 where @xmath50 and @xmath51 are the transaction costs for winner and loser group , respectively . in general , @xmath51 is greater than @xmath50 because the short - selling is usually much more difficult than buying . since the transaction cost is an one - time charge , its effect on the implemented return per unit period becomes smaller as the holding period is lengthened . when the expected return of the momentum portfolio for a given @xmath52 strategy is negative , the strategy can become profitable by simply switching to the contrarian strategy @xmath53 that buys the past loser group and short - sells the past winner group , exactly the opposite position to the momentum portfolio . contrasting to the momentum strategy following the price trend , the contrarian strategy is based on the belief that there is the reversion of price dynamics . if equities have performed well during the past few periods , investors try to sell those stocks to put the profits into their pockets . the investors who bought those equities long time ago are able to make large enough profits even when the price recently has gone slightly downward . however , buyers who recently purchased the equities might not have enough margins yet from their inventories and want not to lose money from the current downward movement because of risk aversion . the only option those investors can take is just selling their holdings off . this herding behavior makes the reversion of price and it is probable to make profits from short - selling if a smarter investor knows when the timing would be . for the opposite case , it is also possible to buy the past losers to get advantages of using the herding because the losers are temporarily undershot by investors massive selling force and the equities tend to recover their intrinsic values . on the way of price recovery , the short - sellers need to buy back what they sold in the past in order to protect their accounts and the serial buy - back can boost the price dynamics to the upward direction which also causes the consequential massive buy - backs by other short - sellers . how much the initial anomaly can be amplified is modeled in shleifer and vishny @xcite . the momentum and contrarian strategies look contradictory to each other but they have only the different time horizons in which each of strategies works well . usually , in three to twelve months scale , the equity follows the trends @xcite but the reversal effect is dominant at the longer and shorter scales than the monthly scale @xcite . for the contrarian strategy , the portfolio return @xmath54 is given by @xmath55 the transaction cost adjusted return @xmath48 for the contrarian strategy is @xmath56 when the implementability of a given strategy in the real markets is the main concern , we need to focus on whether or not it is possible to take actual profits from the strategy among the momentum and contrarian strategies . in this sense , the profitability of the strategy with absolute ( implemented ) return @xmath57 can be measured by @xmath58 and tells whether the potential trading profit can exceed the barrier of the transaction cost . the actual positive return from the momentum / contrarian trading strategies can be taken into the pocket when @xmath57 is positive . however , the transaction cost is not considered because the outperformance / underperfomance with respect to the benchmark strategy is our main concern . as mentioned above , the method for measuring the price momentum is the momentum strategy with the physical momentum as a ranking criterion . there are total eleven types of candidates for physical momentum including the original cumulative return momentum . on the reference day @xmath59 , each physical momentum for equities over the estimation period of 6 weeks is calculated and used for sorting the equities . the ranking for each criterion constructs the momentum portfolios . after holding the portfolio during 6 weeks , it is liquidated to get the momentum profit . the positive implemented returns and sharpe ratios from the implemented return exhibit the robustness of the physical momentum strategies . if their returns beat that of the traditional momentum strategy , it is obvious that the physical momentum definition really has a merit to introduce and there is a practical reason to use the momentum strategies based on the physical momentum as an arbitrage strategy . for the lookback period , some stocks which do nt have enough trading dates are ignored from the analysis . in general , this case happens to companies which are enlisted to the market universe amid of the lookback period . if an equity is traded on only one day during the estimation period , it is neglected from our consideration for the momentum universes because it is impossible to calculate the standard deviation for @xmath60-type momentum for these stocks . since all possible candidates for the physical momentum need to be compared with other criteria over the same sample , it is obvious not to consider these equities with only one trading day in the estimation period . the companies delisted amid of the holding periods do nt cause the same problem because only the lookback return is important in sorting the equities and constructing the momentum portfolios . in this case , the returns for the delisted companies are calculated from the prices on the first and last trading days in the holding periods . in table [ tbl_daily_summary_stat_risk_phys_momentum_weekly_6_6_contrarian_kr_kp200 ] , all weekly contrarian strategies based on physical momentum except for the @xmath60 criteria outperform the traditional contrarian strategy . the @xmath61 portfolio is the best contrarian strategy with the weekly return of 0.261% under the volatility of 2.444% while the benchmark contrarian strategy obtains weekly 0.069% with the standard deviation of 2.846% . the performance of other @xmath22 portfolios is as good as the @xmath61 portfolio and is better than the @xmath62 criterion . although the @xmath63 strategies are slightly worse than the @xmath22 cases , the performance is much better and less volatile than the mean - reversion strategy . although all @xmath60 portfolios exhibit smaller standard deviations , the average returns are all negative . with a few exceptions , the skewness levels of the alternative weekly contrarian strategies are higher and the kurtosis is lower than those of the cumulative return strategy . the historical performance of the portfolios is given in fig . [ grp_acc_return_phys_momentum_weekly_6_6_contrarian_kr_kp200 ] . l l r r r r r r r r r criterion & portfolio & & + & & mean & std . dev . & skewness & kurtosis & fin . wealth & sharpe & @xmath64 & @xmath65 & mdd + @xmath62 & winner ( w ) & 0.1977&3.8044&-1.2951&6.7842&1.0181&0.0689&1.7492&2.5255&67.27 + & loser ( l ) & 0.2662&4.3629&-1.1757&7.6803&1.3708&0.0818&1.6779&2.5040&63.00 + & l w & 0.0685&2.8461&0.1907&1.8784&0.3527&0.0149&1.6930&2.4344&33.40 + + @xmath66 & winner ( w ) & 0.0365&4.3618&-1.2676&6.7068&0.1880&0.0479&2.0039&2.9571&71.44 + & loser ( l ) & 0.2781&4.2244&-1.2474&8.0932&1.4323&0.0888&1.6588&2.3910&64.92 + & l w & 0.2416&2.3988&0.2320&1.0899&1.2443&0.0422&1.5481&2.1681&20.36 + + @xmath67 & winner ( w ) & 0.0433&4.3384&-1.2263&6.4733&0.2229&0.0487&2.0007&2.9610&71.20 + & loser ( l ) & 0.2986&4.2776&-1.2293&8.1616&1.5376&0.0891&1.6558&2.3985&65.17 + & l w & 0.2553&2.4203&0.2482&1.3236&1.3147&0.0431&1.5482&2.1747&21.30 + + @xmath61 & winner ( w ) & 0.0399&4.3078&-1.2111&6.2480&0.2053&0.0482&1.9968&2.9472&70.77 + & loser ( l ) & 0.3006&4.3159&-1.2610&8.4086&1.5479&0.0892&1.6627&2.4087&65.53 + & l w & 0.2607&2.4439&0.2286&1.4620&1.3427&0.0450&1.5289&2.1522&21.51 + + @xmath68 & winner ( w ) & 0.0481&4.2810&-1.1999&6.4031&0.2479&0.0490&1.9824&2.9382&70.31 + & loser ( l ) & 0.3002&4.3536&-1.2939&8.5965&1.5462&0.0886&1.6676&2.4188&66.07 + & l w & 0.2521&2.4656&0.1705&1.6283&1.2983&0.0422&1.5293&2.1521&23.29 + + @xmath69 & winner ( w ) & 0.1214&3.8635&-1.3671&7.6297&0.6251&0.0610&1.8278&2.6834&68.73 + & loser ( l ) & 0.2842&3.9333&-1.6138&11.3616&1.4637&0.0951&1.4908&2.1750&65.07 + & l w & 0.1628&2.1709&0.2235&1.4173&0.8386&0.0281&1.4383&2.0149&21.29 + + @xmath70 & winner ( w ) & 0.1407&3.8520&-1.3704&7.7786&0.7244&0.0624&1.8156&2.6577&68.59 + & loser ( l ) & 0.2891&3.9623&-1.5753&11.1774&1.4889&0.0945&1.4799&2.1615&65.38 + & l w & 0.1484&2.1738&0.2044&1.4014&0.7645&0.0280&1.4775&2.0766&20.54 + + @xmath71 & winner ( w ) & 0.1389&3.8274&-1.3849&7.7040&0.7151&0.0624&1.8149&2.6574&68.95 + & loser ( l ) & 0.2875&3.9804&-1.5728&11.0874&1.4807&0.0940&1.5239&2.2234&65.53 + & l w & 0.1487&2.1910&0.2024&1.6035&0.7657&0.0288&1.5139&2.1327&20.77 + + @xmath72 & winner ( w ) & 0.1597&3.8158&-1.4078&7.9932&0.8223&0.0654&1.8028&2.6433&69.02 + & loser ( l ) & 0.2886&4.0066&-1.5608&10.9274&1.4865&0.0932&1.5209&2.2186&65.49 + & l w & 0.1290&2.2188&0.1553&1.6056&0.6642&0.0258&1.5409&2.1789&21.88 + + @xmath73 & winner ( w ) & 0.2722&3.4612&-1.4260&7.9900&1.4019&0.0794&1.5697&2.2775&62.39 + & loser ( l ) & 0.2486&3.8970&-1.3369&8.6249&1.2801&0.0887&1.4988&2.1919&59.58 + & l w & -0.0237&2.5536&0.3605&2.1385&-0.1218&0.0055&1.5935&2.2586&49.46 + + @xmath74 & winner ( w ) & 0.2871&3.4144&-1.3862&7.6921&1.4783&0.0813&1.5474&2.2447&61.26 + & loser ( l ) & 0.2401&3.9418&-1.3097&8.6519&1.2364&0.0865&1.5137&2.2094&59.65 + & l w & -0.0470&2.6091&0.3298&2.3515&-0.2419&0.0036&1.6149&2.2948&55.65 + [ tbl_daily_summary_stat_risk_phys_momentum_weekly_6_6_contrarian_kr_kp200 ] ( blue ) , @xmath71 ( red ) , and @xmath74 ( green ) in south korea kospi 200.,width=302 ] the outperformance of the @xmath22 and @xmath63 portfolios is achieved by the stronger reversal in each ranking group . the average returns of the loser groups in those strategies are in the range of 0.278%0.301% comparing with weekly 0.266% by the benchmark contrarian portfolio . additionally , the volatility levels of the loser groups are lower than the loser in cumulative return . meanwhile , the winner baskets in the alternative portfolios underperform its competitive winner basket . while the traditional mean - reversion winner group obtains weekly 0.198% , the performance of the winners in @xmath22 is in the range of 0.037%0.048% and the @xmath63 winners gain weekly 0.121%0.160% with the smaller standard deviations than the other strategies . in the cases of the @xmath60 criteria , the loser groups are as good as the other loser baskets but the winner groups show the strongest momentum . an interesting finding in table [ tbl_daily_summary_stat_risk_phys_momentum_weekly_6_6_contrarian_kr_kp200 ] is that the outperformance of all the @xmath22 and @xmath63 portfolios is achieved by taking low risk . all the strategies in these classes are less riskier in every risk measures such as 95% var , cvar , and maximum drawdown . the maximum drawdowns are almost 50% decreased with respect to the benchmark strategy . additionally , the portfolios in the both categories exhibit much higher sharpe ratios . in particular , the risk measures of every selection rules in @xmath22 are at the lowest levels . the risk measures and sharpe ratios of these alternative portfolios are in the narrow ranges . this fact indicates that the consistent risk management is valid for any choices of price momentum definition . the @xmath63 strategies also show the same pattern : higher sharpe ratios and lower risk measures in the narrow ranges . additionally , the risk measures are slightly lower than any other contrarian portfolios . the risk profile of each ranking basket is also consistent with the purpose of the ranking group . every alternative losers are less riskier than the benchmark loser and the winner groups are much exposed to the risk . in particular , the 95% var and cvar levels of the alternative loser groups are lower than those of the loser in cumulative return . meanwhile , the winner groups are under the greater exposures to the risk than the winner basket in the traditional contrarian portfolio . the same pattern is observed for the sharpe ratio . the sharpe ratios of the loser ( winner ) baskets in @xmath22 and @xmath63 portfolios are greater ( smaller ) than that of cumulative return loser . the @xmath63 losers ( winners ) exhibit the slightly better ( worse ) reward - risk measures than @xmath22 . in table [ tbl_daily_summary_stat_risk_phys_momentum_weekly_6_6_contrarian_us_spx ] , every alternative momentum strategies outperform the traditional contrarian strategy . the best portfolios are from the @xmath60 criteria with weekly 0.107% and 0.106% , respectively . these average returns are almost seven - times greater than the performance of the original mean - reversion portfolio . additionally , the volatility levels of the portfolio performance are almost 50% decreased with respect to the benchmark strategy . the @xmath60 portfolios also exhibit the higher skewness and lower kurtosis . the strategies constructed by @xmath22 and @xmath63 also obtain the better performance under the smaller standard deviation . however , the skewness is decreased with respect to the cumulative return strategy . there is no significant improvement in kurtosis . the performance of the alternative portfolios can be found in fig . [ grp_acc_return_phys_momentum_weekly_6_6_contrarian_us_spx ] . l l r r r r r r r r r criterion & portfolio & & + & & mean & std . dev . & skewness & kurtosis & fin . wealth & sharpe & @xmath64 & @xmath65 & mdd + @xmath62 & winner ( w ) & 0.1545&3.4269&-0.8186&6.8648&0.7970&0.0560&1.3386&1.7016&63.77 + & loser ( l ) & 0.1685&4.6130&-0.1908&11.7318&0.8693&0.0494&1.3727&1.7022&81.06 + & l w & 0.0140&2.7235&0.5559&18.0504&0.0722&-0.0018&0.5749&0.7774&68.14 + + @xmath66 & winner ( w ) & 0.1480&4.0676&-0.5063&7.2687&0.7636&0.0507&1.4203&1.8365&71.54 + & loser ( l ) & 0.1807&4.3508&-0.0870&9.3735&0.9326&0.0541&1.4668&1.8387&77.10 + & l w & 0.0327&1.9516&-0.5294&17.1004&0.1690&-0.0013&0.6181&0.8700&42.67 + + @xmath67 & winner ( w ) & 0.1488&3.9203&-0.5343&6.6807&0.7679&0.0507&1.4777&1.9006&70.37 + & loser ( l ) & 0.1877&4.4446&-0.0486&9.7044&0.9684&0.0530&1.4598&1.8298&77.38 + & l w & 0.0388&2.0502&0.1013&14.3828&0.2004&-0.0004&0.6148&0.8658&47.63 + + @xmath61 & winner ( w ) & 0.1500&3.8309&-0.5550&6.2724&0.7738&0.0517&1.4125&1.8306&69.49 + & loser ( l ) & 0.1930&4.4955&-0.0530&9.7263&0.9960&0.0522&1.4650&1.8419&78.09 + & l w & 0.0431&2.1074&0.3413&14.9428&0.2221&-0.0002&0.6133&0.8613&51.78 + + @xmath68 & winner ( w ) & 0.1537&3.6880&-0.5759&5.5509&0.7931&0.0518&1.4104&1.8286&67.46 + & loser ( l ) & 0.1937&4.5815&-0.0799&10.3110&0.9997&0.0515&1.4634&1.8409&78.35 + & l w & 0.0401&2.2530&0.4616&20.3251&0.2067&0.0003&0.6102&0.8587&56.16 + + @xmath69 & winner ( w ) & 0.1402&3.7924&-0.5654&8.0133&0.7236&0.0519&1.3162&1.6957&71.08 + & loser ( l ) & 0.1786&4.0685&-0.1258&9.3141&0.9217&0.0564&1.4230&1.7917&76.01 + & l w & 0.0384&1.8332&-0.3775&16.0465&0.1981&0.0011&0.5720&0.7977&42.57 + + @xmath70 & winner ( w ) & 0.1435&3.6434&-0.6414&7.2830&0.7404&0.0526&1.3151&1.6962&69.64 + & loser ( l ) & 0.1805&4.1670&-0.1287&9.9820&0.9314&0.0559&1.4263&1.7964&76.92 + & l w & 0.0370&1.9170&0.0761&13.1334&0.1910&0.0017&0.5696&0.7971&49.00 + + @xmath71 & winner ( w ) & 0.1416&3.5544&-0.6954&6.9265&0.7308&0.0530&1.3084&1.6894&68.87 + & loser ( l ) & 0.1834&4.2449&-0.1108&10.3969&0.9462&0.0556&1.4250&1.7955&77.25 + & l w & 0.0417&1.9983&0.3715&14.2958&0.2154&0.0047&0.5727&0.7993&51.69 + + @xmath72 & winner ( w ) & 0.1504&3.4075&-0.7225&6.0400&0.7759&0.0544&1.3142&1.7000&67.00 + & loser ( l ) & 0.1877&4.3447&-0.1401&11.1255&0.9687&0.0555&1.4290&1.8006&77.82 + & l w & 0.0374&2.1262&0.4476&19.4093&0.1927&0.0047&0.5662&0.7880&56.10 + + @xmath73 & winner ( w ) & 0.0920&2.8331&-1.0490&8.6160&0.4746&0.0591&1.1677&1.4724&61.18 + & loser ( l ) & 0.1992&3.4731&-0.4319&8.3765&1.0281&0.0635&1.2578&1.5768&70.11 + & l w & 0.1073&1.7319&0.7020&5.0542&0.5535&0.0158&0.4963&0.6816&40.08 + + @xmath74 & winner ( w ) & 0.0968&2.6887&-1.0728&7.7348&0.4995&0.0598&1.1644&1.4695&59.05 + & loser ( l ) & 0.2024&3.5789&-0.3583&8.8179&1.0444&0.0620&1.2602&1.5817&71.17 + & l w & 0.1056&1.8802&0.8292&7.1623&0.5449&0.0145&0.5023&0.6938&45.90 + [ tbl_daily_summary_stat_risk_phys_momentum_weekly_6_6_contrarian_us_spx ] ( blue ) , @xmath71 ( red ) , and @xmath74 ( green ) in u.s . s&p 500.,width=302 ] in weekly scale , the contrarian strategies exhibit the strong reversal in each ranking group . first of all , all loser groups outperform the traditional contrarian loser group . in particular , the performance of the two @xmath60 loser baskets is much better than any other alternative and benchmark loser baskets . additionally , the standard deviations of these @xmath60 loser groups are much lower than those of the other losers constructed by the @xmath62 , @xmath22 , and @xmath63 physical momentum . opposite to the loser groups , the winner groups underperform the winner basket in cumulative return . although the performance of other winner groups is slightly worse than the short basket in cumulative return portfolio , the winner baskets in the @xmath60 criteria achieve the lowest average returns . according to table [ tbl_daily_summary_stat_risk_phys_momentum_weekly_6_6_contrarian_us_spx ] , the @xmath60 portfolios are less riskier in 95% var , cvar , and maximum drawdown . the portfolio with inverse volatility and raw return for the physical momentum definition achieves the lowest risk measures . additionally , its sharpe ratio is the largest sharpe ratio among all the alternative portfolios including the benchmark . another @xmath60 strategy is ranked to the next in every measures to the @xmath73 portfolio . meanwhile , other alternative contrarian portfolios constructed by the physical momentum are riskier in 95% var and cvar although the maximum drawdowns are improved . moreover , the sharpe ratios are better than the benchmark case but much lower than the @xmath60 portfolios . the @xmath60 portfolios are less riskier at the level of each ranking basket . the winner and lower baskets in the @xmath60 portfolios achieve the lowest 95% var , cvar , and maximum drawdown among every competitive baskets including the traditional contrarian portfolio . additionally , the sharpe ratios of the loser groups in the @xmath60 portfolios are the largest sharpe ratios . the winner groups of the @xmath60 portfolios also obtain the highest reward - risk measures . the ranking baskets in the @xmath63 portfolios are slightly less risker but the winner and loser groups in the @xmath22 definitions exhibit the worse risk measures . the intercepts and factor exposures of the fama - french three - factor analysis on the s&p 500 results are given in table [ tbl_ff_regression_phys_momentum_weekly_6_6_contrarian_us_spx ] . all the intercepts by the physical momentum definitions are greater than the three - factor alpha of the traditional contrarian strategy . in particular , the @xmath60 portfolios achieve the largest and positive three - factor alphas . l l r @ l r @ l r @ l r @ l r criterion & portfolio & + & & & & & & @xmath75 + @xmath62 & winner ( w)&-0.0494&&1.1237&@xmath76&0.2532&@xmath76&0.2906&@xmath76&0.8744 + & loser ( l)&-0.0965&&1.4406&@xmath76&0.1721&@xmath77&0.6591&@xmath76&0.8445 + & l w&-0.0471&&0.3169&@xmath76&-0.0810&&0.3685&@xmath76&0.1602 + + @xmath66&winner ( w)&-0.1003&&1.2949&@xmath76&0.3612&@xmath76&0.5236&@xmath76&0.8936 + & loser ( l)&-0.0762&&1.4045&@xmath76&0.2632&@xmath76&0.4818&@xmath76&0.8704 + & l w&0.0241&&0.1096&@xmath76&-0.0980&&-0.0418&&0.0172 + + @xmath67&winner ( w)&-0.0901&&1.2575&@xmath76&0.3567&@xmath76&0.4547&@xmath76&0.8911 + & loser ( l)&-0.0744&&1.4219&@xmath76&0.2594&@xmath76&0.5389&@xmath76&0.8680 + & l w&0.0157&&0.1644&@xmath76&-0.0973&&0.0841&&0.0518 + + @xmath61&winner ( w)&-0.0837&&1.2310&@xmath76&0.3670&@xmath76&0.4165&@xmath76&0.8890 + & loser ( l)&-0.0719&&1.4297&@xmath76&0.2567&@xmath76&0.5782&@xmath76&0.8673 + & l w&0.0118&&0.1986&@xmath76&-0.1102&&0.1617&@xmath77&0.0849 + + @xmath68&winner ( w)&-0.0708&&1.1925&@xmath76&0.3693&@xmath76&0.3448&@xmath76&0.8842 + & loser ( l)&-0.0762&&1.4482&@xmath76&0.2549&@xmath76&0.6252&@xmath76&0.8671 + & l w&-0.0053&&0.2557&@xmath76&-0.1144&&0.2805&@xmath76&0.1450 + + @xmath69&winner ( w)&-0.0885&&1.2362&@xmath76&0.2405&@xmath76&0.4664&@xmath76&0.9034 + & loser ( l)&-0.0603&&1.3269&@xmath76&0.1972&@xmath76&0.4441&@xmath76&0.8757 + & l w&0.0282&&0.0907&@xmath77&-0.0433&&-0.0222&&0.0139 + + @xmath70&winner ( w)&-0.0764&&1.2063&@xmath76&0.2370&@xmath76&0.3831&@xmath76&0.9074 + & loser ( l)&-0.0638&&1.3386&@xmath76&0.2013&@xmath76&0.5146&@xmath76&0.8699 + & l w&0.0127&&0.1323&@xmath76&-0.0357&&0.1315&&0.0513 + + @xmath71&winner ( w)&-0.0731&&1.1870&@xmath76&0.2396&@xmath76&0.3311&@xmath76&0.9088 + & loser ( l)&-0.0650&&1.3536&@xmath76&0.1942&@xmath76&0.5621&@xmath76&0.8681 + & l w&0.0080&&0.1666&@xmath76&-0.0454&&0.2310&@xmath76&0.0928 + + @xmath72&winner ( w)&-0.0553&&1.1538&@xmath76&0.2384&@xmath76&0.2512&@xmath76&0.9110 + & loser ( l)&-0.0660&&1.3701&@xmath76&0.1897&@xmath76&0.6280&@xmath76&0.8656 + & l w&-0.0107&&0.2163&@xmath76&-0.0487&&0.3768&@xmath76&0.1676 + + @xmath73&winner ( w)&-0.0721&&1.0048&@xmath76&0.0652&&0.0936&@xmath76&0.9099 + & loser ( l)&-0.0012&&1.1691&@xmath76&0.0816&&0.3046&@xmath76&0.8830 + & l w&0.0709&&0.1642&@xmath76&0.0164&&0.2111&@xmath76&0.1188 + + @xmath74&winner ( w)&-0.0582&&0.9689&@xmath76&0.0661&@xmath77&0.0187&&0.9125 + & loser ( l)&-0.0042&&1.1906&@xmath76&0.0865&&0.3578&@xmath76&0.8791 + & l w&0.0540&&0.2217&@xmath76&0.0204&&0.3391&@xmath76&0.2082 + @xmath76 1% significance & @xmath77 5% significance [ tbl_ff_regression_phys_momentum_weekly_6_6_contrarian_us_spx ] the market and value factors are statistically significant . meanwhile , the size factor does nt show any significance in any portfolios . the factor loadings on the size factor are positive only in the @xmath60 portfolios . for any contrarian portfolios in the s&p 500 universe , the performance of the contrarian portfolios are not explicable with the fama - french three - factor model . for each ranking basket , the intercepts of the regression are all negative . the stronger reversal in the ranking baskets is also found in the fama - french three - factor analysis . the largest loser alphas are achieved by the winner groups in the @xmath60 portfolios . the alphas of the loser groups are exceptionally smaller than other loser baskets . the worst alphas are obtained by the @xmath22 portfolios . the three - factor alphas of the @xmath60 losers are also worse than the cumulative return case . the factor exposures of the contrarian portfolios are similar to other strategies except for @xmath60 . although all the factor loadings for the winner and loser baskets in @xmath78 and @xmath63 are positive and statistically significant , the @xmath60 strategies exhibit the weaker dependence on the size factor . the fama - french three - factor model can explain the large parts of the portfolio performance with high @xmath75 values . in this paper , the various definitions of the physical momentum on equity price are introduced . using the mapping between the price of an financial instrument and position of a particle in the one - dimensional space , the log return corresponds to the velocity in the log - price space . up to the higher - order correction terms , the cumulative return is also considered as the velocity . the candidates for the financial mass to define the equity momentum quantitatively are the fractional volume , fractional transaction amount in cash , and the inverse of volatility . these definitions have the plausible origins not only from the viewpoint of physics but also based on financial viewpoint . with the financial mass and velocity concepts , it is capable of defining the physical momentum in asset price that is called as the price momentum in finance . measuring the physical momentum for each equity , the contrarian strategies using the physical momentum as a ranking criteria are implemented in the kospi 200 and s&p 500 universes . its performance and reward - risk ratios surpass those of the traditional contrarian strategy in the weekly level . the outperformance of the physical momentum definition is based on the strong mean - reversion in each ranking basket . the winner in physical momentum definition underperform the winner group of the traditional contrarian portfolio . the performance of the physical momentum portfolios is not explained by the fama - french three - factor model . the intercepts are higher than the cumulative return strategy and the @xmath75 values are much lower . in future work , the similar test will be conducted in different markets and asset classes . it will be interesting to implement the physical momentum portfolios in different trading strategies such as high frequency . we are thankful to sungsoo choi , jonghyoun eun , wonseok kang , and svetlozar rachev for useful discussions . 99 lo , a. and mackinlay , a. , a non - random walk down wall street , princeton university press , 2001 singal , v. , beyond the random walk : a guide to stock market anomalies and low risk investing , oxford university press , 2003 fama , e. , the behavior of stock market prices , journal of business 38 ( 1965 ) 34 - 105 samuelson , p. , proof that properly anticipated prices fluctuate randomly , industrial management review 6 ( 1965 ) 41 - 49 farmer , j. d. and lo , a. w. , frontiers of finance : evolution and efficient markets , proc . 96 ( 1999 ) 9991 - 9992 lo , a. w. , the adaptive markets hypothesis : market efficiency from an evolutionary perspective , journal of portfolio management , forthcoming lo , a. w. , reconciling efficient markets with behavioral finance : the adaptive markets hypothesis , journal of investment consulting 7 ( 2005 ) 21 - 44 kahneman , d. , slovic , p. , and tversky , a. , judgment under uncertainty : heuristics and biases , cambridge university press , 1982 shleifer , a. and vishny , r. w. , the limits of arbitrage , journal of finance 52 ( 1997 ) 35 - 55 shleifer , a. , inefficient markets : an introduction to behavioral finance , oxford university press , 2000 kahneman , d. and tversky , a. , choices , values and frames , cambridge university press , 2000 jegadeesh , n. and titman , s. , returns to buying winners and selling losers : implications for stock market efficiency , journal of finance 48 ( 1993 ) 65 - 91 rouwenhorst , k. g. , international momentum strategies , journal of finance 53 ( 1998 ) 267 - 284 rouwenhorst , k. g. , local return factors and turnover in emerging stock markets , journal of finance 54 ( 1999 ) 1439 - 1464 okunev , j. and derek w. , do momentum - based strategies still work in foreign currency markets ? , journal of financial and quantitative analysis 38 ( 2003 ) 425 - 447 asness , c. s. , moskowitz , t. , and pedersen , l. h. , value and momentum everywhere , journal of finance 68 ( 2013 ) 929 - 985 moskowitz , t. j. , ooi , y. h. , and pedersen , l. h. , time series momentum , journal of financial economics 104 ( 2012 ) 228 - 250 erb , c. b. , and harvey , c. r. , the strategic and tactical value of commodity futures , financial analysts journal 62 ( 2006 ) 69 - 97 fama , e. f. and french , k. r. , multifactor explanations of asset pricing anomalies , journal of finance 51 ( 1996 ) 55 - 84 lo , a. w. and mackinlay , a. , when are contrarian profits due to stock market overreaction ? , review of financial studies 3 ( 1990 ) 175 - 205 lewellen , j. , momentum and autocorrelation in stock return , review of financial studies 15 ( 2002 ) 533 - 563 moskowitz , t. j. and grinblatt m. , do industries explain momentum ? , journal of finance 54 ( 1999 ) 1249 - 1290 hong , h. and stein , j. c. , a unified theory of underreaction , momentum trading and overreaction in asset markets , journal of finance 54 ( 1999 ) 2143 - 2184 terence , h. , hong , h. , lim , t. , and stein , j. , bad news travels slowly : size , analyst coverage , and the profitability of momentum strategies , journal of finance 55 ( 2000 ) 265 - 295 daniel , k. , hirshleifer , d. , and subrahmanyam , a. , investor psychology and security market under- and over - reactions , journal of finance 53 ( 1998 ) 1839 - 1886 barberis , n. , shleifer , a. , and vishny , r. , a model of investor sentiment , journal of financial economics 49 ( 1998 ) 307 - 343 lesmond , d. a. , schill , m. j. , and zhou , c. , the illusory nature of momentum profits , journal of financial economics 71 ( 2004 ) 349 - 380 koh , b. c. , risk premium and profitability of relative strength strategies , korean journal of financial management 14 ( 1997 ) 1 - 21 ahn , y. g. , and lee , j. d. , investment strategy based on past stock returns and trading volume , asia - pacific journal of financial studies 33 ( 2004 ) 105 - 137 chae , j. , and eom , y. , negative momentum profit in korea and its source , asia - pacific journal of financial studies 38 ( 2009 ) 211 - 236 asness , c. s. , momentum in japan : the exception that proves the rule , the journal of portfolio management 37 ( 2011 ) 67 - 75 liu , m. , liu , q. , and ma , t. , the 52-week high momentum strategy in international stock markets , journal of international money and finance 30 ( 2011 ) 180 - 204 george , t. j. , and hwang , c. , the 52-week high and momentum investing , journal of finance 59 ( 2004 ) 2145 - 2175 rachev , s. , jasic , t. , stoyanov , s. , and fabozzi , f. , momentum strategies based on reward - risk stock selection criteria , journal of banking and finance 31 ( 2007 ) 2325 - 2346 choi , j. , spontaneous symmetry breaking of arbitrage , physica a 391 ( 2012 ) 3206 - 3218 sornette , d. , stock market speculation : spontaneous symmetry breaking of economic valuation , physica a 284 ( 2000 ) 355 - 375 baaquie , b. e. , a path integral approach to option pricing with stochastic volatility : some exact results , j. phys . i france 7 ( 1997 ) 1733 - 1753 baaquie , b. e. , quantum finance : path integrals and hamiltonians for options and interest rates , cambridge university press ( 2004 ) datar , v. , naik , n. , and radcliffe , r. , liquidity and asset returns : an alternative test , journal of financial markets 1 ( 1998 ) 203 - 220 amihud , y. and mendelson , h. , asset pricing and the bid - ask spread , journal of financial economics 17 ( 1986 ) 223 - 249 hu , s. , trading turnover and expected stock returns : the trading frequency hypothesis and evidence from the tokyo stock exchange , working paper , university of chicago ( 1997 ) lee , c. and swaminathan , b. , price momentum and trading volume , journal of finance 55 ( 2000 ) 2017 - 2070 amihud , y. , illiquidity and stock returns : cross - section and time - series effect , journal of finanicial markets 5 ( 2002 ) 31 - 56 kim , s. , kim , d. , and shin , h. , evaluating asset pricing models in the korean stock market , pacific - basin finance journal 20 ( 2012 ) 198 - 227 glosten , l. r. , jagannathan , r. , and runkle , d. e. , on the relation between the expected value and the volatility of the nominal excess return on stocks , journal of finance 48 ( 1993 ) 1779 - 1801 french , k. r. , schwert , g. w. , stambaugh , r. f. , expected stock returns and volatility , journal of financial economics 19 ( 1987 ) 3 - 29 griblatt , m. , and han , b. , the disposition effect and momentum , working paper , ucla ( 2002 ) de bondt , w. f. m. and thaler , r. , does the stock market overreact ? , journal of finance 40 ( 1985 ) 793 - 805

we introduce the concept of `` negative bubbles '' as the mirror ( but not necessarily exactly symmetric ) image of standard financial bubbles , in which positive feedback mechanisms may lead to transient accelerating price falls . to model these negative bubbles , we adapt the johansen - ledoit - sornette ( jls ) model of rational expectation bubbles with a hazard rate describing the collective buying pressure of noise traders . the price fall occurring during a transient negative bubble can be interpreted as an effective random down payment that rational agents accept to pay in the hope of profiting from the expected occurrence of a possible rally . we validate the model by showing that it has significant predictive power in identifying the times of major market rebounds . this result is obtained by using a general pattern recognition method that combines the information obtained at multiple times from a dynamical calibration of the jls model . error diagrams , bayesian inference and trading strategies suggest that one can extract genuine information and obtain real skill from the calibration of negative bubbles with the jls model . we conclude that negative bubbles are in general predictably associated with large rebounds or rallies , which are the mirror images of the crashes terminating standard bubbles . keywords : negative bubble , rebound , positive feedback , pattern recognition , trading strategy , error diagram , prediction , bayesian methods , financial markets , price forecasting , econophysics , complex system , critical point phenomena

1003.5926

financial bubbles are generally defined as transient upward acceleration of prices above fundamental value @xcite . however , identifying unambiguously the presence of a bubble remains an unsolved problem in standard econometric and financial economic approaches @xcite , due to the fact that the fundamental value is in general poorly constrained and it is not possible to distinguish between exponentially growing fundamental price and exponentially growing bubble price . to break this stalemate , sornette and co - workers have proposed that bubbles are actually not characterized by exponential prices ( sometimes referred to as `` explosive '' ) , but rather by faster - than - exponential growth of price ( that should therefore be referred to as `` super - explosive '' ) . see @xcite and references therein . the reason for such faster - than - exponential regimes is that imitation and herding behavior of noise traders and of boundedly rational agents create positive feedback in the valuation of assets , resulting in price processes that exhibit a finite - time singularity at some future time @xmath0 . see @xcite for a general theory of finite - time singularities in ordinary differential equations , @xcite for a classification and @xcite for applications . this critical time @xmath0 is interpreted as the end of the bubble , which is often but not necessarily the time when a crash occurs @xcite . thus , the main difference with standard bubble models is that the underlying price process is considered to be intrinsically transient due to positive feedback mechanisms that create an unsustainable regime . furthermore , the tension and competition between the value investors and the noise traders may create deviations around the finite - time singular growth in the form of oscillations that are periodic in the logarithm of the time to @xmath0 . log - periodic oscillations appear to our clocks as peaks and valleys with progressively greater frequencies that eventually reach a point of no return , where the unsustainable growth has the highest probability of ending in a violent crash or gentle deflation of the bubble . log - periodic oscillations are associated with the symmetry of discrete scale invariance , a partial breaking of the symmetry of continuous scale invariance , and occurs in complex systems characterized by a hierarchy of scales . see @xcite for a general review and references therein . recent literatures on bubbles and crashes can be summarized as the following kinds : first , the combined effects of heterogeneous beliefs and short - sales constraints may cause large movements in asset . in this kind of models , the asset prices are determined at equilibrium to the extent that they reflect the heterogeneous beliefs about payoffs . but short sales restrictions force the pessimistic investors out of the market , leaving only optimistic investors and thus inflated asset price levels . however , when short sales restrictions no longer bind investors , then prices fall back down @xcite . while in the second type , the role of `` noise traders '' in fostering positive feedback trading has been emphasized . these models says trend chasing by one class of agents produces momentum in stock prices @xcite . the empirical evidence on momentum strategies can be found in @xcite . after the discussion on bubbles and crashes , the literatures on rebound should be summarized also . on the theoretical side , there are several competing explanations for price decreases followed by reversals : liquidity and time - varying risk . @xcite stresses the importance of liquidity : as more people sell , agents who borrowed money to buy assets are forced to sell too . when forced selling stops , this trend reverses . @xcite shows that it is risky to be a fundamental trader in this environment and that price reversals after declines are likely to be higher when there is more risk in the price , as measured by volatility . on the empirical front concerning the forecast of reversals in price drops , @xcite shows that the simplest way to predict prices is to look at past performance . @xcite shows that price - dividend ratios forecast future returns for the market as a whole . however , these two approaches do not aim at predicting and can not determine the most probable rebound time for a single ticker of the stock . the innovation of our methodology in this respect is to provide a very detailed method to detect rebound of any given ticker . in this paper , we explore the hypothesis that financial bubbles have mirror images in the form of `` negative bubbles '' in which positive feedback mechanisms may lead to transient accelerating price falls . we adapt the johansen - ledoit - sornette ( jls ) model of rational expectation bubbles @xcite to negative bubbles . the crash hazard rate becomes the rally hazard rate , which quantifies the probability per unit time that the market rebounds in a strong rally . the upward accelerating bullish price characterizing a bubble , which was the return that rational investors require as a remuneration for being exposed to crash risk , becomes a downward accelerating bearish price of the negative bubble , which can be interpreted as the cost that rational agents accept to pay to profit from a possible future rally . during this accelerating downward trend , a tiny reversal could be a strong signal for all the investors who are seeking the profit from the possible future rally . these investors will long the stock immediately after this tiny reversal . as a consequence , the price rebounds very rapidly . this paper contributes to the literature by augmenting the evidence for transient pockets of predictability that are characterized by faster - than - exponential growth or decay . this is done by adding the phenomenology and modeling of `` negative bubbles '' to the evidence for characteristic signatures of ( positive ) bubbles . both positive and negative bubbles are suggested to result from the same fundamental mechanisms , involving imitation and herding behavior which create positive feedbacks . by such a generalization within the same theoretical framework , we hope to contribute to the development of a genuine science of bubbles . the rest of the paper is organized as follows . section 2.1 summarizes the main definitions and properties of the johansen - ledoit - sornette ( jls ) for ( positive ) bubbles and their associated crashes . section 2.2 presents the modified jls model for negative bubbles and their associated rebounds ( or rallies ) . the subsequent sections test the jls model for negative bubbles by providing different validation steps , in terms of prediction skills of actual rebounds and of abnormal returns of trading strategies derived from the model . section 3 describes the method we have developed to test whether the adapted jls model for negative bubbles has indeed skills in forecasting large rebounds . this method uses a robust pattern recognition framework build on the information obtained from the calibration of the adapted jls model to the financial prices . section 4 presents the results of the tests concerning the performance of the method of section 3 with respect to the advanced diagnostic of large rebounds . section 5 develops simple trading strategies based on the method of section 3 , which are shown to exhibit statistically significant returns , when compared with random strategies without skills with otherwise comparable attributes . section 6 concludes . @xcite , @xcite , @xcite developed a model ( referred to below as the jls model ) of financial bubbles and crashes , which is an extension of the rational expectation bubble model of @xcite . in this model , a crash is seen as an event potentially terminating the run - up of a bubble . a financial bubble is modeled as a regime of accelerating ( super - exponential power law ) growth punctuated by short - lived corrections organized according the symmetry of discrete scale invariance @xcite . the super - exponential power law is argued to result from positive feedback resulting from noise trader decisions that tend to enhance deviations from fundamental valuation in an accelerating spiral . in the jls model , the dynamics of stock markets is described as @xmath1 where @xmath2 is the stock market price , @xmath3 is the drift ( or trend ) and @xmath4 is the increment of a wiener process ( with zero mean and unit variance ) . the term @xmath5 represents a discontinuous jump such that @xmath6 before the crash and @xmath7 after the crash occurs . the loss amplitude associated with the occurrence of a crash is determined by the parameter @xmath8 . the assumption of the constant jump size is easily relaxed by considering a distribution of jump sizes , with the condition that its first moment exists . then , the no - arbitrage condition is expressed similarly with @xmath8 replaced by its mean . each successive crash corresponds to a jump of @xmath5 by one unit . the dynamics of the jumps is governed by a crash hazard rate @xmath9 . since @xmath10 is the probability that the crash occurs between @xmath11 and @xmath12 conditional on the fact that it has not yet happened , we have @xmath13 = 1 \times h(t ) dt + 0 \times ( 1- h(t ) dt)$ ] and therefore @xmath14 = h(t)dt~. \label{theyjytuj}\ ] ] under the assumption of the jls model , noise traders exhibit collective herding behaviors that may destabilize the market . the jls model assumes that the aggregate effect of noise traders can be accounted for by the following dynamics of the crash hazard rate @xmath15 the intuition behind this specification ( [ eq : hazard ] ) has been presented at length in @xcite , among others , and further developed in ( sornette and johansen , 2002 ) for the power law part and by @xcite and ( zhou et al . , 2005 ) for the second term in the right - hand - side of expression ( [ eq : hazard ] ) . in a nutshell , the power law behavior @xmath16 embodies the mechanism of positive feedback posited to be at the source of the bubbles . if the exponent @xmath17 , the crash hazard may diverge as @xmath11 approaches a critical time @xmath0 , corresponding to the end of the bubble . the cosine term in the r.h.s . of ( [ eq : hazard ] ) takes into account the existence of a possible hierarchical cascade of panic acceleration punctuating the course of the bubble , resulting either from a preexisting hierarchy in noise trader sizes @xcite and/or from the interplay between market price impact inertia and nonlinear fundamental value investing @xcite . the no - arbitrage condition reads @xmath18=0 $ ] , where the expectation is performed with respect to the risk - neutral measure , and in the frame of the risk - free rate . this is the standard condition that the price process is a martingale . taking the expectation of expression ( [ eq : dynamic ] ) under the filtration ( or history ) until time @xmath11 reads @xmath19= \mu(t ) p(t ) dt + \sigma(t ) p(t ) { \rm e}_t[dw ] - \kappa p(t ) { \rm e}_t[dj]~. \label{thetyjye}\ ] ] since @xmath20 = 0 $ ] and @xmath21 = h(t)dt$ ] ( equation ( [ theyjytuj ] ) ) , together with the no - arbitrage condition @xmath18=0 $ ] , this yields @xmath22 this result ( [ tjyj4n ] ) expresses that the return @xmath23 is controlled by the risk of the crash quantified by its crash hazard rate @xmath9 . now , conditioned on the fact that no crash occurs , equation ( [ eq : dynamic ] ) is simply @xmath24 its conditional expectation leads to @xmath25 = \kappa h(t ) dt~.\ ] ] substituting with the expression ( [ eq : hazard ] ) for @xmath9 and integrating yields the so - called log - periodic power law ( lppl ) equation : @xmath26 = a + b(t_c - t)^m + c(t_c - t)^m\cos(\omega\ln ( t_c - t ) - \phi ) \label{eq : lppl}\ ] ] where @xmath27 and @xmath28 . note that this expression ( [ eq : lppl ] ) describes the average price dynamics only up to the end of the bubble . the jls model does not specify what happens beyond @xmath0 . this critical @xmath0 is the termination of the bubble regime and the transition time to another regime . this regime could be a big crash or a change of the growth rate of the market . merrill lynch emu ( european monetary union ) corporates non - financial index in 2009 @xcite provides a vivid example of a change of regime characterized by a change of growth rate rather than by a crash or rebound . for @xmath17 , the crash hazard rate accelerates up to @xmath0 but its integral up to @xmath11 which controls the total probability for a crash to occur up to @xmath11 remains finite and less than @xmath29 for all times @xmath30 . it is this property that makes it rational for investors to remain invested knowing that a bubble is developing and that a crash is looming . indeed , there is still a finite probability that no crash will occur during the lifetime of the bubble . the excess return @xmath31 is the remuneration that investors require to remain invested in the bubbly asset , which is exposed to a crash risk . the condition that the price remains finite at all time , including @xmath0 , imposes that @xmath32 . within the jls framework , a bubble is qualified when the crash hazard rate accelerates . according to ( [ eq : hazard ] ) , this imposes @xmath17 and @xmath33 , hence @xmath34 since @xmath32 by the condition that the price remains finite . we thus have a first condition for a bubble to occur @xmath35 by definition , the crash rate should be non - negative . this imposes @xcite @xmath36 as recalled above , in the jls framework , financial bubbles are defined as transient regimes of faster - than - exponential price growth resulting from positive feedbacks . we refer to these regimes as `` positive bubbles . '' we propose that positive feedbacks leading to increasing amplitude of the price momentum can also occur in a downward price regime and that transient regimes of faster - than - exponential _ downward _ acceleration can exist . we refer to these regimes as `` negative bubbles . '' in a `` positive '' bubble regime , the larger the price is , the larger the increase of future price . in a `` negative bubble '' regime , the smaller the price , the larger is the decrease of future price . in a positive bubble , the positive feedback results from over - optimistic expectations of future returns leading to self - fulfilling but transient unsustainable price appreciations . in a negative bubble , the positive feedbacks reflect the rampant pessimism fueled by short positions leading investors to run away from the market which spirals downwards also in a self - fulfilling process . the symmetry between positive and negative bubbles is obvious for currencies . if a currency a appreciates abnormally against another currency b following a faster - than - exponential trajectory , the value of currency b expressed in currency a will correspondingly fall faster - than - exponentially in a downward spiral . in this example , the negative bubble is simply obtained by taking the inverse of the price , since the value of currency a in units of b is the inverse of the value of currency b in units of a. using logarithm of prices , this corresponds to a change of sign , hence the `` mirror '' effect mentioned above . the jls model provides a suitable framework to describe negative bubbles , with the only modifications that both the expected excess return @xmath23 and the crash amplitude @xmath8 become negative ( hence the term `` negative '' bubble ) . thus , @xmath3 becomes the expected ( negative ) return ( i.e. , loss ) that investors accept to bear , given that they anticipate a potential rebound or rally of amplitude @xmath37 . symmetrically to the case of positive bubbles , the price loss before the potential rebound plays the role of a random payment that the investors honor in order to remain invested and profit from the possible rally . the hazard rate @xmath9 now describes the probability per unit time for the rebound to occur . the fundamental equations ( [ eq : hazard ] ) and ( [ eq : lppl ] ) then hold mutatis mutandis with the inequalities @xmath38 being the opposite to those corresponding to a positive bubble as described in the preceding subsection . an example of the calibration of a negative bubble with the jls model ( 4 ) to the s&p 500 index from 1973 - 01 - 01 to 1974 - 10 - 01 is shown in the upper panel of figure 1 . during this period , the s&p 500 index decreased at an accelerating pace . this price fall was accompanied by very clear oscillations that are log - periodic in time , as described by the cosine term in formula ( 4 ) . notice that the end of the decreasing market is followed by a dramatic rebound in index price . we hypothesize that , similar to a crash following an unsustainable super - exponential price appreciation ( a positive bubble ) , an accelerating downward price trajectory ( a negative bubble ) is in general followed by a strong rebound . furthermore , in order to suggest that this phenomenon is not an isolated phenomenon but actually happens widely in all kinds of markets , another example in the foreign exchange market is presented in the lower panel of figure 1 . the usd / eur change rate from 2006 - 07 - 01 to 2008 - 04 - 01 also underwent a significant drawdown with very clear log - periodic oscillations , followed by a strong positive rebound . one of the goals of this paper is to identify such regions of negative bubbles in financial time series and then use a pattern recognition method to distinguish ones that were ( in a back - testing framework ) followed by significant price rises . in financial markets , large positive returns are less frequent than large negative returns , as expressed for instance in the skewness of return distributions . however , when studying drawdowns and drawups ( i.e. , runs of same sign returns ) . johansen and sornette found that , for individual companies , there are approximately twice as many large rallies as crashes with amplitude larger than 20% with durations of a few days @xcite . we adapt the pattern recognition method of @xcite to generate predictions of rebound times in financial markets on the basis of the detection and calibration of negative bubbles , defined in the previous section . we analyze the s&p 500 index prices , obtained from yahoo ! finance for ticker ` gspc ' ( adjusted close price ) . the start time of our time series is 1950 - 01 - 05 , which is very close to the first day when the s&p 500 index became available ( 1950 - 01 - 03 ) . the last day of our tested time series is 2009 - 06 - 03 . we first divide our s&p 500 index time series into different sub - windows @xmath39 of length @xmath40 according to the following rules : 1 . the earliest start time of the windows is @xmath41 1950 - 01 - 03 . other start times @xmath42 are calculated using a step size of @xmath43 calendar days . 2 . the latest end time of the windows is @xmath44 2009 - 06 - 03 . other end times @xmath45 are calculated with a negative step size @xmath46 calendar days . the minimum window size @xmath47 calendar days . the maximum window size @xmath48 calendar days . these rules lead to 11,662 windows in the s&p 500 time series . for each window , the log of the s&p 500 index is fit with the jls equation ( [ eq : lppl ] ) . the fit is performed in two steps . first , the linear parameters @xmath49 and @xmath50 are slaved to the non - linear parameters by solving them analytically as a function of the nonlinear parameters . we refer to @xcite ( page 238 and following ones ) , which gives the detailed equations and procedure . then , the search space is obtained as a 4 dimensional parameter space representing @xmath51 . a heuristic search implementing the tabu algorithm @xcite is used to find initial estimates of the parameters which are then passed to a levenberg - marquardt algorithm @xcite to minimize the residuals ( the sum of the squares of the differences ) between the model and the data . the bounds of the search space are : @xmath52 \label{eq : rangem}\\ \omega & \in & [ 0.01 , 40 ] \\ \phi & \in & [ 0.001 , 2\pi]\\\label{eq : rangetc } t_c & \in & [ t_2 , t_2 + 0.375(t_2 - t_1)]\end{aligned}\ ] ] we choose these bounds because @xmath53 has to be between @xmath54 and @xmath29 according to the discussion before ; the log - angular frequency @xmath55 should be greater than @xmath54 . the upper bound @xmath56 is large enough to catch high - frequency oscillations ( though we later discard fits with @xmath57 ) ; phase @xmath58 should be between 0 and @xmath59 ; as we are predicting a critical time in financial markets , the critical time should be after the end of the time series we are fitting . finally , the upper bound of the critical time should not be too far away from the end of the time series since predictive capacity degrades far beyond @xmath45 . we have empirically found elsewhere @xcite one - third of the interval width to be a good cut - off . the combination of the heuristic and optimization results in a set of parameters @xmath60 and @xmath0 for each of the 11,662 windows . of these parameter sets , 2,568 satisfy the negative bubble condition ( [ eq : rb ] ) . in figure [ fg : histo ] , we plot the histogram of critical time @xmath0 for these negative bubble fits and the _ negative _ logarithm of the s&p 500 time series . peaks in this time series , then , indicate minima of the prices , many of these peaks being preceded by a fast acceleration with upward curvature indicating visually a faster - than - exponential growth of @xmath61 . this translates into accelerating downward prices . notice that many of these peaks of @xmath62 are followed by sharp drops , that is , fast rebounds in the regular @xmath63 . we see that peaks in @xmath62 correspond to peaks in the negative bubble @xmath0 histogram , implying that the negative bubbles qualified by the jls model are often followed by rebounds . this suggests the possibility to diagnose negative bubbles and their demise in the form of a rebound or rally . if correct , this hypothesis would extend the proposition @xcite , that financial bubbles can be diagnosed before their end and their termination time can be determined with an accuracy better than chance , to negative bubble regimes associated with downward price regimes . we quantify this observation below . the aim is first to recognize different patterns in the s&p 500 index from the 11,662 fits and then use the subset of 2,568 negative bubble fits to identify specific negative bubble characteristics . these characteristics will then be used to ` predict ' ( in a back - testing sense ) negative bubbles and rebounds in the future . we first define a rebound , note as @xmath64 . a day @xmath65 is a rebound @xmath64 if the price on that day is the minimum price in a window of 200 days before and 200 days after it . that is , @xmath66\}~ \label{defreboundh2y}\ ] ] where @xmath67 is the adjusted closing price on day @xmath65 . we find 19 rebounds of the @xmath68-days type in the 59 year s&p 500 index history . our task is to diagnose such rebounds in advance . we could also use other numbers instead of 200 to define a rebound . the predictability is stable with respect to a change of this number . this is because we learn from the learning set with a certain number type of rebounds and try to predict the rebounds of the same type . later we will also show the results for @xmath69-days type of rebounds . in what follows we describe a hierarchy of descriptive and quantitative terms as follows . * * learning set . * a subset of the whole set which only contains the fits with critical times in the past . we learn the properties of historical rebounds from this set and develop the predictions based on these properties . * * classes . * two classes of fits are defined according to whether the critical time of a given fit is near some rebound or not , where ` near ' will be defined below . * * groups . * a given group contains all fits of a given window size . * * informative parameters . * informative parameters are the distinguishing parameters of fits in the same group but different classes . * * questionnaires . * based on the value of an informative parameter , one can ask if a certain trading day is a start of rebound or not . the answer series generated by all the informative parameters is called questionnaire . * * traits . * traits are extracted from questionnaire . they are short and contain crucial information and properties of a questionnaire . * * features . * traits showing the specific property of a single class are selected to be the feature of that class . * * rebound alarm index . * an index developed from features to show the probability that a certain day is a rebound . in this paper , we will show how all the above objects are constructed . our final goal is to make predictions for the rebound time . the development of the rebound alarm index will enable us to achieve our goal . several methodologies are presented to quantify the performance of the predictions . in the pattern recognition method of @xcite , one should define the learning set to find characteristics that will then be used to make predictions . we designate all fits before jan . 1 , 1975 as the learning set @xmath70 : @xmath71 there are 4,591 fits in this set , which we all use without any pre - selection . no pre - selection for instance using eq . ( [ eq : rb ] ) is applied , on the basis of the robustness of the pattern recognition method . we then distinguish two different classes from @xmath70 based on the critical time @xmath0 of the fits . for a single fit @xmath72 with critical time @xmath73 , if this critical time is within @xmath74 days of a rebound , then we assign fit @xmath72 to class i , represented by the symbol @xmath75 . otherwise , @xmath72 is assigned to class ii , represented by the symbol @xmath76 . for this study , we chose @xmath77 days because @xmath74 too big will lose precision and @xmath74 too small will take the noise into account . in this case , class i fits are those with @xmath0 within 10 days of one of the 19 rebounds . we formalize this rule as : @xmath78 to be clear , class i is formed by all the fits in learning set @xmath70 which have a critical time @xmath0 within 10 days of one of the rebounds . all of the fits in the learning set which are not in class i are in class ii . we also categorize all fits into separate _ groups _ ( in addition to the two _ classes _ defined above ) based on the length of the fit interval , @xmath79 . we generate 14 groups , where a given group @xmath80 is defined by : @xmath81 , i = 1 , 2 , ... , 14 , f \in \sigma_1\ } \label{eq : groups}\ ] ] all 4,591 fits in the learning set are placed into one of these 14 groups . for each fit in the learning set , we take 6 parameters to construct a flag that determines the characteristics of classes . these 6 parameters are @xmath82 and @xmath83 from eq . ( [ eq : lppl ] ) , @xmath84 ( the negative bubble condition ) from eq . ( [ eq : bg0 ] ) and @xmath85 as the residual of the fit . we categorize these sets of 6 parameters for fits which are in the same group and same class . then for each class - group combination , we calculate the probability density function ( pdf ) of each parameter using the adaptive kernel method @xcite , generating 168 pdfs ( 6 parameters @xmath86 2 classes @xmath86 14 groups ) . we compare the similarity ( defined below ) of the pdfs of each of the six parameters that are in the same group ( window length ) but different classes ( proximity of @xmath0 to a rebound date ) . if these two pdfs are similar , then we ignore this parameter in this group . if the pdfs are different , we record this parameter of this group as an _ informative parameter_. the maximum number of possible informative parameters is 84 ( 6 parameters @xmath86 14 groups ) . we use the kolmogorov - smirnov method @xcite to detect the difference between pdfs . if the maximum difference of the cumulative distribution functions ( integral of pdf ) between two classes exceeds 5% , then this is an informative parameter . we want to assign a uniquely determined integer @xmath87 to each informative parameter . we can do so by using three indexes , @xmath88 and @xmath89 . the index @xmath90 indicates which group , with @xmath91 $ ] . the index @xmath92 indicates the parameter , where @xmath93 refer to @xmath94 respectively . finally , @xmath89 represents the actual informative parameter . assuming that there are @xmath95 informative parameters in total and using the indexes , @xmath87 is then calculated via @xmath96 for @xmath97 $ ] . given the @xmath95 informative parameters @xmath87 , we consider the pdfs for the two different classes of a single informative parameter . the set of abscissa values within the allowed range given by equations ( [ eq : rangem ] - [ eq : rangetc ] ) , for which the pdf of class i is larger than the pdf of class ii , defines the domain @xmath98 ( ` good region ' ) of this informative parameter which is associated with class i. the other values of the informative parameters for which the pdf of class i is smaller than the pdf of class ii define the domain @xmath99 which is associated with class ii . these regions play a crucial role in the generation of questionnaires in the next section . our hypothesis is that many `` positive '' and `` negative bubbles '' share the same structure described by the jls model , because they result from the same underlying herding mechanism . however , nothing a priori imposes that the control parameters should be identical . note that our pattern recognition methodology specifically extract the typical informative parameter ranges that characterize the `` negative bubbles '' . we realize that many new terms are being introduced , so in an attempt to be absolutely clear , we briefly summarize the method to this point . we sub - divide a time series into many windows @xmath39 of length @xmath100 . for each window , we obtain a set of parameters that best fit the model ( [ eq : lppl ] ) . each of these windows will be assigned one of two _ classes _ and one of 14 _ groups_. classes indicate how close the modeled critical time @xmath0 is to a historical rebound , where class i indicates ` close ' and class ii indicates ` not close ' . groups indicate the length of the window . for each fit , we create a set of six parameters : @xmath82 and @xmath83 from eq . ( [ eq : lppl ] ) , @xmath84 ( the negative bubble condition ) from eq . ( [ eq : bg0 ] ) and @xmath85 as the residual of the fit . we create the pdfs of each of these parameters for each fit and define _ informative parameters _ as those parameters for which the pdfs differ significantly according to a kolmogorov - smirnov test . for each informative parameter , we find the regions of the abscissa of the pdf for which the class i pdf ( fits with @xmath0 close to a rebound ) is greater than the class ii pdf . for informative parameter @xmath89 ( defined in ( [ eq : ip ] ) ) , this region is designated as @xmath98 . in the next section , we will use these _ regions _ to create _ questionnaires _ that will be used to predictively identify negative bubbles that will be followed by rebounds . another important distinction to remember at this point is that the above method has been used to find _ informative parameters _ that will be used below . informative parameters are associated with a _ class _ and a _ group_. using the informative parameters and their pdfs described above , we can generate _ questionnaires _ for each day of the learning or testing set . questionnaires will be used to identify negative bubbles that will be followed by rebounds . the algorithm for generating questionnaires is the following : 1 . obtain the maximum ( @xmath101 ) and minimum ( @xmath102 ) values of @xmath0 from some subset @xmath103 , either the ` learning ' set or the ` predicting ( testing ) ' set of all 11,662 fits . 2 . scan each day @xmath104 from @xmath102 to @xmath101 . there will be @xmath105 days to scan . for each scan day , create a new set @xmath106 consisting of _ all _ fits in subset @xmath103 that have a @xmath0 near the scan day @xmath104 , where ` near ' is defined using the same criterion used for defining the two classes , namely @xmath77 days : @xmath107 the number @xmath108 of fits in each set can be @xmath54 or greater . the sum of the number of fits found in all of the sets @xmath109 can actually be greater than the total number of fits in @xmath103 since some fits can be in multiple sets . notice that the fits in each set @xmath106 can ( and do ) have varying window lengths . at this point , only the proximity to a scan day is used to determine inclusion in a scan set . 3 . assign a group to each of the fits in @xmath110 . recall that groups are defined in eq . ( [ eq : groups ] ) and are based on the window length @xmath111 . 4 . using all sets @xmath110 , for each informative parameter @xmath87 found in sec . [ sec : ip ] , determine if it belongs to class i ( close to a rebound ) or class ii ( not close to a rebound ) . there are 3 possible answers : 1 = ` belongs to class i ' , -1 = ` belongs to class ii ' or 0 = ` undetermined ' . the status of ` belonging to class i ' or not is determined as follows . first , find all values of the informative parameter @xmath87 in a particular scan set @xmath106 . for instance , if for a particular scan day @xmath104 , there are @xmath112 fits in the subset @xmath103 that have @xmath0 ` near ' @xmath104 , then the set @xmath106 contains those @xmath112 fits . these @xmath112 fits include windows of varying lengths so that the windows themselves are likely associated with different groups . now consider a given informative parameter @xmath87 and its underlying parameter @xmath92 ( described in sec . [ sec : ip ] ) that has an associated ` good region ' , @xmath98 . remember that this informative parameter @xmath87 has an associated group . count the number @xmath2 of the @xmath112 fits whose lengths belong to the associated group of @xmath87 . if more of the values of the underlying parameter of @xmath2 lie within @xmath98 than outside of it , then @xmath87 belongs to class i and , thus , the ` answer ' to the question of ` belonging to class i ' is @xmath113 . if , on the other hand , more values lie outside the ` good region ' @xmath98 than in it , the answer is @xmath114 . if the same number of values are inside and outside of @xmath98 then @xmath115 . also , if no members of @xmath106 belong to the associated group of @xmath87 then @xmath115 . to assist more in that understanding , let us have a look at an example . assume that the informative parameter information tells us parameter @xmath53 in group 3 is the informative parameter @xmath116 and @xmath117 $ ] is the ` good region ' @xmath98 of class i. we consider a single @xmath104 and find that there are two fits in @xmath106 in this group with parameter @xmath53 values of @xmath118 and @xmath119 . we determine the ` answer ' @xmath120 as follows : * if @xmath121 $ ] , we say that based on @xmath116 ( group 3 , parameter @xmath53 ) that fits near @xmath104 belong to class i. mark this answer as @xmath122 . * if @xmath123 $ ] and @xmath124 $ ] , we say that fits near @xmath104 can not be identified and so @xmath125 . * if @xmath126 $ ] , fits near @xmath104 belong to class ii and @xmath127 . more succinctly , @xmath128 $ } \\ 0 & \mbox{if $ m_i \in [ a , b ] , m_j \notin [ a , b ] , i \neq j , i , j \in \{1,2\}$}\\ -1 & \mbox{if $ m_1,m_2 \notin [ a , b]$ } \end{array } \right.\ ] ] for each of the informative parameters , we get an answer @xmath129 that says that fits near @xmath104 belong to class i or ii ( or can not be determined ) . for a total of @xmath95 informative parameters , we get a questionnaire @xmath130 of length @xmath95 : @xmath131 qualitatively , these questionnaires describe our judgement to whether @xmath104 is a rebound or not . this judgement depends on the observations of informative parameters . the concept of a _ trait _ is developed to describe the property of the questionnaire for each @xmath104 . each questionnaire can be decomposed into a fixed number of traits if the length of questionnaire is fixed . from any questionnaire with length @xmath95 , we generate a series of traits by the following method . every trait is a series of 4 to 6 integers , @xmath132 . the first three terms @xmath133 and @xmath134 are simply integers . the term @xmath135 represents a string of 1 to 3 integers . we first describe @xmath133 and @xmath134 and then the @xmath135 term . the integers @xmath133 and @xmath134 have limits : @xmath136 . we select all the possible combinations of bits from the questionnaire @xmath137 with the condition that each time the number of selected questions is at most 3 . we record the numbers of the selected positions and sort them . the terms @xmath138 and @xmath134 are selected position numbers and defined as follows : * if only one position @xmath139 is selected : @xmath140 * if two @xmath141 are selected : @xmath142 * if three @xmath143 are selected : @xmath144 the term @xmath135 is defined as follows : @xmath145 as an example , a = ( 0,1,-1,-1 ) has traits in table [ tb : trait ] . for a questionnaire with length l , there are @xmath146 possible traits . however , a single questionnaire has only @xmath147 traits , because ( p , q , r ) is defined by p , q and r. in this example , there are 14 traits for questionnaire ( 0,1,-1,-1 ) and 174 total traits for all possible @xmath148 questionnaires . at the risk of being redundant , it is worth briefly summarizing again . until now we have : @xmath95 informative parameters @xmath149 from 84 different parameters ( @xmath150 ) and a series of questionnaires @xmath137 for each @xmath104 from @xmath151 to @xmath152 using set @xmath106 . these questionnaires depend upon which subset @xmath103 of fits is chosen . each questionnaire has a sequence of traits that describe the property of this questionnaire in a short and clear way . now we generate _ features _ for both classes . recall that the subset of fits @xmath153 that we use here is that which contains all fits which have a critical time @xmath0 earlier than @xmath154 1975 - 01 - 01 , @xmath155 . by imposing that @xmath45 and @xmath73 are both smaller than @xmath156 , we do not use any future information . considering the boundary condition of critical times in eq . ( [ eq : rangetc ] ) , the end time of a certain fit @xmath45 is less than or equal to @xmath0 . additionally , we select only those critical times such that @xmath157 . assume that there are two sets of traits @xmath158 and @xmath159 corresponding to class i and class ii , respectively . scan day by day the date @xmath11 from the smallest @xmath0 in @xmath153 until @xmath156 . if @xmath11 is near a rebound ( using the same @xmath77 day criterion as before ) , then all traits generated by questionnaire @xmath160 belong to @xmath158 . otherwise , all traits generated by @xmath160 belong to @xmath159 . count the frequencies of a single trait @xmath161 in @xmath158 and @xmath159 . if @xmath161 is in @xmath158 for more than @xmath162 times and in @xmath159 for less than @xmath163 times , then we call this trait @xmath161 a _ feature _ @xmath164 of class i. similarly , if @xmath161 is in @xmath165 for less than @xmath162 times and in @xmath159 for more than @xmath163 times , then we call @xmath161 a _ feature _ @xmath166 of class ii . the pair @xmath167 is defined as a _ feature qualification_. we will vary this qualification to optimize the back tests and predictions . the final piece in our methodology is to define a _ rebound alarm index _ that will be used in the forward testing to ` predict ' rebounds . two types of rebound alarm index are developed . one is for the back tests before 1975 - 01 - 01 , as we have already used the information before this time to generate informative parameters and features . the other alarm index is for the prediction tests . we generate this prediction rebound alarm index using only the information before a certain time and then try to predict rebounds in the ` future ' beyond that time . recall that a _ feature _ is a trait which frequently appears in one class but rarely in the other class . features are associated with feature qualification pairs @xmath167 . using all the fits from subset @xmath153 found in sec . [ sec : features ] , we generate the questionnaires for each day in the learning set , i.e. , the fits with @xmath0 before 1975 - 01 - 01 . take all traits from the questionnaire @xmath160 for a particular day @xmath11 and compare them with features @xmath164 and @xmath166 . the number of traits in @xmath164 and @xmath166 are called @xmath168 and @xmath169 . then we define : @xmath170 from the definition , we can see that @xmath171 $ ] . if @xmath172 is high , then we expect that this day has a high probability that the rebound will start . we choose feature qualification pair ( 10 , 200 ) here , meaning that a certain trait must appear in trait class i at least 11 times _ and _ must appear in trait class ii less than 200 times . if so , then we say that this trait is _ a feature of class i_. if , on the other hand , the trait appears 10 times or less in class i _ or _ appears 200 times or more in class ii , then this trait is _ a feature of class ii_. the result of this feature qualification is shown in figure [ fg : f2212 ] . note that the choice ( 10 , 200 ) is somewhat arbitrary and does not constitute an in - sample optimization on our part . this can be checked from the error diagrams presented below , which scan these numbers : one can observe in particular that the pair ( 10 , 200 ) does not give the best performance . we have also investigated the impact of changing other parameters and find a strong robustness . with this feature qualification , the rebound alarm index can distinguish rebounds with high significance . if the first number @xmath162 is too big and the second number @xmath163 is too small , then the total number of class i features will be very small and the number of features in class ii will be large . this makes the rebound alarm index always close to 0 . in contrast , if @xmath162 is too small and @xmath163 is too large , the rebound alarm index will often be close to 1 . neither of these cases , then , is qualified to be a good rebound alarm index to indicate the start of the next rebound . however , the absolute values of feature qualification pair are not very sensitive within a large range . only the ratio @xmath173 plays an important role . figures [ fg:22p ] - [ fg:33b ] show that varying @xmath162 and @xmath163 in the intervals @xmath174 and @xmath175 does not change the result much . for the sake of conciseness , only the rebound alarm index of feature qualification pair ( 10 , 200 ) is shown in this paper . once we generate the class i and ii features of the learning set for values of @xmath0 before @xmath156 ( jan . 1 , 1975 ) , we then use these features to generate the predictions on the data after @xmath156 . recall that the windows that we fit are defined such that the end time @xmath45 increases 50 days from one window to the next . also note that all predictions made on days between these 50 days will be the same because there is no new fit information between , say , @xmath176 and @xmath177 . assume that we make a prediction at time @xmath11 : @xmath178 , ~~t > t_p\ ] ] then the fits set @xmath179 is made using the past information before prediction day @xmath11 . we use @xmath180 as the subset @xmath103 mentioned in sec . [ sub : question ] to generate the questionnaire on day @xmath11 and the traits for this questionnaire . comparing these traits with features @xmath164 and @xmath166 allows us to generate a rebound alarm index @xmath172 using the same method as described in sec . [ sec : features - learning ] . using this technique , the prediction day @xmath45 is scanned from 1975 - 01 - 01 until 2009 - 07 - 22 in steps of 50 days . we then construct the time series of the _ rebound alarm index _ over this period and with this resolution of 50 days . the comparison of this rebound alarm index with the historical financial index ( figure [ fg : p2212 ] ) shows a good correlation , but there are also some false positive alarms ( 1977 , 1998 , 2006 ) , as well as some false negative missed rebounds ( 1990 ) . many false positive alarms such as in 1998 and 2006 are actually associated with rebounds . but these rebounds have smaller amplitudes than our qualifying threshold targets . concerning the false negative ( missed rebound ) in 1990 , the explanation is probably that the historical prices preceeding this rebound does not follow the jls model specification . rebounds may result from several mechanisms and the jls model only provides one of them , arguably the most important . overall , the predictability of the rebound alarm index shown in figure [ fg : p2212 ] , as well as the relative cost of the two types of errors ( false positives and false negatives ) can be quantified systematically , as explained in the following sections . the major conclusion is that the rebound alarm index has a prediction skill much better than luck , as quantified by error diagrams . we have qualitatively seen that the feature qualifications method using back testing and forward prediction can generate a rebound alarm index that seems to detect and predict well observed rebounds in the s&p 500 index . we now quantify the quality of these predictions with the use of error diagrams @xcite . we create an error diagram for predictions after 1975 - 01 - 01 with a certain feature qualification in the following way : 1 . count the number of rebounds after 1975 - 01 - 01 as defined in section [ defreboundh2ysec ] and expression ( [ defreboundh2y ] ) . there are 9 rebounds . 2 . take the rebound alarm index time series ( after 1975 - 01 - 01 ) and sort the set of all alarm index values in decreasing order . there are 12,600 points in this series and the sorting operation delivers a list of 12,600 index values , from the largest to the smallest one . the largest value of this sorted series defines the first threshold . 4 . using this threshold , we declare that an alarm starts on the first day that the unsorted rebound alarm index time series exceeds this threshold . the duration of this alarm @xmath181 is set to 41 days , since the longest distance between a rebound and the day with index greater than the threshold is 20 days . then , a prediction is deemed successful when a rebound falls inside that window of 41 days . if there are no successful predictions at this threshold , move the threshold down to the next value in the sorted series of alarm index . once a rebound is predicted with a new value of the threshold , count the ratio of unpredicted rebounds ( unpredicted rebounds / total rebounds in set ) and the ratio of alarms used ( duration of alarm period / 12,600 prediction days ) . mark this as a single point in the error diagram . in this way , we will mark 9 points in the error diagram for the 9 rebounds . the aim of using such an error diagram in general is to show that a given prediction scheme performs better than random . a random prediction follows the line @xmath182 in the error diagram . a set of points below this line indicates that the prediction is better than randomly choosing alarms . the prediction is seen to improve as more error diagram points are found near the origin ( 0 , 0 ) . the advantage of error diagrams is to avoid discussing how different observers would rate the quality of predictions in terms of the relative importance of avoiding the occurrence of false positive alarms and of false negative missed rebounds . by presenting the full error diagram , we thus sample all possible preferences and the unique criterion is that the error diagram curve be shown to be statistically significantly below the anti - diagonal @xmath182 . in figure [ fg:22p ] , we show error diagrams for different feature qualification pairs @xmath167 . note the 9 points representing the 9 rebounds in the prediction set . we also plot the 11 points of the error diagrams for the learning set in figure [ fg:22b ] . as a different test of the quality of this pattern recognition procedure , we repeated the entire process but with a rebound now defined as the minimum price within a window of @xmath183 days instead of @xmath184 days , as before . these results are shown in figures [ fg:33p]-[fg:33b ] . given a value of the _ predictive _ rebound alarm index , we can also use the _ historical _ rebound alarm index combined with bayesian inference to calculate the probability that this value of the rebound alarm index will actually be followed by a rebound . we use predictions near the end of november , 2008 as an example . from figure [ fg : p2212 ] , we can see there is a strong rebound signal in that period . we determine if this is a true rebound signal by the following method : 1 . find the highest rebound alarm index @xmath185 around the end of november 2008 . calculate @xmath186 , the number of days in the interval from 1975 - 01 - 01 until the end of the prediction set , 2009 - 07 - 22 . calculate @xmath187 , the number of days which have a rebound alarm index greater than or equal to @xmath185 . the probability that the rebound alarm index is higher than @xmath185 is estimated by @xmath188 5 . the probability of a day being near the bottom of a rebound is estimated as the number of days near real rebounds over the total number of days in the predicting set : @xmath189 where @xmath190 is the number of rebounds we can detect after 1975 - 01 - 01 and @xmath191 is the rebound width , i.e. the number of days near the real rebound in which we can say that this is a successful prediction . for example , if we say that the prediction is good when the predicted rebound time and real rebound time are within 10 days of each other , then the rebound width @xmath192 . the probability that the neighbor of a rebound has a rebound alarm index larger than @xmath185 is estimated as @xmath193 where @xmath194 is the number of rebounds in which @xmath195 7 . given that the rebound alarm index is higher than @xmath185 , the probability that the rebound will happen in this period is given by bayesian inference : @xmath196 averaging @xmath197 for all the different feature qualifications gives the probability that the end of november 2008 is a rebound as 0.044 . by comparing with observations , we see that this period is not a rebound . we obtain a similar result by increasing the definition of rebound from 200 days before and after a local minimum to 365 days , yielding a probability of 0.060 . when we _ decrease _ the definition to 100 days , the probability that this period is a rebound jumps to 0.597 . the reason for this sudden jump is shown in figure [ fg : bn08 ] where we see the index around this period and the s&p 500 index value . from the figure , we find that this period is a local minimum within 100 days , not more . this is consistent with what bayesian inference tells us . however , we have to address that the more obvious rebound in march 2009 is missing in our rebound alarm index . technically , one can easily find that this is because the end of crash is not consistent with the beginning of rebound in this special period . in this case , we then test all the days after 1985 - 01 - 01 systematically by bayesian inference using only prediction data ( rebound alarm index ) after 1975 - 01 - 01 . to show that the probability that @xmath198 is stable , we can not start bayesian inference too close to the initial predictions so we choose 1985 - 01 - 01 as the beginning time . we have 5 ` bottoms ' ( troughs ) after this date , using the definition of a minimum within @xmath68 days . for a given day @xmath65 after 1985 - 01 - 01 , we know all values of the rebound alarm index from 1975 - 01 - 01 to that day . then we use this index and historical data of the asset price time series in this time range to calculate the probability that @xmath65 is the bottom of the trough , given that the rebound alarm index is larger than @xmath185 , where @xmath185 is defined as @xmath199 to simplify the test , we only consider the case of feature qualification pair ( 10 , 200 ) , meaning that the trait is a feature of class i only if it shows in class i more than 10 times and in class ii less than 200 times . figure [ fg : bayes ] shows that the actual rebounds occur near the local highest probability of rebound calculated by bayesian inference . this figure also illustrates the existence of false positive alarms , i.e. , large peaks of the probability not associated with rebounds that we have characterized unambiguously at the time scale of @xmath68 days . in order to determine if the predictive power of our method provides a genuine and useful information gain , it is necessary to estimate the excess return it could generate . the excess return is the real return minus the risk free rate transformed from annualized to the duration of this period . the annualized 3-month us treasury bill rate is used as the risk free rate in this paper . we thus develop a trading strategy based on the rebound alarm index as follows . when the rebound alarm index rises higher than a threshold value @xmath200 , then with a lag of @xmath201 days , we buy the asset . this entry strategy is complemented by the following exit strategy . when the rebound alarm index goes below @xmath200 , we still hold the stock for another @xmath202 days , with one exception . consider the case that the rebound alarm index goes below @xmath200 at time @xmath42 and then rises above @xmath200 again at time @xmath45 . if @xmath203 is smaller than the holding period @xmath202 , then we continue to hold the stock until the next time when the rebound alarm index remains below @xmath200 for @xmath202 days . the performance of this strategy for some fixed values of the parameters is compared with random strategies , which share all the properties except for the timing of entries and exits determined by the rebound alarm index and the above rules . the random strategies consist in buying and selling at random times , with the constraint that the total holding period ( sum of the holding days over all trades in a given strategy ) is the same as in the realized strategy that we test . implementing 1000 times these constrained random strategies with different random number realizations provide the confidence intervals to assess whether the performance of our strategy can be attributed to real skill or just to luck . results of this comparison are shown in table [ tb : performance ] for two sets of parameter values . the p - value is a measure of the strategies performance , calculated as the fraction of corresponding random strategies that are better than or equal to our strategies . the lower the p - value is , the better the strategy is compared to the random portfolios . we see that all of our strategies cumulative excess returns are among the top 5 - 6% out of 1000 corresponding random strategies cumulative excess returns . box plots for each of the strategies are also presented in figures [ fg : bp1]-[fg : bp2 ] . the cumulative returns as well as the cumulative excess returns obtained with the two strategies as a function of time are shown in figures [ fg : wt1]-[fg : wt2 ] . these results suggest that these two strategies would provide significant positive excess return . of course , the performance obtained here are smaller than the naive buy - and - hold strategy , consisting in buying at the beginning of the period and just holding the position . the comparison with the buy - and - hold strategy would be however unfair as our strategy is quite seldom invested in the market . our goal here is not to do better than any other strategy but to determine the statistical significance of a specific signal . for this , the correct method is to compare with random strategies that are invested in the market the same fraction of time . it is obvious that we could improve the performance of our strategy by combining the alarm indexes of bubbles and of negative bubbles , for instance , but this is not the goal here . we also provide the sharpe ratio as a measure of the excess return ( or risk premium ) per unit of risk . we define it per trade as follows @xmath204}{\sigma}\ ] ] where @xmath205 is the return of a trade , @xmath206 is the risk free rate ( we use the 3-month us treasury bill rate ) transformed from annualized to the duration of this trade given in table [ tb : performance ] and @xmath207 is the standard deviation of the returns per trade . the higher the sharpe ratio is , the higher the excess return under the same risk . the bias ratio is defined as the number of trades with a positive return within one standard deviation divided by one plus the number of trades which have a negative return within one standard deviation : @xmath208\}}{1 + \ # \{r | r \in [ -\sigma , 0)\ } } \label{eq : br}\ ] ] in eq . ( [ eq : br ] ) , @xmath134 is the excess return of a trade and @xmath207 is the standard deviation of the excess returns . this ratio detects valuation bias . to see the performance of our strategies , we also check all the possible random trades with a holding period equals to the average duration of our strategies , namely 25 days and 17 days for strategy i and ii respectively . the average sharpe and bias ratios of these random trades are shown in table [ tb : performance ] . both sharpe and bias ratios of our strategies are greater than those of the random trades , confirming that our strategies deliver a larger excess return with a stronger asymmetry towards positive versus negative returns . as another test , we select randomly the same number of random trades as in our strategies , making sure that there is no overlap between the selected trades . we calculate the sharpe and bias ratios for these random trades . repeating this random comparative selection 1000 times provides us with p - values for the sharpe ratio and for bias ratio of our strategies . the results are presented in table [ tb : performance ] . all the p - values are found quite small , confirming that our strategies perform well . we have developed a systematic method to detect rebounds in financial markets using `` negative bubbles , '' defined as the symmetric of bubbles with respect to a horizontal line , i.e. , downward accelerated price drops . the aggregation of thousands of calibrations in running windows of the negative bubble model on financial data has been performed using a general pattern recognition method , leading to the calculation of a rebound alarm index . performance metrics have been presented in the form of error diagrams , of bayesian inference to determine the probability of rebounds and of trading strategies derived from the rebound alarm index dynamics . these different measures suggest that the rebound alarm index provides genuine information and suggest predictive ability . the implemented trading strategies outperform randomly chosen portfolios constructed with the same statistical characteristics . this suggests that financial markets may be characterized by transient positive feedbacks leading to accelerated drawdowns , which develop similarly to but as mirror images of upward accelerating bubbles . our key result is that these negative bubbles have been shown to be predictably associated with large rebounds or rallies . in summary , we have expanded the evidence for the possibility to diagnose bubbles before they terminate @xcite , by adding the phenomenology and modeling of `` negative bubbles '' and their anticipatory relationship with rebounds . the present paper contributes to improving our understanding of the most dramatic anomalies exhibited by financial markets in the form of extraordinary deviations from fundamental prices ( both upward and downward ) and of extreme crashes and rallies . our results suggest a common underlying origin to both positive and negative bubbles in the form of transient positive feedbacks leading to identifiable and reproducible faster - than - exponential price signatures . .[tb : symbol ] list of symbols [ cols="^ , < " , ] 49 natexlab#1#1url # 1`#1`urlprefix blanchard , o. , watson , m. , 1982 . bubbles , rational expectations and speculative markets . in : wachtel , , crisis in economic and financial structure : bubbles , bursts , and shocks . lexington books : lexington . gelfand , i. , guberman , s. , keilis - borok , v. , knopoff , l. , press , f. , e.ya.ranzman , rotwain , i. , sadovsky , a. , 1976 . pattern recognition applied to earthquake epicenters in california . physics of the earth and planetary interiors 11 ( 3 ) , 227283 . jiang , z .- q . , zhou , w .- x . , sornette , d. , woodard , r. , bastiaensen , k. , cauwels , p. , 2010 . bubble diagnosis and prediction of the 2005 - 2007 and 2008 - 2009 chinese stock market bubbles . journal of economic behavior and organization 74 , 149162 . sornette , d. , woodard , r. , fedorovsky , m. , reimann , s. , woodard , h. , zhou , w .- x . , 2010 . the financial bubble experiment : advanced diagnostics and forecasts of bubble terminations ( the financial crisis observatory ) . sornette , d. , woodard , r. , fedorovsky , m. , reimann , s. , woodard , h. , zhou , w .- x . , 2010 . the financial bubble experiment : advanced diagnostics and forecasts of bubble terminations volume ii master document . http://arxiv.org/abs/1005.5675 . from 1973 - 01 - 01 to 1974 - 10 - 01 ( time window delineated by the two black dashed vertical lines ) with very clear log - periodic oscillations , followed by a strong positive rebound . the best fits from taboo search are used to form a 90% confidence interval for the critical time @xmath0 shown by the light shadow area . the dark shadow area corresponds to the 20 - 80 quantiles region of the predicted rebounds . lower panel : the same phenomenon is observed in foreign exchange market . the plot shows the fitted results for usd / eur change rate from 2006 - 07 - 01 to 2008 - 04 - 01 . the usd / eur change rate performed a significant drawdown with very clear log - periodic oscillations , followed by a strong positive rebound.,title="fig:",scaledwidth=90.0% ] from 1973 - 01 - 01 to 1974 - 10 - 01 ( time window delineated by the two black dashed vertical lines ) with very clear log - periodic oscillations , followed by a strong positive rebound . the best fits from taboo search are used to form a 90% confidence interval for the critical time @xmath0 shown by the light shadow area . the dark shadow area corresponds to the 20 - 80 quantiles region of the predicted rebounds . lower panel : the same phenomenon is observed in foreign exchange market . the plot shows the fitted results for usd / eur change rate from 2006 - 07 - 01 to 2008 - 04 - 01 . the usd / eur change rate performed a significant drawdown with very clear log - periodic oscillations , followed by a strong positive rebound.,title="fig:",scaledwidth=90.0% ] over the set of 2,568 time intervals for which negative bubbles are detected by the condition that the fits of @xmath209 by expression ( [ eq : lppl ] ) satisfy condition ( [ eq : rb ] ) . ( lower ) plot of @xmath210 versus time for the s&p 500 index . note that peaks in this figure correspond to valleys in actual price . ] and @xmath0 are both before jan . 1 , 1975 . ( upper ) rebound alarm index for the learning set using feature qualification pair @xmath211 . the rebound alarm index is in the range @xmath212 $ ] . the higher the rebound alarm index , the more likely is the occurrence of a rebound . ( lower ) plot of @xmath209 versus time of s&p index . red vertical lines indicate rebounds defined by local minima within plus and minus 200 days around them . note that these rebounds are the historical `` change of regime '' rather than only the jump - like reversals . the jump - like reversals , 1972 , 1974 as examples , are included in these rebounds . they are located near clusters of high values of the rebound alarm index of the upper figure . ] . the rebound alarm index is in the range [ 0,1 ] . the higher the rebound alarm index , the more likely is the occurrence of a rebound . ( lower ) plot of @xmath209 versus time of the s&p index . red vertical lines indicate rebounds defined by local minima within in plus and minus 200 days . they are located near clusters of high values of the rebound alarm index of the upper figure . ] means that , if the occurrence of a certain trait in class i is larger than @xmath162 and less than @xmath163 , then we call this trait a feature of class i and vice versa . see text for more information . ] , given the value of the rebound alarm index at @xmath11 , derived by bayesian inference applied to bottoms at the time scale of @xmath68 days . the feature qualification is ( 10 , 200 ) . @xmath185 is the largest rebound index in the past 50 days . the vertical red lines show the locations of the realized rebounds in the history of the s&p500 index . ] ) . lower and upper horizontal edges ( blue lines ) of box represent the first and third quartiles . the red line in the middle is the median . the lower and upper black lines are the 1.5 interquartile range away from quartiles . points between quartiles and black lines are outliers and points out of black lines are extreme outliers . our strategy return is marked by the red circle . this shows our strategy is an outlier among the set of random strategies . the log - return ranked 55 out of 1000 random strategies.,scaledwidth=90.0% ] ) . major performance parameters of this strategy are : 77 trading times ; 66.2% trades have positive return ; 1894 total holding days , which is 15.0% of total time . accumulated log - return is 95% and average return per trade is 1.23% . average trade length is 24.60 days . p - value of this strategy is 0.055,scaledwidth=90.0% ] ) . major performance parameters of this strategy are : 38 trading times ; 65.8% trades have positive return ; 656 total holding days , which is 5.2% of total time . accumulated log - return is 45% and average return per trade is 1.19% . average trade length is 17.26 days . p - value of this strategy is 0.058,scaledwidth=90.0% ]

we report the use of broadband heterodyne spectroscopy to perform continuous measurement of the interaction energy between one atom and a high - finesse optical cavity , during individual transit events ofs duration . we achieve a fractional sensitivity to variations in within a measurement bandwidth that covers 2.5 decades of frequency ( 1300 khz ) . our basic procedure is to drop cold cesium atoms into the cavity from a magnetooptic trap while monitoring the cavity s complex optical susceptibility with a weak probe laser . the instantaneous value of the atom - cavity interaction energy , which in turn determines the coupled system s optical susceptibility , depends on both the atomic position and ( zeeman ) internal state . measurements over a wide range of atom - cavity detunings reveal the transition from resonant to dispersive coupling , via the transfer of atom - induced signals from the amplitude to the phase of light transmitted through the cavity . by suppressing all sources of excess technical noise , we approach a measurement regime in which the broadband photocurrent may be interpreted as a classical record of _ conditional _ quantum evolution in the sense of recently - developed quantum trajectory theories . = 10000 optical cavity quantum electrodynamics ( qed ) in the strong coupling regime provides a unique experimental paradigm for _ real - time _ observation of quantum dynamical processes at the _ single - atom _ level . while spectacular advances have certainly been made in the preparation and tomography of quantum states of motion for a single trapped ion , all such experiments have involved the accumulation of ensemble - averaged data over many successive realizations of the process being studied . recent studies of single - molecule dynamics have likewise demonstrated the `` immediate '' detection of photochemical or conformational events , but such experiments presently lack the potential that cavity qed provides for observing quantum processes on a timescale that makes coherent control / intervention a tangible possibility . we wish here to look beyond the mere detection of quantum jumps , and to focus on the development of a broadband , _ single - shot _ measurement technique that achieves signal - to - noise ratio over a bandwidth that includes all characterstic frequencies of a quantum dynamical process . real - time observation of quantum dynamics in _ many - atom _ systems has recently become an important theme in atomic physics , with notable demonstrations involving vibrational excitations of a trapped bose - einstein condensate and the decay of coherent oscillations of an ensemble of atoms in an optical lattice . in contrast to programs like these , for which the scientific emphasis lies on noninvasive observation of a system s intrinsic dynamical processes , experiments in single - atom cavity qed hold great potential for enabling precise investigations of how measurement backaction _ alters _ the dynamical behavior of a continuously - observed open quantum system . a sophisticated theoretical basis for understanding such issues is presently maturing in the form of quantum trajectory theories , but significant technical challenges remain to be solved before definitive experiments can be performed in the lab . our purpose in the present work is to report substantial progress towards surmounting such obstacles in the context of cavity qed , and hence towards achieving the essential experimental capabilities required to perform quantitative tests of measurement - based stochastic master equations . we ultimately hope to be able to implement some recently - proposed `` applications '' of the continuous observation of dissipative quantum dynamics , in fields such as quantum measurement , quantum chaos , and quantum feedback control . this article focuses on a detailed description of our recent experiments that record the complete time - evolution of interaction energy between one atom and a high - finesse optical cavity , during individual transit events ofs duration . with characteristic atom - cavity interaction energies mhz , we achieve measurement sensitivities khz over a bandwidth that covers the dominant rates of variation in ( 1300 khz ) . unlike typical pump - probe measurements of scattering dynamics in real ( _ e.g. _ diatomic ) molecular systems , our experiments on the jaynes - cummings `` molecule '' yield a continuous time - domain record of the atom - cavity coupling during each individual `` scattering '' event ( transit ) . the data clearly illustrate variations caused by atomic motion through the spatial structure of the cavity eigenmode and/or optical pumping among the atomic internal ( zeeman ) states . in certain parameter regimes of the detuning and probe power , distinctive indications of the quantum - mechanical nature of the atom - cavity coupling can be seen in the photocurrent recorded from just a _ single _ atomic transit . for large ( mhz ) atom - cavity detunings we obtain phase - contrast signals induced by individual atomic transits , corresponding to a regime of strong but dispersive coupling . phase - quadrature measurements of atomic motion have been widely discussed in the quantum optics literature , but the present work provides the first experimental demonstration at the single - atom level . because of the standing - wave spatial structure of the cavity eigenmode , and the corresponding rapid varation of the atom - cavity coupling strength over sub - wavelength distances , our data should in principle display a _ sensitivity _ of m/ to atomic displacements along the cavity axis . unfortunately we can not realize this figure as a _ precision _ for monitoring the atomic position , as we do not presently have any means of separating signal variations due to motion through the standing wave from `` background '' contributions due to transverse motion or optical pumping . in our concluding section , we shall briefly discuss our motivations for further work to disambiguate the nature of rapid variations in our data .

quant-ph9805076

in simple terms , our experimental procedure is to drop a cloud of cold cesium atoms from a magnetooptic trap ( mot ) into a high - finesse optical cavity , while continuously monitoring the cavity s complex susceptibility with a weak probe laser @xcite . by limiting the number of atoms in the initial cloud , we can easily reach an operating regime in which atoms transit the cavity only one at a time . using broadband heterodyne detection and a high - speed digitizer , we continuously record both the amplitude and phase of the transmitted probe beam during @xmath12 ms time windows . each window typically contains from zero to five atom transit signals . the elementary theoretical description of such a measurement employs steady - state solutions of the nonselective master equation for a _ stationary _ two - level atom coupled to a single electromagnetic mode via the jaynes - cummings interaction hamiltonian . atomic center - of - mass motion and optical pumping among zeeman states can only be included in this treatment by allowing for a time - dependent atom - cavity coupling strength . although this type of approach can not make predictions about dynamical variations in the coupling strength , it does provides a quantitative basis for interpreting some time - independent features of our data . our use of this `` adiabatic '' model may be justified to a certain extent by the separation of timescales that we achieve in optical cavity qed with laser - cooled atoms . in optimal cases , the atom - cavity coupling strength should vary by as little as a factor of @xmath13 over the system damping time @xmath14 ns . if we let @xmath15 denote the density operator for the joint state of the atom and cavity , the nonselective master equation ( in the electric dipole and rotating - wave approximations ) reads : @xmath16 + \gamma_\perp\left(2{\hat\sigma}\rho { \hat\sigma}^\dagger - { \hat\sigma}^\dagger{\hat\sigma}\rho - \rho{\hat\sigma}^\dagger { \hat\sigma}\right)\nonumber\\ & & + \left(\kappa_a+\kappa_b+\kappa_c\right)\left(2{\hat a}\rho{\hat a}^\dagger - { \hat a}^\dagger { \hat a}\rho - { \hat a}^\dagger{\hat a}\rho\right),\label{eq : me}\\ { \hat h}_0&=&\hbar\delta{\hat a}^\dagger{\hat a } + \hbar\theta{\hat\sigma}^\dagger{\hat\sigma } + \sqrt{2\kappa_a}{\cal e}\left({\hat a}+ { \hat a^\dagger}\right ) + h_{\rm int},\\ h_{\rm int}&=&\hbar g_0 e^{-\left(y^2+z^2\right)/w^2 } \cos\left(k_lx\right)\left[{\hat a}{\hat\sigma}^\dagger+{\hat a}^\dagger { \hat\sigma}\right].\end{aligned}\ ] ] here @xmath17 is the atomic dipole decay rate , @xmath18 is the cavity field decay rate through the input mirror ( through which the probe laser is injected ) , @xmath19 is the cavity field decay rate through the output mirror , @xmath20 is the cavity field decay rate due to intracavity scattering / absorption losses , @xmath21 is the atom - probe detuning , @xmath22 is the cavity - probe detuning , and the coupling strength @xmath23 is equal to half the maximum single - photon rabi frequency . we treat the atomic center - of - mass coordinates @xmath24 as c - number parameters , with the @xmath25-axis conciding with the cavity axis and @xmath26 parallel to gravity . the gaussian waist of our cavity mode is @xmath27 @xmath2 m . note that we have written the master equation in a frame rotating at the drive frequency , so @xmath28 is a constant term proportional to the complex amplitude of the driving field . to find the steady - state density operator as a function of driving strength and various detunings , we simply set @xmath29 and solve for @xmath30 using linear algebra . the expected amplitude and phase of the heterodyne photocurrent may then be computed as @xcite @xmath31{\rm tr}\left[\rho_{ss}{\hat a}\right ] , \label{eq : hetphot}\ ] ] where @xmath32 represents the overall photodetection efficiency ( including propagation losses between the cavity and photodetectors , heterodyne efficiency , and detector quantum efficiency ) , @xmath33 and @xmath34 represent the photon flux and phase of the ( optical ) local oscillator , and @xmath35 is the frequency of the optical local oscillator relative to the rotating frame ( cavity driving field ) . in the experiment , we mix @xmath36 with an rf local oscillator at the frequency @xmath35 ( which ranges between 40 and 190 mhz ) and separately record the in - phase and quadrature components of the slowly - varying envelope ( with an analog bandwidth of 300 khz ) . note that the presence of an intracavity atom can influence the heterodyne photocurrent only via the interaction hamiltonian @xmath37 , where @xmath38 in a two - level approximation for the atomic internal dynamics , and for a classical treatment of the atomic center - of - mass degrees of freedom , _ all _ steady - state properties of the atom - cavity system are strictly determined by the value of @xmath39 once the parameters ( @xmath18,@xmath19,@xmath17,@xmath40,@xmath41,@xmath28 ) have been specified . this includes the quantity @xmath42 , which represents one possible measure of the `` interaction energy '' between atom and cavity . given that @xmath39 appears to be a more fundamental measure of the interaction strength , however , we have adopted the convention @xmath43 . the atom - cavity interaction can be treated semiclassically using the optical bistability state equation ( obse ) @xcite . the obse is traditionally written in terms of the scaled field variables @xmath25 and @xmath44 ( not to be confused with the atomic coordinates ) , with the correspondence @xmath45 where the saturation photon number @xmath46 is given by @xmath47 . for a given driving strength @xmath28 , the expected intracavity field amplitude can be found by inverting the equation @xmath48 where we again work in a rotating frame at the drive frequency , @xmath49 represents the scaled atom - probe detuning , @xmath50 represents the scaled cavity - probe detuning , and the `` cooperativity '' @xmath51 is defined by @xmath52 note that the dependence of @xmath53 ( and thereby @xmath51 and @xmath46 ) on the atomic position and internal state is implicit . a semiclassical prediction for the heterodyne photocurrent is obtained by substituting @xmath54 for @xmath55 $ ] in equation ( [ eq : hetphot ] ) . figure [ fig : apparat ] provides a general overview of the apparatus , indicating the schematic arrangement of various components to be described below . the diode laser setup for forming the cs mot is not shown . we use a fabry - perot high - finesse microcavity ( `` physics cavity '' ) consisting of two spherical mirrors with 1 m radius of curvature @xcite . the cavity was constructed with a mean length @xmath56 @xmath2 m , which we inferred from the cavity s measured free spectral range of @xmath57 hz . the measured @xmath58 and specified radii of curvature geometrically determine the cavity s electromagnetic mode volume for tem@xmath59 modes near 852 nm @xcite . together with the dipole decay rate @xmath60 mhz for the cs 6p@xmath61 level @xcite , this determines our optimal coupling constant @xmath62 to be @xmath63 mhz for @xmath64 transistions ( specifically the @xmath65 ) and @xmath66 mhz for @xmath67 transitions ( @xmath68 ) within the d2 zeeman manifold @xcite . in order to allow cold atoms to fall into such a short cavity , we found it necessary to have the mirror manufacturer reduce the substrate diameters from the standard value of 7.75 mm down to 3 mm @xcite . this reduced the `` sagittal depth '' of the curved mirror substrates and allowed us to maintain a gap of @xmath69 @xmath2 m around the edge of the cavity . machining of the mirror substrates was performed _ after _ they had been superpolished and coated , but this process did not seem to degrade the mirror reflectivities significantly . the nominal combined transmission and loss per mirror , before machining , was @xmath70 . direct measurements of the cavity finesse yield @xmath71 at an optical wavelength of 852.36 nm , consistent with a combined mirror transmission and loss @xmath72 . this value of @xmath73 is inferred from the measured value of @xmath58 and the measured cavity hwhm @xmath74 mhz . the cavity used throughout the work described in this paper suffers from a rather pronounced birefringence , which for tem@xmath59 modes near 852 nm induced a splitting of @xmath75 mhz between linearly - polarized eigenmodes . it is not entirely clear whether this birefrigence is a result of the substrate machining for diameter - reduction , a property of the coatings , or something associated with the mirror - mounting procedure used for this particular cavity . we note that recent efforts by other members of our group @xcite have produced a cavity of approximately the same finesse with greatly - reduced birefringence ( by a factor @xmath76 ) , using mirrors from a different coating run and with great care taken to minimize cavity misalignments and stress on the mirror substrates . our two mirror substrates are mounted in vee - grooves atop independent aluminum blocks , with a piezoelectric actuator between the blocks for active servo - control of the mirror separation ( cavity length ) . the cavity mount sits on a stack of alternating ofhc copper blocks and viton o - rings for passive vibration isolation , all within an ion - pumped vacuum chamber whose background pressure was typically @xmath77 torr ( inferred from the ion pump current ) . in order to bring the mot as close as possible to the central axis of the physics cavity , we had to use a rather open ( and non - magnetic ) mount design , leading to some compromises in the way of mechanical stability . sitting on the vibration - isolation stack and under vacuum , we found that the native noise spectrum of the cavity length extended out to about 4 khz , with one prominent resonance at 50 hz ( which we attribute to a transmission resonance of the isolation stack ) . due to a set of pzt - actuator resonances above 10 khz , we have ultimately been limited to a unity - gain bandwidth @xmath78 khz for the cavity stabilization servo . although the principle aim for this experiment is to stabilize the cavity length at some precise offset @xmath79 mhz from the cs d2 resonance at 852.359 nm , the strong atom - cavity coupling places severe restrictions on the optical power that can be used for the purpose of generating an error signal . on resonance , the saturation intracavity photon number for our cavity is as small as @xmath80 which sets a fiducial cavity throughput of @xmath81 pw . with such low optical power it would be extremely difficult to obtain a high - quality error signal for locking the physics cavity . other experiments performed in our group have circumvented this problem by using a chopped locking scheme , in which a strong `` lock beam '' alternates with a weak `` probe beam '' at 50% duty cycle and @xmath82 khz frequency @xcite . however , such a strategy inherently limits the servo unity - gain bandwidths to @xmath69 hz at best , and would not be appropriate for future experiments with atoms trapped inside the cavity for long periods of time . in this work we have developed an alternative locking scheme for the physics cavity , which employs an auxilliary diode laser at 836 nm to monitor the cavity length on a different longitudinal mode than that which couples strongly to intracavity cs atoms . we use a commercial grating - stabilized diode produced by new focus ( santa clara , california ) . with a 16 nm detuning , we can send @xmath83 nw through the cavity and incur an ac stark shift of only @xmath84 khz for the atomic resonance at @xmath85 . using an eg&g avalanche - photodiode / transimpedance amplifier module ( model c30998 ) for ac detection of transmitted 836 nm light , we obtain an fm error signal ( modulation frequency 3.8 mhz ) with signal - to - noise ratio ( snr ) @xmath86 at 30 khz bandwidth . the 836 nm diode laser is stabilized to an auxilliary `` transfer cavity , '' which consists of a pair of 25 cm radius - of - curvature mirrors at @xmath87 cm separation . one of the mirrors is mounted on a piezoelectric actuator to allow cancellation of dc drift and low - frequency noise . the transfer cavity has a linewidth @xmath88 khz at both 836 nm and 852 nm , with an overall mode spacing @xmath89 mhz . the transfer cavity is also used for pre - stabilization of the ti : sapphire laser , and some of the ti : sapphire light is used in a cs modulation - transfer spectrometer @xcite to provide an absolute reference for the transfer cavity length . from run to run we used one or two acoustooptic modulators to offset the ti : sapphire / transfer - cavity lock point by + 140 , + 87 , or + 43 mhz relative to the cs 6s@xmath90 6p@xmath91 transition . by comparison of the ti : sapphire and diode laser error signals in their respective locks to the transfer cavity , we infer that the relative rms jitter between them is @xmath92 khz . for both laser locks we use the pound - drever - hall technique @xcite of detecting an fm signal in reflection from the tranfser cavity . the stability of the transfer cavity resonances with respect to atomic cs lines in a vapor cell was such that we did not see any relative jitter beyond the measurement noise in our modulation - transfer spectrometer ( snr @xmath93 in 30 khz bandwidth ) . the ti : sapphire stabilization employs two feedback loops , one with @xmath94 khz bandwidth to the tweeter inside the ring laser and another with @xmath69 khz bandwidth to a vco - driven , double - passed acoustooptic modulator ( aom ) just outside the laser cavity @xcite . the mean frequency of the aom is 76 mhz , and the error signal going to the vco has a lower cutoff of @xmath94 khz to prevent dc drifts . we note that use of the aom is crucial for achieving high stability of the ti : sapphire frequency . the diode laser servo utilizes both feedback to the grating pzt and direct modulation of the injection current , achieving an overall unity - gain bandwidth @xmath95 mhz . having locked both the diode laser and ti : sapphire to modes of the transfer cavity , which itself is locked to cs , we use a travelling - wave electrooptic modulator to generate an rf sideband of the diode laser at @xmath96 mhz . either the upper or lower sideband is used to derive an fm error signal for locking the physics cavity by dithering @xmath97 at 3.8 mhz , thus allowing us to achieve arbitrary placement of the physics cavity mode near 852 nm via the tunability of @xmath97 . our basic requirement for the quality of the physics - cavity servo was that relative jitter of the cavity resonance and the probe laser frequency should not contribute a significant amount of noise in the heterodyne photocurrent . hence the relevant comparison to make is between the noise in both quadratures of a demodulated beatnote and the photocurrent fluctuations produced by the local oscillator alone . at 200 khz bandwidth and with a probe beam strength such that @xmath98 in the empty cavity , the standard deviations of the phase and amplitude quadratures of the transmitted probe beam were measured to be 1.01 and 1.39 ( respectively ) relative to those of the local oscillator alone . note that we have estimated the standard deviation of the quadrature - amplitude signals produced by our optical local oscillator to be only a factor of 1.05 above the theoretical shot - noise limit ( see below ) . we therefore believe that our overall excess noise factor @xmath99 . by taking some simultaneous recordings of the heterodyne photocurrent and the physics - cavity error signal , we were able to verify directly that the atomic transits do not affect the physics cavity servo . to provide a source of cold cs atoms , we used a standard magnetooptic trap loaded directly from a thermal beam @xcite . our choice of thermal - beam loading , as opposed to loading from a background vapor , was driven by an attempt to prevent accidental coating of the physics cavity mirrors with cs . in more than two years of service , we did not detect any significant ( @xmath100 ) change in the cavity finesse . pre - cooling of the cs beam was not necessary for this experiment , as we required only a very low rate of delivering single cold atoms into the cavity mode volume . our mot employs a six - beam configuration , and we orient the anti - helmholtz coils for the trap so that their symmetry axis is parallel to that of the optical cavity . this leads to a mot laser beam geometry with one beam axis running parallel to and just above the cavity , plus two beam axes in the plane of the mirror surfaces ( figure [ fig : motconfig ] ) . the light for the mot was provided by a pair of grating - stabilized diode lasers ( sdl 5421-g2 ) , one tuned to the cs 6s@xmath906p@xmath91 cycling transition for trapping and the other to 6s@xmath1016p@xmath102 for repumping . each trapping beam had @xmath95 cm diameter and anywhere from 40 @xmath2w to 4 mw of optical power , depending on how many atoms we were trying to send into the physics cavity . we typically used a cs reservoir temperature of 60 - 80 c for the thermal beam , which effused through a 200 @xmath2 pinhole and travelled an overall distance of @xmath84 cm to the trapping region ( with a cold mechanical collimator in the way to reduce loading of the ion pump ) . with an anti - helmholtz field of around 25 g / cm , we could load up to @xmath103 atoms into a millimeter - sized cloud , whose mean temperature we estimate to be @xmath69 @xmath2k based on fluorescent imaging of free expansion . this temperature estimate is also supported by the spread in arrival times of individual atoms falling into the cavity . when running the experiment we would load the mot for about 0.5 s , then drop it by quickly turning off the trapping beams with an aom ( using an rf switch with @xmath104 db attenuation ) . after the trapping beams were thus extinguished , we would ramp down the anti - helmholtz field according to an rc - filtered step with @xmath105 ms time constant . the repumping beam was left on all the time , so that falling atoms would be shelved in the @xmath106 ground hyperfine level before entering the cavity . no specific preparation was performed with respect to the atomic zeeman states . dropping @xmath103 atoms we would generally see 30 - 50 atoms falling through the central part of the cavity mode volume , so for single - atom transit data we had to reduce the trapping beam power by a factor of @xmath107 to reach 0 - 2 atoms per drop . the overall repitition rate for the trap - drop cycle was typically 0.6 hz . we used a balanced - heterodyne setup in order to achieve high - efficiency , zero - background photodetection of @xmath95 pw levels of 852 nm light transmitted through the physics cavity . the frequency difference between cavity probe light and the optical local oscillator for heterodyne detection was between 40 - 190 mhz , depending upon our choice for the atom - probe detuning . the probe light was generated from the ti : sapphire output by cascading a + 200 mhz aom and a tunable travelling - wave electrooptic modulator , which was driven between -245 mhz and -440 mhz to produce the desired atom - probe detuning . this indirect method was required to prevent contamination of the heterodyne photocurrent by electronic noise at the heterodyne frequency . light leaving the physics cavity first hit a color - separation mirror which reflected @xmath108 of the 852 nm light but transmited @xmath109 of the 836 nm light , allowing us to recover an error signal for locking the physics cavity ( see above ) without compromising the overall detection efficiency for the probe field . residual 836 nm light going to the heterodyne setup amounted to only @xmath14 nw and had negligible effect on the photocurrent of interest . the local oscillator ( lo ) for the optical heterodyne was spatially cleaned by a @xmath110 fabry - perot cavity ( linewidth @xmath95 mhz ) , which also served to strip off spectral noise at 76 mhz associated with the aom servo for stabilization of the ti : sapphire frequency . the cleaning cavity was locked using the pound - drever - hall method @xcite with fm sidebands at 24 mhz , which likewise had to be kept weak in order not to saturate the ac gain of the heterodyne photodetectors . we used a total of @xmath111 mw in the lo , which generated a shot - noise level @xmath112 db above the electronic noise of the photodetectors in the frequency range of interest . the difference photocurrent from the balanced heterodyne detectors was amplified up to -50 dbm or higher , then divided by a @xmath113 rf splitter . an independent signal generator was used to produce an rf local oscillator at the heterodyne frequency , and it was halved using a @xmath114 rf splitter . the two identical copies of the photocurrent were mixed with the in - phase and quadrature copies of the rf lo to produce an orthogonal pair of quadrature amplitude ( qa ) signals at baseband . the qa signals were further amplified , and passed through 300 khz analog filters with a roll - off of 12 db / octave . we used a 12-bit adc to sample both qa s simultaneously at a rate of 10 mhz per channel , which is sufficiently high to avoid signal aliasing completely . following each drop of the trap , we continuously recorded both qa s for a data acquisition window of 50 ms and streamed the data to a hard drive for offline processing following the experimental run . ideally , we would like the data acquisition procedure just described to yield directly the amplitude and phase quadrature amplitudes of light transmitted through the cavity . if we write the transmitted optical field as @xmath115 , where @xmath116 is a slowly - varying complex amplitude , the amplitude @xmath117 and phase @xmath118 quadrature - amplitudes are defined by @xmath119 . with respect to the quantum - mechanical theory of the master equation ( [ eq : me ] ) , @xmath120 . note that we define @xmath121 to have zero phase when the cavity is empty , so that @xmath118 should have zero mean when there are no intracavity atoms . given the way that we generate the probe beam , however , we have no way of generating a phase - locked rf local oscillator to recover @xmath117 and @xmath118 directly . the phase of the heterodyne photocurrent differs from the phase that the light has just after it leaves the cavity because of fluctuations in the relative optical path length travelled by the signal beam and optical local oscilator in reaching the photodetectors . so the two signals produced by mixing the photocurrent with the shifted and unshifted copies of our rf local oscillator correspond to an orthognal , but _ rotated _ pair of quadrature amplitudes @xmath122 , @xmath123 : @xmath124 luckily , the characteristic timescales for fluctuations in the phase @xmath125 are quite long ( @xmath126 ms , corresponding to acoustic disturbances ) compared to the 250 @xmath2s duration of an individual atom - transit signal . in processing the recorded data to produce the plots discussed below , we have therefore used an `` adaptive '' definition of the amplitude and phase quadrature amplitudes . within a window of 2 ms preceding the signal of interest , we estimate the instantaneous value of @xmath125 by determining the rotation of @xmath122,@xmath123 that produces one quadrature @xmath127 with zero mean and one quadrature @xmath128 with positive mean . then @xmath127 is operationally defined to be the phase quadrature photocurrent , and @xmath128 is the amplitude quadrature photocurrent . figure ( [ fig : hmtlong ] ) shows an example of a 15 ms segment of our quadrature amplitude data . note that some excess low - frequency noise can still be seen in the phase quadrature , which constrains the lower end of our measurement bandwidth to @xmath95 khz . six prominent atom - transit signals , characterized by a sharp drop in the amplitude quadrature and a simultaneous increase in the phase quadrature , can be seen between @xmath129 sec and @xmath130 sec . in our subsequent discussions of the data , we shall focus on individual signal `` events '' of this type . in order to make quantitative comparisons between data and theory , we need to calibrate the strength of the driving field used in each experimental run . as we will ultimately choose to ignore dc optical phase offsets , the relevant quantity for us will be the number of photons @xmath131 that builds up when the cavity is empty . this corresponds to @xmath132 in the master equation case and @xmath133 for the obse . our strategy for determining the intracavity photon number during experimental runs has been to work backwards from the heterodyne photocurrent observed with no atoms in the cavity , using the expression @xcite @xmath134 where @xmath135 is the mean value of the demodulated amplitude quadrature signal ( i.e. the heterodyne photocurrent is of the form @xmath136 ) , @xmath137 is the mean - squared power of fluctuations in the same signal due to ( optical ) shot noise , @xmath138 is the measurement interval , and @xmath139 . note that the above expression is valid for a coherent signal beam , which we assume to be the state of the light transmitted through the cavity when no atoms are present . for an accurate calibration , we thus need to know the output mirror transmission @xmath19 , the total cavity loss ( which can be determined from measurements of the cavity finesse ) , and the overall photodetection efficiency @xmath32 . in determining @xmath131 from the data , we typically chose @xmath138 on the order of 1 ms . the following subsections provide further detail for each critical aspect of the calibration . ideally , the quantity @xmath137 appearing in equation ( [ eq : hetsn ] ) should coincide with the photocurrent noise power @xmath140 observed when the signal beam is blocked . this would allow us to determine the intracavity photon number without having to calibrate the exact gains of the photocurrent amplifiers , etc . but the procedure is invalid if the optical local oscillator carries excessive technical noise , so we have empirically checked the scaling of our nominal @xmath137 with the dc optical power of the lo . a fit of the data to the functional form @xmath141 yields @xmath142 mw@xmath143 . given our typical operating lo power of 1 mw per detector , and considering the relative magnitudes of other uncertainties , we approximate @xmath144 . note that the lo does pass through an intensity stabilizer with @xmath145 hz bandwidth just before entering the cleaning cavity . we directly determined the total cavity field decay rate @xmath146 by making a calibrated measurement of the hwhm of a cavity tem@xmath59 resonance . in order to minimize systematic errors , we did this by using the heterodyne detectors to monitor the transmitted optical power of a fixed - frequency ti : sapph probe beam while scanning the cavity length . the cavity length was always under servocontrol during the measurement , as we generated the scan by stepping the frequency of the rf going to the travelling wave modulator for the locking diode laser . the total cavity field decay rate is then given by the resonance hwhm measured in terms of the modulator rf scan , times a correction factor of the ratio of the diode laser and ti : sapph wavelengths . we find @xmath147 mhz . as the mirrors used to construct the cavity should be identical , we assign half the total losses to each mirror . unfortunately , we did not manage to characterize the ratio of intracavity losses @xmath20 to transmission losses @xmath148 before we accidentaly damaged the cavity . given that the mirror coatings we have previously received from the same manufacturer have displayed very low scattering / absorption loss ( @xmath149 ) in the wavelength range of interest @xcite , we have assumed @xmath150 mhz . three principal factors determine the value of @xmath151 : the spatial overlap @xmath152 between the signal beam and the optical local oscillator , the photodetector quantum efficiency @xmath153 , and the ( power ) efficiency @xmath154 with which we transfer light from the cavity ouput to the photodetectors . using an optical powermeter we measured @xmath155 , as well as @xmath156 for our two photodetectors . we measured @xmath157 by producing a dc fringe between the signal beam and the optical local oscillator , with both beams adjusted to a power level of 1 @xmath2w . the power adjustments were made upstream of the physics cavity and lo cleaning cavity , ensuring that no optical misalignments were introduced in the process . we obtained an independent measurement of @xmath158 from the heterodyne signal - to - noise ratio obtained with a measured signal power of @xmath159 nw . note that the reasonable agreement between this latter value and the directly - measured @xmath160 provides further indication that our optical local oscillator bears minimal excess noise , and that our assumptions @xmath161 , @xmath162 are valid . although it could have been avoided , we did suffer one additional loss of detection efficiency due to gain and phase imbalance in the subtraction of heterodyne photocurrents . several weeks after we took the data sets shown below , we realized that one photodetector output had an rf signal 4 db higher than the other ( this figure includes both the deviation of our heterodyne beamsplitter from being 50/50 , the difference in photodiode quantum efficiencies , and the difference in transimpedance gains ) , and a phase offset of 1 radian . if we write the two photocurrents ( for a coherent signal beam ) as @xmath163 and @xmath164 ( where @xmath165 and @xmath166 are uncorrelated gaussian white noises with zero mean and unit variance ) , the difference photocurrent is @xmath167 where @xmath168 is again a gaussian white noise process with zero mean and unit variance . the complex imbalance @xmath169 should thus reduce the effective photodetection efficiency ( for the purpose of evaluating the heterodyne signal - to - noise ratio ( [ eq : hetsn ] ) ) by @xmath170 using the measured @xmath171 and @xmath172 rad , @xmath173 , and we quote an overall value of @xmath174 . in order to facilitate the interpretation of our data , we generated some rudimentary monte carlo simulations of the heterodyne signals that we should see as atoms fall through the cavity . our code simulates three - dimensional , classical center - of - mass motion of individual atoms under the influence of fluctuating forces due to their strong interactions with the cavity field . gravity is also included in the kinematics , but plays only a minor role over the @xmath95 ms duration of the simulations . the basic scheme of the simulations is to precompute the values of the mean cavity - induced force , the cavity - field - induced diffusion coefficient , and the steady - state ( complex ) amplitude for the intracavity field as a function of atomic position . hence , _ every _ other degree of freedom in the simulation is adiabatically eliminated and slaved to the atomic motion , which is assumed to be the slowest and the `` stiffest '' process in the dynamics . the mean force and intracavity field are derived from steady - state solutions of the master equation ( [ eq : me ] ) , and the cavity diffusion is computed using the quantum regression theorem @xcite . in each timestep , the code first performs an interpolation on the precomputed tables of values to determine the appropriate change in atomic momentum ( which includes a stochastic increment consistent with the local value of the diffusion constant ) , records an appropriate value for the cavity output field , and then updates the atomic position and momentum . the simplest possible integration scheme is used ( explicit ito - euler ) , yielding order @xmath175 convergence in the timestep . we chose a timestep of 7.5 ns in order to keep the run - times for the simulations reasonable under matlab on our pentium ii workstations , and this should have been sufficiently small to keep the integration error below other sources of inaccuracy . the principle shortcoming of this scheme is that it assumes the atomic velocity will remain small enough that variations in the coupling strength @xmath53 will be negligible over timescales on the order of @xmath176 and @xmath177 . if this condition is violated , then the steady - state values of the mean cavity force and cavity output are no longer appropriate . another major approximation was made in treating only a two - level atom optical pumping and/or the associated opto - mechanical effects ( _ e.g. _ , spontaneous forces ) could certainly play an important role in determining the shapes of our observed transit signals . a third , less severe approximation is that we have treated the stochastic process associated with recoil from spontaneous - emissions as being statistically independent from the one associated with dipole - force fluctuations . the possibility that interesting effects could arise due to quantization of the atomic motion is of course ignored as well . given the relative simplicity of our simulation scheme , it would be inappropriate to draw any strong conclusions about our experimental data soley on the basis of apparent similarities with the numerical results . however , the simulations do provide an appealing _ model _ for the atomic center - of - mass dynamics and associated heterodyne signals , with predictions that seem to be fully consistent with what we see in the actual experiment . this model suggests some interesting interpretations for qualitative features of the data . for example , the marked asymmetry ( in time ) of most of our experimental transit signals may be associated with an initial phase of atomic `` channeling '' in the dipole potential provided by the intracavity optical field , followed by a sudden escape from local confinement due to cavity - field - induced momentum diffusion . such a dynamical process , if it could be confirmed in some way , would bear a strong resemblence to the levy walk behavior predicted in @xcite and observed in @xcite . our simulations also suggest that isolated dips and steps in the observed transit signals may be associated with sudden changes in atomic localization relative to the cavity standing wave . before presenting some results of the atomic - motion simulations , let us first describe the numerical method used to compute the cavity diffusion coefficients . following doherty _ et al _ @xcite , we computed the diffusion coefficient associated with dipole - force fluctuations according to @xmath178 where @xmath179 is the force operator @xmath180 and we define @xmath181 . \label{eq : gofr}\ ] ] note that the vector nature of the force operator comes only from the gradient @xmath182 . unlike the computation of mean forces and the expected intracavity field amplitude , the evaluation of ( [ eq : diffco ] ) requires an actual time - integration of the cavity - qed master equation ( [ eq : me ] ) . note that @xmath183 - \langle{\bf f}\rangle^2 , \label{eq : heisen}\ ] ] where @xmath184 is the liouville superoperator defined by @xmath185 with reference to the cavity - qed master equation ( [ eq : me ] ) . knowing the steady - state atom - cavity density matrix @xmath30 , one can evaluate the correlation function in ( [ eq : heisen ] ) by integrating the master equation for a time @xmath154 with @xmath186 as an initial condition , multiplying the result by @xmath30 , and finally taking the trace . the numerical integration was performed by a c++/mpi code on an sgi / cray origin-2000 cluster . we used a truncated basis of 25 fock states for the cavity mode , using an explicit euler integration with a 1 ps timestep and 5 @xmath2s total integration time . figure [ fig : cavdiff1 ] shows the results of such a calculation for the diffusion coefficient as a function of atomic position along the standing wave , for an atom located on the cavity axis . recall that the cavity mode function varies as @xmath187 along the cavity axis , and as a gaussian in the transverse dimensions . starting from the top , the four curves represent @xmath188 , @xmath159 , @xmath189 , and @xmath190 mhz , all with a probe strength such that an average of 2 photons would build up in the cavity if it were empty and with @xmath191 . figure [ fig : cavdiff2 ] shows similar results , all for @xmath192 mhz , but with an average intracavity photon number ( for the empty cavity ) of @xmath193 for the top curve , 4 for the middle , and 2 for the lowest . as the variation of @xmath53 along the standing wave is much more rapid than in the transverse ( gaussian ) directions , we approximated @xmath194 in expression ( [ eq : diffco ] ) . the computation of @xmath179 for the mean force field was of course three - dimensional . note that expression ( [ eq : diffco ] ) gives the diffusion coefficient associated with dipole force fluctuations only it takes no account of recoils from spontaneous emission . we therefore incorporate a second diffusion process in the simulations , whose coefficient is computed from the expectation value of the atomic excitation at each point in space : @xmath195 here @xmath196 is the atomic spontaneous emission rate ( einstein a coefficient ) , and the number @xmath197 comes from averaging over the angular distribution pattern for dipole radiation @xcite . figure [ fig : simex1 ] shows an example of the atomic position along the standing - wave @xmath25 and the atomic velocity along the standing - wave @xmath198 from one typical simulation . also shown are the mean value of the intracavity phase quadrature amplitude @xmath199 , and a simulated heterodyne signal with gaussian noise added at the level appropriate to our experimental parameters ( see equation ( [ eq : hetsn ] ) ) . note that the quantity ( @xmath200 plus noise ) is proportional to the photocurrent @xmath127 that would be recorded in our experiment . in the bottom row of the figure , @xmath200 and @xmath200 plus shot - noise are shown at an analog bandwidth 300 khz in order to illustrate the effects of finite detection bandwidth on the qualitative features of the data @xcite . the results clearly suggests that the gross features of the simulated signals are most strongly affected by atomic motion along the standing wave , in that the overall gaussian profile associated with motion in the @xmath26-direction becomes distorted by the `` envelope '' of the oscillatory variations due to motion along @xmath25 ( recall @xmath201 $ ] ) . looking at the results for @xmath202 and @xmath203 , we see that the atom in this particular simulation was mechanically confined within one well of the cavity standing wave until @xmath204 @xmath2s . that is , @xmath25 did not vary by more than @xmath205 and @xmath198 displays the oscillations that one would expect to see for an atom trapped within a potential well . at time @xmath204 @xmath2s , however , we see that momentum diffusion finally pushes @xmath198 above some threshold value such that the atom is able to escape from local confinement and `` fly above '' the periodic dipole potential associated with the cavity standing wave . generally speaking , one expects this type of escape to occur as the atom passes through the cavity axis in the @xmath206-direction and the standing - wave potential wells begin to decrease in depth . the expectation value @xmath200 clearly reflects the qualitative features of the atomic trajectory @xmath202 , although we also see that the details are lost in the measured signal due to heterodyne shot - noise . what survives in the simulated heterodyne signal is an overall asymmetry in the envelope of the transit - signal , with a sudden `` step '' downwards at @xmath207 @xmath2s ( as suggested by the dotted vertical line ) . figure [ fig : sims ] shows four simulated signals with @xmath208 mhz and an average intracavity photon number of @xmath209 , and two simulated signals with @xmath192 mhz and @xmath210 ( @xmath191 in all cases ) . in each subplot , the upper trace shows the intracavity amplitude quadrature @xmath211 versus time , and the lower trace ( with zero mean value ) shows the intracavity phase quadrature @xmath212 versus time . again , gaussian noise has been added to the simulated signals in order to reproduce the overall signal - to - noise ratio predicted by equation ( [ eq : hetsn ] ) . in all simulations , the atom was started at a position 7 gaussian waists above the cavity with a vertical velocity of -47 cm / s . the initial transverse position , transverse velocity , position along the cavity standing - wave axis , and velocity along the standing - wave were randomly assigned within parameter ranges that were narrow enough to ensure that most simulated atom - drops produced a sizeable signal . the simulated signals in figure [ fig : sims ] may be compared directly to the real data shown in figure [ fig : hmthbw ] . like the simulation shown in figure [ fig : simex1 ] , the signals in figure [ fig : sims ] display a generic asymmetry and often contain an isolated dip or step . the features of this type that are marked by an arrow near the horizontal axis are again associated with sudden changes in the atomic localization relative to the cavity standing wave ( as determined by examining @xmath202 from the simulations ) . although we have the benefit of knowing both the `` actual '' trajectory of an atom and the corresponding heterodyne signal in our simulations , _ a priori _ knowledge of the atomic trajectory is of course unavailable in our experiment . hence , the results of our numerical simulations can only provide general guidelines for how we might try to infer something about individual atomic trajectories from our heterodyne transit signals . first of all , it appears that signficant asymmetries in the observed signals could provide evidence that we do in fact observe mechanical aspects of the atom - cavity coupling within individual atomic transits . relative to the approximations involved in our simulations however , one should bear in mind that such asymmetries could also arise from optical pumping processes that may occur during at atomic transit through the cavity . second , the simulations suggest that the transit signal envelopes are most strongly influenced by atomic motion along the cavity standing wave , as opposed to details of the motion along @xmath44 or @xmath26 . third , isolated dips or steps in the signal could be indicative of sudden changes in the atomic localization with respect to the standing wave . again however , we should be careful to note that such features might also be caused by intracavity optical pumping . moving on to the experimental data , let us first discuss some transit signals displayed in the same fashion as the simulations of figure [ fig : sims ] . the quality of this data illustrates our experimental ability to perform continuous , nearly quantum - noise limited measurement of the atom - cavity interaction energy during individual scattering events . in the second subsection we shall display transit data on the complex plane , and present a quantitative comparison to theoretical predictions of the atom - cavity system s complex optical susceptibility . we shall see that in optimal cases , sufficient information may be obtained from an _ individual _ atom - transit signal to distinguish unambiguously between quantum and semi - classical models of cavity qed . note that all of the data shown here were taken with cavity - probe detuning @xmath213 , while the atom - probe detuning @xmath214 and probe power @xmath131 are varied . the basic unit of our experimental data is a continuous stream of quadrature - amplitude values versus time , recorded for @xmath12 ms windows following each dropping of the mot . one @xmath215 ms segment of such data is shown in figure [ fig : hmtlong ] . this particular data segment was taken with detunings @xmath216 mhz and with a probe power such that an average of @xmath217 photons would build up in the cavity if it were empty ( @xmath218 ) . figure [ fig : hmthbw ] shows six of the largest single - atom transit signals from our entire data set , which covers detunings @xmath40 from @xmath219 to @xmath220 mhz and probe powers @xmath131 from @xmath217 to @xmath221 photons . the particular values of @xmath40 and @xmath131 for each signal in figure [ fig : hmthbw ] are displayed by the y - axis . the data are shown at our full analog bandwidth of 300 khz and sampling rate of 10 mhz ( 12 bit resolution ) . note that we have displaced the amplitude quadrature signals @xmath128 by @xmath222 in order to prevent them from overlapping with the phase quadrature signals @xmath127 . we have found that transit signals of maximal contrast tend not to have very much internal structure , although there is a fairly prominant dip at the point indicated by an arrow in subplot `` dat4 , '' a set of three dips indicated by the arrow in subplot `` dat6 , '' and oscillatory structure in the signal of subplot `` dat5 . '' the signal of subplot `` dat4 '' is really not so different in its overall structure than the simulations shown in figure [ fig : simex1 ] and in the `` traj10 '' subplot of figure [ fig : sims ] . one can see that the shapes of the overall signal envelopes do vary substantially and are generally asymmetric in time . the distribution of the atom - transit signal between amplitude and phase quadratures clearly depends upon the probe detuning , as will be discussed in greater detail below . for the data event recorded with @xmath188 , one sees that there is only a reduction in the power of the transmitted probe without any shift in its phase . at all other detunings , the atom - induced reduction of amplitude quadrature is partly offset by an increase in the phase quadrature , indicating a significant phase shift of the transmitted probe beam . in subplot `` dat6 '' taken with @xmath192 mhz , the signal is seen to primarly reside in the phase quadrature . for signals such as those shown in figure [ fig : hmthbw ] , we may estimate a full - signal to rms - noise ratio of approximately 2.5 for the phase quadrature and 4 for the amplitude quadrature . combining the two signals , we have an overall ratio of 4.5 . the signal bandwidth is 300 khz , implying a relative sensitivity of @xmath223 hz@xmath224 . assuming that the largest signals correspond to atoms reaching the maximal coupling strength of @xmath225 mhz , this sets our broadband sensitivity @xmath226 to time - varations in @xmath227 to be @xmath7 khz/@xmath11 . given our overall photodetection efficiency @xmath174 and our independent assesment of the excess noise factor @xmath99 ( corresponding to technical noise on the optical local oscillator power and on the physics cavity lock ) , we find that this value lies only a factor of @xmath228 above the fundamental quantum noise floor . we maintain this high sensitivity over 2.5 decades of signal frequency , from our full photodetection bandwidth of 300 khz down to 1 khz ( at which point we are limited by residual technical noise in our measurements of the phase quadrature ) . although a complete theoretical description of our experiment at the level of a stochastic master equation @xcite could in principle be formulated to include three - dimensional quantized atomic center - of - mass motion , cavity birefringence , and the full manifold of internal atomic zeeman states , the resulting model would almost certainly be too complex to be useful for quantitative interpretations of our existing data . nevertheless , we wish here to stress that only a theory that explicitly treats the _ quantum conditioning _ @xcite of the atom - cavity evolution on the stochastic component of the heterodyne photocurrent could provide a full account of the statistics and autocorrelation properties of our atom - transit data . in this crucial sense , we would claim that our recent work may be distinguished from all other efforts in contemporary experimental physics . of course , much remains to be done before we can offer definitive proof , so our continuing efforts will largely be directed towards the unambiguous demonstration of quantum conditional dynamics in cavity qed . in addition to displaying our atom - transit signals in the format of photocurrent versus time , we can also construct parametric plots of @xmath128 versus @xmath127 to examine the correlation induced between these two quantities by the atom - cavity interaction . this is equivalent to compiling a histogram of the complex amplitude of the optical field transmitted through the cavity during an atomic transit , and hence to a continuous monitoring of the atom - cavity system s complex optical susceptibility . the underlying probability distributions for such histograms are determined by the ( evolving ) husimi q - function for the intracavity field @xcite . many of our measurements have been conducted in a regime of strong cavity - driving fields ( @xmath229 ) , with the consequence that significant effects of saturation and optical nonlinearity can be seen in our experimental data . as has been extensively discussed in the cavity - qed literature , the nonlinear optical response of the atom - cavity system represents an important experimental signature that may be used to distinguish between quantum and semiclassical theoretical models for the atom - cavity interaction . prior to this work , four groups @xcite have reported nonlinear measurements in cavity qed . in each of these previous experiments , the information gained per atomic transit about the details of the system s nonlinear response was far less than in the data we shall present here . hence , we wish again to stress that the techniques developed in our most recent experimental work have opened new possibilities for exploring quantum dynamics at the level of single quantum realizations ( trajectories ) , as opposed to the level of ensemble averages . in figures [ fig : hmtfig3 ] and [ fig : hmtfig2 ] , we display single - transit data on the complex plane . the experimental data are displayed as gray dots , with each dot representing the values of @xmath128 and @xmath127 at some particular time during an atomic transit . in order to produce a set of dots from a transit signal such as those shown in figure [ fig : hmthbw ] , the quadrature amplitude signals were first passed through an anti - aliasing filter and then resampled every 10 @xmath2s to produce a discrete set of points @xmath230 . for every @xmath231 , a gray dot is placed at radius @xmath232 and polar angle @xmath233 . these transit phasors represent a simple way of looking at the correlation between the amplitude and phase of light transmitted through the cavity , with time removed from the picture . we believe that this type of plot is the best format for comparisons to theory , because atomic internal ( optical pumping among zeeman states ) and external ( motion through the cavity eigenmode ) dynamics should be factored out . in the simplest approximation , these two types of processes merely induce fluctuations in the atom - cavity coupling @xmath53 , which should only move the locus of our gray dots in and out along a curve parametrized by @xmath53 . the overall shape of this curve should be dictated by the interaction hamiltonian for the atom and cavity mode , and is therefore quite easy to compute . in figure [ fig : hmtfig3 ] , we show a set of transit phasors taken at various probe detunings . the data overlay theoretical curves predicted by the quantum and semiclassical theories . the quantum - mechanical predictions are computed by finding the steady - state solution of the master equation ( [ eq : me ] ) for values of @xmath53 in the range @xmath234 $ ] and appropriate values of @xmath40 and @xmath131 . the solid curves represent an interpolation through the discrete set of computed values for @xmath212 and @xmath200 as a function of @xmath53 . the semiclassical prediction is computed in the analogous manner , using the optical bistability state equation ( [ eq : obse ] ) rather than the quantum master equation . note that both curves must agree for @xmath235 , and this common point of origin is marked in each subplot by a triangle . the @xmath236 endpoint of the quantum curve is marked by a circle , and the @xmath236 endpoint of the semiclassical curve is marked by an @xmath237 . in the subplots of figure [ fig : hmtfig3 ] , the grey dots represent an overlay of data from two individual atom - transit signals . the two particular signals displayed in each subplot were selected on the basis of having maximal `` contrast '' among all the signals from a given data set , under the assumption that the atoms causing maximal signals should have come the closest to actually achieving @xmath236 on their way through the cavity . quantum theory is seen to predict the observations quite well for all the detunings shown in figure [ fig : hmtfig3 ] . the quantum and semiclassical predictions shown in figure [ fig : hmtfig3 ] do not differ significantly except for the case of @xmath188 . generally speaking , the two theories are known to agree in their predictions for weak - field response of the atom - cavity system but to differ in their predictions regarding _ saturation _ of the optical response . we here present experimental evidence _ from single atom transits _ for the quantum character of saturation in the atom - cavity system s response to near - resonant driving fields . in figure [ fig : hmtfig2 ] we show a set of transit phasors for fixed detuning @xmath208 but variable probe strength @xmath238 , @xmath239 , @xmath240 , and @xmath221 . a clear discrepancy between the quantum and semiclassical theory may be seen in this sequence of plots , with the data ( gray dots ) showing significant preference for the quantum - mechanical predictions . each subplot in figure [ fig : hmtfig2 ] represents an overlay of data from three individual atom - transit signals , in order to fill in the overall `` shape '' of the experimental transit phasors . it should be clear , however , that just a single data trace would suffice to determine that the experiment matches much more closely with the quantum theory than the semiclassical theory . we plan to elaborate this result in separate publication @xcite . in summary , we have described the details of our recent experimental work to perform continuous measurement of the interaction energy @xmath241 between one atom and an optical cavity during individual transit events . we displayed heterodyne transit signals in two complementary formats , one of which highlights the large bandwidth and nearly quantum - noise limited signal - to - noise ratio achieved in tracking the time - evolution of the amplitude and phase quadratures ( @xmath242 ) of light transmitted through the cavity . the second data format ( transit phasors ) displays the correlation induced between @xmath128 and @xmath127 by the atom - cavity interaction , and our experimental results show that we are able to distinguish unambiguously between quantum and semiclassical models of cavity qed in only a few ( or even just one ) atomic transits . we have furthermore presented the results of rudimentary numerical simulations of atomic motion under the influence of mechanical forces and momentum diffusion associated with the strong atom - cavity coupling , and we examined the interpretations suggested by these simulations for certain qualitative features of our experimental data ( _ i.e. _ steps and asymmetries ) . the primary conclusion we wish to draw in comparing the simulation results with our experimental data is that we have experimentally reached a regime of measurement sensitivity and bandwidth in which details of the atomic center - of - mass trajectories really should be visible . even though we have no incontrovertible means of proving that steps and asymmetries seen in our experimental data should be associated with dynamical processes like channeling and diffusive escape , we are now motivated in our continuing work to develop some means of _ actively influencing _ the atomic motion _ while _ it is still inside the cavity . this would allow us to produce deliberate displacements of an atom along the cavity standing wave , and to examine the induced variations of the heterodyne signal in order to verify our inferred displacement sensitivity of @xmath10 m/@xmath11 . we anticipate that the combined abilities of monitoring and influencing atomic position relative to the cavity standing wave will enable the investigation of schemes for _ real - time feedback control _ of quantized atomic center - of - mass motion . finally , let us note that the standard quantum limit for overall observation time of the position of a single atom with sensitivity @xmath10 m/@xmath11 and 300 khz bandwidth should be @xmath243 @xmath2s @xcite . this implies that quantitative experimental investigations of conditional quantum dynamics ( as described in @xcite ) should indeed become possible in our cavity qed system once we are able to reliably prepare individual atoms in well - defined initial states of motion . the ideal situation in this regard would be to trap and localize atoms _ within _ the cavity , releasing them at a node or antinode of the standing wave , on the cavity axis , and with an initial position uncertainty that is small compared to @xmath205 . current efforts in our group focus on trying to achieve this level of control via optical dipole - force traps and/or far - detuned optical lattices inside the cavity . we wish to acknowledge q. a. turchette s vital participation in the early stages of this work , and to thank a. c. doherty , c. j. hood , and t. w. lynn for valuable discussions . much of the numerical work described in this paper was performed on resources located at the advanced computing laboratory of the los alamos national laboratory . this research was supported by the national science foundation under grant no . phy97 - 22674 , by the onr , and by darpa through the quic initiative ( administered by aro ) . jun ye is supported by an r. a. millikan fellowship from the california institute of technology . h. j. kimble : in _ cavity quantum electrodynamics_. san deigo , academic press : 1994 d. m. meekhof , c. monroe , b. e. king , w. m. itano , d. j. wineland : phys . lett . * 76 * , 1796 ( 1996 ) d. leibfried , d. m. meekhof , b. e. king , c. monroe , w. m. itano , d. j. wineland : phys . rev . lett . * 77 * , 4281 ( 1996 ) w. e. moerner : science * 265 * , 46 ( 1994 ) b. c. stipe , m. a. rezaei , w. ho : science * 279 * , 1907 ( 1998 ) m. r. andrews , d. m. kurn , h. j. miesner , d. s. durfee , c. g. townsend , s. inouye , w. ketterle : phys . rev . * 79 * , 553 ( 1997 ) m. kozuma , k. nakagawa , w. jhe , m. ohtsu : phys . a * 57 * , r24 ( 1998 ) s. guibal , c. triche , l. guidoni , p. verkerk , g. grynberg : opt * 131 * , 61 ( 1996 ) c. m. caves , g. j. milburn : phys . a * 36 * , 5543 ( 1987 ) h. m. wiseman : quantum semiclass * 8 * , 205 ( 1996 ) a. m. herkommer , h. j. carmichael , w. p. schleich : quantum semiclass * 8 * , 189 ( 1996 ) a. c. doherty , a. s. parkins , s. m. tan , d. f. walls : phys . a ( in press ) h. j. carmichael , _ an open systems approach to quantum optics_. berlin , springer : 1993 h. m. wiseman , g. j. milburn : phys . a * 47 * , 642 ( 1993 ) h. m. wiseman , g. j. milburn : phys . a * 47 * , 1652 ( 1993 ) p. zoller , c. w. gardiner : in _ quantum fluctuations , session lxiii of lecole dete de physique des houches_. amsterdam , north - holland : 1997 h. m. wiseman : phys . lett . * 75 * , 4587 ( 1995 ) r. alicki , d. makowiec , w. miklaszewski : phys . 77 * , 838 ( 1996 ) r. schack , c. m. caves : phys . e * 53 * , 3257 ( 1996 ) h. mabuchi : quantum semiclass * 8 * , 1103 ( 1996 ) h. m. wiseman : phys . rev . a * 49 * , 2133 ( 1994 ) h. mabuchi , p. zoller : phys . lett . * 76 * , 3108 ( 1996 ) j. a. dunningham , h. m. wiseman , d. f. walls : phys . a * 55 * , 1398 ( 1997 ) k. s. wong , m. j. collett , d. f. walls : opt . * 137 * , 269 ( 1997 ) a. h. zewail : j. phys 100 * , 12701 ( 1996 ) s. d. gensemer , p. l. gould : phys * 80 * , 936 ( 1998 ) m. a. m. marte , p. zoller : appl . b * 54 * , 477 ( 1992 ) p. storey , t. sleator , m. collett , d. walls : phys . rev . a * 49 * , 2322 ( 1994 ) c. c. gerry : phys . a * 53 * , 1179 ( 1996 ) h. mabuchi , q. a. turchette , m. s. chapman , h. j. kimble : opt * 21 * , 1393 ( 1996 ) c. j. hood , m. s. chapman , t. w. lynn , h. j. kimble : phys . lett . * 80 * , 4157 ( 1998 ) m. j. collett , r. loudon , c. w. gardiner : j. mod . opt . * 34 * , 881 ( 1987 ) l. a. lugiato : prog . opt . * 21 * , 71 ( 1984 ) superpolished bk7 substrates with multi - layer dielectric coatings fabricated by research electrooptics in boulder , co. a. e. siegman : _ lasers_. mill valley , university science books : 1986 . r. j. rafac , c. e. tanner , a. e. livingston , k. w. kukla , h. g. berry , c. a. kurtz : phys . a * 50 * , r1976 ( 1994 ) o. schmidt , k .- knaak , r. wynands , d. meschede : appl . b * 59 * , 167 ( 1994 ) q. a. turchette , r. j. thompson h. j. kimble : appl . phys . b * 60 * , s1 ( 1995 ) j. h. shirley : opt . * 7 * 537 ( 1982 ) r. w. p. drever , j. l. hall , f. v. kowalski , j. hough , g. m. ford , a. j. munley , h. ward : appl . b * 31 * , 97 ( 1983 ) w. vassen , c. zimmermann , r. kallenbach , t. w. hnsch : opt . commun . * 75 * , 455 ( 1990 ) t. l. boyd , h. j. kimble : opt * 16 * , 808 ( 1991 ) a. cable , m. prentiss , n. p. bigelow : opt . lett . * 15 * , 507 ( 1990 ) e. s. polzik , j. carri , h. j. kimble : appl b * 55 * , 279 ( 1992 ) g. rempe , r. j. thompson , h. j. kimble : opt . * 17 * , 363 ( 1992 ) c. w. gardiner : _ quantum noise_. berlin , springer : ( 1991 ) s. marksteiner , k. ellinger , p. zoller : phys . rev . a * 53 * , 3409 ( 1996 ) a. c. doherty : private communication ( 1996 ) h. katori , s. schlipf , h. walther : phys . lett . * 79 * , 2221 ( 1997 ) a. c. doherty , a. s. parkins , s. m. tan , d. f. walls : phys . rev . a * 56 * , 833 ( 1997 ) q. a. turchette , c. j. hood , w. f. lange , h. mabuchi , h. j. kimble : phys . lett . * 75 * 4710 ( 1995 ) m. brune , f. schmidt - kaler , a. maali , j. dreyer , e. hagley , j. m. raimond , s. haroche : phys . lett . * 76 * , 1800 ( 1996 ) r. j. thompson , q. a. turchette , o. carnal , h. j. kimble : phys . a * 57 * , 3084 ( 1998 ) h. mabuchi , j. ye , and h. j. kimble : in preparation h. mabuchi : phys . a ( in press )

we explore the onset of star formation in the early universe , exploiting the observations of high - redshift lyman - break galaxies ( lbgs ) and ly emitters ( laes ) , in the framework of the galaxy formation scenario elaborated by granato et al . ( 2004 ) already successfully tested against the wealth of data on later evolutionary stages . complementing the model with a simple , physically plausible , recipe for the evolution of dust attenuation in metal poor galaxies we reproduce the luminosity functions ( lfs ) of lbgs and of laes at different redshifts . this recipe yields a much faster increase with galactic age of attenuation in more massive galaxies , endowed with higher star formation rates . these objects have therefore shorter lifetimes in the lae and lbg phases , and are more easily detected in the dusty submillimeter bright ( smb ) phase . the short uv bright lifetimes of massive objects strongly mitigate the effect of the fast increase of the massive halo density with decreasing redshift , thus accounting for the weaker evolution of the lbg lf , compared to that of the halo mass function , and the even weaker evolution between and of the lae lf . the much lower fraction of lbgs hosting detectable nuclear activity , compared to smb galaxies , comes out naturally from the evolutionary sequence yielded by the model , which features the coevolution of galaxies and active nuclei . in this framework laes are on the average expected to be younger , with lower stellar masses , more compact , and associated to less massive halos than lbgs . finally , we show that the intergalactic medium can be completely reionized at redshift by massive stars shining in protogalactic spheroids with halo masses from a few to a few , showing up as faint lbgs with magnitude in the range , without resorting to any special stellar initial mass function .

astro-ph0611799

the impressive recent advances in the observational studies of high redshift galaxies and quasars have brought us up to the outskirts of ` dark ages ' , when the amount of neutral intergalactic hydrogen starts becoming significant and tentative indications of the presence of metal - free population iii stars begin to show up . this may give us some confidence that we are approaching the onset of the galaxy formation history and provides key information on physical processes driving the early galaxy evolution and on their influence on the intergalactic medium ( igm ) . different techniques and selection criteria have been exploited to find high-@xmath7 galaxies . a most efficient method is the lyman break dropout technique ( steidel & hamilton 1993 ) , first applied to select @xmath2 galaxies ( steidel et al . 1996 ; lowenthal et al . 1997 ) and subsequently extended to redshifts of up to @xmath8 ( steidel et al . 1999 , 2003 ; dickinson et al . 2004 ; giavalisco et al . 2004 ; ouchi et al . 2004 ; bunker et al . 2004 ; vanzella et al . 2005 , 2006 ; bouwens et al . 2005 ; thompson et al . 2006 ; bouwens & illingworth 2006a , b ) . very deep narrow band imaging at selected redshift windows ( cowie & hu 1998 ) has also been successfully used to search for ly@xmath0 emission up to @xmath9 ( ouchi et al . 2003 ; kashikawa et al . 2006b ; shimasaku et al . 2006 ; iye et al . 2006 ) . while these techniques are sensitive to unobscured or moderately obscured objects , very dusty distant galaxies with extremely high star formation rates ( sfrs ) have been detected by scuba @xmath10 m surveys ( see e.g. chapman et al . 2005 ; aretxaga et al . 2007 ) . substantial numbers of passively evolving high-@xmath7 galaxies have also been found ( fontana et al . 2006 ; grazian et al . 2007 ) . the different selections emphasize different aspects of the early galaxy evolution that need to be integrated in a unified view . in this paper we present a simple physical model featuring an evolutionary link between ly@xmath0 selected ( lae ) , lyman break ( lbg ) , submillimeter bright ( smb ) , and passively evolving galaxies , establishing quantitative relations between luminosities and basic physical quantities such as stellar and halo masses , star formation rates , galactic ages , metal abundances and dust attenuation . the model extends to earlier evolutionary phases the approach by granato et al . ( 2004 ) , according to which the evolution of massive galaxies is tightly linked to the growth of the supermassive bhs at their centers . the model has successfully accounted for an extremely large amount of observational data on galaxies and quasars ( granato et al . 2004 , 2006 ; cirasuolo et al . 2005 ; silva et al . 2005 ; lapi et al . 2006 ; see appendix a for details ) . an extension of the model to the earliest evolutionary phases , when metals were first produced , requires a specific treatment of dust attenuation , which plays an important role in the derivation of high-@xmath7 galaxy properties , since even small amounts of dust strongly affect the rest - frame uv emission . for example , a correction for dust attenuation corresponding to @xmath11 would increase the estimate of the sfr at @xmath12 by a factor of about @xmath13 . the presence of significant attenuation in high-@xmath7 lbgs has indeed been reported by many authors ( e.g. , steidel et al . 1999 ; shapley et al . 2001 , 2006 ; huang et al . 2005 ; ando et al . 2006 ; yan et al . 2006 ; burgarella et al . 2006 ; eyles et al . 2007 ; stark et al . 2007b ) , not surprisingly since the dust formation in early galaxies is expected to occur on very short timescales ( morgan & edmunds 2003 ) . the outline of the paper is the following . in 2 we present an overview of the model and show that , with a simple recipe for the evolution of dust attenuation during the earliest phases of the galaxy lifetime , accounting for the luminosity vs. @xmath14 correlation found by shapley et al . ( 2001 ) , it nicely reproduces the observational estimates of the lbg luminosity functions ( lfs ) at different redshifts . in 3 we investigate the absorption effects on the lyman continuum emission , the evolution of the ly@xmath0 lf , and the contributions of young galaxies to the cosmic re - ionization . finally , in 4 we discuss our results and summarize our conclusions . throughout the paper we adopt a flat cosmology with matter density @xmath15 , hubble constant @xmath16 km s@xmath17 mpc@xmath17 , and normalization of the mass variance @xmath18 . unless otherwise specified , magnitudes are in the ab system . the rest - frame uv emission is proportional to the sfr @xmath19 . this is in turn determined by a number of physical processes ( such as cloud - cloud collisions , cooling , fragmentation , energy inputs from quasars , supernovae [ snae ] and stellar winds , etc . etc . ) which occur on different linear and time scales . even with the more sophisticated numerical techniques , one has to resort to many simplifying assumptions to derive the star formation history , since these processes are extremely complex and occur on scales well below the currently achievable resolution of @xmath20-body@xmath21hydro simulations ( iliev et al . 2006 ) . in this paper we exploit the physical recipe elaborated by granato et al . ( 2004 ) , which relates the average sfr to the distribution of gas and dark matter ( dm ) in proto - galaxies , taking into account the effects of the energy fed back to the intra - galactic gas by sn explosions and by accretion onto the nuclear , supermassive black hole ( bh ) . the model envisages that during or soon after the formation of the host dm halo , the baryons falling into the newly created potential well are shock - heated to the virial temperature . the hot gas is ( moderately ) clumpy and cools fast especially in the denser central regions , where rapid mergers of a fraction less than @xmath22 of the total mass yield strong bursts of star formation that can reach up to thousands solar masses per yr in massive halos at high redshift . such a huge star formation activity also promotes the storage of the cooled gas into a reservoir around the central supermassive bh , eventually leading to accretion onto it . the ensuing sn explosions and the nuclear activity feed energy back to the gaseous baryons , and regulate the sfr and the bh growth . these mutual energy feedbacks actually _ reverse _ the formation sequence of the stellar component of galaxies compared to that of dm halos : the star formation and the buildup of central bhs are completed more rapidly in the more massive haloes , thus accounting for the phenomenon now commonly referred to as _ downsizing_. the basic equations of the model and their analytical solutions are reported in appendix a. the model predicts the time evolution of the sfr , of the metal abundance @xmath23 ( taking into account the chemical enrichment of the primordial infalling gas and the gas outflow due to energy injection by snae and active nuclei ) , of the stellar mass ( see fig . [ fig|sfr : mstar ] ) , and of the mass stored in the central supermassive bh . when the star formation and the chemical evolution history of a galactic halo of mass @xmath24 have been computed , the spectral energy distribution ( sed ) as function of time from extreme uv to radio frequencies is estimated through the grasil code ( silva et al . 1998 ) . coupling these results with the halo formation rate , we can obtain the lfs of galaxies as a function of cosmic time . at @xmath25 , the halo formation rate can be reasonably well approximated by the positive part of the derivative of the halo mass function with respect to cosmic time ( e.g. , haehnelt & rees 1993 ; sasaki 1994 ) . the halo mass function derived from numerical simulations ( jenkins et al . 2001 ) is well fit by the sheth & tormen ( 1999 , 2002 ) formula , that improves over the original press & schechter ( 1974 ) expression , which is well known to under - predict by a large factor the massive halo abundance at high redshift . adopting the sheth & tormen ( 1999 ) mass function @xmath26 , @xmath27 denoting the cosmic time , the formation rate of dm halos is given by @xmath28\ , \left|{\mathrm{d}\delta_c\over \mathrm{d } t}\right|\ , n_{\mathrm{st}}(m_h , t)~,\ ] ] where @xmath29 and @xmath30 are constants obtained by comparison with @xmath20-body simulations , @xmath31 is the mass variance of the primordial perturbation field , computed from the bardeen et al . ( 1986 ) power spectrum with correction for baryons ( sugiyama 1995 ) and normalized to @xmath32 on a scale of @xmath33 mpc , and @xmath34 is the critical threshold for collapse , extrapolated from the linear perturbation theory . briefly , the model requires @xmath13 parameters ( all of which are constrained within rather small ranges by independent data or physical arguments ) to describe the physical processes ruling the star formation , the bh growth and their feedback . with the parameter values reported in table 1 of lapi et al . ( 2006 ) , the model fits the available data on the lfs of protospheroidal galaxies at different redshifts and in different spectral bands , and the local relationship between central bh mass and spheroidal luminosity ( hundreds of data points ) . among others , the model reproduces the 850 @xmath35 m extragalactic counts ( coppin et al . 2006 ; scott et al . 2006 ) with their redshift distributions ( chapman et al . 2005 ; aretxaga et al . 2007 ) , and correctly predicted ( silva et al . 2004 , 2005 ) the fraction of high-@xmath7 galaxies in deep @xmath36 m surveys ( prez - gonzlez et al . 2005 ; caputi et al . 2006 ; see magliocchetti et al . 2007 ) , quantities that all proved to be very challenging for competing models . the evolving lfs of optical and x - ray selected quasars at @xmath37 are also well fitted by adding @xmath38 parameters , namely the visibility times in the optical , @xmath39yr , and in x - rays , @xmath40yr , and a dispersion of @xmath41 dex around the mean relationship between the halo and the central bh mass derived from the model ( see lapi et al . 2006 ) . a summary of the model results and of their comparisons with observational data is presented in table 2 of lapi et al . ( 2006 ) . to sum up , the model traces very well the formation and coevolution of quasars and of their spheroidal hosts . the absolute magnitude at @xmath42 , including attenuation , can be written as @xmath43 where @xmath44 erg s@xmath17 hz@xmath17 is the monochromatic luminosity for @xmath45 ( kennicutt 1998 , with the calibration converted to the romano et al . ( 2002 ) initial mass function [ imf ] adopted here , see appendix a ) . the dust attenuation is expected to be proportional both to some power of star - forming gas mass ( or equivalently of the sfr , see eq . [ [ eq|asfr ] ] ) that reflects the dust column density , and to some power of the metallicity , that reflects the fraction of metals locked into dust grains ( see also jonsson et al . 2006 ) . as shown by fig . [ fig|ebmv ] , a simple , average relation between uv attenuation , sfr and @xmath23 @xmath46 consistent with expectations from the model , provides a very good fit of the luminosity - reddening relation found by shapley et al . variations of the normalization coefficient by @xmath22 , of the exponent of the sfr by @xmath47 , and of the exponent of the metallicity by @xmath48 are still compatible with the data . using eq . ( [ eq|extgigi ] ) and the quantities @xmath49 and @xmath50 provided by our model ( see appendix a for handy approximations ) , we compute the evolution with galactic age of the attenuation for lbgs with different halo masses , see fig . [ fig|ext_magn ] ( left panel ) . the attenuation is predicted to be small for lbgs hosted by low - mass halos ( @xmath51 ) , while it quickly increases with time for higher masses , so that more massive galaxies are uv - bright for shorter times . the more massive lbgs spend approximately @xmath52 of their burst time in an interstellar medium optically thick to their uv emission . the right panel of fig . [ fig|ext_magn ] shows the predicted evolution of uv luminosity with galactic age , including the attenuation given by eq . ( [ eq|extgigi ] ) , at @xmath53 for several halo masses . the uv lf of lbgs can be estimated coupling the uv luminosity and halo mass derived from the our model ( cf . [ fig|ext_magn ] ) with the halo formation rate . the rest - frame uv lf at a cosmic time @xmath27 is then @xmath54~,\ ] ] where @xmath55 $ ] is the time spent at magnitudes brighter than @xmath56 . as illustrated in fig . [ fig|uvcounts_extgigi ] , good agreement with the observational data is obtained . neglecting attenuation would increase the prediction at the bright end by more than an order of magnitude . in fact dust attenuation is drastically shortening the time @xmath55 $ ] [ eq . ( [ eq|uvcounts ] ) ] for large halos ( cf . fig . [ fig|ext_magn ] ) . in this context it is worth noticing that the unattenuated lf can be depressed also by assuming short star formation timescales ( around @xmath57myr at @xmath1 ) , as advocated by stark et al . ( 2007a ) . on the other hand , such short timescales would imply stellar masses and stellar ages much lower than observed ; eyles et al . ( 2005 , 2007 ) find stellar masses around @xmath58 and ages of hundreds myrs . these findings are quite naturally reproduced in our framework . the approximate proportionality of the initial ( galaxy age @xmath59 ) sfr with the halo initial gas mass which , in turn , is proportional to the halo mass [ eq . ( [ eq|sfr ] ) ] implies that the highest luminosity tail of the lf is associated to the most massive galaxies . these are however very rare at high-@xmath7 and , furthermore , their uv - bright phase is very short because of the quick increase of their attenuation ( fig . [ fig|ext_magn ] ) ; they are thus easily missed by the available surveys , covering very small areas . on the other hand , low mass galaxies are too faint to be detectable . hence , according to the present model , the available data only sample the halo mass range @xmath60 , with sfrs of @xmath61 , and stellar masses ranging from @xmath62 to @xmath63 . the strong dependence of @xmath64 on halo mass has clearly a key role in shaping the uv lf . the information on the lbg space densities at @xmath65 is scanty . [ fig|uvcounts_extgigi ] shows that we expect a drastic decrease from @xmath66 ( dashed line ) to @xmath67 ( dotted line ) . bouwens & illingworth ( 2006b ) estimate that there are @xmath68 sources brighter than @xmath69 at @xmath70 on an area of @xmath71 ; we predict @xmath72 sources . bouwens et al . ( 2005 ) obtained an upper limit of @xmath38 galaxies at @xmath73 on an area of @xmath74 , to a magnitude limit of 28.5 in @xmath75 ; our model yields @xmath76 . the relative importance of dust attenuation and hi absorption in determining the escape fraction of ionizing photons and the ly@xmath0 luminosity depends on the complex geometry of the distribution of gas , dust and stars . we assume that most of the ionizing and of the ly@xmath0 photons originate in the very central regions of the galaxy and see the same amount of dust as the photons at the reference wavelength of @xmath77 ; we extrapolate the @xmath77 attenuation to shorter wavelengths adopting the attenuation law proposed by calzetti et al . this yields @xmath78 , @xmath79 , and , extrapolating from @xmath80 to @xmath81 ( cf . draine 2003 ) , @xmath82 . the observed luminosity at @xmath81 is given by @xmath83 where @xmath84 is the intrinsic luminosity at @xmath81 , @xmath85 and @xmath86 are the fractions of @xmath81 photons surviving @xmath87 and dust absorption within the galaxy respectively , and @xmath88 is fraction surviving absorption in the igm along the line of sight . the product @xmath89 is the fraction of @xmath81 photons emerging at the galaxy boundary . following inoue et al . ( 2006 ) we define the ratio @xmath90 between the lyman continuum and uv luminosity at the galaxy boundary as @xmath91 the factor @xmath92 , which in principle should appear in this equation , is very close to unity and is therefore usually omitted . rest - frame ultraviolet spectroscopic observations in the lyman continuum region for samples of @xmath93 and @xmath94 lbgs at @xmath2 have been obtained by steidel et al . ( 2001 ) and by shapley et al . ( 2006 ) . while steidel et al . ( 2001 ) found @xmath95 , shapley et al . ( 2006 ) found an average value about @xmath96 times lower , @xmath97 , yet with a substantial variance and with @xmath72 objects showing significant emission below the lyman limit , corresponding to @xmath98 and @xmath99 . note that these authors determine @xmath88 based on a sample of quasar spectra at the same average redshift of the lbgs , and/or performing numerical simulations of the line - of - sight absorption . a large scatter of the observed ratio is expected , since the uv and the ionizing radiation are emerging from central regions of galaxies with quite complex gas and dust distributions . in fact , regions of strong star formation in local galaxies exhibit line of sights with quite variable dust attenuation and ly@xmath0 emission . in the following we adopt @xmath100 as a reference value , but we explore the range @xmath101 . considering that @xmath102 ( recall that @xmath103 ) and setting @xmath104 , the escape fraction writes @xmath105 the ratio between the intrinsic luminosities at @xmath81 and at @xmath77 for continuous star formation depends only on the imf and on the metal content . for the romano et al . ( 2002 ) imf adopted here ( see appendix a for details ) , we find @xmath106 ( see also leitherer et al . 1999 ; steidel et al . 2001 ; inoue et al . 2005 ) . in fig . [ fig|fesc_fhi ] ( top panel ) we plot @xmath107 as a function of galactic age , computed according to the above equation and assuming @xmath108 ; it is apparent that for massive halos ( @xmath109 ) the escape fraction is negligible over most of the burst duration . on the other hand , the low mass ( i.e. low intrinsic luminosity ) lbgs have large escape fractions and can thus release substantial amounts of ionizing photons ( fig . [ fig|reion_extgigi ] , left panel ) . the emission rate of ionizing @xmath81 photons at the galaxy boundary reads @xmath110 the contribution of these galaxies to the cosmic reionization is estimated in appendix b. the filling factor of hii regions is computed as a function of redshift ( see fig . [ fig|reion_extgigi ] , right panel ) for our reference attenuation [ eq . ( [ eq|extgigi ] ) ] and @xmath100 . it turns out that lbgs are able to completely reionize the universe since @xmath111 . the corresponding optical depth to electron scattering is @xmath112 , within the range of the estimate from wmap three year data ( @xmath113 , page et al . note that lowering @xmath114 to @xmath115 and @xmath116 would move the completion of reionization to @xmath117 and @xmath118 , respectively . most of the ionizing photons are produced by galaxies with halos masses in the range @xmath119 at @xmath120 . the contribution of massive galaxies is depressed by their low number density at the relevant redshifts and by their internal dust absorption . on the other hand , the contribution of the numerous low mass galaxies is damped by the decrease , driven by stellar feedback , of their sfrs . lbgs in the above mass range have very low attenuation and contribute to the faint part of the uv lf ( @xmath121 ) and to the bright end of the observed ly@xmath0 lf at @xmath122 ( see below ) . the ly@xmath0 line typically arises from ionizing photons absorbed by nearby hydrogen gas . if the fraction of absorbed ionizing photons within the galaxy is ( @xmath123 ) and @xmath124 of them are converted into ly@xmath0 photons ( osterbrock 1989 ; see also santos 2004 ) , one gets the unattenuated ly@xmath0 luminosity : @xmath125 the interstellar dust attenuates @xmath126 by a factor @xmath127 ; furthermore , only a fraction @xmath128 of ly@xmath0 photons survives the passage through the igm . therefore the luminosity we see is @xmath129 for the adopted imf ; for a salpeter imf the normalization should be reduced by a factor of about @xmath130 . since the fraction of ionizing photons surviving hi absorption is @xmath131 , eq . ( [ eq|fesc ] ) yields @xmath132 the fraction @xmath85 depends on the attenuation at @xmath77 and on the attenuation law used to extrapolate it to @xmath81 . the middle and bottom panels of fig . [ fig|fesc_fhi ] show the absorbed fraction ( @xmath123 ) and its product by @xmath127 as a function of the galactic age for the attenuation eq . ( [ eq|extgigi ] ) and the calzetti et al . ( 2000 ) attenuation law ; note that for the dust surviving fraction we have assumed @xmath133 . it is apparent that lbgs in large halos , @xmath134 , are laes only for a short time ( less than @xmath135 yr ) . the figure also illustrates the difficulty of inferring the sfr from the ly@xmath0 luminosity : even for relatively low mass halos ( @xmath51 ) the effect of hi absorption and dust attenuation may be very large . this implies that the ly@xmath0 emission from lbgs and in general the statistics of laes at high redshifts are important absorption / attenuation probes . exploiting the relationship between the ly@xmath0 luminosity , the sfr and the absorption [ eq . ( [ eq|llyaobs ] ) ] , the lfs of laes at high redshifts are estimated in the same way as the lbg lfs , and are given by an equation strictly analogous to eq . ( [ eq|uvcounts ] ) . our results are presented in fig . [ fig|lya ] ( left panel ) . it is clear from the above that the lf estimates rely on some quantities that are not very well constrained observationally . still , it may be worth noticing that the data on the lf at @xmath136 and @xmath137 are consistent with a transmission factor @xmath138 , for @xmath139 , while the lf at @xmath140 seems to require @xmath141 . this may suggest that the ionization state of the igm is changing between @xmath137 and @xmath142 ( see also kashikawa et al . we need , however , to keep in mind that the determination of @xmath128 is quite complex and implies assumptions on the relative velocity of the emitting galaxies and the igm , on the presence of galactic winds , of neutral hydrogen , etc . ( see haiman 2002 , santos 2004 and dijkstra , lidz & wyithe 2007 for comprehensive discussions ) . on the other hand , a fit of the observed lf with @xmath143 and ignoring dust attenuation yields a conservative lower limit of @xmath144 at @xmath145 ( see dijkstra , wyithe & haiman 2007 ) . figure [ fig|lya ] also shows our predictions for the lae lf at @xmath146 , when the ly@xmath0 line lies in the @xmath147 band ( see barton et al . 2006 ) , assuming @xmath148 . however , if indeed the reionization is due to lbgs , we expect that the igm ionization level can be significantly lower at this redshift , and the dimming of the ly@xmath0 luminosity substantially stronger . a clear - cut prediction of our model is that the ly@xmath0 luminosity of the most massive galaxies decreases abruptly after few @xmath149 yr , implying that , with rare exceptions , bright lbgs should exhibit relatively low ly@xmath0 luminosity ( see fig . [ fig|lya ] , right panel ) . a deficiency of objects with large ly@xmath0 equivalent width among bright lbgs has recently been reported by ando et al . ( 2006 ) . according to the model , most of the laes with ly@xmath0 luminosities sampled by the available surveys ( @xmath150 erg s@xmath151 erg s@xmath17 at @xmath152 ) are hosted by galactic halos within a rather narrow mass range @xmath153 and shining only for a short time @xmath154 yr . [ fig|sfr : mstar ] shows that the corresponding stellar masses range from a few @xmath155 to a few @xmath156 , in nice agreement with the findings by filkenstein et al . ( 2007 ) and pirzkal et al . ( 2007 ) . indeed , gawiser et al . ( 2006 ) noted that 80@xmath157 of the objects in a sample of laes at @xmath158 have the right colors to be selected as lbgs . they also inferred an average sfr around @xmath159 yr@xmath17 and an average mass in stars @xmath160 , assuming a salpeter imf . applying the correction by a factor of about @xmath72 to account for the different imfs , our model yields sfrs and stellar masses consistent with these estimates . gawiser et al . ( 2006 ) also claim that the dust attenuation is on the average minimal ( @xmath161 ) in their stacked sample . according to our model ( cf . figs . [ fig|sfr : mstar ] , [ fig|ext_magn ] , and [ fig|lya ] ) , the attenuation is almost negligible for low stellar mass objects , since both the sfr and the metal abundance are small . only for objects with @xmath162 and age larger than @xmath163 yr the attenuation becomes non negligible , @xmath164 . the dust in these objects may play an important role in determining the ew of the ly@xmath0 ( filkenstein et al . however the effect is very sensitive to the spatial distribution of dust , stars and hi . lai et al . ( 2007 ) found larger masses , from a few @xmath165 to a few @xmath166 , for @xmath38 laes at @xmath1 . the corresponding ages vary from several myr to several hundreds myr , depending on chemical composition . it has to be noticed that these authors analyzed the 3 objects with the largest fluxes at @xmath167 and @xmath96 @xmath35 m . as a consequence we expect that the selection is biased in favor of the objects with the highest stellar mass . in our framework this selection corresponds to objects with halo masses of @xmath168 , ages of about @xmath169 myr ( cf . [ fig|sfr : mstar ] ) , metallicities around @xmath170 [ cf . ( [ eq|metallicity ] ) ] and @xmath171 ( cf . [ fig|ebmv ] ) ; the predicted mass in stars amounts to @xmath172 . in conclusion , the dust prescription we derive from observations of lbgs is consistent also with the observations of laes at high redshift . we have extended to the earliest evolutionary phases of high redshift galaxies the physical model worked out by granato et al . ( 2004 ) which proved to be capable of accounting for the wealth of data on the heavily obscured active star formation phase probed by ( sub)mm surveys and on the subsequent passive evolution of spheroidal galaxies . lapi et al . ( 2006 ) have shown that this model , which features a coevolution of galaxies and active nuclei , can also provide good fits of the cosmic epoch dependent optical and x - ray lfs of active galactic nuclei . we have directly borrowed from the granato et al . ( 2004 ) model three basic ingredients : ( i ) the formation rate of galactic halos ; ( ii ) the star formation history as a function of halo mass and redshift , and the imf of stars ; ( iii ) the chemical evolution . these ingredients allow us to obtain the unattenuated spectral energy distribution as a function of galactic age . for the galaxies of interest here , which just begin to form their metals , we need a specific recipe for the attenuation . based on simple physical arguments , we have adopted a power - law relation between uv attenuation , sfr and metallicity [ eq . ( [ eq|extgigi ] ) ] , tuned to fit the uv luminosity vs. @xmath14 correlation found by shapley et al . ( 2001 ) ; this uv attenuation turns out to be increasing with galactic age and with increasing galaxy mass . with this additional ingredient the model accounts also for the observed lfs of lbgs at different redshifts . the larger attenuation for more massive objects implies that the uv lf evolves between @xmath1 and @xmath173 much less than the halo mass function . this is due to the fact that the higher dust attenuation mitigates the fast increase with decreasing redshift of the number density of massive halos . as a consequence , the uv lf is only weakly sensitive to the values of cosmological parameters , and in particular of @xmath174 . however , if lf estimates will be extended to @xmath73 , the effect of @xmath174 could be visible down to small enough masses for which the effect of attenuation is small . the derived uv lfs imply that lbg can account for a complete reionization at @xmath111 . most ionizing photons are produced by galaxies with halo masses in the range @xmath175 at @xmath176 , that dominate the faint portion of the uv lf . we expect that lbgs with detectable ly@xmath0 emission are on average much younger , less massive and less dusty , in agreement with the analysis of @xmath177 lbgs observed in the goods - s survey by pentericci et al . note , however , that a second ly@xmath0 bright phase may be expected at much later times , when the interstellar medium is being swept away by super - winds driven by the quasar feedback , allowing ly@xmath0 photons to escape from the galaxy ( see also thommes & meisenheimer 2005 ) . only a small fraction ( @xmath178 ) of lbgs at @xmath2 are detected by deep x - ray surveys ( laird et al . 2006 ; lehmer et al . x - ray luminosities estimated using stacking techniques are generally below @xmath179 , consistent with the luminosities due to the starburst activity ( type - ii sn remnants , hot gas , and high mass x - ray binaries ) . the small fraction of @xmath2 lbgs with @xmath180 generally shows properties of active nuclei . a single faint active nucleus was found by douglas et al . ( 2007 ) in a large sample of @xmath181 lbgs . our model implies that intense star formation activity is accompanied by an exponential growth of a supermassive bh at the galaxy center . the corresponding nuclear x - ray emission ( cf . granato et al . 2006 ) can be written as : @xmath182 where @xmath183 is the mass - to - radiation conversion efficiency , @xmath184 is the bolometric correction for the @xmath185 kev band and @xmath186 is the accretion rate onto the supermassive bh . in the lbg phase the nuclear bhs are still very far from their final masses ( proportional to the halo masses ) and only those associated to halos at the upper limit of the range represented in the available samples ( @xmath134 , see fig . [ fig|ext_magn ] ) can reach eddington limited accretion rates large enough ( @xmath187 yr@xmath17 ) to yield @xmath188 erg s@xmath17 ( see fig . 1 of lapi et al . 2006 ) ; the sfrs of these objects exceed @xmath189 ( fig . [ fig|sfr : mstar ] ) . interestingly , the model predicts that in more massive primeval galaxies during the late smb phase , the sfrs may reach thousands @xmath190 yr@xmath17 and correspondingly the nuclear accretion rates may attain tens @xmath190 yr@xmath17 . in fact , granato et al . ( 2006 ) have shown that the model successfully reproduces the x - ray observations of smb galaxies by alexander et al . ( 2005a , b ) and borys et al . ( 2005 ) . we find that the model reproduces the ly@xmath0 lfs at @xmath191 , @xmath192 and @xmath122 ( ouchi et al . 2003 ; shimasaku et al . 2006 ; kashikawa et al . the effect of increasing opacity with increasing mass is even stronger for laes and is responsible for the weaker evolution of the lae lf , compared to the lbg s , between @xmath53 and @xmath193 , clearly discernible in fig . [ fig|lya ] . evidences of only weak evolution of the lae lf between @xmath194 and @xmath195 have been noted by taniguchi et al . ( 2005 ; see also murayama et al . 2007 ) . according to this model , the galaxies currently selected in ly@xmath0 ( @xmath196 erg s@xmath17 ) are very young ( cf . fig . [ fig|lya ] ) , with typical ages up to @xmath197yr . the ly@xmath0 luminosity range sampled by surveys carried out so far corresponds to a narrow distribution of halo masses around @xmath198 , and stellar masses ranging from @xmath199 to @xmath200 ( see fig . [ fig|sfr : mstar ] ) . for given halo mass , ly@xmath0-selected galaxies have lower stellar masses than lbgs and their stars are preferentially formed in the densest central regions ; they are therefore expected to be more compact than lbgs , as is indeed observed ( dow - hygelund et al . galaxies with higher halo masses have higher ly@xmath0 luminosities but are rare because of the low number density of massive halos at high redshifts and of their short lifetime in the lae phase ; they are thus easily missed by the available surveys , covering very small areas . these rare objects would however have very interesting properties : high ly@xmath0 luminosity with very young ages , low metallicities and correspondingly high ly@xmath0 equivalent widths , low masses in stars within large halo masses . on the other hand , the lower luminosity portion of the lae lf is dominated by relatively older objects , more chemically evolved and with lower ly@xmath0 equivalent widths . therefore the extension to lower luminosities of the lae lf will depend rather critically on the minimum detectable equivalent width . both ly@xmath0-selected galaxies and lbgs are associated to relatively small dm halos and so that they are not expected to show strong clustering , consistent with the results by murayama et al . ( 2007 ) . [ sfrcounts ] shows the cumulative distribution of sfrs for lbgs , laes , and smbs at different redshifts . although we must be aware that there is a significant overlap among these populations , some trends clearly emerge . at high @xmath7 most of the star formation is associated to the lbg phase , but a substantial fraction of the galaxies with the highest star formation rates are predicted to be dust - obscured ( and therefore smb ) . this is because , in the present framework , the more massive galaxies have higher sfrs , yielding higher metallicities and higher dust attenuations . at redshift @xmath201 , however , the space density of smbs is very low . only later on , an increasing fraction of the cosmic sfr occurs in very dusty galaxies . since sfrs are strongly correlated with halo masses , smb galaxies are expected to show stronger clustering , consistent with the results by blain et al . observations of the lbg angular correlation function at @xmath202 ( kashikawa et al . 2006a ; lee et al . 2006 ; hildebrandt et al . 2007 ) find evidences of substantial clustering , yet not as strong as indicated by data on smb galaxies . in particular , the clustering was found to depend on luminosity , suggesting a close correlation between halo mass and sfr , in keeping with our expectations . to sum up , our model establishes a coherent scenario encompassing the variety of galaxy populations selected by different techniques : ly@xmath0 emitters , lyman break galaxies , submillimeter bright , and passively evolving galaxies . in this scenario , these populations correspond to subsequent steps in the evolutionary sequence of high-@xmath7 galaxies . we warmly thank a. bressan for enlightening discussions , and the anonymous referee for very useful comments and suggestions . this work is partially supported from asi , inaf and miur grants . alexander , d.m . 2005a , apj , 632 , 736 alexander , d.m . , et al . 2005b , nature , 434 , 738 ando , m. , et al . 2006 , apj , 645 , l9 aretxaga , i. , et al . 2007 , mnras , submitted [ preprint astro - ph/0702503 ] bardeen j. , bond j. , kaiser n. & szalay a. , 1986 , apj , 304 , 15 barton , e.j . , bullock , j.s . , cooray , a. , & kaplinghat , m. 2006 , newar , 50 , 1 blain , a.w . , chapman , s.c . , smail , i. , & ivison , r. 2004 , apj , 611 , 725 borys , c. , smail , i. , chapman , s.c . , blain , a.w . , alexander , d.m . , & ivison , r.j . 2005 , apj , 635 , 853 bouwens , r.j . , & illingworth , g.d . 2006a , newar , 50 , 152 bouwens , r.j . , & illingworth , g.d . 2006b , nature , 443 , 189 bouwens , r.j . , illingworth , g.d . , thompson , r.i . , & franx , m. 2005 , apj , 624 , l5 bunker , a.j . , stanway , e.r . , ellis , r.s . , & mcmahon , r.g . 2004 , mnras , 355 , 374 burgarella , d. , et al . 2006 , a&a , 450 , 69 calzetti , d. , armus , l. , bohlin , r.c . , kinney , a.l . , koornneef , j. , & storchi - bergmann , t. 2000 , apj , 533 , 682 caputi , k. i. , et al . 2006 , apj , 637 , 727 chabrier , g. 2005 , in the initial mass function 50 years later , eds . e. corbelli and f. palla , astrophysics and space science library vol . 327 ( dordrecht : springer ) , p. 41 chapman , s.c . , blain , a.w . , smail , i. , & ivison , r.j . 2005 , apj , 622 , 772 cirasuolo , m. , shankar , f. , granato , g.l . , de zotti , g. , & danese , l. 2005 , apj , 629 , 816 coppin , k. , et al . 2006 , mnras , 372 , 1621 cowie , l.l . , & 1998 , aj , 115 , 1319 dijkstra , m. , lidz , a. , & wyithe , s. , 2007 , mnras , submitted [ preprint astro - ph/0701667 ] dijkstra , m. , wyithe , s. , & haiman , z. 2007 , mnras , submitted [ preprint astro - ph/0611195 ] dickinson , m. , et al . 2004 , apj , 600 , l99 douglas , l.s . , bremer , m.n . , stanway , e.r . , & lehnert , m.d . 2007 , mnras , in press dow - hygelund , c.c . 2007 , apj , accepted [ preprint astro - ph/0612454 ] draine , b.t . 2003 , ara&a , 41 , 241 eyles , l.p . , 2007 , mnras , 374 , 910 eyles , l.p . , bunker , a.j . , stanway , e.r . , lacy , m. , ellis , r.s . , & doherty , m. 2005 , mnras , 364 , 443 finkelstein , s.l . , rhoads , j.e . , malhotra , s. , pirzkal , n. , & wang , j. 2007 , apj , submitted [ preprint astro - ph/0612511 ] fontana , a. , et al . 2006 , a&a , 459 , 745 gawiser , e. , et al . 2006 , apj , 642 , l13 giavalisco , m. , et al . 2004 , apj , 600 , l103 gnedin , n.y . , & ostriker , j.p . 1997 , apj , 486 , 581 granato , g.l . , silva , l. , lapi , a. , shankar , f. , de zotti , g. , & danese , l. 2006 , mnras , 368 , l72 granato , g.l . , de zotti , g. , silva , l. , bressan , a. , & danese , l. 2004 , apj , 600 , 580 grazian , a. , et al . 2007 , a&a , 465 , 393 haehnelt , & m.g . , rees , m.j . 1993 , mnras , 263 , 168 haiman , z. 2002 , apj , 576 , l1 hildebrandt , h. , et al . 2007 , a&a , 462 , 865 huang , j .- s . , rigopoulou , d. , willner , s.p . , & papovich , c. 2005 , apj , 634 , 137 iliev , i.t . , mellema , g. , pen , u .- l . , merz , h. , shapiro , p.r . , & alvarez , m.a . 2006 , mnras , 369 , 1625 inoue , a.k . , iwata , i. , & deharveng , j .- . , 2006 , mnras , 371 , l1 inoue , a.k . , iwata , i. , deharveng , j .- buat , v. , & burgarella , d. 2005 , a&a , 435 , 471o iwata , i. , et al . 2007 , mnras , in press [ preprint astro - ph/0701841 ] iye , m. , et al . 2006 , , 443 , 186 jenkins , a. , et al . 2001 , mnras , 321 , 372 jonsson , p. , cox , t.j . , primack , j.r . , & somerville , r.s . 2006 , apj , 637 , 255 kashikawa , n. , et al . 2006a , apj , 637 , 631 kashikawa , n. , et al . 2006b , apj , 648 , 7 kennicutt , r.c . 1998 , ara&a , 36 , 189 lai , k. , et al . 2007 , apj , 655 , 704 laird , e.s . , nandra , k. , hobbs , a. , & steidel , c.c . 2006 , mnras , 373 , 217 lapi , a. , shankar , f. , mao , j. , granato , g.l . , silva , l. , de zotti , g. , & danese , l. 2006 , apj , 650 , 42 lee , k .- s . 2006 , apj , 642 , 63 lehmer , b.d . , 2005 , aj , 129 , 1 leitherer , c. , et al . 1999 , apjs , 123 , 3 lowenthal , j.d . , et al . 1997 , apj , 481 , 673 madau , p. , haardt , f. , & rees , m.j . 1999 , apj , 514 , 648 magliocchetti , m. , silva , l. , lapi , a. , de zotti , g. , granato , g.l . , fadda , d. , & danese , l. 2007 , mnras , 375 , 1121 mo , h.j . , & mao , s. 2004 , mnras , 353 , 829 morgan , h.l . , & edmunds , m.g . 2003 , mnras , 343 , 427 murayama , t. , et al . 2007 , apjs , in press [ preprint astro - ph/0702458 ] osterbrock , d.e . 1989 , astrophysics of gaseous nebulae and active galactic nuclei ( univ . science books : mill valley , ca ) ouchi , m. , et al . 2004 , apj , 611 , 685 ouchi , m. , et al . 2003 , apj , 582 , 60 page , l. , et al . 2007 , apj , submitted [ preprint astro - ph/0603450 ] pentericci , l. , et al . 2007 , a&a , in press [ preprint astro - ph/0703013 ] prez - gonzlez , p.g . , et al . 2005 , apj , 630 , 82 pirzkal , n. , malhotra , s. , rhoads , j.e . , & xu , c. 2007 , apj , submitted [ preprint astro - ph/061251 ] press , w.h . , & schechter , p. , 1974 , apj , 187 , 425 romano , d. , silva , l. , matteucci , f. , & danese , l. 2002 , mnras , 334 , 444 santos , m.r . 2004 , mnras , 349 , 1137 sasaki , s. 1994 , pasj , 46 , 427 scott , s.e . , dunlop , j.s . , & serjeant , s. 2006 , mnras , 370 , 1057 shankar , f. , lapi , a. , salucci , p. , de zotti , g. , & danese , l. 2006 , apj , 643 , 14 shapiro , p.r . , & giroux , m.l . 1987 , apj , 321 , l107 shapley , a.e . , steidel , c.c . , pettini , m. , adelberger , k.l . , & erb , d.k . 2006 , apj , 651 , 688 shapley , a.e . , steidel , c.c . , adelberger , k.l . , dickinson , m. , giavalisco , m. , & pettini , m. 2001 , apj , 562 , 95 sheth , r.k . , & tormen , g. , 1999 , mnras , 308 , 119 sheth , r.k . , & tormen , g. , 2002 , mnras , 329 , 61 shimasaku , k. , et al . 2006 , pasj , 58 , 313 silva , l. , de zotti , g. , granato , g.l . , maiolino , r. , & danese , l. 2004 , a&a , submitted [ preprint astro - ph/0403166 ] silva , l. , de zotti , g. , granato , g.l . , maiolino , r. , & danese , l. 2005 , mnras , 357 , 1295 silva , l. , granato , g.l . , bressan , a. , & danese , l. 1998 , apj , 509 , 103 stark d.p . , loeb , a. , & ellis , r.s . 2007a , apj , submitted [ preprint astro - ph/0701822 ] stark d.p . , bunker , a.j . , ellis , r.s . , eyles , l.p . , & lacy , m. 2007b , apj , in press [ preprint astro - ph/0604250 ] steidel , c.c . , adelberger , k.l . , shapley , a.e . , pettini , m. , dickinson , m. , & giavalisco , m. 2003 , apj , 592 , 728 steidel , c.c . , pettini , m. , & adelberger , k.l . 2001 , apj , 546 , 665 steidel , c.c . , adelberger , k.l . , giavalisco , m. , dickinson , m. , & pettini , m. , 1999 , apj , 519 , 1 steidel , c.c . , giavalisco , m. , dickinson , m. , & adelberger , k.l . 1996 , aj , 112 , 352 steidel , c.c . , & hamilton , d. 1993 , aj , 105 , 2017 sugiyama , n. , 1995 , apjs , 100 , 281 sutherland , r.s . , & dopita , m.a . 1993 , apjs , 88 , 253 taniguchi , y. , et al . 2005 , pasj , 57 , 165 thommes , e. , & meisenheimer , k. 2005 , a&a , 430 , 877 thompson , r.i . , eisenstein , d. , fan , x. , dickinson , m. , illingworth , g. , & kennicutt , r.c . 2006 , apj , 647 , 787 vanzella , e. , et al . 2005 , a&a , 434 , 53 vanzella , e. , et al . 2006 , a&a , 454 , 423 yan , h. et al . 2006 , apj , 651 , 24 yoshida , m. , shimasaku , k. , kashikawa , n. , ouchi , m. , & okamura , s. , 2006 , apj , 653 , 988 zhao , d.h . , mo , h.j . , jing , y.p . , & brner , g. 2003 , mnras , 339 , 12 the history of star formation , of fueling of the mass reservoir around the active nucleus , and of accretion into the central bh can be easily computed by numerically solving the set of differential equations written down by granato et al . ( 2004 ; see also the appendix a of lapi et al . 2006 ) . in this appendix we summarize the results concerning the sfr , which are the relevant ones for this paper , and , whenever possible , we give simple analytical approximations for them . at the virialization time a dm halo of mass @xmath24 hosting a galaxy contains a mass @xmath203 of hot gas at the virial temperature , @xmath204 being the mean cosmic baryon to dm mass - density ratio . the gas cools and flows toward the central regions of the halo at a rate @xmath205 where @xmath206 , and the ` condensation ' timescale @xmath207 $ ] is the maximum between the dynamical time and the cooling time at the halo virial radius @xmath208 . the cooling time includes the appropriate cooling function ( sutherland & dopita 1993 ) and allows for a clumping factor @xmath209 in the baryonic component . the second term on the right hand side describes the influence of the quasar kinetic energy output on the hot gas distributed throughout the dm halo . note that in the above equation the effect of the angular momentum is neglected , since it is lost by dynamical friction through mergers of mass clouds @xmath210 on a time scale @xmath211 , where @xmath212 ( e.g. , mo & mao 2004 ) ; major mergers , which are very frequent at high redshift , imply @xmath213 a few . the mass of cold gas is increased by cooling of the hot gas ( @xmath214 ) , and decreased by star formation ( @xmath19 ) and by the energy feedback from snae ( @xmath215 ) and quasar activity ( @xmath216 ) . its evolution thus obeys the equation : @xmath217 where @xmath218 is the fraction of gas restituted to the cold component by the evolved stars , amounting to @xmath219 under the assumption of instantaneous recycling . strictly speaking , this value of @xmath218 is an upper limit , since only a fraction of evolved stars have a significant mass loss in the evolutionary phases considered here ; the results , however , are only very weakly sensitive to the chosen value , in the allowed range . the cold mass ending up in the reservoir around the central super - massive bh is lost at a very low rate ( see granato et al . 2004 ) and has been neglected in the above equation . the energy feedback from snae is parameterized as @xmath220 where the efficiency of gas removal @xmath221 depends on the number of snae per unit solar mass of condensed stars @xmath222 , on the energy per sn available to remove the cold gas @xmath223 , and on the specific binding energy of the gas within the dm halo , @xmath224 . following zhao et al . ( 2003 ) and mo & mao ( 2004 ) , the latter quantity has been estimated for @xmath225 as @xmath226\ , { \mathrm{cm}^2~\mathrm{s}^{-2}}$ ] ; here @xmath227 is the halo circular velocity at the virial radius and @xmath228 is a weak function of the halo concentration @xmath229 . we adopt the strength of sn feedback @xmath230 used by lapi et al . ( 2006 ) in order to reproduce lfs of galaxies and quasars at high redshifts , and also to reproduce the fundamental correlations between local elliptical galaxies and dormant bhs . the cold gas turns into stars at a rate @xmath231 } \approx { m_{\mathrm{cold}}\over t_{\star}}~,\ ] ] where now @xmath232 and @xmath233 refer to the mass shell @xmath234 , and @xmath235 is the star formation timescale averaged over the mass distribution . ( [ eq|mdotinf ] ) and ( [ eq|mdotcold ] ) can be easily solved setting @xmath236 and neglecting the effects of the energy injected in the gas by the accretion onto the central bh . with this approximation the infalling mass declines exponentially as @xmath237 , while the sfr evolves according to @xmath238~,\ ] ] with @xmath239 . the quantity @xmath240 is the ratio between the timescale for the large - scale infall estimated at the virial radius , and the star formation timescale in the central region . an isothermal density profile yields @xmath241 . although eq . ( [ eq|sfr ] ) has been obtained neglecting the quasar feedback , it turns out to be , at the redshifts relevant here ( @xmath25 ) , a good approximation of the results of the full numerical integration of the system of differential equations given by lapi et al . ( 2006 ) , provided that : ( i ) we take into account that the star formation is strongly suppressed by the quasar feedback after a time @xmath242 where @xmath243 for @xmath244 and @xmath245 for @xmath246 ; ( ii ) we adopt the following expression for the condensation timescale @xmath247 in the latter expression , the scaling with the redshift mirrors the dependence of the cooling and/or dynamical time on redshift . moreover , the weak dependence on @xmath24 renders the impact of the energy feedback from the quasar on the infalling gas ; its effect is mimicked by allowing longer condensation timescales for the gas in more massive halos , exposed to stronger quasar feedback from more massive bhs . note that the sfr evolves differently for different halo masses , see fig . [ fig|sfr : mstar ] . for large masses @xmath134 , it increases almost linearly with galaxy age in the initial stages @xmath248 , and then it is suddendly stopped by the energy feedback from the quasar at @xmath249 . on the contrary , for low - mass halos @xmath250 it is first almost constant and then slowly declining ; indeed , in such low mass galaxies the quasar feedback is very mild because the mutual effects of star formation and gas flow toward the reservoir around the central bh result in a strong decline of the bh mass with decreasing halo mass [ e.g. , shankar et al . 2006 , their eq . ( 16 ) ] ; thus the sfr can proceed for much longer times ( note from eq . ( a7 ) that @xmath251 gyr at @xmath1 for @xmath51 ) . note that fig . [ fig|sfr : mstar ] refers to redshift @xmath53 ; at @xmath2 the sfr retains the same time dependence as at @xmath53 , but is on average lower by a factor of @xmath252 for a given halo mass . the total mass @xmath253 cycled through stars during the time @xmath254 can be estimated by integrating the sfr [ eq . ( [ eq|sfr ] ) ] . the present - day total mass of surviving stars @xmath255 can then be derived by assuming a specific imf ; @xmath256 is around @xmath257 after @xmath13 gyr from the burst for a salpeter imf , and is around @xmath22 for the imf adopted here . the latter is a double power - law with slope @xmath258 from @xmath259 to @xmath260 and @xmath261 from @xmath260 down to @xmath262 ( romano et al . 2002 ; for a review on the imf see chabrier 2005 ) . an interesting outcome of our model is that the fraction of the original baryons associated to the galactic halo that is trapped in present - day stars stays almost constant at @xmath263 for @xmath264 , and decreases very rapidly below this threshold . as discussed by shankar et al . ( 2006 ) , this is in keeping with results obtained through x - ray and weak lensing estimates of halos around galaxies . when computing the metal content of the cold gas , one has to take into account the enrichment of the primordial infalling gas due to earlier generations of stars , and the gas outflows due to the energy injection by snae and quasars . in the first stages , the chemical enrichment of the cold gas component is very rapid ; e.g. , @xmath265 and @xmath266 the solar abundance in the gaseous component is attained after @xmath267 and @xmath268 yr , almost independently of the halo mass and redshift . only less than @xmath269 and @xmath270 of the total stellar mass has been built up at these times and therefore the expected final mass fraction of metal - poor stars is tiny . this rapid enrichment is due to the short lifetime of massive stars @xmath271 ( that are efficient metal producers ) relative to the timescale @xmath272 of the infall of the diffuse medium with primordial composition , which dilutes the cold star - forming gas . a simple approximation of the subsequent evolution reads @xmath273~,\ ] ] where @xmath274 yr is a characteristic time after which @xmath23 keeps almost constant . in this appendix we briefly describe the computation of the reionization redshift predicted by our model . the average global production rate of ionizing photons per unit volume at the cosmic time @xmath27 is @xmath275 where @xmath276 is given by eq . ( [ eq|ndotion ] ) and its distribution function @xmath277 can be derived in strict analogy to eq . ( [ eq|uvcounts ] ) . the transition from a neutral to a reionized igm can be described in terms of the evolution of the volume filling factor of hii regions , @xmath278 , which is ruled by the equation ( shapiro & giroux 1987 ; madau , haardt & rees 1999 ) @xmath279 here @xmath280 is the comoving @xmath87 number density , and the recombination time is given by ( madau , haardt & rees 1999 ) @xmath281 in terms of the clumping factor of the igm , @xmath282 . the igm clumping factor has been investigated through hydrodynamical simulations ( e.g. , gnedin & ostriker 1997 ) . iliev et al . ( 2006 ) proposed an analytic description of the simulation results : @xmath283 . in the relevant redshift interval @xmath284 one finds @xmath285 . here we adopt @xmath286 as a reference value . reionization is completed at the redshift when @xmath287 ( right panel of fig . [ fig|reion_extgigi ] ) . finally , the optical depth to electron scattering is given by @xmath288 where @xmath289 is the electron density , @xmath290 @xmath291 is the present day baryon density , and @xmath292 @xmath293 is the thomson cross section .

we give a global picture of the ricci flow on the space of three - dimensional , unimodular , nonabelian metric lie algebras considered up to isometry and scaling . the ricci flow is viewed as a two - dimensional dynamical system for the evolution of structure constants of the metric lie algebra with respect to an evolving orthonormal frame . this system is amenable to direct phase plane analysis , and we find that the fixed points and special trajectories in the phase plane correspond to special metric lie algebras , including ricci solitons and special riemannian submersions . these results are one way to unify the study of ricci flow on left invariant metrics on three - dimensional , simply - connected , unimodular lie groups , which had previously been studied by a case - by - case analysis of the different bianchi classes . in an appendix , we prove a characterization of the space of three - dimensional , unimodular , nonabelian metric lie algebras modulo isometry and scaling . = 1

0909.0938

in this paper we study the ricci flow on three - dimensional , unimodular metric lie algebras . metric lie algebras are in one - to - one correspondence with left - invariant riemannian metrics on simply - connected lie groups , and ricci flow on such metrics has been studied by a number of authors ( e.g. , @xcite @xcite @xcite @xcite @xcite @xcite @xcite @xcite ) . the major advances in this paper are ( 1 ) a unification of the trajectories for the ricci flow , previously viewed individually in case - by - case studies of bianchi classes , into a single global topological picture , and ( 2 ) use of a new technique of flowing the lie structure constants , which highlights different features of the system than the usual evolution of metric coefficients . the space of metric lie algebras has been studied by a number of authors ( e.g. , @xcite @xcite @xcite ) . understanding ricci flow on the space of metric lie algebras is important for studying both homogeneous spaces and ricci flow of general manifolds . a number of ricci soliton metrics ( fixed points of the ricci flow up to diffeomorphism invariance and rescaling ) have been found on homogeneous spaces ( see , e.g. , @xcite @xcite @xcite @xcite @xcite @xcite ) , and it has been suggested that finding ricci solitons may be a promising way to attack alekseevskii s conjecture ( see @xcite and @xcite ) . lott has shown that three - dimensional , type iii solutions to ricci flow converge to the known homogeneous expanding solitons as they collapse in the limit ( see @xcite ) . ricci flow on homogeneous spaces is also useful in constructing self - dual solutions of euclidean vacuum einstein s equations ( see @xcite ) . we will consider the set @xmath0 of three - dimensional , nonabelian , unimodular metric lie algebras modulo isometry and scaling . milnor gives an excellent description of such metric lie algebras in @xcite , in particular showing that there exists a special orthonormal basis @xmath1 which diagonalizes both the ricci endomorphism and the lie bracket ( we say that the lie bracket is diagonalized if @xmath2 $ ] is a scalar multiple of @xmath3 ) . thus the set of three - dimensional , unimodular metric lie algebras depends only on three parameters . in fact , there are two natural choices of those three parameters , and the ricci flow through these parameter spaces takes one of the following forms : 1 . fix the lie algebra and let the metric vary . 2 . evolve the frame to keep it orthonormal and let the structure constants vary . in both cases , the lie bracket and inner product remain diagonal with respect to the frame . however , in the first case the lie bracket coefficients are fixed and the lengths of basis elements change . in the second case , the lie bracket coefficients change but the lengths of basis elements do not change ( since the basis evolves to stay orthonormal ) . it is extremely important that the frame remains orthogonal under the flow , which follows from the fact that both the structure constants and the ricci curvature can always be diagonalized at the same time as the metric . this is true for three - dimensional , unimodular metric lie algebras , but not in general . the lack of such a frame is the major obstacle for classifying ricci flow on four - dimensional , simply - connected homogeneous spaces ; isenberg - jackson - lu @xcite classify ricci flow for some riemannian homogeneous spaces which do admit such a frame . since @xmath0 is a three - dimensional space considered modulo rescaling , we have a two - dimensional system of odes , which is reasonable to analyze as a dynamical system in the plane . most previous work on ricci flow on homogeneous spaces takes the first parametrization ( e.g. , @xcite , @xcite , @xcite ) . in contrast , we will take the second parametrization , and consider @xmath0 as a quotient of the space of structure constants . this method has previously been used by the second author to study ricci flow on nilmanifolds @xcite ( see also @xcite and @xcite ) . let @xmath4 denote the flow on @xmath0 determined by the ricci flow . theorem a describes the topological dynamics of the flow @xmath5 [ theorem a]the phase space @xmath0 is the disjoint union of the following invariant sets ( see figures [ figure1 ] and [ figure2 ] ) : * four points @xmath6 and @xmath7 * six one - dimensional trajectories @xmath8 and @xmath9 ; and * three connected two - dimensional open sets @xmath10 and @xmath11 ; such that * the points @xmath6 and @xmath12 are fixed by @xmath13 * the orbit of a point @xmath14 in a @xmath15 or @xmath16 has @xmath17 and @xmath18 and * the orbit of a point @xmath14 in @xmath19 or @xmath20 has @xmath17 and @xmath21 theorem b interprets theorem a geometrically . in the sequel , we will say a point in the phase space represents a particular metric , although we actually mean that it represents the homothety class of the metric , i.e. , the equivalence class of the metric up to isometry and scaling . the decomposition of the phase space @xmath0 in theorem a corresponds geometrically as follows . 1 . each of the four fixed points @xmath6 and @xmath12 represents a soliton metric : * @xmath22 represents the soliton metric on the three - dimensional heisenberg group @xmath23 * @xmath24 represents the soliton metric on the three - dimensional solvable group @xmath25 * @xmath26 represents the flat metric on the three - dimensional euclidean group @xmath27 * @xmath12 represents the round metric on the group @xmath28 . 2 . the five trajectories @xmath29 @xmath30 @xmath31 @xmath32 and @xmath9 have riemannian submersion structures : * @xmath33 consists of left - invariant metrics on @xmath34 ( often denoted @xmath35 ) . these metrics fiber as riemannian submersions over @xmath36 . * @xmath37 consists of left - invariant metrics on @xmath38 . these metrics fiber as riemannian submersions over @xmath36 . * @xmath39 consists of left - invariant metrics on @xmath40 which fiber as riemannian submersions over the hyperbolic plane @xmath41 * @xmath42 and @xmath9 consist of left - invariant metrics on @xmath43 which fiber as riemannian submersions over the round sphere @xmath44 ( these riemannian manifolds are often called berger spheres ) . the trajectory @xmath42 corresponds to submersions whose fibers are larger than those of the round 3-sphere ( corresponding to the point @xmath12 ) and the trajectory @xmath9 corresponds to submersions whose fibers are smaller than those of the round 3-sphere . 3 . the three connected open sets @xmath10 and @xmath11 have the structures : * @xmath45 consists of left - invariant metrics on @xmath28 . * @xmath46 and @xmath11 consist of left - invariant metrics on @xmath47 . note that the trajectory @xmath48 is still somewhat mysterious . this trajectory was discovered independently by cao , guckenheimer , and saloff - coste @xcite , and evidence for it was present in @xcite . preliminary evidence suggests that this trajectory is not invariant under cross curvature flow , which may indicate it does not arise from extra symmetries of the riemannian metric , as the other special orbits do ( see remark [ remark symmetry ] ) . the organization of this paper is as follows . in section [ section space of unimod ] , we introduce notation and discuss the space of three - dimensional , unimodular metric lie algebras and their curvatures . in section [ section rf ] we derive the ricci flow equations on the space of structure constants . in section [ dynamics ] we analyze the dynamics of the ricci flow equations on a natural phase space and then on @xmath0 , completing the proof of theorem a. in section [ geometry section ] we analyze the dynamics geometrically , relating fixed points and special trajectories to ricci solitons and riemannian submersions , proving theorem b. in section [ section convergence ] we discuss how our convergence results relate to the current literature on ricci flows on three - dimensional , unimodular metric lie algebras and lie groups . finally , we include an appendix which give the details of characterizing the space @xmath0 . ( 40,60)(0,0 ) ( 10,0)(0,1)30 ( 30,30)(0,1)20 ( 10,30)(1,0)20 ( 10,0)(2,3)20 ( 10,30)(1,1)20 ( 10,15)(4,3)20 ( 10,0 ) ( 10,15 ) ( 10,30 ) ( 30,30 ) ( 30,50 ) ( 10,8)(0,1)0.8 ( 10,22)(0,-1)0.8 ( 30,40)(0,1)0.8 ( 20,30)(1,0)0.8 ( 22,24)(4,3)0.8 ( 20,15)(2,3)0.8 ( 20,40)(1,1)0.8 ( 6,0)@xmath49 ( 6,15)@xmath50 ( 6,30)@xmath49 ( 32,30)@xmath51 ( 32,50)@xmath52 ( 4,22)@xmath33 ( 4,8)@xmath33 ( 20,12)@xmath53 ( 14,22)@xmath48 ( 20,31)@xmath37 ( 16,42)@xmath42 ( 31,40)@xmath9 ( 22,37)@xmath45 ( 16,27)@xmath46 ( 12,13)@xmath54 with ricci flow lines . ] consider the following definitions . a _ metric lie algebra _ @xmath55 is a lie algebra @xmath56 together with an inner product @xmath57 on @xmath56 . the dimension of the metric lie algebra is the dimension of @xmath56 , and it is _ unimodular _ or _ nonabelian _ if the lie algebra @xmath56 is unimodular ( i.e. , @xmath58 is trace free for all @xmath59 ) or nonabelian , respectively . recall that there is a one - to - one correspondence between lie algebras @xmath56 and simply - connected lie groups @xmath60 furthermore , there is a one - to - one correspondence between metric lie algebras @xmath55 and simply - connected lie groups @xmath61 with a left - invariant riemannian metric @xmath62 we will use the riemannian manifold @xmath63 corresponding to the metric lie algebra @xmath64 for the following definition . [ isometry def]a linear map @xmath65 is an _ isometry _ of metric lie algebras @xmath66 and @xmath67 if it the differential of a riemannian isometry @xmath68 between the corresponding simply - connected lie groups with induced left - invariant metrics . two metrics @xmath57 and @xmath69 on @xmath56 are _ homothetic _ if @xmath70 for some @xmath71 we say @xmath66 and @xmath67 are _ equivalent up to isometry and scaling _ if there exists a diffeomorphism @xmath72 and @xmath73 such that @xmath74 the space of three - dimensional , nonabelian , unimodular metric lie algebras modulo isometry and scaling will be denoted by @xmath0 . [ remark 1]a related definition is that of isomorphism between metric lie algebras . one says two metric lie algebras @xmath75 and @xmath76 are _ isomorphic _ if there is a linear map @xmath65 which is a lie algebra isomorphism and satisfies @xmath77 the notion of isomorphism in this definition is stronger than the notion of isometry in definition [ isometry def ] . that is , given an isomorphism of metric lie algebras , one can use the group action to extend this to an isometry of the corresponding simply - connected lie groups with corresponding left - invariant metrics . however , it is possible to have isometries of the groups which are not isomorphisms of metric lie algebras . for instance , there is a flat metric @xmath78 on the lie group @xmath79 of euclidean transformations ( as well as its universal cover ) , and there is a flat metric @xmath80 on the abelian group @xmath81 there is an isometry between @xmath82 and @xmath83 but this isometry is not a group isomorphism , and so its differential is not an isomorphism of metric lie algebras . in the beautiful paper of milnor @xcite , geometric properties of left - invariant metrics on three - dimensional lie groups are studied in detail . milnor computes the curvatures of three - dimensional , unimodular metric lie algebras : [ ( * ? ? ? * theorem 4.3)][mil curvature thm]suppose @xmath64 is a three - dimensional , unimodular metric lie algebra . then there exists a @xmath57-orthonormal frame @xmath84 for @xmath56 such that lie brackets for @xmath56 are determined by@xmath85=a_{1}e_{1},\qquad\lbrack e_{3},e_{1}]=a_{2}e_{2},\qquad\text{and}\qquad\lbrack e_{1},e_{2}]=a_{3}e_{3 } , \label{diagonalize}\ ] ] for some constants @xmath86 furthermore , this basis diagonalizes the ricci endomorphism @xmath87 such that @xmath88_{\mathcal{b}}=2\begin{bmatrix } \mu_{2}\mu_{3 } & & \\ & \mu_{1}\mu_{3 } & \\ & & \mu_{1}\mu_{2}\end{bmatrix } , \ ] ] where @xmath89 the sectional curvatures are given by @xmath90 scalar curvature @xmath91 is @xmath92 milnor also describes the isomorphism type of a lie algebra determined by equations ( [ diagonalize ] ) based on the signs of @xmath93 ; see table [ milnor table ] ( where if the signs are all multiplied by @xmath94 the lie algebra is the same ) . [ c]|l|l|l|signs of @xmath95 & associated lie algebra & associated lie groups + @xmath96 & @xmath97 & @xmath43 or @xmath98 + @xmath99 & @xmath100 & @xmath101 or @xmath102 or @xmath103 + @xmath104 & @xmath105 & @xmath106 + @xmath107 & @xmath108 & @xmath109 + @xmath110 & @xmath111 & @xmath112 + recalling the space @xmath0 of metric lie algebras from definition [ isometry def ] , according to theorem [ mil curvature thm ] we have the following lemma . [ lemma m tilde]let @xmath113 be the map which takes @xmath114 to the equivalence class of the metric lie algebra defined by an orthonormal basis @xmath1 with lie bracket determined by ( [ diagonalize ] ) . the map @xmath115 is surjective . notice that we have excluded the point @xmath116 representing the abelian lie algebra @xmath117 , from the domain of @xmath115 . this will allow @xmath115 to descend to a map from @xmath118 to @xmath0 so that we can consider metric lie algebra equivalence up to scaling . certainly the above definition defines a lie bracket , and theorem [ mil curvature thm ] shows that every three - dimensional , unimodular lie algebra can be written in this way . since @xmath0 is a quotient of the set of all such lie algebras , the result follows . we would like to describe the space @xmath0 using a fundamental domain . since @xmath115 descends to a map from @xmath118 to @xmath119 we start with the coordinates @xmath120 on @xmath118 . let @xmath121 and let @xmath122 be the equivalence relation on @xmath123 that is determined by @xmath124 if @xmath125 since a fundamental domain should be compact , we need to compactify @xmath126 and so we also introduce the compact set @xmath127 which is given the one - point - compactification topology , i.e. , open neighborhoods of @xmath128 consist of the complements of compact subsets of @xmath129 we can extend @xmath122 to an equivalence relation on @xmath130 by adding the equivalence @xmath131 in the appendix , we prove the following : [ fundamental domain theorem]there is a bijection between @xmath132 and @xmath0 . using the quotient topology on @xmath133 there is a natural topology on @xmath0 which makes this map a homeomorphism . the proof of theorem [ fundamental domain theorem ] , as well as some of discussion in section [ section convergence ] , requires the covariant derivatives of the ricci tensor , which can be derived in a straightforward way from the formulas in theorem [ mil curvature thm ] . they also appear in @xcite . [ drc]suppose @xmath55 is a three - dimensional , unimodular metric lie algebra . the covariant derivative of the ricci operator satisfies@xmath134 where @xmath135 are as in theorem [ mil curvature thm ] . we now derive the equations for ricci flow on @xmath0 . recall that on a riemannian manifold @xmath136 the ricci flow is the solution to the equations @xmath137 for a left - invariant metric on a lie group @xmath61 , this flow reduces to a flow of the inner product @xmath57 on the lie algebra @xmath56 of @xmath61 ; that is , it reduces to a flow of metric lie algebras @xmath138 recall that theorem [ mil curvature thm ] implies that for any inner product @xmath57 on a three - dimensional , unimodular lie algebra , we can find a @xmath57-orthonormal basis @xmath139 for @xmath56 which diagonalizes the ricci tensor . we will see two ways to formulate the ricci flow : 1 . [ case 1 evol]fix the basis @xmath140 from theorem [ mil curvature thm ] which is orthonormal with respect to @xmath141 the initial inner product , and consider the evolution of the metric coefficients @xmath142 in this case , the structure constants @xmath143 with respect to the basis @xmath144 , i.e. , @xmath145=\lambda_{1}\bar{e}_{1},\qquad\lbrack\bar { e}_{3},\bar{e}_{1}]=\lambda_{2}\bar{e}_{2},\qquad\text{and}\qquad\lbrack \bar{e}_{1},\bar{e}_{2}]=\lambda_{3}\bar{e}_{3 } , \label{lambda identity}\ ] ] are fixed in time . 2 . [ case 2 evol]evolve the basis @xmath146 to be orthonormal with respect to @xmath147 so that the metric is the identity in this frame , but the structure constants @xmath93 with respect to @xmath148 , which are of the form ( [ diagonalize ] ) , depend on time . in general , if @xmath140 is a basis which is orthogonal with respect to a metric @xmath149 and satisfies ( [ lambda identity ] ) , and @xmath150 for @xmath151 then we see that @xmath152 , where @xmath153 is orthonormal and the structure constants @xmath154 ( as in theorem [ mil curvature thm ] ) are related by @xmath155 this is how one can relate the solution flows in formulations [ case 1 evol ] and [ case 2 evol ] . as first described in @xcite , in formulation [ case 1 evol ] the ricci curvature is diagonal with respect to the initial metric @xmath141 and so the following ricci flow evolution can be derived for @xmath156 ( see formulas from theorem [ mil curvature thm]):@xmath157 where the @xmath154 are explicit functions of the @xmath158 ( with fixed parameters @xmath159 ) defined by ( [ a lambda corresp ] ) . thus the equations ( [ q - rf equation ] ) form an autonomous system of odes in the variables @xmath160 since the fixed basis @xmath144 is orthogonal with respect to @xmath161 ( as determined by @xmath162 ) and ( [ lambda identity ] ) continues to be satisfied at each time @xmath163 , we see that the flow ( [ q - rf equation ] ) really is the ricci flow for all time . the fact that the flow remains diagonal is a special property of three - dimensional , unimodular metric lie algebras , and is not true in general ( see , e.g. , @xcite ) . noting that the right sides of the odes ( [ q - rf equation ] ) only contain the @xmath154 , without explicitly containing the @xmath158 , it is natural to consider formulation [ case 2 evol ] . the evolution for @xmath154 are easily derived using ( [ q - rf equation ] ) and ( [ a lambda corresp ] ) . due to theorem [ fundamental domain theorem ] ( and the preceding discussion from section [ section space of unimod ] ) , we will also be interested in @xmath164 we have the following . [ flow - equations - unimodular ] let @xmath61 be a simply - connected , three - dimensional , nonabelian , unimodular lie group with left - invariant metric @xmath62 then the ricci flow on @xmath61 with initial metric @xmath165 corresponds to a flow of metric lie algebras @xmath166 where @xmath56 is the lie algebra of @xmath61 and @xmath167 is @xmath165 restricted to @xmath168 . this flow can be realized as a flow of structure constants @xmath93 ( as determined by theorem [ mil curvature thm ] ) , and , if we suppose that @xmath169 the ratios @xmath170 and @xmath171 ( as defined in ( [ m definition ] ) ) obey the equations@xmath172 we also note that if @xmath169 @xmath173 = 2k(e_{2}\wedge e_{3}).\ ] ] the reader may be troubled that the expressions for @xmath174 and @xmath175 in the proposition are not solely functions of @xmath170 and @xmath171 ; however , they are useful , because we will be interested in imagining slope fields for the flow of @xmath170 and @xmath171 in the @xmath170-@xmath171 plane . the common ( positive ) term @xmath176 simply affects the speed of motion and not the direction of motion , so the trajectories of @xmath177 under the odes ( [ m ode bad ] ) are the same as the trajectories of the autonomous odes @xmath178 the translation to the metric lie algebra was described at the beginning of this section . observe that @xmath179 now calculate @xmath174 using ( [ q - rf equation ] ) as : @xmath180 the formula for @xmath175 follows analogously . it is clear that the ricci flow equations ( [ q - rf equation ] ) for the @xmath158 determine the @xmath154 by ( [ a lambda corresp ] ) , however one might ask if the ricci flow equations for the @xmath154 determine the @xmath181 this is , in fact , true , since once the @xmath154 are an explicit function of @xmath182 one can determine the @xmath158 by explicitly integrating ( [ q - rf equation ] ) , which are now explicit functions of @xmath183 this was first observed in @xcite . ) . ] in this section we will look at the qualitative behavior of the dynamical system ( [ rf equation ] ) . the phase plane for the system of odes ( [ rf equation ] ) is displayed in figure [ figure3 ] , as computed in matlab . first , we consider the fixed points . since the right sides of each equation factor into linear terms , it is easy to see that the fixed points of @xmath185 are @xmath186 also , it is not hard to see that the following curves are preserved by the flow : ( i ) @xmath187 ( ii ) @xmath188 ( iii ) @xmath189 ( iv ) @xmath190 and ( v ) @xmath191 the linearization of the system ( [ rf equation ] ) is the matrix @xmath192{cc}2m_{2}m_{3}-m_{3}-3m_{2}^{2}+1 & m_{2}\left ( m_{2}-1\right ) \\ m_{3}\left ( m_{3}-1\right ) & 2m_{2}m_{3}-m_{2}-3m_{3}^{2}+1 \end{array } \right)\ ] ] and so we see that : 1 . @xmath193 is an unstable fixed point . 2 . @xmath194 is a stable fixed point . 3 . @xmath195 and @xmath196 are saddle points . they have stable manifolds tangent to the lines determined by the eigenvectors @xmath197 and @xmath198 respectively . they have unstable manifolds tangent to the lines determined by the eigenvectors @xmath199 and @xmath200 4 . @xmath197 and @xmath198 are degenerate fixed points . they have stable manifolds tangent to the lines determined by the eigenvectors @xmath197 and @xmath198 respectively . they have a zero eigenvalue corresponding to eigenvectors @xmath201 and @xmath197 ; furthermore , one can look in the zero directions by considering the taylor series of solutions near @xmath202 for the flow along curves @xmath203 and @xmath204 . we see that , for instance , points near @xmath198 and below the @xmath205-axis approach the fixed point , while points above the @xmath205-axis move away from the fixed point , as seen in figure [ figure3 ] . in addition , since @xmath195 and @xmath196 are saddle points ( i.e. , the linearizations at these points each have two distinct eigenvalues of opposite sign ) , the stable manifold theorem ( see , e.g. , @xcite ) implies each has a one - dimensional unstable manifold . although we are unable to calculate the trajectories explicitly , we can , for instance , compute a taylor approximation of the curve at @xmath206 to be@xmath207 furthermore , by the hartman - grobman theorem ( @xcite ) , the trajectories of the differential equation in a neighborhood of @xmath208 are homeomorphic to the trajectories of the linearization around @xmath209 and so this curve contains the only trajectory in the fourth quadrant with @xmath210 which contains @xmath211 we expect that the taylor series ( [ taylor series ] ) has radius of convergence 1 , since the curve has a vertical tangent at the point @xmath212 in this section we prove theorem a. [ proof of theorem a]by theorem [ fundamental domain theorem ] , we can restrict our attention to @xmath123 and consider it up to the equivalence @xmath122 determined by equation ( [ tilde def ] ) . the discussion in section [ section r^3 dynamics ] implies that we have the following fixed points in @xmath126 none of which are equivalent in @xmath213 : @xmath214 , @xmath215 @xmath216 and @xmath217 it is also clear that for any sequence @xmath218 with @xmath219 and @xmath220 we have @xmath221 in @xmath222 define the sets @xmath223 notice that these are , in fact , trajectories of the odes ( [ rf equation ] ) , and correspond to the invariant sets described in section [ section r^3 dynamics ] . the special trajectory @xmath48 is defined as the unstable manifold of the point @xmath24 restricted to the set @xmath224 as an unstable manifold , it must be invariant . consider the set @xmath225 on this set , it follows from ( [ rf equation ] ) that if @xmath226 then @xmath227 is in the positive horizontal direction ( i.e. , @xmath228 and @xmath229 ) . thus the set @xmath230 is invariant . since on this set we have@xmath231 we see that the trajectory @xmath48 can be written as @xmath232 where @xmath233 is a continuous , increasing function for @xmath234 such that @xmath235 and @xmath236 the remaining sets in the partition of @xmath123 are @xmath237 these sets are invariant under the flow of the odes since their boundaries are invariant . it remains to show that the sets have the appropriate forward and backwards limit properties ; a straightforward analysis of the phase diagram completes the proof . in this section we compare the results of theorem a with the known geometry of three - dimensional , unimodular metric lie algebras , thereby proving theorem b. [ proof of theorem b]we first look at the fixed points @xmath238 recall that for a three - dimensional , unimodular metric lie algebra @xmath239 we have a basis @xmath240 as described in theorem [ mil curvature thm ] , and recall that @xmath241 1 . @xmath242 this corresponds to the lie algebra with structure constants @xmath243 and @xmath244 by table [ milnor table ] we see that this point corresponds to the heisenberg lie algebra @xmath245 up to rescaling , there is only one metric lie algebra corresponding to @xmath111 , and so we see that @xmath22 must be that point . this point corresponds to the ricci soliton on @xmath246 found by lauret @xcite , baird - danielo @xcite , and lott @xcite . 2 . @xmath247 this corresponds to the lie algebra with structure constants @xmath243 , @xmath248 and @xmath249 by table [ milnor table ] we see that @xmath250 and @xmath251 determines that this point corresponds to the solvable lie algebra @xmath252 consulting @xcite , we see that the left - invariant riemannian metrics on @xmath34 ( referred to as @xmath35 in the reference ) have the form @xmath253 and the soliton found by baird - danielo @xcite and lott @xcite occurs when @xmath254 since at the point @xmath255 we have @xmath256 we see that @xmath257 and so @xmath24 corresponds to this soliton metric . note that among metrics ( [ sol metrics ] ) , the soliton has the form @xmath258 which has the additional symmetry of switching @xmath205 and @xmath259 also note that switching @xmath205 and @xmath260 corresponds with switching @xmath261 and @xmath262 and thus gives precisely the isometry that identifies the metric lie algebras corresponding to the sets @xmath263 and @xmath264 when using the equivalence relation on @xmath123 from ( [ tilde def ] ) . 3 . @xmath265 this corresponds to the lie algebra with structure constants @xmath266 and @xmath267 by table [ milnor table ] we see that @xmath268 and @xmath269 determines that this point corresponds to the solvable lie algebra @xmath270 consulting @xcite , we see that the left - invariant riemannian metrics on @xmath271 ( referred to as @xmath272 in the reference ) have the form @xmath273 with @xmath274 determining the flat metric . we see that @xmath26 corresponds to the flat metric . note that the flat metric is the maximally symmetric metric of the type ( [ e2 metrics ] ) . 4 . @xmath217 this corresponds to the lie algebra @xmath275 with @xmath276 implying that @xmath277 one easily sees from theorem [ mil curvature thm ] that this metric has constant sectional curvature , and thus it corresponds to the round metric on the 3-sphere . again , we notice that the round metric is maximally symmetric among all left - invariant metrics on @xmath278 we now look at the special trajectories . it is clear that @xmath33 and @xmath37 correspond to metrics on @xmath34 and @xmath271 respectively , and the explicit metrics shown in ( [ sol metrics ] ) and ( [ e2 metrics ] ) show the riemannian submersion structures . in @xcite , the left - invariant metrics on @xmath279 are given explicitly , and one sees that on the trajectory @xmath39 we have @xmath280 indicating that @xmath274 and that the metrics have the form @xmath281 where @xmath282 . these metrics clearly have the form of riemannian submersions over the hyperbolic plane . now consider the trajectories @xmath42 and @xmath283 which we see correspond to metrics on @xmath278 each element in the basis @xmath1 exponentiates to a compact group @xmath284 of rotations in @xmath278 the quotient @xmath285 is diffeomorphic to the sphere @xmath286 and the map @xmath287 is precisely the hopf fibration ( see , e.g. , @xcite ) . using 9.79 and 9.80 in @xcite , this can be made into a riemannian submersion from a left - invariant metric on @xmath43 to a @xmath43-invariant metric on @xmath288 with totally geodesic fibers . the berger spheres are the metrics on @xmath43 which make @xmath289 a riemannian submersion , where @xmath285 is given the round metric on @xmath290 the remaining basis elements span the horizontal subspace of the submersion , and thus must have equal length for @xmath289 to be a riemannian submersion . thus the submersions are represented only when @xmath291 ( so @xmath274 ) or @xmath292 ( so @xmath293 ) or @xmath294 ( so @xmath295 ) . the round sphere is when @xmath277 it is now easy to see that @xmath42 corresponds to the fibers being larger than the fibers for the round sphere : @xmath295 and @xmath296 so we have @xmath297 similarly , on @xmath283 we have @xmath274 and @xmath298 so @xmath299 and the fibers are smaller than the fibers of the round sphere . the descriptions of the basins @xmath300 @xmath11 , and @xmath45 follow immediately from table [ milnor table ] . we could have constructed the riemannian submersions on @xmath279 in the same way we constructed the ones for @xmath43 as follows . instead of considering @xmath301 consider @xmath302 the orientation preserving isometries of @xmath41 there is a compact subgroup @xmath284 acting on the upper half - plane by the isometries @xmath303 for all angles @xmath304 note that this is the isotropy group of the point @xmath305 again , by 9.79 and 9.80 in @xcite , there is a riemannian submersion @xmath306 with totally geodesic fibers , where @xmath307 is given the geometry of @xmath41 this submersion may be lifted to the universal cover to get a line bundle over @xmath41 [ remark symmetry]it is an interesting fact that each of the special points and trajectories except for @xmath48 correspond to metrics with additional symmetry . generically , the left - invariant metrics have isometry groups of dimension @xmath308 the point @xmath24 is maximally symmetric among left - invariant metrics on @xmath34 , containing one extra symmetry , as described in the proof of theorem b. the points @xmath26 and @xmath12 have @xmath309-dimensional symmetry groups . the trajectories @xmath31 @xmath32 and @xmath9 have the additional symmetries of reparametrizing the fibers : if the fibration structure is locally trivialized as @xmath310 then the map @xmath311 is an isometry . note that reparametrizing the fiber is different than multiplication by the generator of the fiber in the group . these extra symmetries give @xmath31 @xmath32 and @xmath9 four - dimensional isometry groups . in this section , we compare the convergence results of this paper with other convergence results for ricci flow on three - dimensional , unimodular lie groups . let @xmath55 be a metric lie algebra , and let @xmath63 be the corresponding simply - connected lie group with left - invariant riemannian metric @xmath62 any riemannian manifold @xmath312 determines a metric space @xmath313 where @xmath314 is the induced riemannian distance function . often it will be relevant to consider quotients @xmath315 which are manifolds , and in the sequel , we may use @xmath165 to denote a metric on @xmath315 as well as on its universal cover @xmath60 the following are all relevant notions of convergence : * if the coefficients of the metrics @xmath161 on the lie algebra , which satisfy the ricci flow odes , converge as @xmath316 then the corresponding metrics @xmath317 converge in @xmath318 or @xmath319 as tensors . we call this @xmath318 or @xmath320 convergence . * the metric spaces @xmath321 converge uniformly as metric spaces ( see @xcite ) . * the metric spaces @xmath321 converge in the pointed gromov - hausdorff topology ( see @xcite ) . * the riemannian groupoids @xmath322 converge in @xmath318 or @xmath320 as riemannian groupoids ( see @xcite and @xcite ) . before we describe the previous work , let s consider the convergence in the present paper . we have looked at convergence of a system of odes for a normalized ricci flow equation in the space of metric lie algebras . in particular , the convergence is for the structure constants @xmath323 which implies convergence of the connection since the connection is determined by the lie brackets ; in fact , if @xmath324 is the riemannian connection and @xmath1 is an orthonormal frame as in theorem [ mil curvature thm ] , then @xmath325 for @xmath326 ( see ( [ mu def ] ) ) . convergence in @xmath327 of the connections implies convergence in @xmath328 of the riemannian metrics ( see , e.g. , ( * ? ? ? * chapter 3 ) ) . the normalization is not given explicitly , but must be such that none of the lie bracket coefficients ( for an orthonormal frame ) become infinite and the lie algebra does not become abelian . this type of convergence is considered on higher dimensional nilpotent metric lie algebras in @xcite . the earliest works consider either the ( forward ) ricci flow equation with no scaling @xcite or the normalized ricci flow equation where the rescaling is based on the scalar curvature @xcite , and prove @xmath327 convergence of the associated left - invariant metrics on the simply - connected lie group . the normalized ricci flow is helpful for @xmath329 since it prevents the sphere from shrinking to a point in finite time , but otherwise the ricci flows exist for all time @xmath163 ( for both unnormalized and normalized ) . some of these geometries collapse , in the sense that some of the metric coefficients @xmath156 go to zero , indicating a compact quotient will have the volume go to zero , as the sectional curvatures go to zero at a rate of @xmath330 . in @xcite , @xmath327 convergence of the backward direction ( of normalized ricci flow ) is considered , and it is found that the curvatures may go to infinity and convergence is often in a gromov - hausdorff sense to a sub - riemannian geometry . similar work is done for the cross curvature flow in @xcite and @xcite . in each of these cases , compact quotients are considered , and limits are considered collapsing if a fundamental domain has injectivity radius going to zero as time goes to infinity . in many cases , compact quotients collapse with bounded curvature , i.e. , the injectivity radius goes to zero while the sectional curvatures stay bounded . if one is interested in the simply - connected lie groups , which are diffeomorphic to @xmath117 in all cases except @xmath43 , the collapsing does not occur . instead , one can show pointed gromov - hausdorff convergence to euclidean space ( see @xcite ) . this convergence is sometimes due to the fact that the flow is stretching the geometry , revealing only that the riemannian manifold is locally euclidean . to counteract this effect , lott @xcite considers convergence when the metric is rescaled by @xmath330 to try to prevent the sectional curvatures from decaying to zero . in addition , lott introduces @xmath318 convergence on riemannian groupoids ( see also @xcite ) , which essentially means that the universal cover converges in @xmath318 while the actions of the fundamental group change as the metric evolves . in this setting , one finds that the rescaled solutions on the universal covers converge to expanding soliton metrics , getting results similar to those in this paper . it would not be hard to carry out a similar analysis in the backward time direction with the use of @xcite . if one is not concerned with a change of topology , the methods of uniform convergence of metric spaces will suffice , as in @xcite . we note that in @xcite and @xcite , only @xmath327 convergence of the metrics is shown explicitly , although @xmath320 convergence of riemannian groupoids is claimed . it is also shown that the sectional curvatures converge , and so using standard compactness arguments ( see , for instance , @xcite @xcite ( * ? ? ? * chapter 3 ) ) , this implies that the convergence is @xmath331 under ricci flow , shi s work ( @xcite and @xcite ) implies that if the curvature is bounded at @xmath332 then all covariant derivatives of curvature converge at all positive times , uniformly for @xmath333 however , since we are considering rescaling at time @xmath332 shi s estimates do not directly apply ( unless we assume uniform backwards existence ) , so an additional argument is needed to show that the metrics converge in @xmath318 for @xmath334 we would like to have at least convergence in @xmath335 . the cases in @xcite and @xcite can be shown to converge in @xmath318 with additional work bounding the covariant derivative . we do not pursue this here , but do present an interesting example below concerning a rescaling that does not produce @xmath320 convergence . there is one major difference between the results in this paper and the results in @xcite and @xcite . metric lie algebras of type @xmath336 converge to flat metrics in our setting , while they converge to @xmath337 in the setting of @xcite and @xcite . we note that @xmath337 can not be realized as a three - dimensional , unimodular lie group with a left - invariant metric . we see that , if we rescale the metric to ensure that the sectional curvatures do not go to zero , the lie algebra coefficients can not stay bounded : using the computations in @xcite , we see that @xmath338 for some constant @xmath339 and two sectional curvatures are asymptotically like @xmath340 while one is asymptotically like @xmath341 the best we can do by rescaling is to maintain the negative sectional curvature and let the other two curvatures go to zero . this suggests scaling the metric by @xmath342 under this rescaling , using ( [ a lambda corresp ] ) we can calculate that @xmath343 thus we see that this rescaling would not stay in the space of unimodular metric lie algebras , and we must take a different scaling to stay in this space . a different scaling will cause the sectional curvatures all to go to zero ( for instance no scaling at all ) , revealing euclidean space . metric lie algebras of the type @xmath344 also converge to a flat metric , so one might ask if rescaling by the maximum curvature can create a non - flat limit . we will use the notation of @xcite . note that@xmath345 and recall that in this case , ricci flow has @xmath346 for constants @xmath347 and sectional curvatures are all proportional to @xmath348 if we rescale , replacing @xmath158 with @xmath349 , we see that @xmath350 thus for the rescaled solution , @xmath351 and thus , using proposition [ drc],@xmath352 and so under this rescaling , @xmath353 as @xmath354 since @xmath355 is not bounded , we can not use arzela - ascoli - type compactness theorems to get convergence of the curvatures ( see @xcite or @xcite ) and the curvatures may not converge . thus this kind of rescaling does not necessarily result in @xmath335 and definitely not @xmath356 convergence . thus there is no natural rescaling which results in a non - flat limit . we can also investigate what happens in the backwards time limit and compare to the results of @xcite . the results are slightly different , since once again the rescaling in @xcite is chosen in a particular way , while we have chosen a different rescaling . a comparison can be made in a straightforward way , which we leave to the reader . however , we would like to point out the particularly interesting case of the backwards limit of the trajectory @xmath283 which appears to converge , in our setting , to the point @xmath26 ( see figure [ figure3 ] ) , which represents the flat metric on @xmath271 ( see theorem b ) . as in the case of convergence of the forward evolution of @xmath336 metrics ( i.e. , those represented by @xmath300 @xmath357 @xmath358 and @xmath39 as in theorem a ) , the fact that we see convergence to the flat metric indicates that our rescaling may be revealing only the infinitesimal euclidean character of the limit . in fact , as we saw in section [ geometry section ] , the trajectory @xmath9 corresponds to berger metrics which are riemannian submersions over the round @xmath359-sphere . as we follow the trajectory backwards , the fibers shrink and so we expect convergence to the @xmath359-sphere . the formalism in @xcite and @xcite together with the calculations in @xcite allow one to make this convergence precise in the sense of riemannian groupoids ( and , in particular , pointed gromov - hausdorff ) . finally , we note that the stability analysis of ricci solitons differs from that in @xcite . in @xcite , only compactly - supported variations are considered , and it was found , for instance , that the soliton metric on @xmath246 is linearly stable . we found that in the space of three - dimensional , nonabelian , unimodular metric lie algebras up to scaling , the soliton metric on @xmath246 is unstable ( it corresponds to a repulsive fixed point ) . these two results are not contradictory , since the variations we consider are certainly not compactly supported . in this appendix , we prove a characterization of the space of three - dimensional , nonabelian , unimodular metric lie algebras , considered up to isometry and scaling . under the correspondence of lemma [ lemma m tilde ] , three - dimensional , unimodular metric lie algebras correspond to vectors @xmath360 by the formula ( [ diagonalize ] ) . to account for equivalence under scaling , we consider two - dimensional real projective space @xmath361 and denote the image of @xmath362 in @xmath118 under the quotient map as @xmath363 permuting the components induces an isometry of the metric lie algebra , and so we define the action of @xmath364 , the group of permutations of three elements , on @xmath118 by @xmath365 notice that the actions of @xmath366 and @xmath367 on @xmath368 are commutative with respect to each other , and thus@xmath369 we denote the image of @xmath370 in @xmath371 as @xmath372 , $ ] where the use of square brackets instead of round brackets indicates the further equivalence using the @xmath366 action . we will now introduce a fundamental domain for @xmath373 recall the sets @xmath123 and @xmath130 defined in ( [ sm ] ) and ( [ smbar ] ) and the equivalence relation @xmath122 determined by ( [ tilde def ] ) and ( [ tilde def plus ] ) . [ proposition about s]the map@xmath374 defined by@xmath375 , \\ \tilde{\phi}\left ( \infty\right ) & = \left [ 1:0:0\right]\end{aligned}\ ] ] is surjective . moreover , @xmath376 induces a homeomorphism @xmath377 consider a point @xmath372 \in\mathbb{rp}^{2}/s_{3}$ ] . we can use multiplication by @xmath378 to ensure that at least one entry is positive and another is nonnegative . then we can use the permutations to ensure that @xmath379 , @xmath380 and @xmath381 in this case we have that @xmath382 = \left [ 1:\frac{y}{x}:\frac{z}{x}\right]\ ] ] with @xmath383 and @xmath384 which proves the first statement . to see that @xmath385 is well - defined , we must check that @xmath386 for @xmath125 we see that @xmath387 = \left [ \frac{1}{m_{3}}:0:1\right ] = \left [ 1:0:\frac{1}{m_{3}}\right ] .\ ] ] to see @xmath385 is a bijection , we note that if @xmath388 then @xmath389 = \left [ 1:m_{2}^{\prime}:m_{3}^{\prime}\right ] , \ ] ] with @xmath390 thus @xmath391 is equal to one of the following , where @xmath392 : @xmath393 , @xmath394 , @xmath395 , @xmath396 , @xmath397 , or @xmath398 . by checking each of the cases , one can verify that @xmath385 is a bijection . to see the continuity , we note that for any sequence @xmath399 satisfying @xmath219 and @xmath220 we have @xmath400 \\ & = \lim_{i\rightarrow\infty}\left [ 1:\frac{m_{2}^{\left ( i\right ) } } { m_{3}^{\left ( i\right ) } } : \frac{1}{m_{3}^{\left ( i\right ) } } \right ] \\ & = \left [ 1:0:0\right ] .\end{aligned}\ ] ] certainly there is a map @xmath401 induced by the map @xmath115 defined in lemma [ lemma m tilde ] . we wish to show that this map is bijection . it is immediate that the map is surjective , so we need only show that it is injective . the main ideas for the proof are in the paper of lastaria @xcite , which constructs families of nonisometric metric lie algebras with the same ricci tensors . we will need two invariants for @xmath0 . it is easy to see that the following quantities are invariants of @xmath0 if @xmath402 ( and the case @xmath403 is distinguished from these cases as well):@xmath404 , \lambda_{2}\left [ \operatorname{rc}\right ] , \lambda_{3}\left [ \operatorname{rc}\right ] \right ) } { \left\vert \operatorname{rc}\right\vert } & \in s^{2}\\ \frac{\left\vert d\operatorname{rc}\right\vert ^{2}}{\left\vert \operatorname{rc}\right\vert ^{3 } } & \in\mathbb{r}\text{,}\ ] ] where @xmath405 \leq\lambda_{2}\left [ \operatorname{rc}\right ] \leq\lambda_{3}\left [ \operatorname{rc}\right ] $ ] are the eigenvalues of the ricci operator put into ascending order . notice that @xmath406 only if @xmath407 and so the corresponding metric lie algebra is not isometric to any metric lie algebra induced by another element of @xmath408 . we will consider the map @xmath409 where @xmath410 gives the two invariants above . recall that the ricci eigenvalues are @xmath411 where @xmath412 @xmath413 @xmath414 are as in theorem [ mil curvature thm ] . we want to express these in terms of @xmath170 and @xmath415 so we introduce the following functions which are closely related to the @xmath135 : @xmath416 recall that on @xmath123 we have @xmath417 and so @xmath418 also , since @xmath419 we have @xmath420 we define the partition@xmath421 of @xmath123 by looking at the signs of @xmath422 and @xmath423:@xmath424 using theorem [ mil curvature thm ] and proposition [ drc ] we can write @xmath425 which is defined on @xmath426 explicitly as follows : @xmath427{cc}\left ( \left ( 0,0,1\right ) , \frac{\left ( \nu_{2}^{2}+\nu_{3}^{2}\right ) } { \nu_{2}\nu_{3}}\right ) & \text{if } \left ( m_{2},m_{3}\right ) \in s_{1}\\ \left ( \left ( -1,0,0\right ) , \frac{\left ( \nu_{1}^{2}+\nu_{3}^{2}\right ) } { \left\vert \nu_{1}\nu_{3}\right\vert } \right ) & \text{if } \left ( m_{2},m_{3}\right ) \in s_{2}\\ \left ( \frac{\left ( \nu_{1}\nu_{2},\nu_{1}\nu_{3},\nu_{2}\nu_{3}\right ) } { \left\vert \nu\nu\right\vert } , \ast\right ) & \text{if } \left ( m_{2},m_{3}\right ) \in s_{++}\\ \left ( \frac{\left ( \nu_{1}\nu_{3},\nu_{1}\nu_{2},\nu_{2}\nu_{3}\right ) } { \left\vert \nu\nu\right\vert } , \ast\right ) & \text{if } \left ( m_{2},m_{3}\right ) \in s_{-+}\\ \left ( \frac{\nu_{1}\nu_{3},\nu_{2}\nu_{3},\nu_{1}\nu_{2}}{\left\vert \nu \nu\right\vert } , \ast\right ) & \text{if } \left ( m_{2},m_{3}\right ) \in s_{-- } , \end{array } \right.\ ] ] where @xmath428 and @xmath429 now we will show that @xmath430 is injective . first we will show that no two points in different partitions in @xmath431 correspond to equivalent metric lie algebras ( equivalent up to isometry and scaling ) . then we will show that within each partition , no two points correspond to equivalent metric lie algebras . since @xmath430 restricted to @xmath438 does not have any zeroes in the first component , then certainly we have @xmath439 if @xmath185 or @xmath432 is in @xmath440 or @xmath441 furthermore , since the first component has all positive entries if @xmath442 it is distinguished from @xmath443 and @xmath444 both of which have two negative entries . this shows that @xmath445 for all cases except if @xmath185 or @xmath432 in @xmath443 and the other in @xmath446 say @xmath447 and @xmath448 and @xmath449 thus we have that @xmath450 with @xmath451 i.e. , there exists @xmath73 such that @xmath452 which implies@xmath453 using ( [ eqn nus ratios ] ) , we find that@xmath454 and , using ( [ nu signs ] ) and ( [ eqn nus ratios ] ) again , we arrive at @xmath455 it is a consequence of ( [ nu inequality ] ) that @xmath456 so @xmath457 thus , using ( [ nu def ] ) , @xmath458 or @xmath459 again by ( [ nu def ] ) , we have the following : @xmath460 inserting these into ( [ eqn nus ratios ] ) gives @xmath461 which implies that @xmath462 or @xmath463 since neither @xmath171 nor @xmath464 can equal one , we must have @xmath465 on @xmath468 we have@xmath469 which implies that @xmath470 notice that this implies that @xmath471 or@xmath472 ( using that @xmath291 implies @xmath473 now we can write @xmath423 and @xmath474 in terms of @xmath170 : @xmath475 so @xmath476 this function is one - to - one on @xmath477 thus @xmath430 is injective . on @xmath478 we have @xmath479 which implies that @xmath480 now we can write @xmath422 and @xmath474 in terms of @xmath170 @xmath481 so @xmath482 this function is strictly increasing for @xmath234 . the result follows . let @xmath488 be the projection onto the first component and let @xmath489 on @xmath486 @xmath490 is @xmath491 we compute the jacobian determinant:@xmath492{cc}\frac{2\left ( 1-m_{3}\right ) } { \nu_{3}^{2 } } & \frac{2m_{2}}{\nu_{3}^{2}}\\ \frac{2m_{3}}{\nu_{2}^{2 } } & \frac{2\left ( 1-m_{2}\right ) } { \nu_{2}^{2}}\end{array } \right ) = \frac{-4\nu_{1}}{\nu_{2}^{2}\nu_{3}^{2}}<0\ ] ] since @xmath493 on @xmath494 on @xmath487 @xmath490 is @xmath495 its jacobian determinant is @xmath492{cc}\frac{2m_{3}}{\nu_{2}^{2 } } & \frac{2\left ( 1-m_{2}\right ) } { \nu_{2}^{2}}\\ \frac{2\left ( 1-m_{3}\right ) } { \nu_{3}^{2 } } & \frac{2m_{2}}{\nu_{3}^{2}}\end{array } \right ) = \frac{4\nu_{1}}{\nu_{2}^{2}\nu_{3}^{2}}<0\ ] ] since @xmath496 on @xmath497 on @xmath444 @xmath490 is @xmath498 its jacobian determinant is @xmath492{cc}\frac{2}{\nu_{2}^{2 } } & \frac{-2}{\nu_{2}^{2}}\\ \frac{2\left ( m_{3}-1\right ) } { \nu_{1}^{2 } } & \frac{-2m_{2}}{\nu_{1}^{2}}\end{array } \right ) = \frac{-4\nu_{3}}{\nu_{1}^{2}\nu_{2}^{2}}>0.\ ] ] in each case we see that @xmath490 is injective , and thus so is @xmath499 as a consequence of all of these propositions , we can conclude that @xmath500 is injective , implying the following theorem , which implies theorem [ fundamental domain theorem ] . note that we have not yet defined a topology on @xmath0 ; it will be defined to allow for the theorem . we have just shown that @xmath501 gives a bijection between @xmath371 and @xmath0 , and we can give @xmath0 a topology which makes @xmath501 a homeomorphism . proposition [ proposition about s ] shows that @xmath385 is a homeomorphism between @xmath408 and @xmath373 b. chow , s .- c . chu , d. glickenstein , c. guenther , j. isenberg , t. ivey , d. knopf , p. lu , f. luo , and l. ni . the ricci flow : techniques and applications . part i. geometric aspects . mathematical surveys and monographs , 135 . american mathematical society , providence , ri , 2007 . xxiv+536 pp . m. gromov . metric structures for riemannian and non - riemannian spaces . based on the 1981 french original . with appendices by m. katz , p. pansu and s. semmes . translated from the french by sean michael bates . reprint of the 2001 english edition . modern birkhuser classics . birkhuser boston , inc . , boston , ma , 2007 . xx+585 pp .

in certain cases the communication time required to deterministically implement a nonlocal bipartite unitary using prior entanglement and locc ( local operations and classical communication ) can be reduced by a factor of two . we introduce two such `` fast '' protocols and illustrate them with various examples . for some simple unitaries , the entanglement resource is used quite efficiently . the problem of exactly which unitaries can be implemented by these two protocols remains unsolved , though there is some evidence that the set of implementable unitaries may expand at the cost of using more entanglement .

1109.5013

a central problem in quantum information theory is that of interconversion between resources , for example , how to use communication channels to produce entanglement between distant parties , or how to use such entanglement to carry out nonlocal operations . in particular , the use of prior entanglement assisted by classical communication to carry out nonlocal unitaries has been the subject of various studies @xcite ; for a more extensive list see ref . @xcite . in this paper we add _ time _ as a resource to be considered along with entanglement cost when constructing protocols for bipartite nonlocal unitaries ( nonlocal gates ) . the ability to implement nonlocal unitaries rapidly may be particularly relevant in the context of distributed quantum computation @xcite , where less time consumption means less decoherence ; or in position - based quantum cryptography @xcite , where it may allow certain position verification schemes to be broken . the usual protocols for bipartite unitaries , such as those in ref . @xcite , have the following general structure : alice carries out local operations and measurements , and sends the measurement results through a classical communication channel to bob , who then carries out corresponding operations and measurements , and sends the measurement results back to alice using classical communication . finally , alice performs additional local operations that may depend on the previous measurement results of both parties . when the distance between the two parties is large the total time required for the protocol will be dominated by the two rounds of communication , thus double the minimum time for a signal to pass from one to the other . however , there exist nonlocal unitaries which can be implemented by a protocol in which alice and bob carry out local operations and measurements at the same time , and then simultaneously send the results to the other party , and finally perform local operations depending on the received messages to complete the protocol . this reduces the total communication time by a factor of two . we are interested in identifying which bipartite unitaries can be carried out using such a _ fast _ protocol , and also in finding the associated entanglement cost . the crucial distinction between a fast and slow protocol of the form considered here is that for the latter , bob needs to wait for a message from alice before choosing the basis in which to carry out his measurement ( `` choosing the measurement basis '' is equivalent to choosing what local gates to do before his measurement ) , whereas in the former this basis can be fixed in advance . we have identified two classes of nonlocal unitary that lend themselves to a fast protocol : _ controlled _ unitaries of the form shown in below , and _ group _ unitaries of the form shown in . the slow versions of both were considered in our previous work @xcite , where we showed that controlled unitary protocols , while useful for understanding how such protocols work , can always be replaced by group unitary protocols that make use of the same resources . our fast protocols represent special cases ( i.e. , special groups and parameter choices ) of the slow protocols discussed previously , and once again the controlled kind can be replaced by the group kind . by increasing the amount of entanglement expended , additional unitaries can be carried out using these fast protocols . in some cases this allows an arbitrarily close approximation to a unitary which can not be carried out exactly by these methods . a still more general class of slow protocols , corresponding to eq . ( 18 ) in @xcite , also has a fast version , but we have yet to find examples of unitaries it can carry out that can not be implemented by our other fast protocols . the protocols we consider are deterministic they succeed with probability one and use a definite amount of entanglement determined in advance . such deterministic fast protocols have previously been studied by groisman and reznik @xcite for a controlled - not ( cnot ) gate on two qubits , and by dang and fan @xcite its counterpart on two qudits . in addition , buhrman _ _ @xcite and beigi and knig @xcite have published approximate schemes for what they call `` instantaneous quantum computation , '' equivalent to a fast bipartite unitary in our language . these protocols , unlike ours , can be used , to approximately carry out any bipartite unitary . the one in @xcite , which is based on the nonlocal measurement protocol in @xcite , has a probability of success less than 1 , so it is not deterministic , but this probability can be made arbitrarily close to 1 by using sufficient entanglement . the protocol in @xcite uses a fixed amount of entanglement to implement with probability 1 a bipartite quantum operation ( completely positive trace preserving map ) which is close to the desired unitary , and it can be made arbitrarily close by using sufficient entanglement . the term `` instantaneous '' is not unrelated to the idea of an `` instantaneous measurement '' as discussed in @xcite , where the terminology seems somewhat misleading in that completing their protocols actually requires a finite communication time , e.g. , the parties must send the results to headquarters ( or to each other ) in order to complete the identification of the measured state . in the same way , `` instantaneous quantum computation '' actually requires a finite communication time , the same as in our fast protocol . the paper is organized as follows . in sec . [ sct2 ] we consider controlled unitaries of the form eq . , where the unitaries being controlled form an abelian group . ( appendix [ sct_extend ] contains an argument , which may be of more general interest , that allows projectors @xmath0 in this formula to be replaced by projectors of rank 1 . ) in addition we show how subsets of the collection of unitaries representing an abelian group can be employed to generate fast unitaries otherwise not accessible by our protocol . section [ sct3 ] is devoted to group unitaries of the form , including a significant number of examples . we also present an argument showing that the controlled - abelian - group unitaries of sec . [ sct2 ] can be transformed to group unitaries of the form . the concluding sec . [ sct4 ] contains a brief summary along with an indication of some open problems . in this section we construct a fast protocol for any controlled - abelian - group unitary of the form @xmath1 where the @xmath0 are orthogonal projectors , possibly of rank greater than 1 , on a hilbert space @xmath2 of dimension @xmath3 . the @xmath4 are unitary operators on a hilbert space @xmath5 of dimension @xmath6 , that form a representation of an abelian group @xmath7 of order @xmath8 . as shown in appendix [ sbct_control_rank ] , it suffices to consider projectors of rank 1 . that is , a scheme for implementing @xmath9 where @xmath10 denotes a ket belonging to a standard ( or computational ) orthonormal basis , is easily extended to one that carries out the more general . in addition we shall consider cases , sec . [ sbct2.3 ] , in which the @xmath4 form a _ subset _ of an abelian group , with the sum in restricted to a subset @xmath11 of the integers from @xmath12 to @xmath13 . the simplest abelian group is a cyclic group , so we start with the case where the @xmath4 in are a representation of such a group . ( it suffices to consider ordinary representations , since a projective representation of a cyclic group is equivalent to an ordinary representation ; see sec . 12.2.4 of @xcite . ) it will be convenient to let @xmath14 be the identity and @xmath15 . the slow protocol @xcite for this case , which works for any collection @xmath4 of unitaries on @xmath16 , is shown in fig . [ fgr1 ] , where @xmath17 is a fully entangled state on the ancillary systems @xmath18 and @xmath19 associated with @xmath20 and @xmath16 , respectively , and the gates @xmath21 , @xmath22 , and @xmath23 ( the fourier matrix ) are defined by @xmath24 here @xmath25 denotes @xmath26 , so @xmath27 . the symbols resembling `` d '' in fig . [ fgr1 ] represent measurements in the standard basis . the slow protocol proceeds by alice carrying out the operations indicated on the left side of fig . [ fgr1 ] , and then sending the outcome @xmath28 of the measurement on @xmath18 to bob over a classical channel . he uses it to carry out a gate @xmath29 , followed by the other operations in the center of the figure . his final measurement outcome @xmath30 is sent to alice over another classical channel , who uses it to perform an additional gate @xmath31 that completes the protocol . a faster protocol can be constructed if the two rounds of classical communication can be carried out simultaneously instead of consecutively . this is possible if bob can carry out various operations , including a measurement , in advance of receiving the value of @xmath28 from alice , as in fig . the classical signals can then be sent simultaneously , and the protocol is completed when both alice and bob make final corrections that depend on the signals they receive . in order to change the slow protocol into a fast protocol , one must , in effect , push the @xmath29 gate in fig . [ fgr1 ] through the two gates that follow it in order to arrive at the situation in fig . the two steps are as follows : 1 . commute @xmath29 with the controlled-@xmath32 gate : the @xmath29 itself passes through the control node unchanged , but leaves the @xmath33 gate controlled by state @xmath34 instead of @xmath35 , so @xmath36 instead of @xmath32 acts on @xmath16 . this can be compensated at the end of the protocol by a local unitary correction of @xmath37 , with @xmath38 ( see the discussion below ) . 2 . commute @xmath29 with @xmath23 : we have that @xmath39 , and since the @xmath40 are diagonal unitaries , they do not affect the measurement result @xmath30 in the standard basis , and thus @xmath40 is absent from the fast protocol in fig . ( due to the removal of @xmath40 there is an unimportant global phase , dependent on @xmath28 and @xmath30 , that is introduced in the implementation of @xmath41 . ) the final correction @xmath42 is possible because the @xmath4 form a cyclic group : the net operation on @xmath16 is @xmath43 , where @xmath38 follows from fig . [ fgr1 ] and the definition of the @xmath29 gate . this is an extra restriction over the slow protocol , where the @xmath32 can be arbitrary unitaries . * example 1 . * case ( i ) . @xmath41 is the @xmath8-dimensional cnot gate , with @xmath44 , and @xmath45 form a cyclic group . this is the class of unitaries implementable by dang and fan s protocol shown in fig . 2 of @xcite . case ( ii ) . @xmath41 is a controlled unitary of the form with @xmath46 , @xmath47 , @xmath48 , @xmath49 . ( the @xmath32 are not shift operators , so this is more general than the protocol in @xcite . ) the fast protocol for controlled - cyclic - group unitaries is easily generalized to the case where the @xmath32 in form an ordinary representation of an abelian group @xmath7 of order @xmath8 . again it suffices ( appendix [ sbct_control_rank ] ) to consider the case where the @xmath0 are rank 1 projectors . any finite abelian group is the direct sum ( direct product ) of @xmath50 cycles , and it is convenient to adopt a label @xmath51 for elements of @xmath7 that reflects this structure , by thinking of it as an @xmath50-tuple of integers , @xmath52 with @xmath53 , where @xmath54 is the length of the @xmath55-th cycle . in this way group multiplication , with @xmath56 the identity , is the same as vector addition , modulo @xmath54 for the @xmath55-th component . similarly , the @xmath57 labels on the systems @xmath18 and @xmath19 in fig . [ fgr2 ] , and the measurement outcomes @xmath28 and @xmath30 , can also written as @xmath50-tuples : @xmath58 , etc . in the following we will make use of the inner product of two @xmath50-tuples such as @xmath59 . the @xmath21 , @xmath22 , and @xmath23 gates are now appropriate tensor products of the cyclic group gates in , for example , @xmath60 understood as @xmath61 and @xmath62 , using the obvious @xmath50-tuple definition of @xmath63 . the @xmath64 gate in fig . [ fgr2 ] is the tensor product of the @xmath65 gates for the different cycles , @xmath66 here is why the protocol works . assume an initial product state @xmath67 on @xmath68 . then the operator implemented on @xmath16 is @xmath69 . the @xmath23 gate on @xmath19 before the measurement gives rise to a phase @xmath70 , which is partially compensated by the phase @xmath71 in the @xmath64 gate on @xmath20 , and since @xmath72 , we are left with an overall phase of @xmath73 . since this phase is independent of @xmath51 , a superposition of initial product states of this form for different @xmath51 will also be transformed by @xmath41 , up to an overall phase that is of no concern . note that the @xmath32 themselves may , but need not , be tensor products , as illustrated in the following example . * example 2 . * case ( i ) : @xmath74 . the @xmath32 defined by @xmath75 are tensor products , and form a group @xmath76 . if one regards @xmath2 as well as @xmath5 as a tensor product of two qubits , the @xmath41 defined by is itself a tensor product @xmath77 , with each factor a controlled - cyclic group unitary with one qubit on the @xmath20 and the other on the @xmath16 side . thus @xmath41 can be implemented by an overall protocol which is just two smaller protocols running in parallel with each other , one for @xmath78 and the other for @xmath79 . case ( ii ) : @xmath80 . modify the @xmath32 in by keeping only the first three rows and columns , so they are no longer tensor products , though they still form a group @xmath76 . consequently , the protocol that carries out @xmath41 can no longer be viewed as two smaller protocols running in parallel . assume that the @xmath4 in form an ordinary representation of an abelian group of order @xmath8 , but the sum over @xmath51 is restricted to some subset @xmath11 of the set of @xmath8 @xmath50-tuples defined in the last subsection . it will suffice once again to consider the case of rank - one projectors , i.e. , . the @xmath57 , @xmath28 , and @xmath30 in fig . [ fgr2 ] run over the same range as before , but @xmath51 is restricted to the set @xmath11 . therefore the dimension @xmath81 of @xmath2 is less than the schmidt rank of @xmath82 , which is the order @xmath8 of the group . it is convenient to use the elements of @xmath11 to label the kets that form the basis of @xmath2 in [ corresponding to the projectors in ] . the operator @xmath64 is now given by , but with @xmath51 restricted to @xmath11 . the reason that the fast protocol in fig . [ fgr2 ] will work in this case is the same as given above in sec . [ sbct2.2 ] ; the fact that @xmath51 is restricted to a subset makes no difference . the significance of this extension of the result in sec . [ sbct2.2 ] is that it enlarges the class of fast unitaries that can be carried out using a protocol of this sort , though perhaps with a significant increase in the entanglement cost . this is illustrated by the following example , which shows that in certain cases one can approximate a continuous family of unitaries using sufficient entanglement ( a large enough @xmath8 ) . * example 3 . * consider a unitary on two qubits @xmath20 and @xmath16 of the form @xmath83 where @xmath84 for some integer @xmath85 . by relabeling @xmath86 on @xmath20 as @xmath87 , we see that is of the form with @xmath51 in the sum restricted to the two values in @xmath88 . thus @xmath41 can be carried out in a fast way at an entanglement cost of @xmath89 ebits . in general , any two - qubit controlled unitary is of the form ( up to local unitaries on @xmath20 and @xmath16 , before and after @xmath41 ) with @xmath90 for some real number @xmath91 . since @xmath91 can be approximated by @xmath92 multiplied by a rational number @xmath93 with large enough @xmath8 , any two - qubit controlled unitary can be approximately implemented ( up to local unitaries ) using this fast protocol by setting @xmath94 equal to @xmath95 for suitable @xmath30 and @xmath8 ; the entanglement cost is again @xmath89 ebits . a further generalization to arbitrary @xmath3 and @xmath6 is possible for any unitary @xmath41 which is diagonal in a product of bases , a basis on @xmath2 and another on @xmath5 . when such a diagonal @xmath41 is written in the form , the unitaries @xmath32 are diagonal . each diagonal element of @xmath32 is approximately an integer root of unity , hence each @xmath32 is approximately an integer root of the identity operator , and the whole set @xmath4 can be approximated by a subset of an ordinary representation of an abelian group of sufficient size . thus any bipartite unitary diagonal in a product of bases can be approximately implemented by a fast protocol . in this section we consider a fast protocol for `` double - group '' unitaries of the form @xmath96 where @xmath7 is a group of order @xmath8 , @xmath97 and @xmath98 are unitaries on hilbert spaces @xmath2 and @xmath5 of dimension @xmath3 and @xmath6 , respectively , and the operators @xmath99 form a projective representation of @xmath7 , in the sense that @xmath100 the collection @xmath101 of complex numbers of unit magnitude is known as the factor system . similarly , @xmath102 and @xmath103 each form a projective unitary representation of @xmath7 with individual factor systems which may differ from one another , whose product for a given @xmath104 and @xmath105 is @xmath106 . from sec . 12.2.1 of @xcite , for our purposes we can assume the factor system @xmath101 is _ standard _ , that is , @xmath107 where @xmath108 is the identity element in @xmath7 . the slow protocol in sec . iv d of @xcite for implementing unitaries of the type is shown in fig . the two parties share a maximally entangled state @xmath109 on the ancillary systems @xmath18 , @xmath19 , each of dimension @xmath8 , the order of @xmath7 . alice and bob perform controlled-@xmath97 and controlled-@xmath98 gates on @xmath110 and @xmath111 , respectively . alice follows this with a @xmath112 gate on @xmath18 , where in the standard basis @xmath112 is a complex hadamard matrix @xmath113 divided by @xmath114 , that is , a unitary with all elements of the same magnitude @xmath115 . then she does a measurement on @xmath18 in the standard basis and sends the result @xmath28 to bob . bob carries out a @xmath116 gate on @xmath19 , where each @xmath116 is a diagonal unitary matrix whose diagonal elements are the complex conjugates of those in the @xmath28-th row of @xmath113 . thus @xmath112 and @xmath116 generalize the fourier gate @xmath23 and the @xmath40 gate in our previous paper @xcite . this more general choice does not extend the set of unitaries the slow protocol can implement , since the phases in @xmath112 and @xmath116 cancel each other , but it allows the fast protocol to implement a larger set of unitaries than would otherwise be possible . next , bob applies a unitary gate @xmath117 to @xmath19 , where @xmath106 is defined in and the coefficients @xmath118 are those in . ( the coefficients @xmath118 are not uniquely defined by @xmath41 if the @xmath119 are linearly dependent , but there is always at least one choice for which @xmath120 is unitary , theorem 7 of @xcite . ) then he measures @xmath19 in the standard basis and sends the result @xmath30 to alice . to complete the protocol alice and bob apply unitary corrections @xmath121 and @xmath122 to their respective systems . the slow protocol in fig . [ fgr3 ] can be replaced with the fast protocol in fig . [ fgr4 ] provided the @xmath116 gate in the former , which in effect determines the basis for bob s measurement , can be eliminated at the cost of re - interpreting the outcome @xmath30 of his measurement . a sufficient condition for this is that for every @xmath28 there exists a _ complex permutation _ matrix @xmath123 ( exactly one nonzero element of magnitude 1 in each row and in each column ) such that @xmath124 the effect of @xmath123 is simply to permute the measurement outcomes , which can then be re - interpreted once @xmath28 is known . the phases in @xmath123 would only introduce a global phase for the implemented @xmath41 ( dependent on @xmath28 and @xmath30 ) and are therefore of no concern . a useful procedure for generating @xmath120 and @xmath112 matrices for which holds employs the _ character table _ @xmath125 of an abelian group @xmath126 of order @xmath8 . this is an @xmath127 matrix , all elements of which are of magnitude 1 , with columns labeled by elements of @xmath126 and rows by its distinct irreducible representations , all of which are one - dimensional . because the representations are one - dimensional , each row is itself a representation ( i.e. , the character is the representation `` matrix '' ) . the element - wise product of two columns is another column , since each row is a representation of the group ; likewise the element wise product of two rows is another row , since the ( tensor ) product of two representations is a representation . thus the transpose of a character table is again a character table . the complex conjugate of any column ( or row ) is another column ( row ) of @xmath125 corresponding to the inverse of the group element . the actual order of the columns or the rows is arbitrary , though it is often convenient to assume that the first row and the first column contain only @xmath128 s . since @xmath129 is a unitary matrix , a character table @xmath125 is a special case of a complex hadamard matrix : one whose elements are all of magnitude 1 , and whose rows ( and columns ) are mutually orthogonal . if @xmath126 is a cyclic group @xmath130 , then up to permutations of rows or columns @xmath131 , with @xmath23 the fourier matrix ; similarly if @xmath126 is the direct product ( sum ) of cycles , @xmath125 is the tensor product of the corresponding fourier matrices @xcite . [ thm1 ] ( a ) let @xmath125 be the character table of an abelian group @xmath126 of order @xmath8 , and define @xmath132 where @xmath133 and @xmath134 are complex permutation matrices , and @xmath135 a diagonal matrix with diagonal elements of magnitude 1 , thus a diagonal unitary . let @xmath116 be the diagonal matrix with diagonal elements equal to the complex conjugates of those forming row @xmath28 of @xmath113 . then there exist complex permutation matrices @xmath136 such that is satisfied for every @xmath28 . \(b ) if the rows of an @xmath127 matrix @xmath113 are linearly independent and form a ( necessarily abelian ) group @xmath126 up to phases under element - wise multiplication , then @xmath113 is of the form given in . the proof is in appendix [ appthm1 ] . note that the group @xmath126 need not be isomorphic to the group @xmath7 represented by @xmath137 although they are of the same order see example 8 in sec . [ sbct3.3 ] with @xmath138 . the matrix @xmath113 defined in has the property that the rows under element - wise products form the abelian group @xmath126 up to a possible phase factor determined by @xmath133 . one consequence of is @xmath139 where @xmath140 is a complex permutation matrix and @xmath135 is a diagonal unitary . when the matrix @xmath120 is of the form , the fact that all its elements are of the same magnitude @xmath115 , as implied by , means that the same is true of the @xmath118 . a partial answer to the question of whether implies that @xmath112 and @xmath120 have the form given in is provided by the following theorem , whose proof is also in appendix [ appthm1 ] . [ thm2 ] for each @xmath28 let @xmath116 be the diagonal matrix whose diagonal elements are the complex conjugates of those forming row @xmath28 of a complex hadamard matrix @xmath141 . if there exists a unitary matrix @xmath120 without a zero element in its first row , and complex permutation matrices @xmath142 such that eq . holds for every @xmath28 , then the matrices @xmath116 form an abelian group up to phases , and and hold . it is worth noting that with @xmath112 and @xmath120 of the form there is a symmetrical version of the fast protocol in fig . [ fgr4 ] in which the @xmath120 gate on the @xmath16 side is replaced with @xmath112 . this requires that the entangled resource @xmath82 be changed to @xmath143 the reason this works is that the changes produced by @xmath140 in in the measurement outcome @xmath30 can always be compensated by altering the function @xmath144 that determines the final corrections . the following is a simple example . * example 4 . * the two - qubit unitary , @xmath145 where @xmath146 and @xmath147 are pauli @xmath148 gates on @xmath20 and @xmath16 , is equivalent under local unitaries to a cnot gate . it is of the form with @xmath7 the cyclic group @xmath149 of order 2 , and can be implemented by the fast protocol in fig . [ fgr4 ] using the matrices @xmath150 where the rows of @xmath112 multiplied by @xmath151 form a group @xmath152 . because @xmath112 and @xmath120 change the measurement basis , alice and bob effectively perform measurements of @xmath153 and @xmath154 , respectively ; this is the same thing ( with the parties interchanged ) as the fast protocol in @xcite . an equivalent symmetrized protocol in which @xmath120 is replaced by @xmath112 employs a resource state @xmath155 , and both parties perform a @xmath153 measurement . given a particular bipartite unitary @xmath41 , can it be implemented using the fast double - group protocol ? any such @xmath41 can always be written in the form using a sufficiently large group ( see sec . v a of @xcite ) , and typically there are many different ways of constructing such an expansion . however , for our fast protocol to work , assuming that @xmath112 and @xmath120 are of the form given in , one must find a _ particular _ expansion , a particular group @xmath7 and unitaries @xmath97 and @xmath98 along with expansion coefficients @xmath118 , that satisfy appropriate conditions . in particular , ( i ) the @xmath118 must all be of the same magnitude @xmath115 , as noted following . but two additional conditions must be checked : ( ii ) the matrix @xmath120 defined in must be unitary , and ( iii ) @xmath120 must be related to the character table of some group @xmath126 as in . condition ( iii ) can be checked in the following way . multiply each row and then each column of @xmath156 by a suitable phase such that the resulting matrix @xmath157 , where @xmath158 and @xmath94 are diagonal unitaries , has @xmath128 s in the first row and first column . then check whether its rows ( alternatively , its columns ) form a group @xmath126 under component - wise multiplication . if this is so , then @xmath159 is a character table of @xmath126 . equating it to @xmath125 in , letting @xmath160 and @xmath161 , and choosing any complex permutation matrix @xmath133 , we arrive at a @xmath112 which along with this @xmath120 satisfies the conditions of theorem [ thm1 ] . thus , provided ( i ) , ( ii ) , and ( iii ) are satisfied there is in fact a fast unitary @xmath41 . the scheme just described provides a useful approach for constructing examples . start with a group @xmath7 and ( projective ) unitary representations @xmath102 and @xmath103 on @xmath2 and @xmath5 , and look for a set of coefficients @xmath162 of equal magnitude such that @xmath120 given by is unitary , and satisfies condition ( iii ) in the preceding paragraph . the search is aided by noting that any factor system , see sec . 12.2.2 of @xcite , is equivalent to a _ normalized _ factor system in which each @xmath163 is an @xmath8-th root of 1 . a consequence , proved in appendix [ appthm1 ] , is the following : [ thm3 ] let @xmath119 in be a projective representation of a group @xmath7 of order @xmath8 with a normalized ( see above ) factor system @xmath106 , . assume the matrix @xmath120 defined in is of the form given in . then the coefficients @xmath162 in can be written in the form @xmath164 , \label{eqn20}\ ] ] where @xmath165 is an integer that depends upon @xmath166 , and @xmath167 is a phase factor independent of @xmath166 . the theorem justifies the following exhaustive , albeit tedious , search procedure for possible sets of coefficients @xmath118 , assuming they all have the same magnitude , once a group @xmath7 , a projective representation of @xmath7 , and a normalized factor system have been chosen . consider all possible sets of coefficients of the form , setting @xmath168 and @xmath169 for the identity @xmath108 of @xmath7 , as the global phase of @xmath41 is unimportant . for each set , check that the matrix @xmath120 given by is unitary . then see if the rows of the corresponding @xmath159 , constructed as described above , form a group under component - wise multiplication . using this procedure we have been able to show that if @xmath170 , so the group is @xmath149 , the only two - qubit unitaries that can be implemented by our fast protocol are either trivial products of unitaries or else equivalent under local unitaries to a cnot gate . ( note that there are additional fast two - qubit unitaries that can be carried out using a bigger group , thus larger @xmath8 and more entanglement . ) both conditions ( ii ) and ( iii ) are nontrivial requirements . not every case in which the @xmath118 are of equal magnitude will lead to a unitary matrix @xmath120 . for example , if @xmath171 for every @xmath166 and @xmath137 is an ordinary representation of @xmath7 , so @xmath172 , then obviously does not define a unitary matrix . and even if @xmath120 is unitary , condition ( iii ) may not hold . for example , the unitary in eq . ( 58 ) of @xcite with @xmath173 , assuming @xmath174 is not an integer multiple of @xmath175 , results in a @xmath159 matrix whose rows do not form a group . for every ordinary representation of an abelian group @xmath7 there is a corresponding fast protocol , as the group is a direct product ( sum ) of cycles , and one can apply the construction in example 6 below . for general projective representations or non - abelian groups the matter remains open . the examples which follow represent just a few of the unitaries that can be carried out by our double - unitary fast protocol . examples 5 and 6 make relatively efficient use of entanglement resources , in that the order of the group , which is the rank of the fully - entangled resource state , is equal to the schmidt rank ( or schmidt number @xcite ) of @xmath41the minimum number of summands required to represent it as a sum of products of operators on @xmath20 and @xmath16 . examples 7 and 8 , the latter involving a non - abelian group @xmath7 , illustrate how the class of fast unitaries can be significantly expanded by using more entanglement . note that any two - qubit unitary is equivalent under local unitaries to one of the form ( see @xcite ) , @xmath176 , \label{eqn21}\ ] ] where @xmath174 , @xmath177 , and @xmath167 are real numbers that can be calculated from the matrix of @xmath41 ( see , e.g. , the appendix of @xcite for the method of calculation ) . for the two - qubit examples below we give the values of @xmath174 , @xmath177 , and @xmath167 . * example 5 . * in the two - qubit unitary @xmath178 with @xmath179 the identity , @xmath21 and @xmath22 the pauli operators @xmath153 and @xmath148 , and @xmath7 the group @xmath76 , the method of search indicated in sec . [ sbct3.2 ] yields the following possibilities for @xmath180 . \(a ) the case @xmath181 is equivalent to the swap gate defined in @xcite , in which the two qubits are interchanged ; @xmath182 in . an alternative fast protocol for this gate consists of teleportation done simultaneously in both directions . \(b ) the case @xmath183 implements the @xmath184 gate as defined in @xcite , equivalent under local unitaries to the double - cnot ( dcnot ) gate defined in @xcite ; @xmath185 . \(c ) the case @xmath186 , where @xmath187 ; @xmath188 . in each case the entanglement resource of two ebits required to carry out the protocol is the minimum possible amount , since the unitary is capable of creating two ebits of entanglement . * example 6 . * when the @xmath137 , with @xmath166 an integer between 0 and @xmath13 , form an ordinary representation of the cyclic group @xmath130 of order @xmath8 , the coefficients @xmath189 will provide a fast implementation of . in particular with @xmath190 this becomes @xmath191 the method of proof of theorem [ thm4 ] can be used to show the equivalence of with @xmath192 , which in turn is locally equivalent to the @xmath8-dimensional cnot gate of example 1 . * example 7 . * the unitary @xmath193 of schmidt rank 4 on two qubits , where @xmath187 and the operators are the same as in example 5 , employs an unfaithful ( each operator , e.g. @xmath194 , appears twice in the sum ) representation of the abelian group @xmath195 , with the eight coefficients being the corresponding @xmath118 . it can be verified that this @xmath162 set satisfies the requirements for the fast protocol . as this group is of order 8 the protocol requires a resource of 3 ebits , and we have not found any fast protocol which can implement this unitary using less entanglement . it corresponds to @xmath196 in ( the b gate of @xcite ) . * example 8 . * for any given integer @xmath197 , let @xmath198 ( @xmath199 ) be the @xmath200 matrices @xmath201 they form an irreducible ordinary representation of the dihedral group @xmath202 of order @xmath203 , where the first kind in correspond to rotations and the second kind to reflections . let @xmath204 , & \text { $ n$ even,}\\ ( \epsilon(f)/\sqrt{2n})\exp[\,i\pi m f(f+1)/n\ , ] , & \text { $ n$ odd , } \end{cases}\ ] ] where @xmath30 is any positive integer coprime with @xmath205 , and @xmath206 is 1 for @xmath207 and @xmath55 for @xmath208 . it can be verified that these @xmath162 sets satisfy the requirements for the fast protocol . the two - qubit unitary constructed in this way is locally equivalent to with @xmath209 , @xmath210 , and @xmath177 , which necessarily lies in the interval @xmath211 $ ] , depending on @xmath30 and @xmath205 in a manner we have not studied in detail . it may be that the possible set of @xmath177 values is dense in @xmath211 $ ] . the following theorem , proved in appendix [ appcnot ] , shows that the family of unitaries for which fast protocols were constructed in sec . [ sct2 ] can also be realized using our fast protocol for double - group unitaries . the converse is not true , since , for instance , the 2-qubit swap gate in example 5 can not be realized as a controlled - abelian - group unitary , as it is of schmidt rank 4 , while a controlled unitary on 2 qubits can not have schmidt rank greater than 2 . [ thm4 ] let @xmath41 be a controlled - abelian - group unitary of the form , where the @xmath32 are a subset of an ordinary representation of an abelian group @xmath7 of order @xmath8 . then @xmath41 is equivalent under local unitaries to @xmath212 where the @xmath118 are complex coefficients , the @xmath213 are linear combinations of @xmath0 s , and @xmath214 , @xmath215 , @xmath216 are all ordinary representations of the group @xmath7 . in addition the @xmath118 can be chosen to satisfy the requirements for the fast protocol as given in sec . [ sbct3.2 ] . hence all controlled - abelian - group unitaries of the form discussed in sec . [ sct2 ] can be implemented by our fast double - group unitary protocol , without using more entanglement . any nonlocal unitary can be carried out deterministically by means of local operations and classical communication provided an appropriate entangled resource is available . however , teleportation and various more efficient schemes typically require two rounds of classical communication , and hence the minimum total amount of time required to complete the protocol is twice the time required for one - way communication . in certain cases there are fast protocols in which the minimum total time is only half as long , and in this paper we have discussed two protocols for fast bipartite unitaries . the first is shown in fig . [ fgr2 ] : it carries out a controlled - abelian - group unitary of the form , including cases in which only a subset of the collection @xmath4 that forms an abelian group appear in the sum . the second , shown in fig . [ fgr4 ] , will carry out a double - group unitary of the form , provided the coefficients @xmath118 satisfy appropriate conditions . we have shown , sec . [ sbct3.4 ] , that unitaries which can be carried out by the first protocol can also be carried out by the second , though the converse is not true ( e.g. , example 5 ) . we have constructed some examples for both protocols . note , however , that we have not been able to answer the fundamental question as to precisely _ which _ unitaries can be carried out _ exactly _ using a fast protocol and a fixed entanglement resource specified in advance . we do not know the answer even for fast protocols of the two types considered in this paper . finding examples for our double - group protocol is not at all trivial ; see the discussion in sec . [ sbct3.2 ] . in sec . [ sbct2.3 ] we discussed cases in which subsets of a group can be used to carry out a fast controlled unitary protocol at the cost of greater entanglement . see in particular example 3 , where we showed that any unitary in a particular continuous family can be approximated arbitrarily closely by a unitary implementable by a deterministic fast protocol , provided one is willing to use up enough entanglement . this is similar in spirit to the results in @xcite and @xcite . their protocols may need less or more entanglement than our protocols , depending on the form of the unitary . in certain situations ( e.g. , example 5 ) , our protocol uses the minimum possible entanglement . it would be nice if these issues could be clarified in terms of some basic principle(s ) of quantum information theory . another question we have not been able to answer is whether unitaries of the more general form @xmath217 , where the @xmath97 form an ordinary or projective representation of a group , but @xmath218 need not do so , can be carried out by means of a fast protocol . a slow protocol was found in our earlier work @xcite , and we have constructed a fast version for that protocol , but it seems to only work for those unitaries implementable by our fast double - group protocol of sec . [ sct3 ] . again , this may reflect some fundamental principle of quantum information , but if so we have not been able to identify it . we thank serge fehr , hoi kwan lau , and shiang yong looi for helpful discussions , including those on the connections of this work with the topics of instantaneous measurement and position - based quantum cryptography . patrick coles read some of the appendices and made helpful suggestions . this work has been supported in part by the national science foundation through grants no . phy-0456951 and no . s.m.c . has also been supported by a grant from the research corporation . in this section , we consider implementation of @xmath41 on @xmath219 using a different @xmath220 and use these ideas below to show that consideration of controlled unitaries can be restricted to those with rank-@xmath128 projectors . a scheme is shown in fig . [ figextend ] , which is valid for both the fast and slow protocols . if the protocol for @xmath220 is fast , then the whole protocol is fast . the circuit in fig . [ figextend ] can be used for the following two situations , to be discussed in detail in the two subsections below . the first situation , called `` extension '' , is that we extend the space of @xmath2 to @xmath221 , and the unitary @xmath222 is an extension of @xmath41 , where @xmath41 is any unitary on @xmath219 . the second situation , called `` compression with extension , '' is only for general controlled unitaries @xmath41 of the form . the protocol replaces the higher - rank projectors on @xmath2 in @xmath41 with rank - one projectors on @xmath221 in @xmath220 , while adding more projectors if needed . applications are found in sec . [ sct2 ] . note that while fig . [ figextend ] shows an extension ( and compression in the case of controlled unitaries ) on the @xmath20 side , one can just as well do this on the @xmath16 side , or both sides . the input state for the whole protocol is any state on @xmath223 together with some fixed state on the ancilla @xmath224 denoted by @xmath225 . the map @xmath226 is unitary , and @xmath227 is its inverse . the unitary @xmath11 obviously has its input dimension equal to its output dimension : @xmath228 , where @xmath229 is determined by @xmath220 ( see below ) , @xmath3 may be unequal to @xmath230 , and @xmath231 may be unequal to @xmath229 . here we consider the first type of extension , where @xmath220 is the direct sum of @xmath232 ( the same as @xmath41 but on a different space ) and another unitary @xmath233 . dimension @xmath229 is fixed by @xmath220 and is greater than @xmath3 . one can always choose @xmath231 to be equal to @xmath229 , but in general may choose @xmath231 to be less than @xmath229 . the action of the unitary @xmath11 on the actual input space is determined by the following equation : @xmath234 where @xmath235 is an orthonormal basis of @xmath2 . the requirements for @xmath11 in this equation can be extended to a full definition of a unitary . the effect of @xmath11 is to transfer alice s input state into @xmath221 . define @xmath236 to be the span of @xmath237 . then @xmath238 , where @xmath239 is a space orthogonal to @xmath236 . then the correct form of @xmath220 should be @xmath240 , where @xmath241 is the same as the original unitary @xmath41 except it is on a different space , and @xmath242 is an arbitrary unitary . now we prove that the circuit defined in fig . [ figextend ] applied to @xmath243 yields @xmath244 . suppose the input state on @xmath245 is @xmath246 . then @xmath247 the argument can be extended by linearity to superpositions of the states @xmath248 . as a side remark , the derivations above should still work if we replace the unitary @xmath11 by an isometry @xmath249 , replace @xmath227 by @xmath250 , and remove system @xmath224 from the circuit in fig . [ figextend ] . it can be verified that the overall operation of the circuit is @xmath251 . we chose to present the argument using the unitary @xmath11 rather than the isometry @xmath252 in order to show that the scheme has no trouble finding an experimental implementation . the same remark also applies to the next subsection . the current extension technique was useful in finding the protocols in secs . [ sbct2.3 ] and example 8 in [ sbct3.3 ] , but it turns out that those protocols ( for the particular types of unitaries ) can be simplified such that no extension is needed , which is why we have not explicitly mentioned this idea of extension in those sections . for general controlled unitaries @xmath41 of the form ( not limited to those implementable by the fast protocols in this paper ) , we now consider a procedure that compresses the higher - rank projectors on @xmath2 into rank - one projectors on @xmath221 , while adding more projectors if needed . the form of @xmath220 is @xmath253 where @xmath254 . apparently @xmath255 . the steps of the protocol are similar to those in appendix [ sbct_extend1 ] , but with the following change to the requirements on @xmath11 ( and accordingly @xmath227 ) : @xmath256 where @xmath257 is the label for the states in a specific basis of @xmath2 , with @xmath51 ( @xmath258 ) labeling which projector @xmath0 , and @xmath259 labeling which basis state in the support of @xmath0 . note the range of @xmath260 depends on @xmath51 , and because of this , @xmath230 should be at least the maximum rank among the @xmath0 s ( @xmath258 ) , while satisfying @xmath228 . the requirements for @xmath11 in can be extended to a full definition of a unitary . the effect of @xmath11 is to transfer the information about `` which @xmath51 '' into @xmath221 , and that information is used in the controlled unitary @xmath220 , and then transferred back to @xmath2 by @xmath227 . the final state of the protocol is @xmath261 , and the proof for the correctness of the protocol is similar to that in appendix [ sbct_extend1 ] . suppose the input state on @xmath245 is @xmath262 . then @xmath263 the argument can be extended by linearity to superpositions of the states @xmath264 . as noted above , for the current subsection it is also plausible to replace the unitary @xmath11 by an isometry @xmath265 , replace @xmath227 by @xmath250 , and remove system @xmath224 from the circuit in fig . [ figextend ] . using the definition of @xmath220 in , it can be verified that the overall operation of the circuit is @xmath251 , where @xmath41 is of the form . * proof of theorem [ thm1 ] . * \(a ) we need to show [ see and note that both @xmath135 and @xmath116 are diagonal matrices ] that @xmath266 is a complex permutation matrix . the diagonal elements of @xmath116 are complex conjugates of the elements in a row of @xmath113 and thus [ eq . ] equal to a common phase factor times those in a particular row , say row @xmath30 , of the character table @xmath125 ; recall that the complex conjugate of a row in @xmath125 is always another row of @xmath125 . now the matrix @xmath267 is the matrix @xmath125 with each column multiplied by the corresponding diagonal element of @xmath116 . thus the @xmath57-th row of @xmath268 is the element - wise product of row @xmath57 of @xmath125 with row @xmath30 of @xmath125 , up to an overall phase that depends on @xmath57 . but since the rows of @xmath125 form a group under element - wise products , this means that @xmath269 for a suitable complex permutation matrix @xmath270 . since @xmath271 , tells us that @xmath272 , and because @xmath134 , @xmath270 , and @xmath273 are all complex permutation matrices , so is @xmath274 . \(b ) the rows of @xmath113 are the elements of a group @xmath126 in the following sense . the element - wise product of any two rows is , up to a phase , a third row , and the fact that the rows are linearly independent means that this third row is uniquely determined . that is , there is a well - defined associative group multiplication , which is commutative , so the group @xmath126 is abelian . there is necessarily one row consisting of identical elements ; this is the identity element of @xmath126 . given any row , there is another row which is , element by element , its complex conjugate , up to a single phase for the whole row ; these two rows are inverses of each other . hence the group @xmath126 is well defined . obviously , each column of @xmath113 consists of elements ( viewed as @xmath275 matrices ) that form an irreducible representation of @xmath126 under the multiplication of complex numbers , and all the elements of @xmath113 are of magnitude 1 . divide each row by its first element to form the matrix @xmath276 , whose rows again form the group @xmath126 , but now without additional phases since the first element of each row is 1 . since the rows of @xmath113 are linearly independent , so are the rows of @xmath276 , and hence also its columns . thus each column of @xmath276 forms a distinct irreducible representation of the group @xmath126 . all the @xmath8 distinct irreducible representations of @xmath126 are included in the columns of @xmath276 , thus @xmath276 is the transpose of a character table of the abelian group @xmath126 , and such a transpose is itself a character table of @xmath126 ( see the discussion preceding theorem [ thm1 ] ) . hence we can identify @xmath276 as the @xmath125 in , and the phases used to change @xmath113 to @xmath276 can be included in the matrix @xmath133 in . * proof of theorem [ thm2 ] . * let @xmath277 and rewrite in the equivalent form @xmath278 the matrix on the left is obtained from @xmath279 by multiplying each column by the corresponding diagonal element of @xmath116 , while the one on the right is obtained by some permutation of the rows of @xmath279 , with an additional overall phase for each row . consider the special case in which the first row of @xmath279 consists of 1 s . then the first row of @xmath280 is @xmath281 , the row vector whose elements are the diagonal elements of @xmath116 , and according to , it is equal to the first row of @xmath282 ( i.e. , a phase times some other row of @xmath279 ) . as the @xmath283 are linearly independent ( they are complex conjugates of the rows of the hadamard matrix @xmath113 ) , it follows by equating , for each @xmath28 , the first row on the left side of with that on the right side , that the element - wise product of the first row of @xmath279 and the @xmath283 for all possible values of @xmath28 generates all the rows of @xmath279 up to a phase ( i.e. , each row of @xmath279 is a phase times one of the @xmath283 ) . then according to , the element - wise product of any row of @xmath279 with any @xmath283 ( i.e. the element - wise product of any two rows of @xmath279 up to a phase ) , is always a third row of @xmath279 up to a phase . similarly the product of any two of the matrices in the set @xmath284 is a third , up to a phase ; this is an associative group product . the first row of @xmath279 is equal to one of the @xmath283 up to a phase , and this @xmath116 corresponds to the group identity . because the first row of @xmath279 consists of 1 s , there is a row of @xmath285 , say row @xmath30 , which consists of equal elements . then the @xmath30-th row of @xmath286 [ see ] , identifies the group inverse of @xmath116 . thus the @xmath116 under matrix products form a group ( denoted by @xmath126 ) up to phases . for any @xmath116 , its complex conjugate @xmath287 is the matrix inverse of @xmath116 , and hence some @xmath288 up to a phase . consequently all the different rows of @xmath113 , the complex conjugates of @xmath283 , are equal to the different @xmath289 up to phases and a permutation of the ordering , hence these rows form the group @xmath126 up to phases under element - wise multiplication . then according to theorem [ thm1](b ) , @xmath126 must be abelian , and @xmath113 is of the form given in . since the rows of @xmath113 are a permutation of the different @xmath289 up to phases , they are also a permutation of the rows of @xmath279 up to phases . thus @xmath279 is of the form with @xmath290 . hence for the special case under consideration , is satisfied , and then follows from . next consider the more general case in which the elements of the first row of @xmath279 are all nonzero . form @xmath159 from @xmath279 by dividing every column of @xmath279 by the corresponding element on the first row . this means that @xmath291 , where @xmath158 is a diagonal matrix . since @xmath158 commutes with every @xmath116 , we can replace @xmath279 on both sides of with @xmath159 . as the first row of @xmath159 consists of 1 s , the argument given above shows that the @xmath284 form an abelian group up to phases , and the rows of @xmath159 are , up to phases , some permutation of the rows of @xmath113 , and @xmath113 is of the form given in , again according to theorem [ thm1](b ) . thus the different columns of @xmath159 all have the same normalization , and since the same is true of @xmath279 , as @xmath120 is assumed to be a unitary matrix , it follows that the diagonal elements of @xmath158 all have magnitude 1 , and one can set @xmath292 in . then follows from . * proof of theorem [ thm3 ] . * the coefficients @xmath118 must all be of the same magnitude , @xmath115 ( see the remark following theorem [ thm1 ] ) , and without loss of generality @xmath167 can be chosen such that @xmath169 in , where @xmath108 is the group identity . first consider the case that @xmath137 is an ordinary representation of @xmath7 , so that @xmath172 . choose any group element @xmath293 and assume it has order @xmath294 , which means that @xmath294 divides @xmath8 and @xmath295 . the first row of @xmath120 , corresponding to @xmath296 in , contains @xmath297 in some order , interspersed with other coefficients @xmath118 . now consider the row of @xmath120 corresponding to @xmath298 in . it is related to the first row as indicated here , @xmath299 where only the relevant columns are shown , rearranged in a convenient order . let us define the @xmath159 matrix , as in the first paragraph of sec . [ sbct3.2 ] , to be the one obtained from @xmath156 by multiplying each row and each column by some phase , so that all the elements in the first row and first column are equal to 1 . this is a character table , so every element is an @xmath8-th root of 1 . equivalently , @xmath159 is obtained from @xmath120 by dividing each column of @xmath120 by the corresponding element in the first row , and then in the resulting matrix dividing each row by its first element . consequently , applying this process to the rows and columns shown in , we conclude that @xmath300 where each @xmath301 is an @xmath8-th root of 1 . the product of these @xmath302 equations , @xmath303^p=\phi_1 \phi_2 \cdots \phi_{p-1}(\sqrt{n})^{p-1},\ ] ] implies , since @xmath169 , that @xmath304 must be a @xmath294-th root of a number which is itself an @xmath8-th root of 1 , and because @xmath294 divides @xmath8 , @xmath305 is of the form . this completes the argument for an ordinary representation . when the @xmath306 form a projective representation of @xmath7 with a standard factor system , the first row in is the same , but the second row will be multiplied by appropriate factors @xmath106 . since we are assuming a normalized factor system , all these additional factors are themselves @xmath8-th roots of 1 , so still holds for @xmath301 which are @xmath8-th roots of 1 , and the rest of the argument is the same as before . in this appendix , we prove theorem [ thm4 ] , which says that the controlled unitary @xmath307 given by , where @xmath308 are orthogonal projectors , and @xmath309 form a subset of an ordinary representation of an abelian group @xmath7 , is equivalent to @xmath310 under local unitaries , where @xmath118 are complex coefficients [ will be defined in ] , and @xmath213 are linear combinations of @xmath308 , and @xmath311 is an ordinary representation of @xmath7 . in addition , the @xmath118 can be chosen to satisfy the requirements for the fast protocol , hence all controlled - abelian - group unitaries can be implemented by the fast double - group unitary protocol . + we first prove the case that @xmath309 form a whole representation , not a subset , and at the end we will remark that the proof also works in the `` subset '' case . the proof is by explicitly constructing a @xmath312 and showing that it is equivalent to @xmath41 under local unitaries . any abelian group is a direct sum of cyclic groups , so @xmath313 , where @xmath314 , and @xmath54 is the order of the cyclic group @xmath315 . then @xmath316 . the group element @xmath51 is relabeled by a vector @xmath317 , where @xmath53 . we use the convention that @xmath51 is the sequential numbering ( starting from 0 ) for the vectors in lexicographical order , so that @xmath318 corresponds to @xmath319 , and @xmath320 corresponds to @xmath321 , etc . suppose @xmath309 has been diagonalized under a suitable unitary similarity transform , then @xmath309 is the direct sum of some irreducible representations ( possibly with redundancy ) . all possible irreducible representations of @xmath7 are one - dimensional , and have the form @xmath322 where @xmath323 is the label for irreducible representations ( some may be missing from @xmath309 , but we still include them in this labeling scheme for convenience ) . denote the computational basis of @xmath5 by @xmath324 , @xmath325 , then the @xmath19-th diagonal elements in @xmath326 determine an irreducible representation labeled by @xmath327 , @xmath328 . the @xmath327 can be written in the vector form [ see and the sentence after that ] , and the components in the vector @xmath327 will be denoted by @xmath329 . as discussed above , for every @xmath330 we can represent @xmath166 using a set of integers : @xmath331 , with group multiplication corresponding to vector addition ( modulo @xmath332 for the @xmath333-th element of the vector ) . define @xmath334 , where @xmath335 is defined by ( basically the same as in example 6 ) @xmath336 , \notag\\ \,\,\,f_s=0,1,\cdots , r_s-1.\end{aligned}\ ] ] we choose the @xmath213 to be @xmath337 p^a_k,\ ] ] where @xmath338 is the vector labeling for @xmath51 . define @xmath339 as @xmath340 \dyad{b}{b},\ ] ] where @xmath329 are the components in the vector labeling for @xmath327 . it is not hard to verify that @xmath311 is an ordinary representation of @xmath7 , and the coefficients @xmath118 form a unitary @xmath120 matrix of the type , hence @xmath312 is unitary . where @xmath258 , @xmath342 , and @xmath10 is any eigenstate of @xmath308 . denote the phase factor in front of @xmath343 in the above equation by @xmath344 , then @xmath345 ^ 2+[k_s+q_{b , s}+(r_s\mbox { mod } 2)/2]^2\}/r_s \big ) \notag\\ & = \frac{1}{\sqrt{n}}\prod_{s=1}^{\eta}\left[\left(\sum_{j=0}^{r_s-1}\exp\{-\pi i [ j+k_s+q_{b , s}+(r_s\mbox { mod } 2)/2]^2/r_s\}\right ) \exp\{\pi i [ k_s+q_{b , s}+(r_s\mbox { mod } 2)/2]^2/r_s\}\right]\notag\\ & = \frac{\alpha}{\sqrt{n}}\prod_{s=1}^{\eta}\exp\{\pi i[k_s+q_{b , s}+(r_s\mbox { mod } 2)/2]^2/r_s\}\end{aligned}\ ] ] where @xmath174 is a constant independent of @xmath51 and @xmath346 . in deriving the last line above , we have used @xmath347 for even @xmath260 , and @xmath348 for odd @xmath260 , which make the substitution @xmath349 possible , and obtained @xmath350 , where @xmath351 ^ 2/r_s\}$ ] . define the local operators @xmath352 and @xmath353 on @xmath2 and @xmath5 , respectively , as follows : @xmath354 from the unitarity of @xmath312 , @xmath344 is always a phase factor with magnitude 1 , hence @xmath352 and @xmath353 are unitary operators . then for @xmath10 chosen arbitrarily from the eigenstates of @xmath308 , we have @xmath355 \ket{k}\ket{b } \notag\\ & = \sum_{k=0}^{n-1}p^a_k\otimes v^b_k \ket{k}\ket{b } \notag\\ & = \uc \ket{k}\ket{b}\end{aligned}\ ] ] where we have used to derive the third line , and used to derive the fourth line . since @xmath308 are of finite rank , there exists a finite collection of states of the form @xmath343 to make a complete basis of @xmath245 . the actions of @xmath356 and @xmath41 are the same on all states in a complete basis , hence they must be identical operators . therefore @xmath41 is equivalent to @xmath312 under local unitaries . using the algorithm in sec . [ sbct3.2 ] , it can be verified that the choice of coefficients @xmath118 given above ( which can be viewed as the choice in example 6 generalized to the non - cyclic abelian groups ) satisfies the requirements for the fast protocol . hence the double - group protocol for @xmath312 is fast . the proof above can basically be applied to the case that @xmath309 form a subset of a representation . in general some @xmath308 do not occur in the expressions for @xmath213 and @xmath352 ; those @xmath308 can be safely removed because @xmath213 and @xmath352 are block diagonal , where the blocks are determined from the support of the @xmath308 s . the coefficients @xmath118 are still the same as above , so the protocol is still fast . hence the proof still works . a. yimsiriwattana and s. j. lomonaco jr . , in _ coding theory and quantum computing _ , edited by d. evans _ et al . _ , volume 381 of contemporary mathematics , ( american mathematical society , 2005 ) , p. 131 . e - print arxiv : quant - ph/0402148v3 . h. buhrman , n. chandran , s. fehr , r. gelles , v. goyal , r. ostrovsky , and c. schaffner , in _ advances in cryptology crypto 2011 _ , edited by p. rogaway , ( springer , new york , 2011 ) , p. 423 . e - print arxiv:1009.2490v4 [ quant - ph ] .

a _ resolving set _ for a graph is a collection of vertices , chosen so that for each vertex , the list of distances from to the members of uniquely specifies . the _ dimension _ of is the smallest size of a resolving set for . much attention has been paid to the metric dimension of distance - regular graphs . work of babai from the early 1980s yields general bounds on the metric dimension of primitive distance - regular graphs in terms of their parameters . we show how the metric dimension of an imprimitive distance - regular graph can be related to that of its halved and folded graphs , but also consider infinite families ( including taylor graphs and the incidence graphs of symmetric designs ) where more precise results are possible .

1312.4971

let @xmath3 be a finite , undirected graph without loops or multiple edges . for @xmath4 , the _ distance _ from @xmath5 to @xmath2 is the least number of edges in a path from @xmath5 to @xmath2 , and is denoted @xmath6 ( or simply @xmath7 if @xmath0 is clear from the context ) . a _ resolving set _ for a graph @xmath3 is a set of vertices @xmath8 such that for each vertex @xmath9 , the list of distances @xmath10 uniquely determines @xmath11 . equivalently , @xmath12 is a resolving set for @xmath0 if , for any pair of vertices @xmath13 , there exists @xmath14 such that @xmath15 ; we say that @xmath16 _ resolves _ @xmath5 and @xmath11 . the _ metric dimension _ of @xmath0 is the smallest size of a resolving set for @xmath0 . this concept was introduced to the graph theory literature in the 1970s by harary and melter @xcite and , independently , slater @xcite ; however , in the context of arbitrary metric spaces , the concept dates back at least as far as the 1950s ( see blumenthal @xcite , for instance ) . for further details , the reader is referred to the survey paper @xcite . when studying metric dimension , distance - regular graphs are a natural class of graphs to consider . a graph @xmath0 with diameter @xmath17 is _ distance - regular _ if , for all @xmath18 with @xmath19 and any vertices @xmath20 with @xmath21 , the number of neighbours of @xmath2 at distances @xmath22 , @xmath18 and @xmath23 from @xmath5 depend only on the distance @xmath18 , and not on the choices of @xmath5 and @xmath2 . these numbers are denoted by @xmath24 , @xmath25 and @xmath26 respectively , and are known as the _ parameters _ of @xmath0 . it is easy to see that @xmath27 , @xmath28 are undefined , @xmath29 , @xmath30 and @xmath31 ( where @xmath32 is the valency of @xmath0 ) . we put the parameters into an array , called the _ intersection array _ of @xmath0 , @xmath33 in the case where @xmath0 has diameter @xmath34 , we have a _ strongly regular _ graph , and the intersection array may be determined from the number of vertices @xmath35 , valency @xmath32 , and the parameters @xmath36 and @xmath37 ; in this case , we say @xmath38 are the parameters of the strongly regular graph . another important special case of distance - regular graphs are the _ distance - transitive _ graphs , i.e. those graphs @xmath0 with the property that for any vertices @xmath39 such that @xmath40 , there exists an automorphism @xmath41 such that @xmath42 and @xmath43 . for more information about distance - regular graphs , see the book of brouwer , cohen and neumaier @xcite and the forthcoming survey paper by van dam , koolen and tanaka @xcite . in recent years , a number of papers have been written on the subject of the metric dimension of distance - regular graphs ( and on the related problem of class dimension of association schemes ) , by the present author and others : see @xcite , for instance . in this paper , we shall focus on various classes of _ imprimitive _ distance - regular graphs , which are explained below . a distance regular graph @xmath0 with diameter @xmath17 is _ primitive _ if and only if each of its distance-@xmath18 graphs ( for @xmath44 ) is connected , and is _ imprimitive _ otherwise . for @xmath45 , i.e. strongly regular graphs , the only imprimitive examples are the complete multipartite graphs @xmath46 ( with @xmath47 parts of size @xmath48 ) . for valency @xmath49 , a result known as _ smith s theorem _ ( after d. h. smith , who proved it for the distance - transitive case @xcite ) states that there are two ways for a distance - regular graph to be imprimitive : either the graph is bipartite , or is _ antipodal_. the latter case arises when the distance-@xmath17 graph consists of a disjoint union of cliques , so that the relation of being at distance @xmath50 or @xmath17 in @xmath0 is an equivalence relation on the vertex set . the vertices of these cliques are referred to as _ antipodal classes _ ; if the antipodal classes have size @xmath48 , then we say @xmath0 is _ @xmath48-antipodal_. it is possible for a graph to be both bipartite and antipodal , with the hypercubes providing straightforward examples . if @xmath0 is a bipartite distance - regular graph , the distance-@xmath34 graph has two connected components ; these components are called the _ halved graphs _ of @xmath0 . if @xmath0 is @xmath48-antipodal , the _ folded graph _ , denoted @xmath51 , of @xmath0 is defined as having the antipodal classes of @xmath0 as vertices , with two classes being adjacent in @xmath51 if and only if they contain adjacent vertices in @xmath0 . the folded graph @xmath51 is also known as an _ antipodal quotient _ of @xmath0 ; conversely , @xmath0 is an _ antipodal @xmath48-cover _ of @xmath51 . we note that @xmath0 and @xmath51 have equal valency ; a result of gardiner ( * ? ? ? * corollary 4.4 ) shows that @xmath48 is at most this valency . the operations of halving and folding may be used to reduce imprimitive graphs to primitive ones : see ( * ? ? ? * 4.2a ) for details . in particular , an imprimitive distance - regular graph with valency @xmath49 may be reduced to a primitive one by halving at most once and folding at most once . also , the intersection arrays of the halved and/or folded graphs of @xmath0 may be obtained from that of @xmath0 . in a 2006 paper of alfuraidan and hall ( * ? ? * theorem 2.9 ) , a refinement of smith s theorem is obtained which will be especially useful to us . we summarize their result below . [ thm : ah ] let @xmath0 be a connected distance - regular graph with @xmath35 vertices , diameter @xmath17 and valency @xmath32 . then one of the following occurs : 1 . @xmath0 is primitive , with @xmath52 and @xmath49 ; 2 . @xmath53 , and @xmath0 is a cycle @xmath54 ; 3 . @xmath55 , and @xmath0 is a complete graph @xmath56 ; 4 . @xmath45 , and @xmath0 is a complete multipartite graph @xmath46 ( where @xmath57 ; 5 . @xmath58 , @xmath0 is both bipartite and @xmath34-antipodal , @xmath59 , and @xmath0 is @xmath60(a complete bipartite graph with a perfect matching deleted ) ; 6 . @xmath58 , @xmath0 is bipartite but not antipodal , @xmath59 , and @xmath0 is the incidence graph of a symmetric design with @xmath2 points and block size @xmath61 ; 7 . @xmath58 , @xmath0 is antipodal but not bipartite , and @xmath0 is an antipodal cover of a complete graph @xmath62 ; 8 . @xmath63 , @xmath0 is both bipartite and antipodal , @xmath0 is an antipodal cover of a complete bipartite graph @xmath64 , and the halved graph is complete multipartite ; 9 . @xmath65 , @xmath0 is both bipartite and antipodal , the halved graphs are antipodal of diameter @xmath66 , and the folded graph is bipartite of diameter @xmath66 ; 10 . @xmath67 , @xmath0 is antipodal but not bipartite , and the folded graph is primitive with diameter @xmath68 and valency @xmath49 ; 11 . @xmath67 , @xmath0 is bipartite but not antipodal , and the halved graphs are primitive with diameter @xmath68 and valency at least @xmath66 ; 12 . @xmath17 is odd and @xmath69 , @xmath0 is bipartite and @xmath34-antipodal , the folded graph is primitive with diameter @xmath70 and valency @xmath49 , and the halved graphs are primitive with diameter @xmath70 and valency at least @xmath66 ; 13 . @xmath17 is even and @xmath71 , @xmath0 is bipartite and antipodal , and the graphs obtained by successive halving and folding are primitive with diameter @xmath72 and valency at least @xmath66 . we shall refer to these classes as ah1ah13 . the numbering is not exactly the same as that given in @xcite . we have separated the case of @xmath60 from the other imprimitive graphs of diameter @xmath66 ( cf . * corollary 2.10 ) ) ; as we shall see , it is something of a special case . apart from @xmath73 and @xmath74 , no graph appears in more than one class . one of main aims of this paper is to consider the asymptotic behaviour of the metric dimension of imprimitive distance - regular graphs . we use knuth s convention for asymptotic notation ( from @xcite ; see also ( * ? ? ? * 9.2 ) ) : for @xmath35 sufficiently large , we say that @xmath75 if there is a constant @xmath76 such that @xmath77 , that @xmath78 if there is a constant @xmath79 with @xmath80 , and @xmath81 if both of these happen . for any graph with @xmath35 vertices and diameter @xmath17 , it is straightforward to see that the metric dimension @xmath82 must satisfy the inequality @xmath83 ( see ( * ? ? ? * proposition 3.6 ) ) ; when considering families of fixed diameter ( such as strongly regular graphs ) , this gives a lower bound on @xmath82 of @xmath84 . like alfuraidan and hall @xcite , we regard class ah1 as the `` generic '' class of distance - regular graphs . for graphs in this class , namely primitive distance - regular graphs of diameter @xmath52 and valency @xmath49 , the pioneering work of babai @xcite in the early 1980s ( in a different context : see @xcite for details ) yields the following . [ thm : babai ] suppose that @xmath0 is a primitive distance - regular graph with @xmath35 vertices , valency @xmath49 and diameter @xmath52 . then : * @xmath85 ; * if @xmath45 ( i.e. @xmath0 is strongly regular ) , we have ( a ) @xmath86 and ( b ) @xmath87 ( where @xmath88 ) ; * if @xmath89 is the maximum size of a set of vertices at a given distance from any vertex of @xmath0 , we have @xmath90 some remarks about theorem [ thm : babai ] are in order . while the bound ( b ) in part ( ii ) appears weaker than ( a ) , if @xmath35 and @xmath32 have a linear relationship ( for example , in the case of paley graphs where @xmath91 ) this may be combined with the lower bound above to obtain @xmath92 . on the other hand , the strongly regular johnson and kneser graphs @xmath93 and @xmath94 have metric dimension @xmath95 ( see ( * ? ? ? * corollary 3.33 ) ) , so the @xmath96 factor can not be eliminated in general . in part ( iii ) the value of @xmath97 may be calculated from the intersection array ; for families where @xmath17 is fixed we may also obtain an improvement on the @xmath98 upper bound . for classes ah2ah4 , the metric dimension is easy to determine exactly , as we summarise below . [ prop : easy ] for the graphs in families ah2ah4 , we have @xmath99 , @xmath100 and @xmath101 . for cycles and complete graphs , this is trivial ; for complete multipartite graphs , it is straightforward ( see ( * ? ? ? * proposition 1 ) ) . asymptotically , for both complete graphs and complete multipartite graphs on @xmath35 vertices , we have @xmath102 . however , we suggest that classes ah2ah4 are pathological cases which should be disregarded as atypical . for the remaining classes ah5ah13 , much more work is required . given that imprimitive distance - regular graphs may be reduced to primitive graphs by the operations of halving and folding , it is desirable to obtain relationships between resolving sets and metric dimension for imprimitive graphs and their halved and/or folded graphs . in this section , we obtain such relationships . first , we consider the halving operation when @xmath0 is bipartite . our first result does not assume that @xmath0 is distance - regular . [ prop : halving ] let @xmath3 be a connected bipartite graph with bipartition @xmath103 , and let @xmath104 and @xmath105 be its halved graphs . then @xmath106 . let @xmath107 and @xmath108 be resolving sets for @xmath109 and @xmath110 , respectively . we will show that @xmath111 is a resolving set for @xmath0 . let @xmath112 . if one of these is in @xmath113 and the other in @xmath114 , then for any vertex @xmath115 , we must have that one of @xmath116 and @xmath117 is odd and the other is even , and thus @xmath118 , i.e. @xmath11 resolves @xmath119 and @xmath120 . in particular , we are free to choose @xmath121 . so we must consider the case where @xmath119 and @xmath120 are both in the same bipartite half . if @xmath122 , then there exists @xmath123 such that @xmath124 , and thus @xmath125 i.e. @xmath11 resolves @xmath119 and @xmath120 ( in @xmath0 ) . the case where @xmath126 is similar . hence @xmath12 is a resolving set for @xmath0 , and thus @xmath106 . a simple application of this theorem is when @xmath127 , so @xmath128 ; clearly @xmath129 and the bound holds with equality . even in the case of distance - regular graphs , the halved graphs need not be isomorphic ( although they must have the same parameters ) , so we can not assume that @xmath130 and @xmath131 should be equal . indeed , a counterexample is provided by the incidence graph of the unique ( up to duality ) generalized quadrangle @xmath132 : its halved graphs ( i.e. the point graphs of the @xmath132 and its dual ) have metric dimension @xmath133 and @xmath134 respectively ( see ( * ? ? ? * table 11 ) ) ; the incidence graph itself has metric dimension @xmath135 ( see ( * ? ? ? * table 4 ) ) . this also shows that the upper bound in proposition [ prop : halving ] need not be met with equality . obtaining bounds on the metric dimension of an antipodal distance - regular graph @xmath0 in terms of its folded graph @xmath51 is less straightforward . suppose @xmath3 is @xmath48-antipodal , with @xmath136 partitioned into @xmath137 antipodal classes @xmath138 . we define a _ @xmath48-antipodal partition _ to be a partition of @xmath136 into transversals of @xmath138 . ( note that some authors use the term `` antipodal partition '' to refer to the partition of @xmath136 into antipodal classes , so care is required here . ) we remark that a @xmath48-antipodal graph has many @xmath48-antipodal partitions , although in certain cases there natural partition arising from how the graph is constructed . we note that in an antipodal distance - regular graph @xmath0 with diameter @xmath17 , for @xmath4 with @xmath139 we have @xmath140 for any @xmath141 , and the diameter of @xmath51 is @xmath142 . [ prop : folding ] let @xmath3 be an @xmath143-antipodal distance - regular graph with diameter @xmath17 , and let @xmath144 be an @xmath143-antipodal partition of @xmath136 . suppose @xmath145 is a resolving set for the folded graph @xmath146 whose diameter is @xmath147 . for @xmath148 , let @xmath149 be its inverse image in @xmath136 with @xmath150 for each @xmath151 . let @xmath152 . * if @xmath153 with @xmath154 , @xmath155 and @xmath156 , then there exists @xmath157 with @xmath158 and @xmath159 such that @xmath160 ( i.e. @xmath161 resolves @xmath162 and @xmath163 ) . * if @xmath164 is odd , or for every @xmath165 there exists @xmath166 such that @xmath167 , then @xmath12 is a resolving set for @xmath0 . in particular , @xmath168 . * if @xmath169 is even and there exists @xmath165 such that @xmath170 for all @xmath121 , then @xmath171 is a resolving set for @xmath0 . in particular , @xmath172 . we start by proving part ( i ) of the theorem . suppose that @xmath153 with @xmath154 , @xmath155 and @xmath156 . since @xmath145 is a resolving set for @xmath51 , there exists @xmath166 such that @xmath173 . consequently , there exist indices @xmath174 such that @xmath175 we wish to find some @xmath157 that resolves @xmath162 and @xmath163 ; we consider the following cases . 1 . if @xmath176 , let @xmath177 , so @xmath157 and @xmath178 . 2 . if @xmath179 , we have @xmath180 and similarly @xmath181 by the properties of antipodal vertices in @xmath0 . in particular , we have @xmath182 , so @xmath183 and resolves @xmath162 and @xmath163 . otherwise , we have @xmath184 . at most one of @xmath174 can be @xmath50 ; without loss of generality we suppose that @xmath185 , and so @xmath186 . for a contradiction , suppose that @xmath187 does not resolve @xmath188 and @xmath163 , i.e. @xmath189 . since @xmath190 , we then have @xmath191 @xmath192 in particular , this implies that @xmath193 , a contradiction . thus @xmath187 resolves @xmath188 and @xmath163 . to prove parts ( ii ) and ( iii ) , we must construct a resolving set for @xmath0 . by part ( i ) , we know that for vertices @xmath194 , if @xmath156 then there is a vertex in @xmath12 which resolves them . it remains to consider the case where @xmath195 and @xmath196 , i.e. @xmath162 and @xmath163 are antipodal vertices in @xmath0 . first , we suppose that there exists @xmath158 such that @xmath197 . then if @xmath196 , for any @xmath198 , we must have that @xmath199 since @xmath200 and @xmath201 , we have @xmath202 , and thus @xmath161 resolves @xmath162 and @xmath203 . combined with part ( i ) , this shows that @xmath12 is a resolving set for @xmath0 of size @xmath204 , and thus ( ii ) holds . otherwise , we must have that @xmath170 for all @xmath158 . since @xmath145 is a resolving set for @xmath51 , there can be at most one vertex of @xmath51 with this property . if @xmath164 is odd , for any @xmath198 , we must have that @xmath205 and so @xmath161 resolves @xmath162 and @xmath203 . thus @xmath12 is a resolving set for @xmath0 of size @xmath204 , and again ( ii ) holds . if @xmath169 is even , we have @xmath206 for all @xmath207 , and thus @xmath208 are not resolved by any @xmath161 . however , by taking all but one of @xmath209 along with @xmath12 , we have that @xmath171 is a resolving set for @xmath0 of size @xmath210 , and thus ( iii ) holds . this completes the proof . straightforward examples are provided by the complete multipartite graphs @xmath46 : these are @xmath48-antipodal covers of the complete graph @xmath211 , and have diameter @xmath34 . since any resolving set @xmath145 for @xmath212 contains @xmath213 vertices , the remaining vertex is adjacent to all of @xmath145 , so case ( iii ) of proposition [ prop : folding ] applies . this gives @xmath214 , which we know from proposition [ prop : easy ] to be the exact value , so the upper bound is achieved . further such cases are discussed in section [ subsection:2antipodal ] below . some examples of where the bounds in proposition [ prop : folding ] are not achieved can be found in the tables in @xcite . first , the conway smith graph @xmath215 on @xmath216 vertices and with diameter @xmath217 is a @xmath66-antipodal cover of the kneser graph @xmath218 ; from ( * ? ? ? * table 10 ) we have @xmath219 and @xmath220 , so @xmath221 . second , the foster graph @xmath222 on @xmath223 vertices and with diameter @xmath134 is a @xmath66-antipodal cover of tutte s @xmath134-cage @xmath224 ; from ( * ? ? ? * table 3 ) we have @xmath225 and @xmath226 , so @xmath227 . in fact , in this latter case the covering graph has smaller metric dimension than its folded graph . when considering the classification of theorem [ thm : ah ] , the most immediate applications of propositions [ prop : halving ] and [ prop : folding ] are to graphs when halving and/or folding yields a primitive graph with diameter at least @xmath34 ; these are precisely the graphs in classes ah10ah13 . combined with babai s theorem [ thm : babai ] , we can obtain upper bounds on the metric dimension of any graph in those classes in terms of its parameters . first , we have the following result for bipartite graphs in classes ah1113 . [ cor : bipartite - babai ] suppose @xmath0 is a bipartite distance - regular graph of diameter @xmath67 whose halved graphs are primitive and have diameter at least @xmath34 . then we have : * @xmath228 ; * @xmath229 ( where @xmath89 is the maximum size of a set of vertices at a given distance from any vertex of @xmath0 ) . using proposition [ prop : halving ] , we know that @xmath230 . since the halved graphs are primitive distance - regular graphs with diameter @xmath68 , we can apply theorem [ thm : babai ] to them . we note that the distance classes of @xmath109 and @xmath110 are formed from the distance classes of @xmath0 , so we have @xmath231 . in case ah10 , we have that @xmath0 is antipodal but not bipartite , so we need to apply proposition [ prop : folding ] instead . [ cor : antipodal - babai ] suppose @xmath0 is a @xmath48-antipodal distance - regular graph of diameter @xmath67 whose folded graph @xmath51 is primitive and has diameter at least @xmath34 . then we have : * @xmath232 ; * @xmath233 ( where @xmath234 is the maximum size of a set of vertices at a given distance from any vertex of @xmath51 ) . for graphs in classes ah5ah9 , halving and/or folding yields either a complete or complete bipartite graph , and so we obtain upper bounds on the metric dimension of such graphs which are linear in the number of vertices . much of the remainder of this paper is devoted to improving upon this . however , we shall first consider some properties of @xmath34-antipodal graphs . in the case where @xmath0 is @xmath34-antipodal with diameter @xmath17 and folded graph @xmath51 , proposition [ prop : folding ] shows that @xmath235 if @xmath17 is odd , and @xmath236 if @xmath17 is even . however , we can obtain more detailed results in this case : the following lemma will be especially useful . for any vertex @xmath2 of @xmath0 , we denote by @xmath237 the set of vertices of @xmath0 that are at distance @xmath18 from @xmath2 . [ lemma : plusminus ] suppose that @xmath0 is a @xmath34-antipodal distance regular graph of diameter @xmath17 , whose vertex set has a @xmath34-antipodal partition @xmath238 . then , without loss of generality , a resolving set for @xmath0 can be chosen just from vertices in @xmath113 . we claim that if @xmath12 is any resolving set for @xmath0 and @xmath239 , then @xmath240 is also a resolving set . to show this , suppose that @xmath241 are resolved by @xmath242 , i.e. @xmath243 . suppose that @xmath244 , i.e. @xmath245 . since @xmath0 is distance - regular , there exists a path of length @xmath246 to some vertex in @xmath247 ; however , as @xmath242 is the unique vertex in @xmath247 , it follows that @xmath248 , and so @xmath249 . therefore , @xmath250 , and hence @xmath251 also resolves @xmath241 . by repeating the above process as required , an arbitrary resolving set for @xmath0 may be transformed into a resolving set consisting only of vertices in @xmath113 , and the result follows . if @xmath0 is bipartite as well as antipodal , and has odd diameter @xmath252 , then it is necessarily @xmath34-antipodal ( otherwise , it would contain an odd cycle of length @xmath253 ) ; such graphs form classes ah5 and ah12 . we have the following theorem . [ thm : bip_2ant_odd ] let @xmath3 be a bipartite , @xmath34-antipodal distance - regular graph with odd diameter @xmath254 , with folded graph @xmath146 of diameter @xmath255 . then @xmath256 . by proposition [ prop : folding ] we have @xmath235 . to show the converse , we consider the bipartition @xmath103 , which is also a @xmath34-antipodal partition , so for any @xmath148 , its preimages in @xmath136 are @xmath257 and @xmath258 . by lemma [ lemma : plusminus ] , there exists a resolving set @xmath259 for @xmath0 with @xmath107 . let @xmath260 . for all @xmath261 , there exists @xmath262 with @xmath263 . if these distances both lie in the interval @xmath264 , we must have @xmath265 similarly , if these distances both lie in the interval we have @xmath266 otherwise , we have ( without loss of generality ) that and @xmath267 , which implies that @xmath268 is even and @xmath269 is odd . therefore , @xmath11 resolves @xmath270 , and thus @xmath145 is a resolving set for @xmath51 . hence @xmath271 . this completes the proof . immediately , we have the following corollary about the graphs in class ah5 . [ cor : doubledclique ] the metric dimension of the graph @xmath60 , i.e. a complete bipartite graph with a @xmath272-factor removed , is @xmath273 . we have that @xmath60 is a bipartite , @xmath34-antipodal distance - regular graph with diameter @xmath58 , and its folded graph is the complete graph @xmath274 which has metric dimension @xmath273 . then we apply theorem [ thm : bip_2ant_odd ] . in terms of the asymptotic behaviour of metric dimension , this tells us that for graphs in class ah5 , namely @xmath275 with @xmath59 , we have @xmath102 , in common with classes ah3 and ah4 . so we may regard this class as another pathological case . for graphs in class ah12 , we have no change in the asymptotics from what we saw in the previous subsection , although we can be more precise . the following definition gives us an alternative interpretation of theorem [ thm : bip_2ant_odd ] . [ defn : bipdouble ] let @xmath3 be a graph . then the _ bipartite double _ ( or _ bipartite cover _ ) of @xmath0 is the bipartite graph @xmath276 whose vertex set consists of two disjoint copies of @xmath136 , labelled @xmath113 and @xmath114 , and where @xmath277 and @xmath278 are adjacent in @xmath276 if and only if @xmath5 and @xmath11 are adjacent in @xmath0 . for example , the bipartite double of a complete graph @xmath279 is the graph @xmath60 from corollary [ cor : doubledclique ] above . more generally , if @xmath0 is distance - regular with diameter @xmath17 and odd girth @xmath280 , then @xmath276 is distance - regular with diameter @xmath280 , and is an antipodal @xmath34-cover of @xmath0 . furthermore , any distance - regular graph of odd diameter which is both bipartite and antipodal must arise this way ( see ( * ? ? ? * 4.2d ) ) , and the bipartition @xmath238 is also a @xmath34-antipodal partition . thus theorem [ thm : bip_2ant_odd ] may be rephrased as follows . [ thm : bipdouble ] suppose that @xmath0 is a distance - regular graph of diameter @xmath17 with odd girth @xmath280 . then the metric dimension of its bipartite double @xmath276 is equal to the metric dimension of @xmath0 . as an example of its applications , theorem [ thm : bipdouble ] may be applied to the following infinite family . the _ odd graph _ @xmath281 has as its vertex set the collection of all @xmath282-subsets of a @xmath283-set , with two vertices adjacent if and only if the corresponding @xmath282-sets are disjoint . ( the odd graph @xmath284 is the petersen graph . ) this graph is distance - regular , has diameter @xmath285 and odd girth @xmath286 , so therefore satisfies the conditions of theorem [ thm : bipdouble ] ; its bipartite double is known as the _ doubled odd graph_. ( see ( * ? ? ? * 9.1d ) for further details . ) consequently , we have another corollary . [ cor : doubledodd ] the odd graph @xmath281 and doubled odd graph @xmath287 have equal metric dimension , which is at most @xmath288 . it follows from ( * ? ? ? * theorem 6 ) that the odd graph @xmath281 has metric dimension at most @xmath288 . since this graph satisfies the hypotheses of theorem [ thm : bipdouble ] , the result follows . this latter corollary provides a slight improvement on theorem 3.1 of guo , wang and li @xcite , who showed that @xmath289 . however , it removes the requirement to consider the doubled odd graph separately from the odd graph . for a graph @xmath0 in class ah7 of theorem [ thm : ah ] , i.e. @xmath0 is antipodal , has diameter @xmath66 and is not bipartite , the folded graph is a complete graph , and proposition [ prop : folding ] gives a bound on @xmath290 of @xmath291 . however , a stronger result is desirable , and is possible in the case where @xmath0 is @xmath34-antipodal : such graphs are called _ taylor graphs _ and are discussed below . we shall see that , rather than using the folded graph , we can @xmath0 to a primitive strongly regular graph in a different way , which then gives us a suitable relationship to bound the metric dimension . a _ taylor graph _ is @xmath34-antipodal distance - regular graph on @xmath292 vertices , obtained via the following construction , due to taylor and levingston @xcite . suppose that @xmath293 is a strongly regular graph with parameters @xmath294 . construct a new graph @xmath0 by taking two copies of the set @xmath136 labelled as @xmath295 , along with two new vertices @xmath296 , and defining adjacency as follows : let @xmath297 be adjacent to all of @xmath113 , @xmath298 be adjacent to all of @xmath114 , @xmath299 and @xmath300 ( in @xmath0 ) if and only if @xmath301 ( in @xmath302 ) , and @xmath303 if and only if @xmath156 and @xmath304 ( where @xmath305 denotes adjacency ) . from the construction , one may verify that @xmath0 is indeed distance - regular , @xmath34-antipodal , and that the folded graph is a complete graph @xmath306 . the given labelling of the vertices ensures that @xmath251 is the unique antipode of @xmath242 , for all @xmath307 . for any vertex @xmath119 of @xmath0 , let @xmath308 $ ] denote the subgraph of @xmath0 induced on the set of neighbours of @xmath119 . the construction ensures that @xmath302 is isomorphic to both @xmath309 $ ] and @xmath310 $ ] ; for any other vertex @xmath119 , @xmath308 $ ] is also strongly regular with the same parameters , but need not be isomorphic to @xmath302 . as a simple example , one may use this construction to obtain the icosahedron from a @xmath311-cycle , which has parameters @xmath312 . for further examples , we refer to the table of strongly regular graphs in brouwer and haemers ( * ? ? ? * 9.9 ) . a _ two - graph _ @xmath313 is a pair @xmath314 , where @xmath136 is a set and @xmath315 is a collection of @xmath66-subsets of @xmath136 , with the property that any @xmath217-subset of @xmath136 contains an even number of members of @xmath315 . from any element @xmath141 , one may form a graph with vertex set @xmath316 by deleting @xmath119 from all triples which contain it , and taking the resulting pairs as edges ; such a graph is a _ descendant _ of @xmath313 . the collection of all descendants of @xmath313 is referred to as a _ switching class _ , because of the relationship with the operation of seidel switching ; for more information on two - graphs and switching classes , see @xcite . a two - graph is _ regular _ if every @xmath34-subset of @xmath136 occurs in a constant number of members of @xmath315 . in @xcite , taylor proved that the descendants of a regular two - graph on @xmath317 points are necessarily strongly regular graphs with parameters @xmath294 . taylor and levingston @xcite subsequently showed the following ; see also ( * ? ? ? * 1.5 ) for an account of their work . [ thm : taylev ] @xmath318 * an antipodal @xmath34-cover of @xmath306 is necessarily a taylor graph . * there exists a one - to - one correspondence between taylor graphs and regular two - graphs on @xmath317 points . * the isomorphism classes of descendants of a regular two - graph @xmath313 , i.e. the members of a switching class of strongly regular graphs with parameters @xmath294 , are precisely the isomorphism classes of induced subgraphs @xmath319 $ ] of the corresponding taylor graph @xmath0 . to confuse matters , the strongly regular graphs arising as the descendants of a regular two - graph associated with the unitary group @xmath320 , as discovered by taylor @xcite , are sometimes referred to as `` taylor s graph '' : see @xcite . distance - transitive taylor graphs were classified in 1992 @xcite . the main result of this section is to relate the resolving sets for a taylor graph with those for the descendants of the corresponding regular two - graph . [ thm : taylor ] let @xmath313 be a regular two - graph with corresponding taylor graph @xmath0 , and let @xmath321 be the switching class of descendants of @xmath313 . choose a descendant @xmath302 with the smallest metric dimension , i.e. @xmath322 for all descendants @xmath323 . then we have : * @xmath324 ; and * @xmath325 for all descendants @xmath323 . first , we show that @xmath326 . label the vertices of @xmath0 as @xmath327 , as described above , and choose a smallest resolving set @xmath328 for @xmath302 . we will show that @xmath329 is a resolving set for @xmath0 . since @xmath12 is a resolving set for @xmath302 , then for any pair of distinct vertices @xmath4 , there exists @xmath330 such that @xmath331 . since @xmath332 and @xmath333 , it follows that @xmath334 resolves the pair @xmath335 . likewise , @xmath336 resolves the pair @xmath337 ; however , since @xmath0 is @xmath34-antipodal , lemma [ lemma : plusminus ] shows that @xmath334 will also resolve the pair @xmath337 . any pair of vertices of the form @xmath338 will be resolved by @xmath297 , as @xmath339 for any @xmath340 , and @xmath341 for any @xmath258 . finally , any pair involving one of @xmath297 or @xmath298 will be resolved by @xmath297 , since @xmath298 is the unique vertex at distance @xmath66 from @xmath297 . now we will establish the reverse inequality , i.e. @xmath342 . choose a resolving set @xmath1 for @xmath0 of size @xmath290 . now choose some vertex @xmath343 , and consider the subgraph @xmath308 $ ] induced on the set @xmath344 of neighbours of @xmath119 . since @xmath0 is a taylor graph , @xmath308 $ ] must be isomorphic to a descendent @xmath323 of the regular two - graph @xmath313 , and thus has diameter @xmath34 . furthermore , the vertices in @xmath345 form one part of a @xmath34-antipodal partition , so by applying lemma [ lemma : plusminus ] , we may assume that the remaining vertices of @xmath1 are all neighbours of @xmath119 . since @xmath1 is a resolving set for @xmath0 , then for any @xmath346 , there exists a vertex @xmath347 that resolves the pair @xmath348 ; note that @xmath349 , as @xmath119 is clearly adjacent to all of its neighbours . furthermore , for any pair of vertices @xmath346 , we have that @xmath350}(u , v)$ ] : since @xmath308 $ ] is an induced subgraph , @xmath5 and @xmath2 are adjacent in @xmath0 if and only if they are adjacent in @xmath308 $ ] , while if @xmath5 and @xmath2 are not adjacent , they have distance @xmath34 in @xmath0 ( in a path through @xmath119 ) and distance @xmath34 in @xmath308 $ ] ( since it has diameter @xmath34 ) . as we assumed that @xmath351 , this shows that @xmath352 is a resolving set of size @xmath353 for @xmath308 $ ] consequently , we have @xmath354 ) \leq \mu(\gamma)-1,\ ] ] as required , and this concludes the proof of part ( i ) . to prove part ( ii ) , we note that a given descendant @xmath323 need not arise in the manner described above , i.e. induced on the set of neighbours of a vertex @xmath119 of a minimum resolving set for @xmath0 . however , any resolving set for @xmath0 may be used to construct a resolving set of the same size for @xmath323 . suppose that @xmath355 $ ] for some vertex @xmath11 . if @xmath1 is a minimum resolving set for @xmath0 that does not contain @xmath11 , then we can still apply lemma [ lemma : plusminus ] to assume that @xmath356 , and the same argument as above shows that @xmath1 is also a resolving set for @xmath357 $ ] . therefore , @xmath358 , and we have @xmath359 and part ( ii ) follows . we remark that in the case of vertex - transitive taylor graphs ( such as those obtained from paley graphs ) , all descendants are isomorphic , and the result simply states @xmath324 . the result in part ( ii ) of theorem [ thm : taylor ] seems a little unsatisfactory : a better result would be that all descendants of a given taylor graph ( i.e. all strongly graphs in the same switching class ) have the same metric dimension , although the author was unable to show this . there is computational evidence to support such a claim . it is known that strongly graphs with the same parameters need not have the same metric the paley graph on 29 vertices has metric dimension @xmath360 , while the other strongly regular graphs with parameters @xmath361 , which fall into five switching classes , all have metric dimension @xmath311 ( see ( * ? ? ? * table 2 ) ) . furthermore , the @xmath362 strongly regular graphs with parameters @xmath363 , which fall into exactly @xmath364 switching classes @xcite , all have metric dimension @xmath360 ( see ( * ? ? ? * table 13 ) ) . ( as an application of theorem [ thm : taylor ] , we know that all @xmath364 taylor graphs on @xmath365 vertices have metric dimension @xmath133 . ) given what we know about the metric dimension of primitive strongly regular graphs from theorem [ thm : babai ] , we can combine this with theorem [ thm : taylor ] to obtain bounds on the metric dimension of taylor graphs . [ cor : taylorbabai ] suppose that @xmath0 is a taylor graph with @xmath366 vertices . then ( a ) @xmath367 and ( b ) @xmath368 ( where @xmath32 is the valency of a descendant of @xmath0 ) . in particular , @xmath369 . of course , if the descendants of a taylor graph @xmath0 are strongly regular graphs with logarithmic metric dimension , then this carries over to @xmath0 . for example , if @xmath0 has a paley graph as a descendant , then this has metric dimension at most @xmath370 ( as shown by fijav and mohar @xcite ) ; is this case it follows that @xmath371 . what can be said about the metric dimension of @xmath48-antipodal distance - regular graphs of diameter @xmath66 , where @xmath372 ? in this section , we will consider graphs in class ah6 , namely bipartite distance - regular graphs with diameter @xmath66 . now , if @xmath0 is such a graph , its halved graphs of @xmath0 will be complete graphs , and so proposition [ prop : folding ] gives an upper bound of @xmath373 on its metric dimension ; as in the previous section , it would be desirable to improve on this . a _ symmetric design _ ( or _ square @xmath34-design _ ) with parameters @xmath374 is a pair @xmath375 , where @xmath376 is a set of @xmath2 _ points _ , and @xmath315 is a family of @xmath32-subsets of @xmath376 , called _ blocks _ , such that any pair of distinct points are contained in exactly @xmath377 blocks , and that any pair of distinct blocks intersect in exactly @xmath377 points . it follows that @xmath378 . a symmetric design with @xmath379 is a _ projective plane _ , while a symmetric design with @xmath380 is known as a _ biplane _ @xcite . the _ incidence graph _ of a symmetric design is the bipartite graph with vertex set @xmath381 , with a point @xmath382 adjacent to block @xmath383 if and only if @xmath384 . it is straightforward to show that the incidence graph of a symmetric design is a bipartite distance - regular graph with diameter @xmath66 . the converse is also true ( see ( * ? ? ? * 1.6 ) ) : any bipartite distance - regular graph of diameter @xmath66 gives rise to a symmetric design . the _ dual _ of a symmetric design is the design obtained from the incidence graph by reversing the roles of points and blocks ; @xmath375 and its dual both have the same parameters . the _ complement _ of a symmetric design @xmath375 has the same point set @xmath376 , and block set @xmath385 . the incidence graph of @xmath386 is obtained from that of @xmath375 by interchanging edges and non - edges across the bipartition . if @xmath375 has parameters @xmath374 , then @xmath386 has parameters @xmath387 . suppose @xmath0 is the incidence graph of @xmath375 . for any distinct points @xmath388 and any distinct blocks @xmath389 , we have @xmath390 , while @xmath391 if @xmath384 and @xmath392 if @xmath393 . it follows that the incidence graph of a symmetric design and that of its complement have the same metric dimension ; clearly , this holds for the incidence graph of a symmetric design and its dual , as the incidence graphs are isomorphic . we observe that if a resolving set @xmath12 for @xmath0 is contained entirely within @xmath376 or entirely within @xmath315 , we have @xmath394 . in the case of the unique symmetric design with @xmath395 , where the blocks are all the @xmath396-subsets , the graph obtained is @xmath60 , which has metric dimension @xmath273 by corollary [ cor : doubledclique ] . from now on , we shall assume that @xmath61 , where it is natural to ask if smaller resolving sets exist which must therefore contain both types of vertex . one approach for constructing resolving sets is as follows . suppose @xmath0 is the incidence graph of a symmetric design @xmath375 . a _ split resolving set _ for @xmath0 is a set @xmath397 , where @xmath398 and @xmath399 , chosen so that any two points @xmath241 are resolved by a vertex in @xmath400 , and any two blocks @xmath401 are resolved by a vertex in @xmath402 . we call @xmath402 and @xmath400 _ semi - resolving sets _ for the blocks and points of the design . the smallest size of a split resolving set will be denoted by @xmath403 . we note that a split resolving set is itself a resolving set , as any vertex will resolve a pair @xmath404 , given that the parities of the distances to @xmath119 and to @xmath405 will be different ; therefore , we only need consider resolving point / block pairs . clearly , we have @xmath406 . a straightforward observation is that , for any two blocks @xmath401 , the point @xmath119 resolves the blocks @xmath401 if and only if @xmath119 lies in exactly one of the two blocks ( i.e. @xmath407 and @xmath393 , or vice - versa ) , and a block @xmath405 resolves the points @xmath241 if and only if exactly one of @xmath241 lies in @xmath405 . in the case of projective planes , the blocks of the design are usually referred to as _ lines _ , and are denoted by @xmath408 . it is known that for a projective plane to exist , we have @xmath409 and @xmath410 for some integer @xmath411 , called the _ order _ of the projective plane . we let @xmath412 denote the incidence graph of a projective plane @xmath413 . a _ blocking set _ for a projective plane @xmath414 of order @xmath411 is a subset of points @xmath415 chosen so that every line @xmath416 contains at least one point in @xmath1 ; moreover , @xmath1 is a _ double blocking set _ if every line @xmath417 contains at least two points in @xmath1 . ball and blokhuis @xcite showed that , for @xmath418 , a double blocking set has size at least @xmath419 , with equality occurring in the plane @xmath420 when @xmath411 is a square . also , one can easily construct a double blocking set of size @xmath421 by taking the points of three non - concurrent lines . double blocking sets and semi - resolving sets are related by the following straightforward proposition . [ prop : blocking ] a double blocking set with a single point removed forms a semi - resolving set for the lines of a projective plane . let @xmath1 be a double blocking set for @xmath414 . any pair of distinct lines @xmath422 intersects in a unique point @xmath119 . since @xmath1 is a double blocking set , there exists @xmath423 such that @xmath424 and @xmath425 . hence @xmath120 resolves the lines @xmath422 . by the same argument , there also exists @xmath426 such that @xmath427 and @xmath428 . this redundancy allows us to delete a point from @xmath1 and still have a semi - resolving set ; however , deleting two points from @xmath1 may prevent us from resolving some pairs of lines . using proposition [ prop : blocking ] , we may obtain a split resolving set for @xmath412 of size @xmath429 by taking a semi - resolving set of this form for the points along with the dual of such a set for the lines ( where @xmath430 denotes the smallest size of a double blocking set in @xmath413 , and @xmath431 denotes the dual plane ) . if @xmath413 is self - dual then this simplifies as @xmath432 . at the problem session of the 2011 british combinatorial conference , the author asked whether this was best possible . in 2012 , the question was answered by hger and takts @xcite for the desarguesian plane @xmath420 . [ thm : ht1 ] a semi - resolving set for @xmath420 has size at least @xmath433 ; for a square prime power @xmath434 , this is at least @xmath435 . of course , a minimum resolving set for @xmath412 need not be a split resolving set . w. j. martin ( personal comunication ) was able to construct a non - split resolving set for @xmath412 of size @xmath436 ( see ( * ? ? ? * figure 1 ) ) , and conjectured that this was best possible ( except for small orders ) . this conjecture was also proved in the 2012 paper of hger and takts @xcite . [ thm : ht2 ] for a projective plane @xmath413 of order @xmath437 , the metric dimension of its incidence graph @xmath412 is @xmath438 . hger and takts also gave a complete description of all resolving sets of this size : see ( * ? ? ? asymptotically , their result gives the following . [ cor : pgasymp ] for a projective plane @xmath413 whose incidence graph @xmath412 has @xmath439 vertices , we have @xmath440 . for a projective plane @xmath413 of order @xmath411 , we know that the number of vertices of @xmath412 is @xmath441 , and by theorem [ thm : ht2 ] ( for @xmath437 ) , the metric dimension is @xmath442 . one might ask if this result holds for symmetric designs with @xmath443 . in the next subsection , we consider this possibility . a _ polarity _ of a symmetric design @xmath375 is a bijection @xmath444 which preserves the point / block incidence relation . it is straightforward to see that @xmath375 admits a polarity if and only if there is an ordering of the points and blocks so that the incidence matrix of the design is symmetric . a point is called _ absolute _ if it is incident to its image under @xmath445 . a polarity @xmath445 is said to be _ null _ if no points are absolute . in this situation , the incidence matrix has zero diagonal , and so is the adjacency matrix of a graph @xmath302 ; this graph is strongly regular with parameters @xmath446 ( see ( * ? ? ? * theorem 2.1 ) ) . we observe that a symmetric design @xmath375 may admit more than one null polarity , and the corresponding strongly regular graphs need not be isomorphic . the reader is referred to the book of ionin and shrikhande ( * ? ? ? * 7.4 ) for more details , and for several constructions of infinite families of such designs , in particular families arising from hadamard matrices . conversely , if one has a strongly regular graph @xmath293 with parameters @xmath446 , the bipartite double of that graph ( recall definition [ defn : bipdouble ] ) is the incidence graph of a symmetric design with parameters @xmath374 , which a null polarity in an obvious way : the points and blocks may be by @xmath113 and @xmath114 respectively , and the map @xmath447 is a null polarity . we note that non - isomorphic graphs may give rise to the same symmetric design : for instance , the @xmath448 lattice @xmath449 and the shrikhande graph are non - isomorphic strongly regular graphs with parameters @xmath450 , yet their bipartite doubles are isomorphic ( and give rise to a @xmath451-biplane ) . given this relationship with bipartite doubles , one may ask if there is a result similar to theorem [ thm : bipdouble ] which can be applied here to find the metric dimension of @xmath0 , and we have the following theorem . [ thm : nullpolarity ] let @xmath0 be the incidence graph of a non - trivial @xmath374 symmetric design with a null polarity , and let @xmath302 be a corresponding strongly regular graph with parameters @xmath446 . then @xmath452 . since @xmath0 is the bipartite double of @xmath302 , we will label the points and blocks of the design by @xmath113 and @xmath114 respectively . suppose @xmath328 is a resolving set for @xmath302 ; we will show that @xmath453 is a resolving set for @xmath0 . now , for any distinct @xmath454 , we have @xmath455 , while @xmath456 if @xmath457 in @xmath302 and @xmath458 if not . clearly , any vertex resolves @xmath459 ( as the distances will have different parities ) , so it suffices to consider resolving pairs of vertices of the form @xmath460 and @xmath461 . if @xmath462 , then clearly @xmath463 resolves the pair @xmath460 and @xmath464 resolves the pair @xmath461 ( and likewise if @xmath121 ) , so we assume that @xmath465 . since @xmath12 is a resolving set for @xmath302 , there exists @xmath330 where @xmath466 ; without loss of generality , this implies that @xmath467 and @xmath468 , so therefore @xmath469 and @xmath470 , and thus @xmath336 resolves the pair @xmath460 . similarly , @xmath334 resolves the pair @xmath461 . hence any pair of vertices of @xmath0 is resolved by a vertex in @xmath453 , and we are done . immediately , we have the following corollary , which is reminiscent of corollary [ cor : taylorbabai ] for taylor graphs . [ cor : symmbabai ] let @xmath0 be the incidence graph of a non - trivial @xmath374 symmetric design where @xmath471 and which has a null polarity . then we have ( a ) @xmath472 and ( b ) @xmath473 . let @xmath302 be the strongly regular graph associated with the design ( as in theorem [ thm : nullpolarity ] ) . theorem [ thm : babai ] gives us upper bounds on @xmath474 , and the result follows . since @xmath0 has @xmath475 vertices , corollary [ cor : symmbabai ] gives @xmath476 . in cases where @xmath2 and @xmath32 have a linear relationship , babai s stronger bound yields an upper bound of @xmath477 on @xmath474 . there are a number of infinite families of such designs , in particular arising from hadamard matrices ( see ( * ? ? ? * 7.4 ) , and also @xcite ) ; the incidence graphs of such designs therefore have metric dimension @xmath478 . we remark , however , that theorem [ thm : nullpolarity ] can never be applied to the incidence graphs of projective planes , since it is known that a projective plane can not admit a null polarity ( see ( * ? ? ? * proposition 4.10.1 ) ) . what happens if we remove the hypothesis that the design has a null polarity ? is it still true that the incidence graph @xmath0 of a non - trivial symmetric design with @xmath443 has metric dimension @xmath479 ? do the bounds of corollary [ cor : symmbabai ] still hold ? while the method used in the proof of theorem [ thm : nullpolarity ] depends on the existence of the null polarity ( to exploit the relationship with strongly regular graphs ) , there are instances of parameter sets @xmath374 with multiple isomorphism classes of designs , of which only some have a null polarity , yet all of their incidence graphs have the same metric dimension . for example , there are five non - isomorphic symmetric designs with parameters @xmath480 , giving rise to four non - isomorphic graphs ; only one of these designs ( the complementary design of the projective geometry @xmath481 ) has a null polarity , yet all four graphs have metric dimension @xmath134 ( see ( * ? ? ? * table 2 ) ) . also , there are three non - isomorphic @xmath451-biplanes , with three non - isomorphic incidence graphs ; again , only one has a null polarity , yet all three have metric dimension @xmath134 ( see ( * ? ? ? * table 2 ) ) . in this paper , we have begun the systematic study of the metric dimension of imprimitive distance - regular graphs . using propositions [ prop : halving ] and [ prop : folding ] , combined with the results of babai ( theorem [ thm : babai ] ) for primitive graphs , we saw that bounds on the metric dimension of graphs in classes ah912 ( of theorem [ thm : ah ] ) follow immediately . the difficult cases are bipartite or antipodal graphs of diameter @xmath66 ( whose halved or folded graphs are complete ) or graphs of diameter @xmath217 which are both bipartite and antipodal ( where the folded graph is complete bipartite ) . in sections [ section : antipodal3 ] and [ section : incidence ] we obtained some results on these classes , where the diameter is @xmath66 ; obtaining general bounds for the metric dimension of graphs in these classes remains open . it seems plausible that the bounds in corollary [ cor : symmbabai ] should hold for symmetric designs in general , if the requirement on the existence of a null polarity is removed . the other cases ( namely antipodal covers of complete and complete bipartite graphs ) also have connections with design theory ( see @xcite ) and this relationship may prove to be useful when studying these cases . the author would like to thank the anonymous referees whose suggestions led to significant improvements to this paper . this research was begun while the author was a postdoctoral fellow at the university of regina and ryerson university . the author would like to thank the faculty of science at each university for financial support . j. j. seidel and d. e. taylor , two - graphs : a second survey , in _ algebraic methods in graph theory ( szeged , 1978 ) _ , pp . 689711 , colloq . jnos bolyai * 25 * , north - holland , amsterdam / new york , 1981 . d. e. taylor and r. levingston , distance - regular graphs , in _ combinatorial mathematics ( proc . internat . combinatorial theory , australian nat . canberra , 1977 ) _ , pp . 313323 , lecture notes in math . * 686 * , springer - verlag , berlin , 1978 .

when finding the nonzero eigenvalues for hamiltonian eigenvalue problems it is especially important to locate not only the unstable eigenvalues ( i.e. , those with positive real part ) , but also those which are purely imaginary but have negative krein signature . these latter eigenvalues have the property that they can become unstable upon collision with other purely imaginary eigenvalues , i.e. , they are a necessary building block in the mechanism leading to the so - called hamiltonian - hopf bifurcation . in this paper we review a general theory for constructing a meromorphic matrix - valued function , the so - called krein matrix , which has the property of not only locating the unstable eigenvalues , but also those with negative krein signature . these eigenvalues are realized as zeros of the determinant . the resulting finite dimensional problem obtained by setting the determinant of the krein matrix to zero presents a valuable simplification . in this paper the usefulness of the technique is illustrated through prototypical examples of spectral analysis of states that have arisen in recent experimental and theoretical studies of atomic bose - einstein condensates . in particular , we consider one - dimensional settings ( the cigar trap ) possessing real - valued multi - dark - soliton solutions , and two - dimensional settings ( the pancake trap ) admitting complex multi - vortex stationary waveforms .

1212.2951

hamiltonian eigenvalue problems have a time - honored history , as they arise in numerous applications stemming from fluid mechanics , celestial mechanics , optical and atomic physics among many other disciplines ; see for some recent examples the books @xcite . especially in higher dimensional settings these problems can rapidly become fairly computationally intractable , at least as concerns providing the full diagonalization of the relevant matrix ( e.g. , when it stems from the linearization around two - dimensional vortex structures or three - dimensional vortex - rings @xcite ) . it is therefore highly desirable to be able to reduce the dimensionality of the calculation by providing a technique that can capture the main features of the linearization spectrum through suitable reductions to a finite dimensional eigenvalue problem . it is the aim of the present work to provide a general overview , as well as a systematic set of case examples of such a method . the approach that will be developed will be based on the so - called krein matrix @xcite . the krein matrix is a meromorphic matrix - valued function constructed via a lyapunov - schmidt reduction , and consequently recasts the infinite - dimensional eigenvalues problem as a finite - dimensional one . the construction is such that the eigenvalues are realized as points for which the krein matrix is singular . the computation and visualization of the determinant of this matrix can serve as a tool to identify the eigenvalues of the original problem . when determining the spectrum for the hamiltonian eigenvalue problem , there are two spectral sets to consider : those with positive real part , and those with negative krein index ( signature ) . the latter eigenvalues are purely imaginary ; however , upon collision with eigenvalues of positive krein signature it will generically be the case that a hamiltonian - hopf bifurcation will occur , which leads to an oscillatory instability . while the eigenvalues with positive real part are easy to visually identify , an additional calculation is necessary in order to identify the signature of a purely imaginary eigenvalue . as we will see later in this paper , the krein matrix is constructed in such a manner that the eigenvalues with negative signature can be identified _ graphically_. consequently , all of the ( potentially ) unstable eigenvalues can be identified visually . a related question is : how many ( potentially ) unstable eigenvalues are there to locate ? for a given problem it may be theoretically possible to establish an upper bound on the real part of all eigenvalues ; however , in many problems of interest there is no upper bound on the imaginary part of the eigenvalue . the hamiltonian - krein index , which will be discussed in detail later in this paper , counts the number of eigenvalues with positive real part , as well as the number of purely imaginary eigenvalues with negative krein signature . in the problems presented in this paper this index will be finite , and it will be related to the number of negative directions of the constrained second variation of the energy ( the underlying wave is realized as a critical point of the constrained energy ) . thus , while no bound is present on the imaginary part of all of the eigenvalues , since there are only a finite number of ( potentially ) unstable eigenvalues , there will be an upper bound for this set . a theoretical determination of this bound is probably not possible ; however , it can be determined numerically through the krein matrix . the computation of the krein matrix in this paper will be numerical for each case study . unfortunately , at this point in time we do not know of any examples for which the krein matrix can be explicitly computed . our hope is that for special problems , e.g. , the nonlinear schrdinger equation with the 1-soliton potential , such a calculation may be possible . this will be the topic of future research . we note that there is reason to be optimistic that this will be a fruitful direction for research , in that it is possible in some special cases to explicitly construct the evans function ( another eigenvalue counting tool ) , e.g. , see ( * ? ? ? * chapters 9.3 and 10.4 ) and the references therein . our presentation will be structured as follows . in , we first provide an overview of the hamiltonian - krein index theory for hamiltonian eigenvalue problems . afterwards , we focus on the construction of the krein matrix and provide a summary of its properties . in we consider specific examples stemming from the application of the krein matrix analysis to vortex dynamical states that are of intense recent interest in the field of atomic physics . in particular , we consider single ( unit charge ) vortex states that are presently fairly routine to produce / observe @xcite , but also examine the case of the recently studied vortex dipoles . for the sake of completeness , we also examine in one - dimensional analogs of such vortex states ; namely , multi - dark soliton structures that have been studied theoretically ( see e.g. @xcite and references therein ) and also have been observed experimentally in @xcite . finally , in we summarize our findings and present our conclusions , as well as some potential topics for future study . regarding the states studied in , it is known that the precessional dynamics of a single vortex is connected with the negative krein signature eigenvalues of the corresponding linearization spectrum , see @xcite . the vortex dipoles were produced by dynamical experiments involving flow past an obstacle @xcite , or quenching through the bose - einstein condensation ( bec ) transition @xcite . furthermore , their dynamical properties and structural robustness were studied in considerable length in recent works in the physical literature @xcite . experimental works on this theme - especially , @xcite - were concerned with issues such as : 1 . the equilibrium configuration , which is also explored herein ; 2 . the epicyclic motion around this equilibrium , which was analyzed through the negative krein signature eigenvalues of the dipole s linearization spectrum . it should be noted in passing that the study of spectral stability properties in bec settings ( and particularly for single- and multi - charge vortices ) is a subject rapidly gaining momentum , as evidenced by the recent work of @xcite on the subject . tk gratefully acknowledges the support of the jack and lois kuipers applied mathematics endowment , a calvin research fellowship , and the national science foundation under grant dms-1108783 . pgk and dy gratefully acknowledge support from afosr grant fa9550 - 12 - 1 - 0332 , from nsf - dms grant of number 0806762 , and pgk also from the alexander von humboldt foundation and the binational science foundation , grant number 2010239 . consider the hamiltonian eigenvalue problem @xmath0 acting on a hilbert space @xmath1 with inner - product @xmath2 . the operator @xmath3 is skew - symmetric bounded operator with a bounded inverse , and the operator @xmath4 is hermitian with a compact resolvent . the space @xmath5 is assumed to be dense . under these assumptions it is well - known that the spectra of @xmath6 , namely @xmath7 , is all point spectra , each eigenvalue has finite multiplicity , and infinity is the only possible accumulation point of the eigenvalues . if we further assume that each of the operators has zero imaginary part , i.e. , @xmath8 , then the eigenvalues satisfy the quartet symmetry that if @xmath9 , then the set @xmath10 . furthermore , the algebraic multiplicities of each of the eigenvalues in the quartet matches , e.g. , @xmath11 . the hamiltonian - krein index theory has a long history ( see @xcite and the references therein ) . the index theory is used to relate @xmath12 , which is the total number ( including multiplicity ) of negative eigenvalues of @xmath13 , to the total number of eigenvalues @xmath9 with positive real part . in general , for hermitian operators @xmath14 we will denote the number of negative eigenvalues including multiplicity ) by @xmath15 . the details for the following discussion can be found in , e.g. , @xcite , and an abbreviated version is included here for the sake of completeness . the eigenvalue problem arises from linearizing a particular hamiltonian system about some type of steady state solution ( solitary wave , spatially periodic wave , etc . ) . the underlying system has @xmath16 symmetries , which means that @xmath17\ge n$ ] , as each symmetry generates an element of the kernel , and the kernel elements generated in this fashion are linearly independent . we will henceforth assume that @xmath17=n$ ] , with @xmath18 . now , the generalized kernel is found by solving @xmath19 , where @xmath20 . now , the symmetries generate conserved quantities , and it turns out to be the case the manner in which these quantities are generated allows us to solve @xmath21 for each @xmath22 : denote these solutions as @xmath23 ( see @xcite ) . thus , @xmath24 , and @xmath25 . if we set @xmath26 as @xmath27 then by the fredholm alternative it will be the case that @xmath28 if and only if @xmath29 is nonsingular : this will henceforth be assumed . in @xmath30 is a solution to for @xmath31 , then by the fredholm alternative it must be the case that for any @xmath20 , @xmath32 consequently , the eigenvalue problem is not solved on all of @xmath33 , but is instead solved on the @xmath16 co - dimensional constrained space @xmath34 , where @xmath35 . thus , it will be important not to calculate @xmath12 , but instead @xmath36 , where @xmath37 , and @xmath38 is the orthogonal projection . it was most recently shown in @xcite that @xmath39 thus , the matrix @xmath29 also plays a significant role in determining the number of negative eigenvalues for the constrained operator . we are now ready to state the main result regarding the number of eigenvalues for @xmath6 which have positive real part . let * @xmath40 : the total number of positive real - valued eigenvalues ( including algebraic multiplicity ) * @xmath41 : the total number of complex eigenvalues with positive real and imaginary part ( including algebraic multiplicity ) our assumptions on @xmath3 and @xmath13 imply the four - fold eigenvalue symmetry @xmath42 , so there will be @xmath43 eigenvalues with positive real part . there is one more set of eigenvalues which we wish to count . first , for a self - adjoint operator @xmath44 and a subspace @xmath45 with basis @xmath46 , let the hermitian matrix @xmath47 be given by @xmath48 with this notation , for each nonzero eigenvalue @xmath49 with associated eigenspace @xmath50 , let @xmath51 the quantity @xmath52 is known as the negative krein index of the eigenvalue , and if @xmath53 the eigenvalue is said to have negative krein signature . counting only those purely imaginary eigenvalues with positive imaginary part , we say the total negative krein index is given by @xmath54 although we will not prove it here , it can easily be shown that @xmath55 for @xmath56 . consequently , there will be @xmath57 purely imaginary eigenvalues with negative krein signature . the hamiltonian - krein index is the weighted sum of these indices ; namely ; @xmath58 the first two terms in the index count the total number of eigenvalues with positive real part , and the last term counts all those purely imaginary eigenvalues with negative krein signature . the major result is that the hamiltonian - krein index is intimately related to the number of negative eigenvalues of the constrained operator , @xmath59 @xcite . as a consequence of we know that there is a finite and prescribed number of ( potentially ) unstable eigenvalues , which , as we will see in subsequent sections , greatly assists in a numerical search for these eigenvalues . additional implications of are : 1 . if @xmath60 is odd , then @xmath61 , so that the underlying wave is spectrally unstable 2 . if @xmath62 , then it is generically the case that the wave is orbitally stable 3 . if @xmath60 is even , then the wave may be spectrally stable with @xmath63 ; however , the orbital stability of the wave is generally not known the krein signature has important implications beyond what is present in the hamiltonian - krein index . if two purely imaginary eigenvalues collide , then after the collision they can attain a nonzero real part ( this is the so - called hamiltonian - hopf bifurcation ) if and only if they have opposite signature . if the two eigenvalues have the same signature , they will simply pass through each other . for a more detailed discussion see ( * ? ? ? * chapter 7.1 ) . consequently , we can think of eigenvalues having negative krein signature as being potentially unstable eigenvalues . [ r:23 ] if the eigenvalue problem is canonical , i.e. , of the form @xmath64 where @xmath65 are self - adjoint operators , then it is possible to derive a lower bound on @xmath40 . for example , if @xmath66=1 $ ] and @xmath67=0 $ ] , then it is true that if @xmath68 , then @xmath61 @xcite . this result will be discussed in further detail in . before we can construct the krein matrix associated with the general hamiltonian spectral problem , we must first find an equivalent self - adjoint pencil . in addition , we must relate the hamiltonian - krein index for the original problem to that of the pencil problem . first suppose that @xmath69 . upon setting @xmath70 , and defining @xmath71 it is not difficult to see that solving is equivalent to solving the canonical system @xmath72 we continue by writing down an equivalent eigenvalue problem for which the operators involved no longer have a nontrivial kernel . regarding @xmath73 we have @xmath74 the discussion in the previous subsection allows us to say that @xmath75 . upon setting @xmath76^\perp$ ] to be the orthogonal projection , it is known from ( * * section 3 ) that for nonzero eigenvalues is equivalent to the system @xmath77 each of the operators @xmath78 are self - adjoint , and the assumption that @xmath29 is nonsingular implies that each is also nonsingular on the range of @xmath79 . this allows us to introduce the invertible self - adjoint operators @xmath80 and note that @xmath81 being a bounded operators implies that from our original assumptions on @xmath13 we have @xmath82 . and rewrite as the quadratic pencil @xmath83 in a similar fashion , if we initially assumed that @xmath49 , the the equivalent pencil would be @xmath84=0,\quad \re u=(\im\lambda)\,{\mathcal{j}}{\mathcal{s}}[\im u].\ ] ] in conclusion , we have that solving the original eigenvalue problem is equivalent to solving the linear pencil @xmath85 which is precisely the spectral problem that was studied by @xcite . the effect of the eigenvalue mapping is illustrated in . eigenvalues with positive real part and nonzero imaginary part are mapped in a one - to - one fashion to eigenvalues with nonzero imaginary part , and the four - fold symmetry is reduced to the reflection symmetry @xmath86 . the system has an unstable eigenvalue @xmath87 with positive real part if and only if the system has an eigenvalue @xmath88 with @xmath89 or with @xmath90 . let us conclude by computing the hamiltonian - krein index for the pencil . first consider the purely imaginary eigenvalues . it is straightforward to show that @xmath91 where @xmath92 , and @xmath93 . since @xmath94 , it is consequently the case that @xmath95 where for @xmath96 the krein index is found by computing @xmath97 thus , the total negative krein index is the same for the pencil as it is for the original problem . now consider those eigenvalues for the original problem with nonzero real part . set @xmath98 to be the multiplicity of the real - valued eigenvalue @xmath96 , and let @xmath99 be the multiplicity of the eigenvalue for @xmath90 ( it is clearly the case that @xmath100 ) . since @xmath101 we clearly have that @xmath102 here we are using the notation that @xmath103 is the multiplicity of the positive real - valued eigenvalue for , and @xmath104 is the multiplicity of an eigenvalue with positive real and imaginary parts . upon summing over all of the eigenvalues , and using , we get the new result that the hamiltonian - krein index for the original problem is related to that of the pencil in the following manner : [ lem : kreinindex ] consider the linear pencil as derived from the eigenvalue problem . for the pencil let @xmath40 denote the total number of negative real - valued eigenvalues ( counting multiplicity ) , @xmath41 the total number of eigenvalues with positive imaginary part ( counting multiplicity ) , and @xmath105 the total negative krein index of all the positive real - valued eigenvalues , where the index for a single eigenvalue @xmath96 with associated eigenspace @xmath106 is given by @xmath97 then with the hamiltonian - krein index given as in , the eigenvalues for the pencil satisfy @xmath107 [ r:25 ] abusing notation a bit , we will , e.g. , denote @xmath98 as the total number of negative real eigenvalues for the pencil , and @xmath103 the total number of positive real eigenvalues for the original eigenvalue problem . we can then summarize the above discussion to say @xmath108 and that the indices are @xmath109{(\ref*{e:11})}}\\ \left(k_{\mathrm{r}}+2k_{\mathrm{c}}+2k_{\mathrm{i}}^-\right)(z)&=2k_{\mathrm{ham}},\quad\mathrm{problem}\,\,{\hyperref[{e:16}]{(\ref*{e:16})}}. \end{split}\ ] ] it will be convenient to relate the hamiltonian - krein index to @xmath110 ( the equality follows from the fact that @xmath3 has bounded inverse ) . as for the number of negative directions for @xmath111 , we have @xmath112^\perp})= { \mathrm{n}}({\mathcal{l}})-{\mathrm{n}}({\bm{\mathit{d}}})=k_{\mathrm{ham}}.\ ] ] the first equality follows from the definition of @xmath111 and the fact that constrained operator maps the subspace @xmath113 to itself , the second equality follows from a simple change of variables , the third equality follows from , and the fourth equality is from . thus , since the number of negative directions is invariant under inversion , with respect to the pencil alone we can rewrite the conclusion of as @xmath114 it can be concluded that if the underlying wave is orbitally stable , then @xmath115 is positive definite ; otherwise , the operator must be indefinite , although it will necessarily have only a finite number of negative directions . and their images , @xmath116 . the ( green ) crosses denote two pairs of real eigenvalues @xmath117 and their images on the negative real axis , @xmath118 . the ( blue ) boxes denote two pairs of purely imaginary eigenvalues @xmath117 and their images on the positive real axis . the filled square has a positive krein signature , while the empty square has a negative krein signature , so that @xmath119 . ] we now turn to the problem of constructing a meromorphic matrix - valued function , the krein matrix , which has the property that it is singular precisely for those values of @xmath88 which correspond to nonzero eigenvalues for the pencil @xmath120 where @xmath121 are defined in . the krein matrix was introduced in @xcite , and the interested reader should consult that work for the details associated with the following discussion ( also see ( * ? ? ? * section 3 ) ) . we construct the krein matrix by projecting off the finite - dimensional negative space of the operator @xmath115 , which as we have already seen in has dimension @xmath60 , and then using a lyapunov - schmidt reduction to compute an equivalent eigenvalue problem . let @xmath122 denote the @xmath60-dimensional negative subspace of @xmath123 , and let @xmath124 be the orthogonal projection . define the constrained operators @xmath125 the operator @xmath126 is positive definite and symmetric , whereas from the index theorem ( * ? ? ? * theorem 2.1 ) the symmetric operator @xmath127 satisfies @xmath128 since @xmath126 is positive definite we can define the conjugated operator @xmath129 and note that the invertibility of @xmath115 yields @xmath130 upon writing a potential eigenfunction to the linear pencil as @xmath131 where @xmath132 is an orthonormal spectral basis for @xmath122 , and applying the projection @xmath133 to , the pencil problem becomes @xmath134 on the other hand , upon applying @xmath135 to , where @xmath136 is the identity operator , and taking the inner product of the resulting equation with @xmath137 for @xmath138 , yields the system of equations @xmath139 where @xmath140 . solving the first equation for @xmath141 and plugging this result into the second equation yields the problem @xmath142 here we are using the notation @xmath143 the matrix @xmath144 is known as the _ krein matrix_. we now relate the properties of the problem to those for the original pencil . by construction @xmath88 is an eigenvalue for the pencil with corresponding eigenfunction @xmath145 if and only if @xmath146 , where @xmath147 . if for a particular eigenvalue @xmath88 it is true that @xmath148 , then it is necessarily true that @xmath149 . on the other hand , if @xmath150 , then @xmath151 , so it can be the case that @xmath152 . now , if @xmath90 , then it will always be the case that @xmath88 is an eigenvalue if and only if @xmath149 . on the other hand , if @xmath153 is an eigenvalue with @xmath152 , then it is the case that for the krein matrix constructed by projecting off of the negative directions of @xmath154 , say @xmath155 , we would necessarily have @xmath156 . finally , if @xmath96 is an eigenvalue with @xmath152 , then the eigenvalue has positive krein signature . in this case the eigenvalue is realized as a removable singularity of the krein matrix , i.e. , @xmath157 is a pole of the krein matrix for which the residue is the zero matrix . if the eigenvalue has negative krein signature , it will necessarily be the case that @xmath149 . in conclusion , we can use @xmath158 as a meromorphic function whose zeros correspond to eigenvalues . the ( potential ) singularities of the krein matrix arise through @xmath159 at the eigenvalues of the self - adjoint operator @xmath160 . in order to use the krein matrix to say something about the krein signature of a real positive eigenvalue , it will be helpful to factor the determinant as a finite product . one can easily observe that @xmath144 is symmetric for all @xmath88 , i.e. , @xmath161 ; in particular , it is hermitian for @xmath162 . this allows us for @xmath162 to extract the @xmath60 eigenvalues of the krein matrix , @xmath163 , hereafter called the _ krein eigenvalues _ , in a meromorphic fashion . thus , instead of finding eigenvalues by looking for the zeros of the determinant of the krein matrix , we can look for the zeros of each individual krein eigenvalue . there will be precisely @xmath60 of these krein eigenvalues . although we will not pursue this line of thought herein , each krein eigenvalue can be thought of as a real meromorphic analogue of the evans function @xcite . both krein eigenvalues and the evans function detect eigenvalue for a linear eigenvalue problem through the zeros . until recently the evans function was constructed solely via a dynamical systems argument , which necessitated that the eigenvalue problem be essentially in one space variable . since the krein matrix is constructed via a functional analytic argument , the spatial dimension associated with the eigenvalue problem is not relevant . in the case that @xmath164 . the krein eigenvalue @xmath165 is denoted by a thick ( green ) dashed curve , and its vertical asymptotes are given by a thin ( green ) dashed curve . the krein eigenvalue @xmath166 is denoted by a thick ( black ) curve , and its vertical asymptotes are given by a thin ( black ) curve . the eigenvalues for the pencil with positive krein index are denoted by ( blue ) squares , and those with negative index are shown as ( red ) circles . ] for real - valued @xmath88 , the properties of the krein eigenvalues are as follows . first , @xmath167 so that each krein eigenvalue is negative for large negative @xmath88 . this follows from the fact that @xmath168 is a negative definite matrix . second , if @xmath169 is a pole of the the krein eigenvalue @xmath163 , it will then be the case that @xmath170 furthermore , if @xmath171 is a simple eigenvalue of @xmath160 , so that @xmath171 is a pole of the krein matrix , then it will be the case that @xmath171 is a simple pole for only one of the krein eigenvalues . in other words , for all but one of the krein eigenvalues the pole @xmath171 of the krein matrix is a removable singularity . finally , if @xmath172 is a simple positive zero of the krein eigenvalue @xmath163 , then it is true that @xmath173;\ ] ] in other words , the slope of the krein eigenvalue at a zero gives definitive information regarding the krein index of the eigenvalue . this is the most useful of the properties of the krein eigenvalues , for it allows us graphically locate those eigenvalues which have negative krein signature . this discussion is summarized in . if the zero of a krein eigenvalue is not simple , then for the eigenvalue in question there will be an associated jordan chain which has length equal to the order of the zero . furthermore , this situation can only arise upon the collision of an ( almost ) equal number of purely imaginary eigenvalues with positive and negative krein signature ( see ( * ? ? ? * section 2.2 ) for the details ) . following we know that for the original eigenvalue problem the hamiltonian - krein index is @xmath174 while for the pencil the index is @xmath175 the individual indices are related via @xmath176 for the original eigenvalue problem there will be an infinite number of purely imaginary eigenvalues , all of which but a finite number will have positive krein index . the purely imaginary eigenvalues with negative krein signature can be determined by first constructing the krein matrix of for the pencil , and then plotting the resultant krein eigenvalues for @xmath162 . in particular , the eigenvalues with negative signature will correspond to those values of @xmath162 such that for some @xmath177 , @xmath178 if the order of a zero of a krein eigenvalue is two or higher , then a collision of eigenvalues of opposite krein signature has occurred . the discussion in this section assumed that the original hamiltonian eigenvalue problem is not in the canonical form of . if the problem is in canonical form , adjustments must be made : this is discussed in the application presented in . we now wish to use the krein matrix to identify the eigenvalues of which have nonzero real part , or which are purely imaginary and have negative krein signature . we intend to explore these spectral features and confirm them against a full linear stability analysis for an example of significant interest to recent experimental applications , namely the study of a single vortex @xcite and of a pair of vortices @xcite in two - dimensional bose - einstein condensates as described by the gp equation . the hamiltonian - krein index of tells us how many of these zeros for the pencil we need to find in order to fully capture all of the ( potentially ) unstable eigenvalues . the real eigenvalues for the linearized problem correspond to negative real eigenvalues for the pencil , and the eigenvalues with nonzero real part correspond to eigenvalues with nonzero imaginary part for the pencil . regarding the eigenvalues with negative signature , from the discussion of the previous section this means that we need to identify the real and positive eigenvalues for the pencil for which a krein eigenvalue has a zero , and the slope of the curve at the zero is positive . the model under consideration for the case of pancake - shaped bose - einstein condensates @xcite is the ( 2 + 1)-dimensional gross - pitaevskii equation in the dimensionless form @xmath179 here , @xmath145 is the macroscopic wave function , @xmath180 is the external harmonic potential , @xmath181 is the frequency of the trap , @xmath182 is the chemical potential , and @xmath183 is the laplace operator , i.e. , @xmath184 . for the problem discussed herein , we assume @xmath185 . we begin by assuming that the steady - state problem is solved , and we will denote that solution by @xmath186 . here @xmath187 are real - valued functions . writing @xmath188 where @xmath189 , the wave will have the property that @xmath190 exponentially fast as @xmath191 . as for the associated eigenvalue problem , abusing notation a bit and writing @xmath192 we see that at @xmath193 we have the linear problem @xmath194 where @xmath195 the eigenvalue problem of the form , i.e. , @xmath196 arises upon using the separation of variables ansatz @xmath197 . with respect to the standard inner - product on @xmath198 the operator @xmath3 is boundedly invertible and skew - symmetric , while the operator @xmath13 is self - adjoint . furthermore , since the potential @xmath199 grows quadratically , and the magnitude of the solution @xmath200 decays exponentially fast as @xmath191 , it is the case that @xmath13 has a compact resolvent . in order to construct the krein matrix for the spectral problem we first follow and set @xmath201 the construction of the krein matrix requires that we consider the spectral problem on the space perpendicular to the kernel of both operators . the fact that solutions to are invariant under @xmath202 and the spatial rotation @xmath203 means that ( generically ) the kernel of @xmath13 will be two - dimensional , so that ( generically ) @xmath204 if the solution has a radially symmetric density , i.e. , @xmath205 , then the dimension of each kernel is ( generically ) one with @xmath206 with this information at hand , and setting @xmath207^\perp$ ] to the the orthogonal projection , we can now compute the restricted operators @xmath208 to create the pencil @xmath209 the krein matrix @xmath210 ( see ) will be created from this pencil using the algorithm described in the previous section . here we assume that the solution is a vortex of charge one . the solution is of the form @xmath211 where the density profile satisfies @xmath212 , and the phase rotates by @xmath213 around the vortex core , justifying the topological charge of the structure @xcite . the density and phase profiles for the single vortex state are shown in , and the corresponding linearization spectrum is shown in . the absolute value of this configuration is radially symmetric ( a feature not shared by the vortex dipole state considered below ) , while its decay for @xmath214 is dictated by the underlying linear problem being exponential for small @xmath182 @xcite and resembling an inverted parabola for large @xmath182 @xcite ( the latter decay features are also shared by the vortex dipoles below ) . the dependence of the relevant eigenvalues as a function of the canonical parameter of the system , namely the chemical potential @xmath182 associated with the number of atoms in the condensate , has been quantified previously ; see e.g. @xcite . we will confirm these findings via the krein matrix , and showcase particular examples for a few representative values of @xmath182 . and chemical potential @xmath215 . the left panel shows the contour plot of the density , and the right panel is the phase plot for the wave function . in this case , the total number of grid points is @xmath216 , and the spatial step size is @xmath217 . , width=340 ] for vortex dipole state when @xmath216 and @xmath217 as a function of the chemical potential @xmath182 . since the ( red ) dashed curve emanating from @xmath218 corresponds to an eigenvalue with negative krein signature , the fact that @xmath219 means that there are no other eigenvalues with positive real part ( see ) . the wave is spectrally stable for all considered values of the chemical potential . the eigenvalue which has negative krein signature is highlighted as the ( red ) dashed line . larger ( in magnitude ) eigenvalues along the imaginary axis have a positive krein signature and do not lead to instabilities.,width=340 ] as we saw above , the kernels of the operators @xmath65 each have dimension one . for small amplitude vortices where the nonlinear interactions are ( almost ) negligible it can be shown via an analysis similar to that presented in @xcite that the hamiltonian - krein index satisfies @xmath219 . following the discussion in we then know that for the pencil @xmath220 while for the original problem @xmath221 thus , if for the pencil there are two positive real zeros of the krein eigenvalues which correspond to an eigenvalue with negative krein index , the rest of the spectrum for the pencil must be positive and purely real . in other words , if for the original eigenvalue problem there is a purely imaginary eigenvalue with negative krein index , the rest of the spectrum must be purely imaginary with positive krein index . consequently , once we numerically find one purely imaginary eigenvalue with negative krein signature , or one set of eigenvalues with nonzero real part which satisfy the hamiltonian eigenvalue symmetry , we need not search for any additional unstable eigenvalues . they simply do not exist . regarding the numerical solution and analysis of the problem , we first discretize the differential operator via a centered finite difference scheme using @xmath222 data points and a spatial interval of @xmath223 . the single vortex state - indeed , any stationary solution - is found by applying newton s method to the discretized problem , treating it as a two - dimensional boundary value problem and starting from a suitably proximal to it initial guess . as for the spectral problem , the smallest ( in norm ) eigenvalues are computed using the matlab function eigs " on the discretized version of the operator @xmath6 ; i.e. , we construct the linearization operator on the same domain as used for the fixed point iteration and utilize standard routines ( such as an implicitly restarted arnoldi method within matlab ) to compute some of the smallest magnitude eigenvalues of the corresponding spectral problem . this provides us with the linear stability results that will be compared with the krein matrix ones in the figures that follow . it is worth noting here that these spectral linearization results can only be obtained by taking advantage of the sparse structure of the underlying discretization matrix . should the full eigenvalue spectrum of the linearization problem be sought in this highly - demanding two - dimensional computation , matlab s routine `` eig '' would have been unable to produce the corresponding numerical results . . in the left panel the left plot is the numerically generated plot of the @xmath219 krein eigenvalues . for this problem the krein matrix is a meromorphic multiple of the identity ; hence , the two krein eigenvalues coincide . the ( red ) crosses are the poles of the krein matrix . if the pole is removable ( e.g. , for @xmath224 ) , then it corresponds to an eigenvalue with positive krein signature . here we see two positive real zeros of the krein eigenvalues for which the functions have positive slope . thus , for the pencil @xmath225 with @xmath226 , which means that for the original problem @xmath227 with @xmath228 . the wave is spectrally stable , but is not a ( local ) minimizer for the constrained hamiltonian . in the right figure the eigenvalues for @xmath6 are denoted by ( blue ) circles . the ( green ) crosses represent the poles of the krein matrix , and the ( red ) triangles are the eigenvalues of @xmath6 which are realized as zeros of the krein eigenvalues . the purely imaginary eigenvalue with negative krein signature is shown as a _ filled _ ( red ) triangle . , width=302 ] we now turn to a direct comparison of the numerical results for eigenvalues obtained from the linear stability analysis ( which has partially been presented in earlier publications , see e.g. @xcite ) and from the krein matrix , presented for the first time for vortex patterns of the gp equation herein . here , we have confirmed the above findings with @xmath229 in , and @xmath230 in . in both of these figures a spatial discretization spacing of @xmath231 was used . similar results were found with smaller values of @xmath223 . in the right panel of each figure the spectrum of @xmath6 was computed for both the original formulation of and the corresponding pencil formulation , and the results of the two were found to agree within the accuracy of the eigenvalue solver . these spectral results were then compared with the prediction of the krein matrix in the right panel . the agreement is excellent . unexpectedly , it turns out to be the case that for the problem at hand the krein matrix is a meromorphic multiple of the identity ; hence , the two krein eigenvalues coincide . in both figures it is clear that there are then two zeros of the krein eigenvalues at which the slope of the curve is positive : this corresponds to an purely imaginary eigenvalue with negative krein signature . as stipulated by hamiltonian - krein index in , the rest of the spectrum must be purely imaginary , and this is indeed seen to be the case . for the spectrum there is not only the eigenvalue at @xmath232 which is due to the phase u@xmath233 gauge invariance of the model , there is also a double eigenvalue at @xmath234 . the latter frequency of double multiplicity is the so - called dipolar mode of the condensate and pertains to a symmetry ( oscillation of the entire condensate cloud in the @xmath235 or @xmath236 direction with the trap frequency ) , and is hence invariant with respect to variations in @xmath182 . one of these modes pertains to a pole of the krein matrix , while the other is a zero of a krein eigenvalue . the eigenvalue with negative krein signature always lies between the origin and this double pair and is known to tend to the origin as the chemical potential @xmath182 increases @xcite . since the eigenvalue has negative signature , as predicted by the theory it is realized as a zero of a krein eigenvalue . . in the left panel the left plot is the numerically generated plot of the @xmath219 krein eigenvalues . for this problem the krein matrix is a meromorphic multiple of the identity ; hence , the two krein eigenvalues coincide . the ( red ) crosses are the poles of the krein matrix , and the removable singularities ( e.g. , @xmath237 ) correspond to eigenvalues with positive krein signature . here we again see two positive real zeros of the krein eigenvalues for which the functions have positive slope . the wave is then spectrally stable with @xmath227 for the problem . the notation used in the right figure is similar to that in .,width=302 ] to showcase the relevance of this pole , the residues of the poles of the krein matrix are computed . if the numerically computed residue of the pole is of @xmath238 , then we say that the pole is removable , and hence it corresponds to an eigenvalue with positive krein signature . the residue is computed via the numerical integration @xmath239here @xmath240 is the relevant pole of the krein matrix , @xmath222 is the number of integration points on the simple positively oriented closed contour @xmath241 which surrounds the pole ( here the contour is without loss of generality chosen to be a square centered on the pole ) , @xmath242 is the point on the integration contour around the pole , and @xmath243 is the segment on the integration path . when the chemical potential @xmath229 the first relevant pole we choose is located at @xmath244 . for the increment of @xmath245 it is seen that @xmath246 so that @xmath247 . the simple pole is removable , and corresponds to a purely imaginary eigenvalue ( the first red cross above zero in ) . the fact that the singularity is removable is evidenced by the fact that neither of the krein eigenvalues has a singularity at the point . now consider the pole at @xmath248 . it is seen that @xmath249 so that @xmath250 , which is nonzero by our criterion . alternatively , we see that the pole is not removable because the krein eigenvalues have a singularity at that point . since the singularity is not removable , this point does not correspond to an eigenvalue . and chemical potential @xmath215 . the left panel shows the contour plot of the density , and the right panel is the phase plot for the wave function . in this case , the total number of grid points is @xmath216 , and the spatial step size is @xmath251 . , width=340 ] for vortex dipole state when @xmath252 and @xmath253 as a function of the chemical potential @xmath182 . the wave is spectrally unstable with @xmath254 when @xmath255 . when @xmath256 we know by the hamiltonian - krein index for the wave that there are no other unstable eigenvalues . when @xmath257 in the figure , the eigenvalue associated with ( red ) circles has negative krein signature . by the hamiltonian - krein index we then know that there are no other non - plotted eigenvalues with positive real part , so that in this case the wave is spectrally stable.,width=340 ] and @xmath258 . the left panel is the plot of the two krein eigenvalues : the ( red ) crosses represent the poles of the krein matrix , and the removable singularities correspond to eigenvalues with positive krein signature . note that @xmath225 for the pencil , and that as expected the zeros of the krein eigenvalues which correspond to eigenvalues with negative krein signature coincide . in the right panel the ( blue ) circles are the eigenvalues for @xmath6 , the ( red ) triangles are the eigenvalues which correspond to zeros of the krein eigenvalues , and the ( green ) crosses are the eigenvalues which correspond to removable singularities of the krein matrix - these eigenvalues have positive krein signature . the purely imaginary eigenvalue with negative krein signature is shown as a _ filled _ ( red ) triangle . the labeling of eigenvalues in the right panel is the same as for .,width=340 ] we now turn to a spectral analysis for the vortex dipole state , which is a stationary vortex - antivortex state ( see for a typical example of its density and phase ) . when @xmath185 the vortex dipole state exists for @xmath259 it is interesting to note here that , as shown in @xcite , such states do not exist at the linear limit , but only bifurcate through a supercritical pitchfork ( symmetry - breaking ) event at a critical point from the dark soliton , which corresponds to the real - valued ( first ) excited state of the two - dimensional harmonic oscillator . since the density is not radially symmetric , it will ( generically ) be the case that @xmath260=2 $ ] . furthermore , it is seen numerically that @xmath261 , so that by we have the same index count for the pencil as ; namely , @xmath262 as in the previous example , there will be two krein eigenvalues to be plotted . unlike the previous example , the krein matrix will not be a meromorphic multiple of the identity ; hence , the krein eigenvalues will not generically overlap . if there are two positive real zeros of the krein eigenvalues for which the curves have positive slope , then the spectrum for the pencil will be positive and purely real with @xmath263 . otherwise the underlying wave will be unstable . as we will see , in our examples this instability will arise when @xmath264 . it is important to note here that for the pencil it must be the case that if @xmath225 , then each krein eigenvalue has a positive zero at precisely the same point . this fact is a consequence of the relationship of the spectrum between the original eigenvalue problem and the pencil ; in particular , the fact that the spectrum of the pencil doubles up " the nonzero spectrum for the original problem . and @xmath265 . the left panel is the plot of the two krein eigenvalues : the ( red ) crosses represent the poles of the krein matrix , and the removable singularities correspond to eigenvalues with positive krein signature . since there are no positive real zeros with positive slope , it must be the case that for the pencil @xmath266 . in the right panel the ( blue ) circles are the eigenvalues for @xmath6 , the ( red ) triangles are the eigenvalues which correspond to zeros of the krein eigenvalues , and the ( green ) crosses are the eigenvalues which correspond to removable singularities of the krein matrix - these eigenvalues have positive krein signature . the labeling of eigenvalues in the right panel is the same as for . the only difference is that the ( blue ) stars represent the complex eigenvalues with nonzero real part as found by the zeros of the krein eigenvalues . , width=340 ] the spectrum once again features the twofold degenerate dipolar modes which are associated with the oscillation with the trap frequency of the whole condensate in the @xmath235 and @xmath236 directions . in addition to these modes , there exists a negative krein signature mode , which in this case collides with the mode departing from zero . this , in turn , leads to the formation of an interval of @xmath182-values where the spectrum possesses complex eigenfrequencies associated with oscillatory instability . therefore , when studying the computation of the krein matrix of the vortex dipole state and comparing its eigenvalues against the linearization analysis , we select two cases , namely @xmath258 and @xmath265 . as we see in , the former is before the collision of the two modes ( i.e. , @xmath267 ) , and the latter is after the collision has occurred ( i.e. , @xmath256 ) . for larger values of @xmath182 the complex - valued eigenvalues return to the imaginary axis . when @xmath258 we show the computation of the krein eigenvalues ( left panel ) and the linearization spectrum ( right panel ) in . in the right panel it can be observed that the zeros of the krein eigenvalues are consistent with the eigenvalues found via the linear stability analysis , except for the real eigenvalues , which are @xmath268 and are the numerical approximation of the known zero eigenvalues . regarding the left panel it is seen from the inset that @xmath269 . from we know that the hamiltonian - krein index for the pencil is @xmath270 hence , for the pencil it is the case that @xmath225 with @xmath226 . for the original problem it is then true that @xmath227 with @xmath228 . the eigenvalue with negative krein signature is denoted by a ( red ) filled triangle in the right panel . as a consequence of the index theory we know that all other eigenvalues must be purely imaginary with positive krein signature . when @xmath265 we show the computation of the krein eigenvalues ( left panel ) and the linearization spectrum ( right panel ) in . again , the zeros of the krein eigenvalues are consistent with the eigenvalues found via the linear stability analysis except for the real eigenvalues , which are @xmath268 and are the numerical approximation of the known zero eigenvalues . from the right panel we see that @xmath271 . upon using we then have that @xmath272 with @xmath273 . as we see in the left panel , as expected the krein eigenvalues have no negative zeros , and no positive zeros with positive slope ; in other words , all of the zeros of the krein eigenvalues correspond to purely imaginary eigenvalues with positive krein signature . in conclusion , as a consequence of the index theory we know that except for the one quartet of simple eigenvalues with nonzero real and imaginary parts , all other eigenvalues must be purely imaginary with positive krein signature . the unstable eigenvalues can not be detected via the graph of the krein eigenvalues in the left panel of . however , they can be found by providing a phase plot for each krein eigenvalue , which is done in when @xmath265 . the axes denote the complex @xmath88-plane , and the colorbar corresponds to the phase of the eigenvalues of the krein matrix . the points where the phase becomes singular correspond to the location of the complex eigenvalues , and are labeled by the white spots . it is clear from a standard winding number argument that each zero of the krein eigenvalue is simple , which is a verification of the fact that @xmath266 . . the colors represent the argument of the krein eigenvalues , @xmath274 and @xmath275 . the two white spots on each of the two panels represent the complex eigenvalues , corresponding to @xmath276.,width=340 ] the model under consideration is the ( 1 + 1)-dimensional gross - pitaevskii equation in the dimensionless form @xmath277 where now @xmath278 is the external harmonic potential . when doing numerical computations for this problem we assume @xmath279 . for this problem we are interested in studying the spectrum associated with real - valued solutions . these will be denoted by @xmath280 , and they will be realized as solutions of the ode @xmath281 the solutions will have the property that @xmath282 exponentially fast as @xmath283 . the eigenvalue problem associated with the real - valued solution will be exactly that as given in , except that now @xmath13 will be diagonal , i.e. , @xmath284 , with @xmath285 the fact that @xmath13 is diagonal , i.e. , the eigenvalue problem is in the canonical form @xmath286 means that not only can more be said about the hamiltonian - krein index for the original problem , but the index for the associated pencil will also change . this latter amendment follows from the fact that the pencil can be formed directly without first passing to the intermediate stage of doubling - up " the eigenvalue problem . because the operators @xmath65 1 . are sturmian ( e.g. , see ( * ? ? ? * chapter 2.3 ) ) 2 . contain the unbounded potential @xmath287 3 . contain a potential which decays exponentially fast as @xmath283 , the general properties of @xmath288 are well - understood : 1 . @xmath289 is composed solely of point eigenvalues 2 . each eigenvalue is geometrically and algebraically simple . the operator @xmath290 will ( generically ) be invertible , while @xmath291 implies that @xmath67=1 $ ] . if we let @xmath292 denote the orthogonal ( spectral ) projection , and set @xmath293 then the search for nonzero eigenvalues for the original problem is equivalent to finding the spectrum of the pencil @xmath294 [ cols="^,^ " , ] the hamiltonian - krein index for the original eigenvalue problem is given by @xmath295 ( compare with ) . since @xmath79 is a spectral projection , it will be the case that @xmath296 . as for @xmath297 , by using the index theorem in @xcite it is the case that @xmath298 upon using the fact that @xmath299 we have the rewritten expression @xmath300 in other words , the slope of the power curve @xmath301 has a direct influence on the number of negative directions of the operator @xmath290 ( this is the so - called vakhitov kolokolov stability criterion for canonical hamiltonian eigenvalue problems ) . in conclusion , the hamiltonian - krein index for the canonical problem is @xmath302 there is also an instability criterion . the classical result of @xcite ( recently reproven in @xcite using the krein matrix ) allows us to say that a lower bound on @xmath103 is caused by a difference in the number of negative directions of @xmath154 and @xmath115 . in other words , @xmath303|,\ ] ] which in particular implies the previously stated result that @xmath304 if @xmath305 . ( see ) versus the chemical potential @xmath182 : the power plotted here is scaled by the @xmath306 . the right panel demonstrates the stationary solutions as @xmath182 varies . the solid blue , dashed green , and dash - dotted red curves represent the cases of @xmath307 respectively . here the grid size is @xmath308 and the spatial interval is @xmath309.,width=340 ] regarding the hamiltonian - krein index for the pencil , the fact that the original system no longer needs to be doubled - up " in order to be put into canonical form means that we just need to count eigenvalues with respect to the eigenvalue mapping @xmath310 ( see ) . in particular , @xmath311 which in the end means that the index for the pencil is precisely that for the original problem , and is given in . additionally , in the construction of the krein matrix the size no longer directly depends on @xmath312 ; instead , it will be of size @xmath296 . , for which @xmath313 and @xmath227 . in the left panel is the plot of the three krein eigenvalues . as predicted by the theory , there is only one positive zero of a krein eigenvalue for which the slope is positive . in the right panel the eigenvalues for @xmath6 are denoted by ( blue ) circles . the ( red ) triangles are the eigenvalues of @xmath6 which are realized as zeros of the krein eigenvalues . the ( blue ) stars represent the complex eigenvalues with nonzero real part as found by the zeros of the krein eigenvalues . the eigenvalue with negative krein signature is marked with a ( red ) filled triangle . note that this eigenvalue is close to an eigenvalue with positive krein signature , which implies that the system is close to a hamiltonian - hopf bifurcation point.,width=340 ] in the previous section we considered the spectral stability of solutions for which two krein eigenvalues could be used to locate the spectrum . we now consider an example for which three krein eigenvalues are needed . the steady - state solution @xmath314 to be considered , hereafter called a 3-dark - soliton , is realized as a continuation of the gauss - hermite function @xmath315 from the chemical potential @xmath316 ( recall that we assume @xmath279 ) . let us first determine the hamiltonian - krein index associated with this solution . a weakly nonlinear analysis along the lines of that presented in , e.g. , @xcite , shows that for @xmath317 small , the power satisfies the relationship @xmath318 ( the details are left for the interested reader ) . as we see from the numerics ( see ) , this relationship holds for all values of @xmath182 under consideration . regarding the indices of the operators @xmath65 , the combination of @xmath314 having three zeros and sturm - liouville theory implies that @xmath319 ; consequently , in the krein matrix analysis there will be three krein eigenvalues to follow . regarding the operator @xmath290 , it is the case that in the weakly nonlinear limit @xmath320 : again , this relationship holds for all @xmath182 under consideration . appealing to we see that @xmath321 ; unfortunately , the lower bound of on @xmath103 leads to no definitive conclusion . regarding the types of ( potentially ) unstable eigenvalues , we have the following possibilities : ' '' '' @xmath103 & 0&0&0&0&2&2&2&4&4&6 + ' '' '' @xmath104 & 0&1&2&3&0&1&2&0&1&0 + ' '' '' @xmath52 & 3&2&1&0&2&1&0&1&0&0 + and @xmath256 with @xmath322.,width=340 ] in we show the location of the first few eigenvalues from the linear stability analysis as a function of @xmath182 . it is always the case that @xmath323 , while @xmath104 decreases from two to one to zero . the decreasing of @xmath104 implies that @xmath52 increases from one to two to three . the critical values of @xmath182 for which @xmath104 decreases are approximately @xmath324 and @xmath325 . in we show the plot of the krein eigenvalues ( left panel ) and full linearization spectrum ( right panel ) when @xmath326 , when @xmath313 and @xmath227 . the eigenvalue in the upper half - plane with negative krein signature is denoted with a ( red ) filled triangles . in we show the plot of the krein eigenvalues ( left panel ) and full linearization spectrum ( right panel ) when @xmath327 , when @xmath256 and @xmath322 . the two eigenvalues in the upper half - plane with negative krein signature are denoted with ( red ) filled triangles . finally , in we show the plot of the krein eigenvalues ( left panel ) and full linearization spectrum ( right panel ) when @xmath328 , when @xmath329 . the three eigenvalues in the upper half - plane with negative krein signature are denoted with ( red ) filled triangles . in general , when considering @xmath16-dark - solitons to it will be the case that @xmath318 with @xmath330 . consequently , when looking for the @xmath331 ( potentially ) unstable eigenvalues for the linearized problem it will be the case that @xmath16 krein eigenvalues must be plotted . and @xmath329.,width=340 ] in the present work we have revisited hamiltonian skew - adjoint eigenvalue problems that typically arise in the linearization around a stationary state of a hamiltonian nonlinear partial differential equation . we presented a brief overview of the known facts for the eigenvalue counts of the corresponding ( potentially ) unstable spectra . we especially focused on a novel , but straightforward , plan to implement finite dimensional techniques to locate this spectrum via the singular points of the meromorphic krein matrix . we illustrate the value of the approach by considering realistic problems for recently observed experimentally multi - vortex and multi - soliton solutions in atomic bose - einstein condensates . we believe that this approach can provide a valuable alternative to the highly - demanding computations that require a diagonalization of the linearization matrix , especially in two- and three - dimensional settings . naturally , there are numerous possibilities for further exploration . it would be especially relevant for the atomic physics applications to use the present method to examine the details of the spectra of three - dimensional solutions such as vortex rings @xcite . on the other hand , from a methodological perspective , it would be also relevant to consider this approach for the case of solutions of other classes of hamiltonian problems such as the korteweg - de vries equations and its generalizations , or nonlinear klein - gordon equations . the latter also offer possibilities ( e.g. in the realm of stability of traveling waves etc . ) to consider quadratic pencils instead of linear ones , whereby extensions of the krein matrix method would be particularly desirable to develop from a rigorous mathematical viewpoint . these themes are currently under study and will be reported in future publications . s. middelkamp , p. g. kevrekidis , d. j. frantzeskakis , r. carretero - gonzlez , and p. schmelcher , bifurcations , stability , and dynamics of multiple matter - wave vortex states , phys . a * 82 * , 013646 ( 2010 ) . s. middelkamp , p. j. torres , p. g. kevrekidis , d. j. frantzeskakis , r. carretero - gonzlez , p. schmelcher , d. v. freilich , and d. s. hall , guiding - center dynamics of vortex dipoles in bose - einstein condensates , phys . a * 84 * , 011605 ( 2011 ) . g. theocharis , a. weller , j. p. ronzheimer , c. gross , m. k. oberthaler , p. g. kevrekidis , and d. j. frantzeskakis , multiple atomic dark solitons in cigar - shaped bose - einstein condensates , phys . a 81 , 063604 ( 2010 ) . p. j. torres , p. g. kevrekidis , d. j. frantzeskakis , r. carretero - gonzlez , p. schmelcher , and d. s. hall , dynamics of vortex dipoles in confined bose - einstein condensates , phys . a * 375 * , 3044 ( 2011 ) . a. weller , j. p. ronzheimer , c. gross , j. esteve , m. k. oberthaler , d. j. frantzeskakis , g. theocharis , and p. g. kevrekidis , experimental observation of oscillating and interacting matter wave dark solitons , phys . * 101 * , 130401 ( 2008 ) .

magnetic flux quanta , of value , in long josephson junctions behave as ( quasi ) solitons . fluxon dynamical states are well described by a perturbed sine - gordon equation model , with boundary conditions determined by the junction geometry and by externally applied magnetic fields , and they give rise to readily measurable physical phenomena , such as step structure in current - voltage characteristics and microwave radiation emission . devices based on fluxon propagation offer potentially interesting applications as oscillators and amplifiers as well as digital applications , described elsewhere in this volume in high performance integrated superconductive circuits . [ [ section ] ]

patt-sol9303003

the combination of ` solitons ' and ` long josephson junctions ' brings together the interests of a broad and varied mix of mathematicians , physicists and device engineers . although the underlying mathematical ideas had been around for some time , the term ` soliton ' itself was coined in 1965 by zabusky and kruskal [ 1 ] , just three years after the original work by josephson [ 2 ] . it required some time , however , for the practical importance of the soliton concept to be fully appreciated by the applied science community certainly more time than was required to absorb the significance of josephson s work . during the late 1960 s and early 1970 s ` solitons ' and ` long josephson junctions ' began to attract increasing research attention . starting from key papers by kulik [ 3 ] , scott [ 4 ] , and fulton and dynes [ 5 ] ( i warn the reader that my literature citations are subjective , partial and incomplete ) , the literature gradually developed , bringing about an improved level of understanding of the basic dynamics of solitons in long junctions . a landmark paper that was particularly influential in focusing interest on the significance of solitons in physical systems was scott _ about the same time , a number of people ( see , _ e.g. _ , fulton _ [ 7 ] ) began to realize that solitons in long junctions offer an exciting potential for practical device applications . in the intervening years research in the field has continued unabated , both ` basic ' research , with the development of refined approximate analytical tools and an ever more detailed exploration of ` exotic ' topics such as chaos , and ` applied ' research , with the development of novel devices such as phase locked arrays of oscillators and the vortex flow transistor . in this present review my essential task is to attempt an up - date of material presented in two previous reviews , one written by me in 1978 [ 8 ] and the other by niels pedersen in 1986 [ 9 ] . to make this present chapter reasonably coherent and self - consistent it will be necessary to repeat a certain amount of material already presented in these two previous reviews ; my main emphasis , however , will be on material that has appeared later . by ` long josephson junction ' we mean simply one that has one dimension , along , say , the @xmath1-coordinate , long with respect to the josephson penetration length , @xmath2 ( see , _ e.g. _ , bruynseraede _ et al . _ [ 10 ] for a concise description of the fundamentals of josephson tunneling ; for a more detailed description see barone and patern [ 11 ] ) , and the other dimension , along , say , the @xmath3-coordinate , short with respect to @xmath2 . one might well ask why we choose to focus attention on what is , after all , a rather particular geometrical configuration . i suspect that the true answer here is more a question of analytical convenience than it is of physical importance : a ` small ' josephson junction is described mathematically by an ordinary differential equation ( ode ) ; a ` long ' junction is described by a partial differential equation ( pde ) in one space dimension and time ; a ` large ' ( in both the @xmath1- and @xmath3- dimensions ) junction is described by a pde in two space dimensions and time . a @xmath4-dimensional pde is a much more complicated object than is a @xmath5-dimensional pde . long josephson junctions can be constructed in different geometries , and the junction geometry has a significant effect on the junction dynamics . the most extensively studied geometries are the overlap geometry , the inline geometry and the annular geometry [ 9 ] . the first two are somewhat similar in that they both consist of finite - length , quasi - one - dimensional strips ; the essential difference between them is the manner in which the bias current is applied to them : in the overlap geometry the current is applied perpendicular to the long dimension of the junction , whereas in the inline geometry it is applied parallel to the long dimension . in fact , overlap and inline are idealized limiting cases ; real physical finite - length junctions are almost always some combination of the overlap and the inline geometries [ 9 ] . the annular geometry , instead , is qualitatively different , consisting of a strip closed upon itself in an annulus . although somewhat more difficult to construct and control from the point of view of fabrication technology , the annular geometry junction offers the interesting possibility of studying soliton dynamics in the absence of boundary reflection effects . the electrodynamics of a josephson junction is , in general , described by a @xmath4-dimensional pde for the phase difference @xmath6 between the junction electrodes . in normalized form this equation can be written [ 12 ] @xmath7 ( we ignore here the so - called @xmath8 term ) . here , @xmath1 and @xmath3 are the spatial coordinates normalized to @xmath2 , @xmath9 is time normalized to the inverse of the josephson angular plasma frequency @xmath10 ( @xmath11 , the maximum electromagnetic propagation velocity in the junction , called also swihart velocity ) , the term in @xmath12 represents shunt dissipation due to quasi - particle tunneling ( normally assumed ohmic , but see scheuermann and chi [ 13 ] for the effect of a physically more realistic , non - ohmic term ) , the term in @xmath13 represents dissipation due to the surface resistance of the superconducting junction electrodes [ 14 ] , and subscripts denote partial derivatives . details of the normalizations may be found in ref . boundary conditions for eq . ( 1 ) , at @xmath14 and @xmath15 , are determined , respectively , by the @xmath3- and @xmath1-components of the magnetic field to which the junction is subjected [ 11 ] . for the case of a ` long ' junction , _ i.e. _ , one for which @xmath16 and @xmath17 , it is relatively simple to calculate the relevant component of the magnetic field generated by the junction bias current [ 11,12 ] . for the overlap geometry , assuming a constant ( physical ) bias current @xmath18 in the @xmath3-direction and a spatially uniform , but possibly time - varying , external magnetic field applied also in the @xmath3-direction , the model reduces to @xmath19 with the boundary conditions @xmath20 here , @xmath21 , where @xmath22 is the ( physical ) maximum pair - current density per unit length in the @xmath1-direction , and @xmath23 is a normalized measure of the @xmath3-component of the external magnetic field ( the normalization is such that , with @xmath24 , @xmath25 is the static threshold for magnetic flux entry into the junction ) . a condition for the validity of the reduction to one dimension is [ 12 ] : @xmath26 . for the inline geometry the bias enters through the boundary conditions rather than as a term in the pde , since @xmath18 is now applied in the @xmath1-direction . thus , the pde becomes @xmath27 and the boundary conditions become @xmath28 where @xmath29 . finally , for an annular junction in the absence of an externally applied magnetic field , assuming a spatially uniform bias current density , the governing pde is once again eq . ( 2 ) , where @xmath1 is now to be interpreted as the dimension along the mean circumference of the annulus , and the boundary conditions simply reflect the spatial periodicity of the system : @xmath30 where @xmath31 is an integer ( the physical significance of which is the difference between the number of solitons and the number of antisolitons trapped in the junction ; it is also known as the ` winding number ' , for reasons that will become clear during the discussion of the mechanical analog ) . in the presence of an externally applied magnetic field the right hand side of eq . ( 2 ) must be augmented by the term @xmath32 , as discussed by grnbech - jensen [ 15 ] , where @xmath33 is the normalized applied magnetic induction , @xmath34 is a unit normal ( radial ) vector , and @xmath35 is a coupling constant . the analogy between the dynamical equation describing a small josephson junction and that describing a simple plane pendulum subjected to viscous damping and an applied mechanical torque is well known and has been utilized by many authors . the extension of this analogy to the long junction case was quickly recognized and exploited see , _ e.g. _ , scott [ 4 ] , fulton [ 16 ] , and cirillo _ et al . _ [ 17 ] ; the extension consists of an elastic coupling between adjacent pendula , thus contributing a ` @xmath36 ' term to the dynamical equation ; a very simple ` pocket calculator ' version can easily be constructed by inserting dress - maker s pins into a rubber band at regular intervals . in spite of the development of high - speed digital computers in recent years , it is my opinion that the mechanical model of the long junction remains still today an extremely effective stimulus for physical intuition . it renders immediately evident , for example , the nature of the three fundamental classes of excitations on long junctions , _ viz . _ , plasma oscillations , breathers and kinks a fact that was recognized , and carefully illustrated , by fulton [ 16 ] . plasma oscillations , called also ` phonons ' or ` radiation ' in other contexts , stem from oscillatory motions of the single pendulum ( see fig . 20 of [ 16 ] ) . the lowest - order , @xmath37 ( @xmath38 is the wave number ) , mode has all of the pendula in the chain oscillating in a spatially uniform motion ; higher - order , @xmath39 , modes have a standing wave structure , with one or more spatial nodes along the pendulum chain . for the pure sine - gordon equation ( eq . ( 2 ) with loss and bias terms set to zero ) , small - amplitude plasma oscillations are characterized by the dispersion relation @xmath40 ; thus , @xmath41 is the lowest possible frequency for such oscillations . breather oscillations , called also ` bions ' in earlier literature , also stem from oscillatory motions of the single pendulum ( see fig . 28 of [ 16 ] ) . they differ from plasma oscillations in that the oscillation amplitude varies with position along the chain , but there are no spatial nodes along the chain . for the pure sine - gordon system of infinite length , breathers can have frequencies in the range @xmath42 . in contrast , kink oscillations stem from rotary , ` over - the - top ' , motions of the single pendulum ( see figs . 25 and 26 of [ 16 ] ) ; a single isolated kink is just a @xmath43 twist in the pendulum chain . viewing fulton s careful illustrations also renders intuitive the idea that a breather may be considered to be an oscillatory bound state of a kink and an anti - kink , an idea that has significant consequences for sine - gordon dynamics . in fact , thoughtful study of fulton s figures was instrumental in deriving the corresponding exact analytic solutions of the pure sine - gordon equation on the finite interval , for arbitrary oscillation amplitudes , by costabile _ [ 18 ] . at this point i should mention that whereas both breathers and kinks of the pure sine - gordon equation are properly classified as ` solitons ' from a strictly mathematical point of view , people who study applications of long josephson junctions normally reserve the term ` soliton ' to mean only kinks . the reason for this is undoubtedly that , due to the dissipative terms those in @xmath12 and @xmath13in the long junction model , plasma oscillations and breathers can exist only as short - lived transients , unless very special forms of driving terms are employed , whereas kinks are highly robust objects that can emerge from a power balance between the @xmath12- and @xmath13-dissipative terms and even a simple constant bias current . kinks are also often called ` fluxons ' in the josephson junction context because each one contains exactly one quantum of magnetic flux , @xmath0 ( @xmath44 is planck s constant and @xmath45 the electron charge ) ; they are also called ` vortices ' because the electric current surrounding a fluxon has the form of a vortex . i shall , in the following , use interchangeably even if somewhat improperly the terms ` soliton ' , ` kink ' , ` fluxon ' , and ` vortex ' to refer to these robust , particle - like objects . finally , the significance of the ` winding number ' @xmath31 in eq . ( 7 ) is now clear : if we introduce @xmath31 complete @xmath46 twists , _ i.e. _ , ` kinks ' , into the pendulum chain before closing the two ends into an annulus , the number @xmath31 is evidently a topological invariant of the system . let us suppose now that we have introduced never mind how ( for the moment)a single kink into the interior of a long junction . how will this object behave ? the answer evidently depends on the type of junction involved , _ i.e. _ , the boundary conditions , and on the values of the various parameters of the system ; the mechanical model will serve as an aid to the imagination here . to be specific , suppose that we have an overlap geometry junction with @xmath47 and @xmath48 , in the absence of an external magnetic field , _ i.e. _ , @xmath49 , with a bias current , @xmath50 . in the absence of the kink , the bias current , _ i.e. _ , torque , causes a uniform angular offset of the static equilibrium position of the pendulum chain , of value @xmath51 . when the kink is introduced , this applied torque pushes it toward one end of the junction , which end depending on the sense of winding of the kink with respect to that of the applied torque . with @xmath52 the two ends of the pendulum chain are free . if the kink arrives at the junction end with enough kinetic energy , the pendulum chain will continue to wind up , _ i.e. _ , the end pendulum will undergo a rotation of @xmath53 , and the resulting anti - kink will be pushed toward the other junction end , where the reflection mechanism will repeat . if , on the other hand , the kink s kinetic energy is not sufficient to survive the reflection , _ i.e. _ , the end pendulum does not receive enough energy to continue winding up , the kink will ` die ' at the first end , and , after a transient oscillation , the junction will relax to the spatially uniform static state , @xmath54 . the former case evidently leads to a dynamic steady state , with the kink bouncing resonantly back and forth across the junction , and each pendulum in the chain increasing its angular position by @xmath55 after each complete back and forth period of the kink . since angular velocity , @xmath56 , in the mechanical model corresponds to voltage in the junction , a state with @xmath57 should be manifested as structure in the current - voltage characteristic of the junction , a fact recognized by fulton and dynes [ 5 ] , who named such structures ` zero - field steps ' . since a kink behaves in much the same way as does a relativistic particle having a ( normalized ) limiting velocity equal to unity ( physical limiting velocity @xmath58 ) , it follows that these steps tend asymptotically to a constant voltage . an experimental current - voltage characteristic of an overlap geometry junction , showing six such zero - field steps , is shown in fig . the step index corresponds to the number of fluxons participating in the dynamic state ; the internal symmetry of a multi - fluxon state can obviously be more or less complicated , but the basic dynamic mechanism remains that of resonant back and forth propagation . let us consider now the question , left unanswered above , of how to introduce a fluxon into an experimental junction . to observe zero - field steps , the experimentalist increases the bias current from zero up to the maximum critical value , at which point the junction switches from the zero - voltage state to the gap - voltage state . the equivalent pendulum chain switches from the static state , @xmath51 , to a spatially uniform rotating state when @xmath6 reaches @xmath59 . the experimentalist then progressively reduces the bias current , tracing out the mccumber background curve of the characteristic , a piece of which is indicated as ` mcb ' in fig . correspondingly , the pendulum chain rotates at an ever slower angular velocity . if the bias current is then increased again before the junction returns to the zero - voltage state , it is possible to trace out the zero - field steps , as shown in fig . the basic mechanism involved was noted by fulton [ 16 ] through observation of the dynamics of a pendulum chain and elaborated analytically only some time later by costabile _ et al . _ [ 19 ] : at a sufficiently slow rotation velocity the pendulum chain can develop spontaneously a dynamic instability that results in the ` birth ' of one or more kinks . suppose now that to the situation described above we add a small dc magnetic field , _ i.e. _ , @xmath60 . this term corresponds to a small torque applied to just the two end pendula of the chain ; at one end the @xmath23-torque is in the same sense as the @xmath61-torque , and at the other end the two torques are opposed . the consequence of adding this term is immediately apparent : at the end where the two torques add , the reflection of fluxons is enhanced , _ i.e. _ , energy is added to the system ; at the other end energy is subtracted from the system , and fluxon reflection is impeded . increasing gradually the value of @xmath23 , this asymmetry between the two junction ends is progressively accentuated until , with an appropriate combination of parameters , a different dynamic steady state configuration can ensue : when the fluxon reaches the energy - subtracting end it ` dies ' , giving rise to a packet of small - amplitude oscillations ( a more or less complicated combination of breathers and plasma waves ) , which in turn propagates back toward the energy - adding end ; at that end , the energy of the packet is sufficient to ` give birth ' to a new fluxon , which again propagates toward the energy - substracting end , _ etc_. this is the mechanism responsible for the appearance of ` fiske steps ' in long josephson junctions [ 20 ] . in the dynamic state corresponding to the first zero - field step each pendulum in the chain advances by @xmath55 during one complete period ; in the dynamic state corresponding to the first fiske step each pendulum advances by @xmath43 during one complete period . this implies that the asymptotic limiting voltage of the first fiske step is just one - half that of the first zero - field step . analogously to the situation for zero - field steps , there also exist higher - order fiske steps having a progressively more complicated internal structure [ 20 ] . as the value of @xmath23 is increased still further , yet another dynamic steady state configuration emerges : that of flux flow . as mentioned above , @xmath25 is the static threshold for the penetration of magnetic flux into the junction in the absence of a bias current . with @xmath62 , this threshold value is reduced ( at the energy - adding end ) ; furthermore , a fluxon injected into the junction at this end is accelerated into the interior of the junction by the bias current . when the combination of @xmath23 and @xmath61 is appropriately above threshold , a dynamic steady state configuration ensues in which fluxons are injected at the energy - adding end as fast as the junction can ` swallow ' them ; these are accelerated toward the other junction end , where they may be reflected or annihilated , depending on the boundary conditions at that end . a computer simulation illustrating this situation is shown in fig . the average junction voltage in such a state is proportional to the average number of fluxons inside the junction times their average velocity . once again , since a fluxon behaves much like a relativistic particle , the current - voltage characteristic tends toward a constant voltage as the average fluxon propagation velocity approaches its limiting value . the resulting steps in the characteristic are commonly called ` flux - flow steps ' or ` velocity - matching steps ' . the potential interest of flux flow dynamics for practical electronic device applications was also recognized quite early see , _ e.g. _ , yoshida and irie [ 22 ] . the basic dynamical phenomena in an inline geometry junction are qualitatively quite similar to those in an overlap geometry junction , _ viz . _ , zero - field steps , fiske steps , and flux flow . quantitative differences are due to the difference in the effects of the bias current . whereas in an overlap junction each pendulum in the chain is subjected to a torque corresponding to @xmath61 , in an inline junction only the two end pendula are subjected to a torque corresponding to @xmath63 , according to eqs . ( 5 ) and ( 6 ) . in the absence of kinks , each pendulum in an overlap chain is offset in static equilibrium position by the same angle , @xmath54 , whereas the angular offset relaxes toward zero over a physical distance @xmath64 ( normalized distance @xmath65 ) proceeding from the ends toward the interior of an inline chain ; the corresponding analytic solution for this latter case was derived by owen and scalapino [ 23 ] in terms of jacobian elliptic functions . the most significant consequence of this difference for fluxon dynamics is that a fluxon in an overlap junction feels a constant accelerating force due to @xmath61 along the entire length of the junction , whereas a fluxon in an inline junction receives an accelerating ` kick ' due to @xmath63 only near the two junction ends . the situation for an annular geometry junction is quite different because there are no junction ends and because the winding number is fixed for a given dynamical configuration ; this means , _ e.g. _ , that fiske steps are not present , and that the distinction between zero - field steps and velocity - matching steps tends to be obscured . on the other hand , the possibility of studying fluxon dynamics in the absence of boundary reflection effects is often of considerable interest ; in this sense the annular geometry offers a ` cleaner ' environment than does either the overlap or the inline geometries . an interesting feature of fluxon dynamics in annular junctions is that for winding numbers different from zero and in the absence of external magnetic fields there is no zero - voltage current in the current - voltage characteristic of the junction . the reason for this is that as soon as a @xmath62 is applied , any fluxon that is locked into the junction begins to move , which implies that @xmath57 . this situation changes if a small dc magnetic field is applied in the plane of the junction : as mentioned in sect . 2.2 , the presence of an externally applied magnetic field is modelled by adding the term @xmath32 to the right hand side of eq . if @xmath33 is time - independent and spatially homogeneous , this term gives rise to a spatially sinusoidal potential function which can effectively pin fluxon motion [ 24,15 ] . very recently , ustinov _ [ 25 ] have devised an ingenious technique for adding fluxons in a controlled way into an annular junction ; this will undoubtedly stimulate further experimental study of this geometry . the only completely general technique ( aside from experimental measurement ) for studying soliton dynamics in long junctions is that of detailed simulation , _ i.e. _ , numerical integration of the appropriate pde model . it is frequently convenient , however , to be able to employ approximate techniques to obtain approximate , even though incomplete , information . soliton perturbation theory furnishes such a technique . since this topic was discussed quite thoroughly in the 1986 review by pedersen [ 9 ] i shall limit myself here simply to recalling a few salient points . the interested reader should consult mclaughlin and scott [ 26 ] for the mathematical basis of this theory , and the review by kivshar and malomed [ 27 ] for a detailed discussion of applications to long josephson junctions . the basic physical idea involved is that a kink is a highly robust , particle - like object that , even though acted upon by perturbations , tends to maintain its identity ; the mathematical foundation underlying this physical fact is that the pure sine - gordon equation is an exactly integrable system [ 6 ] . accordingly , it is possible to describe the dynamics of this object in terms of its energy , defined as @xmath66\,dx % \ ] ] and its momentum , defined as @xmath67 the approximation involved in using the soliton perturbation approach derives from the fact that , since we obviously do not know the unknown solution of the perturbed equation , we use instead the known solution of some ` nearby ' unperturbed equation in eqs . ( 8) and ( 9 ) . as a simple illustration of this procedure , let us consider the problem of describing the motion of a single fluxon in an overlap junction , described by eq . ( 2 ) , of infinite length . in this case , the time derivative of eq . ( 8) may be calculated as @xmath68\,dx\ , . % \ ] ] we use as a solution ansatz the function @xmath69/[1-u^{2}(t)]^{1/2})]\ , , % \ ] ] which , with @xmath70 and @xmath71 , is just the well - known single soliton solution of the pure sine - gordon equation on the infinite interval . for this unperturbed problem , the energy of the solution described by eq . ( 11 ) may be calculated from eq . ( 8) as @xmath72 by using the ansatz of eq.(11 ) , we are implicitly assuming that the essential effect of the perturbing terms those in @xmath12 , @xmath13 , and @xmath61on the dynamics of the system is to cause a slow modulation of the parameters @xmath73 and @xmath74 . this modulation is obtained by inserting eq . ( 11 ) into eq . ( 10 ) , and using eq . ( 12 ) , which gives the result @xmath75 and @xmath76 where @xmath77 is the position of the fluxon at time @xmath78 . in this way , we have reduced the problem from the integration of a pde to the integration of an ode , which provides a substantial computational simplification . the accuracy of results obtained using the perturbation theory approach depends on how ` close ' ( in some function - space sense ) the solution of the perturbed equation is to the solution of the unperturbed equation used to approximate it in the energy and momentum integrals . a practical example in which the perturbation approach breaks down due to significant discrepancies between the two solutions , caused by the effects of the @xmath13-loss term , was described in detail by davidson _ _ [ 28 ] . the only generally valid way to verify the accuracy of perturbation theory results is to compare them _ a posteriori _ with pde simulation results . in spite of this inherent weakness , the perturbation theory approach has repeatedly proved to be an extremely useful tool for the study of soliton dynamics . so far , we have considered only constant , time - independent bias currents and magnetic fields . in many cases of practical interest , however , the behavior of a long junction in the presence of a time - varying driver is of considerable interest . as a simple illustration of this class of problems , let us consider the inline junction described by eqs . ( 4)-(6 ) , subjected to a sinusoidally varying magnetic field , _ i.e. _ , with @xmath23 in eqs . ( 5 ) and ( 6 ) given by @xmath79 ; this might model a situation in which such a junction is exposed to a microwave field . the perturbation theory developed in the previous section offers a convenient tool for analyzing this problem . for simplicity we will treat only the case @xmath80 , which was solved by salerno _ [ 29 ] ; the general case , with @xmath81 , was solved by filatrella _ [ 30 ] . for an inline junction with @xmath80 the time derivative of eq . ( 9 ) is readily calculated to be @xmath82 the general solution of which is obtained by a trivial integration . moreover , for the unperturbed equation , we have that @xmath83 , with @xmath84 given by eq . ( 12 ) . from these facts the soliton trajectory may be found explicitly from eq . ( 14 ) as @xmath85\ , , % \ ] ] where @xmath86 and @xmath87 . the other essential ingredient of the analysis is the treatment of fluxon reflections at the boundaries . this problem was solved in the context of the perturbation theory by levring _ [ 31 ] , who showed that during a boundary reflection , due to eqs . ( 5 ) and ( 6 ) ( with @xmath80 ) , a fluxon undergoes an energy variation @xmath88 , given by @xmath89 where the plus sign is taken at one boundary and the minus sign at the other . this boundary effect can be incorporated into the description of the system dynamics very simply by rewriting eq . ( 15 ) as @xmath90\,\delta ( x(t)-kl)\ , , % \ ] ] in which @xmath91 is the dirac delta and @xmath92 is given by eq . ( 16 ) . in writing eq . ( 18 ) we are constructing a mathematical artifice consisting of a periodically extended junction structure lying along the positive @xmath1-axis , between @xmath93 and @xmath94 , in such a way that the back and forth shuttling motion of the fluxon in the physical junction is transformed into a unidirectional , left - to - right motion on the extended structure , with boundary reflection effects taking place at spatial points equal to integer multiples of @xmath95 . as shown in detail by salerno _ et al . _ [ 29 ] , the system description can be simplified still further , from the ode of eq . ( 18 ) to an explicit , two - dimensional functional map whose variables are @xmath96 , the time , modulo the period of the applied microwave field , of the @xmath38th boundary reflection , and @xmath97 , the fluxon energy at the @xmath38th reflection . in this way we reduce enormously the computational difficulty of the problem : from integrating the pde system of eqs . ( 4)-(6 ) , to integrating the ode of eq . ( 18 ) , to iterating a two - dimensional functional map . the new phenomenon that emerges in this case is that of phase locking : over a certain range of parameters the back and forth motion of the fluxon in the junction becomes locked in phase to the external microwave field ; this is most readily manifested as the appearance of a constant - voltage current step in the current - voltage characteristic of the junction , at a voltage @xmath98 , where @xmath99 is the angular frequency of the applied field , and @xmath31 and @xmath100 are integers , as shown in fig . reduction of the system dynamics to a functional map is particularly convenient for studying the phase locking phenomenon inasmuch as one can use standard , well known techniques to study the existence and stability of fixed points of the map , which corresponds directly to studying the existence and stability of phase - locked states of the fluxon dynamics . the salient features that emerge from such an analysis , applied to the system described by eq . ( 18 ) , are [ 29 ] : ( a ) phase - locked steps similar to the one shown in fig . 3 exist at the fundamental frequency @xmath101 and at all odd subharmonics @xmath102 . ( b ) the height in current of such a step is @xmath103 , _ i.e. _ , equal to the peak - to - peak amplitude of the ac magnetic field ; the step is centered around the unperturbed current - voltage characteristic . ( c ) the stability range in field amplitude for the steps decreases rapidly with increasing subharmonic order , _ i.e. _ , with increasing @xmath100 . ( d ) instability appears first at the center of a step and not , as one might expect , at the extremities . ( e ) for field amplitude values beyond the stability limit , a feigenbaum - like bifurcation cascade leading to chaos in the fluxon times of flight across the junction is observed . ( f ) for appropriate model parameters the phase locking steps can cross the zero - current axis , extending to negative current values ; such zero - crossing steps are normally associated with a hysteresis in the step amplitude . to check the validity of the particle - map formalism used here , it is useful to compare results with those obtained from a full simulation of the original pde model , eqs . ( 4)-(6 ) , which has a fairly long and well tested history of accounting for experimental results . such a comparison shows that the strongest limitation of the particle - map approach stems from the assumption of single fluxon dynamics . physically , a sufficiently large energy exchange term , eq . ( 17 ) , can give rise to the creation of additional fluxons in the junction . such effects are clearly manifested in pde simulations [ 32 ] , but they are not contained in the particle - map model ; their principal consequence is that the linear growth of the step height in current with @xmath104 holds only up to a certain limit , beyond which the step height saturates , or even decreases , with increasing field amplitude . quantitative discrepancies between map and pde results can often be attributed to the fact that a fluxon in the particle - map formalism is a point particle with no spatial extension , whereas a pde fluxon has a non - zero width . consequently , whereas energy exchange at the junction boundaries occurs instantaneously in the map , the exchange interaction is smeared out over a certain time interval in the pde . this difference tends to disappear for junction lengths @xmath16 , but it may be significant for length values of practical interest , especially when subharmonic locking is considered ; this problem has been considered in detail , in the context of a perturbation - theory approach , by grnbech - jensen [ 33 ] . nonetheless , within its range of validity , the simple map approach predicts results to a truly surprising level of detail and accuracy . as mentioned briefly in the previous section , solitons in long junctions can also undergo chaotic motion . this is a topic that has attracted a large amount of attention in recent years , and i refer the reader to the literature ( see , _ e.g. _ , guerrero and octavio [ 34 ] , plus references therein ) for detailed discussions ; i present here only a few particular points . chaotic effects are observed especially when a junction is subjected to time - varying currents and/or magnetic fields , but the presence of such an ac driver is not strictly necessary to bring about chaos : soerensen _ et al . _ [ 35 ] described computer simulations in which a junction that is dc - biased in the fiske - step regime could , with appropriate parameter values , exhibit chaotically intermittent switching between the dynamic states corresponding to the first and the second fiske steps ; further confirmation and analysis of this phenomenon were subsequently reported by yeh _ [ 36 ] and by filatrella _ [ 37 ] . a case in which chaos is brought about by the presence of an ac driver is shown in fig . this figure shows how the fluxon time of flight , defined as @xmath105 , changes with the amplitude of the applied magnetic field , for an inline junction similar to that discussed in the previous section , using the particle - map formalism [ 29 ] ; that the phenomenon is real and not just an artifact of the map formalism was demonstrated analytically by malomed [ 38 ] and and via pde simulation by rotoli and filatrella [ 39 ] . the field value at which the first bifurcation in fig . 4 occurs is just the one for which a stability analysis indicates that the phase - locked state loses stability [ 29 ] ; the values for the subsequent bifurcations follow quite closely those of a feigenbaum cascade [ 40 ] . an interesting fact that emerges from the map calculations is that throughout the bifurcation region , and even for some distance into the chaotic region , the average junction voltage remains exactly that of the phase - locked state , below the bifurcation tree , even though the fluxon is certainly no longer locked to the driver ; this fact may be established by calculating ( numerically ) the average time of flight in these regions . this means that , in these regions , the complication in the underlying dynamics leaves no signature in the current - voltage characteristic of the junction . consequently , the only way to reveal such phenomena experimentally would be through a spectral analysis of the radiation emitted by a junction in a state involving fluxon propagation ; this poses an interesting challenge to experimentalists . for most practical electronic device applications the presence of chaotic phenomena is highly undesirable inasmuch as these constitute simply another source of electrical noise , thereby degrading some desired performance characteristic . for this reason , researchers are pressed not only to identify parameter regions where chaos might exist , in order to avoid them , but in fact to devise stratagems to suppress chaos . a step in this direction was recently reported by salerno [ 41 ] , who showed , using the particle - map formalism , that the type of chaos described in the preceding paragraph can indeed be suppressed by the addition of a small subharmonic component to the ac magnetic field driving term . [ 42 ] have provided further credibility for this result via pde simulation ; in particular , they showed that , in a situation very similar to that indicated in fig . 4 , the addition of an @xmath106 subharmonic component , _ i.e. _ , a component whose frequency would correspond to the fundamental for a step at that voltage , approximately one order of magnitude smaller in amplitude than the ( true ) fundamental , can be sufficient to suppress chaos of the type indicated in fig . soliton propagation phenomena in long josephson junctions , both resonant , back and forth propagation and unidirectional flux flow , give rise naturally to practical applications as electronic oscillators . the reason for this is that a soliton is amonst other things an electromagnetic pulse , which , when it impinges on the free end of a junction , gives rise to the emission of a pulse of electromagnetic energy . if the arrival of solitons at the junction end is time - periodic , then so is the radiation emission . for typical experimental junctions made with currently available fabrication technologies , _ i.e. _ , based on low critical temperature materials , this radiation lies somewhere within the microwave to millimeter - wave region of the electromagnetic spectrum . the power emitted is , in absolute terms , rather low : in this context , one microwatt is a ` large ' power ; on the other hand , existing devices such as sis mixers , that might conceivably be employed in integrated , all - josephson millimeter - wave receivers , require not more than this amount of power . comparison of the relative merits of oscillators based on the two propagation modes , _ viz . _ , resonant back and forth propagation and unidirectional flux flow , shows that they have complementary strengths and weaknesses . single - junction resonant propagation oscillators typically show very narrow output linewidths about @xmath107 khz at oscillation frequencies of @xmath108 ghz have been measured [ 43]but observed output powers into an external load have ranged from picowatts up to something less than about a nanowatt really too small to be practically useful ( however , cirillo _ et al . _ [ 44 ] have estimated a power of @xmath109 nw at @xmath110 ghz into a tightly coupled small junction used as a detector ) . in contrast , for single - junction flux flow oscillators , powers of about a microwatt ( into a small - junction detector ) at frequencies up to @xmath111 ghz were obtained quite early [ 45 ] , and recently much larger values for both power and frequency have been measured [ 46 ] ; however , output spectral linewidths for this configuration are considerably larger than for the resonant propagation configuration . the reason for this relative performance difference between the two configurations is readily understood in an intuitive way : a resonant propagation oscillator is normally operated in zero , or in any case small , external magnetic field . its oscillation frequency is therefore determined essentially by the junction length and , to a lesser extent , by the dc bias current . this suggests a rather sharply defined frequency , but , since radiation is emitted only during a fluxon reflection from the output end ( after a back and forth traversal of the junction ) , it seems plausible that the output power and fundamental frequency might be relatively low ; furthermore , for this same reason , the output voltage waveform of a resonant propagation oscillator consists of a train of fairly sharp spikes , which suggests that output power is delivered into a broad range of harmonics rather than being concentrated at the fundamental frequency . in contrast , the oscillation frequency of a flux flow oscillator depends strongly on the value of the external magnetic field [ 21 ] , which means that there exists another source of fluctuations that might broaden the output spectral linewidth . however , since fluxons impinge on the output end in a tightly packed , unidirectional train , as shown in fig . 2 , it seems plausible that the output power and fundamental frequency might be relatively high ; furthermore , as is also apparent from fig . 2 , the output voltage waveform is quite nearly sinusoidal , which suggests that output power is largely concentrated at the fundamental frequency . in an effort to boost output power levels and to reduce oscillator linewidths , several groups have begun to explore the idea of employing phase - locked arrays of long - junction oscillators , following the lead of corresponding work done in the context of small - junction oscillators , in large measure by the stony brook group [ 47 ] . so far , more attention has been dedicated to arrays of resonant propagation oscillators [ 48 - 51 ] , in an attempt to raise output powers to practically useful levels , but quite recently the problem of phase locking of a flux flow oscillator has also been addressed [ 52 ] . in particular , in the work reported by monaco _ et al . _ [ 50 ] , phase locking of a controlled number , from one to ten , of series - biased resonant propagation oscillators was obtained at @xmath108 ghz by capacitively coupling the oscillator array to a high - q linear distributed resonator . the output power up to @xmath112 nw in an external , room temperature , @xmath113-ohm load was found to vary quadratically with the number of locked junctions , as shown in fig . 5 , in accordance with theoretical expectations . monaco _ et al . _ [ 50 ] also showed that the behavior of their device can be described , at least to a first level of approximation , by a simple model in which the physical , distributed resonator is replaced by a lumped , linear tank circuit . in this model each junction is coupled to the tank through a capacitance which also comprises a part of the resonator , and the dynamics of the oscillating fluxon in each junction is described by the particle - map approach appropriate to overlap geometry junctions , as outlined above in sect . 3 . to calculate the effect of the junctions on the tank one may assume , consistently with the assumed particle nature of the fluxons , that each fluxon reflecting at the tank boundary can be described as a delta function voltage pulse of weight @xmath55 . in this way the tank voltage may be calculated analytically as a sum of impulse responses . the effect of the tank on each junction is to modify the junction boundary condition at the tank boundary ; specifically , the current fed into each junction by the tank can be calculated knowing the tank voltage and the junction coupling capacitance . the resulting model is formally equivalent to an @xmath114-dimensional functional map , in which @xmath31 is the number of locked junctions . although not readily tractable analytically , the system dynamics can easily be solved by iterating the map numerically . one salient feature that emerges from such a solution is that the tank voltage amplitude varies linearly with the number of locked junctions , @xmath31 ; this implies that the output power from the array varies as @xmath115 , in agreement with the experimental result shown in fig . although apparently quite promising as an approach to obtain a narrow linewidth oscillator capable of furnishing a reasonable amount of power , the use of arrays of resonant propagation oscillators is subject to some intrinsic limitations regarding the number of junctions that can be locked together [ 53 ] ; moreover , the practical problems involved in biasing such an array in a simple and controllable manner might well limit the feasibility of this approach . nonetheless , more work is needed in this direction in order to fully clarify the situation . several attempts to design and construct integrated receivers consisting of sis mixers and long - junction oscillators have been described in the literature . the first of these , to my knowledge , was reported by crete _ [ 54 ] in 1990 . this project consisted of a bow - tie antenna , an sis mixer , and a @xmath112-junction , resonant propagation oscillator array , designed to operate at @xmath116 ghz , a region of considerable interest for radio astronomy applications ; no experimental measurements on the complete receiver were reported , however . zhang _ et al . _ [ 55 ] presented results on a receiver consisting of an sis mixer and a long - junction oscillator that could be operated in either the resonant propagation mode or the flux flow mode , depending on the oscillator junction current density a fabrication parameter subject to direct experimental control . estimated output powers of @xmath117 nw at @xmath118 ghz with the oscillator operating in the resonant propagation mode and @xmath119 nw at @xmath120 ghz with the oscillator operating in the flux flow mode were found for this device . in a successive work zhang _ [ 56 ] reported measurements on a very similar receiver in which two flux flow oscillators were coupled to the same sis mixer one oscillator being coupled fairly strongly and the other fairly weakly so as to measure the output linewidths of the oscillators by mixing them down to a lower intermediate frequency . in this way a composite linewidth of @xmath121 mhz in the band @xmath122 ghz was measured . moreover , an available power incident on the mixer of @xmath123 nw at @xmath124 ghz was estimated for this device , which is probably more than adequate for optimal mixer performance . also at 1992 asc , koshelets _ [ 57 ] presented results on two receiver designs , both based on niobium trilayer technology . the first of these consisted of a flux flow oscillator coupled to a single small - junction sis mixer . for this design , an output power of more than @xmath125 nw at @xmath126 ghz was estimated . the oscillator output linewidth was measured at @xmath127 and @xmath128 ghz by mixing the flux flow oscillator signal with the signal from a backward wave oscillator and analyzing the resulting intermediate frequency spectrum . a value of @xmath65 mhz was found , which , however , was believed to be an upper limit due to the measurement instrumentation . the second receiver consisted of a flux flow oscillator coupled to a multi - junction sis mixer array . for this design , a minimum receiver noise temperature of @xmath129 k at @xmath130 ghz was measured , with an oscillator output power of @xmath131 nw . these results , together with those presented by andy smith elsewhere in this volume [ 58 ] , clearly demonstrate that superconducting technology for the construction of low - noise , integrated millimeter - wave receivers is rapidly approaching maturity . as mentioned briefly above in sect . 3.1 , that the flux flow mode of propagation offers potential applications for amplification was recognized quite early [ 22 ] . in the intervening years , much of the pioneering work in this area has been spearheaded by the group working at the university of wisconsin in madison though it is certainly true that work by the japanese groups has proceeded continuously . in recent times interest in this application has enjoyed a widespread diffusion , so that today new results are developing rapidly . the basic idea underlying the operation of a flux flow amplifier is quite readily grasped from the general description of the flux flow phenomenon presented in sect . 3.1 : in the flux flow mode the number of fluxons injected into the junction per unit time depends essentially on the strength of the externally applied magnetic field . since the average junction voltage is proportional to the product of the number of fluxons in the junction and their average propagation velocity , an increase of magnetic field , at constant bias current , causes an increase of the junction voltage . with an appropriate choice of parameters in particular , with the junction biased on a velocity matching step , and with an appropriate load line this mechanism can give rise to amplification . in one of the earliest configurations studied [ 59 ] , called the ` vortex flow transistor ' ( vft ) , the driving magnetic field was furnished by a superconducting control line lying either above or below the junction and electrically insulated from it . in this configuration a time - varying current in the control line causes a time - varying voltage across the junction , so that the basic gain mechanism can be characterized as a transresistance . properly designed , this device can give significant values of power gain [ 60 ] . an alternative procedure for introducing an input signal into such a device is to modulate a current injected directly into the junction counterelectrode [ 61 - 63 ] ; this configuration was given the name ` superconducting current injection transistor ' ( supercit ) . the basic idea here is that , as suggested by eqs . ( 5 ) and ( 6 ) , such an injected current plays a rle in the junction dynamics that is very similar to that played by a magnetic field . in contrast with the vft , which is normally based on a junction having a low - loss , hysteretic current - voltage characteristic , the supercit is normally based on a junction having a high - loss , non - hysteretic current - voltage characteristic . for this configuration , the device transresistance can be enhanced by constructing the junction with a thin counterelectrode and exploiting the resulting kinetic inductance [ 64 ] . however , the supercit has the disadvantage with respect to the vft of being intrinsically a three - terminal , rather than a four - terminal , device , a fact which leads to obvious grounding problems for the eventual construction of arrays of such devices . the performance features of both flux flow oscillators and flux flow amplifiers can be enhanced if sharp , _ i.e. _ , low dynamic - resistance velocity - matching steps can be achieved . a number of attempts to realize such low resistance steps have been based on appropriate tailoring of the junction geometry , in particular , by the addition of ` ears ' or ` wings ' to the junction structure [ 21 ] . an alternative approach , devised by nevirkovets _ et al . _ [ 65 ] , utilizes a dc edge injection current proportional to the junction bias current , with relative polarities chosen so that the effect of the injection current is to oppose that of the applied magnetic field as regards the injection of fluxons into the junction . the result is a sort of negative feedback which tends to stabilize the number of fluxons in the junction , and hence the average junction voltage . analytical results [ 65 ] and numerical simulations [ 65,66 ] suggest that this might prove to be a simple and effective technique for improving device performance . a related approach to the realization of flux flow amplification has been the so - called ` superconducting flux flow transistor ' ( sfft ) [ 67 - 69 ] . this consists of an array of weak links not tunnel junctions patterned into a single superconducting film , with device input provided by the local magnetic field generated by a separate control line located near one end of the array . although the dynamic behavior of weak link arrays has been modelled in terms of a discretized version of eq . ( 2 ) [ 70 ] , it is not clear that this is entirely correct inasmuch as magnetic flux is carried through such a discrete array by abrikosov vortices rather than by josephson vortices ( see , _ e.g. _ , likharev [ 71 ] for a detailed discussion regarding the similarities and differences between these two types of magnetic flux structures ) . the original motivation for studying the sfft was probably the desire to create novel active electronic applications for the new high - temperature superconductors , since good tunnel junction characteristics in high-@xmath132 materials are as yet difficult to realize ( but see the paper by alex braginski [ 72 ] for recent progress on this topic ) ; lately , however , the sfft has begun to assume an increasing importance in its own right inasmuch as there are indications that it may display some performance characteristics that are superior to its long - junction counterparts . an indicative example is the sfft amplifier described by martens _ [ 73 ] at 1992 asc : this device showed a gain of @xmath133 db over a bandwidth of @xmath113 ghz . the central theme of the nato asi on which this present volume is based was that the time now seems ripe for the field of superconducting electronics , which has long played a cinderella rle , to make its debut at the ball . my task has been to present the notion that devices based on soliton propagation effects , which have been considered a particularly esoteric aspect of a generally esoteric field , deserve serious attention in this context . i have tried to show that the basic ideas involved are , in fact , quite readily accessible , and that they do offer a rich potential for practical applications . here , i have focused on just two fairly simple analog applications oscillators and amplifiers leaving the exciting area of digital applications to the able exposition of kostya likharev [ 74 ] . if i succeed in communicating to the reader even a part of the enthusiasm that was prevalent during the asi , then my efforts will have been amply repaid . the work described herein is largely the fruit of a long - standing and felicitous collaboration between the josephson junction groups working at the university of salerno , the cybernetics institute of the italian cnr , and the technical university of denmark ; i wish to express my gratitude to my many friends and colleagues working in these groups . special thanks go to valery koshelets , jim nordman , and yongming zhang for sending me pre - publication material regarding their work on flux flow devices , and to jim nordman , boris malomed , niels pedersen , and alexey ustinov for illuminating comments and a critical reading of the manuscript . financial support from murst and from the progetto finalizzato `` tecnologie superconduttive e criogeniche '' del cnr ( italy ) is gratefully acknowledged . 99 [ 1 ] n. j. zabusky and m. d. kruskal , phys . ( 1965 ) 240 . [ 2 ] b. d. josephson , phys . ( 1962 ) 251 . [ 3 ] i. o. kulik , zh . ( 1966 ) 1952 ; sov . jetp ( 1967 ) 1307 . [ 4 ] a. c. scott , am . j. phys . ( 1969 ) 52 . [ 5 ] t. a. fulton and r. c. dynes , solid state comm . ( 1973 ) 57 . [ 6 ] a. c. scott , f. y. f. chu , and d.w . mclaughlin , proc . ieee ( 1973 ) 1443 . [ 7 ] t. a. fulton , r. c. dynes , and p. w. anderson , proc . ieee ( 1973 ) 28 . [ 8 ] r. d. parmentier , in , k. lonngren and a. scott , eds . ( academic press , new york , 1978 ) , pp . 173 - 199 . [ 9 ] n. f. pedersen , in , s. e. trullinger , v. e. zakharov , and v. l. pokrovsky , eds . ( elsevier , amsterdam , 1986 ) , pp . 469 - 501 . [ 10 ] y. bruynseraede , c. van haesendonck , and v. v. moshchalkov , _ this volume_. [ 11 ] a. barone and g. patern , ( wiley , new york , 1982 ) . [ 12 ] j. c. eilbeck , p. s. lomdahl , o. h. olsen , and m. r. samuelsen , j. appl . ( 1985 ) 861 . [ 13 ] m. scheuermann and c. c. chi , in , h. d. hahlbohm and h. lbbig , eds . ( de gruyter , berlin , 1985 ) , pp . 477 - 481 . [ 14 ] a. c. scott , solid - state electronics ( 1964 ) 137 . [ 15 ] n. grnbech - jensen , phys b ( 1992 ) 7315 . [ 16 ] t. a. fulton , in , b. b. schwartz and s. foner , eds . ( plenum , new york , 1977 ) , pp . 125 - 187 . [ 17 ] m. cirillo , r. d. parmentier , and b. savo , physica d ( 1981 ) 565 . [ 18 ] g. costabile , r. d. parmentier , b. savo , d. w. mclaughlin , and a. c. scott , appl . ( 1978 ) 587 . [ 19 ] g. costabile , s. pagano , and r. d. parmentier , phys . b ( 1987 ) 5225 . [ 20 ] s. n. ern , a. ferrigno , and r. d. parmentier , phys . b ( 1983 ) 5440 . [ 21 ] y. zhang , `` theoretical and experimental studies of the flux - flow type josephson oscillator '' , licentiate thesis , dept . physics , chalmers univ . tech . , gteborg , sweden , 1991 ( unpublished ) . [ 22 ] k. yoshida and f. irie , in , vol . 4 , m. krusius and m. vuorio , eds . ( north - holland , amsterdam , 1975 ) , pp . 176 - 179 . [ 23 ] c. s. owen and d. j. scalapino , phys ( 1967 ) 538 . [ 24 ] n. grnbech - jensen , p. s. lomdahl , and m. r. samuelsen , phys . b ( 1991 ) 12799 . [ 25 ] a. v. ustinov , t. doderer , r. p. huebener , n. f. pedersen , b. mayer , and v. a. oboznov , phys . ( 1992 ) 1815 . [ 26 ] d. w. mclaughlin and a. c. scott , phys . rev . a ( 1978 ) 1652 . [ 27 ] yu . s. kivshar and b. a. malomed , rev . ( 1989 ) 763 . [ 28 ] a. davidson , n. f. pedersen , and s. pagano , appl . ( 1986 ) 1306 ; see also : s. pagano , licentiate thesis , dcamm report s-42 , tech . univ . denmark , lyngby , denmark , 1987 ( unpublished ) . [ 29 ] m. salerno , m. r. samuelsen , g. filatrella , s. pagano , and r. d. parmentier , phys . b ( 1990 ) 6641 . [ 30 ] g. filatrella , g. rotoli , and r. d. parmentier , phys . a ( 1990 ) 122 . [ 31 ] o. a. levring , n. f. pedersen , and m. r. samuelsen , j. appl . ( 1983 ) 987 . [ 32 ] n. f. pedersen and a. davidson , phys . b ( 1990 ) 178 . [ 33 ] n. grnbech - jensen , `` coupling of microwaves to fluxon motion in josephson junctions '' , phys . b , _ in press_. [ 34 ] l. e. guerrero and m. octavio , phys . rev . a ( 1989 ) 3371 . [ 35 ] m. p. soerensen , n. arley , p. l. christiansen , r. d. parmentier , and o. skovgaard , phys . rev . lett . ( 1983 ) 1919 . [ 36 ] w. j. yeh , o. g. symko , and d. j. zheng , phys . b ( 1990 ) 4080 . [ 37 ] g. filatrella , r. d. parmentier , s. pagano , p. l. christiansen , m. p. srensen , and n. grnbech - jensen , `` on the switching between soliton dynamic states in long josephson junctions '' , phys . a , _ in press_. [ 38 ] b. a. malomed , phys . b ( 1990 ) 2037 . [ 39 ] g. rotoli and g. filatrella , phys . a ( 1990 ) 211 ; g. rotoli and g. filatrella , in , m. remoissenet and m. peyrard , eds . ( springer - verlag , berlin , 1991 ) , pp . 284 - 291 . [ 40 ] see , _ e.g. _ : e. ott , rev . ( 1981 ) 655 . [ 41 ] m. salerno , phys . b ( 1991 ) 2720 . [ 42 ] g. filatrella , g. rotoli , and m. salerno , `` chaos suppression in fluxon dynamics of long josephson junctions '' , _ preprint , 1992_. [ 43 ] e. joergensen , v. p. koshelets , r. monaco , j. mygind , m. r. samuelsen , and m. salerno , phys . ( 1982 ) 1093 . [ 44 ] m. cirillo , i. modena , p. carelli , g. s. spagnolo , r. leoni , and m. pullano , ieee trans ( 1991 ) 3335 . [ 45 ] t. nagatsuma , k. enpuku , f. irie , and k. yoshida , j. appl . ( 1983 ) 3302 . [ 46 ] s. kiryu , s. kohjiro , a. shoji , s. kodaira , j. itanani , and s. sato , 1992 applied superconductivity conference , chicago , u.s.a . , august 1992 , paper eld-9 . [ 47 ] j. e. lukens , in , s. t. ruggiero and d. a. rudman , eds . ( academic press , new york , 1990 ) , pp . [ 48 ] m. cirillo and f. l. lloyd , j. appl . ( 1987 ) 2581 . [ 49 ] s. pagano , r. monaco , and g. costabile , ieee trans . ( 1989 ) 1080 . [ 50 ] r. monaco , n. grnbech - jensen , and r. d. parmentier , phys . a ( 1990 ) 195 . [ 51 ] p. barbara , a. davidson , g. filatrella , j. holm , j. mygind , and n. f. pedersen , phys . a ( 1992 ) 241 . [ 52 ] a. v. ustinov , j . mygind , and v. a. oboznov , j. appl . ( 1992 ) 1203 . [ 53 ] g. filatrella , g. rotoli , n. grnbech - jensen , r. d. parmentier , and n. f. pedersen , j. appl ( 1992 ) 3179 . [ 54 ] d .- g . crete , p. feautrier , m. hanus , r. monaco , and p. j. encrenaz , in , m. russo and g. costabile , eds . ( world scientific , singapore , 1990 ) , pp . [ 55 ] y. m. zhang , d. winkler , and t. claeson , 1992 applied superconductivity conference , chicago , u.s.a . , august 1992 , paper eld-7 . [ 56 ] y. m. zhang , d. winkler , and t. claeson , `` linewidth measurements of josephson flux - flow oscillators in the band 280 - 330 ghz '' , _ preprint , 1992_. [ 57 ] v. p. koshelets , a. v. shchukin , s. v. shitov , and l. v. filippenko , 1992 applied superconductivity conference , chicago , u.s.a . , august 1992 , paper eld-8 . [ 58 ] a. d. smith , _ this volume_. [ 59 ] t. v. rajeevakumar , appl . ( 1981 ) 439 . [ 60 ] d. p. mcginnis , j. b. beyer , and j. e. nordman , j. appl . ( 1986 ) 3917 . [ 61 ] k. k. likharev , v. k. semenov , o. v. snigirev , and b. n. todorov , ieee trans . ( 1979 ) 420 . [ 62 ] b. j. van zeghbroeck , appl . ( 1983 ) 736 . [ 63 ] d. p. mcginnis , j. b. beyer , and j. e. nordman , ieee trans . magn . ( 1989 ) 1262 . [ 64 ] t. hashimoto , k. enpuku , k. yoshida , and k. yamafuji , japan . j. appl . ( 1987 ) l1874 . [ 65 ] i. p. nevirkovets , e. m. rudenko , and a. v. ustinov , fiz . nizk . temp . ( 1988 ) 1134 . [ 66 ] a. petraglia , `` non linear dynamics in long josephson junctions '' , graduation thesis , dept . physics , univ . salerno , salerno , italy , 1992 ( unpublished ) . [ 67 ] k. k. likharev , sov . jetp ( 1972 ) 906 . [ 68 ] j. s. martens , d. s. ginley , j. b. beyer , j. e. nordman , and g. k. g. hohenwarter , ieee trans . ( 1991 ) 95 . [ 69 ] j. s. martens , v. m. hietala , t. e. zipperian , d. s. ginley , and c. p. tigges , in , h. koch and h. lbbig , eds . ( springer - verlag , berlin , 1992 ) , pp . 186 - 191 . [ 70 ] s. hontsu and j. ishii , j. appl . ( 1988 ) 2021 . [ 71 ] k. k. likharev , rev . ( 1979 ) 101 . [ 72 ] a. i. braginski , _ this volume_. [ 73 ] j. s. martens , v. m. hietala , t. a. plut , d. s. ginley , g. a. vawter , c. p. tigges , m. p. siegal , j. m. phillips , and s. y. hou , 1992 applied superconductivity conference , chicago , u.s.a . , august 1992 , paper ej-2 . [ 74 ] k. k. likharev , _ this volume_.

let be a finite set , be a reticulation - visible network on , and be a rooted binary phylogenetic tree . we show that there is a polynomial - time algorithm for deciding whether or not displays . furthermore , for all , we show that has at most vertices in total and at most reticulation vertices , and that these upper bounds are sharp .

1508.05424

phylogenetic networks have become increasingly more prominent in the literature as they correctly allow the evolution of certain collections of present - day species to be described with reticulation ( non - tree - like ) events . however , the evolution of a particular gene can generally be described without reticulation events . as a result , analysing the tree - like information in a phylogenetic network has become a common task . central to this task is that of deciding if a given phylogenetic network @xmath1 infers a given rooted binary phylogenetic tree on the same collection of taxa . in this paper , we show that if @xmath1 is a so - called reticulation - visible network , then there is a polynomial - time algorithm for making this decision . this resolves a problem left open in @xcite and @xcite . in the rest of the introduction , we formally state this result as well as the other main result which concerns the number of vertices in a reticulation - visible network . throughout the paper , @xmath0 denotes a non - empty finite set . for @xmath6 , a _ phylogenetic network on @xmath0 _ is a rooted acyclic digraph with no parallel arcs and the following properties : a. the root has out - degree two ; b. vertices of out - degree zero have in - degree one , and the set of vertices with out - degree zero is @xmath0 ; and c. all other vertices either have in - degree one and out - degree two , or in - degree two and out - degree one . if @xmath7 , then @xmath1 consists of the single vertex in @xmath0 . the vertices in @xmath1 of out - degree zero are called _ leaves_. furthermore , the vertices of @xmath1 with in - degree two and out - degree one are called _ reticulations _ , while the vertices of in - degree one and out - degree two are called _ tree vertices_. the arcs directed into a reticulation are _ reticulation arcs _ ; all other arcs are called _ tree arcs_. note that , what we have called a phylogenetic network is sometimes referred to as a _ binary _ phylogenetic network . a _ rooted binary phylogenetic @xmath0-tree _ is a phylogenetic network on @xmath0 with no reticulations . let @xmath1 be a phylogenetic network on @xmath0 and let @xmath2 be a rooted binary phylogenetic @xmath0-tree . we say that @xmath1 _ displays _ @xmath2 if @xmath2 can be obtained from @xmath1 by deleting arcs and vertices , and contracting degree - two vertices . to illustrate , in fig . [ display ] , the phylogenetic network @xmath1 on @xmath8 displays the rooted binary phylogenetic @xmath0-tree @xmath2 . the particular problem of interest is the following : tree containment + _ instance : _ a phylogenetic network @xmath1 on @xmath0 and a rooted binary phylogenetic @xmath0-tree @xmath2 . + _ question : _ does @xmath1 display @xmath2 ? in general , tree containment is np - complete @xcite even when the instance is highly constrained @xcite . let @xmath1 be a phylogenetic network on @xmath0 with root @xmath9 . a vertex @xmath10 in @xmath1 is _ visible _ if there is a leaf @xmath11 in @xmath0 with the property that every directed path from the root to @xmath11 traverses @xmath10 , in which case , we say _ @xmath11 verifies the visibility of @xmath10 _ ( or , more briefly , _ @xmath11 verifies @xmath10 _ ) . if every reticulation in @xmath1 is visible , then @xmath1 is a _ reticulation - visible network_. in fig . [ display ] , @xmath1 is a reticulation - visible network . for example , the visibility of the leftmost reticulation is verified by @xmath12 . observe that this reticulation is not verified by @xmath13 as there is a path from the root of @xmath1 to @xmath13 that avoids it . for the reader familiar with tree - child networks , @xmath1 is tree - child network if and only if every vertex in @xmath1 is visible ( * ? ? ? * lemma 2 ) . thus tree - child networks are a proper subclass of reticulation - visible networks . it was shown in @xcite that there exists a polynomial - time algorithm for tree containment if @xmath1 is a tree - child network . here we generalise that result to reticulation - visible networks . the next theorem is the first main result of the paper . let @xmath1 be a phylogenetic network on @xmath0 and let @xmath2 be a rooted binary phylogenetic @xmath0-tree . then tree containment for @xmath1 and @xmath2 can be decided in polynomial time . [ decision ] it is shown in @xcite that a reticulation - visible network on @xmath0 has at most @xmath14 reticulations . the second of the two main results sharpens this result . let @xmath1 be a reticulation - visible network @xmath1 on @xmath0 and let @xmath15 . then @xmath1 has at most @xmath16 vertices in total and at most @xmath17 reticulations . moreover , these bounds are sharp for all integers @xmath18 . [ sharp ] the paper is organised as follows . the next section consists of concepts that will be used in the description of the algorithm that establishes theorem [ decision ] . called treedetection , the description of this algorithm and the proof of its correctness is given in section [ algorithm ] . in section [ running ] , we analyse the algorithm s running time and show that treedetection completes in @xmath19 steps . together sections [ algorithm ] and [ running ] constitute the proof of theorem [ decision ] . the last section , section [ visible ] , contains the proof of theorem [ sharp ] . throughout the paper , notation and terminology follows semple and steel @xcite . let @xmath1 be a phylogenetic network , and let @xmath20 and @xmath10 be vertices in @xmath1 . if @xmath20 and @xmath10 are joined by an arc @xmath21 , we say @xmath20 is a _ parent _ of @xmath10 and , conversely , @xmath10 is a _ child _ of @xmath20 . more generally , if @xmath20 and @xmath10 are joined by a directed path from @xmath20 to @xmath10 , we say @xmath20 is an _ ancestor _ of @xmath10 and , conversely , @xmath10 is a _ descendant _ of @xmath20 . furthermore , if @xmath20 is neither an ancestor nor descendant of @xmath10 , we say @xmath20 and @xmath10 are _ non - comparable_. a _ backward path _ in @xmath1 from @xmath10 to @xmath20 is an underlying path @xmath22 such that , for all @xmath23 , we have that @xmath1 contains the arc @xmath24 . observe that if there is a backward path in @xmath1 from @xmath10 to @xmath20 , then @xmath20 is an ancestor of @xmath10 . an _ up - down path _ @xmath25 in @xmath1 from @xmath10 to @xmath20 is an underlying path @xmath26 such that , for some @xmath27 , we have that @xmath28 is a backward path from @xmath10 to @xmath29 and @xmath30 is a ( directed ) path from @xmath29 to @xmath10 . the vertex @xmath29 is the _ peak _ of @xmath25 . also , a _ tree path _ in @xmath1 from @xmath20 to @xmath10 is a ( directed ) path such that , except possibly @xmath20 , every vertex on the path is either a tree vertex or a leaf . lastly , the _ length _ of a ( directed ) path , a backward path , an up - down path , and a tree path is the number of edges in the path . let @xmath1 be a phylogenetic network on @xmath0 . a @xmath31-element subset @xmath32 of @xmath0 is a _ cherry _ in @xmath1 if there is an up - down path of length two between @xmath33 and @xmath34 . equivalently , @xmath32 is a cherry if the parent of @xmath33 and the parent of @xmath34 are the same . for a cherry @xmath32 in @xmath1 , let @xmath35 be obtained from @xmath1 by deleting @xmath33 and @xmath34 , and their incident arcs , and labelling their common parent ( now itself a leaf ) with an element not in @xmath0 . we say that @xmath35 has been obtained from @xmath1 by _ reducing the cherry @xmath32_. let @xmath1 be a phylogenetic network on @xmath0 and let @xmath2 be a rooted binary phylogenetic @xmath0-tree . suppose that @xmath1 displays @xmath2 . then there is a subgraph @xmath36 of @xmath1 that is a subdivision of @xmath2 . we say @xmath36 is an _ embedding _ of @xmath2 in @xmath1 . observe that any embedding of @xmath2 in @xmath1 can be formed by deleting exactly one incoming arc at each reticulation and deleting any resulting degree - one vertex that is not a leaf of @xmath1 . we refer to the action of deleting one of the two incoming arcs at a reticulation as _ resolving the reticulation_. in this section , we present the algorithm treedetection . this algorithm takes as input a reticulation - visible network @xmath1 on @xmath0 and a rooted binary phylogenetic @xmath0-tree @xmath2 and , as we establish in this section , outputs * yes * if @xmath1 displays @xmath2 and * no * if @xmath1 does not display @xmath2 . we begin with some further preliminaries . let @xmath1 be a phylogenetic network on @xmath0 with root @xmath9 . let @xmath37 and @xmath38 be distinct elements in @xmath0 . we define the vertex @xmath39 of @xmath1 to be the reticulation at minimum path length from @xmath9 such that @xmath37 verifies @xmath39 and no other element of @xmath0 verifies @xmath39 . if there is no such reticulation , we define @xmath39 to be @xmath37 . furthermore , we define @xmath40 to be the vertex at maximum path length from @xmath9 such that both @xmath37 and @xmath38 verify @xmath40 . note that @xmath39 and @xmath40 are both well defined since if @xmath41 and each element of @xmath42 verifies vertices @xmath20 and @xmath10 of @xmath1 , then it must be that either @xmath20 is an ancestor of @xmath10 or @xmath10 is an ancestor of @xmath20 . briefly , treedetection proceeds by picking a cherry @xmath43 of the targeted rooted binary phylogenetic tree @xmath2 and then considering how the leaves @xmath37 and @xmath38 are related in @xmath1 . there are various cases to consider , but in most cases we can either declare * no * directly , or we find an arc of @xmath1 that can be deleted so that the resulting phylogenetic network displays @xmath2 if and only if @xmath1 displays @xmath2 . in the remaining cases , we can delete the leaf @xmath38 from both @xmath1 and @xmath2 so that the resulting phylogenetic network displays the resulting rooted binary phylogenetic tree if and only if @xmath1 displays @xmath2 . in any case , if we can not immediately declare * no * , we reduce the size of @xmath1 by either a vertex or an arc and , by iterating the procedure , we eventually reduce the problem to the trivial case @xmath7 . the algorithm treedetection consists of a number of subroutines . these subroutines are partitioned into three types . the first type consists of @xmath44 subroutines which we refer to as the _ easy cases_. the other two types are referred to as _ special cases_. the second type consists of @xmath45 subroutines and the third type consists of @xmath46 subroutines . we first give a top - level description of treedetection before detailing the cases and their subroutines . * input : * a reticulation - visible network @xmath1 on @xmath0 and a rooted binary phylogenetic @xmath0-tree @xmath2 . * output : * * yes * if @xmath1 displays @xmath2 , and * no * otherwise . with the top - level description of the algorithm established , we now turn to the details of the various cases . if @xmath47 , let @xmath43 be a cherry in @xmath2 whose distance from its root is maximised . we then have the following ` easy ' cases ( each one holding only if the preceeding cases do not hold ) . subroutines for each easy case as well as their justifications are given in section [ easycases ] . easy cases * the @xmath31-element subset @xmath43 is a cherry in @xmath1 . * for some @xmath48 , we have @xmath49 . * for @xmath50 , we have @xmath51 is an ancestor of @xmath52 . * for @xmath53 , there is an ancestor @xmath54 of @xmath51 that is verified by @xmath55 but not by @xmath56 . * for some @xmath57 , there is a descendant @xmath20 of @xmath51 that is a reticulation not verified by @xmath55 . * for @xmath50 , there is a reticulation @xmath20 that is an ancestor of @xmath51 but not of @xmath56 . * for some @xmath58 , there are two non - comparable reticulations @xmath20 and @xmath54 that are descendants of @xmath40 and ancestors of @xmath51 . we establish in lemma [ inspecialcase ] below that if we are not in one of ( ec1)(ec7 ) , then we must be in one of the three situations illustrated in fig . [ situations ] . here , figs [ situations](i)(iii ) is a subgraph of the phylogenetic network at the completion of ( ec1)(ec7 ) . more precisely , each of these subgraphs consists of all ( directed ) paths from @xmath40 to either @xmath37 or @xmath38 . solid lines correspond to edges . dashed lines between named vertices correspond to ( directed ) paths with no reticulations unless one of the named vertices is @xmath20 . note that @xmath39 and @xmath59 are the parents of @xmath37 and @xmath38 , respectively . in situations 2 and 3 , the structure of the subgraph immediately ` below ' @xmath40 is not completely determined . two key vertices in the analysis to follow are those labelled @xmath60 and @xmath61 in fig . [ situations ] . if there is a reticulation on any path between @xmath40 and @xmath39 , then there is a unqiue maximal reticulation @xmath20 by ( ec7 ) . in this case we define @xmath62 to be the parent of @xmath39 on the path from @xmath20 to @xmath39 and @xmath63 to be the other parent of @xmath39 . if there is no reticulation , let the children of @xmath40 be @xmath64 and @xmath65 and define @xmath62 to be the parent of @xmath39 on the path from @xmath64 to @xmath39 , @xmath63 to be the parent of @xmath39 on the path from @xmath65 to @xmath39 . in each case define @xmath66 likewise . then @xmath60 is the ancestor of @xmath62 and @xmath67 at maximal distance from @xmath40 and @xmath68 is the ancestor of @xmath63 and @xmath69 at maximal distance from @xmath40 . ( the proof of lemma [ inspecialcase ] shows that these are well defined . ) for each situation , the crucial feature we will need to consider is the structure outside of the cherry @xmath43 in @xmath2 . since @xmath43 was chosen to be at maximal distance from the root , the sibling of the parent of @xmath37 and @xmath38 in @xmath2 is either a leaf @xmath70 or the parent of leaves @xmath70 and @xmath71 , where @xmath72 is a cherry . first , we consider the case that the sibling is a single leaf @xmath70 . let @xmath73 be the reticulation in @xmath1 at minimum path length from the root such that @xmath70 verifies @xmath73 and no other element verifies @xmath70 . if there is no such reticulation , define @xmath73 to be @xmath70 . let @xmath74 be the first reticulation on a path from @xmath73 to @xmath70 ( including @xmath73 itself ) that is not an ancestor of @xmath37 or @xmath38 . note that @xmath75 if no such vertex exists . furthermore , @xmath70 verifies @xmath74 , otherwise some other leaf would verify it , and this leaf would also verify @xmath73 . the first of the special cases is detailed below . the corresponding subroutines and their justifications are given in section [ specialcasealgs ] . special case : single leaf sibling * the leaf @xmath70 is not a descendant of @xmath40 . * there is a descendant @xmath20 of @xmath74 that is a reticulation and not verified by @xmath70 . * there is an ancestor @xmath20 of @xmath74 that is a reticulation and not an ancestor of @xmath37 or @xmath38 . * the leaf @xmath70 does not verify @xmath40 . the fifth subroutine deals with the case when none of ( sc1.1)(sc1.4 ) hold , in which case , @xmath76 . denote the parents of @xmath74 as @xmath77 and @xmath78 . let @xmath79 be the unique vertex at maximal distance from @xmath40 that is an ancestor of @xmath77 and an ancestor of either @xmath37 or @xmath38 . likewise , let @xmath80 be the unique vertex at maximal distance from @xmath40 that is an ancestor of @xmath78 and an ancestor of either @xmath37 or @xmath38 . for each of @xmath79 and @xmath80 either it is an ancestor of @xmath60 or an ancestor of @xmath68 , or it lies on one of the following ( directed ) paths : @xmath60 to @xmath62 , call it @xmath81 ; @xmath60 to @xmath67 , call it @xmath82 ; @xmath68 to @xmath63 , call it @xmath83 ; and @xmath68 to @xmath69 , call it @xmath84 . with these definitions , we identify an arc to delete from @xmath1 , or a vertex to delete from both @xmath2 and @xmath1 . the appropriate deletions are given in table [ singlectable ] , while justification for these deletions is given in section [ specialcasealgs ] . in the table , we denote that @xmath79 ( respectively , @xmath80 ) is an ancestor of a vertex @xmath20 by @xmath85 . .table showing the actions to perform on @xmath1 and @xmath2 to create @xmath35 and @xmath36 depending on the locations of @xmath79 and @xmath80 in each of situations 13 when the sibling of the parent of @xmath37 and @xmath38 is a single leaf @xmath70 . if the entry corresponding to the subcase we are in contains a single arc @xmath86 , then @xmath87 and @xmath88 . if the entry contains the leaf @xmath38 , then @xmath89 and @xmath90 . [ cols="^,^,^,^,^",options="header " , ] [ cherryctable ] here we present the subroutines for each of ( ec1)(ec7 ) with justification following each subroutine . throughout the presentation of these subroutines , as well as those presented for the two special cases , whenever we delete an arc or a leaf from @xmath1 or @xmath2 , we additionally contract any resulting degree - two vertex . suppose @xmath43 is a cherry in @xmath1 . then any embedding of @xmath36 in @xmath35 can be extended to an embedding of @xmath2 in @xmath1 by appending @xmath38 to the arc incident with @xmath37 in both @xmath35 and @xmath36 , and including the new edge in the embedding . conversely , any embedding of @xmath2 in @xmath1 gives rise to an embedding of @xmath36 in @xmath35 by deleting @xmath38 , its incident arc , and contracting the resulting degree - two vertex . * requirement : * easy cases 1 does not apply , and @xmath91 for some @xmath48 . [ algstep5a ] [ algstep5b ] [ algstep5bii ] if there is a tree path from @xmath92 to a leaf @xmath93 , then every rooted binary phylogenetic @xmath0-tree embedded in @xmath1 must display the rooted triple @xmath94 . since @xmath50 is a cherry in @xmath2 , no embedding of @xmath2 in @xmath1 can exist . thus returning * no * in line 3 is correct . so assume that there is no such tree path from @xmath92 . then there is a tree path from @xmath92 to a parent @xmath95 of a reticulation @xmath20 . assume @xmath96 . observe that @xmath20 is not a descendant of @xmath52 , otherwise @xmath55 would also verify @xmath52 contradicting the definition of @xmath52 . thus @xmath20 is verified by some leaf @xmath93 . note that we can choose @xmath97 since @xmath96 and @xmath20 is not a descendant of @xmath52 . now an embedding of a rooted binary phylogenetic @xmath0-tree in @xmath1 that uses arc @xmath98 would display the rooted triple @xmath94 if @xmath52 is not a descendant of @xmath20 , and @xmath99 if @xmath52 is a descendant of @xmath20 . on the other hand , if @xmath100 , then it is easily seen that if there is an embedding of @xmath2 that does not use arc @xmath98 , then there is also an embedding of @xmath2 that uses arc @xmath98 . thus in both cases there is an embedding of @xmath2 in @xmath1 if and only if there is an embedding of @xmath2 in @xmath35 . the justifications for the next three subroutines are similar to that for subroutine 2 and are omitted . * requirement : * easy cases 12 do not apply , and @xmath51 is an ancestor of @xmath52 for @xmath101 . * requirement : * easy cases 13 do not apply and , for some @xmath48 , there is an ancestor @xmath54 of @xmath51 which is verified by @xmath55 but not by @xmath56 . * requirement : * easy cases 14 do not apply and , for some @xmath48 , there is a path @xmath25 from @xmath51 to a reticulation not verified by @xmath55 . * requirement : * easy cases 15 do not apply and , for @xmath50 , there is a reticulation @xmath20 that is an ancestor of @xmath51 but not an ancestor of @xmath56 . since @xmath20 is an ancestor of @xmath51 it must be verified by some leaf @xmath102 . but @xmath20 is not an ancestor of @xmath56 , so @xmath97 . if an embedding of @xmath2 uses the arc @xmath103 , then it must display the rooted triple @xmath94 ; a contradiction . thus there is an embedding of @xmath2 in @xmath1 if and only if there is an embedding of @xmath2 in @xmath35 . * requirement : * easy cases 16 do not apply , and there are two non - comparable reticulations @xmath20 and @xmath54 that are descendants of @xmath40 and , for some @xmath48 , ancestors of @xmath51 . without loss of generality , we may assume @xmath20 and @xmath54 are at maximal distance from @xmath40 , that is , there is no reticulation @xmath104 that is a descendant of @xmath20 such that @xmath104 and @xmath54 satisfy the conditions of the requirement and , likewise , no reticulation @xmath105 that is a descendant of @xmath54 such that @xmath20 and @xmath105 satisfy the conditions of the requirement . since @xmath20 and @xmath54 are descendants of @xmath40 , neither is verified by both @xmath37 and @xmath38 and so , by ( ec4 ) , neither is verified by @xmath37 or by @xmath38 . thus there are leaves @xmath106 such that @xmath107 verifies @xmath20 and @xmath108 verifies @xmath54 . since @xmath20 and @xmath54 are non - comparable , @xmath109 as not all paths to a single leaf can go through two non - comparable vertices . by ( ec6 ) , each of @xmath20 and @xmath54 is an ancestor of both @xmath37 and @xmath38 . suppose first that there is a reticulation @xmath104 that is on a path from @xmath20 to @xmath51 . take @xmath104 to be such a reticulation at shortest distance to @xmath20 . by our assumption , @xmath104 can not be non - comparable with @xmath54 , so it must be a descendant of both @xmath54 and @xmath20 . furthermore , @xmath104 must be verified by some @xmath110 . since @xmath2 must display at most one of @xmath111 and @xmath112 ; we delete the appropriate reticulation arc incident with @xmath104 so that there is an embedding of @xmath2 in @xmath1 if and only if there is an embedding of @xmath2 in @xmath35 . by symmetry , we may now assume that there is no reticulation that is on a path from @xmath20 to @xmath51 or from @xmath54 to @xmath51 . let @xmath92 be the parent of @xmath51 that is a descendant of @xmath20 , and let @xmath113 be the parent of @xmath51 that is a descendant of @xmath54 . note that these are well defined by our latest assumption and , also , @xmath114 since @xmath20 and @xmath54 are non - comparable . if an embedding of @xmath2 uses arc @xmath115 , it must display the rooted triple @xmath116 , and if it uses arc @xmath117 , it must display the rooted triple @xmath118 . at most one of these rooted triples is displayed by @xmath2 . thus , with @xmath35 as constructed , there is an embedding of @xmath2 in @xmath1 if and only if there is an embedding of @xmath2 in @xmath35 . we first show that if none of ( ec1)(ec7 ) hold , then we are indeed in one of situations 1 , 2 , and 3 . thereafter , we present the subroutines and their justifications for the special cases . [ inspecialcase ] let @xmath6 , and let @xmath1 be a phylogenetic network on @xmath0 and let @xmath2 be a rooted binary phylogenetic @xmath0-tree . let @xmath43 be a cherry in @xmath2 whose distance from its root is maximised . if none of ( ec1)(ec7 ) applies , then one of situations @xmath119 , @xmath31 , and @xmath120 applies . let @xmath48 . suppose first that the parent of @xmath55 is a tree vertex @xmath10 . then there is a tree - path from @xmath10 that reaches either a parent of a reticulation @xmath20 not verified by @xmath55 , or a leaf @xmath102 . by ( ec2 ) , @xmath121 , so ec5 excludes the tree - path reaching a parent of a reticulation @xmath20 . but then @xmath11 verifies every vertex that @xmath55 verifies ( except @xmath55 itself ) , contradicting the definition of @xmath51 . now suppose that the parent of @xmath55 is a reticulation @xmath122 . let the parents of @xmath54 be @xmath123 and @xmath124 , and note that @xmath123 and @xmath124 are both tree vertices since parents of reticulations are not visible . the vertex @xmath51 must be an ancestor of both @xmath123 and @xmath124 . by ( ec5 ) , @xmath123 can not have a path to a reticulation other than @xmath54 . but then @xmath123 must have a tree path to a leaf @xmath11 that verifies @xmath51 ; a contradiction . it now follows that @xmath51 is the parent of @xmath55 for all @xmath125 . the vertex @xmath40 can not be a reticulation , as its child would also be verified by @xmath37 and @xmath38 . thus @xmath40 has two children @xmath64 and @xmath65 , neither of which is verified by both @xmath37 and @xmath38 . furthermore , by ( ec4 ) , neither @xmath64 nor @xmath65 is verified by exactly one of @xmath37 and @xmath38 . hence there is a ( directed ) path from @xmath64 to each of @xmath37 and @xmath38 , and also from @xmath65 to each of @xmath37 and @xmath38 . except for @xmath39 and @xmath59 , suppose that there are no reticulations on any path between @xmath40 and @xmath39 and between @xmath40 and @xmath59 . then the paths from @xmath64 to each of @xmath37 and @xmath38 are unique , and the paths from @xmath65 to each of @xmath37 and @xmath38 are unique . denote the parent of @xmath39 on the path from @xmath64 to @xmath37 by @xmath62 , and the parent of @xmath39 on the path from @xmath65 to @xmath37 by @xmath63 . similarly , denote by @xmath67 and @xmath69 the parents of @xmath59 . let @xmath60 be the last vertex on the path from @xmath64 to @xmath62 that is an ancestor of @xmath38 , and let @xmath68 be the last vertex on the path from @xmath68 to @xmath63 that is an ancestor of @xmath38 . ( possibly @xmath126 or @xmath127 . ) thus , if , apart from @xmath39 and @xmath59 , there are no reticulations on any path between @xmath40 and @xmath37 , and between @xmath40 and @xmath38 , then we are in situation 1 . now suppose that , in addition to @xmath39 and @xmath59 , there is a reticulation on a path between @xmath40 and @xmath39 or between @xmath40 and @xmath59 . let @xmath20 be such a reticulation at maximal distance from @xmath40 . by ( ec6 ) , @xmath20 is an ancestor of both @xmath37 and @xmath38 . by ( ec7 ) , every other reticulation on a path from @xmath40 to @xmath39 or from @xmath40 to @xmath59 that is not @xmath39 or @xmath59 is an ancestor of @xmath20 . by maximality , the paths from @xmath20 to @xmath37 and from @xmath20 to @xmath38 are unique . let the parent of @xmath39 on the path from @xmath20 to @xmath37 be @xmath62 and let the parent of @xmath59 on the path from @xmath20 to @xmath38 be @xmath67 . let @xmath60 be the last vertex on the path from @xmath20 to @xmath62 that is an ancestor of @xmath38 . let @xmath63 and @xmath69 be the other parents of @xmath39 and @xmath59 , respectively . let @xmath68 be the ancestor of @xmath63 and @xmath69 at maximal distance from @xmath40 . note that @xmath68 is well defined . to see this , assume that there were two possibilities @xmath68 and @xmath128 , both ancestors of @xmath63 and @xmath69 , and at maximal distance from @xmath40 . by maximality , @xmath68 and @xmath128 are non - comparable , so there would have to be a reticulation where the paths from @xmath68 and @xmath128 to @xmath63 merge , and where the paths from @xmath68 and @xmath128 to @xmath69 merge . these two reticulations would also have to be non - comparable , contradicting the exclusion of ( ec7 ) . if @xmath68 is not an ancestor of @xmath20 , then we are in situation 2 . if @xmath68 is an ancestor of @xmath20 , then we are in situation 3 . we now present the subroutines for the special case when the sibling of the parent of @xmath37 and @xmath38 in @xmath2 is a single leaf @xmath70 . [ sc:1 ] * requirement : * the leaf @xmath70 is not a descendant of @xmath40 . this is valid since any embedding of @xmath2 in @xmath1 must have the last common ancestor of @xmath37 and @xmath70 as an ancestor of @xmath40 . thus there are no paths ( in the embedding ) from @xmath40 to any leaves other than @xmath37 and @xmath38 , so we can resolve @xmath39 either way without changing the result of the embedding . in particular , @xmath1 displays @xmath2 if and only if @xmath35 displays @xmath2 . the justification of the next subroutine is similar to that given for easy case 2 and is omitted . * requirement : * there is a path @xmath25 from @xmath74 to a reticulation not verified by @xmath70 . * requirement : * there is an ancestor @xmath20 of @xmath74 that is a reticulation and not an ancestor of @xmath37 or @xmath38 . since @xmath20 is an ancestor of @xmath74 , but not an ancestor of @xmath37 or @xmath38 , we must have @xmath129 , and so @xmath20 must be verified by some leaf @xmath130 . since @xmath20 is not an ancestor of @xmath37 or @xmath38 , it follows that @xmath93 . if an embedding of @xmath2 uses the arc @xmath131 , then it must display the rooted triple @xmath132 ; a contradiction . thus there is an embedding of @xmath2 in @xmath1 if and only if there is an embedding of @xmath2 in @xmath35 . * requirement : * the leaf @xmath70 does not verify @xmath40 . if @xmath70 does not verify @xmath40 , then there is a ( directed ) path @xmath25 from an ancestor of @xmath40 to @xmath74 not via @xmath40 , and @xmath77 is on this path . if we are in situation 2 or situation 3 , then there is no embedding of @xmath2 in @xmath1 that uses arc @xmath131 . otherwise , the embedding of @xmath2 displays the rooted triple @xmath133 , where @xmath11 is a leaf verifying @xmath20 ; a contradiction . if we are in situation 1 and there is an embedding of @xmath2 that uses arc @xmath131 , then in the embedding the last common ancestor of @xmath37 and @xmath70 is an ancestor of @xmath40 , so ( as above ) we could adjust the embedding by resolving @xmath134 either way and it would still be valid . by ( sc1.3 ) , there is an unique path @xmath135 from @xmath40 to @xmath74 . now @xmath135 can not use vertex @xmath60 and vertex @xmath68 , so we can resolve @xmath39 or @xmath59 towards the ancestor not on @xmath135 and @xmath74 towards its parent on @xmath135 , and still have a valid embedding , but one not using arc @xmath131 . thus there is an embedding of @xmath2 in @xmath1 if and only if there is an embedding of @xmath2 in @xmath35 . to complete the analysis of the first special case , we present that last of its subroutines and justify the actions in table [ singlectable ] . [ sc1.5 ] * requirement : * easy cases 17 and special cases 1.11.4 do not apply . the sibling of the parent of @xmath37 and @xmath38 in @xmath2 is a single leaf @xmath70 . if an embedding of @xmath2 in @xmath1 uses the arc @xmath131 and @xmath79 is on the path @xmath81 or @xmath82 , then @xmath39 and @xmath59 will have to resolve towards @xmath63 and @xmath69 . otherwise , the embedding does not have a vertex whose leaf descendants are just @xmath37 and @xmath38 . hence , if each of @xmath79 and @xmath80 lie on @xmath81 or @xmath82 , then we can delete @xmath136 . similarly , if each of @xmath79 and @xmath80 lie on @xmath83 or @xmath84 , then we can delete @xmath137 . if we are in situation 1 , and @xmath79 lies on @xmath81 or @xmath82 and @xmath80 lies on @xmath83 or @xmath84 , then , in an embedding of @xmath2 in @xmath1 , the last common ancestor of @xmath37 and @xmath70 would be @xmath40 which ever way we resolved @xmath74 . thus , in any of these cases , there is no path to a leaf other than @xmath37 , @xmath38 , and @xmath70 from @xmath40 and so , if an embedding exists , one exists with @xmath74 resolved each way ( and @xmath39 and @xmath59 resolved accordingly ) . thus , we can delete @xmath138 . similar argument holds if we are in situation 1 , and @xmath79 lies on @xmath83 or @xmath84 and @xmath80 lies on @xmath81 or @xmath82 . if we are in situations 2 or 3 , and @xmath79 lies on @xmath83 or @xmath84 , no embedding of @xmath2 can resolve @xmath74 towards @xmath77 . this follows as then @xmath39 and @xmath59 would have to resolve towards @xmath62 and @xmath67 and , if @xmath107 is a leaf that verifies @xmath20 ( note @xmath139 ) , then the embedding of @xmath2 in @xmath1 would display the rooted triple @xmath140 ; a contradiction . so we can delete arc @xmath131 . similarly , if we are in situations 2 or 3 , and @xmath80 lies on @xmath83 or @xmath84 , then we can delete arc @xmath141 . now suppose @xmath79 is an ancestor of @xmath60 ( denoted @xmath142 in the table ) . then , if an embedding of @xmath2 in @xmath1 uses the arc @xmath131 and arc @xmath117 , the last common ancestor of @xmath37 and @xmath70 in the embedding would be an ancestor of @xmath79 . thus there is no path to a leaf other than @xmath37 , @xmath38 , and @xmath70 from @xmath79 and so an embedding exists with @xmath39 and @xmath59 resolved towards @xmath62 and @xmath67 , respectively . similarly , if @xmath79 is an ancestor of @xmath68 and an embedding of @xmath2 in @xmath1 exists using arcs @xmath138 and @xmath136 , then one exists with @xmath39 and @xmath59 resolved towards @xmath63 and @xmath69 , respectively . a symmetric argument holds with @xmath80 instead of @xmath79 . thus if @xmath79 is an ancestor of @xmath60 and @xmath80 lies on @xmath83 or @xmath84 ( or vice versa ) , then we can delete arc @xmath137 and the resulting phylogenetic network on @xmath0 displays @xmath2 if and only if @xmath1 displays @xmath2 . likewise , if @xmath79 is an ancestor of @xmath68 and @xmath80 lies on @xmath81 or @xmath82 , then we can delete arc @xmath136 . also , if both @xmath79 and @xmath80 are ancestors of @xmath60 , we can delete arc @xmath117 , and if both are ancestors of @xmath68 , we can delete arc @xmath115 . in each of the last three subcases , the resulting phylogenetic network on @xmath0 displays @xmath2 if and only if @xmath1 displays @xmath2 . if @xmath79 is an ancestor of @xmath60 and @xmath80 lies on @xmath81 or @xmath82 , and an embedding of @xmath2 in @xmath1 exists that resolves @xmath74 towards @xmath78 , then in this embedding @xmath39 and @xmath59 must resolve towards @xmath63 and @xmath69 , respectively , in which case , in the embedding , the last common ancestor of @xmath37 and @xmath70 is @xmath40 in situation 1 , and an ancestor of @xmath20 in situations 2 and 3 . in situation 1 , we could therefore resolve @xmath74 towards @xmath77 without changing the topology of the embedded tree . in situations 2 and 3 , if @xmath139 , where @xmath107 is a leaf that verifies @xmath20 , then the embedding displays the rooted triple @xmath143 ; a contradiction . thus @xmath144 , and therefore @xmath79 lies on the path from @xmath20 to @xmath60 . we can therefore resolve @xmath74 towards @xmath77 without changing the topology of the embedded tree . it follows that in either case , we can delete arc @xmath141 and the resulting phylogenetic network on @xmath0 displays @xmath2 if and only if @xmath1 displays @xmath2 . a similar argument holds if we interchange the roles of @xmath79 and @xmath80 , in which case , we delete the arc @xmath145 . a symmetric argument holds in situation 1 if @xmath79 is an ancestor of @xmath68 and @xmath80 lies on @xmath83 or @xmath84 ( or vice versa ) . the remaining subcase is that @xmath79 is an ancestor of @xmath60 and @xmath80 is an ancestor of @xmath68 , or vice versa . we now show that @xmath146 displays @xmath147 if and only if @xmath1 displays @xmath2 . in any embedding of @xmath147 in @xmath146 , there is a vertex whose leaf descendants are precisely @xmath37 and @xmath70 . also , in the embedding , the peak of the embedded path from @xmath37 to @xmath70 must be an ancestor of either @xmath60 or @xmath68 , since there is no ( directed ) path from either to @xmath70 . if this peak is an ancestor of @xmath60 , then we can extend this embedding to an embedding of @xmath2 in @xmath1 by adding arcs @xmath148 and @xmath149 , while if it is an ancestor of @xmath68 , then we extend the embedding to such an embedding by adding arcs @xmath150 and @xmath149 . these must still be valid embeddings since there is a tree path from @xmath60 to @xmath67 , and from @xmath68 to @xmath69 , so in the embedding @xmath38 is in the correct location . conversely , any embedding of @xmath2 in @xmath1 gives rise to an embedding of @xmath147 in @xmath146 by deleting @xmath38 and its incoming arc . this completes the justification of table [ singlectable ] . finally , we describe the subroutines for the special case when the sibling of the parent of @xmath37 and @xmath38 in @xmath2 is the parent of a second cherry @xmath72 . * requirement : * the vertices @xmath151 and @xmath40 are non - comparable . if @xmath151 and @xmath40 are non - comparable , then , in any embedding of @xmath2 in @xmath1 , the last common ancestor of @xmath37 and @xmath70 is an ancestor of @xmath40 , and so there are no paths in the embedding from @xmath40 to any leaf @xmath11 other than @xmath37 and @xmath38 ; otherwise , the embedding displays the rooted triple @xmath133 . thus , we can resolve @xmath39 either way , and the resulting phylogenetic network on @xmath0 displays @xmath2 if and only if @xmath1 displays @xmath2 . * requirement : * the vertex @xmath151 is not on one of the paths from @xmath40 to either @xmath39 or @xmath59 . if @xmath151 is not on one of the paths from @xmath40 to either @xmath39 or @xmath59 , then , in any embedding of @xmath2 in @xmath1 , the last common ancestor of @xmath37 and @xmath70 is an ancestor of @xmath151 , and so there are no paths in the embedding from @xmath151 to a leaf @xmath11 other than @xmath70 and @xmath71 ; otherwise the embedding displays the rooted triple @xmath132 . therefore we can resolve @xmath73 either way , and the resulting phylogenetic network on @xmath0 displays @xmath2 if and only if @xmath1 displays @xmath152 . * requirement : * there is a reticulation @xmath20 that is an ancestor of @xmath153 or @xmath154 but it is not an ancestor of @xmath37 or @xmath38 . reticulation @xmath20 must be verified by some leaf @xmath155 since @xmath20 is a descendant of @xmath151 and an ancestor of @xmath73 and @xmath156 , but not an ancestor of either @xmath37 or @xmath38 . thus any embedding that resolves @xmath73 towards @xmath20 would display the rooted triple @xmath132 ; a contradiction . we now justify the final actions in table [ cherryctable ] . [ sc2.4 ] * requirement : * easy cases 17 and special cases 2.12.3 do not apply . the sibling of the parent of @xmath37 and @xmath38 in @xmath2 is the parent of a cherry @xmath72 . if an embedding of @xmath2 in @xmath1 uses the arc @xmath157 and @xmath158 is on the path @xmath81 or @xmath82 , then @xmath39 and @xmath59 will have to resolve towards @xmath63 and @xmath69 , respectively , so that the embedding contains a vertex whose leaf descendants are @xmath37 and @xmath38 . likewise , if @xmath158 is an ancestor of @xmath60 and @xmath159 , so that @xmath70 and @xmath71 do not meet before reaching an ancestor of @xmath37 , then an embedding of @xmath2 in @xmath1 that uses arc @xmath157 implies @xmath39 and @xmath59 will have to resolve towards @xmath63 and @xmath69 , respectively ; otherwise , one of the rooted triples @xmath160 and @xmath161 would be displayed by the embedding . the same outcomes hold if replace @xmath162 and @xmath158 with @xmath163 and @xmath164 , respectively , in the hypotheses . thus if both @xmath158 and @xmath164 are of type 1 , we can delete @xmath136 since , either way @xmath73 resolves , the embedding can not use that arc , and the resulting phylogenetic network on @xmath0 displays @xmath2 if and only if @xmath1 displays @xmath2 . similarly , if both @xmath158 and @xmath164 are of type 2 , we can delete @xmath117 . if we are in situation 1 and , without loss of generality , @xmath158 is type 1 and @xmath164 is type 2 , then , in an embedding of @xmath2 in @xmath1 , the last common ancestor of @xmath37 and @xmath70 would be @xmath40 which ever way we resolved @xmath73 . thus , in any such embedding , there is no path to a leaf other than @xmath37 , @xmath38 , @xmath70 , and @xmath71 from @xmath40 and so , if an embedding exists , one exists with @xmath73 resolved each way ( and @xmath39 and @xmath59 resolved accordingly ) . thus we can delete @xmath157 , and the resulting phylogenetic network on @xmath0 displays @xmath2 if and only if @xmath1 displays @xmath2 . if we are in situations 2 or 3 , and @xmath158 is of type 2 , then no embedding of @xmath2 in @xmath1 can resolve @xmath165 towards @xmath77 . this follows as then @xmath39 and @xmath59 would have to resolve towards @xmath62 and @xmath67 , respectively . but then if @xmath107 is a leaf that verifies @xmath20 , and noting that @xmath166 , the embedding would display the rooted triple @xmath140 ; a contradiction . so we can delete arc @xmath157 , and the resulting phylogenetic network on @xmath0 displays @xmath2 if and only if @xmath1 displays @xmath2 . similarly , if @xmath164 is of type 2 , we can delete arc @xmath163 . now suppose @xmath158 is of type 3 . then if an embedding of @xmath2 in @xmath1 uses the arc @xmath157 and arc @xmath117 , it follows that the last common ancestor of @xmath37 and @xmath70 in this embedding would be an ancestor of @xmath158 . thus in such an embedding there is no path to a leaf other than @xmath37 , @xmath38 , @xmath70 , and @xmath71 from @xmath158 , and so an embedding also exists with @xmath39 and @xmath59 resolved towards @xmath62 and @xmath67 . similarly , if @xmath158 is of type 4 and an embedding of @xmath2 in @xmath1 exists using arcs @xmath157 and @xmath136 , then one exists with @xmath39 and @xmath59 resolved towards @xmath63 and @xmath69 . a symmetric argument holds with @xmath164 instead of @xmath158 . thus if @xmath158 is of type 3 and @xmath80 is of type 2 ( or vice versa ) , we can delete arc @xmath137 , and the resulting phylogenetic network on @xmath0 displays @xmath2 if and only if @xmath1 displays @xmath2 . likewise , if @xmath158 is of type 4 and @xmath164 is of type 1 ( or vice versa ) , we can delete arc @xmath115 . also , if both @xmath158 and @xmath164 are of type 3 , we can delete arc @xmath137 , and if both are of type 4 , we can delete arc @xmath136 . in each of the last three subcases , the resulting phylogenetic network displays @xmath2 if and only if @xmath1 displays @xmath2 . suppose that @xmath158 is of type 3 and @xmath164 is of type 1 , and that an embedding of @xmath2 in @xmath1 exists that resolves @xmath73 and @xmath156 towards @xmath78 and @xmath167 , respectively . then , in this embedding , @xmath39 and @xmath59 must resolve towards @xmath63 and @xmath69 , respectively , and the last common ancestor of @xmath37 and @xmath70 is @xmath40 in situation 1 , and an ancestor of @xmath20 in situation 2 and 3 . in situation 1 , we could therefore resolve @xmath73 and @xmath156 towards @xmath77 and @xmath168 without changing the topology of the embedded tree . in situations 2 and 3 , if @xmath169 , where @xmath107 is a leaf that verifies @xmath20 , then the embedding displays the rooted triple @xmath143 ; a contradiction . thus , in situations 2 and 3 , we may assume @xmath170 , and therefore @xmath158 lies on the path from @xmath20 to @xmath60 . we can therefore resolve @xmath73 to @xmath77 without changing the topology of the embedding of @xmath2 . regardless of the situation , we can delete arc @xmath163 , and the resulting phylogenetic network on @xmath0 displays @xmath2 if and only if @xmath1 displays @xmath2 . a similar argument holds if @xmath158 is of type 1 and @xmath164 is of type 3 , in which case , we delete @xmath157 . a symmetric argument holds in situation 1 if @xmath158 is of type 4 and @xmath164 is of type 2 ( or vice versa ) . the remaining case is that @xmath158 is of type 3 and @xmath164 is of type 2 ( or vice versa ) . an identical argument to that given to justify the entries in table [ singlectable ] corrsponding to deleting the leaf @xmath38 in the first special case applies . in particular , this gives that @xmath146 displays @xmath147 if and only if @xmath1 displays @xmath2 . in this section we consider the running time of the treedetection algorithm . the input to the algorithm is a reticulation - visible network @xmath1 on a finite set @xmath0 and a rooted binary phylogenetic @xmath0-tree @xmath2 . traditionally running times in phylogenetics are given in terms of the size of the taxa set , @xmath171 , although the representation of a general phylogenetic network may be much larger that @xmath172 . however in the case of reticulation - visible networks , the number of vertices of the network is at most @xmath4 , and so the input in this case is of size @xmath172 ( see theorem [ sharp ] ) . clearly determining if @xmath7 can be done in constant time , and finding a cherry @xmath173 in @xmath2 can be done in linear time . the main body of the algorithm then boils down to determining which of several cases occurs , and then running a specific subroutine . we first show that we can determine which case occurs in @xmath174 steps . to check the easy cases we need to know the vertices @xmath134 and @xmath40 , and their ancestors and descendants . for each vertex of @xmath1 we prepare a list of : * all ancestors * all descendants * all those descendants that verify the vertex * all those descendants reachable by a tree - path . we also order the vertices by distance of the vertex from the root ; since there are less than @xmath175 vertices altogether , this can be done in @xmath174 steps using a depth first search . we can then scan through the list in order to determine @xmath134 and @xmath40 , by checking whether each vertex satisfies the conditions . further scanning through the list and applying simple checks at each vertex can determine which , if any , of the easy cases occurs , and since we only need compare at most two vertices ( ec7 ) , this can all be achieved in @xmath174 steps . depending on whether the sibling of the cherry @xmath173 is a single leaf @xmath70 or another cherry @xmath176 , we must then check which of several special cases occurs . however , once we have identified @xmath74 , or @xmath177 , by looking through the lists above , it is again a matter of checking whether vertices with certain straightforward conditions on their ancestors and descendants exist or not , and this can be done in @xmath174 steps . finally , each of the subroutines either returns * no * or modifies @xmath1 and possibly @xmath2 before recursively calling treedetection . determining whether to return * no * or make the modifications can be accomplished in @xmath174 steps , since they again require at most checking the existence of vertices with specific ancestor or descendant conditions , possibly including reachability by a tree - path , but since the network is of linear size , this is can be done , and then any modifications of @xmath1 and @xmath2 are only by deleting a constant number of arcs or vertices . this analysis includes determining which row of the final tables is appropriate , since we can identify the vertices @xmath178 , or @xmath179 by scanning through the vertex list and checking straightforward ancestor / descendant conditions , and then determining which of the rows is applicable is again checking ancestor and descendant conditions on these vertices , e.g. we can tell if a vertex is on a path from @xmath60 to @xmath62 by checking if it is a descendant of @xmath60 and an ancestor of @xmath62 . thus the entire algorithm breaks down into a constant number of checks , each of which can be accomplished in @xmath174 steps , followed by a small modification to @xmath1 and possibly @xmath2 , and a recursive call to treedetection on an input that is smaller in either the number of reticulations or the size of the leaf set . since the number of reticulations is linear in @xmath171 , there can be at most @xmath172 recursive calls , and so the entire algorithm completes in @xmath19 steps . in this section , we establish theorem [ sharp ] . let @xmath1 be a reticulation - visible network on @xmath0 with @xmath180 vertices in total and @xmath181 reticulations . let @xmath15 . we first show that @xmath182 and @xmath183 the proof of these two inequalities is by induction on @xmath184 . if @xmath185 , then @xmath1 consists of a single vertex , and ( [ ineq1 ] ) and ( [ ineq2 ] ) hold . suppose that @xmath186 , and that ( [ ineq1 ] ) and ( [ ineq2 ] ) hold for all reticulation - visible networks with fewer leaves . first assume that @xmath1 has a cherry @xmath43 . let @xmath35 be the network obtained from @xmath1 by reducing @xmath43 . since every vertex in @xmath1 is visible , it follows that every vertex in @xmath35 is visible , and so @xmath35 is a reticulation - visible network . therefore , as @xmath35 has @xmath187 vertices , @xmath181 reticulations , and @xmath188 leaves , it follows by the induction assumption that @xmath189 and @xmath190 , so @xmath191 and @xmath192 thus ( [ ineq1 ] ) and ( [ ineq2 ] ) hold . now assume that @xmath1 does not contain a cherry . let @xmath10 be a reticulation in @xmath1 such that amongst all reticulations in @xmath1 it is at maximum distance from the root @xmath9 . let @xmath25 be a path from @xmath9 to @xmath10 that realises this maximum distance . by maximality and the assumption that @xmath1 has no cherry , the child vertex of @xmath10 is a leaf , @xmath11 say . also , note that , as @xmath1 is reticulation visible , neither parent of @xmath10 is a reticulation . let @xmath35 be the network obtained from @xmath1 by deleting the vertices @xmath10 and @xmath11 and their incident arcs , contracting any resulting degree - two vertices , and then replacing any parallel arcs with a single arc and contracting any degree - two vertices resulting from this replacement . since the final step in this process could not have created any further parallel arcs , it is easily seen that @xmath35 is a phylogenetic network on @xmath193 . furthermore , in the process of obtaining @xmath1 from @xmath35 , we initially lose @xmath119 reticulation and @xmath46 vertices in total . additionally , at most two pairs of parallel arcs are replaced with a single arc . each such replacement , loses @xmath119 reticulation and @xmath31 vertices in total . thus , if @xmath194 and @xmath195 denotes the total number of vertices and the number of reticulations in @xmath35 , we have @xmath196 and @xmath197 we next show that either @xmath35 or a phylogenetic network obtained from @xmath1 in an analogous way is reticulation visible , thereby obtaining a phylogenetic network that satisfies the induction assumptions . let @xmath64 and @xmath65 denote the parents of @xmath10 . if @xmath11 does not verify the visibility of any reticulation other than @xmath10 in @xmath1 , then @xmath35 is reticulation visible . therefore , suppose that @xmath11 also verifies the visibility of a reticulation @xmath54 in @xmath1 , where @xmath198 . without loss of generality , we may assume that @xmath64 is the last parent of @xmath10 on @xmath25 . if @xmath64 is an ancestor of @xmath65 , then @xmath25 is not a path of maximum distance from @xmath9 to @xmath10 , so either @xmath64 is a descendant of @xmath65 or @xmath64 is non - comparable to @xmath65 . in either case , this implies that @xmath64 has a child vertex , not equal @xmath10 , that is not an ancestor of @xmath10 . let @xmath199 denote this child vertex . by the maximality of @xmath25 and the assumption that @xmath1 has no cherries , @xmath199 is either a leaf or a reticulation . if @xmath199 is a leaf , then , as every path in @xmath1 from @xmath9 to @xmath11 passes through @xmath54 , every path in @xmath1 from @xmath9 to @xmath199 also passes through @xmath54 . it now follows that if @xmath199 is a leaf , @xmath35 is reticulation visible . therefore suppose that @xmath199 is a reticulation . by the maximality of @xmath25 and the assumption that @xmath1 has no cherries , the child of @xmath199 is a leaf , @xmath200 say . let @xmath201 be the network obtained from @xmath1 by deleting @xmath199 and @xmath200 and their incident arcs , contracting any resulting degree - two vertices , and then replacing any parallel arcs with a single arc and contracting any degree - two vertex resulting from this replacement . as above , if @xmath202 and @xmath203 denote the total number of vertices and the number of reticulations in @xmath201 , then @xmath204 and @xmath205 if @xmath200 does not verify the visibility of any reticulation other than @xmath199 in @xmath1 , then , instead of applying the inductive step to @xmath10 , apply the inductive step to @xmath199 . in particular , @xmath201 is reticulation visible and therefore satisfies the induction assumptions . thus we may assume that @xmath200 also verifies the visibility of a reticulation @xmath105 in @xmath1 , where @xmath206 . now every path in @xmath1 from @xmath9 to @xmath11 passes through @xmath54 , in particular , every path in @xmath1 from @xmath9 to @xmath11 using the arc directed into @xmath64 passes through @xmath54 . in turn , this implies that every path in @xmath1 from @xmath9 to @xmath200 using the arc directed into @xmath64 passes through @xmath54 . we deduce that either @xmath54 is an ancestor of @xmath105 or @xmath105 is an ancestor of @xmath54 in @xmath1 . if @xmath54 is an ancestor @xmath105 in @xmath1 , then every path in @xmath1 from @xmath9 to @xmath200 passes through @xmath54 , and so @xmath35 is reticulation visible , in which case , @xmath35 satisfies the induction assumptions . on the other hand , if @xmath105 is an ancestor of @xmath54 in @xmath1 , then every path in @xmath1 from @xmath9 to @xmath11 passes through @xmath105 , and so @xmath201 is reticulation visible , in which case , @xmath201 satisfies the induction assumptions . without loss of generality , we may assume that the induction assumptions hold for @xmath35 . by induction , and ( [ eqn1 ] ) and ( [ eqn2 ] ) , it follows that @xmath207 so @xmath208 , and @xmath209 so @xmath210 . hence ( [ ineq1 ] ) and ( [ ineq2 ] ) hold . to show that the upper bounds ( [ ineq1 ] ) and ( [ ineq2 ] ) are sharp for all integers @xmath18 , consider the reticulation - visible network shown in fig . [ 2sharp ] . here the bound is tight for when @xmath211 . for all integers @xmath212 , an analogous example can be constructed by replacing a leaf with a complete copy of this network . the resulting network is reticulation - visible , and the number of leaves has increased by @xmath119 , the number of reticulations has increased by @xmath120 , and the total number of vertices has increased by @xmath213 , thereby retaining the tightness of the bound . of course , if @xmath185 , then @xmath1 consists of a single vertex and the bounds hold exactly . p. gambette , a. d. m. gunawan , a. labarre , s. vialette , l. zhang . locating a tree in a phylogenetic network in quadratic time . in : _ 19th ann . ( recomb15 ) _ , lecture notes in computer science , vol . 9029 , pp . 96107 , 2015 .

pulsars are rapidly rotating neutron stars and are the outcome of the collapse of the core of a massive star with a mass of the order of or larger than eight solar masses . this process releases a huge gravitational energy of about , mainly in the form of neutrinos . during the collapse the density increases , and so does the magnetic field due to the trapping of the flux lines of the progenitor star by the high conductivity plasma . when the density reaches a value of around neutrinos become trapped within the protoneutron star and a neutrinosphere , characterized inside by a diffusive transport of neutrinos and outside by a free streaming of neutrinos , is formed and lasts for a few seconds here we focus on the structure of the neutrinosphere , the resonant flavor conversion that can happen in its interior , and the neutrino flux anisotropies induced by this phenomena in the presence of a strong magnetic field . we present a detailed discussion in the context of the spherical eddington model , which provides a simple but reasonable description of a static neutrino atmosphere , locally homogenous and isotropic . energy and momentum are transported by neutrinos and antineutrinos flowing through an ideal gas of nonrelativistic , nondegenerate nucleons and relativistic , degenerate electrons and positrons . we examine the details of the asymmetric neutrino emission driven by active - sterile neutrino oscillations in the magnetized protoneutron star , and the possibility for this mechanism to explain the intrinsic large velocities of pulsars respect to nearby stars and associated supernova remnants . departamento de fsica de altas energas , instituto de ciencias nucleares , universidad nacional autnoma de mxico , apartado postal 70 - 543 , 04510 mxico , distrito federal , mxico . 0.3 cm instituto balseiro and centro atmico bariloche , universidad nacional de cuyo and cnea , 8400 s. c. de bariloche , ro negro , argentina .

hep-ph0503113

the finding of pulsars in the sixties was one of the most astonishing discoveries in astrophysics@xcite . soon after their observation they were identified with neutron stars@xcite , whose existence had been theoretically predicted decades before . neutron stars are the densest , most rapidly rotating , and most strongly magnetized objects in the galaxy , features well understood by the standard physics of gravitational and elementary particle physics@xcite . they are unique laboratories , which give us access to extreme conditions that are virtually impossible to obtain on earth . neutron stars are of particular interest for neutrino physics because of the key role that these particles play in the very first stages of their formation@xcite . there is a remarkable characteristic of pulsars not satisfactorily understood : their large drift velocities with respect to nearby stars . the average value of pulsar velocities is in the range of @xmath4 , an order of magnitude larger than the mean velocity of ordinary stars in our galaxy ( @xmath5)@xcite . in addition , the distribution of these velocities is not gaussian . it seems to have a bimodal structure@xcite , with a significant fraction ( @xmath6 ) of the pulsar population having velocities higher than @xmath7 and up to a maximum of @xmath8 . neutron stars are formed in type - ii supernova explosions and , given the enormous energy liberated during these processes , it is quite natural to look for an explanation in terms of an impulse or _ kick _ received during their birth . this hypothesis is supported by recent analysis of individual pulsar motions and of associations between supernovae remnants and pulsars . it is also consistent with observed characteristics of binary systems with one or both constituents being a neutron star@xcite . despite this evidence , the physical origin of the kick is one of the most important unsolved puzzles in supernovae research . if the distribution is bimodal , then more than one mechanism could be responsible for the natal kicks . two classes of mechanism have been considered@xcite . one of them invokes an asymmetric mass ejection during the supernova explosion as the result of hydrodynamic perturbations caused by global density inhomogeneities prior to the core collapse@xcite or convective instabilities in the neutrino - heated layer behind the shock@xcite . recoil speeds as high as @xmath9 have been obtained with two - dimensional simulations . however , a recent three - dimensional calculation shows that even the most extreme asymmetric collapse does not produce final neutron star velocities above @xmath10@xcite , rendering the situation unclear . only a small fraction of the gravitational binding energy ( @xmath11 ) liberated during a supernova explosion is required to account for the huge electromagnetic luminosity observed and the ejection of the envelope . more than @xmath12 of the energy is released in the form of neutrinos , which are copiously produced when the iron core turns into a neutron core . a small asymmetry ( @xmath13 ) in the momentum taken away by neutrinos should be enough to give the nascent pulsar an acceleration consistent with the measured velocities . the asymmetric neutrino emission induced by strong magnetic fields is the basis of the second class of kick mechanisms , and effects such as parity violation@xcite , asymmetric field distributions@xcite , and dark spots@xcite have been considered in this context . the density in the interior of a protoneutron star is so high , of the order of the nuclear density , that even neutrinos remain trapped , being emitted from an approximately well defined surface , the neutrinosphere . in a magnetized stellar plasma there are anisotropic contributions to the neutrino refraction index@xcite and , as a consequence , the resonant flavor transformations of neutrinos diffusing in different directions respect to the magnetic field occur at different depths within the protostar . this phenomenon can induce a momentum flux asymmetry and is the basis of the neutrino - driven kick mechanism proposed by kusenko and segr ( ks)@xcite . with standard ( active ) neutrinos the mechanism works when the resonance region lies between the @xmath14 and @xmath15 ( or @xmath16 ) neutrinospheres , but requires an exceedingly high square mass difference , in conflict with the existing limits . a variant avoiding such a limitation can be implemented in terms of active - sterile neutrino oscillation@xcite . the measurement of the z - boson width at lep , limiting to three the number of light neutrinos having weak interactions , does not preclude the existence of additional neutral fermions that do not fill the standard model gauge interactions . they have been invoked to explain several questions in astronomy and cosmology and may becomes a necessity if the claimed lsnd evidence for oscillations were confirmed . the range of the oscillation parameters ( mass and mixing angle ) required by the pulsar kick mechanism overlaps with the allowed region for sterile neutrinos being candidates to cosmological dark matter . alternative kick mechanisms driven by neutrino oscillations have been proposed invoking the possible existence of transition magnetic moments@xcite , non orthonormality of the flavor neutrinos@xcite , violations of the equivalence principle@xcite , or off - resonant emission of sterile neutrinos@xcite . in this article we review the explanation of the pulsar kick in terms of the resonant neutrino oscillations . in section ii , we outline the creation of a protoneutron star as a result of a supernova event , and the formation in its interior of a gas of trapped neutrinos . the momentum flux carried by neutrinos is analyzed in section iii . section iv is devoted to the discussion of the neutrino oscillations in matter incorporating the effect of strong magnetic fields . the implication of this phenomenon as source of the asymmetry in the momentum flux transported by neutrinos emitted during the protoneutron star cooling is examined in section v. numerical results are obtained in section vi within the context of the eddington model , where the kick effect is explicitly evaluated . in the final section we summarize the outstanding features of the mechanism . neutron stars are extremely compact objects born in type ii supernova explosions@xcite , as the aftermath of the gravitational collapse of a massive star ( @xmath17 when it exhausts the sources of thermonuclear energy and its iron core reaches the chandrasekhar limit , @xmath18 . the process is triggered by the photodissociation of iron nuclei , @xmath19mev , which consumes energy reducing the pressure of the degenerated electron gas . when the collapse proceeds , pressure support decreases because of the electron capture by nuclei . electron neutrinos are copiously produced through the reaction @xmath20 and initially they freely escape from the core , but as density increases the medium becomes less transparent for them . the inner core ( @xmath21 ) collapses in a homologous way at a subsonic velocity proportional to the radius ( @xmath22 , with the density and temperature profiles remaining similar and only the scale changing with time@xcite . the external shells collapse at a supersonic velocity of the order of free fall velocity ( @xmath23 ) . during the first part of the collapse the main source of opacity for neutrinos is the coherent scattering by heavy nuclei , mediated by the neutral current weak interaction . the cross section for this reaction is proportional to @xmath24 , where @xmath25 is the number of nucleons in the nuclei , and is very effective in trapping neutrinos@xcite . if @xmath26 is the mean free path of neutrinos , then the time for them to diffuse out of the core , estimated from a random walk through a distance of the order of the core radius @xmath27 , is @xmath28 . on the other hand , the core dynamical time scale is of the order of the free fall time @xmath29 . at a certain density , of about @xmath30 for 10 mev neutrinos , both times become comparable and neutrinos are trapped@xcite . @xmath31-equilibrium between @xmath32 and @xmath14 is set up and electron neutrinos also become degenerated . in this way , the collapsing core is the only place , besides the early universe , where neutrinos are in thermal equilibrium . during the collapse the lepton fraction @xmath33 is kept almost constant near the value of the electron fraction at the beginning of the trapping ( @xmath34 ) , with only @xmath35 corresponding to neutrinos@xcite . the degenerate gas of trapped neutrinos diffuses towards the exterior , reaching regions of decreasing density where the collisions are less frequent , until they finally decouple from the star . although this process is continuous , it is convenient to introduce an effective emission surface called the neutrinosphere , characterized as the surface where the optical depth becomes approximately one . more accurately , the radius of the neutrinosphere @xmath36 is defined by the condition @xmath37 the medium is considered opaque for neutrinos inside the neutrinosphere and transparent outside . we can also define the energy sphere , where the energy exchanging reactions freeze out while energy conserving collisions may still be important . the collapse stops suddenly when , after a fraction of seconds , densities of around @xmath38 , exceeding nuclear density , are reached@xcite . then , a shock wave is built up near the interface of the halted inner core and the supersonically free falling outer part . the shock wave propagates outwards and eventually ejects the stellar mantle of the star . at its passage through the neutrinosphere nuclei are dissociated decreasing the neutrino opacity and causing a recession of the neutrinosphere . the @xmath14 produced by the reaction @xmath39 can escape freely building up a short ( @xmath4010msec ) @xmath14 burst , called the neutronization burst " or deleptonization burst " . the star remnant gravitationally decouples from the expanding ejecta and a protoneutron star has been formed@xcite . the neutrinos continue trapped due to the interaction with the residual particles of the medium , mainly nucleons , electrons , and neutrinos . for @xmath14 the dominant reactions are the @xmath31 and @xmath31-inverse , which mantain them in local thermodynamical equilibrium@xcite . in this case , the distinction between neutrinosphere and energy sphere is not crucial . the electron neutrinos also interact with nucleons through the neutral current , but the cross sections for these reactions are smaller . although the energy dependence of the process involving @xmath14 and @xmath41 is the same , in a protoneutron star the number of neutrons is larger than the number of protons , and therefore the @xmath14 suffer more collisions than the @xmath41 . as a consequence , the neutrinosphere of the @xmath14 is a bit further out than the one of the @xmath41 , and the mean energy per emitted particle is smaller . after the achievement of @xmath31 equilibrium , the total entropy is conserved and the collapse becomes adiabatic . in the case of the nonelectron ( active ) neutrinos and antineutrinos , the absence of muons and taus does not allow the establishment of chemical equilibrium with the stellar plasma . therefore , they do not transport leptonic number outside the star and their chemical potentials vanish . their interactions with the background are the same and no distinction needs to be made between @xmath15 and @xmath16 . deeper in the protostar the main interactions for these neutrinos are pair processes such as @xmath42 , etc , where @xmath43 denotes a nucleon , and the scattering reactions with nucleons and electrons . the scattering on electrons are less frequent , but the energy transferred is greater . this process dominates the energy exchange up to the energy sphere where it freezes out . beyond this radius , muon and tau neutrinos still scatter off neutrons until they reach the neutrinosphere . the cross section for @xmath44 is smaller than the one of the @xmath31 reaction , and hence the radius of the muon ( tau ) neutrinosphere @xmath45 is shorter than the radius of the electron neutrinosphere @xmath46 . the protoneutron star contracts , cools , and deleptonizes due to the emission of neutrinos and antineutrinos of all the flavors in a scale of time of the order of 10 sec . the released energy is equally shared by all neutrino and antineutrino species . this phase is known as the kelving - helmholtz phase and ends up in the formation of a neutron star . during this time the radius of the protostar varies from around @xmath47 to @xmath48 , while the mass remains of the order of @xmath49 . after approximately @xmath50 minute the neutrino mean free path becomes comparable to the stellar radius and neutrino luminosity decreases quickly . another important property predicted for neutron stars is that they should have an extremely strong magnetic field . since the conductivity of the iron core is very large , the magnetic flux is essentially trapped and it is conserved as it collapses to form a neutron star@xcite . adopting for the initial magnetic field the typical values observed in some white dwarfs , from the flux conservation one can easily estimate that the surface field of a protoneutron star has to be in the range of @xmath51 . this is supported by evidence inferred from measurements of the spin - down rates and observations of the x - ray spectra features@xcite . magnetic fields of the order of @xmath52 or stronger can be generated by dynamo processes , and in the central regions intensities up to @xmath53 have been considered@xcite . in a protoneutron star the momentum flux from the core to the external regions is dominated by neutrinos , which provide the most efficient mechanism to release the energy generated during stellar collapse . as explained in the previous section , matter densities are so high that neutrinos are trapped forming a degenerated gas at quasi - equilibrium diffusing radially . the statistical description of this gas is made in terms of a distribution function @xmath54 . if the action of external fields is negligible , then the neutrino momentum between two consecutive collisions remains constant . the change of the distribution function is due to neutrino collisions with the particles in the medium and is described by the boltzmann equation @xmath55 where @xmath56 is the collision integral and @xmath57 when the regime is close to equilibrium we can approximate the collision integral as follows @xmath58 where @xmath59 is the average time between collisions and @xmath60 is the fermi - dirac distribution function for the gas of relativistic neutrinos ( @xmath61 ) with chemical potential @xmath62 and temperature @xmath63 . in the static situation ( @xmath64 ) , ( [ boltz ] ) with @xmath65 ) given by eq . ( [ zz ] ) can be solved iteratively . proceeding in this way we obtain @xmath66 which constitutes the diffusion approximation for the distribution function . the second term in this expression gives the correction to equilibrium and allows the transport of the energy and momentum towards the exterior of the protostar . once the distribution function is known the neutrino energy - momentum tensor can be computed as @xmath67 from this formula , for the energy density we have @xmath68}{2\pi^{2}}t^{4}\ , , \label{u}%\ ] ] where @xmath69 is the degeneracy parameter and @xmath70 $ ] is a fermi integral @xmath71=\int_{0}^{\infty}dxx^{n}[1+\exp(x-\eta_{\nu_{l}})]^{-1}.\ ] ] at each point within the protostar we choose a local frame ( @xmath72 ) with @xmath73 , such that @xmath74 . the stress tensor can be expressed in terms of the energy density as @xmath75 the equilibrium component of the distribution function is isotropic and gives a null contribution to the momentum flux ( @xmath76 @xmath77 ) , which is generated exclusively by the diffusive term : @xmath78 to complete the calculation of @xmath79 we need the mean free paths of the neutrinos or equivalently , the opacities @xmath80 . the neutral current reactions equally affect all flavors . since the energies available in the system are much smaller than the mass of the @xmath81 lepton , the @xmath16 does not participate in the charge - current interactions . for the same reason , and for simplicity , we also ignore the presence of muons . accordingly , the main sources of neutrino opacities are @xmath82 in terms of the cross sections for these reactions the @xmath83 are given by @xmath84 where @xmath85 are the neutron and proton number densities and @xmath86 \,,\\ \sigma_{abs } & = 4\,\sigma_{n}\ , . \label{abs}%\end{aligned}\ ] ] here , @xmath87 and @xmath88 is the renormalization for the axial - vector current of the nucleons . in eq . ( [ abs ] ) we neglected contributions of order @xmath89 , where @xmath90 and @xmath91 are the neutron and proton mass , respectively . from the previous formulae we immediately see that @xmath92 where the coefficients @xmath93 are constants . by making @xmath94 and taking the typical values @xmath95 and @xmath96 for the nucleon fractions @xmath97 , we get @xmath98 when eq . ( [ lll ] ) is substituted in the integrand of eq . ( [ mflux ] ) , the momentum flux transported by neutrinos diffusing from the core to the exterior of the protostar becomes @xmath99 \,t^{2}\right ) . \label{ff1}%\ ] ] in what follows we assume that the neutrino chemical potentials are negligible , and using @xmath100=\pi^{2}/12 $ ] and @xmath101=7\pi^{4}/120 $ ] , from eqs . ( [ u ] ) and ( [ ff1 ] ) , we obtain @xmath102 the last results will be used in section 6 , when we calculate the asymmetric neutrino emission in the context of a specific model for the protoneutron star . a notable consequence of neutrino mixing is the phenomena of neutrino oscillations established five decades ago by pontecorvo@xcite . it corresponds to the periodic variation in the flavor content of a neutrino beam as it evolves from the production point . in the last years compelling evidence of neutrino oscillations has been obtained in experiments with atmospheric , solar , accelerator , and reactor neutrinos@xcite . the most plausible interpretation for these new phenomena is that neutrinos are massive particles and the weak - interaction states @xmath103 ( @xmath104 ) are linear combinations of the mass eigenstates @xmath105 ( @xmath106 ) , with masses @xmath107 : @xmath108 where @xmath109 are the elements of a unitary mixing matrix , the analogue of the ckm matrix in the quark sector . consider a beam of @xmath110 created at a certain point with a 3-momentum @xmath111 . when the state given in eq . ( [ mix ] ) evolves , each of the mass eigenstates acquires a different phase @xmath112 , with @xmath113 ( @xmath114 ) . as a consequence , at a certain distance @xmath115 from the source the neutrino beam will not correspond to pure @xmath116 but turns partially into other flavors . the pattern of neutrino oscillations is modified when neutrinos propagate through a material medium . the basic reason is that the energy - momentum relation of an active neutrino is affected by its coherent interactions with the particles that constitute the medium . these modifications are described in terms of a potential energy @xmath117 , which is related to the real part of the refraction index @xmath118 according to @xmath119 and can be calculated from the background contributions to the neutrino self - energy@xcite . if , like in normal matter , electrons but not heavier leptons are present in the background , then only @xmath14 scatter through charge current interactions , while all the flavors interact via the neutral current . therefore , the refraction index differs for neutrinos of different flavors and they will acquire distinct phases when propagating with the same momentum through the medium . this fact can have impressive consequences in the case of mixed neutrinos and , under favorable conditions , flavor transformations are significantly enhanced in a medium with a varying density even when mixing in vacuum is small . this is the essence of the msw mechanism@xcite whose large mixing realization provides the solution to the solar neutrino problem@xcite . besides the modifications of the dispersion relation , neutrinos acquire in matter an effective coupling to the electromagnetic field by means of their weak interaction with the background particles@xcite . as a result , in the presence of an external magnetic field there are additional contributions to the neutrino refraction index and flavor transformations are affected , but in a way that preserves quirality@xcite , contrary to what happens for neutrino oscillations driven by transition magnetic moments@xcite . in what follows , we will discuss neutrino oscillations in a magnetized medium for the simplest situation of mixing between two neutrino species , that is , @xmath116 and another active neutrino @xmath120 or a hypothetical sterile neutrino @xmath121 . in any case we can write @xmath122{cc}% \text{cos } \theta & \text{sin } \theta\\ -\text{sin } \theta & \text{cos } \theta \end{array } \right ) , \ ] ] where @xmath123 is the vacuum mixing angle , a parameter to be fixed from the experimental results . the dirac wave function for relativistic ( mixed ) neutrinos with momentum @xmath111 propagating in matter in the presence of a static uniform magnetic field is well approximated as follows@xcite : @xmath124{c}% 0\\ \phi_{-}% \end{array } \right ) \chi(t)\ , , \label{diracwf}%\ ] ] where we use the weyl representation , @xmath125 is the pauli spinor of negative helicity ( @xmath126 ) and @xmath127 is a flavor - space spinor that satisfies @xmath128 the hamiltonian @xmath129 is a @xmath130 matrix that , in the basis of the flavor states , takes the form @xmath131 with @xmath132{cc}% m_{1 } & 0\\ 0 & m_{2}% \end{array } \right ) \mathbb{u}^{\dagger}%\ ] ] and @xmath133 . the effective potential for active neutrinos can be written as@xcite @xmath134 where @xmath135 is the proton charge . the coefficients @xmath136 and @xmath137 depend on the properties of the thermal background . to order @xmath138 the expression for @xmath136 is independent of @xmath139 , but this is not true in general for @xmath137 . in our analysis we will work within the weak - field limit , i.e. , @xmath140@xcite , and ignore any possible dependence of this coefficient on the magnetic field . the effect of strong magnetic fields on the neutrino propagation in a medium have been considered by some authors@xcite . for ( active ) antineutrinos the potentials change in sign , i.e. , @xmath141 , while in the case of sterile neutrinos , since they do not interact with the background , we have @xmath142 eq . ( [ wolfeq ] ) is the extension of the msw equation to the situation we are considering , and was first derived in ref . the components @xmath143 and @xmath144 of @xmath127 give the amplitude to find the neutrino in the corresponding flavor state . for a uniform background @xmath127 = @xmath145 , where @xmath146 are the energies of the neutrino modes in the medium and the vectors @xmath147 are the solutions of the eigenvalue condition @xmath148 . for oscillations the relevant quantity is @xmath149 where @xmath150 and @xmath151 . when the medium is inhomogenous , along the neutrino path ( @xmath115 ) the elements of @xmath152 are functions of @xmath153 , and @xmath127 has to be determined by solving eq . ( [ wolfeq ] ) with @xmath129 a time dependent matrix . at each point @xmath154 , @xmath129 can be diagonalized by the unitary transformation @xmath155{cc}% \cos\theta_{m } & \sin\theta_{m}\\ -\sin\theta_{m } & \cos\theta_{m}% \end{array } \right ) \,,\label{umatter}%\ ] ] with the mixing angle in matter @xmath156 given by @xmath157 a similar expression is valid in the case of antineutrinos with @xmath158 in the denominator . in general @xmath159 and consequently the mixing angle in matter depends on the magnetic field . as a function of @xmath159 , @xmath160 exhibits the characteristic form of a breit - wigner resonance . for neutrinos ( antineutrinos ) with momentum @xmath111 the resonance condition ( @xmath161 ) is @xmath162 where @xmath163 and the sign @xmath164 ( @xmath165 ) corresponds to neutrinos ( antineutrinos ) . notice that for active - active oscillations the neutral current contributions to @xmath166 cancel since they are the same for both flavors . this is not true in the case of active - sterile oscillations where @xmath167 , and such contributions play a relevant role in the resonant transformations for @xmath168 ( or @xmath169).the efficiency of the flavor transformation depends on the adiabaticity of the process . the conversion will be adiabatic when the oscillation length at the resonance is smaller than the width of the enhancement region . this imposes the condition@xmath170 the dispersion relation of neutrinos in a magnetized plasma made of electrons , protons , and nucleons have been calculated in ref . @xcite , including the contributions due to the anomalous magnetic moment coupling of the nucleons to the photon . from the results given there we have @xmath171 + \tilde{b}_{_{e}},\\ b_{\mu,\tau } & = \sqrt{2}g_{f}\left [ \left ( \frac{1}{2}-2\sin^{2}\theta _ { w}\right ) \left ( n_{p}-n_{e}\right ) -\frac{1}{2}n_{n}\right ] + \tilde { b}_{_{_{\mu,\tau}}},\\ c_{e } & = 2\sqrt{2}g_{f}\left [ g_{a}\left ( 1+\,2m_{p}\frac{\kappa_{p}}% { e}\right ) c_{p}-g_{a}2m_{n}\frac{\kappa_{n}}{e}c_{n}-c_{e}\right ] , \\ c_{\mu,\tau } & = 2\sqrt{2}g_{f}\left [ g_{a}\left ( 1+\,2m_{p}\frac{\kappa _ { p}}{e}\right ) c_{p}-g_{a}2m_{n}\frac{\kappa_{n}}{e}c_{n}+c_{e}\right ] , \end{aligned}\ ] ] where @xmath172 are the anomalous part of the nucleon magnetic moments given by @xmath173 and @xmath174 \,,\\ c_{\flat } & = \frac{1}{2}\int\frac{d^{3}p}{(2\pi)^{3}}\frac{1}{2e}\frac { d}{de}f_{\flat}\ \ \ \ ( \flat = e , n , p)\,.\end{aligned}\ ] ] in these equations @xmath175 and @xmath176(@xmath177 ) are the number densities of electrons and neutrinos ( antineutrinos ) of flavor @xmath178 , respectively and @xmath179 are the distribution functions of the electrons , protons , and neutrons . we model the atmosphere of a protoneutron star as a neutrino gas diffusing through a plasma made of relativistic degenerate electrons and classical nonrelativistic nucleons . in such conditions @xmath180 and @xmath181@xcite , where @xmath63 is the background temperature and @xmath182 is the electron chemical potential . the neutrino fractions are much smaller than the electron fraction @xmath183 @xmath184 and , in the context of our analysis , the contributions to @xmath185 coming from the neutrino - neutrino interactions ( denoted by @xmath186 ) can be neglected . with this approximation in mind , putting @xmath187 , and taking into account that by electrical neutrality @xmath188 @xmath189 , the coefficients in the effective potentials become : @xmath190 it should be noticed that in ref . @xcite @xmath191 vanishes because all the components of the stellar plasma are assumed to be degenerated , and the contributions due to the anomalous magnetic moment of the nucleons are not included . we now use the effective potentials @xmath117 with @xmath136 and @xmath137 given by eqs . ( [ bn])-([cn ] ) to examine the magnetic field effect on the resonant neutrino oscillations within a protoneutron star . in a linear approximation , we can write @xmath192 where @xmath193 and @xmath194 are ordinary functions of the magnitude of the neutrino momentum and @xmath195 is the angle between @xmath196 and @xmath197 . for a certain value of @xmath198 , eq . ( [ rr ] ) determines a spherical shell limited by the spheres of radii @xmath199 , where the resonance condition is verified for neutrinos ( or antineutrinos ) moving in different directions with respect to the magnetic field . the quantity @xmath193 corresponds to the radius of the resonance sphere when @xmath200 and in the case of oscillations between @xmath14 and @xmath15 ( or @xmath16 ) is given by @xmath201 the left - hand side corresponds to the difference @xmath202 evaluated at @xmath203 and is positive definite . therefore , assuming @xmath204 , the above condition can be satisfied for @xmath205 but not for @xmath206 . to be effective as a kick mechanism the resonance transformations of active neutrinos have to take place in the region between the electron and muon ( tau ) neutrinospheres in this case , the @xmath14 are trapped by the medium , but the @xmath207 produced this way are above their neutrinosphere and , when moving outside , can freely escape from the star . in the presence of a magnetic field , the emission points for @xmath207 having the same @xmath198 but different directions are not at the same radius , which originates an asymmetry in the momentum they carried away . replacing @xmath198 in eq . ( [ recon ] ) by the thermal average @xmath208 and taking into account that @xmath209 and @xmath210 at @xmath46 , the requirement of having the resonance within the electron neutrinosphere implies @xmath211 , which is excluded by the experimental results on neutrino oscillations and the cosmological limits . for this reason , in what follows we concentrate only on the active - sterile oscillations . in this case , the resonance condition when @xmath200 takes the form @xmath212 for @xmath168 ( @xmath213 ) . in the static model we use for the protoneutron star , a good approximation is to take @xmath214 . therefore , if @xmath204 , the resonant condition is verified only by antineutrinos and from now on we restrict ourselves to this situation . to determine the quantity @xmath194 in eq . ( [ rr ] ) we substitute @xmath117 as given by eq . ( [ v2 ] ) in eq . ( [ deltam ] ) and expand @xmath136 up to first order in @xmath215 . proceeding in this way we get @xmath216 with @xmath217 where we have defined @xmath218 for any function @xmath219 . using eqs . ( [ bn])-([cn ] ) we find explicitly @xmath220 finally from eq . ( [ adcon ] ) , discarding terms proportional to the magnetic field , the following restriction to the vacuum mixing angle results @xmath221 which guarantees that the evolution will be adiabatic and an almost complete flavor transformation for small values of @xmath123 . if the temperature and density profiles are isotropic , then the momentum flux in the protoneutron star atmosphere is radial ( see eq . ( [ ff2 ] ) ) . as a consequence , in most of the works on the kick mechanism driven by matter neutrino oscillations the resonance condition was evaluated assuming that neutrinos move outside with a momentum @xmath222 pointing in the radial direction . in presence of a magnetic field this approach leads to the concept of a deformed resonance surface , which acts as an effective neutrinosphere . however , at each point within the neutrinosphere there are neutrinos with momentum @xmath111 pointing in every directions and the resonance condition as given by eq . ( [ deltam ] ) actually defines a spherical surface for each value of @xmath223 . this surface acts as a source of sterile neutrinos produced through the resonant @xmath224 transformations , which move in the @xmath225 direction . therefore , as was mentioned above , for a given @xmath198 the resonant transformations take place within a spherical shell , and not on a deformed spherical surface . a more careful calculation of the pulsar kick along this line has been carried out in ref . @xcite and here we follow the same approach . let us consider the situation where the resonance shell is totally contained within the @xmath224 neutrinosphere . the sterile neutrinos produced there that move out freely escape from the protostar , but those directed toward the inside cross the resonance surface again and reconvert into @xmath224 which , being within their own neutrinosphere , are thermalized . consequently , only those @xmath226 going outward can leave the star and the resonance surfaces for different momentum directions behave as effective emission semispheres of sterile antineutrinos . for @xmath111 pointing in opposite directions the radii of the respective semispheres are @xmath227 , and therefore they have different areas that generate a difference in the momentum carried away by the sterile neutrinos leaving the star in opposite directions . there also exists a temperature variation within the resonance shell such that the temperatures of the emission semispheres for @xmath226 going outside in opposite directions are different ( @xmath228 ) . the variations in the area and the temperature tend to compensate each other when the contributions of @xmath226 emitted in every direction are added up . in the calculation below we take both semispheres into account when explicitly computing the asymmetry in the total momentum @xmath229 emitted by the cooling protoneutron star . in this way , we improve the calculation of ref . @xcite where the temperature within the resonance shell was assumed to be uniform . the active neutrinos and antineutrinos are emitted isotropically from their respective neutrinospheres and do not contribute to the momentum asymmetry @xmath230 . therefore , the only nonvanishing contribution can come from the @xmath226 . if their emission lasts for an interval @xmath231 of the order of a few seconds , then in the static model we are using for the protostar @xmath232 , where @xmath233 is the component along the magnetic field direction of the neutrino momentum emitted per time unit . according to eq . ( [ mflux ] ) we can write @xmath234 where @xmath235 and the distribution function for sterile neutrinos @xmath236 is evaluated at @xmath237 . we have chosen a reference frame ( @xmath238 , @xmath239 , @xmath240 ) fixed to the protostar where the magnetic field coincides with the @xmath240-axis , @xmath241 . in addition , at each point within the resonance shell we use the local frame introduced in section 2 to evaluate the components of the energy - momentum tensor . here , the upper limit for the integration on @xmath242 is @xmath243 instead of @xmath244 , because we have to include only the contributions from the sterile neutrinos going outside . to proceed further with the calculation of @xmath233 we need to know @xmath236 . the proper treatment of the interplay of collisions and oscillations for trapped neutrinos requires the use of the density matrix . for our purpose a less elaborate description will suffice and we put @xmath245 , where @xmath246 is the probability for @xmath224 to convert into @xmath247 . let us assume that the adiabaticity condition given by eq . ( [ adcon2 ] ) is verified in the interior of the @xmath224 neutrinosphere , and hence for small mixing @xmath248 . accordingly in eq . ( [ kb ] ) we take @xmath249 , with @xmath250 where a term proportional to the derivative of @xmath251 has been neglected , in agreement with the diffusive approximation we are employing . substituting the expansion ( [ expf ] ) into eq . ( [ kb ] ) and keeping terms that are at most linear in @xmath215 we find @xmath252 the energy released during the core collapse of the progenitor star is approximately equipartitioned among all the neutrino and antineutrino flavors . in the context of our analysis this means that @xmath253 , where @xmath254 represents the momentum rate of the sterile neutrinos in the radial direction and is given by an expression similar to eq . ( [ kb ] ) with @xmath255 replaced by @xmath256 . following the same procedure outlined above we find @xmath257 in terms of @xmath233 and @xmath254 the fractional asymmetry in the total momentum neutrinos carry away becomes @xmath258 to evaluate the remaining integrals in eqs . ( [ er ] ) and ( [ eb ] ) the explicit dependence on @xmath198 of the functions @xmath259 and @xmath260 have to be known . simple analytical results can be derived by replacing these functions by the constant quantity @xmath261 and @xmath262 , where @xmath261 is determined from the resonance condition when @xmath198 is replaced by its thermal average @xmath263 : @xmath264 proceeding in this manner we get @xmath265 in these expressions the terms that depend on @xmath266 come from the diffusive part of the distribution function . in the regime considered here they are smaller than the other contributions and in what follows we will neglect them . suppose that the pulsar momentum @xmath267 is entirely due to the active - sterile oscillation driven kick , then @xmath268 . for a pulsar with a mass @xmath269 and a translational velocity @xmath270 we have @xmath271 . since neutrinos are relativistic , the total amount of momentum they carry is equal to the gravitational energy liberated by means of them , that is @xmath272 . therefore , as mentioned in the introduction , @xmath273 must be of the order of 0.01 to obtain the required kick . in this case eq . ( [ asym ] ) yields @xmath274 with @xmath275 denoting the logarithmic derivative of @xmath276 evaluated at @xmath277 . this formula allows us to determine the magnitude of the magnetic field once the temperature , the baryon density , and the electron fraction profiles are known . below we do it for the @xmath278 conversion under the assumption that the resonance layer is entirely within the @xmath279-sphere . however , we will first indicate how @xmath280 is determined by means of @xmath276 and @xmath281 within our approximate description of a protoneutron star . for @xmath282 , the @xmath31 equilibrium gives the relation @xmath283 among the chemical potentials of the background fermions . hence , for non - relativistic nucleons @xmath284 $ ] and @xmath285 where we used the fact that in the condition prevailing in a protoneutron star @xmath286 is much larger than the difference between the neutron and the proton masses . since electrons are degenerated @xmath287 and eq . ( [ ye ] ) is an implicit equation for @xmath288 to be solved numerically once @xmath289 and @xmath63 are known as a function of @xmath154 . taking the derivative of eq . ( [ ye ] ) with respect to @xmath154 , @xmath290 can be expressed in terms of @xmath291 and @xmath292 as follows @xmath293 to estimate @xmath139 we now assume that the density and the temperature are related according to @xmath294@xcite , where @xmath295 and @xmath296 are the respective values of these quantities at the core radius @xmath297 . from this expression we see that @xmath298 which implies that @xmath299 ( see eq . ( [ hy ] ) ) . therefore , the electron fraction @xmath288 and as a consequence the ratio @xmath300 are constants . as core parameters we take @xmath301 , @xmath302 , and @xmath303 , while for the density profile we adopt the potential law @xmath304 which yields @xmath305 and @xmath306 . the @xmath307-sphere is taken stationary at a density of about @xmath1 that corresponds to a radius @xmath308 and a temperature @xmath309 . using these results as well as the formula for @xmath310 given in eq . ( [ d2 ] ) , from eq . ( [ momas ] ) we obtain the following simple result : @xmath311 where @xmath312 , with @xmath313 . we see that the required intensity of the magnetic field decreases monotonically from @xmath314 at the core radius to @xmath315 at the @xmath279-sphere . according to eqs . ( [ recons1 ] ) and ( [ bn2 ] ) , within the specified interval for @xmath316 the range of the allowed values of the oscillation parameters are @xmath317 , and for small mixing we have @xmath318 @xmath319 . the requirement that the resonant conversion be adiabatic imposes the condition @xmath320 on the mixing angle . a last comment is in order , the ratio of the diffusive term to the isotropic one in eqs . ( [ erad ] ) and ( [ brad ] ) can be expressed as @xmath321 , which illustrates the validity of the approximation done in writing eq . ( [ momas ] ) . in the next section we repeat these calculations in the realm of the spherical eddington model , which provides a simple but quite physically reasonable description of a static neutrino atmosphere . the actual configuration of the neutrinosphere and the resonance region depend on details of the protoneutron star , such as the density , temperature , pressure , and leptonic fraction profiles . the asymmetry in the momentum flux also depends on these features and on the configuration of the magnetic field . the present knowledge about protoneutron stars is not enough to single out a definite model for their structure , but there are some well established general characteristics that help towards proposing models that are both workable and plausible . an important feature is that the size and the shape of the different profiles do not suffer sudden changes during the time interval when the main neutrino emission takes place , i.e. @xmath322 . thus , as a first approximation we can consider a stationary regime that greatly simplifies calculations . another characteristic to be taken into account is that the bulk of neutrinos is produced in the inner core . consequently , we assume that there is no neutrino production outside the core and hence the flux of diffusing neutrinos is conserved in this region . this flux is responsible for the transport of the energy liberated by the protostar . as mentioned before , the system is supposed to be constituted by nucleons , with a density @xmath289 , electrons , neutrinos and antineutrinos . neutrinos are in thermal equilibrium with nucleons , satisfying a transport regime consistent with the diffusion approximation , and are assumed to have a vanishing chemical potential , @xmath323 . the total energy density , the radial energy flux , and the pressure for neutrinos and antineutrinos in the case of spherical symmetry , can be obtained from eqs . ( [ u1 ] ) , ( [ ff2 ] ) and ( [ ttt])@xmath324 where @xmath325 . the main contributions to the neutrino opacities were discussed in section iii . using the results given there for @xmath326 and @xmath327 ( eqs . ( [ ce1 ] ) and ( [ ce2 ] ) ) we obtain @xmath328 . if in addition we assume that baryons constitute a nonrelativistic ideal gas , we have the eddington model . it is simple and physically well justified , allowing a detailed discussion of the relevant characteristics of the protostar and the geometry of the resonance region . originally proposed to describe a stellar photosphere , this model was adapted by schinder and shapiro@xcite to a neutrino atmosphere with a plane geometry . the more realistic case of a spherical atmosphere was considered in ref . @xcite to analyze the geometrical effect on the asymmetric neutrino emission induced by resonant oscillations . photons and electrons are of course present and , in fact , electrons make the leading contribution to the effective potential in the case of oscillations among active neutrinos . however , we can ignore both photons and electrons for the hydrodynamical description of the system . the state equation that defines the system is given by the sum of the contributions from nucleons and neutrinos to the pressure gas , which is @xmath329 in the newtonian limit for the metric ( @xmath330 , @xmath331 , and @xmath332 , with @xmath333 ) the conservation of the total ( neutrinos+matter ) energy - momentum tensor yields the energy flux conservation @xmath334 and the hydrostatic equilibrium equation @xmath335 here , @xmath336 is the mass enclosed up to a distance @xmath154 from the center . through the atmosphere the baryon density is @xmath337 and @xmath338 , and thus we have @xmath339 and @xmath340 . this means that on the right - hand side of eqs . ( [ eqst ] ) and ( [ gradp ] ) we can ignore the contributions coming from the energy density of neutrinos and the baryonic density . in addition , eq . ( [ consf ] ) implies that @xmath341 , where @xmath342 is the luminosity of the protostar . taking all the above into account we arrive at the following set of equations for an isotropic atmosphere of neutrinos in thermal equilibrium with an ideal gas of nucleons : @xmath343 solving this system of equations yields the profiles of four functions throughout the atmosphere : pressure @xmath344 , temperature @xmath276 , baryonic density @xmath345 , and the enclosed mass @xmath346 . from them any other functions of interest can be calculated . to analyze the system it is convenient to introduce adimensional variables normalized to the corresponding values at the core : @xmath347 , @xmath348 , @xmath349 , @xmath350 , and @xmath351 . proceeding in this way , eqs . ( [ m2])-([m5 ] ) can be rewritten @xmath352 where @xmath353 are adimensional constants . thus the set of functions that solve the previous system of equations depends only on these three combinations of the core parameters . by eliminating @xmath354 in eqs . ( [ a1 ] ) and ( [ a3 ] ) we arrive to @xmath355 the last equation is easily integrated as follows : @xmath356 where @xmath357 and @xmath358 is an effective mass defined by @xmath359 from eqs . ( [ a2 ] ) and ( [ p ] ) we can also express the density in terms of the temperature as @xmath360 which , together with eq . ( [ a1 ] ) , yields a first order differential equation for @xmath153 @xmath361 at @xmath362 ( @xmath363 ) we have @xmath364 , which is independent of @xmath365 . for an idealized atmosphere , where the model would apply to the whole space , the physical solutions would correspond to an infinite protostar where the temperature has an asymptotic behavior for @xmath366 , such that the adimensional temperature tends to @xmath367 . thus , the function @xmath365 varies in the range @xmath368 for @xmath369 . ( [ dif ] ) has no analytical solution when @xmath370 is a function of @xmath238 . an approximate solution can be found by the following procedure . let us consider the differential equation ( [ dif ] ) with @xmath370 constant . then , an analytical ( implicit ) solution is given by @xmath371 = \frac{b_{c } } { 1-a}\left(\frac{1}{x}-1\right)\ , . \label{tem}\ ] ] with @xmath372 . if we replace the constant @xmath370 by a well behaved function of @xmath238 , the above expression still satisfies eq . ( [ dif ] ) at @xmath362 . for an infinite atmosphere , a good approximation to the exact solution is given by eq . ( [ tem ] ) , with @xmath370 now a function of @xmath373 : @xmath374 where @xmath375 . from eq . ( [ at ] ) , we see that @xmath376 and @xmath377 , which means that this ansatz fits the extreme values of @xmath365 . once @xmath289 and @xmath63 are known , we can find the electron fraction @xmath378 as a function of @xmath154 by solving the implicit expression derived in eq . ( [ ye ] ) . to examine the prediction of the model in connection with the kick mechanism , we consider a protoneutron star with reasonable values for its core parameters , the same as in ref . @xcite . thus we take @xmath379 , @xmath380 , @xmath381 , @xmath382 , and @xmath383 . in this case , @xmath384 a numerical estimation for the asymptotic value of the solution of eqs . ( [ a1])-([a4 ] ) gives @xmath385 . the electron fraction takes the value @xmath386 at the core . according to eq . ( [ lll ] ) in terms of the adimensional functions the mean free path for the electron neutrino becomes @xmath387 inserting the above expression in the integrand of eq . ( [ meanfp ] ) and evaluating the resulting integral numerically we obtain @xmath388 for the radius of the @xmath389-sphere . this yields @xmath390 , @xmath391 , and @xmath392 , which are in reasonable agreement with typical values from numerical simulations@xcite . in the same way , for the muon ( tau ) neutrino we get @xmath393 . the profiles that correspond to the solution of this model are given in fig . 1 . from eq . ( [ momas ] ) we can estimate again the intensity of the magnetic field required for the kick in the case of @xmath394 . besides the temperature , density , and electron fraction profiles we also need their logarithmic derivatives . for @xmath63 and @xmath289 they are immediately calculated from eqs . ( [ a1])-([a3 ] ) , and the corresponding results are @xmath395 the formula for @xmath396 is obtained substituting the above expressions into eq . ( [ hy ] ) . now we have all the quantities needed to determine the magnetic field from eq . ( [ momas ] ) in the spherical eddington model . the result is plotted in fig . 2 . from the curve presented in this figure we see that @xmath139 decreases from a value of @xmath397 in the core up to a minimum of @xmath398 at @xmath399 . from this point the magnetic field increases steadily as we approach the surface of the neutrinosphere . in contrast with the model used at the end of section 5 , here at @xmath400 the contributions from the geometrical and the temperature variations compensate each other . at this radius the factor @xmath401 in the denominator of eq . ( [ momas ] ) vanishes and the magnetic field takes arbitrarily large values . in other words , no kick can be generated by this mechanism in the regions near the surface of the @xmath15-sphere . in the region between the core and the surface of the neutrinosphere @xmath402 , the range of values allowed for the oscillation parameters is @xmath403 for a small mixing angle this inequality translates into the condition @xmath404 for the sterile neutrino mass . small mixing is allowed by the adiabaticity condition , which warrants an efficient flavor neutrino conversion for @xmath405 at @xmath406 and @xmath407 at @xmath408 . the results for @xmath409 and the mixing angle agree with those derived in the previous section and are compatible with a sterile neutrino that is a viable dark matter candidate@xcite . in this work we have examined in detail a possible explanation of the large drift velocities observed in pulsars in terms of an asymmetric neutrino emission . the asymmetry is a consequence of the resonant neutrino conversions affected by the strong magnetic field characteristic of protoneutron stars . we have shown that for active - sterile neutrino oscillations this is a feasible kick mechanism . two conditions must be simultaneously satisfied to generate a natal kick by this mechanism . first , the conversion has to take place at the interior of the neutrinosphere of the active neutrino , and second the magnetic field has to be intense enough in order to induce a momentum asymmetry of the required magnitude . we have made detailed calculations by means of the neutrino distribution function in the diffusion approximation , combined with the idea of a spherical resonance shell for neutrinos with momentum @xmath198 that move in different directions relative to @xmath410 . this scheme provides a better description of the problem than the one formulated in terms of a single deformed sphere acting as an effective emission surface of sterile neutrinos . in our approach the sterile neutrinos produced in the resonance shell move in every direction relative to @xmath197 . however , those going toward the interior of the protoneutron star cross again the resonance region and are reconverted into active neutrinos . these are within their own neutrinosphere and become thermalized . as a consequence , only the outgoing sterile neutrinos contribute to the total momentum emitted by the protostar . two opposite effects have to be taken into account when computing the fractional momentum asymmetry . one is purely geometric , and comes from the difference in the areas of the semispherical emission surface for neutrinos moving in opposite directions with the same momentum magnitude . the other is due to the radial temperature gradient , by which sterile neutrinos produced at different depths have unequal energy . in general both contributions do not compensate and thus we obtain a non null fractional asymmetry in the total momentum . explicit results have been obtained for two simple models of a protostar , with the matter background assumed to be composed by nonrelativistic nucleons and degenerated electrons . one of them , which we worked with in detail , is the eddington model for a spherical neutrino atmosphere . magnetic fields of the order of @xmath411 are needed to reproduce the observed pulsar velocities . at first sight these fields could seem rather large , compared with the estimated values at the surface of a protoneutron star ( @xmath412 ) , but in fact intensities as high as @xmath413 are possible at the core . a given parametrization for the magnetic field at the interior of a protoneutron star is@xcite @xmath414,\ ] ] with @xmath415 and @xmath416 . by adopting this profile as an upper boundary for the magnetic field in the protostar , we can see that the required field remains well below this boundary in most of the region within the neutrinosphere . in addition , the oscillation parameters are compatible with the allowed region for sterile neutrinos to be warm dark matter , leaving the mechanism as an attractive possibility to explain the proper motion of pulsars . this work was partially supported by conicet - argentina , conacyt- mxico , and dgapa - unam ( mxico ) under grant papiit - in109001 . 99 a. hewish , s. j. bell , j. d. h. pikington , p. f. scott , and r. a. collons , nature * 217 * , 709 ( 1968 ) . f. pacini , nature * 216 * , 567 ( 1966 ) ; t. gold , nature * 218 * , 731 ( 1966 ) . s. l. shapiro and s. a. teukolsky , _ black holes , white dwarfs and neutron stars : the physics of compact objects _ ( wiley , 1983 ) ; n. k. glendenning , compact stars : nuclear physics , particle physics , and general relativity , 2nd edition ( springer - verlag , 2000 ) . g. g. raffelt , _ stars as laboratories for fundamental physics _ ( the university of chicago press , chicago , 1996 ) ; h. suzuki , in _ physics and astrophysics of neutrinos _ , edited by m. fukugita and a. suszuki ( springer - verlag , 1994 ) . a. g. lyne and d. r. lorimer , nature * 369 * , 127 ( 1994 ) ; d. r. lorimer , m. bailes , and p. a. harrison , mon . not . soc . * 289 * , 592 ( 1997 ) ; b. m. s. hansen and e. s. phinney , mon . not . * 291 * , 569 ( 1997 ) ; j. m. cordes and d. f. chernoff , ap . j. * 505 * , 315 ( 1998 ) ; c. fryer , a. burrows , and w. benz , ap . j. * 496 * , 333 ( 1998 ) . z. arzoumanian , d. f. chernoff , and j. m. cordes , ap . j. * 568 * , 289 ( 2002 ) . see d. lai , d. f. chernoff , and j. m. cordes , ap . j * 549 * , 1111 ( 2001 ) and references therein . see , for example , d. lai , _ neutron star kicks and supernova asymmetry _ , to appear in the proceedings of 3-d signatures in stellar explosions : a workshop honoring j. craig wheeler s 60th birthday , austin , texas , 10 - 13 june 2003 , astro - ph/0312542 , and references therein . a. burrows and j. hayes , phys . lett . * 76 * , 352 ( 1996 ) . m. herant , w. benz , w. r. hix , c. l. fryer . , and s. a. colgate , ap . j. * 435 * , 339 ( 1994 ) ; a. burrows , j. hayes , b. a. fryxell , ap . j. 450 , 830 ( 1995 ) ; h - th . janka and e. muller , astron . and astrophys . * 290 * , 496 ( 1994 ) ; _ ibid . _ * 306 * , 167 ( 1996 ) ; l. scheck , t. plewa , h .- th . janka , k. kifonidis and e. mller , phys . lett . * 92 * , 011103 ( 2004 ) . c. l. fryer , ap . j. * 601 * , l175-l178 ( 2004 ) . n. n. chugai , sov . 10 * , 87 ( 1984 ) ; o. f. dorofeev _ et al_. sov . . lett . * 11 * , 123 ( 1985 ) ; a. vilenkin ap . j. * 451 * , 700 ( 1995 ) ; c. j. horowitz and g. li , phys . lett . * 80 * , 3694 ( 1998 ) ; a. kusenko , g. segr , and a. vilenkin phys . lett . * b437 * , 359 ( 1998 ) ; p. arras and d. lai , phys . rev . * d60 * , 043001 ( 1999 ) . g. s. bisnovatyi - kogan , astron . trans . * 3 * , 287 ( 1993 ) ; d. lai and y .- z . quian , ap . j. * 505 * , 844 ( 1998 ) ; e. roulet , j. high energy phys . * 01 * , 013 ( 1998 ) . r. c. duncan and c. thompson , ap . j. * 392 * , l9 ( 1992 ) . v. b. semikoz , sov . . phys . * 46 * , 946 ( 1987 ) ; j. c. dolivo , j. f. nieves , and p. b. pal , phys . rev . * d40 * , 3679 ( 1989 ) . j. c. dolivo and j. f. nieves , phys . rev . * d56 * , 5898 ( 1997 ) . a. kusenko and g. segr , phys . lett . * 77 * , 4872 ( 1996 ) ; _ ibid . _ * 79 * , 2751 ( 1997 ) ; phys . rev . * d59 * , 061302 ( 1999 ) . y. z. qian , phys . lett . * 79 * , 2750 ( 1997 ) , c. w. kim , j. d. kim , and j. song , phys . b419 * , 279 ( 1998 ) . m. barkovich , j. c. dolivo , r. montemayor , and j. f. zanella , phys . rev . * d66 * , 123005 ( 2002 ) . a. kusenko and g. segr , phys . lett . * b396 * , 197 ( 1997 ) . g. m. fuller , a. kusenko , i. mocioiu , and s. pascoli , phys . rev . * d68 * , 103002 ( 2003 ) ; a. kusenko , int . j. mod d13 * , 2065 ( 2004 ) . m. barkovich , j. c. dolivo , and r. montemayor , phys . rev . * d70 * , 043005 ( 2004 ) . for a recent review see m. cirelli , astro - ph/0410122 , to appear in the proceedings of 10th international symposium on particles , strings and cosmology ( pascos 04 and pran nath fest ) , boston , massachusetts , 16 - 22 august 2004 . e. k. akhmedov , a. lanza , and d. w. sciama , phys . rev . * d56 * , 6117 ( 1997 ) ; e. nardi and j. i. zuluaga , ap . j. * 549 * , 1076 ( 2001 ) ; g. lambiase , astro - ph/0411242 ; g. lambiase , astro - ph/0412408 . d. grasso , h. nunokawa , and j. w. f. valle , phys . lett . * 81 * , 2412 ( 1998 ) . r. horvat , mod . a13 * , 2379 ( 1998 ) ; m. barkovich , h. casini , j. c. dolivo , and r. montemayor , phys . lett . * b506 * , 20 ( 2001 ) . h. a. bethe , rev phys * 62 * , 801 ( 1990 ) . p. goldreich and s. v. weber , astrophys . j. * 238 * , 991 ( 1980 ) . d. z. freedman , phys . d 9 * , 1389 ( 1974 ) . t. j. mazurek , nature * 252 * , 287 ( 1974 ) ; j. cooperstein , phys . 163 * , 95 ( 1988 ) . s. a. colgate and m. h. johnson , phys . lett . * 5 * , 235 ( 1960 ) . a. burrows and j. lattimer , ap . j. * 307 * , 178 ( 1986 ) . a. burrows and t. j. mazurek , ap . j. * 259 * , 330 ( 1982 ) . p. mszros , _ high - energy radiation from magnetized neutron stars _ ( the university of chicago press , 1992 ) . v. urpin and j. gil , astron . astrophys . * 415 * , 305 ( 2004 ) . d. bandyopadhyay , s. chakrabarty and s. pal , phys . * 79 * , 2176 ( 1997 ) . r. manka , m. zastawny - kubika , a. brzezina , i. bednarek , astro - ph/0202303 . b. pontecorvo , zh . . fiz . * 33 * , 549 ( 1957 ) ; _ _ ibid.__**34 * * , 247 ( 1958 ) ; z. maki , m. nakagawa , and s. sakata , prog . * 28 * , 870 ( 1962 ) . r. n. mohapatra and p. pal , _ massive neutrinos in physics and astrophysics _ , 3rd . edition ( world scientific , 2004 ) . y. hayato , the super - kamiokande and k2k collaborations , eur . j. * c33 * , s829 ( 2004 ) . a. b. mc donald , nucl - ex/0412005 . d. notzold and g. raffelt , nucl . phys . * b307 * , 924 ( 1988 ) ; j. c. dolivo , j. f. nieves , and m. torres , phys . rev . * d46 * , 1172 ( 1992 ) . l.wolfenstein , phys . rev * d17 * , 2369 ( 1978 ) ; s. p. mikheyev y a. yu . smirnov , nuovo cim . * 9c * , 17 ( 1986 ) . a. yu . smirnov , hep - ph/0412391 . j. c. dolivo and j. f. nieves , phys . b383 * , 87 ( 1996 ) . e. kh.akhmedov , phys . lett . * b213 * , 64 ( 1988 ) ; c. -s . lim and w. j. marciano , phys . rev . * d37 * , 1368 ( 1988 ) ; m. m. guzzo and h. nonokawa , astrop . * 11 * , 317 ( 1999 ) ; o. g. miranda _ et al . _ , b595 * , 360 ( 2001 ) . h. nunokawa , v. b. semikoz , a. yu . smirnov , and j. w. f. valle , nucl . b501 * , 17 ( 1997 ) . v. b. semikoz and j. w. f. valle , nucl . phys . * b425 * , 651 ( 1994 ) ; erratum nucl . phys . * b485 * , 651 ( 1997 ) ; p. elmfors , d. grasso , and g. raffelt , _ ibid . _ * b479 * , 3 ( 1996 ) ; a. erdas , c. w. kim , and t. h. lee , phys . rev . * d58 * , 085016 ( 1998 ) . p. j. schinder and s. l. shapiro , ap.j . * 259 * , 311 ( 1982 ) . a. d. dolgov and s. h. hansen , astroparticle physics * 16 * , 339 ( 2001 ) . 1 : the different profiles normalized to the corresponding value at the core for the eddington model discussed in section vi . 0.5 cm fig . 2 : the magnetic field @xmath139 required to produce the kick by the @xmath417 resonant conversion for the eddington model discussed in section vi , normalized to the field required at the core , @xmath418 .

this work identifies geometric effects on dynamics due to nonadiabatic couplings in born oppenheimer systems and provides a systematic method for deriving corrections to mixed quantum - classical methods . specifically , an exact path integral formulation of the quantum nonadiabatic dynamics of born oppenheimer systems is described . stationary phase approximations to the propagator for full quantum dynamics are derived . it is shown that quantum corrections to mixed quantum classical methods can be obtained through stationary phase approximations to the full quantum dynamics . a rigorous description of the quantum corrections due to electronic nonadiabatic coupling on the nuclear dynamics within the ehrenfest framework is obtained . the fewest switches surface hopping method is shown to be obtained as a quasiclassical approximation to the dynamics and natural semiclassical extensions to include classically forbidden nonadiabatic transitions are suggested . .4 in .1 in -2 in 6.35 in 8.75 in

physics0702220

the born - oppenheimer approximation@xmath0 is fundamental to studies of a wide variety of molecular systems . however , violations of the born - oppenheimer approximation@xmath1 are ubiquitous in many molecular and condensed phase phenomena , and are responsible for several relaxation processes . despite their importance , fully quantum treatments of nonadiabatic corrections to born oppenheimer dynamics are limited and no perturbative formulation exists to account for them . it is practically impossible to compute the full quantum nonadiabatic dynamics for realistic systems due to the prohibitive computational cost associated with such a calculation . to circumvent this difficulty , a class of mixed quantum classical methods have been invented@xmath2 . these methods can be broadly classified into mean - field@xmath3 and trajectory surface hopping methods@xmath4 . mean field methods originate from work by ehrenfest@xmath5 , where the nuclear degrees of freedom , treated as classical mechanical variables , experience a potential that is the expectation value of the electronic hamiltonian with respect to the instantaneous electronic wavefunctions obtained by solving the electronic schrodinger equation . on the other hand , surface hopping methods@xmath4 treat the nuclear dynamics as being localized to a single born - oppenheimer surface at any time , with nonadiabatic couplings between surfaces leading to instantaneous jumps , or `` hops '' between born - oppenheimer surfaces . common to all mixed quantum - classical methods is the identification of the nuclear ( or slow ) degrees of freedom as behaving as essentially classical quantities , while interacting with a quantum system consisting of the electronic degrees of freedom . this identification is usually justified by a mixture of mathematical and phenomenological arguments . as a result , while these methods have been investigated by intuitive tests and computational simulations over the years , there is no rigorous theory to understand situations where this separation into classical and quantum subsystems breaks down and quantum effects on the classical degrees of freedom have to be considered . this problem is also evident in studies of the dynamics of molecular systems on higher - dimensional born - oppenheimer surfaces , where geometric features , like conical intersections between surfaces , have a significant physical effect on the dynamics . thus , given the somewhat ad - hoc nature of most mixed quantum classical methods , a central problem is to develop a framework which would incorporate these methods as an approximation and would enable the systematic study of corrections to the mixed quantum classical picture of nonadiabatic dynamics . this work aims to systematically construct mixed quantum classical methods as semiclassical limits of quantum nonadiabatic dynamics . such an approach enables the systematic improvement of mixed quantum classical methods to include further quantum corrections as well as the development of new and hybrid mixed quantum classical methods . exact path integral descriptions of the quantum nonadiabatic dynamics are provided and stationary phase approximations for the dynamics are derived . the resulting dynamical picture is a consistent means of obtaining a mixed quantum classical approximation to the nonadiabatic dynamics . it is demonstrated that both the ehrenfest approach and the surface hopping approach correspond to stationary phase approximations of the quantum nonadiabatic dynamics , when studied in complementary representations of the electronic system . furthermore , the approximate equations of motion obtained in the ehrenfest picture contain additional contributions to the effective nuclear force . these contributions are a direct , and nontrivial consequence of the geometry of the born - oppenheimer potential energy surfaces . for systems with only one nuclear dimension , these contributions are zero , and the dynamics is of pure ehrenfest type , while for higher dimensional surfaces , they are non - zero , and represent corrections to the quasiclassical dynamics due to the topological features of the surface , like the presence of conical intersections and degeneracies . in the complementary picture from which surface hopping methods are obtained as an approximation , quantum corrections to the nonadiabatic nuclear dynamics can be built in by considering semiclassical expansions of the nuclear propagator in the born - oppenheimer approximation . in addition to this , the approach outlined in this paper provides a new interpretation of nonadiabatic effects as a consequence of the geometry of the electronic hilbert space , and thus as arising due to a generalization of berry s phase for parametrized quantum systems . the outline of the paper is as follows . in sec.ii , a path integral description of the quantum nonadiabatic dynamics is introduced . issues concerning path integral approaches to nonadiabatic dynamics are discussed and the derivation of the approach used in this paper are outlined in sec.ii.a . in sec.ii.b , the path integral description is derived . it is based on integrating out the electronic degrees of freedom using a coherent state basis . furthermore , the path integral so obtained is compared to other path integral formulations constructed within a born - oppenheimer basis . it is shown in this section that nonadiabatic couplings between states are due to the geometry of the electronic hilbert space , and can be regarded as generalizations of berry s phase . in sec.iii , the path integral approach is analysed and stationary phase approximations are derived . sec.iii.a demonstrates that the stationary phase approximation to the coherent state path integral corresponds to the ehrenfest formulation of mixed quantum classical dynamics . furthermore , geometric corrections to the nuclear force in the ehrenfest approach are obtained . these corrections are related to the phase coherence between different states , and are most significant when different born oppenheimer energy surfaces become close in energy and are strongly coupled due to nonadiabatic effects . sec.iii.b studies stationary phase approximations to the complementary path integral description constructed using a born - oppenheimer basis for the electronic system . it is shown that the equations of motion so obtained correspond in the semiclassical limit , to the fewest switches surface hopping approach to nonadiabatic dynamics . the theory in sec.iii.b also points to the construction of semiclassical propagators that can extend descriptions of nonadiabatic dynamics beyond the mixed quantum - classical limit . sec.iii.c describes semiclassical quantization rules to approximate the eigenvalue spectrum for the full quantum dynamics . the paper then concludes with a discussion of the results and future work in sec.iv . the problems with path integral approaches to quantum nonadiabatic dynamics are briefly discussed , and a short outline of the consequent derivation of the path integral based theory is given . the detailed mathematical treatment of a path integral approach based on coherent states is then subsequently discussed . it should also be noted that this text uses the units convention where planck s constant is set to one , i.e @xmath6 in this paper . the nonadiabatic corrections to the born oppenheimer approximation can be viewed as a result of a coupling between a set of `` fast '' or bath degrees of freedom and slow or `` system '' degrees of freedom . however , the mechanisms that result from this coupling are different from the usual problem of a system interacting with a dissipative bath , since the hilbert space enclosing the bath degrees of freedom in the born oppenheimer theory is parametrized by the slow degrees of freedom . as a consequence , the time dependent evolution of such a system involves complex geometric phase effects@xmath7 . pechukas@xmath8 first developed a path integral prescription for nonadiabatic dynamics and derived a set of nonlocal equations of motion for the semiclassical nuclear path . later work@xmath9 on a path integral description clarified the interpretation of the nonadiabatic coupling vector potentials as nonabelian gauge fields . however , these descriptions are not easily amenable to approximations which can be studied computationally . in the pechukas formulation , the effect of the electronic degrees of freedom on system dynamics enters through a nonlocal potential which is a function of the paths taken by the system , and similarly in the formulation of moody , shapere and wilczek@xmath10 , nonadiabatic couplings between born - oppenheimer states enter as path ordered matrix exponentials . thus , an approach that adopts the path integral framework for the exact nonadiabatic dynamics , but which is also amenable to systematic and computationally tractable approximations would be of considerable theoretical and practical utility . the problems with regards to nonlocality and path ordered exponentials encountered in a path integral description of the quantum nonadiabatic dynamics can be partially resolved by an appropriate choice of the path integral . to this end , a local basis of continuous , @xmath11-particle coherent state slater determinants is chosen to describe the electronic system . the use of coherent states enables the straightforward construction of stationary phase approximations to the quantum dynamics and corresponding semiclassical quantization rules . furthermore , since the propagator can be written as a functional integral over a continuous coherent state space , issues relating to path ordering and nonlocality become considerably simplified . the coherent state basis is parametrized by the nuclear coordinates and is overcomplete . since the basis is parametrized by the nuclear coordinates , a local hilbert space is defined at each point on a given nuclear path . thus , intuitively , motion along a nuclear path involves the following : the first contribution is due to the propagation of the nuclear part of the hamiltonian , and the second is due to a rotation of the hilbert space from one point into the next one on the path ( fig 1 ) . this rotation takes basis states at one point into the basis states at the next , and hence changes the electronic contribution to the effective nuclear hamiltonian . in addition , the rotation of the basis in general involves `` twists '' and skews resulting in torsional contributions to the force acting on the nuclear degrees of freedom . to explicitly describe this , the derivation of the coherent state path integral proceeds as follows . the matrix elements of the time propagation operator for the full ( nuclear plus electronic ) hamiltonian between initial and final nuclear positions are written down as a path integral . the time for propagation is broken up into a large number of ( infinitesimal ) time segments and the propagation operator is written in a trotter product expansion . complete sets of nuclear position eigenstates are then introduced for each time segment , and an infinitesimal propagator between successive nuclear positions is written down . the resulting expression is a path integral of an operator acting on the electronic degrees of freedom over each nuclear path . for a given nuclear path , this quantity simply corresponds to the electronic propagation operator parametrized by the nuclear path . the matrix elements of the quantum nonadiabatic propagation operator between initial and final electronic states and nuclear positions are then evaluated through the matrix elements of this path dependent quantity between initial and final electronic states . this is done by adding overcomplete sets of electronic @xmath11 particle coherent state slater determinants at each time slice and evaluating matrix elements of the infinitesimal propagation operator . the overlap between these basis states belonging to consecutive timesteps gives rise to the nonadiabatic coupling contribution to the dynamics . the resulting path integral is then further analysed . the full hamiltonian for a system with nuclear and electronic degrees of freedom can be written as @xmath12 the nuclear hamiltonian @xmath13 is given by @xmath14 the coordinates @xmath15 denote the @xmath16 nuclear degrees of freedom with momenta given by @xmath17 . the potential energy function , @xmath18 is the bare nuclear potential . the electronic hamiltonian can be similarly defined and the electronic coordinates , @xmath19 , are elements of the vector @xmath20 . the electronic potential energy @xmath21 is parametrically dependent on the nuclear positions . @xmath22 for simplicity of presentation , the nuclei and electrons are assumed to be spinless . this assumption is reasonable for the nonadiabatic effects considered here , and the treatment that follows can be extended to include spin effects . also it is assumed that the nuclear masses , @xmath23 are all equal and have the value @xmath24 . this is solely for notational convenience and does not significantly affect the conclusions of this work . a born oppenheimer basis of eigenstates and eigenvalues is defined by @xmath25\mid\psi_{\mu};{\bf r}\rangle = e_{\mu}({\bf{r}})\mid\psi_{\mu } ; { \bf{r}}\rangle\ ] ] the unitary time propagation operator for this hamiltonian can be expanded in a trotter product expansion as described below . @xmath26 } \exp{\big[-i\epsilon\frac{{\bf p}^{2}}{2m}\big]}\exp{\big[-i\frac{\epsilon}{2}(h_{e } + u)\big]}}\end{aligned}\ ] ] the propagation operator for a given set of initial and final nuclear positions @xmath27 and @xmath28 can be written as @xmath29}\langle{\bf r}_{k}\mid\exp{\big[-i\epsilon\frac{{\bf p}^{2}}{2m}\big ] } \mid{\bf r}_{k+1}\rangle}}\nonumber\\ \exp{\big[-\frac{i\epsilon}{2}\{h_{e}({\bf r , r}_{k+1})+u({\bf r}_{k+1})\}\big ] } & & \end{aligned}\ ] ] where @xmath30 and @xmath31 . it is important to note that the matrix elements of the propagation operator described in eq.(6 ) are still operators acting on the electronic hilbert space . this is because the matrix elements are a partial trace with respect to the eigenstates @xmath32 and @xmath33 of the nuclear position operator @xmath34 . this partial trace can be rewritten as a path integral over the set of nuclear paths @xmath35 . @xmath36 } g_{na}[{\bf r}(\tau)]}\ ] ] @xmath37 $ ] is the bare nuclear action corresponding to the potential @xmath18 . the quantity @xmath38 is an operator quantity in the electronic degrees of freedom , and is a nonlocal functional of the nuclear paths . physically , the operator @xmath39 $ ] includes the `` connection '' between electronic hilbert spaces defined at each point on a given nuclear path @xmath40 . thus , a complete solution of the nonadiabatic problem would require that electronic matrix elements of @xmath38 be evaluated . to perform such an evaluation , overcomplete sets of antisymmetric @xmath11 electron slater determinant coherent state wavefunctions are introduced at each nuclear position along a given nuclear path@xmath41 . a general @xmath11 electron antisymmetric wavefunction can be written as a linear combination of slater determinants . this basis of @xmath11 electron slater determinants , @xmath42 is overcomplete and has the closure property : @xmath43 here , a local basis of @xmath11 electron coherent state slater determinants@xmath41 is defined at each point on a given nuclear path . hence , this local electronic basis is parametrized by the nuclear position to account for the fact that a local electronic hilbert space exists at each nuclear position . for such a basis defined for a nuclear position @xmath44 , the completeness relation reads : @xmath45 for notational convenience , the integration measure in the overcompleteness relation will be written as @xmath46 = \frac{d\phi^{*}({\bf r})d\phi({\bf r})}{(2\pi i)^{n}}e^{-\mid\phi({\bf r})\mid^{2}}\ ] ] with the differentials and the absolute value of @xmath47 defined by @xmath48 the parametrization of the electronic hilbert space by the nuclear position is responsible for geometric phase effects . as will be shown below , these effects enter through the definition of a local basis of wavefunctions at each nuclear position . the operator @xmath38 can be explicitly defined to be @xmath49=\prod_{k=1}^{p}{\exp{[-i\epsilon h_{e}({\bf r , r}_{k})]}}\ ] ] this operator is a path ordered quantity because the electron hamiltonian does not commute with itself when evaluated at different points on a nuclear path . to evaluate the matrix elements of this operator , overcomplete sets of @xmath11-electron slater determinant wavefunctions are introduced at each timestep . on doing so , the matrix elements of @xmath38 between an initial electronic state @xmath50 and a final state @xmath51 can be written as @xmath52d\mu[\phi_{j}({\bf r}_{j})]\langle\phi^{0};{\bf q ' } \mid\psi({\bf q}')\rangle\langle\psi({\bf q}')\mid e^{-i\epsilon h_{e}({\bf r , q } ' ) } \mid\phi({\bf q}')\rangle}}\nonumber\\ \langle\phi({\bf q}')\mid\psi_{1}({\bf r}_{1})\rangle\langle\psi_{1}({\bf r}_{1})\mid e^{-i\epsilon h_{e}(\bf{r , r}_{1})}\mid\phi_{1}({\bf r}_{1})\rangle ... \langle\phi_{p}({\bf r}_{p})\mid\phi^{f};{\bf q}\rangle\nonumber\\\end{aligned}\ ] ] two sets of overcomplete basis states are introduced at each timestep , because the time evolution operator is in general not diagonal in this basis of states , and the overlap between basis states corresponding to different positions on the nuclear path is not unity . more specifically , the coherent states inserted at successive timesteps , @xmath53 , are parametrized by different sets of nuclear positions , @xmath54 . since there is no constraint on the basis states at different nuclear positions , all possible matrix elements of the time evolution operator between these basis states will have to be considered . furthermore , to evaluate this off - diagonal matrix element , an additional set of basis states parametrized by the position @xmath55 is introduced at time @xmath56 . since the path integral is discontinuous in the coherent state space , all possible overlaps of this new basis element with the basis element parametrized by @xmath57 will have to be considered , and the matrix elements of the time evolution operator are not constrained to be diagonal . this lack of constraint is unlike the cases encountered with more traditional coherent state path integrals where only diagonal matrix elements between coherent states suffice , and is a consequence of the parametrization by the nuclear coordinates . it is clear that the above path integral involves matrix elements of the form @xmath58 = \langle\psi({\bf r}_{k})\mid\exp{(-i\epsilon h({\bf r , r}_{k}))}\mid\phi({\bf r}_{k})\rangle\ ] ] if the exponent on the right hand side of eq.(14 ) is expanded in powers of @xmath59 and terms higher than the first order are neglected , the matrix elements @xmath60 can be evaluated to be approximately @xmath58 = \langle\psi({\bf r}_{k})\mid\big[1 - i\epsilon h({\bf r , r}_{k})\big]\mid\phi({\bf r}_{k})\rangle\ ] ] making the same assumption as before , the right hand side of eq.(15 ) can be reexponentiated to yield @xmath58 = \langle\psi({\bf r}_{k})\mid\phi({\bf r}_{k})\rangle\exp{\big[-i\epsilon\frac{\langle \psi({\bf r}_{k})\mid h({\bf r , r}_{k})\mid\phi({\bf r}_{k})\rangle}{\langle\psi({\bf r}_{k } ) \mid\phi({\bf r}_{k})\rangle}\big]}\ ] ] in addition to the matrix elements @xmath60 , the path integral contains `` connection '' factors of the form @xmath61 these connection factors are geometric in character and are due to the parallel transport of the local electronic basis from one nuclear position to the next . thus , these factors correspond to a generalization of the canonical , or berry s phase for simple , parameter dependent hamiltonians@xmath62 . to obtain a more familiar form for this phase dependence , the connection factors can be rewritten as follows : @xmath63 the difference @xmath64 is the difference between the wavefunctions @xmath65 evaluated at successive time steps . @xmath66 on exponentiating and making the assumption that the basis wavefunctions @xmath65 vary continuously , the connection factor takes the form @xmath67 where the time derivative acts on the wavefunction @xmath65 . thus , on multiplying the matrix elements and correction factors together , the following path integral expression for the electronic contribution @xmath38 can be obtained as below : @xmath68d[\phi^{*}({\bf r}(\tau)),\phi({\bf r}(\tau ) ) ] \langle\psi^{0};{\bf q}'\mid\psi({\bf q}')\rangle}\nonumber\\ \langle\phi({\bf q})\mid\psi^{f};{\bf q}\rangle\exp{\big[i\int_{0}^{t } { \frac{\langle\phi({\bf r}_{\tau})\mid i\overleftrightarrow{\partial}_{\tau } - h_{e } \mid\psi({\bf r}_{\tau})\rangle}{\langle\phi({\bf r}_{\tau})\mid\psi({\bf r}_{\tau})\rangle } d\tau}\big]}\end{aligned}\ ] ] the quantities @xmath69 = \prod_{j=1}^{p}{d\mu[\psi_{j}({\bf r}_{j})]}$ ] are the measures for the coherent state path integral . the time derivative is symmetrized by integrating by parts and ignoring boundary terms . thus , the full path integral between two states @xmath70 and @xmath71 is given by a total action which of the form : @xmath72 = \int_{0}^{t}{\{l_{n}({\bf r},\dot{\bf r } ) + \frac{\langle\phi({\bf r})\mid i\overleftrightarrow{\partial_{t } } - h_{e}\mid\psi({\bf r})\rangle}{\langle\phi({\bf r})\mid\psi({\bf r } ) \rangle}\}dt}\ ] ] here the time derivative , @xmath73 , is the symmetrized derivative defined as : @xmath74\ ] ] the purely nuclear contribution to the action integral is contained in the lagrangian @xmath75 which is defined as follows : @xmath76 here the quantities @xmath44 correspond to the set of nuclear coordinates , and @xmath24 is the mass of the nuclei . @xmath18 is the bare nuclear potential as defined earlier in eq.(2 ) . the path integral in terms of this action can be written as @xmath77\big]}\nonumber\\\end{aligned}\ ] ] since , in general , off diagonal matrix elements of the time evolution operator in the coherent state basis are being considered , the boundary conditions on the path integral are that the coherent states defined at the end points @xmath78 and @xmath27 are allowed to vary without constraint . however , in the semiclassical limit , which corresponds to choosing only the subset of continuous paths in coherent state space , the coherent states at the boundaries are constrained to have unit overlaps with the initial and final electronic wavefunctions respectively . as noted earlier , the necessity of choosing two sets of coherent states at each time slice in the path integral means that the paths in general are discontinuous . this is because , in the limit where the discrete time slicing in the path integral is made infinitesimally small , the choice of different states at each such time slicing corresponds to a discontinuous jump in the space of coherent states . however , the stationary phase limit of the path integral involves continuous paths in state space . hence , it is reasonable to reduce the unconstrained sum over all ( discontinuous and continuous ) paths to one where the coherent states are constrained to be similar to each other . a simple way of doing this is to assume the overlap , @xmath79 , to be nearly unity . this is a reasonable approximation also because contributions to the path integral when this overlap is small are negligible . this can be seen from the discrete version of the path integral , eq.(13 ) and eq.(16 ) , from which it is clear that as the overlap goes to zero , the exponent in the path integral becomes highly oscillatory , and in addition is multiplied by the small overlap . the action when the overlap is assumed to be unity can be considered to be a zeroth order approximation to the full action , when expanded in terms of the overlap . with this simplification , the path integral action becomes : @xmath72 = \int_{0}^{t}{\big[l_{n}({\bf r},\dot{\bf r } ) + \langle\phi({\bf r})\mid i\overleftrightarrow{\partial}_{\tau } - h_{e}\mid\psi({\bf r})\rangle\big]d\tau}\ ] ] this action has a simple interpretation . the time derivative part of the action corresponds to the nonabelian generalization of berry s phase , while the other parts correspond to the dynamical evolution of the system . it is now possible to show that the generalized berry s phase in this action corresponds to nonadiabatic coupling terms between born - oppenheimer energy levels . to show this , the slater determinant states @xmath65 and @xmath80 are expanded in a local basis of born oppenheimer electronic eigenstates as below : @xmath81\mid\psi_{\mu};{\bf r}}\rangle\\ \mid\phi({\bf r})\rangle = \sum_{\mu}{w_{\mu}[\phi,{\bf r}]\mid\psi_{\mu};{\bf r}}\rangle\end{aligned}\ ] ] the states @xmath82 are eigenstates of the electron hamiltonian @xmath83\mid\psi_{\mu};{\bf r}\rangle = e_{\mu}({\bf r})\mid\psi_{\mu};{\bf r } \rangle\ ] ] in this basis , the phase of the connection factor over the entire path , @xmath84 can be rewritten as @xmath85 = \sum_{\mu\nu}{\int_{0}^{t}{d\tau\frac{1}{2}(w^{*}_{\mu}\langle\psi_{\mu};{\bf r}\mid i\partial_{\tau}[w_{\nu}\mid\psi_{\nu};{\bf r}\rangle ] - i\partial_{\tau}[w^{*}_{\mu } \langle\psi_{\mu};{\bf r}\mid]w_{\nu}\mid\psi_{\nu};{\bf r}\rangle)}}\ ] ] this connection factor is the phase obtained by multiplying the connection factors @xmath86 , in eq.(20 ) over the entire path . a little algebra can be used to demonstrate that the connection factor can be further simplified into the following form : @xmath85 = \int_{0}^{t}{d\tau\sum_{\mu}{\frac{1}{2}(w^{*}_{\mu}[\phi,{\bf r } ] i\dot{w}_{\mu}[\psi,{\bf r } ] + h.c ) } } - \int_{c_{{\bf r}}}{d{\bf r}\cdot \sum_{\mu\nu}{{\bf a}_{\mu\nu}({\bf r})w^{*}_{\mu}[\phi,{\bf r}]w_{\nu}[\psi,{\bf r}]}}\ ] ] here , the coupling term @xmath87 is the familiar nonadiabatic coupling vector in the born oppenheimer basis : @xmath88 the second term that contributes to the connection factor has the form of an integral over the gauge potential @xmath87 . in contrast to the case of a simple berry phase , this `` gauge '' potential is an infinite dimensional matrix and corresponds to a nonabelian geometric phase . thus the nonadiabatic coupling between born oppenheimer surfaces has a purely geometric interpretation as arising due to the parallel transport of a local basis of electronic wavefunctions along a nuclear path . it is therefore a natural generalization of the berry phase for parameter dependent quantum systems@xmath62 . this generalized geometric phase also depends on the off - diagonal elements of the electronic density matrix , and consequently couples the slow , nuclear and fast , electronic degrees of freedom . in the limit where an infinite number of adiabatic states corresponding to the fast modes can be accessed , this phase represents a dissipative coupling between the system of slow nuclear modes and the bath of fast electronic modes@xmath89 . it should be noted that generalizations of berry s phase to the case of nondegenerate electronic levels have previously been obtained by mead@xmath90 and by zygelman @xmath91 . a related path integral description of the nonadiabatic dynamics has also previously been derived by moody , shapere and wilczek@xmath92 . in their work , a path integral was obtained by introducing born - oppenheimer electronic eigenstates at each nuclear position , in the expression for @xmath93 $ ] given by eq.(12 ) . the electronic matrix elements of @xmath93 $ ] between initial and final born oppenheimer eigenstates @xmath94 and @xmath95 are given by @xmath96\mid\psi_{n};{\bf q}\rangle = \sum_{m_{1},m_{2}, .. }{\exp[-i\epsilon e_{m}({\bf q}')]\langle\psi_{m};{\bf q}'\mid \psi_{m_{1}};{\bf r}_{1}\rangle\exp[-i\epsilon e_{m_{1}}({\bf r}_{1})]}\nonumber\\ \langle\psi_{m_{1}};{\bf r}_{1}\mid\psi_{m_{2}};{\bf r}_{1}\rangle .... \langle \psi_{m_{n}};{\bf r}_{n}\mid\psi_{n};{\bf q}\rangle\nonumber\\ \end{aligned}\ ] ] this expression contains connection factors of the form : @xmath97 it is easy to see that this connection factor can be manipulated as with the coherent state case to give @xmath98\ ] ] with @xmath99 . this expression can be rewritten in terms of the nonadiabatic coupling to give @xmath100\ } _ { m_{k},m_{k+1}}\ ] ] this connection factor is similar to the connection factor , eq.(20 ) derived in the coherent state electronic basis . it is however harder to handle for the purpose of approximations and computations , due to the discrete nature of the summation over born - oppenheimer eigenstates . although the approach presented here is physically equivalent to that of moody , shapere and wilczek , the use of electronic coherent states enables semiclassical approximations to the nonadiabatic quantum dynamics . this is possible because the restricted sum over only born oppenheimer eigenstates in eq.(33 ) is replaced with an integral over continuous paths in the coherent state space . this provides a practical advantage , in that computationally tractable semiclassical approximations to the full path integral become possible . it is shown later in this work that stationary phase approximations to the nonadiabatic path integral so obtained , are closely related to the ehrenfest formulation of mixed quantum - classical dynamics . the path integral representation derived here is an exact reformulation of the quantum dynamics of the system . however , due to the continuum nature of the coherent state basis , it is difficult to numerically evaluate the path integral propagator and obtain the full quantum dynamics . the path integral derived in the previous section can be used to develop a semiclassical formulation of the nonadiabatic dynamics . in this section , stationary phase approximations to the path integral are derived and a semiclassical quantization rule is described for the nonadiabatic dynamics . the stationary phase equations so obtained form a natural quasiclassical limit for the full quantum dynamics . for the coherent state path integral , this approximation is shown to yield the ehrenfest equations of motion with an additional correction term . this correction term corresponds to the geometric contribution to the equations of motion , and is non - zero in general when the nuclear subspace is three dimensional or higher . for the sake of completeness , the stationary phase approximation to the coherent state path integral is compared to a generalized stationary phase type approximation to the nonadiabatic path integral when written out in the basis of electronic born - oppenheimer eigenstates . it is shown that a generalization of the stationary phase approximation in this case yields a quasiclassical theory which is the target of the fewest switches trajectory surface hopping method . a natural semiclassical extension of this approximation yields a generalization of the fewest switches method which can account for hops in classically forbidden regions . finally , the derivation of general mixed quantum classical methods that combine the ehrenfest and surface hopping methods is discussed . thus , this section clarifies and lays down the neccesary theoretical foundations for the systematic improvement of mixed quantum classical methods . a stationary phase approximation to the path integral can be made to yield quasiclassical equations of motion for the coupled dynamics of the nuclear and electronic degrees of freedom . such approximations are derived here to obtain quasiclassical equations of motion for the system . it is shown that the equations so obtained are a generalization of ehrenfest s equations of motion . a density matrix for the electronic subsystem can be defined as @xmath101 = w^{*}_{\beta}[\phi,{\bf r}]w_{\alpha}[\psi,{\bf r}]\ ] ] where the quantities @xmath102 are as defined in eq.(27 ) and eq.(28 ) . to express the nonadiabatic action in terms of this density matrix , the explicit form of the action from eqs.(25 ) and ( 30 ) is written to be @xmath103 = \int_{0}^{t}{d\tau\big[l_{n}({\bf r},\dot{\bf r } ) + \frac{i}{2 } [ w_{\mu}^{*}\dot{w}_{\nu } - \dot{w}_{\mu}^{*}w_{\nu } ] - \dot{\bf r}\cdot\sum_{\mu\nu}{{\bf a}_{\mu\nu}w_{\nu}[\phi , t]w_{\mu}^{*}[\psi , t]}\big]}\ ] ] with this form of the action , the stationary phase approximation corresponds to the condition @xmath104 = 0 .\ ] ] a detailed derivation of the stationary phase equations is given in appendix a. at the end of this work . on performing the appropriate manipulations , the stationary phase condition for the electronic density matrix , @xmath105 , can be shown to be the following equation of motion : @xmath106}\ ] ] here , the energies @xmath107 are the eigenvalues of the born oppenheimer eigenstates as defined in eq.(29 ) . the corresponding equation of motion for the nuclear coordinates , @xmath44 with mass @xmath24 and the bare nuclear potential @xmath18 can be written to be @xmath108 the stationary phase equation for the nuclear coordinates , @xmath44 , contains a term that is proportional to the time derivative of the off diagonal electron density matrix @xmath109 . in addition to the other terms which are proportional to the off - diagonal density matrix , this is an explicit contribution to the force acting on the nuclei due to the time variation of the electronic coherence . hence , this accounts for changes in the effective nuclear force due to electronic dephasing . eq.(41 ) can be further simplified and written explicitly in terms of the elements of the electronic density matrix , @xmath105 and the nonadiabatic couplings @xmath110 as , @xmath111 if a nonabelian `` vector potential '' is defined as an infinite dimensional matrix with elements @xmath112_{\mu\nu } = { \bf a}_{\mu\nu}\ ] ] then the corresponding gauge invariant , nonabelian `` magnetic field '' is given by @xmath113 from these definitions and eq.(42 ) , the stationary phase equations for the nuclear coordinates can be written in the form @xmath114 eq.(40 ) corresponds to the familiar liouville equation for the electronic density matrix , while eq.(45 ) is different because it contains corrections due to the nonadiabatic coupling . the first three force terms correspond to the effective force in the ehrenfest formulation of mixed quantum classical dynamics , while the last term is a geometric correction to the ehrenfest equations . this term therefore constitutes a new addition to the equations of motion and will need to be retained for consistency with the full quantum dynamics . when this term is ignored , the equations of motion so obtained correspond to the ehrenfest equations of motion . the equation , eq.(45 ) , for the nuclear coordinates satisfies the invariance under general unitary transformations of the electronic basis . to see this , note that the nonadiabatic coupling matrix @xmath115 is defined through the nonadiabatic coupling matrix elements as @xmath112_{\mu\nu } = { \bf a}_{\mu\nu}({\bf r } ) = -i\langle\psi_{\mu}\mid\nabla_{\bf r}\psi_{\nu}\rangle\ ] ] to write this in a more transparent form , infinite dimensional column vectors @xmath116 can be constructed with elements given by the electronic basis wavefunctions @xmath117 . the matrix can then be rewritten as a product of column vectors @xmath118 it is straightforward to show that under a unitary transformation , @xmath119 of the electronic wavefunctions , @xmath120 the nonadiabatic coupling matrix transforms as @xmath121 the `` magnetic field '' , @xmath122 with matrix elements @xmath123 correspondingly transforms as @xmath124 the density matrix elements @xmath125 also transform similarly under the unitary transformation , and hence the `` magnetic field '' contributions to the equations of motion are invariant under such transformations . this invariance is a general version of the the usual invariance under gauge transformations , of vector potentials and magnetic fields in classical electromagnetism . it should be noted that the unitary transformations under which the electronic wavefunction basis transforms are infinite dimensional matrices , and hence the group , @xmath126 that these unitary transformations belong to is the group of symmetries for the nonadiabatic corrections . on the other hand , the terms in the equations of motion which are gradients of the electronic energy eigenstates break this symmetry , since the energy eigenvalues are , in general , not completely degenerate . when degeneracies exist for a subset of the electronic energy levels , the contribution to the nuclear force from the degenerate states is symmetric under unitary transformations corresponding to the degenerate subspace . the group corresponding to this symmetry is @xmath127 , where @xmath128 is the degree of the degeneracy . when this symmetry group exists , the time evolution of the degenerate electronic subspace picks up a berry phase@xmath129 which contributes to the nuclear force in eq.(45 ) in a fashion identical to that of the nonadiabatic coupling between nondegenerate states . thus , the equation of motion for the nuclei , eq.(45 ) , is valid irrespective of whether the states in the electronic subspace are degenerate or not . for example , in the case of a conical intersection between two potential energy surfaces , the wavefunction becomes multivalued . to resolve this multivaluedness , a geometric phase due to a vector potential can be added to the wavefunctions . the geometric forces in eq.(45 ) are the classical analog of this procedure . it should be noted that the geometrical , `` magnetic field '' contributions in eq.(45 ) are zero for one dimensional nuclear potential energy surfaces . this can be intuitively understood through the discussion in sec.ii.a . this magnetic contribution is due to the `` rotation '' of the local electronic hilbert space along the nuclear path . for one dimensional nuclear potential energy surfaces , this `` rotation '' does not have any torsional components , and is simply a pure parallel transport of the basis . thus , for single nuclear dimensions , these forces are zero and do not contribute to the dynamics . however , for potential energy surfaces parametrized by higher dimensional nuclear coordinates , these forces are non - zero and of importance to the nonadiabatic dynamics . these couplings therefore cause an effective magnetic force contribution to be added to the usual ehrenfest force , in the mixed quantum classical dynamics . the magnetic forces are the quasiclassical analog of the additional berry - mead phase which the electronic wavefunctions are multiplied by , to maintain their single valuedness . a complementary description of the full nonadiabatic quantum propagation of the nuclei can be obtained by writing the nonadiabatic propagation operator @xmath130 $ ] from eq.(12 ) in a basis of electronic born oppenheimer eigenstates . as mentioned previously such a path integral expression for the propagator has been written down by moody , shapere and wilczek . explicitly , if the wavefunction for the combined system is given by @xmath131 then the time evolution of the nuclear states @xmath132 can be written as @xmath133 the propagator @xmath134 is the matrix element of the time evolution operator when evaluated between an initial electronic born - oppenheimer state , @xmath135 , defined for an initial position @xmath27 and a final electronic born - oppenheimer state , @xmath136 , defined for a final nuclear position @xmath28 . this matrix element can be evaluated using the trotter product formula to slice the time @xmath137 into infinitesimal time intervals , and introducing a basis of electronic born oppenheimer states at each time slice . the resulting propagator is the same as the propagator defined in eq.(33 ) . explicitly , this propagator is the matrix element @xmath138 if the born - oppenheimer limit is assumed to be valid , then the propagator is diagonal in the indices @xmath139 . @xmath140 here , @xmath141 is the nuclear propagator in a potential given by @xmath142 @xmath143 the nonadiabatic nuclear propagation can be thought of as a sequence of propagation along born - oppenheimer paths with the propagator defined above in eq.(54 ) , and `` hops '' occurring in between to switch between born - oppenheimer paths . this is illustrated in fig.2 . detailed considerations of the structure of the paths that contribute to the transition amplitude @xmath134 thus point to the origins of surface hopping type approximations . to derive explicitly quasiclassical techniques based on the discrete path integral , consider the effective propagation operator defined in eq.(12 ) , which corresponds to matrix elements of the full propagation operator between initial and final nuclear position eigenstates . to recapitulate , the matrix elements are given as an integral of a path ordered exponential @xmath144 $ ] . for a given nuclear path @xmath35 the matrix elements of @xmath38 between an initial born - oppenheimer electronic state @xmath135 and a final state @xmath136 are given by @xmath145\mid{\bf q};\psi_{n}\rangle = \langle{\bf q}';\psi_{m}\mid e^{-ih({\bf q}')\epsilon}e^{-ih({\bf r}_{1})\epsilon} ... e^{-ih({\bf q } ) \epsilon}\mid{\bf q};\psi_{n}\rangle\ ] ] from this relationship , the quantum nonadiabatic propagator between an initial born oppenheimer electronic wavefunction @xmath95 defined at an initial nuclear position @xmath27 and a final born oppenheimer electronic wavefunction , @xmath94 defined for a position @xmath28 can be written as @xmath146\mid{\bf q};\psi_{n}\rangle}\ ] ] if a nonadiabatic transition takes place during the interval @xmath147 for sufficiently small values of @xmath59 , the propagator has the composition property @xmath148 the overlap between electronic wavefunctions at @xmath149 and @xmath150 is the connection factor , @xmath151 , discussed in the previous section , and is responsible for the nonadiabatic coupling between levels . as before , it can be rewritten as : @xmath152 with the nonadiabatic coupling @xmath153 given by eq.(32 ) as before . this composition law for the exact nonadiabatic propagator can be used to derive a born series expansion for the propagator in terms of the born oppenheimer propagator . the physics behind this is illustrated in fig.2 . @xmath154 here , the propagator is written as a sum over paths where there are successively higher number of nonadiabatic transitions . to obtain a classical nuclear propagation scheme , the nuclei can be propagated along the stationary phase solutions for those pieces of the nuclear path where the propagator is of the born oppenheimer type . in other words , the born - oppenheimer limit of the nonadiabatic propagator can be approximated with a semiclassical expansion around the stationary phase path as discussed below . it is easy to see that along a born oppenheimer path , @xmath155 furthermore , if a single nonadiabatic transition occurs from @xmath156 on a sufficiently short timescale @xmath59 , the amplitude @xmath132 changes by @xmath157 putting the two equations together , the total time derivative of the wavefunction @xmath158 is recovered to be : @xmath159 it is important to note that this equation is valid under the assumption that nonadiabatic transitions occur on a very short timescale in comparison to the timescales that govern the rest of the dynamics . in general , there is a nonzero probability for nonadiabatic transitions occuring over longer timescales and accompanied by finite changes , @xmath160 in the nuclear coordinates . this equation is derived assuming that the contribution from such transitions is small , and that dominant features of the nonadiabatic dynamics can be captured by considering transitions that occur on infinitesimal timescales . to derive a semiclassical limit for the nuclear evolution , the nuclear wavefunctions @xmath161 are approximated as gaussian wavefunctions centered around the instantaneous classical nuclear path on the born oppenheimer surface , according to the prescription of heller@xmath162 . @xmath163\ ] ] the functions @xmath164 are instantaneous gaussian wavepackets centered around the classical born - oppenheimer trajectories given by @xmath165 furthermore , this gaussian wavepacket approximately satisfies ( when the born oppenheimer potential energy , @xmath166 , is expanded to quadratic order in @xmath167 ) the time dependent schrodinger equation : @xmath168 substituting the ansatz , eq.(64 ) , into eq.(63 ) , one obtains for the time evolution of the coefficients , @xmath169 @xmath170 making use of the approximation in eq.(66 ) , the time evolution of @xmath169 becomes , in the semiclassical limit : @xmath171 multiplying by @xmath172 and integrating over the coordinates @xmath167 gives @xmath173 this can be rewritten in terms of a density matrix @xmath174 as @xmath175}\ ] ] furthermore , if the gaussians @xmath176 are assumed to be delta functions in the positions @xmath177 , the above equations simplifies further into : @xmath178}\ ] ] where the variable @xmath44 and its velocity , @xmath179 are classical quantities corresponding to the center of the gaussian function @xmath176 . since the functions @xmath176 are gaussians centered on the classical nuclear born - oppenheimer trajectories , the classical variables @xmath44 satisfy newton s laws : @xmath180\ ] ] when the trajectory being considered is that belonging to the @xmath181 born - oppenheimer surface . further , the rate of change of the diagonal elements , @xmath182 is given by @xmath183}\ ] ] it is clear from this equation and the fact that the nuclear propagation always occurs on a single born - oppenheimer surface , that these equations describe the surface hopping method@xmath4 , for which the fewest switches algorithm is a monte carlo solution . thus , the fewest switches surface hopping ( fssh ) approach@xmath184 corresponds to a semiclassical approximation to the exact quantum nonadiabatic dynamics , similar to the ehrenfest method described in sec iii.a . it is also interesting to note that eq.(70 ) can be used instead of eq.(73 ) to generate the surface hopping criterion . eq.(70 ) provides for a nonzero hopping probability even in regions where transitions are classically forbidden , unlike the usual fssh method . a similar extension has also been proposed previously by neria and nitzan@xmath185 , on physical grounds through the construction of a semiclassical golden rule . in addition to this , it is clear from this discussion that semiclassical propagation of the nuclear wavefunction , either using the gaussian form , eq.(65 ) and eq.(66 ) proposed by heller@xmath162 , or by the use of a different basis , lead to generalized schemes for nonadiabatic dynamics that go beyond the mixed quantum classical limit . the use of frozen gaussians within this framework correspond to a scheme for nonadiabatic dynamics that is similar to the multiple spawning method proposed by martinez and co - workers@xmath186 . the approximation of hops occurring over infinitesimal timescales that is required to derive both eq.(70 ) and eq.(73 ) causes phase changes in the wavefunction due to the hops to be neglected . furthermore , hops need not occur on infinitesimal timescales , and the quantum nuclear paths which contribute to the dynamics are in general not differentiable , rendering the nuclear velocity to be a poorly defined quantity . in circumstances where such paths contribute significantly to the dynamics surface hopping type methods will be invalid . additionally , in the case where two ( or more ) electronic potential energy surfaces become degenerate , the nonadiabatic coupling also picks up nonzero diagonal elements which could act on the dynamics along significant sections of the nuclear path . in such situations , an approximate means of treating the dynamics would be to add the diagonal terms in the nonadiabatic coupling to the nuclear hamiltonian and perform the nuclear propagation on each born oppenheimer surface with an additional magnetic force due to the diagonal terms in the nonadiabatic coupling . the off diagonal nonadiabatic couplings can then still be treated as acting instantaneously to cause hops . it is likely that such modifications could be of value in improving the accuracy of the fssh method in such situations . finally , it is useful to discuss the relative utility of the discrete basis path integral presented here versus the coherent state approach that leads to the ehrenfest equations in the previous section . the natural stationary phase limit of the coherent state path integral is the mean field ehrenfest theory , while the discrete path integral has a corresponding stochastic limit . these two limits can be thought of as corresponding to two different types of statistics in the hopping events . the coherent state path integral is a better starting point for approximations when there is significant nonadiabatic coupling between multiple born - oppenheimer surfaces , i.e when hops can occur with high frequency . in such a situation , a mean field treatment is a good first approximation to the dynamics . however , the discrete basis approach is better suited to studying the dynamics in the presence of nonadiabatic dynamics between a few levels , and when born oppenheimer surfaces begin to diverge . in such a situation , the hopping events are relatively infrequent , and nonadiabatic behaviour is sensitive to the detailed structure of neighbouring born - oppenheimer surfaces . the discrete basis approach , for which the surface hopping algorithm is an approximation is a good starting point for approximation in this situation . the nonadiabatic action eq.(38 ) can be used to describe a semiclassical quantization condition when the system is in a bound state . this condition is derived in a fashion similar to that used to study the semiclassical quantization of coherent state path integrals@xmath187 , and in quantum field theories@xmath188 . if an assumption is made that the system executes periodic orbits , then a quantization condition can be written . to derive this quantization condition , the trace of the propagator is written as @xmath189\ ] ] the trace is related to the density of states as below . @xmath190 = i\int_{0}^{\infty}{e^{iet}k_{na}(t ) dt}\ ] ] if the trace corresponding to @xmath191 is expanded in terms of the eigenstates of the entire ( the combined nuclear and electronic ) system , @xmath191 can be written in terms of the system energy eigenvalues @xmath192 as @xmath193 it is clear from this that the poles of the function @xmath191 correspond to the energy eigenvalues of the system . hence , the poles of the fourier transform of the propagator trace are related to the eigenstates of the combined system in this fashion . the fourier transform of the trace of the propagator can be written in terms of the nonadiabatic coherent state path integral as follows : @xmath194d[\psi(t ) ] \exp{(is_{t}[{\bf r},\psi,\phi ] + et)}}}\ ] ] here , the quantities @xmath195 are coherent state basis elements and the action @xmath196 is the same as defined in eq.(26 ) . to obtain an approximate quantization rule , only diagonal coherent state matrix elements in eq.(26 ) are considered . eq.(75 ) which relates the fourier transform @xmath191 to the energy spectrum of the system also implies that @xmath191 is singular when the energy @xmath197 is equal to any of the system eigenenergies . this implies that the path integral of the propagator in eq.(77 ) is a maximum for such values of the energy @xmath197 . the propagator also acquires its maximal values when its phase is stationary . consequently , the density of states , @xmath191 , is dominated by contributions to the path integral from stationary phase paths . furthermore , if it is assumed that periodic orbits exist for the combined dynamics of the system , then the density of states in eq.(75 ) can be approximately written as @xmath198 = \sum_{p.o}{c_{k}\exp{[is[t_{k}(e ) ] + iet_{k}(e)]}}\ ] ] where the time @xmath199 is the time period for the classical traversal of the @xmath200th periodic orbit . since , multiple traversals of a given periodic orbit also constitutes a periodic orbit , this sum can be simplified into a sum over periodic orbits @xmath201 which can not be reduced further into smaller sets of periodic orbits : @xmath202 + iet_{\mu}(e)]}}{1 - \exp{[is[t_{\mu}(e ) ] + iet_{\mu}(e)]}}}\ ] ] this expression has poles given by @xmath203\ ] ] this quantization condition can be explicitly written out to be @xmath204 } = 2\pi(n + \frac{1}{2})\ ] ] it should be noted that this quantization condition is valid under the assumption that periodic orbits corresponding to the effective classical system defined by the action @xmath205 exist . for nuclear motion coupled to a large number of electronic levels , finding the periodic orbits of the system is a difficult task . however , for small values of the nonadiabatic coupling , this quantization rule provides an approximate means of estimating the total energy spectrum of the system , given an initial set of born - oppenheimer surfaces . furthermore , if the nonadiabatic transitions couple only a finite set of electronic born - oppenheimer surfaces , this rule may have validity because periodic orbits for the combined nuclear and electronic density matrix variables could exist . this rule can be written in terms of the total connection factors @xmath206 $ ] for the closed path @xmath207 as @xmath208 = 2\pi(n + \frac{1}{2})\ ] ] a special case of the born oppenheimer problem , for which the semiclassical quantization rule is useful , is that of a single vibrational coordinate , @xmath209 , coupled to a set of electronic levels . in this situation , the quantization rule reads @xmath210 in the absence of the nonadiabatic couplings , @xmath211 , the quantization rule gives the exact vibrational energy levels ( assuming that the vibrations are purely harmonic ) . the contribution due to nonzero nonadiabatic coupling can be interpreted as a shifting of the energy levels of the oscillator to account for energy exchange with the electronic bath . if an orbit is executed , nonadiabatic transitions between vibrational states can occur , which leads to energy transfer between the oscillator and the electronic system . this work has presented an exact first principles path integral formulation of quantum nonadiabatic dynamics . the path integral approach developed here has the advantage that it enables computationally tractable stationary phase approximations to the full path integral . previous path integral based approaches to this problem have stationary phase approximations that lead to nonlocal theories of mixed quantum classical dynamics@xmath8 , or require considering a constrained set of paths for the purposes of approximation@xmath10 . the method presented here therefore represents an advance over previous approaches in this regard . a second consequence of this work is that corrections to the quasiclassical nuclear dynamics have been obtained . these corrections reflect underlying geometric features of the electronic hilbert space . the corrections so obtained have the advantage that they can be calculated through straightforward extensions of current mixed quantum - classical simulation methodologies . applications of this approach to study the dynamics of real systems in the presence of conical intersections and other degeneracies in the electronic energy levels are currently being considered . finally , semiclassical quantization rules have been proposed for nonadiabatic dynamics in the quasiclassical limit . these rules are potentially of utility in determining approximate energy spectra when the born - oppenheimer approximation is not valid . a third consequence of this work has been to provide a link between exact quantum theories of nonadiabatic dynamics and semiempirical surface hopping methods . this work establishes a rigorous theoretical basis for the surface hopping method . it has been demonstrated that the fewest switches surface hopping approach can be regarded as a generalized stationary phase approximation to the exact quantum dynamics , in complement to the ehrenfest formulation of mixed quantum classical dynamics . natural extensions of the surface hopping method to include classically forbidden nonadiabatic transitions , as well as the inclusion of nuclear quantum effects through the construction of rigorous multiple spawning type algorithms have been demonstrated . this work also allows for the construction of semiclassical golden rules to describe nonadiabatic transitions . the present work also provides a clear perspective on issues concerning equilibrium between the quasiclassical nuclear and quantum , electronic subsystems . it is clear from the theory presented here that traditional ehrenfest methods do not satisfy equilibrium conditions if only single mean field trajectories are considered . each such trajectory corresponds to a stationary phase solution for the nonadiabatic propagator at a given set of initial conditions . however , thermal properties of the system enter through sampling of initial conditions . hence , if appropriately sampled ensembles of mean field trajectories are chosen , a mixed quantum - classical equilibrium can be established . correct sampling of mean field trajectories will require an initial sampling of the reduced density matrix for the nuclei and should also correctly account for initial phase correlations according to eq.(26 ) . it is possible that observed violations of mixed quantum classical equilibrium by the ehrenfest method are due to an incorrect sampling of either of these contributions . since the surface hopping approximation is obtained by looking at the propagation of the system in each of the states occupied by it , an ensemble averaging of surface hopping trajectories will sample the initial density matrix of the system , and hence this approach and any systematic generalizations based on it can be expected to reproduce the correct mixed quantum classical equilibrium . this has been demonstrated through a separate set of arguments@xmath212 . this work provides a different conceptual perspective on the nature of nonadiabatic dynamics . nonadiabatic couplings can be regarded as geometrical objects which are generalizations of berry s geometric phase . thus , they can be regarded on the same footing as the geometrical phases that arise due to conical intersections@xmath213 . this similarity between nonadiabatic couplings and geometric phases that arise in degenerate systems has also been noted by previous authors@xmath214 . in principle , the methods described in this work do not differentiate between systems which have degeneracies in energy and those which are nondegenerate . therefore , the nonadiabatic path integral formulation presented here is of considerable generality and may be of practical utility in computational studies of nonadiabatic dynamics . a potentially useful approach to study exact quantum nonadiabatic dynamics is to consider quantum monte carlo schemes to evaluate the coherent state path integral . direct applications of quantum monte carlo schemes are likely to face difficulties due to the sign problem . however , simplifications could be made by using an effective single electron ansatz to replace the matrix elements in the action eq.(26 ) . such an ansatz could be rigorously constructed by using the runge - gross theorem@xmath215 which relates the exact one electron time dependent density to the expectation value of the electronic hamiltonian in eq.(26 ) . extensions of the time dependent density functional approach to one electron density matrices can then be used to accurately evaluate the path integral due to the action in eq.(26 ) . this has the added advantage of naturally incorporating time dependent density functional theory with the nonadiabatic propagation of a quantum system . this approach will be explored in later work . although the exact coherent state path integral presented here is a challenge for numerical computation , it is possible to develop numerically more tractable semiclassical propagators that describe the nonadiabatic dynamics as an approximation to the coherent state path integral . one such approach would involve a stationary phase evaluation of the electronic dynamics ( the ehrenfest equations for the electron density matrix ) , while propagating the nuclear motion using a semiclassical initial value representation . more promising applications of this work are in developing better approximate methods to describe quantum nonadiabatic dynamics . one potential application is to develop methods which systematically combine ehrenfest and surface hopping type algorithms . to enable this the propagation of the nuclear dynamics can be divided into sections wherein different basis sets are used depending on physical considerations regarding the nature of nonadiabatic interactions along the nuclear paths . another natural extension is in refining the surface hopping method to include classically forbidden nonadiabatic transitions by approximating the nuclear wavefunction with a gaussian wavepacket . alternatively , more general semiclassical propagation schemes can be used to propagate nuclear dynamics in the discrete representation of the path integral . this would facilitate the inclusion of nuclear quantum effects as well as more accurate treatment of dynamics in the presence of strong nonadiabatic coupling between born - oppenheimer surfaces . the use of coherent states , and the derivation of the path integral presented here is analogous to path integral treatments of quantum spin systems . this analogy with quantum spin systems also forms the basis for classical mapping techniques@xmath216 that have been developed to study mixed quantum classical dynamics . finally , the path integral methods presented here are apparently amenable to perturbative approximations around its stationary phase value . however difficulties remain due to the nature of the nonadiabatic corrections to born oppenheimer dynamics . the nonadiabatic coupling between born oppenheimer levels consists of matrix elements of the nuclear momentum operator . the nuclear momentum operator is unbounded in the hilbert space , and further , the nuclear paths that contribute to the path integral for the system are in general not differentiable . hence , a perturbation expansion in the nonadiabatic coupling element is not well defined in the path integral framework@xmath92 . in the case of ehrenfest dynamics , this issue also manifests through the assumptions made in constructing the coherent state path integral . the assumption made in coherent state path integrals is usually that the paths chosen in the coherent state space are continuous . this is not valid in general for the problem of nonadiabatic dynamics . similarly , extensions of surface hopping type methods will need to account for non - differentiable nuclear paths where the nuclear velocity is not defined . a means of overcoming this problem for the coherent state path integral is to regularize the path integral by considering higher order time derivatives in the action , in analogy to the procedure followed for spin systems@xmath217 . a perturbative treatment of nonadiabatic dynamics that takes these issues into account is being worked on and will be presented at a later date . the major part of this work was done when the author was at yale university . financial support from a yale university graduate teaching fellowship is gratefully acknowledged . the author thanks professor j.c tully for his support and encouragement , and for discussions that motivated this work . the author acknowledges dr.a.v madhav and professor j. simons for critical comments and suggestions regarding the manuscript . the author is grateful to an anonymous referee for remarks and criticism which greatly improved and clarified the content of this paper . the derivation of the stationary phase equations in sec.iii a. is described here . the effective action in the coherent state representation is given by @xmath218 with a lagrangian @xmath219 given by @xmath220 } - \dot{\bf r}\cdot\sum_{mn}{{\bf a}_{mn}w^{*}_{m}w_{n}}\ ] ] for the stationary phase path , the action is stationary when varied with respect to the variables @xmath44 and @xmath221 which determine the path . in other words the stationary phase condition : @xmath222 implies the equations : @xmath223 the first of the three equations above is the usual lagrangian equation of motion for newtonian mechanics which governs the classical dynamics of the variable @xmath35 , while the next two equations render the action stationary with respect to variations of the electronic density matrix . the equations for the variables @xmath224 are obtained by using the appropriate unsymmetrized form of their kinetic energy term ( the fourth term in the right hand side of eq.(a2 ) ) . on applying these equations of motion to the lagrangian , one obtains @xmath225 the second and third equations , eq.(a7 ) and eq.(a8 ) , can be combined to give the liouville von - neumann equations for the electronic density matrix , @xmath226 : @xmath227}\ ] ] further , the identity @xmath228 can be used to simplify the last term in eq.(a6 ) to @xmath229 using this , the equation of motion for the nuclear coordinates becomes @xmath230 } -\sum_{m}{\sigma_{mm}\nabla_{\bf r}e_{m}({\bf r } ) } -\nabla_{\bf r}u({\bf r})\nonumber\\ -\sum_{m , n}{\sigma_{nm}[(\dot{\bf r}\cdot\nabla){\bf a}_{mn } + \dot{\bf r}\times(\nabla\times{\bf a}_{nm})]}\end{aligned}\ ] ] the time derivative of @xmath110 can be rewritten as @xmath231 and hence cancels with the corresponding term on the right hand side . thus , the equation of motion simplifies to : @xmath232}\ ] ] furthermore , the time derivative of the electronic density matrix , @xmath233 , can be replaced by using the liouville - von neumann equations of motion for the electronic density matrix , to obtain @xmath234_{nm}\ } } - \sum_{mn}{\sigma_{nm}\dot{\bf r}\times(\nabla_{\bf r}\times{\bf a}_{mn})}\end{aligned}\ ] ] the first three terms are the standard ehrenfest force acting on the nuclear variables , while the last two terms are new . further , using the vector identity @xmath235 the penultimate term on the right hand side can be simplified and rewritten as @xmath236_{nm}\ } = -i\sigma_{nm}\dot{\bf r } \times({\bf a}\times{\bf a})_{mn}\ ] ] putting this together with eq.(a14 ) and defining an effective magnetic field by @xmath237 the equation of motion for the nuclear coordinate is obtained as @xmath238 this is the equation of motion discussed in the text of the paper , and is the ehrenfest equation of motion with an additional contribution from the `` magnetic field '' , @xmath239 .

we present results from a mosaic of nine _ chandra _ observations of m86 and the surrounding field . we detect three main diffuse components : the virgo icm at.4 kev , the extended halo of m86 at.2 kev , and the cooler central and stripped gas of m86 at.8 kev . the most striking feature is a long tail of emission , which consists of a plume 4north of m86 and two main extensions emanating from the plume . based on the morphology and temperature structure of the tail , we conclude that it is formed by ram pressure stripping of m86 as it falls into the virgo cluster and interacts with the virgo icm , in agreement with earlier work . the tail is 150 kpc in projection , and a simple estimate gives a lower limit on the true length of the tail of 380 kpc , making this the longest ram pressure stripped tail presently known . the total gas mass in the plume ( ) and tail ( ) is about three times that in the core of m86 , which supports the scenario where most of the gas was stripped rapidly and recently . the projected position of the plume can be understood if m86 has an aspherical potential , as suggested by optical isophotes . ram pressure stripping from an aspherical potential can also explain the split `` double tails '' seen in m86 and in other virgo cluster galaxies in the field . the large line - of - sight velocity of m86 ( 1550 with respect to m87 ) , its position relative to the virgo cluster , and the orientation of the tail tightly constrain its orbital parameters . the data are inconsistent with a radial orbit , and imply inner and outer turning radii of kpc and mpc , indicating that m86 is , at best , only weakly bound to the virgo cluster .

0806.0866

m86 ( ngc 4406 ) is a bright elliptical ( e3/s0 ) galaxy in the virgo cluster of galaxies . it is the dominant member of one of the larger subgroups within virgo ( binggeli et al . 1993 ; bhringer et al . 1994 ; schindler et al . its line - of - sight velocity relative to m87 , the dominant member of the virgo cluster , is -1550 km s@xmath6 , much higher than the average cluster velocity dispersion ( smith et al . the x - ray surface brightness distribution is unusual , with a large `` plume '' extending to the northwest from m86 , which was first noticed as part of a survey of virgo cluster galaxies undertaken with the _ einstein observatory _ ( forman et al . 1979 ) . the galaxy has an optical asymmetry that extends in a direction similar to the direction of the plume ( nulsen & carter 1987 ; mihos et al . several authors have interpreted this plume as arising from ram pressure stripping due to strong interactions with the virgo icm ( forman et al . 1979 ; fabian et al . 1980 ; takeda et al . 1984 ; knapp et al . 1989 ; bregman & roberts 1990 ; white et al . 1991 ; rangarajan et al . 1995 ) . elmegreen et al . ( 2000 ) find dust streamers in the core of m86 that connect to the nucleated dwarf galaxy vcc 882 , and suggest a recent interaction between the pair that may have contributed to the asymmetry of m86 s optical isophotes . at x - ray energies , m86 has been observed by _ einstein _ ( forman et al.1979 ; white et al . 1991 ) , _ ginga _ ( takano et al . 1989 ) , _ exosat _ ( edge 1990 ) , _ asca _ ( matsushita et al . 1994 ) , and _ rosat _ ( bhringer et al . 1994 ; rangarajan et al.1995 ) . recently , finoguenov et al . ( 2004 ) presented _ xmm - newton _ observations of m86 and concluded that its unusual morphology is due to an interaction with an x - ray filament rather than the virgo icm . however , their conclusions were based on the large separation between m86 and m87 along the line of sight of @xmath7 mpc reported by neilsen & tsvetanov ( 2000 ) , which is inconsistent with the more recent result from mei et al . ( 2007 ) who find a separation of @xmath8 mpc . we report here on a mosaic of nine _ chandra _ observations of m86 and the surrounding field , totaling 240 ksec of exposure . the observations and data reduction techniques are described in [ sec : obs ] . the x - ray image is presented in [ sec : img ] , and results on temperature and abundance structure from spectral analysis are given in [ sec : spec ] . in [ sec : discuss ] , we discuss some of the more interesting features of m86 , and place constraints on its orbit . in particular , we give new results on the extent of the stripped tail , and measure the density profile of an x - ray brightness edge seen to the southeast . our results are summarized in [ sec : summary ] . we assume a distance to the virgo cluster of 16 mpc throughout , consistent with the latest results from mei et al . ( 2007 ) , which gives a scale of 0.08 kpc/ for @xmath9 , @xmath10 , and @xmath11 km s@xmath6 mpc@xmath6 . all error ranges are 90% confidence intervals , unless otherwise stated . table [ tab : obs ] summarizes the nine _ chandra _ observations of m86 , which completely cover the center of the galaxy , as well as the plume and long tail . all data were reprocessed from the level 1 events files using the latest calibration files ( as of ciao3.3 ) . cti and time - dependent gain corrections were applied where applicable . lc_clean was used to remove background flares . after periods with obvious flares were removed , the mean rate was calculated for some observed quiescent time interval , and the program was re - run , forcing the global mean to equal this quiescent rate . for observations without obvious flares , time bins that were not within 3@xmath12 of the mean were discarded . the final cleaned exposure times are given in column ( 5 ) of table [ tab : obs ] . the emission from m86 and the surrounding virgo cluster fills the image field of view for each observation . we therefore used the standard caldb blank sky background files appropriate for each observation , normalized to our observations in the 10 - 12 kev energy band . we combined background maps for each pointing separately , and generated exposure maps to account for the pointing offsets and the different ccd responses . to generate exposure maps , we assumed a mekal model with @xmath13 kev , galactic absorption , and abundance of 30% solar at a redshift @xmath14 , which is consistent with typical results from detailed spectral fits ( see [ sec : spec ] ) . the exposure corrected , background subtracted , smoothed mosaic image is shown in figure [ fig : fullimg ] . several patches of diffuse emission are seen in m86 , as is an impressive tail extending to the northwest . to enhance the visibility of the diffuse emission we created an image with bright point sources removed . for each _ chandra _ pointing , regions containing point sources were `` filled in '' using a poisson distribution whose mean was equal to that of a local annular background region . the resulting smoothed image ( figure [ fig : smoimg ] ) shows : * a long tail extending to the nw , first detected by forman et al.(1979 ) . the tail is bifurcated at its base , with the brighter ( northern ) extension at a position angle of @xmath15 ( measured from north to east ) and the shorter extension at @xmath16 . at 18.4(88.4 kpc ) from the center of the bright emission at the base of the tail ( or 21.1 ( 101.5 kpc ) from the center of m86 ) , the tail turns directly north for 6.8 ( 32.5 kpc ) . it then curves to the west at 26.7 ( 128.1 kpc ) from m86 , and finally is no longer detected at 30.6 ( 150 kpc ) . there is evidence for a fainter , parallel tail along the northeastern edge of the tip of the bright tail , discussed below . * a large plume of emission , directly north of m86 , at the root of the long tail ( noted previously by forman et al . the tail emanates from this clump , rather than from m86 itself as one would expect for a continuous stripping process . * a sharp boundary along the northeastern edge of the tail , and an apparent `` void '' in the diffuse emission just north of the tail . this is in stark contrast to the southwestern edge of the tail , where the surface brightness falls off smoothly . rangarajan et al . ( 1995 ) first noted this void , and suggested that it is an evacuated mach cone left from m86 s passage through the outer regions of the virgo icm . * an extended halo of x - ray emission associated with m86 with an edge 13 ( 62.4 kpc ) to the southeast . this edge also is visible in _ rosat _ observations ( see figure [ fig : rosat ] ) . * an asymmetry in the diffuse emission near the core of m86 , with an extension to the south . rangarajan et al . ( 1995 ) suggest that this is a high pressure region , formed as the leading edge of m86 plows into the virgo icm . * several nearby galaxies . the elliptical galaxy m84 , west of m86 , shows complicated structure in the diffuse emission in the core related to the radio outburst , and a tail extending to the south ( finoguenov & jones 2002 ) . also visible are ngc 4388 ( south - southwest of m86 ; beckmann et al . 2004 ) and ngc 4438 ( east of m86 ; machacek et al . 2004 ; vollmer et al.2005 ) . both also have diffuse tails , which likely indicate their projected direction of motion with respect to the virgo icm . most tails split into two main streams . a small patch of faint emission also can be seen in the region of the disk galaxy ngc 4402 ( crowl et al . 2005 ) . for comparison , the ( heavily binned ) _ chandra _ mosaic image is plotted alongside the _ rosat _ and _ dss _ images in figure [ fig : rosat ] . the _ rosat _ image clearly shows the distribution of diffuse emission on larger scales . in particular , the contribution from the m87 halo is seen in the southeast , as is a large halo of diffuse emission around m86 itself , outside the brightest central regions . these components are individually analyzed in our detailed spectral fits ( see [ sec : dspec ] ) . results from these fits indicate that the extended m86 halo is composed of group gas associated with m86 , at a temperature of @xmath17 kev . a close up view of the stripped tail , with scaling and binning chosen to show its structure more clearly , is presented in figure [ fig : tail ] . the twisting and curving of the tail at the faint tip may be caused by turbulence or similar `` weather '' in the virgo icm . the tail is clearly bifurcated at its base . additionally , a fainter tail , split from the first , is barely visible along the northern edge of the faint end of the main tail . to test the significance of this feature , we extracted the total count rate in evenly spaced bins across the width of the box shown in figure [ fig : tail ] . figure [ fig : tailproj ] shows two distinct peaks . if we take an average count rate from the highest three bins in each peak and compare it with the average count rate from the lowest two bins in the gap , we find that the gap is significant at about 2.5@xmath12 . as described below in [ sec : displace ] , these double streams , as seen clearly at the base of the tail and possibly at the faint end of the tail , as well as in other virgo cluster galaxies in the field , may be explained as stripping from an inclined aspherical potential . in summary , the x - ray images show three diffuse emission components : the virgo icm ( centered on m87 ) , the extended m86 halo , and the core and tail of m86 . we generate a temperature map as a guide for detailed spectral fitting to disentangle the various components . we assume a galactic absorption of @xmath18 @xmath19 throughout . allowing the absorption to vary did not significantly improve any of the spectral fits . the temperature map was derived using a method employed by osullivan et al . ( 2005 ) and maughan et al . for each temperature map pixel ( 19.7/pix ) , we extracted a spectrum from a circular region containing 1000 net counts ( after subtracting the blank sky background and a component to account for emission from m87 , see [ sec : dspec ] ) , up to a maximum radius of 3.3 ( on the order of the chip size ) . this large maximum radius was used to obtain sufficient source counts to extend the temperature map to faint regions of the tail . the resulting spectrum was fit in the 0.6 2.0 kev range with an absorbed apec model using xspec . data from back - side- ( bi ) and front - side - illuminated ( fi ) ccds were treated as separate data groups for each observation . for each fit , the abundance was fixed at 26% solar ( consistent with most of the detailed spectral fits ; see table [ tab : spectra ] ) . the resulting temperature map is shown in figure [ fig : tmap ] ( middle ) . regions that did not have at least 1000 source counts within the maximum 3.3 radius were excluded from the map , as were regions in the southeast that were dominated by virgo cluster emission . for comparison , we also show the tessellated temperature map in figure [ fig : tmap ] . each bin was fit using only counts from that area , so the extraction regions are well - defined . the bins were generated using the algorithm provided by diehl & statler ( 2006 ) , which is a generalization of cappellari & copin s ( 2003 ) voronoi binning algorithm , and requiring roughly 1100 net counts per bin . each bin was fit with a single apec model , without a component to model the virgo icm background , in contrast to the `` smoothed '' temperature map . the size of the bins roughly indicate the size of the extraction regions for the smoothed temperature map pixels in the same area . the tessellated and smoothed temperature maps are in good agreement , except in the regions far from m86 where the contribution from the virgo cluster is more important . in these temperature maps , the remarkably sharp boundary between the cool galaxy gas and that of the cluster on the northern edge of the tail is apparent ( figure [ fig : tmap ] ) . at least four distinct cold clumps can be seen : one centered on m86 , one 1.41 ( 6.8 kpc ) south of m86 , one elongated clump 2.45 ( 11.7 kpc ) east of m86 , and a larger cool plume 3.63 ( 17.4 kpc ) to the north . each clump corresponds to a bright peak in the x - ray image . additionally , the cooler tail extends to the northwest . to the southeast , in the region of the brightness edge discussed in [ sec : halo ] , the gas is somewhat hotter ( @xmath20 kev ) in the southern half of this feature than in the northeastern half at the same distance from m86 . this temperature difference is detected in spectra from non - overlapping regions in these two areas , indicating that the difference is not an artifact of the technique used to generate the temperature map . more detailed fits are discussed in [ sec : dspec ] . based on the derived temperature map and the x - ray image , we defined 23 regions for detailed spectral analysis ( see figure [ fig : regmap ] ) . to include data from a particular observation in the spectral analysis , we required that it contribute at least 200 counts ( each data group was binned such that there were at least 40 counts per bin ) . a summary of the regions and spectral model for each region are given in table [ tab : spectra ] . r1 - r4 are centered on the individual clumps of emission near m86 , while r5 -r10 trace the long tail nw of m86 ( hereafter regions will be labeled as r# , with the number defined by column 1 of table [ tab : spectra ] ) . r11 roughly corresponds to the `` shocked region '' indicated by finoguenov et al . ( 2004 ; see their figure 7 ) . there are several regions to either side of the m86 tail . r17 , r18 , r19 , & r23 measure the large scale diffuse components from m86 and m87 , outside the central region where the cool gas is observed ( note that r23 is a combination of five disjoint regions , see figure [ fig : regmap ] ) . finally , r14 - r16 contain m84 and its tail of diffuse emission . each region was initially fit with an absorbed apec model over the 0.6 - 5.0 kev range , with an additional apec component to model emission from the virgo icm as in [ sec : tmap ] , with the abundance allowed to vary . additional components were added , as needed , to improve the fits ( see table [ tab : spectra ] . for example , a single temperature model gave large residuals at high energies when fit to the regions corresponding to the cores of m86 and m84 ( r1 and r14 , respectively ) . since each of these regions is expected to have a contribution from unresolved x - ray point sources , a power law component was added to the model , which greatly improved the fits ( and gave a photon index consistent with @xmath21 , as expected , see e.g. sarazin et al . 2003 ) . for regions where this power law component could not be well - constrained , we fixed the photon index at @xmath22 . nine regions showed an improved fit when another apec model was added ( in such cases the abundances of each component were constrained to vary together ) . data from each pointing were treated as separate data groups , with normalizations for each pointing allowed to vary independently . as an example , the two temperature plus power law fit to the spectra for the core of m86 ( r1 ) is shown in figure [ fig : m86spec ] . each of the three main components of the diffuse emission ( the virgo icm , m86 group gas , and the cooler central m86 gas ) was detectable in some subset of the regions shown in figure [ fig : regmap ] . in r13 it was possible to fit all three components simultaneously . the virgo and m86 halos cover most , if not all of the fov . it was therefore necessary to include these components when doing detailed spectral fits , even in regions where they could not be modeled directly due to inadequate statistics . to account for the varying virgo contribution , we took the normalizations of the best fit apec models to r18 and r23 and linearly interpolated across the field based on the distance from the center of m87 . although this only gives a rough estimate , we note that it does not affect our results significantly . for regions closer to m87 ( e.g. , r19 ) the interpolation gives a value very close to that measured from r18 , which is an appropriate local background . similarly , regions near the faint end of the m86 tail ( e.g. , r10 ) have normalizations close to that of r23 , which again , is an appropriate local background for this area . for most regions of interest , m86 is bright enough that the contribution from virgo is negligible , and the accuracy of interpolating is therefore sufficient . a similar approach was used to model the contribution from the halo around m86 . although , in general , the normalization of this component decreased with distance from the core of m86 , the variation across the field was more complicated than for virgo , which is expected since this gas is in the process of being stripped from m86 . we included a fixed component , with @xmath23 kev , abundance set equal to that of the main component , and variable normalization , in regions where the m86 halo temperature could not be determined directly ( see column 2 of table [ tab : spectra ] ) . the effects of accounting for this component can be seen , for example , by comparing lines 1 & 2 in table [ tab : spectra ] . including the hotter 1.2 kev component results in a lower best - fit temperature for the cooler gas , and gives a better statistical fit to the data . in general , our results are consistent with finoguenov et al.(2004 ) , who derive x - ray temperatures and abundances for the central region , including m86 and the plume from _ xmm - newton _ observations . we note that for the cores of m86 and m84 ( r1 and r14 in table [ tab : spectra ] ) the abundances are poorly determined , despite the large number of net counts . using a vapec model to allow for non - solar abundance ratios did not improve the fits . the failure of these models may be due to abundance variations within the defined regions . for instance , the abundance map given by finoguenov et al . ( 2004 ) shows significant variation near the core of m86 , even on relatively small scales . additionally , the models do not allow for multiphase gas , although the temperature most likely varies significantly in the cores . using the detailed spectra , we can derive the approximate mass in cool gas for the base of the tail ( the `` plume '' ) and the core of m86 . using the temperature and abundance measured in r4 , and assuming a prolate spheroid geometry with major and minor axis lengths of 12.3 kpc and 7.9 kpc , we find a density of @xmath24 @xmath25 and a mass @xmath26 . similarly , in the core of m86 , we find @xmath27 @xmath25 and @xmath28 within a sphere of radius 9.6 kpc . assuming a cylindrical geometry for the tail , and considering only the cool gas components in r5 - r9 , we find a typical density of @xmath29 @xmath25 and a total gas mass of @xmath30 . the mass in stripped gas is therefore more than three times the gas mass in the core , which supports the scenario where a significant fraction of the m86 ism is removed rapidly due to ram pressure stripping ( takeda et al . 1984 ) . rangarajan et al . ( 1995 ) find similar results for the plume and the core . however , we find consistent abundance values in the core and in the plume within the errors , while rangarajan et al . ( 1995 ) find a higher abundance in the plume . this disagreement is likely due to the multiphase model we use , which includes a separate component for the 1.2 kev m86 halo emission . the most striking feature in figure [ fig : smoimg ] is the long tail of emission extending to the nw from m86 . m86 has a line - of - sight velocity of -244@xmath31 km s@xmath6 , while m87 s is 1307@xmath32 km s@xmath6 ( smith et al . 2000 ) . therefore , m86 is traversing the virgo cluster at @xmath33 km s@xmath6 ( about mach 2 for @xmath34 kev ) . the tail naturally forms due to ram pressure stripping of the m86 corona by the virgo cluster icm . finoguenov et al . ( 2004 ) suggested that the tail formed due to interactions with a filament rather than the virgo icm , though their result was based on older distance measurements that placed m86 outside the virial radius of the virgo cluster . if we assume that m86 is bound to the virgo cluster , its large line - of - sight relative velocity allows us to constrain the separation between m86 and m87 . using the m87 mass profile detailed in [ sec : orbit ] , we find that a free - fall velocity of 1500 km s@xmath6 corresponds to a separation from m87 of 0.5 mpc ( the projected separation is 0.35 mpc ) . this is an upper - limit , since the total relative velocity must be at least as large as the line - of - sight relative velocity . this separation is consistent with recent surface brightness fluctuation distances of virgo cluster galaxies , which give a distance between m86 and m87 of @xmath35 mpc ( mei et al.2007 ) . we also can place a lower - limit on the length of the long stripped tail using the mass profile of m87 and the line - of - sight velocity . the maximum free - fall velocity of m86 from infinity at the projected separation of 0.35 mpc is about 1680 km s@xmath6 . using this as an upper - limit on its current 3d velocity , we find that the angle between the direction of motion of m86 and the line - of - sight is @xmath36 . assuming that the stripped tail is aligned with m86 s current direction of motion , and given that the length of the tail in the plane of the sky is 150 kpc ( 0.51@xmath37 ) , we find a lower - limit on the actual length of the tail of @xmath38kpc , making this the longest ram pressure stripped tail presently known . the long stripped tail originates in the plume of diffuse emission located directly north of m86 . for a continuous stripping process , one would expect the tail to extend from m86 itself . this separation between m86 and the plume can be explained by rapid recent stripping , in which a significant fraction of the remaining gas in m86 is rapidly stripped when the ram pressure stripping condition is met . a detailed discussion of this process is given in [ sec : displace ] as seen from figure [ fig : regmap ] and table [ tab : spectra ] , the stripped tail shows a general trend of cooler gas in and near the plume ( with temperatures in the 0.80.85 kev range ) and warmer gas at the tip ( in the 0.91.2 kev range ) . the temperature structure of the tail is consistent with a ram pressure stripping model , where , as m86 falls into the virgo cluster , the hotter , higher entropy group gas is stripped first due to interactions with the virgo icm , followed by the cooler , lower entropy m86 ism , which is removed rapidly once the stripping condition is met . the _ chandra _ and _ rosat _ x - ray images ( figure [ fig : rosat ] ) show an extended halo associated with m86 , with a sharp brightness edge to the southeast in the direction of m87 . since m86 is traversing the virgo cluster supersonically ( see [ sec : tail ] ) we expect a shock to be driven in the virgo icm , possibly producing a brightness edge similar to that seen in the southeast . we attempted to measure the density jump associated with the shock in the following way . we extracted the _ chandra _ 0.52.0 kev surface brightness profile of this edge in two roughly equal sectors since the morphology of this feature is irregular in the south ( see figure [ fig : rosat ] ) , and since the gas temperature differs in these two regions ( see [ sec : tmap ] ) . the emission measure profile for the northern sector is shown in figure [ fig : nedge ] . distance is measured from the center of curvature of the apparent edge , which is 10 kpc ( 2.16 ) east of m86 . this profile was fit with a spherical gas density model consisting of two power laws . the free parameters were the normalization , the inner ( @xmath39 ) and outer ( @xmath40 ) slopes , the position of the density discontinuity ( @xmath41 ) , and the amplitude of the jump ( @xmath42 ) . we assumed that the gas is isothermal with @xmath13 kev and that the abundance is constant at 30% solar , consistent with results from spectral fits ( see [ sec : spec ] ) . for the best fit model ( see figure [ fig : nedge ] ) , we find @xmath43 , @xmath44 , @xmath45 kpc , and @xmath46 . similar results are found from the southern sector , but with larger errors . the lack of a well - defined edge is consistent with our findings for the orbit of m86 ( see [ sec : orbit ] ) . in particular , for all of the likely orbits , m86 is moving to the southeast , such that our lines of sight pass through the mach cone . to see a sharp edge , our line of sight must be tangent to the shock front . knowing the orbit of m86 is the key to understanding its interaction with the virgo cluster . the ram pressure stripped tail reveals the motion of the galaxy on the sky . the length and direction of the tail , together with the large line - of - sight speed of m86 , constrain its orbit . the large luminosity of m86 suggests that it dominates the associated in - falling subgroup ( schindler et al . 1999 ) . in the following , we assume m86 is bound to the virgo cluster . for the purpose of calculation , the gravitational potential of the virgo cluster is treated as a spherical nfw potential ( navarro , frenk & white 1997 ) , with a virial radius of 1.3 mpc ( evrard , metzler & navarro 1996 ) and a concentration parameter of 4.5 ( neto et al.2007 ) , appropriate for a @xmath47 kev cluster . the gravitating mass was normalized to match the total mass within 320 kpc of @xmath48 ( schindler et al . 1999 ; scaled to a distance of 16 mpc ) . for this potential , the escape speed from a radius of 351 kpc ( the projected separation of m86 from m87 ) is 1677 @xmath3 , not much larger than m86 s line - of - sight speed of 1550 @xmath3 relative to m87 . the empirical mass distribution of schindler ( 1999 ) gives the same escape speed at 351 kpc as our model potential . we note that the gravitational potential of the virgo cluster is not well constrained at distances from the cluster center comparable to or greater than its virial radius . furthermore , the truncated nfw potential is highly simplified , ignoring mass beyond the virial radius , departures from spherical symmetry , and the dynamic state of the cluster due to continuing in - fall . since some of the limits we derive here are sensitive to the poorly known potential at large radii , they should be treated as indicative rather than quantitative ( outside the context of the model ) . the material in this section is supplemented by a more detailed discussion in appendix [ ap : orbit ] . we consider radial orbits first , as suggested by previous studies ( e.g. , forman et al . 1979 ; white et al . 1991 ) . in order for the gas tail to point away from the cluster center , the galaxy must be inbound . it can easily be shown that the line - of sight velocity , @xmath49 , is maximized for some radius @xmath50 greater than the observed separation @xmath51 ( see appendix [ ap : orbit ] ) . for a marginally bound ( zero energy ) radial orbit in the potential described above , the maximum value of @xmath52 occurs when m86 is close to twice its projected distance from the cluster center , giving @xmath53 , less than the observed value of @xmath54 . thus , unless m86 is significantly unbound from the virgo cluster ( or the potential is incorrect ) , its line - of - sight speed is inconsistent with radial and nearly radial orbits . it is convenient to specify more general orbits in terms of their inner and outer turning radii , @xmath55 and @xmath56 , respectively . we constrain the possible orbits of m86 by placing limits on these parameters . coarse limits can be placed on these radii by considering energy arguments alone . assuming that m86 is bound to the virgo cluster , we find @xmath57 kpc and @xmath58 mpc , or @xmath59 times the virial radius of the virgo cluster ( see appendix [ ap : orbit ] ) . these limits can be further restricted by considering the range of possible viewing directions for each point on an orbit . there can be zero , two , or four possible viewing directions that would place m86 at the observed separation from the cluster and give it the observed line - of - sight velocity ( see appendix [ ap : orbit ] ) . figure [ fig : mb ] shows a range of marginally bound orbits ( @xmath60 ) , with regions that meet these conditions marked in color . the range of an orbit where these conditions are met shrinks as @xmath56 decreases , i.e. , as the orbit becomes more tightly bound . for the marginally bound orbits , the full range of @xmath55 for which @xmath49 can attain its observed value is @xmath61 kpc ( the range represented in figure [ fig : mb ] ) . for more tightly bound orbits , the acceptable range of @xmath55 is reduced . the lower limit on @xmath56 is increased if we consider the extent of the ram pressure stripped tail . the projected orbit must extend to at least the distance ( @xmath62 kpc ) that the tail projects beyond m86 in the direction away from the cluster center . locations on the orbits where the line - of - sight speed can attain @xmath63 and these conditions also are met are shown in red in figure [ fig : mb ] . we see that the range of possible locations for m86 on these orbits is tightly constrained . we repeated this analysis for a number of values for the outer turning radius . the range of potential orbits and locations for m86 shrinks with decreasing @xmath56 and no orbits were found to meet these criteria for @xmath64 mpc ( 6.3 virial radii ) . figure [ fig : projmb ] shows projections onto the sky of the marginally bound orbits ( corresponding to the midpoints of the red regions of figure [ fig : mb ] ) overlaid on the _ chandra _ 0.5 2 kev image , with m86 at the representative location for each orbit . it is evident that the possible orbits for m86 can be pruned further . for example , the orbit on the lower right in figure [ fig : projmb ] does not reach far enough north to produce the remote part of the gas tail . at the other limit , the position angle of m87 measured from m86 is @xmath65 ( east from north ) , generally eastward of these orbits . overdense stripped gas tends to fall towards the cluster center , in that direction , so that some stripped gas can reasonably lie to the east of the orbit . stripped gas can not lie west of the orbit . thus the orbits that cross the region to the east of the tail and north of m86 , where there is no sign of stripped gas , are unlikely candidates for the orbit of m86 . note that the orbits in figure [ fig : projmb ] are not simply related to their inner turning radii . the orbit corresponding to the smallest value of the inner turning radius appears second from the right at the top of the figure . at first , orbits for increasing values of @xmath55 lie to the left of this , but at a value of @xmath55 approaching ( but less than ) @xmath51 , the acceptable viewing direction flips from outward to inward ( i.e. , the acceptable sign of @xmath66 changes , see appendix [ ap : orbit ] ) . at this point , the projected position of the orbit shifts from leftmost to rightmost at the top of figure [ fig : projmb ] . whereas the apparent location where the orbit passes through the virial radius was moving outward , it now moves inward as @xmath55 increases , coming into the field of view for the largest values of @xmath55 here . this results in disjoint ranges of possible orbits for m86 . thus the marginally bound orbits with inner turning radii close to @xmath67 kpc and those with inner turning radii in the range @xmath68 kpc are best suited to model the orbit of m86 . applying the same criteria to the more tightly bound orbits yields the collection of possible orbits shown in figure [ fig : m86orb ] . we have not attempted to be highly selective or exhaustive , since that requires a well defined model for the formation of the gas trail . for values of the outer turning radius in the lower end of the acceptable range , the shrinking range of possible orbits and locations excludes orbits with inner turning radii @xmath69 . all of the acceptable orbits for m86 are weakly bound to the virgo cluster , with the most tightly bound having outer turning radii at @xmath70 mpc . all of the possible locations for m86 lie only a little farther from m87 than m86 does in projection . all are close to the plane of the sky , ranging from 167 kpc closer than m87 to 263 kpc farther than m87 from the sun at the extremes . m86 must also be close to the pericenter of its orbit . the direction of motion of m86 is close to our line of sight , within @xmath71 @xmath72 of it , for all of the cases illustrated . m86 is traversing the virgo cluster supersonically ( @xmath73 @xmath3 alone is almost twice the sound speed @xmath74 , which is taken to be 850 @xmath3 ) . therefore , we expect a shock to be driven in the virgo icm . for the orbits we derive , the angle of inclination is smaller than the opening angle of the mach cone ( @xmath75 ) and we should not expect shock fronts to be visible on the plane of the sky ( cf . rangarajan et al . this is illustrated in figure [ fig : mach ] , which shows the marginally bound orbit with @xmath76 kpc . each circle is drawn centered on a point where m86 was at a time @xmath77 in the past , with a radius of @xmath78 . this is done at equally spaced times to indicate the shape of the mach cone . the absence of caustics in figure [ fig : mach ] shows that we would not see the shock front in projection . ( while this statement is accurate , except possibly in the small region where the shock front is highly supersonic , the diagram is schematic . for example , if m86 were coming directly towards the earth , radii of sections of the mach cone would be @xmath79 larger than drawn . ) our lines of sight through the compressed gas behind the shock are longest where the circles pile up to the southeast of m86 , consistent with the enhancement in x - ray surface brightness seen @xmath80 southeast of m86 ( figure [ fig : mach ] ) . the prominent plume of gas lying 3 4(@xmath81 kpc ) north of m86 , noted previously ( e.g. , forman et al . 1979 ) , appears to be physically separated from the dense gas remaining in the galaxy and it is centered some distance east of the orbits illustrated in figure [ fig : m86orb ] . the removal of a significant fraction of the ism in a single blob is expected from rapid ram pressure stripping ( takeda et al . 1984 ) . a highly simplified model for ram pressure stripping treats the plume as a single particle subject to gravity and drag due to its motion through the surrounding gas . allowing for buoyancy , the net force on the plume due to gravity is @xmath82 , where @xmath83 is its density , @xmath84 is its volume , @xmath85 is the density of the ambient icm , and @xmath86 is the local acceleration due to gravity . the drag force on the plume is @xmath87 , where @xmath88 is the drag coefficient , @xmath42 is the plume cross section , @xmath89 is its velocity , and @xmath90 . thus , the net acceleration of the plume is @xmath91 { { \bf g}}- [ { c_{\rm d}}{\rho_{\rm e}}a / ( \rho v ) ] v { { \bf v}}$ ] . two of the main parameters of this model are the density contrast , @xmath92 , and the factor @xmath93 . for the density of the icm , @xmath85 , we use the beta model of schindler et al . ( 1999 ; @xmath94 , core radius @xmath95 ) . on the grounds that the density of the plume is likely to be lower now ( due to its stripping and ejection from the confining potential of m86 ) , we set its density contrast at the projected radius of m86 to be 18 ( cf . @xmath81 from the numbers above , see [ sec : dspec ] ) and treat the density of the plume as constant . the other factor is determined by treating the plume as a sphere with a constant radius of 10 kpc , with @xmath96 . as above , this radius is a little larger than the observed radius . to simulate stripping , a gas blob is placed at the location and velocity of m86 at the time it is released from the cluster virial radius . the orbit of the blob is followed , subject to the gravitational acceleration of the cluster and the moving galaxy . this model can not account for the current location of the plume if the gravitational potential of m86 is spherical . the path of the blob is determined by the competition between gravity and drag . if the drag is large , the blob is ejected from m86 early on its orbit , and slows quickly until it is falling towards the cluster center at its terminal speed , and ultimately lies farther away from m86 than observed . its early ejection also leaves it too far back along the orbit of m86 . reducing the relative significance of the drag causes the blob to be ejected later , bringing it closer to its observed position but not far enough from the orbit in the direction of the cluster center . this issue can be resolved if the gravitational potential of m86 is aspherical . consider a small gas blob in an aspherical galaxy moving at an inclined angle through the icm , illustrated schematically in figure [ fig : inclined ] . while ram pressure is insufficient to eject the blob , it is driven to an equilibrium position where ram pressure is balanced by gravity . since the direction of the gravitational field must oppose the drag , the external flow must be perpendicular to the equipotential surface . the equilibrium position therefore lies at a location away from the axis passing through the center of the galaxy and parallel to the flow ( e.g. , near the point labeled `` equipotential '' in figure [ fig : inclined ] ) . as the drag starts to overwhelm gravity and displaces the blob in the direction of the flow , the equipotentials the blob encounters tilt , so that it is subject to a component of the gravitational force directed away from the axis . because forces in the direction of the flow are nearly balanced , this off - axis component of gravity drives the blob farther away from the axis of the flow . thus , a blob stripped from such a galaxy tends to emerge away from the axis of the flow . treating the gas as a fluid , interstellar gas that is pushed inward by ram pressure along the leading edge of the galaxy flows out preferentially along the long axis of the potential , in the direction of the weakest gravitational force . the model for the aspherical gravitational potential of m86 used here has the form @xmath97 , where @xmath98 / w$ ] is a minor modification of the standard nfw form ( @xmath99 / w$ ] in the same notation ) . the extra factor of @xmath100 , where @xmath101 is the concentration parameter , makes the total mass converge , avoiding the need to truncate the mass distribution at the virial radius ( which would create problems for our fourth order integrator ) . the coordinate @xmath102 , where @xmath103 is the nfw potential scale length and @xmath104 is chosen to give the desired ellipticity . the additional 1 under the square root flattens the potential at small @xmath50 , providing a better model for the gravitational force on an extended gas blob ( of size comparable to @xmath103 ) when it is close to the center of the galaxy . this modification also makes the integrator behave better near the center of the galaxy . the virial radius of m86 was set to 100 kpc and its concentration parameter to 8 , roughly the values expected for a massive galaxy . the normalizing factor , @xmath105 , was expressed as @xmath106 , so that @xmath12 is a rough measure of the line - of - sight velocity dispersion for m86 . the value used below is @xmath107 . we note that for the model what matters is the ratio of the drag force on the blob to the binding force of the m86 potential , allowing for a trade off between the choice of potential and the density ratio and size of the blob . as a result , the choice of potential is not very critical . the two remaining parameters are the axial ratio , @xmath108 , and the orientation of m86 . consistent with its s0/e3 classification , optical images show that m86 is highly flattened on scales comparable to the projected distance of the plume ( e.g. @xmath109 for the isophote with @xmath110 in the deep image of nulsen & carter 1987 ) . based on the discussion above , the potential of m86 would need to have a long axis pointing roughly northward from our line - of - sight . therefore , we assume it is oblate , using @xmath111 . the orientation of m86 is then determined by the direction of its minor axis . from the optical images , this has a position angle of @xmath112 on the sky . however , its tilt with respect to the plane of the sky is unknown . we have set the minor axis to point towards us in the north , at an angle of @xmath113 from the plane of the sky this is roughly the orientation that maximizes the transverse displacement of the blob . figure [ fig : blob ] shows the path of the blob on the sky for this set of model parameters for the orbit with @xmath56 = 9.1 mpc and @xmath55 = 314 kpc . in this model , the blob is currently slower than m86 by 355 @xmath3along our line of sight and trails it by 38 kpc . around the edges of the gas halo in m86 , where the shear in the external flow is strong ( figure [ fig : inclined ] ) , if the effective viscosity is high , viscous stresses pull the interstellar gas out of the galaxy . alternatively , if the viscosity is low , shear instabilities mix the interstellar gas with the icm , also stripping it from the galaxy ( nulsen 1982 ) . this stripping is aided by the low pressure due to the bernoulli effect around the edges of the galaxy , which tends to pull gas into the path of the flow . this process works around all edges of the inclined galaxy , as seen from the direction of the flow . however , it is expected to be greatest at the leading and trailing edges . the shear in the external flow is expected to be greatest near the leading edge , favoring stripping there . the large ram pressure at the leading edge displaces the edge of the interstellar gas deeper into the potential of the galaxy , pushing it towards the trailing edge of the galaxy , where the pressure is lower . thus , gas at the trailing edge sits higher in the gravitational potential of the galaxy , favoring its removal from the galaxy . this may account for the apparent double streams of gas seen trailing m86 and several other galaxies in the composite image . `` viscous '' stripping from the main body of m86 as well as the plume can explain the broad features of the gas trail , though modeling the finer features , such as the smaller blobs of gas lying to the east of the galaxy , requires a more detailed treatment of the gas dynamics ( e.g. , numerical simulations ) and is beyond the scope of this paper . the proton mean free path due to coulomb collisions is given approximately by @xmath114 pc , where @xmath115 is the gas temperature in kev and the electron density is @xmath116 . in terms of this , the reynolds number is @xmath117 , where @xmath118 is the flow speed , @xmath51 is the sound speed , and @xmath119 is a relevant length scale . taking @xmath120 and @xmath121 kpc gives @xmath122 in the 2.4 kev icm , but @xmath123 for the gas in the 0.77 kev plume ( and interstellar gas ) if it is exposed directly to the external flow . these values suggest that the external flow can be largely laminar , while the flow in the cool interstellar gas is relatively turbulent . further complicating matters , the effective mean free path may be significantly smaller than the coulomb mean free path ( e.g. , schekochinin et al . the reynolds number for this flow is therefore not well defined , due to the range in gas temperature and uncertainty in the mean free path . we have argued that the plume and long tail of m86 formed due to ram pressure stripping forces generated as m86 falls into the virgo cluster . several studies have found a similar interpretation for the formation of this feature ( e.g. , forman et al . 1979 ; fabian et al.1980 ; rangarajan et al . 1995 , however see bregman & roberts 1990 ; finoguenov et al . we concentrate on these main results : * the plume and long tail observed in the diffuse emission are created by ram pressure stripping as m86 falls into the virgo cluster . the tail is 150 kpc in projection ( a simple estimate , which assumes free - fall velocity for m86 and an nfw potential for m87 , gives a lower - limt on the true length of the tail of 380 kpc ) , making this the longest ram pressure stripped tail presently known . * based on the x - ray spectra , we detect three distinct components associated with the m86/virgo cluster system : the virgo icm , with @xmath124 kev ; the extended halo of m86 , with @xmath125 kev ; and the cooler central and stripped gas of m86 , with @xmath126 kev . the temperature structure of the tail is consistent with ram pressure stripping , where the higher entropy m86 halo gas is stripped first and deposited in the tip of the tail , and the lower entropy m86 ism is stripped more recently , constituting the base of the tail and the plume . * the large line - of - sight velocity of m86 , and its position relative to the virgo cluster , tightly constrain its orbit , especially if it is assumed that the gas tail traces the orbit . in particular , the observations are inconsistent with a radial orbit . we show that m86 is at best only marginally bound to the virgo cluster , with an inner turning radius on the order of 300 kpc as expected from its recent in - fall . our best - fitting orbital model requires that m86 be close to m87 , less than 167 kpc closer than or 263 kpc farther than m87 along our line of sight , which is consistent with the most recent distance estimates based on surface brightness fluctuations ( mei et al . 2007 ) which give a line - of - sight separation of @xmath127 mpc . * the prominent plume of gas lying 3 4north of m86 appears to have been rapidly driven from m86 by ram pressure stripping . the projected position of the plume , which does not lie directly on our best - fit model orbit for m86 , can be understood if m86 has an aspherical potential ( as suggested by optical isophotes ) . if m86 moves through the virgo icm at an inclination angle relative to the local flow , the gas at the trailing edge is more easily stripped , thereby displacing the gas from the nominal orbit of m86 itself . this model may also explain the apparent double streams of gas seen trailing m86 , as well as those in other virgo galaxies . * the apparent brightness edge to the southeast seen in _ rosat _ observations is also seen in the _ chandra _ images . the edge is well fit with a two power law gas density model , with an abrupt jump in density by a factor of @xmath128 at the edge ( consistent with no jump ) . assuming that this brightness edge is the shock generated as m86 supersonically falls into the virgo cluster , the lack of a well - defined density jump is consistent with what is expected from our orbital model , which suggests that the orientation of the mach cone would make it difficult to detect the shock edge . the financial support for this work was provided by nasa contracts nas8 - 39073 , nas8 - 38248 , nas8 - 01130 , nas8 - 03060 , the chandra science center , and the smithsonian institution . beckmann , v. , gehrels , n. , favre , p. , walter , r. , courvoisier , t. j .- l . , petrucci , p .- o . , & malzac , j. 2004 , apj , 614 , 641 binggeli , b. , popescu , c. c. , & tammann , g. a. 1993 , a&as , 98 , 297 birzan , l. , rafferty , d. a. , mcnamara , b. r. , wise , m. w. , & nulsen , p. e. j. 2004 , apj , 607 , 800 bhringer , h. , briel , u. g. , schwarz r. a. , voges , w. , hartner , g. , & trmper , j. 1994 , nature , 368 , 828 bregman , j. n. , & roberts , m. s. 1990 , apj , 362 , 468 cappellari , m. , & copin , y. 2003 , mnras , 342 , 345 crowl , h. h. , kenney , j. d. p. , van gorkom , j. h. , & vollmer , b. 2005 , aj , 130 , 65 diehl , s. , & statler , t. s. 2006 , mnras , 368 , 497 edge , a. c. 1990 , phd thesis , univ . leicester fabian , a. c. , schwarz , j. , & forman , w. 1980 , mnras , 192 , 135 finoguenov , a. , & jones , c. 2002 , apj , 574 , 74 finoguenov , a. , pietsch , w. , aschenbach , b. , & miniati , f. 2004 , a&a , 415 , 415 forman , w. , schwarz , j. , jones , c. , liller , w. , & fabian , a. c. 1979 , apj , 234 , l27 machacek , m. e. , jones , c. , & forman , w. r. 2004 , apj , 610 , 183 matsushita , k. et al . 1994 , apjl , 436 , 41 maughan , b. j. , ellis , s. c. , jones , l. r. , mason , k. o. , crdova , f. a. , & priedhorsky , w. 2006 , apj , 640 , 219 mei , s. , et al . 2007 , apj , 655 , 144 mihos , c. , harding , p. , feldmeier , j. , & morrison , h. 2005 ; apjl , 631 , 41 navarro , j. s. , frenk , c. s. , & white , s. d. m. 1995 , mnras , 275 , 720 neilsen , e. h. , & tsvetanov , z. i. 2000 , apj , 536 , 255 neto et al . 2007 , mnras , 381 , 1450 nulsen , p. e. j. 1982 , mnras , 198 , 1007 nulsen , p. e. & carter , d. 1987 , mnras , 225 , 939 osullivan , e. , vrtilek , j. m. , kempner , j. c. , david , l. p. , & houck , j. c. 2005 , mnras , 357 , 1134 rangarajan , f. v. n. , white , d. a. , ebeling , h. , & fabian , a. c. 1995 , mnras , 277 , 1047 sarazin , c. l. , kundu , a. , irwin , j. a. , sivakoff , g. r. , blanton , e. l. , & randall , s. w. 2003 , apj , 595 , 743 schekochinin , a. a. , cowley , s. c. , r. m. kulsrud r. m. , rosin , m. s. , & heinemann , t. 2007 , preprint ( astro - ph/0709.3828 ) schindler , s. , binggeli , b. , & bhringer , h. 1999 , a&a , 343 , 420 smith , r. j. , lucey , j. r. , hudson , m. j. , schlegel , d.j . , & davies . r. l. 2000 , mnras , 313 , 469 takano , s. , awaki , h. , koyama , k. , kunieda , h. , tawara , y. 1989 , nautre , 340 , 289 takeda , h. , nulsen , p. e. j. , & fabian , a. c. 1984 , mnras , 208 , 261 vollmer , b. , braine , j. , combes , f. , sofue , y. 2005 , a&a , 441 , 473 white , d. a. , fabian , a. c. , forman , w. , jones , c. , & stern , c. 1991 , apj , 375 , 35 lcccc 318&2000 - 04 - 07&m86 & s3 , s2 , i2 , i3&12.964 + 803&2000 - 05 - 19&m84 & s3 , s2 , i2 , i3&26.699 + 963&2000 - 04 - 07&m86 & s3 , s1 , s2 , i2 , i3&13.220 + 1619&2001 - 06 - 08&ngc4388 & s3 , s2 , i2 , i3&19.705 + 2882&2002 - 01 - 29&ngc4438 & s3 , s2 , i2 , i3&24.891 + 5908&2005 - 05 - 01&m84 & s3 , s1 , s2 , i2 , i3&36.299 + 5912&2005 - 03 - 09&se of m86 & i0 , i1 , i2 , i3 , s2&31.370 + 5913&2005 - 03 - 19&m86 tail tip & i0 , i1 , i2 , i3 , s2&34.744 + 6131&2005 - 11 - 07&m84 & s3 , s1 , s2 , i2 , i3&37.855 + lccccc 1&0.724@xmath129//(2.4)&0.42@xmath130&1.71@xmath131&108/91=1.18&6235 + 1&0.677@xmath132/1.291@xmath133/(2.4)&0.64@xmath134&(1.5)&90/89=1.01&6235 + 2&0.697@xmath135/(1.2)/(2.4)&0.31@xmath136&&56/57=0.98&3086 + 3&0.653@xmath137/(1.2)/(2.4)&0.50@xmath138&(1.5)&88/83=1.06&4273 + 4&0.773@xmath139/(1.2)/(2.4)&0.47@xmath140&&109/84=1.29&8630 + 5&0.854@xmath141/(1.2)/(2.4)&0.40@xmath142&&94/93=1.01&4461 + 6&0.849@xmath143/(1.2)/(2.4)&0.31@xmath144&&89/88=1.01&5059 + 7&0.860@xmath145/(1.2)/(2.4)&0.43@xmath146&&87/74=1.18&2843 + 8&0.785@xmath147/(1.2)/(2.4)&0.45@xmath148&&44/40=1.09&1749 + 9&0.924@xmath149/(1.2)/(2.4)&0.50@xmath150&&30/29=1.03&1043 + 10&1.191@xmath151/(2.4)&0.42@xmath152&&20/25=0.73&739 + 11&0.865@xmath153/1.290@xmath154/(2.4)&0.55@xmath155&&163/176=0.92&8795 + 12&0.871@xmath156/1.176@xmath157/(2.4)&0.62@xmath158&&240/187=1.29&8914 + 13&0.675@xmath159/1.256@xmath160/2.723@xmath161&0.72@xmath162&&343/293=1.17&11033 + 14&0.625@xmath163/(1.2)/(2.4)&1.17@xmath164&1.82@xmath165&459/327=1.40&42002 + 15&0.790@xmath166/(1.2)/(2.4)&0.60@xmath162&&93/90=1.03&3666 + 16&1.355@xmath167/(2.4)&0.49@xmath168&&57/59=0.96&1746 + 17&1.091@xmath169/2.422@xmath170&0.40@xmath171&&53/45=1.18&3054 + 18&1.085@xmath172/2.107@xmath173&0.22@xmath174&&118/120=0.98&4643 + 19&0.651@xmath175/1.065@xmath176/(2.4)&0.25@xmath177&&147/128=1.15&7325 + 20&1.319@xmath178/(2.4)&0.39@xmath179&&30/23=1.29&1087 + 21&1.023@xmath180/(2.4)&0.22@xmath181&&147/105=1.39&2912 + 22&0.881@xmath182/1.236@xmath183/(2.4)&0.32@xmath136&&77/75=1.03&2646 + 23&1.186@xmath184/2.712@xmath185&0.45@xmath186&&214/236=0.91&5193 + = .31 . the model is a two - temperature thermal plasma with fitted temperatures and abundance ( given in the second row of table [ tab : spectra ] ) , plus an unresolved power law component with photon index of 1.5 and a fixed thermal plasma component to model emission from the virgo icm , all with galactic absorption . the data are from two s3 observations , obs - ids 318 and 963 ( see table [ tab : obs ] ) . [ fig : m86spec ] ] ) . the x - axis gives the radius from the apparent center of curvature defined by the feature . the best fit two power law density jump model is given by the solid line . [ fig : nedge ] ] towards us . locations on the orbits where the past orbit would also appear to project 100 kpc outside the current position of m86 are shown in red ( solid lines ; the midpoint of each of these regions is marked with an asterisk ) . both regions shrink for orbits with smaller outer turning radii . as plotted , all orbits pass through the virial radius at @xmath187 kpc on the @xmath188 axis . ] ( marked with an asterisk ) . orbits are plotted from the point where they cross the virial radius ( sometimes outside the field shown ) . ] , one representative orbit is shown projected onto the smoothed 0.5 2 kev image of m86 for the remaining plausible orbits . colors indicate the outer turning radius as follows : @xmath189 mpc , yellow ; @xmath190 mpc , blue ; @xmath191 mpc , green ; @xmath60 , red . ] kpc , one of the better candidates for the orbit of m86 . centered on the orbit , at equally spaced times , circles are plotted on the plane of the sky with radii @xmath192 , where @xmath77 is the time before the present when m86 was at that location and @xmath193 is the sound speed ( @xmath194 ) . ] in cylindrical polar coordinates , with the origin at the cluster center and the orbit ( instantaneously ) in the @xmath195 @xmath196 plane , the position of m86 can be expressed as @xmath197 and its velocity as @xmath198 , where @xmath199 , @xmath200 , @xmath201 are the coordinate basis vectors . when m86 is viewed along the unit vector , @xmath202 , its location projected onto the plane of the sky is @xmath203 . expressing @xmath202 in terms of the coordinate vectors as @xmath204 , the projected separation , @xmath51 , of m86 from the cluster center is given by @xmath205 , which requires @xmath206 ( and @xmath207 ) . since @xmath202 is a unit vector , @xmath208 , further requiring @xmath209 . the remaining parameter , @xmath210 , is the cosine of the polar angle of the viewing direction ( i.e. , the orbit inclination ) . with the two values of @xmath66 above , the parametrization @xmath211 , for @xmath212 , specifies the complete range of viewing directions that project m86 at the observed distance from the cluster center . in this notation , the velocity of m86 along our line - of sight is @xmath213 . for radial orbits with m86 inbound ( @xmath214 ) , @xmath215 . this vanishes for @xmath216 ( where a radial orbit must be perpendicular to our line - of - sight ) , while the radial speed , @xmath217 , decreases with increasing @xmath50 , so that @xmath52 is maximized for some @xmath207 . the inner and outer turning radii , @xmath55 and @xmath56 , respectively , are related to the specific energy of an orbit by @xmath218 , where @xmath219 is the gravitational potential at @xmath55 and @xmath220 is that at @xmath56 . the specific angular momentum of the orbit , @xmath221 , is given by @xmath222 . if m86 is bound to the virgo cluster , then , since @xmath223 and its speed , @xmath224 , which requires @xmath225 . taking @xmath226 , this gives an upper limit on the inner turning radius of @xmath227 kpc . to obtain a lower limit on @xmath56 , note that @xmath228 , where the kepler speed at @xmath56 is given by @xmath229 and the speed at the outer turning radius is @xmath230 . the net acceleration at the outer turning radius is inward , requiring @xmath231 , so that @xmath232 is an increasing function of @xmath56 . interchanging @xmath55 and @xmath56 in this argument and noting that the net acceleration is outward at the inner turning radius ( @xmath233 ) shows that @xmath232 is also an increasing function of @xmath55 . since @xmath234 , for a fixed energy , maximizing @xmath55 minimizes @xmath56 . for m86 , @xmath235 and @xmath236 , so that its energy @xmath237 . to obtain @xmath238 with @xmath239 requires @xmath240 mpc , or @xmath59 times the virial radius of the cluster . for values of @xmath55 larger than @xmath51 , the total energy would need to match or exceed @xmath241 . the value of @xmath56 required to attain this energy is an increasing function of @xmath55 , unless the mean density of the cluster increases with radius . thus , @xmath242 provides a lower limit on the outer turning radius of m86 s orbit . at a fixed point on an orbit , the requirement @xmath243 determines the possible viewing directions , if any , that could model m86 . for each of the two solutions @xmath244 , the requirement @xmath245 gives @xmath246 . if either of these gives @xmath247 in the range @xmath248 $ ] , then it corresponds to two possible viewing directions , with @xmath249 ( symmetrically placed above and below the plane of the orbit ) . thus , for each point on an orbit , there can be zero , two , or four potential viewing directions that would place m86 at the observed separation from the cluster center and give it the observed line - of - sight velocity . figure [ fig : mb ] shows a range of marginally bound orbits ( @xmath60 ) , with regions that meet these conditions marked in color . since latexmath:[$|{v_{\rm { v_{\rm los , max}}= \sqrt{1 - s^2/r^2 } if the trail of stripped gas lies along the orbit of m86 , then the past orbit must project away from the cluster center at least as far as the gas tail extends . from above , @xmath251 is a unit vector in the plane of the sky that points from the cluster center towards m86 . for each potential location of m86 on an orbit , we have computed @xmath252 for values of @xmath253 ranging over the past orbit . if stripping is to account for the location of the gas tail , @xmath254 must increase initially as @xmath253 moves backward along the orbit and the maximum value of @xmath255 must exceed the distance , @xmath62 kpc , that the tail projects beyond m86 in the direction away from the cluster center . locations on the orbits where the line - of - sight speed can attain @xmath63 and these conditions are also met are shown in red in figure [ fig : mb ] .

we report two spectropolarimetric observations of sn 2005hk , which is a close copy of the very peculiar " sn 2002cx , showing low peak luminosity , slow decline , high ionization near peak and an unusually low expansion velocity of only about 7,000 . further to the data presented by chornock et al . , ( 2006 ) , at -4 days before maximum , we present data of this object taken on 9 november 2005 ( near maximum ) and 23 november ( + two weeks ) that show the continuum and most of the spectral lines to be polarized at levels of about 0.2 - 0.3% . at both epochs the data corresponds to the spectropolarimetric type d1 . the general low level of line polarization suggests that the line forming regions for most species observed in the spectrum have a similar shape to that of the photosphere , which deviates from a spherical symmetry by . in comparison with spectropolarimetry of type ia and core - collapse sne at similar epochs , we find that the properties of sn 2005hk are most similar to those of type ia sne . in particular , we find the low levels of continuum and line polarization to indicate that the explosion mechanism is approximately spherical , with homogeneous ejecta ( unlike the chemically segregated ejecta of ccsne ) . we discuss the possibility that sn 2005hk was the result of the pure deflagration of a white dwarf and note the issues concerning this interpretation .

1008.3985

the nature of the complex explosion mechanisms thought to be behind supernovae ( sne ) , of all types , is expected to leave a signature in the resulting geometry of the explosion . the study of the polarized light arising from sn explosions provides a unique and powerful method for directly measuring asymmetries in sne . the power of spectropolarimetry is demonstrated by its application to the study of the shapes of sne at large distances , providing an effective spatial resolution far superior to any current or currently planned imaging technique . + in sn photospheres , the predominant form of opacity is due to electron scattering , which is also a polarizing process . in a spherical symmetry , light originating from different quadrants of the photosphere have opposite but equal polarization components , leading to a zero net observed polarization . in the presence of asymmetries , the cancellation of the polarization components arising from different portions of the ejecta is incomplete , leading to an observable polarization whose magnitude is related to the degree of the asymmetry . in addition to the polarization of continuum light , the absorption due to line forming species exterior to the photosphere `` cast shadows '' on the photosphere . if the photosphere is not completely covered , a polarization signature will be associated with wavelength - dependent absorption features , with the degree of polarization related to the amount and orientation of the coverage of the photosphere by the line forming region . + previous polarimetric studies of sne have shown that type ia sne and core - collapse supernovae ( ccsne ) have different polarization properties . type ia sne are observed to have generally low polarizations , indicative of a close - to - spherical symmetry . the degree of polarization of these events decreases with time , as deeper layers of ejecta are revealed , implying the ionization/ density structure becomes more spherically symmetric with depth . in contrast , ccsne have been observed with a wide variety of polarizations , generally larger than those observed for type ia sne . in addition , the degree of polarization is observed to increase with time for ccsne , suggesting the ejecta become more asymmetric at increasing depth . + in the classical observing modes of photometry and spectroscopy , type ia sne at certain phases can be confused upon casual inspection with type ic sne at other phases due to the common display of lines . special care must be given when trying to determine the type and the physics of peculiar " type ia and peculiar " type ic sne . the application of spectropolarimetry can provide additional constraints , both in the abstract sense of the stokes parameters and in the inferred geometries , that can help separate events belonging to the different distinct classifications . + one particular class of sne , which have been previously classified as type ia sne ( albeit the most peculiar type ia " ; @xcite ) , are the so - called `` 2002cx - like '' sne . it has recently been suggested that members of this class may be representative of type ia sn events that result from a pure - deflagration ( see * ? ? ? * and references therein ) or instead may be peculiar , low - luminosity type ic sne @xcite . + here we report two epochs of spectropolarimetry of the 2002cx - like sn 2005hk , acquired at approximately maximum light and two weeks later using the very large telescope ( vlt ) of the european southern observatory ( eso ) . sn 2005hk was discovered in ugc 272 by @xcite . the location of sn 2005hk , relative to its host galaxy , is shown as fig . [ fig : obs : hostimage ] . early spectroscopy was presented by @xcite , @xcite , @xcite and @xcite . they identify the early high ionization ( e.g. ) species typical of sn 2002cx - like sne , as well as the bright type ia sn 1991 t ( see also * ? ? ? * ) , and the low velocities characteristic of sn 2002cx ( @xmath2 6000 @xmath0 ) . the spectra of @xcite and @xcite showed intermediate - mass elements with weaker features than regular sne ia , but stronger than sn 1991 t . @xcite suggested that ni and co may be present in the early spectra and hence in the outermost layers of the ejecta . the red portions of the spectra show appreciable features of and between 7500 and 9000 , whereas sn 1991 t , a possible analog , showed a basically featureless spectrum in that phase and wavelength range . 2002cx and sn 2005hk showed similar fine scale structure suggesting that the structure was systematic and hence not the product of random clumping in the ejecta of either event . @xcite present extensive _ ugribvrijhk _ photometry and optical spectroscopy of sn 2005hk , including the spectral data of @xcite and @xcite . importantly , they note that the light curves of these two events are very similar , with neither showing the typical secondary maxima in r and i displayed by normal sne ia , but light curves that are also distinctly different than under - luminous sne ia such as sn 1991bg . in addition , the bolometric light curve , including swift uv data , shows a rather slow light curve ( @xmath3 ; @xcite ) , uncharacteristic of under - luminous sn ia and similar to normal type ia sne over a magnitude brighter at maximum . + @xcite presented spectropolarimetry of sn 2005hk , at 4 days prior to our first observation . they determined that the continuum was polarized @xmath4 in the red , with the data represented by a dominant axis . they also identified a weak modulation of the polarization of the line at @xmath5 . + in section [ sec : obs ] we present our spectropolarimetric observations and our adopted reduction procedure . the results from these observations are presented in [ sec : results ] and these results are discussed in [ sec : disc ] . in [ sec : conc ] we present our conclusions . + spectropolarimetry of sn 2005hk was acquired using the eso - vlt on 2005 nov 9 and 23 , with the fors1 instrument in polarization pmos mode . a log of these observations is presented in table [ tab : obs : log ] . the data were reduced following the scheme outlined in @xcite . at each epoch , the observations consisted of four exposures with the half - wavelength retarder plate positioned at four position angles @xmath6 , @xmath7 , @xmath8 , and @xmath9 , to derive the two linear stokes parameters @xmath10 and @xmath11 and @xmath12 . ] . both observations were conducted with the 300v grism , without an order separation filter , providing a spectral resolution of @xmath13 as measured from arc line calibration data . telluric lines were removed from flux spectra in the standard manner . flux standard stars were observed , with the full polarimetry optics in place , to calibrate the flux levels of the sn data and remove telluric features . the telluric features are not expected to affect the polarimetry , except to decrease the level of signal - to - noise across the absorption , provided there is no significant change in the strength of absorptions between exposures with the retarder plate rotated by @xmath7 . + the instrumental stability between the observations at the two epochs was assessed by calculating the difference between the instrumental signature corrections @xmath14 and @xmath15 . at both epochs the difference was found to be @xmath16 , implying a limit on the instrumental stability and the measurement of the stokes parameters to @xmath17 . this is consistent with previous reported measurements of the instrumental signature corrections and limits on stability of the vlt fors instrument @xcite . + the degree of polarization was corrected for biases due to observational noise using the equations derived in @xcite , and , additionally , using a monte carlo simulation of the fors1 instruments in the manner of @xcite and . the wavelength scales were corrected for the heliocentric recessional velocity of the host galaxy of 3895 @xmath18 . the spectroscopic properties of sn 2005hk have already been described in some detail in other works @xcite . here the prominent spectroscopic features , which characterise sn 2005hk at the epochs at which spectropolarimetry was acquired , will be briefly reviewed to orient the reader . the flux spectra from our two spectropolarimetric observations are presented as fig . [ fig : obs : spec ] . + our data from 2005 november 9 are interleaved with the coverage presented by @xcite who give spectra from 7 and 14 november that they assign to -3 and + 4 days with respect to b maximum which they determine to be jd2,453,685.1 . this means our data are from very near b maximum , perhaps a day earlier . we confirm that the spectral features are very similar to those of sn 2002cx at the same epoch @xcite . furthermore , @xcite acquired a flux spectrum of sn 2005hk in the course of acquiring spectropolarimetry at november 5.4 , 4 days prior to our first observation . @xcite performed line identifications using the synow code for their spectrum . they found that the strongest lines were from with other lines of intermediate mass elements being present , but relatively weaker . they obtained a better fit in the blue by adding lines of , , and . they used an excitation temperature of 10,000 k for the low - ionization stage elements and 15,000 k for the higher ionization stages , and . our flux spectrum from the 2005 november 09 observation , despite being acquired 4 days later , requires similar identifications to those made by @xcite . in the red portion of the spectrum , we identify the ir triplet , although the synthetic spectra of @xcite suggested that at the epoch of their observations this was mostly due to and . the feature at @xmath19 is identified by @xcite as , while @xcite suggested it was a mixture of and . @xcite observed the appearance of @xmath20 after maximum in spectra of sn 2002cx , which suggests that at 0 days for sn 2005hk this feature could be at least partially due to . + our spectrum from 2005 november 23 corresponds to an epoch two weeks after b maximum . our data are very similar to spectra of that date presented by @xcite , but our data have higher signal to noise and broader wavelength coverage . the spectrum of @xcite from november 25 ( + 15 days ) has similar wavelength coverage to ours of november 23 and remains very similar to our flux spectrum . at this epoch , the underlying continuum is cooler with a relative paucity of flux in the blue compared to the spectrum of 9 november . the spectrum is characterized by the growth in amplitude of small scale features . synow was used by us , in conjunction with previously published line identifications , to determine line identifications specific to the spectrum at this epoch . the degree of blending of multiple lines is lower at the second epoch , due to the relative decrease in the expansion velocity . the truncation of the red - side of the @xmath20 emission line profile at @xmath21 may be evidence for , but it is located close to strong telluric absorption features . the identification of @xmath22 at + 14 days is debatable , and is likely to be due to an increase in @xmath23 @xcite . @xcite similarly found that was no longer required to fit spectra of sn 2002cx at a similar epoch . the photospheric velocity was estimated , using lines of ( 42 ) , to be @xmath24@xmath0 . + at the second epoch , there are also two distinct absorption features that are likely to be caused by the ir triplet . the ir triplet was identified at 12 days after maximum in the spectrum of sn 2002cx @xcite . the two features might be due to partially resolving the triplet in this low velocity photosphere ( @xmath25 @xmath0 ) , or there could be two kinematically separate components such as those observed in normal type ia sne @xcite . if the latter , the absorption minima correspond to velocities of @xmath26 and @xmath27 @xmath0 , respectively , taking the mean wavelength of the blend to be 8579 . the bluer of these two components is moving somewhat more rapidly than the typical range identified for photospheric features for type ia sne at similar epochs @xcite . comparison with sn 2002cx at 12 days @xcite leads to the identification of the strong absorption observed at 5730 as intrinsic to the sn , corresponding to a velocity of @xmath28@xmath0(which is similar to the high velocity feature ) . + at the first epoch , there is a feature at @xmath29 . @xcite argue it is unlikely to be due to due to absence of other stronger lines that should occur in the wavelength range of these observations ( e.g. @xmath30 ) . rather @xcite suggests the @xmath31 may arise from @xmath32 . @xcite could not identify features in spectra of sn 2002cx . at the second epoch , however , there is a suggestion of an absorption ( blended with an line giving rise to a flat topped emission line profile for that line ) which is consistent with the expected position of @xmath30 with a velocity of @xmath27 @xmath0 ( similar to the low velocity component of ) . spectropolarimetry of sn 2005hk at the two epochs is presented in figs . [ fig : obs : panela ] and [ fig : obs : panelb ] . + the correct measurement of the intrinsic polarization of sne requires the subtraction of the polarization contribution of the interstellar medium , both in the galaxy and in the host . in the first instance , a general limit can be placed on the degree of the interstellar polarization ( isp ) by the amount of reddening measured towards sn 2005hk . for galactic - type dust the maximum degree of the isp is related to the reddening by the limit @xmath33 @xcite . the degree of galactic foreground reddening in the direction of sn 2005hk has been reported to be @xmath34 @xcite . the limit on isp arising in the foreground is @xmath35 . + the reddening towards sn 2005hk can be estimated , if we assume that the narrow d absorption component arises in the interstellar medium ( ism ) of the host . the equivalent width of the d absorption feature at the first epoch was measured to be @xmath36 . this value is smaller than the value reported by @xcite , for observations 4 days prior to our first observations , although the difference between the two measured values is @xmath37 . we note @xcite do not quote errors on their determined equivalent width . it is not clear if this difference is within the measurement errors or if it reflects evolution triggered by the sn @xcite . we follow @xcite , who adopt the results of for evaluating the relationship between d absorption and the reddening . this relationship is based on galactic - type extinction laws ( e.g. * ? ? ? it is , therefore , implicitly assumed that the reddening component , measured from the narrow absorption component in the host galaxy , follows a similar law . as the d component lines are not resolved in our data , we also adopt a range of flux ratios between d1 and d2 of 1.1 to 2 @xcite . this limits the reddening of the host to @xmath38 . the total reddening , for galactic type reddening and isp law , yields a constraint on the total isp of @xmath39 . + @xcite present a technique for the determination of the isp for specific wavelength ranges which have null intrinsic polarization . this assumes that , for these specific wavelength ranges , there is sufficient overlap between groups of lines such that there is complete intrinsic depolarization over these wavelengths . @xcite suggest that the wavelength region of 4800 - 5600 is intrinsically depolarized , where line blanketing opacity dominates over electron scattering opacity . any polarization , observed over this wavelength range , is expected to be due to the isp alone . for sn 2005hk , at both epochs , the lack of significant overlap and blending of lines leads to the complex behaviour of the stokes parameters . @xcite find that the wavelength range of 4000 - 4200 satisfies the criteria for intrinsic depolarization , namely significant line blanketing . the observation of @xcite at november 5.4 is sufficiently close to our observation at the first epoch that we observed similar features in the polarization spectrum and the stokes parameter spectra . importantly , we too determine that this wavelength range provides the best approximation to a line blanketed regime . the measured stokes parameters across this range at the first epoch are @xmath40 and @xmath41 . the uncertainties on these quantities are the combined systematic and statistical uncertainties . + the spectral evolution of sn 2005hk by the second epoch , with the increase in the strengths of individual and lines in the range of 4000 - 4200 , leads to the presence of intrinsic polarization ( @xmath42 and @xmath43 ) . instead , we identify the region of 4250 - 4450 as being spectroscopically similar to a line blanketed regime with no unblended line structure . over this wavelength range the average values of the stokes parameters are measured as @xmath44 and @xmath45 . the measured stokes parameters , over the depolarized wavelength range , agree with those determined at the first epoch . + we adopt values of the isp stokes parameters of @xmath46 and @xmath47 , for both the galactic and host isp components , which are approximately consistent with the values measured by @xcite ( who calculate separate galactic and host components ) . the value of the isp relative to the observed stokes parameters is shown on figs . [ fig : obs : panela ] and [ fig : obs : panelb ] . these values correspond to a polarization of @xmath48 with an angle of @xmath49 . the orientation of the isp component is shown on fig . [ fig : obs : hostimage ] . as previously observed for other sne , the isp component is aligned with the spiral arm pattern at the sn location , due to alignment of magnetic field lines and , hence , dust grains along spiral arms @xcite . + given the limited wavelength range over which it can be determined , it is not clear if and how the isp component varies with wavelength . although a standard serkowski - type law could be applied , it is unclear what the appropriate values of @xmath50 and @xmath51 would be . furthermore , it has been observed for a number of sne with significant isps that the isp does not necessarily even follow a galactic - serkowski type law of wavelength dependence . given the small value of the isp , we subtract , therefore , a constant isp component from the observed data , to recover the intrinsic polarization of the sn ( as shown by @xmath52 in figs . [ fig : obs : panela ] and [ fig : obs : panelb ] ) , under the assumption that any wavelength dependence of the isp is smaller than the uncertainties derived above . + on the stokes @xmath53 plane ( see figs . [ fig : obs : qupanela ] and [ fig : obs : qupanelb ] ) , the isp component is located close to the centroid of the sn 2005hk data . the calculated weighted centroids for the isp - corrected data @xmath54 are listed in table [ tab : obs : domaxis ] . + after correction for the isp , the intrinsic polarization of sn 2005hk is observed to be extremely low across the observed wavelength range . the degree of polarization of a number of features observed to be relatively strongly polarized in the uncorrected polarization spectrum , are no longer as polarized once the data have been corrected for the isp . utilising the data across the range @xmath55 as representative of a region of pure continuum emission , the level of the continuum polarization is measured as @xmath56 and @xmath57 at the first and second epochs , respectively . at the second epoch the assumption that this wavelength range represents a true continuum region is most likely invalid , due to the presence of a number of minor , broad undulations which are most likely due to the lines ( rather than incorrect flatfielding of the flux spectrum ) . this wavelength range avoids any polarization associated with the feature at @xmath31 , and is less than the continuum polarization measured by @xcite at 2005 nov 5.4 of @xmath58 . this level of polarization is consistent with a spheroidal configuration with axial ratio @xmath59 . + the significance of the polarization signal is given by @xmath60 @xcite . at both epochs there are no features , at either @xmath61 or @xmath62 binning , which are detected in the intrinsic polarization spectrum @xmath52 at @xmath63 and only a small number of bins detected at @xmath64 . @xcite argue that in their data from 5 november , only the line of @xmath65 is polarized and at a rather small level . on inspection , their data suggest that @xmath66 is also somewhat polarized . there is another feature of similar amplitude showing a minimum at @xmath67 in the polarized flux that has no strong component in the total flux spectrum , but is likely to correspond to @xmath68 . + our data at the first epoch also show possible modulation in the polarization spectrum associated with @xmath65 , as well as @xmath22 and the @xmath31 feature . at redder wavelengths there are a number of peaks which are associated with telluric features , suggesting that these were variable over the course of the observations . the apparent significance of these features is therefore overestimated , as the possible variability of telluric features was not considered in the reduction process and not accounted for in the final error analysis . + by the second epoch , there are possible peaks in the intrinsic polarization associated with the absorptions of the multiplet 42 , the intrinsic feature and the absorption of @xmath69 . + the absence of a significant signal in the polarization spectrum does not necessarily imply that the intrinsic polarization of spectral features is also not significant . @xcite and @xcite studied the change in polarization @xmath70 across particular spectral features , by comparing the modulation in the stokes @xmath10 and @xmath11 parameters from the approximate `` continuum '' level and at the wavelength of absorption minimum , where : @xmath71 @xmath72 is independent of the choice of the isp component . the values of @xmath72 for particular spectral features , at both epochs , are presented in table [ tab : res : delp ] . + even when assessed by modulation of the stokes parameters , the data at both epochs show few significant features . at the first epoch , h&k , and , importantly , @xmath73 show significant modulation , whereas at the second epoch the significant modulation is associated with the , now stronger , lines . the interpretation of the data presented in table [ tab : res : delp ] requires some caution , however , as the selection of @xmath74 and @xmath75 , by eye , can lead to a bias in favour of selecting those wavelengths that yield the maximum change in the stokes parameters . the principal differences between the data presented by @xcite at -4 days , compared to our data at 0 days , is due to much larger s / n achieved in their observation . we estimate that at @xmath31 , at 50 binning , @xcite achieved a final @xmath76 , whereas our observation reached @xmath77 . + at the first epoch , two sets of two contiguous bins ( at 30 ) are observed at both the 7200 feature and immediately redward at 7315 . an obvious line feature is associated with the former spectral line in the flux spectrum . the latter feature corresponds to a telluric feature that , although it should be equally cancelled between the two beams at each retarder plate position , may appear polarized due to variability in the telluric line between the pair of observations used to derive each stokes parameter . + on the stokes @xmath53 plane ( see figs . [ fig : obs : qupanela ] and [ fig : obs : qupanelb ] ) , the data at both epochs assume an approximate elliptical distribution , with no major structure deviating from the main concentration of points . at the first epoch , the data have no obvious preferred direction . by the second epoch the data are more extended across the @xmath53 plane , although again with no major deviations from the main concentration of data points . from fig . [ fig : obs : qupanelb ] it is also evident that , by the second epoch , the degree of polarization has become more wavelength dependent , with data at bluer wavelengths being concentrated at lower values of @xmath10 and @xmath11 than the data at the red extreme of the observed wavelength range . our lower @xmath78 data does not show the wavelength dependent elongation to the same degree as observed at -4 days , for which a dominant axis is clear to the eye . the behaviour of the data on the stokes @xmath53 plane , at both epochs , is quantified and discussed in section [ sec : res : decomp ] . the determination of the dominant axis , and the decomposition of the stokes vector , permits a correction for the random position angle at which target sne have been observed , as well as determination of the principal axis of symmetry of the data , the relative orientations of the different line forming species and , ultimately , the underlying geometry . the dominant axis forms a line across the stokes @xmath53 plane , and the decomposition process involves rotating the data so that the principal rotated stokes axis is aligned with the axis of greatest variance for the data on the stokes plane . the polarization signal is measured in the two orthogonal directions , which are effectively the `` rotated stokes parameters '' , along the dominant axis ( @xmath79 ) and along the orthogonal axis ( @xmath80 ) . a polarization component in the orthogonal direction permits the identification of deviations from the single axial symmetry represented by the dominant axis . + here we explore two complementary techniques for determining the dominant axis : least - squares fitting of a straight line and a principal components analysis of the data on the stokes @xmath53 plane . importantly , these two different analyses _ should _ give identical measures of the dominant axis , but provide different parametrisation of the data on the stokes @xmath53 plane . this latter point is particularly crucial for applying the sn spectropolarimetric classification scheme of . + straight lines , of the form @xmath81 , were fit to the data of both epochs , using a weighted least - squares fitting technique , where the data were weighted according to the errors in both @xmath10 and @xmath11 @xcite . in order to confirm the validity of this technique , the form of the straight line was reversed and the best fit was calculated for @xmath82 , giving identical results . the value of @xmath83 yields the angle of the dominant axis with respect to the stokes @xmath10-axis , through which the data are to be rotated to decompose the stokes vector onto the dominant and orthogonal axes . for a single axial symmetry , the dominant axis will pass through the origin of the stokes @xmath53 plane . in instances where the data are offset from the origin ( e.g. by a wavelength - independent continuum polarization component ) the offset will be _ partially _ manifested in the measured @xmath84 parameter . + dominant axes were calculated for the data at both epochs over two different wavelength ranges : 3700 to 8600 and 4000 - 7000 . the former range wavelength encompasses the entire wavelength range of the observed data , whereas the latter excludes regimes with poor levels of signal - to - noise ( @xmath85 ) . the regions of low signal - to - noise arise at the blue and red extremes of the data , due to line - blanketing and the blue spectral energy distribution significantly reducing the observed flux in the blue and red , respectively , and the declining detector response ( in the red ) . + the results of the least - squares analysis and the properties of the determined dominant axes are presented in table [ tab : obs : domaxis ] . the values of @xmath84 , @xmath83 , and the polarization angle of the dominant axis ( @xmath86 ) were found to be relatively insensitive , within the uncertainties , to the choice of the wavelength range over which the dominant axis was calculated . the dominant axes are shown on the stokes @xmath53 plane on figs . [ fig : obs : qupanela ] and [ fig : obs : qupanelb ] . the large values of @xmath87 for the straight line fit , at both epochs , indicate that the data are poorly described by a straight line , with a significant polarization component in the orthogonal direction . at the first epoch , simple inspection of the data on the stokes @xmath53 plane would suggest that the data are evenly distributed in angle about the centroid ( as is illustrated in section [ sec : res : pca ] ) . at the second epoch , a dominant axis is more clearly observed . + in order to better understand and quantify the distribution of the data on the @xmath53 stokes plane and , in particular , the degree of the orthogonal polarization component , a principal component analysis of the data was conducted . + the elements of the @xmath88 @xmath89 covariance matrix take the form : @xmath90 where @xmath91 , @xmath92 , @xmath93 ( see * ? ? ? * ) , @xmath94 and the weighted average of the stokes parameters are @xmath95 , as shown in table [ tab : obs : domaxis ] , over the @xmath96 wavelength bins . the eigenvectors and eigenvalues of the covariance matrix provide an alternative measure of the angles of the dominant and orthogonal axes ( @xmath97 ) and the ratio of the degrees of polarization in the dominant and orthogonal directions ( an axial ratio for the resulting ellipse @xmath98 ) , respectively . this latter quantity can be used to describe an ellipse on the stokes @xmath53 plane which contains the data . the results of the principal component analysis are presented in table [ tab : obs : domaxis ] , and corresponding ellipses are presented on figs . [ fig : obs : qupanela ] and [ fig : obs : qupanelb ] . as expected , the polarization angle of the dominant axis determined using this analysis agrees , within the uncertainties , with the result from the least - squares fit . + the non - zero values of @xmath98 determined from the principal components analysis show that the data at the two epochs are inconsistent with both straight lines and null polarization . this suggests that , although the polarization measured at both epochs is low , we have measured a real polarization signal for sn 2005hk . the ratio of the dominant and orthogonal components measured at the second epoch is smaller than that measured at the first , implying that at the first epoch the ejecta were approximately round and , with time , departed more significantly from the spherical symmetry . + this interpretation requires some caution , however , as @xcite observed data that corresponded to a dominant axis four days prior to our first observation , which we do not see in our data at the first epoch . this may reflect the low @xmath78 of our first observation , rather than an `` oscillating '' evolution in shape of the structure of the ejecta ; with the ejecta appearing approximately spherical at maximum but departing from a spherical symmetry both before and after maximum light . the data were rotated onto the dominant and orthogonal axes , using the angles determined from the least - squares fit analysis ( for the limited wavelength range of 4000 - 7000 ) . the rotated data are presented as figs . [ fig : obs : doma ] and [ fig : obs : domb ] . despite the apparent increase in the polarization angle from the first to second epochs , the determined polarization angles for the dominant axes are different from the angle measured at -4 days , for which @xcite determined @xmath99 . + at the first epoch , the dominant and orthogonal polarization components are similar to stokes @xmath10 and @xmath11 , due to the small angle through which the data were rotated . the polarization modulations associated with , , are predominantly observed in the dominant polarization component , and over the range 4000 - 7000 the deviations observed in the orthogonal polarization spectrum are consistent with the expected noise level . similar to the @xcite observations at -4 days , an important exception is the 7200 absorption feature ( and possibly @xmath65 ) which has both dominant and orthogonal polarization components ( although @xcite refer to the rotated stokes parameters @xmath100 and @xmath101 , these are equivalent to our dominant and orthogonal axes ) . at our second epoch , the majority of features , associated with , and , particularly , the @xmath31 feature show some modulation in both the dominant and orthogonal polarization components . while @xcite observed an obvious wavelength dependence in the dominant polarization component , with increasing @xmath79 at redder wavelengths ( which we estimate to be @xmath102 ) , 4 days later we found the wavelength dependence to be shallower @xmath103 . at @xmath104 days we measure a similar wavelength dependence @xmath105 . this wavelength dependence is clearly seen on fig . [ fig : obs : qupanelb ] . the change may not be evidence for evolution in the wavelength dependence of the dominant polarization component , in itself , as the observations of @xcite were acquired with higher @xmath78 . such wavelength dependence of the polarization , increasing at redder wavelengths was observed for sn 1999by @xcite . at the temperatures determined for sn 2005hk , and for fe - rich matter , the opacity at the photosphere is dominated by blends of line transitions for @xmath106 . at redder wavelengths , the opacity is primarily due to electron scattering . the continuum polarization may indicate an aspherical thomson scattering region , e.g. by a global asymmetry induced by the rotation of the white dwarf progenitor @xcite or off - center delayed detonations @xcite . because multiple line scatterings are effectively depolarizing , the observation of increasing polarization with wavelength may be understood in terms of the decrease of the role of line opacity at redder wavelengths . if the dominant axis corresponds to a principal axial symmetry , then the line forming regions of the species observed in the flux spectrum follow the same axial symmetry as that of the continuum at the first epoch , but not at the second epoch . on the basis of the continuum polarizations measured at 0 and 14 days , any asymmetry in the shape of the photosphere ( assuming a spheroidal configuration ) requires an axial ratio @xmath59 . a similar limit on the asphericity of the photosphere is achieved if one uses the higher value of continuum polarization ( @xmath58 ) determined by @xcite + the general level of line polarization is low , suggesting that the line forming regions cover an approximately spherically symmetric photosphere homogeneously . limits on the amount of polarization expected to be observed can be calculated , given a level of continuum polarization and line depth , under the assumption that the line forming region blocks only _ unpolarized light _ from the photosphere , while not blocking polarized light arising from the limb ( i.e. that the photosphere can be separated into discrete polarized and unpolarized zones ) : @xmath107 where @xmath108 and @xmath109 are the observed fluxes in the continuum and in the absorption line minimum , respectively ( i.e. such that the line depth is @xmath110 - this convention is chosen as the observed polarization arises from the light that is not absorbed ) . as @xcite calculated for their observation of sn 2005hk , we also calculate the theoretical limits on the line polarization for measured absorption line depths in our observations . + at the first epoch , the maximum polarization for the @xmath22 line is only 0.14% . as @xcite suggest , that the lines are not strongly polarized is commensurate with their apparent weakness . this is true at the first epoch , where line strengths of only @xmath111 are observed . at the second epoch , however , absorption line depths of @xmath112 are observed , and with the increase in depth of these lines it might be expected that , should the line forming regions be asymmetric , that there would also be a commensurate increase in the polarization observed across these lines ( see fig . [ fig : obs : monte ] ) [ eqn : res : linepollim ] assumes that all of the polarization is due to the continuum , which implicitly assumes that the photosphere itself is asymmetric . in a number of cases , in particular for core - collapse sne , the classical limit for line polarization is exceeded ( see e.g. * ? ? ? * ; * ? ? ? * ) , for low values of continuum polarization which suggest approximately spherical photospheres . violations of this criterion can be explained by large scale structure , with a dissimilar geometry to that of the photosphere , or clumping , which may produce polarization irrespective of the shape of the photosphere or the degree of continuum polarization . the degree of observed polarization , for a clumpy line forming region , should depend on the size of the clumps : for a given line absorption depth , fewer larger clumps block larger areas with contiguous polarization vectors , whereas more smaller clumps can be distributed more evenly across the photosphere . + in order to explore the possible role of clumping , and the expectation for the production of polarization across shallow lines , a monte - carlo toy model of a simple photosphere and obscuring clumped line forming region was constructed following the scheme of @xcite . a circular continuum emitting photosphere was adopted , with the luminosity as a function of projected radius @xmath113 across a photosphere of radius @xmath114 set to : @xmath115 with limb darkening modelled using the coefficient @xmath116 , which for @xmath117 yields no limb darkening . for a three - dimensional spherical photosphere , this would correspond to to a limb darkening law of @xmath118 . the radial dependence of the polarization was assumed to follow a quadratic form @xmath119 , with the constant of proportionality chosen to yield 15% polarization for light emitted at the limb of the photosphere @xcite . opaque circular clumps were assumed , with @xmath120 , such that all photospheric flux from the region underlying the clump was blocked . for each iteration of the model , @xmath96 clumps of size @xmath121 were randomly positioned across the photosphere and the corresponding flux of the uncovered regions , and their polarization components , were summed . clumps were permitted to overlap and only partially cover the photosphere . for each size of clumps , 3000 simulations were conducted . this permits a comparison of the expected polarization for a given absorption line depth , assuming a particular clump size in the line forming region . the results of this model ( for @xmath122 , @xmath123 , @xmath124 and @xmath125 ) are shown as fig . [ fig : obs : monte ] . + for some of the models , zero polarization was achieved for uniform coverage of the photosphere by the clumps , however the difference between the average polarizations expected ( as shown in fig . [ fig : obs : monte ] ) and null polarization was @xmath126 for all the calculated models . these models suggest that , for a clumpy line forming region , the clump scale for sn 2005hk is very small ( @xmath127 ) to produce lines of the depths observed , in the flux spectra at both epochs , with low polarization . the uniformly low polarization observed for lines arising from different species suggest that the ejecta are generally mixed , with no separate structures _ on the plane of the sky _ for different elements or ionization stages(e.g and ) . the element responsible for the absorption at 7200 may be an exception , with a line forming region that is more aspherical than other species . the data for sn 2005hk , at the two epochs presented here and those presented by @xcite , are consistent with an approximately spherical symmetric photosphere surrounded by similar line forming regions for all species . + the principal assumption of the model presented above is that the line forming region and the photosphere are separate . as discussed in [ sec : res : domort ] , the opacity at the photosphere ( where the quasi - continuum forms ) is not due to electron scattering alone , but rather includes a significant wavelength dependent contribution from atomic transitions ; this implies that line forming regions are not , in reality , separate entities except in the most high - velocity regions such as those discussed by @xcite . + the size of the clumps and their relation to `` large - scale '' structure within the ejecta , leads to an interesting problem of semantics : since very large clumps ( @xmath128 ) may be considered large scale structure and very small clumps ( @xmath129 ) tend towards constituting an effectively homogeneous medium . large opacities require that a large fraction of the photosphere , the last thomson scattering distance , are reprocessed by lines , mostly due to iron - group elements ( which will result in a significantly altered spectrum ) . the class of sne similar to sn 2002cx currently has 15 members , including 2005hk and 2008 ha @xcite . as discussed in detail by @xcite and @xcite , both sn 2002cx and sn 2005hk were of especially low velocity at the photosphere near maximum light , @xmath130 @xmath0 , had early relatively high ionization , and low peak luminosity . neither showed the typical secondary maxima in r and i of normal sn ia , but were also dissimilar to previously studied under - luminous sn ia that do not show the secondary maxima ( e.g. * ? ? ? there is also extreme diversity within the class , with sn 2008 ha being @xmath131 magnitudes fainter than sn 2002cx @xcite . the light curve decline is rather slow , unlike previous under - luminous sn ia that serve to define the light curve brightness / decline relation prominently used in cosmology studies . + @xcite pointed out that no published model of a sn ia corresponds to all the behavior of sn 2002cx , but suggested that a pulsating delayed detonation model might be worth considering further . @xcite suggested that the nebular spectral characteristics that they determined might be consistent with a pure deflagration model since the models in the literature have insufficient energy to correspond to normal sne ia and hence tend to have smaller expansion velocities and produce less @xmath132ni and hence tend to be dim . @xcite suggest that the low expansion velocities , low peak luminosity , and the absence of secondary maxima in the nir light curves may be reproduced by a 3d pure - deflagration model that produces 0.25@xmath133 of @xmath132ni . + a key issue with the hypothesis that this class of explosions represents pure deflagrations of white dwarfs is that the pure deflagration models leave a substantial amount of unburned c and o in the outer layers . this c and o should show up in the early phases and while @xmath20 is clearly present at the second epoch ( and possibly the first ) , the case for @xmath30 ( and possibly for a contribution to the @xmath31 feature ) is less clear but it is not conclusively absent . in addition , suggest that the late - time spectra of pure deflagration models show no similarity to that of sn 2002cx either , in particular the absence of strong features due to c and o. @xcite claim a marginal detection of in late - time nebular spectra of sn 2002cx , with the suggestion that more unburned c and o may be contained inside the dense , still optically thick core at late times . @xcite claim that at early times @xmath20 is suppressed by high temperatures and , at late times , nebular lines are also suppressed by high temperatures . the lack of observation of @xmath20 for sn 2005hk at very early times ( before our observations ) is uncertain , due to poor spectral coverage at early times @xcite and low s / n @xcite . @xcite are able to approximately model the photospheric spectrum of sn 2005hk , based on the w7 deflagration explosion model @xcite . importantly , @xcite find that , given the conditions of the sn , their spectral synthesis suggests large amounts of unburned o. their model also assumes spherical symmetry , for which they achieve a reasonable approximation to the line profiles ( which are somewhat sensitive to any departures from spherical symmetry ; see e.g. @xcite for a counter example ) . the features remain too broad , are forbidden rather than permitted , and show prominent emission of [ ] @xmath134 . high densities , as evidenced by the low expansion velocities and p cygni profiles observed at late nebular epochs , might suppress the [ ] feature , but it is predicted in the models and so its absence is a definite constraint on the nature of sn 2005hk . + @xcite consider the possibility that these events may be related to normal sne ia , but viewed along the hole " left by having the binary companion nearly directly along the line of sight @xcite . such an aspect might increase the prominence of high - ionization species , weaken absorption lines , reduce the photospheric velocity and induce some polarization due to the non - spherical geometry of the ejecta . one issue with this model for sn 2002cx and related explosions is that it would result in higher luminosities than when viewed from other aspect angles . even if the underlying explosion were a subluminous event such as sn 1991bg , if the view were down the hole , " the result would be brighter than sn 1991bg . both sn 2002cx and sn 2005hk were subluminous by about 1 magnitude in all bands ( except h ; * ? ? ? it is certainly not clear that this is consistent with a hole " model . @xcite comment that their measurement of a small but appreciable continuum polarization and weak line features was inconsistent with the prediction of @xcite of a continuum polarization of @xmath2 0.1 percent and strong line polarization peaks for an explosion viewed slightly off the axis of the hole . as the sn evolves , the relative size of the hole and the photosphere will change ( in particular if the hole closes , which @xcite expect to be a slow process if it occurs ) , which should lead to a commensurate change in the polarization properties . importantly , we see only moderate change in the polarization , despite the significant changes in the flux spectrum , between the epochs at which sn 2005hk was observed . in addition , it is not at all clear how such a model would account for the late - time permitted line spectrum delineated by @xcite . our observations do suggest some line polarization . this demands some asymmetry in the ejecta , consistent with a single axial symmetry , but does not spell out quantitative agreement with the hole " model . + an alternative model might be a plume " model @xcite in which a single plume ( or very few plumes ) reach the surface , spreading fresh @xmath132ni . if the plume burned little , the expansion might be slow and leave high densities at late times . our spectropolarimetry suggests that the outer layers of the ejecta are approximately homogeneously mixed , although stratified , which suggests mixing in a large number of smaller plumes ( with size @xmath135 at the first epoch ) . + this class of objects thus presents many conundra . they show high excitation early on but otherwise have none of the properties of sne ia like sn 1991 t that show early . their light curves are dim but broad , violating the `` phillips relation '' . they have very slow expansion velocities and may remain partially optically thick at nearly 300 days after the explosion . they are mildly asymmetric in the continuum , in the early on and in near maximum . the polarization of sn 2005hk is completely compatible with the observed range of polarization properties for type ia sne ( see table [ tab : disc : comp ] ) . the level of polarization of sn 2005hk is not the lowest _ detected _ for a sn , with sn 1996x ( a normal type ia sn ; @xcite ) showing polarization just above statistical significance and with no apparent dominant axis . the observations of sn 2005hk show it to have a spectropolarimetric type of d1 , similar to most type ia sne , indicating that there is a principal axis of symmetry to the ejecta . importantly , there are no apparent loop like features , even for the strongest lines ( which is particularly pertinent for the observations at the second epoch ) which would indicate a significant departure from this main axial symmetry . normal type ia sne , such as sn 2001el @xcite and sn 2004s @xcite , show some structure in their ejecta principally in the , as well as the high velocity component of the ir3 . conversely , there are type ia sne , such as sn 2004dt @xcite , whose line polarization is principally in the dominant polarization component which is consistent with a single axial symmetry . in the case of sn 2004dt , however , the observed line polarization was @xmath136 . in summary , the similarity between the spectropolarimetry of sn 2005hk and type ia sne indicates that their respective explosion mechanisms are both approximately spherical ( see table [ tab : disc : comp ] ) . + @xcite identified a correlation between the degree of polarization of , at -5 days , with the decline rate ( @xmath137 ) of the light curve . at some wavelengths sn 2005hk is an outlier from the phillips brightness - decline relationship , although it is consistent with the relationship in the @xmath138 and @xmath139 bands @xcite . while the polarization measurements of sn 2005hk is offset from the measured relation derived for normal type ia sne by @xcite , it is not as significant an outlier as the subluminous type ia sne 1999by and 2005ke ( see fig . [ fig : disc : wang ] ) . + @xcite , @xcite and @xcite consider the possibility that events such as sn 2002cx and 2005hk are not related to type ia sne , but may instead be a type of ccsn . based on the photometric and spectroscopic similarity between sn 2008 ha and sn 2005hk with low - luminosity type iip and type ic sne ( 2005cs and 2007gr , respectively ) and the occurrence of these events in late - type galaxies , @xcite suggest 2002cx - like sne are low - luminosity type ic sne , potentially associated with gamma ray bursts ( grbs ) . @xcite present additional events in this class , which are claimed to have taken place in lenticular s0 galaxies , which may argue against massive stars as the universal progenitors for this type of event . spectropolarimetry provides additional parameter space in which to examine the similarities between different types of sne . + @xcite and @xcite note the spectroscopic similarity between sn 2005hk and sn 2008 ha at + 15 and + 8 days ( after optical maximum ) , respectively , as well as their similarity with the subluminous type iip sn 2005cs at + 44 days , whilst it was still on the photometric plateau @xcite . at such times type iip sne have generally low polarizations @xcite , seemingly similar to that observed by us for sn 2005hk . from a spectropolarimetric perspective , however , comparison of sn 2005hk with sn 2005cs whilst on the plateau is inappropriate . during the plateau phase , in type iip sne , the photosphere is located in the hydrogen envelope . due to the low ionization energy of h , the shape of the photosphere is approximately spherical . once the photosphere has descended below the h envelope , into the he layer ( he being harder to ionize and excite ) , the shape of the photosphere more closely follows the shape of the underlying exciting material . despite an asymmetric explosion , @xcite observed that during the plateau sn 2004dj had a low polarization , which increased significantly once the plateau phase had ended . given the h - deficient nature of sn 2002cx - like sne , a more appropriate comparison is with the h - deficient ccsne , which have been observed ( as listed in table [ tab : disc : comp ] ) to have significant continuum and line polarization for different species indicating the role of highly asymmetric explosion mechanisms . highly asymmetric interiors have also been observed for the type iip sne 2001dh and 2003gd ( maund et al . , 2010 , in prep . ) , which suggest that an asymmetric explosion mechanism is common to all these ccsn events . + it is interesting to note another ongoing controversy involving sne ia and sne ic . sn 2002ic was revealed by @xcite to resemble a sn ia embedded in a hydrogen - rich medium . @xcite interpreted this as a sn ia exploding with an evolved binary companion . @xcite showed that the hydrogen component was strongly polarized and argued that the geometry was consistent with an extended , clumpy , disk - like structure such as observed in proto - planetary nebula stars . @xcite and @xcite identified other events of this category that were originally identified as type iin events , but which , on closer inspection and comparison with sn 2002ic , revealed an underlying type i spectrum . more recently , @xcite have challenged this interpretation , arguing that the underlying supernova in sn 2002ic more closely matched a sn ic than a sn ia . + in summary , the nature of the polarization of sn 2005hk is inconsistent with the polarization measured for ccsne , in particular those of type ibc which have been generally observed to have very large polarization ( especially line polarization ) that increases with time . sn 2005hk has been observed at two epochs , corresponding to maximum light and two weeks after maximum . after subtraction of the isp component , the sn is observed to have a very low to null intrinsic polarization . a modulation in the stokes parameters is observed across the @xmath22 line being the most significant feature , at the first epoch , with @xmath140 . + at both epochs , a dominant axis can be fit to the data , although the data are poorly described by simple straight line fits on the stokes @xmath53 plane . while there is modulation in the dominant polarization component , at the first epoch , that is associated with line features in the flux spectrum , the absence of similar modulations in the orthogonal polarization component implies that the line forming regions of the species in the flux spectrum share the same axial symmetry as that of the photosphere . the data at both epochs is classified as being of spectropolarimetric type d1 . + in comparison with previously published high s / n observations of sn 2005hk at -4 days @xcite , we observe a possible rotation of the polarization angle of the dominant axis , but it is unclear ( due to the differences in s / n ) if this represents a real rotation of the principal axis of symmetry with depth into the ejecta . + the low level of continuum polarization limits the asymmetry of the photosphere , at both epochs , to @xmath1 . the low line polarization is suggestive that the line forming regions are mixed , that different species occupy similar locations within the ejecta , and that they have a similar geometry to that of the photosphere . this is similar to the general behaviour of type ia sne , and different from the chemically segregated ejecta observed for ccsne . + we conclude that the spectropolarimetry of sn 2005hk is not inconsistent with some of the expectations of the deflagration explosion model . we find that , in order to preserve the approximate spherical symmetry of the photosphere and the line forming regions , any plume - like structures would be required to have a scale , in the plane of the sky , of @xmath135 at both epochs . this would give the overall appearance of homogeneously mixed ejecta containing both the products of nucleosynthesis ( fe , ni , co ) and unburned o at all depths . + spectral similarities between sn 2002cx - like events and low - luminosity type iip sne are likely to be merely superficial . the differences in the observed spectropolarimetry of sn 2005hk with ccsne suggests , from a geometry perspective , that these events have different underlying explosion mechanisms . unfortunately , with spectropolarimetry of only one sn from the 2002cx - like class it is not possible to determine if these results are applicable to the entire class , in particular to the most peculiar member of this class , sn 2008 ha . the research of jrm is funded through the sophie & tycho brahe fellowship . the dark cosmology centre is supported by the dnrf . the research of jcw is supported in part by nsf grant ast-0707769 . pah is supported by the nsf grants ast 0708855 and 0703902 . the authors are grateful to the european organisation for astronomical research in the southern hemisphere for the generous allocation of observing time . they especially thank the staff of the paranal observatory for their competent and never - tiring support of this project in service mode . the authors thank the anonymous referee for useful comments and suggestions that have improved the manuscript . , a. v. , richmond , m. w. , branch , d. , gaskell , m. , herbst , w. , ford , c. h. , treffers , r. r. , matheson , t. , ho , l. c. , dey , a. , sargent , w. l. w. , small , t. a. , & van breugel , w. j. m. 1992 , , 104 , 1543 , w. , filippenko , a. v. , chornock , r. , berger , e. , berlind , p. , calkins , m. l. , challis , p. , fassnacht , c. , jha , s. , kirshner , r. p. , matheson , t. , sargent , w. l. w. , simcoe , r. a. , smith , g. h. , & squires , g. 2003 , , 115 , 453 , p. a. , benetti , s. , altavilla , g. , blanc , g. , cappellaro , e. , elias - rosa , n. , garavini , g. , goobar , a. , harutyunyan , a. , kotak , r. , leibundgut , b. , lundqvist , p. , mattila , s. , mendez , j. , nobili , s. , pain , r. , pastorello , a. , patat , f. , pignata , g. , podsiadlowski , p. , ruiz - lapuente , p. , salvo , m. , schmidt , b. p. , sollerman , j. , stanishev , v. , stehle , m. , tout , c. , turatto , m. , & hillebrandt , w. 2005 , , 623 , l37 , f. , chandra , p. , chevalier , r. , justham , s. , podsiadlowski , p. , wolf , c. , gal - yam , a. , pasquini , l. , crawford , i. a. , mazzali , p. a. , pauldrach , a. w. a. , nomoto , k. , benetti , s. , cappellaro , e. , elias - rosa , n. , hillebrandt , w. , leonard , d. c. , pastorello , a. , renzini , a. , sabbadin , f. , simon , j. d. , & turatto , m. 2007 , science , 317 , 924 , m. m. , li , w. , frieman , j. a. , blinnikov , s. i. , depoy , d. , prieto , j. l. , milne , p. , contreras , c. , folatelli , g. , morrell , n. , hamuy , m. , suntzeff , n. b. , roth , m. , gonzlez , s. , krzeminski , w. , filippenko , a. v. , freedman , w. l. , chornock , r. , jha , s. , madore , b. f. , persson , s. e. , burns , c. r. , wyatt , p. , murphy , d. , foley , r. j. , ganeshalingam , m. , serduke , f. j. d. , krisciunas , k. , bassett , b. , becker , a. , dilday , b. , eastman , j. , garnavich , p. m. , holtzman , j. , kessler , r. , lampeitl , h. , marriner , j. , frank , s. , marshall , j. l. , miknaitis , g. , sako , m. , schneider , d. p. , van der heyden , k. , & yasuda , n. 2007 , , 119 , 360 , m. , tominaga , n. , nomoto , k. , valenti , s. , sahu , d. k. , minezaki , t. , yoshii , y. , yoshida , m. , anupama , g. c. , benetti , s. , chincarini , g. , della valle , m. , mazzali , p. a. , & pian , e. 2009 , , 692 , 1131 , s. , pastorello , a. , cappellaro , e. , benetti , s. , mazzali , p. a. , manteca , j. , taubenberger , s. , elias - rosa , n. , ferrando , r. , harutyunyan , a. , hentunen , v. p. , nissinen , m. , pian , e. , turatto , m. , zampieri , l. , & smartt , s. j. 2009 , , 459 , 674 ccccc object & date & exposure & median & type + & ut & ( s ) & airmass & + sn 2005hk&2005 nov 09.163 & @xmath141 & 1.255 & sn + hd 49798 & 2005 nov 09.363 & 5 & 1.067 & flux std . + + sn 2005hk&2005 nov 23.096 & @xmath141 & 1.160 & sn + ltt 1020 & 2005 nov 23.132 & 5 & 1.139 & flux std . + lccc species & @xmath142 & @xmath143 & @xmath72 + + h k & 3780 & 3960 & @xmath144 + & 4225 & 4315 & @xmath145 + & 6120 & 6270 & @xmath146 + ? & 7070 & 7215 & @xmath147 + ir3 & 8340 & 8490 & @xmath148 + + h k & 3750 & 3930 & @xmath149 + & 4017 & 4077 & @xmath150 + & 4610 & 4910 & @xmath151 + & 5910 & 6030 & @xmath152 + ? & 7130 & 7220 & @xmath153 + cccccccccccc & & & & & & + date & & @xmath84 & @xmath83 & @xmath154 & @xmath155 & & axial ratio & @xmath97 & & @xmath156 & @xmath157 + & & & & & & & @xmath98 & & & & + 09 nov 2005@xmath158 & & @xmath159 & @xmath160 & 347.2/154 & @xmath161 & & 0.87 & 3.7 & & @xmath162&@xmath163 + 09 nov 2005@xmath164 & & @xmath165 & @xmath166 & 202.3/98 & @xmath167 & & 0.78 & 8.7 & & @xmath168&@xmath169 + + 23 nov 2005@xmath158 & & @xmath170 & @xmath171 & 288.8/154 & @xmath172 & & 0.47 & 14.0 & & @xmath173&@xmath174 + 23 nov 2005@xmath164 & & @xmath175 & @xmath176 & 160.9/98 & @xmath177 & & 0.42 & 18.55 & & @xmath178 & @xmath179 + + + + @xmath180 relative to optical maximum ; @xmath181 sn spectropolarimetric types see ; @xmath182 relative to the second maximum ; references ( 1 ) @xcite ; ( 2 ) @xcite ; ( 3 ) @xcite ; ( 4 ) @xcite ; ( 5 ) @xcite ; ( 6 ) @xcite ; ( 7 ) ; ( 8) @xcite ; ( 9 ) @xcite , @xcite , @xcite ; ( 10 ) @xcite ; ( 11 ) @xcite ; ( 12 ) @xcite . + -band image of the site of sn 2005hk , indicated by the filled circle , relative to its host galaxy ugc 272 . the orientation of the determined isp component , at @xmath183 , is shown by the arrow.,width=321 ] ) ; the observed polarization spectrum ( @xmath184 ) ; the observed polarization spectrum corrected for the isp component ( @xmath52 ) ; the stokes @xmath10 and @xmath11 spectra uncorrected for the isp ( the value of the isp is shown for each parameter by the grey line ) ; and the polarization angle @xmath185 . line identifications are provided in the top panel , and are based on the identifications by @xcite , @xcite , @xcite and comparison with synthetic spectra produced using the synow code . the data have been rebinned to 30@xmath186 for clarity . the wavelength range selected for the determination of the isp are shown by the heavy lines.,width=340 ] plane . the data have been binned to 30@xmath186 for clarity , and have been corrected for the isp component , which is indicated on the stokes plane by the filled black circle , with the radius of the circle corresponding the measured uncertainties . the data are colour coded according to wavelength following the scheme of the colour bar . the origin of the stokes plane is indicated by the cross ( @xmath187 ) . the black dashed circle represents the null polarization envelope , at 6000@xmath186 . the heavy black line indicates the best fit to the `` dominant axis . '' the ellipse shows the apparent axial ratio and orientation of the data , as determined using the principal components analysis , with the size of the ellipse arbitrarily scaled.,width=321 ] ) and the orthogonal axis ( @xmath80 ) . the data have been binned to 30@xmath186 for clarity , and have been corrected for the isp component . the weighted average of the isp , rotated onto the dominant and orthogonal axes , is indicated by the heavy grey lines.,width=283 ] ( blue @xmath188 ) , @xmath123 ( red @xmath189 ) , @xmath124 ( green @xmath190 ) and @xmath125 ( black @xmath191 ) . solid lines are for simulations with @xmath192 , while dashed lines indicate simulations with @xmath117 . the shaded area indicates the permitted values of line polarization , according to eqn . [ eqn : res : linepollim ] , for a continuum polarization of @xmath58 determined by @xcite . lower values of the continuum polarization ( @xmath193 determined here ) place even stricter upper limits on the scale of clumping . minor fluctuations in the curves are due to the statistical nature of the monte carlo simulations ; except for the case of @xmath128 for @xmath194 , where the results are dominated by the finite size of the `` pixels '' that compose the model photosphere.,width=226 ] as a function of light curve decline rate . our measurements of sn 2005hk are shown in red , with the value of @xmath72 ( ) indicated by the point with the heavy outline , data in black are taken from @xcite , the measurement for sn 2006x ( @xmath195 ) is from , the measurement for the luminous event sn 2009dc ( @xmath188 ) is taken from @xcite , and the measurement for sn 2005ke ( @xmath189 ) is taken from patat et al . , ( 2010 , in prep.).,width=321 ]

in this paper we study new - physics contributions to the top - quark decay . we search for ways of detecting such new physics via measurements at the lhc . as top quarks are mainly produced at the lhc in production via gluon fusion , we analyze the process . we find six observables that can be used to reveal the presence of new physics in . three are invariant mass - squared distributions involving two of the final - state particles in the top decay , and three are angular correlations between the final - state quarks coming from the decay and the coming from the decay . the angular correlations are related to the spin correlation .

1407.1724

physics beyond the standard model ( sm ) is expected to exist at energies above the weak scale . while successive experiments at lep , the tevatron and the lhc have served to validate the sm over the past few decades , no direct evidence of new physics ( np ) has been found yet . clearly , np either exists at an energy scale higher than what has been probed , or its hints are subtler than we envision . the lhc , which is currently operational , is essentially a top - quark factory . the properties of the @xmath3 can therefore be measured with good precision . now , the mass of the top quark is more than an order of magnitude larger than that of all other fermions . as such , it may be affected by np in ways that do not manifest themselves in the interactions of the lighter fermions . in addition , its large mass causes the top to decay before it can hadronize , so that it can be studied more or less as a free quark . in this paper we study np contributions to top - quark decay . the dominant @xmath3 decay modes in the sm involve @xmath6 , with @xmath7 , @xmath8 or @xmath9 . since the experimentally - measured value of the top decay width is in good agreement with the sm prediction @xcite , it is evident that the np contribution to the dominant decay modes , if any , is very small compared to that of the sm . on the other hand , in the case of decay modes that are suppressed in the sm , an np contribution that is comparable to that of the sm in that mode may go unnoticed simply because its impact on the total width is small . this makes it interesting to probe rare decays , as these could well be where the new physics is lurking . one such decay is @xmath10 . it is suppressed in the sm because it involves the small element @xmath11 ( @xmath12 ) of the cabibbo - kobayashi - maskawa ( ckm ) quark mixing matrix . there are other suppressed decays ( e.g. , @xmath13 ) , but here we focus on @xmath0 . single - top production is rather suppressed at the lhc @xcite , so that it is difficult to isolate the decay @xmath14 experimentally and analyze it on its own . the most significant production mode for top quarks at the lhc is pair ( @xmath1 ) production . at lhc energies , this is dominated by gluon fusion ( @xmath15 ) , as opposed to quark - antiquark annihilation ( @xmath16 ) . in order to search for np in top decay , the full process @xmath17 , with @xmath14 and @xmath18 , must be analyzed . apart from the usual difficulties of studying a multi - particle final state , this channel suffers from another complication the @xmath5 decay leads to a second @xmath19 in the final state , providing an additional background that must be taken into account . the main purpose of this paper is to analyze the process @xmath20 , and to look for observables that can reveal the presence of np in top decay , along the lines of ref . @xcite , will be examined elsewhere @xcite . ] . we will show that there are two types of observables that can be used . the first is simply an invariant mass - squared distribution involving two of the final - state particles in @xmath14 . with np , its form is altered compared to that of the sm . note , however , that this type of observable is entirely related to the decay of the @xmath3 itself . the associated production of the @xmath5 is unimportant , except insofar that one must distinguish the @xmath19 quarks coming from the @xmath3 and @xmath5 decays . the second type of observable does rely on the fact that a @xmath1 pair has been produced . the key point is that , in @xmath1 production , the spins of the @xmath3 and @xmath5 are correlated @xcite . the spin - correlation coefficient for the produced @xmath1 pair can be defined as @xmath21 where @xmath22 and @xmath23 denote the alignment of the spins of the top and antitop with respect to the chosen spin - quantization axis . the spin of the @xmath3 itself is related to the angular distribution of its decay products through the relation @xmath24 where @xmath25 is the angle between the direction of the @xmath26 decay product and the spin quantization axis in the rest frame of the top , and @xmath27 is a numerical coefficient whose value depends on the identity of this decay product . the spin of the @xmath5 is related to the angular distribution of its decay products through a similar relation , with @xmath28 and @xmath29 . naturally then , the spin correlation between the pair - produced top and antitop is manifested in the angular correlation between the decay products of the two particles . that relation is given as follows @xcite : @xmath30 its measurement permits the extraction of @xmath31 . if the measured value differs from the prediction of the sm , it would indicate the presence of np . one point should be noted at this juncture . the spin - correlation coefficient @xmath31 is , by definition , a property of the @xmath1 production process . however , its experimental determination depends on the decay . equation ( [ eq : angcorr ] ) assumes that the @xmath3 and @xmath5 decay via sm interactions only . if there are np contributions in top decay , the value of @xmath31 extracted from the angular correlations of the top and antitop decay products will be different from the sm prediction . this would not be due to a change in the value of @xmath31 itself , but rather to a change in the form of eq . ( [ eq : angcorr ] ) . while there have been several studies of the effect of np on @xmath1 spin correlations , most of them have focused on np that affects @xmath32 production . these span both cp - conserving @xcite and cp - violating @xcite np scenarios . possibilities include non - standard @xmath33 couplings in the form of anomalous chromomagnetic dipole or chromoelectric dipole interactions , as well as many of the np models proposed to explain the large @xmath1 forward - backward asymmetry observed at the tevatron @xcite . of course , np contributions may be present in both @xmath1 production and in the decay . however , np in the production is much easier to detect , in that it should be observable even in the dominant decay modes of the top . for this reason we ignore the possibility of np in @xmath34 production in our analysis . we assume it will have been detected or ruled out before the study of np in the decay is done . once the observables that carry the signature of np have been pinpointed , the next question is : to what extent can they realistically be used to probe np in top decay ? can they be used to identify , even partially , the type of np present ? this is examined in the companion paper @xcite . there we show that it is likely that there will be enough events at the lhc to measure these observables reasonably precisely and extract information about the nature of np at play . in this paper , we begin in sec . ii by examining how np in top decay can affect @xmath14 . iii we briefly discuss the full pair production and decay chain @xmath2 ( full details are given in the appendix ) . the observables that can be used to search for np in top decay are described in sec . iv . in sec . v we perform a numerical simulation of @xmath2 at the lhc , including np , and compare the results for the observables with our analytical calculations . we conclude in sec . as detailed in the introduction , this work focuses on the search for new physics in rare decays of the top quark . in this paper , we examine the decay @xmath14 . however , the method described here can also be applied to other suppressed decays such as @xmath35 , etc . while examining a suppressed decay mode , one must consider the most dominant production mode in order to have sufficient statistics . hence , the search for np in this top decay mode must involve the process @xmath17 . even there , one may have chosen to ignore the details of the production process and focus only on the decay . however , as we show in the following sections , there is something to be gained by considering the full process @xmath2 , in that the @xmath34 spin correlations can be put to use in the identification of np . in the sm , the decay @xmath14 arises via @xmath36 , followed by @xmath37 . np contributions to @xmath14 can be parameterized via an effective lagrangian @xmath38 , with @xmath39 in the above expressions , colour indices are not shown , but are assumed to contract in the same manner as those of the sm ( i.e. , the fields @xmath19 with @xmath3 and @xmath40 with @xmath41 ) . in some np models , the colour indices would contract in the opposite manner ( i.e. , the fields @xmath40 with @xmath3 and @xmath19 with @xmath41 ) . however , with fierz transformations it is straightforward to incorporate colour - mismatched terms into the effective lagrangian @xcite . in general , the np couplings ( the @xmath42 s in the above equations ) have both weak and strong phases . however , as argued in ref . @xcite , since the np strong phases can only be generated by self - rescattering from the np operators , they are very small . for this reason , we neglect all np strong phases , so that the @xmath42 s contain only weak phases . furthermore , the np couplings can all reasonably be assumed to be of order unity , so that the sm and np contributions to @xmath14 can very well be about the same size . when computing the effect of np on a particular observable , it is therefore important to include both the sm - np and np - np interference pieces . we calculate the square of the matrix element for @xmath14 as a function of the top - quark spin ( @xmath44 ) , including the sm and all the np contributions . we find @xmath45 ~ , \label{eq : ampsquaredtbbbarc}\end{aligned}\ ] ] where @xmath46 , with @xmath47 and where @xmath44 is the spin four - vector of the top quark . above , @xmath48 , @xmath49 and @xmath50 . @xmath51 is defined as @xmath52 ~ , \label{eq : abbar}\end{aligned}\ ] ] where @xmath53 and @xmath54 . the remaining @xmath55 are defined as @xmath56 in the above , @xmath57 note that @xmath51 contains both the sm and np contributions , whereas the other @xmath55 contain only np contributions . the term proportional to @xmath58 in eq . ( [ eq : ampsquaredtbbbarc ] ) describes the triple product ( tp ) in the decay . because the @xmath42 s contain only weak phases , the tp is purely cp - violating . furthermore , eq . ( [ eq : abbar ] ) contains terms proportional to @xmath59 and @xmath60 . of these , @xmath60 is also cp - violating . now , if one adds eq . ( [ eq : ampsquaredtbbbarc ] ) to its cp - conjugate counterpart , all cp - violating terms cancel , leaving the remaining terms unchanged ( apart from a normalization factor of @xmath61 ) . in focusing on cp - conserving observables , we implicitly assume that this cp averaging has been performed . the main point to be retained from eq . ( [ eq : ampsquaredtbbbarc ] ) is that the amplitude squared depends on seven different combinations of np couplings six @xmath62 s and re@xmath63 . thus , there are a number of independent observables that , in principle , can provide information about the np . while we can hope to measure all seven of these quantities , we can not measure all of the individual @xmath42 parameters . in the remainder of this paper ( and in the companion paper ) , when we refer to `` identifying '' the np , what is meant is this partial identification of the six @xmath62 s and re@xmath63 , not the complete identification of all of the @xmath42 parameters . as a first step , we calculate the cross - section for @xmath1 pair production followed by the decay chain @xmath64 , @xmath65 . we present an outline of the analysis in what follows ; the more technical details can be found in the appendix . briefly , the analysis proceeds as follows . the process is represented in fig . [ fig : kinematics ] . the six - body phase space is decomposed into five solid angles and four invariant masses . the narrow - width approximation and @xmath5 quarks produced via gluon fusion , to the @xmath66 produced in the @xmath5 decay , and generally to the @xmath66 produced in the @xmath3 decay . ] is then used for the @xmath3 and @xmath5 quarks to eliminate two of the invariant - mass degrees of freedom . the solid angles @xmath67 , @xmath68 , @xmath69 , @xmath70 and @xmath71 are defined in five different rest frames , as indicated in fig . [ fig : kinematics ] . the @xmath72 and @xmath73 superscripts indicate that these angles are defined in reference frames that are , respectively , one and two boosts away from the @xmath34 rest frame . the invariant masses @xmath74 and @xmath75 are defined through the relations @xmath76 and @xmath77 . in the end , the differential cross section is a complicated function of the final - state momenta @xmath78 ( @xmath79 - 6 ) and the couplings , and is defined with respect to @xmath80 . we stress that fig . [ fig : kinematics ] represents only the kinematics of @xmath2 . it is not a feynman diagram . in particular , @xmath81 does not necessarily correspond to the @xmath82 resonance in the @xmath5 decay , and @xmath83 does not necessarily correspond to the @xmath84 resonance in the sm part of the @xmath3 decay . rather , @xmath85 , @xmath86 and @xmath87 are the momenta of the @xmath41 , @xmath19 and @xmath88 quarks in @xmath14 , with all permutations being allowed . that is , @xmath85 , @xmath86 and @xmath87 can each stand for @xmath89 , @xmath90 or @xmath91 , and similarly for the particles in the @xmath5 decay . in constructing the observables , we consider several of these possibilities . the first step in finding observables that can yield information about np in top decay is to define which final - state particles correspond to @xmath85-@xmath92 . there are several choices possible , corresponding to different observables . throughout this work the momenta for the @xmath5 decay products are assigned as follows : @xmath93 , @xmath94 and @xmath95 . taking @xmath96 , @xmath97 and integrating eq . ( [ eq : dsigma2 ] ) from the appendix over @xmath81 and over all angles except @xmath98 and @xmath99 , we find is `` @xmath100 '' in this case ( see the caption of fig . [ fig : kinematics ] for a precise definition ) . this angle is associated with the direction of the @xmath41-@xmath88 center of mass in the top rest frame , which is opposite to the direction of the @xmath19 in this frame . similarly , @xmath99 is `` @xmath101 '' . ] @xmath102\nonumber\\ & & \!\!\!\!\!\!\!+~\frac{3 g_f m_t^2 } { \sqrt{2}\pi^2 \left(1-\zeta_w^2\right)^2 \left(1 + 2\zeta_w^2\right ) } \sum_{i,\sigma } \hat{a}_i^\sigma\left[h_i^{bc}\left(\zeta_{bc}^2\right ) + \widetilde{h}_i^{bc } \left(\zeta_{bc}^2\right ) \xi^\sigma \kappa(r ) \cos\theta_{\bbar}^*\cos\theta_\ell^*\right]\!\ ! \bigg\},~~ \label{eq : dsigdcosdcosdm}\end{aligned}\ ] ] where @xmath103 and @xmath104 . @xmath105 is defined in eq . ( [ eq : sigmasm ] ) of the appendix , the @xmath106 s are given in eq . ( [ ahatdefs ] ) , and @xmath107 is defined as @xmath108 } ~,\end{aligned}\ ] ] where @xmath109 is defined in eq . ( [ eq : pt ] ) in the appendix . note that @xmath110 = -@xmath31 as defined in eq . ( [ eq : spincorrcoeff ] ) . the functions @xmath111 and @xmath112 are defined in table [ tab : hs ] , and h_^bc(_bc^2 ) = ( 1-_bc^2)_bc^2 ( 1-_w^2-_bc^2 ) . [ eq : hsm ] in writing down eq . ( [ eq : dsigdcosdcosdm ] ) , we have dropped a contribution proportional to re@xmath113 , which tends to yield a somewhat small effect in practice . this contribution is not difficult to calculate , but its inclusion makes the expression for the differential cross section somewhat cumbersome . also , since we are only considering cp - even contributions , we have dropped a term proportional to im@xmath113 . .definitions of the @xmath114 ( @xmath115 ) and @xmath112 functions . the columns correspond to @xmath116 , @xmath117 , @xmath88 . [ cols="^,^,^,^",options="header " , ] [ tab : hs ] starting from eq . ( [ eq : dsigdcosdcosdm ] ) , we can integrate one or two more times to obtain differential cross sections in terms of the two angles or in terms of the invariant mass squared , respectively . these are the two types of observables we focus on in this paper : * invariant mass - squared distribution . * integrating over the angles @xmath98 and @xmath99 in eq . ( [ eq : dsigdcosdcosdm ] ) yields @xmath118 the above contains three functions of @xmath119 that multiply the various sm and np terms : @xmath120 , @xmath121 and @xmath122 . the three functions are qualitatively different from each other , so that the measurement of the invariant mass - squared distribution permits the extraction of the np parameters @xmath123 and @xmath124 . * angular correlation . * integrating over @xmath119 in eq . ( [ eq : dsigdcosdcosdm ] ) , we obtain @xmath125\nonumber\\ & + & \frac{3 g_f m_t^2 } { 4\sqrt{2}\pi^2 \left(1-\zeta_w^2\right)^2 \left(1 + 2\zeta_w^2\right)}\bigg [ \bigg(\sum_{i,\sigma } \hat{a}_i^\sigma\bigg ) \\ & + & \bigg(\hat{a}_{\bbar}^+-\hat{a}_{\bbar}^- -\frac{1}{3}\big ( \hat{a}_b^+-\hat{a}_b^-+\hat{a}_c^+-\hat{a}_c^- \big)\bigg)\kappa(r ) \cos\theta_{\bbar}^*\cos\theta_\ell^*\bigg ] \bigg\}. \nn\end{aligned}\ ] ] by measuring this differential cross section and comparing it to the sm prediction , one can extract the sum of np parameters @xmath126 and a linear combination of the differences @xmath127 ( @xmath50 ) . note that this observable is sensitive to the @xmath1 spin correlation . for the sm , this is just the coefficient of the term proportional to @xmath128 , up to an overall normalization factor . once np is included , this term gets an additional contribution proportional to a combination of differences of the np parameters . the corresponding angular correlation is given by @xmath137\nonumber\\ & + & \frac{3 g_f m_t^2 } { 4\sqrt{2}\pi^2 \left(1-\zeta_w^2\right)^2 \left(1 + 2\zeta_w^2\right)}\bigg [ \bigg(\sum_{i,\sigma } \hat{a}_i^\sigma\bigg ) \nonumber\\ & & \hskip-2truecm + ~\bigg(\hat{a}_b^+-\hat{a}_b^- -\frac{1}{3}\big ( \hat{a}_{\bbar}^+-\hat{a}_{\bbar}^-+\hat{a}_c^+-\hat{a}_c^- \big)\bigg)\kappa(r ) \cos\theta_b^*\cos\theta_\ell^*\bigg ] \bigg\ } ~ , \label{eq : dsigdcosdcos2}\end{aligned}\ ] ] where @xmath138 the measurement of this angular correlation allows one to extract the sum of np parameters @xmath126 and a different linear combination of the differences @xmath139 ( @xmath50 ) as compared to eq . ( [ eq : dsigdcosdcos ] ) . finally , we take @xmath140 and @xmath141 . in this case , @xmath142 where the @xmath143 are defined in table [ tab : hs ] , and h_^b ( _ b^2 ) = ( 1-_w^2-_b ^2 ) ( _ w^2+_b ^2 ) ( 1-_w^2-_b^2 ) . [ eq : hsmbbarb ] we have dropped a contribution proportional to re@xmath113 in eq . ( [ eq : dsigdm3 ] ) , because its effect is somewhat small in practice . the measurement of the invariant mass - squared distribution permits the extraction of the np parameters @xmath144 and @xmath145 . the angular correlation is given by @xmath146\nonumber\\ & + & \frac{3 g_f m_t^2 } { 4\sqrt{2}\pi^2 \left(1-\zeta_w^2\right)^2 \left(1 + 2\zeta_w^2\right)}\bigg [ \bigg(\sum_{i,\sigma } \hat{a}_i^\sigma\bigg ) \nonumber\\ & & \hskip-2truecm + ~\bigg(\hat{a}_c^+-\hat{a}_c^- -\frac{1}{3}\big ( \hat{a}_{b}^+-\hat{a}_{b}^-+\hat{a}_{\bbar}^+-\hat{a}_{\bbar}^- \big)\bigg)\kappa(r ) \cos\theta_c^*\cos\theta_\ell^*\bigg ] \bigg\ } \ , , \label{eq : dsigdcosdcos3}\end{aligned}\ ] ] where @xmath147 here the measurement of the angular correlation allows one to extract the sum of np parameters @xmath126 and a third distinct linear combination of the differences @xmath127 ( @xmath50 ) . the measurement of any of these observables allows one to detect the presence of np in top decay . if all three angular correlations and invariant mass - squared distributions can be measured , the results can be combined to give measurements of all six np parameters @xmath62 , as well as re@xmath113 . furthermore , there are numerous measurements , providing significant redundancy . this is discussed in detail in the companion paper , ref . @xcite . one can also perform all of the integrations , giving the total cross section @xcite : @xmath148 the measurement of @xmath149 is , in principle , the most straightforward way to detect np . any disagreement between the measured total cross section and its sm value would indicate np . the downside of this approach , however , is that the absolute size of the cross section might be difficult to determine due to qcd corrections , etc . pair production can be found in ref . these include contributions from both @xmath150 and @xmath151 , as well as higher - order corrections . the @xmath1 cross section at the lhc at a centre - of - mass energy of 14 tev is @xmath152 pb . ] for this reason it may be better to use the invariant mass - squared distributions and/or the angular correlations . the measurement of the triple - differential distribution of eq . ( [ eq : dsigdcosdcosdm ] ) would give a great deal of information about the np parameters . however , it is unlikely there will be sufficient statistics to allow this measurement to be carried out . the expressions in the previous section provide a clear picture of the corrections to the various observables introduced by the new - physics contributions . in order to obtain meaningful projections in the context of the lhc , we perform a numerical simulation using @xcite . the new couplings due to the effective lagrangian [ eqs . ( [ eq : eff1])-([eq : eff3 ] ) ] are incorporated into via feynrules @xcite . we compute @xmath153 and obtain the @xmath154 distributions and the angular correlations discussed in the previous section . this naturally involves the convolution of the cross sections and differential cross sections calculated at the parton level with the appropriate parton densities . we use cteq6l1 pdfs @xcite with the factorization and renormalization scales set to @xmath155 = 172 gev . in fig . [ fig : zetasq_bc ] , we compare the normalized @xmath156 distribution obtained using for a @xmath157 collider with a centre - of - mass energy of 14 tev with that obtained for gluons colliding at a fixed centre - of - mass energy of 600 gev using the analytic expression in eq . ( [ eq : dsigdm ] ) . this is done for both the sm and a particular np scenario . note that , although these np parameters are complex , there are no sm - np or np - np interference effects . as such , they do not lead to cp violation . ] . in both cases there is remarkable agreement between the two methods of obtaining @xmath156 . at first glance , this may seem extremely surprising , but a slightly closer look at the issue reveals that it is not really so . the @xmath154 distributions involve only the decay products of the top . any observable that involves only particles coming from a single decay can be computed in the rest frame of the decaying particle and converted to its laboratory - frame equivalent by applying a lorentz boost . at a @xmath157 collider , each event would be associated with a different boost . but since @xmath158 is lorentz - invariant by construction , the distributions can be expected to look identical in both the top rest frame and the laboratory frame , which is what is seen in fig . [ fig : zetasq_bc ] . note , however , that the observed @xmath154 distribution is the result of an ensemble of top decays in which the top quarks are not all identical to begin with . while most of the top quarks are produced on - shell , the ensemble also includes top quarks that are off - shell to varying degrees . moreover , the virtuality of the tops is distributed differently in the fixed - energy and variable - energy cases : in the fixed - energy case one has the additional condition that @xmath159 is fixed . nevertheless , it turns out that this is a small effect . the normalized distributions for the two cases look almost identical , and the inclusion of pdfs does not lead to any significant change in their shape . the slight ( noticeable ) difference in the region @xmath160 0.8 is due to the following . in the analytic expressions , the widths of the @xmath3 and the @xmath66 are dealt with in slightly different ways . for the @xmath3 , the narrow - width approximation is incorporated by making the substitution ( see the appendix ) @xmath161 for the @xmath66 , the result of applying the narrow - width approximation is encapsulated in the factor @xmath162 appearing in the definition of @xmath120 in eq . ( [ eq : hsm ] ) . the finite width of the @xmath66 can be approximated to some extent by making the replacement @xmath163 \ , .\ ] ] this is included in the curves in fig . [ fig : zetasq_bc ] . on the other hand , in both the @xmath3 and @xmath66 widths are dealt with identically with the integral covering an interval of @xmath164 in each case . next we turn to the angular correlation ( fig . [ fig : ctheta_bbarl ] ) . this observable involves decay products coming from two different decays . it is therefore sensitive not only to the physics in the two decays ( whether new or standard ) , but also to the correlations in the production of the two decaying particles ( i.e. , the @xmath1 spin correlations ) . this information is contained in the factor @xmath107 . in a fixed - energy gluon - gluon collison , @xmath107 is fixed . in our expression ( [ eq : dsigdcosdcos ] ) ] , @xmath107 is replaced by its expectation value @xmath110 . when this is calculated over the energy range sampled in 14 tev @xmath157 collisions , we find that , once again , the normalized distributions obtained using this expression agree very well with those obtained from the full numerical simulation using . the fact that the analytical expressions for the observables agree with numerical simulations suggests that it is possible to extract some of the new - physics parameters by fitting the shapes of these distributions . we present the results of these fits in the companion paper @xcite . note that , in comparing the analytical expressions with the simulation , we have taken the @xmath19 quark to be that coming from the @xmath3 decay . however , as noted in the introduction , there is also a @xmath19 coming from the @xmath5 decay , and this background must be taken into account . this issue , along with other complications , is addressed in ref . in this paper we study new - physics ( np ) contributions to top - quark decay . such effects can be significant only for decays that are suppressed in the sm . here we focus on @xmath0 , whose sm amplitude involves the small element @xmath11 ( @xmath12 ) of the ckm matrix . allowing for all lorentz structures , there are ten possible dimension-6 np operators that can contribute to this decay . the goal is to find ways of detecting the presence of such np in @xmath14 . since the lhc produces top quarks copiously , it is an excellent place to search for signals of np in @xmath0 . however , the dominant mode for top - quark production is pair ( @xmath1 ) production via gluon fusion : @xmath165 . this makes it difficult to study @xmath14 on its own . in order to search for np in top decay , the full process @xmath2 must be analyzed . we consider only cp - conserving np , and find that there are two types of observables that can be used to reveal the presence of np in top decay . the first is an invariant mass - squared distribution involving two of the final - state particles in @xmath14 . there are three such distributions . the second is an angular correlation between the decay products of the @xmath3 and @xmath5 . this is related to the @xmath1 spin correlation . we consider the angular correlation between one of the final - state quarks in @xmath14 and the @xmath4 coming from the @xmath5 decay . there are three such correlations . the six observables depend on different combinations of the coefficients of the ten np operators . we compare the analytical expressions for the observables with the results of a numerical simulation of the lhc using . we find that the agreement between the two is excellent . this suggests that the measurement of these observables can indeed be used to extract some of the new - physics parameters . in the companion paper , @xcite , we demonstrate this explicitly by performing fits of such measurements . we also show how to deal with complications such as the background due to the @xmath19 coming from the @xmath5 decay . * acknowledgments * : the authors wish to thank the madgraph and feynrules teams for extensive discussions about madgraph and feynrules , respectively . the authors are also indebted to german valencia and howard baer for helpful discussions and to zach bethel and carl daudt for technical support . this work was financially supported by nserc of canada ( dl , ps ) . in addition , this work has been partially supported by anpcyt under grant no . pict - prh 2009 - 0054 and by conicet ( as ) . the work of sj and jm was supported by the u.s . national science foundation under grant phy1215785 . the work of kk was supported by the u.s . national science foundation under grants phy0900914 and phy1215785 . kk also acknowledges sabbatical support from taylor university . in this appendix we work out an expression for the differential cross section for @xmath2 . our main result may be found below in eqs . ( [ eq : dsigma2])-([eq : dlambda ] ) . as an intermediate step , we write the differential cross section for @xmath2 in a quasi - factorized form that makes use of expressions for @xmath166 , @xmath167 and @xmath168 ( see eq . ( [ eq : diffcrosssectiondoublespin2 ] ) , below ) . throughout , we assume that np is present only in @xmath14 ; @xmath169 and @xmath168 are purely sm in nature . furthermore , we always employ the narrow - width approximation for the @xmath3 and @xmath5 , which is equivalent to assuming that they are on - shell . we begin with @xmath165 ( see fig . [ fig : gluonfusion ] ) . the amplitude squared for @xmath169 , including the @xmath3 and @xmath5 polarizations , was computed in ref . it is useful to define the following quantities : @xmath170 where @xmath171 and @xmath172 are the @xmath3 and @xmath5 momenta , and @xmath173 and @xmath174 are the momenta of the initial gluons . the matrix element squared is then given by ( see also fig . [ fig : gg->4and6fermions ] and eq . ( [ eq : mabij ] ) below ) , @xmath175 \big\ } ~ , \label{eq : ggttbar}\end{aligned}\ ] ] in which @xmath176 integrating the amplitude squared over phase space and summing over the @xmath3 and @xmath5 spins yields the following expression for the parton - level scattering cross - section : @xmath177 ~. \label{eq : sigggttbar}\end{aligned}\ ] ] we now derive expressions that can be used to translate @xmath3-spin - dependent observables into a form that may be more useful to experimentalists . our starting point is the observation that the spins of the @xmath3 and the @xmath5 are correlated in @xmath178 [ see eq . ( [ eq : ggttbar ] ) ] . thus , @xmath3-spin observables can in principle be translated into observables that employ the spin of the @xmath5 . this is shown schematically in fig . [ fig : gg->4and6fermions ] ( a ) . of course , the spin of the @xmath5 is itself not directly measurable . fortunately , however , the momentum of the charged lepton in @xmath179 is correlated with the spin of the @xmath5 . thus , in order to consider @xmath3-spin - dependent observables in @xmath0 , we can study the full process @xmath2 , as is indicated in fig . [ fig : gg->4and6fermions ] ( b ) . consider the two diagrams shown in fig . [ fig : gg->4and6fermions ] . the matrices @xmath180 , @xmath181 and @xmath182 indicated there are defined via the production and decay amplitudes as follows : @xmath183 in which @xmath184 are colour indices . we assume that colour indices contract as in the sm , so that @xmath185 note that @xmath186 is assumed to contain all of the np effects . explicit calculation , starting from the effective lagrangian given in eqs ( [ eq : eff1])-([eq : eff3 ] ) , yields @xmath187\end{aligned}\ ] ] furthermore , we define @xmath188 , @xmath189 and @xmath190 via the following relations @xmath191 let us begin by considering the diagram in fig . [ fig : gg->4and6fermions ] ( a ) . the amplitude for this process may be written as follows @xmath192 multiplying the above expression by its complex conjugate and making the substitution @xmath193^{-1 } \simeq \frac{\pi}{\gamma_t m_t}\delta\!\!\left(p_t^2-m_t^2\right ) ~,\end{aligned}\ ] ] we have @xmath194\ , , \label{eq : largetracesinglespin}\end{aligned}\ ] ] where the sum over spins includes the gluon spins as well as those of the @xmath41 , @xmath19 and @xmath88 . note that the @xmath3 quark only shows up via its propagator in this expression . thus , the spin of the @xmath3 is summed over , as it should be . the spin of the @xmath5 , however , appears explicitly . this can be simplified by using the following identity , which is similar to an expression in ref . @xcite ( see also ref . @xcite ) : @xmath195 & = & \frac{1}{2}\big\{\mbox{tr}\left[\mathbb{x}\left(\pslash \pm m\right)\right ] \mbox{tr}\left[\mathbb{y}\left(\pslash \pm m\right)\right ] \nonumber \\ & & \hskip-1truecm -~\eta_{\mu\nu}\mbox{tr}\left[\mathbb{x}\left(\pslash \pm m\right)\gamma^5\gamma^\mu\right ] \mbox{tr}\left[\mathbb{y}\left(\pslash \pm m\right)\gamma^5\gamma^\nu\right ] \big\ } ~ , \label{eq : xytrick}\end{aligned}\ ] ] where @xmath196 and where it is assumed that @xmath197 . setting @xmath198 we can split the trace in eq . ( [ eq : largetracesinglespin ] ) into two pieces , one corresponding to the @xmath34 production ( @xmath199 ) and one to the @xmath3 decay ( @xmath200 ) . finally , defining @xmath201/ \mbox{tr}\!\left[b{\overline{b}}\left(\pslash_t + m_t\right)\right ] , \label{eq : nt}\end{aligned}\ ] ] we find that we can write the differential cross section corresponding to fig . [ fig : gg->4and6fermions ] ( a ) in the following suggestive form @xcite : @xmath202 1 . the @xmath3 polarization , @xmath203 , is a very particular four - vector , defined in eq . ( [ eq : nt ] ) . 2 . while @xmath204 is calculated for a particular spin four - vector for the @xmath3 , the @xmath3 spin is averaged in @xmath205 . although eq . ( [ eq : diffcrosssectionsinglespin ] ) has the appearance of being factorized cleanly into two pieces , the @xmath3-polarization four - vector contained in @xmath206 depends on the phase - space variables contained in @xmath207 . similarly , the spin four - vectors for the @xmath41 , @xmath19 and @xmath88 appear both in @xmath203 and in @xmath207 . 4 . given the preceding comment , one must exercise some caution when integrating over phase space and summing over the @xmath41 , @xmath19 and @xmath88 spins . in particular , one must do so for the product of @xmath208 and @xmath209 , and not for the two quantities separately . for the spin sum , the @xmath3-polarization - dependent quantity that appears in calculations is always @xmath210 n_{t\mu}$ ] . it is safe to sum this quantity over spins . the above approach can be generalized to the scenario indicated in fig . [ fig : gg->4and6fermions ] ( b ) by applying the trick in eq . ( [ eq : xytrick ] ) twice in succession , once for the @xmath3 and once for the @xmath5 . one new subtlety in this case is that the final state contains two identical @xmath19 antiquarks . one should therefore antisymmetrize the total amplitude under the exchange of the two @xmath19 s . in practice , we implement cuts in such a way that the two @xmath19 s can effectively be distinguished . in particular , in @xmath14 , we have @xmath211 . but this relation will not , in general , be satisfied if the @xmath19 comes from the decay of the @xmath5 . thus , the two @xmath19 s can be distinguished using experimental cuts , and we therefore treat them as non - identical . further discussion on this point is included in the companion paper @xcite . defining @xmath212/ \mbox{tr}\!\left[c\overline{c}\left(\pslash_{\tbar } - m_t\right)\right ] , \label{eq : ntbar}\end{aligned}\ ] ] where @xmath213 and proceeding as above , we find @xcite @xmath214 use of the above expression requires some care , since the same subtle issues are present as were noted above for the analogous expression in eq . ( [ eq : diffcrosssectionsinglespin ] ) . at this stage , let us work out expressions for the `` special '' @xmath3 and @xmath5 polarization four - vectors , @xmath215 and @xmath216 , respectively . the quantity that is of interest in the calculation is @xmath217 n_t^\alpha & = & -\eta^{\alpha\beta}\sum_{b , \bbar , c \mbox{\scriptsize ~spins } } \mbox{tr}\!\left[b{\overline{b } } \left(\pslash_t + m_t\right)\gamma^5\gamma_\beta\right ] \nonumber\\ & = & -\left(4\sqrt{2}g_f v_{tb}v_{cb}\right)^2 \bigg[\sum_{i,\sigma } \frac{2\,p_i\cdot p_t}{m_t}\left(p_t^\alpha - \frac{m_t^2 p_i^\alpha}{p_i\cdot p_t}\right ) \!\xi^\sigma \!a_i^\sigma \nonumber \\ & & ~~~~ + 32\,m_t \,\mbox{im}\left(x^t_{ll}x^{s*}_{ll}+x^t_{rr}x^{s*}_{rr}\right ) \epsilon^{\alpha\beta\gamma\delta}p_{b\beta } p_{\bbar\gamma}p_{c\delta}\bigg ] .~~ \label{eq : dgam_ntalpha}\end{aligned}\ ] ] note that this `` special '' polarization four - vector for the @xmath3 quark , which will eventually be incorporated into the expression for @xmath2 , contains all of the relevant information and correlations related to the decay of the @xmath3 . since the semileptonic decay of the @xmath5 is assumed to be sm - like , the expression for @xmath216 is much simpler . defining @xmath218 [ in analogy with the sm part of eq . ( [ eq : abbar ] ) ] , we find @xmath219 \tilde{n}_\tbar^\alpha & = & -\overline{\eta}^{\alpha\beta}\sum_{\bbar , \ell , \nubar \mbox{\scriptsize ~spins } } \mbox{tr}\!\left[c{\overline{c } } \left(\pslash_\tbar - m_t\right)\gamma^5\gamma_\beta\right ] \nonumber\\ & = & \left(4\sqrt{2}g_f v_{tb}\right)^2 \frac{2\ , p_{\ell}\cdot p_\tbar}{m_t}\left(p_{\tbar}^\alpha - \frac{m_t^2 p_\ell^\alpha}{p_{\ell}\cdot p_\tbar}\right)\!a_\ell \,.~~~ \label{eq : dgam_ntbaralpha}\end{aligned}\ ] ] equations ( [ eq : dgam_ntalpha ] ) and ( [ eq : dgam_ntbaralpha ] ) may be compared to related expressions in ref . the following expressions are also useful : @xmath217 & = & \left(4\sqrt{2}g_f v_{tb}v_{cb}\right)^2 \sum_{i,\sigma } 2\ , p_i\cdot p_t \,a_i^\sigma \ , , \\ \sum_{\bbar , \ell , \nubar \mbox{\scriptsize ~spins } } \!\!\!\mbox{tr}\!\left[c \overline{c}\left(\pslash_{\tbar}-m_t\right)\right ] & = & \left(4\sqrt{2}g_f v_{tb}\right)^2 2\ , p_{\ell}\cdot p_\tbar \,a_\ell \,.~~~\end{aligned}\ ] ] with these expressions in hand , we can now work out the final expression for the differential cross - section . using eqs . ( [ eq : dgam_ntalpha ] ) and ( [ eq : dgam_ntbaralpha ] ) in eq . ( [ eq : diffcrosssectiondoublespin2 ] ) , we have @xmath220 where @xmath221 \!-\ ! \xi^\sigma p_i\!\cdot\ ! p_\ell \ , g(r , z ) \nonumber\\ & & ~~~~~~~~~~~~~~-\frac{\left(r^2 - 1\right ) \left[r^2\left(z^2 - 2\right)+1\right ] \xi^\sigma}{2 m_t^2 } \left(p_i\!\cdot\!q \,q\!\cdot\!p_\ell + p_i\!\cdot\!p_t \,p_t\!\cdot\!p_\ell\right)\nonumber\\ & & ~~~~~~~~~~~~~~- \frac{r^2(r^2 - 1)(z^2 - 1)\xi^\sigma}{2 m_t^2}\big [ p_i\!\cdot\ ! p_g \left ( p_g\!\cdot\ ! p_\ell - q\!\cdot\ ! p_\ell\,rz \right ) \nonumber\\ & & ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~+ p_i\!\cdot\ ! q \left ( p_g\!\cdot\ ! p_\ell\,rz - q\!\cdot\ ! p_\ell \right ) \big ] \big\ } ~ , \\ { \cal b}_{\mbox{\scriptsize tp } } & = & 16 a_\ell \,\mbox{im}\left(x^t_{ll}x^{s*}_{ll}+x^t_{rr}x^{s*}_{rr}\right ) \big\ { -g(r , z)\epsilon\left(p_b , p_{\overline{b}},p_c , p_\ell\right ) \nonumber\\ & & - \frac{\left(r^2 - 1\right)p_\ell\!\cdot\!p_{\overline{t } } } { m_t^2}\left[r^2\left(z^2 - 2\right)+1\right ] \epsilon\left(p_b , p_{\overline{b}},p_c , q\right)\nonumber\\ & & -\frac{r^2(r^2 - 1)(z^2 - 1)}{2 m_t^2}\big [ \left ( p_g\!\cdot\ ! p_\ell - q\!\cdot\ ! p_\ell\,rz \right)\epsilon\left(p_b , p_{\overline{b}},p_c , p_g\right ) \nonumber\\ & & ~~~~~~~~~~~~~~~~~~~~~~~~+ \left ( p_g\!\cdot\ ! p_\ell\,rz - q\!\cdot\ ! p_\ell \right)\epsilon\left(p_b , p_{\overline{b}},p_c , q\right ) \big ] \big\ } ~ , \label{eq : btp2}\end{aligned}\ ] ] and @xmath222 in the above , the @xmath78 are the momenta of the final - state quarks coming from the top decay ( i.e. , @xmath41 , @xmath19 and @xmath88 ) ; also , @xmath223 , @xmath224 , @xmath225 , @xmath109 , @xmath226 , @xmath227 and @xmath228 were defined in eqs . ( [ eq : pt ] ) , ( [ eq : f ] ) and ( [ eq : g ] ) . in arriving at the above expression for @xmath229 , we have decomposed the six - body phase space into five solid angles and four invariant masses ( see fig . [ fig : kinematics ] ) , and then have used the narrow - width approximation for the @xmath3 and @xmath5 quarks to eliminate two of the invariant - mass degrees of freedom . the solid angles @xmath67-@xmath71 and the invariant masses @xmath74 and @xmath75 are discussed in sec . [ sec : ggprocess ] . inspection of eqs . ( [ eq : bnon - tp2 ] ) and ( [ eq : btp2 ] ) reveals elements that are a combination of expressions coming from the production and decay of the @xmath3 and @xmath5 quarks . the @xmath55 are related to the decay @xmath14 [ see eqs . ( [ eq : abbar])-([ahatdefs ] ) ] . @xmath230 is similarly related to the semileptonic decay of the @xmath5 . in the sm , @xmath231 \quad & & = \quad \sigma \!\left(gg\to t\tbar\right ) \mbox{br}\!\left.\left(t\to b\bbar c \right)\right|_{\mbox{\scriptsize sm } } \mbox{br}\!\left(\tbar\to \bbar\ell\nubar\right ) , \label{eq : sigmasm}\end{aligned}\ ] ] in which @xmath232 is defined in eq . ( [ eq : sigggttbar ] ) , br@xmath233 and br@xmath234 . after the inclusion of new physics , @xmath235 & & = \sigma_{\mbox{\scriptsize sm } } \bigg\ { 1 + \frac{4\gamma_w}{m_w}\mbox{im}\left(x^{v*}_{ll}\right ) + \frac{3 g_f m_t^2 } { 4\sqrt{2}\pi^2 \left(1-\zeta_w^2\right)^2 \left(1 + 2\zeta_w^2\right ) } \sum_{i,\sigma } \hat{a}_i^\sigma\bigg\}.\end{aligned}\ ] ] v. m. abazov _ et al . _ [ d0 collaboration ] , phys . rev . d * 85 * , 091104 ( 2012 ) [ arxiv:1201.4156 [ hep - ex ] ] ; t. a. aaltonen _ et al . _ [ cdf collaboration ] , phys . lett . * 111 * , 202001 ( 2013 ) [ arxiv:1308.4050 [ hep - ex ] ] . g. aad _ et al . _ [ atlas collaboration ] , phys . b * 717 * , 330 ( 2012 ) [ arxiv:1205.3130 [ hep - ex ] ] . k. kiers , t. knighton , d. london , m. russell , a. szynkman and k. webster , phys . d * 84 * , 074018 ( 2011 ) [ arxiv:1107.0754 [ hep - ph ] ] . k. kiers , d. london , p. saha and a. szynkman , in preparation . _ experimental measurements of the @xmath34 spin correlation involving four jets can be found in _ + t. aaltonen _ et al . _ [ cdf collaboration ] , phys . rev . d * 83 * , 031104 ( 2011 ) [ arxiv:1012.3093 [ hep - ex ] ] ; v. m. abazov _ et al . _ [ d0 collaboration ] , phys . lett . * 108 * , 032004 ( 2012 ) [ arxiv:1110.4194 [ hep - ex ] ] ; g. aad _ et al . _ [ atlas collaboration ] , arxiv:1407.4314 [ hep - ex ] . _ the cms collaboration has made a measurement using two jets : _ + s. chatrchyan _ et al . _ [ cms collaboration ] , phys . lett . * 112 * , 182001 ( 2014 ) [ arxiv:1311.3924 [ hep - ex ] ] . g. mahlon and s. j. parke , phys . d * 81 * , 074024 ( 2010 ) [ arxiv:1001.3422 [ hep - ph ] ] . m. arai , n. okada , k. smolek and v. simak , phys . d * 70 * , 115015 ( 2004 ) [ hep - ph/0409273 ] ; m. arai , n. okada , k. smolek and v. simak , phys . d * 75 * , 095008 ( 2007 ) [ hep - ph/0701155 ] ; r. frederix and f. maltoni , jhep * 0901 * , 047 ( 2009 ) [ arxiv:0712.2355 [ hep - ph ] ] ; m. arai , n. okada and k. smolek , phys . rev . d * 79 * , 074019 ( 2009 ) [ arxiv:0902.0418 [ hep - ph ] ] ; c. -x . yue , t. -t . zhang and j. -y . liu , j. phys . g * 37 * , 075016 ( 2010 ) [ arxiv:1003.2770 [ hep - ph ] ] ; c. degrande , j. -m . gerard , c. grojean , f. maltoni and g. servant , jhep * 1103 * , 125 ( 2011 ) [ arxiv:1010.6304 [ hep - ph ] ] ; j. cao , l. wu and j. m. yang , phys . d * 83 * , 034024 ( 2011 ) [ arxiv:1011.5564 [ hep - ph ] ] ; m. baumgart and b. tweedie , jhep * 1109 * , 049 ( 2011 ) [ arxiv:1104.2043 [ hep - ph ] ] ; d. krohn , t. liu , j. shelton and l. -t . wang , phys . d * 84 * , 074034 ( 2011 ) [ arxiv:1105.3743 [ hep - ph ] ] ; y. bai and z. han , jhep * 1202 * , 135 ( 2012 ) [ arxiv:1106.5071 [ hep - ph ] ] ; v. barger , w. -y . keung and b. yencho , phys . d * 85 * , 034016 ( 2012 ) [ arxiv:1112.5173 [ hep - ph ] ] ; s. fajfer , j. f. kamenik and b. melic , jhep * 1208 * , 114 ( 2012 ) [ arxiv:1205.0264 [ hep - ph ] ] ; z. han , a. katz , d. krohn and m. reece , jhep * 1208 * , 083 ( 2012 ) [ arxiv:1205.5808 [ hep - ph ] ] ; m. baumgart and b. tweedie , jhep * 1303 * , 117 ( 2013 ) [ arxiv:1212.4888 [ hep - ph ] ] . d. atwood , a. aeppli and a. soni , phys . * 69 * , 2754 ( 1992 ) ; w. bernreuther and a. brandenburg , phys . rev . d * 49 * , 4481 ( 1994 ) [ hep - ph/9312210 ] ; k. -m . cheung , phys . d * 55 * ( 1997 ) 4430 [ hep - ph/9610368 ] ; h. -y . zhou , phys . d * 58 * , 114002 ( 1998 ) [ hep - ph/9805358 ] ; m. beneke , i. efthymiopoulos , m. l. mangano , j. womersley , a. ahmadov , g. azuelos , u. baur and a. belyaev _ et al . _ , in * geneva 1999 , standard model physics ( and more ) at the lhc * 419 - 529 [ hep - ph/0003033 ] ; o. antipin and g. valencia , phys . rev . d * 79 * , 013013 ( 2009 ) [ arxiv:0807.1295 [ hep - ph ] ] ; s. k. gupta , a. s. mete and g. valencia , phys . d * 80 * , 034013 ( 2009 ) [ arxiv:0905.1074 [ hep - ph ] ] ; s. k. gupta and g. valencia , phys . d * 81 * , 034013 ( 2010 ) [ arxiv:0912.0707 [ hep - ph ] ] . x. -g . he , g. valencia and h. yokoya , jhep * 1112 * , 030 ( 2011 ) [ arxiv:1110.2588 [ hep - ph ] ] . t. aaltonen _ et al . _ [ cdf collaboration ] , phys . d * 83 * , 112003 ( 2011 ) [ arxiv:1101.0034 [ hep - ex ] ] ; v. m. abazov _ et al . _ [ d0 collaboration ] , phys . rev . d * 84 * , 112005 ( 2011 ) [ arxiv:1107.4995 [ hep - ex ] ] . p. saha , k. kiers , d. london and a. szynkman , arxiv:1407.1725 [ hep - ph ] . a. datta and d. london , phys . b * 595 * , 453 ( 2004 ) [ hep - ph/0404130 ] . e. byckling and k. kajantie , _ particle kinematics _ ( wiley , new york , 1973 ) . n. kidonakis , arxiv:1304.7775 [ hep - ph ] ; s. moch and p. uwer , phys . rev . d * 78 * , 034003 ( 2008 ) [ arxiv:0804.1476 [ hep - ph ] ] . j. alwall , m. herquet , f. maltoni , o. mattelaer and t. stelzer , jhep * 1106 * , 128 ( 2011 ) [ arxiv:1106.0522 [ hep - ph ] ] ; http://madgraph.hep.uiuc.edu/.

it is shown by the author that if gravitons are super - strong interacting particles and the low - temperature graviton background exists , the basic cosmological conjecture about the dopplerian nature of redshifts may be false . in this case , a full magnitude of cosmological redshift would be caused by interactions of photons with gravitons . a new dimensional constant which characterizes one act of interaction is introduced and estimated . non - forehead collisions with gravitons will lead to a very specific additional relaxation of any photonic flux . it gives a possibility of another interpretation of supernovae 1a data - without any kinematics . of course , all of these facts may implicate a necessity to change the standard cosmological paradigm . some features of a new paradigm are discussed here , too . a quantum mechanism of classical gravity based on an existence of this sea of gravitons is described for the newtonian limit . this mechanism needs graviton pairing and `` an atomic structure '' of matter for working it , and leads to the time asymmetry . if the considered quantum mechanism of classical gravity is realized in the nature , then an existence of black holes contradicts to einstein s equivalence principle . it is shown that in this approach the two fundamental constants - hubble s and newton s ones - should be connected between themselves . the theoretical value of the hubble constant is computed . in this approach , every massive body would be decelerated due to collisions with gravitons that may be connected with the pioneer 10 anomaly . it is shown that the predicted and observed values of deceleration are in good agreement . some unsolved problems are discussed , so as possibilities to verify some conjectures in laser - based experiments .

hep-th0506189

by a full coincidence of the forms of coulomb s and newton s laws , which describe an interaction of electric charges and a gravitational interaction of bodies , we see a dramatic difference in developing the pictures of these interactions on a quantum level . constructed with pillars on multiple experiments qed is one of the most exact physical theories and an archetype for imitation by creation of new models . while the attempts to quantize the remarkable in its logical beauty theory of general relativity , which describes gravitation on a classical level so fully and delicately , ( see review @xcite ) not only have not a hit until today but gave a specific side psychological effect - there exists conceptualization that quantum gravity may be described only by some sophisticated theory . an opinion is commonly accepted , too , that quantum gravity should manifest itself only on the planck scale of energies , i.e. it is a high - energy phenomenon . the value of the planck energy @xmath0 gev has been got from dimensional reasonings . still one wide - spread opinion is that we know a mechanism of gravity ( bodies are exchanging with gravitons of spin 2 ) but can not correctly describe it . if an apple from the legend about newton s afflatus can imagine all these complications , it would hesitate to fall to the ground so artlessly as it is accepted among the apples . perhaps , physicists would be able to refuse easier the preconceived stereotypes which balk - as it seems to me - to go ahead in understanding quantum gravity if experiments or observations would give more essential meat for reasonings . but in this area , at least up to recent years , there was observed nothing that may serve if not ariadne s clew but such a simple physical contradiction that an aspiration to overcome it would advantage introduction of new ideas and revision of the `` inviolable '' . in a few last years , the situation has been abruptly changed . i enumerate those discoveries and observations which may force , in my opinion , the ice to break up . \1 . in 1998 , anderson s team reported about the discovery of anomalous acceleration of nasa s probes pioneer 10/11 @xcite ; this effect is not embedded in a frame of the general relativity , and its magnitude is somehow equal to @xmath1 , where @xmath2 is the hubble constant , @xmath3 is the light velocity . \2 . in the same 1998 , two teams of astrophysicists , which were collecting supernovae 1a data with the aim to specificate parameters of cosmological expansion , reported about dimming remote supernovae @xcite ; the one would be explained on a basis of the doppler effect if at present epoch the universe expands with acceleration . this explanation needs an introduction of some `` dark energy '' which is unknown from any laboratory experiments . \3 . in january 2002 , nesvizhevsky s team reported about discovery of quantum states of ultra - cold neutrons in the earth s gravitational field @xcite . observed energies of levels ( it means that and their differences too ) in full agreement with quantum - mechanical calculations turned out to be equal to @xmath4 ev . the formula for energy levels had been found still by bohr and sommerfeld . if transitions between these levels are accompanied with irradiation of gravitons then energies of irradiated gravitons should have the same order - but it is of 40 orders lesser than the planck energy by which one waits quantum manifestations of gravity . the first of these discoveries obliges to muse about the borders of applicability of the general relativity , the third - about that quantum gravity would be a high - energy phenomenon . it seems that the second discovery is far from quantum gravity but it obliges us to look at the traditional interpretation of the nature of cosmological redshift critically . an introduction into consideration of an alternative model of redshifts @xcite which is based on a conjecture about an existence of the graviton background gives us odds to see in the effect of dimming supernovae an additional manifestation of low - energy quantum gravity . under the definite conditions , an effective temperature of the background may be the same one as a temperature of the cosmic microwave background , with an average graviton energy of the order @xmath5 ev . in this chapter , the main results of author s research in this approach are described . starting from a statistical model of the graviton background with low temperature , it is shown - under the very important condition that gravitons are super - strong interacting particles - that if a redshift would be a quantum gravitational effect then one can get from its magnitude an estimate of a new dimensional constant characterizing a single act of interaction in this model . it is possible to calculate theoretically a dependence of a light flux relaxation on a redshift value , and this dependence fits supernova observational data very well at least for @xmath6 . further it is possible to find a pressure of single gravitons of the background which acts on any pair of bodies due to screening the graviton background with the bodies @xcite . it turns out that the pressure is huge ( a corresponding force is @xmath7 times stronger than the newtonian attraction ) but it is compensated with a pressure of gravitons which are re - scattered by the bodies . the newtonian attraction arises if a part of gravitons of the background forms pairs which are destructed by interaction with bodies . it is interesting that both hubble s and newton s constants may be computed in this approach with the ones being connected between themselves . it allows us to get a theoretical estimate of the hubble constant . an unexpected feature of this mechanism of gravity is a necessity of `` an atomic structure '' of matter - the mechanism does nt work without the one . collisions with gravitons should also call forth a deceleration of massive bodies of order @xmath1 - namely the same as of nasa s probes . but at present stage it turns out unclear why such the deceleration has _ not _ been observed for planets . the situation reminds by something of the one that took place in physics before the creation of quantum mechanics when a motion of electrons should , as it seemed by canons of classical physics , lead to their fall to a nucleus . because the very unexpected hypothesis founded into the basis of this approach is a super - strong character of gravitational interaction on a quantum level , i would like to explain my motivation which conduced namely to such the choice . learning symmetries of the quantum two - component composite system @xcite , i have found that its discrete symmetries in an 8-space may be interpreted by an observer from a 4-dimensional world as the exact global symmetries of the standard model of particle physics if internal coordinates of the system ( the composite fermion ) are rigidly fixed . this conclusion was hard for me and took much enough time . but to ensure almost full fixation of components of the system , an interaction connecting them should be very strong . because of it , when a choice arise - an amount of gravitons or an intensity of the interaction , - i have remembered this overpassed earlier barrier and chose namely the super - strong interaction . without this property , the graviton background would not be in the thermodynamical equilibrium with the cosmic microwave background that could entail big difficulties in the model . so , in this approach we deal with the following small quantum effects of low - energy gravity : redshifts , its analog - a deceleration of massive bodies , and an additional relaxation of any light flux . the newtonian attraction turns out to be the main statistical effect , with bodies themselves being not sources of gravitons - only correlational properties of _ in _ and _ out _ fluxes of gravitons in their neighbourhood are changed due to an interaction with bodies . there does still not exist a full and closed theory in this approach , but even the initial researches in this direction show that in this case quantum gravity can not be described separately of other interactions , and also manifest the boundaries of applicability of a geometrical language in gravity . let us introduce the hypothesis , which is considered in this approach as independent from the standard cosmological model : there exists the isotropic graviton background . photon scattering is possible on gravitons @xmath8 where @xmath9 is a photon and @xmath10 is a graviton , if one of the gravitons is virtual . the energy - momentum conservation law prohibits energy transfer to free gravitons . due to forehead collisions with gravitons , an energy of any photon should decrease when it passes through the sea of gravitons . from another side , none - forehead collisions of photons with gravitons of the background will lead to an additional relaxation of a photon flux , caused by transmission of a momentum transversal component to some photons . it will lead to an additional dimming of any remote objects , and may be connected with supernova dimming . we deal here with the uniform non - expanding universe with the euclidean space , and there are not any cosmological kinematic effects in this model . we shall take into account that a gravitational `` charge '' of a photon must be proportional to @xmath11 ( it gives the factor @xmath12 in a cross - section ) and a normalization of a photon wave function gives the factor @xmath13 in the cross - section . also we assume here that a photon average energy loss @xmath14 in one act of interaction is relatively small to a photon energy @xmath15 then average energy losses of a photon with an energy @xmath16 on a way @xmath17 will be equal to @xcite : @xmath18 where @xmath19 is a constant . if a _ whole _ redshift magnitude is caused by this effect , we must identify @xmath20 where @xmath3 is the light velocity , to have the hubble law for small distances @xcite . a photon energy @xmath11 should depend on a distance from a source @xmath21 as @xmath22 where @xmath23 is an initial value of energy . the expression ( 2 ) is just only so far as the condition @xmath24 takes place . photons with a very small energy may lose or acquire an energy changing their direction of propagation after scattering . early or late such photons should turn out in the thermodynamic equilibrium with the graviton background , flowing into their own background . decay of virtual gravitons should give photon pairs for this background , too . perhaps , the last one is the cosmic microwave background @xcite . it follows from the expression ( 2 ) that an exact dependence @xmath25 is the following one : @xmath26 if an interaction with the graviton background is the only cause of redshifts . it is very important , that this redshift does not depend on a light frequency . for small @xmath27 the dependence @xmath25 will be linear . the expressions ( 1 ) - ( 3 ) are the same that appear in other tired - light models ( compare with @xcite ) . in this approach , the ones follow from a possible existence of the isotropic graviton background , from quantum electrodynamics , and from the fact that a gravitational `` charge '' of a photon must be proportional to @xmath15 photon flux s average energy losses on a way @xmath28 due to non - forehead collisions with gravitons should be proportional to @xmath29 where @xmath30 is a new constant of the order @xmath31 these losses are connected with a rejection of a part of photons from a source - observer direction . such the relaxation together with the redshift will give a connection between visible object s diameter and its luminosity ( i.e. the ratio of an object visible angular diameter to a square root of visible luminosity ) , distinguishing from the one of the standard cosmological model . let us consider that in a case of a non - forehead collision of a graviton with a photon , the latter leaves a photon flux detected by a remote observer ( an assumption of a narrow beam of rays ) . the details of calculation of the theoretical value of relaxation factor @xmath30 which was used in author s paper @xcite were given later in the preprint @xcite . so as both particles have velocities @xmath32 a cross - section of interaction , which is `` visible '' under an angle @xmath33 ( see fig . 1 ) , will be equal to @xmath34 if @xmath35 is a cross - section by forehead collisions . the function @xmath36 allows to take into account both front and back hemispheres for riding gravitons . additionally , a graviton flux , which falls on a picked out area ( cross - section ) , depends on the angle @xmath37 we have for the ratio of fluxes : @xmath38 where @xmath39 and @xmath40 are the fluxes which fall on @xmath35 under the angle @xmath33 and normally , @xmath41 is a square of side surface of a truncated cone with a base @xmath35 ( see fig . = 12.98 cm finally , we get for the factor @xmath42 @xmath43 by @xmath44 a formed cone contains self - intersections , and it is @xmath45 . by @xmath46 we have @xmath47 . after computation of simple integrals , we get : @xmath48 in the considered simplest case of the uniform non - expanding universe with the euclidean space , we shall have the quantity @xmath49 in a visible object diameter - luminosity connection if a whole redshift magnitude would caused by such an interaction with the background ( instead of @xmath50 for the expanding uniform universe ) . for near sources , the estimate of the factor @xmath30 will be some increased one . the luminosity distance ( see @xcite ) is a convenient quantity for astrophysical observations . both redshifts and the additional relaxation of any photonic flux due to non - forehead collisions of gravitons with photons lead in our model to the following luminosity distance @xmath51 @xmath52 where @xmath53 . to compare a form of this predicted dependence @xmath54 by unknown , but constant @xmath2 , with the latest observational supernova data by riess et al . @xcite , one can introduce distance moduli @xmath55 , where @xmath56 is an unknown constant ( it is a single free parameter to fit the data ) ; @xmath57 is the luminosity distance in units of @xmath58 . = 12.98 cm in figure 2 , the hubble diagram @xmath59 is shown with @xmath60 to fit observations for low redshifts ; observational data ( 82 points ) are taken from table 5 of @xcite . the predictions fit observations very well for roughly @xmath6 . it excludes a need of any dark energy to explain supernovae dimming . discrepancies between predicted and observed values of @xmath59 are obvious for higher @xmath61 : we see that observations show brighter sne that the theory allows , and a difference increases with @xmath61 . it is better seen on figure 3 with a linear scale for @xmath57 ; observations are transformed as @xmath62 with the same @xmath60 . ; it might be partially caused by quickly raising contribution of a dispersion of measured flux : it should be proportional to @xmath63 . ] = 12.98 cm it would be explained in the model as a result of specific deformation of sn spectra due to a discrete character of photon energy losses . today , a theory of this effect does not exist , and i explain its origin only qualitatively @xcite . for very small redshifts @xmath27 only a small part of photons transmits its energy to the background ( see below fig . 8 in section 6 ) . therefore any red - shifted narrow spectral strip will be a superposition of two strips . one of them has a form which is identical with an initial one , its space is proportional to @xmath64 where @xmath65 is an average number of interactions of a single photon with the background , and its center s shift is negligible ( for a narrow strip ) . another part is expand , its space is proportional to @xmath66 and its center s shift is equal to @xmath67 where @xmath68 is an average energy loss in one act of interaction . an amplitude of the red - shifted step should linear raise with a redshift . for big @xmath27 spectra of remote objects of the universe would be deformed . a deformation would appear because of multifold interactions of a initially - red - shifted part of photons with the graviton background . it means that the observed flux within a given passband would depend on a form of spectrum : the flux may be larger than an expected one without this effect if an initial flux within a next - blue neighbour band is big enough - due to a superposition of red - shifted parts of spectrum . some other evidences of this effect would be an apparent variance of the fine structure constant @xcite or of the cmb temperature @xcite with epochs . in both cases , a ratio of red - shifted spectral line s intensities may be sensitive to the effect . this comparison with observations is very important ; to see some additional details , we can compute and graph the ratio @xmath69 , where @xmath70 is an observed analog of @xmath71 ( see fig . 4 ) @xcite . an expected value of the ratio should be equal to 1 for any @xmath61 . ( dots ) ; observational data are taken from table 5 of @xcite . if this model is true , the ratio should be equal to 1 for any @xmath61 ( solid line).,width=490 ] let us consider that a full redshift magnitude is caused by an interaction with single gravitons . if @xmath72 is a cross - section of interaction by forehead collisions of a photon with an energy @xmath11 with a graviton , having an energy @xmath73 we consider really ( see ( 1 ) ) , that @xmath74 where @xmath75 is a space angle element , and the function @xmath76 has a constant value for any @xmath77 . if @xmath78 is a spectral density of graviton flux in the limits of @xmath75 in some direction ( @xmath79 is a graviton frequency , @xmath80 ) , i.e. an intensity of a graviton flux is equal to the integral @xmath81 @xmath82 is an equivalent temperature of the graviton background , we can write for the hubble constant @xmath83 introduced in the expression ( 1 ) : @xmath84 if @xmath85 can be described by the planck formula for equilibrium radiation , then @xmath86 where @xmath87 is the stephan- boltzmann constant . as carriers of a gravitational `` charge '' ( without consideration of spin properties ) , gravitons should be described in the same manner as photons ( compare with @xcite ) , i.e. one can write for them : @xmath88 now let us introduce a new dimensional constant @xmath89 , so that for forehead collisions : @xmath90 then @xmath91 where @xmath92 is an average graviton energy . assuming @xmath93 and @xmath94 we get the following rough estimate for @xmath95 @xmath96 ( see below section 4.3 for more exact estimate of @xmath89 and for a theoretical estimate of @xmath2 ) that gives us the phenomenological estimate of cross - section by the same and equal @xmath11 and @xmath92 : @xmath97 we can introduce the following new constants ( see @xcite ) : @xmath98 which are analogues , on this new scale , of classical constants : the newton constant @xmath99 the planck length @xmath100 and the planck energy @xmath101 correspondingly . let us accept from dimensional reasonings that @xmath102 where @xmath103 then we have for these new constants : @xmath104 if one would replace @xmath105 with @xmath106 then an electrostatic force , acting between two protons , will be only @xmath107 times smaller than a gravitational one by the same distance ( a theoretical finding of the newton constant @xmath105 is given below in section 4.3 ) . using @xmath23 instead of @xmath108 we can evaluate the new non - dimensional `` constant '' ( a bilinear function of @xmath11 and @xmath109 ) , @xmath110 which would characterize one act of interaction : @xmath111 we must remember here , that a universality of gravitational interaction allows to expect that this floating coupling `` constant '' @xmath112 should characterize interactions of any particles with an energy @xmath113 including gravitons , with single gravitons . for @xmath114 and @xmath115 we have @xmath116 but for @xmath117 and @xmath118 we shall have @xmath119 i.e. @xmath112 will be comparable with qed s constant @xmath120 already by @xmath121 such an interaction would have the same intensity as the strong interaction ( @xmath122 ) . as it was reported by anderson s team @xcite , nasa deep - space probes ( pioneer 10/11 , galileo , and ulysses ) experience a small additional constant acceleration , directed towards the sun ( the pioneer anomaly ) . today , a possible origin of the effect is unknown . it must be noted here that the reported direction of additional acceleration may be a result of the simplest conjecture , which was accepted by the authors to provide a good fit for all probes . one should compare different conjectures to choose the one giving the best fit . we consider here a deceleration of massive bodies , which would give a similar deformation of cosmic probes trajectories @xcite . the one would be a result of interaction of a massive body with the graviton background , but such an additional acceleration will be directed against a body velocity . it follows from a universality of gravitational interaction , that not only photons , but all other objects , moving relative to the background , should lose their energy , too , due to such a quantum interaction with gravitons . if @xmath20 it turns out that massive bodies must feel a constant deceleration of the same order of magnitude as a small additional acceleration of cosmic probes . let us now denote as @xmath11 a full energy of a moving body which has a velocity @xmath123 relative to the background . then energy losses of the body by an interaction with the graviton background ( due to forehead collisions with gravitons ) on the way @xmath28 must be expressed by the same formula ( 1 ) : @xmath18 where @xmath124 if @xmath125 where @xmath126 is a time , and @xmath127 then we get for the body acceleration @xmath128 by a non - zero velocity : @xmath129 we assume here , that non - forehead collisions with gravitons give only stochastic deviations of a massive body s velocity direction , which are negligible . for small velocities : @xmath130 if the hubble constant @xmath2 is equal to @xmath131 ( it is the theoretical estimate of @xmath2 in this approach , see below section 4.3 ) , a modulus of the acceleration will be equal to @xmath132 that has the same order of magnitude as a value of the observed additional acceleration @xmath133 for nasa probes . we must emphasize here that the acceleration @xmath134 is directed against a body velocity only in a special frame of reference ( in which the graviton background is isotropic ) . in other frames , we may find its direction , using transformation formulae for an acceleration ( see @xcite ) . we can assume that the graviton background and the microwave one are isotropic in one frame ( the earth velocity relative to the microwave background was determined in @xcite ) . to verify my conjecture about the origin of probes additional acceleration , one could re - analyze radio doppler data for the probes . one should find a velocity of the special frame of reference and a constant probe acceleration @xmath134 in this frame which must be negative , as it is described above . these two parameters must provide the best fit for all probes , if the conjecture is true . in such a case , one can get an independent estimate of the hubble constant , based on the measured value of probe s additional acceleration : @xmath135 i would like to note that a deep - space mission to test the discovered anomaly is planned now at nasa by the authors of this very important discovery @xcite . under influence of such a small additional acceleration @xmath134 , a probe must move on a deformed trajectory . its view will be determined by small seeming deviations from exact conservation laws for energy and angular momentum of a not - fully reserved body system which one has in a case of neglecting with the graviton background . for example , ulysses should go some nearer to the sun when the one rounds it . it may be interpreted as an additional acceleration , directed towards the sun , if we shall think that one deals with a reserved body system . it is very important to understand , why such an acceleration has not been observed for planets . this acceleration will have different directions by motion of a body on a closed orbit , and one must take into account a solar system motion , too . as a result , an orbit should be deformed . the observed value of anomalous acceleration of pioneer 10 should represent the vector difference of the two accelerations @xcite : an acceleration of pioneer 10 relative to the graviton background , and an acceleration of the earth relative to the background . possibly , the last is displayed as an annual periodic term in the residuals of pioneer 10 @xcite . if the solar system moves with a noticeable velocity relative to the background , the earth s anomalous acceleration projection on the direction of this velocity will be smaller than for the sun - because of the earth s orbital motion . it means that in a frame of reference , connected with the sun , the earth should move with an anomalous acceleration having non - zero projections as well on the orbital velocity direction as on the direction of solar system motion relative to the background . under some conditions , the earth s anomalous acceleration in this frame of reference may be periodic . the axis of earth s orbit should feel an annual precession by it . this question needs a further consideration . it was shown by the author @xcite that screening the background of super - strong interacting gravitons creates for any pair of bodies both attraction and repulsion forces due to pressure of gravitons . for single gravitons , these forces are approximately balanced , but each of them is much bigger than a force of newtonian attraction . if single gravitons are pairing , an attraction force due to pressure of such graviton pairs is twice exceeding a corresponding repulsion force if graviton pairs are destructed by collisions with a body . in such the model , the newton constant is connected with the hubble constant that gives a possibility to obtain a theoretical estimate of the last . we deal here with a flat non - expanding universe fulfilled with super - strong interacting gravitons ; it changes the meaning of the hubble constant which describes magnitudes of three small effects of quantum gravity but not any expansion or an age of the universe . if gravitons of the background run against a pair of bodies with masses @xmath136 and @xmath137 ( and energies @xmath138 and @xmath139 ) from infinity , then a part of gravitons is screened . let @xmath140 is a cross - section of interaction of body @xmath141 with a graviton with an energy @xmath142 where @xmath79 is a graviton frequency , @xmath143 is the same cross - section for body @xmath144 in absence of body @xmath145 a whole modulus of a gravitonic pressure force acting on body @xmath141 would be equal to : @xmath146 where @xmath147 is a graviton spectrum with a temperature @xmath82 ( assuming to be planckian ) , the factor @xmath148 in front of @xmath149 is introduced to allow all possible directions of graviton running , @xmath150 is another average energy of running gravitons with a frequency @xmath79 taking into account a probability of that in a realization of flat wave a number of gravitons may be equal to zero , and that not all of gravitons ride at a body . body @xmath145 placed on a distance @xmath21 from body @xmath151 will screen a portion of running against body @xmath141 gravitons which is equal for big distances between the bodies ( i.e. by @xmath152 ) to : @xmath153 taking into account all frequencies @xmath154 the following attractive force will act between bodies @xmath141 and @xmath155 @xmath156 let @xmath147 is described with the planck formula : @xmath157 let @xmath158 and @xmath159 is an average number of gravitons in a flat wave with a frequency @xmath79 ( on one mode of two distinguishing with a projection of particle spin ) . let @xmath160 is a probability of that in a realization of flat wave a number of gravitons is equal to @xmath161 for example @xmath162 a quantity @xmath150 must contain the factor @xmath163 i.e. it should be : @xmath164 that let us to reject flat wave realizations with zero number of gravitons . but attempting to define other factors in @xmath165 we find the difficult place in our reasoning . on this stage , it is necessary to introduce some new assumption to find the factors . perhaps , this assumption will be well - founded in a future theory - or would be rejected . if a flat wave realization , running against a finite size body from infinity , contains one graviton , then one can not consider that it must stringent ride at a body to interact with some probability with the one . it would break the uncertainty principle by w. heisenberg . we should admit that we know a graviton trajectory . the same is pertaining to gravitons scattered by one of bodies by big distances between bodies . what is a probability that a single graviton will ride namely at the body ? if one denotes this probability as @xmath166 then for a wave with @xmath167 gravitons their chances to ride at the body must be equal to @xmath168 taking into account the probabilities of values of @xmath167 for the poisson flux of events , an additional factor in @xmath150 should be equal to @xmath169 i have admitted in @xcite that @xmath170 where @xmath171 ( below it is admitted for pairing gravitons : @xmath172 - see section 4.3 ) . in such the case , we have for @xmath150 the following expression : @xmath173 then we get for an attractive force @xmath174 @xmath175 @xmath176 where @xmath177 @xmath178 this and all other integrals were found with the mathcad software . if @xmath179 then the constant @xmath180 is equal to : @xmath181 by @xmath182 @xmath183 that is three order greater than the newton constant , @xmath184 but if single gravitons are elastically scattered with body @xmath151 then our reasoning may be reversed : the same portion ( 13 ) of scattered gravitons will create a repulsive force @xmath185 acting on body @xmath186 and equal to @xmath187 if one neglects with small allowances which are proportional to @xmath188 so , for bodies which elastically scatter gravitons , screening a flux of single gravitons does not ensure newtonian attraction . but for gravitonic black holes which absorb any particles and do not re - emit them ( by the meaning of a concept , the ones are usual black holes ; i introduce a redundant adjective only from a caution ) , we will have @xmath189 it means that such the object would attract other bodies with a force which is proportional to @xmath180 but not to @xmath99 i.e. einstein s equivalence principle would be violated for them . this conclusion , as we shall see below , stays in force for the case of graviton pairing , too . the conclusion can not be changed with taking into account of hawking s quantum effect of evaporation of black holes @xcite . to ensure an attractive force which is not equal to a repulsive one , particle correlations should differ for _ in _ and _ out _ flux . for example , single gravitons of running flux may associate in pairs @xcite . if such pairs are destructed by collision with a body , then quantities @xmath150 will be distinguished for running and scattered particles . graviton pairing may be caused with graviton s own gravitational attraction or gravitonic spin - spin interaction . left an analysis of the nature of graviton pairing for the future ; let us see that gives such the pairing . to find an average number of pairs @xmath190 in a wave with a frequency @xmath79 for the state of thermodynamic equilibrium , one may replace @xmath191 by deducing the planck formula . then an average number of pairs will be equal to : @xmath192 and an energy of one pair will be equal to @xmath193 it is important that graviton pairing does not change a number of stationary waves , so as pairs nucleate from existing gravitons . the question arises : how many different modes , i.e. spin projections , may graviton pairs have ? we consider that the background of initial gravitons consists of two modes . for massless transverse bosons , it takes place as by spin @xmath141 as by spin @xmath144 if graviton pairs have maximum spin @xmath145 then single gravitons should have spin @xmath31 but from such particles one may constitute four combinations : @xmath194 ( with total spin @xmath186 ) , and @xmath195 ( with total spin @xmath196 all these four combinations will be equiprobable if spin projections @xmath197 and @xmath198 are equiprobable in a flat wave ( without taking into account a probable spin - spin interaction ) . but it is happened that , if expression ( 24 ) is true , it follows from the energy conservation law that composite gravitons should be distributed only in two modes . so as @xmath199 then by @xmath200 we have @xmath201 i.e. all of gravitons are pairing by low frequencies . an average energy on every mode of pairing gravitons is equal to @xmath202 the one on every mode of single gravitons - to @xmath203 these energies are equal by @xmath204 because of that , the numbers of modes are equal , too , if the background is in the thermodynamic equilibrium with surrounding bodies . the above reasoning does not allow to choose a spin value @xmath186 or @xmath205 for composite gravitons . a choice of namely spin @xmath186 would ensure the following proposition : all of gravitons in one realization of flat wave have the same spin projections . from another side , a spin - spin interaction would cause it . the spectrum of composite gravitons is also the planckian one , but with a smaller temperature ; it has the view : @xmath206 it means that an absolute luminosity for the sub - system of composite gravitons is equal to : @xmath207 where @xmath87 is the stephan - boltzmann constant ; i.e. an equivalent temperature of this sub - system is @xmath208 the portion of pairing gravitons , @xmath209 a spectrum of single gravitons , @xmath210 and a spectrum of subsystem of pairing gravitons , @xmath211 are shown on fig . 5 as functions of the dimensionless parameter @xmath212 . ( solid line ) , a spectrum of single gravitons , @xmath210 ( dashed line ) , and a spectrum of graviton pairs , @xmath211 ( dotted line ) versus the dimensionless parameter @xmath77.,width=490 ] it is important that the graviton pairing effect does not change computed values of the hubble constant and of anomalous deceleration of massive bodies : twice decreasing of a sub - system particle number due to the pairing effect is compensated with twice increasing the cross - section of interaction of a photon or any body with such the composite gravitons . non - pairing gravitons with spin @xmath141 give also its contribution in values of redshifts , an additional relaxation of light intensity due to non - forehead collisions with gravitons , and anomalous deceleration of massive bodies moving relative to the background . if running graviton pairs ensure for two bodies an attractive force @xmath213 then a repulsive force due to re - emission of gravitons of a pair alone will be equal to @xmath214 it follows from that the cross - section for _ single additional scattered _ gravitons of destructed pairs will be twice smaller than for pairs themselves ( the leading factor @xmath215 for pairs should be replaced with @xmath216 for single gravitons ) . for pairs , we introduce here the cross - section @xmath217 where @xmath218 is an average pair energy with taking into account a probability of that in a realization of flat wave a number of graviton pairs may be equal to zero , and that not all of graviton pairs ride at a body ( @xmath218 is an analog of @xmath150 ) . this equality is true in neglecting with small allowances which are proportional to @xmath219 ( see section 4.4 ) . replacing @xmath220 and @xmath221 where @xmath222 we get for graviton pairs : @xmath223 this expression does not take into account only that beside pairs there may be single gravitons in a realization of flat wave . to reject cases when , instead of a pair , a single graviton runs against a body ( a contribution of such gravitons in attraction and repulsion is the same ) , we add the factor @xmath224 into @xmath225 @xmath226 then a force of attraction of two bodies due to pressure of graviton pairs , @xmath227 , - in the full analogy with ( 19 ) - will be equal to : and @xmath2 were overestimated of @xmath228 times ] @xmath229 @xmath230 where @xmath231 @xmath232 the difference @xmath233 between attractive and repulsive forces will be equal to : @xmath234 where the constant @xmath235 is equal to : @xmath236 both @xmath180 and @xmath235 are proportional to @xmath237 ( and @xmath238 so as @xmath239 ) . if one assumes that @xmath240 then it follows from ( 34 ) that by @xmath241 the constant @xmath89 should have the value : @xmath242 an average graviton energy of the background is equal to : @xmath243 where @xmath244 ( it is @xmath245 by @xmath241 ) . we can use ( 8) and ( 34 ) to establish a connection between the two fundamental constants , @xmath105 and @xmath2 , under the condition that @xmath246 we have for @xmath95 @xmath247 then @xmath248 so as the value of @xmath105 is known much better than the value of @xmath249 let us express @xmath2 via @xmath250 @xmath251 or in the units which are more familiar for many of us : @xmath252 this value of @xmath2 is in the good accordance with the majority of present astrophysical estimations @xcite ( for example , the estimate @xmath253 km / s / mpc has been got from sn1a cosmological distance determinations in @xcite ) , but it is lesser than some of them @xcite and than it follows from the observed value of anomalous acceleration of pioneer 10 @xcite . the described quantum mechanism of classical gravity gives newton s law with the constant @xmath235 value ( 34 ) and the connection ( 38 ) for the constants @xmath235 and @xmath254 we have obtained the rational value of @xmath2 ( 39 ) by @xmath255 if the condition of big distances is fulfilled : @xmath256 because it is known from experience that for big bodies of the solar system , newton s law is a very good approximation , one would expect that the condition ( 40 ) is fulfilled , for example , for the pair sun - earth . but assuming @xmath257 and @xmath258 we obtain assuming for rough estimation @xmath259 @xmath260 it means that in the case of interaction of gravitons or graviton pairs with the sun in the aggregate , the considered quantum mechanism of classical gravity could not lead to newton s law as a good approximation . this `` contradiction '' with experience is eliminated if one assumes that gravitons interact with `` small particles '' of matter - for example , with atoms . if the sun contains of @xmath261 atoms , then @xmath262 where @xmath263 is an average energy of one atom . for rough estimation we assume here that @xmath264 where @xmath265 is a proton rest energy ; then it is @xmath266 i.e. @xmath267 this necessity of `` atomic structure '' of matter for working the described quantum mechanism is natural relative to usual bodies . but would one expect that black holes have a similar structure ? if any radiation can not be emitted with a black hole , a black hole should interact with gravitons as an aggregated object , i.e. the condition ( 40 ) for a black hole of sun mass has not been fulfilled even at distances @xmath268 for bodies without an atomic structure , the allowances , which are proportional to @xmath219 and are caused by decreasing a gravitonic flux due to the screening effect , will have a factor @xmath269 or @xmath270 these allowances break the equivalence principle for such the bodies . for bodies with an atomic structure , a force of interaction is added up from small forces of interaction of their `` atoms '' : @xmath271 where @xmath272 and @xmath273 are numbers of atoms for bodies @xmath141 and @xmath186 . the allowances to full forces due to the screening effect will be proportional to the quantity : @xmath274 which can be expressed via the full masses of bodies as @xmath275 or @xmath276 by big numbers @xmath272 and @xmath273 the allowances will be small . the allowance to the force @xmath277 acting on body @xmath145 will be equal to : @xmath278 @xmath279 ( for body @xmath141 we shall have the similar expression if replace @xmath280 ) , where @xmath281 let us find the ratio : @xmath282 using this formula , we can find by @xmath283 @xmath284 an analogical allowance to the force @xmath285 has by the same conditions the order @xmath286 or @xmath287 one can replace @xmath265 with a rest energy of very big atom - the geometrical approach will left a very good language to describe the solar system . we see that for bodies with an atomic structure the considered mechanism leads to very small deviations from einstein s equivalence principle , if the condition ( 40 ) is fulfilled for microparticles , which prompt interact with gravitons . for small distances we shall have : @xmath288 it takes place by @xmath289 for @xmath290 this quantity is many orders larger than the planck length . the equivalence principle should be broken at such distances . under the condition ( 44 ) , big digressions from newton s law will be caused with two factors : 1 ) a screening portion of a running flux of gravitons is not small and it should be taken into account by computation of the repulsive force ; 2 ) a value of this portion can not be defined by the expression ( 13 ) . instead of ( 13 ) , one might describe this portion at small distances with an expression of the kind : @xmath291 ( the formula for a spheric segment area is used here @xcite ) . formally , by @xmath292 we shall have for the portion ( 45 ) : @xmath293 where the second term shows that the interaction should be weaker at small distances . we might expect that a screening portion may tend to a fixing value at super - short distances , and it will be something similar to asymptotic freedom of strong interactions . but , of course , at such distances the interaction will be super - strong and our naive approach would be not valid . if the described model of redshifts is true , what is a picture of the universe ? it is interesting that in a frame of this model , every observer has two own spheres of observability in the universe ( two different cosmological horizons exist for any observer ) @xcite . one of them is defined by maximum existing temperatures of remote sources - by big enough distances , all of them will be masked with the cmb radiation . another , and much smaller , sphere depends on their maximum luminosity - the luminosity distance increases with a redshift much quickly than the geometrical one . the ratio of the luminosity distance to the geometrical one is the quickly increasing function of @xmath294 @xmath295 which does not depend on the hubble constant . an outer part of the universe will drown in a darkness . by the found theoretical value of the hubble constant : @xmath296 ( then a natural light unit of distances is equal to @xmath297 light gyr ) , plots of two theoretical functions of @xmath61 in this model - the geometrical distance @xmath25 and the luminosity distance @xmath54 - are shown on fig . 6 @xcite . as one can see , for objects with @xmath298 , which are observable now , we should anticipate geometrical distances of the order @xmath299 light gyr and luminosity distances of the order @xmath300 light gyr in a frame of this model . an estimate of distances to objects with given @xmath61 is changed , too : for example , the quasar with @xmath301 @xcite should be in a distance approximately of 2.8 times bigger than the one expected in the model based on the doppler effect . = 12.98 cm we can assume that the graviton background and the cosmic microwave one are in a state of thermodynamical equilibrium , and have the same temperatures . cmb itself may arise as a result of cooling any light radiation up to reaching this equilibrium . then it needs @xmath302 to get through the very edge of our cosmic `` ecumene '' ( see fig . = 12.98 cm some other possible cosmological consequences of an existence of the graviton background were described in @xcite . observations of last years give us strong evidences for supermassive and compact objects ( named now supermassive black holes ) in active and normal galactic nuclei @xcite . massive nuclear `` black holes '' of @xmath303 solar masses may be responsible for the energy production in quasars and active galaxies @xcite . in a frame of this model , an existence of black holes contradicts to the equivalence principle . it means that these objects should have another nature ; one must remember that we know only that these objects are supermassive and compact . there should be two opposite processes of heating and cooling the graviton background @xcite which may have a big impact on cosmology . unlike models of expanding universe , in any tired light model one has a problem of utilization of energy , lost by radiation of remote objects . in the considered model , a virtual graviton forms under collision of a photon with a graviton of the graviton background . it should be massive if an initial graviton transfers its total momentum to a photon ; it follows from the energy conservation law that its energy @xmath304 must be equal to @xmath305 if @xmath109 is an initial graviton energy . in force of the uncertainty relation , one has for a virtual graviton lifetime @xmath306 @xmath307 i.e. for @xmath308 it is @xmath309 in force of conservation laws for energy , momentum and angular momentum , a virtual graviton may decay into no less than three real gravitons . in a case of decay into three gravitons , its energies should be equal to @xmath310 with @xmath311 so , after this decay , two new gravitons with @xmath312 inflow into the graviton background . it is a source of adjunction of the graviton background . from another side , an interaction of gravitons of the background between themselves should lead to the formation of virtual massive gravitons , too , with energies less than @xmath313 where @xmath313 is a minimal energy of one graviton of an initial interacting pair . if gravitons with energies @xmath314 wear out a file of collisions with gravitons of the background , its lifetime increases . in every such a collision - decay cycle , an average energy of `` redundant '' gravitons will double decrease , and its lifetime will double increase . only for @xmath315 cycles , a lifetime will increase from @xmath316 to @xmath317 gyr . such virtual massive gravitons , with a lifetime increasing from one collision to another , would duly serve dark matter particles . having a zero ( or near to zero ) initial velocity relative to the graviton background , the ones will not interact with matter in any manner excepting usual gravitation . an ultra - cold gas of such gravitons will condense under influence of gravitational attraction into `` black holes '' or other massive objects . additionally to it , even in absence of initial heterogeneity , the one will easy arise in such the gas that would lead to arising of super compact massive objects , which will be able to turn out `` germs '' of `` black holes '' . it is a method `` to cool '' the graviton background . so , the graviton background may turn up `` a perpetual engine '' of the universe , pumping energy from any radiation to massive objects . an equilibrium state of the background will be ensured by such a temperature @xmath318 for which an energy profit of the background due to an influx of energy from radiation will be equal to a loss of its energy due to a catch of virtual massive gravitons with `` black holes '' or other massive objects . in such the picture , the chances are that `` black holes '' would turn out `` germs '' of galaxies . after accumulation of a big enough energy by a `` black hole '' ( to be more exact , by a super - compact massive object ) by means of a catch of virtual massive gravitons , the one would be absolved from an energy excess in via ejection of matter , from which stars of galaxy should form . it awaits to understand else in such the approach how usual matter particles form from virtual massive gravitons . there is a very interesting but non - researched possibility : due to relative decreasing of an intensity of graviton pair flux in an internal area of galaxies ( pairs are destructed under collisions with matter particles ) , the effective newton constant may turn out to be running on galactic scales . it might lead to something like to the modified newtonian dynamics ( mond ) by mordehai milgrom ( about mond , for example , see @xcite ) . but to evaluate this effect , one should take into account a relaxation process for pairs , about which we know nothing today . it is obvious only that gravity should be stronger on a galactic periphery . the renormalization group approach to gravity leads to modifications of the theory of general relativity on galactic scales @xcite , and a growth of newton s constant at large distances takes place , too . kepler s third law receives quantum corrections that may explain the flat rotation curves of the galaxies . i would like to show here ( see @xcite ) a full realizability at present time of verifying my basic conjecture about the quantum gravitational nature of redshifts in a ground - based laser experiment . of course , many details of this precision experiment will be in full authority of experimentalists . it was not clear in 1995 how big is a temperature of the graviton background , and my proposal @xcite to verify the conjecture about the described local quantum character of redshifts turned out to be very rigid : a laser with instability of @xmath319 hasnt appeared after 9 years . but if @xmath320 , the satellite of main laser line of frequency @xmath321 after passing the delay line will be red - shifted at @xmath5 ev / h and its position will be fixed ( see fig . 8) . it will be , and its intensity should linear rise with a path of photons in a delay line , @xmath322 . a center - of - mass of both lines is expected to be approximately near @xmath323.,width=490 ] caused by the fact that on a very small way in the delay line only a small part of photons may collide with gravitons of the background . the rest of them will have unchanged energies . the center - of - mass of laser radiation spectrum should be shifted proportionally to a photon path . then due to the quantum nature of shifting process , the ratio of satellite s intensity to main line s intensity should have the order : @xmath324 where @xmath322 is a path of laser photons in a vacuum tube of delay line . it gives us a possibility to plan a laser - based experiment to verify the basic conjecture of this approach with much softer demands to the equipment . an instability of a laser of a power @xmath325 must be only @xmath326 if a photon energy is of @xmath327 . it will be necessary to compare intensities of the red - shifted satellite at the very beginning of the path @xmath322 and after it . given a very low signal - to - noise ratio , one could use a single photon counter to measure the intensities . when @xmath328 is a quantum output of a cathode of the used photomultiplier ( a number of photoelectrons is @xmath328 times smaller than a number of photons falling to the cathode ) , @xmath329 is a frequency of its noise pulses , and @xmath167 is a desired ratio of a signal to noise s standard deviation , then an evaluated time duration @xmath126 of data acquisition would have the order : @xmath330 assuming @xmath331 we would have the estimate : @xmath332 years , that is unacceptable . but given @xmath333 , we get : @xmath334 days , that is acceptable for the experiment of such the potential importance . of course , one will rather choose a bigger value of @xmath322 by a small laser power forcing a laser beam to whipsaw many times between mirrors in a delay line - it is a challenge for experimentalists . from thermodynamic reasons , it is assumed here that the graviton background has the same temperature as the microwave background . also it follows from the condition of detail equilibrium , that both backgrounds should have the planckian spectra . composite gravitons will have spin @xmath186 , if single gravitons have the same spin as photons . the question arise , of course : how are gravitons and photons connected ? has the conjecture by adler et al . @xcite ( that a graviton with spin @xmath186 is composed with two photons ) chances to be true ? intuitive demur calls forth a huge self - action , photons should be endued with which if one unifies the main conjecture of this approach with the one by adler et al . - but one may get a unified theory on this way . to verify this combined conjecture in experiment , one would search for transitions in interstellar gas molecules caused by the microwave background , with an angular momentum change corresponding to absorption of spin @xmath186 particles ( photon pairs ) . a frequency of such the transitions should correspond to an equivalent temperature of the sub - system of these composite particles @xmath335 if @xmath82 is a temperature of the microwave background . from another side , one might check this conjecture in a laser experiment , too . taking two lasers with photon energies @xmath336 and @xmath337 , one may force laser beams to collide on a way @xmath338 ( see fig . 9 ) . if photons are self - interacting particles , we might wait that photons with energies @xmath339 , if @xmath340 , would arise after collisions of initial photons . . an expected beam of photons with energies @xmath339 falls to a photoreceiver.,width=490 ] if we _ assume ( only here ) _ that single gravitons are identical to photons , it will be necessary to take into account the following circumstances to calculate an analog of the hubble constant for this experiment : an average graviton energy should be replaced with @xmath337 , the factor @xmath341 in ( 8) should be replaced with @xmath342 , where @xmath343 is a divergence of laser beam 2 , and one must use a quantity @xmath344 instead of @xmath345 in ( 8) , where @xmath325 is a laser 2 power and @xmath346 is a cross - section of its beam . together all it means that we should replace the hubble constant with its analog for a laser beam collision , @xmath347 : @xmath348 taken @xmath349 , @xmath350 , @xmath351 , and @xmath352 , that is characterizing a he - ne laser , we get the estimate : @xmath353 . then photons with energies @xmath339 would fall to a photoreceiver with a frequency which should linearly rise with @xmath338 ( proportionally to @xmath354 ) , and it would be of @xmath355 if both lasers have equal powers @xmath356 , and @xmath357 . it is a big enough frequency to give us a possibility to detect easy a flux of these expected photons in ir band . i think there is not any sense to try to analyze theoretically consequences of this conjecture - it will be easier to verify it experimentally . the nature may answer the question if we ask correctly . all that was said in the above sections does nt depend on the answer , but it would be very important for our understanding of known interactions . if this tentative non - linear vacuum effect exists , it would lead us far beyond standard quantum electrodynamics to take into account new non - linearities ( which are not connected with electron - positron pair creation ) and an essential impact of such a non - locally born object as the graviton background . it follows from the above consideration that the geometrical description of gravity should be a good idealization for any pair of bodies at a big distance by the condition of an `` atomic structure '' of matter . this condition can not be accepted only for black holes which must interact with gravitons as aggregated objects . in addition , the equivalence principle is roughly broken for black holes , if the described quantum mechanism of classical gravity is realized in the nature . because attracting bodies are not initial sources of gravitons , a future theory must be non - local in this sense to describe gravitons running from infinity . non - local models were considered by g.v . efimov in his book @xcite . the le sage s idea to describe gravity as caused by running _ particles was criticized by the great physicist richard feynman in his public lectures at cornell university @xcite , but the pioneer 10 anomaly @xcite , perhaps , is a good contra argument pro this idea . the described quantum mechanism of classical gravity is obviously asymmetric relative to the time inversion . by the time inversion , single gravitons would run against bodies to form pairs after collisions with bodies . it would lead to replacing a body attraction with a repulsion . but such the change will do impossible the graviton pairing . cosmological models with the inversion of the time arrow were considered by sakharov @xcite . penrose has noted that a hidden physical law may determine the time arrow direction @xcite ; it will be very interesting if namely realization in the nature of newton s law determines this direction . a future theory dealing with gravitons as usual particles should have a number of features which are not characterizing any existing model to image the considered here features of the possible quantum mechanism of gravity . if this mechanism is realized in the nature , both the general relativity and quantum mechanics should be modified . any divergencies , perhaps , would be not possible in such the model because of natural smooth cut - offs of the graviton spectrum from both sides . gravity at short distances , which are much bigger than the planck length , needs to be described only in some unified manner . references carlip , s. _ rept . _ 2001 , _ 64 , _ 885 ; [ gr - qc/0108040 ] . anderson , j.d . _ 1998 , _ 81 , _ 2858 ; _ phys . d65 , _ 082004 ; [ gr - qc/0104064 v4 ] . riess , a.g . aj _ 1998 , _ 116 , _ 1009 . perlmutter , s. et al . _ apj _ 1999 , _ 517 , _ 565 . nesvizhevsky , v.v . _ nature _ 2002 , _ 415 , _ 297 . ivanov , m.a . _ general relativity and gravitation _ 2001 , _ 33 , _ 479 ; erratum : 2003 , _ 35 , _ 939 ; [ astro - ph/0005084 v2 ] . ivanov , m.a . [ gr - qc/0009043 ] ; proc . of the int . `` frontiers of fundamental physics 4 '' ( 9 - 13 dec 2000 , hyderabad , india ) , sidharth , b.g . , altaisky , m.v . kluwer academic / plenum publishers : august 2001 ; proc . of the 4th edoardo amaldi conference on gravitational waves ( perth , western australia , 8 - 13 july 2001 ) _ class . quantum grav . _ 2002 , _ 19 , _ 1351 . ivanov , m.a . screening the graviton background , graviton pairing , and newtonian gravity [ gr - qc/0207006 ] . ivanov , m.a . _ acta phys . polonica _ 1990 , _ b21 , _ 25 and 813 ; _ nuovo cimento _ 1992 , _ 105a , _ 77 ; [ hep - th/0207210 ] . hubble , e.p . , and humason , m. _ apj , _ 1931 , _ 74 , _ 43 . penzias , a.a . , wilson , r.w . _ apj , _ 1965 , _ 142 , _ 419 . smoot , g.f . _ apj ( letters ) , _ 1992 , _ 396 , _ l1 . laviolette , p.a . _ apj _ 1986 , _ 301 , _ 544 . ivanov , m.a . another origin of cosmological redshifts [ astro - ph-0405083 ] . riess , a.g . type ia supernova discoveries at @xmath358 from the hubble space telescope ... [ astro - ph/0402512 ] ( to appear in _ apj , _ 2004 ) . ivanov , m.a . contribution to the quantum electronics and laser science conference ( qels95 ) , may 21 - 26 , 1995 , baltimore , usa ; paper number : qthg1 . murphy , m. t. et al . time evolution of the fine structure constant [ astro - ph/0209488 ] . srianand , r. et al . _ nature _ 2000 , _ 408 , _ 931 ; [ astro - ph/0012222 ] . ivanov , m.a . contribution to the sixth international symposium `` frontiers of fundamental and computational physics '' ( ffp6 ) , 26 - 29 september 2004 , udine , italy ; [ astro - ph/0409631 ] . choi , s.y . d48 , _ 2953 ; _ d48 , _ r5465 ; 1995 , _ d51 , _ 2751 . ivanov , m.a . contribution to the tenth marcel grossmann meeting ( mg10 ) , 20 - 26 july 2003 , rio de janeiro , brazil ; [ gr - qc/0307093 ] . ivanov , m.a . contribution to the conference `` thinking , observing and mining the universe '' ( thinking2003 ) , 22 - 27 sep 2003 , sorrento , italy ; [ astro - ph/0309566 ] . hawking , s.w . _ _ 1975 , _ 43 , _ 199 . tolman , r.c . _ relativity , thermodynamics and cosmology . _ clarendon press : oxford , 1969 . conklin , e.k . _ nature _ 1969 , _ 222 , _ 971 . anderson , j.d . , nieto , m.m . , turyshev , s.g . d11 , _ 1545 ; [ gr - qc/0205059 v3 ] . turyshev , s.g . xxxiv - th rencontres de moriond meeting on gravitational waves and experimental gravity . les arcs , savoi , france ( january 23 - 30 , 1999 ) ; [ gr - qc/9903024 v2 ] . freedman , w. l. et al . _ apj _ 2001 , _ 553 , _ 47 . filippenko , a.v . [ astro - ph/0410609 ] ; to be published in _ white dwarfs : probes of galactic structure and cosmology , _ sion , e. m. , shipman , h. l. , vennes , s. , eds . ; kluwer : dordrecht . willick , j.a . , puneet batra . _ apj _ 2001 , _ 548 , _ 564 . korn , g.a . , korn , t.m . _ mathematical handbook , _ second edition , mcgraw - hill book co. : new york , 1968 , table 1.10 - 2 . ivanov , m.a . in _ searches for a mechanism of gravity , _ ivanov , m.a . , savrov , l.a . nikolaev , yu.a . publisher : nizhny novgorod , 2004 , pp 266 - 273 ( in russian ) . ivanov , m.a . a quantum gravitational model of redshifts [ astro - ph/0409111 ] . ivanov , m.a . model of graviton - dusty universe [ gr - qc/0107047 ] . fan , x. et al . _ aj _ 2000 , _ 120 , _ 1167 . van der marel , r. p. in _ new light on galaxy evolution , _ proc . heidelberg , june 1995 , r. bender and r.l . davies , eds . ; kluwer academic publishers . kleijn , g. a. v. et al . _ aj _ 2000 , _ 120 , _ 1221 . peterson , b. m. , wandel , a. _ apj _ 2000 , _ 540 , _ l13 . di matteo , t. , carilli , c.l . , fabian , a.c . _ apj _ 2001 , _ 547 , _ 731 . kenji bekki . _ apj _ 2000 , _ 540 , _ l79 . milgrom , m. in the proceedings of the ii int . workshop on the identification of dark matter , buxton , england , 1998 ; world scientific : singapore , 1999 ; [ astro - ph/9810302 ] . shapiro , i.l . , sola , j. , stefancic , h. running g and @xmath359 at low energies from physics at @xmath360 : possible cosmological and astrophysical implications [ hep - ph/0410095 ] . reuter , m. , weyer , h. running newton constant , improved gravitational actions , and galaxy rotation curves [ hep - th/0410117 ] . ivanov , m.a . contribution to the 14th workshop on general relativity and gravitation ( jgrg14 ) , which will be held in nov 29 - dec 3 2004 , kyoto , japan ; [ gr - qc/0410076 ] . adler , s.l . d14 , _ 359 . adler , s.l . d14 , _ 379 . efimov , g.v . _ non - local interactions of quantized fields . _ nauka : moscow , 1968 . feynman , r. _ the character of physical law . _ cox and wyman ltd . : london , 1965 . sakharov , a.d . _ jetp _ 1980 , _ 79 , _ 689 . r. in the einstein survey _ general relativity , _ hawking , s.w . , israel , w. , eds . ; cambridge university press : 1979 .

we report the distance and [ fe / h ] value for the globular cluster ngc 5904 ( m5 ) derived from the fourier decomposition of the light curves of selected rrab and rrc stars . the aim in doing this was to bring these parameters into the homogeneous scales established by our previous work on numerous other globular clusters , allowing a direct comparison of the horizontal branch luminosity in clusters with a wide range of metallicities . our ccd photometry of the large variable star population of this cluster is used to discuss light curve peculiarities , like blazhko modulations , on an individual basis . new blazhko variables are reported . from the rrab stars we found [ fe / h] , and a distance of kpc , and from the rrc stars we found [ fe / h] = and a distance of kpc . the results for rrab and rrc stars should be considered independent since they come from different calibrations and zero points . absolute magnitudes , radii and masses are also reported for individual rr lyrae stars . the distance to the cluster was also calculated by alternative methods like the period - luminosity relation of sx phe and the luminosity of the stars at the tip of the red giant branch , and we obtained the results and kpc respectively . the distribution of rr lyrae stars in the instability strip is discussed and compared with other clusters in connection with the oosterhoff and horizontal branch type . the oosterhoff type ii clusters systematically show a rrab - rrc segregation about the instability strip first - overtone red edge , while the oosterhoff type i clusters may or may not display this feature . a group of rr lyrae stars is identified in an advanced evolutionary stage , and two of them are likely binaries with unseen companions .

1604.03981

the globular cluster ngc 5904 ( m5 , or c1645 + 476 in the iau nomenclature ) ( @xmath7 , @xmath8 , j2000 ; @xmath9 , @xmath10 ) is among the closest globular clusters ( gcs ) to the sun and hence it is very bright . its horizontal branch is at about @xmath1115 mag . being a nearby cluster , it is affected by very little reddening , @xmath12 mag ( harris 1996 ) . m5 has a very rich population of variable stars and no doubt its proximity has contributed to the very early discovery of numerous variables . the first 46 variables were discovered by solon j. bailey in the last decade of the xix century , on photographs taken with the 13-inch boyden telescope at arequipa , peru , and they were announced by pickering ( 1896a ) . periods of some of these variables were calculated by pickering ( 1896b ) and barnard ( 1898 ) . bailey himself reported the periods for 63 of the nearly 90 variables then known ( bailey & leland 1899 ) and in 1902 he listed x , y positions for 92 variables ( v1-v92 , his table xxiv ) and offered an identification chart ( his fig . 2 , bailey 1902 ) . despite the richness of the cluster in variable stars , the next batch of discoveries only happened about forty years later when oosterhoff ( 1941 ) found v93-v103 . yet another 46 years later , kadla et al . ( 1987 ) found the variables v104-v114 . a year later variables v115-v131 were announced by kravtsov ( 1988 ) while v132-v133 were found by kravtsov ( 1991 ) . intense ccd photometry of the cluster led to the discovery of 35 more variables between 1996 and 2000 ; v134-v141 were announced by sandquist et al . ( 1996 ) although v134 and v135 were the already known variables v129 and v36 respectively . in some cases the names of the new variables given in the original papers were modified by caputo et al . ( 1999 ) or in the catalogue of variable stars in globular clusters ( cvsgc ; clement et al.2001 ) to yield a consistent list of variable names . thus v142 - 143 were found by brocato et al . ( 1996 ) , v144-v148 by reid ( 1996 ) , v149-v154 by yan & reid ( 1996 ) , v155 - 159 by drissen & shara ( 1998 ) , v160-v163 by olech et al . ( 1999 ) and v164-v168 by kaluzny et al . v169 was noted by rees ( 1993 ) and confirmed by kaluzny et al . the last batch of variables , v170-v181 , one sx phe ( v170 ) and 11 semi - regular late - type ( sra ) variables , was recently announced by arellano ferro et al . ( 2015a ) . this makes nearly 120 years of variable star discoveries in this cluster . as in most of our recent papers we have employed the dandia implementation of difference image analysis ( dia ) ( bramich 2008 ; bramich et al . 2013 ) to extract high - precision photometry for all of the point sources in the field of m5 . we collected 6890 light cuves in the @xmath13 and @xmath14 bandpasses with the aim of building up a colour - magnitude diagram ( cmd ) and discussing the horizontal branch ( hb ) structure as compared to other oosterhoff type i ( ooi ) and oosterhoff type ii ( ooii ) clusters . days in the type i or ooi and @xmath15 days in the type ii or ooii . also the percentage of first overtone pulsators or rrc stars is higher in ooi clusters . ] we also fourier decompose the light curves of the rr lyrae stars ( rrl ) to calculate their metallicity and luminosity in order to provide independent and homogeneous estimates of the cluster mean metallicity and distance . the scheme of the paper is as follows : in @xmath16 2 we describe the observations , data reduction and calibration to the standard system . in @xmath16 3 the periods and phased light curves of rrl stars are displayed and the fourier light curve decomposition of stable rrl is described . the corresponding individual values of [ fe / h ] and @xmath17 are reported . @xmath16 4 deals with the discussion of the distribution of rrl in the hb . in @xmath16 5 the metallicity and distance of the parent cluster are inferred from the rrl and independent distance calculations from the p - l relation of sx phe stars , and the luminosity of the tip of the red giant branch ( trgb ) . in this section we also give comments on peculiar stars . finally , in @xmath16 6 we summarize our conclusions . .the distribution of observations of m5 . columns @xmath18 and @xmath19 give the number of images taken with the @xmath13 and @xmath14 filters respectively . columns @xmath20 and @xmath21 provide the exposure time , or range of exposure times . the average seeing is listed in the last column . [ cols="<,^,^,^,^,^ " , ] the physical parameters for the rrl and the inverse - variance weighted means are reported in table [ fisicos ] . two independent estimations of [ fe / h]@xmath22 were found from 38 rrab and 24 rrc stars in m5 : @xmath23 and @xmath24 respectively , where the uncertainties are the standard deviations of the inverse - variance weighted means . morgan ( 2013 ) published a new calibration of [ fe / h]@xmath22 for rrc stars , calculated for an extended number of stars and clusters , and noted that the results of the new calibration are in general consistent with the old one from morgan et al . ( 2007 ) , and that the average differences are of the order of 0.02 dex . we have also carried out the calculation using the new calibration and found [ fe / h]@xmath25 , i.e. 0.04 dex more deficient than the value derived with the calibration of morgan et al . ( 2007 ) which is in better agreement with the result from the rrab calibration . a similar conclusion was found by arellano ferro et al . ( 2015b ) for the case of ngc 6229 . the weighted mean @xmath17 values for the rrab and rrc stars are [email protected] mag and [email protected] mag respectively and will be used in section [ sec : distance ] to estimate the mean distance to the parent cluster . recently we analysed in detail the hb structure of ngc 6229 ( arellano ferro et al . 2015b ) , a cluster that apparently has a similar history and evolution to that of m5 . we comment here on their similarities and some important differences . from an analysis of the proper motion of m5 , cudworth & hanson ( 1993 ) , cudworth ( 1997 ) and scholz et al . ( 1996 ) argued that m5 , which is currently in the inner - halo , has a wide galactic orbit reaching as far as 40 - 50 kpc from the galactic center , which reveals m5 to actually be , like ngc 6229 , an outer - halo cluster . the similarity of metallicity and age between m5 and ngc 6229 has been noted in the past ( borissova et al . in fact they have essentially the same metallicity ; [ fe / h]@xmath22 , -1.36 ( m5 ) and -1.30 ( ngc 6229 ) and their cmds indeed resemble each other , both having a long blue tail and a well populated hb ; their values of the lee - zinn parameter @xmath27 ( @xmath28 where @xmath29 refer to the number of stars to the blue , inside , and to the red of the is ) are + 0.31 and + 0.24 respectively . borissova et al . ( 1999 ) argued indetail in favour of the two clusters being coeval to within 1 gyr . ngc 6229 is an ooi cluster , as recently discussed by arellano ferro et al . ( 2015b ) . in m5 , the periods of the 79 rrab and 34 rrc stars listed in table [ variables ] average 0.553d and 0.311d , respectively ; these values , along with the ratio of the number of rrab to rrc stars , clearly confirm that m5 is also an oosterhoff type i cluster . both clusters have a substantial number of sr variables and are poor in sx phe stars , since three sx phe are known in m5 and only one in ngc 6229 . however , when it comes to the instability strip ( is ) in the hb , there are important differences : m5 has about twice as many stars in the hb as ngc 6629 , which is a consequence of m5 being brighter and then more massive , and hence it is not a surprise that it has about twice as many rrl . nearly a hundred rrl in m5 show secular period variations ( szeidl et al . 2011 ) while these have been detected and measured in some 16 stars in ngc 6229 ( arellano ferro et al . we must note however that , although the time base of available data is similar for m5 ( at least 100 years ) and ngc 6229 ( @xmath3083 years ) , the latter has not been observed as heavily as the former over the last century , which may have limited the discovery of secular period changes . a close comparison of the hb in m5 and ngc 6229 has led borisova et al . ( 1999 ) to conclude that the blue hb in ngc 6229 is brighter than in m5 , although the red hb matches very well . the bailey , or period - amplitude diagram , of m5 for the @xmath13 and @xmath14 filters is shown in fig . [ bailey ] . the scatter of the rrab stars about the general trends represented by the continuous lines derived for the ooi clusters m3 by cacciari et al . ( 2005 ) in @xmath13-band , and ngc 2808 by kunder et al . ( 2013a ) in @xmath14-band , is quite considerable and a number of outliers sitting on the dashed sequence can be identified and they have been labelled in the figure . according to cacciari et al . ( 2005 ) this is the locus for the evolved stars . the bailey diagram further confirms m5 as an ooi type cluster . and @xmath14 amplitudes of the rrl in m5 . filled and open symbols represent rrab and rrc stars respectively . triangles are used for stars with blazhko modulations . in the top panel the continuous and segmented lines are the loci found by cacciari et al . ( 2005 ) for unevolved and evolved stars , respectively , in the ooi cluster m3 . in the bottom panel the black segmented locus was found by arellano ferro et al . ( 2011 , 2013 ) for the ooii clusters ngc 5024 and ngc 6333 . the blue loci are from kunder et al . ( 2013 ) for the ooi cluster ngc 2808 . the red parabolas were calculated by arellano ferro et al . ( 2015b ) for a sample of rrc stars in five ooi clusters . the black parabola was found by kunder et al . ( 2013b ) for 14 ooii clusters . ] several features can be highlighted from the distribution of rrl in the hb from its expanded version in the bottom panel of fig . [ cmd ] , in combination with the bailey diagram of fig . [ bailey ] and the period - colour diagram displayed in fig . [ lp_col ] , very useful for the task of interpreting the distribution of rrl stars in the instability strip . in this last diagram , the period @xmath31 , where 15.021 mag is the average @xmath32 for all the stable rrl , is plotted as a function of @xmath33 . the use of @xmath34 significantly reduces the scatter in the diagram ( van albada & baker 1971 ; bingham et al . 1984 ) . stable rrab and rrc stars as well as those exhibiting blazhko modulations have been distinguished in these three figures . let us note first that all rrab stars falling in the evolved sequence of the top panel of fig . [ bailey ] have been plotted with red filled circles and red labels in the bottom panel of fig.[cmd ] , and they are @xmath35 the brightest stars , as expected for stars in an advanced evolutionary state towards the agb . however a higher luminosity can be attained by non - evolutionary means , like the presence of an unseen companion or the existence of atmospheric enrichment of helium due to extra mixing at the rgb stage ( sweigart 1997 ) . let us note the labelled outlier stars in fig . [ lp_col ] ( v93 , v107 , v108 , v118 , v119 , v122 , v125 , v158 ) ( other peculiar outliers v25 , v104 and v142 are discussed below in section [ sec : ind_stars ] ) . these stars are labelled in the bottom panel of fig . [ cmd ] in black . these stars have too short a period for their colour which implies that they have larger gravity and should be of lower luminosity . if in spite of this they happen to be among the most luminous stars in the is , then they are good candidates to have a companion . this seems to be the case of v107 , v122 and v158 . similar cases were identified by cacciari et al . ( 2005 ) among the rrl in m3 ( v48 , v58 , and v146 ) and by arellano ferro et al . ( 2015b ) in ngc 6229 ( v14 , v31 , v54 and v55 ) . thus , these stars with shorter periods compared to others of similar colour , could have enhanced helium in their atmospheres or have become luminous due to the presence of an unseen companion . the remainder of the stars with red symbols might be truly evolved stars . unfortunately none of this was included in the secular period change analysis of szeidl et al . ( 2011 ) perhaps because most of them were only discovered after 1987 and the time - base of existing data is not long enough . it has been noted and discussed in several of our recent papers that some clusters show a neat splitting in the distribution of stable rrab and rrc stars in the cmd with the border at @xmath36 . in fig . [ cmd ] the corresponding border lines for ngc 6229 , ngc 5024 and ngc 4590 duly reddened are indicated by vertical black dashed lines . a detailed discussion of this fact has been given by arellano ferro et al . ( 2015b ) ( their section 5 ) . in brief , a clear rrab - rrc splitting is distinguished in the ooii clusters ngc 288 , ngc 1904 , ngc 5024 , ngc 5053 , ngc 5466 , ngc 6333 and ngc 7099 , all with rather blue hbs ( @xmath37 0.4 ) . it is also observed in the ooii cluster ngc 4590 which has a red hb ( @xmath27=0.17)(see cmd in fig . 11 of kains et al . among ooi clusters , ngc 3201 does not present the splitting ( arellano ferro et al . 2014 ) while ngc 6229 does , very clearly ( arellano ferro et al . 2015b ) . in fig . [ fel ] we have plotted the lee - zinn parameter @xmath27 as a function of [ fe / h]@xmath38 taken mostly from table 2 in catelan ( 2009 ) . the ooii clusters all show blue tails , except ngc 4590 . the more metallic ooi clusters may have both blue and red hbs and display a large dispersion . those clusters for which we have explored the rr lyrae distribution are plotted with filled symbols while empty symbols are used otherwise . three ooi clusters are also labelled ; despite their closeness in the plot the only one with a clear rrab - rrc segregation is ngc 6229 , while ngc 3201 , and now m5 ( see fig . [ cmd ] ) , show both blazhko and stable rrab stars well distributed all across the is , in the either - or region , i.e. to the blue of the first - overtone red border . it should be noted however that in all cases stable rrc stars are well confined to the blue of the first - overtone red border as expected . in passing let us mention that this is yet another important difference between m5 and `` its twin '' ngc 6229 . it is true that we have explored only three ooi clusters and that an analysis of a larger number of both types of oosterhoff clusters is desirable ; however , we might preliminarily conclude that blue - tailed ooii clusters seem to have their rrab and rrc stars systematically well segregated around the red border of the first - overtone instability strip . however , ooi clusters may or may not have this property . . the reduced period @xmath34 is defined as @xmath39.,width=313,height=207 ] as a function of [ fe / h]@xmath38 for galactic gcs of different oosterhoff types , according to the legend , with data taken from catelan ( 2009 ) . inner and outer halo clusters are represented by circles and squares respectivelly , width=302,height=283 ] . [ fel ] the distribution of rrl on the hb may be explained by the arguments of caputo et al . ( 1978 ) which involve the occurrence of a hysteresis mechanism ( van albada & baker 1973 ) for stars crossing the is . according to this mechanism the stars in the `` either - or '' region can retain the mode they were pulsating in before entering the region , which depends on whether the star is coming from the blue side as an rrc or from the red side as an rrab . caputo et al . ( 1978 ) also suggest that the original mode in which a rrl is pulsating depends on the location of the starting point at the zero age horizontal branch ( zahb ) , which in turn depends on the mass and the chemical composition of the star ; they used these ideas to explain the existence of the two oosterhoff groups as follows ; 1 ) the zahb point is in the fundamental zone , leading to an assortment of rrc and rrab in the `` either - or '' region and a lower value of the average period for both rrc and rrab and a lower proportion of rrc stars , hence an ooi cluster , and 2 ) the zahb point is bluer than the fundamental zone , in the `` either - or '' or in the first overtone regions , then the `` either - or '' region is populated exclusively by rrc stars and the rrab stars are to be found only in the fundamental region , producing larger averages of periods and a higher proportion of rrc and hence an ooii type cluster . the above scenario explains well the clean segregation of rrc and rrab seen in the ooii clusters , or the definitive lack of segregation in ooi clusters like ngc 3201 or m5 , but , as commented by arellano ferro et al . ( 2015b ) , the clean segregation observed in the ooi cluster ngc 6229 is at odds with the above picture . furthermore , rrab - rrc segregation in ooii clusters is also favoured by the arguments of pritzl et al . ( 2002 ) that in clusters with blue hbs , i.e. large values of @xmath27 , stars with masses below a critical mass on the zahb to the blue of the is , evolve redwards and spend sufficient time to contribute to the population of rrl . thus , clusters with a blue hb morphology are ooii , with redwards evolution , which in turn favours the segregation between rrc and rrab on the is which , in passing , should at least statistically display increasing secular period variations . unfortunately there are few clusters with a large enough number of rrl with secularly changing periods that have been studied ; e.g. the ooii clusters omega cen ( martin 1938 ) and m15 ( silbermann & smith 1995 ) and the ooi clusters m3 ( corwin & carney 2001 ) and m5 ( szeidl et al . 2011 ) ; in all these clusters only a small surplus , and probably not statistically significant , of rrl are reported as having increasing periods . growing evidence of the existence of multiple stellar populations in gcs ( see for instance jang & lee ( 2015 ) and the references therein ) invites us to consider whether this is connected with the observed rrab - rrc splitting discussed above . according to jang & lee ( 2015 ) , in inner - halo gcs the time elapsed between the first stellar generation ( g1 ) and a helium ( @xmath40 ) and cno enhanced second generation ( g2 ) is @xmath30 0.5 gyr while in the outer - halo gc s , g1 has been delayed by @xmath30 0.8 gyr and the time between g1 and g2 has been longer , @xmath30 1.4 gyr . they have demonstrated that for metallicities [ fe / h]@xmath41 , in outer gc s the is is mostly ( but not exclusively ) populated by g1 while the @xmath40- and cno - enhanced g2 rr lyrae stars must be more luminous ( see their figure 7 ) with an average @xmath42d , which agrees very well with the observed @xmath43d in m5 . for inner gc s of similar metallicity the mix of g1 and g2 in the is is likely more due to the shorter time between generations in which case @xmath44d . for lower metallicity outer gc s , [ fe / h]@xmath45 , the is is populated basically by g2 stars . we shall recall that for a @xmath40- and cno - enhanced g2 , the masses on the zahb are shifted towards lower temperatures ( jang et al . 2014 , their figure 3 ) , hence displacing the first overtone red - edge ( fre ) to the red ( bono caputo & marconi 1995 ) . more than one generation sharing the is will have different fre boundaries contributing to a mix of rrc and rrab stars . what effect the above scenario might have on the presence or not of the rrab - rrc splitting is not clear at present as several factors seem to play a role ; the overall metallicity , the presence of more than one generation in the is and the time elapsed between generations . we could add to this the possible presence of pre - zahb rr lyrae stars ( silva aguirre et al . 2008 ) . clearly more investigation of a larger sample of gc s is necessary to confront the observations with theoretical predictions . we speculate that in outer - halo ooii clusters the larger time difference between generations favours the existence of an rrab - rrc splitting as we have generally observed . the distance to m5 can be estimated from our data using different approaches ; firstly , from the weighted mean @xmath17 , calculated for the rrab and rrc from the fourier light curve decomposition ( table [ fisicos ] ) , which can be considered as independent estimates since they come from different empirical calibrations and zero points . secondly , we can use the @xmath14-band rr lyrae p - l relation derived by catelan , pritzl & smith ( 2004 ) . as a third approach we can use the three known sx phe stars and their p - l relation . and finally a fourth approach is via the bolometric magnitude for stars at the tip of the rgb . below we expand on these solutions . being m5 a nearby cluster , its reddening is small , we adopted @xmath46=0.03 mag ( harris 1996 ) . given the mean @xmath17 for rrl in table [ fisicos ] we found a true distance modulus of @xmath47 mag and @xmath48 mag using 38 rrab and 24 rrc stars respectively , which correspond to the distances @xmath49 and @xmath4 kpc . the quoted uncertainties are the standard deviations of the corresponding means . the distance to m5 listed in the catalogue of harris ( 1996 ) ( 2010 edition ) is 7.5 kpc in good agreement with our calculations . the @xmath14-band rr lyrae p - l relation derived by catelan , pritzl & smith ( 2004 ) is of the form : @xmath50 with @xmath51 -1.765 $ ] . we applied these equations to all 62 rrab and rrc stars in table [ fisicos ] . the periods for the rrc stars were fundamentalized following the period ratio @xmath52 in double mode stars ( catelan 2009 ) . we found an average distance of [email protected] kpc . an independent estimate of the distance can be obtained from the three sx phe known in the cluster . [ sxpl ] shows the p - l relationship of cohen & sarajedini ( 2012 ) ; @xmath53 positioned for the rrl mean distance . the first - overtone and second - overtone lines were positioned adopting the frequency ratios @xmath54= 0.783 and @xmath55 = 0.571 ( see santolamazza et al . 2001 or jeon et al . 2003 ; poretti et al . 2005 ) . it seems clear from the plot that v164 is a fundamental pulsator while v160 and v170 pulsate in the first - overtone . then , adopting eq . [ eqn : pl_cs ] we find an average true distance modulus of @xmath56 or a distance of @xmath5 kpc . the uncertainty was calculated from calibration errors neglecting the uncertainty in the period . yet another approach to the cluster distance determination is by using the tip of the rgb . this method was developed for estimating distances to nearby galaxies ( lee , freedman & madore , 1993 ) . one possibility is to use the specific calibration of the bolometric magnitude of the tip of the rgb , of m5 by viaux et al . ( 2013 ) under the arguments that the neutrino magnetic dipole moment enhances the plasma decay process , postpones helium ignition in low - mass stars , and therefore extends the red giant branch ( rgb ) in gcs . these authors came to the conclusion that in m5 @xmath57= -4.17 @xmath26 0.13 mag and we adopted that value . as in arellano ferro et al . ( 2015b ) for the case of ngc 6229 , we found that the method is extremely sensitive to the star selection . reasonable results are found for the stars nearest to the tip of the rgb , i.e. the reddest and brightest in this region of the cmd . we restricted our calculation to the two brightest rgb stars in our sample v50 and v174 . according to viaux et al . ( 2013 ) , the trgb is between 0.05 and 0.16 mag brighter than the brightest stars on the rgb . the fact that v50 and v174 may not be the brightest stars on the rgb implies that their magntiudes would have to be corrected to bring them to the trgb by at least the above quantities ; the larger the correction the smaller the resulting distance . we applied a correction between 0.05 and 0.16 mag to v50 and v174 and found a mean distance of between 7.5 and 7.2 kpc respectively . the above independent estimations of the cluster distance are , within their uncertainties , satisfactorily in agreement . however , given the sensitivity of the sx phe p - l and the trgb approaches to the number of stars involved as well as to the sample selection , we believe that the distance determinations from the rrl stars , which deal with much larger samples and carefully calibrated zero points of the luminosity scale are the best achieved . for the metallicity , the overall average value from the rrl is [ fe / h]@xmath58 or [ fe / h]@xmath59 found from the fourier decomposition of the light curves of 38 rrab and 24 rrc . these values can be compared with previous values in the literature : [ fe / h]@xmath60 ( harris 1996 ) ; [ fe / h]@xmath61 ( kaluzny et al . 2000 ) from an independent fourier decomposition analysis ; [ fe / h]@xmath62 ( zinn 1985 ) ; [ fe / h]@xmath63 ( carretta et al . 2009 ) the age of m5 has been discussed in numerous works . we have not attempted an independent estimation . the age of m5 was determined by jimenez & padoan ( 1998 ) via its luminosity function and they found an age of @xmath64 gyr assuming [ @xmath65/fe]=+0.4 . two more recent determinations of the age of m5 are by dotter et al . ( 2010 ) using relative ages from isochrone fitting , and by vandenberg et al . ( 2013 ) using an improved calibration of the vertical method or the magnitude difference between the turn off point and the hb ; these authors find [email protected] gyr and [email protected] gyr respectively . thus , in the cmd of fig . 3 we have overlayed the corresponding isochrones for 12.0 gyr from the victoria - regina evolutionary models ( vandenberg et al . 2014 ) and for the metallicities [ fe / h]=@xmath66 ( red ) and [ fe / h]=@xmath67 ( blue ) and [ @xmath65/fe]=+0.4 , shifted for the average apparent distance modulus @xmath68=14.395 mag found from the rrab and rrc stars , and @xmath69 . the isochrones for these two metallicities are very similar except perhaps at the tip of the rgb where the isochrone for the more metal - rich composition gets less luminous . however , if shifts are applied to these isochrones within the uncertainty of the distance modulus and reddening , @xmath70 mag , the two cases are indistinguishable . we can stress then that our cmd is consistent with the metallicity and distance derived in this paper and an age of @xmath71 gyr found in the above papers for m5 . the fourier decomposition of the light curves of stable rrl , and the calibrations and zero points available in the recent literature allowed us to determine the mean metallicity and distance to m5 as [ fe / h]@xmath72 and @xmath24 and @xmath1 kpc and @xmath73 kpc , from the rrab and rrc stars respectively . also individual values of radius and mass are provided . the employment of the @xmath74 rr lyrae p - l relation leads to a distance of [email protected] kpc . the distance to the cluster was also estimated from two independent methods ; the p - l relation of sx phe and the luminosity at the tip of the rgb finding [email protected] kpc and [email protected] kpc respectively . however we have shown that in the case of m5 the accuracy of these two alternative methods does not compete with the accuracy attained with the rr lyraes and the fourier light curve decomposition . a group of 16 evolved stars have been identified from their distribution in the amplitude - period plane . evolved stars should be moving to the red in the cmd , hence their periods should be increasing . from the period change study of szeidl et al . ( 2011 ) and our own period change analysis , which will be published elsewhere , we can see that some stars have increasing periods ( v11 , v16 , v39 , v45 , v77 , v87 and v90 ) and some have clearly decreasing periods ( v9 , v12 and v17 ) , which might be unexpected in truly evolved stars . however it is understood now that stochastic processes may produce both positive , negative or irregular period variation , e.g. mixing events in the core of a star at the hb ( sweigart & renzini 1979 ) or , as suggested more recently by silva aguirre et al . ( 2008 ) , period decreases can occur in pre - zahb rr lyrae stars on their final approach to the zahb . in the bottom panel of fig . [ cmd ] we have drawn a small line segment from the corresponding circles , to the right or left to indicate period increase or decrease respectively . the remainder of the evolved stars were not found to be undergoing secular period variations in our analysis . from their position on the cmd , their long period value and large period change rate , we conclude that v77 and v87 are truly advanced in their evolution towards the agb . for the most luminous stars in the is ( v107 , v122 and v158 ) we argue that they are not truly evolved stars but owe their overluminosity to a helium enhanced atmosphere or to the presence of an unseen companion . we note that in the hb instability strip of m5 , rrab stars share the either - or region with rrc stars , like in the case of the ooi cluster ngc 3201 , but in contrast with ngc 6229 , a cluster presumably nearly a twin of m5 , where the rrab - rrc splitting is noticeable . we also highlight the fact that the rrab - rrc splitting is a common feature in ooii clusters with a blue tail . our ccd @xmath75 time - series , spanning a little more than two years , allowed us to detect amplitude and phase modulations in 14 rrab and 9 rrc stars not previously reported as having the blazhko effect . these new findings account for incidence rates of at least 38% and 26% of the rrab and rrc respectively showing the blazhko effect in m5 . this comes as a natural result since in the last few years the detection of blazhko variables in gcs has been increasing , most likely due to the improvement in the quality of the ccd observations and reduction techniques . take for example the cluster ngc 5024 with the largest known population of blazhko variables with 66% and 37% of rrab and rrc respectively ( arellano ferro et al . 2012 ) . finally , it is well known from key studies in the past ( e.g. oosterhoff 1941 , szeidl et al . 2011 ) that a large fraction of the rrl population of m5 is undergoing secular period variations . our present data add up to 20 years to the time - base of available data for m5 and a re - analysis of the period change rates is deferred to a forthcoming paper . to the memory of janusz kaluzny whose numerous contributions to the field of variable stars in globular clusters have been a permanent source of inspiration . numerous suggestions and comments from an anonymous referee have been very valuable and are warmly acknowledged . it is a pleasure to thank no kains for his comments and suggestions . this publication was made possible by grant in106615 - 17 from the dgapa - unam ( mexico ) , the collaborative program conacyt - mincyt ( mexico - argentina ) # 188769 and by nprp grant # x-019 - 1 - 006 from the national research fund ( a member of qatar foundation ) . we have made an extensive use of the simbad and ads services , for which we are thankful . arellano ferro , a. , ahumada , j. a. , caldern , j.h . , kains , n. , 2014 , rmaa , 50 , 307 arellano ferro , a. ; bramich , d. m. ; figuera jaimes , r. ; giridhar , sunetra ; kuppuswamy , k. , 2012 , mnras , 420 , 1333 arellano ferro , a. , bramich , d. m. , giridhar , s. , luna , a. , muneer , s. , 2015a , ibvs 6137 arellano ferro , a. , mancera pia , p. e. , bramich , d. m. , giridhar , s. , ahumada , j.h . , kains , n. , k. kuppuswamy , 2015b , mnras , 452 , 727 arellano ferro , a. , bramich , d. m. , figuera jaimes , r. , et al . , 2013 , mnras , 434 , 1220 arellano ferro , a. , figuera jaimes , r. , giridhar , sunetra , bramich , d. m. , hernndez santisteban , j. v. , kuppuswamy , k. , 2011 , mnras , 416 , 2265 barnard , e.e . , 1898 , an , 147 , 243 bailey , s. i. , 1902 , apj , 10 , 255 bailey , s. i. , leland , e. f. , 1899 , apj , 10 , 255 bingham e. a. , cacciari c. , dickens r. f. , fusi pecci f. , 1984 , mnras , 209 , 765 bono , g. , castellani , v. , marconi , m. , 2000 , apj , 532 , l129 bono , g. , caputo , f. , marconi , m. , 1995 , aj , 110 , 2365 borissova , j. , catelan , m. , ferraro , f. r. , spassova , n. , buonanno , r. , iannicola , g. , richtler , t. , sweigart , a. v. , 1999 , a&a , 343 , 813 bramich d. m. , 2008 , mnras , 386 , l77 bramich , d. m. , bachelet , e. , alsubai , k. a. , mislis , d. , parley , n. , 2015 , a&a , 577 , a108 bramich d. m. , figuera jaimes r. , giridhar s. , arellano ferro a. , 2011 , mnras , 413 , 1275 bramich d. m. , freudling w. , 2012 , mnras , 424 , 1584 bramich d. m. , horne , k. , albrow , m. d. , and 8 coauthors , 2013 , mnras , 428 , 2275 brocato , e. , castellani , v. , ripepi , v. , 1996 , aj , 111 , 809 burke e.w . , rolland w.w . , , 1970 , jrasc , 64 , 353 cacciari , c. , corwin , t. m. , carney , b. w. , 2005 , aj , 129 , 267 caputo , f. , castellani , v. , marconi , m. , ripepi , v. , 1999 , mnras , 306 , 815 caputo f. , castellani v. , tornamb a. , 1978 , a&a , 67 , 107 carretta , e. , bragaglia , a. , gratton , r. , dorazi , v. , lucatello , s. , 2009 , a&a , 508 , 695 catelan m. , 2004 , in @xmath76 , asp conference series ; editors d.w . kurtz & k.r . pollard , 310 , 113 . catelan m. , 2009 , ap&ss , 320 , 261 catelan , m. , pritzl , b. j. , smith , h. a. , 2004 , apjs , 154 , 633 clement , c. m. , muzzin , a. , dufton , q. , ponnampalam , t. , wang , j. , burford , j. , richardson , a. , rosebery , t. , rowe , j. , hogg , h. s. , 2001 , aj , 122 , 2587 cohen , r. e. , sarajedini , a. , 2012 , mnras , 419 , 342 cudworth k.m . , 1997 , in : humphreys r.m . ( ed . ) proper motions and galactic astronomy . 127 , asp , san francisco , 91 cudworth k.m . , hanson , r. , 1993 , aj , 105 , 168 corwin , t. m. , carney , b. w. , 2001 , aj , 122 , 3138 dotter a. et al . , 2010 , apj , 708 , 698 draper p. w. , 2000 , in manset n. , veillet c. , crabtree d. , eds . 216 , astronomical data analysis software and systems ix . , san francisco , p. 615 drissen , l. , shara , m. m. , 1998 , aj , 115 , 725 dworetsky m. m. , 1983 , mnras , 203 , 917 evstigneeva , n. m. , shokin , yu . a. , samus , n. n. , tsevetkova , t. m. , 1995 , sov . , 21 , 451 harris , w. e. , 1996 , aj , 112 , 1487 honeycutt , r. k. , 1992 , pasp , 104 , 435 jang , s. , lee , y - w . , 2015 , apjs , 218 , 31 jang , s. , lee , y - w . , joo , s - j . , na , cch . , 2014 , mnras , 4433 , l15 jeon y .- b . , lee m. g. , kim s .- , lee h. , 2003 , aj , 125 , 3165 jimenez , r , padoan , p. , 1998 , apj , 498 , 704 jurcsik , j. , 1998 , a&a , 333 , 571 jurcsik , j. , kovcs g. , 1996 , a&a , 312 , 111 jurcsik , j. , szeidl , b. , clement , c. , hurta , zs . , lovas , m. , 2011 , mnras , 411 , 1763 kadla , z. i. , gerashchenko , a. n. , yablokova , n. v. , irkaev , b.n . , 1987 , ast . tsirk . , 1502 , 7 kains , n. et al . 2015 , a&a , 578 , a128 kaluzny , j. , thompson , i. b. , krzeminski , w. , pych , w. , 1999 , a&a , 350 , 469 kaluzny , j. , olech , a. thompson , i. , pych , w. , krzeminski , w. , schwarzenberg , a. , 2000 , a&as , 143 , 215 kovcs , g. , 1998 , mmsai 69 , 49 kovcs , g. , kanbur , s. m. , 1998 , mnras , 295 , 834 kovcs , g. , walker , a. r. , 2001 , a&a 371 , 579 kravtsov , v.v . , 1988 , 1526 , 6 kravtsov , v. v. , 1991 , sov . , 17 , 455 kunder a. , stetson p. b. , catelan m. , walker a. r. , amigo p. , 2013a , aj , 145 , 33 kunder a. et al . , 2013b , aj , 146 , 119 lee , m.g . , freedman , w. , madore , b.f . , 1993 , apj , 417 , 553 lenz p. , breger m. , 2005 , communications in asteroseismology , 146 , 53 martin , w. chr . , 1938 , leiden ann , 17b , 1 morgan s. , 2013 , in guzik j. a. , chaplinw . j. , handler g. , pigulski a. , eds , proc . 301 , precision asteroseismology . cambridge univ . press , cambridge , morgan , s. , wahl , j. n. , wieckhorts , r. m. , 2007 , mnras , 374 , 1421 olech , a. , wozniak , p. r. , alard , c. , kaluzny , j. , thompson , i. b. , 1999 , mnras , 310 , 759 oosterhoff , p. th . , 1939 , observatory , 62 , 104 oosterhoff , p. th . , 1941 , leiden ann . , 17 , part 4 padmanabhan , n. , schlegel , d.j . , finkbeiner , d. p. and 21 authors , 2008 , apj , 674 , 1217 pickering , e.c . , 1896a , a. n. , 139 , 137 pickering , e.c . , 1896b , a. n. , 140 , 285 poretti e. et al . , 2005 , a&a , 440 , 1097 pritzl b. j. , armandroff t. e. , jacoby g. h. , da costa g. s. , 2002 , aj , 124 , 1464 rees , r. f. , 1993 , aj , 106 , 1524 regnault n. , conley , a. , guy , j. , and 13 authors , 2009 , a&a , 506 , 999 reid , n. , 1996 , mnras , 278 , 367 sandquist , e. l. , bolte , m. , stetson , p. b. , hesser , j. e. , 1996 , apj , 470 , 910 santolamazza p. , marconi m. , bono g. , caputo f. , cassisi s. , gilliland r. l. , 2001 , apj , 554 , 1124 samus n.n . , durlevich o.v . , kazarovets e v. , kireeva n.n . , pastukhova e.n . , zharova a.v . , et al . , 2009 , general catalog ofvariable stars scholz , r .- d . , odenkirchen , m. , hirte , s. , irwin , m. j. , borngen , f. , ziener , r. , 1996 , mnras , 278 , 251 silbermann , n. a. , smith , h. a. , 1995 , aj , 109 , 1119 silva aguirre , v. , catelan , m. , weiss , a. , valcarce , a.a.r . , 2008 , a&a , 489 , 1201 stetson , p. b. , 2000 , pasp , 112 , 925 storm , j. , carney , b.w . , beck , j. a. , 1991 , pasp , 1033 , 1264 sweigart , a. v. , 1997 , apj , 474 , l23 sweigart , a.v . , renzini , a. , 1979 , a&a , 71 , 66 szeidl , b. , hurta , zs . , jurcsick , j. , clement , c. , lovas , m. , 2011 , mnras , 411 , 1744 van albada t. s. , baker n. , 1971 , apj , 169 , 311 van albada , t.s . , baker , n. , 1973 , apj , 185 , 447 vandenberg d. a. , brogaard k. , leaman r. , casagrande l. , 2013 , apj , 755 , 134 vandenberg , d. a. , bergbusch , p. a. , ferguson , j. w. , edvardsson , b. , 2014 , apj , 794 , 72 viaux , n. , catelan , m. , stetson , p. b. , raffelt , g. g. , redondo , j. , valcarce , a. a. r. ; weiss , a. , 2013 , a&a , 558 , a12 yan , l. , reid , n. , 1996 , mnras , 279 , 751 zacharias , n. , finch , c. t. , girard , t. m. , henden , a. , bartlett , j. l. , monet , d. g. , zacharias , m. i. , 2013 , aj , 145 , 44 zinn , r. , 1985 , apj 293 , 424 zinn , r. , west , m. j. , 1984 , apjs , 55 , 45 in this section we comment on the light curves , variable types and nature of some interesting or peculiar variables in table [ variables ] and figs . [ varsaba ] and [ varsc ] . we put some emphasis on the amplitude and phase modulations of the blazhko type in specific stars . in all the stars labelled @xmath77 in table [ variables ] the amplitude variations are neatly distinguished in the light curves in figs . [ varsaba ] and [ varsc ] in both the @xmath13 and @xmath14 filters . * v25 , v36 , v53 , v74 , v102 , v108 , v140 and v159*. all these stars were found by arellano ferro et al . ( 2015a ) to be misidentified in the literature . in that paper the identifications have been discussed and corrected and a detailed identification chart is given . * v14*. in the cvsgc ( 2014 update ) it is noted as an unconfirmed variable , probably after evstigneeva et al . ( 1995 ) . however the star is clearly variable in the study of kaluzny et al . ( 2000 ) and in the present work . it displays some amplitude modulation already noted by kaluzny et al . ( 2000 ) . * v18*. it displays a tremendous amplitude variation from 0.617 mag in 2012 to 1.218 mag in 2013 and 2014 . the observations on january 23 , 2013 are already consistent with the large amplitude , meaning that the star underwent the amplitude change between may 2012 and january 2013 . the photographic light curve from 1934 from oosterhoff ( 1941 ) , although a bit scattered , shows an amplitude of 0.76 mag and while not directly comparable with our @xmath13 or @xmath14 light curve amplitudes we mention that oosterhoff himsef noted considerable light curve variations and suggested to classifiy it as an irregular variable . the data from 1997 of kaluzny et al . ( 2000 ) show an amplitude of 1.27 mag and a mild suggestion of the blazhko effect . the repetitivity of our light curves from february to may in 2012 at the low amplitude and again during 2013 - 2014 at the large amplitude , suggests that the star remains with a constant amplitude before it goes through the amplitude variation episodes : it is therefore a very strong blazhko modulator with a rather long period . according to szeidl et al . ( 2011 ) the blazhko period is longer than 500d . * v27*. despite its relative isolation it shows a very peculiar light curve . we have not been able to identify more than one period . thus the observed amplitude and phase modulations must be due to the blazhko effect . we adopted the period of szeidl et al . ( 2011 ) that phases the light curve best . they did not find secular period changes for this star . the blazhko modulations are very prominent . * v28*. similar to v18 , this star shows a very pronounced amplitude modulation from 1.120 mag to 0.653 mag with much different time distributions than v18 . the amplitudes in the light curves of oosterhoff ( 1941 ) and kaluzny et al . ( 2000 ) are in between the above two extremes . * v42 , v84*. these two w virginis ( cw ) stars are shown in the cmd of fig . [ cmd ] and their light curves are displayed in fig . the data are included in the electronic table [ tab : vi_phot ] . their periods are listed in table [ variables ] . for v42 the period 25.735d given in the cvsgc phases our data correctly . for v84 the period 53.95d in the cvsgc does not phase our data properly but we found that about half of it , 26.49d , produces a nice light curve . * v77*. this is the rrab with the largest period in m5 ( 0.845158d ) . it has a low amplitude and a peculiar shape of the light curve with a roundish maximum . in the amplitude - period diagram ( fig . [ bailey ] ) the star falls in the extreme right of the diagram , much beyond the evolved stars sequence . in the cmd ( bottom of fig . [ cmd ] ) it is one of the brightest stars . the star is undergoing a secular period increase with a rate among the largest in m5 ( szeidl et al . 2011 ) . all these properties are consistent with the star being in an advanced evolutionary state towards the agb . * v78*. this star has the shortest period among the known rrl stars in m5 ( 0.264820d ) . kaluzny et al . ( 2000 ) have argued that it may be a second overtone pulsator ( rre ) and noted the slight asymmetry of the light curve . we point out that while the asymmetry is there , it appears as prominent as in the other rrc stars with longer periods ( e.g. v15 , v55 , v60 etc . ) . in fact , an inspection of the light curves of the rrc stars in fig . [ varsc ] reveals that a slight asymmetry is present in most of them , so this seems to be more the rule than the exception in this cluster . however , following catelan s ( 2004 ) practice , we have checked the log p - @xmath78 plane and confirmed that , after fundamentalizing the period of v78 assuming that it is a second - overtone pulsator , the star falls in the natural extension to short period of the rrab distribution . this seem to support the idea that v78 is a rre . * v104*. it has been noticed by several authors that the light curve of this peculiar star shows very prominent amplitude and phase modulations . it has been suggested to be a double mode star ( reid 1996 ) and a semi - detached binary ( drissen & shara 1998 ) . its rather long period for an rrc star and its position among rrab stars on the log p - @xmath78 plane make it more likely to be an rrab star . we used period04 to search for traces of double mode activity but found no significant secondary frequencies . our light curve is very peculiar indeed and shows multiple rr lyrae - like curves , all well phased with a period of 0.486748 , with phase and mean brightness displacements . it is probably an rrab in a binary system , unfortunatelly we could only estimate five times of maximum on our data and two in the light curve of drissen & shara ( 1998 ) and thus , with the handful of times of maximum the corresponding o - c diagram does not shed clear signs of duplicity . * v121*. this star is close to the center of the cluster but it is not blended with another star of similar magnitude or brighter . despite this , it shows large modulations in phase and amplitude probably due to the presence of blazhko effect . the light curves in caputo el al . ( 1999 ) and drissen & shara ( 1998 ) look stable but their time - base are very short . * v127*. this star is very close to another of similar brightness . it is identified in the discovering paper ( kravtsov 1988 ) and we confirm that the variable is the nw of the pair ( see fig . [ chart ] ) . in the paper by olech ( 1999 ) its light curve appears as that of a small amplitude rrc - like with a period 0.544965d . however our period 0.540366d produces a clear rrab light curve with amplitude apparently diminished due to flux contamination by the neighbour . on the bailey diagram ( fig . [ bailey ] ) the star clearly sits among the rrab stars . it also shows amplitude modulations . * v156*. the variability of this star and a nice looking rrab light curve were reported by drissen & shara ( 1998 ) from their high resolution images of the hubble space telescope ( hst ) , and they labelled it as v15 . the star identified by these authors is strongly blended in our images and we were not able to either calculate a period at confidence or recover a rrl - like light curve , hence it was not included in fig . [ varsaba ] . in fig . [ v156 ] we show the nightly light variations and note the changes of the maximum brightness , inviting the suggestion of blazhko modulations . caputo et al . ( 1999 ) offered an alternative identification and named the star v156 , however the light curve of that star does not show variations in our collection . * v157*. the light curve of this star has an anomalously low amplitude , likely due to blending in our images with a close bright star . its position in the cmd ( fig . [ cmd ] ) and bailey diagram ( fig . [ bailey ] ) is therefore peculiar . * v158*. it is noted in the cvsgc as `` rrc ? '' indeed its period 0.442627d is rather long for an rrc , however , its light curve is a little asymmetric and resembles that of the other rrc stars in the cluster . in the bailey s diagram ( fig . [ bailey ] ) , the star falls halfway between the rrc and the rrab distributions . we have labelled it as rrc . we have identified a few close pairs of variable stars in our images that , depending on the seeing conditions may or may not have contaminated each other s light curves . we list them below and briefly discuss them . * v109-v142*. these two rrab stars are separated by only 4.0 pixels or 1.2 arcsec in our images and hence the interference of each other s light curves which is evident as noise ( see table [ tab : observations ] ) . the light curve of v142 shows very prominent phase and amplitude modulations . the variability was discovered by brocato , castellani & ripepi ( 1996 ) but the light curve was not made available . partial light curves were displayed by caputo et al . ( 1999 ) and drissen & shara ( 1998 ) where the variations are clear but no modulation was detected , most likely due to the limited time - base of these data sets . * v112-v113*. the separation of these stars is 7.4 pixels or 2.2 arcsec which means that in nights of poorer seeing their light curves are contaminated by the light of the neighbour producing noisier light curves . * v130-v131*. these stars are separated by 8.4 pixels or 2.5 arcsec , i.e. ocassionally , on nights of poorer seeing there is mutual light contamination . scatter in the light curve of v130 was noted by kaluzny et al . ( 2000 ) but their search for a secondary frequency was futile . our light curve shows clear amplitude modulations which could be attributed to the blazhko effect . v131 is not included by kaluzny et al . both stars included by caputo et al . ( 1999 ) show no amplitude modulations but their light curves are sparse . in our data v131 also shows mild amplitude modulations . it is possible that the observed amplitude modulations are due , at least partially , to the light contamination from the neighbour . . a systematic and thorough analysis of 50 rrab stars in m5 was conducted by jurcsik et al . ( 2011 ) in search of blazhko - type variability and its characterization , using a collection of photometry of m5 covering nearly 100 years . they identified 20 blazhko stars among the rrab population and calculated the modulation periods ( their table 1 ) . the modulation periods range mostly between 40 and 600 days , with the exception of v72 for which the modulation is about 1200 days . the time - base of our data is 770 days ( see table [ tab : observations ] ) . this means that based only on our observations we should be able to detect the modulations in all those variables , provided that the modulations are strong enough and the photometry is of good quality . of the 20 stars found by jurcsik et al . ( 2011 ) four ( v2 , v29 , v58 and v72 ) are not in the fov of our images . the remaining 16 stars are labelled @xmath79 in table [ variables ] , however v5 , v30 and v38 , look rather stable in our light curves , thus we would have been unable to identify these as blazhko variables based exclusively on our data . we find it difficult to judge the real blazhko nature of these three stars . we note first that the blazhko modulations reported by jurcsik et al . ( 2011 ) are in all cases very mild and that for v5 and v38 the blazhko classification relies on a comparison between the data from reid ( 1996 ) , whose light curves are a bit scattered , and the data from kaluzny et al . ( 1999 ; 2000 ) separated by 5 years ; both stars are in crowded fields and kaluzny et al . 2000 has pointed out that neighbours at 5 - 10 arcsec affect some of the photometric data . for the more isolated star v30 the comparison was between the data of storm et al . ( 1991 ) and kaluzny et al . ( 1999 ; 2000 ) separated by 10 years , reduced and transformed to the standard system by different approaches . while these details may contribute to apparent amplitude differences , it is also possible that our data time - base is not long enough for the detection of a long - period modulation . for the other 13 stars the modulations are also evident in our data . besides the above mentioned stars , we have also been able to detect amplitude - phase modulations in the light curves of another 14 rrab stars , some of which are subtle but visible , hence probably undetectable in old photographic data . these 14 newly identified blazhko rrab stars are labelled @xmath77 in table [ variables ] . this makes a total of 30 blazhko stars in m5 in a sample of 79 rrab stars , i.e. an occurrence rate of at least 38% . . no blazkho rrc variables were reported by jurcsik et al . however , the modulations are clearly visible in 9 stars of the 34 in our sample ; v35 , v40 , v44 , v53 , v55 , v99 , v120 , v130 and v131 . this makes an incidence rate of at least 26% for rrc blazhko variables in m5 . . the presence of bumps near the maximum light in rrc stars is clear in v31 , v35 , v57 , v60 , v62 , v95 , v100 , v105 , v108 , v113 , v116 , v130 and v133 . bumps were noted before by kaluzny et al . ( 2000 ) in v31 , v35 , v57 and v62 and highlighted that bumps do not occur for stars with periods above 0.28 - 0.30 d , which is true also for our extended list of `` bumpy '' stars with the exception of v108 and v130 with [email protected] , although we may concede that the bump in these cases is rather mild . kaluzny et al . ( 2000 ) venture the possibility that the bump is associated with some sort of interaction between the first and second overtones . the bump was actually reproduced in the rrc star u com by hydrodynamical models constraining the metallicity and calibrating the adopted turbulent convection model . the bump was found to be sensitive to the parameters governing the coupling between convection and pulsation ( bono , castellani & marconi 2000 ) .

_ the ejecta of the first probably very massive stars polluted the big bang primordial element mix with the first heavier elements . the resulting ultra metal - poor abundance distribution provided the initial conditions for the second stars of a wide range of initial masses reaching down to intermediate and low masses . the importance of these second stars for understanding the origin of the elements in the early universe are manifold . while the massive first stars have long vanished the second stars are still around and currently observed . they are the carriers of the information about the first stars , but they are also capable of nuclear production themselves . for example , in order to use ultra or extremely metal - poor stars as a probe for the r - process in the early universe a reliable model of the s - process in the second stars is needed . eventually , the second stars may provide us with important clues on questions ranging from structure formation to how the stars actually make the elements , not only in the early but also in the present universe . in particular the c - rich extremely metal - poor stars , most of which show the signature , are thought to be associated with chemical yields from the evolved giant phase of intermediate mass stars . models of such agb stars at extremely low metallicity now exist , and comparison with observation show important discrepancies , for example with regard to the synthesis of nitrogen . this may hint at burning and mixing aspects of extremely metal - poor evolved stars that are not yet included in the standard picture of evolution , as for example the hydrogen - ingestion flash . the second stars of intermediate mass may have also played an important role in the formation of heavy elements that form through slow neutron capture reaction chains ( s - process ) . comparison of models with observations reveal which aspects of the physics input and assumptions need to be improved . the s - process is a particularly useful diagnostic tool for probing the physical processes that are responsible for the creation of elements in stars , like for example rotation . as new observational techniques and strategies continue to penetrate the field , for example the multi - object spectroscopy , or the future spectroscopic surveys , the extremely metal - poor stars will play an increasingly important role to address some of the most fundamental and challenging , current questions of astronomy . _ * the second stars * + falk herwig + los alamos national laboratory + los alamos , theoretical astrophysics group , nm 87544 , usa + [email protected]

0708.3848

one of the most intriguing questions of astrophysics and astronomy is the origin of the elements in the early universe , and how this relates to the first formation of structure . about 200,000 years after the big bang the epoch of structure formation emerged and the first stars were born from the initial , pristine baryonic matter . without any elements heavier than helium to provide cooling , the first stars that formed from the baryonic matter trapped in the emerging mini - dark matter halos were probably very massive , greater than 30@xmath0 ( e.g. * ? ? ? these massive stars burned through their available fuel in about one to two million years , exploded as supernovae , and dispersed the first elements heavier than helium into the nascent universe , or collapsed into black holes . these first events of stellar evolution influenced their early universe neighborhood , and determined under which conditions and with which initial abundance low - mass stars with masses like the sun eventually formed . these take about 100 to 1000 times longer to form than more massive stars . thus , by the time the first low - mass stars formed there was already a small amount of heavier elements present . such low - mass stars are really the second stars , stars which formed from a non - primordial abundance distribution . these second stars are important because while the massive first stars have long since vanished , some second stars are still around . their importance lies in their capacity as carriers of the information about the formation and evolution of those long gone first generations of stars . the second stars are also important because of the nuclear production they are capable of by themselves ( for example the first s - process in the second stars ) . observations of such second stars are now emerging in increasing quantity and with high - resolution spectroscopic abundance determination @xcite . there are a number of strategies to find ultra and extremely metal - poor stars . the most effective appears to be spectroscopic surveys . most of the emp and ump stars have been discovered in the hk survey @xcite and the hamburg / eso survey ( he s * ? ? ? * ) . here , i designate somewhat arbitrarily ultra metal - poor ( ump ) stars as those with @xmath1 } } } \leq -3.5 $ ] ( @xmath2 ) and extremely metal - poor ( emp ) stars as those with @xmath1 } } } \leq -1.8 $ ] ( @xmath3 ) . emp stars include the most metal - poor globular clusters . ump stars are rare and hard to find in the spectroscopic surveys . only approximately a dozen have been reported so far , with the present record holder he 0107 - 5240 at @xmath1 } } } = -5.4 $ ] @xcite . at and below this metallicity spectral features are so weak that even the 10 m class telescopes are barely suitable for the job . the formation mode of the second stars depends on a complex history of events , each of which may leave signatures in the observed abundances of emp and in particular the ump stars . these events include the ionizing radiation of the very first primordial stars and the expansion of the associated h ii regions , their effect on possibly primordial star formation in neighboring mini - dark matter halos , as well as possibly the metal enrichment from a pair - instability supernova , depending on the mass of the first primordial stars . these processes and the cosmological assumption that fix the underlying dark - matter structure will eventually determine the abundance distribution of the emp and ump stars now discovered and analyzed . the most recent studies confirm that the very first primordial stars are massive @xcite . however in these calculations the effect of the ionization from star formation in nearby mini - dark matter halos is taken into account , and it is found that the following generation of primordial stars may include intermediate mass stars . the most metal - poor stars are potentially an extremely powerful tool to study star formation and evolution in the early universe @xcite . these stars with metallicities reaching in excess of 5 orders of magnitude in metallicity below the solar metallicity often show abundance distributions that are very different from the solar abundance distribution . the solar abundance distribution is the result of a long galactic chemical evolution , reflecting the cumulative result of mixing and nuclear processing of many generations of stars of a wide range of masses . the most metal - poor stars are very old and have formed very early , at the onset of galactic chemical evolution . it is therefore reasonable to assume that the abundance pattern in these second stars are the result of only few different sources and corresponding stellar nuclear production sites . the goal is to identify these nuclear production sites and the mechanisms and processes through which the corresponding stellar ejecta have been brought together in the star formation cloud of the second stars . an important part of the problem is to account for self - pollution or external pollution events that may have altered the stellar abundances between the time of second star formation and observation . as the number of emp and ump stars with detailed abundance continuously grows one can discern certain abundance patterns . a very unusual sub - group are the carbon - rich emp and ump stars . observations indicate that many of these stars are polluted by asymptotic giant branch ( agb ) binary companions . this implies that reliable models of agb stars including their heavy element production through the s - process at extremely low metallicity are required to understand these stars , and to disentangle the agb contribution from other sources that have contributed to the observed abundance patterns . in the following section the carbon - rich emp and ump stars will be discussed in more detail . in [ sec : improve ] current models of emp intermediate mass stars and their uncertainties are presented . such models have important applications for the emerging field of near - field cosmology , which uses galactic and extra - galactic stellar abundances to reconstruct the processes that lead to the formation of our and other galaxies ( [ sec : nfc ] ) . then , ongoing work both observational and theoretical to determine the nucleosynthesis of nitrogen at extremely low metallicity is described ( [ sec : nitrogen ] ) and [ sec : hingba ] covers the hydrogen - ingestion flash in born - again giants , because this type of events may be important for emp agb stars as well , and may indeed be a major source of nitrogen in these stars . [ sec : spr ] deals with the , and how it can be used as a diagnostic tool to improve the physics understanding of stellar mixing . finally , and before some forward lookgin concluding remarks ( [ sec : concl ] ) the importance of nuclear physics input for the question of c - star formation will be discussed in [ sec : enrt ] . about @xmath4 of all emp and ump stars show conspicuous enrichment of the cno elements , most notably c. for example , cs 29497 - 030 @xcite has @xmath1 } } } = -2.9 $ ] , @xmath5 } } } = 2.4 $ ] , @xmath6 } } } = 1.9 $ ] , @xmath7 } } } = 1.7 $ ] , and @xmath8 } } } = 0.5 $ ] . others , like lp 625 - 44 ( @xmath1 } } } = -2.7 $ ] * ? ? ? * ) or cs 22942 - 019 ( @xmath1 } } } = -2.7 $ ] * ? ? ? * ) , have a similar overabundance of c , but smaller n ( @xmath6 } } } = 1.0 $ ] for lp 625 - 44 ) and larger na ( @xmath8 } } } = 1.8 $ ] for lp 625 - 44 ) . many of the heavier elements , like ti , cr , mn or zn in cs 29497 - 030 are either not overabundant or rather somewhat underabundant compared to the solar distribution . large overabundances of n and na are not predicted by standard models of massive stars , neither for pop iii @xcite nor at larger metallicity @xcite . primary production of n requires to expose the he - burning product c to h - burning again . such a condition is encountered in massive agb stars during hot - bottom burning @xcite . na is made in c - burning in massive stars but its final abundance is sensitive to the neutron excess , and thus scales less than linear with metallicity . if primary @xmath9 is exposed to he - burning it will lead to the production of @xmath10 . either an additional proton collected in hot - bottom burning , or a neutron from the @xmath11 reaction in the he - shell flash of agb stars can then lead to primary @xmath12 production . this process leads to significant na overabundances in standard agb stellar models of very low or zero metallicity @xcite . the overabundances of c , n and na are not the only indication that the abundances in c - rich emp stars are at least partly caused by agb stellar evolution . all of the objects mentioned above , and in fact most c - rich emp stars show in some cases significant overabundances of the s - process elements . in addition they reveal their binarity through radial velocity variations , corresponding to binary periods of a few days in the case of he 0024 - 2523 @xcite to more typically one to a dozen years . in fact , in their statistical analysis @xcite find that observations of radial velocities of cemp - s stars ( c - rich emp stars with s - process signature ) obtained to date is consistent with a @xmath13 binarity rate of cemp - s stars . he 0107 - 5420 is currently the intrinsically most metal - poor star with @xmath1 } } } = -5.3 $ ] . however , the total metal abundance in terms of elements heavier than he is quite large . in particular the cno elements are significantly overabundant compared to the solar abundance ratios , with @xmath5 } } } = + 4.0 $ ] , @xmath6 } } } = + 2.4 $ ] @xcite , and @xmath7 } } } = + 2.3 $ ] @xcite . the large overabundance alone indicate that only a few nuclear sources were involved in creating this abundance pattern . together with the low fe abundance the large cno abundances require a primary production of the cno elements . a primary nucleosynthesis production chain in a star is based only on the primordial elements h and he . a secondary production in contrast requires heavier elements already to be present in the initial element mix at the time of star formation . for example , standard models of massive stars predict a secondary production of n from cno cycling of c and o that is initially present from earlier generations of stars . he 0107 - 5420 is a good example to discuss the possible nuclear production sites that may be important at the lowest metallicities . certainly the production of the first massive pop iii stars plays an important role . @xcite have discussed the observed abundance pattern within the framework of mixing and fallback of the ejecta during the explosion of supernovae that eventlually form a black hole . they show that the mixing and fallback parameters can be chosen in a way that accounts for much of the observed pattern . for example , their model correctly reflects the large observed fe / ni ratio , and the large c / fe and c / mg ratios . the model does predict a significant n overabundance but it quantitatively falls short of matching the observed value . their model predicts the charactereistic odd - even effect that reflects the small ( or missing ) neutron - excess of the primordial nuclear fuel . one of the observed species that hints that he 0107 - 5420 may not have such a pronounced odd - even pattern is na . the observerved overabundance is @xmath8 } } } = 0.8 $ ] , more than 1.5 dex above the model prediction from the mixing and fallback supernova . unfortunately , for the heavy trans - iron elements only upper limits could be derived so far , which prohibits definite access to the s- and r - process abundances in he 0107 - 5420 . at this point na and to a lesser extent n are the major indicators that some additional sources are part of the nuclear production site inventory that caused the abundance pattern in this most metal - poor star . without better guidance from ab - initio coupled structure and star formation simulations such as shown in fig.[fig : dmhalo ] several scenarios have to be considered . material from the first supernova could be ejected from the corresponding halo and injected into a neighbouring halo . this metal enrichment could lead to the formation of an intermediate mass star , for example a very massive agb star , maybe even a super - agb star that eventually ends in a onemg core - collapse supernova . as shown in [ sec : improve ] a contribution from such stars can account for both na as well as n , but more detailed models of this particular scenario need to be done for quantitative comparison . a low - mass star like he 0107 - 5420 could then form from the combined nuclear production of these two sources , a roughly @xmath14 black hole forming supernova and a massive agb or super - agb star . alternatively the low - mass star forms from the ejecta of just one supernova , and this initial abundance pattern is modified by self - polution . he 0107 - 5420 is in fact a giant on the first ascent . observations , in particular of globular cluster red giants show that mixing and nuclear processing of c into n below the deep convective envelope lead to secondary n production @xcite . this could explain the quantitative discrepancy between the observed n abundance and the model by @xcite . however , the rather large carbon isotopic ratio @xmath15 implies that self - polution can not account for the observed na abundance , in particular assuming the absence of @xmath10which is a secondary nucleosynthesis product in massive stars . this leads to the external polution scenario that is frequently considered in this context . as mentioned above many cemp stars are in fact in binary systems , and the abundance signatures seen in these c - rich very metal - poor stars could be in part the result of mass transfer from the agb progenitor of a white dwarf companion . for he 0107 - 5420 @xcite discuss this possibility in detail . for the evolution of primordial agb stars they consider the possibility of the h - ingestion flash , a process in which the he - shell flash convection zone reaches outward into the h - rich envelope , resulting in a peculiar nucleosynthesis and mixing regime . such events are known at solar metallicity to cause a significant fraction of young white dwarfs to evolve for a short period of time back to the agb ( born - again evolution , [ sec : hingba ] ) , and the most prominent representative of this class of objects is sakurai s object @xcite . as discussed in the previous section the evolution of low- and intermediate mass stars at zero or extremely or ultra metal - poor abundance is essential for studying the most metal - poor stars and their cosmological origin and environment . this section deals first with the evolution of agb stars that may have formed from big bang material . a possibly important evolutionary phase - the h - ingestion flash triggered by the he - shell flash - has been observed in models of pop iii and even in models of non - zero but ultra - low metal content ( [ sec : hflash ] ) . finally the evolution and yields of extremely metal - poor tp - agb stars is described . the evolution of pop iii massive stars is discussed , for example , by @xcite . simulations show that many of the overall properties of pop iii intermediate mass stellar models @xcite probably apply to ultra metal - poor stars as well . at @xmath16 the initial thermal pulses may show peculiar convective mixing events at the core - envelope boundary during or after the actual he - shell flash . but eventually the models enter a phase of rather normal thermal pulse agb evolution , with regular thermal pulse cycles , third dredge - up , hot - bottom burning and mass loss . many uncertainties of the pop iii stellar evolution is related to the possibility of flash - burning that is not yet well understood ( [ sec : hflash ] ) . @xmath16 stellar evolution is different from evolution at higher metal content because cno catalytic material for h - burning is initially absent . burning hydrogen via the pp - chain requires a higher temperature than h - burning via the cno cycle . eventually the triple-@xmath17 process will provide some c , and a mass fraction as low as @xmath18 is sufficient to switch h - core burning to cno cycling . during he - core burning some c is produced outside the convective core and a tail of carbon reaches from the convective core out toward the h - shell . after the end of he - core burning , and depending on the initial mass , this carbon and some nitrogen is mixed into the envelope . according to @xcite this second dredge - up raises the envelope c - abundance above @xmath19 for @xmath20 . for such a c - abundance the h - shell is fully supported by cno cycling . as a result the thermal pulse agb evolution is like that of the ultra metal - poor cases . models with initial masses below @xmath21 show a peculiarity that is unknown in more metal - rich models . after an initial series of weak thermal pulses the h - shell forms a convection zone when it re - ignites after the he - shell flash @xcite . the lower boundary of this convection zone is highly unstable , because it coincides with the opacity discontinuity that marks the core - envelope interface . even small amounts of mixing will drive flash - like burning and deep mixing , as protons enter a @xmath22-rich zone . it seems that the details of the evolution of the h - shell convection zone is not important as this one - time event leads to deep dredge - up and enrichment of the envelope with a sufficient amount to support regular cno cylce burning in the h - shell thereafter . accordingly , these lower - mass cases will evolve like extremely metal - poor stars too , and the @xmath16 models discussed here can be used as proxies for what is in reality born as an ultra metal - poor star . the effect of rotationally induced mixing processes may alter the evolution of low- and intermediate - mass stars and very low metallicity even before the the agb and lead to a qualitatively different thermal pulse evolution . @xcite describe models for metallicity @xmath23 in which @xmath22 and @xmath24 is mixed out of the he - burning core and up to the location of the h - shell . this leads to an increase of cno material , in particular @xmath9 in the envelope that could result in a rather normal thermal pulse evolution without any peculiar hydrogen convection zone . whether the rotationally induced mixing processes before the agb or the subsequent thermal pulse agb evolution with the associated dredge - up events dominate the final cno yields is not yet clear . the difference between a real pop iii and an extremely metal - poor thermal pulse agb star is the initial absence of fe and other elements heavier than al in the pop iii star . this does not preclude that a pop iii agb stars can not generate -elements by the slow neutron capture process . @xcite have used the -framework established for solar - like metallicities ( [ sec : spr ] ) and ran network calculations with an initial abundance appropriate for the interior of a @xmath16 tp - agb star . they used the models of @xcite as stellar evolution input . the major uncertainty was the mixing for the formation of the neutron donor species @xmath25 , and the unknown feedback of such mixing to the thermodynamic structure evolution . @xcite and @xcite have shown that this feedback is likely very important . nevertheless , the -models for @xmath16 are instructive , because they show that if a sufficient neutron source is available , the can be based on seed nuclei with lower mass number than fe , for example c. c is produced in a primary mode in tp - agb stars . for such a @xmath16 the fe / ni ratio should be markedly sub - solar , approximately close to the quasi steady - state value of @xmath26 , that is given by the ratio of the neutron cross sections of ni and fe ( the isotopic minimum for each element ) . he 0107 - 5240 with @xmath1}}}=-5.3 $ ] @xcite shows a larger than solar fe / ni ratio , which is evidence for the contribution of supernovae to the abundace distribution of this star . the peculiar h - convective episode described in the previous section was observed in agb models with of @xmath27 and @xmath16 by @xcite . during the following thermal pulse the he - shell flash convection zone reachea out into the h - rich envelope . this leads to the h - ingestion flash ( hif ) . protons are mixed on the convective time scale into the he - shell flash region which is with depth increasingly hot . the hydrogen flash - burning leads to a separate convection zone as shown in fig.[fig : ff08348 ] . convective h - burning in this layer is characterized by a large abundance of @xmath22 and protons . the protons have been mixed into this region from the envelope . accordingly the @xmath9 abundance in this layer is very large , up to @xmath28 by mass in a layer of @xmath29 . the models show that after both the he - shell flash and the hif have subsided a deep dredge - up episode will mix this @xmath9-rich layer into the envelope . this may account for a substantial @xmath9 production and part of the observational pattern observed in emp stars that are believed to be poluted by emp agb stars ( [ sec : nitrogen ] ) . a hif was also found in the @xmath30 , @xmath16 model , but @xcite did not find the hif in models of @xmath31 . agb stars are highly non - linear systems . the occurrence of the hif depends on many details of the models , including the physics of mixing , opacities , the numerical resolution and nuclear reaction rates . small changes in any of these ingredients , that may go unnoticed during a more robust stellar evolution phase ( like the main - sequence ) , can be the deciding factor here whether a hif occurs or not . currently , theoretical models do not agree if and in which initial mass and metallicity regime the hif occurs @xcite . however , it is important to better understand this situation because the hif may play an important role in the in extremely metal - poor stars @xcite . hydrogen ingested into the he - shell flash or pulse - driven convection zone ( pdcz ) may lead to @xmath25 production which could release neutrons . @xcite propose that the abundance pattern of the currently most metal - poor star he 0107 - 5420 is in part due to the nucleosynthesis in a thermal pulse agb star with hif . the details of the nucleosynthesis in a hif depend sensitively on the physics of simultaneous rapid nuclear burning and convective mixing . the hydrogen - ingestion flash can also be a source of lithium , distinctly different from conditions during hot - bottom burning @xcite . the evolution of born - again stars , like sakurai s object , provide valuable additional constraints with regard to the hif in agb stars ( [ sec : hingba ] ) . the very late thermal pulse is associated with a hif @xcite . by reproducing the evolution of stars like sakurai s object one can gain some confidence in models of the hif in metal - poor agb stars . @xcite showed that models in which the convective hydrogen ingestion occurs with lower velocities than predicted by the mixing - length theory @xcite feature a faster born - again evolution than mlt models , quantitatively in agreement with observations . the physical interpretation is that rapid nuclear burning on the convective time scale releases energy and adds buoyancy to down - flowing convective bubbles , leading to additional breaking of the plumes . the temperature and time scales for nucleosynthesis during the hif would be significantly different . this has not yet been applied to hif models at @xmath16 or extremely low metallicity . by definition all yields of pop iii stars are primary , because the initial composition contains only big bang material . nuclear production is secondary when nuclei heavier than h and he are already present in the initial abundance distribution and transformed in other species . pop iii or emp agb stars can produce a large number of species in primary mode , including the cno elements , ne , na , mg and al . the production of heavier elements depends on the availabilitity of neutrons . some of these primary species , like oxygen , are usually not considered a product of agb evolution at moderate metal - deficiency or solar metallicity . however , the initial envelope mass fraction of oxygen for models with @xmath32 is @xmath33 . @xmath24 in the intershell material ( cf . fig.[fig : kipp ] ) that is dredged - up is primary and even at this low metallicity of the order @xmath34 by mass . dredge - up of such material will enhance the envelope abundance significantly . a set of low- and intermediate mass agb stellar evolution models with @xmath32 ( [ fe / h ] = -2.3 ) with detailed structure , nucleosynthesis and yield predictions has been presented by @xcite . the oxygen overabundance in these models is @xmath35 , depending on mass . the abundance evolution for c , n , o and na is shown in fig.[fig : nacno ] for two initial masses . the evolution of c and n is qualitatively different in the two cases . the lower mass sequence shows the increase of c and some o due to the repeated dredge - up events . no significant amount of @xmath9 is dredged - up . the result is a large c / n ratio . the @xmath27 track shows the opposite behaviour . the much lower c / n ratio is the result of hot - bottom burning @xcite . c and o are dredged - up after the thermal pulse as in the lower mass case , but during the following quiescent intperpulse phase the envelope convection reaches into the region hot enough for h - burning and the envelope c is transformed into n. the initial - mass transition between hot - bottom burning models with large n abdundances and c dredge - up models without hot - bottom burning is very sharp . this precludes the notion that emp stars with simultaneously large n and c overabundances , like cs 29497 - 030 with @xmath5}}}=2.4 $ ] and @xmath6}}}=1.9 $ ] @xcite , may be poluted by such agb stars in the transition regime where hot - bottom is only partially efficent ( see [ sec : nitrogen ] ) . the signature of hot - bottom burning in fig.[fig : nacno ] is a gradual change of abundance between the steps caused by dredge - up . from this it can be seen that na has a different nuclear production site depending on initial mass . in the @xmath30 case na is produced in the he - shell flash , by n - captures on @xmath10 . the neutrons come from the @xmath36 reaction . @xmath12 is the dredged - up but not produced during the interpulse phase . the neutron - heavy mg isotope @xmath37 is produced in a similar way . in the @xmath27 case na is mainly produced in hot - bottom burning , by p - capture on dredged - up @xmath10 . in the previous section it was argued that due to the large primary production of cno and many other species the @xmath16 model predictions would also apply approximately to stars with ultra - low metallicity as well . neutron capture reactions play an important role , and need to be included in yield - predictions , together with a neutron - sink approximation for species not explicitely included in the network . this has been done in a qualitatively similar way by both @xcite for the @xmath32 models and @xcite for their @xmath16 models , and the abundance evolution is quantitatively similar . both model sets predict for non - hot - bottom burning cases large primary production of c and o , while the @xmath9 abundance does not change . @xmath10 is produced in both cases in the he - shell flashes , and in both cases the final @xmath10 exceeds that of @xmath9 . the evolution of @xmath12 , @xmath38 and @xmath37 is qualitatively the same too . in particular the ratio of these three isotopes is quantitatively the same in the last computed model of the @xmath16 sequence and the final agb model at @xmath32 . this is in particular interesting as these low - mass agb models predict that @xmath39 , contrary to the signature of hot - bottom burning . one of the largest uncertainties is the adopted mass loss . while mass loss in agb stars of solar metallicity can be constraint observationally , this has not been possible for extremely low metallicity . other uncertainties relate to mixing in emp agb stars . @xcite has shown that convection induced extra - mixing , like exponential overshoot that is used to generate a @xmath25 pocket for the in higher - metallicity models , may lead to vigorous h - burning during the third dredge - up . depending on the efficiency of such mixing the third dredge - up may turn into a flame - like burning front , leading to very deep core penetration . this may significantly impact the formation and effectiveness of a @xmath25 pocket for the @xcite . accurate stellar yields for intermediate mass stars are requested by another emerging field , near - field cosmology @xcite . the baryon ( stellar ) halo of the milky way retains a fossil imprint of the merging history of the galaxy . different merging components , like the infalling dwarf galaxies with a range of masses , have different star formation histories that translate into different abundance signatures of the member stars of these components . the current surge of spectroscopic multi - object capabilities at large telescopes will likely significantly enhance the importance of this approach . an example is the recent work by @xcite in which abundances of halo stars are compared with abundances of stars in satellites , dwarf galaxies trapped in the potential well of the galaxy . here , the basic idea is that dwarfs are merging with a galaxy at different times at which point star formation and chemical evolution stops . satellites that survive until the present day should show the signature of more evolved chemical evolution , for example including the elements associated with the long - lived low - mass stars . in contrast , halo stars that are the dispersed members of dwarf galaxies that have merged at an earlier time in the evolution of the galaxy , have less evolved chemical evolution patterns , for example showing more clearly the patterns of nucleosynthesis in massive star evolution . potentially , the implications that can be derived from comparison of halo stars and satellite members may include some fundamental questions of nucleosynthesis itself . the data presented in @xcite show that @xmath40}}}$ ] is systematically smaller in stars belonging to satellites compared to their halo counterparts . this can be explained in the @xmath41 cold dark matter model that predicts that most halo stars have formed in rather massive dwarf galaxies , which merged with the galaxy a long time ago @xcite . these stars show the signature of truncated chemical evolution , dominated by the yields of supernovae type ii and their significant @xmath17-element contribution and moderate fe ejecta . stars in present , less massive satellite galaxies show the signature of chemical evolution components that take more time , like the sn ia . these events add fe but little @xmath17-elements , and their @xmath40}}}$ ] ratio is therefore smaller . abundances of other elements could be compared too . for example , @xcite compare among others the abundances of y , ba and eu in halo stars and satellite dwarf galaxies . ba is typically considered an main - component element ( at least at @xmath1 } } } > -2 $ ] ) because the elemental solar abundance has a @xmath42 contribution @xcite . only the smaller remaining fraction is made in the r - process . as discussed above , the nuclear production site of the main - component of the s - process are low- and intermediate mass agb stars , in the initial mass range @xmath43 . this component of galaxy chemical evolution needs even more time to contribute than the sn ia . and indeed , [ ba / fe ] behaves differently from @xmath40}}}$ ] in halo stars and satellite stars . [ ba / fe ] is on average higher in the satellite stars than in halo stars , with a considerable spread . this is consistent with the framework of truncated chemical evolution of systems that were early disrupted in merger events , and with the understanding of stellar evolution and nucleosynthesis that @xmath17-elements are predominanty made in short - lived massive stars , while ba originates in rather old populations . it is then extremely interesting to consider the r - process element eu , which in the solar abundance distribution has a very small contribution of only @xmath44 . this element does not behave like @xmath17-elements . instead , [ eu / fe ] is on average the same in halo stars and dwarf galaxy stars , however with a larger spread in the latter . in order to remain in the proposed scenario of why abundances in the halo differ from those in the nearby dwarf galaxies one would then have to assume that at least an important fraction of the r - process elements does not originate in sn ii , which eject their yields on a short time scale . instead , one would have to assume that eu in the satellite stars comes from a source that releases the ejecta on a time scale comparable to or longer than the fe production in sn ia . such sources could be the accretion induced collapse @xcite or the collapse of a super - agb star in the initial mass range @xmath45 , if the core mass grows to the chandrasekhar limit @xcite . it would certainly go to far at this point to draw any further conclusion , for example in conjunction with the two r - porcess source model proposed by @xcite . it is clear , that the full potential of using the abundances patterns of stars to reconstruct the formation of galaxies in their cosmological context can only be reached with detailed yield predictions including all masses and reaching down to extremely low metallicity . y is another example that reinforces this notion . in the solar abundance distribution it has a @xmath46 fraction from the main - component originating in low - mass stars . however , it is also the termination point of the weak from massive stars , and it does have some r - porcess contribution . in the data presented by @xcite it behaves with regard to the difference between halo and satellite stars like the @xmath17-elements , which in this context would imply an important contribution from massive stars . nitrogen is the fifth most common element in the universe . in stellar evolution of very low metallicity the primary production is of particular importance . strictly from a nucleosynthesis point of view nitrogen is always made in a well established sequence of events , involving first the triple-@xmath17 process that makes @xmath22 and then two subsequent proton captures that are part of the cn - cylce . in a stellar model sequence the problem is to identify the mixing processes that can account for this sequence of events . how can @xmath22 , that is made in the he - burning that requires higher temperatures , be brought into layers of lower temperature where h - burning still takes place . or , how can protons be mixed down into the he - burning layers and be captured by @xmath22 to eventually form @xmath9 , but then not be exposed to @xmath17-captures that would convert @xmath9 into @xmath10 . because the primary production of nitrogen depends so much on stellar mixing , this element assumes a key role in understanding stellar nucleosynthesis at extremely- and ultra - low metallicity . nitrogen is certainly present in the most metal - poor stars in the milky way @xcite . according to standard stellar evolution and nucleosynthesis models primary n comes from intermediate mass ( i m ) agb stars @xcite , with perhaps a small contribution from rotating massive stars @xcite . in agb stars the repeating sequence of thermal pulses induce convective mixing events that lead to the prodution of n. c is made in the he - shell flash . dredge - up after the flash mixes that c into the envelope , and in massive agb stars the c in the envelope is then transformed into n. the low [ n/@xmath17 ] abundances observed in some damped lyman-@xmath17 systems @xcite have prompted suggestions that the imf may be biased in favor of massive stars in some systems @xcite , or that these systems represent the earliest stages of chemical evolution of massive stars only @xcite . observations of n and c abundances in c - rich emp stars with -signature provide another way to study the n production at low metallicity . about 30@xmath48 of all extremely metal - poor stars ( [ fe / h ] < -2 ) are strongly c- and to a lesser extent n - enhanced ( [ sec : cemp ] ) . in particular the overabundance of c and of n in addition to their s - process signature has lead to the assumption that the agb star progenitor of the current white dwarf companion to the cemp - s star is responsible for the observed abundance pattern . because of the established binary nature and the -signature the cemp - s stars are assumed to be poluted by the individual agb stars that are the progenitors of their present white dwarfs companions . in fig.[fig : nitrogen ] the [ c / n ] and [ c / fe ] ratios of literature data are shown , together with the agb model prediction of @xmath49 - @xmath21 tracks . the literature data show systematically lower [ c / n ] ratios than what is expected by @xmath49 or @xmath50 models that dredge - up c , but do not produce n. this poses the question of the primary origin of the n in theses stars . models of initial masses between 4 and 6 feature efficient hot - bottom burning ( hbb ) which turns most dredged - up primary carbon into nitrogen resulting in [ c / n]<-1 and [ c / fe]<1.5 . however , none of the literature data seem to show the very low [ c / n ] ratio that would be expected for an emp star that happened to have an agb companion in the @xmath51 initial mass range . we have carried out observations of emp targets with 0.5<[c / fe]<1 using the ch and the nh band for abundance determination @xcite . specifically we wanted to find out whether the paucity of emp stars with [ c / n]<-1 is a selection effect or a systematic observational bias imposed by the abundance indicators employed in previous studies . while c - strong stars can be identified by the g - band at @xmath52 , the strong nh band at @xmath53is never observed in the medium - resolution surveys that provide the targets for detailed abundance studies ( e.g. * ? ? ? only the cn bands at @xmath54and @xmath55are included , and cn lines are not strong in n - rich stars unless c is also enhanced by a large amount . therefore , studies of c and n abundances have favored c - rich stars that are easily identified , and may have missed stars that have @xmath5}}}\sim 0 - 1 $ ] , but are more rich in n. thus the literature data in fig.[fig : nitrogen ] , drawn from high - resolution follow - up of the medium - resolution candidates , may be the result of a strong observational bias against finding n - rich stars . in order to overcome this bias we obtained spectra for 18 new stars . in fig.[fig : nitrogen ] , we plot the preliminary estimates of the [ c / n ] ratios of 18 stars , based on observations of the nh bands during 6 nights at ctio / kpno . we did not find any stars with low [ c / n ] , and the analysis of additional data is underway to put our findings on a more robust statistical basis . this leaves us with two important open questions : ( 1 ) where are the emp stars polluted by massive agb stars ? and ( 2 ) where does the n in the cemp stars come from that we do observe ? while we do not have any idea at this point of what the answer to the first question could be , there are a number of possibilities to address the absence of primary n. these include rotationally induced mixing before the agb phase ( [ sec : z00 ] ) , the h - ingestion flash ( [ sec : hflash ] ) , and extra - mixing in agb stars @xcite as envoked in models of rgb stars to account for abundance anomalies in globular cluster member stars @xcite . in any of these cases the interpretation of n abundances in metal - poor systems of all kinds may have to be re - evaluated . nuclear production in the agb stellar interior can become observable at a later time in the evolution , during the post - agb phase , on the surface of those central stars of planetary nebulae that become h - deficient . here h - deficient means in most cases that the mass fraction of h as observed in the stellar spectrum is less than @xmath56 . this leads to additional important constraints on the evolution of agb stars , which is in particular important to study agb evolution at extremely low metallicitiy . h - deficient central stars of planetary nebulae ( cspn ) as well as very young white dwarfs have been analysed in detail ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) and new important information about the abundances of these stars is yet to be revealed . stellar evolution predicts that about @xmath57 of all post - agb stars will eventually loose all their small remaining h - rich envelope mass of the order @xmath58 and expose the bare h - free cores . the carbon , oxygen and helium rich nature of these objects is evident from their wolf - rayet or pg1159 type emission line spectra , and reflects nuclear processed material that has been built up during the stars progenitor evolution . although there is considerable spread in the observed abundances of these stars , the pattern can be summarized in mass fraction as he @xmath59 c @xmath60 and @xmath61 . these h - deficient cspn are important for agb evolution modeling , because they provide a unique opportunity to study directly the nuclear processing shells in agb stars . in order to explain the evolutionary origin @xcite introduced the born - again evolution scenario . the star evolves off the agb , and becomes a hydrogen - rich central star and eventually a very hot , young white dwarf . however , the he - shell may still be capable to ignite a late he - shell flash , and in that case the star retraces its evolution in the hrd , back into the giant region ( fig.[fig : hrdsak ] ) . as a result of hydrogen - ingestion into the he - convection and rapid burning , or by mixing from the emerging convective envelope ( or because of both ) the surface abundance of such born - again stars will be extremely h - deficient or even h - free . stellar models of this evolutionary origin scenario connect the surface abundance of the wolf - rayet central stars and the pg1159 stars with the intershell abundance of the progenitor agb star . the intershell layer between the he- and the h - shell is well mixed during each he - shell flash . this zone contains the main nuclear production site of the progenitor agb star , and the abundance of this zone reflects contributions from both he- and h - shell burning ( fig.[fig : kipp ] ) . as a result of the born - again evolution these layers become visible at the surface of the resulting h - deficient cspn . the initial models of the h - deficient central stars by iben and collaborators showed qualitatively that the born - again scenario could account for high he and c abundances , as these elements were abundant in the intershell of the agb progenitor model they used . however , they could not account for the high observed oxygen abundance . @xcite were the first to propose that the solution could be non - standard mixing during the agb evolution . models with overshooting at the bottom of the pulse - driven convection zone do not only feature higher temperatures at the bottom of that layer ( [ sec : spr - mix ] ) , but they also show higher c and in particular o abundances in the pdcz . subsequently , @xcite has explored in detail how the various abundances depend on the overshoot efficiency and other details . in essence , overshooting brings agb intershell abundances of he , c and o in very good quantitative agreement with the observed abundances of wolf - rayet type central stars and pg1159 stars , in the framework of the born - again evolution . more effort is needed to consolidate these constraints on the intershell abundance with the possibly tight upper limits on overshooting at the bottom of the pdcz that the may provide ( [ sec : spr - mix ] ) . the connection of the surface abundances of the hot pg1159 stars and the progenitor agb intershell abundance has been reinforced recently by several new observational findings . these include a substantial overabundance of ne @xcite , f abundances ranging from solar up to 250 times solar @xcite in good agreement with agb nucleosynthesis predictions by @xcite , and fe - deficiencies of at least 1 dex compared to solar @xcite , which may reflect the depeletion of fe in the agb intershell due to n - captures . in that case the fe / ni ratio should be low , as in fact observed in sakurai s object . this born - again star is the smoking gun of the very late thermal pulse scenario and links this evolutionary scenario with the h - deficient pg 1159 and wolf - rayet type central stars @xcite . much of the observed abundance pattern of sakurai s object as well as the rapid evolutionary time scale has been reproduced quantitatively with stellar models @xcite . similar to the pg 1159 stars , sakurai s object shows fe depletion of about 1dex compared to solar . the low ratio of @xmath62 , shows that sakurai s object shows surface material that has been directly irradiated with neutrons @xcite . new radio observations show that sakurai s object has started reheating again , on its second evolution into the cspn @xcite . this marks a new phase of the evolution of this star , and confirms the concept of convective mixing efficiency modified by nuclear burning @xcite . the is the origin of half of all elements heavier than iron @xcite . it is also important for isotopic ratios and in some cases elemental abundances of lighter elements - in particular at extremely low metallicity . the heavy elements are made by the through neutron captures that are slow compared to the competing @xmath63-decay . starting from the abundant iron group elements it follows closely the valley of stability in the chart of isotopes . it is characterized by neutron densities @xmath64 . observations and theory agree that the nuclear production site of the main and strong component of the ( @xmath65 ) originates in low mass asymptotic giant branch ( agb ) stars . the weak component below the first peak at @xmath66 is produced during he and c burning in massive stars . here we deal only with the in agb stars . the is important in the context of the second stars for two reasons . first , the observed elemental abundance distribution of the trans - iron elements is at any given metallicitiy a mix of s- and r - process contribution . in order to disentangle the two contributions in very - low metallicity stars , the in this metallicity regime needs to be understood . this may than enable progress in identifying the conditions for the r - process . second , -branchings can be used to probe the physics for stellar mixing ( [ sec : spr - diag ] ) , which is particularly usefull when modeling the evolution of emp stars . the in low - mass agb stars has two neutron sources . the main source is @xmath25 , that forms via the @xmath67 reaction . at the end of the dredge - up phase ( fig.[fig : kipp ] ) after the he - shell flash , when the bottom of the h - rich convective envelope has penetrated into the @xmath22 rich intershell layer , partial mixing at this interface would create a thin layer providing simultaneously the required protons and @xmath22 . the in the @xmath25 pocket is characterized by low neutron densities ( @xmath68 ) that last for several thousand years under radiative , convectively stable conditions during the quiescent interpulse phase . the physics of mixing at the h/@xmath22 interface at the end of the third dredge - up phase has not yet been clearly identified ( see below ) . most likely it is some type of convection induced mixing beyond the convection boundary . the @xmath11 reaction requires the high temperatures that can be found at the bottom of the pulse - driven convection zone ( pdcz ) during the he - shell flash ( @xmath69 ) . the neutrons are released with high density ( @xmath70 ) in a short burst @xcite . these peak neutron densities are realised for only about a year , followed by a neutron density tail that lasts a few years , depending on the stellar model assumptions . the current quantitative modeling of the uses the thermodynamic output from a stellar evolution calculation including mass loss as input for nucleosynthesis calculations with a complete network @xcite . the post - processing accounts for both the @xmath25 neutron source as well as for the @xmath10 source , and mixes the different contributions according to the information provided by the stellar evolution calculations . the free parameter of the model is the @xmath25 abundance in the @xmath25 pocket that is proportional to the neutron exposure that results from burning the @xmath25 in the @xmath71 reaction . physical mixing processes which are responsible for bringing protons down from the envelope into the @xmath22-rich core to enable @xmath25 formation , are not explicitely included in this model . observationally , models have to account for observed spread in observables that are related to the neutron exposure ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? this spread is not only evident from stellar spectroscopy , but also from sic grain data , that indicate that a spread by a factor of five is necessary for the neutron exposure for a given mass and metallicity @xcite . currently , this spread is accounted for by a range of different cases in each of which the @xmath25 abundance in the pocket is assumed to be different @xcite . however , the physics of the mixing that is associated with the range of neutron exposures has yet to be identified . it is useful to distinguish -mixing for solar and moderately metal - poor stellar evolution , and the extremely and ultra metal - poor cases . in the first , the assumption that -mixing does not feedback strongly into the thermodynamic evolution is generally valid , in the latter it is generally not . many more observational constraints exist for the solar and moderately metal - poor , in particular the isotopic information from the pre - solar sic grains . mixing for the for the solar and moderately metal - poor has to satisfy two general constraints : ( 1 ) how is the partial mixing zone of h and @xmath22 generated that eventually forms the neutron source species @xmath25 , and ( 2 ) what is the origin of the observed spread in neutron exposures . mixing for the @xmath25 pocket is probably related to the penetrative evolution of the bottom of the convective envelope during the third dredge - up . possible mechanisms include exponential diffusive overshooting @xcite , mixing induced by rotation @xcite , and mixing by internal gravity waves @xcite . each of these effectively leads to a continuously and quickly decreasing mixing efficiency from the h - rich convection zone into the radiative @xmath22-rich layer , and each of these will lead to the formation of two pockets which are overlapping @xcite . at low h/@xmath22 ratios h - burning is proton - limited , and protons will make @xmath25 via the @xmath72 reaction but no ( or little ) @xmath9 . at larger h/@xmath22 ratio a @xmath9pocket forms the maximum abundance of both @xmath25 and @xmath9 depends on the @xmath22 abundance in the intershell . therefore , the conditions in the @xmath25 pocket are not independent of , for example , the mixing at the bottom of the pdcz . @xcite derive the relationship @xmath73 , where @xmath74 is the intershell @xmath22 mass fraction and @xmath75 is the maximum neutron exposure reached in the @xmath25pocket . the observations of h - deficient post - agb stars described in [ sec : hingba ] require that the @xmath22 abundance in the pdcz of agb stars is about @xmath76 . this value is reproduced in agb stellar evolution models with overshooting at the bottom of the pdcz , and about twice as large as in models without overshooting . apart from the maximum amount of @xmath25 in the pocket the mixing process must produce a partial mixing layer of the right mass @xmath77 which eventually contains enriched material . in the exponential overshooting model of @xcite the mixing coefficient is written as @xmath78 , where @xmath79 is the mixing - length theory mixing coefficient at the base of the convection zone , @xmath80 is the geometric distance to the convective boundary , @xmath81 is the pressure scale height at the convective boundary , and @xmath82 is the overshooting parameter . if applied to core convection @xmath83 reproduces the observed width of the main sequence . @xcite finds that @xmath84 at the bottom of the convective envelope generates a large enough @xmath25 pocket . however , the maximum neutron exposure in the @xmath25 pocket of the overshooting model is @xmath85 , while in the non - overshooting models this value is @xmath86 . correspondingly @xcite obtained only negative values for the logarithmic ratio [ hs / ls ] , while the overshooting model with the larger neutron exposure predicts [ hs / ls]@xmath87 . this compares to an observed range of @xmath88}}}<0.0 $ ] for stars of solar metallicity ( see * ? ? ? * for a compilation of observational data ) . models with overshooting at all convective boundaries can reproduce only the largest observed hs / ls ratios , indicating that the neutron exposure in the @xmath25 pocket in these models is at the maximum of the observationally bounded range . overshooting alone is not able to account for all features of -mixing . in particular there is no mechanism to acount for the spread in neutron exposures within the overshooting framework . rotation , however , may induce a range of mixing efficiencies for a sample of stars with otherwise identical parameters . models of rotating agb stars were presented by @xcite , and the was analyzed in detail by @xcite and @xcite . the implementation of rotation for the agb models was the same as the one that had been used previously to construct rotating models of massive stars @xcite . this implementation yields a @xmath25 pocket generated by shear mixing below the envelope convection base , however an order of magnitude smaller than what is needed in the partial mixing zone of a non - rotating model to reproduce the observed overproduction in stars . the second important finding is that shear mixing which initially generates the small @xmath25 pocket , prevails throughout the interpulse phase , even when the base of the convection is receding in mass after h - shell burning has resumed . when the dredge - up ends , the low - density slowly rotating convective envelope and the fast rotating compact radiative core are in contact and mixing is induced through shear at this location of large differential rotation . this radial velocity gradient remains as a source for shear mixing at exactly the mass coordinate of the @xmath25 pocket with important consequence for the . shear mixing during the interpulse phase swamps the @xmath25 pocket with @xmath9 from the pocket just above . by the time the temperature has reached about @xmath89 and @xmath25 starts to release neutrons via @xmath90 , @xmath9 is in fact more abundant than @xmath25 in all layers of the @xmath25 pocket . @xmath9 is a very efficient neutron poison . it has a very large @xmath91 rate and simply steels neutrons from the iron seed . as a result , the neutron exposure is only @xmath92 , about a factor of ten too small to generate the observed abundance distribution . the detailed post - processing models of current rotating agb stars showed that they are not capable to account for the observed overabundances . parametric models show that weaker shear mixing during the interpulse leads to a weaker poisoning effect . for very small poisoning effects the neutron exposures are still large enough to reproduce some observations . while neither rotation or overshooting alone provide the right amount of -mixing it is interesting to consider a combination of these . in essence the idea is that overshooting would provide mixing for the formation for the @xmath25 pocket and for a larger @xmath22abundance in the pdcz to obtain a large neutron exposure in the @xmath25pocket . then , shear mixing during the interpulse could add some poison and result in the observed spread . magnetic fields have not yet been considered in agb stellar evolution . however , one may assume that qualitatively magnetic fields will add coupling between the fast rotating core and the slowly rotating envelope and provide additional angular momentum transport . models including this effect could result in smaller shear mixing than predict by the current rotating agb stellar models . it is interesting to note that such an effect of magnetic fields may also help to reconcile the predicted rotation rates of agb cores of @xmath93 with the rotation rate determinations of white dwarfs . spectroscopic determinations of rotation rates of white dwarfs of spectral type da can not rule out such values , but most are also consistent with zero or very low rotation @xcite . however , asterioseismological measurements of wd rotation rates clearly yield smaller values in the range @xmath94 ( see references in * ? ? ? * ) . the high precision information on the pre - solar meteoritic sic grains provide isotope ratio measurements that allow to probe the conditions at the nuclear production site in more detail . one example is the highly temperature dependent nucleosynthesis triggered by the release of neutrons from @xmath10 at the bottom of the pdcz . the temperature at the bottom of the pdcz correlates with the efficiency of extra mixing like overshooting at the bottom of this convection zone @xcite . neutrons in the pdcz are generated by the @xmath11 reaction . for larger temperatures the neutron density is higher . isotopic ratios that enclose a branch point isotope in the path will be more neutron heavy for higher neutron densities . an example is the @xmath95/@xmath96-ratio , that is set by the branching at the radioactive @xmath97 ( fig.[fig : zr ] ) . for low neutron densities @xmath97 decays . for @xmath98 @xmath99 becomes significant , hence @xmath95 is produced . if the temperature is larger the neutron density is larger and the @xmath95/@xmath96 ratio , which can be measured in sic grains , is larger as well . @xcite have studied the measured isotopic ratios of mo and zr as well as sr and ba from sic grains . they evaluate the sensitivity of their results in terms of nuclear reaction rate uncertainties . all branchings that are activated by the @xmath10 neutron source depend on the still uncertain @xmath11 rate . in addition , the @xmath95/@xmath96 ratio depends on the neutron cross - section of the unstable isotope @xmath97 . like for almost all radioactive nuclei the @xmath100 rate of @xmath97 is not measured . the theoretical estimates vary from maxwellian - averaged cross sections of @xmath101 @xcite to @xmath102 ( jendl-3.2 ) . in order to make full use of the potent method of using the as a diagnostic tool it is critical that n - cpature rates of the radioactive -branch point nuclides are measured . different -branchings are sensititv to different mixing processes , including those that may be induced by rotation . the as a diagnostic tool can provide information on mixing processes that are potentially relevant for the evolution of stars of all masses , including the progenitors of supernovae . in particular the details of the initial model for a supernova calculations determines important properties of the explosion , like asymmetries , or the final fate as black hole or neutron star @xcite . in order to understand c - rich extremely metal - poor ( cemp ) stars the formation of c - rich agb stars has to be understood in a quantitative way . agb stars become c - rich because of the third dredge - up , which mixes c - rich material from the intershell into the envelope ( fig.[fig : kipp ] ) . the evolution of c and o in agb stars and the problems related to modelling the third dredge - up are discussed in @xcite . in summary , the third dredge - up is now obtained in 1d stellar evolution calculations in sufficient amount and at sufficient low core mass . these calculations take into account convection - induced mixing into the stable layers in a time- and depth dependent way , and use high numerical resolution . uncertainty is introduced by the mixing length parameter @xcite . understanding the dredge - up properties of agb stars is important , because the dredge - up dependnent yield predictions for low and intermediate mass stars enter models for galaxy chemical evolution . agb stars serve as diagnostics for extragalactic populations , and for this purpose the conditions of the o - rich to c - rich transitions needs to be known . c - rich giants are the brightest infrared population in extra - galactic systems . finally , the envelope enrichment of agb stars with the s - process elements is intimately related to the dredge - up properties of the models . recently , it has been shown that uncertainties in nuclear reaction rates propagate in a significant way into the dredge - up and thereby yield predictions . @xcite calculated an extensive grid of @xmath30 , @xmath103 tracks for combinations of rates for the @xmath104 , the triple-@xmath17 and the @xmath105 rate within the errors given in the nacre compilation @xcite . the main result was that dredge - up and c yields are larger for lower @xmath106 rate and for larger triple-@xmath17 rate . the @xmath107 rate plays a less important role . it was also found that nuclear physics work since the 1999 nacre compilation require a downward revision of the @xmath108 rate , by almost a factor of 2 . fig.[fig : enrt ] shows the revised range for this rate and the yields from the calculations assuming the uncertainty range of the nuclear reaction rates . extremely metal - poor stars are an emerging field of astrophysical and astronomical research , pushing the limits of observations , and theory and numerical simulation . potentially , much can be learned about challenging questions of astrophysics . how do galaxies like our milky way form ? how did the first stars and their cosmological environment form and evolve ? some exciting clues about the evolution of stars come from the smallest astrophysical bodies , pre - solar stardust extracted from primitive meteorites . thus stellar evolution and nucleosynthesis connects nearby phenomena of planetary system formation with star formation and evolution in the earliest time of the universe as recorded in the element distribution patterns of the most metal - poor stars . in the future the quantity and quality of spectroscopic data of stars , in particular the valuable most metal - poor stars , will increase dramatically , due to multi - object spectroscopy and large spectroscopic surveys . it is already becoming clear that this data can be put to full use only with qualitatively and quantitatively improved simulations of nuclear production in stars , including low and intermediate mass and massive stars . i would like to thank d. schoenberner for nominating me for the ludwig - biermann award , and for directing my interest to stellar evolution and nucleosynthesis as my _ doktorvater_. i am very grateful to n. langer and d. vandenberg for their continuing support . i would also like to thank my colleagues at the theoretical astrophysics group ( t-6 ) at los alamos national laboratory , in particular c. fryer , a. heger , and f. timmes , as well as r.reifarth at lansce-3 , who contribute to a very stimulating atmosphere . finally , i would like to thank a. font and b. oshea for stimulating discussions , that have contributed to some views expressed in this article . this work was funded under the auspices of the u.s.dept . of energy , and supported by its contract w-7405-eng-36 to los alamos national laboratory . , w. , ando , h. , honda , s. , iye , m. , izumiura , h. , kajino , t. , kambe , e. , kawanomonoto , s. , noguchi , k. , okita , k. , sadakane , k. , sato , b. , shelton , i. , takada - hidai , m. , takeda , y. , watanabe , e. , & yoshida , m. 2002 , pasj , 54 , 427 , t. , bonifacio , p. , molaro , p. , cayrel , r. , spite , m. , spite , f. , plez , b. , andersen , j. , barbuy , b. , beers , t. c. , depagne , e. , hill , v. , franois , p. , nordstr " om , b. , & primas , f. 2004 , a&a , 413 , 1073

recent results on classical and quantum strings in a variety of black hole and cosmological spacetimes , in various dimensions , are presented . the curved backgrounds under consideration include the black hole anti de sitter spacetime and its dual , the black string , the ordinary black holes with or without a cosmological constant , the de sitter and anti de sitter spacetimes and static robertson - walker spacetimes . exact solutions to the string equations of motion and constraints , representing circular strings , stationary open strings and dynamical straight strings , are obtained in these backgrounds and their physical properties ( length , energy , pressure ) are described . the existence of _ multi - string _ solutions , describing finitely or infinitely many strings , is shown to be a general feature of spacetimes with a positive or negative cosmological constant . generic approximative solutions are obtained using the string perturbation series approach , and the question of the stability of the solutions is addressed . furthermore , using a canonical quantization procedure , we find the string mass spectrum in de sitter and anti de sitter spacetimes . new features as compared to the string spectrum in flat minkowski spacetime appear , for instance the _ fine - structure effect _ at all levels beyond the graviton in both de sitter and anti de sitter spacetimes , and the _ non - existence _ of a hagedorn temperature in anti de sitter spacetime . we discuss the physical implications of these results . finally , we consider the effect of spatial curvature on the string dynamics in robertson - walker spacetimes .

hep-th9504007

the classical and quantum propagation of strings in curved spacetimes has attracted a great deal of interest in recent years . the main complication , as compared to the case of flat minkowski spacetime , is related to the non - linearity of the equations of motion . it makes it possible to obtain the complete analytic solution only in a very few special cases like conical spacetime @xcite and plane - wave / shock - wave backgrounds @xcite . there are however also very general results concerning integrability and solvability for maximally symmetric spacetimes @xcite and gauged wzw models @xcite . these are the exceptional cases ; generally the string equations of motion in curved spacetimes are not integrable and even if they are , it is usually an extremely difficult task to actually separate the equations , integrate them and finally write down the complete solution in closed form . fortunately , there are several different ways to `` attack '' a system of coupled non - linear partial differential equations . the systematic study of string dynamics in curved spacetimes and its associated physical phenomena was started in refs.@xcite . besides numerical methods , which will not be discussed here , approximative [ 7 - 10 ] and exact [ 11 - 14 ] methods for solving the string equations of motion and constraints in curved spacetimes , have been developed . classical and quantum string dynamics have been investigated in black hole backgrounds [ 15 - 18 ] , cosmological spacetimes [ 7 , 11 - 14 , 17 - 21 ] , cosmic string spacetime @xcite , gravitational wave backgrounds @xcite , supergravity backgrounds ( which are necessary for fermionic strings ) @xcite , and near spacetime singularities @xcite . physical phenomena like the hawking - unruh effect in string theory @xcite , horizon string stretching @xcite , particle transmutation @xcite , string scattering @xcite , mass spectrum and critical dimension @xcite , string instability [ 7 , 11 - 13 , 16 , 21 ] and multi - string solutions [ 11 - 13 , 28 ] have been found . in a generic @xmath2-dimensional curved spacetime with metric @xmath3 and coordinates @xmath4 the string equations of motion and constraints are : @xmath5 @xmath6 where dot and prime stand for derivative with respect to the world - sheet coordinates @xmath7 and @xmath8 respectively and @xmath9 are the christoffel symbols with respect to the metric @xmath10 in the following we present recent results [ 26 - 29 ] on solutions to eqs.(1.1)-(1.2 ) in a variety of curved spacetimes from cosmology , gravitation and string theory . we obtain explicit ( exact and/or approximate ) mathematical solutions , discuss the corresponding physical properties , we quantize the solutions in different ways ( canonical quantization , semi - classical quantization ) and find the physical content : the mass spectrum . the presentation is organized as follows : in section 2 , we discuss the classical string dynamics in the @xmath0 black hole anti de sitter ( bh - ads ) spacetime , recently found by baados et al @xcite . we compare with the string dynamics in ordinary cosmological and black hole spacetimes . this clarifies the geometry ( as seen by a string ) of the bh - ads spacetime . in section 3 , generalizing results from section 2 , we derive the quantum string mass spectrum in ordinary @xmath2-dimensional anti de sitter spacetime . we discuss , in particular , the sectors of low and very high mass states . new physical phenomena arise like the fine - structure effect at all levels beyond the graviton and the non - existence of a hagedorn critical temperature . the results are compared with corresponding results obtained in minkowski and de sitter spacetimes . sections 4 and 5 are devoted to the investigation of the more general underlying structure of solutions to eqs.(1.1)-(1.2 ) . in section 4 , we find new classes of exact string and multi - string solutions in cosmological and black hole spacetimes , while in section 5 , we consider the effects of a non - zero spatial curvature on the classical and quantum string dynamics in friedmann - robertson - walker ( frw ) universes . anti de sitter ( ads ) spacetime is often considered to be a spacetime of minor importance in a cosmological context . however , first of all , it _ is _ a frw universe and as such should not be neglected , secondly ; it serves as a simple and convenient spacetime for comparison and understanding of results obtained in ( say ) minkowski or de sitter spacetimes and third , it has a tendency to show up ( in disguise ) as a solution in various models of dilaton - gravity and string theory . an example of the latter is represented by the @xmath0 bh - ads spacetime of baados et al @xcite . this spacetime background has arised much interest recently [ 31 - 35 ] . it describes a two - parameter family ( mass @xmath11 and angular momentum @xmath12 ) of black holes in 2 + 1 dimensional general relativity with metric : @xmath13 it has two horizons @xmath14 and a static limit @xmath15 defining an ergosphere , as for ordinary kerr black holes . the spacetime is not asymptotically flat ; it approaches anti de sitter spacetime asymptotically with cosmological constant @xmath16 the curvature is constant @xmath17 everywhere , except probably at @xmath18 where it has at most a delta - function singularity . notice the weak nature of the singularity at @xmath19 in 2 + 1 dimensions as compared with the power law divergence of curvature scalars in @xmath20 ( we will not discuss here the geometry near @xmath21 for a discusion , see refs.@xcite ) . the spacetime , eq.(2.1 ) , is also a solution of the low energy effective action of string theory with zero dilaton field @xmath22 anti - symmetric tensor field @xmath23 and @xmath24 @xcite . moreover , it yields an exact solution of string theory in 2 + 1 dimensions , obtained by gauging the wzwn sigma model of the group @xmath25 at level @xmath26 @xcite ( for non - compact groups , @xmath26 does not need to be an integer , so the central charge @xmath27 when k=52/23 ) . this solution is the black string background @xcite : @xmath28 @xmath29 which is related by duality @xcite to the 2 + 1 bh - ads spacetime , eq.(2.1 ) . it has two horizons @xmath30 the same as the metric , eq.(2.1 ) , while the static limit is @xmath31 we first investigate the string propagation in the @xmath0 bh - ads background by considering the perturbation series around the exact center of mass of the string : @xmath32 the original method of refs.@xcite can be conveniently formulated in covariant form . it is useful to introduce @xmath33 normal vectors @xmath34 ( which can be chosen to be covariantly constant by gauge fixing ) , and consider comoving perturbations @xmath35 i.e. those seen by an observer travelling with the center of mass , thus @xmath36 after fourier expansion , @xmath37 the first order perturbations satisfy the matrix schrdinger - type equation in @xmath7 : @xmath38 second order perturbations @xmath39 and constraints are similarly covariantly treated , @xmath39 also satisfying schrdinger - type equations with source terms , see ref.@xcite . for our purposes here it is enough to consider the non - rotating @xmath40 2 + 1 bh - ads background and a radially infalling string . we have solved completely the c.m . motion @xmath41 and the first and second order perturbations @xmath42 and @xmath43 in this background . eqs.(2.4 ) become : @xmath44 the first order perturbations are independent of the black hole mass , only the anti de sitter ( ads ) part emerges . all oscillation frequencies @xmath45 are real ; there are no unstable modes in this case . the perturbations : @xmath46\ ] ] are completely finite and regular . this is also true for the second order perturbations which are bounded everywhere , even for @xmath47 we have also computed the conformal generators @xmath48 ( see ref.@xcite ) , and the string mass : @xmath49.\ ] ] the mass formula is modified ( by the term @xmath50 ) with respect to the usual flat spacetime expression . this is due to the asymptotic ( here ads ) behaviour of the spacetime . in ordinary @xmath1 black hole spacetimes ( without cosmological constant ) , which are asymptotically flat , the mass spectrum is the same as in flat minkowski spacetime @xcite . we compare with the string perturbations in the ordinary ( @xmath1 ) black hole anti de sitter spacetime . in this case eqs.(2.4 ) become : @xmath51 @xmath52 the transverse @xmath53-perturbations are oscillating with real frequencies and are bounded even for @xmath54 for longitudinal @xmath55-perturbations , however , imaginary frequencies arise and instabilities develop . the @xmath56-instability sets in at : @xmath57 lower modes become unstable even outside the horizon , while higher modes develop instabilities at smaller @xmath58 and eventually only for @xmath59 for @xmath60 ( which implies @xmath61 ) we find @xmath62 and : @xmath63 @xmath64 for @xmath61 the @xmath55-perturbations blow up while the string ends trapped into the @xmath19 singularity . we see the important difference between the string evolution in the 2 + 1 bh - ads background and the ordinary 3 + 1 ( or higher dimensional ) black hole anti de sitter spacetime . we also compare with the string propagation in the 2 + 1 black string background , eq.(2.2 ) ( with @xmath65 ) . in this case , eqs.(2.4 ) become : @xmath66 @xmath67 the @xmath53-modes are stable , while @xmath68 develop instabilities . for @xmath60 ( which implies @xmath61 ) we find @xmath69 and : @xmath70 with similar conclusions as for the ordinary 3 + 1 ( or higher dimensional ) black hole anti de sitter spacetime . in order to extract more information about the string evolution in these backgrounds , in particular exact properties , we consider the circular string ansatz : @xmath71 in the equatorial plane @xmath72 of the stationary axially symmetric backgrounds : @xmath73 this includes all the cases of interest here : the 2 + 1 bh - ads spacetime , the black string , as well as the equatorial plane of ordinary einstein black holes . the string dynamics , determined by eqs.(1.1)-(1.2 ) , is then reduced to a system of second order ordinary differential equations and constraints , also described as a hamiltonian system : @xmath74 @xmath75 which in all backgrounds considered here are solved in terms of either elementary or elliptic functions . the dynamics of the circular strings takes place at the @xmath58-axis in the @xmath76-diagram and from the properties of the potential @xmath77 ( minima , zeroes , asymptotic behaviour for large @xmath58 and the value @xmath78 ) , general knowledge about the string motion can be obtained . on the other hand , the line element of the circular string turns out to be : @xmath79 @xmath80 being the invariant string size . for all the static black hole ads spacetimes ( 2 + 1 and higher dimensional ) : @xmath81 for the rotating 2 + 1 bh - ads spacetime : @xmath82 ( see fig.1 ) . @xmath77 has a global minimum @xmath83 between the two horizons @xmath84 ( for @xmath85 otherwise there are no horizons ) . the vanishing of @xmath77 at @xmath86 ( see ref.@xcite ) determines three different types of solutions : ( i ) for @xmath87 there are two positive zeroes @xmath88 the string never comes outside the static limit , never falls into @xmath19 neither ( there is a barrier between @xmath89 and @xmath19 ) . the mathematical solution oscillates between @xmath90 and @xmath91 with @xmath92 it may be interpreted as a string travelling between the different universes described by the maximal analytic extension of the manifold . ( ii ) for @xmath93 there is only one positive zero @xmath94 outside the static limit and there is no barrier preventing the string from collapsing into @xmath19 . the string starts at @xmath95 with maximal size @xmath96 outside the static limit , it then contracts through the ergosphere and the two horizons and eventually collapses into a point @xmath21 for @xmath97 it may be still possible to continue this solution into another universe like in the case ( i ) . ( iii ) @xmath98 is the limiting case where the maximal string size equals the static limit : @xmath99 in this case @xmath100 thus the string contracts through the two horizons and eventually collapses into a point @xmath21 the exact general solution in the three cases ( i)-(iii ) is given by : @xmath101 where : @xmath102 @xmath103 are constants in terms of @xmath104 given in ref.@xcite , and @xmath105 is the weierstrass elliptic @xmath105-function with invariants @xmath106 discriminant @xmath107 and roots @xmath108 also given in ref.@xcite . the three cases ( i)-(iii ) correspond to the cases @xmath109 @xmath110 and @xmath111 respectively . notice that @xmath112 in the case ( i ) , @xmath113 can be written in terms of the jacobian elliptic function @xmath114,\;\tau^*=\sqrt{e_1-e_3}\;\tau,\;k=\sqrt{(e_2- e_3)/(e_1-e_3)}.$ ] it follows that the solution ( i ) oscillates between the two zeroes @xmath90 and @xmath91 of @xmath115 with period @xmath116 where @xmath117 is the real semi - period of the weierstrass function : @xmath118 in terms of the complete elliptic integral of the first kind @xmath119 we have : @xmath120 in the case ( ii ) ( @xmath110 ) two roots ( @xmath121 ) become complex , the string collapses into a point @xmath19 and we have : @xmath122 where @xmath123 is the real semi - period of the weierstrass function for this case . in the case ( iii ) ( @xmath124 ) the elliptic functions reduce to hyperbolic functions : @xmath125 so that : @xmath126 here , the string starts as a point , grows until @xmath127 ( at @xmath95 ) , and then it contracts until it collapses again ( @xmath19 ) at @xmath128 in this case the string makes only one oscillation between @xmath19 and @xmath129 notice that for the static background ( @xmath65 ) , the only allowed motion is ( ii ) , i.e. @xmath130 ( there is no ergosphere and only one horizon in this case ) , with : @xmath131 for @xmath132 the string collapses into @xmath19 and stops there . the penrose diagram of the 2 + 1 bh - ads spacetime for @xmath65 is very similar to the penrose diagram of the ordinary ( @xmath1 ) schwarzschild spacetime , so the string motion outwards from @xmath19 is unphysical because of the causal structure . the coordinate time @xmath133 can be expressed in terms of the incomplete elliptic integral of the third kind @xmath134 see ref.@xcite . the string has its maximal size @xmath135 at @xmath136 passes the horizon at @xmath137 ( expressed in terms of an incomplete elliptic integral of the first kind ) and falls into @xmath19 for @xmath138 @xmath123 being the real semi - period of the weierstrass function . that is , we have : @xmath139 and @xmath140 can be expressed in terms of the jacobian zeta function @xmath141 see ref.@xcite . we also study the circular strings in the ordinary @xmath1 spacetimes . in the generic 3 + 1 kerr anti de sitter ( or kerr de sitter ) spacetime , the potential @xmath77 is however quite complicated , see ref.@xcite . the general circular string solution involves higher genus elliptic functions and it is not necessary to go into details here . we will compare with the non - rotating cases , only . it is instructive to recall @xcite the circular string in minkowski spacetime ( min ) , for which @xmath142 ( fig.2a ) , the string oscillates between its maximal size @xmath143 and @xmath19 with the solution @xmath144 in the schwarzschild black hole ( s ) @xmath145 ( fig.2b ) , the solution is remarkably simple : @xmath146 the mathematical solution oscillates between @xmath147 and @xmath148 but because of the causal structure and the curvature singularity the motion can not be continued after the string has collapsed into @xmath21 for anti de sitter spacetime ( ads ) , we find : @xmath149 ( fig.2c ) . the string oscillates between @xmath150 and @xmath19 with the solution : @xmath151\mid,\ ] ] which is periodic with period @xmath152 @xmath153 for schwarzschild anti de sitter spacetime ( s - ads ) , we find @xmath154 ( fig.2d ) and : @xmath155 @xmath156 are constants given in terms of @xmath157 see ref.@xcite ( @xmath135 is the root of the equation @xmath158 , which has in this case exactly one positive solution ) . the invariants , the discriminant and the roots are also given explicitly in ref.@xcite . the string starts with @xmath127 at @xmath136 it then contracts and eventually collapses into the @xmath19 singularity . the existence of elliptic function solutions for the string motion is characteristic of the presence of a cosmological constant . for @xmath159 the circular string motion reduces to simple trigonometric functions . from fig.2 and our analysis we see that the circular string motion is qualitatively very similar in all these backgrounds ( min , s , ads , s - ads ) : the string has a maximal _ bounded _ size and then it contracts towards @xmath21 there are however physical and quantitative differences : in minkowski and pure anti de sitter spacetimes , the string truly oscillates between @xmath135 and @xmath18 while in the black hole cases ( s , s - ads ) , there is only one half oscillation , the string motion stops at @xmath21 this also holds for the 2 + 1 bh - ads spacetime with @xmath65 ( fig.1b ) . notice also that in all these cases , @xmath160 and @xmath161 for @xmath162 the similarity can be pushed one step further by considering small perturbations around the circular strings . we find @xcite : @xmath163 determining the fourier components of the comoving perturbations . for the spacetimes of interest here , @xmath164 ( min , s , ads , s - ads ) , or @xmath165 ( 2 + 1 bh - ads ) , the comoving perturbations are regular except near @xmath18 where we find ( for all cases ) @xmath166 and : @xmath167 it follows that not only the unperturbed circular strings , but also the comoving perturbations around them behave in a similar way in all these non - rotating backgrounds ( 2 + 1 and higher dimensional ) . this should be contrasted with the string perturbations around the center of mass , which behave differently in these backgrounds . it must be noticed that for rotating ( @xmath168 ) spacetimes , the circular string behaviour is qualitatively different from the non - rotating ( @xmath65 ) spacetimes . for large @xmath169 both in the 2 + 1 bh - ads as well as in the 3 + 1 ordinary kerr - ads spacetimes , non - collapsing circular string solutions exist . the potential @xmath170 for @xmath60 and no collapse into @xmath19 is possible . the dynamics of circular strings in curved spacetimes is determined by the string tension , which tends to contract the string , and by the local gravity ( which may be attractive or repulsive ) . in all the previous non - rotating backgrounds , the local gravity is attractive ( i.e. @xmath171 ) , and it acts together with the string tension in the sense of contraction . but in spacetimes with regions in which repulsion ( i.e. @xmath172 ) dominates , the strings can expand with unbounded radius ( unstable strings @xcite ) . it may also happen that the string tension and the local gravity be of the same order , i.e. the two opposite effects can balance , and the string is stationary . de sitter spacetime provides an example in which all such type of solutions exist @xcite . in de sitter spacetime , @xmath77 is unbounded from below for @xmath173 ( @xmath174 ) and unbounded expanding circular strings are present . in addition , an interesting new feature appears in the presence of a positive cosmological constant : the existence of _ multi - string _ solutions [ 9 - 11 ] . the world - sheet time @xmath7 turns out to be a multi ( finite or infinite ) valued function of the physical time . that is , one single world - sheet where @xmath175 can describe many ( even infinitely many @xcite ) different and independent strings ( in flat spacetime , one single world - sheet describes only one string ) . in the s , ads and s - ads spacetimes , the multi - string feature for circular strings is absent . the main conclusions of this section are given in tables 1 , 2 . further details can be found in ref.@xcite in the first part of the previous section , the classical string dynamics was studied in the @xmath0 bh - ads spacetime . in this section we go further in the investigation of the physical properties of strings in ads spacetime , by performing string quantization . since the @xmath0 bh - ads spacetime is locally ads , the results of section 2 can be extended to classical strings in @xmath2-dimensional ads spacetime as well . in ads spacetime , the string motion is oscillatory in time and is _ stable _ ; all fluctuations around the string center of mass are well behaved and bounded . local gravity of ads spacetime is always negative and string instabilities do not develop . the string perturbation series approach , considering fluctuations around the center of mass , is particularly appropriate in ads spacetime , the natural dimensionless expansion parameter being @xmath176 where @xmath177 is the string tension and the ` hubble constant ' @xmath178 the negative cosmological constant of ads spacetime is related to the ` hubble constant ' @xmath179 by @xmath180 all the spatial ( @xmath181 ) modes in @xmath2-dimensional ads , oscillate with frequency @xmath182 which are real for all @xmath183 ( @xmath184 being the string mass ) . in this section , we perform a canonical quantization procedure . from the conformal generators @xmath185 and the constraints @xmath186 imposed at the quantum level , we obtain the mass formula ( the quantum version of eq.(2.7 ) written in units where @xmath177 appears explicitly ) : @xmath187,\ ] ] where : @xmath188 and we have applied symmetric ordering of the operators . the operators @xmath189 satisfy : @xmath190=[\tilde{\alpha}^r_n,\;(\tilde{\alpha}^r_n)^ { \dag}]=1,\;\;\;\;\;\;\mbox{for all}\;\;n>0.\ ] ] for @xmath191 which is clearly fulfilled in most interesting cases , we have found the lower mass states @xmath192 and the quantum mass spectrum @xcite . physical states are characterized by the eigenvalue of the number operator : @xmath193,\ ] ] and the ground state is defined by : @xmath194 we find that @xmath195 is an _ exact _ solution of the mass formula in @xmath196 and that there is a graviton at @xmath197 which indicates , as in de sitter space @xcite , that the critical dimension of ads is 25 ( although it should be stressed that the question whether these spacetimes are solutions to the full @xmath198-function equations remains open ) . as in minkowski spacetime , the ground state is a tachyon . remarkably enough , for @xmath199 we find that a generic feature of all excited states beyond the graviton , is the presence of a _ fine structure _ effect : for a given eigenvalue @xmath200 the corresponding states have different masses . for the lower mass states the expectation value of the mass operator in the corresponding states ( generically labelled @xmath201 ) turns out to have the form ( see ref.@xcite ) : @xmath202 the collective index `` j '' generically labels the state @xmath201 and the coefficients @xmath203 are all well computed numbers , different for each state , their precise values are given in ref.@xcite . the corrections to the mass in minkowski spacetime appear to order @xmath204 therefore , the leading regge trajectory for the lower mass states is : @xmath205 in minkowski spacetime the mass and number operator of the string are related by @xmath206 in ads ( as well as in de sitter ( ds ) ) , there is no such simple relation between the mass and the number operators ; the splitting of levels increases considerably for very large @xmath207 the _ fine structure _ effect we find here is also present in de sitter space . up to order @xmath208 the lower mass states in ds and in ads are the same , the differences appear to the order @xmath209 the lower mass states in de sitter spacetime are given by eq.(3.6 ) but with the @xmath210-term getting an opposite sign ( @xmath211 ) . for the very high mass spectrum , we find more drastic effects . states with very large eigenvalue @xmath212 namely @xmath213 have masses : @xmath214 and angular momentum : @xmath215 where @xmath216 are well computed numbers different for each state , their precise values are given in ref.@xcite . since @xmath217 we see from eq.(3.8 ) that the masses of the high mass states are _ independent _ of @xmath218 in minkowski spacetime , very large @xmath219 states all have the same mass @xmath220 but here in ads the masses of the high mass states with the same eigenvalue @xmath219 are _ different _ by factors @xmath221 in addition , because of the fine structure effect , states with different @xmath219 can get mixed up . for high mass states , the level spacing _ grows _ with @xmath219 ( instead of being constant as in minkowski spacetime ) . as a consequence , the density of states @xmath222 as a function of mass grows like @xmath223 $ ] ( instead of @xmath224 $ ] as in minkowski spacetime ) , and _ independently _ of @xmath218 the partition function for a gas of strings at a temperature @xmath225 in ads spacetime is well defined for all finite temperatures @xmath226 discarding the existence of the hagedorn temperature . for the low mass states ( @xmath192 ) in anti de sitter spacetime , our results can be written as @xcite : @xmath227 where `` @xmath228 '' is a collective index labelling the state @xmath229 it is now important to notice that @xmath230 for _ all _ the low mass states @xcite , i.e. there is no `` constant '' term on the right hand side of eq.(3.10 ) . a non - zero @xmath231-term would give rise to a @xmath177-independent contribution to the string mass . its absense , on the other hand , means that the first term on the right hand side of eq.(3.10 ) is super - dominant ( since , in all cases , @xmath232 ) and that the string scale is therefore set by @xmath233 for the high mass states ( @xmath234 ) we found instead @xcite : @xmath235 where the number @xmath236 depends on the state . the right hand side of eq.(3.11 ) is exactly like a non - zero dominant @xmath231-term in eq.(3.10 ) . for the high mass states the scale is therefore set by @xmath237 which is equal to the absolute value of the cosmological constant @xmath238 ( up to a geometrical factor ) and _ independent _ of @xmath218 this suggests that for @xmath191 the masses of _ all _ string states can be represented by a formula of the form ( 3.10 ) . for the low mass states @xmath239 while for the high mass states @xmath231 becomes a large positive number . in the black string background , eq.(2.2 ) , we have calculated explicitly the first and second order string fluctuations around the center of mass @xcite . we then determined the world - sheet energy - momentum tensor and we derived the mass formula in the asymptotic region . the mass spectrum is equal to the mass spectrum in flat minkowski spacetime . therefore , for a gas of strings at temperature @xmath225 in the asymptotic region of the black string background , the partition function goes like : @xmath240 which is only defined for @xmath241 i.e. there is a hagedorn temperature : @xmath242 in higher dimensional ( @xmath1 ) black hole spacetimes the next step now would be to set up a scattering formalism , where a string from an asymptotic in - state interacts with the gravitational field of the black hole and reappears in an asymptotic out - state @xcite . however , this is not possible in the black string background . in the uncharged black string background under consideration here , _ all _ null and timelike geodesics incoming from spatial infinity pass through the horizon and fall into the physical singularity @xcite . no `` angular momentum '' , as in the case of scattering off the ordinary schwarzschild black hole , can prevent a point particle from falling into the singularity . the string solutions considered here are based on perturbations around the string center of mass which follows , at least approximately , a point particle geodesic . therefore , a string incoming from spatial infinity inevitably falls into the singularity in the black string background . in this section we return to the string equations of motion and constraints , eqs.(1.1)-(1.2 ) , to look for new classes of exact solutions . in most spacetimes , quite general families of exact solutions can be found by making an appropriate ansatz , which exploits the symmetries of the underlying curved spacetime . in axially symmetric spacetimes , a convenient ansatz corresponds to circular strings , as we saw in section 2 . such an ansatz effectively decouples the dependence on the spatial world - sheet coordinate @xmath8 and the string equations of motion and constraints reduce to non - linear coupled ordinary differential equations , eqs.(2.18)-(2.19 ) . in this section we will make instead an ansatz which effectively decouples the dependence on the temporal world - sheet coordinate @xmath243 this ansatz , which we call the `` stationary string ansatz '' is dual to the `` circular string ansatz '' in the sense that it corresponds to a formal interchange of the world - sheet coordinates @xmath244 as well as of the azimuthal angle @xmath245 and the stationary time @xmath246 in the target space : @xmath247 the stationary string ansatz will describe stationary strings when @xmath246 ( and @xmath7 ) are timelike , for instance in anti de sitter spacetime ( in static coordinates ) and outside the horizon of a schwarzschild black hole . on the other hand , if @xmath246 ( and @xmath7 ) are spacelike , for instance inside the horizon of a schwarzschild black hole or outside the horizon of de sitter spacetime ( in static coordinates ) , the stationary string ansatz will describe dynamical propagating strings . considering for simplicity a static line element in the form : @xmath248 the stationary string ansatz reads explicitly : @xmath249 the string equations of motion and constraints , eqs.(1.1)-(1.2 ) , reduce to two separated first order ordinary differential equations : @xmath250,\ ] ] where @xmath251 is an integration constant . the qualitative features of the possible string configurations can be read off directly from the shape of the potential @xmath252 thereafter , the detailed analysis of the quantitative features can be performed by explicitly solving the ( integrable ) system of equations ( 4.4 ) . the induced line element on the world - sheet is given by : @xmath253 thus , if @xmath254 is negative , the world - sheet coordinate @xmath7 becomes spacelike while @xmath255 becomes timelike and the stationary string ansatz ( 4.3 ) describes a dynamical string . if @xmath254 is positive , the ansatz describes a stationary string . in this section we solve explicitly eqs.(4.4 ) and we analyze in detail the solutions and their physical interpretation in minkowski , de sitter , anti de sitter , schwarzschild and @xmath0 black hole anti de sitter spacetimes . in all these cases , the solutions are expressed in terms of elliptic ( or elementary ) functions . we furthermore analyze the physical properties , energy and pressure , of these solutions . the energy and pressure densities of the strings can be obtained from the @xmath256 dimensional spacetime energy - momentum tensor : @xmath257 after integration over a spatial volume that completely encloses the string @xcite , the energy - momentum tensor takes the form of a fluid : @xmath258 in minkowski spacetime ( min ) , the potential defined in eqs.(4.4 ) is given by : @xmath259 and the solution of eqs.(4.4 ) describes one infinitely long straight string with `` impact - parameter '' @xmath260 the equation of state , relating energy and pressure densities , takes the well - known form ref.@xcite : @xmath261 in anti de sitter spacetime ( ads ) , the potential is given by : @xmath262.\ ] ] the radial coordinate @xmath263 is periodic with finite period @xmath264 which can be expressed in terms of a complete elliptic integral , see ref.@xcite . for @xmath265,$ ] the solution describes an infinitely long stationary string in the wedge @xmath2660,\delta\phi[\;,$ ] where : @xmath267\;\in\;\;]0,\;\pi[\ ] ] the elliptic modulus @xmath268 which is a function of @xmath269 @xcite , parametrizes the solutions , @xmath2700,1/\sqrt{2}[\;.$ ] the azimuthal angle is generally not a periodic function of @xmath8 thus when the spacelike world - sheet coordinate @xmath255 runs through the range @xmath271-\infty,+\infty[\;,$ ] the solution describes an _ infinite number _ of infinitely long stationary open strings . the general solution is therefore a _ multi - string _ solution . until now multi - string solutions were only found in de sitter spacetime [ 11 - 13 ] . our results show that multi - string solutions are a general feature of spacetimes with a cosmological constant ( positive or negative ) . the solution in anti de sitter spacetime describes a _ finite _ number of strings if the following relation holds : @xmath272 here @xmath219 and @xmath11 are integers , determining the number of strings and the winding in azimuthal angle , respectively , for the multi - string solution , see fig.3 . the equation of state for a full multi - string solution takes the form @xmath273 : @xmath274 corresponding to extremely unstable strings @xcite . in de sitter spacetime ( ds ) , the potential is given by : @xmath275.\ ] ] in this case we have to distinguish between solutions inside the horizon ( where @xmath7 is timelike ) and solutions outside the horizon ( where @xmath7 is spacelike ) . inside the horizon , the generic solution describes one infinitely long open stationary string winding around @xmath21 for special values of the constants of motion , corresponding to a relation , which formally takes the same form as eq.(4.12 ) , the solution describes a closed string of finite length @xmath276 the integer @xmath219 in this case determines the number of `` leaves '' , see fig.4 . the energy is positive and finite and grows with @xmath207 the pressure turns out to vanish identically , thus the equation of state corresponds to cold matter . outside the horizon , the world - sheet coordinate @xmath7 becomes spacelike while @xmath255 becomes timelike , thus we define : @xmath277 and the string solution is conveniently expressed in hyperboloid coordinates . the radial coordinate @xmath278 is periodic with a finite period @xmath279 for @xmath280,$ ] the solution describes a straight string incoming non - radially from spatial infinity , scattering at the horizon and escaping towards infinity again , fig.5 . the string length is zero at the horizon and grows indefinetely in the asymptotic regions . as in the case of anti de sitter spacetime , the azimuthal angle is generally not a periodic function , thus when the timelike world - sheet coordinate @xmath281 runs through the range @xmath271-\infty,+\infty[\;,$ ] the solution describes an _ infinite _ number of dynamical straight strings scattering at the horizon at different angles . the general solution is therefore a multi - string solution . in particular , a multi - string solution describing a _ finite _ number of strings is obtained if a relation of the form ( 4.12 ) is fulfilled . it turns out that the solution describes _ at least _ three strings . the energy and pressures of a full multi - string solution have also been computed . in the asymptotic region they fulfill an equation of state corresponding to extremely unstable strings , i.e. like eq.(4.13 ) . in the schwarzschild black hole background ( s ) , the potential is given by : @xmath282.\ ] ] no multi - string solutions are found in this case . outside the horizon the solution of eqs.(4.4 ) describes one infinitely long stationary open string . this solution was already derived in ref.@xcite , and we shall not discuss it here . inside the horizon , where @xmath7 becomes spacelike while @xmath255 becomes timelike , we make the redefinitions ( 4.15 ) and the solution is conveniently expressed in terms of kruskal coordinates , see ref.@xcite . the solution describes one straight string infalling _ non - radially _ towards the singularity , see fig.6 . at the horizon , the string length is zero and it grows _ indefinitely _ when the string approaches the spacetime singularity . thereafter , the solution can not be continued , see ref.@xcite . in the @xmath0 black hole anti de sitter spacetime ( bh - ads ) @xcite , the potential is given by : @xmath283.\ ] ] outside the horizon , the solutions `` interpolate '' between the solutions found in anti de sitter spacetime and outside the horizon of the schwarzschild black hole . the solutions thus describe infinitely long stationary open strings . as in anti de sitter spacetime , the general solution is a _ multi - string _ describing _ infinitely _ many strings . in particular , for certain values of the constants of motion , corresponding to the condition of the form ( 4.12 ) , the solution describes a finite number of strings . in the simplest version of the @xmath0 bh - ads background ( @xmath284 ) , it turns out that the solution describes _ at least _ seven strings , see ref.@xcite . inside the horizon , we make the redefinitions ( 4.15 ) and the solution is conveniently expressed in terms of kruskal - like coordinates , see ref.@xcite . the solution is similar to the solution found inside the horizon of the schwarzschild black hole , but there is one important difference at @xmath21 as in the schwarzschild black hole background , the solution describes one straight string infalling non - radially towards @xmath18 and beyond this point the solution can not be continued because of the global structure of the spacetime . at the horizon the string size is zero and during the fall towards @xmath18 the string size _ grows _ but stays _ finite_. this should be compared with the straight string inside the horizon of the schwarzschild black hole , where the string size grows indefinetely . the physical reason for this difference is that the point @xmath19 is not a strong curvature singularity in the @xmath0 bh - ads spacetime . we have considered also the schwarzschild de sitter and schwarzschild anti de sitter spacetimes . these spacetimes contain all the features of the spacetimes already discussed : singularities , horizons , positive or negative cosmological constants . all the various types of string solutions found in the other spacetimes ( open , closed , straight , finitely and infinitely long , multi - strings ) are therefore present in the different regions of the schwarzschild - de sitter and schwarzschild - anti de sitter spacetimes . the details are given in ref.@xcite . we close this section with a few remarks on the stability of the solutions . generally , the question of stability must be addressed by considering small perturbations around the exact solutions . in ref.@xcite , a covariant formalism describing physical perturbations propagating along an arbitrary string configuration embedded in an arbitrary curved spacetime , was developed . the resulting equations determining the evolution of the perturbations are however very complicated in the general case , although partial ( analytical ) results have been obtained in special cases for de sitter @xcite and schwarzschild black hole @xcite spacetimes . the exact solutions found in this section fall essentially into two classes : dynamical and stationary . the dynamical string solutions outside the horizon of de sitter ( or s - ds ) and inside the horizon of schwarzschild ( or s - ds , s - ads ) spacetimes , are already unstable at the zeroth order approximation ( i.e. without including small perturbations ) , in the sense that their physical length grows indefinetely . for the stationary string solutions the situation is more delicate . the existence of the stationary configurations is based on an exact balance between the string tension and the local attractive or repulsive gravity . for that reason , it can be expected that the configurations are actually unstable for certain modes of perturbation , especially in strong curvature regions . the main conclusions of this section are given in table 4 . further details can be found in ref.@xcite . the propagation of strings in friedmann - robertson - walker ( frw ) cosmologies has been investigated using both exact and approximative methods , see for example refs.[7 , 11 - 14 , 17 - 20 ] ( as well as numerical methods , which shall not be discussed here ) . except for anti de sitter spacetime , which has negative spatial curvature , the cosmologies that have been considered until now , have all been spatially flat . in this section we will consider the physical effects of a non - zero ( positive or negative ) curvature index on the classical and quantum strings . the non - vanishing components of the riemann tensor for the generic d - dimensional frw line element , in comoving coordinates : @xmath285 are given by : @xmath286 where @xmath287 is the scale factor and @xmath288 is the curvature index . clearly , a non - zero curvature index introduces a non - zero spacetime curvature ; the exceptional case provided by @xmath289 corresponds to the milne - universe . from eqs.(5.2 ) , it is also seen that the curvature index has to compete with the first derivative of the scale factor . the effects of the curvature index are therefore most conveniently discussed in the family of frw - universes with constant scale factor , the so - called static robertson - walker spacetimes . this is the point of view we take in the present section . we consider both the closed ( @xmath290 ) and the hyperbolic ( @xmath291 ) static robertson - walker spacetimes , and all our results are compared with the already known results in the flat ( @xmath292 ) minkowski spacetime . we determine the evolution of circular strings , derive the corresponding equations of state ( using eq.(4.6 ) ) , discuss the question of strings as self - consistent solutions to the einstein equations @xcite , and we perform a semi - classical quantization . we also find all the stationary string configurations in these spacetimes . the radius of a classical circular string in the spacetime ( 5.1 ) , for @xmath293 is determined by : @xmath294 where @xmath295 is an integration constant ( @xmath296 in the notation of eqs.(2.18)-(2.19 ) ) . this equation is solved in terms of elliptic functions and all solutions describe oscillating strings ( fig.7 shows the potential @xmath77 for @xmath297 ) . for @xmath298 when the spatial section is a hypersphere , the string either oscillates on one hemisphere or on the full hypersphere . the energy is always positive while the average pressure can be positive , negative or zero , depending on the precise value of an elliptic modulus ; the equation of state is given explicitly in table 5 , in the different cases . interestingly enough , we find that the circular strings provide a self - consistent solution to the einstein equations with a selected value of the curvature index . self - consistent solutions to the einstein equations with string sources have been found previously in the form of power law expanding universes @xcite . we have semi - classically quantized the circular string solutions of eq.(5.3 ) . we used an approach developed in field theory by dashen et . @xcite , based on the stationary phase approximation of the partition function . the result of the stationary phase integration is expressed in terms of the function : @xmath299 where @xmath300 is the action of the classical solution , @xmath184 is the mass and the period @xmath301 is implicitly given by : @xmath302 the bound state quantization condition then becomes @xcite : @xmath303 for @xmath183 ` large ' . parametric plots of @xmath304 as a function of @xmath305 in the different cases , are shown in fig.8 . the strings oscillating on one hemisphere give rise to a finite number @xmath306 of states with the following mass - formula ref.@xcite : @xmath307 as in flat minkowski spacetime , the scale of these string states is set by @xmath218 the strings oscillating on the full hypersphere give rise to an infinity of more and more massive states with the asymptotic mass - formula ref.@xcite : @xmath308 the masses of these states are independent of @xmath309 the scale is set by the curvature index @xmath310 notice also that the level spacing grows with @xmath311 a similar result was found recently for strings in anti de sitter spacetime @xcite . for @xmath312 when the spatial section is a hyperboloid , both the energy and the average pressure of the oscillating strings are positive . the equation of state is given in table 5 . in this case , the strings can not provide a self - consistent solution to the einstein equations . after semi - classical quantization , we find an infinity of more and more massive states . the mass - formula is given by ( see ref.@xcite ) : @xmath313 notice that the level spacing grows faster than in minkowski spacetime but slower than in the closed static robertson - walker spacetime . on the other hand , the stationary strings are determined by eqs.(4.4 ) : @xmath314 where @xmath251 is an integration constant ( fig.9 shows the potential @xmath77 for @xmath297 ) . for @xmath298 all the stationary string solutions describe circular strings winding around the hypersphere . the equation of state is of the extremely unstable string type @xcite . for @xmath312 the stationary strings are represented by infinitely long open configurations with an angle between the two `` arms '' given by : @xmath315 the energy density is positive while the pressure densities are negative . no simple equation of state is found for these solutions . we also computed the first and second order fluctuations around a static string center of mass , using the string perturbation series approach @xcite and its covariant versions @xcite . up to second order , the mass - formula for arbitrary values of the curvature index ( positive or negative ) is identical to the well - known flat spacetime mass - formula ; all dependence on @xmath288 cancels out . we first studied the string propagation in the 2 + 1 bh - ads background . we found the first and second order perturbations around the string center of mass as well as the mass formula , and compared with the ordinary black hole ads spacetime and with the black string background as well . these results were then generalized to ordinary @xmath2-dimensional anti de sitter spacetime . the classical string motion in anti de sitter spacetime is stable in the sense that it is oscillatory in time with real frequencies and the string size and energy are bounded . quantum mechanically , this reflects in the mass operator , which is well defined for any value of the wave number @xmath316 and arbitrary high mass states ( and therefore an infinite number of states ) can be constructed . this is to be contrasted with de sitter spacetime , where string instabilities develop , in the sense that the string size and energy become unbounded for large de sitter radius . for low mass states ( the stable regime ) , the mass operator in de sitter spacetime is given by eq.(3.1 ) but with @xmath317 real mass solutions can be defined only up to some _ maximal mass _ of the order @xmath318 @xcite . for @xmath191 real mass solutions can be defined only for @xmath319 ( where @xmath219 is the eigenvalue of the number operator ) and therefore there exists a _ finite _ number of states only . these features of strings in de sitter spacetime have been recently confirmed within a different ( semi - classical ) quantization approach based on _ exact _ circular string solutions @xcite . the presence of a cosmological constant @xmath238 ( positive or negative ) increases considerably the number of levels of different eigenvalue of the mass operator ( there is a splitting of levels ) with respect to flat spacetime . that is , a non - zero cosmological constant _ decreases _ ( although does not remove ) the degeneracy of the string mass states , introducing a _ fine structure effect_. for the low mass states the level spacing is approximately constant ( up to corrections of the order @xmath320 ) . for the high mass states , the changes are more drastic and they depend crucially on the sign of @xmath321 a value @xmath322 causes the _ growing _ of the level spacing linearly with @xmath219 instead of being constant as in minkowski space . consequently , the density of states @xmath222 grows with the exponential of @xmath323 ( instead of @xmath184 as in minkowski space ) discarding the existence of a hagedorn temperature in ads spacetime , and the possibility of a phase transition . in addition , another important feature of the high mass string spectrum in ads spacetime is that it becomes independent of @xmath218 the string scale for the high mass states is given by @xmath324 instead of @xmath325 for the low mass states . we have found new classes of exact string solutions obtained either by the circular string ansatz , eq.(2.16 ) , or by the stationary string ansatz , eq.(4.3 ) , in a variety of curved backgrounds including schwarzschild , de sitter and anti de sitter spacetimes . many different types of solutions have been found : oscillating circular strings , closed stationary strings , infinitely long stationary strings , dynamical straight strings and multi - string solutions describing finitely or infinitely many stationary or dynamical strings . in all the cases we have obtained the exact solutions in terms of either elementary or elliptic functions . furthermore , we have analyzed the physical properties ( length , energy , pressure ) of the string solutions . finally , we solved the equations of motion and constraints for circular strings in static robertson - walker spacetimes . we computed the equations of state and found that there exists a self - consistent solution to the einstein equations in the case of positive spatial curvature . the solutions were quantized semi - classically using the stationary phase approximation method , and the resulting spectra were analyzed and discussed . we have also found all stationary string configurations in these spacetimes and we computed the corresponding physical quantities , string length , energy and pressure . larsen is supported by the danish natural science research council under grant no . 11 - 1231 - 1se 11 h.j . de vega and n. snchez , phys . d42 ( 1990 ) 3969 de vega , m. ramn - medrano and n. snchez , nucl . b374 ( 1992 ) 405 . de vega and n. snchez , phys . b244 ( 1990 ) 215 , phys . 65c ( 1990 ) 1517 , ijmp a7 ( 1992 ) 3043 , nucl . phys . b317 ( 1989 ) 706 and ibid 731 . + d. amati and c. klimcik , phys . b210 ( 1988 ) 92 . + m. costa and h.j . de vega , ann . 211 ( 1991 ) 223 and ibid 235 . + c. loust and n. snchez , phys . d46 ( 1992 ) 4520 . v.e . zakharov and a.v . mikhailov , jetp 47 ( 1979 ) 1017 . + h. eichenherr , in `` integrable quantum field theories '' , ed . j. hietarinta and c. montonen ( springer , berlin , 1982 ) . de vega and n. snchez , phys . d47 ( 1993 ) 3394 . i. bars and k. sfetsos , mod . a7 ( 1992 ) 1091 . de vega , j.r . mittelbrunn , m.r . medrano and n. snchez , phys . lett . b232 ( 1994 ) 133 and `` the two - dimensional stringy black hole : a new approach and a pathology '' , par - lpthe 93/14 . h.j . de vega and n. snchez , phys b197 ( 1987 ) 320 . . de vega and n. snchez , nucl . b299 ( 1988 ) 818 . mende , in `` string quantum gravity and physics at the planck energy scale '' . proceedings of the erice workshop held in june 1992 , ed . n. snchez ( world scientific , 1993 ) . a.l . larsen and v.p . frolov , nucl . b414 ( 1994 ) 129 . de vega , a.v . mikhailov and n. snchez , teor . 94 ( 1993 ) 232 , mod . a29 ( 1994 ) 2745 . f. combes , h.j . de vega , a.v . mikhailov and n. snchez , phys . d50 ( 1994 ) 2754 . . de vega , a.l . larsen and n. snchez , nucl . b427 ( 1994 ) 643 . i. krichever , `` two - dimensional algebraic - geometrical operators with self - consistent potentials '' , landau institute preprint , march , 1994 , to appear in funct . analisis and appl . de vega and n. snchez , nucl . b309 ( 1988 ) 552 and ibid , 577 . loust and n. snchez , phys . d47 ( 1993 ) 4498 . de vega and i.l . egusquiza , phys . d50 ( 1994 ) 763 . larsen , phys . d50 ( 1994 ) 2623 . a.l . larsen and m. axenides , b318 ( 1993 ) 47 . de vega and n. snchez , phys . d50 ( 1994 ) 7202 . n. snchez and g. veneziano , nucl . b333 ( 1990 ) 253 . + m. gasperini , n. snchez and g. veneziano , ijmp a6 ( 1991 ) 3853 , nucl . b364 ( 1991 ) 365 . . de vega , m. ramn - medrano and n. snchez , nucl . b374 ( 1992 ) 425 . de vega and n. snchez , phys . d45 ( 1992 ) 2783 . + m. ramn - medrano and n. snchez , class . quantum grav . 10 ( 1993 ) 2007 . n. snchez , phys . b195 ( 1987 ) 160 . de vega , m. ramn - medrano and n. snchez , nucl . b351 ( 1991 ) 277 . a.l . larsen and n. snchez , d50 ( 1994 ) 7493 . a.l . larsen and n. snchez , `` mass spectrum of strings in anti de sitter spacetime '' , demirm - paris-94048 , to appear in phys . d. a.l . larsen and n. snchez , `` new classes of exact multi - string solutions in curved spacetimes '' , demirm - paris-95003 , to appear in phys . d. a.l . larsen and n. snchez , `` the effect of spatial curvature on the classical and quantum strings '' , demirm - paris-95004 . m. baados , c. teitelboim and j. zanelli , phys . 69 ( 1992 ) 1849 . g.t . horowitz and d.l . welch , phys . 71 ( 1993 ) 328 . n. kaloper , phys . d48 ( 1993 ) 2598 . m. baados , m. henneaux , c. teitelboim and j. zanelli , phys . d48 ( 1993 ) 1506 . c. farina , j. gamboa and a.j . segu - santonja , class . quantum grav . 10 ( 1993 ) l193 . + n. cruz , c. martnez and l. pea , class . quantum grav . 11 ( 1994 ) 2731 . a. achcarro and m.e . ortiz , phys . d48 ( 1993 ) 3600 . + d. cangemi , m. leblanc and r.b . mann , phys . d48 ( 1993 ) 3606 . j. horne and g.t . horowitz , nucl . b368 ( 1992 ) 444 . t. buscher , phys . b201 ( 1988 ) 466 , b194 ( 1987 ) 59 . a. vilenkin , phys . d24 ( 1981 ) 2082 . de vega and n. snchez , int . j. mod . a7 ( 1992 ) 3043 . frolov , v.d . skarzhinsky , a.i . zelnikov and o. heinrich , b224 ( 1989 ) 255 . r. dashen , b. hasslacher and a. neveu , phys . d11 ( 1975 ) 3424 . . de vega , a.l . larsen and n. snchez , `` semi - classical quantization of circular strings in de sitter and anti de sitter spacetimes '' , demirm - paris-94049 , submitted to phys . larsen , `` stable and unstable circular strings in inflationary universes '' , nordita-94/14-p , to appear in phys . larsen , nucl . b412 ( 1994 ) 372 . the potential @xmath115 eq.(2.21 ) , for a circular string in the @xmath0 bh - ads spacetime . in ( a ) we have @xmath326 and a barrier between the inner horizon and @xmath19 , while ( b ) represents a case where @xmath327 and a string will always fall into @xmath19 . the static limit is @xmath328 . the potential @xmath115 eq.(2.18 ) , for a circular string in the equatorial plane of the four @xmath256 dimensional spacetimes : ( a ) minkowski ( min ) , ( b ) schwarzschild black hole ( s ) , ( c ) anti de sitter ( ads ) , and ( d ) schwarzschild anti de sitter space ( s - ads ) . figure 3 . the @xmath329 multi - string solution in anti de sitter spacetime . the @xmath330 multi - string solutions describe @xmath219 stationary strings with @xmath11 windings in the azimuthal angle @xmath331 in anti de sitter spacetime . figure 4 . the @xmath332 stationary string solution inside the horizon of de sitter spacetime . besides the circular string , this is the simplest stationary closed string configuration in de sitter spacetime . schematic representation of the time evolution of the @xmath329 dynamical multi - string solution , outside the horizon of de sitter spacetime . only one of the 5 strings is shown ; the others are obtained by rotating the figure by the angles @xmath333 and @xmath334 during the `` scattering '' at the horizon , the string collapses to a point and re - expands . the dynamical straight string inside the horizon of the schwarzschild black hole . the string infalls non - radially towards the singularity with indefinetely growing length . the potential @xmath77 introduced in eqs.(5.3 ) for a circular string in the static robertson - walker spacetimes : ( a ) flat @xmath335 ( b ) closed @xmath336 ( c ) closed @xmath337 ( d ) hyperbolic @xmath338 figure 8 . parametric plot of @xmath304 as a function of @xmath305 in the three cases : ( a ) @xmath339 ( strings oscillating on one hemisphere , only finitely many states ) , ( b ) @xmath340 ( strings oscillating on the full hypersphere , infinitely many states ) , ( c ) @xmath291 ( strings oscillating on the hyperboloid , infinitely many states ) . the potential @xmath77 introduced in eqs.(5.10 ) for a stationary string in the static robertson - walker spacetimes : ( a ) flat @xmath335 ( b ) closed @xmath341 ( c ) hyperbolic @xmath338 table 1 . string motion described by the string perturbation series approach in the 2 + 1 bh - ads , ordinary black hole - ads , de sitter ( ds ) and black string backgrounds . circular exact string solutions in the indicated backgrounds . @xmath80 is the invariant string size . characteristic features of the quantum string mass spectrum in anti de sitter ( ads ) and de sitter ( ds ) spacetimes . short summary of the features of the string solutions found in this section . in anti de sitter spacetime and outside the horizon of de sitter and @xmath0 bh - ads spacetimes , the solutions describe a _ finite _ number of strings provided a condition of the form @xmath342 is fulfilled , where @xmath343 is the angle betwen the `` arms '' of the string and @xmath330 are integers . classical circular strings in the static robertson - walker spacetimes . notice that a self - consistent solution to the einstein equations , with the string back - reaction included , can be obtained only for @xmath344 table 6 . semi - classical quantization of the circular strings in the static robertson - walker spacetimes . notice in particular the different behaviour of the high mass spectrum of strings in the three cases . stationary strings in the static robertson - walker spacetimes . notice that the pressure densities are always negative in all three cases .

_ x - shooter _ and isis wht spectra of the starforming galaxy _ phl 293b _ also known as _ a2228 - 00 _ and _ sdss j223036.79 - 000636.9 _ are presented in this paper . we find broad ( fwhm = 1000km / s ) and very broad ( fwzi = 4000km / s ) components in the balmer lines , narrow absorption components in the balmer series blueshifted by 800km / s , previously undetected feii multiplet ( 42 ) absorptions also blueshifted by 800km / s , ir caii triplet stellar absorptions consistent with [ fe / h ] and no broad components or blushifted absorptions in the hei lines . based on historical records , we found no optical variability at the 5 level of 0.02 mag between 2005 and 2013 and no optical variability at the level of 0.1mag for the past 24 years . the lack of variability rules out transient phenomena like luminous blue variables or sn iin as the origin of the blue shifted absorptions of hi and feii . the evidence points to either a young and dense expanding supershell or a stationary cooling wind , in both cases driven by the young cluster wind . [ firstpage ] galaxies : abundances galaxies : dwarfs

1409.1189

early studies of emission line galaxies have shown that objects that spectroscopically resembled hii regions both in line emission intensities and widths , constituted about 80 per cent of total samples , the rest being galaxies of seyfert type ( see e.g. * ? ? ? * and references therein ) . some of these objects are also characterised by their compacticity and blue excess as shown on photographic plates and represent the overlap between blue compact galaxies ( bcg ) and hii galaxies . _ phl 293 b _ is one of these objects . it was found by @xcite in the `` palomar - haro - luyten '' survey of faint blue objects . @xcite obtained its optical spectrum and described it as having a faint continuum with unresolved emission in the balmer series and [ oiii ] @xmath25007,4959 and 4363 that makes of it , to our knowledge , one of the first detections of the [ oiii]@xmath34363 auroral line in a star forming galaxy . we nowadays recognize this fact as the one allowing a trustworthy derivation of the gas elemental abundances in regions of star formation showing an emission line spectrum . in the case of _ phl 293 b _ , these abundances are amongst the lowest known , less than one tenth of the solar value @xcite and in the border line of what is considered to be an extremely metal deficient galaxy ( e.g. * ? ? ? the galaxy was included in the sample studied by @xcite who lists its absolute magnitude and size as -13.6 and 0.2 kpc respectively . _ phl 293b _ also looks compact in the images obtained by @xcite , and @xcite give an effective radius for this galaxy of only 0.4 kpc . its absolute magnitude according to the sloan digital sky survey ( sdss ) is , m@xmath4 = -14.77 . these facts make of it a very low luminosity and compact hii galaxy . one of the characteristics of hii galaxies is their high star formation rate that takes place in a very small volume and probably in short duration episodes , thus making these galaxies easily observable . given the large value of the equivalent width of their emission lines , it is the current burst of star formation that dominates their luminosity at blue and visible wavelengths . the low metallicity of these objects guarantees that they are in a chemically unevolved stage probably similar to what is expected in galaxies at early cosmological times . the evolution of their massive young stars , responsible for the gas ionisation , is conditioned by their low metallicity as is probably the case for pop iii stars . the evolution of high mass stars is short and encompasses phases with intense episodes of mass loss , therefore the presence of low intensity broad components or wings in the otherwise narrow emission lines typical of star - forming regions ( sfr ) , are relatively common in high s / n medium resolution spectra . well known examples are : ngc 604 , a giant hii region in the spiral galaxy m 33 @xcite ; ngc 5471 , a giant hii region in the spiral galaxy m 101 @xcite , ngc 2363 , a giant hii region in the irregular galaxy ngc 2366 @xcite and 30 dor in the lmc , one of the largest extragalactic hii regions in the local universe @xcite . this is also the case , as expected , for hii galaxies and strong line bcg ( see e.g. * ? ? ? * ) , since it is the dominant star forming region that dominates their integrated spectra . the most likely origin of the extended line wings observed in sfr is related to the evolution of very massive stars and the feedback processes between them and the surrounding interstellar medium . given the large number of these stars present in a relatively small volume , it may be expected that the combination of their powerful winds gave rise to complex kinematical components in both permitted and forbidden lines . also multiple supernova events have been invoked as the possible cause ( see * ? ? ? * for the case of ngc 2363 ) . however , some of these massive stars like luminous blue variables ( lbv ) , can be so luminous as to be capable of producing visible effects by themselves particularly in nearby resolved system and some of them may even end their lives as supernovae hence dominating for some time the galaxy luminosity output . this evolutionary path may be the one followed by the most massive stars of low metallicity for which the inefficiency of line - driven stellar winds would translate in a low mass - loss rate and failure in becoming a wr star . some of these stars might be the progenitors of type iin supernovae in which the narrow line spectrum arises from the interaction of the supernova blast wave with the circumstellar shell . these supernovae can be very luminous and also very long lived ( aretxaga et al . 1999 ; smith et al . 2007 ) . another mechanism accounting for the presence of low intensity broad line components in hii galaxies or bcg spectra may be mass accretion onto an intermediate - mass black hole ( 10@xmath5 - 10@xmath6 m@xmath7 ) . up to now there are no _ bona fide _ low metallicity objects ( about one tenth of solar ) harbouring massive black holes . in fact , it is intrinsically difficult to identify these objects in commonly used diagnostic diagrams ( e.g. * ? ? ? * ) . the overlapping of their broad component h@xmath8 luminosities and those of supernovae and even stellar winds , makes this identification even harder . yet , from the point of view of galactic evolution to firmly establish the existence of these objects would be of the greatest importance . in this sense the detection of x rays emission would be a major discriminant . in this work we analyse recently obtained moderate to high resolution spectra of _ phl 293b _ which is a low luminosity , low metallicity hii galaxy that shows low intensity broad wings and blue shifted narrow absorptions in the hydrogen recombination lines , in order to try to shed some light on their origin . 2 gives a description of the data , 3 presents the results obtained from them and 4 is devoted to their analysis . a discussion is presented in 5 and the conclusions of our work are given in 6 . _ phl 293b _ , also known as hl 293b , kinman s dwarf , _ a2228 - 00 _ and _ sdss j223036.79 - 000636.9 _ is a very low luminosity galaxy ( m@xmath9 = -14.37 ; * ? ? ? * ) at a distance of 23.1 mpc obtained from the radial velocity @xcite taken from nasa / ipac extragalactic database ( ned ) corrected for virgo infall , great attractor and shapley , with a hubble constant of 73 km s@xmath10 mpc@xmath10 . its metallicity was first measured by @xcite who gave a value of 12+log(o / h ) = 7.78 and , more recently , by @xcite who derive a value of 12+log(o / h ) = 7.66 from sdss data . we have compiled spectroscopic data of _ phl 293b _ for 4 different epochs spanning more than 10 years . archival spectroscopic data have been extracted from dr7 of the sloan digital sky survey @xcite , _ sdss_(j223036.79 - 000636.9 ) ; eso _ vlt - uves _ [ programme i d 70.b-0717(a ) ] ; science verification ( sv ) _ vlt - x - shooter _ [ eso program 60.a-9442(a ) ] . to these data we have added new observations obtained with the isis spectrograph attached to the 4.2 m wht at the observatorio del roque de los muchachos in the island of la palma ( spain ) . the data from sdss were taken on the 22nd of august 2001 and cover the spectral range from 3800 to 9200 at a resolution r = 1800 - 2200 . the uves data were obtained on the 18th of november 2002 in both the blue and red arms covering a total wavelength range from 3100 to 6800 with 0.2 spectral resolution . _ x - shooter _ @xcite observations were performed during science verification on the nights of the 16th august and 28th september 2009 . the data simultaneously cover the spectral range from uv(@xmath11 ) to k@xmath12 ( @xmath13 ) and were reduced using the eso recipe execution tool ( esorex ) following standard procedures with minor adaptations for these sv data . isis data were obtained on the 30th of november 2011 using an eev12 detector attached to the blue arm of the spectrograph . the r300b grating was used covering the unvignetted wavelength range 4800 - 7200 ( centred at @xmath14 = 6000 ) , giving a spectral dispersion of 0.86 / pixel which combined with a slit width of 1 arcsec yields a resolution of about 3.4 . the observations were made at paralactic angle , at an airmass of 1.3 and with a seeing of 0.5 arcsec . several bias and sky flat - field frames were taken at the beginning and at the end of the night . in addition , two lamp flat - fields , before and after the observation , and one calibration lamp exposure were performed . the calibration lamp used was cune+cuar . the spectra were processed and analysed with iraf routines following standard procedures that include removal of cosmic rays , bias subtraction , division by a normalized flat - field and wavelength calibration . in the last step , the spectra were corrected for atmospheric extinction and flux calibrated . four standard star observations were used : bd+17 4708 , wolf1346 , g191b2b and feige34 allowing a good spectrophotometric calibration with an estimated accuracy of about 5 per cent . the photometric optical data was compiled from 3 different sources : the catalina sky survey ( css ) second release @xcite provides for _ phl 293b _ , accurate magnitudes for 80 nights from april 2005 to october 2012 . most nights have 4 different observations inside one hour total observing time . in addition , the sdss provides photometric parameters for the night of 22/08/2001 , and @xcite published their photometry for _ phl 293b _ obtained in october 1988 with the 3.5 m telescope of the calar alto observatory ( spain ) . @xcite provides the earliest brightness estimates based on the palomar sky survey ( pss ) plates and photographic images taken with the lick observatory 20-inch astrograph in 1949 and 1965 and at the prime focus of the lick 120-inch telescope . _ phl 293b _ was observed also using the acis - s camera on board the chandra x - rays observatory in 2009 september 25th . data cover the energy range 0.4 - 10kev and the exposure time was about 7.7 ks ( obsid 11294 ) . the blue to visual spectrum of _ phl 293b _ shows the usual strong emission lines found in star - forming galaxies : recombination lines from hydrogen and helium and collisionally excited lines of different ions : [ oii ] , [ oiii ] , [ nii ] , [ sii ] , [ siii ] , etc . the most striking feature of the spectral lines is the presence of broad components in the hydrogen recombination lines , already reported by izotov et al . ( 2007 ) , which are undetected in the corresponding helium lines . these broad components that can be seen in figures [ fig : xsh_halpha ] through [ fig : xsh_hgamma ] for the balmer lines : h@xmath8 , h@xmath15 , h@xmath16 and h@xmath17 , have fwhm between 1000 and 400 km / s @xcite and relatively low intensities . they are redshifted by about 50 km / s with respect to the narrow emission lines reference frame as determined by the centroid of the [ oiii ] @xmath3 4959 line . figure [ halphafit ] shows a detail of the multigaussian fit performed on the _ x - shooter _ spectrum h@xmath8 profile . details of measurements performed in sdss and _ x - shooter _ spectra are listed in tables [ sdssspec ] and [ xshspec ] , using the normal notation for equivalent widths as positive for lines in absorption , negative for emission . in what follows , we will call the reference frame defined by the [ oiii ] @xmath3 4959 line , the galaxy reference frame . the fluxes in the narrow component of h@xmath8 and h@xmath15 ( measured from the sdss spectrum ) are 4.86 and 1.54 @xmath18 10@xmath19 erg s@xmath10 @xmath20 respectively and the ratio of broad to narrow components in both lines is about 0.25 . errors in measurements depend strongly on flux uncertainties . individual errors are determined from profile fitting routines . these errors rely on a good determination of errors present in the spectra , which is directly obtained for the sdss spectra and had to be modeled for the _ x - shooter _ data . they are below 1% for radial velocity and 10% for dispersion determinations in strong narrow line profiles but can raise to tens of km s@xmath10for broader components . uncertainties in flux range from 3% in strong lines and/or components but can grow up to 20% in weak absorption lines . equivalent widths share a few percent to 20% uncertainty based on relatively low counts level for the continuum flux . from the _ x - shooter _ data the measured narrow line h@xmath8 and h@xmath15 fluxes are @xmath21 and @xmath22 erg s@xmath10 @xmath20 , respectively . the ratio h@xmath8/h@xmath15between the narrow line components is the same in both spectra within the observational errors . the ratio between the narrow and broad components for these lines are also the same within the observational errors . hence the differences in the flux measurements between the sdss and the _ x - shooter _ spectra can be attributed to aperture effects . since the aperture used for the sdss observations is larger than the one used for the _ x - shooter _ one , we hereafter use the sdss measurements for the line fluxes and ews . additional features of the balmer lines are an ultra - broad component with fwzi @xmath13 4000km / s redshifted by about 500 km / s and a p cygni like absorption blueshifted by 800 km / s with respect to the reference frame ( see table [ xshspec ] ) . [ emission line profiles ] to summarise , the complex h@xmath8 line profile shows a narrow component at the galaxy reference frame , a broad component fwhm @xmath13 1000 km / s redshifted by about 50 km / s with respect to the galaxy reference frame , an ultra - broad component with fwzi @xmath13 4000 km / s redshifted by about 500 km / s and a p cygni like narrow absorption blueshifted by 800 km / s . these features are also seen in the other balmer lines , except for the ultra - broad component which is seen only as a red wing . a weak red wing is also visible in [ oiii]@xmath35007 corresponding possibly to the ultra - broad component . all these features are visible both in the _ x - shooter _ data analysed here and in the sdss spectrum . the broad component represents about 20% of the total h@xmath8 emision flux . the blue shifted absorption is weaker in h@xmath8 than in the other balmer lines , although the continuum is difficult to fit in the blue wing of h@xmath8 . the hei lines show only narrow components . . here and in the following spectra , flux is measured in ergs@xmath10@xmath20.,scaledwidth=50.0% ] line profile . four components are clearly identified : a blue wing absorption blueshifted by 800 km s@xmath10 , a strong narrow component , a strong broad component and a relatively fainter ultra broad component , redshifted by 900 km s@xmath10 . the narrow spike at @xmath36620 is an artifact from the reduction process and discarded during profile fitting . , scaledwidth=50.0% ] , [ oiii ] @xmath35007 region . note the possibly very broad blue wing in [ oiii ] . the position of the feii multiplet 42 blueshifted by 800 km s@xmath10with respect to the galaxy rest frame is shown.,scaledwidth=50.0% ] . the positions of h@xmath16 and hei @xmath34471 blueshifted by 800 km s@xmath10with respect to the galaxy rest frame are shown.,scaledwidth=50.0% ] we have identified a conspicuous absorption feature at @xmath3 5183 . this absorption line has no emission counterpart and coincides with the wavelength of the feii multiplet 42 ( laboratory wavelength 5169.03 ) line blueshifted by the same 800 km / s displayed by the balmer p cygni like components . companion lines of the same multiplet blueshifted by the same velocity have also been found at @xmath2 5031 , 4939 and they are indicated in figure [ hbeta ] . further absorption lines are observed in the far red spectral region , corresponding to the caii triplet ( cat ) lines at @xmath2 8498 , 8542 , 8662 in the galaxy rest frame ( see figure [ catspec ] ) . although these lines , being close to the left of the adjacent paschen lines at @xmath2 8505 , 8548 , 8667 , can mimic p cygni profiles , the caii lines are clearly separated from the paschen ones . furthermore , there is no hint of a p cygni like component in the @xmath3 8601 paschen line . although the _ x - shooter _ infrared spectrum is not a very good one , we can identify hydrogen recombination lines of the paschen and brackett series , hei lines and a hint of the h@xmath23 line at 2.122 @xmath24 m rest frame . there is no evidence for broad components or p cygni like profiles in any of these lines . the wht - isis spectrum covering from @xmath34300 to @xmath37500 ( observations set - up described in 2.1 ) is shown in figure [ isisv ] . the strongest emission lines are labelled . . the positions of h@xmath17 and hei @xmath3 4026 blueshifted by 800 km s@xmath10 with respect to the galaxy rest frame are shown.,scaledwidth=50.0% ] [ cols="^,^,^,^,^,^ " , ] @|l|ccc|cccccc|ccc|@ + spectra & & h@xmath15 & & & & feii(42 ) & & & & & [ oiii]4959 & + & & & & & 1 & & & 3 & & & & + & wav . & eqw & vr & wav . & eqw & vr & wav . & eqw & vr & wav . & flux & eqw + & ( ) & ( ) & km s@xmath10 & ( ) . & ( ) & km s@xmath10 & ( ) . & ( ) & km s@xmath10 & ( ) . & & ( ) + sdss & 4847.3 & 1.1 & -866 . & 4912.2 & 1.0 & -715 . & 5154.5 & 0.67 & -843 . & 4959.1 & 164 & -171 . + uves & 4848.1 & 2.0 & -816 . & -&- & & 5154.7 & 0.88 & -832 . & 4959.0 & 192 & -208 . + xshooter & 4848.0 & 1.1 & -822 . & 4910.8 & 0.91 & -800 . & 5154.5 & 0.68 & -843 . & 4959.0 & 186 & -197 . + isis & 4847.9 & 1.0 & -829 . & 4912.0 & 0.71&-727 & 5154.7 & 0.57 & -832 . & 4959.1 & 220 & -211 . + we have extracted from the catalina sky surveys ( css ) data release 2 database @xcite 353 photometric points for _ phl 293b _ corresponding to 91 observing nights over the 8.4 years from 16/5/2005 to 27/9/2013 . most nights have 4 independent observations inside a 60 minutes total span . nightly averages are shown ( plus symbols ) in figure [ catalina ] and have a typical r.m.s . scatter of 0.040 magnitudes . the yearly averages are shown with filled rhomboids . the average brightness for the whole dataset , indicated with a thin line , is @xmath2517.056 @xmath26 0.004 . the sdss published petrosian r magnitude is [email protected] @xmath28 0.01 which combined with the colour g - r=0.03 and the sdss colour transformations yield a petrosian v magnitude [email protected] @xmath28 0.01 observed on the 22/08/2001 . the sdss value is indicated with an x in figure [ catalina ] . @xcite provided an earlier photometric point for the object . their october 1988 observation with the 3.5 m telescope of the calar alto observatory gives v=17.02@xmath28 0.01 . @xcite published the earliest magnitude estimates for _ phl 293b _ : [email protected] . we quote from his paper : a very rough estimate of the b magnitude is 17.7 on the sky survey plates and about a half magnitude fainter on the 120-inch plates " . kinman attributes this difference to the scale difference between the pss and the lick 120-inch plates given that there is no variability between 1949 and 1965 in plates taken with the lick 20-inch astrograph . bearing in mind that they are based on eye estimates of the pss plates from 1949 and photographic images taken at the prime focus of the lick observatory 120-inch telescope in 1965 , errors of about @xmath280.5 magnitudes are expected . there is no detection of x - rays emission in the acis - chandra image . the point source sensitivity of this camera in the 0.4 to 6 kev energy range is @xmath31 erg@xmath20s@xmath10 for an exposure time of 10 ks ( table 6.1 of * ? ? ? it gives for the _ phl 293b _ chandra observation of 7.7 ks a point source sensitivity of about @xmath32 erg@xmath20s@xmath10 . _ phl 293b _ is a low luminosity ( m@xmath9 = -14.37 ; * ? ? ? * ) , low metallicity [ 12+log(o / h ) = 7.61 ; izotov et al . 2007 ] bcg . the comparison of our measured narrow h@xmath8 to h@xmath15 flux ratio with the case b recombination theoretical value yields a logarithmic extinction at h@xmath15 of c(h@xmath15 ) = 0.12 . for our sdss flux measurements and the assumed distance of 23.1 mpc ( ned ) this gives an extinction corrected h@xmath15 luminosity of 1.3 @xmath18 10@xmath33 erg s@xmath10 . the total lyman continuum photon rate for a given region can be derived from the h@xmath8 luminosity ( e.g. * ? ? ? * ) : @xmath34 taking into account that the number of lyman continuum photons per unit mass of the ionising population decreases with age , and using the equivalent width of h@xmath15 ( ew(h@xmath15 ) ) as an age parameter @xcite , it is possible to calculate the mass of the ionising star cluster under certain assumptions about the initial mass function ( imf ) . for a salpeter imf with lower and upper mass limits of 0.8 and 120 m@xmath7 respectively , we have @xcite : @xmath35=44.48 + 0.86 ~log[|ew(h\beta)|]\ ] ] where the vertical bars indicates modulus . for an ew(h@xmath15 ) = -100 , that represents the total ew of h@xmath15 ( see table [ sdssspec ] ) , the resulting ionising star cluster mass is m@xmath36= 1.66 @xmath1810@xmath6 m@xmath7 . according to the models of @xcite , an ew(h@xmath15 ) = -100 may be produced by a cluster of 3.43@xmath1810@xmath6 m@xmath7 with an age of about 7@xmath1810@xmath37 yr , for a metallicity of z=0.0004 or a cluster of 2.06@xmath1810@xmath6 m@xmath7 with an age of about 5@xmath1810@xmath37 yr , for a metallicity of z=0.004 . the ionised gas metallicity of _ phl 293b _ is about z=0.002 , i.e. in between the values of the two models quoted , hence intermediate values for the mass of the ionising cluster and the age can be expected . the observed equivalent width may be affected by the continuum of an underlying older stellar population or a bright stellar component like for example a sn or an lbv ; therefore it has to be considered as a lower limit for the equivalent width of the ionising cluster , and our mass and age estimates are in consequence upper limits . the ir caii triplet at nominal @xmath2 8498 , 8542 , 8662 is clearly seen in absorption in the spectrum of _ phl 293b _ ( see figure [ catspec ] ) at the same redshift shown by the narrow emission lines . the cat index definition @xcite that uses the two strongest lines of the triplet , measures approximately 2 , which , if taken literally , suggests [ fe / h]@xmath38 for giant or supergiant stars ( see e.g. * ? ? ? this index has to be corrected for the contribution of the nebular continuum . for the model of log(age)=6.86 and z=0.0004 chosen above , this contribution amounts to 0.55 per line and hence the corrected cat index would be 3.1 , consistent with the values calculated by @xcite from single stellar population synthesis models for the lowest abundance they consider , z=0.004 . for the assumed mass , age and metallicity of the ionising cluster the models by @xcite predict no wr stars and only one or two rsg stars . _ phl 293b _ is unusual as an hii galaxy due to the presence of broad permitted and narrow forbidden lines in its optical spectrum . the absence of a strong broad forbidden [ oiii]@xmath35007 component indicates that the density of the gas producing the broad component is high enough to collisionally de - excite [ oiii]@xmath35007 ( a condition for which the electron density has to be n@xmath39 ) . an upper limit to the mass of the ionised gas in the broad line region can be estimated using the observed l(h@xmath8 ) @xcite and the electron density lower limit , @xmath40 therefore , @xmath41 the ionized gas mass of the narrow line region is , assuming an electron density of 100 @xmath42 ( derived from the [ sii ] doublet ratio in our isis data ) , @xmath43 similar to the ionising star cluster mass determined in 4.1 . from the nightly averaged data ( pluses ) in figure [ catalina ] no variability is detected . dividing the data into 9 groups of one year duration each ( filled diamonds in the same figure ) reduces the r.m.s . scatter to 0.009 per group as expected for a non variable sample and again no systematic trend in the luminosity is apparent ; furthermore a cursory analysis indicates that there is no periodic variation . based on the css data we can infer that at 3 @xmath1 level any long term variability , i.e. over a few years , can not be larger than 0.02 magnitudes and that any medium term variability , i.e. over a few months , should be less than 0.04 magnitudes . further still , including the data from @xcite , @xcite and sdss we can safely conclude that there is no variability at the level of few tenths of magnitude over a period of 25 years . to generate relatively narrow p cygni like absorption profiles such as the ones observed in _ phl 293b _ , the absorbing material must cover a substantial fraction of the continuum source , in this case a young stellar cluster several parsecs in size . these absorptions may be produced in an expanding dense supershell created by the interaction of combined stellar winds with the circumcluster medium . an expanding supershell has been postulated to explain the blueshifted metal absorption detected in the uv spectrum of several nearby and high redshift hii galaxies @xcite . alternatively the observed blueshifted narrow absorptions could be associated with a single star in a very luminous transient phase like an lbv or a sn iin . the possibility that the p cygni like profiles in _ phl 293b _ are originated in the powerful wind of an lbv has been extensively discussed by @xcite . these authors suggested that the absorptions seen at @xmath3 4939 and @xmath35031 are in fact the blue shifted absorptions associated with the hei lines at rest wavelengths @xmath34921 and @xmath35016 in accordance with the p cygni profile of the balmer lines . figure [ multiple230 ] shows the spectral regions around the hei optical lines from the _ x - shooter _ data . the dotted line corresponds to a blueshift of 800 km s@xmath10 consistent with what is observed in the balmer lines . the top row displays all the hei lines with the same flux scale in order to show the relative importance of the emissions . the bottom row shows the spectra normalised to the same amplitude for the emission line and the continuum to illustrate the differences in strength of the associated absorptions . it is clear that while the hei @xmath3 4921 emission is the weakest , the absorption to the blue is the strongest and furthermore it is redshifted with respect to the 800 km s@xmath10 line . @xcite attributed this difference in velocity between balmer and hei lines p cygni like profiles to the different formation depth of the core of each line in an accelerated wind . however there are several difficulties with this suggestion . firstly the broad absorption to the blue of hei @xmath34921 has an equivalent width similar to that of the p cygni like absorption of @xmath44 , probably too large to be associated with such a weak hei line . secondly , it is not clear why much stronger hei lines like @xmath25876 , 6678 and 7065 do not show a blueshifted absorption as can be seen in figure [ multiple230 ] , although hei 5876 seems to have , at the noise level , an associated weak absorption . an important clue for an alternative interpretation of the absorptions seen at @xmath3 4939 and @xmath35031 is given by the discovery of a previously unidentified absorption at @xmath455183 ( fig . [ hbeta ] ) corresponding to @xmath35155 in the rest frame of the forbidden lines and @xmath35169 in the rest frame of the blue shifted balmer absorptions . this absorption is real and is present in the spectra from sdss , vlt - uves , vlt-_x - shooter _ and wht - isis that were taken in four different epochs ( see fig.[multiple2 ] ) . although the @xmath455183 absorption could be attributed to mg2 @xmath35172 present in late star spectra , this possibility can be readily dismissed given that the observed wavelength would imply a shift of 1050 km s@xmath10with respect to the galaxy rest frame and definitely inconsistent with the presence of the caii triplet lines at the galaxy velocity rest frame . a more plausible interpretation is that we are in fact detecting the feii multiplet ( 42 ) @xmath2 4923.93,5018.44,5169.03 blueshifted by the same amount as the balmer absorption features i.e. @xmath13800 km / s to @xmath2 4911,5005,5155 . these absorptions have already been detected in a variety of transient luminous objects like the lbv candidate ngc 2366v1 or sn iin like 1995 g , 1999el or 1999eb among others ( see figure 2 of * ? ? ? lbvs represent a short ( @xmath13 10@xmath46 yr ) phase in the evolution of massive stars during which the star undergoes luminous outbursts associated with large mass loss events . during these short phases lasting few years the lbv can reach , in extreme cases , 10@xmath37 l@xmath47 or mv @xmath13 -10 and luminosity variations of @xmath13 1 mag . the luminosity changes are in general correlated with spectral changes . the typical absolute magnitude of an lbv in quiescent state is mv @xmath13 -6 . in general information about lbvs is very sparse given that only 35 candidates have been identified in our galaxy @xcite . only two of these , @xmath48 car and p cygni , were observed during an outburst . extragalactic examples are known due to either photometric or specific spectral variability . the mechanism responsible for the outbursts remains unidentified . it is important to mention that the wind terminal velocities quoted by ( * ? ? ? * km s@xmath10 ) are significantly larger than the usual range for lbvs ( 100 - 250 km s@xmath10 up to 500 km s@xmath10 for @xmath49 car ) @xcite . this discrepancy is attributed by izotov et al . to _ phl 293b _ very low metallicity . there are two striking examples of lbvs in low metallicity dwarf galaxies , one in ngc 2366 and the other one in ic1613 . ngc 2366 is a giant hii region ( similar to 30 dor in the lmc ) located in the metal poor ( 12+log(o / h)=7.9 , i.e. o / h @xmath500.16 solar , * ? ? ? * ) dwarf galaxy ngc 2363 , a member of the m81 group . the variable star ngc 2366-v1 showed a big eruption similar to an lbv with an amplitude of 3.5 mag in 4 years reaching a maximum of about mv=-10.2 @xcite . after the outburst it started a decline of about 0.25 magnitudes in 6 years accompanied by @xmath510.2 mag dips lasting about a year each . regarding its spectral evolution , the fe ii lines ( 4923 , 5018 , and 5171 ) present between 1997 and 2000 disappeared later while the hei transitted from absorption to emission and the p cygni feature in h@xmath8 disappeared @xcite . these authors reported an average wind terminal velocity of @xmath13 300 km s@xmath10 for ngc 2366-v1 . ic 1613 is a dwarf irregular galaxy in the local group with a metallicity of log(o / h)@xmath5212 = [email protected] as determined from its b - supergiants @xcite . nebular abundance estimates vary between 7.60 and 7.90 @xcite , similar to the metallicity values shown by ngc2363 and _ phl 293b_. the peculiar variable star v39 @xcite was studied in detail by @xcite . its spectrum shows strong balmer and feii p cygni profiles combined with weak hei emission similar to the early stages of the evolution of the ngc2366-v1 lbv star . the equivalent width of the h@xmath8 p cygni profile was @xmath5145 in emission and @xmath132.3 in absorption with a blue shift or wind terminal velocity of @xmath51400 km / s . its absolute magnitude reached mv = -8.1 . the feii ( 42 ) absorptions in these two transient objects show equivalent widths of about 2 while the observed value in the spectrum of _ phl 293b _ is just under 1 . the contribution of the nebular continuum at the feii wavelengths is 25 per cent ( see above ) , which increases this value to 1.3 . this implies that under this hypothesis , about 30 percent of the observed continuum should be due to the transient object itself . but , the absolute magnitude of _ phl 293b _ is mv = -14.5 while a very luminous lbv during outburst would reach mv @xmath50 -10 , thus contributing less than 2 percent of the total luminosity i.e. more than one order of magnitude smaller than what is needed to explain the observed strength of the feii absorptions ( see table [ feii ] ) . it seems important to notice that these two lbvs , both located in low metallicity environments , show wind terminal velocities that are typical of lbv ( 100 - 250 km s@xmath10 ) while @xmath49 car , belonging to a region that is not metal poor ( 12 + log o / h=8.36 @xmath260.03 ; * ? ? ? * ) , has a larger wind terminal velocity ( up to 500 km s@xmath10 ) . wind velocities between 100 km / s and 1000 km / s are reported for transient events or supernova impostors " that could be related to powerful eruptions of lbvs @xcite , and in a very interesting recent work @xcite report an unusual variable source in the nearby dwarf galaxy mrk 177 ( ugc 239 ) that they suggest can be explained as an lbv eruption followed by a sn iin like event in 2001 . @xcite discussed the x - rays properties of the galactic lbvs . they found that their x - ray luminosities in the 0.5 - 8 kev range are between @xmath54 and @xmath55 ergs@xmath10 . the estimated minimum luminosity for a point source to be detected in the acis - chandra image of _ phl 293b _ is about @xmath56 ergs@xmath10 ( see section [ photometry ] ) about three orders of magnitud higher than the largest reported values for galactic objects . one fact supporting the hypothesis of the presence of a transient object would be the detection of long term variability in the flux from _ phl 293b_. however , as stated in [ photvar ] , we can conclude that there is no variability at the level of a few hundredths of magnitude over 25 years . this implies that any long term variable component has to be fainter than v @xmath5721.5 . for a distance modulus ( m - m)@xmath5831.8 this upper limit means that any long term variable component has to be fainter than m@xmath59 - 10.5 . however , this does not exclude the possibility that we are observing such a transient object during a non variable phase . the caveat in that case is that , if the variable component is that faint in relation with the ionising cluster , it will be difficult to explain the observed strength of the balmer and feii absorptions . an alternative scenario for the broad lines and pseudo p cygni profiles in _ phl 293b _ is that they are originated in an evolved sn type iin . it is usually assumed that sn iin explode inside the circumstellar material previously ejected in the form of a slow dense wind by the red supergiant sn progenitor . what is observed is the result of the interaction between the ejecta and this slowly expanding dense circumstellar medium ( csm ) which transforms the mechanical energy of the ejecta into radiation . these events are known as sn iin or compact supernova remnants ( see e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? . in this interaction , radiative cooling can become catastrophic and these remnants rapidly radiate most of the mechanical energy in a relatively short time scale and so they are highly luminous . the energy output is mostly in the extreme uv and x - ray regions of the spectrum . the optical continuum is a mixture of young stars and the emission from the sn iin with fast moving high density gas producing the broad line emission . the spectra of some sn iin associated with hii regions like 1995 g , 1999el or 1999eb among others , present broad balmer components with relatively narrow p cygni like absorptions plus narrowish feii absorption with multiplet ( 42 ) being the strongest ( see figure 2 of * ? ? ? * ) . a significant constraint on the sn iin / compact remnant scenario is the long term lack of variability in _ phl 293b_. the secular variability of the broad lines in this scenario comes from the evolution of the interaction of the ejecta and the circumstellar medium , i.e. the evolution of the compact remnant . the evolution of the blue light - curve can be parametrized as @xmath60 @xcite with its peak luminosity being a function of the circumstellar density ( @xmath61 ) . these are analytical approximations and their predicted values should be taken as such . we note that the observed absolute magnitude of _ phl 293b _ ( @xmath62 ) is fainter than the absolute magnitude of a single compact remnant near peak ( @xmath63 for @xmath64 @xmath42 ) . this , together with the upper limit for the rate of change for the last @xmath50 30 years , is only consistent with the very late stages of the evolution of a compact remnant , i.e. when its light curve flattens . using the @xmath65 time dependence of the luminosity as an indicator , we find that a compact remnant of age @xmath66 years evolving in a medium with density @xmath67 @xmath42 would have shown a decline of about 1.5 magnitudes since maximum and of about 0.1 magnitudes in the last 10 years . we can estimate its line luminosities using the models of @xcite . the observed luminosity of the broad component of h@xmath8 is : @xmath68 from the analytical solutions of @xcite and tables 1.1 and 1.2 of @xcite it is possible to estimate that a compact remnant evolving in a medium with @xmath67 @xmath42 would have a shock velocity of 800 km / s at an age defined by @xmath69 of about 112 yr . predicted values for the shock parameters and h@xmath8 corresponding to such age are : @xmath70 @xmath71 @xmath72 @xmath73 @xmath74 and the fwzi of the broad component is about twice the shock velocity or 1600 km s@xmath10 . given its temperature , the remnant is expected to emit in x rays up to about few kev with a luminosity just below @xmath75erg s@xmath10 . these values are too large compared with the observed upper limit of @xmath76erg s@xmath10 . furthermore , the expected drop in luminosity 7.5 years later i.e. at an age of 119.5 yr is 10 % or about 0.1 magnitudes and the drop in luminosity in 20 years is about 0.2 magnitudes , thus we can safely conclude that the observed upper limit on the variability of _ phl 293b _ makes it very difficult for the observed absorptions and broad emission to be originated in a sn type iin . in summary , we have evidence that an old compact remnant could explain the observed broad lines and the dense shell responsible for the blue - shifted absorptions in _ phl 293b_. but , while the observed line properties are in rough agreement with a compact remnant slightly older than 100 years , this age is much smaller than the age derived from the absolute blue magnitude limit imposed by the lack of variability and furthermore the expected variability in h@xmath8 at 100 years is also too large compared with the limits deduced from the data . to see if there is some information in the data regarding the spatial distribution of the absorbers , we compared in table [ feii ] the ew measurements of the hydrogen absorptions that are supposed to cover most of the source of the continuum , with the ones of the feii lines taking as reference the ew of [ oiii]@xmath35007 that is expected to extend beyond most of the ionising cluster continuum but to be still inside the host galaxy continuum . the ew of the h@xmath15 and feii absorptions are , within errors , basically the same in all four sets of data . to check for aperture effects we measured the ew of [ oiii]@xmath35007 and the results are displayed in table [ feii ] . as expected there is a slight increase in the ew measured with the smaller _ x - shooter _ aperture ( 1 arcsec ) compared with the 3 arcsec circular aperture of the sdss spectrum ( although determination of faint continuum contribution can also account for such variations among spectra ) . we have also checked for possible variability in the broad emission components . the broad and narrow components of h@xmath8 are listed in tables [ sdssspec ] and [ xshspec ] . their ratio of the broad to total h@xmath8 flux is , for both sdss and x - shooter observations , the same within errors , about 0.21 showing no indication of variability as already discussed . the lack of variability imposes a strong constraint for scenarios involving discrete transient sources like lbv or sn iin . at the same time the lack of variability suggests that the origin of the blue shifted absorptions could be associated with a global process involving a large part of the ionising cluster . furthermore we have to address the fact that what we are witnessing in the feii lines and perhaps also in the balmer series in _ phl 293b _ are blue shifted absorptions without the associated narrow emissions typical of classical p cygni profiles . this lack of associated emissions indicates that the material responsible for the detected absorptions , while covering a large part of the continuum source , is not an extended circum - cluster medium ( ccm ) . if it was it would have produced the so far undetected associated emission lines . luminous regions of star formation in the nearby universe like orion or ngc 3606 in our galaxy , 30-doradus in the lmc or ngc 604 in m33 , to name just a few of the nearest ones , all show gaseous filaments surrounding the ionising cluster . this fact and the observed lack of variability over many years suggest a scenario where the blueshifted absorptions found in _ phl 293b _ are formed in an expanding supershell generated by the cluster wind interacting with a moderately dense gas closely surrounding the star - forming region . the cluster wind drives a shock into the ccm and a recombined dense shell is formed in the post - shock region . this dense shell expanding at 800 km / s produces the observed blue shifted absorptions in h and fe following catastrophic radiative cooling in a scenario similar to the case of the compact supernova remnant described above . the main difference being that instead of an instantaneous input of energy and mass the wind scenario involves a constant rate . as the cluster wind shocks the surrounding ism it will generate , given adequate conditions , a post shock high density supershell that will be ionised by the cluster uv output modified by the radiative shock emission . as already mentioned , a supershell scenario has been postulated to explain the diversity of ly@xmath8 profiles observed in starbursts ( see * ? ? ? * ; * ? ? ? * and references therein ) where also blueshifted uv ism metal absorption lines are observed . in this scenario the ly@xmath8 emission and the blueshifted metal absorption lines are formed in an expanding supershell generated by the interaction of the combined stellar winds and supernova ejecta from the young starburst , with an extended gaseous halo . among the many intrinsic parameters of a star - forming region that can affect the properties of the observed emission line profiles , velocity , density and ionisation distributions of the gas along the line of sight are by far the dominant ones . the expulsion of dust and gas from young clusters due to the action of stellar winds and supernovae has been discussed by many authors ( see for example * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) as well as the resultant cluster infant mortality " effect ( see for example * ? ? ? [ can we estimate the supershell column densities , its mass or fe abundance ? ] the strength of a weak absorption line depends on its oscillator strength @xmath77 , its wavelength @xmath3 and column density n. if the element is mostly in one ion state and there is no line saturation the relation between the column density n and the ew@xmath78 is given by @xcite : @xmath79 from the _ x - shooter _ spectra of _ phl 293b _ we have measured ew = 0.91 and 0.68 for the feii lines 4923 and 5169 respectively . the corresponding oscillator strengths are 0.0104 and 0.0226 @xcite . with these values for the oscillator strength we can compute the @xmath80 column density , @xmath81 and @xmath82 for the feii lines 4923 and 5169 respectively . the @xmath80 column density values are in good agreement with each other within observational errors . if the supershell is formed from shocked ccm , it must have abundances similar to those observed in the ionised gas . if fe / o is solar , this implies for the supershell fe / h @xmath83 and therefore the hydrogen column density associated with @xmath80 is , @xmath84 or @xmath85 if fe / o is half solar . given that we do not know the hydrogen ionization fraction , these values should be taken as an estimate of the total , i.e. neutral plus ionized , hydrogen column density . following the same procedure we can estimate from the balmer absorptions the column density of neutral hydrogen . for a measured ew = 6 and 2 for h@xmath8 and h@xmath15 the column densities estimates of neutral hydrogen are respectively : @xmath86 and @xmath87 the comparison of the column density of `` total '' and neutral hydrogen suggests a high degree of ionisation in the gas that is producing the blueshifted absorptions . an interesting variant of this scenario is provided by the super star cluster ( ssc ) cooling wind model of @xcite , @xcite and collaborators . this group has shown that in the case of a very massive and extremely compact young cluster its wind may radiatively cool in a stationary condition close to the outer radius of the cluster . this cool ejecta should be the responsible agent for the broad emission and blue shifted absorptions observed ( see figure 4a of * ? ? ? we expect to see soon detailed model calculations for the supershell and stationary cooling wind scenarios and the comparison of the theoretical predictions with the observed parameters of _ phl 293b_. if the feii absorptions are formed in a stationary cooling wind the metal abundances will be those of the combined ejecta of stellar winds and sne . in this scenario fe / h will be much higher than that of the ism in _ phl 293b _ and consequently the estimated @xmath88 column density will be proportionally lower than those estimated above . assuming solar abundances for the cluster wind , the estimated ionized hydrogen column density will be : @xmath89 a question that is raised in these relatively long lived cluster wind scenarios is : why is it that among several thousand hii galaxies known , only _ phl 293b _ is known to have simultanously narrow blue shifted balmer and feii absorptions ? or equivalently , why is this type of event so rare ? the answer may be related to the fact that we are witnessing an event that produces weak narrow absorptions that are detectable only with high dispersion and high s / n spectra . we have analysed spectra of the low metallicity starforming galaxy _ phl 293b _ corresponding to four epochs obtained in four different combinations of telescopes and spectrographs , the sdss , vlt - uves , vlt-_x - shooter _ and wht - isis . we find moderate narrow absorption components in the balmer series blueshifted by 800km / s . we detected also the ir caii triplet lines at the galaxy velocity rest frame , i.e. the rest frame defined by the ionised gas narrow emission lines . we also find narrow absorptions at @xmath45 4911 at @xmath45 5004 ( partially filled up by [ oiii ] @xmath3 5007 ) and a previously unidentified absorption at @xmath455183 . we interpret these narrow absorptions as the feii multiplet ( 42 ) @xmath2 4923.93,5018.44,5169.03 , similar to those detected in a variety of transient luminous objects like the lbv candidate ngc 2366 v1 or sn type iin 1995 g , 1999el or 1999eb , blueshifted by the same amount as the balmer absorption features i.e. @xmath13800 km / s to @xmath2 4911,5005,5155 . the analysis of the photometric data provided by the css puts a strong upper limit to the possible variability of _ phl 293b_. basically any optical yearly variability allowed by the data should be smaller than 0.02 magnitudes at the 3 sigma level in the 8.5 years between april 2005 and september 2013 . the possibility of any secular trend in the luminosity of _ phl 293b _ in the last 15 years ( considering only ccd data ) is also limited to at most 0.02 magnitudes at the 3 sigma level . this lack of variability and the observed strength of the balmer and feii absorptions rule out any transient of the type of an lbv or sn type iin as the origin of the blue shifted absorptions of h and feii . the evidence points to either a young and dense expanding supershell or a stationary cooling wind , both driven by the young cluster wind . we suggest that the observed absorptions and broad balmer emissions are originated in one of these scenarios which seem capable of explaining the observed spectral features , the constant ( within errors ) photometric history and the rarity of the phenomenon . many starbursts , both nearby and at high z , show blueshifted far uv ism narrow absorption lines . on that basis , coupled with the observational evidence , we expect that the far uv spectrum of _ phl 293b _ will show blue shifted ism lines and moderate ly@xmath8 emission with perhaps a p cygni - like profile . we have to bear in mind that there are not many hii galaxies known to show prominent broad wings in their emission lines and none with the quality of the data presented here for _ phl 293b_. this leaves open the possibility that we are witnessing an event not detected in other systems due to low s / n data . even if it is true that we havent been actively looking for them , and that data of the quality and variety discussed here is not widely available , it is still puzzling that fe absorptions have not been detected in other star forming hii galaxies . a search should be performed in high dispersion , high signal - to - noise spectra of hii galaxies to investigate the presence of supershells in starbursts with or without strong broad components in the balmer lines . until such data is gathered and analysed , the fact that we do not see many hii galaxies showing spectra similar to _ phl 293b _ means that this may be a relatively short duration stage in the evolution of compact and massive stellar clusters , lasting perhaps only a few thousand years . et and rt are grateful to the mexican research council ( conacyt ) for support under grants 2008 - 103365 and 2010 - 01 - 155046 . financial support for this work has also been provided by the spanish _ ministerios de educacin y ciencia and ciencia e innovacin _ under grants aya2007 - 67965-c03 - 03 and aya2010 - 21887-c04 - 03 . our team enjoyed the hospitality of the institute of astronomy , cambridge , of the departamento de fsica terica of the universidad autnoma de madrid and of the observatorio astronmico of the universidad nacional de la plata ( argentina ) during fruitful visits when this paper was started and developed . we thank vahram chavushyan for suggesting the identification of the feii absorption features in @xmath24911 , 5155 and discussions with daniel kunth helped to clarify the estimates of the analysis of these feii features . greatly enjoyed were discussions with guillermo tenorio - tagle , sergiy silich and sergio martnez - gonzlez , from which a picture of the stationary wind scenario started to emerge . we are grateful to vital fernndez for helping with the reduction of the wht data . we thank the _ x - shooter _ data reduction team for valuable help and suggestions and an anonymous referee whose questionings greatly improved the clarity of the paper . the css survey is funded by the national aeronautics and space administration under grant no . nng05gf22 g issued through the science mission directorate near - earth objects observations program . the crts survey is supported by the u.s . national science foundation under grants ast-0909182 . 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 . 99 abazajian k. n. , et al . , 2009 , apjs , 182 , 543 aretxaga i. , terlevich r.,1994 , mnras , 269 , 462 aretxaga i. , cid fernandes r. , terlevich r. , 1997 , mnras , 286 , 271 aretxaga i. , benetti s. , terlevich r.j . , fabian a.c . , cappellaro e. , turatto m. , della valle m. , 1999 , mnras , 309 , 343 asplund m. , grevesse n. , sauval a. j. , scott p. , 2009 , ara&a , 47 , 481 bastian n. , goodwin s. p. , 2006 , mnras , 369 , l9 bresolin f. , urbaneja m. a. , gieren w. , pietrzyski g. , kudritzki r .- p . , 2007 , apj , 671 , 2028 cairs l. m. , vlchez j.m . , gonzlez prez j.n . , iglesias - pramo j. , caon n. , 2001 , apjs , 133 , 321 castaeda h. o. , vilchez j. m. , copetti m. v. f. , 1990 , apj , 365 , 164 chugai n.n . , 1991 , mnras , 250 , 513 chugai n.n . , blinnikov s. i. , cumming r. j. , lundqvist p. , bragaglia a.,filippenko a. v. , leonard d. c. , matheson t. , 2004 , mnras , 352 , 1213 clark j.s . , larionov v.m . , arkharov a. , 2005 , a&a , 435 , 239 daz a. i. , terlevich e. , pagel b. e. j. , vilchez j. m. , edmunds m. g. , 1987 , mnras , 226 , 19 daz a.i . , terlevich e. , terlevich r. , 1989,mnras , 239 , 325 daz a.i . , terlevich e. , terlevich r. , 1989,mnras , 239 , 325 daz a. i. , castellanos m. , terlevich e. , luisa garca - vargas m. , 2000 , mnras , 318 , 462 di carlo e. , massi f. , valentini g. , et al . , 2002 , apj , 573 , 144 dodorico s. et al . , 2006 , spie , 6269 , 626933 dottori h. a. , 1981 , ap&ss , 80 , 267 drake , a.j . 2009 , apj , 696 , 870 drissen l. , roy j .- , robert c. , 1997 , apj , 474 , l35 fassia , a. , et al . 2001 , mnras , 325 , 907 filippenko a.v.,1989 , aj , 97 , 726 filippenko a.v.,1997 , ara & a , 35 , 309 french h. b. , 1980 , apj , 240 , 41 garca - vargas m. l. , moll m. , bressan a. , 1998 , a&as , 130 , 513 geha m. , blanton m. r. , masjedi m. , west a. a. , 2006 , apj , 653 , 240 giridhar s. , arellano ferro a. , 1995 , rev.mex.a.a . , 31,23gtt07 gonzlez - delgado r. m. , et al . , 1994 , apj , 437 , 239 goodwin s. p. , bastian n. , 2006 , mnras , 373 , 752 grosbl p. , dottori h. , 2013 , a&a , 551 , l13 hgele g.f . , daz a.i . , cardaci m.v . , terlevich e. , terlevich r. , 2009 , mnras , 396 , 2295 haro g. , luyten w. j. , 1962 , biton , 3 , 37 herrero a. , garcia1 m. , uytterhoeven k. , najarro f. , lennon d. j. , vink j. s. , castro n. , 2010 , a&a , 513 , a70 izotov y. i. , dyak a. b. , chaffee f. h. , foltz c. b. , kniazev a. y. , lipovetsky v. a. , 1996 , apj , 458 , 524 izotov y. i. , thuan t.x . , 2009 , apj , 690 , 1797 izotov y. i. , thuan t.x . , stasiska g. , 2007 , apj , 662 , 15 izotov y. i. , guseva n. g. , fricke k. j. , henkel c. , 2011 , a&a , 533 , a25 kinman , t.d . , 1965 , apj , 142 , 1241 koss m. , blecha l. , mushotzky r. , hung ch.l . , veilleux s. , trakhtenbrot b. , schawinski k. , stern d. , smith n. , li y. , and 5 coauthors , 2014 , arxiv:1401:6798v1 kramida , a. , ralchenko , yu . , reader , j. and nist asd team ( 2012 ) . nist atomic spectra database ( version 5.0 ) , [ online ] . available : http://physics.nist.gov/asd kunth d. , & stlin g. , 2000 , a&ar , 10 , 1 lee h. , grebel e. k. , hodge p. w. , 2003 , a&a , 401 , 141 leitherer , c. , allen , r. , altmer , b. , et al . 1994 , apj , 428 , 292 leitherer c. , heckman t. m. , 1995 , apjs , 96 , 9 macchetto f. , colina l. , golombek d. , perryman m. a. c. , di serego alighieri s. , 1990 , apj , 356 , 389 mas - hesse , m.j . , kunth , d . , tenorio - tagle , g . , leitherer , c . , terlevich r.j . , terlevich , e . , 2003 , apj , 598 , 858 melnick j. , tenorio - tagle g. , terlevich r. , 1999 , mnras , 302 , 677 moll m. , garca - vargas m. l. , bressan a. , 2009 , mnras , 398 , 451 moll m. , garca - vargas m. l. , bressan a. , 2009 , mnras , 398 , 451 morton , d.,1991 , apjs , 77 , 119 mould j. r. , et al . , 2000 , apj , 529 , 786 naz y. , rauw g. , hutsemkers d. , 2012 , a&a , 538 , a47 petit v. , drissen l. , crowther p. a. , 2006 , aj , 132 , 1756 pilyugin l. s. , ferrini f. , shkvarun r. v. , 2003 , a&a , 401 , 557 roy j .- r . , aube m. , mccall m. l. , dufour r. j. , 1992 , apj , 386 , 498 silich s.,tenorio - tagle g.,rodrguez - gonzlez a. 2007 , apj , 610 , 226 sandage a. , 1971 , apj , 166 , a70 smith , n. , li , w. , silverman , j. m. , ganeshalingam , m. , & filippenko , a. v. 2011 , mnras , 415 , 773 stasiska g. , cid fernandes r. , mateus a. , sodr l. , asari n. v. , 2006 , mnras , 371 , 972 tenorio - tagle , g. , silich , s. a. , kunth , d. , terlevich , e. , terlevich , r. 1999 , mnras , 309 , 332 tenorio - tagle g. , wnsch r. , silich s. , palous j. 2007,apj 658 , 1196 terlevich e. , daz a.i . , terlevich r. , 1990,mnras , 242 , 271 terlevich e. , daz a.i . , terlevich r. , 1990,mnras , 242 , 271 terlevich e. , daz a. i. , terlevich r. , gonzlez - delgado r. m. , prez e. , garca vargas m. l. , 1996 , mnras , 279,1219 terlevich r. , tenorio - tagle g. , franco j. , melnick j.,1992 , mnras , 255 , 713 terlevich r. , 1994 , in : circumstellar media in late stages of stellar evolution , proceedings of the 34th herstmonceux conference held in cambridge , july 12 - 16 , 1993 . edited by r.e.s . clegg , i.r . stevens and w.p.s . cambridge , uk : cambridge university press , p.153 turatto m. , 2003 , in : supernovae and gamma - ray bursters . ed . k. weiler . , lecture notes in physics , vol . 598 , p.21 - 36 turatto m. , 2005 , in : cosmic explosions , on the 10th anniversary of sn1993j . proceedings of iau colloquium 192 . edited by j.m . marcaide and kurt w. weiler . springer proceedings in physics , vol . berlin : springer , 2005 . , p.151 tutukov a. v. , 1978 , a&a , 70 , 57 vernet j. , dekker h. , dodorico s. , kaper l. , kjaergaard p. , hammer f. , randich s. , zerbi f. , groot p. j. , hjorth j. , and 80 coauthors , 2011 , a&a , 536a , 105 vestergaard m. , peterson b.m . , 2006 , apj , 641 , 689

this paper introduces a new importance sampling scheme , called adaptive twisted importance sampling , which is adequate for the improved estimation of rare event probabilities in he range of moderate deviations pertaining to the empirical mean of real i.i.d . summands . it is based on a sharp approximation of the density of long runs extracted from a random walk conditioned on its end value .

0910.1819

importance sampling procedures aim at reducing the calculation time which is necessary in order to evaluate integrals , often in large dimension . we consider the case when the integral to be numerically computed is the probability of an event defined by a large number of random components ; this case has received quite a lot of attention , above all when the event is of _ small _ probability , typically of order @xmath0 or so , as occurs frequently in industrial applications or in communication devices . the order of magnitude of the probability to be estimated is here somehow larger , and aims at coping with `` moderate probabilities '' as dealt with in statistics . the basic situation in is can be stated as follows . let @xmath1 be some random variable , say on @xmath2 with probability measure @xmath3 and density @xmath4 . let @xmath5 be a subset of @xmath6 with @xmath7 let @xmath8 denote a sample of i.i.d . observations of @xmath1 . by the law of large numbers @xmath9estimates @xmath10 without bias , when the @xmath11 are sampled under the density @xmath12 an altenative unbiased estimate of @xmath10 can be defined through @xmath13for all density @xmath14 when the support of @xmath4 is a subset of the support of @xmath14 , and the @xmath15are i.i.d . observations of a r.v . @xmath16 with density @xmath17 as is well known the optimal choice for the is sampling density @xmath14 is @xmath18 , the density of @xmath1 conditioned upon the event @xmath19 unfortunately an unpracticable choice which presumes the knowledge of @xmath20 the quantity to be estimated . would this sampling density be at hand , the required number @xmath21 of replications of @xmath16 to be performed would reduce to @xmath22 and the estimate would be exactly @xmath23 this fact motivates efforts in order to approximate @xmath24 in the case when the variable @xmath1 has a distribution which allows it . sometimes the random variable @xmath1 is obtained as a function of a large number of random variables , say @xmath25 and the event @xmath26 is of small or moderate probability . also the density of @xmath27 can not be evaluated analytically , due to the very definition of @xmath27 , but the random variables @xmath28 s have known distribution . this happens for instance when @xmath1 is a moment estimator or when it is the linear part of the expansion of an m or l -estimate ( see section 4 ) . the example which we have in mind is the following , which helps as a benchmark case in the is literature . the r.vs @xmath29 are i.i.d . , are centered with variance 1 , with common density @xmath30 on @xmath6 , and @xmath31is the empirical mean of the @xmath32 the set @xmath5 is @xmath33 where @xmath34 tends slowly to @xmath35 from above and we intend to estimate@xmath36for large but fixed @xmath37 many asymptotic results provide sharp estimates for @xmath38 but it is a known fact that asymptotic expansions are not always good tools when dealing with numerical approximations for fixed ( even large ) @xmath37 for example , citing ermakov ( 2004 , p 624 , ermakov2003 ) , the berry - esseen approximation for the evaluation of risks of order @xmath39 in testing is pertinent for sample sizes of order 5000 - 10000 ; also the accuracy of available moderate deviation probabilities as developped by inglot , kallenberg and ledwina in inglotkallenbergledwina1992 has not been investigated . this motivates our interest in numerical techniques in this field . according to ( [ lln ] ) the basic estimate of @xmath38 is defined as follows : generate @xmath21 i.i.d . samples @xmath40 with underlying density @xmath41 and define @xmath42where @xmath43here @xmath44 @xmath45 the statistics @xmath46 estimates the _ moderate deviation _ probability of the sample mean of the @xmath32 also denoting @xmath14 a sampling density of the vector @xmath47 the associated is estimate is @xmath48 in the range of moderate deviations the two major contributions to is schemes for the estimation of @xmath49 are fuh and hu @xcite and ermakov @xcite . the paper by fuh and hu does not consider events of moderate deviations as intended here ; it focuses on is schemes for the estimation of @xmath50 where @xmath51 is a given multinormal random vector and @xmath5 is a fixed set in @xmath52 the authors consider efficiency with respect to the variance of the estimate and state that for the case of interest the efficient sampling scheme is deduced from the distribution of @xmath53 by a shift in the mean inside the set @xmath54 the papers by ermakov instead handle similar problems as we do . ermakov s 2007 paper @xcite considers a sampling scheme where @xmath14 is the density of i.i.d . components . he proves that this scheme is efficient in the sense that the computational burden necessary to obtain a relative precision of the estimate with respect to @xmath49 does not grow exponentially as a function of @xmath37 he considers statistics of greater generality than the sample mean , such as m and l estimators ; in the range of moderate deviations the asymptotic behavior of those objects is captured however through their linear part which is the empirical mean of their influence function , which puts the basic situation back at the center of the scene . we discuss efficiency in section 3 and present some results in connection with ermakov s pertaining to m and l estimators in section 4 . the numerator in the expression ( [ form is ] ) is the product of the @xmath55 while the denominator need not be a density of i.i.d . copies evaluated on the @xmath56 . indeed the optimal choice for @xmath14 is the density of @xmath57 conditioned upon @xmath58 say @xmath59 since the optimal solution is known to be @xmath60 , the best its approximation , the best the sampling scheme , at least when it does not impose a large calculation burden ; classical sampling schemes consist in simulation of independent copies of r.v.s @xmath61 , @xmath62 , and efficiency is defined in terms of variance of the estimate inside this class of sampling , which , by nature , is suboptimal with respect to sampling under good approximations of @xmath63 for long runs , i.e. for large @xmath64the present paper explores the choice of good sampling schemes from this standpoint . obviously mimicking the optimal scheme results in a net gain on the number @xmath21 of replications of the runs which are necessary to obtain a given accuracy of the estimate with respect to @xmath49 @xmath45 however the criterion which we consider is different from the variance , and results as an evaluation of the mse of our estimate on specific subsets of the runs generated by the sampling scheme , which we call typical subsets , namely having probability going to @xmath22 under the sampling scheme as @xmath65 increases . on such sets , the mse is proved to be of very small order with respect to the variance of the classical estimate , whose mse can not be diminuished on any such typical subsets . we believe that this definition makes sense and prove it also numerically . this is the scope of section 3 in which it will be shown that the relative gain in terms of simulation runs necessary to perform an @xmath66 relative error on @xmath49 drops by a factor @xmath67 with respect to the classical is scheme . our proposal therefore hinges on the local approximation of the conditional distribution of longs runs @xmath68 from @xmath69 this can not be achieved through the classical theory of moderate deviations , first developped by de acosta and more recently by ermakov ; at the contrary the ad hoc procedure developped in the range of large deviations by diaconis and freedman @xcite for the local approximation of the conditional distribution of @xmath68 given the value of @xmath70 is the starting point of the present approach . we find it useful to briefly expose these two different points of view . we also mention the approximation technique for moderate deviations of sub linear functionals of the empirical measure by inglot , kallenberg and ledwina inglotkallenbergledwina1992 , based on strong approximation techniques ; these results provide explicit equivalents for the probability of moderate deviations , but do not lead to adequate approximations for the obtention of their numerical counterparts by is methods . the following notation and assumptions will be kept throughout this paper . we assume that @xmath71 satisfies the cramer condition , i.e. @xmath72 has a finite moment generating function @xmath73 in a non void neighborhood of @xmath74 denote @xmath75and @xmath76when defined . the values of @xmath77 and @xmath78 are the expectation and the variance of the _ tilted _ density @xmath79where @xmath80 is the only solution of the equation @xmath81 when @xmath82 belongs to the support of @xmath12 denote @xmath83 the probability measure with density @xmath84 . the _ chernoff function _ of @xmath71 is@xmath85for @xmath86 in the support of @xmath71 and it holds@xmath87where @xmath88 denotes the reciprocal function of @xmath89 denote @xmath90the characteristic function of @xmath91 assume that @xmath92for some @xmath93 this condition entails the validity of the edgeworth expansions to be used in the sequel ( see e.g. feller @xcite ) . the notation @xmath94 is used to denote the value of the density @xmath4 of the r.v . @xmath95 at point @xmath96 the notation @xmath97 is used to define the value of the density of the r.v . @xmath98 under @xmath4 , i.e. when the summands are i.i.d . with density @xmath12 also we may write @xmath99 to denote the density ( on the corresponding image space ) of some function @xmath100 of the sample @xmath101 we write @xmath102 the distribution of @xmath57 given @xmath103 and @xmath104 its density . the symbol @xmath105 denotes the standard normal density on @xmath106 a basic requirement for a good is sampling scheme is that it mimicks the conditional _ density _ @xmath59 we first expose a general argument in this direction in order to clarify that there is no bypass through the general theory of large or moderate deviations to achieve this goal . also the present discussion motivates the choices of classical is sampling schemes ( ermakov ) , emphasizing that the general theory provides the proof that the _ marginal _ conditional distribution of @xmath57 under @xmath103 is well approximated by @xmath107 a statement which is usually refered to as a _ gibbs conditional principle . _ we need some tools from the moderate deviation principle as developped by @xcite following deacosta1992 . let @xmath108 be a class of measurable functions defined on @xmath6 and @xmath109 be the class of all signed finite measures on @xmath6 which satisfy@xmath110on @xmath111 define the @xmath112 topology , which is the coarsest for which all mappings @xmath113 ( @xmath114 ) are continuous for all @xmath115 in @xmath108 . for @xmath3 a probability measure and @xmath116 in @xmath117 the so - called chi - square distance between @xmath3 and @xmath116 is defined through@xmath118whenever @xmath116 is absolutely continuous with respect to @xmath119 and equals @xmath120 otherwise . @xmath121the following _ moderate deviation _ sanov result holds ; see ermakov2007 . assume that @xmath34 tends to @xmath122 and @xmath123 tends to infinity . let @xmath124 denote the empirical measure pertaining to an i.i.d . sample @xmath125 write @xmath126 it holds @xmath127 where the interior and closure of the set @xmath128 refer to the @xmath112 topology on @xmath129 consider now the asymptotic distribution of @xmath71 conditionally upon the sequence of events @xmath130 , so - called moderate deviation events . with @xmath131 and @xmath132 the class of all bounded measurable functions , @xmath133 holds with @xmath128 substitued by @xmath134 the subset of @xmath111 defined through@xmath135 with @xmath3 the probability measure of the r.v . @xmath71 denote @xmath136 the @xmath137 _ projection of _ @xmath3 on @xmath138 namely @xmath139the set @xmath134 is closed in @xmath111 @xmath140 equipped with the @xmath141 @xmath121topology . existence of a @xmath137 projection of @xmath3 on a @xmath142closed subset of @xmath143 holds as a consequence of theorem 2.6 in @xcite when @xmath144 is finite for all @xmath100 in @xmath145 which clearly holds since @xmath146 is finite . uniqueness follows from the convexity of @xmath134 and the strict convexity of @xmath147 from ( [ sanov mdp ] ) it can easily be obtained that@xmath148with @xmath149 which in turn yields the following [ prop gibbs moderate ] with the above notation@xmath150 the proofs of ( [ cond sanov mdp ] ) and of the above proposition are differed to the appendix . this way can not provide an equivalent expression for the conditional _ density _ of @xmath71 which requires strong regularity assumptions . furthermore it can not be extended to the case of interest here , when @xmath71 is substituted by @xmath151 for large values of @xmath152 , i.e. when an approximation of the law of the path @xmath57 is needed , at least on long runs . however the result in proposition [ prop gibbs moderate ] is a strong argument in favor of ermakov s sampling scheme , namely simulating i.i.d . r.v.s with common density @xmath153 in ( [ form is ] ) . the other way follows zabell @xcite and diaconis and freedman @xcite approaches , which were developped in the range of large deviations . see also van camperhout and cover vancamperhoutcover1981 , who considered the density or the c.d.f . of @xmath154 conditioned on the value of @xmath98 for fixed @xmath155 it is restricted in essence to the context of the sample mean . the sketch of the method is as follows . the density of @xmath71 given @xmath156 writes@xmath157where we used the symbol @xmath4 to emphasize that the @xmath158 are i.i.d . with common density @xmath159 it is a known fact , and easy to establish , that the density defined in ( [ cond dens ] ) is invariant when sampling from any density of the form ( [ tilted density ] ) instead of @xmath159 this yields , selecting @xmath160 @xmath161when the r.vs @xmath28 s obey a local central limit theorem under the sampling density @xmath162 it can be proved that @xmath163as @xmath65 tends to @xmath164 diaconis and freedman obtain such a statement when @xmath71 is substituted by @xmath68 with @xmath165 we will continue this approach in the range of moderate deviations , enhancing it to the density of @xmath166 with @xmath167 integrating with respect to the conditional distribution of @xmath98 under the event @xmath168 provides the required approximation@xmath45 the scope of the present paper is to present some technique which provides typical realisations of runs @xmath68 under the conditional event @xmath103 for very large @xmath155 therefore it aims at the exploration of the support of the distribution of @xmath57 under @xmath169 the application which is presented pertains to importance sampling for the estimation of rare events probabilities through the adaptive twisted is scheme . section 2 of this paper is devoted to the approximation of the conditional density of @xmath68 under @xmath169 section 3 presents the atis algorithm , a number of remarks for its practical implementation , and discusses efficiency . section 4 is devoted to m and l estimates and their moderate deviation probabilities . we have postponed many proofs to the appendix , but the main one of section 2 . moderate deviations results for sums of i.i.d . real valued random variables under our assumptions have been studied since the 50 s by many authors . we will make use of a _ local _ result , due to richter @xcite , which we state as [ lemma richter local ] under the general hypotheses and notation of this paper , when @xmath34 is a sequence satisfying @xmath170 together with @xmath171 it holds@xmath172 the _ global _ counterpart of lemma [ lemma richter local ] in the form used here is due to jensen ( see @xcite , corollary 6.4.1 ) and states [ lemma jensen ] under the same hypotheses as above@xmath173where @xmath174 the following known fact is used repetedly . it sets that the conditional densities of sub - partial sums given the partial sum is invariant through any tilting . assume @xmath175 i.i.d . with density @xmath4 and note @xmath176 the corresponding tilted density for some parameter @xmath177 [ lemma inv conditional]for @xmath178 for all @xmath179 in the support of @xmath119 for all @xmath180 and @xmath181 @xmath182 the sequence of constants @xmath34 defining @xmath5 in ( [ a ] ) and ( [ e_n ] ) satisfies @xmath183 in this section we obtain a close approximation for @xmath184 for @xmath185 where @xmath34 and @xmath186 satisfy the following set of conditions . the value of @xmath187 satisfies @xmath188 with @xmath189 we denote ( a1), ... ,(a4 ) , ( c1), ... ,(c4 ) the above conditions . it appears clearly from ( [ form is ] ) that the optimal choice @xmath190 need only to hold on paths @xmath191 sampled under @xmath14 and not on all @xmath192 in a similar way the approximation of the optimal density need to be realized only when evaluated on samples @xmath193 generated according to this approximation , and approximation of @xmath194 on the entire space @xmath195 is not needed@xmath45 the approximation of @xmath196 by such a density @xmath197 is difficult to obtain on realizations under @xmath197 and much easier under @xmath198 the following lemma proves that approximating @xmath196 by @xmath197 under @xmath104 is similar to approximating @xmath104 by @xmath197 under @xmath199 let @xmath200 and @xmath201 denote two p.ms on @xmath202 with respective densities @xmath203 and @xmath204 [ lemmafrompntogn ] suppose that for some sequence @xmath205 @xmath121which tends to @xmath122 as @xmath65 tends to infinity@xmath206as @xmath65 tends to @xmath164 then @xmath207 denote @xmath208it holds for all positive @xmath209@xmath210where @xmath211since @xmath212it follows that @xmath213which proves the claim . this shows that the approximation of @xmath104 need not to be achieved on the whole space @xmath195 but only on _ typical paths _ under the conditionning event @xmath169 it appears thatsuch a sharp approximation is possible on quite long portions @xmath214of sample paths generated under @xmath102,@xmath121when @xmath186 tends to @xmath215 together with @xmath65 and @xmath216 goes to @xmath217 let @xmath218 such that @xmath219 with @xmath220 small enough . we prove that the sequence of conditional densities @xmath221 is closely approximated by a sequence of suitably modified tilted densities when evaluated at @xmath222 a realization under the density @xmath223 this is the scope of proposition [ prop approx local cond density ] hereunder . the size of @xmath220 is such that @xmath224 can be substituted by an integral of @xmath225 with respect to the distribution of @xmath98 conditionally on @xmath226 this is the scope of proposition [ prop p_n equiv g under s_n > na_n ] . define @xmath227 @xmath228 and @xmath229 through@xmath230@xmath231and@xmath232which are the variance and the kurtosis of @xmath233 reflecting the corresponding characteristics of @xmath4 , since @xmath229 is close to @xmath122 as shown in the following result . [ lemmaorderoftinetc ] let @xmath218 belong to @xmath234 and assume that ( a ) holds together with ( c2 ) and ( c3 ) . then under @xmath235 , @xmath229 tends to @xmath122 , @xmath236 tends to @xmath22 and @xmath237 tends to the third centered moment of @xmath4 uniformly upon @xmath238 in @xmath239 write @xmath240which goes to @xmath122 under @xmath102 uniformly upon @xmath218 under ( c2 ) and ( c3 ) where we used lemma [ lemma m_i , n under conditioning ] ; therefore @xmath229 goes to @xmath122 uniformly in @xmath218 which concludes the proof . the following density @xmath241 defined in ( [ g_s ] ) on @xmath242 provides the sharp approximation of @xmath243 this density is defined on @xmath244 as a product of conditional densities which are set in the following displays . it only approximates @xmath245 on typical vectors @xmath246 which are realizations of @xmath68 under @xmath169 chose any density @xmath247 ( for convenience denoted @xmath248 in ( g_s).and for @xmath249 define recursively the sequence of conditional densities @xmath250 through @xmath251and @xmath252a density on @xmath6 , with @xmath229 the unique solution of the equation@xmath253where @xmath254 the normalizing factor @xmath255 is @xmath256define @xmath257 the density on @xmath244 through@xmath258the definition in,([g_i ] ) can also be stated as @xmath259where @xmath260 is the normal density with mean @xmath261 and variance @xmath262 at @xmath86 . here @xmath263@xmath264and the constant @xmath265 is @xmath266 . this form is appropriate for the simulation . the density @xmath250 is a slight modification from @xmath267 . it approximates sharply p@xmath268 . for small values of @xmath269 , the contribution of @xmath270 and of @xmath271 is small and @xmath250 fits nearly with @xmath272 , when @xmath218 is close to @xmath34 , which is in accordance both with diaconis and freedman s approximation when translated in the moderate deviation range and with ermakov s is scheme . when the @xmath273 s are i.i.d . normal then @xmath274 for all @xmath275 we then have [ prop approx local cond density]set @xmath218 with @xmath276 and assume ( a ) together with ( c2 ) and ( c3 ) . let @xmath47 be a sample with distribution @xmath277 then uniformly upon @xmath218 @xmath278 the proof uses a bayes formula to write @xmath279 as a product of @xmath186 conditional densities of individual terms of the trajectory evaluated at @xmath214 , and the invariance property stated in lemma [ lemma inv conditional ] . each term of this product is approximated through an edgeworth expansion which together with the three preceeding lemmas , conclude the proof . it holds @xmath280by the independence of the r.vs @xmath28 ; we have set @xmath281 by lemma [ lemma inv conditional]@xmath282where we used bayes formula and the independence of the @xmath283 s under @xmath284 a precise evaluation of the dominating terms in this lattest expression is needed in order to handle the product ( [ joint density ] ) . under the sequence of densities @xmath285 the i.i.d . r.vs @xmath286 define a triangular array which satisfies a local central limit theorem , and an edgeworth expansion . under @xmath285 , @xmath287 has expectation @xmath288 and variance @xmath289 center and normalize both the numerator and denominator in the fraction which appears in the last display . denote @xmath290 the density of the normalized partial sum @xmath291 when the summands are i.i.d . with common density @xmath284 hence , evaluating both @xmath290 and its normal approximation at point @xmath292 @xmath293the sequence of densities @xmath290 converges pointwise to the standard normal density under the assumptions , when @xmath294 tends to infinity , i.e. when @xmath295 tends to infinity , and an edgeworth expansion to the order 5 is performed for the numerator and the denominator . set @xmath296 using lemma [ lemmaminunderconditioning ] we have @xmath297it then holds@xmath298 \label{hermite } \\ & & + o_{\mathfrak{p}_{n}}\left ( \frac{1}{\left ( n - i-1\right ) ^{3/2}}\right ) . \notag\end{aligned}\ ] ] we perform an expansion in @xmath299 up to the order @xmath300 with a first order term @xmath301 namely@xmath302where @xmath303 with @xmath304 only the first order term is relevant when handling the conditional density of the sub trajectory @xmath305 write @xmath306and use lemmas [ lemma m_i , n under conditioning ] and [ lemma max y_i under e_n ] to obtain@xmath307and@xmath308where the @xmath309 terms stem from the convergence of @xmath310 to @xmath22 by lemma [ lemmaorderoftinetc]@xmath45 assuming ( c2 ) it follows that @xmath311and@xmath312which yields @xmath313 the hermite polynomials depend upon the moments of the underlying density @xmath314 since @xmath315 has expectation @xmath122 and variance @xmath22 the terms corresponding to @xmath316 and @xmath317 vanish . up to the order @xmath318 the polynomials write @xmath319 , @xmath320 . in order to obtain a development of the polynomial bracket in ( [ hermite ] ) in terms of powers of @xmath321 only the term in @xmath86 from @xmath322 and the constant term from @xmath323 are relevant . it holds@xmath324when ( c3 ) holds then @xmath325for the term of order @xmath318 it holds @xmath326when ( c2 ) and ( c3 ) hold it follows that @xmath327the fifth term in the expansion plays no role in the asymptotics , under ( a ) . to sum up and using ( a ) and lemma [ lemma max y_i under e_n ] we get@xmath328 turn back to ( [ condtilt ] ) and do the same edgeworth expansion in the demominator , which writes@xmath329summarizing and using both ( [ p3 ] ) and ( [ p4 ] ) we obtain @xmath330 the term @xmath331 in @xmath332 comes from the ratio of the two gaussian densities @xmath299 and @xmath333 taking logarithms and using standard calculus provides the result in ( [ g_i ] ) ; indeed the constant term @xmath334 in ( [ p4 ] ) combines with the corresponding one in ( [ pi 0 ] ) to produce a term of order @xmath335 whose sum is @xmath336 . we now prove that @xmath337 as defined in ( [ k_i ] ) satisfies @xmath338 this will conclude the proof . use the classical bounds @xmath339to obtain on both sides of the above inequalities the second order approximation of @xmath340 the upper bound is @xmath341 .\end{aligned}\]]the lower bound is the same up to order 2 and the third order term plays no role . use lemma [ lemmaorderoftinetc ] to conclude , making a taylor expansion in @xmath342 @xmath343 and @xmath344 the dominating terms are due to @xmath345 and @xmath346 which yield the @xmath347 term in ( [ kiapproximation ] ) . the other terms are indeed @xmath348 using lemma [ lemmaminunderconditioning ] , leading to ( [ kiapproximation ] ) . hence ( [ ratiofixed i ] ) writes as @xmath349 putting the pieces together yields under ( a ) @xmath350uniformity upon @xmath218 is a consequence of lemma [ lemmaorderoftinetc ] . this closes the proof of the proposition . when the @xmath273 s are i.i.d . normal , then the result in the above proposition holds with @xmath351 stating that @xmath352 for all @xmath353 in @xmath195 . the density in ( [ g_i ] ) is a slight modification of @xmath284 however second order terms are required here in order to handle the approximation of the density of @xmath354conditioned upon @xmath355 and @xmath356 the modification from @xmath357 to @xmath358 is a small shift in the location parameter , which reflects the asymmetry of the underlying distribution @xmath359 and a change in the variance : large values of @xmath360 have smaller weight for large @xmath361 which is to say that the distribution of @xmath360 tends to concentrate around @xmath288 as @xmath269 approaches @xmath155 the `` moderate deviation '' case is typically @xmath362 for @xmath363 in @xmath364 in this case the condition @xmath365 holds for all values of @xmath366 the other case is when @xmath34 is `` nearly constant '' , in the range @xmath367,@xmath368 decreasing very slowly to @xmath369 with @xmath370 [ remark edgeworth array]in lemmas [ lemma m_i , n under conditioning ] and [ lemma max x_i under conditioning ] , as in the previous proposition , we use an edgeworth expansion for the density of the normalized sum of the @xmath371th row of some triangular array of row - wise independent r.vs with common density . consider the i.i.d . r.vs @xmath175 with common density @xmath372 where @xmath218 may depend on @xmath65 but remains bounded@xmath45 the edgeworth expansion pertaining to @xmath373 can be derived following closely the proof given for example in @xcite , pp 532 and followings substituting the cumulants of @xmath4 by those of @xmath374 . denote @xmath375 the characteristic function of @xmath376 clearly for any @xmath377 there exists @xmath378 such that @xmath379 @xmath380 and since @xmath34 is bounded , @xmath381 therefore the inequality ( 2.5 ) in @xcite p533 holds . with @xmath382 defined as in feller1971 ( 2.6 ) holds with @xmath383 replaced by @xmath384 and @xmath218 by @xmath385 ( 2.9 ) holds , which completes the proof of the edgeworth expansion in the simple case . the proof goes in the same way for higher order expansions . this justifies our argument in the lemmas cited above . in the proofs of proposition [ prop approx local cond density ] we made use of such expansions when the r.vs @xmath386 are i.i.d . with common density @xmath387the same argument as sketched hereabove applies in this case also . let @xmath388 @xmath389 with distribution under the conditioning event @xmath169 hence for any borel set @xmath5 @xmath390the distribution of @xmath391 is concentrated on a small neighborhood of @xmath392 indeed we have [ lemma magnitude of t ] assume that ( a1 ) holds@xmath45 for any sequence @xmath393 such that ( c1 ) holds@xmath394@xmath395 use lemma [ lemma jensen ] . moreover @xmath391 is asymptotically exponentially distributed . the asymptotic distribution of @xmath391 is captured in the following [ lemma approx exponential for tbold]when ( a1 ) holds then for all @xmath180 in @xmath396 the r.v . @xmath397 satisfies@xmath398where @xmath399 and therefore @xmath400 write @xmath401and use lemmas [ lemma richter local]and [ lemma jensen ] . a first order expansion yields @xmath402 which proves the claim . in this section proposition [ prop approx local cond density ] is extended in order to provide an approximation of @xmath403 when @xmath404 is a random vector generated under @xmath223 this is obtained through an integration w.r.t . @xmath218 in ( [ local approx under exact value ] ) ; indeed it holds @xmath405 and the domain of integration can be reduced to a small neighborhood of @xmath406 which contains nearly all the realizations of @xmath391 under @xmath407 . this argument allows the interchange of asymptotic equivalents and integration . define @xmath408where @xmath409 is defined in ( [ g_s ] ) . when @xmath257 is substituted by @xmath197 then @xmath410does not stand . let @xmath411 where @xmath412 is fitted compatibly with proposition [ prop approx local cond density ] . define @xmath413 [ prop p_n equiv g under s_n > na_n ] when @xmath214 is a random vector generated with density @xmath104 and @xmath121(a ) and ( c ) hold then @xmath414 the proof of proposition [ prop p_n equiv g under s_n > na_n ] relies upon the following lemma , whose proof is postponed to the appendix . [ lemma from local cond to global cond ] let @xmath415 satisfy @xmath416 and ( a ) and ( c ) hold then when @xmath47 is generated under @xmath104 it holds@xmath417 we now prove proposition [ prop p_n equiv g under s_n > na_n ] through an integration of the local approximation given in proposition [ prop approx local cond density ] . for all @xmath218 in @xmath234@xmath418uniformly on @xmath218 when @xmath47 is sampled under @xmath223 it then holds @xmath419where we used lemmas [ lemma from local cond to global cond ] together with ( c4 ) which helps to keep the @xmath420 term . this concludes the proof of proposition [ prop p_n equiv g under s_n > na_n ] . as a consequence of lemma [ lemmafrompntogn ] the following result holds , which asseses that when sampled under @xmath421 @xmath121the likelihood of the random vector @xmath422 approximates @xmath423 [ prop global approx inverse ] assume ( a ) and ( c ) . let @xmath422 be a random vector with p.m. @xmath424 with density @xmath421 on @xmath244 defined in ( [ g global ] ) . it holds @xmath425as @xmath426 the last result in proposition [ prop global approx inverse ] above suggests that an importance sampling density deduced from @xmath421 would benefit from some optimality as defined in the introduction since it fits with the conditional density on long runs . it is enough to approximate the conditional distribution of @xmath427 under @xmath407 by lemma [ lemma approx exponential for tbold ] and to plug in this approximation in ( [ g global ] ) . let @xmath428 denote a r.v . with exponential distribution with parameter @xmath429 on @xmath430@xmath431 using again lemmas [ lemma richter local ] and [ lemma jensen ] it is easily checked that @xmath432for some sequence @xmath433 whinch tends to @xmath369 from which @xmath434with @xmath435 , which proves that we may substitute @xmath391 by the exponential r.v . @xmath428 while keeping the properties of the is procedure . we denote @xmath436the sampling scheme under which the estimate ( [ form is ] ) is computed ; in ( [ new sampling ] ) the value of @xmath437 is defined through@xmath438with @xmath439 since the r.v . @xmath428 is highly concentrated in a small neighborhood of @xmath34 we suggest to forget about @xmath440 in the definition ( [ new sampling ] ) of @xmath441 above and to integrate on @xmath442 instead of @xmath239 numerical experiments argue in favor of this heuristic . the remarks at the end of this paragraph provide simple and efficient solutions for the effective calculation of the estimate . 1- draw @xmath443 independent random variables @xmath444 with distribution ( [ densityt ] ) and define the density on @xmath195@xmath445where @xmath446 is defined as @xmath447where @xmath448 is defined in ( [ g_i ] ) for @xmath449 , @xmath450 and @xmath451where @xmath452 and @xmath453 @xmath121is the only solution of the equation @xmath454 with @xmath455 with @xmath456 2-define @xmath21 which is the number of replications of the simulated random trajectory to be performed 3-for @xmath457 between @xmath22 and @xmath21 do \ { draw a random variable @xmath458 with distribution ( densityt ) draw the first @xmath186 variables @xmath459 recursively with density @xmath460 as defined in ( [ g_t ] ) with @xmath461 substituted by @xmath458 . draw the @xmath462 random variables @xmath463 independently with common density @xmath464 defined in ( pi^alpha_k ) with @xmath461 substituted by @xmath465 } 4- define @xmath466 where @xmath467 a number of remarks hereunder show that atis is not difficult to implement . since the first order efficiency of i.i.d sample schemes is reached if and only if the sampling distribution is the twisted one with parameter @xmath34 ( see @xcite ) , the present algorithm should be compared with it . the classical is scheme which uses i.i.d . replicates with density @xmath468 is easy to implement but may lead to biased estimates of @xmath49 ; the simulation of a r.v . with density @xmath153 is difficult in non standard cases when @xmath4 is easy to simulate then an acceptance / rejection algorithm can be used ; however this requires to truncate the support of @xmath4 , what should precisely be avoided in order to obtain unbiaised estimates ; see @xcite . when @xmath153 is easy to simulate , atis may take more time to run , due to the various intermediate calculations which are required at each stage of the algorithm . the generation of the r.v . @xmath459 above is easy and fast and does not require any simulation according to a twisted density . it holds @xmath469where @xmath260 is the normal density with mean @xmath261 and variance @xmath262 at @xmath86 . here @xmath263@xmath470a r.v . @xmath471 with density @xmath472 with @xmath473 and where @xmath4 is a given density and @xmath474 is easy to simulate : denote @xmath475 the c.d.f . with density @xmath476 it is easily checked that @xmath477 is the density of the r.v . @xmath478 where @xmath479 is a r.v . on @xmath480 $ ] with density @xmath481 denotes the reciprocal function of @xmath475 @xmath45 now an acceptance / rejection algorithm provides a realisation of @xmath479 . indeed let @xmath482 be a density such that @xmath483 for some constant @xmath484 and all @xmath86 in @xmath480 $ ] ; let @xmath485 be uniformly distributed on the hypograph of @xmath486 namely @xmath487 where @xmath488 has density @xmath115 and @xmath489 is uniform @xmath480 $ ] independent of @xmath490 when @xmath491 is less than @xmath492 then @xmath488 has density @xmath493 the calculation of @xmath494 above requires the value of @xmath495 in ( [ g_i+1(x_i+1/x_i)numerically ] ) . a monte carlo technique can be used : simulate @xmath496 i.i.d . r.vs @xmath497 with density @xmath498 , which is fast , and substitute @xmath265 by @xmath499 , which provides a very accurate approximation to be inserted in the calculation of the estimate . it may seem that this algorithm requires to solve @xmath500 equations of the form @xmath501 in order to obtain the @xmath229 which are necessary to perform the simulation of @xmath502 as described above as well as the calculation of @xmath503 . such is is not the case , and only @xmath21 equations have to be solved . consider for example the simulation of @xmath459 with density @xmath504 this is achieved as follows : 1- solve the equation @xmath505whose solution is @xmath506 generate @xmath507 according to @xmath508 2- since @xmath509use a first order approximation to derive@xmath510from which ( [ g_i+1(x_i+1/x_i)numerically ] ) is derived and @xmath511 can be simulated as mentioned above . in the moderate deviation scale the function @xmath512 does not vary from @xmath22 and the above approximation is fair . the density @xmath513 on @xmath195 is a monte carlo approximation of @xmath197 defined by@xmath514where @xmath515 is replaced by @xmath516 and the integral is replaced by a finite mixture . @xmath443 is a free parameter . also notice that the @xmath462 i.i.d . r.vs have common tilted density @xmath464 with parameter given by ( [ start remaining tilted n - k rv s ] ) , thus identical to ermakov s sampling scheme with end point in@xmath517 and not in @xmath518 the critical parameter @xmath186 is the length of the partial sum run which is to be simulated according to the density @xmath519 as defined in ( [ g(x_1^k ) ] ) . by ( [ g(x_1^k ) ] ) it would be enough to establish some statistics averaging the estimate ratios @xmath520 on a set of runs , and to select @xmath521 as some @xmath522 ensuring that this ratio keeps close to @xmath217 in the case when the r.vs @xmath273 are normally distributed the density @xmath523 as defined in ( [ g_i ] ) coincides with @xmath524 for all value of @xmath269 between @xmath22 and @xmath525 which entails that @xmath186 can be set equal to @xmath526 this very peculiar case is illustrated in figure 1 , for @xmath527 and @xmath49 is close to @xmath528 . we can see that atis produces a very sharp estimate of @xmath49 for a small value of @xmath21 when compared to the classical is scheme . in the other cases , when @xmath250 approximates @xmath524 only under some conditions on @xmath186 as described in conditions ( a ) , we propose the following heuristics , which works well and is easy to implement ; other choices are possible , which provide similar acceptable results . instead of @xmath441 consider the following construction , which will also be used in the is algorithm : simulate @xmath444 , i.i.d . with distribution ( [ densityt ] ) and define the density on @xmath529@xmath530where @xmath446 is defined as @xmath531where @xmath448 is defined in ( [ g_i ] ) for @xmath449 . the density @xmath532 is a monte carlo approximation of @xmath533 . by ( [ p_n fraktur ] ) and following the same heuristics as for @xmath534 define , with a new set of i.i.d . @xmath461 s @xmath535we use lemma [ lemma richter local ] in order to obtain an explicit approximation for @xmath536 it holds @xmath537define therefore @xmath538and @xmath539fix some integer @xmath21 which is the number of runs to be simulated in order to fix @xmath540 @xmath21 need not be large . for all @xmath457 between @xmath122 and @xmath21 draw independently a random variable @xmath541 with density ( [ densityt ] ) and the run @xmath542 with density @xmath543 defined as in ( [ g_s ] ) with @xmath186 substituted by @xmath522 and @xmath218 by @xmath544 @xmath545fix @xmath186 as the smallest @xmath522 which indicates a departure of this statistics from @xmath217 in atis the distribution in ( [ g(x_1^k ) ] ) is substituted by a numerical approximation of @xmath546which is suboptimal with respect to ( [ g(x_1^k ) ] ) but is easily implemented . a monte carlo procedure produces @xmath547 as described above in ( [ gbar ] ) . it appears that @xmath443 should be large when @xmath186 is large . for example in the normal case with @xmath527 for @xmath548 , then @xmath549 produces excellent estimates for values of @xmath21 of order @xmath550 whereas for @xmath551 the value of @xmath443 should increase up to @xmath552 with the same @xmath21 .as seen in figure2@xmath45 the reason for this increase in @xmath443 is that ( [ gintegrale ] ) is a mixture of densities in very high dimension , which seems very sensitive with respect to the approximation of the mixture measure . this point should deserve a specific study , out of the scope of the present paper . however the normal case is quite specific , since it allows @xmath186 to be as close to @xmath65 as wanted . in the other cases , as examplified in the figures pertaining to the exponential case , @xmath186 is resticted to lower values and @xmath443 is rather low . the evaluation of the performances of is algorithm is a controversal argument . many criterions are at hand , for example the _ probability of hits _ which counts the relative number of simulations hitting the target @xmath553 or the _ variance _ of the estimator . we refer to the book by bucklew @xcite for a discussion on the relative merits of each approach . the variance of an is estimate of @xmath49 under the sampling density @xmath14 writes@xmath554with @xmath555 the situation which we face with our proposal lacks the possibility to provide an order of magnitude of the variance our our is estimate , since the properties necessary to define it have been obtained only on _ typical paths _ under the sampling density @xmath441 defined in ( [ new sampling ] ) and not on the whole space @xmath195 ( but in the case when the @xmath273 s are normally distributed)@xmath45 we will prove , however , that the performance of this new procedure can be considered favorably . not surprisingly the loss of performance with respect to the optimal sampling density @xmath556 is due to the @xmath462 last i.i.d . simulations , leading a quasi- mse of the estimate proportional to @xmath557 in order to discuss this we first go back to the classical is scheme , for which we evaluate the asymptotic variance . the asymptotic variance of the estimate of @xmath558 can be evaluated as follows . the classical is is defined simulating @xmath21 times a random sample of @xmath65 i.i.d . r.vs @xmath559 , @xmath560 with tilted density @xmath468 . the standard is estimate is defined through@xmath561where the @xmath562 are i.i.d . with density @xmath153 and @xmath563 is as in ( [ indics_n / n > a_n])@xmath45 set @xmath564the variance of @xmath565 is given by @xmath566the _ relative accuracy _ of the estimate @xmath567 is defined through @xmath568it holds [ prop rel efficiency standard is]the relative accuracy of the estimate @xmath567 is given by @xmath569 it holds , omitting the index @xmath457 for brevity and noting @xmath179 for @xmath34 @xmath570the laplace integral above satisfies@xmath571as @xmath65 tends to infinity , which , together with the expansion @xmath572(which holds when @xmath573 @xmath574 ) concludes the proof . we have used lemma [ lemmaorderoftinetc ] to assess that @xmath575 we now come to a discussion of the above result . it is well known that the variance is not a satisfactory criterion to describe the variability of the outcomes of a random phenomenon : for example , a sequence of symmetric r.vs @xmath576 taking values @xmath577 with relative frequencies defined through @xmath578 has variance going to @xmath579 while being concentrated at @xmath580 in this case we can define an increasing family of sets @xmath581 with @xmath582 on which @xmath583 a much better indicator , obtained through trimming . we will prove that such an indicator can not be defined for the classical is scheme , stating therefore that the variance rate obtained in proposition prop rel efficiency standard is is indeed meaningfull . the easy case when @xmath175 are i.i.d . with standard normal distribution is sufficient for our need . the variance of the is estimate is proportional to @xmath584a set @xmath581 resulting as reducing the mse should penalize large values of @xmath585 while bearing nearly all the realizations of @xmath586 under the i.i.d . sampling scheme @xmath153 as @xmath65 tends to infinity . it should therefore be of the form @xmath587 for some @xmath440 so that \(a ) @xmath588and ( b)@xmath589which means that the is sampling density @xmath153 can lead a mse defined by @xmath590with a clear gain over the variance indicator . however when @xmath591 ( b ) does not hold and when @xmath592 ( a ) does not hold . so no reduction of this variance can be obtained by taking into account the properties of the _ typical paths _ generated under the sampling density : a reduction of the variance is possible only by conditioning on `` small '' subsets of the sample paths space . on no classes of subsets of @xmath593 with probability going to @xmath22 under the sampling is it possible to reduce the variability of the estimate , whose rate is definitely proportional to @xmath594 imposing a burden of order @xmath595 in order to achieve a relative efficiency of @xmath596 with respect to @xmath597 we will evaluate the performance of our estimate under @xmath441 since the algorithm envolves technical parameters ( typically @xmath443 ) ; in practice the monte carlo approximation introduces no significant bias . at the contrary to just evidenced hereabove , the procedure which we propose has a small asymptotic variability when evaluated through trimming on classes of subsets of @xmath195 whose probability goes to @xmath22 under the sampling @xmath441 . these subsets of @xmath195 get smaller and smaller as @xmath65 increases as measured through the mse of the estimate with respect to the mse of the classical is estimate . we prove the existence of these trimming sets in the present section and state that the resulting gain in terms of the mse of our estimate is the proper measure of its performance . these sets are the @xmath598 described in the following lemma , whose proof is differed to the appendix . for sake of notational simplicity denote @xmath205 the @xmath599 defined in ( [ g(x_1^k ) ] ) . [ lemma set c_n for efficiency ] with the just mentioned @xmath600 define the family of sets @xmath598 in @xmath195 such that for all @xmath601 in @xmath602@xmath603and @xmath604where @xmath453 is defined through@xmath605and @xmath606 satisfies @xmath607together with@xmath608then @xmath609furthermore on @xmath598@xmath610 we now prove that our is algorithm provides a net improvement over the classical is scheme in terms of mean square error when evaluated on this family of sets . define@xmath611@xmath612we prove that [ prop rel efficiency is]the relative accuracy of the estimate @xmath613 is given by @xmath614 denote @xmath615 the expectation with respect to the p.m. @xmath616 of @xmath40 conditioned upon @xmath617 ; we omit the index @xmath457 for brevity . using the definition of @xmath598 we get@xmath618the second line uses @xmath619 the third line is bayes formula . the fourth line is lemma [ lemma jensen ] . the fifth line uses ( [ order of t_k ] ) and uniformity in lemma [ lemma jensen ] , where the conditions in corollary 6.1.4 of jensen ( 1995 ) are easily checked since , in his notation , @xmath620 , condition ( i ) holds for @xmath621 in a neighborhood of @xmath122 ( @xmath622 undeed is resticted to such a set in our case ) , ( ii ) clearly holds and ( iii ) is ( [ f.c . edgeworth ] ) . [ prop rel efficiency]when @xmath623 then under ( a ) the ratio of the relative efficiencies of the adaptive is algorithm with respect to the standard is scheme is of order @xmath67 .. the same result holds when @xmath624 this section provides some application of the previous results for some classical types of estimators for which sharp moderate deviation probabilities can be obtained through linear approximations . we follow closely the work by @xcite ; see also @xcite . let @xmath625 denote a real valued statistical functional defined on the space @xmath626 , where we assume that @xmath625 has an influence function . let @xmath3 be a given p.m. we assume that for all @xmath116 in @xmath111 there exists a function @xmath627 ( depending on @xmath3 ) such that @xmath628 where @xmath496 is a seminorm defined on @xmath111 , continuous in the @xmath112 topology , and @xmath629 is a continuous and strictly monotone function which satisfies @xmath630 as @xmath631 the function @xmath14 is the _ influence function _ of @xmath625 at @xmath632 the class @xmath108 considered here contains @xmath633 let @xmath634 be real valued function defined on @xmath635 and assume that @xmath3 satisfies @xmath636 define @xmath637 as any solution of the equation @xmath638if defined . when @xmath639 depends upon a real valued parameter @xmath640 such that @xmath641then @xmath625 is _ fisher consistent _ and the substitution of @xmath642 by @xmath643 in ( [ m estim])@xmath394 the empirical measure pertaining to an i.i.d . sample with unknown p.m. @xmath644 provides a consistent estimate of @xmath645 under appropriate regularity conditions ; see @xcite . such estimate is an m - estimator . we assume that all conditions m1 to m5 in @xcite hold , which implies that ( [ diff t ] ) above holds ( see @xcite theorem 4.2 ) . also in this case the function @xmath14 writes@xmath646the same situation holds for l - estimators , when ( [ sanov mdp ] ) holds in @xmath111 it can be checked that a strong mdp holds for @xmath647 indeed when @xmath14 belongs to the class @xmath108 and @xmath648 = -\infty\]]then using ( [ diff t ] ) and ( [ sanov mdp ] ) it can be proved that the remaining term in @xmath649 is negligible w.r.t . the linear approximation @xmath650 on the moderate deviation scale , as follows from ( 2.14 ) and ( 2.15 ) in @xcite . furthermore in this case the strong moderate deviation holds for @xmath651 and @xmath652 in the range @xmath653 see also inglot , kallenberg and ledwina @xcite . + this graph illustrates proposition [ prop approx local cond density ] . + the graph shows the role of @xmath186 in the behavior of the estimate . the @xmath273 s are standard normal , @xmath654 and @xmath655 when @xmath186 is less than @xmath656 the new estimate improves on the classical i.i.d . a change in @xmath657 leads no significant change ( here @xmath549 ) . the value of @xmath21 is @xmath658 + the graph illustrates the accuracy of the asymptotic results in propositions [ prop rel efficiency standard is ] and [ prop rel efficiency is ] . the @xmath659 s are standard normal , @xmath527p@xmath660 the graph is an illustration of proposition [ prop rel efficiency ] . the r.vs @xmath273 s are standard normal , @xmath654 and @xmath655 in ordinate is the ratio of the empirical value of the mse of the adaptive estimate w.r.t . the empirical mse of the i.i.d . twisted one . the value of @xmath186 is @xmath661 this ratio stabilizes to @xmath67 for large @xmath662 in full accordance with proposition [ prop rel efficiency ] . + + the graphs above are typical paths under the conditional distribution ( with @xmath663 ) and under the i.i.d . sampling with tilted density . the value of @xmath65 is @xmath664 and the approximation of the conditional density of the random walk is fair up to @xmath665 , as indicated by the fact that the is estimator of @xmath49 is correct up to @xmath665 , which can be seen as a pertinent indicator . + the random variables @xmath666 are i.i.d . with exponential distribution with parameter @xmath22 on @xmath667 the case treated here is @xmath668 with @xmath527 @xmath669 and @xmath670 these values are computed through a very long run of the standard is algorithm ( with i.i.d . sampling according to the twisted ) and are used as a benchmark.the estimates are calculated with @xmath671 and @xmath672 for @xmath673 , i.e. for the classical i.i.d . twisted sampler ( lower values of @xmath21 lead unstable estimates ) by theorem 3.4 ( 2 ) in broniatowski and keziou ( 2006 ) it holds @xmath678 for some constants @xmath82 and @xmath679 the projection @xmath680 satisfies both @xmath681 and @xmath682 which yield @xmath683 and @xmath684 observe that @xmath688 also denote @xmath689 ( resp @xmath690 ) the subset of @xmath691 defined through @xmath692 , resp @xmath693 using bayes formula and the above moderate deviation result ( [ sanov mdp ] ) it follows that for any measurable set @xmath694 in @xmath143 for any positive ( resp . negative ) @xmath640 it then holds @xmath696 if @xmath697 and @xmath698 ( resp @xmath699 if @xmath700 and @xmath701 ) , which is to say , going to the limit in @xmath702 that @xmath703 where @xmath704 is the lebesgue decomposition of @xmath705 this closes the proof of ( [ gibbs mdp ] ) . a second order expansion of @xmath706 in a neighborhood of @xmath707 yields@xmath708hence for all borel set @xmath5 it holds @xmath709 since both @xmath710 tends to @xmath10 and @xmath711 is a finite measure it follows that @xmath712 tends to @xmath580 select @xmath181 in @xmath234 and denote @xmath715 the p.m on @xmath202 conditioned on @xmath716 it holds@xmath717we prove that for @xmath718 @xmath719as @xmath720 where @xmath721 denotes the variance of @xmath53 conditionally on @xmath722 integrating with respect to the distribution of @xmath98 conditioned upon @xmath103 concludes the proof@xmath45 using with @xmath724 , normalizing both @xmath725 and @xmath726 and making use of a first order edgeworth expansion in those expressions yields @xmath727and @xmath728with a similar development for the joint density @xmath729 , using the same tilted distribution @xmath730 it readily follows that @xmath731since @xmath732it follows that when @xmath733 tends to @xmath122 , then @xmath734 since @xmath735 this amounts to @xmath736integration with respect to the distribution of @xmath98 conditioned upon @xmath103 and splitting the integeral on @xmath737 and @xmath738 , using ( c2 ) concludes the proof . it can be proved that @xmath739conditionally on @xmath740 this result is to be compared with the gibbs principle for moderate deviations stated in the introduction which assets that for _ fixed _ @xmath741 the joint distribution of @xmath742 conditioned upon @xmath407 converges weakly , as @xmath743 , to the joint distribution of @xmath741 r.vs @xmath744 which are independent copies of @xmath745 .the above result says that even for sequences depending upon @xmath65 , we may replace the original @xmath741 variables by the @xmath741 _ independent _ tilted ones when exploring the behavior of @xmath746 under @xmath58 since @xmath747 shares the same limit distribution . let @xmath181 such that @xmath750 @xmath751 . denote @xmath715 the probability measure of @xmath57 given the the value of @xmath752 since @xmath753we first state the order of magnitude of @xmath748 under @xmath715 in the next lemma . define @xmath755 for all @xmath640 it holds@xmath756center and normalize both @xmath98 and @xmath757with respect to the density @xmath758 in the last line above , denoting @xmath759 the density of @xmath760 when @xmath761 has density @xmath758 with mean @xmath363 and variance @xmath762 we get @xmath763under the sequence of densities @xmath758 the triangular array @xmath764 obeys a first order edgeworth expansion @xmath765for some constant @xmath766 independent of @xmath65 and @xmath363 and where @xmath767where @xmath768 is the third hermite polynomial ; @xmath769 and @xmath770 are the second and third centered moments of @xmath771 we used uniformity upon @xmath180 in the remaining term of the edgeworth expansions . let @xmath772 such that @xmath773 making use of chernoff inequality@xmath774for any @xmath675 such that @xmath775 is finite . @xmath776it holds @xmath777which proves the lemma . as above write @xmath778where @xmath363 is defined as in the above lemma through @xmath755 use the same argument as in lemma [ lemma max x_i under conditioning ] to assess that when @xmath779 goes to infinity then the.rhs above tends to @xmath580 this closes the proof . it holds@xmath780by lemma [ lemma approx exponential for tbold ] it holds under ( c1 ) @xmath781where @xmath782 denote @xmath783 @xmath45 set @xmath784with @xmath785 and @xmath786 it holds@xmath787use lemma [ lemma m_i , n under conditioning ] to obtain@xmath788where @xmath789 use lemma lemma jensen to obtain @xmath790which tends to @xmath122 under ( c ) . in the above display , @xmath792 by the above definition @xmath793note also that @xmath794which goes to @xmath22 as @xmath65 tends to @xmath795 where we have used proposition [ prop p_n equiv g under s_n > na_n ] . in the above displays @xmath796 is the density of @xmath422 when @xmath797 is sampled under @xmath798 we have just proved that the sequence of sets @xmath799 contains roughly all the sample paths @xmath797 under the importance sampling density @xmath798 there exists @xmath806 such that for any @xmath601 in @xmath581 @xmath807indeed @xmath808and @xmath809 therefore@xmath810since @xmath811 is bounded so is @xmath812 and therefore @xmath813 as @xmath814 which implies ( [ control t_k / a_n ] ) .

analyses of the decay channel of the resonance by the cdf , belle , and lhcb collaborations have established its quantum numbers as . an analysis of the invariant mass distribution in the decay channel by the babar collaboration indicated a preference for over . we point out that a proper evaluation of the in that analysis increases the probability for from 7.1% to about 18.7% . in the case of quantum numbers , where the has an s - wave coupling to , the proximity of the threshold to thresholds and the narrow width of the suggest that the effects of scattering between and charm meson pairs could be significant . we derive invariant mass distributions for and that take into account s - wave scattering between the , , and channels . we also analyze the effects of scattering through the charmonium resonance . we find that scattering effects are unable to produce significant changes in the shape of the invariant mass distribution .

1305.5564

ever since the discovery of the @xmath1 by the belle collaboration in 2003 @xcite , one of the leading interpretations has been a charm meson molecule whose constituents are a superposition of @xmath13 and @xmath16 @xcite . this identification would require the @xmath2 quantum numbers of @xmath1 to be @xmath3 . the observation of its decay into @xmath17 determined the charge conjugation @xmath18 to be @xmath19 @xcite . in 2006 , the cdf collaboration reduced the options for @xmath2 to @xmath3 and @xmath6 by analyzing decays into @xmath0 @xcite . the lhcb collaboration recently ruled out @xmath6 , finally establishing the quantum numbers of @xmath1 as @xmath3 @xcite . the option @xmath6 had been disfavored on various theoretical grounds , especially if the @xmath1 is identified with the @xmath20 charmonium state @xmath21 . the prediction of the mass of @xmath21 in most potentials models is lower than 3872 mev by 40 to 100 mev @xcite . the decay of @xmath21 into @xmath17 should have strong multipole suppression @xcite . the expected production rate for d - wave charmonium in a hadron collider is much smaller than the observed production rate of the @xmath1 at the tevatron @xcite . the decay of d - wave charmonium into @xmath22 should have angular momentum suppression @xcite . finally , the degree of isospin violation required by the observed branching fraction into @xmath0 is difficult to accommodate for d - wave charmonium @xcite . all of these problems are solved , or at least ameliorated , if the quantum numbers are @xmath3 . back in 2010 , the babar collaboration analyzed decays of @xmath1 into @xmath12 and concluded that @xmath6 was preferred over @xmath3 @xcite . they quantified this preference in terms of a probability that was 7.1% for @xmath3 and 61.9% for @xmath6 . we will point out that a proper quantification of the likelihood for the observed result increases the probability for @xmath3 to 18.7% . with the properly calculated probabilities , the preference for @xmath6 over @xmath3 is no longer so significant . however it is still worth considering whether a more accurate description of the resonance in the @xmath12 channel would be important in the babar analysis or in future analyses . since it has quantum numbers @xmath3 , the @xmath8 has an s - wave coupling to @xmath9 . the proximity of the @xmath9 threshold to the @xmath13 and @xmath14 thresholds and the narrow width of the @xmath11 suggest that the effects of scattering between @xmath9 and charm meson pairs could be significant . we therefore study the effects of scattering between these coupled channels on the @xmath1 resonance in the @xmath12 channel . we also analyze the effects of scattering through the @xmath15 charmonium resonance , which has quantum numbers @xmath3 . in section [ sec : widths ] , we introduce our notation for the three coupled channels and for the many masses that are relevant to this problem . in section [ sec : ddscat ] , we derive the scattering amplitudes due to s - wave scattering between the coupled channels . we use them in section [ sec : lineshape ] to determine the inclusive line shape of the @xmath1 resonance and its line shape in the @xmath12 channel . we also determine the effect of the @xmath15 resonance on the line shape . in section [ sec:3pionmass ] , we derive a simple expression for the @xmath4 invariant mass distribution . we examine the babar results in ref . @xcite and point out that a proper evaluation of the @xmath7 significantly increases the probability for the quantum numbers @xmath3 . we show that the experimental resolution , which was ignored in previous theoretical analyses , has a significant effect on the @xmath4 invariant mass distribution . finally we study the effect on that invariant mass distribution of scattering between the three coupled channels and scattering through the @xmath15 charmonium resonance . we consider the effects of scattering between three @xmath23 channels involving the particle pairs @xmath13 , @xmath14 , and @xmath9 . we label the three channels by the integers 0 , 1 , and 2 and a vector index @xmath24 associated with the polarizations of the spin-1 particles : @xmath25 [ ch012 ] the @xmath26 matrices that project these channels onto isospin 0 and isospin 1 are @xmath27 [ ispin01 ] we denote the masses of the charm mesons @xmath28 , @xmath29 , @xmath30 , and @xmath31 by @xmath32 , @xmath33 , @xmath34 , and @xmath35 and the masses of @xmath36 and @xmath11 by @xmath37 and @xmath38 . we denote the reduced masses for the three channels in eqs . ( [ ch012 ] ) by @xmath39 , @xmath40 , and @xmath41 , respectively . the energy differences @xmath42 and @xmath43 between the thresholds for @xmath14 and @xmath9 and the @xmath13 threshold are @xmath44 [ delta012 ] we denote the total energy of the pair of particles in their center - of - momentum frame by @xmath45 . their total energy relative to the @xmath13 threshold is @xmath46 the amplitude for the propagation of a pair of particles between contact interactions involves the square root of their total energy relative to threshold . the appropriate thresholds for the pairs of particles in the channels in eqs . ( [ ch012 ] ) are complex , with imaginary parts given by the sum of the decay widths of the two particles . if one of the widths is much larger than the other one , it is sufficient to only take the larger one into account . the resulting threshold factors for the pairs of particles in the channels in eqs . ( [ ch012 ] ) are @xmath47^{1/2 } , \label{kappa0 } \\ \kappa_1(e ) & = & \left [ -2 \mu_1 ( e - \delta_1 + i \gamma_{*1}/2 ) \right]^{1/2 } , \label{kappa1 } \\ \kappa_{\psi \omega}(e ) & = & \left [ -2 \mu_{\psi \omega } ( e - \delta_{\psi \omega } + i \gamma_\omega/2 ) \right]^{1/2 } , \label{kappav}\end{aligned}\ ] ] [ kappa012 ] where @xmath48 kev , @xmath49 kev , and @xmath50 mev are the decay widths of @xmath28 , @xmath30 , and @xmath11 . the reduced masses are @xmath51 mev , @xmath52 mev , and @xmath53 mev . it is convenient to introduce a @xmath26 matrix @xmath54 whose diagonal entries are the threshold factors in eqs . ( [ kappa012 ] ) : @xmath55 we denote the mass of the @xmath1 by @xmath56 and its width by @xmath57 . the most precise determinations of @xmath56 and @xmath57 come from the @xmath0 decay channel . measurements in this channel avoid biases associated with the @xmath13 threshold that plague some other decay channels , such as @xmath22 @xcite . the most precise measurements of @xmath56 have been made by the cdf , belle , lhcb , and babar collaborations @xcite . the pdg average for the mass is @xmath58 mev @xcite . we denote the binding energy relative to the @xmath13 threshold by @xmath59 . using the pdg averages for @xmath33 and @xmath60 , we obtain the binding energy @xmath61 more precise measurements of @xmath33 by the lhcb collaboration @xcite and by an analysis of data from the cleoc collaboration @xcite have further decreased the uncertainty in @xmath62 , reinforcing the conclusion that @xmath1 is extremely close to the @xmath13 threshold . the best experimental upper bound on @xmath57 comes from measurements in the @xmath0 decay channel by the belle collaboration @xcite : @xmath63 a theoretical lower bound is provided by the width of the constituent @xmath28 : @xmath64 mev . in this section , we derive the low - energy scattering amplitudes for the three coupled channels consisting of neutral and charged charm mesons and @xmath9 . we then write down simpler scattering amplitudes for the charm mesons only in which the effects of the @xmath9 channel are taken into account implicitly through one of the scattering parameters . finally we write down scattering amplitudes for the charm mesons that take into account the @xmath15 resonance . the low - energy scattering amplitudes @xmath66 from s - wave contact interactions between the three coupled channels defined in eqs . ( [ ch012 ] ) can be expressed as a @xmath26 matrix : @xmath67^{-1 } , \label{f - gk}\ ] ] where @xmath54 is defined in eq . ( [ kmatrix ] ) and @xmath68 is a symmetric @xmath26 matrix of coupling constants . imposing the constraints from isospin symmetry , this matrix has the form @xmath69 where @xmath70 and @xmath71 are the isospin projection matrices defined in eqs . ( [ ispin01 ] ) and @xmath72 , @xmath73 , @xmath74 , and @xmath75 are constants with dimensions of momentum . the matrix of amplitudes in eq . ( [ f - gk ] ) satisfies the lippmann - schwinger equation : @xmath76 this can be verified by inserting eq . ( [ f - gk ] ) for @xmath77 , multiplying on the left by @xmath68 , and multiplying on the right by @xmath78 , in which case it reduces to a trivial identity . the explicit expressions for the scattering amplitudes @xmath66 in eq . ( [ f - gk ] ) are @xmath79/d , \label{f00 } \\ f_{01 } & = & [ ( \gamma_1 -\gamma_0 ) ( -\gamma_v + \kappa_{\psi \omega } ) - 2 \gamma_x^2 ] /d , \label{f01 } \\ f_{11 } & = & [ ( -\gamma_0 -\gamma_1 + 2 \kappa ) ( -\gamma_v + \kappa_{\psi \omega } ) - 2 \gamma_x^2 ] /d , \label{f11 } \\ f_{02 } & = & 2 ( -\gamma_1 + \kappa_1 ) \gamma_x / d , \label{f02 } \\ f_{12 } & = & -2 ( -\gamma_1 + \kappa ) \gamma_x / d , \label{f12 } \\ f_{22 } & = & [ 2 \gamma_0 \gamma_1 - ( \gamma_0 + \gamma_1 ) ( \kappa_1 + \kappa ) + 2 \kappa_1 \kappa ] /d , \label{f22}\end{aligned}\ ] ] [ f - entries ] where the denominator is @xmath80 ( -\gamma_v + \kappa_{\psi \omega } ) - 2 ( -2\gamma_1 + \kappa_1 + \kappa ) \gamma_x^2 . \label{den}\ ] ] for energies above the appropriate thresholds , the nonrelativistically normalized t - matrix elements @xmath81 for scattering between the three channels are given by the matrix @xmath82 where @xmath83 is the diagonal matrix of reduced masses . the imaginary parts of the scattering amplitudes @xmath66 in eq . ( [ f - gk ] ) can be expressed as @xmath84 f(e)^ * , \label{imf - gk}\ ] ] the t - matrix elements for elastic scattering between the three coupled channels are exactly unitary if the constants @xmath72 , @xmath73 , @xmath74 , and @xmath75 in eq . ( [ gmatrix ] ) are real and if the widths @xmath85 , @xmath86 , and @xmath87 in eqs . ( [ kappa012 ] ) are set to zero . in this case , the imaginary part of @xmath66 is nonzero only if the energy @xmath88 exceeds one of the thresholds @xmath89 , @xmath42 , and @xmath43 . the effects of additional inelastic scattering channels can be taken into account through the analytic continuation of the parameters @xcite . the dominant effects of inelastic scattering channels that correspond to decay products of @xmath13 , @xmath14 , and @xmath9 , such as @xmath22 , @xmath90 , and @xmath12 are taken into account through the widths @xmath85 , @xmath86 , and @xmath87 in @xmath91 , @xmath92 , and @xmath93 . the dominant effects of other inelastic scattering channels can be taken into account through the coupling constants @xmath72 , @xmath73 , @xmath74 , and @xmath75 , which can have positive imaginary parts . for example , the isospin-1 decay mode @xmath0 , in which @xmath94 is dominated by the @xmath95 resonance , can be taken into account through the positive imaginary part of @xmath73 . as the energy @xmath88 approaches the @xmath13 threshold at @xmath96 , the elastic scattering amplitude for @xmath13 must approach the universal expression @xcite @xmath97 it is easy to identify @xmath98 by exploiting the fact that the denominator @xmath99 in eq . ( [ den ] ) is linear in @xmath91 . since @xmath100 , we can set @xmath96 in @xmath92 and @xmath93 . the resulting expression for the inverse scattering length is @xmath101 \gamma_x^2/[-\gamma_v + \kappa_{\psi \omega}(0 ) ] } { \gamma_0 + \gamma_1 - 2 \kappa_1(0 ) + 2 \gamma_x^2/[-\gamma_v + \kappa_{\psi \omega}(0)]}. \label{gamma}\ ] ] the binding energy and the width of the @xmath1 are determined by the real and imaginary parts of @xmath98 . this puts two constraints on the real parts and the small imaginary parts of the four parameters @xmath102 , @xmath103 , @xmath74 , and @xmath75 . if @xmath104 or if @xmath105 is much larger than @xmath106 , the @xmath9 channel decouples from the charm meson channels . the @xmath9 scattering amplitude @xmath107 in eq . ( [ f22 ] ) reduces to @xmath108 . the scattering amplitudes for the 0 and 1 channels reduce to the scattering amplitudes for charm mesons derived in ref . @xcite : @xmath109 [ fbl - entries ] where the denominator is @xmath110 these are the appropriate scattering amplitudes if the @xmath1 resonance is generated dynamically by attractive interactions between the charm mesons . the charm meson scattering amplitudes @xmath111 , @xmath112 , and @xmath113 in eqs . ( [ f - entries ] ) , which take into account the @xmath9 channel explicitly , can be obtained exactly from the amplitudes in eqs . ( [ fbl - entries ] ) by making the substitution @xmath114 the second term on the right side is the product of the @xmath9 scattering amplitude and transition amplitudes proportional to @xmath75 . thus the only effect of the @xmath9 channel on scattering between the charm mesons is to resolve the isospin-0 inverse scattering length into an energy - dependent term from transitions to @xmath9 and a constant @xmath72 that takes into account shorter - distance effects . the @xmath15 charmonium state has quantum numbers @xmath3 . if its mass is close enough to that of the @xmath1 resonance , it can have a significant effect on the charm meson scattering amplitudes near the @xmath13 threshold . it could be responsible for generating the @xmath1 resonance or it could be a separate resonance with quantum numbers @xmath3 . if the @xmath15 is a separate resonance from the @xmath1 , it is expected to be higher in mass . one former candidate for the @xmath15 is a state labelled @xmath116 by the particle data group @xcite . it was discovered by the babar collaboration through @xmath117 decays into @xmath118 in the decay channel @xmath119 @xcite , which implies that its charge conjugation is @xmath120 . its properties were measured more accurately in ref . the @xmath116 was also observed by the belle collaboration in the production channel @xmath121 @xcite , which would have excluded @xmath3 , but that observation could also be attributed instead to the nearby charmonium state @xmath122 at 3927 mev . however a recent analysis by the babar collaboration of @xmath123 determined the spin - parity to be @xmath124 @xcite . this excludes @xmath116 as a candidate for @xmath15 . at this point , there is no well - established resonance besides the @xmath1 that might be identified with @xmath15 . we will however for completeness consider the possibility of a separate @xmath3 resonance with mass above 3872 mev . the coupled - channel problem for low - energy s - wave interactions of neutral and charged charm meson pairs with a @xmath3 charmonium resonance was solved in ref . the scattering amplitudes are those for charm mesons in eqs . ( [ fbl - entries ] ) with the substitution @xmath125 these scattering amplitudes were also studied in ref . they are exactly unitary for real values of the four parameters @xmath72 , @xmath73 , @xmath126 , and @xmath127 . the combination @xmath128 can be identified as the energy of the @xmath15 resonance . in ref . @xcite , charmonium phenomenology was used to obtain the estimate @xmath129 for the coupling constant . the methods of ref . @xcite could be extended to the case with a third scattering channel @xmath9 that also couples to the @xmath15 . in this section , we present line shapes for the @xmath1 resonance in the @xmath12 channel . we first discuss the short - distance factors in a factorization formula for the line shapes . we give an expression for the line shape in which the @xmath9 channel is taken into account explicitly . we then give an expression for the line shape in which the effects of the @xmath9 channel are taken into account implicitly through the scattering parameter @xmath72 . finally we give an expression for the line shape in which the @xmath15 resonance is taken into account . for a production process that involves an energy transfer that is large compared to the low - energy scales @xmath42 and @xmath43 set by the differences between the thresholds , the inclusive production rate summed over all resonant final states @xmath8 satisfies a factorization formula @xcite . if the production process is a decay , such as @xmath131 or @xmath132 , the differential decay rate can be expressed as @xmath133 where the @xmath134 are short - distance factors that are insensitive to the resonance energy @xmath88 . the expression for the matrix im@xmath77 in eq . ( [ imf - gk ] ) can be used to decompose the differential decay rate in eq . ( [ dgamma / de ] ) into contributions proportional to the imaginary parts of the scattering parameters , which appear in the matrix @xmath68 , and the imaginary parts of the threshold factors , which appear in the matrix @xmath135 . the short - distance factors @xmath134 in eq . ( [ dgamma / de ] ) are entries of a positive - definite hermitian matrix . they can be expressed as sums with positive weights of terms of the form @xmath136 , where @xmath137 is a short - distance amplitude for the creation of a pair of particles in the channel @xmath138 . the sum is over transition channels @xmath139 from the initial state to the additional final - state particles besides those in the resonance channel . constraints on these short - distance amplitudes from the symmetries of qcd imply constraints on the short - distance factors @xmath134 . the decay @xmath140 is particularly simple , because the transition @xmath141 between the two spin-0 particles has a single short - distance amplitude @xmath142 . the short - distance factors can therefore be expressed as @xmath143 the isospin symmetry of decays that proceed at the quark level through the heavy quark decay @xmath144 relates the short - distance amplitudes for @xmath145 and @xmath140 : @xmath146 the constraints on the coefficients @xmath147 and @xmath148 were derived previously @xcite . the constraint on the coefficients @xmath149 follows from the equality of the short - distance amplitudes for @xmath150 and @xmath151 , which is required by isospin symmetry . using the simple form for the short - distance factors for @xmath152 in eq . ( [ gamma - c ] ) , the differential decay rate in eq . ( [ dgamma / de ] ) reduces to @xmath153\ , [ c_j f_{jl}(e)]^*\ , \left [ { \rm i m } g_{kl } - { \rm i m } k_{kl}(e ) \right ] de . \label{dgamma / de : btok}\ ] ] we can obtain order - of - magnitude estimates for the ratios of @xmath154 , @xmath155 , and @xmath156 from measured partial widths of @xmath117 into @xmath135 plus appropriate pairs of mesons . the decay amplitude into three mesons , such as @xmath157 , is a function of two lorentz invariants whose range extends over the dalitz plot for the three mesons . the amplitudes @xmath147 , @xmath148 , and @xmath149 are the short - distance factors of the decay amplitudes in the corner of the dalitz plot corresponding to the threshold for the two mesons other than @xmath135 . our estimates of their ratios are based on the assumption that the short - distance factors do not vary dramatically over the dalitz plot . the partial widths of @xmath117 into @xmath135 plus pairs of charm mesons have been measured by the babar collaboration in ref . the partial widths into @xmath135 plus @xmath9 were measured by the babar collaboration in ref . @xcite . using the data from @xmath158 decays , we estimate @xmath159 by dividing the sum of the partial widths for @xmath14 and @xmath160 by the sum of the partial widths for @xmath13 and @xmath16 . using the data from @xmath161 decays , we must interchange the numerator and denominator . the resulting estimates for @xmath159 are 0.14 from @xmath158 decays and 0.17 from @xmath161 decays . using the data from @xmath158 decays , we estimate @xmath162 by dividing the partial width for @xmath9 by the sum of the partial widths for @xmath13 and @xmath16 . using the data from @xmath161 decays , we must replace the denominator by the sum of the partial widths for @xmath14 and @xmath160 . the resulting estimates for @xmath162 are 0.037 from @xmath158 decays and 0.036 from @xmath161 decays . these estimates suggest that the short - distance production rates for @xmath9 and for @xmath14 and @xmath163 are smaller than those for @xmath13 and @xmath164 by factors of about 30 and 6.5 , respectively . the suppression of @xmath149 relative to @xmath147 and @xmath148 does not necessarily imply that the @xmath149 term in the resonance factor in eq . ( [ dgamma - psiomega ] ) can be neglected . the @xmath147 and @xmath148 terms in the resonance factor are multiplied by @xmath75 , which is an amplitude for a transition between @xmath9 and a pair of charm mesons . since this process involves a rearrangement of constituent charm quarks between the two mesons , @xmath75 could provide a sufficient suppression factor to make the @xmath147 and @xmath148 terms comparable in strength to the @xmath149 term . for production of the resonance in other channels , such as @xmath165 and @xmath22 , the @xmath149 term in the resonance factor should be completely negligible . between contact interactions : ( a ) simple bubble diagram with a @xmath9 cut , ( b ) diagram with a @xmath166 cut . the @xmath36 , @xmath11 , and pions are represented by double solid , wavy , and dashed lines , respectively . , title="fig:",width=226 ] between contact interactions : ( a ) simple bubble diagram with a @xmath9 cut , ( b ) diagram with a @xmath166 cut . the @xmath36 , @xmath11 , and pions are represented by double solid , wavy , and dashed lines , respectively . , title="fig:",width=226 ] the inclusive differential decay rate for @xmath167 in eq . ( [ dgamma / de : btok ] ) can be partially resolved into contributions from individual resonant states by inserting the expressions for @xmath168 in eq . ( [ imf - gk ] ) . the imaginary part of @xmath169 comes from cutting the bubble diagram in fig . [ fig : psiomega](a ) , in which @xmath36 and @xmath11 propagate between points where they are created and annihilated . the term in the differential rate proportional to im@xmath93 therefore represents the contribution from the final state @xmath9 or from decay products of this pair of particles : @xmath170 = \big| \mbox{$\sum$}_{i=0,1,2 } c_i f_{i2}(e ) \big|^2 \left ( - { \rm i m } \kappa_{\psi \omega}(e ) \right ) \ , de , \label{dgamma - psiomega}\ ] ] where @xmath171 are the scattering amplitudes in eqs . ( [ f - entries ] ) . the imaginary part of the function @xmath172 in eq . ( [ kappav ] ) can be expressed in analytic form : @xmath173 if @xmath88 is above the threshold @xmath43 by much more than @xmath174 , the expression in eq . ( [ imkappav ] ) reduces to @xmath175^{1/2}$ ] . in this region of @xmath88 , eq . ( [ dgamma - psiomega ] ) is the differential rate for producing @xmath36 and @xmath11 on their mass shells . if @xmath88 is below the threshold @xmath43 by much more than @xmath176 , the expression in eq . ( [ imkappav ] ) reduces to latexmath:[$ ( \frac18 \mu_{\psi \omega } / @xmath88 , eq . ( [ dgamma - psiomega ] ) is the differential rate for producing @xmath36 plus the decay products of a virtual @xmath11 , such as @xmath4 . the contribution from the specific decay channel @xmath12 can be obtained from eq . ( [ dgamma - psiomega ] ) by multiplying by the branching fraction @xmath178 . as the energy @xmath88 approaches the @xmath13 threshold , the elastic scattering amplitude for @xmath13 approaches the universal expression in eq . ( [ f - gamma ] ) . the pole at @xmath179 arises from the denominator @xmath180 , which is a common factor in all the scattering amplitudes in eq . ( [ f - entries ] ) . thus all the scattering amplitudes have that same energy dependence near the @xmath13 threshold . the linear combination of scattering amplitudes that appears in the resonance factor in eq . ( [ dgamma - psiomega ] ) has the behavior @xmath181 \gamma_x c_0 - 2 \gamma_1 \gamma_x c_1 - [ 2 \gamma_1 \gamma_0 - ( \gamma_1 + \gamma_0 ) \kappa_1 ] c_2 } { [ \gamma_0 + \gamma_1 - 2 \kappa_1 ] [ -\gamma_v + \kappa_{\psi \omega } ] + 2 \gamma_x^2}\ , \frac{1}{-\gamma + \kappa(e ) } , \label{sumcf - gamma}\ ] ] where @xmath92 and @xmath93 are evaluated at @xmath96 . thus the resonance factor in eq . ( [ dgamma - psiomega ] ) has the simple universal form @xmath182 at energies @xmath88 small compared to the thresholds @xmath42 and @xmath43 , which are both approximately 8 mev . if the short - distance factor @xmath149 for the production of @xmath9 is sufficiently small , it is not essential to take the @xmath9 channel into account explicitly . it can be taken into account implicitly through the isospin-0 inverse scattering length @xmath72 . in the factorization formula in eq . ( [ dgamma / de : btok ] ) , @xmath72 appears in the coupling constant matrix @xmath68 . the contribution to im@xmath183 from the @xmath9 channel can be deduced from the substitution for @xmath72 given in eq . ( [ gamma0-psiomega ] ) : @xmath184 where the terms with @xmath185 and @xmath186 have been dropped because they do not contribute to the @xmath9 final state . our final result for the decay rate into the @xmath9 channel is @xmath170 = \big| \mbox{$\sum$}_{i=0,1 } c_i ( f_{i0}(e ) - f_{i1}(e ) ) \big|^2 \frac{\gamma_x^2}{|-\gamma_v+\kappa_{\psi \omega}(e)|^2 } \left ( - { \rm i m } \kappa_{\psi \omega}(e ) \right ) \ , de , \label{dgamma - psiomegabl}\ ] ] where @xmath187 and @xmath188 are the scattering amplitudes in eqs . ( [ fbl - entries ] ) with the substitution for @xmath72 in eq . ( [ gamma0-psiomega ] ) . this result can also be obtained from the expression in eq . ( [ dgamma - psiomega ] ) in which the @xmath9 channel is taken into account explicitly by setting @xmath189 . the scattering amplitudes for charm mesons that take into account the possibility that the @xmath3 charmonium resonance @xmath15 is near the @xmath1 are given by eqs . ( [ fbl - entries ] ) with the substitution for @xmath72 in eq . ( [ gamma0-y ] ) . the corresponding expressions for the line shapes were derived in ref . they take into account the short - distance production of @xmath15 as well as charm mesons . the contributions from the imaginary parts of @xmath72 , @xmath73 , and @xmath126 were taken into account , but the coupling constant @xmath127 was assumed to be real . the line shapes were used to carry out a phenomenological analysis of the @xmath190 , @xmath22 , and @xmath191 channels . we can use the results in ref . @xcite to write down an expression for the line shape in the @xmath9 channel . for simplicity , we ignore the possibility of the short - distance production of @xmath15 . the inclusive line shape in eq . ( [ dgamma / de : btok ] ) for isospin-0 channels produced by @xmath192 reduces to @xmath193 = \big| \mbox{$\sum$}_{i=0,1 } c_i ( f_{i0}(e ) - f_{i1}(e ) ) \big|^2~ { \rm i m } \left ( 1/\gamma_0 + g^2/(e-\nu ) \right)^{-1 } \ , de , \label{dgamma - psiomegachi}\ ] ] where @xmath187 and @xmath188 are the scattering amplitudes in eqs . ( [ fbl - entries ] ) with the substitution for @xmath72 in eq . ( [ gamma0-y ] ) . the expression for the imaginary part in eq . ( [ dgamma - psiomegachi ] ) that corresponds to cutting rules is @xmath194 { \rm im}(g ) \bigg ) . \label{imgammachi}\end{aligned}\ ] ] the first two terms in the parentheses can be interpreted as contributions from inelastic charm meson scattering and from @xmath15 decay , respectively . the imaginary parts of @xmath72 and @xmath195 must be positive . the third term in eq . ( [ imgammachi ] ) can be attributed to interference between inelastic charm meson scattering and @xmath15 decay . positivity of the line shape for all energies @xmath88 requires @xmath196 . the line shape in the @xmath197 channel can be obtained by inserting eq . ( [ imgammachi ] ) into eq . ( [ dgamma - psiomegachi ] ) , by replacing the imaginary parts of @xmath72 , @xmath195 , and @xmath127 by the @xmath9 channel contributions to the imaginary parts , and by multiplying by the branching fraction @xmath198 . in this section , we derive a simple expression for the distribution of the invariant mass @xmath199 of the three pions in the decay channel @xmath12 . we describe the results of the babar analysis of the @xmath199 distribution and point out that the probability for the quantum numbers @xmath3 for the @xmath1 was underestimated . we describe previous theoretical analyses of the @xmath199 distribution , which ignored the effects of experimental resolution . we also study the effects of scattering on the @xmath199 distribution . the differential decay rate in eq . ( [ dgamma - psiomega ] ) is differential only in the energy @xmath88 . we proceed to derive the @xmath201 distribution within the same framework . the factor of @xmath202 in eq . ( [ dgamma - psiomega ] ) comes from cutting the bubble diagram in fig . [ fig : psiomega](a ) . in the case of the final state @xmath12 , we can obtain an expression that is differential in additional variables by replacing that cut diagram by a cut of the diagram in fig . [ fig : psiomega](b ) in which the cut passes through @xmath12 . the distribution of the invariant mass @xmath199 of the pions would be obtained by integrating over all the other pion variables besides @xmath199 . this distribution can also be obtained more simply from the @xmath9 cut diagram in fig . [ fig : psiomega](a ) . in the nonrelativistic limit , the relation between @xmath199 , the total energy @xmath203 , and the relative momentum @xmath204 of the @xmath36 or the virtual @xmath11 is @xmath205 if the width of the @xmath11 is included in its propagator , the momentum integral for the @xmath9 cut diagram multiplied by @xmath206 is @xmath207 the lower limit on the integral over @xmath199 extends below the physical lower limit of @xmath208 , but the unphysical region is strongly suppressed by the breit - wigner factor . upon integrating over @xmath199 , eq . ( [ int3pi ] ) reproduces the expression for @xmath202 in eq . ( [ imkappav ] ) . an expression for the differential decay rate that is differential in both @xmath88 and @xmath199 can therefore be obtained by replacing @xmath202 by the integrand on the right side of eq . ( [ int3pi ] ) : @xmath209 = \big| \mbox{$\sum$}_{i } c_i f_{i2}(e ) \big|^2 \frac{\gamma_\omega \ ; q } { 2 \pi [ ( m_{3\pi } - m_\omega)^2 + \gamma_\omega^2/4]}\ , dm_{3\pi}\ , dm , \label{dgamma / dedm}\end{aligned}\ ] ] where @xmath210 and @xmath211^{1/2}$ ] is the relative momentum of the @xmath36 or the virtual @xmath11 . this expression for the differential decay rate is lorentz invariant . the dependence on @xmath199 is simply the product of @xmath204 and a breit - wigner resonance function . the differential decay rate into @xmath12 can be obtained by multiplying the right side of eq . ( [ dgamma / dedm ] ) by the branching fraction @xmath198 . integrated over the @xmath166 invariant mass @xmath45 from 3862.5 mev to 3882.5 mev . the data points are the babar data from ref . the histograms are for monte carlo events generated by babar under the assumption that the coupling of @xmath1 to @xmath9 is s - wave ( dashed lines ) or p - wave ( solid lines ) @xcite . the vertical dashed line marks the position of the center of the @xmath11 resonance . ] the babar data on the decay of @xmath1 into @xmath9 that favors the quantum numbers @xmath6 over @xmath3 is the @xmath201 distribution shown in fig . [ fig : babar ] @xcite . the @xmath166 invariant mass @xmath45 is integrated over the range from 3862.5 mev to 3882.5 mev , which extends about 10 mev above and below the @xmath13 threshold . the range of @xmath212 in fig . [ fig : babar ] is from 740 mev to 791.8 mev , which is approximately @xmath213 to @xmath214 . the babar data in fig . [ fig : babar ] consists of @xmath215 events including a background of @xmath216 events . also shown in fig . [ fig : babar ] are histograms of monte carlo events generated by the babar collaboration under the assumptions that the coupling of @xmath8 to @xmath9 is either s - wave or p - wave . the histograms are normalized to 34 events . since the p - wave monte carlo gives a better fit to the @xmath201 distribution , the babar collaboration concluded that the quantum numbers @xmath6 are favored over @xmath3 . a quantitative measure of the quality of the fit is @xmath7 of the histogram with respect to the 6 nonzero data points . in ref . @xcite , the values of @xmath7 per degree of freedom were given as @xmath217 the probabilities for @xmath7 to be larger than these values are 7.1% and 61.9% , respectively . this seems to indicate that p - wave coupling of @xmath8 to @xmath9 ( and therefore quantum numbers @xmath6 ) is strongly favored over s - wave coupling ( and quantum numbers @xmath3 ) . however @xmath218 is a function of the normalization @xmath219 of the histograms . the values of @xmath7 given in ref . @xcite were for histograms normalized to @xmath220 events , which is the central value of the sum of the data points . normalizing them in this way is fine for illustrating differences in their qualitative behavior , as in fig . [ fig : babar ] . however it is not appropriate for calculating the @xmath7 , because it does not allow for independent fluctuations in the 6 bins . instead it requires that any downward fluctuations in some bins be compensated by upward fluctuations in other bins . furthermore there is no guarantee that the probability distribution for @xmath221 is the standard @xmath7 probability distribution . the quantity that has the probability distribution of @xmath7 for 5 degrees of freedom in the limit of ideal measurements is @xmath218 minimized with respect to @xmath219 . the minimum @xmath7 per degree of freedom for the babar data is @xmath222 for the s - wave monte carlo , the minimum is at @xmath223 and the probability for @xmath7 to be larger than the observed value is 18.7% . for the p - wave monte carlo , @xmath224 and the probability for @xmath7 to be larger than the observed value is 66.2% . while a p - wave coupling is still favored over s - wave , it is not favored as strongly as reported in ref . @xcite . in order to compare a theoretical distribution to one that is measured , it is essential to take into account the experimental resolution . in the babar experiment , the experimental resolution on @xmath45 is @xmath225 mev @xcite . the experimental resolution on @xmath201 was not given in ref . a reasonable estimate is the resolution of the @xmath11 mass in a study of the decay @xmath226 , which was @xmath227 mev @xcite . these resolutions can be taken into account by convolving the distribution in eq . ( [ dgamma / dedm ] ) with gaussians in @xmath45 and @xmath201 : @xmath228}{dm_{3\pi}'\,dm'}\ , \frac{e^{-(m - m')^2/2\sigma_x^2}}{\sqrt{2\pi}\sigma_x } \frac{e^{-(m_{3\pi}-m'_{3\pi})^2/2\sigma_\omega^2}}{\sqrt{2\pi}\sigma_\omega } \ , . \label{dgamma_smeared}\ ] ] although it was not stated explicitly in ref . @xcite , the babar data and histograms in fig . [ fig : babar ] are uncorrected for acceptances and efficiencies @xcite . this can be deduced from the fact that the central value of each data point for the combined distributions from @xmath158 and @xmath161 decay in fig . 4c of ref . @xcite is equal to the sum of the central values of the data points for the separate distributions from @xmath158 decay and @xmath161 decay in figs . 4a and 4b . since the babar data shown in fig . [ fig : babar ] are uncorrected , direct comparisons with theoretical distributions for @xmath199 are not appropriate . in ref . @xcite , the generator for babar s p - wave monte carlo differs from the generator for the s - wave monte carlo by a multiplicative factor of @xmath229 , where @xmath204 is the relative momentum of the @xmath36 and @xmath230 . however the generator used for babar s s - wave monte carlo is not stated in ref . @xcite . based on the limited information provided , a plausible guess is that the generator is equivalent to eq . ( [ dgamma / dedm ] ) with the resonance factor @xmath231 replaced by @xmath232 , where @xmath233 mev is the central value of the @xmath1 mass from babar s fit to the @xmath12 invariant mass distribution . this central value corresponds to a negative binding energy @xmath234 mev , but it is consistent within errors with the small positive binding energy in eq . ( [ ex ] ) . integrated over the @xmath166 invariant mass @xmath45 from 3862.5 mev to 3882.5 mev . the data points and histograms are as described in fig . [ fig : babar ] . the three curves are calculated from eq . ( [ dgamma / dedm ] ) using ( 1 ) the resonance factor @xmath235 and no experimental resolution ( dotted curve ) , as in ref . @xcite , ( 2 ) a breit - wigner resonance factor with width 1.7 mev and no experimental resolution ( dash - dotted curve ) , as in ref . @xcite , ( 3 ) the resonance factor @xmath235 and experimental resolution of 5.6 mev in @xmath201 ( solid curve ) . the vertical dashed line marks the position of the center of the @xmath11 resonance . ] there have been two previous theoretical analyses @xcite of the @xmath199 distribution shown in fig . [ fig : babar ] . both analyses were based on the incorrect implicit assumption that the babar data were corrected for acceptances and efficiencies . both analyses also ignored the experimental resolution . hanhart et al . carried out a combined analysis of data from the belle collaboration on the decay into @xmath165 and from the babar collaboration on the decay into @xmath9 , comparing the options @xmath3 and @xmath6 for the quantum numbers of the @xmath1 @xcite . the belle data was the distribution of @xmath236 from @xmath237 , and it consisted of approximately 200 events in 19 bins @xcite . the babar data was the distribution of @xmath201 from @xmath238 , and it consisted of only 25 events in 6 bins @xcite . the theoretical distributions for @xmath236 and @xmath201 in ref . @xcite take into account @xmath239 mixing and the energy dependence of the @xmath240 and @xmath11 widths . for the mass of the @xmath1 , the authors used @xmath241 mev , and they ignored its width . they also ignored the experimental resolutions of the invariant masses @xmath242 and @xmath212 of the pions and @xmath45 of the system consisting of @xmath36 and pions . the resulting distribution for @xmath201 drops to 0 sharply at 775 mev . for the s - wave case , it is well - approximated by eq . ( [ dgamma / dedm ] ) with the resonance factor @xmath243 replaced by @xmath235 . this simple distribution is illustrated in fig . [ fig : m3pi ] , where it has been normalized so that the area under the curve is 34 events . the shape of the curve does not resemble that of the babar data or babar s s - wave monte carlo histogram , primarily because the experimental resolution on @xmath199 was ignored . as a measure of the quality of the combined fit , ref . @xcite used the @xmath7 per degree of freedom for the babar and belle data sets . given the large error bars in the babar data and the small number of data points , this measure is sensitive only to the total number of @xmath244 events and not to the shape of the @xmath201 distribution . the authors concluded from their analysis that the combined belle and babar data favor the quantum numbers @xmath3 . faccini et al . carried out an analysis @xcite that also included the belle data on angular distributions for the decay into @xmath245 @xcite . the authors used @xmath246 mev , and they took the width of the @xmath1 to be 1.7 mev . this is larger than the upper bound on the width in eq . ( [ gammax ] ) . they ignored the experimental resolutions of @xmath242 , @xmath201 , and @xmath45 . for the s - wave case , the resulting distribution for @xmath201 is well - approximated by eq . ( [ dgamma / dedm ] ) with @xmath231 replaced by a breit - wigner function of @xmath88 centered at @xmath96 with width 1.7 mev . this distribution is illustrated in fig . [ fig : m3pi ] , where it has been normalized so that the area under the curve is 34 events . the sharp cutoff on @xmath201 at 775 mev in ref . @xcite has been replaced by a tail from the breit - wigner that extends up to about 785 mev . the shape of the curve does not resemble that of the babar data or babar s s - wave monte carlo histogram , primarily because the experimental resolution on @xmath199 was ignored . in the combined fit to the @xmath190 and @xmath244 data , the @xmath7 per degree of freedom favors the quantum numbers @xmath3 , but this measure is sensitive only to the total number of @xmath244 events . according to ref . @xcite , the analysis of the @xmath244 data alone excludes @xmath3 at the 99.9% confidence level . however this probability should not be taken seriously , because experimental resolution was ignored in the analysis and because the babar data were not corrected for acceptances and efficiencies . to take into account the resolution in the experiment of ref . @xcite , the distribution in eq . ( [ dgamma / dedm ] ) should be convoluted with a gaussian in @xmath45 of width @xmath225 mev and a gaussian in @xmath201 of width @xmath247 mev as in eq . ( [ dgamma_smeared ] ) . the @xmath201 distribution is then obtained by integrating over @xmath45 from 3862.5 mev to 3882.5 mev . if we assume the resonance factor @xmath231 is dominated by a region near the @xmath13 threshold whose width is small compared to 6.7 mev , we can replace the resonance factor by a delta function at @xmath96 . the effect of integrating over the 20 mev range of @xmath88 is then to constrain @xmath45 to be equal to the @xmath13 threshold . the experimental resolution on @xmath45 appears only in a multiplicative factor , so it does not affect the shape of the @xmath201 distribution . the resulting @xmath201 distribution is shown in fig . [ fig : m3pi ] , where it has been normalized so that the area under the curve is 34 events . the shape of this distribution is much closer to both the babar data and babar s s - wave monte carlo than the distributions in which energy resolution was ignored . both of the previous theoretical analyses took into account the energy dependence of the width @xmath87 of the @xmath11 resonance @xcite . the energy dependence comes primarily from the total phase space for the decay @xmath248 . the phase space increases by about 2% as the invariant mass increases by @xmath50 mev from @xmath249 mev to @xmath250 . thus the effect of the energy - dependent width is not very dramatic . since the babar data in fig . [ fig : babar ] is uncorrected for acceptances and efficiencies , direct comparisons with theoretical @xmath199 distributions are not appropriate . however , given the relatively low probability for babar s s - wave monte carlo , it is worth asking whether there are aspects of the @xmath1 resonance that could improve the agreement between the s - wave monte carlo and the data . better agreement could have been obtained with a generator that gives an @xmath199 distribution whose peak is shifted lower by about 10 mev by suppressing the distribution above 770 mev . the differential rate for s - wave coupling to @xmath9 in eq . ( [ dgamma / dedm ] ) implies that the @xmath199 distribution has the form @xmath251}{dm_{3\pi}}= \frac{\gamma_\omega } { 2 \pi [ ( m_{3\pi } - m_\omega)^2 + \gamma_\omega^2/4 ] } \int_{e_{\rm min } } \!\!\!\!\!\!de~ \big| \mbox{$\sum$}_{i } c_i f_{i2}(e ) \big|^2 q(e ) , \label{dgamma / dm}\end{aligned}\ ] ] where @xmath252^{1/2}$ ] , @xmath253 , and the upper endpoint of the integral over @xmath88 is well above the @xmath10 threshold region . the last factor in eq . ( [ dgamma / dm ] ) is the integral of the line shape weighted by the relative momentum @xmath204 . we wish to determine whether the dependence of this factor on @xmath199 could improve the agreement between the s - wave monte carlo and the babar data in fig . [ fig : babar ] . the generator for babar s p - wave monte carlo produced a downward shift in the peak of the @xmath199 distribution by about 10 mev through an additional multiplicative factor of @xmath229 . if the line shape can be approximated by a delta function near @xmath96 , the factor of @xmath254 has a zero at @xmath255 mev . we wish to determine whether a comparable shift can be produced instead by a change in the resonance factor . integrated over the @xmath166 invariant mass @xmath45 from 3862.5 mev to 3882.5 mev . the experimental resolutions of 6.7 mev in @xmath45 and 5.6 mev in @xmath201 are taken into account . the curves are calculated using the binding energy @xmath256 mev and a resonance factor that is either a delta function at @xmath257 ( blue solid curve ) or the universal resonance factor with the minimal width @xmath258 mev ( black dashed curves ) . the shaded band takes into account variations in @xmath62 from 0 to 0.65 mev and in @xmath57 from 0.066 to 1.2 mev . ] according to eq . ( [ sumcf - gamma ] ) , the resonance factor sufficiently near the @xmath13 threshold has the universal line shape @xmath182 . this universal resonance factor can not be approximated by a delta function in @xmath88 , because it has power - law tails that decrease as @xmath259 at large @xmath260 . when integrated over a smooth distribution in @xmath88 , it can be approximated by the sum of a lorentzian in @xmath88 and @xmath261 tails at positive and negative energies . the lorentzian is centered at @xmath262 with width @xmath57 , where @xmath62 and @xmath57 are the binding energy and width of the @xmath1 , and it can be approximated by @xmath263 . the tails have the form @xmath259 that extend from small @xmath260 of order @xmath62 to large @xmath260 of order @xmath42 or @xmath43 . the relative sizes of the integrals of the line shape over the peak near @xmath262 and over the tails are approximately @xmath264 and @xmath265 , so the contribution from the tails can be significant . the @xmath201 distributions for the universal resonance factor @xmath182 and for @xmath266 , with the resolutions in @xmath45 and @xmath201 taken into account , are compared in fig . [ fig : m3pib ] . replacing @xmath266 by the universal resonance factor gives a negligible shift in the position of the peak in @xmath201 . the variations in the @xmath201 distribution for the universal resonance factor from the uncertainties in @xmath62 and @xmath57 are also shown in fig . [ fig : m3pib ] . the small shifts in the position of the peak come primarily from the variations in @xmath62 . babar s p - wave monte carlo produced a significant shift in the position of the peak of the @xmath199 distribution through a multiplicative factor of @xmath254 , which has a zero at @xmath267 . it is plausible that a significant shift could also be produced by an approximate zero in the resonance factor instead . an approximate zero of @xmath268 could arise from interference between the scattering channels . to suppress the region of @xmath199 above 765 mev , the approximate zero of @xmath268 would have to be at an energy @xmath269 above the @xmath13 threshold . the shape of the general resonance factor @xmath231 depends on the ratios @xmath270 and @xmath271 of the complex short - distance factors and on the scattering parameters @xmath72 , @xmath73 , @xmath75 , and @xmath74 . the short - distance factors @xmath142 can be complex , so an approximate zero requires a fine tuning of these coefficients . we can examine this possibility with a simplified form of the resonance factor that has fewer adjustable parameters . we set @xmath189 , which implies that production of the @xmath1 resonance is dominated by the creation of charm meson pairs at short distance rather than @xmath272 . this assumption is motivated by the suppression of @xmath149 suggested by the estimates in section [ sec : lineshapesd ] . the resonance factor then reduces to @xmath273 , where the scattering amplitudes @xmath274 and @xmath275 are given in eqs . ( [ f02 ] ) and ( [ f12 ] ) . the numerator of the resonance factor is proportional to @xmath276 . if there is an interference zero at @xmath269 , the short - distance coefficients must satisfy @xmath277 we further assume @xmath278 , which implies that the @xmath272 channel would not have a bound state near threshold in the absence of the @xmath1 resonance . the denominator @xmath180 in @xmath274 and @xmath275 then reduces to the denominator @xmath279 for the 2-channel case in eq . ( [ denbl ] ) , with @xmath72 replaced by @xmath280 the scattering parameter @xmath73 is determined by the value of @xmath98 as in eq . ( [ gamma ] ) , which reduces to @xmath281 thus @xmath282 is the only scattering parameter that affects the shape of the resonance factor . the adjustable parameters in our simplified resonance factor are @xmath282 and @xmath269 . we wish to determine whether the peak in the @xmath199 distribution can be shifted downward by about 10 mev by adjusting these parameters . we therefore consider @xmath269 in the range between 0 and 10 mev . for some values of @xmath282 , there is another resonance with energy below @xmath1 , which is inconsistent with observations . demanding no such resonance constrains @xmath283 mev . there can also be a substantial enhancement near @xmath284 from a cusp associated with the opening up of the @xmath285 threshold . this has the opposite effect of the suppression above the threshold that we want . to avoid the enhancement , we require @xmath286 , which combined with eq . implies @xmath287 mev or @xmath288 mev . for the given regions of @xmath269 and @xmath289 , we are unable to obtain a significant shift in the peak of the @xmath199 distribution to lower mass . another way to suppress the resonance factor for @xmath88 above the @xmath290 threshold is through interference with the low - energy tail of the @xmath15 resonance . the inclusive line shape in the isospin-0 channel is given in eq . it can be resolved into contributions proportional to the imaginary parts of @xmath72 , @xmath195 , and @xmath127 by using eq . . the line shape in the @xmath9 channel can be obtained by replacing the imaginary parts of @xmath72 , @xmath195 , and @xmath127 in eq . ( [ imgammachi ] ) by the contributions to those imaginary parts from the @xmath9 channel , and allowing those imaginary parts to be energy - dependent . for example , the substitution for im@xmath183 is given in eq . ( [ imgamma0-psiomega ] ) . if we use the assumption @xmath291 , the substitution for im@xmath183 reduces to @xmath202 multiplied by a constant . the substitutions for @xmath195 and @xmath127 would reduce to similar forms . the expression in eq . then reduces to the product of the resonance factor @xmath292 , the threshold factor @xmath202 , and a quadratic function of the energy @xmath88 . the maximum possible interference effect corresponds to total destructive interference at some energy @xmath269 above the @xmath13 threshold . in this case , the quadratic function of @xmath88 reduces to @xmath293 and the expression in eq . ( [ imgammachi ] ) reduces to @xmath294_{j/\psi\ , \omega } \longrightarrow\ , \frac { -\text{im}\,\kappa_{\psi \omega } ( e ) } { |e-\nu+g^2 \gamma_0|^2 } \,\frac{2\gamma_x^2}{|\gamma_v|^2 } \ , ( e - e_0)^2 , \label{imgamma0-last}\ ] ] we vary the position of interference zero by changing @xmath269 between 0 and 10 mev . we are unable to obtain a significant shift in the peak of the distributions to smaller @xmath199 by tuning the interference position @xmath269 . we are also unable to obtain a significant shift using zeroes in both the @xmath1 resonance factor @xmath295 and the @xmath15 resonance factor in eq . . the quantum numbers of the @xmath1 have been definitely established as @xmath3 from analyses of the @xmath190 decay channel . this settles an issue raised by a babar analysis of the @xmath199 distribution for the @xmath244 decay channel that preferred @xmath6 over @xmath3 @xcite . we pointed out that in the babar analysis , the quoted values of @xmath7 were not minimized with respect to the adjustable normalizations of the monte carlo distributions . upon minimization of the @xmath7 , the probability for @xmath3 is increased significantly from 7.1% to 18.7% while the probability for @xmath6 is increased only slightly from 61.9% to 66.2% . thus the preference for @xmath6 over @xmath3 was overstated in ref . @xcite . for the benefit of future analyses of the @xmath1 resonance in the @xmath166 decay channel , we considered whether a more accurate description of the resonance could have further improved the agreement between the babar data and the babar s - wave monte carlo for the @xmath3 case . to describe the resonance more accurately , we derived the low - energy scattering amplitudes due to s - wave couplings between the three channels in eq . ( [ ch012 ] ) : neutral charm meson pairs , charged charm meson pairs , and @xmath272 . we also considered how the scattering would be affected by an additional @xmath15 resonance with quantum numbers @xmath3 . we used the scattering amplitudes to derive the line shape for the @xmath244 decay channel and also the @xmath199 distribution . the babar p - wave monte carlo that was preferred by the babar data over the s - wave monte carlo gave an @xmath199 distribution whose peak was about 10 mev lower . we considered several mechanisms for shifting the peak for the s - wave case to lower values of @xmath199 . we considered the effects of the power - law tails of the universal scattering amplitude . we considered interference between the charm - meson scattering channels . we also considered the interference from the tail of a higher @xmath15 resonance . for all these mechanisms , the @xmath199 distribution was robust against a shift in the peak shift to lower values . we conclude that , given the resolution in the babar experiment , a more accurate description of the @xmath1 resonance in the @xmath166 decay channel is not essential . the effects we have considered may however be important in future analyses of this decay channel with higher resolution . this research was supported in part by the u.s . department of energy under grant de - fg02 - 91-er40690 and under contract de - fg02 - 94er40818 . we thank a. mokhtar and w. dunwoody for providing the babar results in fig . [ fig : babar ] .

we specialize a recently - proposed determinant formula for the overlap of the zero - momentum nel state with bethe states of the spin-1/2 xxz chain to the case of an odd number of downturned spins , showing that it is still of `` gaudin - like '' form , similar to the case of an even number of down spins . we generalize this result to the overlap of-raised nel states with parity - invariant bethe states lying in a nonzero magnetization sector . the generalized determinant expression can then be used to derive the corresponding determinants and their prefactors in the scaling limit to the lieb - liniger ( ll ) bose gas . the odd number of down spins directly translates to an odd number of bosons . we furthermore give a proof that the nel state has no overlap with non - parity - invariant bethe states . this is based on a determinant expression for overlaps with general bethe states that was obtained in the context of the xxz chain with open boundary conditions . the statement that overlaps with non - parity - invariant bethe states vanish is still valid in the scaling limit to ll which means that the bec state has zero overlap with non - parity - invariant ll bethe states .

1403.7469

the study of out - of - equilibrium dynamics of isolated many - body quantum systems has seen great progress during the last decade . in particular , much theoretical progress was made in the field of one - dimensional quantum integrable models . out - of - equilibrium realizations of these models , despite being strongly - correlated @xcite and having a complete set of algebraically independent conserved quantities , are believed to exhibit relaxation behaviour . this relaxation can be described , in principle , by a generalized gibbs ensemble ( gge ) @xcite . an alternative way was recently proposed in the context of a quantum quench , the so - called `` quench action approach '' @xcite . this first - principles - based analysis of time - evolved expectation values of physical observables uses as input the overlaps of the initial state with energy eigenstates . after a saddle - point approximation that is exact in the thermodynamic limit , it predicts both the time evolution of expectation values of physical observables and their long - time stationary values . to obtain the latter , one in particular needs the leading order term of the overlaps in the thermodynamic limit . despite the fact that slavnov s theorem @xcite gives overlaps between eigenstates of a bethe ansatz - solvable @xcite hamiltonian with ( in principle ) generic states , until recently very little was known about overlaps between eigenstates of two hamiltonians of the same family but at different values of interaction ( here , we primarily focus on two xxz chains with different anisotropies , one of which is in the ising limit ) . it is these overlaps that are required for the application of the quench action approach to interaction quenches . for the lieb - liniger model , exact overlaps of the ground state for an even number of free bosons ( bose - einstein condensate ( bec ) state ) and energy eigenstates at generic finite interaction parameter were conjectured in @xcite and thereafter proven in @xcite . their form resembled the gaudin determinant form for norms of bethe states @xcite , and was suitable for analysis in the thermodynamic limit @xcite . this made an application of the quench action approach possible , leading to a description of the stationary state at late times after the quench . surprisingly , and in contrast to thermal properties of the lieb - liniger gas , this description turned out to yield a closed - form solution in the thermodyamic limit , for any ( repulsive ) value of the final interaction strength . of particular interest was the conclusion that these results remain inaccessible to gge - based calculations @xcite . in the context of spin chains , in @xcite overlaps between the nel state and parity - invariant xxz bethe states with an even number of down spins were derived from an earlier form of the overlaps @xcite . this new form has the same gaudin - like structure as the lieb - liniger overlaps and therefore can be treated in the thermodynamic limit as done in @xcite . note that this result is still restricted to an even number of down spins . in this paper we present a derivation of the overlaps of the zero - momentum nel state and parity - invariant bethe states , but now in the sectors of odd numbers of down spins . we also take the scaling limit to the lieb - liniger model and present the overlaps between the bec state and bose gas bethe states with an odd number of bosons . the calculations are not trivial and the results show a slightly different structure of the determinant as compared to the even particle case @xcite . these new overlap formulas may be specifically useful in broader contexts , such as in their relation to exact solutions of the 1d kardar - parisi - zhang equation @xcite . furthermore , unlike the lieb - liniger case , up to now a proof that non - parity - invariant xxz bethe states have vanishing overlaps with the initial nel state was lacking . in this paper we present this proof , both for an even and odd number of down spins . the relevance for the quench action approach is clear , as the time evolution of expectation values is governed by a double sum over the complete hilbert space , including non - parity - invariant states . the paper is organized as follows . in section [ sec : xxzchain ] we review the bethe ansatz solution of the xxz chain and recall the gaudin norm formula of bethe states @xcite , which will be used in subsequent sections . in section [ sec : oddandnpi ] we present the derivation of the overlap formula with parity - invariant bethe states with an odd number of down spins , and we prove that non - parity - invariant states have zero overlap with the nel state . in section [ sec : scaling_ll ] we perform the scaling limit to the lieb - liniger bose gas and present the bec overlap formula for an odd number of bosons . the model which we initially focus on is the well - known spin-@xmath1 anisotropic heisenberg chain with hamiltonian @xmath2 since we are interested in antiferromagnetic cases , the overall exchange coupling constant is taken to be positive , @xmath3 . the exchange anisotropy is parametrized as @xmath4 with @xmath5 being a free parameter ( we will in practice of course limit ourselves to real @xmath6 ) . we consider a chain with @xmath7 sites ( which we choose to be an even number ) and impose periodic boundary conditions @xmath8 , @xmath9 . this hamiltonian can be diagonalized by bethe ansatz @xcite . starting from the ferromagnetic reference state @xmath10 , we can construct eigenstates with @xmath11 overturned spins as [ eq : ba_state ] @xmath12 where the positions of the downturned spins are labeled by the indices @xmath13 , @xmath14 , in such a way that @xmath15 for @xmath16 . the amplitudes take the bethe ansatz form @xmath17 } \exp\left\{- i \sum_{a=1}^m p(\lambda_{q_a})s_a - \frac{i}{2 } \sum_{\substack { a , b=1\\ b > a}}^m \theta(\lambda_{q_b } - \lambda_{q_a } ) \right\}\epp\ ] ] here , @xmath18 represents the set of all permutations of @xmath19 , and @xmath20}$ ] represents the parity of the permutation . the state is an eigenstate of the total momentum operator with eigenvalue @xmath21\epp\ ] ] the scattering phase shift in the wave function is given by @xmath22 the branches of the logarithms are fixed by @xmath23 and @xmath24 . the quasi - momenta @xmath25 , which form a self - conjugate set @xcite , are called rapidities or bethe roots , and they specify an eigenstate of hamiltonian with energy eigenvalue @xmath26 under the condition that they fulfill the bethe equations @xmath27 the norm of an eigenstate is given in terms of the gaudin determinant @xcite [ eq : norm_bethe_state ] @xmath28 where @xmath29 . we shall write that a bethe state of the form is `` on - shell '' if the parameters @xmath25 satisfy bethe equations . in contrast , we will write `` off - shell '' if they are arbitrary complex numbers . we further call a bethe state parity - invariant if they obey the symmetry condition @xmath30 , where , for generic @xmath4 , i.e. @xmath31 or @xmath32 , @xmath33 not a root of unity , we identify rapidities when their imaginary parts differ by @xmath34 . this in particular means that the two points @xmath35 are identified . hence , adding a single rapidity at @xmath36 to a parity - invariant set of rapidities does not destroy this symmetry . in the root - of - unity case , e.g. when @xmath37 , we consider rescaled rapidities , e.g. @xmath38 or @xmath39 respectively , and we identify the two points @xmath40 , because in these special cases bethe equations allow for these type of solutions . in the following we are interested in overlaps of zero - magnetization xxz bethe states with the zero - momentum nel state , which is the symmetric combination of nel and anti - nel . whenever we mention in the following the nel state @xmath41 we mean this linear combination with momentum zero . before proceeding further , let us define four classes of zero - momentum bethe states , and mention where they are specifically treated : * parity - invariant bethe states with an even number of down spins ( sec [ sec : pi_even ] , ref . @xcite ) , * parity - invariant bethe states with an odd number of down spins ( sec . [ sec : pi_odd ] ) , * non - parity - invariant bethe states without any bethe root at zero ( sec . [ sec : npi_even ] ) , * non - parity - invariant bethe states with one bethe root at zero ( sec . [ sec : npi_odd ] ) . the number of downturned spins is given by the number of bethe roots @xmath42 . in this section we only consider bethe states lying in the zero - magnetization sector ( @xmath43 ) . the nonzero magnetization sector shall become important for the scaling to the ll model in sec . [ sec : scaling_ll ] . we denote a parity - invariant ( off - shell ) bethe state with an even number of down spins by @xmath44 . in ref . @xcite , an explicit formula for the overlap @xmath45 of the nel state @xmath46 with parity - invariant bethe states in the zero - magnetization sector @xmath43 was given ( see sec . [ sec : pi_even ] ) . this could be extended to the sector of nonzero magnetization ( @xmath47 ) in ref . @xcite , where the overlaps with so - called @xmath0-raised nel states were derived . however , these results are still restricted to even numbers of down spins . in sec . [ sec : pi_odd ] we present an explicit expression for the overlap of the nel state with a bethe state with an odd number of parity - invariant bethe roots valid for generic @xmath6 . the result shows a slightly different structure of the determinant as compared to the case of even number of down spins . note that an odd number of down spins implies that there is one bethe root at the origin or at @xmath36 , the rest being arranged in symmetric pairs ( around the origin ) . thereafter we shall present a proof that non - parity - invariant bethe states have no overlap with the nel state for both cases , even and odd numbers of bethe roots . the overlap of the zero - momentum nel state with an unnormalized parity - invariant xxz off - shell state with an even number of down spins is given by @xcite ( @xmath7 divisible by four , number of particles @xmath43 even ) [ eq : overlap_xxz_offshell ] @xmath48 where the prefactor @xmath49 and the matrix @xmath50 read @xmath51\left[\prod_{\substack{j > k=1\\ \ \sigma=\pm}}^{n/4}\frac{{s_{\lambda_j+\sigma\lambda_k,\eta}}{s_{\lambda_j+\sigma\lambda_k,-\eta}}}{{s_{\lambda_j+\sigma\lambda_k,0}}^2}\right]\epc \\[3ex ] g_{jk}^{+ } & = \delta_{jk}\left(n{s_{0,\eta}}k_{\eta/2}(\lambda_j)-\sum_{l=1}^{n/4}{s_{0,\eta}}k_\eta^{+}(\lambda_j,\lambda_l)\right ) + { s_{0,\eta}}k_\eta^{+}(\lambda_j,\lambda_k)\notag \\ \label{eq : gaudin_plus } & \qquad\quad + \delta_{jk}\frac{{s_{2\lambda_j,\eta}}\,\mathfrak{a}_j+{s_{2\lambda_j,-\eta}}\,\bar{\mathfrak{a}}_j}{{s_{2\lambda_j,0 } } } + ( 1-\delta_{jk})f_{jk}\epc \quad\qquad j , k=1,\ldots , n/4\epc\\[3ex ] \label{eq : f_jk } f_{jk } & = \mathfrak{a}_k\left ( \frac{{s_{2\lambda_j,\eta } } { s_{0,\eta}}}{{s_{\lambda_j+\lambda_k,0}}{s_{\lambda_j-\lambda_k,\eta } } } - \frac{{s_{2\lambda_j,-\eta}}{s_{0,\eta}}}{{s_{\lambda_j-\lambda_k,0}}{s_{\lambda_j+\lambda_k,-\eta } } } \right ) + \mathfrak{a}_k\bar{\mathfrak{a}}_j \left(\frac{{s_{2\lambda_j,-\eta}}{s_{0,\eta}}}{{s_{\lambda_j-\lambda_k,0}}{s_{\lambda_j+\lambda_k,-\eta}}}\right ) \notag\\ & \quad - \bar{\mathfrak{a}}_j\left(\frac{{s_{2\lambda_j,-\eta}}{s_{0,\eta}}}{{s_{\lambda_j-\lambda_k,0}}{s_{\lambda_j+\lambda_k,-\eta } } } + \frac{{s_{2\lambda_j,-\eta}}{s_{0,\eta}}}{{s_{\lambda_j+\lambda_k,0}}{s_{\lambda_j-\lambda_k,-\eta}}}\right)\end{aligned}\ ] ] with @xmath52 we also introduced the abbreviations @xmath53 and @xmath54\left(\frac{{s_{\lambda_j,\eta/2}}}{{s_{\lambda_j,-\eta/2}}}\right)^{n}\epp\ ] ] formula still holds for the bra states @xmath55 or @xmath56 . note that then @xmath49 has to be modified by a factor @xmath57 . we consider the case when an odd number of spins are turned down . we choose @xmath58 divisible by four , @xmath59 odd . a parity - invariant off - shell bethe state has either one bethe root at the origin or at @xmath36 . the former leads to a state with total momentum @xmath60 , the latter to a state with @xmath61 [ see eq . ] . since we are interested in the overlap with the zero - momentum nel state @xmath41 , only the latter matters . we define @xmath62 as the number of bethe roots with positive real part . the overlap of the nel state @xmath41 with these ( unnormalized ) parity - invariant off - shell states is ( see [ sec : odd_proof ] , @xmath63 ) [ eq : overlap_xxz_offshell_odd ] @xmath64 where the prefactor @xmath65 and the matrix @xmath66 read @xmath67\left[\prod_{\substack{j > k=1\\ \ \sigma=\pm}}^{m'}\frac{{s_{\lambda_j+\sigma\lambda_k,\eta}}{s_{\lambda_j+\sigma\lambda_k,-\eta}}}{{s_{\lambda_j+\sigma\lambda_k,0}}^2}\right]\epc\\[3ex ] g_{jk}^{(+,odd ) } & = \delta_{jk}\left(n{s_{0,\eta}}k(j)-\sum_{l=0}^{m'}{s_{0,\eta}}k(j , l)\right ) + { s_{0,\eta}}k(j , k)\notag \\ \label{eq : gaudin_plus_odd } & \quad + \delta_{jk}(1-\delta_{j0})\frac{{s_{2\lambda_j,\eta}}\,\mathfrak{a}_j+{s_{2\lambda_j,-\eta}}\,\bar{\mathfrak{a}}_j}{{s_{2\lambda_j,0 } } } + ( 1-\delta_{jk})f_{jk}\epc \quad j , k=0,\ldots , m'\epc\\[3ex ] \label{eq : f_jk_odd } f_{jk } & = \mathfrak{a}_k\left ( \frac{{s_{2\lambda_j,\eta } } { s_{0,\eta}}}{{s_{\lambda_j+\lambda_k,0}}{s_{\lambda_j-\lambda_k,\eta } } } - \frac{(1-\delta_{j0}){s_{2\lambda_j,-\eta}}{s_{0,\eta}}}{{s_{\lambda_j-\lambda_k,0}}{s_{\lambda_j+\lambda_k,-\eta } } } \right ) + \mathfrak{a}_k\bar{\mathfrak{a}}_j \left(\frac{{s_{2\lambda_j,-\eta}}{s_{0,\eta}}}{{s_{\lambda_j-\lambda_k,0}}{s_{\lambda_j+\lambda_k,-\eta}}}\right ) \notag\\ & \quad - \bar{\mathfrak{a}}_j\left(\frac{{s_{2\lambda_j,-\eta}}{s_{0,\eta}}}{{s_{\lambda_j-\lambda_k,0}}{s_{\lambda_j+\lambda_k,-\eta } } } + \frac{(1-\delta_{0k}){s_{2\lambda_j,-\eta}}{s_{0,\eta}}}{{s_{\lambda_j+\lambda_k,0}}{s_{\lambda_j-\lambda_k,-\eta}}}\right)\epc \\[2ex ] k(j , k ) & = \left\ { \begin{array}{r@{\hspace{4ex}}l } \left(k_{\eta}(\lambda_j-\lambda_k ) + k_{\eta}(\lambda_j+\lambda_k)\right)\epc & 1\leq j , k \leq m'\epc \\ \frac{1}{2 } \left(k_{\eta}(\lambda_j-\lambda_k ) + k_{\eta}(\lambda_j+\lambda_k)\right ) \epc & \text{otherwise}\epc \end{array } \right.\end{aligned}\ ] ] and @xmath68 . the abbreviations @xmath69 and @xmath70 now read @xmath71\left(\frac{{s_{\lambda_j,\eta/2}}}{{s_{\lambda_j,-\eta/2}}}\right)^{n}\epp\ ] ] note the additional factors @xmath72 in the `` on - shell - vanishing '' terms and the factors @xmath1 in the functions @xmath73 and @xmath74 as soon as one of the arguments @xmath75 is @xmath76 which addresses the parameter @xmath77 . for on - shell bethe states all @xmath69 , @xmath70 terms vanish . using norm formula we can write the overlap of the nel state @xmath41 with normalized on - shell states as [ eq : overlap_odd ] @xmath78 \notag\\ & \quad\times\frac{\det{}_{\!m'+1}(g_{jk}^{+})}{\sqrt{\det{}_{\!2m'+1}(g_{jk})}}\epc\end{aligned}\ ] ] where @xmath79 this completes the previous result ( see sec . [ sec : pi_even ] and ref . @xcite ) of an even number of downturned spins . in this subsection we prove that non - parity - invariant on - shell bethe states have no overlap with the nel state @xmath41 in both cases , even and odd numbers of down spins . in refs . @xcite it is shown that overlaps of the states @xmath80 or @xmath81 with an off - shell bethe state @xmath82 are given by the following determinant expression ( note that we here use a different normalization of bethe states as compared to ref . @xcite ) , [ eq : overlap_s3 ] @xmath83 \left[\prod_{j > k=1}^{m}\frac{{s_{\lambda_j+\lambda_k,\eta}}}{{s_{\lambda_j+\lambda_k,0}}}\right]\det{}_{\!m}(\mathds{1}+u)\epc\\ \label{eq : overlap_s3b } u_{jk } & = \frac{{s_{2\lambda_k,\eta}}{s_{2\lambda_k,0}}}{{s_{\lambda_j+\lambda_k,0}}{s_{\lambda_j-\lambda_k,\eta}}}\left[\prod_{\substack{l=1\\ l\neq k}}^{m}\frac{{s_{\lambda_k+\lambda_l,0}}}{{s_{\lambda_k-\lambda_l,0}}}\right]\left[\prod_{l=1}^m\frac{{s_{\lambda_k-\lambda_l,-\eta}}}{{s_{\lambda_k+\lambda_l,\eta}}}\right]\left(\frac{{s_{\lambda_k,\eta/2}}}{{s_{\lambda_k,-\eta/2}}}\right)^{2m}\epc\end{aligned}\ ] ] where @xmath84 is defined in and @xmath85 is the total momentum of the bethe state . note that there is a difference of a factor @xmath86 compared to the usual expression because here we consider overlaps with nel and anti - nel states instead of its symmetric combination . the parameters @xmath87 , @xmath88 , are arbitrary complex numbers . we show that the determinant of @xmath89 is zero for non - parity - invariant on - shell bethe states , _ i.e. _ the set @xmath25 are bethe roots and there is at least one @xmath90 for which @xmath91 for all @xmath92 . we multiply @xmath89 from the right with the diagonal matrix @xmath93\left(\frac{{s_{\lambda_k,-\eta/2}}}{{s_{\lambda_k,\eta/2}}}\right)^{2m}\right\}_{k=1}^m\epp\ ] ] the resulting matrix @xmath94 reads @xmath95 \left[-\prod_{\substack{l=1}}^{m}\frac{{s_{\lambda_j-\lambda_l,\eta}}}{{s_{\lambda_j-\lambda_l,-\eta}}}\right ] \left(\frac{{s_{\lambda_j,-\eta/2}}}{{s_{\lambda_j,\eta/2}}}\right)^{2m}- \frac{{s_{0,\eta}}{s_{2\lambda_k,0}}}{{s_{\lambda_j+\lambda_k,0}}{s_{\lambda_j-\lambda_k,\eta}}}\prod_{\substack{l=1\\ l\neq k}}^m\frac{{s_{\lambda_k+\lambda_l,0}}}{{s_{\lambda_k-\lambda_l,0}}}\notag\\ & = \delta_{jk } b_j\left[\prod_{\substack{l=1\\ l\neq j}}^{m}\frac{{s_{\lambda_j+\lambda_l,\eta}}}{{s_{\lambda_j-\lambda_l,\eta}}}\right ] - \frac{{s_{0,\eta}}{s_{2\lambda_k,0}}}{{s_{\lambda_j+\lambda_k,0}}{s_{\lambda_j-\lambda_k,\eta}}}\prod_{\substack{l=1\\ l\neq k}}^m\frac{{s_{\lambda_k+\lambda_l,0}}}{{s_{\lambda_k-\lambda_l,0}}}\epc\end{aligned}\ ] ] where we defined the abbreviations @xmath96 \left(\frac{{s_{\lambda,-\eta/2}}}{{s_{\lambda,\eta/2}}}\right)^{2m}\epp\ ] ] the overlap prefactor in @xmath97 reads now @xmath98 \left[\prod_{j > k=1}^{m}\frac{{s_{\lambda_j-\lambda_k,\eta}}{s_{\lambda_j-\lambda_k,-\eta}}}{{s_{\lambda_j+\lambda_k,0}}{s_{\lambda_j+\lambda_k,\eta}}}\right]\epp\ ] ] it causes problems if one of the spectral parameters is at the origin ( or at @xmath36 , or both ) or if the set @xmath25 contains one or more pairs of the form @xmath99 . the poles at @xmath100 , @xmath101 are less important because at the end @xmath87 are bethe roots , and string solutions to the bethe equations for finite @xmath7 show always deviations of perfect strings . exceptions are strings with zero real part that belong to parity invariant states , but those are not considered here . in all cases of non - parity - invariant states we shall show that the product of @xmath102 and @xmath103 vanishes if we impose the on - shell condition ( bethe equations ) . for this purpose we pull the divergent parts of @xmath103 into the determinant and explicitly construct an eigenvector of the resulting matrix that has eigenvalue zero . let us first consider the case with @xmath104 pairs @xmath105 , @xmath106 , and all other bethe roots are unpaired : @xmath107 and @xmath108 for all @xmath109 . the case of one bethe root at zero or @xmath36 is treated in the next subsection , sec . [ sec : npi_odd ] . we start with the determinant expression for off - shell bethe states using matrix @xmath110 defined in eq . . we impose bethe equations later . we redefine @xmath111 and @xmath112 for @xmath113 and consider the limits @xmath114 , @xmath113 , when the subset @xmath115 of @xmath116 s becomes parity invariant . we have @xmath117\left[\prod_{\substack{l=1\\j\neq 2j-1,2j}}^m\frac{{s_{\mu_j+\lambda_l,-\eta}}}{{s_{\mu_j-\lambda_l,-\eta}}}\right ] + \mathcal{o}(\epsilon_j)\notag \\ & = p_{+}^{-1}(\mu_j)p_{-}(\mu_j ) + \mathcal{o}(\{\epsilon_j\}_{j=1}^m)\epc\end{aligned}\ ] ] where @xmath118 for @xmath119 . we write down the zeroth order in all @xmath120 of the matrix elements @xmath121 , where `` order @xmath122 '' or @xmath123 means @xmath124 , @xmath125 - \frac{{s_{0,\eta}}{s_{2\lambda_k,0}}}{{s_{\lambda_j+\lambda_k,0}}{s_{\lambda_j-\lambda_k,\eta}}}\prod_{\substack{l=2m+1\\ l\neq k}}^m\frac{{s_{\lambda_k+\lambda_l,0}}}{{s_{\lambda_k-\lambda_l,0 } } } + \mathcal{o}(\epsilon)\epp\end{aligned}\ ] ] it is @xmath126 and @xmath127 everywhere , except for the last equation where @xmath128 . the @xmath90-th @xmath129 diagonal block is [ @xmath130 , @xmath131 , @xmath132 , all these symbols evaluated at @xmath133 up to order @xmath122 @xmath134 the other elements in the corresponding two columns are of order @xmath122 . we insert @xmath135 between the two matrices on the right , which yields again up to order @xmath122 @xmath136 these manipulations of the first @xmath104 diagonal @xmath129 blocks affect also the first @xmath137 columns , whereas the lower right @xmath138 block of the @xmath139 matrix @xmath110 remains unchanged . the important point of these determinant transformations is that , under the determinant , we can absorb the divergent prefactors @xmath140 , @xmath113 , into the columns @xmath141 , @xmath113 , and then send all @xmath142 s to zero afterwards . we write the corresponding elements in columns @xmath141 as @xmath143 because in these limits they are finite , but unimportant numbers . the corresponding elements in columns @xmath144 , @xmath113 , are all of order @xmath122 and hence go to zero in these limits . the determinant then reads @xmath145 { \scriptscriptstyle}b_{2,2m+1}\end{array } & \begin{array}{l } { \scriptscriptstyle}\ldots \\[-1.2ex]{\scriptscriptstyle}\ldots \end{array } & \begin{array}{l } { \scriptscriptstyle}b_{1,m } \\[-1.2ex ] { \scriptscriptstyle}b_{2,m}\end{array } \\ o & \bar{b}_2 & \ldots & o & \begin{array}{l } { \scriptscriptstyle}b_{3,2m+1 } \\[-1.2ex ] { \scriptscriptstyle}b_{4,2m+1}\end{array } & \begin{array}{l } { \scriptscriptstyle}\ldots \\[-1.2ex]{\scriptscriptstyle}\ldots \end{array } & \begin{array}{l } { \scriptscriptstyle}b_{3,m } \\[-1.2ex ] { \scriptscriptstyle}b_{4,m}\end{array } \\ \vdots & & \ddots & \vdots & \vdots & & \vdots \\ o & o & \ldots & \bar{b}_m & \begin{array}{l } { \scriptscriptstyle}b_{2m-1,2m+1 } \\[-1.2ex ] { \scriptscriptstyle}b_{2m,2m+1}\end{array } & \begin{array}{l } { \scriptscriptstyle}\ldots \\[-1.2ex]{\scriptscriptstyle}\ldots \end{array } & \begin{array}{l } { \scriptscriptstyle}b_{2m-1,m } \\[-1.2ex ] { \scriptscriptstyle}b_{2m , m}\end{array } \\ o & o & \ldots & o & \begin{array}{l } { \scriptscriptstyle}b_{2m+1,2m+1 } \\[-1.2ex ] { \scriptscriptstyle}b_{2m+2,2m+1}\end{array } & \begin{array}{l } { \scriptscriptstyle}\ldots \\[-1.2ex]{\scriptscriptstyle}\ldots \end{array } & \begin{array}{l } { \scriptscriptstyle}b_{2m+1,m } \\[-1.2ex ] { \scriptscriptstyle}b_{2m+2,m}\end{array } \\ \vdots & & & \vdots & \vdots & \ddots & \vdots \\ o & o & \ldots & o & \begin{array}{l } { \scriptscriptstyle}b_{m-1,2m+1 } \\[-1.2ex ] { \scriptscriptstyle}b_{m,2m+1}\end{array } & \begin{array}{l } { \scriptscriptstyle}\ldots \\[-1.2ex]{\scriptscriptstyle}\ldots \end{array } & \begin{array}{l } { \scriptscriptstyle}b_{m-1,m } \\[-1.2ex ] { \scriptscriptstyle}b_{m , m}\end{array } \\ \end{pmatrix}\epc\end{aligned}\ ] ] where the symbol @xmath146 stands for @xmath147 and the diagonal blocks are given by @xmath148 this determinant relation is valid in the off - shell sector . the parameters @xmath149 , @xmath113 , can be chosen arbitrarily . note that one has to evaluate the functions @xmath150 , @xmath151 , @xmath152 , and @xmath153 at the corresponding @xmath154 . a determinant is the product of all its eigenvalues . we multiply matrix @xmath155 with one of its eigenvectors @xmath156 . the main idea of the steps from eq . to eq . is to choose the parameters @xmath149 such that @xmath157^{1/2 } + v_{2j}(1-\alpha_j)[b(\mu_j)p_+(\mu_j)p_0(\mu_j)]^{-1/2 } = 0 $ ] . they exist if and only if @xmath158 furthermore , we impose bethe equations , _ i.e. _ @xmath159 for @xmath113 and @xmath160 for @xmath161 . note that the symbols @xmath162 are defined in eq . and correspond to the bethe equations of the non - parity - invariant parameters @xmath87 , @xmath161 whereas @xmath163 are the bethe equations for the parameters @xmath154 , corresponding to the parity - invariant subset of bethe roots @xmath164 . if the conditions @xmath165 are fulfilled for all @xmath113 , we can now show that @xmath166 is an eigenvector with eigenvalue zero . multiplication of @xmath155 with @xmath167 and using @xmath168^{1/2 } + ( 1-\alpha_j)[p_+(\mu_j)p_0(\mu_j)]^{-1/2}= 0 $ ] as well as @xmath169 [ see eq . ] yields @xmath170 and @xmath171 . the functions @xmath172 and @xmath173 are given by @xmath174 we show that they are identically zero for any set of complex parameters @xmath175 . they are holomorphic in @xmath176 because all residues vanish , @xmath177 for all @xmath178 . they are bounded and hence , with liouville s theorem , we conclude that they are constant . furthermore , @xmath179 and therefore @xmath180 and @xmath181 for all @xmath182 . the components @xmath183 , @xmath161 , can be written as @xmath184 with @xmath185 the same reasoning as for @xmath186 applies for @xmath187 . we also have vanishing residues , @xmath188 , and therefore @xmath189 for all @xmath190 . in summary , we obtain @xmath191 . therefore , @xmath192 if the bethe roots fulfill the conditions @xmath193 for all paired bethe roots @xmath194 , @xmath113 , and unpaired roots @xmath195 , @xmath196 . if there are bethe roots @xmath197 that do not fulfill condition , we have @xmath198 . we define the index sets @xmath199 and observe that @xmath200 , @xmath201 , and if @xmath202 then @xmath203 and vice versa . we choose the components of the vector @xmath167 in the following way , @xmath204 the eigenvalue equations for the components @xmath205 are as before . for @xmath206 they read @xmath207 and @xmath208 . the functions @xmath209 and @xmath210 are given by @xmath211 according to @xmath212 for all @xmath213 , we can write them as @xmath214 they have zeroes at @xmath215 . in summary , we again found an eigenvector @xmath216 of the matrix @xmath155 with eigenvalue zero , @xmath217 the condition @xmath216 is equivalent to @xmath218 which is always the case for @xmath219 . in this case , we therefore have @xmath220 if the set @xmath25 satisfies bethe equations . for @xmath221 we have @xmath222 pairs @xmath223 and @xmath224 is always trivially fulfilled . then @xmath225 , @xmath226 , and @xmath227 . thus , one can not show that @xmath228 this way . in fact , for the case of parity - invariant bethe states @xmath44 , it was previously shown @xcite that @xmath229 the explicit formulas are given in eqs . and of sec . [ sec : pi_even ] . parity invariance constrains bethe roots to occur in pairs or to be at zero and/or at @xmath36 . let us consider the case when one bethe root is at zero . if it is at @xmath36 the logic of the proof is exactly the same and all calculations are similar and straightforward . we denote the set of off - shell parameters by @xmath230 and set @xmath231 . we consider again @xmath104 pairs @xmath232 , @xmath233 , @xmath234 , @xmath106 , with a complementary set of nonzero unpaired bethe roots @xmath91 for all @xmath235 as in the previous section . the determinant then reads @xmath236 \ast \end{array } & \bar{b}_1 & o & \ldots & o & \begin{array}{l } { \scriptscriptstyle}b_{1,2m+1 } \\[-1.2ex ] { \scriptscriptstyle}b_{2,2m+1}\end{array } & \begin{array}{l } { \scriptscriptstyle}\ldots \\[-1.2ex]{\scriptscriptstyle}\ldots \end{array } & \begin{array}{l } { \scriptscriptstyle}b_{1,m-1 } \\[-1.2ex ] { \scriptscriptstyle}b_{2,m-1}\end{array } \\ \begin{array}{c } \ast \\[-1.6ex ] \ast \end{array } & o & \bar{b}_2 & \ldots & o & \begin{array}{l } { \scriptscriptstyle}b_{3,2m+1 } \\[-1.2ex ] { \scriptscriptstyle}b_{4,2m+1}\end{array } & \begin{array}{l } { \scriptscriptstyle}\ldots \\[-1.2ex]{\scriptscriptstyle}\ldots \end{array } & \begin{array}{l } { \scriptscriptstyle}b_{3,m-1 } \\[-1.2ex ] { \scriptscriptstyle}b_{4,m-1}\end{array } \\ \vdots & \vdots & & \ddots & \vdots & \vdots & & \vdots \\ \begin{array}{c } \ast \\[-1.6ex ] \ast \end{array } & o & o & \ldots & \bar{b}_m & \begin{array}{l } { \scriptscriptstyle}b_{2m-1,2m+1 } \\[-1.2ex ] { \scriptscriptstyle}b_{2m,2m+1}\end{array } & \begin{array}{l } { \scriptscriptstyle}\ldots \\[-1.2ex]{\scriptscriptstyle}\ldots \end{array } & \begin{array}{l } { \scriptscriptstyle}b_{2m-1,m-1 } \\[-1.2ex ] { \scriptscriptstyle}b_{2m , m-1}\end{array } \\ \begin{array}{c } \ast \\[-1.6ex ] \ast \end{array } & o & o & \ldots & o & \begin{array}{l } { \scriptscriptstyle}b_{2m+1,2m+1 } \\[-1.2ex ] { \scriptscriptstyle}b_{2m+2,2m+1}\end{array } & \begin{array}{l } { \scriptscriptstyle}\ldots \\[-1.2ex]{\scriptscriptstyle}\ldots \end{array } & \begin{array}{l } { \scriptscriptstyle}b_{2m+1,m-1 } \\[-1.2ex ] { \scriptscriptstyle}b_{2m+2,m-1}\end{array } \\ \vdots & \vdots & & & \vdots & \vdots & \ddots & \vdots \\ \begin{array}{c } \ast \\[-1.6ex ] \ast \end{array } & o & o & \ldots & o & \begin{array}{l } { \scriptscriptstyle}b_{m-2,2m+1 } \\[-1.2ex ] { \scriptscriptstyle}b_{m-1,2m+1}\end{array } & \begin{array}{l } { \scriptscriptstyle}\ldots \\[-1.2ex]{\scriptscriptstyle}\ldots \end{array } & \begin{array}{l } { \scriptscriptstyle}b_{m-2,m-1 } \\[-1.2ex ] { \scriptscriptstyle}b_{m-1,m-1}\end{array } \\ \end{pmatrix}\epc\end{aligned}\ ] ] where @xmath237 stands for @xmath238 , and all other blocks are the same as in eq . . now we impose bethe equations , _ @xmath159 for @xmath239 and @xmath160 for @xmath240 , where we set @xmath241 and identify index @xmath42 with index @xmath76 . then , considering the vector @xmath242 , we obtain @xmath243 in the last step we used @xmath244 from eq . with @xmath42 replaced by @xmath245 . under the condition that @xmath246 we can further show that @xmath247 for all other components , analogously to the case of even number of particles in subsection [ sec : npi_even ] . hence , it is shown that both @xmath248 and @xmath249 have zero overlap with all non - parity - invariant on - shell bethe states . therefore , the overlap with the zero - momentum nel state @xmath250 is zero , too . before we consider the scaling limit of the overlap @xmath251 to lieb - liniger @xcite ( @xmath252 , @xmath253 , @xmath254 for all finite bethe roots , and eventually @xmath255 ) , we have to ensure that there are only finitely many downturned spins in the initial state @xmath256 because then the overlap is nonzero only for bethe states with finite @xmath42 and the scaling limit to ll is applicable . the norm of a parity - invariant on - shell bethe state with an odd number @xmath257 of downturned spins is given by [ see eq . ] @xmath258 \\ \times\left[\prod_{j=1}^m \frac{{s_{\lambda_j,\eta } } { s_{\lambda_j,-\eta } } } { { s_{\lambda_j,0}}^2}\right ] \prod_{\substack{j > k=1\\ \sigma = \pm}}^m\frac{{s_{\lambda_j - \sigma \lambda_k,\eta } } { s_{\lambda_j + \sigma \lambda_k,-\eta } } } { { s_{\lambda_j + \sigma \lambda_k,0}}^2 } \sqrt{\det{}_{\!2m+1 } ( g_{jk } ) } \epp\end{gathered}\ ] ] to derive a determinant expression for @xmath259 we follow exactly the logic of ref . @xcite . as initial states we consider @xmath260-fold @xmath0-raised nel states @xmath261 \sigma_n^+ \left[\prod_{j = n+1}^n q^{-\sigma_j^z/2}\right]\epc\quad \tilde{s}_q^+ = \sum_{n=1}^n \left[\prod_{j=1}^{n-1 } q^{-\sigma_j^z/2}\right ] \sigma_n^+ \left[\prod_{j = n+1}^n q^{\sigma_j^z/2}\right]\notag\end{aligned}\ ] ] and @xmath0 is related to the anisotropy via @xmath262 . in the ll scaling limit , these @xmath0-deformed operators act on the nel states as global @xmath263 operators and the state scales to the ( unnormalized ) bec - like state @xcite of the lieb - liniger bose gas . if we set @xmath264 this leads to an odd number of downturned spins in the resulting initial @xmath265 state , which directly corresponds to an odd number of bosons in the bec state . we make use of overlap formula for an odd number of downturned spins with @xmath266 . we consider the overlap of the @xmath260-fold @xmath0-raised nel state with a parity - invariant ( off - shell ) bethe state in the magnetization sector @xmath267 . the global symmetry operators that are needed to construct @xmath0-raised nel states are obtained by sending spectral parameters to infinity and taking the proper normalization into account @xcite . as in the even particle case @xcite we split the prefactor @xmath49 of the determinant formula into two parts , @xmath268 we again have @xmath269 . the other part of @xmath49 is determined by @xmath270\\ \times\frac{1}{\sqrt{\sinh^{2m+1}(\eta)\det{}_{\!2m+1}(g_{jk } ) } } \epp\end{gathered}\ ] ] within the limit of some spectral parameters to infinity the determinant can be simplified . following the reasoning of ref . @xcite the entire upper right @xmath271 block is zero . furthermore , the lower right @xmath272 block becomes a triangular matrix and together with the factor @xmath273 the determinant of this block becomes again @xmath274_q!$ ] ( as in the even particle case ) . then we plug in the on - shell condition and all @xmath275-terms in the upper left @xmath276 block vanish . in summary , [ eq : new_overlap_odd ] @xmath277_q!}{\| |n^{(2n)}\rangle \| } \frac{\hat{\gamma}\det{}_{\!m+1}(\hat{g}_{jk}^{+})}{\|\{\pm\lambda_j\}_{j=1}^m \cup \{0\ } \| } \notag \\ & = \frac{[2n]_q!}{\| |n^{(2n)}\rangle\| } \frac{\sinh(\frac{\eta}{2})}{\sinh(\eta)}\left[\prod_{j=1}^{m}\frac{\sqrt{\tanh(\lambda_j+\frac{\eta}{2 } ) \tanh(\lambda_j-\frac{\eta}{2})}}{2\sinh(2\lambda_j)}\right ] \notag\\ & \quad\times\frac{\det{}_{\!m+1}(\hat{g}_{jk}^{+})}{\sqrt{\det{}_{\!2m+1}(\hat{g}_{jk})}}\epc\end{aligned}\ ] ] where @xmath278 note again that @xmath279 is defined differently for @xmath280 and for @xmath281 ( by a factor @xmath1 ) . the function @xmath282 is defined as usual , @xmath283 . @xmath284 is the norm of the @xmath260-fold @xmath0-raised nel state which , in the limit @xmath285 , can be calculated explicitly @xcite , @xmath286 we shall need this result to evaluate the prefactor of the determinant formula in the scaling limit to ll exactly . we repeat the steps of ref . @xcite for the scaling limit to lieb - liniger . there is a one - to - one correspondence @xcite between xxz and lieb - liniger bethe states as well as between their norm formulas . the matrices @xmath287 in eq . for finite @xmath288 turn into the corresponding lieb - liniger matrices @xcite . since @xmath289 and due to the corrective factor @xmath290 ( for an explanation of this factor see ref . @xcite ) , we obtain for the overlaps with the bec state , @xmath291 ( with @xmath292 , @xmath104 being the number of ll rapidities with positive real part ) @xmath293 with @xmath294 k_{ll}(j , k ) & = \left\ { \begin{array}{r@{\hspace{4ex}}l } \left(k_{ll}(\lambda_j-\lambda_k ) + k_{ll}(\lambda_j+\lambda_k)\right)\epc & 1\leq j , k \leq m \epc \\ \frac{1}{2 } \left(k_{ll}(\lambda_j-\lambda_k ) + k_{ll}(\lambda_j+\lambda_k)\right ) \epc & \text{otherwise}\epc \end{array } \right.\\ k_{ll}(\lambda ) & = \frac{2c^2}{\lambda^2+c^2}\epp\end{aligned}\ ] ] we cross - checked this result analytically for @xmath295 and numerically up to @xmath296 . this completes the result of ref . @xcite where the formula for overlaps of parity - invariant ll bethe states with bec states with even numbers of bosons @xcite was proven . the results of sec . [ sec : npi ] for non - parity - invariant xxz bethe states are valid in the scaling limit to ll which proves that overlaps of the bec state with non - parity - invariant ll bethe states are zero . in this paper we presented a proof that the overlap of the nel state with a non - parity - invariant on - shell bethe state vanishes . we further gave a formula for the overlap with parity - invariant bethe states with an odd number of downturned spins . these results of subsections [ sec : pi_odd ] , [ sec : npi_even ] , and [ sec : npi_odd ] complete the answer to the question : what is the overlap of the nel state with xxz bethe states ? it is now proven that it is zero for any non - parity - invariant on - shell state , and it is nonzero for parity - invariant ( off - shell ) states . the latter can be expressed by `` gaudin - like '' determinants for both cases , even and odd number of down spins . this therefore opens the door to a quench action analysis of the quench starting from the nel state . our results thus pave the way towards a comparison with gge predictions of this quench , as obtained in @xcite , which would be particularly interesting . the scaling to the lieb - linger model and the derivation of the corresponding determinant expressions for even and odd numbers of bosons is straightforward ( for the former see ref . @xcite , for the latter see sec . [ sec : ll_odd ] ) . to summarize , the overlaps of the bec state with bethe states of the lieb - liniger bose gas ( repulsive and attractive ) are now known for all possible cases . they vanish for non - parity - invariant states and are , in general , nonzero for parity - invariant ones . the latter are given by determinants of gaudin type , which have slightly different structures for even and an odd numbers of bosons . these results can be used e.g. in the context of 1d kardar - parisi - zhang equation as in @xcite , which can be mapped to the attractive lieb - liniger bose gas . we would like to thank frank ghmann and pasquale calabrese for useful discussions . we acknowledge financial support from the foundation for fundamental research on matter ( fom ) and the netherlands organisation for scientific research ( nwo ) . we start from equation for the overlap of an off - shell bethe state @xmath297 again with @xmath298 , but now @xmath42 odd . we set @xmath299 and further @xmath300 for @xmath301 , @xmath302 for @xmath303 , and finally @xmath304 . we first consider @xmath266 . the case @xmath305 is discussed at the end of this section . by multiplying the prefactor and the inverse of the determinant with @xmath306 , where @xmath307 , we get regular expressions in the limits @xmath114 , @xmath308 \left[\prod_{j=1}^{m ' } \frac{{s_{\lambda_j,\eta}}{s_{\lambda_j,-\eta}}}{{s_{\lambda_j,0}}^2}\right ] \left[\prod_{\substack{j > k=1\\ \ \sigma=\pm}}^{m'}\frac{{s_{\lambda_j+\sigma\lambda_k,\eta}}{s_{\lambda_j+\sigma\lambda_k,-\eta}}}{{s_{\lambda_j+\sigma\lambda_k,0}}^2}\right]\epc \\ & \det{}_\text{reg } = \lim_{\delta_0 \to 0}\lim\nolimits_{\{\delta_j \to 0\}_{j=1}^{m ' } } \left\{\left[\prod_{j=0}^{m ' } \delta_j \right]^{-1 } \det{}_{\!m}(\delta_{jk}+u_{jk})\right\}\epp \label{eq : dreg}\end{aligned}\ ] ] we order the rows and columns of the matrix in such a way that the first one corresponds to @xmath309 , and that pairs of rapidities @xmath99 always correspond to neighboured rows and columns . we label them by indices @xmath310 . column zero of the matrix is up to first order in @xmath311 given by @xmath312 the second step is only allowed when @xmath313 . for @xmath314 it means that there would be an additional minus sign . now we insert the parity invariance condition for all other @xmath315 rapidities . we can do exactly the same procedure as for the even case @xcite . we redefine the factors @xmath316 as @xmath317\left(\frac{{s_{\lambda_k,\eta/2}}}{{s_{\lambda_k,-\eta/2}}}\right)^{2m}\epc \quad k = 1,\ldots , m'\epc \\ \mathfrak{a}_0 & = -1\epp\end{aligned}\ ] ] the first row of the matrix , row zero , is then given by @xmath318 these steps are allowed for both @xmath266 and @xmath314 . we further define @xmath319 as in ref . @xcite , @xmath320 and we multiply the @xmath139 matrix @xmath89 from the left and from the right respectively with the diagonal matrices @xmath321 the structure of the matrix becomes @xmath322 & \delta_2 \alpha_0^{-1}\left[\!\begin{array}{cc } \tilde{u}_{0,3 } \alpha_2^{-1 } & \tilde{u}_{m,4 } \alpha_2 \end{array}\!\right ] & \dots \\[4ex ] \delta_0 \alpha_0\left[\!\!\begin{array}{c } \alpha_1 \tilde{u}_{1,0 } \\ \alpha_1^{-1 } \tilde{u}_{2,0 } \end{array}\!\!\right ] & \left[\!\!\!\begin{array}{ll } 1-\delta_1\frac{{s_{2\lambda_1,-\eta}}}{{s_{2\lambda_1,0}}}\alpha_1 ^ 2 & 1+\delta_1\mathfrak{b}_1^{-}\alpha_1 ^ 2\\ 1+\delta_1\mathfrak{b}_1^{+}\alpha_1^{-2 } & 1-\delta_1\frac{{s_{2\lambda_1,\eta}}}{{s_{2\lambda_1,0}}}\alpha_1^{-2 } \end{array}\!\!\right ] & \delta_2\left[\!\!\begin{array}{cc}a_{12}&b_{12}\\ c_{12}&d_{12}\end{array}\!\!\right ] & \dots \\[6ex ] \delta_0 \alpha_0 \left[\!\!\begin{array}{c } \alpha_2 \tilde{u}_{3,0 } \\ \alpha_2^{-1 } \tilde{u}_{4,0 } \end{array}\!\!\right ] & \delta_1\left[\!\!\begin{array}{cc}a_{21 } & b_{21}\\ c_{21 } & d_{21}\end{array}\!\!\right ] & \left[\!\!\!\begin{array}{ll } 1-\delta_2\frac{{s_{2\lambda_2,-\eta}}}{{s_{2\lambda_2,0}}}\alpha_2 ^ 2 & 1+\delta_2\mathfrak{b}_2^{-}\alpha_2 ^ 2\\ 1+\delta_2\mathfrak{b}_2^{+}\alpha_2^{-2 } & 1-\delta_2\frac{{s_{2\lambda_2,\eta}}}{{s_{2\lambda_2,0}}}\alpha_2^{-2 } \end{array}\!\!\right ] & \dots \\ \vdots & \vdots & \vdots & \ddots \end{array}\right)\epc \ ] ] where the elements @xmath323 , @xmath324 , @xmath325 , and @xmath326 , @xmath327 are the same as in the even case . analogously , the determinant can be simplified by replacing column @xmath328 by the difference of columns @xmath328 and @xmath329 for all @xmath330 and afterwards by replacing row @xmath141 by the difference of rows @xmath141 and @xmath144 for all @xmath331 as in ref . @xcite which leads up to first order to @xmath332 & \left[\!\!\begin{array}{cc } \delta_2 e_{02 } & 0 \end{array}\!\!\right ] & \dots \\[1ex ] \left[\!\!\begin{array}{c } \delta_0 e_{10 } \\ 0 \end{array}\!\!\right ] & \left[\!\!\begin{array}{c@{\hspace{0.9ex}}c } \delta_1 d_{1 } & 0 \\ 0 & 1 \end{array}\!\!\right ] & \left[\!\!\begin{array}{c@{\hspace{0.9ex}}c } \delta_2 e_{12 } & 0 \\ 0 & 0 \end{array}\!\!\right ] & \dots\\[3ex ] \left[\!\!\begin{array}{c } \delta_0 e_{20 } \\ 0 \end{array}\!\!\right ] & \left[\!\!\begin{array}{c@{\hspace{0.9ex}}c } \delta_1 e_{21 } & 0 \\ 0 & 0 \end{array}\!\!\right ] & \left[\!\!\begin{array}{c@{\hspace{0.9ex}}c } \delta_2 d_{2 } & 0 \\ 0 & 1 \end{array}\!\!\right ] & \\ \vdots & \vdots & & \ddots \end{array}\right ) \\[2ex ] = \left[\prod_{k=0}^{m'}\delta_k\right]\det{}_{\!m'+1 } \hspace{-0.5ex}\left[\begin{array}{cccc } \tilde{u}_{00 } & e_{01 } & e_{02 } & \dots\\ e_{10 } & d_{1 } & e_{12 } & \dots \\ e_{2 m } & e_{21 } & d_{2 } & \\[-0.7ex ] \vdots & \vdots & & \ddots \end{array}\right]\epp\end{gathered}\ ] ] the new matrix elements are ( @xmath313 and @xmath305 ) @xmath333 where we used @xmath334 . if @xmath305 the formula for the overlap is the same and its derivation straightforward . the only difference is a minus sign in eq . , which cancels at the end when we express the corresponding matrix elements by the function @xmath335 . this eventually leads to eqs . .

thermoelectric effects are studied in an aharonov - bohm ( ab ) interferometer with an embedded quantum dot in the kondo regime . the ab flux - dependent transmission probability has an asymmetrical shape arising from the fano interference between the direct tunneling path and the kondo - resonant tunneling path through a quantum dot . the sign and magnitude of thermopower can be modulated by the ab flux and the direct tunneling amplitude . in addition , the thermopower is anomalously enhanced by the kondo correlation in the quantum dot near the kondo temperature ( ) . the kondo correlation in the quantum dot also leads to crossover behavior in diagonal transport coefficients as a function of temperature . the amplitude of an ab oscillation in electric and thermal conductances is small at temperatures far above , but becomes enhanced as the system is cooled below . the ab oscillation is strong in the thermopower and the lorenz number within the crossover region near the kondo temperature .

cond-mat0211554

the phase - coherence of an electron s wave function in mesoscopic systems enables one to observe a wide variety of interesting phenomena in quantum physics . one of them is the phase shift experienced by electrons due to scattering centers . although the phase can not be measured directly in bulk systems , mesoscopic systems provide an opportunity to attain such information . recently the transmission phase shift experienced by electrons passing through a quantum dot was measured using the aharonov - bohm ( ab ) interferometers @xcite . in addition to the direct observation of the phase , phase - coherence leads to such phenomena as the magnetic and electrostatic ab effects , fano interference , persistent currents , universal conductance fluctuations , etc . quantum dots provide a unique opportunity to study nonequilibrium many - body effects due to the strong coulomb interaction and quantization of energy levels , spin , and charge . a fano resonance was observed@xcite in the conductance in transport through a quantum dot which is connected to two leads . the opening of direct tunneling between two leads was believed to be the main reason for the observation of the fano interference pattern . the interference between two current paths , resonant and direct tunneling , leads to the asymmetrical fano resonance in the differential conductance . however , the nature of the direct tunneling path is not clear and the direct tunneling can not be controlled in this work@xcite . inserting a quantum dot in an ab interferometer , the fano shape in the @xmath1 curves was also observed@xcite . free control of the direct tunneling probability is possible using a gate voltage . in addition , the asymmetrical shape of a fano resonance has been observed in other nanoscopic systems . when magnetic adatoms are deposited on a metallic surface , the local electronic density of states ( dos ) on a metallic surface is modified due to the kondo effect occurring at the site of a magnetic adatom . using a scanning tunneling microscope ( stm)@xcite , the differential conductance between the stm tip and the metallic surface was observed to be of asymmetrical fano resonance type close to the magnetic adatom at low temperature . confining electrons to an ellipse@xcite in a quantum corral , the phase coherence of an electron s wave functions was detected by imaging the metallic surface with stm spectroscopy . when a magnetic adatom was positioned at one of the two foci of the ellipse , the stm experiments showed the coherently reproduced image of the magnetic atom at the other empty focus point . a fano interference pattern was also observed in the carbon nanotubes@xcite when co ions were deposited on the surface of carbon nanotubes . the ability to observe the fano resonance shape in all these experiments depends on the phase coherence of an electron s wave functions propagating along a metallic surface . in this paper , we consider a phase - coherent aharonov - bohm interferometer with an embedded quantum dot ( see fig . [ fanodot ] ) and study theoretically the thermoelectric effects of this system @xcite when the embedded quantum dot lies in the kondo regime . the quantum dot in the kondo regime@xcite will be called a kondo dot . though the issue of the transmission phase in this system is a very interesting subject , we focus on the effects of phase - coherence on the transport coefficients . the phase - coherence enables us to study the effects of the aharonov - bohm oscillations as well as the fano interference on the thermoelectric properties of the model system . we use the nonequilibrium keldysh green s function method and the noncrossing approximation for our model study . there are two possible paths for the flow of electrons in our ab interferometer . the direct tunneling amplitude between two leads is a temperature - independent constant , while the tunneling through the kondo dot is strongly temperature dependent due to the kondo effect . different physics is expected above and below the kondo temperature ( @xmath0 ) , due to the crossover behavior in the kondo dot as a function of temperature . since the flow of electrons through a quantum dot is blocked by the strong repulsive coulomb interaction ( coulomb blockade ) at high temperatures above @xmath0 , practically no fano interference arises and the ab oscillations are very _ weak _ in transport coefficients . with lowering temperature below @xmath0 , the kondo resonance peak in a quantum dot develops close to the fermi level . this opening of an additional resonant current path leads to the fano interference with the direct tunneling path . fano interference and ab flux modify the shape of the transmission probability through the system near the fermi level . out of all the transport coefficients , thermopower is one of the most sensitive to the shape or the particle - hole asymmetry of the transmission probability and is an appropriate experimental probe to investigate the fano interference and the ab effect . we find that the sign and magnitude of the thermopower can be modulated by controlling the _ aharonov - bohm phase _ with magnetic fields and the _ tunneling matrices _ with varying gate voltages . in addition , we find that the kondo correlation enhances the amplitude of ab oscillations in diagonal transport coefficients at low @xmath2 compared to @xmath0 . since two diagonal transport coefficients , the electric and thermal conductances , are influenced by fano interference and ab flux in the same way , their ratio , the lorenz number , is insensitive to ab flux near the zero temperature and is fixed at the sommerfeld value . the amplitude of ab oscillations in the thermopower and lorenz number is strong near the kondo temperature . a short paper which presents some of these results appeared elsewhere @xcite . 0.5 cm this paper is organized as follows . in sec . [ form_sect ] , the model hamiltonian is introduced and the formulation of electric and heat currents is summarized leaving the details to appendix [ gdosband ] and [ wideband ] . in sec . [ nca_sect ] , the modified noncrossing approximation ( nca ) is briefly introduced for our model system . numerical results from solving the nca equations self - consistently are presented in sec . [ result_sect ] and a conclusion is included in sec . [ sum_sect ] . in this section , we formulate the electric and heat currents in the aharonov - bohm interferometer which contains a quantum dot in one arm . the model system is shown in fig . [ fanodot ] . using the keldysh s nonequilibrium green s function method @xcite , the currents are derived in a landauer - bttiker form . the phase - coherence of the model system enables us to study the kondo effects of a quantum dot , the fano interference between the direct path and the kondo - resonant path , and the aharonov - bohm effects by applying magnetic fields . when the quantum dot lies in the kondo regime or when there is an odd number of electrons within a dot and the spin of a highest - lying electron is unpaired , the quantum dot can be effectively considered as a magnetic impurity with spin @xmath3 . the highest - lying electron can then be described by the anderson impurity model treating other electrons as a background . the model hamiltonian may be written as @xmath4 where @xmath5 \nonumber\\ & & + \frac{1}{\sqrt{v } } \sum_{p = l , r } \sum_{\vec{k}\alpha } \left [ v_{pd } c_{p\vec{k}\alpha}^{\dag } d_{\alpha}^{\phantom { * } } + h.c . \end{aligned}\ ] ] @xmath6 describes the isolated system of a dot and leads and the tunneling between a dot and two leads is described by the tunneling hamiltonian @xmath7 . here @xmath8 is the electron annihilation operator of spin direction @xmath9 in the @xmath10 ( left ) @xmath11 ( right ) lead . @xmath12 is the annihilation operator of the highest - lying electron with unpaired spin in a quantum dot . @xmath13 is the direct tunneling matrix from the right lead to the left one and @xmath14 is the hopping amplitude from the left or right electrode to a quantum dot(@xmath15 ) . in our model study , the wave vector dependence of these tunneling matrices is neglected . since electrons can flow from left to right in two different paths ( direct tunneling and resonant tunneling via a quantum dot ) , quantum interference is expected when the phase coherence of electrons is retained . applying a magnetic field to the system , we can study the aharonov - bohm(ab ) effect on the transport . the ab phase @xmath16 is incorporated into the tunneling matrices in a manner that @xmath17 . @xmath18 is the magnetic flux passing through the system as shown in fig . [ fanodot ] and @xmath19 is the flux quantum . applied magnetic fields are assumed to be not strong enough to split the doubly degenerate spin states of a quantum dot . [ cols="^,^,^,^ " , ] the electric current operator can be defined as a change in the number of electrons per unit time in the left electrode . @xmath20 \nonumber\\ & = & \frac{e}{i\hbar } \frac{1}{v } \sum_{\vec{k}\vec{k}'\alpha } \left [ t_{lr } c_{l\vec{k}\alpha}^{\dag } c_{r\vec{k}'\alpha } - t_{rl } c_{r\vec{k}'\alpha}^{\dag } c_{l\vec{k}\alpha } \right ] \nonumber\\ & & + \frac{e}{i\hbar } \frac{1}{\sqrt{v } } \sum_{\vec{k}\alpha } \left [ v_{ld } c_{l\vec{k}\alpha}^{\dag } d_{\alpha } - v_{dl } d_{\alpha}^{\dag } c_{l\vec{k}\alpha } \right].\end{aligned}\ ] ] here @xmath21 is the number operator in the left lead . the heat current operator is also defined as a change in the thermal energy per unit time in the left lead . @xmath22.\end{aligned}\ ] ] @xmath23 is the hamiltonian of the left lead without the chemical potential shift due to the source - drain voltage . using the current conservation in a steady state , both currents can be written in a landauer - bttiker form in terms of the green s function of a dot . @xmath24 . \end{aligned}\ ] ] here @xmath25 is the fermi - dirac thermal distribution function in the lead @xmath26 when each lead is in thermal equilibrium at temperature @xmath27 . @xmath28 is the chemical potential shift due to the applied source - drain bias voltage . @xmath29 is the transmission probability spectral function through an ab ring and is related to the green s function @xmath30 of a dot by the equation , @xmath31.\end{aligned}\ ] ] the first term @xmath32 comes from the direct tunneling and is given by the expression . @xmath33 here @xmath34 is the dimensionless measure of direct tunneling of electrons between two leads . @xmath35 is the density of states ( dos ) at the fermi level of the left , right leads , respectively . the overlined dos is normalized such that its value is unity at the fermi energy . @xmath36 and the definitions of reduced green s functions @xmath37 of each lead ( @xmath26 ) can be found in appendix [ gdosband ] . the resonant tunneling through the quantum dot and the interference effect are all included in the second term of @xmath29 . @xmath38 in the absence of the direct tunneling , @xmath39 is reduced to the familiar expression , @xmath40 . the parameters of @xmath41 and @xmath42 are tabulated in the table [ table1 ] . detailed derivations and definitions of new notations can be found in appendix [ gdosband ] . the line - broadening parameter @xmath43 ( @xmath26 ) measures the hopping rate of electrons between the quantum dot and the leads . the complex number @xmath44 contains the effect of applied magnetic fields through the aharonov - bohm phase @xmath16 . this combination of tunneling matrices @xmath45 has the meaning that electrons hop off the quantum dot into the right lead , tunnel from the right lead to the left one via the direct tunneling , and hop back on the quantum dot . one complete circulation of an electron s motion along the aharonov - bohm ring picks up the ab phase @xmath46 generated by the enclosed magnetic flux , @xmath18 . note that @xmath47 and @xmath48 for @xmath26 , but @xmath49 . these relations under the inversion of the magnetic flux leads to the ( broken ) ab phase symmetry in ( out of ) equilibrium , respectively . in the wide conduction band limit , all the relevant green s functions of two electrodes become energy - independent except for the thermal functions . the transmission probability spectral function reduces to ( see appendix [ wideband ] for details ) @xmath50 the first term is the transmission amplitude due to the direct tunneling . the second term is the expected form for transport through a quantum dot with the renormalized hopping rates @xmath51 and @xmath52 . the third term is the interference between two current paths . we can further simplify the transmission probability spectral function @xmath29 as here @xmath54 is equal to @xmath55 . note that @xmath56 is the transmission probability through the direct tunneling , and @xmath57 is the maximum dimensionless linear conductance through a quantum dot in the absence of the direct tunneling . this dimensionless conductance also provides measures of the asymmetry in the coupling of a quantum dot to the left and right electrodes . in the linear response regime , we can expand the electric and heat currents up to the linear terms of @xmath58 and @xmath59 . the transport coefficients @xmath60 are defined by the relations , @xcite @xmath61 and can be expressed in terms of the transport integral , @xmath62 , \end{aligned}\ ] ] as @xmath63 , @xmath64 and @xmath65 . the linear response conductance , @xmath66 , is given by the equation , @xmath67 which measures particle - hole symmetrical part of the transmission probability @xmath29 with respect to the fermi level . the thermopower of a quantum dot in a two - terminal configuration can be found in an open circuit(@xmath68 ) by measuring the induced voltage drop across a quantum dot when the temperature difference between two leads is applied . the thermopower is defined by the relation @xmath69 and can be expressed as @xmath70 here the constant @xmath71 is approximately @xmath72 @xmath73v / k . in most metals , the thermopower is of the order of a few @xmath73v / k . since the integral @xmath74 measures the first moment of energy @xmath75 in the transmission probability , the thermopower probes the particle - hole asymmetric part in @xmath29 with respect to the fermi energy . the thermal conductance @xmath76 through an ab ring can be expressed in terms of these transport integrals as @xmath77.\end{aligned}\ ] ] the thermal conductance probes the particle - hole symmetric part of the transmission probability . to study the thermoelectric effects of our model system , we have to find the green s function , @xmath78 , of a quantum dot . due to the strong repulsive coulomb interaction in a dot , the calculation of @xmath78 is nontrivial and needs a many - body technique . for this purpose , we adopt the noncrossing approximation ( nca ) , which is a self - consistent diagrammatic method ( for details , see refs . @xcite ) . in our model system , the direct tunneling path renormalizes the anderson hybridization functions and leads to a flux - dependent effective continuum band for the anderson impurity ( quantum dot ) , and the nca integral equations are accordingly modified as shown below . the nca has been very successful in studying the anderson impurity models in the kondo regime both in normal metals @xcite and out of equilibrium@xcite . the zero - temperature analysis of the nca self - energies reveals that the nca gives rise to the nonanalytic behavior of the green s functions near the fermi energy . however this nonanalytic behavior shows up in the finite temperature nca below the _ pathological temperature _ @xmath27 . the value of @xmath27 can be estimated analytically@xcite and @xmath79 , where @xmath0 is the kondo temperature , so that the interesting kondo effects in physical quantities can be computed down to temperatures far below @xmath0 but still larger than @xmath27 . due to the direct tunneling between the two electrodes , the effective continuum band for the anderson impurity ( quantum dot ) is modified from two simple leads . in general , the self - energy for an electron on an anderson impurity embedded in an effective electron continuum band can be decomposed into two terms , the one - body and many - body contributions , @xmath80 here @xmath81 is the self - energy due to the on - site coulomb interaction , while the first term @xmath82 comes from hopping into the continuum band . in our model system , the one - body contribution @xmath83 is given by the eq.([oneself ] ) and can be written as @xmath84 the two auxiliary green s functions , @xmath85 and @xmath86 , are introduced in appendix [ gdosband ] . this one - body self - energy shows up as the continuum band propagator in the feynman diagrams of the pseudoparticle green s functions , which will be defined below . 0.3 cm in this paper , we take the limit of @xmath87 and remove the double occupancy in the highest - lying electron s state . since we are interested in the aharonov - bohm effect and the kondo effect in the thermoelectric transport coefficients , the essential physics can be obtained using the infinite @xmath88 nca approach . in general the vertex correction in the self - energy is needed to get the correct kondo energy scale@xcite when all three fock spaces of an electron in a dot ( empty , singly occupied , and doubly occupied configurations ) are included . in the infinite @xmath88 nca , pseudoparticle green s functions are introduced for the empty and singly occupied states , and the self - energy equations for these propagators are solved self - consistently to second order in @xmath89 and @xmath90 . the empty state is represented by a slave boson operator @xmath91 and singly occupied states ( doubly degenerate ) are denoted by the pseudofermion operator @xmath92 . the highest lying electron s annihilation operator in a dot can then be represented as @xmath93 the nca self - energy diagrams for a pseudofermion and a slave boson are displayed in fig . [ ncaself ] for our model system and their equations are @xmath94 respectively . here @xmath95 accounts for the two spin directions or the degeneracy of pseudofermion state . @xmath96 and @xmath97 are the green s functions of the slave boson and pseudofermion operators and are defined as @xmath98 projecting onto the physical hilbert space@xcite , the lesser and retarded self - energies can be shown to be given by the equations , @xmath99 these four equations are our nca integral equations to be solved . the lesser and greater self - energies of a quantum dot due to the hopping into the two leads are @xmath100 where the anderson hybridization functions are given by the equations , @xmath101 , \\ \overline{\gamma}_r ( \epsilon ; -\phi ) & = & \frac{\overline{n}_r ( \epsilon ) } { |d^r(\epsilon)|^2 } \left [ \gamma_r + \gamma\gamma_l |g_l^{r } ( \epsilon)|^2 \right . \nonumber\\ & & \hspace{1.0 cm } \left . + z g_l^{r } ( \epsilon ) + z^ * g_l^{a } ( \epsilon ) \right ] . \end{aligned}\ ] ] two fermi - dirac functions @xmath102 and @xmath103 describe the thermal distribution in the left and right electrodes , respectively and @xmath104 . the anderson hybridization functions , @xmath51 and @xmath52 , are renormalized due to the direct tunneling term . in a wide conduction band limit , the renormalized anderson hybridization becomes independent of energy variable and simplifies to @xmath105 } { ( 1+\gamma)^2}.\end{aligned}\ ] ] this reduction of couplings between the quantum dot and the two leads results in the smaller kondo temperature than in the absence of the direct tunneling . in addition , the effective anderson hybridization functions are explicitly dependent on the aharonov - bohm phase @xmath46 . note that the ab phase dependence of the anderson hybridization functions is different in the above self - energies and in the equations of table [ table1 ] ( c ) and ( d ) defined in the expression for the transmission probability . when a finite source - drain bias voltage is applied , @xmath106 is not equal to @xmath107 , and the lesser and the greater self - energies @xmath108 do not remain invariant under the inversion of the magnetic flux , @xmath109 . out of equilibrium , the ab phase symmetry is broken in the pseudoparticle self - energies and the green s function of a quantum dot so that the transmission probabilities are not the same under @xmath109 . in equilibrium , @xmath110 and the ab phase symmetry under @xmath109 is recovered . although we use the formalism presented in appendix [ gdosband ] which treats electrodes with a general density of states ( dos ) , in actual numerical work a lorentzian dos is adopted , @xmath111 here @xmath112 is the bandwidth of two leads and is used as the energy unit ( @xmath113 ) . the formalism for the flat dos in the leads or the wide conduction band limit is useful in analyzing our numerical results . throughout our numerical works , we use the following set of model parameters for a quantum dot . the energy level of a dot is chosen to be @xmath114 . the values of @xmath51 and @xmath52 are adjusted to satisfy both the total linewidth , @xmath115 , and the chosen value of @xmath57 ( the asymmetry coupling factor ) . since the effective anderson hybridization function or linewidth of the quantum dot gets smaller due to the direct tunneling between two leads , the kondo temperature @xmath0 is also suppressed . in a wide conduction band limit , the linewidth is @xmath116 , which is independent of the ab phase @xmath46 . @xmath0 can be estimated by the equation , @xmath117 where the dimensionless exchange coupling @xmath118 is defined by the relation @xmath119 . one notable point is that the kondo temperature is very sensitive to the value of the direct tunneling probability @xmath56 . in the nca approach to the anderson impurity , the one - body hopping term ( @xmath89 and @xmath90 ) is used as the expansion parameter . since only a subset of the self - energy diagrams is included up to the infinite order , the nca underestimates the one - body contribution ( @xmath82 ) to the self - energy of an impurity site . accordingly , the fermi liquid relation@xcite is not satisfied at @xmath120k and the kondo resonance peak is exaggerated especially for the orbitally nondegenerate @xmath121 anderson model . this may lead to the violation of the causality relation well below @xmath0 , or to the negative spectral weight in other renormalized green s functions@xcite ( the transmission probability @xmath29 near @xmath122 in our case ) . we can remedy this _ situation by exploiting the fermi liquid relation@xcite for the self - energy of a dot , @xmath123 this relation follows from the fact that the dot s self - energy due to the coulomb interaction @xmath88 satisfies the following relation@xcite , @xmath124 ^ 2 , \end{aligned}\ ] ] when @xmath125 . however the self - energy @xmath126 of a dot computed from the nca always satisfies the following inequality , @xmath127 that is , the imaginary part of the nca self - energy is always less than the fermi liquid value , @xmath128 . in fact , the nca does a relatively good job in producing the energy and temperature dependence of the self - energy near the fermi energy , but underestimates the one - body contribution to the dot s self - energy as noted above . note that the one - body contribution is temperature - independent , and its form is known once the shape of the anderson hybridization is given . we can make corrections in the one - body contribution to the nca self - energy of a dot such that the above fermi liquid relation is qualitatively satisfied . with this scheme@xcite , we remedy the _ unphysical _ behavior of the nca results at low temperatures well below @xmath0 in our work . 0.5 cm _ transmission probability spectral function _ , @xmath29 . the transmission probability shows the crossover behavior as a function of temperature due to the kondo correlation in the kondo dot . at high temperatures above @xmath0 , electrons flow from the left lead to the right one predominantly via the direct tunneling since the tunneling of an electron from the leads to the dot is prohibited by the strong coulomb repulsion ( coulomb blockade ) . in this case , the ab effect and the fano interference in the transmission probability are weak . as the system is cooled below @xmath0 , the kondo resonance peak in the quantum dot develops close to the fermi level and provides a new channel for the flow of electrons though the dot . the newly opened current path interferes with the direct tunneling path leading to fano interference and a strong ab effect in the transmission probability . the general structure of the transmission spectral functions near the fermi level ( @xmath122 ) can be read off from the equation ( [ tprobflat ] ) . as shown in fig . [ kondopeak ] @xmath129 develops the kondo resonance peak with its width of the order of @xmath0 near @xmath122 while @xmath130 varies very rapidly over the energy scale of @xmath0 near @xmath122 with a dip just below @xmath131 and a peak above @xmath122 . the overall shape of the transmission spectral function @xmath29 is determined by the value of the ab phase @xmath46 and the sign of @xmath132 [ see eq . ( [ tprobflat ] ) ] , @xmath133 a typical fano interference pattern a dip and peak structure is expected when @xmath134 . at @xmath135 , @xmath29 has a dip(peak ) resonance structure if @xmath136 , respectively . furthermore , we have one exact relation at zero temperature for the transmission probability at the fermi level ( @xmath137 ) , @xmath138 , which derives from the fermi liquid relation@xcite for the green s function of a dot , @xmath139 . in equilibrium , @xmath140 or the transmission probability remains invariant under the inversion of the magnetic flux , @xmath141 . this ab phase symmetry or onsager relation can be deduced from the expression of @xmath29 in eq.([tprobgdos ] ) or eq.([tprobflat ] ) . if @xmath142 remains invariant under the transformation @xmath109 , so is the transmission probability . in equilibrium , the fermi - dirac thermal distribution functions in two leads are identical or @xmath110 . @xmath143 since the total anderson hybridization @xmath144 is an even function of @xmath46 and is reduced to @xmath145 in a wide conduction band limit which does not depend on the ab phase . due to the ab phase symmetry and the periodicity in @xmath46 , we have only to consider @xmath29 in the range , @xmath146 . out of equilibrium or when a finite source - drain voltage is applied , the two thermal distribution functions are not identical so the ab phase symmetry is broken in @xmath78 and @xmath29 . when the direct tunneling is weak , so is the fano interference , and the kondo - related resonance peak persists for the entire range of the ab phase . as the direct tunneling is increased , the overall magnitude of the transmission probability is also increased , and the fano interference becomes more pronounced . when @xmath147 or the quantum dot is coupled symmetrically to the left and right leads ( @xmath148 ) , the transmission probability @xmath29 for the case of @xmath149 , is presented in the short paper@xcite . we briefly summarize the shape of @xmath29 in this case . when @xmath135 , the transmission probability has a kondo - related resonance peak near the fermi level . when @xmath134 , the typical shape of a fano antiresonance shows up in the transmission probability . when @xmath150 , a dip ( peak ) structure develops below ( above ) the fermi level ( @xmath151 ) , respectively . on the other hand , a dip ( peak ) structure arises above ( below ) @xmath151 , respectively , when @xmath152 . 0.5 cm 0.5 cm in figs . [ tprobap ] and [ tprobad ] , we present the computed @xmath29 when the couplings are asymmetrical or @xmath153 . the ab phase dependence of @xmath29 is displayed in fig . [ tprobap ] when @xmath154 and @xmath155 . comparing the curves in fig . [ tprobap ] and in fig . 2 in our short paper@xcite , the overall shape of @xmath29 is not qualitatively different between _ asymmetrically _ and _ symmetrically _ coupled dots when @xmath134 . when @xmath156 [ see fig . [ tprobap ] ( a ) ] , @xmath29 shows a typical fano interference pattern , a dip structure below @xmath122 and a peak above @xmath122 . with further increasing @xmath46 for @xmath157 , the fano resonance pattern is inverted with respect to @xmath122 compared to the case @xmath158 . for a symmetrical kondo resonance peak at @xmath122 or when the energy structure in a quantum dot is particle - hole symmetric , exact inversion is expected , @xmath159 , for @xmath160 . in our case of the infinite @xmath88 calculation , the doubly occupied configuration in the dot is removed from our consideration , and particle - hole symmetry is broken . due to this broken particle - hole symmetry , the spectral function of the dot has more spectral weight in the electron excitations ( @xmath161 ) than the hole excitations ( @xmath162 ) and the relation @xmath159 for @xmath160 does not hold in our case , which can be seen by comparing the curves in fig . [ tprobap ] ( a ) and ( c ) . when @xmath135 or @xmath163 , @xmath132 is equal to @xmath164 and re@xmath30 is absent in @xmath29 . the shape of @xmath29 near @xmath122 is solely determined by the sign of @xmath132 . @xmath132 remains always negative in the case of a _ symmetrically _ coupled dot so that the kondo resonance peak persists in the transmission probability spectral function over the entire range of the direct tunneling amplitude @xmath56 . with increasing @xmath56 , the kondo - related peak becomes smaller and in the end is buried by the magnitude of @xmath56 as @xmath165 . note that @xmath29 is less than or equal to one for all energies in our _ one transport mode _ model . the transmission spectral function reaches its maximum possible value @xmath166 , a unitary kondo resonance tunneling . on the other hand , @xmath132 can change its sign from minus to plus in the case of an _ asymmetrically _ coupled dot . @xmath29 retains its peak structure near @xmath122 until @xmath167 [ see fig . [ tprobad ] ( a ) ] . the maximum value of @xmath29 is set by the value of @xmath57 . when @xmath168 or @xmath169 [ see fig . [ tprobad ] ( b ) ] , @xmath170 is a constant . the interference completely washes out the tunneling probability through the quantum dot . the dip structure develops replacing the peak near @xmath122 as soon as @xmath132 becomes positive with increasing direct tunneling amplitude , @xmath171 [ see fig . [ tprobad ] ( c ) ] . the dip structure becomes more pronounced with further increasing @xmath56 . 0.5 cm _ electric conductance _ , @xmath172 . the temperature and ab phase dependence of the electric conductance is displayed in fig . [ eqg ] for a symmetrically coupled dot . the temperature dependence of the electric conductance is determined by the values of @xmath56 and @xmath46 . when @xmath56 is small , the electric conductance is monotonically increasing with lowering temperature for all ab phases . when @xmath56 is large , @xmath173 is increasing for some range of @xmath46 and decreasing for others ( see fig . [ eqg ] ) . the direct tunneling probability is a temperature - independent constant , while the transmission through the kondo dot is strongly temperature dependent and shows crossover behavior as a function of temperature due to the kondo correlation . since the ab effects derive from the interference between two paths , the amplitude of the ab oscillation depends on the magnitude of the transmission coefficients for the two arms of the ab interferometer . a stronger direct tunneling amplitude , @xmath56 , leads to a larger amplitude of the ab oscillation . the amplitude of the ab oscillation also shows crossover behavior as a function of temperature . the amplitude is weak ( strong ) at high ( low ) temperatures above ( below ) @xmath0 , respectively . this crossover behavior derives from the fano interference and the kondo correlation in the quantum dot . since the quantum dot lies in the coulomb blockade regime at high temperatures above @xmath0 , the green s function of the dot is negligibly small near the fermi level so that the amplitude of the ab oscillation is weak . on the other hand , the kondo correlation with lowering temperature below @xmath0 opens a new transport channel through the quantum dot and the amplitude of the ab oscillation is enhanced . 0.5 cm _ thermal conductance and lorenz ratio _ , @xmath174 and @xmath175 . the diagonal transport coefficients , the electric and thermal conductances , are influenced by the fano interference and the ab flux in the same way . the variation of @xmath174 versus @xmath2 or @xmath46 is very similar to that of electric conductance @xmath173 . the lorenz number @xmath175 is defined by the ratio of two diagonal conductances @xmath176 the variation of @xmath177 is displayed in fig . [ eql ] as a function of temperature @xmath2 and the ab phase @xmath46 . our results obey the wiedeman - franz law satisfying the relation @xmath178 , this relation is independent of the ab flux . the ab oscillation in the lorenz number is strong in the crossover region or near the kondo temperature . 0.5 cm _ thermopower , @xmath179_. since thermopower measures the average of the transmission probability weighted by the electronic excitation energy with respect to the fermi energy , it is a very sensitive probe of the particle - hole asymmetry in the transmission probability @xmath29 . as shown in figs . [ tprobap ] and [ tprobad ] , the shape of @xmath29 can be modulated through the fano interference by the _ ab flux _ and the _ direct tunneling amplitude @xmath56_. the sign of the thermopower can probe this structure . the sign of the thermopower is one of the experimental tools which can determine the main carriers of charge and heat . when more spectral weight lies in the electron excitations than in the hole excitations , the main carriers of charge and heat are electron excitations , and the sign of @xmath179 is negative [ see figs . [ tprobap](a ) and [ eqt](b ) ] . in the opposite case , the main carriers are hole excitations , and the sign of @xmath179 is positive [ see fig . [ tprobap](c ) and [ eqt](b ) ] . in the particle - hole symmetric case ( not shown here ) , electron and hole excitations carry the same amount of electric and heat currents . the signs of electric ( heat ) current are the same ( opposite ) for electron and hole excitations . the net result is that the thermopower is zero . the particle - hole symmetric case can be realized when the quantum dot is located in the center of the coulomb blockade diamond . since the main part of our computed thermopower @xmath179 in the ab interferometer is already reported in our short paper@xcite , we briefly summarize the main results . we find that the sign of @xmath179 can be changed by the energy structure in a quantum dot@xcite or the gate voltage capacitively coupled to the dot and by the ab flux@xcite . the magnitude of @xmath179 is anomalously enhanced and is of the order of @xmath180 within the crossover region or near the kondo temperature . for comparison , @xmath179 is of order @xmath181 in most of normal metals . the same enhancement of @xmath179 is well known in heavy fermion systems and in the bulk kondo systems@xcite . when the direct tunneling amplitude is weak or @xmath182 , the sign of @xmath179 remains invariant under the inversion of ab flux at temperatures near or less than @xmath0 [ see fig . [ eqt](a ) ] . this result follows from the fact that the fano interference is weak and the kondo - related peak persists in @xmath29 over the entire range of ab phase @xmath46 . the sign change in @xmath179 with ab flux can be realized when the direct tunneling @xmath56 is increased [ see fig . [ eqt](b ) and also fig . 3 in our short paper@xcite ] . the amplitude of an ab oscillation in the thermopower is strongest within the crossover temperature region which is similar to the lorenz number . the amplitude is weak ( strong ) when the direct tunneling is small ( large ) , respectively [ see figs . [ eqt](a ) and ( b ) ] . the sign of @xmath179 can be changed by varying the direct tunneling amplitude @xmath56 . when @xmath183 in fig . [ tprobad](a ) , the transmission probability has more spectral weight in the electron excitations than in the hole excitations . in this case , the thermopower is negative . in fig . [ tprobad](b ) or when @xmath184 , @xmath29 is more or less particle - hole symmetric so that @xmath185 . when the direct tunneling is increased or @xmath186 [ fig . [ tprobad](c ) ] , holes are the main carriers since the transmission probability is featured with more spectral weight below the fermi energy . in this case , the thermopower is positive . in this work we studied the effects of fano interference and aharonov - bohm phase on the diagonal and off - diagonal transport coefficients in an ab ring with an embedded quantum dot in the kondo regime . the transport properties shows crossover behavior from a high - temperature regime with no kondo correlation to a low - temperature regime with kondo correlation . the crossover behavior manifests itself in the amplitude of an ab oscillations in the diagonal transport coefficients , the electric and thermal conductances . the amplitude is small at high temperatures above @xmath0 but becomes enhanced below @xmath0 by the kondo correlation in the quantum dot . in the case of thermopower , the amplitude of an ab oscillation is strongest within the crossover region near the kondo temperature . the lorenz number , defined as the ratio of electric and thermal conductances , is fixed at the sommerfeld value near zero temperature , but also shows ab oscillations near the kondo temperature . the effect of the ab flux is more dramatic in the thermopower than in diagonal transport coefficients . in addition to the ab oscillations in magnitude , even the sign of thermopower can be modulated by the ab flux . in the case of weak direct tunneling amplitude , the sign change is not possible under the ab flux since the fano inference is weak . when the direct tunneling amplitude is increased , the ab flux can change the sign of the thermopower . most of studies in quantum dots have focused on the @xmath1 curves . charge confinement is relatively easy with the use of the gate voltages , but the heat confinement is a more difficult job possibly because of the easy transfer of heat via other excitations . with more advances in nanotechnology , thermoelectric transport coefficients will provide additional information about the quantum transport in quantum dot systems . this work is supported in part by the bk21 , in part by the national science foundation under grant no . dmr 9357474 , and in part by grant no . 1999 - 2 - 114 - 005 - 5 from the kosef . in this section , we derive the expression of electric and thermal currents for general shape of the density of states(dos ) for two leads . since the derivation of thermal current is the same as electric current , we derive the expression of electric current below in detail and discuss briefly thermal current at the end . 0.5 cm 0.3 cm using the nonequilibrium green s function method@xcite , the electric current can be expressed in terms of the lesser mixed green s functions , @xmath187 \nonumber\\ & & + \frac{2e}{h } \frac{1}{\sqrt{v } } \sum_{\vec{k}\alpha } \int d\epsilon ~ \mbox{im } \left [ v_{ld } ~ g_{dl}^ { < } ( \vec{k } ; \epsilon ) \right ] , \end{aligned}\ ] ] where two mixed green s functions are defined by the equations @xmath188 the feynman diagrams of these two mixed green s functions are shown in fig . [ mgreen ] . the mixed green s function @xmath189 , shown in fig . [ mgreen](a ) , is given by the equation @xmath190 the subscript @xmath191 in the integral symbol denotes the keldysh contour . the auxiliary green s function @xmath192 is defined in fig . [ dgreen](a ) and is determined by the following dyson - like equation @xmath193 @xmath194 is the green s function of the conduction electrons in the lead @xmath195 when @xmath196 . @xmath197 @xmath78 is the green s function of a quantum dot , @xmath198 and its self - energy ( @xmath199 ) consists of two parts , the one - body contribution(@xmath82 ) coming from tunneling into two reservoirs and the many - body contribution ( @xmath200 ) due to the on - site coulomb interaction @xmath88 . [ dgreen](b ) gives the one - body contribution to the self - energy of a quantum dot . @xmath201 @xmath82 includes all the multiple tunneling processes between two leads . feynman diagrams of the mixed green s function @xmath202 is shown in fig . [ mgreen](b ) and @xmath202 is given by the equation @xmath203 \nonumber\\ & & \hspace{1.0 cm } \times g_d ( t_1 , t_2 ) ~ \left [ v_{d\bar{p } } \overline{g}_{\bar{p } } ( \vec{k}_2t_2 , \vec{k}'t ' ) + v_{dp } g_{p\bar{p}0 } ( \vec{k}_2t_2 , \vec{k}'t ' ) \right ] , \\ g_{p\bar{p}0 } ( \vec{k}t , \vec{k}'t ' ) & = & \frac{1}{v } \sum_{\vec{k}_1 } \int_c dt_1 ~ \overline{g}_{p}(\vec{k}t , \vec{k}_1t_1 ) t_{p\bar{p } } g_{\bar{p } } ( \vec{k } ' ; t_1,t ' ) . \end{aligned}\ ] ] another auxiliary green s function @xmath86 , defined in fig . [ dgreen](c ) , is introduced to facilitate the algebra . introducing the wave - vector - summed green s functions , @xmath204 the electric current can be written as @xmath205 \nonumber\\ & & + \frac{2e } { h } n_s \int d\epsilon ~ \mbox{im } \left [ v_{ld } g_{dl}^ { < } ( \epsilon ) \right ] . \end{aligned}\ ] ] here @xmath95 is two possible spin directions of conduction electrons in the reservoirs . the wave - vector - summed green s functions are determined by the following equations , @xmath206 , \\ \label{grnrl } g_{rl } ( t , t ' ) & = & g_{rl0 } ( t , t ' ) + \int_c dt_1 \int_c dt_2 ~ \left\ { \overline{g}_{r } ( t , t_1 ) ~v_{rd } + g_{rl0 } ( t , t_1 ) ~v_{ld } \right\ } \nonumber\\ & & \hspace{2.5 cm } \times g_d ( t_1 , t_2 ) ~ \left\ { v_{dl } \overline{g}_{l } ( t_2 , t ' ) + v_{dr } g_{rl0 } ( t_2 , t ' ) \right\ } \nonumber\\ & = & g_{rl0 } ( t , t ' ) + \int_c dt_1 ~ \left\ { \overline{g}_{r } ( t , t_1 ) ~v_{rd } + g_{rl0 } ( t , t_1 ) ~v_{ld } \right\}~ g_{dl } ( t_2,t ' ) , \\ \label{indgrn1 } \overline{g}_l ( t , t ' ) & = & g_l(t , t ' ) + \int_c dt_1 \int_c dt_2 ~ \overline{g}_l ( t , t_1 ) t_{lr } g_r ( t_1,t_2 ) t_{rl } g_l(t_2,t ' ) , \\ \label{indgrn2 } g_{rl0 } ( t , t ' ) & = & \int_c dt_1 ~ g_r ( t , t_1 ) t_{rl } \overline{g}_{l } ( t_1 , t ' ) ~=~ \int_c dt_1 ~ \overline{g}_r ( t , t_1 ) t_{rl } g_{l } ( t_1 , t ' ) . \end{aligned}\ ] ] @xmath207 is obtained from the equation ( @xmath208 ) by interchanging @xmath209 . the retarded , advanced , and lesser green s functions of the left and right reservoirs ( summed over the wave vector ) can be written as @xmath210 here @xmath211 is the dos of two leads and is normalized such that @xmath212 . reduced green s function @xmath213 is introduced to simplify our algebra below and @xmath214 . the auxiliary green s functions , given by the equations ( [ indgrn1 ] , [ indgrn2 ] ) , are @xmath215 , \\ g_{p\bar{p}0}^{r , a } ( \epsilon ) & = & \frac{t_{p\bar{p } } } { d^{r , a}(\epsilon ) } ~ g_p^{r , a } ( \epsilon ) g_{\bar{p}}^{r , a } ( \epsilon ) ~=~ \frac{t_{p\bar{p } } } { d^{r , a}(\epsilon ) } ~ \pi^2 n_l n_r g_l^{r , a } ( \epsilon ) g_{r}^{r , a } ( \epsilon ) , \\ g_{p\bar{p}0}^ { < } ( \epsilon ) & = & \frac{t_{p\bar{p } } } { |d^{r}(\epsilon)|^2 } ~ \left [ g_p^ { < } ( \epsilon ) g_{\bar{p}}^{a } ( \epsilon ) + g_p^{r } ( \epsilon ) g_{\bar{p}}^ { < } ( \epsilon ) \right ] \nonumber\\ & = & \frac{t_{p\bar{p } } } { |d^{r}(\epsilon)|^2 } ~ 2\pi^2 n_l n_r \left [ g_{\bar{p}}^{a } ( \epsilon ) f_p ( \epsilon ) + g_p^{r } ( \epsilon ) f_{\bar{p } } ( \epsilon ) \right].\end{aligned}\ ] ] new variables are introduced , @xmath216 the electric current [ eq.([ieqn ] ) ] can be written as @xmath217 note the following reduction of equations @xmath218 ~=~ \frac { g_{l}^{a } } { d^{r } } ~ \overline{v}_{dl}^{a } , \\ 1 + t_{lr } g_{rl0}^{r } & = & \frac{1 } { d^r } , \\ v_{dl } \overline{g}_l^ { < } + v_{dr } g_{rl0}^ { < } & = & \frac{1 } { |d^r|^2 } \left [ \overline{v}_{dl}^{r } g_{l}^ { < } + \overline{v}_{dr}^{r } g_{r}^ { < } t_{rl } g_{l}^{a } \right ] , \\ \overline{g}_r^ { < } v_{rd } + g_{rl0}^ { < } v_{ld } & = & \frac{1 } { |d^r|^2 } \left [ g_{r}^ { < } \overline{v}_{rd}^{a } + g_{r}^{r } t_{rl } g_{l}^ { < } \overline{v}_{ld}^{a } \right ] , \end{aligned}\ ] ] where the renormalized tunneling matrices are introduced to simplify the equations . @xmath219 the current spectral function in the left lead can be written as @xmath220 we also find the current spectral function in the right lead , @xmath221 since only the imaginary parts are relevant to the expression of electric current , the current spectral function in each lead can be further simplified . @xmath222 + \overline{v}_{dr}^{r } g_{r}^ { < } t_{rl } [ g_{l}^{a } - g_{l}^{r } ] \overline{v}_{ld}^{r } \right\ } \right\ } , \\ \mbox{im } g_{ir}^ { < } ( \epsilon ) & = & t_0 ( \epsilon ) \left\ { f_{l}(\epsilon ) - f_{r}(\epsilon ) \right\ } + 2 g_{d}^ { < } \cdot \overline{\gamma}_r ( \epsilon ) \nonumber\\ & & + \frac{2 } { |d^r|^2 } \mbox{im } \left\ { \frac{g_d^{r}}{d^r } \left\ { \overline{v}_{dr}^{r } g_{r}^ { < } \overline{v}_{rd}^{r } [ 1 - |t_{lr}|^2 g_{r}^{r } g_{l}^{a } ] + \overline{v}_{dl}^{r } g_{l}^ { < } t_{lr } [ g_{r}^{a } - g_{r}^{r } ] \overline{v}_{rd}^{r } \right\ } \right\ } , \end{aligned}\ ] ] where we introduced the transmission probability @xmath56 for the direct tunneling and the renormalized anderson hybridization @xmath223 , @xmath224 to simplify the algebra , we write @xmath225 \right\ } , \\ a_{lr } & = & \frac{1 } { |d^r|^2 } \mbox{im } \left\ { \frac{g_d^{r } } { d^r } \overline{v}_{dr}^{r } t_{rl } [ g_{l}^{a } - g_{l}^{r } ] \overline{v}_{ld}^{r } \right\ } , \\ a_{rl } & = & \frac{1 } { |d^r|^2 } \mbox{im } \left\ { \frac{g_d^{r } } { d^r } \overline{v}_{dl}^{r } t_{lr } [ g_{r}^{a } - g_{r}^{r } ] \overline{v}_{rd}^{r } \right\ } , \\ a_{rr } & = & \frac{1 } { |d^r|^2 } \mbox{im } \left\ { \frac{g_d^{r } } { d^r } \overline{v}_{dr}^{r } \overline{v}_{rd}^{r } [ 1 - |t_{lr}|^2 g_{r}^{r } g_{l}^{a } ] \right\}.\end{aligned}\ ] ] using the current conservation in a steady state , @xmath226 , the expression of @xmath227 can be obtained in terms of the retarded and advanced green s functions . @xmath228 . \end{aligned}\ ] ] inserting the lesser green s function of a quantum dot into @xmath229 , we find the current spectral function , @xmath230 \nonumber\\ & & + \frac { 2 } { \overline{\gamma } ( \epsilon ) } \left [ ( \overline{\gamma}_r a_{ll } - \overline{\gamma}_l a_{rl } ) g_{l}^ { < } + ( \overline{\gamma}_r a_{lr } - \overline{\gamma}_l a_{rr } ) g_{r}^ { < } \right ] . \end{aligned}\ ] ] after some algebra , we can readily derive the following identities , @xmath231 using these results , the current spectral function is reduced to a simple form , @xmath232 , \\ t(\epsilon ) & = & t_0 ( \epsilon ) - \mbox{im } \left [ g_d^{r } z_{lr } \right].\end{aligned}\ ] ] newly introduced parameters are given by the equations , @xmath233 ^ 2 + ( z + z^ * ) g_r^{r } ( \epsilon ) \right ] , \\ z_r^{r } ( \epsilon ) & = & \frac{\pi n_r(\epsilon ) \overline{v}_{dr}^{r } \overline{v}_{rd}^{r } } { d^r(\epsilon ) |d^r| } ~=~ \frac{\overline{n}_r ( \epsilon ) } { d^r(\epsilon ) |d^r| } \left [ \gamma_r + \gamma\gamma_l [ g_l^{r } ( \epsilon)]^2 + ( z + z^ * ) g_l^{r } ( \epsilon ) \right ] , \\ \overline{\gamma}_l(\epsilon ) & = & \frac{\pi n_{l } ( \epsilon ) |v_{dl}^{a}|^2 } { |d^r|^2 } ~=~ \frac{\overline{n}_l ( \epsilon ) } { |d^r|^2 } \left [ \gamma_l + \gamma\gamma_r |g_r^{r } ( \epsilon)|^2 + z g_r^{r } ( \epsilon ) + z^ * g_r^{a } ( \epsilon ) \right ] , \\ \overline{\gamma}_r(\epsilon ) & = & \frac{\pi n_{r } ( \epsilon ) |v_{dr}^{a}|^2 } { |d^r|^2 } ~=~ \frac { \overline{n}_r ( \epsilon ) } { |d^r|^2 } \left [ \gamma_r + \gamma\gamma_l |g_l^{r } ( \epsilon)|^2 + z^ * g_l^{r } ( \epsilon ) + z g_l^{a } ( \epsilon ) \right].\end{aligned}\ ] ] here @xmath234 where @xmath46 is the aharonov - bohm phase due to the magnetic flux threading through the ab ring . thermal current can be derived in the same way . the only difference between electric and thermal currents is the multiplying factor in the integrands of the landauer - bttiker formula . the charge of an electron , @xmath235 , in the electric current is replaced by the energy of an electron , @xmath236 , in the thermal current leaving the left lead . the thermal current is given by the expression , @xmath237.\end{aligned}\ ] ] here @xmath29 is the transmission probability which we derived in the above . in a wide conduction band limit , all the green s functions related to the leads become energy - independent constants . especially , @xmath238 , @xmath239 , and @xmath240 . the expression of the current can be highly simplified in this case . all the relevant green s functions of two leads are given by the equations , @xmath241 , \\ g_{p\bar{p}0}^{r , a } & = & - \frac{\pi^2 n_l n_r } { 1 + \gamma } ~t_{p\bar{p } } , \\ g_{p\bar{p}0}^ { < } & = & \frac{2i\pi^2 n_l n_r } { ( 1+\gamma)^2 } ~t_{p\bar{p } } \left [ f_p(\epsilon ) - f_{\bar{p } } ( \epsilon ) \right].\end{aligned}\ ] ] @xmath242 , \\ z_r^{r } & = & \frac{1 } { ( 1 + \gamma)^2 } \left [ \gamma_r - \gamma \gamma_l - i(z+z^ * ) \right ] , \\ \overline{\gamma}_l & = & \frac{1 } { ( 1 + \gamma)^2 } \left [ \gamma_l + \gamma \gamma_r - i(z - z^ * ) \right ] , \\ \overline{\gamma}_r & = & \frac{1 } { ( 1 + \gamma)^2 } \left [ \gamma_r + \gamma \gamma_l + i(z - z^ * ) \right].\end{aligned}\ ] ] the transmission spectral function @xmath29 can be written in another form , @xmath249~ \mbox{im } g_d^r , \\ t_0 & = & \frac{4\gamma } { ( 1+\gamma)^2 } , ~~~ g ~=~ \frac{4\gamma_l \gamma_r } { ( \gamma_l + \gamma_r)^2 } . \end{aligned}\ ] ] @xmath56 and @xmath57 are the dimensionless conductance of direct tunneling and a resonant tunneling through a quantum dot , respectively . we note that the expression of @xmath29 in the wide conduction band limit was derived in the literature.@xcite 100 a. yacoby , m. heiblum , d. mahalu , and h. shtrikman , phys . 74 * , 4047 ( 1995 ) . r. schuster , e. buks , m. heiblum , d. mahalu , v. umansky , and h. shtrikman , nature * 385 * , 417 ( 1997 ) . y. ji , m. heiblum , d. sprinzak , d. mahalu , and h. shtrikman , science * 290 * , 779 ( 2000 ) . y. ji , d. mahalu , and h. shtrikman , phys . lett . * 88 * , 076601 ( 2002 ) . j. gres , d. goldhaber - gordon , s. heemeyer , m.a . kastner , h. shtrikman , d. mahalu , and u. meirav , phys . rev . b * 62 * , 2188 ( 2000 ) . k. kobayashi , h. aikawa , s. katsumoto , and y. iye , phys . lett . * 88 * , 256806 ( 2002 ) . v. madhavan , w. chen , t. jamneala , m.f . crommie , and n.s . wingreen , science * 280 * , 567 ( 1998 ) . j. li , w. schneider , r. berndt , and b. delley , phys . lett . * 80 * , 2893 ( 1998 ) . manoharan , c.p . lutz , and d.m . eigler , nature * 403 * , 512 ( 2000 ) . odom , j .- l . huang , c.l . lieber , science * 290 * , 1549 ( 2000 ) . for the case of noninteracting electrons , see , for example , g. d. guttman , e. ben - jacob , and d. j. bergman , phys . b * 51 * , 17758 ( 1995 ) ; ya . m. blanter , c. bruder , r. fazio , and h. schoeller , phys . rev . b * 55 * , 4069 ( 1997 ) ; m. v. moskalets , zh . ksp . . fiz . * 114 * , 1827 ( 1998 ) . d. goldhaber - gordon , h. shtrikman , d. malahu , d. abusch - magder , u. meirav , and m.a . kastner , nature * 391 * , 156 ( 1998 ) . cronenwett , t.h . oosterkamp , and l.p . kouwenhoven , science * 281 * , 540 ( 1998 ) . j. nygard , d.h . cobden , and p.e . lindelof , nature * 408 * , 342 ( 2000 ) . glazman and m.e . raikh , zh . . fiz . * 47 * , 378 ( 1988 ) [ jetp lett . * 47 * , 452 ( 1988 ) ] ; t.k . ng and p.a . lee , phys . lett . * 61 * , 1768 ( 1988 ) . langreth , 1976 , in _ linear and nonlinear electron transport in solids _ , vol.*17 * of _ nato advanced study institute , series b : physicsi _ , edited by j.t . devreese and v.e . van doren ( plenum , new york , 1976 ) , p. 3 . y. meir , n. s. wingreen , and p. a. lee , phys . lett . * 70 * , 2601 ( 1993 ) ; n.s . wingreen and y. meir , phys . b * 49 * , 11040 ( 1994 ) . hettler , j. kroha , and s. hershfield , phys . 73 * , 1967 ( 1994 ) . kim and s. hershfield , phys . b * 63 * , 245326 ( 2001 ) .

we investigate the nature and spatial variations of turbulence in the small magellanic cloud ( smc ) by applying several statistical methods on the neutral hydrogen ( hi ) column density image of the smc and a database of isothermal numerical simulations . by using the 3rd and 4th statistical moments we derive the spatial distribution of the sonic mach number ( ) across the smc . we find that about 90% of the hi in the smc is subsonic or transonic . however , edges of the smc ` bar ' have and may be tracing shearing or turbulent flows . using numerical simulations we also investigate how the slope of the spatial power spectrum depends on both sonic and alfvn mach numbers . this allows us to gauge the alfvn mach number of the smc and conclude that its gas pressure dominates over the magnetic pressure . the super - alfvnic nature of the hi gas in the smc is also highlighted by the bispectrum , a three - point correlation function which characterizes the level of non - gaussianity in wave modes . we find that the bispectrum of the smc hi column density displays similar large - scale correlations as numerical simulations , however it has localized enhancements of correlations . in addition , we find a break in correlations at a scale of pc . this may be caused by numerous expanding shells of a similar size .

0911.3652

in the recent decade , many advances in both observations and computational models have provided new insights into the workings and evolution of the interstellar medium ( ism ) . the emerging picture is that interstellar turbulence plays the key role in ism structure formation and evolution ( mckee & ostriker 2007 ) . in the galaxy and the magellanic clouds , the ism is turbulent on scales ranging from less than a parsec to a few kiloparsecs ( crovisier & dickey 1983 ; stanimirovic et al . 1999 ; deshpande et al . 2000 ; dickey et al . 2001 ; elmegreen et al . 2001 ; elmegreen & scalo 2004 ) . although the observational evidence for the importance of turbulence in the ism is overwhelming , many questions remain open . for example , what are the dominant energy sources and physical processes that convert kinetic energy into turbulence ( burkert 2006 ) ? at what scales and through which modes is turbulent energy dissipated ( heyer & zweibel 2004 ) ? how do the level and type of turbulence depend on properties of the interstellar gas ( e.g. presence / absence of star formation , presence / absence of tidal effects , or the strength of magnetic field ) ? since no complete theory of astrophysical turbulence exists , studying its effects on the multiphase ism can be challenging and calls for a combination of numerical and observational efforts . statistical studies have proved to be important in characterizing the magnetized turbulent ism @xcite , however the interpretation of results is not always straight forward . several statistical methods have been extensively used for both observational and synthetic data . these statistics include probability density functions ( pdfs ) , wavelets , the principal component analysis , higher order moments , velocity coordinate spectrum ( vcs ) , and velocity channel analysis ( vca ) , to name just a few ( gill & henriksen 1990 ; brunt & heyer 2002 ; kowal , lazarian & pogosyan 2000 , 2004 , 2006 ; lazarian & beresnyak 2007 ) . most of these statistical methods require large datasets with a large spatial or velocity dynamic range , and produce a single , mostly one - dimensional , measure . this results in the lack of spatial information about turbulent properties across a given interstellar cloud , or a galaxy , making a connection with underlying physical properties highly difficult . in this paper we explore a new method for obtaining spatial information about the level and nature of ism turbulence on the neutral hydrogen ( hi ) observations of the small magellanic cloud ( smc ) . the smc , a dwarf irregular galaxy in the local group , has a highly gas rich ism environment ( see , stanimirovic et al . 1999 , henceforth known as sx99 ) , and is an excellent candidate for ism studies . being nearby ( 60 kpc , @xcite ) , the smc is distant enough for all its objects to be treated as having roughly the same distance , unlike the milky way where distance determination is relatively uncertain . the hi observations of the smc , obtained using the australia telescope compact array ( atca ) and the parkes telescope ( sx99 ) , have been used for several investigations , including the hi spatial power spectrum and the kinematic study of hi , which revealed the existence of many expanding shells of gas and three supergiant shells . the power law index of hi density and velocity distributions was derived in sx99 and @xcite , while the genus statistic in @xcite revealed spatial variations of hi morphology . because the sx99 smc data set is well studied , it is a perfect candidate to investigate new statistical methods . we can acquire new information , but also test and confirm past results , as well as validate the promise of these statistical tools for further use in other observational studies . in this study , we investigate turbulent properties of the hi in the smc by applying the higher order moments on the hi column density image . we then use a database of mhd simulations to bootstrap the spatial distribution of the sonic mach number across the smc . the crucial aspect of our approach is the confluence of observations and numerical simulations : only by combining the two we can retrieve the spatial variations of turbulent properties . this is the reason why we oscillate between observational and synthetic data in this paper . we also investigate whether and how interstellar shocks leave footprints on the hi gas by employing the bispectrum , a three point statistical measure , on the smc hi column density image . again , to interpret our results we apply the same statistics on the database of mhd simulated column density images . in particular , the paper is organized as follows . we start with [ sec : background ] by providing a brief summary of previous work regarding the statistical methods used in our study . in [ sec : data ] we describe the smc hi column density map and the database of numerical simulations of compressible mhd turbulence used for the comparison with observations . in 4 we introduce higher order moments and their dependence on the sonic and alfvenic mach numbers . we then apply higher order moments on the smc hi observations to derive an image of the sonic mach number across the smc , in 5 . we compare our results with an observational estimate of the sonic mach number of the cold neutral medium ( cnm ) in the smc , based on a comparison of the spin and kinetic temperature of hi absorption profiles , in [ ratio ] . in [ ps ] we show how the power - law slope of the spatial power spectrum depends on the sonic mach number and use this to gauge the alfvenic mach number of the hi gas seen in emission . in [ sec : bispectrum ] we present an analysis of the bispectrum of the smc , as well as a brief discussion of the noise and windowing effects . in [ sec : discussion ] we provide a discussion of our results , followed by our conclusions . higher order statistical moments of density fluctuations have been studied extensively . for example , the variance of density fluctuations has been shown to increase with the sonic mach number @xmath0 @xcite . therefore , if turbulence is the dominant structuring mechanism , @xmath0 can be estimated from the variance of density fluctuations . however , the observable that is the most easily available from observations for various ism tracers is the column density . while this is a less direct measure of turbulence compared to velocity , a comparison between observations and simulations is the most straight forward in case of column densities . only very recently , kowal et al . ( 2007 ) , henceforth referred to as klb , have investigated how variance , skewness and kurtosis , the 2nd , 3rd and 4th order moments respectively , depend on @xmath0 . they found strong correlations : as the sonic mach number increases , so does the gaussian asymmetry of the column density ( and density ) pdfs due to gas compression via shocks , resulting in the increase of variance , skewness and kurtosis . klb used limited resolution models of 128@xmath3 in their study , while burkhart et al . ( 2009 ) , henceforth known as bfkl , saw the same trends using high resolution isothermal simulations . in both studies , the moments had little dependence on the alfvn mach number ( @xmath4 ) , or the line - of - sight ( los ) orientation used for integrating 3-d simulated data cubes . these studies are motivational , and imply that the sonic mach number of turbulence in interstellar clouds could be characterized by variance , skewness and kurtosis of observed column density distributions . the two - dimensional spatial power spectrum characterizes the energy distribution over spatial scales . sx99 found that the spatial power spectrum of individual hi velocity slices is well fit by a power - law , with an average slope of @xmath5 . stanimirovic & lazarian ( 2001 ) showed that the power - law slope steepens when several channels are integrated together , and used this to estimate the density power - law slope of @xmath6 and the velocity slope of @xmath7 . no obvious breaks or preferred scales were found over the range of 30 pc to 4 kpc . the density and velocity spectral slopes are similar , and in the case of incompressible mhd turbulence , density behaves as a passive scalar and thus scales in the same way as velocity ( monin & yaglom(1967 ) , lithwick & goldreich ( 2001 ) , cho & lazarian 2003 ) . however , the estimated slopes are slightly more shallow than the predictions for the kolmogorov spectrum ( @xmath8 ) , which is expected for incompressible fluids with a weak magnetic field . however , different types of turbulence are expected to have different spectral slopes . for example , in incompressible fluids with a strong magnetic field , the spectrum is expected to be even steeper and scale as @xmath9 ( biskamp 2003 ) . when the medium is supersonic ( as we will see later applies to parts of the smc ) , these relations are no longer valid due to shocks forming highly asymmetric density structures . klb demonstrated that the spectral slope of mhd turbulence becomes more shallow with increasing @xmath10 . this can be understood as shocks in supersonic turbulence create more small scale structure in density @xcite . this behavior was found to be weakly dependent of the alfvn mach number . while the spatial power spectrum has long been applied on both observations and simulations , it essentially uses only the amplitude of the fourier transform of the initial signal , while the phase information is totally ignored . the bispectrum , however , is a three point statistical measure which incorporates both the amplitude _ and phase _ of the correlation of signals in fourier space . because of this , it can be used to search for nonlinear wave - wave interactions and characterize how shocks affect turbulent properties of the ism . the bispectrum has been used in cosmology and gravitational wave studies to detect departures from gaussianity @xcite , and for the characterization of wave - wave interactions in laboratory plasmas @xcite . the first application of the bispectrum on synthetic astrophysical mhd turbulence was in bfkl . bfkl found a general correlation between the bispectrum of 2d column density and 3d density maps for simulated data cubes of 512@xmath3 resolution . also , supersonic models showed a much greater degree of correlation between structures of different scales than subsonic models . while comparing models with the same sonic mach number , models with a stronger magnetic field ( i.e. sub - alfvnic ) showed enhanced correlations . these results are encouraging and suggest that the bispectrum can be also used to provide insight into the nature of the turbulence cascade . the small - scale hi structure of the smc was observed with the atca , a radio interferometer , in a mosaicking mode ( staveley - smith et al . observations of the same area were obtained also with the 64-m parkes telescope , in order to map the distribution of large - scale features . both sets of observations were then combined ( see sx99 ) , resulting in the final hi data cube with angular resolution of 98@xmath11 , velocity resolution of 1.65 , and 1-@xmath12 brightness temperature sensitivity of 1.3 k , to the continuous range of spatial scales between 30 pc and 4.4 kpc . the velocity range covered with these observations is 90 - 215 . for details about the atca and parkes observations , data processing , and data combination ( short spacings correction ) see staveley - smith et al . ( 1997 ) and sx99 . the hi column density image is shown in figure [ fig : smc ] . we corrected the original image by multiplying the hi column density of each pixel ( @xmath13 in atom @xmath14 ) with the correction factor @xmath15 derived by sx99 : @xmath16 as the original position - position - velocity ( ppv ) smc cube has @xmath17 pixels , the resulting column density image is a 2d array with @xmath18 pixels . we generate a database of 13 3d numerical simulations of isothermal compressible ( mhd ) turbulence by using the code of klb and varying the input values for the sonic and alfvnic mach number . the sonic mach number is defined as @xmath19 , where is @xmath20 is the local velocity , @xmath21 is the sound speed , and the averaging is done over the whole simulation . similarly , the alfvnic mach number is @xmath22 , where @xmath23 is the alfvnic velocity , @xmath24 is magnetic field and @xmath25 is density . klb used resolution of 128@xmath3 , while we use 512@xmath3 . details about klb s code were published ( see cho et al . 2003 ) and the code was used in several studies . we briefly outline the major points of their numerical setup . the code is a second - order - accurate hybrid essentially nonoscillatory ( eno ) scheme ( see * ? ? ? * ) which solves the ideal mhd equations in a periodic box : @xmath26 = { \bf f } , \\ \frac{\partial { \bf b}}{\partial t } - \nabla \times ( { \bf v } \times{\bf b } ) = 0,\end{aligned}\ ] ] with zero - divergence condition @xmath27 , and an isothermal equation of state @xmath28 , where @xmath29 is the gas pressure . on the right - hand side , the source term @xmath30 is a random large - scale driving force ] . we drive turbulence solenoidally , at wave scale @xmath31 equal to about 2.5 ( 2.5 times smaller than the size of the box ) . this scale defines the injection scale in our models in fourier space to minimize the influence of the forcing on the generation of density structures . density fluctuations are generated later on by the interaction of mhd waves . the time @xmath32 is in units of the large eddy turnover time ( @xmath33 ) and the length in units of @xmath34 , the scale of energy injection . the scale of energy dissipation is defined by the numerical diffusivity of the scheme . the magnetic field consists of the uniform background field and a fluctuating field : @xmath35 . initially @xmath36 . we divided our models into two groups corresponding to sub - alfvnic ( @xmath37 ) and super - alfvnic ( @xmath38 ) turbulence . for each group we computed several models with different values of gas pressure ( see table [ tab : models ] ) . we ran compressible mhd turbulent models , with 512@xmath3 resolution , for @xmath39 crossing times , to guarantee full development of energy cascade . since the saturation level is similar for all models and we solve the isothermal mhd equations , the sonic mach number is fully determined by the value of the isothermal sound speed , which is our control parameter . the models are listed and described in table [ tab : models ] . .description of the simulations - mhd , 512@xmath3 [ tab : models ] [ cols="^,^,^,^,^,^ " , ] for each model , the results of mhd simulations are : the isothermal 3d density field , three components of velocity ( @xmath40 , @xmath41 , @xmath42 ) , and three components of magnetic field . as an example , figure [ fig : fakecloud ] ( left ) shows a density field for a sub - alfvnic , subsonic simulation . to calculate the column density distribution we integrate the 3d density fields along a given los . we introduce the following nomenclature throughout the paper : x column density " or column density in the x direction " refers to the density cube being integrated along the x direction ( parallel to the mean magnetic field ) . this description is similar for the y and z directions ( perpendicular to @xmath43 ) . in the case of the figure [ fig : fakecloud ] ( left ) , the los is along the z axis , and the magnetic field is oriented along the x axis . in order to compare simulations with the smc data , we apply a simple scaling to the simulated column density distributions : @xmath44 where @xmath45 denotes the standard deviation and @xmath46 denotes the arithmetic mean of the column density image . this method , often referred to as the _ standard score _ , standardizes all data sets used in this study . after the scaling , the new data set has values which represent the difference between the original data values and the sample mean , in units of the standard deviation . the same scaling was applied on the smc hi column density image . another consideration to take into account is the fact that simulated cubes are periodic and do not have boundaries . in other words , simulations do not have a decrease in the column density values that is seen in the smc data as one goes radially out from the center of the image . klb showed that boundaries affect higher statistical moments of density , but have little effect on the spatial power spectrum and the bispectrum , generally only impacting the large scales . to introduce cloud boundaries to our simulations , we multiply the simulated 3d density fields by a 3d spherical function , which has a value of one within a sphere of radius @xmath47 and is decaying outside of this radius with a gaussian profile . we choose a @xmath47 value of 128 pixels , which is 4 times smaller then the box size . an example of the simulation with a boundary is shown in the right - hand panel in figure [ fig : fakecloud ] . please note that we impose cloud boundaries on the simulated cubes of homogeneous turbulence after turbulence has been fully developed . this results in the simulated column density profiles being similar to the smc profile , however , is different from simulations of turbulence being developed within cloud boundaries . the most straightforward statistical properties of a distribution are the mean value and the variance . the mean arithmetic value and the variance of the distribution of , for example , column density @xmath48 are given by : @xmath49 and @xmath50 where @xmath51 is the number of samples or points of the mesh in the case of simulation data . the mean value is a less important property in our studies so we do not consider it here ; however , it is required to calculate higher moments . variance measures the width of the distribution and is always positby the third and fourth - order statistical moment . skewness is defined as : @xmath52 if a distribution @xmath48 is gaussian , its skewness is zero . negative skewness indicates the data are skewed in the left direction ( the tail of the pdf is extended to the left , or towards low density values ) , while positive values imply that the distribution is skewed in the right direction ( the tail of the pdf is extended to the right , or towards high density values ) . kurtosis is a measure of whether a quantity has a distribution that is peaked or flattened compared to a gaussian distribution . kurtosis is defined in a similar manner to skewness , only is derived from the forth order statistical moment . if a data set has positive kurtosis then the pdf of data values will have a distinct sharp peak near the mean and elongated tails . if a data set has a negative kurtosis then its pdf will be flat at the mean . kurtosis is defined as : @xmath53 [ fig : skewplot ] variance , skewness and kurtosis of simulated column densities depend on turbulent properties , specifically the sonic mach number . this is shown in figure [ fig : kowal ] for the simulated column density distributions with cloud boundaries . error bars are computed by estimating the standard deviation between models with the same sonic mach number but differing los orientations . all three moments depend almost linearly on the sonic mach number , for @xmath54 . klb noticed a similar trend for supersonic simulations but focused only on @xmath55 . our simulations extend this result all the way to @xmath56 . for @xmath57 klb found that both skewness and kurtosis have a rather flat dependence on the sonic mach number , while the variance continues to decreases gradually for subsonic models . the increase of higher order moments with @xmath10 can be explained in that as the mach number increases , the abundance of small - scale density structure increases , resulting in a broader density pdf . bfkl showed that the increase of higher - order moments with @xmath10 is not a resolution effect , since they replicated the result of klb , ( who used cubes of 128@xmath3 resolution ) with cubes of 512@xmath3 resolution . which one of three high - order moments represents the best statistics to describe turbulence in observational data ? while variance has a linear dependence over a broad range of @xmath10 values , it depends on the exact scaling of the data set , making a direct comparison between simulations and observations difficult . on the other hand , skewness and kurtosis are , by definition , normalized by the standard deviation and do not have this problem . in addition , variance changes with the length along the los , since perturbations can add up in a random walk fashion . all of this , as well as the result by klb that kurtosis is the least affected by cloud boundaries , suggest that for our immediate comparison of observations with the mhd simulations , the higher order moments , skewness and kurtosis , are more appropriate statistics . while it is useful to know the mach number of the ism in a global sense , even more interesting is to see how it varies spatially and whether it correlates with local star forming regions where high turbulence is expected . to characterize small scale departures from gaussian distributions within the column density distributions , we create moment maps of the simulated column densities using a circular moving kernel . we do this by moving a circle of a given radius @xmath58 across the image and calculating the skewness and kurtosis at all points . first we must decide on a kernel size that will enclose enough pixels to provide reliable statistics . according to tabachnick & fidell ( 1996 ) , the standard error in skewness ( ses ) can be estimated by @xmath59 and the standard error in kurtosis ( sek ) can be estimated by @xmath60 , where @xmath51 is the number of points used to calculate the skewness / kurtosis . generally , if a value of skewness / kurtosis falls within @xmath61ses / sek , then the distribution is considered to be normal , while values of skewness / kurtosis larger than twice the absolute value of ses / sek imply significant non - gaussianity . figure [ fig : skewerror ] shows @xmath62 ses and sek as a function of the kernel radius @xmath58 . clearly , ses / sek is proportional to @xmath63 ; the smaller the radius , the higher the absolute value of ses / sek . thus , we select @xmath64 pixels , which corresponds to the point where @xmath62 ses / sek starts to deviate from zero and ensures that gaussianity is represented by @xmath62 ses / sek@xmath65 . using a circular moving kernel we generate 3rd and 4th moment maps of simulated column density distributions . to ensure there are no resolution artifacts , we briefly explore correlations between skewness and kurtosis of the derived moment maps on a pixel - to - pixel basis . based on the work by klb and bfkl , and also our figure [ fig : kowal ] , we expect for supersonic models a correlation of both skewness and kurtosis with the sonic mach number , and that skewness and kurtosis agree in sign and relative value . however , as subsonic models show unexpected behavior at low resolutions ( from the klb study ) , the dependence of skewness and kurtosis on the sonic mach number is not strong . figure [ evidence ] shows pixel - to - pixel comparison between skewness and kurtosis for two example models : a supersonic with @xmath66 ( left panel ) and a subsonic with @xmath67 ( right panels ) . it is evident that the supersonic model shows a good correlation between kurtosis and skewness , while the subsonic model does not . this is exactly as expected . this result proves that our approach of deriving moment maps is not resolution limited although we use a smaller number statistics within each circular kernel ( @xmath68 pixels ) . this also highlights another interesting property : regions of moment maps with a good correlation between skewness and kurtosis could be interpreted by supersonic mhd turbulence , while a poor correlation may be caused by subsonic turbulence . we present in the appendix another technique that could be used to further identify subsonic regions . [ sec : mmaps ] we first calculate the higher order moments of the whole smc hi column density image . we find that the skewness is 1.52@xmath69 , and the kurtosis is @xmath70 . the estimated uncertainties are determined by calculating twice the ses or sek . figure [ fig : skewplot ] suggests that on average the hi in the smc is supersonic with @xmath71 . to characterize small scale departures from the gaussian distribution within the smc we create 3rd and 4th moment maps of the smc using a circular kernel , as discussed for simulations in section [ sec : mom ] . we repeat a similar process for the smc as we did with the simulations . we use a radius of @xmath64 and calculate moments in a circular kernel moving across the smc . this is equivalent to convolving the original hi image with a gaussian function with a fwhm of 30@xmath72 . the resultant moment images therefore have a lower resolution than the original hi image . after this we create a mask which cuts off pixels below a column density of @xmath73 @xmath14 in the hi column density image , to exclude increase in noise along the edges of the image . this is done to eliminate noisy pixels in moment maps , since pixels along the edge have significantly smaller column density values . figures [ fig : skewmap ] shows the distribution of skewness and kurtosis across the smc ( the hi column density image was standardized using the standard score method ) , with overlaid hi contours smoothed to 30@xmath72 resolution . these maps retain the overall shape of the smc , which can be seen in the hi contours . familiar features of the smc , such as the east wing and bar , regions of high star formation , can be picked out in these maps . generally , skewness ranges from @xmath74 to 1 , with several discrete regions reaching higher positive or negative values . for kurtosis , most pixels range from @xmath75 to 1 , with again several exceptions . based on figure 3 , this suggest that large areas in the smc have @xmath76 and are subsonic or transonic . along most of the bar the skewness and kurtosis correlate well . this suggests that the mhd turbulence could be the cause of local departures from gaussianity in the hi column density distribution . we warn once again that both our study and earlier work find strong dependence of skewness and kurtosis on the sonic mach number only for supersonic models , while the dependence for subsonic models is rather weak . interpreting subsonic turbulence is therefore difficult , and poor correlation between skewness and kurtosis could be due to either subsonic turbulence or some other physical mechanisms . several interesting regions stand out in figure [ fig : skewmap ] . * along the hi bar and the eastern wing region both skewness and kurtosis are negative , reaching values even lower than what is seen in figure 3 for global averages . based on expectations from numerical simulations this could be explained by the subsonic isothermal mhd turbulence , as shown in figure 5 . most of the hi bar and the wing therefore could be explained as being subsonic , with several discrete concentrations with almost no turbulence , probably tracing quiescent regions and potential sites of future star formation . however , as subsonic regions are hard to constrain from simulations , additional processes may be also playing an important role . * from hi peaks radially outward , both skewness and kurtosis gradually increase . the highest values of both skewness ( @xmath77 ) and kurtosis ( @xmath78 ) are found along the hi bar and correspond to areas of compressed hi contours . these regions could be interpreted as having the highest level of turbulence ( with @xmath79 ) in the smc . this suggests that the most turbulent regions may be associated with the shearing flows and/or shocks between the bar and the surrounding hi . * sudden change in the behavior of both skewness and kurtosis can be noticed around ra 01@xmath80 10@xmath81 , dec @xmath82 10@xmath72 . for example , skewness flips from high positive to high negative values over an angular scale of @xmath83 . interestingly , this flip happens in the direction of an hi extension towards the lmc and again may be pointing out streaming or tidal motions caused by the interactions between the smc and the lmc . in order to compare the range of values found in the smc and the simulations we plot several histogram of the skewness and kurtosis values in figure [ fig : kurtsmcsim ] . these figures show that the majority of smc pixels generally match well with the mildly supersonic models with @xmath84 . however , the smc distribution appears to be more narrow in terms of both skewness and kurtosis values than the @xmath85 simulation . for subsonic and very supersonic models , the values of skewness and kurtosis of the smc are not in the range of the simulations . as shown above , with our limited angular resolution of 30@xmath72 we see strong hints that different regions in the smc have different turbulent properties , suggesting spatial variations of @xmath10 . we would now like to combine all this information into an image of @xmath10 for the smc . as already discussed , based on the klb study , kurtosis is the least prone to effects caused by cloud boundaries and as figure 3 suggests has an almost linear shape for @xmath86 . therefore , we fit a linear function for all values of kurtosis @xmath87 . this function can be inverted to estimate @xmath10 : @xmath88 . pixels in the @xmath10 image with corresponding kurtosis @xmath89 are set to zero and most likely mark subsonic regions , or could be caused by additional processes . the resultant map is shown in figure [ fig : ms_var ] ( left ) . as expected already , most of the area of the smc bar and the eastern wing could be interpreted as having subsonic or transonic mach numbers , @xmath90 . several concentrated regions across the smc indicate very quiescent environments , most of them unfortunately have a size close to our angular resolution . regions with the highest sonic mach number , @xmath91 , are found around the bar and correspond to compressed hi contours . we can quantify the fraction of hi with different @xmath10 . the most likely quiescent regions ( set to zero in our map ) comprise 8% of the mapped area . regions with @xmath92 comprise about 40% , while about the same fraction is contained in transonic regions with @xmath93 . regions with higher turbulence , @xmath94 constitute about 10% of the area . figure [ fig : ms_var ] ( right ) shows contours of the h@xmath95 emission ( obtained by kennicutt et al . 1995 ) smoothed to 30@xmath72 and overlaid on the @xmath10 map . sites of the most recent star formation in the bar and the eastern wing have @xmath96 , suggesting that the most turbulent regions are not associated with star formation . the most turbulent regions appear to trace shearing and tidal motions . we note though that our resolution of 30@xmath72 is too low to trace individual star - forming regions . as an estimate of the velocity dispersion along the los is often used as a possible measure of hi turbulence , we investigated how the hi velocity dispersion compares to our @xmath10 map . stanimirovic et al . ( 2004 ) found that regions with the highest dispersion ( @xmath97 ) appear to be associated with the positions of the three largest super - giant shells . we find no obvious correlation between the hi velocity dispersion and our @xmath10 map . this lack of correlation could be due to projection effects , and/or large los depth of the smc . in addition , velocity dispersion ( similar to velocity centroids ) is subject to the interplay of several statistical effects ( lazarian & esquivel 2003 , esquivel & lazarian 2005 , esquivel et al . the most obvious is the contribution of both velocities and densities to the resulting measure . more subtle , but still important , are effects of velocity - density correlations and non - gaussianity . these effects make centroids unreliable when dealing with supersonic turbulence ( esquivel et al . effects of phase transitions resulting in the existence of cold and warm hi and the pliable equation of state corresponding to interstellar hi are likely to act in a similar way . thus the velocity dispersions obtained from spectral lines are substantially affected by densities , which are the quantities that we deal in this study . they should be distinguished from the true velocity dispersions , which , unfortunately , are not available through spectral line observations . constraining the sonic mach number of the warm neutral medium ( wnm ) observationally is highly difficult due to the lack of direct measurements of the hi temperature . this is the main reason why in the previous section we used the hi column density image of the smc and a database of numerical simulations to explore spatial variations of @xmath10 . to investigate reliability of our results we compare our derived @xmath10 distribution with results from an observational method which constrains temperature of the cold hi , and therefore @xmath10 of the cnm this method is based on the comparison of the spin temperature ( @xmath98 ) with the upper limit on the kinetic temperature ( @xmath99 ) for cnm clouds seen in hi absorption . for observational data we use hi absorption and emission observations of 29 radio continuum sources behind the smc by dickey et al . ( 2000 ) . for each of the sources @xcite provided the fwhm linewidth of the absorption spectra , and estimated the spin temperature of the cnm seen in absorption . by assuming doppler broadening of hi velocity profiles , we can estimate the upper limit on the kinetic temperature as @xmath100 . as shown in @xcite , the ratio of @xmath99 and the spin temperature , @xmath98 , is related to the 1d mean square turbulent velocity : @xmath101 where @xmath31 is the boltzmann constant and @xmath102 is the mass of the hydrogen atom . multiplying this by 3 gives the mean square 3d turbulence velocity @xmath103 of the cnm . to estimate the sonic mach number we need to divide @xmath104 by the sound speed @xmath105 we adopt a mean atomic weight of @xmath106 for the ism of smc abundance ( see mao et al . 2008 ) . using this , we can write : @xmath107 for each of 29 radio continuum sources from @xcite we derive @xmath99 and use the estimated @xmath98 to calculate the sonic mach number of the cnm clouds . @xcite used three different methods to estimate the spin temperature . as our aim is to compare @xmath0 with values derived from the los integrated hi emission profiles , we use their los averaged spin temperature ( or @xmath108 ) . the median value of the spin temperature is 43 k. a histogram of derived @xmath0 values is shown in figure [ fig : hist ] . the median mach number for the whole sample of 29 sources is @xmath109 , and the histogram peaks around @xmath110 . this suggests that the internal cnm macroscopic motions are highly supersonic . to compare these values with the ones derived using the method of higher - order statistical moments we show a histogram of data points from the @xmath0 map in the right panel of figure [ fig : hist ] . while the observed histogram suffers from the low number statistics , due to a small number of suitable ( strong ) sources , the two histograms have a similar shape : almost a gaussian central distribution with a tail at higher mach values obviously , the observationally inferred @xmath0 values suggest at least a factor of 2 higher level of turbulence across the smc than what we derived using higher statistical moments . this is not surprising as the observed values trace predominately the cnm , while the @xmath0 map was derived using the hi column density image and therefore is affected by both cnm and wnm . we discuss this further in [ sec : discussion ] . in any case , to sample better cnm turbulence across the smc future sensitive hi absorption observations of continuum sources behind the smc are clearly needed . the slope of the fourier transform of the two point autocorrelation function , or the spatial power spectrum , also provides information on important properties of turbulence flows . in particular , the slope of the power spectrum is known to depend on the sonic mach number . with a database of numerical simulations we explore whether the slope of the spatial power spectrum depends also on the alfvn mach number . the alfvn mach number dependence would be especially interesting as we could use higher - order statistical moments to estimate the sonic mach number , and from there the alfvn mach number and the strength of the magnetic field . the power spectrum is defined as : @xmath111 where @xmath31 is the wavenumber and @xmath112 is the fourier transform . stanimirovic & lazarian ( 2001 ) estimated the power - law slope of @xmath6 for the spatial power spectrum of the hi column density image of the smc . we measure the power spectrum slope for various simulated column densities . figure [ fig : power ] shows how the power spectral slope changes with the sonic mach number . the slope is increasingly shallow for supersonic models and levels off for very high mach number turbulence . this is expected as higher mach number turbulence has more density irregularities and more power on small scales . we have shown in this figure slopes for two values of @xmath113 : 0.7 ( dashed line ) and 2.0 ( dotted line ) . for both strong and weak magnetic fields , the power spectrum slope increases with @xmath10 , however this increase is steeper for super - alfvenic and subsonic turbulence . for all cases the error bars were calculated from the standard deviation in the power spectral slope derived from column density images for three different los orientations . the smc has a power spectrum slope of @xmath6 and spatially mostly @xmath114 . based on the difference between slopes expected for sub - alfvnic and super - alfvnic turbulence in figure [ fig : power ] , it is likely that the super - alfvnic description fits better the slope of the smc . the bispectrum is closely related to the power spectrum . in a discrete system , the power spectrum is defined by equation ( 11 ) . in a similar way , the bispectrum is defined as : @xmath115 where @xmath116 and @xmath117 are the wave numbers of two interacting waves , and @xmath112 is the original discrete time series data with finite number of elements with @xmath118 representing the complex conjugate of @xmath112 . the bispectrum is a complex function which measures both phase and magnitude information between different wave modes . as this is the first application of the bispectrum on the hi data we show in figure [ fig : example ] a visual example of the difference between the power spectrum and the bispectrum this figure shows the original hi column density image of the smc ( top left ) , and the same image but with manipulated phases ( top right ) . to obtain the top right image we fourier transformed the smc image and randomized the phases with a gaussian random distribution but left the _ amplitudes _ the same . even though the phases of the top images are very different , the power spectrum uses only amplitude information and is identical for both images , as shown in the bottom panels of figure [ fig : example ] . however , the bispectrum looks very different for the two images , as expected , and offers an insight into the phase information . in practice , our calculation of the bispectrum involves performing a fast fourier transform ( fft ) of the column density functions and the application of equation [ eq : bispectra ] . we randomly choose wavevectors and their directions , @xmath119 and @xmath120 and iterate over them , calculating @xmath121 . we limit the maximum length of the wave vectors to half of the box size . we normalize direction vectors to unity , calculate positions in fourier space , and finally , compute the bispectrum , which yields a complex 2d array . when plotting the bispectrum ( figures 12 - 15 ) we plot bispectral amplitudes , which give information about the degree of mode correlations in the original column density distribution . in order to interpret the bispectrum of the smc hi column density we first compute the bispectrum of the mhd simulations ( scaled by the standard score method and without applying cloud boundaries ) and look for dependence of wave - wave correlations on the sonic and alfvn mach numbers . figure [ fig : bispsimms ] shows the bispectrum of simulated column density distributions for the case of fixed @xmath122 and two extreme cases of @xmath10 : 0.7 and 7.0 . in figure [ fig : bispsimma ] we fix @xmath123 and look at two cases of @xmath113 : 0.7 and 2.0 . the first thing to notice in the bispectral contour maps is that the @xmath124 case always shows high correlation since this is a trivial case of two wave numbers being the same . for @xmath125 the bispectral amplitude and isocontour shape vary with turbulent properties but generally amplitude decreases gradually radially from @xmath126 . models with circularly / broad - shaped isocontours have high amplitudes and wave - wave correlations , while models with more narrow isocontours have lower amplitudes and therefore weaker correlations . for a fixed @xmath113 , supersonic models show a higher degree of wave - wave correlations over subsonic models . for a fixed @xmath10 , models with a higher magnetic field ( sub - alfvnic , e.g. @xmath1270.7 ) show somewhat stronger correlations then the models with a weaker magnetic field ( super - alfvnic ) although this difference is not striking . the bispectrum of the smc hi column density is shown in figure [ fig : bispectraglobal ] . the top row ( first column ) shows the bispectrum of the column density derived by integrating all 78 velocity channels . the middle and bottom rows show the bispectrum of the first ( for channels 1 to 42 ) and the last half ( for channels 43 to 78 ) of the data cube . to facilitate interpretation the wavenumbers @xmath128 and @xmath129 are shown in terms of linear size kpc corresponds to the largest angular scale covered in the original image used to calculate the bispectrum , @xmath130 . we scale then all wavenumbers @xmath31 by @xmath131 to express wavenumbers in terms of physical linear size at the distance of the smc ( 60 kpc ) . ] . while there are small differences in the bispectral amplitudes , there is essentially no significant difference in the contour shape for the three bispectra . to properly compare the smc bispectrum with simulations we apply a _ windowing function on the smc column density image to simulate the effect of periodic boundaries . we use the hanning window @xmath132 function , defined as : @xmath133 where @xmath51 is the number of pixels along the x or y axis , and @xmath134 ranges from 0 to @xmath135 . this windowing function makes the map periodic in the fourier space . the second column of figure [ fig : bispectraglobal ] shows the resultant bispectra for all three cases of hi column density integrations . the windowing seems to slightly change bispectral amplitudes on the large scales , but leaves the general structure of the bispectral contours intact , although some smoothing effects are observed . the increase in large scale correlations is at a level of @xmath136% . before interpreting the smc bispectrum we briefly investigate how the presence of noise in the observational data affects the bispectrum . we made the noise image of the hi column density by taking a single velocity channel from the smc hi data cube without hi emission and integrating this image over the whole velocity range . the bispectrum of the noise image ( normalized using the standard score method ) is shown in figure [ fig : noise ] ( right ) . in the case of purely gaussian noise , we expect a random distribution over the whole space of wavelengths . as a result , the bispectrum will have a gradual increase in amplitude , with the highest amplitude being at large @xmath31 ( or small scale ) . figure [ fig : noise ] ( left ) illustrates this effect for a pure gaussian distribution . the bispectrum of the smc noise image shows essentially the same trend , suggesting that we are dealing mainly with gaussian ( random ) noise . as is obvious from figures 12 - 14 , the bispectral amplitude increases in the opposite direction ( from large to small scales ) . the effect of noise is therefore the most important at the smallest spatial scales , and we keep this in mind when interpreting the bispectrum . figure [ fig : bispectraglobal ] shows the degree of wave - wave correlations in the hi distribution . modes are obviously strongly correlated at the largest scales and show weaker correlation at small scale . this trend is similar to what was found for various mhd simulations , and obviously very different from the case of gaussian noise . in addition , the @xmath137 line shows the strongest correlations , as is also seen in the simulations . however , contrary to simulations where the level of correlation gradually decreases along this line , we see significant small - scale variations . several local peaks are noticeable and a strong break at mid - scales . the break occurs around @xmath138 pc and is seen in all three integrations of the hi data cube . this could be due to a lack of interactions between turbulent eddies at the scale of @xmath2 pc , caused possibly by numerous expanding shells in the smc . hatzidimitriou et al . ( 2005 ) and stanimirovic ( 2007 ) showed that shells in the smc range in size from 30 to 800 kpc , however the radius distribution peaks at @xmath139 pc . the break in the bispectrum may be signifying the lack of correlations on scales close to the typical shell diameter , or some additional physical processes . we can also compare the bispectrum of the smc with the bispectra of various simulations . by visual inspection of figures 12 - 13 , the closest simulation to the smc in terms of contour shapes is the one for @xmath140 and @xmath141 . further work in this area is required to derive a real measure to quantitatively compare observed and simulated bispectra . using 3rd and 4th statistical moments of the hi column density image and boothstreping turbulent information from a database of isothermal mhd simulations , we have mapped spatial variations of the sonic mach number across the smc . while most of the hi seen in emission in the smc appears to be subsonic or transonic , several supersonic regions have emerged from our study . it is interesting that these regions do not correlate well with the most recent sites of star formation and seem to point out to large scale shearing or tidal flows . commonly , it is believed that supernovae and superbubbles are the main drivers of galactic turbulence ( mccray & snow 1979 ) , with a typical size of @xmath142 pc . while we do not have high enough resolution to see changes on such small scales ( @xmath143 ) in our derived map of @xmath0 , most of the star - forming bar of the smc appears to have subsonic or transonic properties when viewed at resolution of 30@xmath72 . the most turbulent regions in the smc may be tracing some kind of shearing flows between the smc bar and the surrounding hi . this suggests that smc s chaotic history with the lmc and our own milky way has probably left strong turbulent imprints on the hi gas . the lack of a turnover in the hi spatial power spectrum on the largest observed scales is also indicative of the fact that turbulent energy injection happens largely on scales larger than the size of the smc @xcite . similarly , goldman ( 2000 ) suggested that the hi turbulence in the smc was induced by large - scale flows from tidal interactions with the milky way and the lmc about @xmath144 yrs ago . such large - scale bulk flows could have generated turbulence through shear instabilities , and this turbulence has not have had enough time to decay . most of the hi in the smc has a sonic mach number of 1 - 2 . this is on average at least two times smaller than what we inferred from hi absorption observations for the cnm in the smc , @xmath1453.5 - 4 . similarly , for the cnm in the milky way heiles & troland ( 2003 ) found @xmath146 with a large dispersion . a sonic mach number of about 4 - 5 is commonly assumed for cold gas in molecular clouds ( federrath et al . 2009 ) . for example , heyer et al . ( 2006 ) measured from co observations @xmath147 for the rosette molecular cloud , and @xmath148 for g216 - 2.5 . on the other hand , hill et al . ( 2008 ) found that the distribution of the warm ionized medium ( wim ) in the milky way can be best fit by models for mildly supersonic turbulence with @xmath149 . turbulent properties of the hi in emission in the smc are therefore closer to properties of the wim in the milky way than properties of the cnm . this may suggest a large fraction of warm relative to cold hi being traced in hi emission . in the milky way , the hi is known to consist of at least two components with different temperature : the wnm with @xmath150 k and the cnm with @xmath151 k. in addition , there could be a substantial amount of gas at intermediate temperatures @xcite . due to its lower metallicity , the hi in the smc has different properties . dickey et al . ( 2000 ) found @xmath152 k , in agreement with theoretical expectations by wolfire et al . ( 1995 ) whereby the existence of the two - phase medium is possible only at higher pressures compared with the range that applies for solar neighborhood conditions . they also estimated the fraction of cold hi in the smc to be @xmath153% . this is lower than @xmath154% found for the milky way . as our simulations are isothermal it is obvious to wonder how does the multi - phase hi affect our statistics and conclusions . we investigate this in figure [ fig : ex ] by producing a simulated data cube from a weighted combination of two cubes , one subsonic ( @xmath10=0.7 ) and one supersonic . the subsonic cube represents contribution from warmer gas , while the supersonic cube represents the cold gas . we combined the two cubes with different emphasis on warm vs cold gas , obtained the column density image of the resultant cube , and calculated its moments . figure [ fig : ex ] shows that for the case when the supersonic cube has @xmath10=2.0 skewness of the final cube with up to 50% of subsonic gas will be biased towards supersonic gas and will appear dominated by cold gas . if we increase the sonic mach number of the supersonic cube to 4.0 , the dominance of the higher turbulence is even more pronounced . a cube with up to 60 - 70% of subsonic gas and 25% of supersonic gas , will still have high skewness biased by the supersonic contribution . considering that the hi column density image results in the mean @xmath93 , significantly lower than what is expected for the cold hi , the fraction of the cnm along any los is most likely @xmath155% . this supports the dickey et al . ( 2000 ) estimate of the fraction of cold hi in the smc being about 15% . two different statistical approaches in our study suggest that the hi gas in the smc seen in emission is super - alfvnic . as we have shown in figure [ fig : power ] , in addition to the sonic mach number the spectral slope of the spatial power spectrum is sensitive to the alfvenic mach number for @xmath156 . the sub - alfvnic models generally show steeper slopes due to large scale influence of magnetic fields . thus , if one independently knows the sonic mach number , it is possible to estimate the alfvenic one using just the column density data . while the dependence of the spectral slope on the alfvenic mach number has not received much attention in the past , it is somewhat expected . essentially , magnetization decreases compression in the shocks . strong magnetic forces mix up density clumps preventing formation of isolated peaks , which results in a steeper spectrum . in addition , in the sub - alfvnic case we expect oblique shocks to be disrupted by alfvn shearing , which in turn , produces more small scale shocks @xcite . another indication that the hi in the smc is super - alfvnic comes from the bispectrum . the very sharp decrease in the bispectral amplitudes from large to small scales observed for the smc is the closest to the trend found for simulated data for the case of @xmath85 and @xmath157 . detailed comparison between simulated and observed bispectra awaits future work , however this qualitative comparison is certainly encouraging . assuming on average @xmath84 , the power spectrum slope suggests a super - alfvnic hi in the smc with @xmath158 . this is generally in agreement with the observationally inferred strength of the magnetic field by mao et al . ( 2008 ) . using their estimate for @xmath159 @xmath160 g , a radius of the smc of 2 kpc , the total hydrogen mass of @xmath161 m@xmath162 , and a typical velocity dispersion of 20 ( stanimirovic et al . 2004 ) , we estimate @xmath163 . as the alfvenic mach number shows the nature of the interplay between gas pressure and magnetic fields , it appears that the gas pressure in the smc dominates over the magnetic pressure . our bispectrum analysis of the smc hi data was the first attempt to apply bispectrum on observed astrophysical data . while more detailed comparison between observations and simulations awaits future work , we clearly see trends in the bispectral amplitudes similar to what was found for simulations of supersonic mhd turbulence . the most interesting finding is , however , the effect of small - scale variations in the @xmath124 correlations and a strong break in correlations at a scale of 160 pc . such small - scale variations , or jumps , have not been seen in the bispectrum of simulated data . we can speculate about several possible scenarios that could explain their existence . the jumps could be caused by the energy injection due to processes other than turbulence affecting specific spatial scales . alternatively , the jumps may be marking the presence of colder or multi - phase gas . similarly , the observed break in the bispectrum at about 160 pc is intriguing . as we already pointed out , it is interesting that most expanding shells in the smc ( more than 500 were cataloged so far ) have a diameter of @xmath164 pc . the break could be due to the lack of correlations on scales similar to the distance between two shell centers . obviously this will require further studies . a natural question to ask is how results presented in this paper depend on the resolution of numerical simulations . for example , kritsuk et al . ( 2007 ) investigated how resolution of numerical simulations affects the power spectrum of density . these authors found that the spectral index estimates based on low resolution simulations bear large uncertainties due to the bottle neck contamination , and that the power spectra of @xmath165 simulations are substantially shallower then models with resolution of @xmath166 . however , while kritsuk et al . ( 2007 ) only examined hydrodynamic turbulence , beresnyak et al . ( 2008 ) showed that the slopes were very different between the mhd and pure hydrodynamic cases . for instance , the slopes for hydrodynamic simulations showed a pronounced and well defined bottleneck effect , while the mhd slopes were much less affected . this is indicative of mhd turbulence being less local than the kolmogorov turbulence , and suggests that our simulations will be less affected by resolution . in addition , kritsuk et al . ( 2007 ) found a difference in the slope between hydrodynamic @xmath165 and @xmath166 simulations to be 0.17 . this would result in a change of @xmath167 only and will not change our interpretation . we also add that in the case of higher statistical moments and the bispectrum bfkl confirmed trends noticed by klb at lower resolution of 128@xmath3 . another issue that should be further addressed and that could affect our results is the type of numerical forcing of turbulence . federrath et al . ( 2009 ) recently investigated the effects of solenoidal vs. compressive ( divergence - free vs. curl - free ) forcing on a variety of statistics including pdfs and higher order moments . they found that both types of driving mechanisms are compatible with observations of molecular clouds however , depending on the data studied , one type could be superior then the other in terms of the statistics and reproduced observables . this implies that different regions in the smc may exhibit statistical signatures of either compressive or solenoidally driven turbulence . we have investigated a new method for constraining turbulent properties of the ism , specifically the sonic mach number , by using the hi column density image and a database of numerical simulations with a range of sonic and alfvenic mach numbers . by applying the 3rd and 4th statistical moments on both observed and simulated data we have derived the spatial distribution of the sonic mach number across the smc with angular resolution of 30@xmath72 . to provide an estimate of the alfvenic mach number we used two approaches : the spatial power spectrum and the bispectrum . using the database of numerical simulations we have shown that the spatial power spectrum varies with both the sonic and alfvenic mach numbers . if the sonic number is known the alfvenic number can be constrained from this dependence . the bispectrum shows the level of correlation between turbulent eddies of different size and depends greatly on the sonic mach number , and somewhat on the alfvenic mach number . by comparing the bispectra of observations and simulations we have gauged the importance of magnetic fields relative to the gas pressure in the smc . the following results were discussed in the paper . * skewness and kurtosis of the hi column density generally correlate well and are within the range expected from mhd simulations . this suggests that departures from gaussianity could be interpreted as being governed by mhd turbulence . * most of the hi in the smc bar and the eastern wing is subsonic or transonic with @xmath90 . sites of most recent star formation have @xmath96 . regions with the highest skewness and kurtosis , which could be interpreted as having @xmath91 , correspond to the edges of the bar . the most turbulent regions are most likely tracing tidal or shearing flows . the fraction of the smc with different turbulent properties is : 10% with @xmath94 , 80% with @xmath168 , and about 10% with very low values of @xmath10 . * using hi absorption profiles from dickey et al . ( 2000 ) we have estimated that the cnm in the smc is highly supersonic with @xmath169 . this is at least a factor of two higher than what we measured from the higher statistical moments for the hi gas seen in emission . one possible reason for this discrepancy could be that hi emission is dominated by warm gas and the fraction of the cnm in the smc is @xmath170% . * the slope of the spatial power spectrum and the bispectrum suggest that the hi in the smc is super - alfvenic with @xmath158 . this is implies that the gas pressure dominates over the magnetic pressure . * the bispectrum of the hi column density shows large scale wave correlations suggesting a large scale energy injection mechanism . contrary to simulations which show a smooth decrease of wave - wave correlations from large to small scales , the smc bispectrum shows localized enhancements of correlations and at least one prominent break at @xmath2 pc . we speculate that the multi - phase medium , and/or energy injection by processes other than turbulence , could be responsible for the correlation jumps . the break on the other hand appears at a scale similar to the diameter of the majority of expanding shells in the smc . b.b is thankful for valuable discussions with diego falceta - gonalves and jungyeon cho . bb acknowledges the nasa wisconsin space grant consortium and the national science foundation graduate research fellowship . al acknowledges nsf grant ast 0808118 and the center for magnetic self - organization . ss acknowledges support from the nsf grant ast 0707679 and the research corporation . in figure [ evidence2 ] we plot kurtosis vs. skewness in a manner similar to figure [ evidence ] . the difference between this figure and figure [ evidence ] is that we added seven zero points into the beam ( instead of 3841 points there are now 3848 points ) . this produces almost no change in the supersonic models since they already have high positive skewness and kurtosis . supersonic models still show strong correlation between skewness and kurtosis . however a big change is seen in the subsonic skewness vs. kurtosis . the additional zero points shift the distribution from gaussian ( the mean of our simulations with no scaling applied is unity ) to negative skewness and very peaked kurtosis producing a tight anti - correlation . because this technique only strongly affects subsonic areas we can use it to locate subsonic turbulence in the smc with skewness and kurtosis by looking for anti - correlation and very high values . these properties also hold for simulations with cloud boundaries imposed . note that one must use caution here and carefully examine the distribution of data . the application of this technique to the smc data is shown in figure [ fig : compare ] . this plot shows the skewness and kurtosis maps of the smc side by side with the modified beam . indeed , most regions are unchanged from the analysis in section [ sec : mmaps ] . however a few regions stick out with signatures that are subsonic , that is , anti - correlation between skewness and kurtosis . two regions of the highest kurtosis are located in the area between the hi bar and the eastern wing . while kurtosis reaches values of 4 - 8 , the corresponding skewness values are negative , @xmath171 . again , such combination of skewness and kurtosis may correspond to simulations of subsonic isothermal mhd turbulence , and/or points out additional processes at work . beresnyak , a. , lazarian , a. & cho , j. , 2005 , , 624 , l93 biskamp , d. ( 2003 ) , magnetohydrodynamical turbulence , ( cambridge university press , cambridge ) . brunt , c. , & heyer , m.,2002,apj , 566 , 27 burkert , a. , 2006 , c. r. physique 7 burkhart , b. falceta - goncalves , d. , kowal , g. , lazarian , a. , 2009 , , 693 , 250 ( bfkl ) cho , j. & lazarian , a. 2002 , phys . lett . , 88 , 5001 cho , j. & lazarian , a. , 2003 , , 345 , 325 chepurnov , a. , & lazarian , a. , 2009 , apj , 693 , 1074 chepurnov , a. , lazarian , a. , gordon , j. , & stanimirovic . , s. , apj , 688 , 1021 croviser , j. , & dickey , m. , 1983 , a&a , 122 , 282 deshpande et al.,2000 , apj , 543 , 227 dickey et al . , 2000 , apj , 536 , 756 elmegreen , b. , & scalo , j. , ara&a , 42 , 211 esquivel , a. , & lazarian , a. , 2005 , apj , 631 , 320 esquivel , a. , & lazarian , a. , 2009 , apj in press esquivel , a. , lazarian , a. , pogosyan , d. , & cho , j. , 2003 , mnras , 342 , 325 esquivel et al . , 2007 , mnras , 381 , 1733 federrath , c. , et al.,2009 , aa , in press fry , j. n. 1998 , annals of the new york academy of sciences 848 , 62 gill , a.g . , & henriksen , r.n . , 1990 , , 365 , l27 goodman , a. et al.,2009 , apj , 692 , 91 hatzidimitriou et al . , 2005 , mnras , 360,1771 heiles , c. , troland , t. , 2003 , apj , 586 , 1067 heyer , m. , & zwiebel , e. , 2004 , ap&ss , 292 , 9 hill , a. s. , benjamin , r. a. , kowal , g. , reynolds , r. j. , haffner , . l. m. & lazarian , a. , 2008 , apjs intrator , t. , meassick , s. , browning , j. , majeski , r. & hershkowitz , n. 1989 , phys . fluids b , 1 , 271 - 273 kennicutt et al . 1995 , aj , 109 , 594 kowal , g. , lazarian , a. & beresnyak , a. , 2007 , , 658 , 423 ( klb ) kowal et al . , 2009 , apj , 700,63 kritsuk et al . , 2007 , apj , 665 , 416 lazarian , a. , 2007 , jqsrt , 106 , 225 lazarian , a. , 2009 , ssr , 143 , 357 lazarian et al.,2001 , apj , 555 , 130 lazarian , a. & pogosyan , d. , 2004 , , 616 , 943 lazarian , a . & pogosyan , d . , 2006 , , 652 , 1348 lazarian , a. & pogosyan , d . , 2000 , , 537 , 720 lazarian , a. , & vishniac , e. , 1999 , apj , 511 , 193 levy , d. , puppo , g. & russo , g. , 1999 , mathematical modeling and numerical analysis , 33 , 547 liguori , m. , hansen , f. k. , komatsu , e. , matarrese , s. & riotto , a. , 2006 , phrvd , 73 , 3505 lithwick , y. & goldreich , p. , 2001 , , 562 , 279 liu , x .- d . & osher , s. , 1998 , journal of computational physics , 141 , 1 mao et al . 2009 , apj , 688 , 1029 mccray , r. , & snow , p , 1979 , a&aa , 17 , 213 mckee , c. , ostriker , e. , 2007 , monin , a.s . & yaglom , a.m.,1967 , statistical fluid mechanics , mit press montgomery , d. , brown , m. r. , matthaeus , w. h. , 1987 , j. geophys . , 92 , 282 nordlund , a. k. & padoan , p. , 1999 , interstellar turbulence , proceedings of the 2nd guillermo haro conference , cambridge university press ostriker , e. c. , stone , j. m. & gammie , c. f. , 2001 , apj , 546 , 980 scoccimarro , r. , 2000 , , 544 , 597 stanimirovic , s. , staveley - smith , l. , dickey , j. m. , sault , r. j. , & snowden , s . l. , 1999 , mnras , 302 , 417 ( sx99 ) stanimirovic , s. & lazarian , a. , 2001 , apj , 551 , l53 stanimirovic , s. , 2001 , _ astrophysics and space science _ , 277 , 87 staveley - smith , l. , sault , r. j. , hatzidimitriou , d. , kesteven , m. j. , & mcconnell , d. 1997 , mnras , 289 , 225 stanimirovi , s. , staveley - smith , l. , & jones , p. a. 2004 , apj , 604 , 176 stanimirovi , s.,2007 , iau symposium , 237 , 84 stutzki , j. , bensch , f. , heithausen , a. , ossenkopf , v. , & zielinsky , m. , 1998 , , 336 , 697 tabachnick , b. g. , & fidell , l. s. ( 1996 ) . using multivariate statistics ( 3rd ed . ) . new york : harper collins . tynan , g. r. , moyer , r. a. , burlin , j. & holland , c. 2001 , phys . plasmas , 8 , 2691 westerlund , b. e. 1991 , in the magellanic clouds , iau symp . , 148 , 15

we calculate the properties of the two - band hubbard model using the dynamical cluster approximation . the phase diagram resembles the generic phase diagram of the cuprates , showing a strong asymmetry with respect to electron and hole doped regimes , in agreement with experiment . asymmetric features are also seen in one - particle spectral functions and in the charge , spin and d - wave pairing susceptibility functions . we address the possible reduction of the two - band model to a low - energy single - band one , as it was suggested by zhang and rice . comparing the two - band hubbard model properties with the single - band hubbard model ones , we have found similar low - energy physics provided that the next - nearest - neighbor hopping term has a significant value( ) . the parameter is the main culprit for the electron - hole asymmetry . however , a significant value of can not be provided in a strict zhang and rice picture where the extra holes added into the system bind to the existing cu holes forming local singlets . we notice that by considering approximate singlet states , such as plaquette ones , reasonable values of , which capture qualitatively the physics of the two - band model can be obtained . we conclude that a single - band t - t-u hubbard model captures the basic physics of the cuprates concerning superconductivity , antiferromagnetism , pseudogap and electron - hole asymmetry , but is not suitable for a quantitative analysis or to describe physical properties involving energy scales larger than about 0.5 ev .

cond-mat0411092

the theory of the cuprate high temperature superconductors remains one of the most important and daunting problems in condensed matter physics . the high @xmath2 cuprate superconductors are layered materials with relatively complex structures and chemical composition . they are highly correlated , with an effective bandwidth roughly equal to the effective local coulomb interaction . the short - range correlations are known to play a paramount role in these materials . therefore , the dynamical cluster approximation ( dca ) @xcite , which treats short - range correlations explicitly and the long - range physics at the mean - field level , is an ideal tool for the investigation of these systems . a common characteristic all the cuprate materials share is the presence of quasi - two - dimensional @xmath3 planes . these planes are commonly believed to contain the low - energy physics . however , the full complexity of the orbital chemistry of just the @xmath4 planes and the strong coulomb repulsion on the cu ions would lead to models which are very difficult to study with conventional techniques . the cuprates are characterized by a very rich , but also in many respects very intriguing physics . the undoped materials are antiferromagnetic ( afm ) insulators with a gap of approximatively @xmath5 . upon doping the afm is destroyed and the system becomes superconducting ( sc ) . at small doping , in the proximity of the afm phase , the normal state physics can not be described in terms of fermi liquid theory and is characterized by the presence of a pseudogap . an essential demand of every successful theory is to capture all these fundamental features at the same time . experimental data show that the phase diagram and other physical characteristics like the density of states ( dos ) near the fermi level of the hole - doped and electron - doped materials are very different @xcite . there could be many reasons for this asymmetry . the electron and the hole doped materials are physically different and apart from the @xmath4 planes , they contain different elements and chemical structures . these structural and compositional differences can influence the low - energy physics . therefore in this paper , we use dca to address whether the physics of a pure @xmath4 plane contains this asymmetry or if the origin of the asymmetry in real materials comes from other influences . different models for describing the physics of a @xmath3 plane were proposed by various authors . photoemission experiments in the insulating parent material show that the first electron - removal states have primarily oxygen character ; whereas , the first electron - addition states have @xmath6 character , already suggesting a strong asymmetry . this places these materials in the charge - transfer gap region of the zaanen - sawatzky - allen scheme @xcite . early on , considering the ligand field symmetry and band structure calculations @xcite , it was realized that the most important degrees of freedom are the cu @xmath7 which couple with the in - plane o @xmath8 orbitals . therefore , one of the first models proposed to describe the physics of high @xmath2 materials was the so called _ three - band hubbard model _ introduced by varma _ et al . _ @xcite and emery _ et al . _ @xcite , which considers explicitly both the oxygen @xmath9 and the cooper @xmath7 orbitals . in fact , because the direct oxygen - oxygen hopping is neglected , only the combination of oxygen orbitals with @xmath10 symmetry couples with the @xmath6 orbitals , and the above proposed three - band model reduces to a two - band one . however zhang and rice ( zr ) @xcite argue that the low - energy physics of the _ hole doped _ superconductors can be described by a single - band model . starting from the two - band model , zhang and rice claim that an extra hole added into the oxygen band binds strongly with a hole on the @xmath11 , forming an on - site singlet . this singlet state , which has zero spin can be thought as moving through the lattice like a hole in an antiferromagnetic background . consequently , the physics can be described by a one - band t - j model . pertinent criticism to these simplified models were raised by various authors . with respect to cu degrees of freedom , h. eskes _ et al . _ @xcite stressed the possible importance of the other @xmath6 orbitals , showing that they should be explicitly considered when physics which implies excitations with energy larger than @xmath12 is involved . however , these criticisms do not concern us for the present study since we are interested only in physics at energies lower than @xmath13 . investigating the relative importance of various parameters describing the @xmath3 planes it was realized early on that in addition to the cu on - site coulomb repulsion ( @xmath14 ) and cu - o hopping integral ( @xmath15 ) the o - o hopping integrals result in a large o @xmath16 band width ( @xmath17 ) , indicating that these should be included explicitly in any theory @xcite therefore , using the dca technique as a means of including all these most important parameters and bands , we address two major problems in this paper : the physics of the @xmath3 plane ( including a detailed study of the electron - hole asymmetry ) and the reduction of the multi - band model to a single - band model . regarding the reduction to a one - band model , one of the most serious criticisms to zr theory is the neglect of the o @xmath16 band structure @xcite . the natural tendency of the finite oxygen bandwidth is to delocalize and destabilize the zr singlets . the question arises whether the low - energy states ( i.e. the zr singlets ) are still well separated from the higher energy states ( i.e. the non - bonding oxygen states ) . otherwise , the reduction to a single - band model which neglects these high energy states is not possible . this problem was previously considered by eskes and sawatzky @xcite within an impurity calculation approach , but there , unlike in the dca approach , both the spatial correlation effects and the dispersion of the low - energy states were neglected . another important objection to zr theory was first raised by emery and reiter @xcite , and regards the nature of the low - energy states . are these states real singlets which can be mapped onto holes , or does the hole on the o bind into a more complicated state which involves more than one cu hole ? choosing a particular solvable example , which considers the cu spins arranged ferromagnetically , they showed that the low - energy states are in fact an admixture of the zhang - rice singlets and the corresponding triplets . this implies a nonzero value for the oxygen spin , and destroys the equivalence of these states to holes . however , it is not clear if the situation is similar in the cuprates , i.e. if the zr singlet - zr triplet admixture is significant . but the merit of emery and reiter is to emphasize that the fact that , as a consequence of the strong cu - o hybridization low - energy states well separated from the non - bonding oxygen band states appear , does not necessarily mean that the physics can be reduced to a single - band model . the third problem we address regarding the reduction to a single - band model is the estimation of the single - band parameters . we notice that different approximations result in different values of the parameters . especially the magnitude of the next - nearest - neighbor hopping is very dependent of the initial assumptions . for example , if we assume that the hole addition low - energy states are genuine zr singlets , i.e bound states between a cu hole and a orthogonal wannier oxygen orbital , we obtain a negligible next - nearest - neighbor hopping @xcite . on the other hand , if we consider the low - energy states to be plaquette singlets , i.e. bound states between a cu hole and a hole on the state formed by the four oxygen around the cu , the value of the next - nearest - neighbor hopping is significant @xcite . of course , because of the nonorthogonality of the plaquette states , the plaquette singlets are not genuine singlets and therefore they can not be rigorously mapped into holes . however , because their overlap with the local singlets is large ( @xmath18 ) @xcite , it is still possible that this approximation is good . our calculations show that a multi - band model and a single - band t - t-u hubbard model with a significant value of the next - nearest - neighbor hopping exhibit a similar low - energy physics . the essential parameter needed for the agreement is the next - nearest - neighbor hopping , @xmath0 . this parameter is also the main culprit for the observed electron - hole asymmetry . however , as mentioned above , the large value of @xmath0 can not be obtained in a strict zr picture . thus our results also implicitly indicate that the multi - band model can not be rigorously reduced to a single - band model . therefore , besides showing the similarities between the two models , we also point out their significant differences in this paper . the final conclusion is that a single - band t - t-u hubbard model , with a significant value of @xmath0 , captures the basic physics of the cuprates , and thus is suitable to describe the afm , pseudogap and sc physics together with the relevant asymmetries observed in the phase diagram , in the one - particle spectra and in the two - particle response functions . however , we believe that it is not suitable for a quantitative material specific analysis , for describing the higher energy spectroscopic features as in optical spectroscopy or resonant inelastic x - ray scattering , or for studying more subtle features related to the finite value of the spin on the oxygen . this paper is organized as follows . in section [ sec : form ] the two - band hubbard model and the dca technique is introduced . our two - band model takes fully into account the oxygen dispersion and considers only the oxygen degrees of freedom which couple directly to the cu @xmath7 orbitals . the results of the dca calculation applied to the two - band hamiltonian are presented in section [ sec:2bm ] . the possible reduction of the two - band model to a single - band one , together with a detailed analysis of the single - band t - t-u hubbard model , is addressed in section [ sec:1bm ] . a discussion regarding the similarities and the differences between the two - band model and the single - band one is given in section [ sec : disc ] . the conclusions of our study are reviewed in section [ sec : conc ] . band structure calculations @xcite , cluster calculation @xcite , photoemission @xcite and other experiments show that the relevant cu degrees of freedom are the @xmath7 orbitals which couple with the in - plane @xmath19 and @xmath20 o orbitals . all these degrees of freedom result in a five - band ( four oxygen and one copper band ) hamiltonian in general . we have studied the five - band model in detail @xcite and have found that due to the strong cu - o hybridization , only the oxygen degrees of freedom which couple directly with cu are relevant for the low - energy physics . consequently , to a very good approximation , the five - band model can be reduced to a two - band one . the two - band model contains one cu @xmath7 correlated band and one oxygen band . at every site the oxygen states are obtained by taking a linear combination with @xmath21 symmetry of the four o @xmath22 orbitals which form a plaquette around the cu ion . these are the only oxygen states which hybridize directly with cu . however , it should be mentioned that these plaquette states are not orthogonal , two neighboring states sharing a common oxygen atom . an orthogonal basis can be obtained by the procedure described in the original zr paper @xcite . first , applying a fourier transform , translational invariant ( bloch ) states are constructed . the bloch states are orthogonal but not normalized , so they should be multiplied by a normalization factor @xmath23^{-1/2}~.\ ] ] after normalization a complete and orthonormal set of oxygen states is obtained . in this basis the two - band hubbard hamiltonian can be written as : @xmath24 we work in the hole representation and @xmath25 ( @xmath26 ) represents the creation operator of a cu ( o ) hole with spin @xmath27 and momentum @xmath28 . the o band dispersion and the cu - o hybridization are given by @xmath29 @xmath30 with @xmath31 being the o - o hopping integral . the last term in eq.[eq : ham ] represents the coulomb repulsion between two holes on the same @xmath6 orbital . we choose the commonly accepted values of the parameters , based on the band structure calculations of mcmahan _ et al . _ @xcite and hybertsen _ et al . _ @xcite because of the low density of oxygen holes ( @xmath32 ) , we treat the coulomb repulsion on @xmath33 ( given by @xmath34 ) and the repulsion between nearest - neighbor cu and o holes ( given by @xmath35 ) at the mean field level as a reasonable approximation . the effect will be an increases of our estimation for @xmath36 by @xmath37 , where @xmath38 and @xmath39 are the average occupation of cu and respectively o bands . a choice of @xmath40 , @xmath41 and @xmath42 results in a increase of @xmath43 by @xmath44 . to conclude , we take in eq . [ eq : ham ] , @xmath45 and @xmath46 . ) around @xmath47=@xmath48 and @xmath49.,width=240 ] the dca is an extension of the dynamical mean field theory ( dmft)@xcite . the dmft maps the lattice problem to an impurity embedded self - consistently in a host and therefore neglects spatial correlations . the dca maps the lattice to a finite - sized cluster embedded in a host . non - local correlations up to the cluster size are treated explicitly , while the physics on longer length scales is treated at the mean field level . this technique is ideal for the problems where short - ranged correlations are predominant as in the high-@xmath2 materials . here we calculate the properties of the embedded cluster with a quantum monte carlo ( qmc ) algorithm . the cluster self - energy is used to calculate the properties of the host , and this procedure is repeated until a self - consistent convergent solution is reached . the self - energy and vertex functions of the cluster are then used to calculate lattice quantities . a detailed description of the qmc - dca algorithm is given in ref . @xcite in the two - band model the oxygen degrees of freedom are not correlated , and therefore they are not included explicitly in the cluster . their effect is fully contained in the cluster - host hybridization function and in the host green s function . here we consider a @xmath50 cluster of cu ions , which we believe to be large enough to capture the essential physics of hubbard - type models . the @xmath50 cluster will result in a coarse - graining of the bz in four cells , as shown in fig . [ fig : coarse ] . the solid line is the @xmath6 part of the dos calculated at @xmath51 . the value of the parameters is @xmath52 . the dashed line shows the dos when @xmath53 . the chemical potential @xmath54.,width=316,height=211 ] the undoped materials have one hole per @xmath55 unit . for @xmath53 the dos is given by the dashed line in fig . [ fig : zr ] and the hole addition states would be of pure o character . when the cu - o hybridization @xmath56 is switched on , the extra holes added to the oxygen band will scatter with the @xmath11 spins and bound states will appear at the bottom of the oxygen band . this is illustrated by the solid line , which plots the partial @xmath6 dos which was obtained using the maximum entropy method ( mem)@xcite for the analytic continuation of the qmc data to real frequencies . it can be noticed that the first hole addition states have a strongly mixed @xmath6 and @xmath8 character ( the @xmath6 character in the spectrum is large now ) and an energy pushed well below the edge of the initial non - bonding oxygen band . therefore only these states are relevant for the low - energy physics up to @xmath57 hole doping . ] . in the zr theory these low - energy states which appear as a consequence of the strong cu - o hybridization are considered to be local singlets which move through the lattice like holes in an afm background . consequently the claim is that the physics can be described by a one - band t - j model . in order to determine the phase diagram we calculate a large number of susceptibilities which are relevant for spin , charge and superconducting ordering , both at the center and the corner of the bz . for example , the nel and sc critical temperatures , @xmath58 and respectively @xmath2 in the phase diagram presented in fig . [ fig : phase ] are determined from the divergence of the corresponding susceptibilities . the pseudogap crossover temperature @xmath59 is obtained from the maximum in the uniform magnetic susceptibility when accompanied by a suppression of spectral weight in the dos . similar to what was found in the single - band hubbard model @xcite , we find afm and d - wave sc for both electron and hole doped regimes . however the electron - hole symmetry is broken . in the electron doped case afm persists to a much larger doping . on the contrary , sc disappears at a smaller critical doping clusters are considered . ] . these features of the phase diagram are in qualitative agreement with the experimental findings @xcite . the one - electron spectral functions as measured with photoemission are also different . our @xmath60 cluster divides the bz into four cells around @xmath47=@xmath48 and @xmath49 ( see fig . [ fig : coarse ] ) and approximates the lattice self - energy by a constant @xmath61 within a cell . because of this coarse - graining , a comparison with arpes is not possible apart from gross features . however , as the phase diagram shows , we believe that even our small cluster captures much of the physics of the cuprates . here we want to stress the difference between the electron and the hole doped case within our @xmath60 cluster approximation . in fig . [ fig : dos ] a ) and b ) we show the total @xmath6 states dos and the @xmath6 coarse - grained @xmath47 dependent dos ( which would correspond to the average over all @xmath28 belonging to a coarse - graining cell of the single particle spectra @xmath62 ) for the hole and respectively for the electron doped case , at @xmath63 doping . the total dos looks qualitatively similar and at the chemical potential we see in both cases a depletion of states which indicates the presence of the pseudogap . the @xmath47 dependent dos is very different . the important feature which we want to stress is the location of the pseudogap in the bz . in the hole doped case the pseudogap appears around @xmath64 . for the electron doped case we do not detect any suppression of states around @xmath64 even though the pseudogap is clearly present in the total dos . these features are in agreement with the photoemission experiments . the hole doped materials show fermi pockets around @xmath65 and gapped states around @xmath64 @xcite . for the electron doped materials the photoemission spectra @xcite exhibit a gap near @xmath65 and fermi surface pockets around @xmath64 . with the present cluster size the dca can not determine where in @xmath28-space the pseudogap is , but it is interesting that it is not at @xmath64 . the presence of the pseudogap at @xmath65 for the electron doped system can only be checked by increasing the cluster size , and this work is in progress . dos and coarse - grained @xmath47 dependent @xmath6 dos at @xmath63 doping . a ) hole doping case . b ) electron doping case.,width=316,height=240 ] ( upper part ) and charge @xmath66 ( lower part ) susceptibilities versus temperature for different hole densities . @xmath67 in the legend represents the number of holes per unit cell.,width=316,height=240 ] -wave pairing susceptibility @xmath68 versus temperature for different hole densities.,width=316,height=240 ] , the unscreened cu moment , @xmath69 ( eq . [ eq : uscr ] ) versus hole filling . b ) the screened cu moment , @xmath70 ( eq . [ eq : scr ] ) versus hole filling.,width=316,height=240 ] the electron and the hole doped susceptibility functions are also different both for the divergence temperatures and the temperature and doping dependence . in fig . [ fig : sus ] we show the uniform spin and charge susceptibilities versus temperature at @xmath63 and @xmath71 doping . a common feature for all cases is the existence of a characteristic temperature @xmath72 below which the spin response is suppressed and the charge one is enhanced . @xmath72 corresponds to the pseudogap ( seen in the dos ) onset temperature . the suppression of the spin excitations below @xmath72 was also seen in nmr @xcite experiments and it was associated with the pseudogap . besides these common features the electron and the hole doped susceptibilities behave differently . generally , the maximum value of the spin susceptibility increases with hole filling . this means that in the hole doped case , the spin susceptibility at the pseudogap temperature is strongly increasing with doping unlike in the electron doped case where it decreases upon doping . at the same doping the hole doped spin susceptibility is much larger than the electron doped one . another interesting feature is the very strong increase of the charge susceptibility for the electron doped case in the underdoped region ( @xmath63 doping ) , suggesting a tendency towards phase separation @xcite . asymmetric behavior can also be noticed in fig . [ fig : scsus ] , where we plot the _ inverse _ of the @xmath6-wave paring susceptibility . above @xmath73 the pairing susceptibility increases with doping in the electron doped case and remains more or less constant in the hole doped case . because of the large cu - o hybridization the system is strongly covalent . for example in the undoped regime the cu occupation number is only @xmath74 . the fact that the cuprates are strongly covalent was also observed in nmr measurements @xcite . we notice that the system exhibits a slightly doping dependent covalency . this is shown in fig . [ fig : ndmiu]-a where the cu occupation number versus hole density is plotted . a constant covalency , equal to the one in the undoped regime ( i.e. @xmath75 cu holes and @xmath76 o holes per site ) , would correspond to the dashed line . it can be noticed that , for the electron doped regime , the cu hole occupation number is decreasing faster than the hole concentration , which indicates an increasing covalency with increasing electron doping . this happens because at large electron doping , i.e. when the hole - filling of the @xmath3 plane is small , the effective hybridization is a result of a large @xmath77 in the bz ) is strongly @xmath28 dependent , its value taking values from @xmath78 at @xmath49 point to zero at @xmath79 point in the bz . ] . increasing the number of holes , the bz starts to fill up and a smaller @xmath77 will be responsible for the hybridization , and consequently the covalency decreases . for the hole doped regime , the extra holes go primarily on the oxygen band , and therefore we do not have a direct measure of the covalency . in fig . [ fig : ndmiu]-a the unscreened moment on the cu orbitals is shown . it is defined as : @xmath80 the difference between @xmath81 and @xmath82 is a measure of the double occupancy with holes on cu sites . in the electron doped regime the double occupancy is very small , but it increases substantially in the hole doped regime , which indicates that the low - energy hole addition states contain double occupied cu configurations in a significant measure . in fig . [ fig : ndmiu]-b the screened moment on cu , defined as @xmath83 where @xmath84 is cu spin operator at site @xmath85 , is shown . the main effect of the extra holes is to screen the spins on the cu sites . the screening starts to be effective bellow temperatures of about @xmath13 ( not shown ) . in the zhang - rice scenario an extra hole perfectly screens one spin on cu forming a strongly bound on - site singlet which would contain a significant amount of the double occupied cu configuration . so , our results do not contradict the zr theory , but also do not exclude other scenarios where the extra holes form more complicated bound states which involve more than one cu spin . quantitative analysis based on the amount of screening as function of hole doping can not give an answer to the validity of the zr assumption , because , aside from the screening due to the oxygen holes , there are also non - local processes which contribute to the screening of cu moments ( for example , a possibility is the formation of inter - site spin singlets associated with the resonance valence bond scenario ) . concluding that the electron - hole asymmetry is an intrinsic property of the @xmath3 plane , we next address the cause of this asymmetry and the possible reduction to a one - band model . in sec . [ sec:2bm ] we showed that , due to the cu - o hybridization , the addition of holes results in the formation of low - energy states , with an energy well bellow ( @xmath12 ) the initial oxygen band ( see fig . [ fig : zr ] ) . the reduction to a one - band model is based on the zr claim that these states are singlets , i.e. spinless entities which can be regarded as holes moving in an antiferromagnetic background . due to the monte carlo nature of our calculation , which does not provide a wave function for the ground state , we can not determine directly the exact nature of these states . the most we can do is to compare the results of a two - band hubbard model calculation with those of a one - band hubbard model , and based on the similarities and differences that we might find to decide about the validity of the single - band approach . in order to compare the two - band and the one - band models , we should first get an idea about the possible single - band model effective parameters . we discuss here two different approaches for calculating these parameters , both based on the assumption that the low - energy states are localized , and close to the zr proposed singlets . [ cols="<,<,<,<,<,<,<,<,<,<",options="header " , ] the cell - perturbation theory @xcite assumes that the zr mapping is _ strictly _ true and therefore the low - energy states are genuine local singlets . here and everywhere in the paper by local we refer to the oxygen _ orthogonal _ wannier states which are different from the non - orthogonal plaquette states around the cu ions . to deduce the one - band model parameters we work in the site representation . we can fourier transform eq . [ eq : ham ] and write it as @xmath86 where @xmath87 here @xmath85 represents the site index . the oxygen operators @xmath88 describe the orthogonal wannier states . the zr assumption implies that @xmath89 is responsible for the formation of the low - energy states ( local singlets ) , and @xmath90 will determine the hopping parameters . therefore the cell - perturbation theory provides a means to determine the one - band parameters provided that the zr assumption is correct . elaborate calculations along this line were done in @xcite for a variety of multi - band parameters . in a first order approximation in @xmath91 , the effective u is given by @xmath92 where @xmath93 , @xmath94 and @xmath95 represent the energies of the two ( i.e. the zr singlet ) , one ( i.e. the bonding state ) and respectively zero hole states of eq . [ eq : ham1 ] . an important point is that @xmath90 introduce three types of hoppings . if we denote with @xmath96 , @xmath97 and @xmath98 the lowest energy states of @xmath99 corresponding to two , one and respectively zero holes , we have the following hopping integrals @xmath100 @xmath101 ( eq . [ eq : hopph ] ) describes the hopping of the zr singlet , @xmath102 ( eq . [ eq : hoppe ] ) the hopping of the electron , and @xmath103 produces the exchange interaction @xmath104 the cell - perturbation theory applied to our model gives the parameters shown in the first row of table . [ tab : zrparam ] . we want to point out two things . first , the reduced hamiltonian in the cell - perturbation theory is a t - t-j model , @xmath105 with different hopping parameters for the electron and the hole doped regions and with a value of the exchange interaction not determined by the quasiparticle s hopping ( @xmath101 or @xmath102 ) , but , as it is shown in eq . [ eq : jj ] by @xmath103 . therefore , a comparison with a one - band hubbard model , should be done cautiously . second , we want to stress that the value of the next - nearest - neighbor hopping terms ( @xmath106 and @xmath107 ) is very small compared to the nearest - neighbor terms . the reason is that the initial oxygen - oxygen hybridization , @xmath31 , results in an effective hopping term comparable in magnitude with the one resulting from the copper - oxygen hybridization , but with a different sign . this was also remarked in @xcite , and turns out to be an important observation for our final conclusions . the other approach used for determining the parameters of the one - band model is based on a cluster calculation . in order to estimate the nearest - neighbor hopping , the next - nearest - neighbor hopping and the exchange terms , eskes _ @xciteconsidered two clusters , @xmath108 ( which contains two nearest - neighbor cu ions ) and respectively @xmath109 ( which contains two next - nearest - neighbor cu ions ) . the exchange term is determined as the energy difference between the singlet and the triplet state of two holes on a cluster . for three holes on a cluster , the two energetically lowest states can be very well ( @xmath110 ) approximated with the bonding and anti - bonding states of a _ plaquette zr singlet _ hopping between the two cells . therefore , the differences between these two levels is two times the zr singlet hopping , @xmath101 . in an analogous way , considering only one hole on a cluster , the electron hopping , @xmath102 , is determined . using the cluster approach , our two - band model results in the effective parameters shown in the second row of the table [ tab : zrparam ] . it can be immediately noticed that the two approaches produce different parameters , especially regarding the value of the next - nearest - neighbor hoppings . in the cluster calculation we obtain significant next - nearest - neighbor hoping terms , @xmath111 and @xmath112 with different sign for the hole and respectively electron doped case ( @xmath113 , @xmath114 ) . the reason for the discrepancy between the two approaches is that , unlike the cell - perturbation method which considers local singlets , the cluster approach considers singlets between a cu hole and a oxygen state formed on the plaquette around the cu ion . since the oxygen plaquette states are nonorthogonal it is possible to write them as a linear combination of many orthogonal oxygen states at different sites , i.e. the plaquette singlets are nonlocal states ( in the orthogonal base ) . it should be pointed out that at first glance this non - locality seems irrelevant ( the overlap of the local oxygen states with the plaquette states @xcite is about @xmath18 ) , but apparently it turns out to influence the value of the next - nearest - neighbor hopping of the reduced hamiltonian considerably . as pointed out by emery and reiter @xcite , the plaquette singlets are in fact an admixture of zr singlets and zr triplets and this can result in a significant value of the spin - spin correlation on the oxygen sites . of course , if the zr singlet - zr triplet admixture is significant , this will make a rigorous reduction to a one - band model impossible , because now the low - energy states have also a spin component and therefore can not be mapped into holes ( spinless entities ) . it is worth pointing out that , in the cluster approach , the large value of the next - nearest - neighbor hopping terms results solely from the finite oxygen dispersion and the lack of hopping between the copper and the next - nearest neighbor oxygen plaquette state . on the other hand , in the cell - perturbation theory a copper - next - nearest neighbor oxygen hopping term is present . it results in an effective next - nearest - neighbor hopping with a sign different than the one produced via oxygen - oxygen hopping . we believe that a comparison between the two - band hubbard model and a single - band hubbard model should be done with extreme caution . we want to stress the possible problems here . first , the reduction based on the zr approximation , which results in a single - band t - j ( or t - t-j ) model assumes the strong - coupling limit , i.e. a ratio @xmath115 ( the two - dimensional bandwidth is @xmath116 ) . the low - energy density of states of the two - band model shown in fig . [ fig : zr ] and fig . [ fig : dos ] indicates a bandwidth of the order of the gap , showing that we are rather at the intermediate - coupling than at strong - coupling . in the cell - perturbation theory we get @xmath117 , which also suggests intermediate - coupling physics . therefore , the question to be asked is whether the intermediate coupling regime , characterized by an effective repulsion of the same order of magnitude as the bandwidth , can still be well approximated by a second - order perturbation reduced t - j model . second , considering the previous objection , one may think that a reduction to the single - band hubbard model in intermediate coupling regime , rather than to a t - j model , is more appropriate . however , serious problems arise from the fact that , in the zr theory the nature of the antiferromagnetic correlations is different from that in the single - band hubbard model , i.e. it is not directly related with the quasiparticle ( zr singlet or electron ) hopping . therefore , unless both the two - band hubbard model and the one - band hubbard model can be reduced to a t - j model , a comparison between them does not make much sense . nevertheless , we believe that even when the effective repulsion is comparable with the bandwidth the second order perturbation theory which produce the t - j model can be used successfully . we are going to discuss this at the beginning of the next section ( sec . [ ssec : ttu ] ) . third , the non - locality of the low - energy states ( in the sense discussed in sec . [ sssec : comp ] ) can have very serious consequences beyond determining the value of the hopping parameters , making the single - band approach to fail completely . , for different values of the ratio @xmath118 of the single - band hubbard model . , width=316,height=211 ] the t - j model results as a low - energy effective hamiltonian from the hubbard model by projecting out the doubly occupied states . therefore , the double occupancy of the site orbitals constitutes a measure of the validity of this approximation . in fig . [ fig : docc ] we plot the double occupancy of the site orbitals for different values of the ratio @xmath118 . it can be noticed that for @xmath119 the double occupancy is always smaller than @xmath120 beyond this value , the double occupancy increases fast , being about @xmath121 for @xmath122 . ] . this indicates that , even in the intermediate coupling regime the low - energy physics of the one - band hubbard model can be well described by a t - j model . even if , it is more natural to compare the two - band model with a t - t-j ( or a t - t-j - j ) model , this turns out from our perspective to be rather inconvenient due to the technical difficulties encountered by the qmc when applied to such models . therefore , we proceed by comparing the two - band model with a t - t-u hubbard model , focusing on the qualitative features rather than on a quantitative comparison . in the strong - coupling limit , the t - t-u model reduces to a t - t-j - j model , with the constraint @xmath123 . therefore , it is reasonable to assume that if the value of @xmath124 is not too large and the reduction of the two - band model to a single - band model is valid , the two models should exhibit similar physics . assuming that the reduction to a one - band model in the spirit of the zr approximation is possible , we should expect from table . [ tab : zrparam ] the hopping parameters to be different in the hole doped and in the electron doped regions . on the other hand the exchange interaction , @xmath125 should be the same . therefore , we study the single - band t - t-u hubbard model and address the following questions : * how do the system properties depend on the ratio @xmath126 ? * what is the role of the next - nearest - neighbor hoping t ? doping , for three different values of @xmath127 , when @xmath128 is constant.,width=316,height=211 ] ( black ) and @xmath129 ( grey ) doping for @xmath130 ( circle ) and @xmath131 ( square).,width=316,height=211 ] at @xmath63 ( circle ) and @xmath71 ( squares ) doping . inset -b ) inverse of d - wave pairing susceptibility versus temperature at @xmath132 doping , for @xmath130 ( circles ) and @xmath131 ( diamonds ) . , width=316,height=211 ] dependent dos at @xmath63 doping . @xmath133 , @xmath134 , width=316,height=192 ] the values of the parameters in table . [ tab : zrparam ] show that in general the ratio @xmath135 is larger in the hole - doped regime than in the electron - doped case . in order to address the electron - hole asymmetry observed in the two - band model , in this section we study the properties of the single - band hubbard model as a function of @xmath126 , by keeping @xmath128 ( given by eq . [ eq : jju ] ) constant and varying the hopping @xmath127 . the next - nearest - neighbor hopping @xmath0 is set to zero . with respect to antiferromagnetism , with increasing @xmath127 the nel temperature at small doping and the critical doping where the antiferromagnetism disappear decrease . for example , at @xmath63 doping , the antiferromagnetic susceptibility is diverging only for the small value of @xmath127 shown in fig . [ fig : jtafm ] . assuming that the hole doped regime is characterized by a larger value of @xmath126 , this feature is in agreement with the two - band model asymmetric behavior ( see fig . [ fig : phase ] ) . the uniform spin susceptibility is shown in fig . [ fig : jtssus ] . one can notice that an increase of @xmath127 results in an increase of @xmath59 and a decrease of the spin susceptibility at @xmath59 . this together with the behavior of the susceptibility as a function of doping is in contrast to what was observed in the two - band model ( see fig . [ fig : sus ] ) where the spin susceptibility is larger in the hole doped case and an increase ( decrease ) with doping of the susceptibility at @xmath59 for the hole ( electron ) doped regimes is found . the behavior of the @xmath6-wave pairing susceptibility as a function of @xmath127 is shown in fig . [ fig : jtdsus ] . the critical temperature increases with increasing @xmath127 ( the increase of @xmath2 is about @xmath71 of the increase of @xmath127 ) , as can be seen in the inset ( a ) . this increase is much too large to be in agreement with the two - band model results even if , actually , for the two - band model we obtained a hole - doped @xmath2 larger with about @xmath136 than the electron - doped one ) . ] . by extrapolating the inverse of the @xmath6-wave pairing susceptibility at @xmath132 doping ( see inset ( b ) ) it can be concluded that an increase of @xmath127 results in an increase of the critical doping where sc disappears . we also notice that , at small doping and above @xmath2 , a large @xmath127 suppresses the pairing correlations . these features are in agreement with the asymmetry of the two - band model phase diagram . nevertheless , we notice that , above @xmath2 and for both values of @xmath127 , with increasing the doping the pairing correlations increase too . this behavior is characteristic in the electron doped regime of the two - band model , but can not explain the hole doped regime where the pairing does not depend on the doping ( see fig . [ fig : scsus ] ) . the other difference between the two - band and the single - band hubbard model is the value of the sc susceptibility critical exponent @xmath137 , which is much smaller in the two - band model case . the density of states and the @xmath47 dependent dos for the one - band t - u hubbard model at @xmath63 doping is shown in fig . [ fig : jtdos ] . the one - particle spectra exhibit a pseudogap in the total dos and in the @xmath47 dependent dos at @xmath64 point in bz , similar to the hole doped spectra of the two - band hubbard model . the single - band t - u hubbard hamiltonian is particle - hole symmetric and therefore can not explain the one - particle spectra in the electron - doped regime of the two - band hubbard model . at the end of this section we conclude the following : a single - band t - u hubbard model ( i.e. @xmath138 ) with a larger value of the hopping parameter for the hole doped regime _ can not _ explain the electron - hole asymmetries observed in the two - band hubbard model , especially the ones which characterize the one - particle spectral functions and the susceptibility functions . . for the t - t-u hubbard model @xmath139 . , width=316,height=211 ] ( upper part ) and charge @xmath66 ( lower part ) susceptibilities versus temperature for different hole densities.,width=316,height=211 ] -wave pairing susceptibility @xmath68 versus temperature for different hole densities.,width=316,height=211 ] total dos and coarse - grained @xmath47 dependent dos at @xmath63 doping for @xmath140 . a ) hole doping case . b ) electron doping case . , width=316,height=211 ] in this section we study the role of the next - nearest - neighbor hopping , @xmath0 , in the single - band hubbard model @xmath141 we choose the following parameters , @xmath142 , @xmath143 and @xmath144 . these parameters are close to the ones in table . [ tab : zrparam ] , resulting in @xmath145 and @xmath146 . as for the two - band hubbard model , we work in the hole representation , defined as the one where the filling @xmath147 corresponds to a hole doping @xmath148 . values of the filling smaller than one correspond to the electron doped regime . we keep the sign of @xmath0 always positive . in order to avoid confusion we want to point out that in a t - j model the filling is always smaller than one . therefore , in order to describe the electron and the hole doped regimes one has to employ the hole and , respectively , the electron representation . accordingly , the sign of @xmath0 has to be chosen negative in the hole doped regime and positive in the electron doped case and @xmath0 change sign . a change of @xmath127 sign has no physical consequences aside from a translation with @xmath49 in the bz corresponding to a canonical transformation which changes the sign of site orbitals on one sublattice . an equivalent statement is that the particle - hole transformation , as defined in ref . @xcite , change the sign of @xmath0 but not of @xmath127 . ] . in fig . [ fig : ttuphase ] the phase diagram of the t - t-u model is shown with a solid line . for comparison , the phase diagram of t - u hubbard model ( i.e. @xmath138 case ) , which is symmetric with respect to hole and electron doping , is shown with dashed line . at half filling @xmath0 introduces an effective antiferromagnetic exchange @xmath149 between the same sublattice spins and subsequently frustrates the lattice . however at finite electron doping , @xmath0 favors the antiferromagnetism , making it persist up to a larger doping . on the other hand , in the hole doped case , the antiferromagnetism is always suppressed by @xmath0 . with respect to superconductivity , the presence of @xmath0 results in a smaller ( larger ) critical electron ( hole ) doping at which the superconductivity disappears . the asymmetry introduced by @xmath0 is in agreement with the one observed in the two - band model phase diagram . we find that @xmath0 has no major influence on the maximum superconductivity critical temperature @xmath150 . the uniform spin and charge susceptibilities are shown in fig . [ fig : ttuspinsus ] . the spin susceptibility at the pseudogap temperature @xmath72 is strongly increasing with doping for the hole doped case and an opposite effect is seen for the electron doped case . the downturn at @xmath72 in the spin susceptibility is much sharper for the hole doped regime indicating a fast transition to the pseudogap physics . all these features are in very good qualitative agreement with the ones corresponding to the two - band hubbard model . because of the similarity with the two - band model , it is also worth mentioning that in the electron - doped regime the charge susceptibility is strongly increased below @xmath72 in the underdoped region . the @xmath6-wave paring susceptibilities shown in fig . [ fig : ttuscsus ] exhibit asymmetric features , also in a qualitative agreement with those in the two - band model . in the electron doped regime , with increasing the doping , the pairing correlations above @xmath2 increase . in the hole - doped regime close to @xmath2 , the pairing correlations do not significantly depend on the doping . however , contrary to the two - band model behavior , at larger temperature an increase of pairing correlations with doping is observed . the magnitude of this increase is smaller than in the electron - doped case and a larger value of @xmath0 ( e.g. @xmath151 , not shown ) will reduce it further , improving the resemblance with the two - band model . in fig . [ fig : ttudos ] we present the dos of the t - t-u hubbard model at @xmath63 doping . the one - particle spectral functions resemble the corresponding two - band hubbard model ones . the presence of the @xmath0 parameter is responsible for the location of the pseudogap in the bz . the necessity of the @xmath0 in explaining the measured arpes line shape and the electron - hole asymmetry was realized early on @xcite . representing hoppings in the same sublattice , this parameter is not severely renormalized by the afm background , and consequently its influence turns out to be important . exact diagonalization results @xcite of a t - t-j model are in agreement with ours . the @xmath0 hopping process lowers the kinetic energy and moves the quasiparticle position from @xmath65 to @xmath64 in the electron doped case . the nel - like configurations , which do not hinder this process , are stabilized . in the hole doped case the @xmath0 hopping does not lower the kinetic energy of quasiparticles and it is not energetically favorable , therefore leading to a suppression of afm at all dopings . the main conclusion of this section is that a one - band t - t-u hubbard model describes qualitatively well the physics ( i.e. the phase diagram , the one - particle spectra and the two - particle response functions ) of the two - band hubbard model , provided a significant value of the next - nearest neighbor hopping ( @xmath152 ) is considered . however , besides all these similarities there are also some important differences which we are going to emphasize in the next section . in general , the deduction of an effective low - energy hamiltonian implies two steps . first , defining the low - energy states and second , projecting the resolvent operator , @xmath153 , on the subspace spanned by these low - energy states @xcite . the inverse of the projected operator can be written as @xmath154 , where @xmath155 is the low - energy hamiltonian can be considered independent of @xmath156 if the high - energy and the low - energy states are well separated . ] . this procedure is equivalent to finding an hamiltonian which produces the same one - particle , two - particle , three - particle , etc . , spectral functions on the energy range considered to be `` low - energy '' . rigorously , in order to prove that the one - band model is the effective hamiltonian which describes the two - band hubbard model low - energy physics we should compare not only the one - particle and the two - particle spectra , but also all higher order correlation functions . however , we believe that the comparison of only the one - particle and the two - particle spectral functions is compelling enough , especially since the experimental information is also obtained by measuring the response functions behavior ( and in almost all cases the two - particle operators or the one - particle ones , as in photoemission , are involved ) . it is also true that a comparison of the dynamic susceptibilities would be required , but with our quantum monte carlo based algorithm the calculation of these quantities for the two - band model is extremely computational resources consuming and has not been done yet . however partial information about the relevant excited states is contained in the temperature behavior of the static susceptibilities . the main conclusion of sec . [ sec:1bm ] is that a t - t-u hubbard model describes qualitatively well the physics of the two - band hubbard model , but only if a substantial next - nearest - neighbor hopping is considered . however , the calculation in sec . [ sssec : cell ] ( first row of table . [ tab : zrparam ] ) and the more rigorous results by feiner _ et al . _ @xcite , show that in a strict zr picture the next - nearest - neighbor hopping is negligible . therefore it is difficult to explain the two - band hubbard model physics assuming the formation of local zr singlets . for hole doped systems , a significant value of @xmath0 can be obtained only if the extra holes form nonlocal bound states with the existing cu holes , presumably something close to the plaquette singlets . of course we have no reasons to discard other states spread over even more oxygen sites , which can result in a magnitude of the hopping parameters different ( probably not too much ) form the one obtained by cluster calculation ( second row of table . [ tab : zrparam ] ) . in the electron doped systems , the doping dependent covalency shown in fig . [ fig : ndmiu]-a , clearly indicates that the hybridization of the cu with the o states at different sites is important . a doping dependent covalency should also imply doping dependent parameters . the cluster calculation which allows the formation of non - local ( plaquette ) low - energy states , unlike the cell - perturbation ( or strict zr ) approach , provides a value of the hopping parameters which captures qualitatively the physics of the two - band model . however , we do not believe that finding the exact value of the one - band hubbard model parameters is a relevant or even a well addressed problem , because the non - locality of the low - energy states implies that the two models are not equivalent , as was pointed out by emery _ et al . _ @xcite . aside from the similarities between the two - band and the t - t-u hubbard models discussed in sec.[sssec : tprim ] we also find some differences . for example , one important difference can be observed in the @xmath6-wave pairing susceptibility ( fig . [ fig : ttuscsus ] and fig . [ fig : scsus ] ) . in the two - band hubbard model the critical exponent @xmath137 which defines the divergence of the susceptibility at @xmath2 , is much smaller ( around @xmath157 at finite hole doping ) than the one characteristic to the one - band model ( around @xmath158 ) , indicating larger fluctuations from the mean - field value @xmath159 is a measure of fluctuation . ] . both the cell - perturbation and cluster calculation provide a larger nearest - neighbor hopping @xmath127 for the hole doped region . according to the analysis presented in sec . [ sssec : tj ] this should result in both larger @xmath59 and @xmath2 . however the two - band model results does not indicate that this is the case , the respective critical temperatures being not very different in the electron and hole doped regimes . based on our comparison we can draw the following conclusions . the one - band hubbard model retains much of the two - band hubbard model physics , but a significant next - nearest - neighbor hopping ( @xmath1 ) should be provided . if the purpose of the investigation is the study of the basic physics like the sc mechanism , the proximity of afm , sc and pseudogap we believe that a one - band t - t-u hubbard model should be good enough . on the other hand , if the purpose is to describe more subtle features like the ones which may result from the finite value of the spin correlation on oxygen , or if a quantitative material specific calculation is desired , the single - band model approach fails . obviously also the single - band model should not be used to describe spectral features at energies above @xmath160 such as the optical , electron energy loss and inelastic x - ray scattering results . in this paper we use the dca to calculate the properties of the two - band hubbard model . the 2@xmath1612-site cluster phase diagram resembles the generic phase diagram of the cuprates and exhibits electron - hole asymmetry . we also find asymmetric features for the one - particle spectral functions and for the relevant susceptibility functions . these characteristics are in qualitative agreement with experimental findings . we address the validity of the single - band hamiltonian as the effective low - energy model for the cuprates . we discuss the possible problems which may cause the failure of the reduction from two - band to one - band and also show that , depending on the approximations involved , the value of the one - band hubbard parameters ( especially the next - nearest - neighbor hopping ) can be significantly different . we use dca to study the role of the different parameters in the single - band t - t-u hubbard model and compare the phase diagram , the one - particle and the two - particle response functions with the corresponding two - band hubbard model ones . we conclude that the two models exhibit similar low - energy physics provided that a significant next - nearest - neighbor hopping @xmath0 is considered . the parameter @xmath0 is also the main culprit for the electron - hole asymmetry of the cuprates . the large value of @xmath0 needed for a qualitative agreement between the two models can not be obtained in a strict zr picture where the extra holes form local singlets with the existing cu holes . plaquette singlets , which in the oxygen wannier representation are not local , and presumably other spatially extended states can provide a larger value of @xmath0 . the doping dependent covalency in the electron doped case also indicates that the non - local cu - o hybridization is important . however , the formation of non - local low - energy states also implies that they are not real singlets and consequently can not be rigorously mapped into holes and therefore the two models are not equivalent . we also point out some differences between the two models . in the two - band hubbard model the fluctuations in the @xmath6-wave pairing channel above @xmath2 is much stronger . the deduction of the parameters both in cell - perturbation and cluster approach results in a larger nearest - neighbor hopping @xmath127 for the hole doped regime . however the critical temperatures @xmath59 and @xmath2 in the two - band hubbard model are approximatively the same in both regimes , quite different from what should be expected . the conclusion is that a single - band hubbard model with a significant value of the next - nearest - neighbor hopping ( @xmath1 ) , captures the basic physics of the two - band hubbard model , including the proximity of antiferromagnetism , superconductivity and pseudogap and explaining the electron - hole asymmetry seen in the phase diagram , one - particle and two - particle spectral functions . however , the single - band hubbard model is not entirely equivalent to the two - band hubbard one and we believe that it is not suitable for quantitative material specific studies or for describing more subtle features which may results from the non - locality of the low - energy states . it is also not suitable to describe physics which implies excitations with energy scales larger than @xmath13 . we thank f.c . zhang and paul kent for useful discussions . the work was supported by nsf grant dmr-0073308 , by cmsn grant doe de - fg02 - 04er46129 and by the netherlands foundation for fundamental research on matter ( fom ) with financial support from the netherlands organization for scientific research ( nwo ) and the spinoza prize program of nwo . the computation was performed at the pittsburgh supercomputer center , the center for computational sciences at the oak ridge national laboratory and the ohio supercomputer center . part of this research was performed by tm as a eugene p. wigner fellow and staff member at the oak ridge national laboratory , managed by ut - battelle , llc , for the u.s . department of energy under contract de - ac05 - 00or22725 . 99 m. h. hettler , a. n. tahvildar - zadeh , m. jarrell , t. pruschke and h.r . krishnamurthy , phys . b * 58 * , r7475 ( 1998 ) ; m. h. hettler , m . mukherjee , m. jarrell and h.r . krishnamurthy , phys . b * 61*,12739 ( 2000 ) . c. almasan and m.b . maple , _ chemistry of high - temperature superconductors _ , 1991 edited by c.m.r . rao ; e. dagotto , rev . phys * 66 * 763 ( 1994 ) b.o . lett . * 74 * , 964 ( 1995 ) n.p . et al . _ , lett . * 88 * , 257001 ( 2002 ) j. zaanen , g.a.sawatzky and j.w.allen , phys.rev.lett . * 55 * , 418 ( 1985 ) j. d. jorgensen , h. b. schuttler , d.g . hinks , d. w. ii capone , k. zhang , m. b. brodsky , and d. j. scalapino , phys.rev.lett . * 58 * , 1024 ( 1987 ) l. f. mattheiss , phys.rev.lett . * 58 * , 1028 ( 1987 ) j. yu , a. j. freeman and j. h. xu , phys.rev.lett . * 58 * , 1035 ( 1987 ) c. m. varma and s. schmitt - rink , solid state com . * 62 * , 681 ( 1987 ) v.j . emery , phys.rev.lett . * 58 * , 2794 ( 1987 ) ; v.j . emery and g. reiter , phys . rev . * b 38 * , 4547 ( 1988 ) f.c . zhang and t.m . rice , phys . rev . * b 37 * , r3759 ( 1988 ) ; phys . rev.*b 41 * , 7243 ( 1990 ) h. eskes , l.h.tjeng and g.a.sawatzky , phys.rev . * b 41 * , 288 ( 1990 ) e.b . stechel and d.r . jennison , phys.rev . * b 38 * , 4632 ( 1988 ) m.s . hybertsen , m. schluter and n.e . christensen , phys.rev . * b 39 * , 9028 ( 1989 ) j. zaanen and a.m. oles , phys . rev . * b 37 * , 9423 ( 1988 ) h. eskes and g.a . sawatzky , phys . 61 * , 1415 ( 1988 ) v.j . emery and g. reiter , phys . * b 38 * , r11938 ( 1988 ) ; phys . rev.*b 41 * , 7247 ( 1990 ) j. h. jefferson , h. eskes and l. f. feiner , phys . rev.*b 45 * , 7959 ( 1992 ) ; l. f. feiner , j. h. jefferson , and r. raimondi , phys . rev.*b 53 * , 8751 ( 1996 ) h. eskes , g.a.sawatzky and l. feiner , physica * c 160 * , 424 ( 1989 ) a. k. mcmahan , j.f . annett , r.m . martin , phys . rev . * b 42 * , 6268 ( 1990 ) a. macridin , _ ph.d . thesis _ , university of groningen , http://www.ub.rug.nl/eldoc/dis/science/a.macridin . g.kotliar , w.krauth and m.rozenberg , rev.mod.phys . * 68 * , 13 ( 1996 ) m. jarrell , th.maier , c.huscroft , and s.moukouri , phys.rev . * b 64*,195130 ( 2001 ) m. jarrell and j. e.gubernatis , phys.rep . * 269 * , 135 ( 1996 ) m. jarrell , th.maier , m.h.hettler , a.n.tavildarzadeh , europhys.lett . * 56 * , 563 ( 2001 ) w.w . et al . _ , lett . * 62 * , 1193 ( 1989 ) ; m. takigawa _ et al . _ , phys.rev.*b 43 * , 247 ( 1991 ) ; h. alloul , a. mahajan , h.casalta , o. klein , phys . lett . * 70 * , 1171 ( 1993 ) a. macridin _ et al . _ , in preparation . m. takigawa , _ et al . _ physica c * 162 - 164 * , 853 ( 1989 ) ; m. takigawa _ _ , edited by k.bedell , d. coffey , d.e . meltzer , d. pines , and j.r . schriffer ( , ) , p. 236 s.moukouri and m.jarrell , phys.rev.lett . * 87 * , 167010 ( 2001 ) eduardo fradkin , _ _ , ( , 1991 ) , p. 9 - 10 a. nazarenko , k.j.e . vos , s. haas , e. dagotto , r.j . gooding , phys . rev . * b 51 * , r8676 ( 1995 ) ; r. eder , y. ohta , g. a. sawatzky , phys . rev . * b 55 * , 3414 ( 1997 ) ; p.w . leung , b.o . wells , r.j . gooding , phys . rev . * b 56 * , 6320 ( 1997 ) ; o. p. sushkov , g. a. sawatzky , r. eder , h. eskes , phys . * b 56 * , 11769 ( 1997 ) t. tohyama and s. maekawa , phys . rev.*b 64 * , 212505 ( 2001 ) ; supercond . technol.*13 * , r17 a.auerbach , _ _ , ( , ) , p. 25

we define and investigate spectral invariants for floer homology of an open subset in , defined by kasturirangan and oh as a direct limit of floer homologies of approximations . we define a module structure product on and prove the triangle inequality for invariants with respect to this product . we also prove the continuity of these invariants and compare them with spectral invariants for periodic orbit case in . keywords : lagrangian submanifolds , floer homology , spectral invariants +

1411.7807

lagrangian floer homology for open subsets in cotangent bundles was introduced by kasturirangan and oh in @xcite as a part of a project of `` quantization of eilenberg steenrod axioms '' ( see @xcite ) . the construction goes as follows . let @xmath1 be an open subset of a compact smooth manifold @xmath3 , with a smooth compact boundary @xmath4 . the conormal bundle , @xmath5 , defined as @xmath6 is a lagrangian submanifold of the cotangent bundle @xmath2 . define @xmath7 and @xmath8 the set @xmath9 , called _ the conormal to @xmath10 _ , is a singular lagrangian submanifold , but it allows a smooth approximation by exact lagrangian submanifolds . following @xcite , we denote these approximations by @xmath11 . it holds @xmath12 as @xmath13 in lipschitz topology ( see @xcite for the details ) . floer homology for the open set @xmath14 is defined to be a direct limit of floer homologies of approximations . in order to have the latter well defined , one needs to consider a hamiltonian @xmath15\to\mathbb r$ ] such that @xmath16 and , for a fixed @xmath17 , an open set @xmath14 such that @xmath18 which can be obtained by generic choice of @xmath14 . floer homology for the pair @xmath19 is now defined in a standard way , the set of the generators @xmath20 consists of the hamiltonian paths @xmath21 which are critical points of the _ effective action functional : _ @xmath22 here @xmath23 is a canonical liouville form on @xmath2 and @xmath24 is a function satisfying @xmath25 which exists since @xmath11 is exact . the boundary map @xmath26 is defined as the number of perturbed holomorphic discs with boundary on @xmath27 and @xmath11 : @xmath28\to t^*m,\\ \frac{\partial u}{\partial s}+j_\varepsilon\left(\frac{\partial u}{\partial t}-x_h(u)\right)=0,\\ u(s,0)\in o_m , u(s,1)\in\upsilon_\varepsilon . \end{array}\right.\ ] ] here @xmath29 is an almost complex structure compatible to the standard symplectic form @xmath30 , which coincides with the canonical almost complex structure @xmath31 on @xmath2 at infinity . by the canonical almost complex structure @xmath31 we assume the one induced by the levi - civita connection for a fixed metric @xmath32 . denote by @xmath33 the corresponding floer homology . floer homology of the open subset @xmath14 is defined as a direct limit of above floer homologies for the approximations @xmath11 : @xmath34 after defining an appropriate partial ordering to the set of pairs @xmath35 ( see below or @xcite for more details ) . in this paper we construct and investigate spectral invariants for the floer homology of the open subset @xmath0 and establish their compatibility with the direct limit ( [ eq : dir_limit ] ) . the first step in this direction is the construction of piunikhin - salamon - schwarz isomorphism between @xmath0 and singular homology of @xmath14 ( modelled by morse homology ) . more precisely , we prove the following theorem . let @xmath36 denotes the morse homology of @xmath14 ( see section [ sec : pss ] below ) . [ thm : pss ] there exist pss - type isomorphisms @xmath37 which are inverse to each other and natural with respect to canonical isomorphisms in morse and floer theory . more precisely , if @xmath38 denote the canonical isomorphisms in floer and morse theory respectively , and @xmath39 and @xmath40 the corresponding pss isomorphisms , then the diagram @xmath41 commutes . we construct pss isomorphisms and prove theorem [ thm : pss ] in section [ sec : pss ] . first we construct the corresponding isomorphisms for approximations @xmath33 and prove that they commute with the homomorphisms that define the direct limit ( [ eq : dir_limit ] ) . next , we construct three pair - of - pants type products in morse and floer theory for open sets . products in morse and floer theory were studied by various authors : abbondandolo and schwarz @xcite , auroux @xcite , oh @xcite and also in @xcite . here we establish the following products for open subset . [ thm : prod_open ] there exist a pair - of - pants type products : @xmath42 that turns floer homology for an open set @xmath43 into a @xmath44module . theorem [ thm : prod_open ] is proven in section [ sec : prod ] . since floer homology for the open set is defined as a direct limit , the key step is to prove that the products defined on homology for approximation commute with the homomorpshisms that define the direct limit . finally , using the above pss isomorphism , we construct the spectral invariants for lagrangian floer homology of the open subset @xmath0 . spectral invariants were defined by oh in @xcite in lagrangian floer theory , and by schwarz @xcite for periodic orbit case , following the work of viterbo @xcite . they were further studied by leclercq @xcite , by monzner , vichery and zapolsky in @xcite , eliashberg and polterovich in @xcite ( see also a book by polterovich and rosen @xcite ) , oh in @xcite , humilire , leclercq and seyfaddini @xcite and also in @xcite , @xcite . we prove the following properties of these spectral invariants : their continuity with respects to @xmath17 and their subadditivity with respect to the products from theorem [ thm : prod_open ] . we also compare the above spectral invariants with the invariants for periodic orbits case , using the homomorphisms defined via `` chimneys '' introduced by albers @xcite and abbondadolo and schwarz @xcite . further , we prove the inequality of spectral invariants between two open sets @xmath45 and a specific singular homology class ( see subsection [ subsec : inv_open ] ) . this slightly generalizes a result by oh @xcite for a spectral invariant @xmath46 more precisely , in section [ sec : inv ] we prove the following theorem . [ thm : inv_open ] for given singular or morse homology class @xmath47 , the spectral invariant @xmath48 defined via pss isomorphism from theorem [ thm : pss ] has the following properties : * * triangle inequality . * @xmath49 , for @xmath50 * * continuity . * relative spectral invariant @xmath51 is continuous with respect to @xmath17 * * comparison with periodic orbit invariants . * if @xmath52 stands for a spectral invariants for periodic orbit case in @xmath2 and @xmath53 , @xmath54 are homomorphisms in homology induced by the inclusion map , we have : @xmath55 * * invariants for subsets . * let @xmath45 be two open subset of @xmath3 and let @xmath56 ( the homomorphism induced by inclusion @xmath57 ) be surjective . for @xmath58 it holds : @xmath59 * acknowledgement . * the third author would like to thank yong - geun oh for very fruitful discussions during his visit to center for geometry and physics at pohang in summer 2013 . on that occasion he brought to the third author s attentions his joint work with r. kasturirangan on floer homology for open sets . the third author also thanks center for geometry and physics for their hospitality and an enjoyable atmosphere . pss type isomorphism was originally constructed by piunikhin , salamon and schwarz @xcite for periodic orbit case , and later adapted in @xcite for lagrangian case . we first establish the pss isomorphism for approximations . it follows from ( [ prop : trans_cond ] ) that all solutions of hamiltonian equation @xmath60 with @xmath61 satisfy @xmath62 , so by choosing @xmath63 to be close enough to @xmath9 , we may assume that all solutions of ( [ ham_paths ] ) satisfy @xmath64 the grading for @xmath65 is defined to be @xmath66 where @xmath67 is a canonically assigned maslov index , defined for any smooth closed submanifold @xmath68 ( see definition 5.9 in @xcite ) . the dimension of the space @xmath69 of perturbed holomorphic discs the satisfy ( [ d ] ) and the infinity boundary conditions : @xmath70 is @xmath71 for all @xmath72 ( see @xcite ) . consider now a morse function @xmath73 such that the set of critical points of @xmath74 , @xmath75 is contained in a closed set @xmath76 , @xmath77 , and therefore , is finite . for @xmath78 and @xmath65 define the space of mixed objects ( see figure [ pss_pic ] ) : @xmath79\to u,\ ; u:[0,+\infty)\times[0,1]\to t^*m,\\ \dot{\gamma}(s)=-\nabla_g f(\gamma(s ) ) , \\ \frac{\partial u}{\partial s}+ j(\frac{\partial u}{\partial t}-x_{\rho_rh}(u))= 0,\\ u(s,0 ) , u(0,t)\in o_m,\,u(s,1)\in\upsilon , \\ \gamma(-\infty)=p,\ , u(+\infty , t)=x(t ) , \\ u(0,1)=\gamma(0),\\ \end{array}\right.\right\}\end{aligned}\ ] ] where @xmath80 is a smooth function such that @xmath81 that defines pss isomorphism , width=302,height=151 ] let @xmath82 denotes the morse index of a critical point @xmath83 and @xmath84 . [ prop : dim ] for generic choices the set @xmath85 is a smooth manifold of dimension @xmath86 . _ proof : _ denote by @xmath87 the unstable manifold of @xmath83 and by @xmath88 the set of solutions of @xmath89\to t^*m,\\ \frac{\partial u}{\partial s}+ j(\frac{\partial u}{\partial t}-x_{\rho_rh}(u))=0 , \\ u(s,0)\in o_m,\,u(s,1)\in\upsilon,\,u(0,t)\in o_m \\ u(+\infty , t)=x(t ) . \end{array } \right.\ ] ] for a generic choice of parameters the evaluation map @xmath90 is transversal to the diagonal , so @xmath91 is a smooth manifold of codimension @xmath92 in @xmath93 . since @xmath94 ( see @xcite ) and @xmath95 ( see appendix in @xcite and @xcite ) , the dimension of @xmath85 is @xmath96 in the same way we conclude that the set @xmath97\times[0,1]\to t^*m,\;\gamma:[0,+\infty)\to u,\\ \frac{\partial u}{\partial s}+ j(\frac{\partial u}{\partial t}-x_{\tilde{\rho}_rh}(u))= 0,\\ \dot{\gamma}(s)=-\nabla_g f(\gamma(s ) ) , \\ u(s,0 ) , u(0,t)\in o_m,\,u(s,1)\in\upsilon , \\ u(-\infty , t)=x(t),\,\gamma(+\infty)=p , \\ \gamma(0)=u(0,1)\\ \end{array}\right.\right\}\end{aligned}\ ] ] is a smooth manifold of dimension @xmath98 . here @xmath99 . define @xmath100 to be the set of all solutions of the differential equation @xmath101 modulo @xmath102 action and , similarly , denote by @xmath103 by standard arguments using the arzela - ascoli and gromov compactness theorems , one can prove the following proposition . bubbling can not occur due to exactness of @xmath104 and lagrangian boundary conditions . [ prop : comp ] ( 1 ) if @xmath105 , then the zero - dimensional manifolds @xmath85 and @xmath106 are compact , and hence , finite sets . + ( 2 ) for @xmath107 , the topological boundary of the one - dimensional manifold @xmath85 is @xmath108 where the first union is taken over all @xmath109 , with @xmath110 , and the second over all @xmath111 , such that @xmath112 . similarly , when @xmath113 , we have @xmath114 the part ( 1 ) in the previous proposition enables us to define the homomorphism between morse and floer homology . denote by @xmath115 ( i.e. @xmath116vector spaces over the sets of generators of corresponding indices ) . let @xmath117 and @xmath118 denote the corresponding morse and floer homology groups . denote : @xmath119 and define @xmath120 it follows from the standard cobordism arguments and the part ( 2 ) of the proposition [ prop : comp ] that the homomorphisms @xmath121 and @xmath122 define the homomorphisms @xmath123 on the homology level . [ thm : iso_approx ] the homomorphisms @xmath124 and @xmath125 are isomorphisms and it holds @xmath126 _ proof : _ the proof relies on standard cobordism arguments . the auxiliary one - dimensional manifolds are : @xmath127\rightarrow u , \\ \gamma_{+}:[0,+\infty)\rightarrow u , \\ u:{\mathbb r}\times [ 0,1]\rightarrow t^*m \\ \frac{d\gamma_{\pm}}{dt}=-\nabla f(\gamma_{\pm } ) , \\ \frac{\partial u}{\partial s}+ j(\frac{\partial u}{\partial t}-x_{\rho_rh}(u))= 0 , \\ \gamma_-(-\infty)=p , \ ; \gamma_+(+\infty)=q , \\ u(s,0)\in o_m , \ ; u(s,1)\in\upsilon,\\ u(\pm\infty , t)=\gamma_{\pm}(0 ) \end{array } \right . \right\},\ ] ] and @xmath128\times[0,1]\rightarrow t^*m , \\ u_{+}:[0,+\infty)\times[0,1]\rightarrow t^*m , \\ \gamma:[-\varepsilon,\varepsilon]\rightarrow u \\ \frac{d\gamma}{dt}=-\nabla f(\gamma ) , \\ \frac{\partial u_{\pm}}{\partial s}+ j(\frac{\partial u_{\pm}}{\partial t}-x_{\rho_rh}(u_{\pm}))= 0 , \\ u_\pm(s,0),u_\pm(0,t)\in o_m,\;u_\pm(s,1)\in\upsilon , \\ u_{\pm}(0,1)=\gamma(\pm\varepsilon ) , \\ u_-(-\infty , t)=x(t ) , \\ u_+(+\infty , t)=y(t ) , \end{array } \right . \right\}\ ] ] unlike before , in the equations ( [ eq : aux1 ] ) and ( [ eq : aux2 ] ) , the smooth cut - off function @xmath129 $ ] is symmetric : @xmath130 the rest of the proof is similar to the proof of theorem 6 in @xcite . for two morse functions @xmath131 and @xmath132 , morse homologies @xmath133 and @xmath134 are canonically isomorphic ( see @xcite ) . similarly , for two hamiltonians @xmath135 and @xmath136 , the corresponding floer homologies @xmath137 and @xmath138 are isomorphic . denote these canonical isomorphisms by @xmath139 denote by @xmath140 the isomorphisms defined in ( [ eq : hom_h ] ) . [ thm : comm_approx ] the diagram @xmath141 commutes . _ proof : _ the proof is also based on cobordism arguments , so we will only give the description of the auxiliary one - dimensional manifold that we use here : @xmath142\times[0,1]\rightarrow t^*m , \\ \gamma:[0,+\infty)\rightarrow u , \\ \frac{\partial u}{\partial s}+ j(\frac{\partial u}{\partial t}-x_{\rho_rh_{\alpha\beta}^{\lambda}}(u))= 0 , \\ u(-\infty , t)=x_\alpha(t),\\ u(s,0 ) , u(0,t)\in o_m,\,u(s,1)\in\upsilon , \\ \frac{d\gamma}{dt}=-\nabla f_{\alpha\beta}^{\lambda}(\gamma ) , \\ \gamma(+\infty)=p_\beta,\\ \gamma(0)=u(0,1 ) , \end{array } \right . \right\}\ ] ] where @xmath143 is a homotopy connecting @xmath144 and @xmath145 . here @xmath146 is a smooth homotopy connecting @xmath135 and @xmath136 , and similarly for @xmath147 , while @xmath148 is as in ( [ eq : rho_symmetric ] ) . in order to define floer homology for the open set as a direct limit of floer homologies for the approximations , kasturirangan and oh defined a partial ordering on the set of approximations as : @xmath149 the function @xmath150 is defined by @xmath151 on @xmath14 , where @xmath152 is a smooth function such that @xmath153 ( recall that @xmath154 is exact ) and @xmath155 is a canonical projection . since @xmath17 is fixed , one has to vary the almost complex structure @xmath156 to obtain a transversal position . denote by @xmath157 an almost complex structure corresponding to @xmath154 and denote by @xmath158 a canonical homomorphism that satisfies : @xmath159 for given triple @xmath160 sufficiently close to @xmath9 ( see @xcite ) . since we want to establish an isomorphism between floer homology and morse homology for a fixed morse function , we will vary riemannian metric , so the term `` generic choices '' in the proposition [ prop : trans_cond ] refers to an almost complex structure @xmath156 and riemannian metric @xmath161 . fix a hamiltonian function @xmath17 and a morse function @xmath74 . for a lagrangian approximation @xmath154 , choose an almost complex structure @xmath157 and a riemannian metric @xmath162 such that all the transversality conditions are fulfilled , i.e. the sets @xmath163 , @xmath164 , @xmath165 and @xmath106 are manifolds for all @xmath166 and all hamiltonian paths @xmath167 with boundaries on @xmath27 and @xmath154 . since for two riemannian metric @xmath162 and @xmath168 there is a canonical isomorphism @xmath169 satisfying @xmath170 we can define morse homology @xmath171 as a direct limit : @xmath172 consider a diagram : @xmath173 where we use the abbreviations @xmath174 and so on . [ prop : var ] the diagram ( [ eq : diag_ups_g ] ) commutes . _ proof : _ the commutativity of ( [ eq : diag_ups_g ] ) is equivalent to the identity @xmath175 in order to prove ( [ eq : aux_comm ] ) , let us fix @xmath176 and define * @xmath177 to be a monotone homotopy for @xmath178 such that @xmath179 and @xmath180 the corresponding family of almost complex structures such that all transversality conditions hold ; * @xmath181 to be a homotopy of riemannian metrics such that @xmath182 recall that the homomorphism @xmath183 at the chain level ( we denoted by @xmath184 the induced homomorphism in homology ) is defined via the number of the set @xmath185 and the homomorphism @xmath186 via the number of @xmath187\rightarrow t^*m , \\ \frac{\partial u}{\partial s}+ \widetilde j_s(\frac{\partial u}{\partial t}-x_h(u))= 0 , \\ u(s,0)\in o_m,\,u(s,1)\in\widetilde{\upsilon}_s\\ u(-\infty , t)=x(t),\;u(+\infty , t)=y(t ) . \end{array}\right.\right\}\ ] ] for two critical points @xmath188 of morse function @xmath74 define the following auxiliary manifolds @xmath189\rightarrow t^*m \\ \gamma_1:(-\infty,0]\rightarrow u,\gamma_2:[0,+\infty)\rightarrow u \\ \frac{\partial u}{\partial s}+ \widetilde j_s(\frac{\partial u}{\partial t}-x_h(u))= 0 \\ \frac{d\gamma_j}{ds}=-\nabla_{\tilde g_s } f(\gamma ) , \;j=1,2\\ u(s,0),u(\pm\infty , t)\in o_m,\,u(s,1)\in\widetilde\upsilon_s , \\ \gamma_1(0)=u(-\infty,1),\;\gamma_2(0)=u(+\infty,1 ) \end{array}\right.\right\}\ ] ] and @xmath190 for @xmath191 , the set @xmath192 is a zero - dimensional and @xmath193 is an one - dimensional smooth manifold . the topological boundary of @xmath193 consists of the union of the following components : @xmath194 in @xmath195 , @xmath196 and @xmath197 are hamiltonian paths with @xmath198 , @xmath199 , @xmath200 . the loss of compactness corresponding to the boundary points from @xmath201 , @xmath202 and @xmath203 happens when @xmath204 is bounded , and @xmath195 occurs when @xmath205 . counting the objects from @xmath201 and @xmath202 gives us the maps of the type @xmath206 and @xmath207 , for a map @xmath208 defined via the number of zero component of manifold @xmath193 . therefore , the maps @xmath76 and @xmath209 , defined by the number of zero - dimensional boundary components @xmath195 and @xmath203 respectively , are the same in the homology . the map @xmath76 is obviously the same as the map @xmath210 , i.e. the right hand side in ( [ eq : aux_comm ] ) , although the riemannian metrics defining the gradient lines in mixed objects are not the same . this follows from standard parameter independence arguments . so we need to prove that that @xmath211 in the homology level . by standard cobordism arguments , we can conclude that the map @xmath209 does not depend on the choice of @xmath212 . by letting @xmath213 , we conclude that , in the holomoly level , the map @xmath209 is equal to the map @xmath214 , defined by the number of the set @xmath215\rightarrow t^*m \\ \gamma_1:(-\infty,0]\rightarrow u,\gamma_2:[0,+\infty)\rightarrow u \\ \frac{\partial u}{\partial s}+ \widetilde j_s(\frac{\partial u}{\partial t})= 0 \\ \frac{d\gamma_j}{ds}=-\nabla_{\tilde g_s } f(\gamma ) , \;j=1,2\\ u(s,0),u(\pm\infty , t)\in o_m,\,u(s,1)\in\widetilde\upsilon_s , \\ \gamma_1(0)=u(-\infty,1),\;\gamma_2(0)=u(+\infty,1 ) \end{array}\right.\right\}.\ ] ] but since the above @xmath216 is holomorphic , it holds @xmath217 } u^*\omega=\iint_{\partial(\mathbb{r}\times[0,1 ] ) } u^*\theta\stackrel{(i)}{=}\\ & \int_{-\infty}^{+\infty}u(1,s)^*dh_{\widetilde{\upsilon}_s } = h_{{\upsilon}_b}(u(+\infty,1))-h_{{\upsilon}_a}(u(-\infty,1))\stackrel{(ii)}{=}0.\end{aligned}\ ] ] the equality @xmath218 follows from ( [ exactness ] ) and from the fact that @xmath216 maps @xmath219 $ ] , @xmath220 $ ] and @xmath221 to @xmath27 . the equality @xmath222 holds since the function @xmath223 are zero on @xmath27 ( see @xcite ) . we conclude that the maps @xmath214 and @xmath224 are the same , therefore , ( [ eq : aux_comm ] ) holds . from proposition [ prop : var ] and the fact that all the maps in ( [ eq : diag_ups_g ] ) commute , it follows that there exists a direct limit isomorphism @xmath225 similarly , by considering the commutative diagram @xmath226 we obtain the direct limit isomorphism @xmath227 the induced maps @xmath228 and @xmath229 are isomorphisms and it holds @xmath230 _ proof : _ from theorem [ thm : iso_approx ] we have @xmath231 let @xmath232 be the representative of the class @xmath233\in hm_k(f , u)$ ] . we have @xmath234)=\psi([\phi^a(p_a)])=[\psi^a(\phi^a(p_a))]=[p_a],\ ] ] and similarly for the second equality . from the canonical isomorphisms @xmath235 and the commutativity of @xmath236 one obtains an isomorphism : @xmath237 similarly , we have @xmath238 the diagram @xmath239 commutes . _ proof : _ recall that the diagram ( [ diag : s : t ] ) commutes for all approximations close enough to @xmath9 and for generic choices . so we have @xmath240)= \mathbf{t}_{\alpha\beta}\l(\l[\psi_\alpha^a(x_a)\r]\r)= \l[t^a_{\alpha\beta}\l(\psi_\alpha^a(x_a)\r)\r]=\\ & [ \psi_\beta^a\l(s^a_{\alpha\beta}(x_a)\r)]=\psi_\beta\l([s^a_{\alpha\beta}(x_a)]\r)=\psi_\beta\circ \mathbf{s}_{\alpha\beta}([x_a]),\end{aligned}\ ] ] for every @xmath241\in hf_k(h_\alpha , u : m)$ ] . this proves theorem [ thm : pss ] . in this section we construct a product on floer homology for an open subset @xmath242 first , we define @xmath243 @xmath244 for every approximation @xmath63 . product ( [ product * ] ) is defined by counting pair of pants objects . we define a riemannian surface ( with a boundary ) @xmath245 as a disjoint union @xmath246\sqcup{\mathbb r}\times[0,1]\ ] ] with the identification @xmath247 for @xmath248 ( see figure [ pants_pic ] ) . + denote by @xmath249 , @xmath250 , @xmath251 the two `` incoming '' and one `` outgoing '' ends , such that @xmath252\times(-\infty,0],\\ & \sigma_3\approx [ 0,1]\times[0,+\infty),\\ \end{aligned}\ ] ] and by @xmath253 , @xmath254 . let @xmath255 $ ] denote the smooth cut off functions such that @xmath256 for @xmath257 , @xmath258 and @xmath259 we define a moduli space @xmath260 ( see figure [ pants_pic ] ) . that defines product @xmath243,width=302,height=113 ] for generic choices , @xmath261 is a smooth @xmath262dimensional manifold . we define a map @xmath243 at the chain level as @xmath263 for the generators @xmath196 and @xmath197 . here , @xmath264 denotes the ( modulo 2 ) number of elements of a zero - dimensional component of @xmath261 . we extend the product @xmath243 to @xmath265 by bilinearity . one can show that @xmath243 commutes with boundary maps and induces a product in homology ( [ product * ] ) . [ prop : circ_open ] the product @xmath243 defines a product on floer homology for open subset : @xmath242 in order to prove proposition [ prop : circ_open ] , we need the following lemma . recall that we denote by @xmath266 the homomorphism @xmath267 defined by ( [ eq : f_ab ] ) . here @xmath268 , etc . to emphasize the hamiltonian we are considering , we will sometimes write @xmath269 . [ lem ] for @xmath270 , @xmath271 it holds @xmath272 _ proof of proposition [ prop : circ_open ] _ : let us take @xmath273\in hf_*(h_1,u : m)$ ] and @xmath274\in hf_*(h_2,u : m)$ ] . the elements @xmath196 and @xmath197 belong to some spaces @xmath275 and @xmath276 . in general , @xmath277 and @xmath278 are not the same . since @xmath196 and @xmath279 represent the same element in @xmath280 we can take @xmath281 and @xmath282 as representatives of @xmath273 $ ] and @xmath274 $ ] respectively . so we can assume that @xmath196 and @xmath197 belong to some @xmath283 and @xmath284 , for the same @xmath63 . we now define a product @xmath243 on homology as @xmath285\circ[y]:=[x\circ y].\ ] ] we need to show that a product does not depend on representatives of a class . let @xmath286 and @xmath287 represent the same element in @xmath280 and similarly @xmath288 and @xmath289 in @xmath290 . this means that there exist @xmath291 such that @xmath292 let @xmath293 . we have @xmath294 which means that @xmath295 and @xmath296 represent the same element in @xmath297 , i.e. the product @xmath243 is well defined on homology for the open set . _ proof of lemma [ lem ] _ : the homomorphism @xmath266 is an isomorphism for @xmath277 , @xmath298 large enough . the inverse homomorphism is actually @xmath299 . this can be proved using the cobordism arguments similar to ones in the proof of the independence of floer homology with respect to the parameters ( hamiltonian , almost complex structure ) . therefore , ( [ eq : f+circ ] ) is equivalent to @xmath300 the proof of ( [ eq : f^-1 ] ) is similar to the proof of the proposition [ prop : var ] , so let us just specify the auxiliary one - dimensional manifold whose boundaries provide the corresponding homology maps . let @xmath301 be as in ( [ eq : mon_hom ] ) and @xmath286 , @xmath288 , @xmath302 be the solutions of @xmath303 for appropriate maslov indices of @xmath304 we define the auxiliary one - dimensional manifold @xmath305 to be the set of all pairs @xmath306 , where @xmath307 and @xmath308 is the solution of the equation @xmath309 where @xmath310 is depicted in the figure [ aux_mfd_pic ] , as well as corresponding boundary conditions . the almost complex structure @xmath311 is chosen to satisfy all the regularity conditions . , width=491,height=264 ] the following proposition establishes the ring structure pss isomorphism . [ prop : circ_cdot_pss ] for @xmath312 it holds @xmath313 where @xmath228 is the pss isomorphism ( [ pss_denote ] ) and @xmath314 is a product on morse homology ( see subsection [ moduli ] ) . _ proof : _ it follows from the definition of @xmath228 and proposition [ prop : circ_open ] that it is enough to show that @xmath315 for a fixed approximation @xmath63 and fixed riemannian metric defining the product @xmath314 . the equality ( [ eq : circ_cdot_app ] ) is equivalent to @xmath316 the latter equality follows from cobordism arguments similar to ones used in the proof of proposition [ prop : var ] and lemma [ lem ] . the auxiliary one - dimensional manifold we use here is explained by the figure [ aux_mfd2_pic ] . , width=529,height=207 ] let us recall the construction of the homology product on @xmath317 . let @xmath318 be morse functions defined on @xmath14 such that @xmath319 and @xmath320 intersect transversally for every critical point @xmath321 of @xmath322 . for @xmath323 , @xmath324 , we define the moduli space @xmath325 to be the set of all trees @xmath326 such that @xmath327\to u,\ ; \gamma_3:[0,+\infty)\to u,\\ \dot\gamma_j=-\nabla f_j(\gamma_j),\,j=1,2,3,\\ \gamma_i(-\infty)=p_i,\,i=1,2,\\ \gamma_3(+\infty)=p_3,\\ \gamma_1(0)=\gamma_2(0)=\gamma_3(0 ) \end{array}\right.\ ] ] , width=226,height=94 ] for generic choices these spaces are manifolds of dimension @xmath328 if @xmath329 denotes the mod 2 number of a zero dimensional component , then the product @xmath314 is defined at the chain level : @xmath330 as : @xmath331 on generators . since it commutes with the morse boundary operator , it is well defined at the homology level . by standard combinatorial and cobordism arguments one can show that it defines a ring structure on @xmath317 . now , we note that for every approximation @xmath63 , @xmath332 is a @xmath317module . as before , the external product is defined at the chain level @xmath333 let @xmath334 denotes a smooth family of hamiltonians such that @xmath335 for @xmath336 , @xmath337 we define the moduli space @xmath338 as a set of pairs @xmath339 such that @xmath340\to u,\,u:{\mathbb r}\times[0,1]\to t^*m,\\ \dot{\gamma}=-\nabla f(\gamma(t)),\\ \pa_su+j_\upsilon(\pa_tu - x_{h^s}(u))=0,\\ u(s,0)\in \upsilon,\,u(s,1)\in o_m,\,s\le 0,\\ u(s,0)\in o_m,\,u(s,1)\in\upsilon,\,s\ge 0,\\ \gamma(-\infty)=p,\\ u(-\infty , t)=x(1-t),\,u(+\infty , t)=y(t),\\ \gamma(0)=u(0,0 ) . \end{array}\right.\end{aligned}\ ] ] these spaces are @xmath341 dimensional manifolds and their zero dimensional component is compact . we define the product @xmath342 on the set of the generators of chain complexes as : @xmath343 where @xmath344 denotes the cardinality of the zero dimensional component of @xmath338 . using standard cobordism arguments , as above , one can show that @xmath342 induces a product in homology . similarly to @xcite it follows that @xmath332 is a @xmath317module , i.e. @xmath345 for all @xmath346 and @xmath347 . in order to have the products @xmath314 and @xmath342 well defined on a direct limit of morse and floer homology groups , one needs to check their compatibilities with homomorphisms @xmath348 and @xmath349 . let @xmath266 and @xmath224 be the homomorphisms that define the direct limit morse and floer homology groups , obtained by the number of ( [ eq : f_ab ] ) and ( [ eq : g_ab ] ) respectively . then it holds @xmath350 for all @xmath351 and @xmath352 . the proof is similar to the proof of the lemma [ lem ] . from ( [ eq : star_mod ] ) and ( [ eq : star_com ] ) it follows that @xmath314 and @xmath342 are well defined operations on @xmath36 and @xmath353 and that @xmath353 is a @xmath36module . since the action functional decreases along the strips that define the boundary operator @xmath360 the boundary operator descends to @xmath361 and defines @xmath362 denote the corresponding homology groups by @xmath363 . [ limit ] let @xmath374\in hm_*(f , u : g_\upsilon)\setminus\{0\}$ ] . then there exists an approximation @xmath375 such that @xmath376,h)=c_{\overline{\upsilon}}(g_{\upsilon\overline{\upsilon}}(\alpha),h)\ ] ] for all @xmath377 . _ proof : _ we have the following commutative diagram @xmath378 let us take @xmath374\in hm_k(f , u)\setminus\{0\}$ ] and @xmath379 such that @xmath380)\in{\operatorname{im}}(i^\lambda_*).\ ] ] then , there exists @xmath381\in hf_k^\lambda(h , u : m)$ ] such that @xmath380)=i^\lambda_*([\widetilde{\alpha}]).\ ] ] from the definition of a direct limit we conclude that @xmath382 for some @xmath63 and @xmath383 . since @xmath380)=[pss_\upsilon(\alpha)]=[i^\lambda_{\upsilon'*}(\widetilde{\alpha})]=i^\lambda_*[\widetilde{\alpha}],\ ] ] we find that @xmath384 for some @xmath385 which is closer to @xmath9 than @xmath63 and @xmath383 , @xmath386 and @xmath387 . using the commutativity of the above diagram we get @xmath388 thus , @xmath389 we find that @xmath390,h).\ ] ] if we take @xmath391 and @xmath379 such that @xmath392 then @xmath393 for some @xmath394 . again , from the definition of a direct limit we obtain the equalities @xmath380)=[pss_\upsilon(\alpha)]=[i^\lambda_{\upsilon*}(\overline{\alpha})]=i^\lambda_*[\overline{\alpha}].\ ] ] this gives us the inequality @xmath395,h)\leq c_\upsilon(\alpha , h).\ ] ] in the quotient space @xmath171 elements @xmath396 and @xmath397 represent the same element . from ( [ spektar_klase_1 ] ) and ( [ spektar_klase_2 ] ) we derive @xmath398,h)=c_u([g_{\upsilon\overline{\upsilon}}(\alpha)],h)\leq c_{\overline{\upsilon}}(g_{\upsilon\overline{\upsilon}}(\alpha),h),\ ] ] and all inequalities become equalities . since spectral invariants decrease as @xmath399 , i.e. @xmath400 for every @xmath401 , they all become equal to @xmath402,h)$ ] , starting from some @xmath375 . [ thm : cont_invar ] relative spectral invariants for an open set @xmath14 @xmath403 are continuous with respect to the hamiltonian @xmath17 , @xmath404 ( here @xmath405 denotes the generator of zero homology group , @xmath406 ) . _ proof : _ first , we prove that @xmath407 is continuous with respect to hamiltonian @xmath17 . let us fix a good approximation @xmath63 and let @xmath17 and @xmath408 be two hamiltonians satisfying @xmath409 . consider the linear homotopy @xmath410 ( we can approximate this linear homotopy with a regular one ) . the isomorphism @xmath411 is defined by a number of the holomorphic strips that connect @xmath412 and @xmath413 : @xmath414\to t^*m\left|\begin{array}{l } \frac{\partial u}{\partial s}+ j_\upsilon(\frac{\partial u}{\partial t}-x_{h^s}(u))= 0,\\ u(s,0)\in o_m,\,u(s,1)\in\upsilon , \\ u(-\infty , t)=x(t),\ , u(+\infty , t)=y(t)\\ \end{array}\right.\right\}.\end{aligned}\ ] ] if there exists @xmath415 that connects @xmath196 and @xmath197 then it holds @xmath416 ( see @xcite for details ) . it follows @xmath417 for @xmath347 we can define @xmath418 it is obvious that @xmath419 relates @xmath420 and @xmath421 , i.e. @xmath422 from ( [ ineq_inv ] ) it follows @xmath423 since @xmath424 we get the inequality @xmath425 that holds for all @xmath426 . if we write the same inequality for the generator of zero homology group , we derive the continuity of relative spectral invariants for approximations : @xmath427 from this inequality and from the lema [ limit ] it easily follows that relative spectral invariants for an open subset are continuous with respect to the hamiltonian @xmath17 . _ proof : _ choose a hamiltonian @xmath431 that is regular , smooth and close enough to @xmath432 : @xmath433 we prove that a product @xmath243 descends to a product on filtered homologies @xmath434 let us take a smooth family of hamiltonians @xmath435\times t^*m\to{\mathbb r}$ ] such that @xmath436 we can choose @xmath208 such that @xmath437,\ ] ] and @xmath438 elsewhere . assume that for @xmath439 and @xmath440 there exists @xmath441 for some @xmath442 . here , @xmath216 is pseudo holomorphic pants for a hamiltonian @xmath208 @xmath443 using the relations @xmath444 stoke s formula and properties of a hamiltonian @xmath208 it follows @xmath445 now , from proposition [ prop : circ_cdot_pss ] we obtain the inequality @xmath446 since spectral invariants are continuous with respect to the hamiltonian we get the claimed inequality . + taking a direct limit in the previous proposition and using lema [ limit ] we obtain the following theorem . recall the definition of spectral invariants for periodic orbit floer homology . denote by @xmath451 and @xmath452 floer homology for periodic orbits in @xmath2 and morse homology for the morse function @xmath453 respectively . let @xmath454 stands for pss isomorphism for periodic orbits , defined in @xcite @xmath455 and let @xmath456 . the filtration in floer homology for periodic orbits is given by the standard action functional @xmath457 which is well defined in the cotangent bundle setting . filtered floer homology groups are homology groups of a chain complex generated by @xmath458 where @xmath459 denotes the @xmath460vector space over the set of periodic hamiltonian @xmath461orbits in @xmath2 of conley zehnder index @xmath462 . denote by @xmath463 the corresponding filtered group and , again , by @xmath464 the map induced by inclusion . the homomorphisms defined using `` chimneys '' are considered by abbondandolo and schwarz in @xcite ( in the context of floer homology of cotangent bundles and the ring - isomorphism with the homology of the loop space ) and albers in @xcite ( in the construction of the comparison homomorphisms between lagrangian and hamiltonian floer homology ) . the construction of a chimney is different in our situation , due to the boundary conditions . for @xmath468 and @xmath469 define the manifold of chimneys as : @xmath470 for generic choices , @xmath471 is a smooth manifold of dimension @xmath472 , where @xmath473 denotes the conley - zehnder index of a loop @xmath277 . we define conley - zehnder index by requiring that the dimension of caps @xmath474/_{(0,t_1)\sim(0,t_2)}\to t^*m\l|\begin{array}{l}\pa_su+j(\pa_tu - x_{\rho_rh}\circ u)=0,\\ u(+\infty , t)=a(t ) \end{array}\r.\r\}.\ ] ] is given by @xmath475 . let @xmath277 be a periodic orbit . if there exists @xmath480 , let @xmath197 be a loop defined as @xmath481 since @xmath482 , we have @xmath483 . therefore , we have @xmath484 so @xmath485 it follows that @xmath478 defines the mapping @xmath486 which also descends to the homology level : @xmath487 ( see also @xcite ) . the diagram @xmath488_{\imath_*^{\lambda}}\ar[r]^-{\chi^{\lambda } } & hf^{\lambda}_k(t^*m : h , j_\upsilon ) \ar[d]_{\jmath_*^{\lambda}}\\ hf_k(o_m,\upsilon : h , j_\upsilon)\ar[r]^-{\chi } & hf_k(t^*m : h , j_\upsilon)}\ ] ] commutes . define @xmath493 this homomorphism also descends to the homology level @xmath494 since it commutes with the boundary operators . as above , one can show that it also induces a homomorphism on the filtered homology level , and that the corresponding diagram ( analogous to ( [ eq : diag : chi , i , j ] ) ) commutes . let @xmath74 be morse function on @xmath14 as above and @xmath495 morse function on @xmath2 . let @xmath496 be a hamiltonian function and @xmath497 a homomorphism induced by inclusion . suppose all the choices are generic . the diagram @xmath498_{\imath_*^{\lambda}}\ar[r]^-{\chi^{\lambda } } & hf_k^{\lambda}(t^*m : h , j_\upsilon ) \ar[d]_{\jmath_*^{\lambda}}\\ hf_k(o_m,\upsilon : h , j_\upsilon)\ar[d]_{{\operatorname{pss}}_\upsilon^{-1}}\ar[r]^-{\chi } & hf_k(t^*m : h , j_\upsilon ) \ar[d]_{{\operatorname{pss}}^{-1}}\\ hm_k(f , u : g_\upsilon)\ar[r]^{\imath _ * } & hm_k(\tilde f , t^*m : g_\upsilon)}\ ] ] commutes . in the statement of the proposition @xmath53 denotes the composition of the inclusion @xmath499 and the canonical isomorphism @xmath500 where a morse function @xmath501 is the following appropriate extension of @xmath74 to @xmath2 . consider a tubular neighbourhood @xmath502 of @xmath14 . first , we extend @xmath74 to the vector bundle @xmath503 over @xmath14 , and obtain the morse function @xmath504 such that @xmath505 then we extend the morse function @xmath504 defined on the open subset @xmath503 to the morse function @xmath501 on @xmath2 ( see @xcite for details ) . we chose the extension @xmath501 in such way that there are no trajectories for the negative gradient flow of @xmath501 leaving @xmath14 . under this assumptions the morse complex @xmath506 is a subset of the morse complex @xmath507 and the inclusion of these complexes becomes the homomorphism on the homology level @xmath508 . _ proof : _ the upper diagram is ( [ eq : diag : chi , i , j ] ) . the lower diagram is @xmath509_{{\operatorname{pss}}_\upsilon^{-1}}\ar[r]^-{\chi } & hf_k(t^*m : h , j_\upsilon)\ar[d]_{{\operatorname{pss}}^{-1}}\ar[r]^-{{\operatorname{id } } } & hf_k(t^*m : h , j_\upsilon)\ar[d]_{{\operatorname{pss}}^{-1}}\\ hm_k(f , u : g_\upsilon)\ar[r]^{\imath_0 } & hm_k(f , t^*m : g_\upsilon)\ar[r]^{t_{f,\widetilde{f } } } & hm_k(\tilde{f},t^*m : g_\upsilon)}\ ] ] the commutativity of the diagram on the right side was resolved in @xcite . to prove the commutativity of the diagram on the left side we will show that @xmath510 in order to do that using the usual cobordism arguments , we consider the following two auxiliary manifolds . let @xmath148 be the symmetric cut - off function ( [ eq : rho_symmetric ] ) . let @xmath83 be the critical point of a morse function @xmath74 and @xmath511 the critical point of a morse function @xmath501 . fix @xmath512 and define @xmath513\to u,\;\dot{\gamma_1}=-\nabla f(\gamma_1),\\ \gamma_2:[0,+\infty)\to t^*m,\;\dot{\gamma_2}=-\nabla f(\gamma_2),\\ u:\sigma\to t^*m,\;\pa_su+j(\pa_tu - x_{\rho_r(s)h}\circ u)=0,\\ u(s,0)\in o_m,\,u(s,1)\in\upsilon\;\mbox{for}\;s\le 0,\\ \gamma_1(-\infty)=p,\ , \gamma_1(0)=u(-\infty , t),\\ u(+\infty , t)=\gamma_2(0),\ , \gamma_2(+\infty)=q , \end{array}\r.\r\}\end{aligned}\ ] ] ( see figure [ mixed_pic ] ) . define also @xmath514 for @xmath515 and generic choices , @xmath516 is a smooth one - dimensional manifold with topological boundary that can be identified with @xmath517 where @xmath518 here @xmath519\times [ 0,1]\to t^*m,\\ \pa_su+j(\pa_tu - x_{\tilde\rho h}\circ u)=0,\\ u(s,0)=u(s,1),\\ \gamma:[0+\infty)\to t^*m,\,\dot{\gamma}=-\nabla f(\gamma),\\ u(0,t)=\gamma(0),\ , u(-\infty , t)=a(t),\ , \gamma(+\infty)=q , \end{array } \r.\r\}\ ] ] i.e. it is the space of combined object defining a pss isomorphism for periodic orbits and @xmath520 . denote by @xmath521 the homomorphism obtained by counting the elements of the zero dimensional manifold @xmath522 . since the number of the boundary of the one - dimensional manifold @xmath516 is even , i.e. zero in @xmath523 , and the maps @xmath524 and @xmath525 are of the form @xmath526 the homomorphisms @xmath527 and @xmath528 are equal in the homology . by standard cobordism argument one can show that the mapping @xmath527 does not depend on @xmath212 . letting @xmath213 , we obtain that , at the homology level , the mapping @xmath527 counts the objects @xmath529 where @xmath216 is a _ holomorphic _ map from @xmath530 to @xmath2 . the boundary of @xmath216 is on @xmath531 , so it must be constant due to the exactness of both lagrangian manifold and the fact that @xmath532 ( see the proof of lemma [ lem ] for the details ) . hence @xmath527 is chain homotopic to the map obtained by counting the pairs @xmath533 with properties : @xmath534\to u,&\dot{\gamma_1}=-\nabla f(\gamma_1),&\gamma_1(-\infty)=p,\\ \gamma_2:[0,+\infty)\to t^*m , & \dot{\gamma_2}=-\nabla f(\gamma_2),&\gamma_2(+\infty)=q\\ \gamma_1(0)=\gamma_2(0 ) . & & \end{array } \right.\ ] ] since we know that @xmath535 , @xmath536 is a negative gradient trajectory of @xmath501 connecting two critical points of the same morse index . number of such pairs is equal 1 in case @xmath537 and 0 otherwise . thus , @xmath527 is chain homotopic to the homomorphism @xmath508 . on the other hand , the mapping @xmath528 is exactly the homomorphism @xmath538 , so the claim follows . now , for two generic almost complex structures @xmath157 and @xmath539 , denote by @xmath540 a canonical isomorphism of floer homologies for periodic orbits : @xmath541 that satisfies @xmath542 as before , define floer homology for periodic orbits as a direct limit @xmath543 the filtered floer homology is defined as : @xmath544 _ proof . _ the commutativity of ( [ eq : diag_chi_ab ] ) is equivalent to : @xmath548 the proof of the above equality is similar to the proofs of the proposition [ prop : var ] and lemma [ lem ] . the auxiliary one - dimensional manifold will be the set of the pairs @xmath306 , where @xmath307 , @xmath216 is a chimney with the properties depicted in figure [ aux_chimney_pic ] . [ cor : com_dir_lim ] the diagram @xmath551_{\imath_*^{\lambda}}\ar[r]^-{\chi^{\lambda } } & hf_k^{\lambda}(t^*m : h ) \ar[d]_{\jmath_*^{\lambda}}\\ hf_k(h , u : m)\ar[d]_{{\operatorname{pss}}_u^{-1}}\ar[r]^-{\chi } & hf_k(t^*m : h ) \ar[d]_{{\operatorname{pss}}^{-1}}\\ hm_k(f , u)\ar[r]^{\imath _ * } & hm_k(\tilde f , t^*m)}\ ] ] commutes . _ proof : _ from the commutativity of ( [ 2diags_u ] ) ones easily gets @xmath553 ( see the proof of theorem [ thm : c_uv ] below or @xcite for more details ) so the claim follows . one can obtain the inequality of similar type by using the homomorphism @xmath554 . the corresponding commutative diagram is @xmath555_{\jmath_*^{\lambda}}\ar[r]^-{\tau^{\lambda } } & hf_{k - n}^{\lambda}(h , u : m ) \ar[d]_{\imath_*^{\lambda}}\\ hf_k(t^*m : h)\ar[d]_{{\operatorname{pss}}^{-1}}\ar[r]^-{\tau } & hf_{k - n}(h , u : m ) \ar[d]_{{\operatorname{pss}}_u^{-1}}\\ hm_k(\tilde f , t^*m)\ar[r]^{\imath _ ! } & hm_{k - n}(f , u)},\ ] ] where @xmath556 is the map obtained by inclusion map and poincar duality map : @xmath557 from this commutativity , we have the following in @xcite oh considered a spectral invariant @xmath560 ( the notions are the same as in the subsection [ subsec : inv_open ] ) . if @xmath45 are two open subset of @xmath3 and @xmath561 is surjective , oh proved that @xmath562 we can prove slightly more precise statement , the inequality for any homology class ( with @xmath523 coefficients ) , using the pps isomorphism for an open subset in the proof . [ thm : c_uv ] let @xmath45 be two open subset of @xmath3 and let @xmath56 ( the homomorphism induced by inclusion @xmath57 ) be surjective . let @xmath48 be as in ( [ eq : def_inv_c_open ] ) . for @xmath58 it holds : @xmath59 _ proof : _ let @xmath563 denote the inclusion homomorphisms defined by oh in @xcite . the following diagram is commutative : @xmath564_{\imath_{*u}^{\lambda}}\ar[r]^- { \imath_{*uv}^{\lambda } } & hf_k^{\lambda}(h , v : m ) \ar[d]_{\imath_{*v}^{\lambda}}\\ hf_k(h , u : m)\ar[d]_{{\operatorname{pss}}_u^{-1}}\ar[r]^-{\imath_{*uv } } & hf_k(h , v : m ) \ar[d]_{{\operatorname{pss}}^{-1}_v}\\ hm_k(f , u)\ar[r]^{\jmath_{*uv } } & hm_k(f , v).}\ ] ] the commutativity of the upper diagram is proven in @xcite . to prove the commutativity of the lower one , it is enough to prove the commutativity of @xmath565_{{\operatorname{pss}}_{\upsilon^u}^{-1}}\ar[r]^-{\imath^{(h , j)}_{*uv } } & hf_k(o_m,\upsilon^v : h , j ) \ar[d]_{{\operatorname{pss}}^{-1}_{\upsilon^v}}\\ hm_k(f , u : g)\ar[r]^{\jmath_{*uv } } & hm_k(f , v : g),}\ ] ] for all @xmath566 close enough to @xmath9 and @xmath567 close enough to @xmath568 . here @xmath569 is the inclusion map also defined in @xcite . take @xmath273 $ ] in @xmath570 . it holds @xmath571))=\sum_{p\in cm_k(v)}n(x , p)[p].\ ] ] on the other hand , we have @xmath572))= \sum_{p\in cm_k(u)}n(x , p)\jmath_{*uv}([p]),\ ] ] which is the same as ( [ eq : aux4 ] ) if @xmath573 is surjective . let @xmath574 if @xmath575 , then @xmath576 , for @xmath577 , so , from the commutativity of ( [ eq:2diags ] ) we have @xmath578 we conclude that @xmath579 , therefore @xmath580 so by taking an infimum over @xmath581 , we obtain @xmath59 a. abbondadolo , m. schwarz , _ notes on floer homology and loop space homology _ , morse theoretic methods in nonlinear analysis and in symplectic topology , nato sci . ser , ii math . * 217 * , springer , dordrecht , pp . 74 - 10 , 2006 . s. piunikhin , d. salamon , m. schwarz , _ symplectic floer donaldson theory and quantum cohomology _ , in : contact and symplectic geometry , publ . newton instit . 8 , cambridge univ . press , cambridge , pp . 171200 , 1996 .

we use the coordinate bethe ansatz to exactly calculate matrix elements between eigenstates of the lieb liniger model of one - dimensional bosons interacting via a two - body delta - potential . we investigate the static correlation functions of the zero - temperature ground state and their dependence on interaction strength , and analyze the effects of system size in the crossover from few - body to mesoscopic regimes for up to seven particles . we also obtain time - dependent nonequilibrium correlation functions for five particles following quenches of the interaction strength from two distinct initial states . one quench is from the non - interacting ground state and the other from a correlated ground state near the strongly interacting tonks - girardeau regime . the final interaction strength and conserved energy are chosen to be the same for both quenches . the integrability of the model highly constrains its dynamics , and we demonstrate that the time - averaged correlation functions following quenches from these two distinct initial conditions are both nonthermal and moreover distinct from one another .

1601.00434

the lieb liniger model of a one - dimensional bose gas with repulsive delta - function interactions is a paradigmatic example of an exactly solvable continuous , integrable many - body quantum system @xcite . in particular , it has served as the context for the development of theoretical tools that have subsequently been widely applied in the study of integrable systems , such as the so - called `` thermodynamic bethe ansatz '' functional representation , which provides the exact equation of state , excitation spectrum @xcite , and bulk parameters @xcite of the system in the thermodynamic limit . however , the calculation of correlation functions from the exact solutions provided by the bethe ansatz is notoriously difficult . at zero temperature , exact closed - form solutions for some equilibrium correlation functions are known in the tonks girardeau limit of infinite interaction strength @xcite . this comparatively tractable limit also allows for some strong - coupling expansion results for large but finite interactions @xcite . in the opposite weakly interacting quasi - condensate regime , a mean - field approach can be used to describe the system @xcite and a bogoliubov method can be used to determine the low - lying excitation spectrum @xcite , relying on small density fluctuations . fewer results are available for intermediate interaction strengths , away from the strongly - interacting and weakly - interacting regimes . the development of the luttinger liquid description of quantum fluids @xcite and the related formalism of conformal field theory @xcite have lead to the prediction of power - law scaling for first - order correlations at large distances , with an exponent given in terms of the equation of state that is known exactly from the thermodynamic bethe ansatz @xcite . the algebraic bethe ansatz provides a determinantal representation of correlations , from which their asymptotic behavior can be extracted @xcite . more recently , exact expressions for local second- and third - order correlations @xcite , together with exact results for the one - body correlation function at asymptotically short distances @xcite in terms of the equation of state have been derived . away from the asymptotic short- and long - range regimes , the behavior of correlation functions is less well known . for intermediate interaction strengths and arbitrary length scales one must resort to numerics to determine the correlation functions . results for the latter have been obtained using numerical methodologies including quantum monte carlo @xcite , and density matrix renormalization group approaches @xcite . a recently developed , integrability - based approach combines the decomposition of correlation functions into sums over matrix elements ( form factors ) of certain simple operators between bethe ansatz eigenstates @xcite . this approach has generated results , for example , for static and dynamical equilibrium correlations at zero and finite temperature for systems of up to @xmath0 particles @xcite . other finite temperature results for correlation functions have been obtained using imaginary time stochastic gauge methods @xcite , taking the non - relativistic limit of a relativistic field theory @xcite , utilizing fermi bose mapping for the strongly interacting gas @xcite , employing perturbative expansions in temperature and interaction strength @xcite , as well as combining the thermodynamic bethe ansatz with the hellmann feynman theorem @xcite . experiments with ultracold quantum gases are able to realize effectively one - dimensional systems by tightly confining the gas in two of the three spatial dimensions , either using optical lattice potentials or atom - chip traps @xcite . these experiments are now probing the predictions of the lieb liniger model . the configurability of quantum - gas experiments allows for so - called _ quenches _ of the system , in which hamiltonian parameters of the system are abruptly changed , and thus for the study of the lieb - linger model out of equilibrium , providing even greater challenges for theory . the dynamically evolving correlations of the lieb liniger gas in nonequilibrium scenarios are currently a topic of significant interest , and a number of theoretical approaches have been applied . notable examples include exact diagonalization under a low momentum cutoff @xcite , mapping of the hard - core tonks girardeau gas to free spinless fermions @xcite , phase - space methods @xcite , dynamic bogoliubov - like approximations @xcite and tensor - network methods @xcite . references @xcite employed nonperturbative approximative functional - integral methods , while in ref . @xcite a dynamical luttinger - liquid approach was taken . other calculations make explicit use of the integrability of the system . these are based on various bethe ansatz approaches , and include utilizing fermi - bose mapping @xcite and strong coupling expansions of the coordinate bethe ansatz wave function @xcite , combining the algebraic bethe ansatz with other numerical methods @xcite , and using the yudson contour - integral representation for infinite - length systems @xcite . recently , it was conjectured that the dynamics following an interaction strength quench are captured by a thermodynamic bethe ansatz saddle point state and excitations around it the so - called quench action approach @xcite . in the spirit of the methodology of refs . @xcite , gritsev _ et al . _ @xcite investigated a quench from @xmath1 by combining algebraic bethe ansatz expressions for form factors with truncated sums over states , and employing monte carlo summation over the eigenstate components of the initial state . in this paper we take a different approach , and calculate correlation functions of the lieb - linger model , both in and out of equilibrium , by calculating matrix elements between lieb liniger eigenstates directly within the coordinate bethe ansatz formalism . given the known expressions for the coordinate - space forms of lieb liniger eigenstates , we generate symbolic expressions for matrix elements of operators between these states in terms of the bethe rapidities . the numerically obtained values of the rapidities can then be substituted to yield essentially numerically exact values for the matrix elements . in our previous work we applied this methodology to quenches from the ideal gas ground state to positive @xmath2 for up to @xmath3 particles @xcite . in sec . [ sec : model ] we provide the details of the methodology , and describe how it can be used to calculate the matrix elements of the lieb liniger eigenstates . these symbolic expressions , and thus the computational cost of evaluating them , grow combinatorially with particle number , restricting the method to systems of only a few particles . however for small particle numbers @xmath4 we obtain numerically exact results for ground - state correlations , which are described in sec . [ sec : groundstate ] . our results demonstrate that local correlations in the strongly interacting regime are already close to their thermodynamic - limit values for these few - body to mesoscopic systems . an additional advantage of our methodology is that it can also calculate overlaps between lieb liniger eigenstates corresponding to _ any _ two interaction strengths , which allows us to study the dynamics of quenches of the interaction strength between arbitrary values . in sec . [ sec : dynamics ] we utilize this property to study the effects of integrability on the relaxation of the lieb - liniger model following such a quench . in particular , we compare two nonequilibrium quench scenarios with the same final hamiltonian and state energy , but beginning from starkly different initial states . statistical mechanics would predict that the system would relax to the same thermal state in both cases , but due to the integrability of the lieb - linger model not only are the time - averaged states following the two quenches non - thermal , they are also distinct . after characterizing and comparing the nonequilibrium dynamics following both quenches , we conclude in sec . [ sec : conclusions ] . the lieb liniger model @xcite describes a system of @xmath5 indistinguishable bosons subject to a delta - function interaction potential in a periodic one - dimensional ( 1d ) geometry of length @xmath6 . we work in units such that @xmath7 and the particle mass @xmath8 , and so the hamiltonian of this system reads @xmath9 where @xmath10 is the interaction strength . the coordinate bethe ansatz yields eigenstates @xmath11 of hamiltonian with spatial representation @xcite @xmath12 \nonumber \\ & \qquad\times \prod_{k > l } \left ( 1 - \frac{ic \ , \mathrm{sgn}(x_k - x_l ) } { \lambda_{\sigma(k)}-\lambda_{\sigma(l ) } } \right),\end{aligned}\ ] ] where the rapidities @xmath13 ( or quasimomenta ) are solutions of the bethe equations @xmath14 the quantum numbers @xmath15 are any @xmath5 distinct integers ( half - integers ) in the case that @xmath5 is odd ( even ) @xcite , and @xmath16 denotes a sum over all @xmath17 permutations @xmath18 of @xmath19 . the normalization constant reads @xcite @xmath20 \big]^{1/2 } } \ ; , \ ] ] where @xmath21 is the @xmath22 matrix with elements @xmath23_{kl } & = \delta_{kl } \big ( l + \sum_{m=1}^{n } \frac{2c}{c^2 + ( \lambda_k-\lambda_m)^2 } \big)\nonumber \\ & \qquad - \frac{2c}{c^2 + ( \lambda_k-\lambda_l)^2}.\end{aligned}\ ] ] the rapidities determine the total momentum @xmath24 and energy @xmath25 of the system in each eigenstate . the ground state of the system corresponds to the set of @xmath5 rapidities that minimize @xmath26 and constitute the ( pseudo-)fermi sea of the 1d bose gas @xcite . the fermi momentum @xmath27 is the magnitude of the largest rapidity occurring in the ground state in the tonks girardeau limit of strong interactions the only parameter of the lieb liniger model in the thermodynamic limit is the dimensionless coupling @xmath28 , where @xmath29 is the 1d density . in finite systems , physical quantities also depend on the particle number @xmath5 ( see , e.g. , sec . [ subsec : finitesize ] ) , whereas the length @xmath6 of our system , and therefore also the density @xmath30 , are arbitrary . consequently , in this article we will specify both @xmath5 and @xmath2 . unless specified otherwise , we measure time in units of @xmath31 , energy in units of @xmath32 , and length in units of @xmath33 . as the eigenstates @xmath11 form a complete basis @xcite for the state space of the lieb liniger model , the expectation value @xmath34 of an arbitrary operator @xmath35 in a schrdinger - picture density matrix @xmath36 can be expressed as a sum of matrix elements of @xmath35 between the states @xmath11 . in particular , in a pure state @xmath37 we have @xmath38 whereas in a statistical ensemble with density matrix @xmath39 , we find @xmath40 in this article , we focus in particular on the normalized @xmath41-order equal - time correlation functions @xmath42^{1/2}},\end{aligned}\ ] ] where @xmath43 is the annihilation ( creation ) operator for the bose field and @xmath44 . here and in the following we drop the time index @xmath45 of the state vectors . since the hamiltonian we consider in this article is translationally invariant along the periodic volume of length @xmath6 , the mean density @xmath46 is constant in both time and space , and @xmath47 . the correlation functions @xmath48 can therefore be expressed as the expectation values of the operators @xmath49 . we note that for the same reasons as above the matrix elements @xmath50 are invariant under global coordinate shifts @xmath51 and thus , without loss of generality , we can set one of the spatial variables to zero . for the first - order correlation function , the matrix elements are @xmath52 the evaluation of the integral in eq . is complicated by the sign function in eq . and the associated nonanalyticities in @xmath53 where any two particle coordinates @xmath54 and @xmath55 coincide . however , we can use the bose symmetry of the wave function @xmath53 to reexpress this matrix element as a sum of integrals @xmath56 over the ordered domains @xcite @xmath57 substituting the coordinate - space form [ eq . ] of the lieb - liniger eigenfunctions , we obtain @xmath58 where @xmath59 . the matrix elements of the second - order correlation operator @xmath60 are similarly given by @xmath61 where @xmath62 and @xmath63 is defined analogously in terms of the elements of @xmath64 . in the limit @xmath65 this expression simplifies somewhat , and in general the matrix elements of the local @xmath41-order correlation operator @xmath66^m [ \hat{\psi}(0)]^m / n^m$ ] are given by the expression @xmath67 where the domain @xmath68 . we note , moreover , that eqs . include as degenerate cases the diagonal matrix elements ( cf . @xcite ) appropriate to the calculation of correlations in the ground state ( sec . [ sec : groundstate ] ) and in statistical ensembles ( sec . [ sec : dynamics ] ) . the calculation of correlation functions from eqs . involves the evaluation of integrals of the general form @xmath69 where ( for the repulsive interactions @xmath70 considered in this article ) the @xmath71 are real numbers . a single closed form for this integral does not exist , as in general one or more @xmath71 may vanish , and this must be handled separately from the case of @xmath72 . however , given knowledge of the _ particular _ sets of rapidities @xmath73 and @xmath74 ( and permutations @xmath75 and @xmath64 ) , and thus of the locations of zero exponents @xmath76 in eq . , each individual integral of this form can be reduced to an algebraic expression in terms of @xmath77 . more specifically , each successive integration @xmath78 yields a term ( involving , in general , @xmath79 ) arising from the primitive integral @xcite @xmath80 in the case that @xmath71 is nonzero , or from @xmath81 otherwise . in our calculations , the construction of algebraic expressions for the integrals occurring in eqs . in terms of the rapidities @xmath13 is efficiently performed by a simple computer algorithm that accounts for and combines the symbolic terms that arise from these successive reductions . we note that , e.g. , each matrix element @xmath82 is a sum of @xmath5 integrals over @xmath83-dimensional domains and that the integrand in each case comprises @xmath84 terms @xcite , illustrating the dramatically increasing computational cost of evaluating correlation functions with increasing @xmath5 . nevertheless , the explicit closed - form expression for the integral produced by our algorithm can be evaluated to obtain a numerically exact result by substituting in the values of the rapidities . the latter are obtained by solving eq . numerically using newton s method , starting in the tonks girardeau regime of strong interactions @xmath85 and iteratively progressing to smaller values of @xmath2 using initial guesses given by linear extrapolation of the solutions at stronger interaction strengths . we note that this algorithmic approach also provides for the efficient and accurate calculation of the overlaps @xmath86 between eigenstates of hamiltonian corresponding to different values of @xmath2 , which we make use of in our analysis of nonequilibrium dynamics in sec . [ sec : dynamics ] . in particular , the overlap between an arbitrary eigenstate @xmath87 of @xmath88 at a finite interaction strength @xmath89 and the noninteracting ground state @xmath90 , with constant spatial representation @xmath91 , is simply given by @xmath92 which can easily be evaluated semi - analytically using our algorithm . in practice we find that the results we obtain for the overlaps from our evaluation of eq . agree with the recently derived closed - form expressions for these quantities @xcite , which imply in particular that @xmath93 as any @xmath94 . as a first application of our methodology we calculate the correlation functions of the lieb liniger model in the ground state for up to @xmath95 particles . in this case , we need to evaluate only the diagonal elements of eqs . in the ground - state wave function , thereby obtaining exact algebraic expressions for correlation functions in terms of the ground - state rapidities , which are themselves determined to machine precision ( sec . [ subsec : correlation_functions ] ) . the ground - state correlations of the lieb liniger model have been considered extensively in previous works ( see refs . @xcite and references therein ) , and we compare our exact mesoscopic results to those obtained with various other methods and approximations , for finite system sizes as well as in the thermodynamic limit . this allows us to clarify the utility and limitations of calculations , such as ours here and in ref . @xcite , that involve only small particle numbers . we begin by considering the first - order correlation function @xmath96 in the ground state of the lieb liniger model . in fig . [ fig : g1momdist](a ) we plot @xmath97 for @xmath95 particles for a range of interaction strengths @xmath2 , which exhibits the expected decrease in spatial phase coherence with increasing @xmath2 @xcite . ( color online ) one- and two - body correlations in the lieb liniger ground state , for @xmath95 particles . ( a ) non - local first - order coherence @xmath97 . the black dot - dashed line indicates the asymptotic long - range behavior @xmath98 of a tonks girardeau gas in the thermodynamic limit . ( b ) corresponding zero - temperature momentum distribution @xmath99 . the black dot - dashed line indicates the universal high - momentum power - law scaling @xmath100 common to all positive interaction strengths @xcite . ( c ) non - local second - order coherence @xmath101 . ( d ) corresponding static structure factor @xmath102.,scaledwidth=48.0% ] as is well known , true long - range order , @xmath103 @xcite , is prohibited in an interacting homogeneous 1d bose gas in the thermodynamic limit , even at zero temperature ( see ref . @xcite and references therein ) . indeed the lieb liniger system is quantum critical at zero temperature , and the asymptotic long - range behavior of @xmath97 is a power - law decay ( so - called _ quasi_-long - range order ) @xcite . this power - law scaling of @xmath97 is only expected to be realized at separations @xmath104 large compared to the healing length @xmath105 and , in a finite periodic geometry such as we consider here , is curtailed by the finite extent @xmath6 of the system ( see , e.g. , ref . indeed , for @xmath106 , the power - law decay is not visible in our finite - sized calculation , although as the interaction strength @xmath2 increases @xmath97 exhibits behavior consistent with power - law decay over an increasingly large range of @xmath104 , see fig . [ fig : g1momdist](a ) . in particular , for @xmath107 , our results for @xmath97 seem to converge toward the asymptotic scaling of the tonks girardeau limit ( black dot - dashed line ) with increasing @xmath2 . due to the translational invariance of our system , the first - order correlations of the lieb liniger ground state are encoded in the momentum distribution @xmath108 which , in our finite periodic geometry , is only defined for discrete momenta @xmath109 , with @xmath110 an integer . in fig . [ fig : g1momdist](b ) we plot the momentum distributions @xmath99 corresponding to the first - order correlation functions @xmath97 shown in fig . [ fig : g1momdist](a ) . the first feature that we note in fig . [ fig : g1momdist](b ) is that for all interaction strengths , @xmath111 exhibits a power - law decay @xmath112 ( dot - dashed black line ) at high momenta . this is a universal result for delta - function interactions in one dimension @xcite ( and indeed also in higher dimensions @xcite ) . the effects of the finite extent @xmath6 of the system on the first - order correlations are again evident in this momentum - space representation . for @xmath106 , no deviation from the @xmath113 scaling is observed for the smallest ( nonzero ) momenta @xmath114 that can be resolved in the periodic geometry . for larger values of the interaction strength , @xmath111 departs from the @xmath113 scaling at increasingly large values of @xmath115 with increasing @xmath2 , and develops a hump at momenta near @xmath116 for @xmath117 @xcite . we note that although the small-@xmath115 behavior of @xmath118 tends towards the @xmath119 scaling exhibited by the tonks girardeau gas in the thermodynamic limit , the rounding off of the power - law decay of @xmath97 as @xmath120 precludes @xmath118 from reaching the known asymptotic @xmath121 behavior in our finite geometry . in fig . [ fig : g1momdist](c ) , we present the nonlocal second - order coherence @xmath122 , which provides a measure of density - density correlations , for @xmath95 particles at a range of interaction strengths @xmath2 . in the limiting case of an ideal gas ( @xmath123 ) , the ground state of the system is a fock state of @xmath5 particles in the zero - momentum single - particle mode , and the second - order coherence @xmath124 ( horizontal dashed line ) is therefore independent of @xmath104 . as the interaction strength @xmath2 is increased , the second - order coherence is increasingly suppressed at zero spatial separation and correspondingly enhanced at separations @xmath125 . oscillations in @xmath101 develop at finite @xmath104 as the system enters the strongly interacting regime @xmath126 @xcite and , in particular , for @xmath127 ( dashed cyan line ) , our numerical results are practically indistinguishable from the exact tonks girardeau limit result ( solid black line ) @xcite . an alternative representation of the second - order correlations of the ground state is given by the static structure factor @xmath102 , which is related to @xmath101 by @xcite @xmath128.\ ] ] in fig . [ fig : g1momdist](d ) we present the structure factors @xmath102 corresponding to the correlation functions @xmath101 shown in fig . [ fig : g1momdist](c ) . for all values of @xmath2 , @xmath129 due to particle - number conservation and translational invariance . in the ideal - gas limit ( red circles ) @xmath130 for all nonzero @xmath114 . in the opposite limit of a tonks girardeau gas @xmath131 which tends , in the thermodynamic limit , to the well - known result ( see , e.g. , ref . @xcite ) @xmath132 for @xmath133 , and @xmath134 for @xmath135 . just as for @xmath101 , we observe that for @xmath127 ( cyan plus symbols ) , our numerical results for @xmath102 are almost identical to the known exact expression [ eq . ] for the tonks girardeau limit ( black crosses ) . for smaller values of @xmath2 our mesoscopic results for @xmath102 appear consistent with those of refs . @xcite , obtained using quantum monte carlo and algebraic - bethe ansatz techniques , respectively . ( color online ) interaction - strength dependence of the local second- , third- and fourth - order coherence in the lieb liniger ground state , for @xmath95 particles . to aid visibility , we plot @xmath136 scaled by a factor of @xmath137 , and @xmath138 scaled by a factor of @xmath139 . dot - dashed lines indicate asymptotic weak- ( @xmath140 ) and strong - coupling ( @xmath141 ) expressions for @xmath136 , @xmath142 and @xmath138 in the thermodynamic limit ( see text).,scaledwidth=48.0% ] we now focus in more detail on local correlation functions . we note that the local second - order coherence has recently been proposed as a measure of quantum criticality in the 1d boson system @xcite , while the local third - order correlations have received increasing attention both in theory @xcite and experiment @xcite . the local fourth - order correlations for the lieb liniger model have also been investigated @xcite . in fig . [ fig : g23localgs ] , we plot the local second - order coherence @xmath136 ( solid red line ) , together with the local third - order coherence @xmath143 ^ 3 [ \hat{\psi}(0 ) ] ^3 \rangle / n^3 $ ] ( dotted green line ) , and the local fourth - order coherence @xmath144 ^ 4 [ \hat{\psi}(0 ) ] ^4 \rangle / n^4 $ ] ( dashed blue line ) for @xmath95 particles and a broad range of interaction strengths @xmath2 . for comparison , we also plot the asymptotic results obtained in the bogoliubov limit of weak interactions ( @xmath145 ) in the thermodynamic limit @xcite ( left - hand dot - dashed lines ) . the numerical results for small @xmath2 are broadly comparable to these thermodynamic - limit results . however , for the small particle numbers considered here , the suppression of @xmath136 , @xmath142 , and @xmath138 due to interactions in the limit of small @xmath2 is overshadowed by the suppression due to the finite population of the system @xcite . at larger @xmath2 , the effects of interactions dominate , and the numerical results converge closely to the appropriate strong - coupling expressions @xcite ( right - hand dot - dashed lines ) . we note , therefore , that the local correlations of the lieb liniger ground state , and particularly their scaling with @xmath2 , appear to be quite insensitive to the infrared cutoff imposed by the finite extent of our system in the strongly interacting regime @xmath141 . the results we have obtained so far indicate that , as expected , the small size of our system leads to corrections to correlation functions as compared to their known asymptotic forms in the thermodynamic limit . however , our results also suggest that the effects of finite system size are comparatively less important for local correlations , particularly in the limit of large interaction strengths @xmath141 . to further elucidate the potential significance of finite - size effects in our calculations of nonequilibrium dynamics @xcite , here we give a brief characterization of the dependence of correlation functions of the lieb liniger ground state on the particle number @xmath5 at a fixed value of the interaction strength @xmath2 . specifically we consider the case for @xmath146 , as this value places the system in the strongly interacting regime @xmath141 ( which appears less sensitive to finite - size effects than the weakly interacting regime @xmath147 ) , while still exhibiting significant deviations from the tonks girardeau limit ( see , e.g. , ref . @xcite ) . whereas elsewhere in this paper we quote momenta ( lengths ) in units of @xmath116 ( @xmath33 ) , in comparing results between systems with different particle numbers @xmath5 we quote momenta ( lengths ) in units of @xmath148 [ @xmath149 , so as to avoid a potentially misleading dependence of the unit of length on @xmath5 [ cf . ( color online ) dependence of first- and second - order correlations in the lieb liniger ground state on particle number @xmath5 for @xmath146 . ( a ) first - order correlation function @xmath97 . ( b ) corresponding momentum distribution function @xmath99 . black dot - dashed lines in ( a ) and ( b ) indicate the asymptotic infrared scaling of @xmath97 and @xmath111 , respectively , with luttinger parameter @xmath150 ( see text ) . ( c ) second - order correlation function @xmath101 . ( d ) corresponding static structure factor @xmath102 . the black dot - dashed lines in ( c ) and ( d ) represent the phenomenological expressions of ref . @xcite for @xmath101 and @xmath102 in the thermodynamic limit , respectively.,scaledwidth=48.0% ] in fig . [ fig : finitesizegs](a ) we plot @xmath97 for particle numbers @xmath151 and @xmath152 . for small @xmath104 , the curves fall nearly perfectly on one line . the same behavior can be observed for the large-@xmath115 tail of the corresponding momentum distribution @xmath111 , which we plot in fig . [ fig : finitesizegs](b ) . indeed , at larger momenta @xmath153 , @xmath111 appears to exhibit a rapid collapse to a single curve with increasing @xmath5 @xcite . however , the differences in @xmath111 are so small that they can not be seen in fig . [ fig : finitesizegs](b ) . for small momenta , our choice of units implies an increasing resolution with increasing particle number , specifically @xmath154 . however , this lowest resolvable momentum seems to fall on one line for increasing particle number , indicating that the infrared behavior of large systems can be at least partly accessed by our mesoscopic system sizes . luttinger - liquid theory predicts a long - range power - law decay @xmath155 , where the luttinger parameter @xmath156 can be calculated from the thermodynamic limit of the bethe ansatz solution ( see , e.g. , refs . @xcite and references therein ) . for our parameters we have @xmath150 , implying an asymptotic scaling @xmath157 [ black dot - dashed line in fig . [ fig : finitesizegs](a ) ] . this corresponds to a power - law behavior @xmath158 @xcite [ dot - dashed line in fig . [ fig : finitesizegs](b ) ] for small momenta . we note that this infrared scaling is a true many - body effect and as such does not show up for @xmath159 particles . indeed , one can show analytically that , for @xmath159 , the momentum distribution @xmath160 and thus @xmath161 is the highest power in the series expansion of @xmath111 . in fig . [ fig : finitesizegs](c ) we plot the nonlocal second - order coherence @xmath101 for @xmath146 and @xmath151 and @xmath152 . the corresponding static structure factor @xmath102 is shown in fig . [ fig : finitesizegs](d ) . in fig . [ fig : finitesizegs](d ) we also plot ( black dot - dashed line ) the form of @xmath102 resulting from the phenomenological expression proposed in ref . @xcite ( see also ref . this expression involves the limiting dispersions and edge exponents of the lieb liniger model , which we obtain by numerically solving the appropriate integral equations @xcite . we also plot the corresponding prediction for @xmath101 ( black dot - dashed line ) in fig . [ fig : finitesizegs](c ) . we note that the numerical results for our mesoscopic systems are , in general , rather close to the phenomenological thermodynamic - limit expressions even for the relatively small particle numbers considered here . we now apply our methodology to the nonequilibrium dynamics of the lieb liniger model . specifically , we consider the evolution of a system , initially prepared in the ground state of hamiltonian with interaction strength @xmath162 , following an abrupt change , at time @xmath163 , of the interaction strength to a distinct value @xmath164 a so - called `` interaction quench '' . the evolution of the system following such a quench is generated by hamiltonian with interaction strength @xmath2 , which we denote by @xmath165 hereafter . the time - evolving state is given at all times @xmath166 by @xmath167 are the eigenstates of @xmath165 with energies @xmath168 , and @xmath169 are the overlaps of the @xmath170 with the initial state @xmath171 . the expectation value of an arbitrary operator @xmath35 in the state @xmath172 is given by @xmath173 we use the methodology described in sec . [ sec : model ] to evaluate both the overlaps @xmath174 and the matrix elements @xmath175 that appear in eq . . one of the features of our methodology is that it allows us to describe quenches between arbitrary interaction strengths . in this paper we consider two interaction - strength quenches , from different initial interaction strengths @xmath162 , to a common final value of the coupling @xmath2 . specifically , we consider a quench from the non - interacting limit @xmath176 ( similar to those previously studied in refs . @xcite ) and a quench from the correlated ground state obtained for a strong interaction strength @xmath177 . as @xmath165 is time independent following the quench , energy is conserved during the dynamics . we choose the final interaction strength after the two quenches such that the postquench energy is the same in both cases . the statistical description of the dynamics of sufficiently ergodic systems is usually based on the assumption that the energy is the sole integral of motion , such that the equilibrium system is entirely determined by its energy . if this would be the case for our system , the two quenches would lead to the same equilibrium state . however , the dynamics according to the integrable lieb liniger hamiltonian are strongly constrained by the conserved quantities other than the total energy . by performing two different quenches to the same final hamiltonian and energy , we investigate the effects of integrability on the postquench evolution of the lieb liniger system . the conserved energy following the quench is the energy of the system at time @xmath178 , @xmath179 where @xmath180 is the energy of the ground state @xmath181 of the initial hamiltonian @xmath182 and we used the well - known result @xmath183 @xcite , which implies that @xmath184 is given by following the tangent to the curve @xmath185 at @xmath162 out to @xmath2 . here , @xmath186 is the local second - order coherence in the initial state . in the case of a quench from the noninteracting ground state ( @xmath176 ) , eq . reduces to the simple expression @xmath187 @xcite , implying that the energy imparted to the system during the quench diverges as @xmath188 @xcite . by contrast , in a quench from the tonks girardeau limit @xmath189 to a finite interaction strength @xmath2 the final energy is bounded from above , @xmath190 , by the ground - state energy of the tonks girardeau gas . nevertheless , according to eq . ( [ eq : addedegammai ] ) , a final interaction strength @xmath191 such that @xmath192 does exist . here , we consider quenches of @xmath3 particles , and determine this final interaction strength to machine precision , inferring a value @xmath193 from numerical solutions for the energy and local second - order coherence of the ground state at finite @xmath2 ( sec . [ subsec : secondorder ] ) . we note that although the overlaps @xmath194 of the initial state @xmath181 with the eigenstates of @xmath195 can be calculated analytically in the case of the quench from @xmath176 @xcite , for the quench from @xmath177 no closed - form expressions for these quantities are known , and thus their numerical values must be determined using the semi - analytical methodology described in sec . [ subsec : correlation_functions ] . an important summary of the postquench expectation value of an operator [ eq . ] is provided by the time - averaged value @xmath196 neglecting degeneracies in the spectrum of @xmath195 ( see discussion in appendix [ app : de ] ) , such averages are given by the expectation values @xmath197 of operators @xmath35 in the density matrix @xmath198 of the diagonal ensemble @xcite . formally , the sums in eq . , , and range over an infinite number of eigenstates @xmath11 , and thus the basis over which @xmath172 is expanded must be truncated in our numerical calculations . by only including eigenstates with an absolute initial - state overlap @xmath199 larger than some threshold , we consistently neglect small contributions to correlation functions from weakly occupied eigenstates and minimize the truncation error for a given basis size . we quantify this truncation error by the violations of the normalization and energy sum rules , as we discuss in appendix [ app : cutoff ] . ( color online ) time evolution of local second - order correlations for @xmath3 particles following quenches of the interaction strength to a final value @xmath193 from initial values @xmath176 ( red dotted line ) and @xmath177 ( blue dashed line ) . the horizontal solid ( dot - dashed ) line indicates the prediction of the diagonal ensemble for @xmath136 for the quench from @xmath200 ( @xmath201).,scaledwidth=47.0% ] in fig . [ fig : dyng2local ] we plot the time evolution of the local second - order coherence @xmath202 for @xmath3 particles following quenches of the interaction strength from initial values @xmath176 ( red dotted line ) and @xmath177 ( blue dashed line ) to the common final value @xmath203 . for the quench from the noninteracting initial state ( @xmath176 ) , as time evolves the local second - order coherence decays from its initial value @xmath204 before settling down to fluctuate about the diagonal - ensemble expectation value @xmath205 ( horizontal dot - dashed line ) . this behavior is consistent with results obtained for similar quenches of the interaction strength from zero to a positive value in ref . @xcite . for the quench from @xmath177 , the value of @xmath136 in the initial `` fermionized '' state is @xmath206 . in this case @xmath202 rises as time progresses , and then exhibits somewhat irregular oscillations about @xmath205 ( horizontal solid line ) . we observe that the decay ( growth ) of @xmath202 to its diagonal - ensemble value and the onset of irregular oscillations about this value occur on comparable time scales in the two quenches . we note that the predictions of the diagonal ensemble for the local second - order coherence @xmath205 are very similar for the two quenches , despite the significant difference between the values of @xmath136 in the two initial states . however , they are clearly distinct @xmath205 for the quench from the noninteracting state is in fact larger than that for the quench from the correlated state by an amount @xmath207 , demonstrating that the system retains some memory of its initial state in the long time limit as is expected for an integrable system . we analyze this difference in more detail in sec . [ subsec : contrib ] . we now turn our attention to the time evolution of the full non - local second - order correlation function @xmath208 . in fig . [ fig : g2dyn](a ) we show the dependence of @xmath208 on separation @xmath104 for the quench from the noninteracting initial state at four representative times . [ note that the upper limit @xmath209 of the @xmath104 axis in fig . [ fig : g2dyn](a ) corresponds to @xmath210 in the present case of @xmath3 particles . ] at @xmath163 ( horizontal solid line ) , the second - order coherence has the constant form of the noninteracting ground state . at short times ( e.g. , @xmath211 , red dashed line ) a minimum in @xmath101 develops at zero separation , together with the corresponding maximum required by the conservation of @xmath212 @xcite . as time progresses a wave pattern of maxima and minima develops and propagates away from the origin ( e.g. , @xmath213 , green dotted line ) . by time @xmath214 ( blue dot - dashed line ) , the distinct maxima and minima of @xmath208 have broadened in such a way that they are no longer clearly distinguishable and the correlation function agrees reasonably well with its diagonal - ensemble form ( black dot - dashed line ) for small separations @xmath215 . in fig . [ fig : g2dyn](b ) we show the full space and time dependence of @xmath208 following a quench from @xmath176 , which gives a more complete picture of the development of a correlation wave at short length scales and its propagation to larger values of @xmath104 as time progresses . the correlation wave we observe here is consistent with the results of previous investigations of the dynamics following the sudden introduction of repulsive interactions in an initially noninteracting gas @xcite . in fig . [ fig : g2dyn](d ) we plot the spatial form of @xmath208 for the quench from @xmath177 at the same four representative times considered in fig . [ fig : g2dyn](a ) . despite the obvious distinction that the initial ( @xmath163 , solid grey line ) correlation function is in the fermionized regime with @xmath216 , the behavior of @xmath208 for this quench is qualitatively similar to that observed for the quench from @xmath176 , in that at early times ( e.g. , @xmath217 , red dashed line ) , deviations from @xmath218 occur only at small separations @xmath219 . moreover , as time evolves and @xmath202 increases towards @xmath205 , larger modulations of @xmath208 about its initial functional form develop ( e.g. , @xmath220 , green dotted line ) . at later times ( e.g. , @xmath221 , blue dot - dashed line ) , @xmath208 is close to @xmath222 at small separations @xmath215 , but exhibits large excursions away from it at larger @xmath104 . in fig . [ fig : g2dyn](e ) we plot the full space and time dependence of @xmath208 following the quench from @xmath177 . although the behavior of @xmath208 here obviously differs from that following a quench from the noninteracting initial state [ fig . [ fig : g2dyn](b ) ] , with the `` fermionic '' depression around @xmath223 lessening rather than growing in magnitude , a similar pattern of propagating correlation waves in @xmath208 can again be seen . the correlation - wave pattern common to both quenches is more clearly exhibited by the _ change _ @xmath224 in the correlation function following the quench , which we plot in figs . [ fig : g2dyn](c ) and [ fig : g2dyn](f ) . this representation of the postquench second - order coherence of the system reveals a remarkably similar pattern of propagating waves in both cases , although the maxima and minima of the two wave patterns are inverted relative to one another . fitting a power law to the position @xmath225 of the first propagating extremum of each of the two correlation waves , we find @xmath226 for the quench from @xmath176 and @xmath227 for the quench from @xmath177 , which we indicate by the solid black lines in figs . [ fig : g2dyn](c ) and [ fig : g2dyn](f ) . these power - law trajectories are consistent with the `` telescoping '' @xmath228 behavior obtained for a quench @xmath1 in ref . @xcite , and for quenches from finite repulsive interactions to the noninteracting limit in ref . @xcite ( see also ref . the small scale features on top of the main propagating extrema differ for the two quenches , with fast oscillations appearing more pronounced for the quench @xmath229 in fig . [ fig : g2dyn](c ) . even though hardly visible in fig . [ fig : g2dyn](f ) , they are still present for the quench from @xmath230 , but due to the different distribution of overlaps in the final basis compared to the quench from @xmath176 ( cf . sec . [ subsec : contrib ] ) , they contain more high - frequency components and therefore the fine structure differs . ( color online ) time - averaged second - order correlation functions following quenches of the interaction strength to @xmath193 from initial values @xmath176 ( red solid line ) and @xmath177 ( blue dotted line ) . results are for @xmath3 particles . ( a ) the correlation functions @xmath101 in the initial states with @xmath176 ( horizontal solid line ) and @xmath177 ( grey dashed line ) , as well as for the ground state at @xmath231 ( solid black line ) are also indicated for comparison . the black dot - dashed line corresponds to the thermal value of the correlation function following relaxation , as predicted by the canonical ensemble ( see text ) . ( b ) comparison of the time - averaged second order correlation functions to the various ensembles defined in the text : the standard canonical ensemble ( black dot - dashed line ) , the canonical ensemble restricted to zero - momentum eigenstates ( black solid line ) , and the canonical ensemble restricted to parity - invariant states ( grey solid line ) . , scaledwidth=48.0% ] we now compare the time - averaged second - order correlation functions following the two quenches with the form of this function that would be obtained if , following the quench , the system relaxed to thermal equilibrium . as in ref . @xcite we make use of the canonical ensemble , for which the density matrix is given by @xmath232 where the partition function @xmath233 . the inverse temperature @xmath234 is determined implicitly by fixing the mean energy in the state @xmath235 to the common postquench energy , i.e. , @xmath236 . the sum in eq . , like that in eq . , formally ranges over an infinite number of eigenstates . we therefore truncate this sum by applying a cutoff in energy , as described in appendix [ app : cutoff ] . in fig . [ fig : g2dynstat](a ) we plot the second - order correlation function @xmath237 in the canonical ensemble ( black dot - dashed line ) , along with the diagonal - ensemble predictions @xmath222 for the quenches from @xmath176 ( red solid line ) and from @xmath177 ( blue dotted line ) . for comparison we also plot the correlation functions in the initial states with @xmath176 ( horizontal line ) , @xmath177 ( grey dashed line ) , as well as the ground state for @xmath231 ( solid black line ) . for the quench from @xmath176 , the time - averaged value @xmath205 is smaller than the corresponding thermal value @xmath238 , consistent with the results of refs . in fact @xmath222 is suppressed below @xmath239 over a range of separations @xmath240 . correspondingly , @xmath241 at larger separations @xmath104 due to particle number and momentum conservation . for the quench @xmath242 , the diagonal - ensemble coherence function @xmath222 is similar in shape to that of the quench from @xmath176 however , it is somewhat smaller at @xmath223 , and correspondingly larger at large @xmath104 . this indicates some memory of the initial state preserved by the dynamics of the integrable lieb liniger system @xcite . despite these differences , on the whole both functions @xmath222 are comparable to @xmath239 ( cf . also ref . we note , however , that they are also both reasonably close to the ground state result for @xmath101 at interaction strength @xmath203 ( solid black line ) , although the local value @xmath205 for both quenches is much closer to the thermal value than the ground state value . since the system is in its ground state before the quench for both @xmath176 and @xmath177 , and the total momentum operator @xmath243 commutes with the hamiltonian , the postquench states at @xmath203 only have support on eigenstates with total momentum @xmath244 . furthermore , the spatially structureless initial state at @xmath176 implies additional parity - invariance ( @xmath245 ) in bethe rapidity space for the postquench eigenstates @xcite . thus an interesting question to ask is if we constructed a canonical density matrix restricted to @xmath244 states , or one further restricted to parity - invariant states ( which are a subset of the @xmath244 states ) , would these yield better agreement with the diagonal ensemble predictions for the quenches ? we have performed these constructions with the temperature in both cases fixed via the postquench energy in the same way as for the canonical ensemble , cf . eq . and the following text . in fig . [ fig : g2dynstat](b ) , we plot the resulting second - order correlation function @xmath237 for the standard canonical ensemble ( black dot - dashed line ) , as well as in the restricted @xmath244 ensemble ( solid black line ) , and the parity - invariant ensemble ( solid grey line ) . we also include the diagonal - ensemble predictions @xmath222 for the quenches from @xmath176 ( red solid line ) and from @xmath177 ( blue dotted line ) . it can be seen that the restricted ensembles give results for the correlation function that are quite close to the standard canonical ensemble , and are no closer to the diagonal ensemble results . ( color online ) ( a ) populations @xmath246 of eigenstates with energies @xmath247 following quenches to @xmath248 from @xmath176 ( red crosses ) and @xmath177 ( blue squares ) . note that the @xmath249-axis is plotted on a logarithmic scale . for the quench from @xmath177 , additional non - parity - invariant states appear in degenerate , parity - conjugate pairs and since their contributions is identical , the points lie on top of each other . the black dotted line with filled black circles represents the populations @xmath250 of eigenstates with energies @xmath247 for the canonical ensemble . the grey line with grey filled circles , and the black dashed line with empty black circles are the corresponding results for the @xmath244 restricted ensemble , and the parity - restricted ensemble , respectively . ( b ) low - energy part of ( a ) . , scaledwidth=48.0% ] the relaxation of the nonlocal correlations @xmath208 takes place on a similar time scale to that of the local coherence @xmath202 for both of the quenches considered here . this should be contrasted with , e.g. , the behavior following a quench from the noninteracting limit to @xmath127 reported in ref . @xcite , in which @xmath202 decays rapidly and the development and propagation of correlation waves occurs over a significantly longer time scale . we identify the absence of a significant separation of the time scales of local and nonlocal evolution here as a consequence of the fact that only a small number of eigenstates contribute significantly to the postquench dynamics ( cf . @xcite and references therein ) . indeed , we find that the purity @xmath251 of the diagonal - ensemble density matrix takes values @xmath252 for the quench @xmath253 and @xmath254 for the quench @xmath242 , indicating rather weak participation of the eigenstates @xmath11 in the dynamics . the difference in the purities can largely be attributed to the somewhat greater occupation of the ground state of @xmath195 following the quench from the @xmath127 initial state . to further illustrate the difference in the final states , in fig . [ fig : overlapdist ] we plot the occupations of eigenstates with energy @xmath247 for the quenches from @xmath176 ( red crosses ) and @xmath177 ( blue squares ) . for the quench from @xmath177 , significantly more eigenstates have occupations above a given threshold than in the case of @xmath176 , resulting in a much larger basis size in this case . however , the occupation of the ground state of @xmath195 is somewhat larger for the quench from @xmath177 than for @xmath176 , and the low - lying excited states are comparatively weakly occupied for @xmath177 , cf . [ fig : overlapdist](b ) . this result is reasonably intuitive , as the ground state for @xmath255 is moderately correlated , and will be more similar to the @xmath127 than the @xmath123 ground state . the distribution of normalization over eigenstates @xmath11 is thus more sharply `` localized '' on the ground state in this case , resulting in the somewhat larger value of the purity @xmath256 following this quench . for comparison , we also plot the occupations of the three ensembles introduced in sec . [ subsec : relaxed ] in fig . [ fig : overlapdist ] . the restrictions lead to a reduction in available eigenstates for any given energy - window , and correspondingly the temperature of the canonical ensemble is smaller than that of the @xmath244 ensemble , which is in turn smaller than that of the parity - invariant ensemble . the occupations of eigenstates for the quench from @xmath176 ( red crosses ) and from @xmath177 ( blue squares ) are suggestive of power - law decay at high energies . for small energies on the other hand , fig . [ fig : overlapdist](b ) shows that the functional form is not incompatible with exponential decay . we have described a method to calculate matrix elements between eigenstates of the lieb liniger model of one - dimensional delta - interacting bosons . this method is based on the coordinate bethe ansatz , which generates a complete set of energy eigenfunctions for any fixed coupling strength . this allows us to obtain overlaps between eigenstates of different hamiltonians , as well as expressions for correlation functions . by introducing periodic boundary conditions , we obtained expressions amenable to numerical evaluation . we applied our methodology to the evaluation of first- , second- , third- , and fourth - order correlation functions in the ground state of the lieb liniger model for various values of the interparticle interaction strength . our results indicate that although the correlations of the system are in general distorted by the small system size , finite - size effects become increasingly less significant with increasing interaction strength and decreasing spatial separation . out of equilibrium , we investigated the dynamics of relaxation after a quantum quench of the interparticle interaction strength towards a non - thermal steady state . starting from two different initial states , we quenched to a common final interaction strength @xmath203 chosen in such a way that both postquench energies were the same . our calculations reveal a similar relaxation process for the second - order coherence @xmath208 for both initial states : the build - up of correlations on short interparticle distances and their propagation through the system as time progresses . the time - averaged second - order correlation functions in both cases disagreed with the prediction for thermal equilibrium and were biased , relative to one another , towards their pre - quench forms an intuitive result given the integrability of the system . in the future it would be interesting to study quenches from other initial states with the same final energy to explore how the memory of the initial state is manifest in different situations . although our method is restricted to small system sizes due to computational complexity and here only applied to five particles out of equilibrium , we were able to obtain the dynamical evolution as well as time - averaged correlation functions to high precision . finally we note that the evaluation of matrix elements of the lieb liniger model with this method is not restricted to real - valued bethe rapidities , opening the door to investigating the nonequilibrium dynamics of attractively interacting systems ( where the rapidities become complex - valued ) and that following quenches from more complex initial states . this work was supported by arc discovery project , grant no . dp110101047 ( j.c.z . , t.m.w . , k.v.k . , and m.j.d . ) and by the eu - fet proactive grant aqus , project no . 640800 ( tg ) . m.j.d . acknowledges the support of the jila visiting fellows program . the hilbert space of the lieb liniger model is infinite dimensional , and therefore the sums in eqs . , , and must be truncated for numerical purposes . here , we provide details of the truncation scheme for the two different initial states we considered in sec . [ sec : dynamics ] , and explain how we quantify the error resulting from this truncation . for the quench from @xmath176 , the initial state @xmath181 only has nonzero overlap with eigenstates @xmath257 of @xmath195 that are parity invariant ( i.e. , eigenstates for which @xmath245 ) and , _ a fortiori _ , have zero total momentum @xmath258 @xcite . the strongly - correlated initial state of the quench from @xmath177 similarly has zero overlap with eigenstates @xmath257 with nonzero total momentum , but in this case states contributing to @xmath172 , and thus @xmath259 , need not be parity - invariant in general . for @xmath176 our results for the overlaps agree with recently obtained analytical expressions @xcite , which predict real positive overlaps , given the phase convention implicit in eq . , for quenches to @xmath260 . for @xmath177 , we find that the overlaps are still real , but are no longer restricted to positive values . we briefly summarize our procedure to determine the cutoff here see appendix a of ref . @xcite for an extended discussion for the case of parity - invariant states . it can be shown @xcite that the solutions @xmath73 of the bethe equations are in one - to - one correspondence with the numbers @xmath15 that appear in eq . . this allows us to uniquely label states by the set @xmath261 . without loss of generality , we order the numbers @xmath15 such that @xmath262 , and we only need consider states for which @xmath263 , corresponding to zero total momentum @xmath258 . we specialize hereafter to the case @xmath3 , which is the largest @xmath5 for which we consider the dynamics in this article . the states can be grouped into families , labelled by @xmath264 . we have found empirically that within each such family , the eigenstate @xmath265 has the largest absolute overlap @xmath266 with the initial state , for both initial states we consider ( @xmath176 and @xmath177 ) . furthermore , this overlap is larger than that of the most significantly contributing eigenstate @xmath267 of the following family ( @xmath268 ) . we therefore construct the basis by considering in turn each family @xmath264 and including all states within that family for which the overlap with the initial state exceeds our chosen threshold value @xmath269 . eventually , for some value of @xmath264 , even the eigenstate @xmath265 has overlap with @xmath181 smaller than the threshold , at which point all states that meet the threshold have been accounted for . we note that the lieb liniger model has an infinite number of conserved charges @xmath270=0 $ ] ; @xmath271 , with eigenvalues given by @xmath272 . however , for a quench from @xmath176 their expectation values in the diagonal ensemble @xmath273 diverge for all even @xmath274 @xcite . our numerical results suggest that this is also the case for quenches from @xmath275 ( indeed , they diverge for almost all states but eigenstates @xcite ) . for all odd values @xmath276 , the expectation values of the corresponding conserved charges @xmath277 are identically zero for our initial states and quench protocol . thus , the only nontrivial and regular conserved quantities are the particle number ( @xmath278 ) and energy ( @xmath279 ) . as in ref . @xcite , we quantify the saturation of the normalization and energy sum rules by the sum - rule violations @xmath280 respectively , where @xmath184 is the exact postquench energy [ eq . ] . we note that the calculation of time - dependent observables involves a double sum over @xmath73 , and is therefore more numerically demanding than the calculation of expectation values in the de . moreover , the calculation of the local coherence @xmath202 is much less demanding than that of the full nonlocal @xmath208 . we therefore use different thresholds @xmath269 , resulting in different basis sizes and sum - rule violations , in the calculation of @xmath202 , @xmath208 , and @xmath281 , as indicated in table [ tab : cutoffs ] . we note that the energy sum rule is in general less well satisfied than the normalization sum rule , due to the @xmath282 tail of the diagonal - ensemble distribution of eigenstates @xcite . we find also that both sum rules are less well satisfied for the quench @xmath283 , despite the truncation procedure described above resulting in more than five times as many basis states being employed in its solution than are used in the quench @xmath229 . for expectation values in the ce [ eq . ] , we truncate the basis by retaining all states with energies below some cutoff @xmath284 . the inverse temperature @xmath234 is then chosen to minimize the energy sum - rule violation @xmath285 . the normalization sum rule is fulfilled by construction . since all states ( not only those with zero momentum ) contribute to this sum , the number of eigenstates involved in canonical - ensemble calculations is much larger than that in diagonal - ensemble calculations . for the canonical - ensemble correlation function plotted in fig . [ fig : g2dynstat ] we used an energy cutoff of @xmath286 , which yields a basis of @xmath287 eigenstates @xmath11 . we checked that this cutoff is sufficiently large to ensure saturation of @xmath288 ( fig . [ fig : g2dynstat ] ) . for the ensemble restricted to @xmath244 eigenstates [ fig . [ fig : g2dynstat](b ) ] , we used an energy cut - off of @xmath289 , corresponding to @xmath290 eigenstates , while for the parity - invariant ensemble we used an energy cut - off of @xmath291 , corresponding to @xmath292 eigenstates . .[tab : cutoffs ] basis - set sizes and sum - rule violations for full non - local , time - evolving second - order coherence @xmath208 , for local , time - evolving second - order coherence @xmath293 , and for time - averaged second - order coherence @xmath222 following quenches from @xmath176 , and @xmath177 to @xmath248 . [ cols="^,^,^,^,^,^",options="header " , ] @xmath294 occupations of the @xmath281 basis set are used in the calculation of @xmath256 ( sec . [ subsec : contrib ] ) . ( color online ) contributions of degenerate energy eigenstates to the time - averaged second - order correlation function following a quench from @xmath177 to @xmath248 for @xmath3 particles . ( a ) contributions @xmath295 of off - diagonal matrix elements corresponding to the three largest weights @xmath296 . ( b ) total contribution of degenerate energy eigenstates . , scaledwidth=48.0% ] the time - averaged expectation value [ eq . ] of an operator @xmath35 can be expressed as an expectation @xmath297 in the time - averaged density matrix @xmath298 the first term in eq . is simply the diagonal - ensemble density matrix @xmath299 [ eq . ] , to which @xmath300 reduces in the absence of degeneracies in the spectrum of @xmath165 . this is the case for the quench from @xmath176 , as the only eigenstates of @xmath195 with nonvanishing overlaps with @xmath181 in this case are the parity - invariant states @xmath11 with @xmath73=@xmath301 , which are nondegenerate ( see ref . @xcite and references therein ) . by contrast , in a quench from @xmath275 , @xmath172 has support on non - parity - invariant states @xmath11 , which are degenerate with their parity conjugates @xmath302 . in general such degeneracies can have observable consequences for time - averaged expectation values @xcite . however , as can be seen from fig . [ fig : ensemblediff ] , the correction to @xmath281 due to the contributions of degenerate eigenstates in the case of the quench from @xmath177 is small . it is straightforward to show that the elements @xmath303 of the local second - order coherence between parity - conjugate states must vanish due to symmetry considerations . at larger separations @xmath104 , the matrix elements between these pairs of states are nonzero , as illustrated in fig . [ fig : ensemblediff](a ) . however , these contributions are small compared to the diagonal - ensemble result @xmath281 , and indeed the total contribution of all parity - conjugate states in our finite - basis description [ fig . [ fig : ensemblediff](b ) ] would yield a barely visible correction to the function @xmath281 plotted in fig . [ fig : g2dynstat ] . we note also that the substitution of @xmath259 for the time - averaged density matrix @xmath300 introduces negligible error in the calculation of the purity of this matrix ( sec . [ subsec : contrib ] ) .

the motion of driven interfaces in random media at finite temperature and small external force is usually described by a linear displacement at large times , where the velocity vanishes according to the creep formula as for . in this paper , we question this picture on the specific example of the directed polymer in a two dimensional random medium . we have recently shown ( c. monthus and t. garel , arxiv:0802.2502 ) that its dynamics for can be analyzed in terms of a strong disorder renormalization procedure , where the distribution of renormalized barriers flows towards some `` infinite disorder fixed point '' . in the present paper , we obtain that for small , this `` infinite disorder fixed point '' becomes a `` strong disorder fixed point '' with an exponential distribution of renormalized barriers . the corresponding distribution of trapping times then only decays as a power - law , where the exponent vanishes as as . our conclusion is that in the small force region , the divergence of the averaged trapping time induces strong non - self - averaging effects that invalidate the usual creep formula obtained by replacing all trapping times by the typical value . we find instead that the motion is only sub - linearly in time , i.e. the asymptotic velocity vanishes . this analysis is confirmed by numerical simulations of a directed polymer with a metric constraint driven in a traps landscape . we moreover obtain that the roughness exponent , which is governed by the equilibrium value up to some large scale , becomes equal to at the largest scales . # 1#2 # 1#2

0803.4125

transport phenomena in random media have remained a very active field of research since the discovery of the anderson localization fifty years ago @xcite . for classical systems also , the presence of quenched disorder can induce completely new transport behaviors with respect to pure systems in some regions of parameters . it turns out that even the dynamics of a single particle in one - dimensional random media ( see the review @xcite and references therein ) can already present a very rich phase diagram as a function of the temperature @xmath0 and the external applied force @xmath1 . as an example , we shown on fig . 1 the exactly known phase diagram for the biased sinai model @xcite : one needs to introduce a dimensionless parameter @xmath15 ( where we have chosen to fix the disorder strength at some simple value to simplify the notations and emphasize the dependence upon @xmath0 and @xmath1 we are interested in ) . for @xmath5 , the motion is logarithmically slow @xmath16 . for @xmath17 , the diffusion is anomalous with a sub - linear displacement @xmath18 ; for @xmath19 , the velocity becomes finite @xmath20 , but the dispersion remains anomalous of order @xmath21 . finally for @xmath22 , both the velocity and the diffusion coefficient are finite . so already in simple one - dimensional systems , various dynamical properties can undergo phase transitions at various thresholds . and the external force @xmath1 of the one - dimensional sinai model @xcite : the important dimensionless parameter is @xmath23 ( the disorder strength has been fixed at some simple value ) . from the point of view of strong disorder renormalization @xcite , the logarithmic behavior @xmath24 for @xmath5 corresponds to an `` infinite disorder fixed point '' , whereas the anomalous diffusion phase @xmath25 for @xmath26 corresponds to a `` strong disorder fixed point '' . a finite - velocity @xmath27 appears only for @xmath28 , i.e. when the force is above some temperature - dependent value . , height=302 ] and the external force @xmath1 is not exactly known . however one expects physically two very different regimes : ( i ) near the zero - temperature depinning critical point @xmath29 , the transport at small temperature is dominated by depinning events . ( ii ) in the regime where @xmath0 is finite and @xmath1 is small , the roughness exponent @xmath13 of the equilibrium case @xmath5 is expected to describe the polymer roughness up to some large scale ( that diverges as @xmath4 ) . the transport is then dominated by thermal jumps of finite segments of the polymer between quasi - equilibrated metastable states . in this paper , we focus only on the regime ( ii ) and discuss the center - of - mass motion and the roughness at the largest scales . , height=302 ] the case of more complex systems like interfaces or manifolds in random media has attracted a lot of attention in relation with many important applications , and various approaches have been developed to elucidate the structure of the dynamical phase diagram ( see the reviews @xcite and references therein ) . the simplest model within this class of systems is the directed polymer in a two - dimensional random medium ( see @xcite for a review ) , first introduced to model an interface in the low - temperature phase of two - dimensional disordered ferromagnets @xcite . the statics of this model is rather well understood with exactly known critical exponents : at any temperature @xmath0 , the directed polymer is in a disordered - dominated phase characterized by the roughness exponent @xmath13 and the droplet exponent @xmath30 @xcite . in comparison , the dynamics is much harder to study , and no exact result exists . even numerically , the computational complexity changes completely between the statics that can be studied via exact transfer matrix methods that are polynomial in the size of the polymer @xcite , and the dynamics where the determination of barriers is an np - complete problem @xcite . as a consequence , a complete characterization of the dynamical properties as a function of the temperature and external force ( here we consider that the disorder strength is fixed ) has remained a very challenging issue . an essential novelty with respect to the dynamics of the sinai model described above , is the presence of the chain constraint of the polymer , so that even at zero temperature @xmath31 , the dynamics is already non trivial : the competition between interaction and disorder gives rise to a non - equilibrium depinning phase transition , between a pinned phase @xmath32 where the interface remains blocked forever in some configuration , and a moving phase for @xmath33 . this type of depinning transition has motivated a lot of experimental and theoretical studies ( see the reviews @xcite ) , and an exact mean - field solution has been obtained in @xcite . the critical region with a small temperature around the depinning transition of an elastic line has been studied in @xcite . in the present paper , we focus on the opposite region of the dynamical phase diagram ( see fig . 2 ) where the temperature is finite and the external force is small . in this regime , one expects that the dynamics is dominated by thermal activations between locally quasi - equilibrated metastable states . in particular , the polymer is expected to keep during the motion its equilibrium roughness exponent @xmath13 up to some large scale ( that diverges as @xmath4 ) . this equilibrium roughness has been measured in dynamic simulations in @xcite , and in @xcite when the temperature remains larger that some disorder - dependent threshold . however in other regimes , larger roughness exponents have been measured at larger scales @xcite . this issue of the roughness at the largest scales will be rediscussed later in the paper . but we first need to better understand the statistical properties of the barriers between the locally quasi - equilibrated metastable states . in the regime of finite temperature and small external force , the standard picture in the field ( see the review @xcite and references therein ) seems to be that the dynamics then corresponds to a creep motion , where the center - of - mass moves linearly in time as soon as the temperature is positive @xmath34 with a very non - linear force - velocity relation of the form @xmath35 where the exponent @xmath36 involves the dimension @xmath37 of the interface ( the directed polymer we are interested in corresponds to @xmath38 ) , the equilibrium roughness exponent @xmath39 , and the barrier exponent @xmath40 of the dynamics without external force @xmath41 note that this relation is usually written in the form @xmath42 after using the additional assumption concerning the identity between the barrier exponent @xmath40 and the droplet exponent @xmath43 . however , in our opinion , there is no convincing evidence of the equality @xmath44 neither theoretically ( see @xcite for a recent summary on the debate ) nor numerically ( the best numerical results presently available @xcite points towards an exponent of order @xmath45 rather different from the exactly known droplet exponent @xmath46 ) , we will consider in the following that the barrier @xmath40 is an independent exponent satisfying the bound @xmath47 the usual qualitative argument given in favor of a finite velocity in eq . [ defvitesse ] as soon as @xmath48 is that any barrier can be passed by thermal activation. this is of course true , but this is not sufficient to conclude that the asymptotic velocity is finite , since the motion can also exhibit anomalous behavior as shown by the sinai model at small external force ( see the beginning of the introduction ) . to determine the asymptotic motion of the center - of mass at large times , one needs to study the time needed to travel over a given large distance . the more quantitative argument in favor of the finite velocity of eq . [ defvitesse ] is a scaling argument where the barrier landscape existing in the absence of bias is tilted to take into account the bias contribution : this scaling argument yields that the relevant barriers for the large - time dynamics corresponds to a barrier scale @xmath49 and to a length scale @xmath50 that are finite as soon as @xmath51 ( and that diverge as @xmath4 ) . we fully agree with this scaling argument that will be rediscussed below in section [ scalingarg ] . however again , this is not sufficient to conclude that the velocity is finite , since exactly the same argument can be made for the sinai model where an anomalous diffusion phase exists ( see section [ scalingarg ] for more details ) . the crucial property to determine whether the velocity is finite or not is the probability distribution of barriers @xmath52 around this typical value @xmath49 . in the present paper , we explain that within the strong disorder renormalization in configuration space that we have introduced recently @xcite ( see @xcite for a more detailed presentation ) , one obtains an exponential tail for the distribution of renormalized barriers @xmath53 as is well known in the field of disordered systems since derrida s random energy model @xcite , other disordered models sharing the same low - energy states statistics @xcite and related bouchaud s trap models @xcite ( see also @xcite for one of the first mention of exponentially distributed barriers in connection with extremal statistics ) , this innocent exponential distribution for the barriers corresponds via the change of variable @xmath54 to a very broad power - law decay for the distribution of the trapping times @xmath55 @xmath56 the exponent depends continuously on the external force and on the temperature ( we consider here that the disorder remains fixed ) @xmath57 since the characteristic barrier scale @xmath49 grows with @xmath1 and diverges as @xmath4 as @xmath58 the exponent @xmath7 vanishes as @xmath4 as @xmath59 as a consequence , the region of small external force where @xmath60 corresponds to a very broad distribution of trapping times with a diverging averaged value @xmath61 . this invalidates the usual creep formula that is obtained by replacing all trapping times by the typical value @xmath62 to obtain @xmath63 ( see eq . [ vitessecreep ] ) . we obtain instead that for @xmath60 , the center - of - mass displacement then grows only sub - linearly in time @xmath64 i.e. the asymptotic velocity vanishes @xmath12 . although the vast majority of papers on the subject never mentions the possibility of a zero - velocity phase , we are aware of three papers where the question of the probability distribution of barriers around the typical value has been raised : \(i ) twenty years ago , ioffe and vinokur @xcite have proposed the anomalous sub - linear motion of eq . [ powerlaw ] . but for reasons that are very unclear to us , this possibility seems to have completely disappeared in the more recent literature . \(ii ) fifteen years ago , bouchaud has pointed out in @xcite that the usual creep argument might lead to incorrect results because of problem of strong fluctuations , and cites the sinai model as an example where the transport is indeed harder than anticipated from the typical barrier alone . \(ii ) ten years ago , vinokur , marchetti and chen @xcite have proposed an exponential distribution of barriers , based on extremal statistics argument , and the corresponding power - law distribution for trapping times . however , these authors have supplemented this power - law distribution by a sharp cut - off to recover the usual finite velocity behavior of eq . [ defvitesse ] . in our opinion , there is no good reason to impose this sharp cut - off , since the same procedure for the exactly soluble sinai model would give a wrong answer . to summarize this long introduction , the aim of the present paper is to justify the existence of the anomalous diffusion phase of eq . [ sublinear ] in the region of the phase diagram corresponding to finite temperature and small external force ( see fig . 2 ) , via the use of some strong disorder renormalization procedure on the barriers . the paper is organized as follows . in section [ infinite ] , we briefly recall the main idea of strong disorder renormalization , and we describe the statistical properties of the `` infinite disorder '' fixed point that describes the thermal dynamics of the directed polymer without external force . in section [ strong ] , we explain how the presence of a small external force @xmath1 transforms this `` infinite disorder fixed point '' into a `` strong disorder fixed point '' with an exponential distribution of renormalized barriers , and we describe the consequences for distribution of trapping times and the large - time dynamics . in section [ numerics ] , we present detailed numerical simulations for a directed polymer driven in a traps landscape . in section [ previous ] , we compare our results concerning the roughness with previous works . our conclusions are summarized in section [ conclusion ] . in appendix a , we recall the subtleties associated to the type of the interactions ( metric constraint versus elastic energy ) to justify our choice to consider the metric constraint . strong disorder renormalization ( see @xcite for a review ) is a very specific type of renormalization ( rg ) that has appeared in the field of quantum spin chains : this approach introduced by ma and dasgupta @xcite has been developed by d.s . fisher @xcite , who has introduced the crucial notion of `` infinite disorder '' fixed points where the method becomes asymptotically exact , and who has shown how to obtain explicit exact results for critical exponents and scaling functions for zero - temperature quantum critical points . this method has thus generated a lot of activity for various disordered quantum models @xcite . it has been then successfully applied to various classical disordered dynamical models , such as random walks in random media @xcite , reaction - diffusion in a random medium @xcite , coarsening dynamics of classical spin chains @xcite , trap models @xcite , random vibrational networks @xcite absorbing state phase transitions @xcite , zero range processes @xcite and exclusion processes @xcite . in all these cases , the strong disorder rg rules have been formulated _ in real space _ , with specific rules depending on the problem . for more complex systems where the formulation of strong disorder rg rules has not been possible in real space , we have recently proposed in @xcite a strong disorder rg procedure _ in configuration space _ that can be defined for any master equation as we now recall . the starting point of strong disorder renormalization in configuration space is the master equation describing the evolution of the probability @xmath65 to be in a configuration @xmath66 at time t @xmath67 the notation @xmath68 represents the transition rate from configuration @xmath69 to @xmath66 , and the notation @xmath70 represents the total exit rate out of configuration @xmath66 . for dynamical models , the aim of any renormalization procedure is to integrate over fast processes to obtain effective properties of slow processes . the general idea of strong renormalization for dynamical models consists in eliminating iteratively the fastest process . the rg procedure introduced in @xcite consists in the iterative elimination of the state with the highest exit rate . we refer to @xcite for a detailed description and derivation of the renormalization rules on the transition rates . their most important property is their multiplicative structure that suggests that for a very broad class of disordered systems , the distribution of renormalized exit barriers @xmath71 will become broader and broader upon iteration , so that the strong disorder renormalization procedure should become asymptotically exact at large time scales . note that a very important advantage of this formulation in terms of the transition rates of the master equation is that the renormalized barriers take into account the true barriers of the dynamics , whatever their origin which can be either energetic or entropic . after this general presentation ( see @xcite for more details ) , we now turn to the specific problem of a directed polymer in a two - dimensional random medium we are interested in . for the dynamics _ without external force _ , we have followed numerically the rg flow of the renormalized transition rates and we have found some `` infinite disorder '' fixed point @xcite as we now explain . the rg scale @xmath72 is defined as the scale of the last eliminated exit barrier . at time @xmath73 , the appropriate rg scale is thus @xmath74 so that all metastable states of exit barrier @xmath75 , i.e. of exit time @xmath76 have been kept , whereas all metastable states of exit time @xmath77 have been eliminated . at large scale , one expects that the probability distribution of the remaining exit barriers @xmath78 will converge towards some scaling form @xmath79 where @xmath80 is the fixed point probability distribution , and where @xmath81 is the appropriate scaling factor that represents the width of the renormalized distribution . for the directed polymer in a two - dimensional random medium , we have obtained numerically in @xcite that the width @xmath81 grows asymptotically linearly with the rg scale @xmath72 @xmath82 and that the rescaled distribution of eq . [ pgammabout ] is extremely close to the exponential form @xmath83 note that these two properties seem extremely robust within strong disorder rg since they hold for exactly in soluble models in @xmath38 @xcite and have been also found numerically in quantum models in dimension @xmath84 @xcite . to describe the thermal relaxation of disordered systems towards equilibrium when starting at time @xmath85 from a non - equilibrium initial state , it is useful to introduce the notion of some coherence length @xmath86 that grows slowly in time ( see the reviews @xcite and references therein ) . this coherence length separates the smaller lengths @xmath87 which are quasi - equilibrated at time @xmath73 from the larger lengths @xmath88 which are completely out of equilibrium . then full equilibrium is reached only when the coherence length reaches the macroscopic linear size @xmath89 of the system . within the droplet scaling theory proposed both for spin - glasses @xcite and for directed polymers in random media @xcite , the barriers grow as a power law of the length @xmath90 @xmath91 with some barrier exponent @xmath92 . the typical time @xmath93 associated to scale @xmath90 grows as an exponential @xmath94 . equivalently , the coherence length - scale @xmath86 associated to time @xmath73 grows only logarithmically in time @xmath95 in the numerical study of the relaxation towards equilibrium of an elastic chain in a two dimensional random medium , starting from a straight line at @xmath85 , this coherence length can be extracted from the behavior of the structure factor as a function of time @xcite , and the corresponding measure of the barrier exponent yields @xmath45 @xcite ( see the comments on this exponent @xmath40 before eq . [ psibound ] ) . within the strong disorder renormalization procedure , one may either extract a coherence length @xmath96 as a function of the rg scale @xcite , or consider the statistics of the barriers associated to a given length @xcite , and one obtains the scaling @xmath97 where @xmath98 is a random variable of order one . the barrier exponent @xmath40 obtained in @xcite is of order @xmath99 , i.e. close to the estimation @xmath45 measured in @xcite via langevin dynamics . a systematic study of the dependence of the prefactor @xmath100 upon temperature is not yet available . this prefactor would be of the form @xmath101 if the barriers were purely energetic , but since the strong disorder renormalization is defined on the transition rates , the prefactor @xmath100 may also contain entropic contributions and then have a more complicated temperature dependence . the final picture for the dynamics of the directed polymer of length @xmath102 at finite temperature @xmath0 and no external force @xmath5 is thus the following : \(a ) during the regime @xmath103 where the coherence length @xmath86 is smaller than the length @xmath102 of the polymer , the polymer is quasi - equilibrated with its equilibrium roughness exponent @xmath13 only on smaller length scale than the coherence length @xmath104 , whereas larger length scale @xmath105 are still completely out - of - equilibrium . the `` infinite disorder fixed point '' describes the hierarchical structure of growing barriers that have to be passed to equilibrate on larger and larger length scales , and this is why the coherence length grows only logarithmically in time @xmath106 \(b ) when the coherence length @xmath86 reaches the length @xmath102 of the polymer , the full polymer is characterized by the equilibrium roughness exponent @xmath13 . the coherence length @xmath86 can not grow anymore , so that the renormalization procedure has to be stopped at the rg scale of order @xmath107 one then expects that the polymer of roughness exponent @xmath13 will become able to move slowly in the transversal direction between metastable states that are separated by distances of order @xmath108 and by barriers distributed with the fixed point distribution ( eq . [ pgammabout ] , [ sigmalinear2 ] and [ pexp ] ) for the scale @xmath109 , that presents the exponential decay @xmath110 on the renormalized scale @xmath109 , one expects that the transverse motion of the center - of - mass corresponds to an effective one - dimensional sinai model , where the unit distance scale is of order @xmath108 and the unit barrier scale is @xmath111 . as a consequence for @xmath112 , one expects the logarithmically slow behavior @xmath113 the `` infinite disorder fixed point '' described above for @xmath5 characterizes some criticality in the time direction in the following sense : there is no characteristic scale for the barriers except the rg scale @xmath72 itself . this scale invariance will be broken by the introduction of some external force @xmath1 that will introduce some characteristic length - scale and thus some corresponding scale @xmath49 for the barriers . however , if the external force is very small , the scale @xmath49 ( that diverges as @xmath4 ) will be very large . we thus expect that the rg flow can be analyzed in terms of two regimes : \(i ) during the first regime @xmath114 , the distribution of the renormalized exit barriers will follow the same rg flow as in the absence of force , i.e. it will converge towards the scaling of eq . [ pgammabout ] , with the same exponential rescaled distribution of eq . [ pexp ] @xmath115 with a width @xmath116 which grows linearly in @xmath72 as in eq . [ sigmalinear2 ] @xmath117 \(ii ) when the large scale @xmath118 is reached , the width saturates at the finite large value @xmath49 @xmath119 instead of the flow towards infinity that characterizes the critical case @xmath5 . since this width remains finite asymptotically as @xmath120 , one speaks of a ` finite - disorder fixed point ' . however , since this width @xmath49 diverges at small force , the region of small @xmath1 is a strong disorder fixed point , where the asymptotic accuracy of the renormalization approach is of order @xmath121 . this notion is thus very useful to study the vicinity of infinite disorder fixed point in the space of parameters and we refer to the review @xcite for more detailed discussions . in one - dimension , where the strong disorder rg procedure can be followed exactly , the crossover of the width between the regimes ( i ) and ( ii ) described above , is of the form ( translated in our present notation ) : @xmath122 $ ] @xcite . for our present analysis , the important point is that the saturation scale @xmath49 is large enough , so that the rg flow during the first regime @xmath123 , that behaves as the critical flow , contains sufficiently rg steps to have converged towards the scaling form of eq . [ pgammtransi ] . so , when saturation occurs at scale @xmath49 , the probability distribution of renormalized exit barrier follows the exponential form @xmath124 the physical meaning of the two regimes described above is as follows . during the first regime ( i ) , the external force @xmath1 is so small that the flow is very similar to the flow for @xmath5 . in particular , the barriers grows upon iteration on scales @xmath125 and the motion is not yet directed along the bias . when the saturation occurs at the large scale @xmath49 , this means on the contrary that the motion becomes effectively directed in the direction of the external force @xmath1 for scales @xmath126 . the barriers against the bias are not renormalized anymore , and one can stop the renormalization procedure . the appropriate model on this scale is then a directed model along the bias , with barriers distributed as in eq . [ distriexp2 ] . as recalled in the introduction ( see the discussion between eqs . [ distriexp ] and [ powerlaw ] ) , the corresponding distribution of the trapping time @xmath127 is then a broad power law @xmath128 with exponent @xmath129 within our present rg framework , the saturation scale @xmath49 should be determined as the limiting value of the width ( eq . [ sigmasaturation ] ) of the renormalized distribution of barriers . to determine the dependence on the external force at small @xmath1 , we may rephrase the scaling argument which is usually used in the field @xcite as follows . on the length scale @xmath90 , the barriers _ in the absence of external force _ follow the scaling of eq . [ barrierpsi ] with the barrier exponent @xmath40 . if the external force @xmath1 is small , one may take into account its effects by a tilt of the landscape that lowers the barriers against the force in the following way @xmath130 where @xmath39 represents the equilibrium roughness exponent , so that the correction in eq . [ barrierpsidrift ] corresponds to a transversal move of order @xmath131 for the segment of length @xmath90 of the polymer . since @xmath132 , the force term always dominates at sufficiently large length scale . the length scale @xmath50 that will give rise to the biggest barriers can be obtained by differentiating eq . [ barrierpsidrift ] with respect to @xmath90 . dropping constants of order @xmath133 one obtains the length scale @xmath134 and the corresponding barrier scale @xmath135 where @xmath36 is the exponent that usually appear in the creep formula ( see eq . [ muusual ] and associated comments ) @xmath136 so the power - law exponent of eq . [ alphaft2 ] vanishes for @xmath4 as @xmath137 in the case of the sinai model , one may actually use exactly the same arguments with the following changes : the length @xmath90 is now along the direction of the motion , the barrier exponent @xmath138 simply describes the brownian fluctuation of the random energy landscape , the prefactor @xmath100 has the simple @xmath0-dependence @xmath139 because the barriers are purely energetic , and there is no roughness exponent @xmath140 since the model concerns a single particle . for this special case , one obtains @xmath141 and the following expression for the power - law exponent @xmath142 in agreement with the exact results mentioned at the beginning of the introduction . in the case of the sinai model , the saturation scales @xmath50 for the length and @xmath49 for the barrier are the only finite scales present in the problem . the strong disorder renormalization procedure should then be stopped at the scale @xmath49 , and one ends up with an effective directed trap model . we refer to @xcite for a more detailed presentation of the quantitative relations that can be derived at large scales between the sinai model in an external force and the one - dimensional directed trap model . for the present model concerning the directed polymer , the polymer length @xmath102 introduces another length scale in the problem . of course , one is interested into large polymer length @xmath143 , but since the length @xmath50 diverges as @xmath4 ( eq . [ letoite ] ) , one needs to distinguish various regimes in terms of the polymer length @xmath102 in the regime @xmath144 , the relevant barriers for the motion of the polymer as a whole corresponds to the rg scale @xmath109 of eq . [ gammalstop ] , which is well below the saturation scale @xmath49 introduced by the force @xmath1 . this means that the force @xmath1 is then too small to impose a directed motion between two neighboring quasi - equilibrated metastable configurations . the motion will become effectively directed only on larger transversal lengthscales separating many metastable configurations . for a polymer of length of order @xmath145 , the motion of the polymer corresponds to an effectively directed trap model . each trap of trapping time @xmath146 corresponding to a transverse displacements of order @xmath147 ( where @xmath148 is a random variable of order @xmath133 with a finite averaged value @xmath149 ) . after @xmath150 traps , the total displacement of the center of mass will be of order @xmath151^{\zeta } \label{xgtot}\end{aligned}\ ] ] whereas the total time needed to escape from these @xmath150 traps will be of order @xmath152 as is well known in the field of lvy statistics @xcite , the sum of @xmath150 variables distributed with the broad power - law of eq . [ powerlaw2 ] grows linearly in @xmath150 only when the average value @xmath153 remains finite , i.e. for @xmath28 @xmath154 however for @xmath155 , the average value of the trapping time diverges @xmath10 and the sum grows more rapidly as @xmath156 it is then clear that the asymptotic velocity is finite only for @xmath157 . for @xmath9 , the velocity vanishes @xmath12 and the center - of - mass displacement grows only sub - linearly in time @xmath158 finally in the regime @xmath159 , the strong disorder rg procedure has to be stopped at the rg scale @xmath49 of eq . [ betoite ] corresponding to the length scale @xmath50 of eq . [ letoite ] . the picture that emerges is then some kind of directed parallel trap model , where a number of order @xmath160 segments of typical length @xmath50 have to move in parallel in a trap landscape where the statistics of trapping times is given by the broad power - law of eq . [ powerlaw2 ] with exponent given in eq . [ alphafinal ] . the main question is then to determine the form of the effective interaction between these large segments from the knowledge of the microscopic interactions between monomers . this issue is discussed in detail in appendix [ appendix ] , where we first recall why the metric constraint is a safe choice for the microscopic interactions ( in contrast with a quadratic elastic energy that is known to lead to an unphysical roughness exponent in some cases ) , and where we then explain why the metric constraint is also an appropriate choice for the effective interactions between consecutive segments of size @xmath50 . from the two properties : \(i ) each segment of size @xmath50 would follow the sub - linear motion of eq . [ xganomalous ] if it were alone \(ii ) the metric constraint between these segments can only delay ( but not accelerate ) each segment with respect to the motion it would have followed if it had been alone we conclude that the center - of - mass motion will also follow the asymptotic sublinear behavior of eq . [ xganomalous ] @xmath161 this is confirmed by the numerical simulations presented in the next section , where we moreover discuss the behavior of the interface width at the largest scales ( i.e. at scales much larger than @xmath50 for the initial microscopic model ) . in this section , we define and study an effective directed model within a landscape made of traps , which is expected to be the appropriate coarse - grained model of the true microscopic non - directed dynamics in the regime @xmath159 discussed above in section [ interactingtrap ] . we stress that a single monomer of this effective model represents a quasi - equilibrated segment of length @xmath50 of internal roughness @xmath13 of the true microscopic model ( see the previous sections for more explanations ) . we consider a directed polymer of length @xmath102 defined by the heights @xmath162 with cyclic boundary conditions @xmath163 and with the metric constraint @xmath164 the initial configuration is the flat zig - zag configuration @xmath165 the traps landscape is defined as follows : the random trapping times @xmath166 are independent and drawn from the power - law distribution @xmath167 the dynamics is defined by a master equation of the form of eq . [ master ] . to respect the chain constraints of eq . [ chain ] , the monomer @xmath168 of the configuration @xmath169 is movable only if @xmath170 , and the possible movement is then @xmath171 . the total rate out of the configuration @xmath66 is thus given by @xmath172 the two main observables in the field of interface dynamics are @xcite \(i ) the height @xmath173 of the center - of - mass of an interface of length @xmath102 as a function of time @xmath73 @xmath174 \(ii ) the width @xmath175 of the interface of length @xmath102 as a function of time @xmath73 @xmath176 ^ 2 \label{w2t}\end{aligned}\ ] ] in our numerical study , we have used the bortz - kalos - lebowitz algorithm @xcite which is a faster - than - the - clock algorithm where each program iteration computes the time and the site where the next movement occurs @xcite . the idea is that from the knowledge of @xmath177 , the escape time @xmath178 from configuration @xmath66 is a random variable drawn from the law @xmath179 then the monomer @xmath168 which is effectively moved @xmath171 is drawn with the probability @xmath180 in this framework , the total displacement @xmath181 of the center - of - mass is the natural variable , and one computes for a given dynamics in a given disordered sample the time @xmath182 needed to reach a given displacement @xmath181 of the center - of - mass . similarly , one computes the width @xmath183 of the interface for a given displacement @xmath181 of the center - of - mass starting from the flat initial condition of eq . [ initial ] . the numerical results presented below correspond to measures at the values @xmath184 , where the maximal displacement @xmath185 wished at the end of the simulation fixes the cpu time . for the case @xmath186 where the averaged trapping time diverges @xmath10 ( eq [ qtau ] ) , we have made studies with @xmath187 and with @xmath188 for the following sizes and the corresponding number @xmath189 of disordered samples : \(i ) for @xmath187 , we have studied sizes in the range @xmath190 with a statistics between @xmath191 and @xmath192 . \(ii ) for @xmath188 , we have studied sizes in the range @xmath193 with a statistics between @xmath194 and @xmath195 . for the case @xmath196 where the averaged trapping time is finite @xmath197 ( eq [ qtau ] ) , we have made studies with @xmath198 for sizes between @xmath199 with a statistics between @xmath200 and @xmath201 . ( note that all other parameters being the same , the cpu time turns out to be larger for the case @xmath202 than for the case @xmath203 , because the number of movable monomers in the stationary regime is larger for the case @xmath202 ) we give below results for single histories in a given disordered sample , as well as disorder - averaged results denoted by an overbar . in particular , we will be interested in @xmath204 that represents the disorder - average of the logarithm of the time needed to reach a displacement @xmath181 of the center - of - mass . for the width defined in eq . [ w2 t ] , we moreover introduce the following notation @xmath205 and of the interface width @xmath175 as a function of the time @xmath73 time for a polymer of length @xmath206 ( a ) for the case @xmath186 where the averaged trapping time diverges @xmath10 ( eq [ qtau ] ) , the dynamics is very intermittent . ( b ) for the case @xmath196 where the averaged trapping time is finite @xmath207 ( eq [ qtau ] ) , the center - of - mass motion is smooth . , title="fig:",height=226 ] and of the interface width @xmath175 as a function of the time @xmath73 time for a polymer of length @xmath206 ( a ) for the case @xmath186 where the averaged trapping time diverges @xmath10 ( eq [ qtau ] ) , the dynamics is very intermittent . ( b ) for the case @xmath196 where the averaged trapping time is finite @xmath207 ( eq [ qtau ] ) , the center - of - mass motion is smooth . , title="fig:",height=226 ] we first consider a single history in a given disordered sample for a polymer of length @xmath206 . the dynamics of the center - of - mass position @xmath208 ( eq . [ hgt ] ) and of the width @xmath175 ( eq . [ w2 t ] ) are shown on fig . [ singleruns ] ( a ) for the case @xmath186 where the averaged trapping time diverges @xmath10 ( eq [ qtau ] ) , and on fig . [ singleruns ] ( b ) for the case @xmath196 where the averaged trapping time is finite @xmath207 ( eq [ qtau ] ) . this comparison shows that the dynamics is very different in the two cases . for @xmath186 , the stationary regime is characterized by a very intermittent dynamics of the center - of - mass @xmath208 and of the width @xmath175 , which remain pinned for long time intervals separated by rapid avalanches . on the contrary for @xmath196 , the motion of the center - of - mass is smooth , and in the stationary regime , the width @xmath175 fluctuates rapidly around its time - averaged value . ( a ) for the case @xmath186 where the averaged trapping time diverges @xmath10 ( eq [ qtau ] ) , very large regions of slope @xmath209 are present ( b ) for the case @xmath196 where the averaged trapping time is finite @xmath207 ( eq [ qtau ] ) , there are more structures on shorter scales . , title="fig:",height=226 ] ( a ) for the case @xmath186 where the averaged trapping time diverges @xmath10 ( eq [ qtau ] ) , very large regions of slope @xmath209 are present ( b ) for the case @xmath196 where the averaged trapping time is finite @xmath207 ( eq [ qtau ] ) , there are more structures on shorter scales . , title="fig:",height=226 ] we show on fig . [ config ] typical corresponding configurations in the stationary regime . for @xmath186 , one clearly see on fig . [ config ] ( a ) very large regions of slope @xmath209 , whereas for @xmath196 , the configuration of fig . [ config ] ( b ) presents more structures on shorter scales . and of the interface width @xmath210 as a function of time ( log - log plot ) : data averaged over @xmath211 disordered samples for a polymer of length @xmath206 ( a ) for the case @xmath186 where the averaged trapping time diverges @xmath10 ( eq [ qtau ] ) , the center - of - mass motion is sublinear with exponent @xmath212 in the transient regime ( eq . [ slopebeta ] ) and with exponent @xmath186 in the stationary regime ( eq . [ slope0.5 ] ) ( b ) comparison with the case @xmath196 where the averaged trapping time is finite @xmath207 ( eq [ qtau ] ) . , title="fig:",height=226 ] and of the interface width @xmath210 as a function of time ( log - log plot ) : data averaged over @xmath211 disordered samples for a polymer of length @xmath206 ( a ) for the case @xmath186 where the averaged trapping time diverges @xmath10 ( eq [ qtau ] ) , the center - of - mass motion is sublinear with exponent @xmath212 in the transient regime ( eq . [ slopebeta ] ) and with exponent @xmath186 in the stationary regime ( eq . [ slope0.5 ] ) ( b ) comparison with the case @xmath196 where the averaged trapping time is finite @xmath207 ( eq [ qtau ] ) . , title="fig:",height=226 ] we have shown above on fig [ singleruns ] the dynamics of @xmath208 and @xmath175 for a single history of a polymer of size @xmath102 . we show on fig . [ average2000 ] the same observables after averaging over @xmath213 disordered samples . we note @xmath214 the crossover time between the transient regime where the disorder - averaged width @xmath215 ( eq . [ wav ] ) grows and the stationary regime where the disorder - averaged width saturates towards a time - independent value @xmath216 for @xmath186 , we find that in the stationary regime , the plot of @xmath217 as a function of @xmath204 ( see fig . [ average2000 ] a ) corresponds to a slope @xmath218 @xmath219 in agreement with the sub - linear motion predicted in eq . [ xganomalousbis ] . of course for @xmath196 we find that in the stationary regime , the plot of @xmath217 as a function of @xmath220 ( see fig . [ average2000 ] b ) corresponds to a slope @xmath209 corresponding to the usual case of finite velocity @xmath221 . for @xmath186 , we moreover measure that , in the initial transient regime , the center - of - mass and the width both grows with the same exponent @xmath222 in time @xmath223 with a value of order @xmath224 that will be interpreted in section [ argument ] below . : ( a ) @xmath225 as a function of @xmath217 for various sizes @xmath102 ( b ) @xmath226 as a function of @xmath227 : the slope corresponds to the roughness exponent @xmath14 . , title="fig:",height=226 ] : ( a ) @xmath225 as a function of @xmath217 for various sizes @xmath102 ( b ) @xmath226 as a function of @xmath227 : the slope corresponds to the roughness exponent @xmath14 . , title="fig:",height=226 ] we now consider the dynamics of the disorder - averaged width defined in eq . [ wav ] as a function of the center - of - mass displacement @xmath181 . for @xmath186 , we show on fig . [ figroughnessmu0.5 ] ( a ) the dynamics of the width @xmath228 for various polymer sizes @xmath102 . we show on fig . [ figroughnessmu0.5 ] ( b ) the log - log plot of the saturation value @xmath229 defined in eq . [ wavsat ] as a function of @xmath102 : we find a slope of order @xmath209 @xmath230 our conclusion is thus that for @xmath155 , the roughness exponent is @xmath231 , as already suggested by the shape of typical configurations in the stationary regime ( see fig . [ config ] a ) ( a ) @xmath225 as a function of @xmath217 for various sizes @xmath102 ( b ) @xmath226 as a function of @xmath227 : the slope is of order @xmath232 . , title="fig:",height=226 ] ( a ) @xmath225 as a function of @xmath217 for various sizes @xmath102 ( b ) @xmath226 as a function of @xmath227 : the slope is of order @xmath232 . , title="fig:",height=226 ] for comparison , we show on fig . [ figroughnessmu2 ] the same observables for the case @xmath196 . in particular , the log - log plot of fig . [ figroughnessmu2 ] ( b ) corresponds to a slope of order @xmath233 for the width in the stationary regime . this high value can be understood from the typical configuration in the stationary regime shown on fig . [ config ] b. this suggests that even in the finite velocity phase @xmath28 where the averaged trapping time is finite ( eq [ qtau ] ) , it is very difficult for the different regions of the polymer to remain synchronized during the dynamics , so that the the polymer tends to be stretched on large distances with a slope close to the maximal slope @xmath209 allowed in the presence of the metric constraint . and @xmath102 the width of the interface in the case @xmath186 : ( a ) @xmath225 as a function of @xmath220 for various sizes @xmath102 ( b ) data collapse of the same data using the variables @xmath234 as a function of @xmath235 ( see eq . [ scalingwav ] ) with @xmath236 . , title="fig:",height=226 ] and @xmath102 the width of the interface in the case @xmath186 : ( a ) @xmath225 as a function of @xmath220 for various sizes @xmath102 ( b ) data collapse of the same data using the variables @xmath234 as a function of @xmath235 ( see eq . [ scalingwav ] ) with @xmath236 . , title="fig:",height=226 ] in the field of interface dynamics , it is usual to summarize the crossover between the transient dynamics and the stationary dynamics by the following scaling form for the width @xcite @xmath237 where @xmath238 represents the roughness in the stationary regime , where @xmath239 is the dynamical exponent of the crossover time @xmath240 and where the scaling function @xmath241 has the following behaviors : \(i ) it converges towards a constant at infinity @xmath242 to recover the saturation value ( eq . [ wavsat ] ) at large time ; \(ii ) it behaves as a power - law at small argument : @xmath243 where the exponent @xmath222 governs the growth in time during the initial transient behavior @xmath244 where the scaling relation @xmath245 between exponents is expected to obtain a @xmath102-independent behavior in the transcient regime @xcite . we show on fig . [ fsswidth ] that a good data collapse can be achieved with the values @xmath14 and @xmath246 the corresponding exponent of the scaling function @xmath241 at small argument is measured to be of order @xmath247 in agreement with the previous estimate of eq . [ mesurebeta ] . equivalently , as a function of the center - of - mass position @xmath181 , the width satisfies the finite - size scaling form @xmath248 where the scaling function @xmath249 has for asymptotic behaviors @xmath250 and @xmath251 , so that in the transient regime , one has the simple proportionality @xmath252 the compatibility between the finite - size scaling forms of eq . [ scalingwav ] and of eq . [ scalingwavhg ] yields the following finite - size scaling form for the center - of - mass position @xmath253 where @xmath254 is a scaling function ( data not shown ) . we expect that the roughness @xmath14 found above numerically for the special case @xmath255 remains the same in the whole phase @xmath26 where strong non - self - averaging effects of the trapping times ( eq [ qtau ] ) occur @xmath256 we may now use a simple argument to determine the other exponents as a function of @xmath257 in the interval @xmath26 . for @xmath181 large , the time @xmath258 needed to make a center - of - mass displacement of order @xmath259 for a polymer of length @xmath102 may be estimated from the maximal @xmath260 trapping time among @xmath261 independent variables drawn with the probability distribution of eq [ qtau ] @xmath262 from this maximal trapping time , the minimal rate transition rate encountered is from eq . [ wouttrap ] @xmath263 assuming that the time @xmath73 scales as the inverse of this minimal rate @xmath264 , we obtain by inversion the following scaling in the stationary regime @xmath265 the scaling function @xmath254 of eq . [ scalinghg ] thus grows asymptotically as the power - law @xmath266 and the dynamical exponent @xmath239 describing the scaling of the crossover time @xmath240 reads @xmath267 for @xmath255 used in our simulations , this corresponds to @xmath268 in agreement with the value measured above ( eqs [ znume ] ) . to determine the exponent @xmath222 of the initial transient regime , the above argument has to be changed as follows . we expect that in the transient regime , an extensive number @xmath269 of monomers ( with a finite density @xmath270 ) are still pinned in their initial positions . equivalently , there is a number @xmath271 of segments of typical length @xmath272 , that have made a move of typical amplitude @xmath273 , because the dynamics is building the roughness @xmath14 at these small scales . as a consequence , the center - of - mass displacement is of order @xmath274 . since all these segments are independent , we now have to evaluate the time @xmath275 for a single segment of size @xmath276 to move over a distance of order @xmath276 , i.e. the time @xmath275 to overcome @xmath277 trapping times . the maximal trapping time among these scales as ( see eq . [ taumax ] ) @xmath278 the time @xmath275 can be thus estimated as in eq . [ wmin ] : @xmath279 . using @xmath280 derived above , we obtain by inversion the transient behavior @xmath281 the scaling function @xmath254 of eq . [ scalinghg ] thus grows at the origin as the power - law @xmath282 i.e. the transient exponent @xmath222 reads @xmath283 for @xmath186 , this is in agreement with the value @xmath284 measured above ( see eqs [ mesurebeta ] and [ betanume ] ) . finally , with the values of eqs [ zetavalue ] , [ zvalue ] and [ betavalue ] , the transient behavior of the width of eq . [ transientwav ] simplifies into the same scaling behavior as the center - of - mass displacement @xmath181 ( eq . [ hgtransient ] ) @xmath285 as expected from eq . [ scalingwavhginitial ] . in conclusion , the dynamics in the zero - velocity phase @xmath26 can be understood in terms of simple arguments based on the statistics of trapping times . the critical exponents that are computed from these arguments are in agreement with our numerical simulations both in the stationary regime and the transient regime . for a polymer with the metric constraint at finite temperature driven by a small force , we may summarize the behavior of the roughness discussed in the above sections as follows : \(i ) at scales smaller than @xmath286 , the force is too small to really change the energy landscape seen by the polymer at @xmath5 , and thus the roughness is expected to be governed by the equilibrium value @xmath13 . \(ii ) at scales larger than @xmath286 , we have found in our numerical study of the previous section that , at small force where the exponent of the trapping time power law ( eqs [ powerlaw2 ] and [ alphaft2 ] ) is in the interval @xmath17 , the roughness becomes @xmath14 . in this section , we compare this scenario with previous works . we are aware of two works where a roughness exponent @xmath14 has been measured for the dynamics in the presence of a metric constraint : \(i ) sneppen has found a roughness exponent @xmath14 for the following self - organized interface dynamics called model a in @xcite : at each iteration , the site with the smallest pinning force among the sites that are allowed to move without breaking the metric constraint , moves forward by one unit and a new pinning force is drawn for this site . self - organized - criticality models are usually not directly related to finite - temperature dynamics . here for instance , the important differences with the effective directed traps model defined by the master equation of eq . [ master ] with eq . [ wouttrap ] are the following : in sneppen s model , there is no real time , but only a number of iterations that corresponds to the total displacement @xmath181 of the center - of - mass ; and at each iteration , it is the always the least pinned movable site that moves , whereas in the master equation of eq [ master ] , the time and the site of the next moves are drawn with eqs [ pexit ] and [ pconfigjump ] ) . nevertheless , there is clearly a close relationship between the two models , in particular in the mechanism that generates a roughness exponent @xmath14 in the stationary state ( see the typical configurations on fig . [ config ] ( a ) of the present paper and fig . 1a of ref . @xcite ) . \(ii ) tang and leschhorn have found in their analysis of the dynamics slightly above the directed percolation depinning threshold @xcite that the moving interface is not self - affine , but is a mixture of two kinds of behaviors : there exists finite pinned segments that have for roughness the exponent of the depinning transition @xmath287 whereas segments with a slope of order @xmath209 are found between them . since the region discussed by tang and leschhorn is on the horizontal axis of fig . 2 slightly above the depinning critical point , it is very far from the region near the vertical axis of fig . 2 discussed in the present paper . however , the two pictures that emerge have nevertheless some similarities : in our case , the local roughness is determined by the equilibrium exponent @xmath13 up to the length @xmath288 , in their case the local roughness is the depinning roughness @xmath287 , but in both cases the roughness becomes equal to 1 at largest scales , i.e. the maximal roughness which is physically acceptable . this seems to indicate that in both cases , it is very difficult for the different regions of the polymer to remain synchronized during the dynamics , so that the maximal slope 1 allowed by the metric constraint is actually reached to maintain the full polymer together . most of the numerical studies on the dynamics at finite temperature have been done for a langevin dynamics in the presence of an elastic quadratic energy between monomers : \(i ) the equilibrium roughness exponent @xmath13 has been measured during the driven dynamics in @xcite , and in @xcite when the temperature remains larger that some disorder - dependent threshold . in our opinion , this means that in these two simulations , the length @xmath102 of the polymer was smaller or of the order of the length @xmath50 . \(ii ) at lower temperature in @xcite , the authors have measured an effective roughness exponent of order @xmath289 . in the study @xcite concerning the limit @xmath290 , the authors have found a crossover between the equilibrium value @xmath13 at small scales and the critical depinning value @xmath291 at larger scales , and have related this crossover to functional rg calculations @xcite . since any roughness exponent @xmath292 is unphysical ( see @xcite and the discussion in appendix a ) , our opinion is that it would be very interesting if the simulations of @xcite based on a quadratic elastic energy were done with the metric constraint instead , to see what physically acceptable roughness exponent @xmath293 would actually emerge at large scales . in this paper , we have considered the dynamics of the directed polymer in a two - dimensional random potential in the regime where the temperature @xmath0 is finite and the external force @xmath1 is small . we have explained how the `` infinite disorder fixed point '' that describes the dynamics for @xmath5 becomes a `` strong disorder fixed point '' for small @xmath1 with an exponential distribution of renormalized barriers . since the corresponding distribution of trapping times then only decays as a power - law @xmath6 , where the exponent @xmath7 vanishes as @xmath294 as @xmath4 , we have concluded that the motion is only sub - linearly in time @xmath11 in the region @xmath9 , i.e. that the asymptotic velocity vanishes @xmath12 , in contrast with the usual creep scenario where the velocity is finite as soon as @xmath295 . all along the paper , we have discussed the similarities with the sinai model with bias , where an analogous zero - velocity phase has been established long ago by rigorous methods @xcite and where the asymptotic exactness of the strong disorder renormalization has been demonstrated explicitly by a direct comparison with the available rigorous results @xcite . we have then checked the presence of the predicted zero - velocity phase by numerical simulations of a directed polymer with a metric constraint driven in a traps landscape . we have moreover obtained that the roughness , which is governed by the equilibrium exponent @xmath13 up to the large scale @xmath288 , is equal to @xmath14 at the largest scales . an important issue is of course whether such a zero - velocity phase also exists for interfaces of higher dimensionalities @xmath84 driven in a random medium of dimension @xmath296 , and more generally for other classes of driven extended systems . since `` collective transport in random media is an impossibly broad subject '' @xcite , a general answer clearly goes beyond the present work . however we think that the essential property needed to have a zero - velocity phase at small force is the presence of a positive barrier exponent @xmath92 for the dynamics at @xmath5 , and this should be the case for a broad class of disordered systems in finite dimensions . then the dynamics for @xmath5 will be logarithmically slow and should correspond to some `` infinite disorder fixed point '' for the renormalized barriers , that transforms into a `` strong disorder fixed point '' at small @xmath1 , with an exponentially distribution of renormalized barriers . another argument in favor of this general scenario is the extremal statistics argument on the barriers @xcite that also lead to an exponential tail for the probability distribution of large barriers , and thus to a power - law decay for the trapping time distribution . it is a pleasure to thank j.p . bouchaud for sending us a copy of his conference proceeding @xcite . in the text , the interaction used to maintain the continuity of the interface is a metric constraint , both for the microscopic model and for the effective directed trap model described in section [ interactingtrap ] . in the present appendix , we explain this choice . [ [ metric - constraint - between - the - heights - of - two - neighboring - monomers - vert - h_i1 - -h_i - vert - leq-1 . ] ] metric constraint between the heights of two neighboring monomers : @xmath298 . ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ this metric constraint is for instance very much used in numerical studies using transfer matrix methods ( see the review @xcite ) . it can can be moreover justified in various microscopic models , in particular when the directed polymer represents an interface in a two - dimensional disordered ferromagnet well below the critical temperature @xmath299 , which was the original motivation to introduce the directed polymer model @xcite . [ [ elastic - energy - of - the - form - e_el - h_i1 - -h_i - sim - h_i1 - -h_in - with - n2-usually . ] ] elastic energy of the form : @xmath300 , with @xmath301 usually . it is of course very common in physics to replace a hard constraint by a soft constraint . in the present case , the use of an elastic quadratic energy usually comes from a small - gradient expansion , an hypothesis which has to be consistent with the results obtained for the roughness . indeed , the roughness exponent @xmath238 has to satisfy the bound @xmath293 for at least two reasons @xcite : first , if one obtains @xmath292 , the gradient is not small , and thus there is an inconsistency with the small gradient expansion used to obtain the elastic quadratic energy ; second , if one obtains @xmath292 , the elastic energy diverges in the thermodynamic limit @xmath302 . the metric constraint can be seen as the limit @xmath303 of the power @xmath150 of the elastic energy . as a consequence , there is room for various universality classes between @xmath301 and @xmath304 . so one should not assume a priori that all these interactions are equivalent , but test for each case of interest if they lead to the same results or not . for the directed polymer at equilibrium at any temperature @xmath0 and no external force in dimension @xmath305 , the same equilibrium roughness exponent @xmath306 arises if one considers the metric constraint or the elastic energy with @xmath301 ( see the review@xcite ) . however , at the zero - temperature depinning transition ( see fig . 2 ) , some subtleties arise ( see the detailed discussion in @xcite ) : it turns out that the model defined in terms of an elastic quadratic energy ( case @xmath301 ) gives a roughness exponent @xmath307 which is unphysical because it is strictly greater than one @xmath308 @xcite . this is in contrast with the model defined either with the metric constraint or with an elastic energy of at least quartic order ( @xmath309 ) where the roughness exponent is physical @xmath310 and actually takes a value @xmath311 @xcite that can be understood in terms of pinning by directed percolation clusters @xcite . in conclusion , since the microscopic model with an elastic quadratic energy ( case @xmath301 ) is known to lead to unphysical results at the depinning transition ( @xmath308 ) , it is clearly not a good starting point to study the general phase diagram for the dynamics at finite temperature and finite force . on the contrary , the metric constraint is always a safe choice , since by definition it can not produce an unphysical roughness strictly larger than one . [ [ interaction - between - the - segments - of - size - lft - in - the - effective - directed - trap - model ] ] interaction between the segments of size @xmath50 in the effective directed trap model ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ for the reasons explained in details above , we thus consider that the _ microscopic _ interactions between monomers are given by the metric constraint . the question is then : in the effective directed trap model described in section [ interactingtrap ] , what is the corresponding renormalized interactions between two consecutive segments of typical length @xmath50 representing local quasi - equilibrated metastable states ? in the absence of more refined arguments , we feel that the metric constraint is actually also an appropriate renormalized interaction , since it prevents the appearance of an unphysical roughness @xmath292 and it allows all possible physically acceptable values @xmath293 for the roughness . bouchaud , j.phys . i france 2 , 1705 ( 1992 ) ; j.p . bouchaud and d. dean , j. phys . i ( france ) * 5 * 265 ( 1995 ) ; a. compte and j. p. bouchaud , j. phys . a : math . gen . 31 6113 ( 1998 ) ; b. rinn , p. maass , and j.p . bouchaud , phys . 84 , 5403 ( 2000 ) and phys . rev . b 64 , 104417 ( 2001 ) ; e.m . bertin , j .- bouchaud , phys . e 67 , 026128 ( 2003 ) and phys . e 67 , 065105(r ) ( 2003 ) . d. s. fisher phys . 69 , 534 - 537 ( 1992 ) ; d. s. fisher , phys . rev . b 50 , 3799 ( 1994 ) ; d. s. fisher , phys . rev . b 51 , 6411 - 6461 ( 1995 ) ; d.s . fisher and a. p. young , phys . rev . b 58 , 9131 ( 1998 ) . fisher , physica a 263 ( 1999 ) 222 . d. fisher , p. le doussal and c. monthus , phys . 80 ( 1998 ) 3539 ; d. s. fisher , p. le doussal and c. monthus , phys . e 59 ( 1999 ) 4795 ; c. monthus and p. le doussal , physica a 334 ( 2004 ) 78 . bouchaud , cond - mat/9910387 , published in soft and fragile matter : nonequilibrium dynamics , metastability and flow , m. e. cates and m. r. evans , eds . , iop publishing ( bristol and philadelphia ) 2000 , pp 285 - 304 l. berthier , v. viasnoff , o. white , v. orlyanchik , f. krzakala in `` slow relaxations and nonequilibrium dynamics in condensed matter '' ; eds : j .- barrat , j. dalibard , m. feigelman , j. kurchan ( springer , berlin , 2003 ) . w. krauth , `` introduction to monte carlo algorithms '' in advances in computer simulation j. kertesz and i. kondor , eds , lecture notes in physics ( springer verlag , 1998 ) ; w. krauth , `` statistical mechanics : algorithms and computations '' , oxford university press ( 2006 ) .

in , c. giller proposed an invariant of ribbon-knots in based on a type of skein relation for a projection to . in certain cases , this invariant is equal to the alexander polynomial for the-knot . giller s invariant is , however , a symmetric polynomial which the alexander polynomial of a-knot need not be . after modifying a-knot into a montesinos twin in a natural way , we show that giller s invariant is related to the seiberg - witten invariant of the exterior of the twin , glued to the complement of a fiber in .

1212.1147

a smooth _ @xmath0-knot _ @xmath4 in the @xmath5-sphere is an embedded @xmath6 in @xmath1 considered up to smooth isotopy . we say that @xmath4 is unknotted when it bounds a smoothly embedded @xmath7 . in contrast to classical knot theory , @xmath0-knots display many pathologies ; their exteriors need not be a @xmath8 @xcite nor does the homeomorphism type of the knot complement determine the knot s isotopy class @xcite . after a surgery which replaces the regular neighborhood @xmath9 with @xmath10 we obtain a homology @xmath11 . with @xmath12 for such a manifold , current tools for smooth @xmath5-manifolds offer few invariants . however , the natural generalizations of gadgets like the alexander polynomial or ideal can be computed for @xmath0-knots . in @xcite , c. giller proposed a definition for an invariant we will call @xmath13 . the proposed invariant is derived from a kind of skein relation on the projection to @xmath2 of a @xmath0-knot analogous to the conway skein relation for projections of @xmath14-knots to @xmath15 . for a certain class of @xmath0-knots , @xmath13 is known to compute the alexander polynomial . this relation , which we discuss more thoroughly in section [ sec : giller ] , always results in a symmetric polynomial . however , the alexander polynomial of a @xmath0-knot need not be symmetrizable . for example , the knot shown in figure [ fig : giller - ex ] has alexander polynomial @xmath16 while giller s polynomial is @xmath17 . further complicating matters is that giller s polynomial was not actually known to be an invariant . if instead of @xmath0-knots , we look at `` twins '' in @xmath1 , we have objects to which gauge - theoretic methods can be applied . it is in this context we see that giller s polynomial computes , in the relevant cases , the seiberg - witten invariant of the exterior of the twin . in @xcite and @xcite , j. m. montesinos introduced the concept of a twin in @xmath1 . such an object consists of a pair of @xmath0-knots @xmath18 which meet transversely twice . as the second homology of @xmath1 is trivial , these intersection points must come with opposite signs . by `` standard twins '' , we will mean that both @xmath19 are unknotted and that the exterior of the pair is diffeomorphic to @xmath20 . we construct these explicitly below . in general , the exterior of a twin is a homology @xmath21 with boundary diffeomorphic to @xmath22 . as @xmath23 is trivial , an orientation on an @xmath0-knot @xmath4 in @xmath1 determines a trivialization @xmath24 . that is , for a @xmath0-knot @xmath4 , all framings are equivalent to the seifert in @xmath1 is a @xmath25 manifold @xmath26 with boundary diffeomorphic @xmath27 , smoothly embedded in @xmath1 so that @xmath28 . as in @xcite , seifert manifolds exist because of the following : consider the map @xmath29 which is given by a trivialization of the normal bundle of @xmath27 . obstructions to extending this map over all of @xmath30 vanish , giving us a map @xmath31 . we can homotope this map to a smooth map which remains equal to the projection @xmath32 on a tubular neighborhood of the boundary . then as @xmath30 is compact there are a finite number of critical points . let @xmath26 be the preimage of one of the regular points . ] framing . fix orientations on the @xmath0-knots , @xmath18 , in a twin @xmath33 and consider @xmath34 . take a simple closed curve @xmath35 on the twin passing through the intersection points of @xmath18 and lift it to @xmath36 using the framings . all such @xmath35 are isotopic on the twin and , as each @xmath19 has a single framing , this lift is canonical . this decomposes @xmath37 the boundaries of normal discs @xmath38 to @xmath18 refine the decomposition to @xmath39 . with this decomposition in mind , we define a standard surgery for a twin @xmath33 in @xmath1 . let @xmath40 where @xmath41 identifies the @xmath42 from @xmath20 with @xmath35 . up to isotopy , any two such @xmath43 differ by a diffeomorphism @xmath44 . since any such map extends over @xmath21 , the resulting manifold only depends on the embedding of the twin . also , @xmath45 so that the @xmath0-torus core , @xmath46 , of the surgery is identified with @xmath47 as a homology class . let @xmath48 be a smooth fiber in an elliptic fibration of @xmath3 , the underlying smooth manifold for a complex @xmath49 surface . the regular neighborhood of the fiber , @xmath50 , comes with a trivialization @xmath51 induced by the fibration giving us the identification @xmath52 . taking a twin @xmath53 , we also have a decomposition @xmath54 . if we fix an identification of this @xmath55 and @xmath48 and an orientation reversing diffeomorphism between @xmath42 and @xmath35 , we obtain an identification @xmath43 of @xmath56 and @xmath57 . let @xmath58 since any automorphism of @xmath48 exends smoothly over @xmath59 , this construction depends on the smooth isotopy class of @xmath48 but is independant , up to diffeomorphism , of the particular choice of @xmath43 . as @xmath60 and the image of @xmath61 normally generates @xmath62 , @xmath63 is simply connected . then an easy meyer - vietoris argument shows that @xmath3 and @xmath63 have the same cohomology ring . by freedman s theorem , @xmath3 and @xmath63 are homeomorphic . this procedure may also be thought of as the generalized fiber sum of @xmath3 and @xmath64 along @xmath48 and the core @xmath46 of the surgered twin with its default framing . as such , we will sometimes write @xmath63 as @xmath65 . the invariant of the twin @xmath33 which we will consider will be the seiberg - witten invariant of @xmath63 . i.e. @xmath66 thought of as an element of the group ring @xmath67 $ ] . precisely , @xmath68 lives in @xmath69 $ ] which we identify with @xmath70 by a homomorphism which extends the identity map on @xmath71 . the invariant is well defined up to a sign which depends on a homology orientation of @xmath72 as the standard twins @xmath73 have complement diffeomorphic to @xmath21 , @xmath74 is diffeomorphic to @xmath3 and @xmath75 now suppose that we have a twin @xmath33 and a disjoint torus @xmath76 in @xmath1 . as @xmath76 is null - homologous , it has a canonical framing given by the outward normal of its seifert manifold . we can then form the self fiber sum @xmath77 . with a choice of an identification of @xmath78 and @xmath76 , the result is well defined and any choice of identification will have the same seiberg - witten polynomial , due to the gluing formulas of @xcite . then form @xmath79 with @xmath80 a pushoff of @xmath46 . our new manifold @xmath81 is a homology @xmath82 with @xmath83 elliptic fibers and @xmath84 with @xmath85)$ ] . see @xcite . we define @xmath86 thought of as a element of the group ring @xmath87 $ ] . take @xmath88 , a knotted @xmath6 , and a twin @xmath89 in @xmath1 . then , selecting one of the spheres @xmath90 in the twin , we form the connected sum @xmath91 at some point away from the 2 double points of the twins . ( this construction is not independent of the choice of @xmath18 in @xmath33 in general . ) if we take @xmath33 to be the standard twins , then this construction _ is _ independent of the choice of @xmath6s and provides a handy method for studying @xmath0-knots via twins . this independence is due to the existence of a orientation preserving diffeomorphism @xmath92 of @xmath1 which interchanges @xmath18 in standard twins . the diffeomorphism @xmath92 is constructed as follows : view @xmath1 as @xmath93 . define @xmath92 on @xmath94 to be the obvious map which interchanges @xmath90 and @xmath95 . then @xmath92 induces the map @xmath96\ ] ] on @xmath97 under the basis for @xmath98 given by @xmath99 and @xmath100 where @xmath101 is the boundary of the normal bundle to @xmath102 . this map then extends over @xmath21 giving @xmath92 on @xmath103 . this construction is originally due to e. artin in @xcite . to each smooth knot @xmath4 in @xmath103 and let @xmath104 be the corresponding arc in @xmath105 obtained by removing a small @xmath105 surrounding a point of @xmath4 . we can arrange @xmath104 so that it meets @xmath106 perpendicularly at the north and south poles . clearly @xmath107 and @xmath108 are diffeomorphic . using the decomposition @xmath109 , we can identify the @xmath5-sphere with the space @xmath110 where @xmath111 iff @xmath112 . let @xmath113 be , respectively , the images of @xmath114 and the @xmath115 under @xmath116 . each is a smoothly embedded @xmath0-sphere which intersect each other pairwise at the images of the north and south poles of @xmath106 . thus @xmath90 and @xmath95 form twins @xmath117 . the first sphere , @xmath90 , bounds @xmath118 in @xmath1 and if @xmath27 is a seifert surface for @xmath4 , then @xmath95 bounds @xmath119 union a @xmath0-handle attached along @xmath120 . the loop @xmath35 , connecting the intersection points of the @xmath19 , bounds a copy @xmath27 in the complement of the twin . in fact , @xmath121 is diffeomorphic to @xmath122 . in the case that @xmath4 is the unknot , we get the standard twins @xmath123 . see figure [ fig : stdtwins ] . when the standard twin surgery is performed on twins formed by artin spinning @xmath4 , the result is the manifold @xmath124 . ( where @xmath125 is the result of zero surgery on @xmath126 . ) this case is identical to the knot surgery considered by fintushel and stern in @xcite and so we have @xmath127 where @xmath128 is the symmetrized alexander polynomial of @xmath4 and @xmath85)$ ] . as before , given a knot @xmath4 in @xmath103 we obtain a knotted arc @xmath104 in @xmath105 with boundary on the north and south poles . for each @xmath129 let @xmath130 be the image of @xmath104 rotated by @xmath131 radians about the @xmath132-axis . the annulus @xmath133 in @xmath134 is obtained by taking the union of @xmath135 . this annulus descends to a knotted @xmath6 , @xmath95 , in @xmath136 . together with @xmath90 , the image of @xmath137 , we form a twin . we call @xmath95 the @xmath138-twist spin of @xmath4 and write @xmath139 and @xmath140 for the associated twin . as with the artin spin , the first sphere , @xmath90 , bounds @xmath141 in @xmath1 . the loop @xmath35 , connecting the intersection points of the @xmath19 , bounds a copy @xmath27 in the complement of the twin . this construction is due to zeeman , who in @xcite , showed that the @xmath138-twist spin of a knot was not isotopic to any @xmath0-knot obtained by artin spinning when @xmath142 . the @xmath143 case is interesting in that the @xmath14-twist spin , @xmath144 is unknotted independently of choice of @xmath4 . however , the twin @xmath145 is typically knotted . this provides an interesting example of twins in which both @xmath0-knots are unknots but with the twins being knotted as an interchangable pair . see @xcite . similar to the twist spin , this construction involves a deformation of a knotted arc @xmath104 in @xmath105 fixing the north and south poles which returns the arc to the starting point . take a international date line of @xmath106 union @xmath104 and push it into @xmath146 so that it is null homologous . call this @xmath104 . then consider the @xmath14-parameter family of diffeomorphisms given by pushing a base point @xmath147 @xmath138 times along @xmath104 . the return map @xmath43 is then a diffeomorphism of the quadruple @xmath148 which is the identity on all but the first component . since @xmath43 is isotopic to the identity rel @xmath106 , @xmath134 is diffeomorphic to @xmath149 . let the annulus @xmath150 be the image of @xmath151 in @xmath152 . then , as before , @xmath150 becomes a @xmath0-knot @xmath95 in @xmath1 after quotienting by @xmath116 . we call @xmath95 the @xmath138-roll spin of @xmath4 and write @xmath153 and @xmath154 for the associated twin . the roll spinning construction is due to fox who , in @xcite , showed that , for @xmath155 , the knotted @xmath0-sphere @xmath95 coming from the deformed arc is not isotopic to any @xmath138-twist spun knot . in this case , the @xmath14-roll spin had a corresponding visualization , duplicated in figure [ fig : foxroll ] , of the motion of @xmath156 in @xmath105 which explains why the word `` roll '' was chosen to describe this construction . note that both twist and roll spinning can be described in terms of certain diffeomorphisms of @xmath105 which keep @xmath106 and @xmath104 fixed identically . with this in mind , we now consider their mutual generalization , deform spinning : let @xmath43 be a self diffeomorphism of @xmath105 keeping @xmath157 and a base point @xmath158 fixed identically . then the mapping torus of @xmath43 is @xmath159 with an embedded annulus @xmath160 which is the quotient of @xmath161 . then after quotienting by the identification @xmath162 for @xmath163 as before , we obtain a knotted @xmath0-sphere @xmath95 , the image of @xmath160 , which together with @xmath90 , the image of @xmath164 , forms a twin . we write @xmath165 and @xmath166 for the twin pair . the isotopy class of @xmath95 and of @xmath166 is determined solely by the isotopy class of @xmath43 . this construction was introduced by litherland in @xcite . diffeomorphisms such as @xmath43 are called `` deformations '' and form a group @xmath167 of deformations modulo isotopy . @xmath167 is isomorphic to the group @xmath168 of automorphisms of @xmath169 preserving the fixed peripheral subgroup given by the image @xmath170 . in this setup we find that @xmath171 corresponds to conjugation by the meridian of @xmath90 and @xmath92 corresponds to conjugation by the longitude of @xmath90 . then if @xmath90 * is not a torus knot , @xmath172 . * is a @xmath173-torus knot , @xmath174 and @xmath175 . in the case that @xmath4 is a composite knot , @xmath176 , the deformation group @xmath177 may be much larger than the subgroup generated by @xmath178 and includes a copy of the pure braid group on @xmath179 strands ; see @xcite . we say that a 2-knot @xmath4 is ribbon if it is formed by the following construction : let @xmath180 ( bases ) , @xmath181 ( bands ) each be embedded in @xmath182 with @xmath183 . if a band intersects a base elsewhere , @xmath184 and @xmath185 the second type of intersection is called a ribbon intersection of @xmath4 or ribbon singularity of @xmath186 . then @xmath187 is a ribbon knot with ribbon presentation given by @xmath186 . we can define ribbon surfaces of arbitrary genus in the same manner . suppose that we have a twin @xmath89 for which the @xmath19 are ribbon . then for each @xmath19 we have a set of bases @xmath188 and bands @xmath189 . we will say that @xmath33 is ribbon if 1 . @xmath190 2 . @xmath191 , @xmath192 with @xmath193 for each band in @xmath194 ( ribbon intersection ) 3 . @xmath195 , @xmath192 with @xmath196 for each band in @xmath197 ( ribbon intersection ) 4 . @xmath198 . more specifically , @xmath199 and @xmath200 meet only in two balls @xmath201 from each and at these intersections ( corresponding to the intersection points @xmath202 ) , we have the following local model : @xmath203 @xmath204 so that taking the boundary of each gives us the cone of the positive hopf link . the @xmath205 case is the same but with orientations reversed on @xmath206 , giving us a negative hopf link cone boundary . in this case we say that @xmath33 is a _ ribbon twin_. it is known that of the deform spun knots , only the artin and 1-twist spun knots are ribbon . hence , at most artin and 1-twist spun twins may be ribbon . artin spun twins are known to be ribbon but it is not known to the author if 1-twist spun twins are ribbon . two ribbon presentations are _ stably equivalent _ if they are equivalent under the following operations . addition of a trivial base / band pair , sliding the disc to which a band attaches ( band slide ) , and moving a ribbon intersection along a base / band sequence ( band pass ) are shown in figure [ fig : band - stabilization ] and together with isotopy generate _ stable equivalence _ of ribbon presentations . clearly stable equivalence of ribbon presentations generates isotopies of the corresponding ribbon knot but the converse also holds isotopic ribbon knots have stably equivalent ribbon presentations . for a proof of this , see @xcite . it will occasionally be easier to deal with simplified ribbon presentations . let @xmath207 be the graph which has vertices corresponding to bases and edges given by bands , connected in the natural way . it is clear that @xmath208 is the genus of the ribbon surface specified by the ribbon presentation . we restrict ourselves to the case where the ribbon surface is a sphere or torus . suppose that @xmath209 has a vertex @xmath147 of valence @xmath25 or greater . then one of the outgoing edges of @xmath147 has a path which ends at a vertex @xmath210 with a single incoming edge . perform the band slide corresponding to this path to get a new ribbon presentation @xmath211 with the same set of bases . in @xmath211 , the valence of @xmath147 has decreased by @xmath14 and the valence of @xmath210 is now @xmath0 . continue this procedure until we arrive at a graph @xmath212 ( and corresponding ribbon presentation ) for which each vertex has valence at most @xmath0 . then , as cell complexes , @xmath212 is either an interval or a circle as the ribbon surface is a sphere or a torus . we will call such ribbon presentations _ linear_. consider the connect sum of a ribbon @xmath0-knot @xmath88 with standard twins @xmath89 . standard twins have a simple ribbon presentation given by two bases and a band each . ( each base is for one of the twin intersection points . ) stabilize the band in @xmath90 by switching it for two bands and a base . then the connect sum @xmath213 is formed by adding a band from the new base of @xmath88 to any of the terminal bases in a ribbon presentation of @xmath90 . in the study of knots in @xmath2 , a generic projections to @xmath15 , together with crossing information , completely determines the isotopy type of a knot . similarly , there is a theory for decorated projections for twins and surfaces in @xmath214 which determines their embedding up to isotopy . giller proves , in @xcite , that if @xmath27 is a surface in @xmath182 , then up to isotopy @xmath27 admits a projection to @xmath2 with only double and triple points exist . further , such projections are generic . in these generic projections , the double points either exist in families which are either simple closed curves or embedded open intervals whose closed endpoints are triple points . see figure [ fig : dbltrpl - pt ] . in the same paper , he gives methods of decorating these projections with over middle under crossing information and a way of determining if an arbitrary set of crossing information gives a lift of such an immersion of a surface in @xmath2 to an embedding in @xmath182 . we will only consider those knots and twins which admit a projection which contain no triple points . not all twins or surfaces have such a projection and those that do are said to be _ simply knotted_. first examples of simply knotted @xmath0-knots include artin spun knots and ribbon @xmath0-knots . for artin spun knots , we can get a projection with no triple points by doing the same spinning construction ( one dimension down ) to the projection to @xmath15 of the original , classical knot . this creates an @xmath215s worth of double points for each crossing in the classical knot s projection . we will call a twin @xmath89 simply knotted if both @xmath19 are simply knotted and pairwise have no triple points . ribbon knots have embedded projections away from the ribbon singularities the intersections of the interiors of bands with interiors of the bases . ( this is in contrast with ribbon @xmath14-knots , for which projections of non - intersecting bands may have crossings . the analogous situation for @xmath0-knots is an under / over crossing of the whole band which does not result in a crossing in the projection to @xmath2 . ) nearby the ribbon singularities , we have projections which appear as in figure [ fig : ribbon - sing ] . it was proved in @xcite that all simply knotted @xmath6s are ribbon . s of double points . ] when we have a @xmath215 family of double points , we have local neighborhoods around each which appear as in the first picture in figure [ fig : dbltrpl - pt ] . this gives the neighborhood of the family the structure of an bundle over @xmath215 . as the surfaces in @xmath182 are orientable and the two preimages of the double points are separated , the monodromy must be trivial . this means that , local to the @xmath215 family of double points , the projection is that of a classical knot crossing times @xmath215 . then , for a simply knotted projection of a ( oriented ) surface in @xmath182 , it is sufficient to label one of the surfaces as being over crossing at each family of double points . we will use `` + '' to denote this . -knot with alexander polynomial @xmath216 and giller polynomial @xmath217 for a twin , a few additional pieces of information are needed . we need to keep track of the two intersection points of the spheres . in @xmath1 , the neighborhood of each is diffeomorphic to the cone on a positive or negative hopf link . then the ( undecorated ) projection of such a neighborhood appears as does a neighborhood of double points . we decorate the projection with a solid dot to indicate the intersection point of the @xmath6s and @xmath218 signs to indicate over / under crossings on the double point arcs which emanate from it . we switch from over to under at the intersection of the spheres in twins . see figure [ fig : twin - int ] in @xcite , satoh showed how to represent ribbon surfaces of genus @xmath219 and @xmath14 in @xmath182 by means of virtual knots / links . for our purposes , a virtual knot ( or link ) is a diagram in @xmath15 of embedded , oriented arcs which end either at `` crossings '' as in the ( top ) first two pictures in figure [ fig : virtual - knot ] or at endpoints as in the ( top ) third picture . each such diagram corresponds to a collection of immersed surfaces in @xmath2 by replacing each of the crossings and endpoints in figure [ fig : virtual - knot ] with the corresponding surfaces in @xmath2 shown . these are then connected via tubes parallel to the embedded arcs . thus , any virtual link corresponds to the projection , with crossing information , of a collection of ribbon @xmath0-spheres and tori in @xmath182 . ] conversely , linear ribbon presentations of knots correspond to virtual knots . take a projection of @xmath220 having only double points at ribbon singularities . for each band in the linear ribbon presentation , consider the image of its core in @xmath2 extended to the center of the bases to which the band attaches . this gives an immersed ( at ribbon singularities ) arc @xmath104 in @xmath2 . taking a generic projection @xmath221 we get an arc @xmath222 immersed in @xmath15 with two kinds of singularities : * double points of the projection @xmath223 and * projections of immersion points @xmath224 . each of the first kind of double point corresponds to a virtual crossing . for the second kind of crossing we must first consider a diversion about orientations . the endpoints of @xmath222 each correspond to a base with only one band attached here , @xmath4 locally consists of the discs @xmath225 . with a fixed orientation on @xmath4 , we orient the boundary of the @xmath188 with the outward normal . we then say that the endpoint of @xmath222 is out / in as the boundary orientation on the @xmath188 is counterclockwise or clockwise , respectively ( when @xmath188 is orientation - preserving identified with the unit complex disc . ) this orients @xmath222 . then , with @xmath222 oriented , we can check that the second type of immersion point corresponds to the ribbon intersection in figure [ fig : virtual - knot ] . if it does not ( i.e. the two crossings have the opposite under / over information ) then perform the isotopy in figure [ fig : vk - fix ] . once this has been done , we may use our correspondence from figure [ fig : virtual - knot ] to label each of the immersion points of @xmath222 as virtual knot crossings . in addition to the `` classical '' reidemeister moves in figure [ fig : classical - reidemeister ] , associated to a virtual knot , we have the series of `` virtual '' reidemeister moves in figure [ fig : virt - reidemeister ] giving allowable isotopies . notice that move d is one of the forbidden moves of the virtual knots of kauffman . the type of virtual knot we consider here is sometimes referred to a being _ weakly _ virtual but in the spirit of brevity we will omit `` weakly '' in this paper . these concepts of virtual knots are inequivalent as there are virtual knots , in the sense of kauffman , which are knotted and which , when move d is allowed , are unknotted . such an example is given in @xcite . we will add additonal markings to describe ribbon twins . for a ribbon twin @xmath89 , there are two bases @xmath201 in the ribbon representation of each @xmath19 which correspond to the twin intersection points @xmath202 . perform band slides until the ribbon presentations of the @xmath19 are linear with endpoints @xmath201 . then we have corresponding virtual knot representations of the @xmath19 with identical endpoints . we will use @xmath226 to mark each of these as they correspond to the cone on the positive and negative hopf bands at the intersection points . with this in mind , we get the moves in figure [ fig : reidemeister - move - f ] . , @xmath227 for twins , versions for @xmath228 are identical ] now let us consider the connect sum of a ribbon @xmath0-knot @xmath88 with standard twins @xmath89 as described in section [ sec : ribbon ] . recall that , in this construction , we formed the connect sum by adding a single band between bases of @xmath90 and @xmath88 . we can assume that @xmath90 has a ribbon presentation with @xmath25 bases , @xmath229 with the positive / negative twin intersections occuring at @xmath230 and @xmath231 respectively . this ribbon presentation has @xmath0 bands , @xmath232 from @xmath230 to @xmath233 and @xmath234 from @xmath233 to @xmath231 . assume that @xmath88 is given by a virtual knot diagram and hence that it has a linear ribbon decomposition . let @xmath235 be a band connecting the base @xmath233 on @xmath90 and endpoints of the linear ribbon decomposition of @xmath88 . then @xmath236 has a ribbon decomposition in the shape of a @xmath76 , with the top bar cosisting of the bands @xmath237 and the shaft consiting of @xmath235 and the ribbon decomposition for @xmath88 . to get a virtual knot diagram for the twin @xmath238 , we will need to perform a series of band slides sliding one of @xmath237 along the ribbon decomposition of @xmath88 to the other endpoint . this is straightforward and results in a ribbon decomposition whose virtual knot diagram can be obtained from that of @xmath88 by 1 . replace each strand of the virtual knot for @xmath88 with two parallel strands , 2 . at the two endpoints of the virtual knot , replace the endpoint with one of diagrams show in the bottom of the first two columns of figure [ fig : conn - sum ] , using both , 3 . replace crossings for the virtual knot for @xmath88 with configurations as shown in figure [ fig : conn - sum ] . an example is shown in figure [ fig : giller - twin0 ] . another technique for obtaining a ribbon twin from a ribbon @xmath0-knot @xmath90 is to take a virtual knot presentation for @xmath90 and perform virtual reidemesiter moves so that the endpoint - bases sit in the unbounded region of the plane . then connect these using an crossingless arc sitting in the unbounded face this gives the second , unknotted , @xmath0-sphere @xmath95 of the twin . for example , see figure [ fig : giller - twin ] as discussed earlier , a twin in @xmath1 has a canonical surgery associated to it . since our decorated projections determine isotopy type , no additional information is needed to carry out surgery . for a @xmath239 in @xmath1 , however , we will need additional information . as any @xmath240 is nullhomologous , it bounds a seifert manifold which , via its inward normal , gives a seifert framing for the @xmath239 . this gives us a decomposition , @xmath241 . as surgery replacing @xmath242 with @xmath21 is determined by the image of @xmath243 , we see that we can entirely describe surgery by specifying a curve on @xmath239 and an integer giving the winding about a meridian ( boundary of normal disc ) to @xmath239 . see figure [ fig : surgery - info ] . when the @xmath239 is ribbon with a linear ribbon presentation and corresponding virtual knot diagram , we can decompose @xmath239 in the following manner : let @xmath244 be the core of the ribbon presentation , projected to @xmath2 . let @xmath245 be an essential loop on @xmath239 which , when projected to @xmath2 is null - homologous in @xmath246 . any such loop represents the same homology class on @xmath239 . let @xmath247 be @xmath248 in a band in the ribbon presentation . orient @xmath245 to coincide with the orientation of the virtual knot diagram . then orient @xmath247 so that @xmath249 with respect to the orientation on @xmath239 . so @xmath250 . then , in the virtual knot diagram , labeling the knot corresponding to @xmath239 with @xmath251 where @xmath35 is the winding number of the attached @xmath243 with respect to the seifert framing and @xmath252 is the slope of @xmath243 projected to @xmath239 . we will write an @xmath102 together with @xmath253s on the appropriate components when we wish to denote this surgery . in @xcite , c. giller defines a polynomial @xmath254 of simply knotted @xmath6s in @xmath1 . this supposed invariant obeys a skein relation similar to that of the alexander polynomial for classical knots . that is , consider a embedded circle of double points in a projection of a ( collection of ) oriented sphere or torus in @xmath182 . as mentioned before , we can trivialize the neighborhood of the double points so that we have the neighborhood of a classical knot crossing times @xmath215 . all surfaces in question are oriented and so orient the double points of their projection this orients both strands in the classical picture . we can then replace this neighborhood with @xmath215 times any of the @xmath25 options in figure [ fig : resolution ] , obtaining the invariant is then defined by the relation : @xmath255 together with @xmath256 and @xmath257 giller also describes @xmath13 in a manner similar to that of the alexander polynomial . that is , letting @xmath26 be a seifert manifold for @xmath4 , he forms the infinte cyclic cover @xmath258 of @xmath259 and presents @xmath260 as a @xmath261 $ ] module . then @xmath13 is defined by @xmath262/(\delta_g)$ ] where @xmath76 is the @xmath261 $ ] torsion part of @xmath260 . whenever @xmath4 is ribbon , we can choose @xmath26 to be a punctured @xmath263 given by the ribbon presentation . it is easy to verify by standard arguments that isotopies and band - stabilizations of the ribbon presentation yeild the same @xmath13 . therefore , @xmath13 is well - defined for ribbon knots . for artin spun knots , giller s polynomial is the alexander polynomial . in the case that we apply these computations to the projection of the knotted sphere in an artin spun twin , giller s polynomial is the seiberg - witten polynomial . ( as shown in @xcite ) . interestingly , @xmath0-knots and twins need not have a symmetric alexander polynomial . giller s polynomial and the seiberg - witten polynomial , however , are symmetric . for example , see figure [ fig : giller - ex ] , which is the spun right hand trefoil with crossing changes . the natural questions to ask are then : is giller s polynomial an invariant of @xmath0-knots ? if so , is it equal to the seiberg - witten polynomial for the corresponding twin ? for twins , what is the relationship between the alexander polynomial and the seiberg - witten invariant ? our invariant provides suggestive evidence that the second question , at least in some cases , should be answered in the affirmative . the main theorem of fintushel and stern in @xcite gives a way of computing the seiberg - witten invariants of classical - knot surgered @xmath5-manifolds in terms of the symmetrized alexander polynomial of the knot . the proof relies on a technique j. hoste developed in @xcite which is a method for obtaining kirby calculus diagrams for so called `` sewn - up @xmath264-link exteriors '' in @xmath103 . ( like fintushel and stern , we will only consider the case where @xmath264-links are actually knots and links . ) we discuss a simplified but sufficient version of the original move below so to demonstrate the ideas involved . -dimensional hoste move ] a sewn up knot exterior is formed by taking either two oriented knots in one copy of @xmath103 or in two separate copies , excising a normal neighborhood of each knot , and gluing the resulting boundary @xmath239s by a diffeomorphism . for our purposes , we will let the diffeomorphism be the one which identifies oriented meridians and longitudes for the seifert framings of each knot . this procedure does two things , it removes two copies of @xmath265 with a chosen framing and orientation , and it replaces them with an @xmath266 . together , these are the boundary of @xmath267 . thus we may think of forming a sewn up link exterior as the result ( on the boundary ) of adding a round @xmath5-dimensional @xmath14-handle to @xmath268 so that the feet of the round @xmath14-handle are the two knots , each with the proper framing . now , consider a projection of a link @xmath269 in @xmath103 with oriented components @xmath90 , @xmath95 and a small region in the projection where @xmath90 and @xmath95 run parallel but in opposite directions . we can then connect @xmath90 and @xmath95 via an arc . see figure [ fig : hoste2 ] . attach the round handle as above to form the sewn up link exterior for @xmath90 and @xmath95 . note that we can choose the attaching map of the round handle so that in the corresponding morse - bott function , the points @xmath270 on @xmath18 where the arc touches each knot are both connected to the same point @xmath158 on the critical @xmath215 by gradient flow lines . take a perfect morse function on the critical @xmath215 of the round handle so that the index zero critical point is @xmath158 . this decomposes the round @xmath14-handle into a @xmath14-handle and @xmath0-handle corresponding to the @xmath219- and @xmath14-handles of the morse function on @xmath215 . and @xmath0 handle ] attaching the @xmath14-handle to @xmath103 results in self connect summing @xmath103 at the points @xmath270 . by standard tricks , this is the same as zero surgery on the unknot around the arc in the second drawing in figure [ fig : hoste2 ] . the attaching circle of the @xmath0-handle is the `` band sum '' @xmath4 of @xmath90 and @xmath95 as shown in figure [ fig : hoste1 ] and the second picture in figure [ fig : hoste2 ] . attaching the @xmath0-handle to the result ( an @xmath271 ) of the previous surgery is then merely zero surgery on @xmath4 . in @xcite , the hoste move shows up in @xmath5-dimensions with an @xmath215 equivariance as we cross the @xmath25-manifold with @xmath215 . when that is done , the surgeries on knots show up as surgeries on square zero tori which , by using @xcite , are amenable to computations of the seiberg - witten invariant . the @xmath5-dimensional version of the hoste move we will discuss here does not assume this @xmath215 equivariance , although local @xmath215 equivariances will occur . [ prop:4d - hoste1 ] consider two embedded , oriented square zero tori @xmath272 in a @xmath5-manifold @xmath258 . suppose that @xmath272 are connected by an annulus @xmath273 , embedded in @xmath258 , so that @xmath274 consists of an essential curve on each torus . let each @xmath275 be framed so that @xmath276 is in the subspace of @xmath277 generated by the pushoffs of loops on @xmath275 with respect to the framing . let @xmath43 be a diffeomorphism @xmath278 which identifies the components of @xmath274 in @xmath279 and @xmath280 . then the self fiber sum , @xmath281 , is also the result of surgery on two tori : the `` band sum '' of the tori along @xmath150 and torus given by the loop in figure [ fig : band - sum ] in the neighborhood of @xmath282 for each @xmath283 . we can reinterpret the fiber sum as the result ( on the boundary ) of adding a @xmath284-dimensional toric @xmath14-handle ( a @xmath285 ) to @xmath286 so that the attaching region , @xmath287 is identified with the normal bundles to @xmath279 and @xmath280 with their chosen framing . this results in deleting the two @xmath21s and replacing them with a @xmath288 . this identification is determined by choice of framings for @xmath272 and a diffeomorphism @xmath43 between them . choose a factorization @xmath289 so that the @xmath290 factor is @xmath274 in both @xmath275s . the critical @xmath239 for the @xmath284-dimensional toric handle is identified with the @xmath275s by the gradient flow . pick a perfect morse function on @xmath291 and perturb the morse - bott function on the @xmath284-dimensional handle by an extension of it . this gives us a reinterpretation of the @xmath284-dimensional toric @xmath14-handle as two round ( @xmath215 ) handles a round @xmath0-handle and a round @xmath14-handle corresponding to the critical points of the morse function on @xmath291 . consider the round @xmath14-handle first . such a handle is a @xmath292 so that it attaches along @xmath293 . by construction , the two @xmath294s are neighborhoods of the components of @xmath274 , with framing given by the inward normal along @xmath150 , a vector field along @xmath274 parallel to @xmath275 , and a third vector field defined by orthogonality to these and the tangent space to @xmath274 . consider a neighborhood of @xmath150 which is @xmath215 equivariant , matching the @xmath215 equivariance of @xmath295 . when small , such a neighborhood is diffeomorphic to @xmath215 times the `` h '' in figure [ fig : band - sum ] . ( the vertical lines are in @xmath275 ; the horizontal , slices of @xmath150 . ) attaching the round @xmath14-handle is the same as ( equivariantly ) self - connect summing at the places where the vertical lines intersect the horizontal core of the band . in each @xmath25-manifold slice , this is equivalent to performing zero surgery on the loop linking the band in figure [ fig : band - sum ] . then , in turn , this gives us a square zero torus @xmath269 and a surgery to perform on it within the neighborhood of @xmath150 . now , a @xmath284-dimensional round @xmath0-handle is a @xmath296 attached along @xmath297 . outside of the neighborhood of @xmath150 , the attaching torus is equal to the @xmath275 . ( two annuli ) inside the neighborhood of @xmath150 , the attaching torus @xmath76 is given by @xmath215 times the boundary of the band in figure [ fig : band - sum ] . ( two more annuli ) the framing of this torus is given by the framings of the @xmath275 outside the neighborhood of @xmath150 and by the inward normal to the band on ( each slice of ) the inside . this can be seen by band summing pushoffs of the @xmath275 . ( by hypothesis , @xmath150 has zero winding with respect to our framing so this can be done by band summing in the @xmath215 trivialization . ) @xmath76 inherits a factorization @xmath298 from the @xmath275 by noting that the @xmath290 factors of the @xmath275 survive and that the @xmath291 factors are themselves band summed . attaching the round @xmath0-handle then performs a surgery on this torus which sends @xmath243 to the @xmath299 factor . let us now examine how proposition [ prop:4d - hoste1 ] can be described in terms of surgery information on projections . first we will look at the case when the tori come from surgeries on a twin and a torus . let @xmath258 be the result of standard surgery on a ribbon twin @xmath33 and @xmath300 surgery on a ribbon torus @xmath76 in @xmath1 , both specified by a virtual knot diagram as in sections [ subsec : virtual - knot - diag ] and [ subsec : surgery - diagram ] . note that the cores @xmath301 of each surgery inherit preferred framings from their surgery description . suppose that @xmath33 and @xmath76 have a classical knot crossing in the virtual knot diagram and hence a ribbon intersection . then , in a projection to @xmath2 , there is a neighborhood as in figure [ fig : crossing - change - band ] . consider the annuli shown in the figure . each of these annuli are isotopic . this can be seen by the fact that on the left of each picture , the horizontal surface overcrosses the vertical surface so the annulus must lie completely under the horizontal surface to the left of the ribbon singularity . thus we can isotope the annulus freely on the left of the ribbon singularity . similarly , on the right the annulus lies completely above the horizontal surface and so we can isotope it freely on that side . now let @xmath302 be the particular representation of the annulus corresponding to the particular orientations of the virtual knot crossing depicted below it . now , @xmath302 connects an equator @xmath303 to one of @xmath19 in @xmath89 to a essential curve @xmath304 on the torus . since we have done @xmath300 surgery on the torus ( in virtual knot notation ) , the surgery curve ( in the projection notation ) meets @xmath304 once . used for smoothing ] the method of constructing @xmath302 ensures that @xmath305 consists of essential curves on the cores of the twin and torus surgeries when @xmath302 is extended by the projection to the surgered manifold . this means that we can apply proposition [ prop:4d - hoste1 ] once we have chosen the diffeomorphism @xmath43 between @xmath301 . we have already required that @xmath43 identify the components of @xmath305 . @xmath43 is then determined when we require that it identify the following : 1 . the projection to @xmath306 of @xmath307 , where @xmath308 , @xmath309 the normal bundle , and 2 . the projection to @xmath78 of @xmath310 , where @xmath311 and lies in the @xmath6 which does not contain the preimage of the double points . @xmath243 is the boundary of the fiber of the normal bundle to this @xmath6 . with this data fixed , proposition [ prop:4d - hoste1 ] can be applied . the result is that @xmath312 sewn up by @xmath43 is diffeomorphic to @xmath313 where 1 . @xmath314 denotes the `` band sum '' of @xmath33 and @xmath76 along @xmath302 , 2 . the first @xmath21 is glued to @xmath315 by the standard surgery on twins , 3 . @xmath171 is the torus given by the loop in figure [ fig : band - sum ] , and 4 . the second @xmath21 is glued to @xmath316 so that @xmath243 goes to the nullhomologous pushoff of the loop in figure [ fig : band - sum ] . finally , we can isotopy this region in @xmath314 to be as shown in one of the pictures in the top row of figure [ fig : smoothing ] . the appropriate smoothing depends on the orientation of the horizontal surface and corresponds to the selection of band previously . in figure [ fig : smoothing ] , the correspondence of the orientation of the horizontal surface to the virtual knot diagrams is show by the original diagrams to the lower left of each smoothing and the smoothed virtual knot diagrams to the lower right . with corresponding virtual knot diagrams ] we will require the hoste move between two tori in only one case . suppose that both tori lie in a neighborhood diffeomorphic to @xmath294 so that in each @xmath317 the tori are as shown in figure [ fig : hoste3 ] . the we only need to describe one aspect of the @xmath25-dimensional local picture the gluing map for the sewn up exterior for the right hand side of figure [ fig : hoste3 ] . this map is given by identifying meridians to each loop and the pushoffs along the obvious once punctured discs to each . times the left is @xmath215 times the right with the two starred tori sewn up ] consider a classical crossing in a virtual knot diagram for a twin and/or a torus and the corresponding annulus from figure [ fig : crossing - change - band2 ] . notice that the correspondence is reversed from that of the @xmath318 hoste move . as before , both bands shown are isotopic . push the horizontal surface along the annulus in figure [ fig : crossing - change - band2 ] to get the configuration in figure [ fig : ribbon - sing - isotope ] . used for crossing change ] by pushing horizontal tube along band , over crossing the vertical tube in the two new sets of double points . the new pair of crossings is locally modeled on @xmath215 times the diagram in the lower right . the diagram corresponding to the second picture in figure [ fig : crossing - change - band2 ] is similar . ] focus our attention to the lower of the two new loops of crossings in figure [ fig : ribbon - sing - isotope ] ( or the corresponding picture for the other band . ) call this crossing @xmath244 . local to @xmath244 , we have the model of @xmath215 times a @xmath25-dimensional oriented knot crossing . note that if we form an annulus by taking a path from the lower to the upper double point in each @xmath25-manifold picture , we get an annulus @xmath150 which is isotopic to the annulus @xmath302 from before . perform one of the surgeries on a torus indicated by the @xmath25-dimensional pictures in figure [ fig : crossing - change - local ] localized at @xmath244 in figure [ fig : ribbon - sing - isotope ] ( or the corresponding picture for the other band . ) the appropriate surgery is the top for the first picture of figure [ fig : crossing - change - band2 ] and the lower for the second . this changes the crossing @xmath244 from an overcrossing of the horizontal surface to an under crossing , resulting in figures [ fig : crossing - change - isotopy ] and [ fig : crossing - change - isotopy2 ] . finally , we perform the isotopies indicated in these figures and see that our result has changed the crossing type from a classical @xmath218 to @xmath319 or from a classical @xmath319 to @xmath218 in the virtual knot diagram . it is important to note that the torus which we have surgered is isotopic to the surgered torus @xmath171 of section [ sec : twin - torus ] . we identify the two via the isotopy of the band @xmath302 . -dimensional picture ] after surgery and the corresponding virtual knot crossing of result ] after surgery and the corresponding virtual knot crossing of result ] consider a twin @xmath320 given by a virtual knot presentation and the manifold @xmath321 . suppose that the virtual knot presentation for @xmath90 contains a classical crossing ; so @xmath90 has a ribbon intersection with itself . let @xmath322 , @xmath323 and @xmath324 be the results of replacing the crossing in the virtual knot diagram with the three options in figure [ fig : skein - relation ] . note that @xmath324 will actually be a twin and a torus . consider the square zero torus @xmath171 from the previous sections . now , @xmath316 admits a decomposition @xmath325 from it lying in the @xmath215 equivariant neighborhood of the annulus @xmath302 . namely , in each @xmath25-manifold slice of the neighborhood , @xmath171 is given by a loop @xmath245 linking the slice of @xmath302 once . then @xmath326 where @xmath247 is @xmath327 in the equivariant neighborhood of @xmath302 . finally @xmath328 finishes the decomposition . log transform surgery on @xmath171 is defined by removing @xmath329 and replacing with another copy of @xmath20 . such a surgery is uniquely determined up to diffeomorphism by the image of @xmath243 in @xmath330 . using our decomposition above , we can describe such a surgery by a triplet of integers @xmath331 with no common factor . such a triplet gives specifies the isotopy type of the curve to which we will glue @xmath42 . we will write @xmath332 for the result of the @xmath331 log transform surgery on @xmath258 . from section [ sec : crossing - change ] we see that we can move from @xmath323 to @xmath322 by surgery on @xmath333 which in each @xmath25-manifold slice is @xmath334 surgery on the loop which gives @xmath171 . thus we can move from @xmath323 to @xmath322 by a @xmath335 log transform on @xmath171 . note that @xmath336 surgery on @xmath171 is the identity . morgan , mrowka , and szabo s formula in @xcite then gives @xmath337 now by our description in section [ sec : twin - torus ] , @xmath338 is the sewn up twin / torus exterior of @xmath324 fiber summed to @xmath3 . thus , by our definition of @xmath339 , @xmath340 while @xmath341 and @xmath342 therefore , @xmath343 now consider a twin @xmath320 and torus @xmath344 given by a virtual knot presentation and the manifold @xmath321 . suppose that the virtual knot presentation contains a classical crossing between @xmath90 and @xmath344 ; so @xmath90 has a ribbon intersection with @xmath76 . let @xmath322 , @xmath323 and @xmath324 be the results of replacing the crossing in the virtual knot diagram with the three options in figure [ fig : skein - relation ] . note that @xmath345 will each be a twin and a torus while @xmath324 will be a single twin . as before , we consider the torus @xmath346 . from section [ sec : crossing - change ] we see that we can move from @xmath323 to @xmath322 by surgery on @xmath333 which in each @xmath25-manifold slice is @xmath334 surgery on the loop which gives @xmath171 . thus we can move from @xmath323 to @xmath322 by a @xmath335 log transform on @xmath171 . note that @xmath336 surgery on @xmath171 is the identity . morgan , mrowka , and szabo s formula in @xcite then gives @xmath337 now , since each @xmath345 are composed of a torus and a twin , the manifolds @xmath347 are each sewn up twin / torus exteriors fiber summed to @xmath3 . thus @xmath348 . nearby @xmath171 we have a local model for @xmath323 given in the left picture of figure [ fig : hoste4 ] . if we perform the @xmath318 hoste move on @xmath338 at an annulus equal to @xmath215 times the horizontal line in this picture , we obtain the picture to the right in figure [ fig : hoste4 ] . note that this picture is identical to that of figure [ fig : hoste3 ] . applying the version of the @xmath318 hoste move from section [ sec : torus - torus ] , we get a local picture equal to that in the right hand side of figure [ fig : hoste3 ] . the tori in this picture are each isotopic to the torus @xmath349 where @xmath350 is the normal bundle to @xmath19 , the knots which comprise @xmath324 . this torus can also be described as one of the components of @xmath351 . in the fiber sum manifold , @xmath338 , each torus we have just described is isotopic to the fiber @xmath48 . thus , @xmath338 is diffeomorphic to @xmath352 . therefore , @xmath353 so @xmath354 and @xmath355 so @xmath343 recall the definition of giller s polynomial given by equations ( [ eq : giller - conway - relation ] ) , ( [ eq : giller - unknot - relation ] ) , and ( [ eq : giller - split - relation ] ) . these are the conway - style relation , the value on the unknotted sphere , and the vanishing of the polynomial for split links , respectively . we now discuss similar results for our invariant . suppose that @xmath89 and @xmath76 are ribbon with a virtual knot presentation which is split or only has pairwise virtual crossings . then using the virtual reidemeister moves of figures [ fig : virt - reidemeister ] and [ fig : reidemeister - move - f ] , we can separate @xmath33 and @xmath76 . this means that the projections of @xmath33 and @xmath76 are separated by an @xmath103 . so @xmath33 and @xmath76 are separated by an @xmath103 . this means that the manifold formed by surgering @xmath33 and @xmath76 is a connected sum . now , on the @xmath76 side of the connected sum , the intersection form on @xmath357 will be a hyperbolic pair as will the intersection form on the side given by surgery on @xmath33 . thus we are given a manifold which is the connected sum of two manifolds each with @xmath358 . therefore , the seiberg - witten invariant of this surgery vanishes . it follows that @xmath359 and that @xmath360 now let us consider the conway - style relation for giller s polynomial we initially discussed in section [ sec : giller ] . this relation involves crossing changes and resolution at individual loops of double points . now , in our crossing change surgery , we had a similar action of changing the lower crossing from figure [ fig : ribbon - sing - isotope ] . in the diagrams for our @xmath318 hoste move , we took a different local projection to illustrate the appropriate surgery . however , smoothing the lower crossing from figure [ fig : ribbon - sing - isotope ] also yields a smoothing in the virtual knot diagram . in other words , selecting a particular set of double points to apply the relation in equation ( [ eq : giller - conway - relation ] ) to , @xmath361 computes @xmath362 . therefore , @xmath363 for ribbon knots . we can not make a stronger statement of equality however , as the relation from equation ( [ eq : giller - conway - relation ] ) allows us to move into configurations of surfaces which are inaccessible to the invariant @xmath339 . now suppose that @xmath89 is ribbon with a virtual knot presentation which only has virtual crossings . then we can use the virtual reidemeister moves @xmath302 and @xmath48 from figures [ fig : virt - reidemeister ] and [ fig : reidemeister - move - f ] to completely unknot the diagram for @xmath33 . therefore , @xmath364 and so @xmath365 . suppose that @xmath89 is a twin possibly with accompanying torus @xmath76 with the configuration ribbon . suppose that we reverse the orientation of one of the @xmath19 or of @xmath76 . this reverses the orientation of the torus @xmath78 ( or @xmath306 ) and induces a chance in homology orientation from the change of sign in pairing with @xmath78 . then , if @xmath366 , @xmath367 . similarly , @xmath368 . currently , the author is unaware if crossings can be chosen so that the tree of terminates in standard twins and unlinked twin / torus pairs . the previous work of fintushel and stern guarantees that the process terminates when @xmath90 in @xmath89 is knotted with only classical crossings and @xmath95 is unknotted with no ribbon intersections with @xmath90 . the presence of virtual crossings in the diagrams complicates the general case . additionally , the author has yet to find a general method of dealing with ribbon intersections between the @xmath19 . however , there seem to be a fairly large number of new examples which we may compute using the current tools . the first we will compute is the twin version of the example from giller s paper . call this twin @xmath369 this was encountered previously in figure [ fig : giller - twin ] . follow the computation through figures [ fig : computation - giller - twin ] , [ fig : computation - giller - twin2 ] , [ fig : computation - giller - twin3 ] , and [ fig : computation - giller - twin4 ] . in figure [ fig : computation - giller - twin4 ] , we arrive at configurations @xmath370 and @xmath48 . here @xmath244 and @xmath48 are isotopic to standard twins , so @xmath371 . the other configurations @xmath372 differ by the orientation of the torus and so @xmath373 . therefore , @xmath374 now let us look at the twin in figure [ fig : unknot - pair - twin ] , a twin in which both @xmath0-knots are unknots but which pairwise have ribbon intersections . call this twin @xmath375 . our current tools do not allow us to deal directly with pairwise ribbon intersections . follow the computation through figures [ fig : unknot - pair - twin1 ] , [ fig : unknot - pair - twin2 ] , and [ fig : unknot - pair - twin3 ] . we arrive at configurations @xmath376 and @xmath269 . here @xmath377 and @xmath269 are isotopic to standard twins , so @xmath378 . the other configurations @xmath4 contains a separated torus so @xmath379 . therefore , @xmath380 finally , we remark on uniqueness and related topics . in what is our artin spun case , fintushel and stern have conjectured that their knot surgery construction yields nondiffeomorphic manifolds for `` essentially different '' knots . ( here , `` essentially different '' means that two knots are not isotopic , they are not mirrors , nor they isotopic under mirroring of connect - summands . ) the alexander polynomial does not completely distinguish knots however , so the seiberg - witten invariants in their current form do not shed any light on their conjecture . similarly , it seems doubtful that the manifolds @xmath381 and @xmath382 are diffeomorphic , but with the seiberg - witten invariants being equal , we have no obvious way in which to distinguish them . in particular , it seems possible that @xmath382 is diffeomorphic to @xmath383 where @xmath117 is the artin spin of the left handed trefoil . also , results of c. taubes in @xcite show that a manifold @xmath258 with @xmath384 admits a symplectic form , the leading term in @xmath385 will have coefficient equal to one . the converse to this statement is known to be false by work of fintushel and stern in @xcite . in the classical ( or artin spun ) case , it is possible to construct a symplectic form on @xmath63 when the classical knot @xmath4 from which @xmath33 is constructed is a fibered knot . while it may be possible to rephrase this construction in terms of twins , it is unclear what topological conditions are required on @xmath33 to achieve the same result . ( a sufficient condition is that @xmath64 fibers over @xmath239 or @xmath6 . )

this paper presents a new approach for analysing structural properties of time series from complex systems . starting from the concept of recurrences in phase space , the recurrence matrix of a time series is interpreted as the adjacency matrix of an associated complex network which links different points in time if the evolution of the considered states is very similar . a critical comparison of these recurrence networks with similar existing techniques is presented , revealing strong conceptual benefits of the new approach which can be considered as a unifying framework for transforming time series into complex networks that also includes other methods as special cases . it is demonstrated that there are fundamental relationships between the topological properties of recurrence networks and the statistical properties of the phase space density of the underlying dynamical system . hence , the network description yields new quantitative characteristics of the dynamical complexity of a time series , which substantially complement existing measures of recurrence quantification analysis .

0908.3447

since the early stages of quantitative nonlinear sciences , numerous conceptual approaches have been introduced for studying the characteristic features of dynamical systems based on observational time series @xcite . popular methods that are increasingly used in a variety of applications ( see , for example , @xcite ) include ( among others ) lyapunov exponents , fractal dimensions , symbolic discretisation , and measures of complexity such as entropies and quantities derived from them . all these techniques have in common that they quantify certain dynamically invariant phase space properties of the considered system based on temporally discretised realisations of individual trajectories . as a particular concept the basic ideas of which originated in the pioneering work of poincar in the late 19th century @xcite , the quantification of recurrence properties in phase space has recently attracted considerable interest @xcite . one particular reason for this is that these recurrences can be easily visualised ( and subsequently quantified in a natural way ) by means of so - called recurrence plots obtained from a single trajectory of the dynamical system under study @xcite . when observing this trajectory as a scalar time series @xmath0 ( @xmath1 ) , one may use a suitable @xmath2-dimensional time delay embedding of @xmath0 with delay @xmath3 @xcite , @xmath4 , for obtaining a recurrence plot as a graphical representation of the binary recurrence matrix @xmath5 where @xmath6 is the heaviside function , @xmath7 denotes a suitable norm in the considered phase space , and @xmath8 is a threshold distance that should be reasonably smaller than the attractor diameter @xcite . to simplify our notation , we have used the abbreviation @xmath9 ( with @xmath10 being the point in time associated with the @xmath11-th observation recorded in the time series ) wherever appropriate . experimental time series often yield a recurrence plot displaying complex structures , in particular , with different properties of the non - interrupted diagonal and vertical structures ( `` lines '' ) . a variety of statistical characteristics of the length distributions of these lines ( such as maximum , mean , or shannon entropy ) can be used for defining additional quantitative measures that characterise different aspects of dynamic complexity of the studied system in more detail . this conceptual framework is known as recurrence quantification analysis ( rqa ) @xcite and is nowadays frequently applied to a variety of real - world applications of time series analysis in various fields of research @xcite . however , most of these rqa measures are sensitive to the choice of embedding parameters , which are found to sometimes induce spurious correlations in a recurrence plot @xcite . recent studies have revealed that the fundamental invariant properties of a dynamical system ( i.e. , its correlation dimension @xmath12 and correlation entropy @xmath13 ) are conserved in the recurrence matrix @xcite . furthermore , it is found that the estimation of these invariants is independent of the particular embedding parameters . the recurrence plots preserve all the topologically relevant phase space information of the system , such that one can completely reconstruct a time series from its recurrence matrix ( modulo some rescaling of its probability distribution function ) @xcite . a further appealing paradigm for analysing structural features of complex systems is based on their representation as complex networks of passive or active ( i.e. , mutually interacting ) subsystems . for this purpose , classical graph theory has been systematically extended by a large variety of different statistical descriptors of the topological features of such networks on local , intermediate , and global scales @xcite . these measures have been successfully applied for studying real - world networks in various scientific disciplines , including the structural properties of infrastructures @xcite , biological @xcite , ecological @xcite , and climate networks @xcite , to give some prominent examples . the corresponding results have triggered substantial progress in our understanding of the interplay between structure and dynamics of such complex networks @xcite . the great success of network theory in various fields of research has recently motivated first attempts to generalise this concept for a direct application to time series @xcite . however , a substantial number of the recently suggested techniques have certain conceptual limitations , which make them suitable only for dealing with distinct types of problems . as an alternative that may provide a unifying conceptual and practical framework for nonlinear time series analysis using complex networks , we reconsider the concept of recurrences in phase space for defining complex network structures directly based on time series . for this purpose , it is straightforward to interpret the recurrence matrix @xmath14 as the adjacency matrix @xmath15 of an unweighted and undirected complex network , which we suggest to call the _ recurrence network _ associated with a given time series . to be more specific , the associated adjacency matrix is given by @xmath16 where @xmath17 is the kronecker delta introduced here in order to avoid artificial self - loops . a corresponding conceptual idea has recently been independently suggested by different authors @xcite , but not yet systematically studied . in this work , we however aim to give a rigorous derivation and detailed interpretation for a variety of quantitative characteristics of recurrence networks . it shall be noted that a generalisation to weighted networks ( as partially studied in @xcite ) is straightforward if the recurrence matrix is replaced by the associated distance matrix between pairs of states . in any case , recurrence networks referring to the mutual phase space distances of observational points on a single trajectory are spatial networks , i.e. fully embedded into an @xmath2-dimensional space , which has important implications for their specific topological features . we will raise this point in detail within this paper . the consideration of recurrence plots as graphical representations of complex networks allows a reinterpretation of many network - theoretic measures in terms of characteristic phase space properties of a dynamical system . according to the ergodicity hypothesis , we suppose that one may gain full information about these properties by either ensembles of trajectories , or sufficiently long observations of a single trajectory . following this line of ideas , we may approximate the ( usually unknown ) invariant density @xmath18 ( which is related with the associated invariant measure @xmath19 by @xmath20 ) of the studied system by some empirical estimate @xmath21 obtained from a time series , where @xmath8 defines the level of coarse - graining of phase space involved in this procedure . transforming the time series into a recurrence network then allows to quantitatively characterise the higher - order statistical properties of the invariant density @xmath18 by means of complementary methods , i.e. , network - theoretic measures . according to the above argumentation , quantitative descriptors of the topological features of recurrence networks can be considered as novel measures within the framework of rqa . our technique therefore exhibits additional deep insights into the phase space properties of dynamical systems directly related to their complex dynamics . additionally , we will emphasise that we may also take seriously the duality of adjacency matrices of complex networks on the one hand , and recurrence matrices of dynamical systems on the other hand , which would also allow transferring concepts from dynamical systems theory ( given that the corresponding recurrence plot based estimates are invariant under temporal reordering ) to complex network theory . in this work , however , we will concentrate on a detailed discussion of how phase space properties can be further quantified in terms of network theory . the remainder of this paper is organised as follows : section [ sec : review ] presents a critical review of existing approaches for extracting complex networks from time series , including a comprehensive discussion of their potentials and potential problems ( with a special emphasis on how to interpret the resulting networks topological properties ) . the concept of recurrence networks as a natural alternative is further discussed in section [ sec : recnet ] . in particular , it is demonstrated that many network - theoretic measures yield sophisticated quantitative characteristics corresponding to certain phase space properties of a dynamical system that have not yet been explicitly studied in terms of other dynamical invariants or measures of complexity based on rqa . in order to support our theoretical considerations , section [ sec : examples ] provides some examples of how different network - theoretic measures reveal certain phase space properties of various dynamical systems . finally , we summarise our main results and outline some future directions of further methodological developments based on our proposed technique . in this section , a review and classification of existing approaches for studying the properties of time series by means of complex networks approaches is presented ( see tab . [ tab : methods ] ) . in particular , strengths and possible limitations of the existing approaches will be briefly discussed . .[vertexedgetable]summary of the definitions of vertices and the criteria for the existence of edges in existing complex network approaches to time series analysis ( given in chronological order ) . [ cols="<,^,>",options="header " , ] as a first measure that allows to quantify the importance of a vertex in a complex network , the degree centrality of a vertex @xmath22 , @xmath23 , is defined as the number of neighbours , i.e. the number of vertices @xmath24 that are directly connected with @xmath22 : @xmath25 note that in general , the sum is taken over all @xmath24 . however , according to our definition ( [ def : adjacency ] ) , we skip the corresponding condition in the following . normalising this measure by the maximum number of possible connections , @xmath26 , one gets the local connectivity @xmath27 which , from the recurrence plot point of view , corresponds to the local recurrence rate @xmath28 of the state @xmath22 . thus , the degree centrality and local connectivity yield an estimator for the local phase space density , since for a vertex @xmath22 located at position @xmath29 in phase space , @xmath30 ( when using the maximum norm ) and , hence , @xmath31 in complex network studies , one is often interested in the frequency distribution of degree centralities , @xmath32 , in particular , the presence of an algebraic scaling behaviour , which is characteristic for scale - free networks @xcite . however , although several authors have recently focussed their attention on this characteristic obtained from different types of complex networks derived from time series @xcite , we would like to underline that for a complete characterisation of the phase space properties of a dynamical system , one should prefer studying not only degree centralities , but also other higher - order statistical measures . in some situations , it is useful not to consider the full distribution of degree centralities in a network , but to focus on the mean degree of all vertices @xmath33 as a simple characteristic quantity of this distribution , where @xmath34 is the total number of edges in the recurrence network . the mean degree centrality @xmath35 is directly proportional to the edge density @xmath36 of the network or , alternatively , its recurrence plot equivalent , the global recurrence rate @xmath37 , @xmath38 note that the recurrence rate coincides with the definition of the correlation integral @xmath39 , which is commonly used to estimate the correlation dimension @xmath12 , for example , using the grassberger - procaccia algorithm @xcite . the connection between the edge density and the correlation dimension can be understood by the fact that the local recurrence rate @xmath28 of a vertex @xmath22 corresponds to the measure of a @xmath2-dimensional ball @xmath40 of radius @xmath8 centered at the point @xmath29 in the @xmath2-dimensional phase space in the limit that time goes to infinity ( @xmath41 ) . when considering the euclidean norm as a distance measure in phase space , these balls are defined as hyperspheres , for the maximum norm as hypercubes etc . then , the pointwise ( information ) dimension of the probability measure @xmath19 at @xmath42 is defined as @xmath43 @xcite . due to the heterogeneity of the phase space visited by the trajectory ( i.e. , the non - uniform phase space density that results in different degree centralities @xmath23 in different parts of this space ) , the proper estimation of @xmath44 is a nontrivial task and often requires expensive computational power and a high data quality and quantity . thus , one may expect a better statistics for @xmath12 , since it more heavily weights regions of the phase space which have a higher probability measure @xmath19 . though the correlation integral has been well established in the literature for estimating the correlation dimension , we point out the improvement in estimating @xmath12 based on the diagonal lines in @xmath45 , which yields an algorithm that is independent of the embedding parameters @xcite . consequently , the recurrence network representation @xmath46 of a time series fully conserves the geometric properties of the phase space of the underlying dynamical system . the clustering coefficient of a vertex @xmath22 , @xmath47 , characterises the density of connections in the direct neighbourhood of this vertex in terms of the density of connections between all vertices that are incident with @xmath22 . in many networks , such loop structures formed by three vertices occur more often than one would expect for a completely random network . hence , high clustering coefficients reveal a specific type of structure in a network , which is related to the cliquishness of a vertex @xcite . in this work , we consider the definition of clustering coefficient proposed by watts and strogatz @xcite , @xmath48 where @xmath49 is the total number of closed triangles including vertex @xmath22 , which is bound by the maximum possible value of @xmath50 . for vertices of degree @xmath51 or @xmath52 ( isolated or tree - like points , respectively ) , the clustering coefficient is defined as @xmath53 , as such vertices can not participate in triangles by definition . equation ( [ cc_local ] ) can be rewritten in terms of conditional probabilities as @xmath54 using bayes theorem , with @xmath55 and a similar expression for @xmath56 . as for a recurrence network , the value of @xmath46 depends only on the phase space distance and the choice of @xmath8 , the latter relationship may be used to derive analytical results at least for one - dimensional systems based on their invariant density . corresponding details can be found in [ app_1dmap ] , including the corresponding treatment of the bernoulli and logistic maps as specific examples . as for the degree centrality , one may consider the average value of the clustering coefficient taken over all vertices of a network , the so - called global clustering coefficient @xmath57 as a global characteristic parameter of the topology of a network . one expects that the value of @xmath58 is for a given dynamical system with a phase space density @xmath18 in the asymptotic limit @xmath59 exclusively determined by the choice of @xmath8 , which determines the scale of resolution . a more detailed discussion of the corresponding effects and their implications for certain model systems will be given in section [ sec : examples ] . the mean nearest neighbour degree @xmath60 of vertex @xmath22 gives the average degree in the neighbourhood of @xmath22 , @xmath61 the degree centrality @xmath23 is a measure of the density of states in the immediate neighbourhood of state @xmath22 , whereas @xmath60 can be interpreted to indicate the mean density of states in the next neighbourhood ( next topological shell of neighbours ) of state @xmath22 . hence , both measures taken together contain information about the local density anomaly in the vicinity of @xmath22 , which we propose to measure by the local degree anomaly @xmath62 vertices with a positive degree anomaly ( @xmath63 ) hence indicate local maxima of phase space density , while such which @xmath64 correspond to local density minima . hence , the local degree anomaly may be considered as a proxy for the local heterogeneity of the phase space density . in a similar way , the average absolute value of the local degree anomaly , @xmath65 , serves as a measure for the overall spatial heterogeneity of the phase space density profile . a network is called assortative if vertices tend to connect preferentially to vertices of a similar degree @xmath66 . on the other hand , it is called disassortative if vertices of high degree prefer to connect to vertices of low degree , and vice versa . hence , assortativity can be quantified by the pearson correlation coefficient of the vertex degrees on both ends of all edges @xcite , @xmath67 ^ 2}{\frac{1}{l } \sum_{j > i } \frac{1}{2 } ( k_i^2+k_j^2)a_{i , j } - \left [ \frac{1}{l } \sum_{j > i } \frac{1}{2 } ( k_i+k_j)a_{i , j } \right]^2}.\ ] ] if the density of states in phase space hardly varies within an @xmath8-ball , the degrees on either ends of an edge will tend to be similar and hence the assortativity coefficient @xmath68 will be positive . this means , that the more continuous and slowly changing the density of states is , the closer @xmath68 will be to its maximum value one . within the framework of recurrence networks , @xmath68 can hence be interpreted as a measure of the continuity of the density of states or put differently , of the fragmentation of the attractor . note that this aspect has not yet been specifically addressed by other nonlinear measures , in particular , within the rqa framework . the overlap of the neighbourhood spaces of two vertices @xmath69 is measured by the matching index @xmath70 where @xmath71 if there are no common neighbours , and @xmath72 if the neighbourhoods coincide @xcite . using the notion of @xmath8-balls around points in phase space , one may alternatively write @xmath73 due to the spatial constraints of the recurrence network , the neighbourhood spaces of @xmath69 can only overlap if @xmath74 i.e. , @xmath71 for all pairs of vertices @xmath75 with @xmath76 . moreover , the matching index @xmath77 decreases on average with an increasing spatial distance @xmath78 between the two considered states . note that since @xmath79 already for @xmath80 , there may be unconnected points with a matching index @xmath81 . the matching index of pairs of vertices in a recurrence network is closely related to the concept of _ twins _ @xcite , which has recently been successfully applied for constructing surrogate data ( twin surrogates ) in the context of statistical hypothesis testing for the presence of complex synchronisation @xcite . twins are defined as two states of a complex system that share the same neighbourhood in phase space , i.e. , the two vertices of the recurrence network representing these states have a matching index @xmath72 . hence , the matching index can be used for identifying candidates for twins . note that pairs of vertices @xmath11 and @xmath82 in a recurrence network with @xmath83 can still be considered as _ potential _ twins , since @xmath72 may in some cases be approached by only slight changes of the threshold @xmath8 . consequently , we suggest interpreting the matching index as a measure of the _ twinness _ of @xmath11 and @xmath82 . furthermore , it should be noted that adjacent pairs of edges @xmath75 ( @xmath84 ) with a low matching index @xmath85 connect two distinct regions of the attractor and may therefore be indicative of geometrical bottlenecks in the dynamics ( cf . our discussion of the betweenness centrality in section [ sec : bc ] ) . as we consider recurrence networks as undirected and unweighted , we assume all edges to be of unit length in terms of graph ( geodesic ) distance . consequently , the distance between any two vertices of the network is defined as the length of the shortest path between them . note that time information is lost after transforming the trajectory into a network presentation . therefore , the terminology of the shortest path length @xmath86 in the recurrence network reflects the minimum number of edges that have to be passed on a graph between a vertex @xmath11 to a vertex @xmath82 . in the same spirit , @xmath86 is related to the distance of states @xmath11 and @xmath82 in phase space . ] for a general non - periodic trajectory @xmath87 embedded in a two - dimensional phase space @xmath88 . [ nonperiodic_list ] ] in order to better understand the meaning of shortest path lengths , let us study their calculation for two toy model series : first , we consider a periodic trajectory @xmath89 , @xmath90 , with @xmath91 , i.e. there are @xmath92 points in the phase space ( fig . [ periodic_list](a ) ) . the corresponding recurrence plot is shown for @xmath93 ( fig . [ periodic_list](b ) ) . as it has already been mentioned above , the recurrence matrix @xmath94 and the adjacency matrix @xmath46 of the associated recurrence network are basically equivalent . adopting a common visualisation of connectivity patterns from the literature on complex networks , we illustrate the recurrences of the considered model time series by placing the individual observations ( vertices ) on a circle with equal common distances ( fig . [ periodic_list](c ) ) . in this representation , the shortest path length ( in the network sense ) between two vertices @xmath11 and @xmath82 corresponds to the smallest number of `` jumps '' in phase space via pairs of neighbours ( i.e. recurrences ) in phase space . obviously , the number of such jumps is determined by the prescribed value of @xmath8 and the spatial distance between @xmath11 and @xmath82 . for instance , the shortest path from vertex @xmath95 to @xmath96 is @xmath97 as indicated by the matrix of mutual shortest path lengths ( fig . [ periodic_list](d ) ) . note that this list is symmetric by definition , i.e. @xmath98 ) . the same heuristic analysis can also be performed for a general nonperiodic trajectory in phase space as shown in fig . [ nonperiodic_list](a - d ) . we wish to underline that the terminology of shortest path lengths in networks does not have a direct relevance to the dynamical evolution of the observed system . in contrast , @xmath86 measures distances in phase space ( among a discrete set of points on the attractor ) in units of the neighbourhood size @xmath8 . for example , in the periodic case displayed in fig . [ periodic_list ] , it takes @xmath99 iterations ( time points ) from vertex @xmath52 to @xmath100 in the time domain , while the shortest path to cover the phase space distance has only a length of @xmath101 . hence , shortest paths do not allow to infer the temporal evolution of the system . even more , for the path concept in a recurrence network , no information about the temporal order of the individual observations is considered ( for example , the shortest path between vertices @xmath52 and @xmath102 in fig . [ periodic_list ] is given by the sequence ( 1,8,5,2 ) , which is not ordered in time ) . one should note that if the phase space is strongly fragmented ( for instance , in the period-3 window of the logistic map , which has been discussed elsewhere @xcite , the phase space consists of three discrete points ) , the resulting recurrence networks may be composed of different disconnected clusters . furthermore , there might be more than one shortest path connecting two nodes . for example , in the aperiodic example in fig . [ nonperiodic_list ] , the shortest path from node @xmath52 to node @xmath103 , ( @xmath104 as shown in fig . [ nonperiodic_list](d ) ) , can be obtained by three different choices , that are , ( 1,2,6,7 ) , ( 1,5,3,7 ) , and ( 1,5,6,7 ) . the average path length @xmath105 is defined as the mean value of the shortest path lengths @xmath86 taken over all pairs of vertices @xmath75 , @xmath106 here , for disconnected pairs of vertices , the shortest path length is set to zero by definition . note that in most practical applications , this has no major impact on the corresponding statistics . the average phase space separation of states @xmath107 serves as an @xmath8-lower bound to @xmath105 , since @xmath108 due to the triangular inequality , and hence @xmath109 interpreted geometrically , this inequality holds because @xmath105 approximates the average distance of states along geodesics on the recurrence network graph ( which can be considered as the geometric backbone of the attractor ) in multiples of @xmath8 , while @xmath107 gives the mean distance of states in @xmath110 as measured by the norm @xmath111 . by a similar argument as used in eq . ( [ path_distance_inequality ] ) for the average path length , the diameter @xmath112 of the recurrence network ( i.e. the maximum path length ) serves as an @xmath8-upper bound to the estimated diameter @xmath113 of the attractor in phase space : @xmath114 the inverse average shortest path length of vertex @xmath22 to all others in the recurrence network is measured by the closeness centrality @xcite @xmath115 if @xmath11 and @xmath82 are not connected , i.e. , @xmath79 , the maximum shortest path length in the graph , @xmath26 , is used in the sum by definition . in a recurrence network , @xmath116 can be geometrically interpreted as measuring the closeness of @xmath22 to all other states with respect to the average length ( in units of @xmath8 ) of geodesic connections on the recurrence network graph . in other words , @xmath116 is large if most of the other vertices are reachable in a small number of @xmath8-jumps from state to state . from eqs . ( [ path_distance_inequality ] ) and ( [ closeness_equation ] ) , we can see that the inverse closeness @xmath117 is bounded from below by the average phase space distance of vertex ( state ) @xmath22 to all other vertices ( states ) @xmath118 in units of @xmath8 ( geometrical closeness ) , as measured by the norm @xmath111 , @xmath119 put differently , geometrical closeness provides an upper bound for topological closeness , @xmath120 the betweenness centrality @xmath121 has been originally introduced for characterising the importance of individual vertices for the transport of information or matter in general complex networks @xcite . unlike the degree centrality @xmath23 , it is defined locally but depends on global adjacency information . let us assume that information travels through the network on shortest paths . there are @xmath122 shortest paths connecting two nodes @xmath11 and @xmath82 . we then regard a node @xmath22 to be an important mediator for the information transport in the network , if it is traversed by a large number of all existing shortest paths . betweenness centrality is given by @xmath123 where @xmath124 gives the number of shortest paths from @xmath11 to @xmath82 , that include @xmath22 . here the contribution of shortest paths is weighted by their respective multiplicity @xmath122 , the physical rational for this normalisation being that the total volume of information flow between two vertices , when summed over all shortest paths connecting them , should be the same for all pairs in the network . hence , in addition to degree and closeness centralities , betweenness centrality yields another possibility to identify especially relevant vertices . for a recurrence network , the notion of information transfer is not useful anymore . however , one may still argue in a geometric way that high betweenness states are typical for regions of sparse phase space density that separate different high - density clusters ( refering to the information flow analogy mentioned above , one may consider the corresponding vertices as _ geometric bottlenecks _ ) . thus , the occurrence of high betweenness values can be a sign of highly fractionated attractors ( on the scale resolved by the considered threshold @xmath8 ) . a more detailed discussion of the corresponding implications for some simple model systems will be given in section [ sec : examples ] . while betweenness centrality refers to vertex properties of a network , one may define an equivalent measure also based on the number of shortest paths on the network that include a specific edge @xmath75 . we refer to the corresponding property as the edge betweenness @xmath125 . note that though there is a conceptual difference between vertex - related and edge - related betweenness , both quantities are indicators for regions of low phase space density that separate regions with higher density ( or , to say it differently , of regions of high attractor fractionation ) and thus have practically the same dynamical meaning . in the following , we will show the potentials of the network - theoretic measures discussed in the previous section for recurrence networks obtained from three paradigmatic chaotic model systems . basic results for one - dimensional maps have already been described for the logistic map ( see @xcite ) based on numerical calculations and are supplemented by some further computations in the appendix . at this point , we prefer to discuss in some more detail the properties of systems that are defined in somewhat higher dimensions . in particular , we consider the hnon map @xmath126 as an example for a chaotic two - dimensional map , and the rssler system @xmath127 as well as the lorenz system @xmath128 as two examples for three - dimensional chaotic oscillators . in all following considerations , no additional embedding will be used . note , however , that for the continuous systems , temporal correlations between subsequent observations have been excluded by removing all sojourn points @xcite . [ p2p_henon ] and [ p2p_rossler ] show examples of typical trajectories of these three model systems . in addition , the shortest paths between the first and last point of the individual realisation are indicated , underlining the deep conceptual differences between the concepts of trajectory ( in phase space ) and path ( in a recurrence network , see section [ sec : recnet ] ) . ) , indicated by circles . the initial condition is marked by a square the size of which corresponds to the neighbourhood threshold @xmath129 ( maximum norm ) considered in the derivation of the corresponding recurrence network . the shortest path ( with @xmath130 ) between the first and the last point is indicated by a continuous red line . ] let us first consider the dependence of the global network measures @xmath105 , @xmath58 and @xmath68 on the choice of the threshold @xmath8 for our three model systems . the variations of @xmath105 with the threshold @xmath8 are shown in fig . [ avg_path_all ] and verify the existence of an inverse relationship of a corresponding lower bound postulated in eq . ( [ avgl_eq ] ) . + + for the global clustering coefficient , the dependence on @xmath8 is more complicated and depends on the specific properties of the considered system ( fig . [ avg_path_all ] ) . in particular , while for too small @xmath8 , problems may occur , since the recurrence network may decompose into different disconnected clusters for a length @xmath131 of the considered time series , for intermediate threshold values , an approximately linear increase of @xmath58 with @xmath8 seems to be a common feature of all three examples . following the discussion of the behaviour of one - dimensional maps in [ app_1dmap ] , we may argue that this increase is most likely related to the effect of the attractor boundaries . finally , concerning the assortativity coefficient @xmath68 , we observe that for small @xmath8 , the recurrence networks are highly assortative ( e.g. @xmath68 is close to 1 ) . this behaviour can be related to the fact that in case of small neighbourhoods , these phase space regions are usually characterised by only weak variations of the phase space density , so that neighbouring vertices have a tendency to obey a similar degree . as @xmath8 becomes larger , larger regions of the phase space are covered , where the density may vary much stronger , which implies that the degrees of neighbouring vertices become less similar . note , however , that since in this case , the mutual overlap of the different neighbourhoods becomes successively larger , there is still a significantly positive correlation between the degrees of neighbouring vertices . one may further observe that the decrease of @xmath68 with @xmath8 may be interrupted by intermediate increases , which are probably related to some preferred spatial scale of the separation of certain dynamically invariant objects such as unstable periodic orbits ( upos ) . we will come back to this point in sec . [ cluster_spatial ] . in the following , we will study the interrelationships between local network properties and structural features of the phase space for the three considered chaotic model systems . when considering the degree centrality @xmath23 or , equivalently , the local density @xmath132 for all vertices of the network , a broad range of variability is found ( fig . [ degree_all ] ) . in particular , the behaviour follows the expectation that regions with a high phase space density ( for example , the merger of the two scrolls of the lorenz oscillator ) also reveal a high density of vertices and , hence , high degree centralities . note that the calculation of a recurrence plot depends on the parameter @xmath8 , which should be tailored to the considered system under study and the specific questions one wishes to address . several `` rules of thumb '' for the choice of the threshold @xmath8 have been advocated in the literature @xcite . it has been suggested that the choice of @xmath8 to achieve a fixed recurrence rate @xmath37 is helpful for the estimation of dynamical invariants in many systems @xcite . therefore , this procedure will be adopted here to obtain an overall visualisation of the degree centrality @xmath23 in phase space , with @xmath133 ( which lies within the typical scaling region of the correlation integral ) . however , as we will see later , for the local clustering coefficients ( sec . [ cluster_spatial ] ) disclosing local fine structures of the phase space density , it is necessary to choose smaller @xmath8 . figure [ closeness_all ] reflects the spatial distribution of the closeness centrality @xmath116 . in good agreement with our previous theoretical considerations on the geometric meaning of this measure ( sec . [ closeness ] ) , we find high values of @xmath116 near the centre of gravity of the attractor in phase space , and low values at phase space regions that have large distances from this centre . ( eq . ( [ closeness_equation ] ) ) in phase space ( a ) hnon map , ( b ) rssler , and ( c ) lorenz systems ( @xmath131 and @xmath8 as in fig . [ degree_all ] ) . ] concerning the local clustering coefficient , one may suppose that in the case of high - density regions in phase space , there are many vertices located in the vicinity of a specific vertex , in particular , in a distance that does not exceed @xmath134 . by definition , these vertices must then also be adjacent to each other , which gives considerable contributions to the clustering coefficient . in contrast , for low - density regions , one may argue that even if there are more than one vertices in some @xmath8-neighbourhood of a vertex ( i.e. , @xmath135 ) , it is less likely that these are also separated by a distance that is smaller than @xmath8 . following these considerations , one might expect some relationship between the degree centrality and the local clustering coefficient . however , as figs . [ kvcv_all ] , [ cc_henon ] , and [ cc_ros ] demonstrate , @xmath47 does clearly reveal more and different structural properties than the degree centrality alone ( note that here , a smaller value of @xmath8 has been chosen to disclose the local fine structures of the phase space density ) . in particular , a visual comparison with fig . [ degree_all ] reveals that the clustering coefficient characterises some specific higher - order characteristics of the phase space density . and the local clustering coefficient @xmath47 for typical realisations of ( a ) hnon map , ( b ) rssler , and ( c ) lorenz system . the inserted values @xmath136 give the corresponding rank - order correlation coefficient ( spearman s rho ) . ] beside the effect of the local phase space density , we argue that the local clustering coefficient does also depend on the spatial filling ( i.e. , the homogeneity of the phase space density ) in the neighbourhood of the considered point . in particular , in the case of a two - dimensional system , an alignment of vertices along a one - dimensional subspace will produce a clearly lower clustering coefficient than a homogeneous filling of the neighbourhood . this behaviour is underlined in fig . [ cc_henon ] for the hnon map , where maximum values of @xmath137 can be particularly found at the two tips of the attractor . hence , the local clustering coefficient can be considered as an entropy - like characteristic in that it quantifies the homogeneity of the phase space density in the neighbourhood of a vertex . from the theory of spatial random graphs @xcite , which may be assumed to yield the lowest possible clustering coefficients among all spatial networks , it is known that for a given dimension @xmath2 of the considered system , in the asymptotic limit @xmath59 and @xmath138 , the possible values of @xmath47 are bound between the corresponding theoretical value and @xmath52 . note that this lower bound systematically decreases with increasing @xmath2 , which appears to be reasonable if one interpretes @xmath47 as an entropy - like quantity . to be more specific , according to dall and christensen @xcite , this decay follows an exponential function for sufficiently large embedding dimensions . the presence of distinct structures in the spatial profile of the local clustering coefficient is related to the emergence of specific dynamically invariant objects in the considered model systems . in the case of hnon map ( fig . [ cc_henon ] ) , there is a clear tendency that points that are close to the stable manifold associated with the system have remarkably higher values of @xmath47 . note , however , that because of finite size effects , this coincidence can not be found for all corresponding regions of the phase space . for the two continuous systems ( fig . [ cc_ros ] ) , points close to the trapping regimes of upos have higher clustering coefficients . it is , in some sense , trivial to understand the role of upos in forming such regimes of higher clustering . whenever the trajectory of the corresponding systems visits the neighbourhood of an upo , it is captured in this neighbourhood for a certain finite time , during which the probability of recurrences is increased . furthermore , once the trajectory is trapped , the local divergence rate becomes smaller . this smaller local divergence rate is captured by the clustering coefficient ( in terms of higher - order correlations between neighbours of a vertex ) . as for the finite-@xmath8 effect in the hnon map , the regions with increased clustering coefficients in most cases only coincide with upos of lower periods . therefore , in fig . [ cc_ros ] , only a few upos of low order are shown for comparison . note that if two upos are separated by a distance smaller than @xmath8 in phase space , the clustering coefficient is not able to distinguish between these two structures and , hence , shows a broad band with increased values . following this argumentation , @xmath47 is a useful measure for detecting phase space regions with a high density of low - order upos , which is in good agreement with corresponding considerations in @xcite . our interpretation of the betweenness centrality in sec . [ sec : bc ] implies that @xmath121 is a rather sensitive measure of the local fragmentation of the attractor and thus may give complementary information especially on very small scales . unfortunately , numerical limitations in the calculation of this measure did not allow us to explore the limit of small neighbourhoods ( @xmath138 ) . however , from our computations with somewhat larger thresholds ( see fig . [ bc_ros ] ) , we can already derive some general statements about the behaviour of betweenness centrality for the considered model systems . first , note that regions close to the outer boundaries of the attractor ( in contrast to those in the vicinity of the inner boundaries , e.g. , of the rssler oscillator ) are not important for many shortest path connections on the recurrence network . hence , vertices settled in the corresponding parts of the phase space are characterised by low betweenness values . second , if there are pronounced regions with rather few isolated points in between high - density regions ( for example , between two upos in the rssler or lorenz systems ) , there is an increasing number of shortest paths crossing these vertices , which leads to higher values of @xmath121 . in turn , vertices in the vicinity of upos ( i.e. , high - density regions ) show lower betweenness values . therefore , betweenness centrality provides a complementary view on the attractor geometry in comparison to the local clustering coefficient @xmath47 ( fig . [ cc_ros ] ) . similarly to the local vertex properties , one may also study the characteristics of different edges in the recurrence networks . since the resulting structures are more pronounced than for the three model systems considered so far , fig . [ mi_eb ] shows the matching index and edge betweenness for one realisation of the logistic map @xmath139 in the intermittent chaotic regime ( see @xcite ) . the presence of intermittent dynamics can be clearly seen from the recurrence plots in terms of extended square recurrence patterns , which hence lead to mutually connected vertices of the associated recurrence networks that correspond to subsequent points in time . figures [ mi_eb ] and [ mi_dist ] show the complex dependence between phase space distance @xmath78 , matching index @xmath77 , and edge betweenness @xmath125 . for the matching index , the results are consistent with our theoretical considerations presented in sec . [ sec : mi ] . in particular , for @xmath140 , we have @xmath141 , while for @xmath142 , @xmath143 . concerning the temporal evolution during the laminar ( intermittent ) phase , one may recognise that at the beginning , there is hardly any change in the state of the system , hence , @xmath78 is very small for subsequent points in time ( vertices of the recurrence network ) , which relates to large values of the matching index near @xmath52 . as the laminar phases are close to their termination , chaotic variations emerge and rise in amplitude , which leads to a subsequent increase of @xmath78 and , hence , decrease of @xmath77 . concerning the edge betweenness @xmath125 ( the spatial pattern of which is very similar to that of the vertex - based betweenness centrality @xmath121 due to the spatial proximity of edge and corresponding vertices ) , the overall behaviour is opposite to that of @xmath77 . during laminar phases , we find that since all states are very close to each other , possible shortest connections may alternatively pass through a variety of different edges , leading to low values of the edge betweenness . close to the termination , there is in turn an increase of this measure . however , the most interesting feature of the edge betweenness is presented by isolated edges with very high values of @xmath125 , which correspond to rarely visited phase space regions between intervals of higher phase space density . more specifically , the average edge betweenness of vertices in such low - density regions may exceed that of high - density regions by orders of magnitude ( fig . [ mi_eb ] ) . this paper has reconsidered the analysis of time series from complex systems by means of complex network theory . we have argued that most existing approaches for such an analysis suffer from certain methodological limitations or a lack of generality in their applicability . as an appealing solution , we have suggested recurrence networks as a unifying framework for studying time series as complex networks , which is based on the idea of recurrence plots . as we have argued , this specific approach is applicable to univariate as well as multivariate time series without or with embedding . in addition , recurrence networks can be applied for studying time series with non - equidistant time - scales and allow the construction of simple significance tests with respect to the associated network - theoretic measures @xcite . as a main achievement , we have provided a thorough reinterpretation of a variety of statistical measures from network theory computed for recurrence networks in terms of phase space properties of dynamical systems . since all time ordering information is lost in this approach , all complex network characteristics are dynamically invariant , i.e. , they are only sensitive to certain properties of the invariant density of the considered dynamical system . from this invariance , it follows that specific measures such as the local clustering coefficient may be used for detecting dynamically invariant objects like unstable periodic orbits or chaotic saddles . on the other hand , this feature also implies that the proposed method can not be used to distinguish between deterministic ( chaotic ) and stochastic systems , which is exemplified by our comparison between bernoulli map and uniform noise in [ app_1dmap ] . as a possibility to overcome this potential point of criticism to our method , we emphasise that an additional embedding should change the properties of deterministic systems in a different way than for a stochastic system , hence , studying complex network properties in dependence on the embedding dimension might help solving this interpretation problem . we will further elaborate this idea in future research . using wide - spread statistical characteristics of complex networks such as the `` trinity '' of centrality measures ( degree , closeness , betweenness ) and the clustering coefficient , we were able to provide a detailed interpretation of the corresponding results for recurrence networks in terms of higher - order phase space properties . in particular , degree centrality relates to the local density , closeness centrality to the average geometrical proximity of an observation to all other observations , and betweenness centrality to the local fragmentation of points in phase space . for the clustering coefficient , the consideration of different model systems has revealed that apart from possible density effects , the clustering properties are related to the spatial filling of the phase space , which become important close to the attractor boundaries and certain dynamically invariant objects such as invariant manifolds or unstable periodic orbits . additionally , we have presented a rigorous analytical treatment of the local and global clustering coefficients of the recurrence networks of one - dimensional chaotic maps , which perfectly matches our numerical results ( see [ app_1dmap ] ) . a possible further relationship between the local clustering coefficient of a recurrence network and the local lyapunov exponent of the underlying dynamical system remains a topic for future work . with respect to existing recurrence plot based methods of time series analysis , e.g. , recurrence quantification analysis ( rqa ) , we would like to emphasise that our approach yields a complementary view on the phase space properties of the underlying dynamical system . in particular , one should note that nearly all of the considered network - theoretical measures have no direct equivalents in traditional rqa and vice versa . hence , one may think of situations when either one of the two frameworks ( i.e. , rqa or recurrence networks ) may provide better results than the other . in turn , there have been some recent approaches on applying methods of time series analysis to general complex networks ( e.g. @xcite ) . we would like to underline that due to the duality between the recurrence matrix of a time series and the adjacency matrix of the associated recurrence network , rqa might be another promising candidate for this purpose ( as long as one restricts oneself to measures that are invariant under re - ordering of time ) , which might yield interesting complementary insights on complex networks in a variety of different situations . a more detailed investigation of a corresponding approach will be subject of future research . * acknowledgements . * this work was partly supported by the german research foundation ( dfg project no . he 2789/8 - 2 and sfb 555 project c1 ) and the japanese ministry for science and education . in order to compute the clustering coefficient of the recurrence network of a dynamical system , certain system - specific integrals have to be solved . for the case of one - dimensional maps that are defined on the integral @xmath144 $ ] , these integrals can be explicitly expressed and eventually also evaluated analytically . the computation of the local clustering coefficient involves the computation of certain integrals to estimate the required probability terms discussed in section [ sec : recnet ] . in particular , we have that for a given vertex @xmath22 at a point @xmath145 , @xmath146 ^ 2 \end{split}\ ] ] for the condition in the denumerator of eq . ( [ cc_ws ] ) , where we have made use of the fact that the observations @xmath147 and @xmath148 can be considered to be independent of each other . however , this independence condition is not fulfilled in the corresponding numerator @xmath149 , which therefore requires a subtle choice of the integration boundaries to assure a correct treatment of the three - point relationships @xmath150 as one may see in fig . [ clust_1d ] , the correct integration is not trivial . in particular , if we use the abbreviations @xmath151\end{aligned}\ ] ] we obtain the following expressions for the local clustering coefficient @xmath152 if @xmath153 : + @xmath154 understanding the global clustering coefficient as the expectation value of the local one ( taken over the whole possible range in @xmath157 ) , one may use the following expression for deriving its value : @xmath158 among all nonlinear maps defined on the unit interval @xmath144 $ ] , the bernoulli map @xmath159 has the simplest possible invariant density @xmath160 ) . ] . this allows an easy evaluation of the integrals for an analytic computation of the local clustering coefficient , being aware of the restricted integration range . as a result , one finds : @xmath161 in particular , for @xmath138 ( and @xmath59 ) , we have @xmath162 @xmath1630,1[$ ] , which corresponds to the value of random geometric graphs in one dimension @xcite . hence , one may speculate about this value being a universal limit for the recurrence networks of one - dimensional chaotic maps . moreover , for @xmath164 and @xmath165 , we have @xmath166 independent of @xmath8 . this behaviour is consistent with our previous observations concerning the effect of sharp attractor boundaries on the local clustering coefficient , for example , in case of the tips of the hnon attractor ( sec . [ cluster_spatial ] ) . for the global clustering coefficient , the computation of eq . ( [ cc_global ] ) leads to @xmath167 which is in good agreement with numerical results ( see figure [ ccav_bernoulli ] ) . hence , deviations from the theoretical value @xmath168 occur exclusively due to boundary effects , and may even lead to @xmath169 for very large thresholds @xmath8 . the logistic map at @xmath170 is known to have the invariant density @xmath171 with this , we find the following expressions : @xmath172 and @xmath173 note that the integrals @xmath174 and @xmath175 can be transformed into each other by the transformation @xmath176 , and that @xmath177 is invariant under the same transformation , reflecting the corresponding symmetry of the phase space density @xmath178 . since the remaining integrals can only be solved numerically , we are not able to give an explicit equation for the functional dependence of the clustering coefficient on both @xmath157 and @xmath8 . however , comparing the numerical solution of our analytical results with the clustering coefficients of recurrence networks of one realisation of the logistic map at @xmath170 , we find ( apart from remaining fluctuations due to the finite length of the considered time series ) an excellent agreement ( see fig . [ cc_logmap ] ) . in particular , for the attractor boundaries at @xmath179 and @xmath180 , we have @xmath137 independent of @xmath8 as for the bernoulli map . moreover , we observe almost uniform values of the local clustering coefficient close to @xmath168 within the interval @xmath181 $ ] , and a systematic tendency towards larger values close to the boundaries . since the measure of the latter intervals systematically increases with increasing @xmath8 , these boundary effects are again responsible for the systematic increase in the global clustering coefficient @xmath58 as @xmath8 becomes larger . abarbanel , _ analysis of observed chaotic data _ ( springer , new york 1996 ) h. kantz , t. schreiber , _ nonlinear time series analysis _ ( cambridge university press , cambridge 1997 ) c. diks , _ nonlinear time series analysis methods and applications _ ( world scientific , singapore 1999 ) j.c . sprott , _ chaos and time series analysis _ ( oxford university press , oxford 2003 ) r.v . donner , s.m . barbosa ( eds . ) , _ nonlinear time series analysis in the geosciences applications in climatology , geodynamics , and solar - terrestrial physics _ ( springer , berlin 2008 ) h. poincar . sur la problme des trois corps et les quations de la dynamique , acta mathematica 13 , 1271 ( 1890 ) b. saussol , s. troubetzkoy , s. vaienti . recurrence , dimensions , and lyapunov exponents . j. stat . 106 , 623634 ( 2002 ) j .- p . eckmann , s. oliffson kamphorst , d. ruelle . recurrence plots of dynamical systems . europhys . lett . 5 , 973977 ( 1987 ) n. marwan , m.c . romano , m. thiel , j. kurths . recurrence plots for the analysis of complex systems . phys . rep . 438 , 237329 ( 2007 ) s. schinkel , o. dimigen , n. marwan . selection of recurrence threshold for signal detection . j. st 164 , 4553 ( 2008 ) f. takens , detecting strange attractors in turbulence , in : d. rand , l .- s . young ( eds . ) , _ dynamical systems and turbulence _ , lecture notes in mathematics , vol . 898 , 366381 ( springer , berlin 1981 ) j.p . zbilut , c.l . webber jr . embeddings and delays as derived from quantification of recurrence plots . lett . a 171 , 199203 ( 1992 ) c.l . webber jr . , zbilut . dynamical assessment of physiological systems and states using recurrence plot strategies . j. appl . physiol . 76 , 965973 ( 1994 ) n. marwan , n. wessel , j. kurths . recurrence plot based measures of complexity and its applications to heart rate variability data . phys . rev . e 66 , 026702 ( 2002 ) n. marwan . a historical review of recurrence plots . eur . phys . j. spec . 164 , 312 ( 2008 ) m. thiel , m.c . romano , j. kurths . spurious structures in recurrence plots induced by embedding . 44 , 299305 ( 2006 ) m. thiel , m.c . romano , p.l . read , j. kurths . estimation of dynamical invariants without embedding by recurrence plots . chaos 14 , 234243 ( 2004 ) m. thiel , m.c . romano , j. kurths . how much information is contained in a recurrence plot ? phys . lett . a 330 , 343349 ( 2004 ) g. robinson , m. thiel . recurrences determine the dynamics . chaos 19 , 023104 ( 2009 ) r. albert , a .- statistical mechanics of complex networks . 74 , 4797 ( 2002 ) m.e.j . the structure and function of complex networks . siam rev . 45 , 167256 ( 2003 ) l. da f. costa , f.a . rodrigues , g. travieso , p.r . villas boas . characterization of complex networks : a survey of measurements . 56 , 167242 ( 2007 ) p. crucitti , v. latora , m. marchiori . a topological analysis of the italian electric power grid . physica a 338 , 9297 ( 2004 ) c.s . zhou , l. zemanov , g. zamora , c.c . hilgetag , j. kurths . hierarchical organization unveiled by functional connectivity in complex brain networks . 97 , 238103 ( 2006 ) j.a . dunne , r.j . williams , n.d . food - web structure and network theory : the role of connectance and size . usa 99 , 1291712922 ( 2002 ) j.f . donges , y. zou , n. marwan , j. kurths . complex networks in climate dynamics . comparing linear and nonlinear network construction methods . j. st 174 , 157179 ( 2009 ) x.f . complex networks : topology , dynamics and synchronization . j. bifurcat . chaos 12 , 885916 ( 2002 ) s. boccaletti , v. latora , y. moreno , m. chavez , d .- u . complex networks : structure and dynamics . phys . rep . 426 , 175308 ( 2006 ) a. arenas , a. diaz - guilera , j. kurths , y. moreno , c. zhou . synchronization in complex networks . phys . rep . 469 , 93153 ( 2008 ) j. zhang , m. small . complex network from pseudoperiodic time series : topology versus dynamics . 96 , 238701 ( 2006 ) j. zhang , j. sun , x. luo , k. zhang , t. nakamura , m. small . characterizing pseudoperiodic time series through the complex network approach . physica d 237 , 28562865 ( 2008 ) x. xu , j. zhang , m. small . superfamily phenomena and motifs of networks induced from time series . usa 105 , 1960119605 ( 2008 ) y. shimada , t. kimura , t. ikeguchi . analysis of chaotic dynamics using measures of the complex network theory . in . : v. kurkov et al . ( eds . ) , icann 2008 , part i. lecture notes in computer science 5163 . springer , berlin , 2008 , pp . 6170 y. yang , h. yang . complex network - based time series analysis . physica a 387 , 13811386 ( 2008 ) l. lacasa , b. luque , f. ballesteros , j. luque , j.c . nuo . from time series to complex networks : the visibility graph . usa 105 , 49724975 ( 2008 ) j. wu , h. sun , z. gao . mapping to complex networks from chaos time series in the car following model . in : proc . congress traffic and transportation studies . asce , reston , 2008 , pp . . m. small , j. zhang , x. xu . transforming time series into complex networks . in : j. zhou ( ed . ) , complex 2009 , part ii . lecture notes of the institute for computer sciences , social informatics and telecommunications engineering 5 . springer , berlin , 2009 , pp . 20782089 r. donner . complex network approach to geophysical time series analysis . geophys . 11 , egu2009 - 11157 ( 2009 ) z. gao , n. jin . flow - pattern identification and nonlinear dynamics of gas - liquid two - phase flow in complex networks . e 79 , 066303 ( 2009 ) c. robinson , _ dynamical systems . stability , symbolic dynamics , and chaos _ ( 2nd ed . , crc press , boca raton 1999 ) y. hirata , k. judd , d. kilminster . estimating a generating partition from observed time series : symbolic shadowing . e 70 , 016215 ( 2004 ) c.s . daw , c.e.a . finney , e.r . a review of symbolic analysis of experimental data . 74 , 915930 ( 2003 ) r. donner , u. hinrichs , b. scholz - reiter . symbolic recurrence plots : a new quantitative framework for performance analysis of manufacturing networks . phys . j. st 164 , 85104 ( 2008 ) z .- y . evolution of traffic flow with scale - free topology . . phys . lett . 22 , 27112714 ( 2005 ) k.p . gao . a topological approach to traffic dynamics . europhys . 74 , 369375 ( 2006 ) x .- g . li , z .- y . gao , k .- relationship between microscopic dynamics in traffic flow and complexity in networks . e 76 , 016110 ( 2007 ) z .- y . li , h .- j . huang , b .- h . mao , j .- f . zheng . scaling laws of the network traffic flow . physica a 380 , 577584 ( 2007 ) j .- f . zheng , z .- y . gao . a weighted network evolution with traffic flow . physica a 387 , 61776182 ( 2008 ) p. li , b.h . an approach to hang seng index in hong kong stock market based on network topological statistics . 51 , 624629 ( 2006 ) p. li , b .- h . extracting hidden fluctuation patterns of hang seng stock index from network topologies . physica a 378 , 519526 ( 2007 ) z. gao , n. jin . community structure detection in complex networks with applications to gas - liquid two - phase flow . in : j. zhou ( ed . ) , complex 2009 , part ii . lecture notes of the institute for computer sciences , social informatics and telecommunications engineering 5 . springer , berlin , 2009 , pp . 19171928 j.f . donges , y. zou , n. marwan , j. kurths . the backbone of the climate network . europhys . , in press ( 2009 ) r. donner . spatial correlations of river runoffs in a catchment . in : j. kropp , schellnhuber ( eds . ) , _ in extremis : extremes , trends and correlations in hydrology and climate _ ( springer , berlin , forthcoming ) x .- h . ni , z .- q . jiang , w .- x . degree distributions of the visibility graphs mapped from fractional brownian motions and multifractal random walks . arxiv:0812.2099 [ physics.soc-ph ] l. lacasa , b. luque , j. luque , j.c . the visibility graph . a new method for estimating the hurst exponent of fractional brownian motion . europhys . 86 , 30001 ( 2009 ) c. liu , w .- x . zhou , w .- k . statistical properties of visibility graph of energy dissipation rates in three - dimensional fully developed turbulence . arxiv:0905.1831 [ physics.flu-dyn ] y. yang , j. wang , h. yang , j. mang . visibility graph approach to exchange rate series . physica a , in press . doi:10.1016/j.physa.2009.07.016 r. donner , t. sakamoto , n. tanizuka . complexity of spatio - temporal correlations in japanese air temperature records . in : r.v . donner , s.m . barbosa ( eds . ) , _ nonlinear time series analysis in geosciences _ , pp . 125155 , springer , berlin , 2008 p. grassberger , i. procaccia . characterization of strange attractors . 50 , 346349 ( 1983 ) m.e.j . assortative mixing in networks . phys . rev . 89 , 208701 ( 2002 ) m. thiel , m.c . romano , j. kurths , m. rolfs , r. kliegl . twin surrogates to test for complex synchronisation . europhys . lett . 75 , 535541 ( 2006 ) m.c . romano , m. thiel , j. kurths , k. mergenthaler , r. engbert . hypothesis test for synchronization : twin surrogates revisited . chaos 19 , 015108 ( 2009 ) p. van leeuwen , d. geue , m. thiel , d. cysarz , s. lange , m.c . romano , n. wessel , j. kurths , d.h . influence of paced maternal breathing of fetal - maternal heart rate coordination . usa 106 , 1366113666 ( 2009 ) n. marwan , j.f . donges , y. zou , r.v . donner , j. kurths . complex network approach for recurrence analysis of time series . arxiv:0907.3368 [ nlin.cd ] l.c . a set of measures of centrality based on betweenness . sociometry 40 , 3541 ( 1977 ) j.b . recurrence time statistics for chaotic systems and their applications . 83 , 31783181 ( 1999 ) e. ott , _ chaos in dynamical systems _ ( cambridge university press , cambridge , 1993 ) h.e . nusse , j.a . yorke , _ dynamics : numerical explorations _ ( springer , new york , 1994 ) j. dall , m. christensen . random geometric graphs . e 66 , 016121 ( 2002 ) h. yang , f. zhao , l. qi , b. hu . temporal series analysis approach to spectra of complex networks . e 69 , 066104 ( 2004 ) k.p . li , z.y . gao , x.m . zhao . multiple scale analysis of complex networks using the empirical mode decomposition method . physica a 387 , 2981 - 2986 ( 2008 )

we calculate the linear torque exerted by a planet on a circular orbit on a disc containing a toroidal magnetic field . all fluid perturbations are singular at the so called _ magnetic resonances _ , where the doppler shifted frequency of the perturbation matches that of a slow mhd wave propagating along the field line . these lie on both sides of the corotation radius . waves propagate outside the lindblad resonances , and also in a restricted region around the magnetic resonances . the magnetic resonances contribute to a significant global torque which , like the lindblad torque , is negative ( positive ) inside ( outside ) the planet s orbit . since these resonances are closer to the planet than the lindblad resonances , the torque they contribute dominates over the lindblad torque if the magnetic field is large enough . in addition , if increases fast enough with radius , the outer magnetic resonance becomes less important and the total torque is then negative , dominated by the inner magnetic resonance . this leads to outward migration of the planet . even for at corotation , a negative torque may be obtained . a planet migrating inward through a nonmagnetized region of a disc would then stall when reaching a magnetized region . it would then be able to grow to become a terrestrial planet or the core of a giant planet . in a turbulent magnetized disc in which the large scale field structure changes sufficiently slowly , a planet may alternate between inward and outward migration , depending on the gradients of the field encountered . its migration could then become diffusive , or be limited only to small scales . # 1to 0pt#1 accretion , accretion discs mhd waves planetary systems : protoplanetary discs

astro-ph0301556

almost 20% of the extrasolar planets detected so far orbit at a distance between 0.038 and 0.1 astronomical unit ( au ) from their host star . it is very unlikely that these short period giant planets , also called hot jupiters , have formed _ in situ _ ( bodenheimer 1998 ; bodenheimer , hubickyj & lissauer 1999 ) . more likely , they have formed further away in the protoplanetary nebula and have migrated down to small orbital distances . it is also possible that migration and formation were concurrent ( papaloizou & terquem 1999 ) . so far , three mechanisms have been proposed to explain the location of planets at very short orbital distances . one of them relies on the gravitational interaction between two or more giant planets , which may lead to orbit crossing and to the ejection of one planet while the other is left in a smaller orbit ( rasio & ford 1996 ; weidenschilling & marzari 1996 ) . however , this mechanism can not account for the relatively large number of hot jupiters observed . another mechanism is the so called migration instability ( murray et al . 1998 ; malhotra 1993 ) . it involves resonant interactions between the planet and planetesimals located inside its orbit which lead to the ejection of a fraction of them while simultaneously causing the planet to migrate inward . although such interactions may lead to the migration of giant planets on small scales , they would require a very massive disc to move a jupiter mass planet from several astronomical units down to very small radii . such a massive disc is unlikely and furthermore it would be only marginally gravitationally stable . the third mechanism , that we are going to focus on here , involves the tidal interaction between the protoplanet and the gas in the surrounding protoplanetary nebula ( goldreich & tremaine 1979 , 1980 ; lin & papaloizou 1979 , 1993 and references therein ; papaloizou & lin 1984 ; ward 1986 , 1997 ) . here again the protoplanet can move significantly only if there is at least a comparable mass of gas within a radius comparable to that of its orbit . however this is not a problem since this amount of gas is needed anyway in the first place to form the planet . note that hot jupiters can also be produced by the dynamical relaxation of a population of planets on inclined orbits , formed through gravitational instabilities of a circumstellar envelope or a thick disc ( papaloizou & terquem 2001 ) . however , if objects as heavy as @xmath2boo may be produced _ via _ fragmentation , it is unlikely that lower mass objects would form that way . a planet embedded in a gaseous disc launches density waves at the lindblad resonances ( goldreich & tremaine 1979 , hereafter gt79 ) . the torque exerted by the planet on these waves is responsible for the exchange of angular momentum between the disc rotation and the planet s orbital motion . the interaction between the planet and the disc inside / outside its orbit leads to a negative / positive torque on the disc , and therefore to a gain / loss of angular momentum for the planet . in the linear regime , because in a ( even uniform ) keplerian disc the outer lindblad resonances are slightly closer to the planet than the inner lindblad resonances , the interaction with the outer parts of the disc leads to a larger lindblad torque than that with the inner parts ( ward 1986 , 1997 ) . therefore , the net lindblad torque exerted by the planet causes it to lose angular momentum and to move inward relative to the gas ( type i migration ) . note that there is also a torque exerted by the planet on the material which corotates with the perturbation . its sign depends on the gradient of vortensity at corotation ( gt79 ) . however , it is usually found to be less significant than the net lindblad torque . the drift timescale for a 110 m@xmath3 planet undergoing type i migration in a standard gaseous disc is typically between @xmath4 and @xmath5 years , shorter than the disc lifetime or estimated planetary formation timescales . when the mass of the planet becomes large enough , the perturbation becomes nonlinear . feedback torques from the disc on the planet , which have to be taken into account in this regime , then opposes the motion of the planet and stops it altogether when the planet is massive enough ( ward 1997 ) . however , density waves still transport angular momentum outward . if they are dissipated locally , the angular momentum they transport is deposited in the disc in the vicinity of the planet and a gap may be cleared out ( goldreich & tremaine 1980 ; lin & papaloizou 1979 , 1993 and references therein ) . the planet is then locked into the angular momentum transport process of the disc , and migrates inward at a rate controlled by the disc viscous timescale ( type ii migration ) . here again , the drift timescale is rather short , being in the range @xmath6@xmath7 years for standard parameters . type ii migration , which occurs for planets with masses at least comparable to that of jupiter , can be avoided if after the planet forms there is not enough mass left in the disc to absorb its angular momentum . however , type i migration appears inevitable as there has to be enough gas in the disc to push a forming core inward if this core is to accrete a massive envelope to become a giant planet at some point . different scenarii for stopping a planet undergoing inward migration have been considered ( see , e.g. , terquem , papaloizou & nelson 2000 and references therein ) , but none of them can satisfactorily explain the presence of extrasolar planets with semi major axes from a few 0.01 au all the way up to several au . recently , papaloizou ( 2002 ) has shown that type i migration reverses for reasonable disc models once the eccentricity of the orbit becomes comparable to the disc aspect ratio . this is because in that case the planet spends more time near apocentre , where it is being speeded up , than near pericentre , where it is being slowed down . although the interaction with the disc tends to circularize the orbit of the planet ( goldreich & tremaine 1980 ) , significant eccentricity may be maintained by gravitational interactions between different planets forming simultaneously ( papaloizou & larwood 2000 ) . here we investigate the effect of a magnetic field on planet migration in the linear regime . we consider a planet on a circular orbit and , to keep the problem tractable , restrict ourselves to the case of a purely toroidal field . this tends to be the dominant component in discs as it is produced by the shearing of radial field lines . as described above , the torque exerted by a planet on a disc depends mainly on the location of the radii where the perturbation is in resonance with the free oscillations of the disc . we therefore expect a magnetic field to modify the tidal torque , as it introduces more degrees of freedom in the disc . the plan of the paper is as follows . in sections [ sec : basic ] , [ sec : equilibrium ] and [ sec : potential ] we give the basic equations , describe the equilibrium disc model and give the expression of the perturbing potential acting on the disc . in section [ sec : response ] we study the disc response to this perturbing potential . we first linearize the equations and derive the second order differential equation describing the disc response ( [ sec : linearization ] ) . we then carry on a wkb analysis ( [ sec : wkb ] ) , which shows that magnetosonic waves propagate outside the outermost turning points . in [ sec : magneticres ] we study the disc response at the locations where it is singular . we call these radii _ magnetic resonances_. there are two such resonances , located on each side of the planet s orbit and within the lindblad resonances . in [ sec : turning ] we calculate all the turning points associated with the disc response . on each side of the planet there are two or three of them , depending on the value of the azimuthal number @xmath8 . one of these turning points concide with the lindblad resonance . in contrast to the nonmagnetic case , waves are found to propagate in a restricted region inside the outermost turning points , around the magnetic resonances . we then proceed to give an expression for the tidal torque in general and in the vicinity of the magnetic resonances in particular in section [ sec : torque ] . numerical calculations of the torque exerted by the planet on the disc are presented in section [ sec : numerics ] . we show that the torque exerted in the vicinity of the magnetic resonances tends to dominate the disc response when the magnetic field is large enough . this torque , like the lindblad torque , is negative inside the planet s orbit and positive outside the orbit . therefore , if @xmath9 increases fast enough with radius , the outer magnetic resonance becomes less important ( it disappears altogether when there is no magnetic field outside the planet s orbit ) and the total torque becomes negative , dominated by the inner magnetic resonance . this corresponds to a positive torque on the planet , which leads to outward migration . finally , we discuss our results in section [ sec : discussion ] . the disc is described by the equation of motion : @xmath10 the equation of continuity : @xmath11 and the induction equation in the ideal mhd approximation : @xmath12 where @xmath13 is the lorentz force per unit volume , @xmath14 the pressure , @xmath15 the mass density , @xmath16 the flow velocity , @xmath17 the gravitational potential and @xmath18 the magnetic field ( @xmath19 is the permeability of vacuum ) . si units are used throughout the paper . here we neglect self gravity so that the gravitational potential @xmath17 is assumed to be due to a central mass @xmath20 . we consider a thin disc , so that the above equations can be averaged over the disc thickness . to first order we neglect the @xmath21dependence of @xmath16 and @xmath17 , so that the vertically averaged equation of motion and continuity are , respectively : @xmath22 = - { \mbox{\boldmath $ \nabla$}}\langle p \rangle + \langle { \bf f } \rangle - \sigma { \mbox{\boldmath $ \nabla$}}\psi , \label{motion}\ ] ] @xmath23 where @xmath24 is the surface mass density and the brackets denote averaging over the disc thickness . to close the system of equations , we adopt a barotropic equation of state : @xmath25 the sound speed @xmath26 is then given by : @xmath27 we adopt a nonrotating cylindrical polar coordinate system @xmath28 with origin at the central mass . we denote @xmath29 the associated unit vectors . we suppose that at equilibrium the disc is axisymmetric and in rotation around a central mass , so that @xmath30 , where @xmath31 is the angular velocity . furthermore , we assume that the equilibrium configuration contains only a toroidal magnetic field , i.e. @xmath32 . the lorentz force per unit volume is then : @xmath33 if we assume reflection with respect to the disc midplane and that @xmath34 vanishes at the disc surface , this leads to : @xmath35 we consider a planet of mass @xmath36 on a circular orbit with radius @xmath37 and angular velocity @xmath38 . at the location @xmath39 in the disc , it exerts the gravitational potential : @xmath40 where @xmath41 and we have introduced a softening length @xmath42 . we now expand @xmath43 in a fourier series with respect to the variable @xmath44 : @xmath45 with @xmath46 and @xmath47 using the generalized laplace coefficients defined as ( ward 1989 ; korycansky & pollack 93 , hereafter kp93 ) : @xmath48 where : @xmath49 @xmath50 we can rewrite @xmath51 under the form : @xmath52 @xmath53 and the expression on the right hand side has to be divided by a factor two if @xmath54 . the subscript p indicates that the quantities have to be evaluated at @xmath55 . in this paper we consider small perturbations , so that the basic equations can be linearized . we denote eulerian perturbations with a prime . since the equilibrium is axisymmetric and steady , we can expand each of the perturbed quantities in fourier series with respect to the variable @xmath44 and solve separately for each value of @xmath8 . the general problem may then be reduced to calculating the response of the disc to the real part of a complex potential of the form @xmath56 $ ] , where the amplitude @xmath51 is real and the frequency @xmath57 , and summing up over @xmath8 . we make all eulerian fluid state variable perturbations complex by writing : @xmath58 where @xmath59 is any state variable . the physical perturbations will be recovered by taking the real part of these complex quantities . we denote @xmath60 the lagrangian displacement , and write its @xmath8th fourier component as @xmath61 $ ] . since here we are only interested in the perturbations induced by the planet in the plane of the disc , we take @xmath62 . linearization of the induction equation ( [ induction ] ) , or the equivalent statement of the conservation of magnetic flux , leads to an expression for the perturbed magnetic field in the form ( chandrasekhar & fermi 1953 ) : @xmath63 the corresponding perturbation of the lorentz force per unit volume is : @xmath64 . \label{fmp}\ ] ] use of equations ( [ bp ] ) and ( [ fmp ] ) with @xmath62 gives : @xmath65 , \\ \langle f'_{m \varphi } \rangle & = & \frac{imb_1}{2 \mu_0 } \ ; \frac { \langle b^2 \rangle \xi_{mr}}{r^2 } , \label{fpertphi } \\ \langle f'_{mz } \rangle & = & 0 , \end{aligned}\ ] ] where we have defined the dimensionless quantities : @xmath66 we now linearize the @xmath67 and @xmath68components of the equation of motion ( [ motion ] ) and the equation of continuity ( [ mass ] ) . using @xmath69 , with @xmath70 , and equations ( [ state ] ) and ( [ csound ] ) , we get : @xmath71 @xmath72 @xmath73 where we have found it convenient to use the radial component of the lagrangian displacement @xmath74 and the azimuthal component of the eulerian perturbed velocity @xmath75 as variables . here @xmath76 is the linear perturbation of the enthalpy and @xmath77 is the epicyclic frequency , which is defined by : @xmath78 we can further eliminate @xmath75 and @xmath79 to get , after some tedious algebra , the following second order differential equation for @xmath74 : @xmath80 with : @xmath81 @xmath82 @xmath83 \right . \nonumber \\ & & \left . \ ; \ ; \ ; \ ; \ ; \ ; \ ; \ ; \ ; + \frac{b_2}{2 } - m^2 + \frac{c^2}{r^2 \sigma^2 } \left ( m^2 - 1 \right ) - \frac { b_1 ^ 2 c^2}{4 \beta r^2 \sigma^2 } \right\},\end{aligned}\ ] ] @xmath84,\ ] ] where @xmath85 , with @xmath86 being the alfvn speed , and we have defined the dimensionless quantities : @xmath87 the magnetic field comes in through the parameter @xmath88 only . we see by multiplying all the coefficients of equation ( [ eq2 ] ) by @xmath89 that there is no singularity in the coefficients of this equation where @xmath90 , i.e. at corotation , in contrast to the nonmagnetic case ( gt79 ) . when there is no magnetic field , the particles located at corotation can respond secularly to a forcing with frequency @xmath91 and therefore stay exactly in phase with the perturbation ( they surf the tidal wave ) . in the presence of a toroidal field however , the tension of the field line to which the particles are attached provides a restoring force which introduces some inertia in the response of the particles , destroying the resonance . from equations ( [ fpertphi ] ) , ( [ vphim ] ) and ( [ wm ] ) , we can write @xmath79 and @xmath75 as functions of @xmath74 . we give these expressions as they will be needed below : @xmath92 , \label{wm2}\ ] ] with : @xmath93 + \frac{\kappa^2}{2 \omega \sigma } - \frac{b_1 c^2}{2 \beta r^2 \sigma^2},\ ] ] and : @xmath94 , \label{vphim2}\ ] ] with : @xmath95 + \frac{b_1}{2 \beta } \right\ } - \frac{\kappa^2 r}{2 \omega c}.\ ] ] we assume that the first order derivative term in equation ( [ eq2 ] ) can be neglected ( this is verified _ a posteriori _ ) and we define @xmath96 , where @xmath37 is the corotation radius , i.e. , for a circular orbit , the planet orbital radius . then the homogeneous equation associated with equation ( [ eq2 ] ) can be rewritten as : @xmath97 \frac{d^2}{d x^2 } \left ( \frac{\xi_{mr}}{r_p } \right ) + \frac{{\cal a}_0}{\left(1+x \right)^2 } \frac{\xi_{mr}}{r_p } = 0 . \label{eq2wkb}\ ] ] we assume @xmath98 and @xmath99 , which means @xmath100 , where @xmath101 is the disc semi thickness , and @xmath102 . with @xmath103 being typically on the order of unity , the wkb approximation can then be used if @xmath104 . we have : @xmath105 providing @xmath106 , i.e. @xmath107 . when this condition is satisfied then @xmath108 and we may look for homogeneous free wave solutions of the form @xmath109 , where @xmath110 and @xmath111 is a slowly varying amplitude . since @xmath112 is of order @xmath113 , the first order derivative term can be neglected compared to the second order derivative term in equation ( [ eq2 ] ) if @xmath114 . when this is satisfied , the assumption made in this subsection is validated . then equation ( [ eq2wkb ] ) yields : @xmath115 which is the dispersion relation for magnetosonic waves ( e.g. , tagger et al . 1990 ) propagating outside the lindblad resonances ( located at @xmath116 where @xmath117 ) with the group velocity : @xmath118 in the absence of magnetic field , these waves reduce to density waves ( gt79 ) . these solutions describe the response of the disc in the parts where the perturbing potential is negligible , i.e. outside the lindblad resonances and far enough from them . we now consider the response of the disc inside the lindblad resonances . the differential equation ( [ eq2 ] ) is singular at the radii @xmath119 where @xmath120 vanishes , i.e. where @xmath121 /\beta = 0 $ ] . we call these locations _ magnetic resonances_. there are two such radii , the _ inner _ magnetic resonance @xmath122 and the _ outer _ magnetic resonance @xmath123 . at these locations : @xmath124 i.e. the frequency of the perturbation in a frame rotating with the fluid matches that of a slow mhd wave propagating along the field line . using @xmath125 $ ] , where @xmath126 is the keplerian angular velocity , and @xmath127 we get , to first order in @xmath128 : @xmath129 where @xmath130 and @xmath88 are evaluated in the vicinity of @xmath37 . we suppose here that these quantities vary on a scale large compared to @xmath130 . as the field becomes weaker ( @xmath131 ) , the magnetic resonances converge toward the corotation radius . note that since the effective lindblad resonances are all located beyond @xmath132 from corotation ( see section [ sec : turning ] ) , they are beyond the magnetic resonances . we define the new variable @xmath133 , where @xmath119 is either the inner or outer resonance , and we do a local analysis of equation ( [ eq2 ] ) around @xmath134 . for @xmath135 , we have @xmath136 . here we assume that @xmath137 , the parameters @xmath138 , @xmath139 , @xmath140 , @xmath141 and @xmath142 are at most of order unity and @xmath102 . this amounts to assuming that only @xmath143 varies in the expression ( [ coefa2 ] ) of @xmath120 . for @xmath144 , we then have : @xmath145 where the subscript @xmath146 indicates that the corresponding quantities are taken at @xmath147 , and @xmath148 or @xmath149 depending on whether @xmath150 or @xmath151 . we now approximate @xmath112 , @xmath152 and @xmath153 by their values at @xmath147 . this requires @xmath154 . if this condition is not satisfied , @xmath155 can not be replaced by its value at @xmath147 . to the first non zero order in @xmath156 and allowing for @xmath157 , we can then rewrite equation ( [ eq2 ] ) under the form : @xmath158 where we have defined the dimensionless quantity @xmath159 and where @xmath160 and @xmath161 are given by : @xmath162 @xmath163.\ ] ] we next introduce the new variable @xmath164 , so that equation ( [ eq2b ] ) becomes : @xmath165 note that @xmath166 is real outside the magnetic resonances , where @xmath167 , whereas it is imaginary inside the resonances , where @xmath168 . the solutions of this equation can be written as : @xmath169 where @xmath170 and @xmath171 are the bessel functions of first and second kind , respectively , and @xmath172 and @xmath173 are ( complex ) constants which _ a priori _ depend on @xmath174 , @xmath156 and @xmath8 . the factor @xmath175 has been introduced to simplify the asymptotic expression of @xmath176 below . when @xmath177 ( i.e. @xmath154 ) , @xmath178 and @xmath179 ( abramowitz & stegun 1972 ) . to deal with the singularity at @xmath180 , we use the landau prescription , i.e. we displace the pole slightly from the real axis by replacing @xmath181 with @xmath182 , where @xmath183 is a small positive real constant . this amounts to multiplying all the perturbed quantities by @xmath184 , and therefore to considering the response of the disc to a slowly increasing perturbation . this procedure can also be thought of as introducing a viscous force , or as shifting the resonances progressively with time . note that such a shift occurs when the planet drifts through the disc , since the material in the resonances then changes with time . shifting @xmath181 below the real axis is equivalent to shifting @xmath185 above this axis in equation ( [ eq2b ] ) , since only @xmath155 was allowed to vary in the expression of @xmath120 . therefore , we replace @xmath185 with @xmath186 , where @xmath187 is a small positive real constant . for @xmath154 , the solutions of equation ( [ eq2c ] ) can then be written as : @xmath188 , \label{sol0}\ ] ] so that , near @xmath134 : @xmath189 note that there is a phase shift of @xmath190 on passing through @xmath134 . since @xmath191 has opposite sign at @xmath192 and @xmath193 ( @xmath194 changes sign on passing through corotation whereas @xmath195 does not ) and is real , the real part of the solutions of equation ( [ eq2c ] ) should have opposite sign at @xmath192 and @xmath193 whereas their imaginary parts should have the same sign . therefore , from equation ( [ sol00 ] ) we get @xmath196 , where the asterisk denotes the complex conjugate . equations ( [ wm2 ] ) and ( [ vphim2 ] ) can now be used to evaluate @xmath75 and @xmath79 for @xmath154 : @xmath197 @xmath198 where @xmath199 is @xmath31 taken at @xmath147 . finally , @xmath69 yields , for @xmath200 : @xmath201 . \label{vmr}\ ] ] we note that , in a nonmagnetized disc , the azimuthal perturbed velocity varies as @xmath202 with @xmath203 near corotation ( gt79 ) , whereas here it varies as @xmath204 near the magnetic resonances . at the locations where the solutions of equation ( [ eq2 ] ) change from wave like to non wave like ( turning points ) , the disc response varies on a large scale , and therefore it couples well to the perturbing potential . this results in a large contribution to the tidal torque ( see gt79 and [ sec : torque ] ) . it is therefore important to calculate the location of these turning points . the wkb analysis above ( see [ sec : wkb ] ) indicates that the differential equation ( [ eq2 ] ) has turning points at the locations of the so called _ nominal _ lindblad resonances @xmath116 . waves propagate outside these resonances , which are located on both sides of the corotation radius . in the nonmagnetic case , it is known that the turning points do no concide with the nominal lindblad resonances for values of @xmath8 in excess of @xmath205 ( goldreich & tremaine 1980 , artymowicz 1993 ) . if they did concide , the turning points ( also called _ effective _ lindblad resonances ) would converge toward @xmath37 as @xmath8 increases . instead , they converge toward a radius which is located a finite distance ( equal to @xmath132 ) away from @xmath37 . since the potential is more and more localized around @xmath37 at large @xmath8 , coupling between the perturbation and the disc response is lost , and the contribution to the total torque becomes negligible . this is the torque cutoff first discussed by goldreich & tremaine ( 1980 ) . the wkb analysis above indicates that for values of @xmath8 up to about @xmath205 , the location of the turning points does not depend on whether there is a magnetic field or not . in both cases they concide with the nominal lindblad resonances . however , for larger values of @xmath8 , there are differences between a magnetized and a nonmagnetized disc , as the analysis below shows . we now calculate the location of the turning points . there are no singularities in the coefficients of equation ( [ eq2 ] ) at @xmath116 . this is similar to the nonmagnetic case , where only the second order differential equation for @xmath206 has ( removable ) singularities there ( narayan , goldreich & goodman 1987 ) . all the physical quantities are regular at these locations . to identify the turning points , we eliminate the first order derivative term in this equation . we define @xmath207 , @xmath208 being either the inner or outer nominal lindblad resonance , and introduce the new dependent variable : @xmath209 so that the homogeneous equation associated with equation ( [ eq2 ] ) can be written under the exact form : @xmath210 where : @xmath211 we now expand @xmath212 around @xmath134 . at @xmath116 , @xmath213 . we neglect here the departure from keplerian rotation ( which is second order in @xmath128 ) so that @xmath214 . then @xmath215 , with @xmath216 and @xmath217 or @xmath218 depending on whether we consider the outer or inner resonance , respectively . we assume @xmath88 and @xmath219 constant . note that for values of @xmath220 and for @xmath221 , we can not approximate @xmath143 by its value at @xmath208 . after some algebra , we find , to the first non zero order : @xmath222 \right . \nonumber \\ & & \left . \ ; \ ; \ ; \ ; \ ; \ ; - \frac{9 m^2 h^2 } { 2 x^2 \left ( x^2 - m^2 h^2 \right ) } \left ( \frac{3}{2 } + \frac{m^2 h^2}{2 x^2 } \right ) \right\ } , \label{kcoef}\end{aligned}\ ] ] where we have assumed @xmath223 , since these are the values of interest for the tidal torque ( see gt79 ) . this expression is valid even for @xmath8 larger than or very large compared to @xmath224 , and whatever the values of @xmath225 and @xmath142 compatible with @xmath88 and @xmath226 being constant and provided they are not large compared to unity . note that the expression above can also be seen as an expansion around the corotation radius , which corresponds to @xmath227 . the turning points are the radii where @xmath228 . we solve this equation numerically and display the results in figure [ fig1 ] , where we plot @xmath229 ( where @xmath230 ) as a function of @xmath8 for the turning points located beyond corotation , and for @xmath231 , 1 and 10 and @xmath232 . the outer magnetic resonance is also represented . to each of these outer turning points and resonance corresponds an inner turning point or resonance located at the same distance from corotation to first order in @xmath128 or @xmath233 . similar plots are obtained for other values of @xmath128 . artymowicz ( 1993 ) has calculated that , in a nonmagnetized disc , the location of the turning points is given by ( his eq . [ 39 ] ) : @xmath234 so that , for @xmath235 , the distance to corotation is @xmath132 . in a magnetized , we identify up to three turning points , that we label r1 , r2 and r3 : 1 . the outermost turning point r1 concides with the nominal lindblad resonance for small values of @xmath8 . when @xmath236 , its distance to corotation is @xmath132 for @xmath237 ( like in the nonmagnetic case ) and @xmath238 for @xmath239 , i.e. @xmath240 with @xmath241 . the intermediate turning point r2 is located at a distance from corotation equal to @xmath238 for @xmath237 and @xmath132 for @xmath239 . 3 . the innermost turning point r3 exists only above some critical value of @xmath8 , @xmath242 , which depends on @xmath88 and @xmath128 . interpolation of the numerical results gives @xmath243 . when @xmath235 , the distance of this turning point to corotation is @xmath244 , i.e. it concides with the magnetic resonance . for either @xmath245 or @xmath246 , r2 and r3 merge at large @xmath8 . in figure [ fig1 ] , the shaded areas indicate where the waves are evanescent ( i.e. @xmath247 ) . in contrast to the nonmagnetic case , there is a region inside the outermost turning points where waves propagate . this region moves toward the corotation radius and decreases in size as @xmath88 increases . it disappears altogether for infinite @xmath88 . note that the magnetic resonances are contained within this region , so that the singular slow modes excited at the magnetic resonances can propagate . since waves do not propagate around corotation for @xmath248 , we expect a torque cutoff at large @xmath8 like in the nonmagnetic case . we can check whether waves propagate or not around corotation by taking the limit @xmath249 ( i.e. @xmath250 ) in the expression ( [ kcoef ] ) above . in the nonmagnetic case ( @xmath251 ) : @xmath252 since this is negative , the waves are evanescent around corotation in that case . when @xmath88 is finite : @xmath253 this is positive when @xmath254 with @xmath255 , so that waves propagate around corotation for these values of @xmath8 . note that the value of @xmath242 found by interpolation of the numerical results agrees perfectly with the analytical expression . in the linear regime , the net torque exerted by the planet on the disc between the radii @xmath256 and @xmath257 is : @xmath258 with : @xmath259 { \bf r } \mbox{\boldmath $ \times$ } { \rm re } \left [ \mbox{\boldmath $ \nabla$ } \left ( \psi'_m e^{i m \left ( \varphi - \omega_p t \right ) } \right ) \right ] \ ; r d\varphi dr .\ ] ] because of the @xmath68periodicity , the first order term is zero and the torque has only a @xmath21component which can be written as : @xmath260 where , following kp93 , we have defined the `` torque density '' : @xmath261 the asterisk denotes the complex conjugate . in a nonmagnetized disc , another useful quantity is the angular momentum flux : @xmath262 \ ; { \rm re } \left [ v'_{m r } e^{i m \left ( \varphi - \omega_p t \right ) } \right ] \ ; d \varphi , \ ] ] which can also be written under the form : @xmath263 in a nonmagnetized disc , inside the lindblad resonances and away from corotation , the response of the disc is non wave like and in phase with the perturbation , so that @xmath264 is real . outside the lindblad resonances , the wavelength of the perturbation is small compared to the scale on which the perturbing potential varies so that @xmath265 integrated over radius is small . therefore , most of the contribution to the torque comes from the resonances ( lynden bell & kalnajs 1972 , gt79 ) . there are two contributions to the total torque . these are the lindblad torque , which is equal to the angular momentum advected through the disc , and the corotation torque , which accumulates at the corotation resonance ( gt79 ) and is associated with a discontinuity of the angular momentum flux there . in the presence of a magnetic field , it is not clear it is possible to calculate the torque separately at the different turning points , because the scale on which the disc response varies there may become larger than the distance between the turning points . in addition , the torque exerted around corotation for the values of @xmath8 for which waves propagate there may be significant , as the scale on which the perturbing potential varies and that on which the disc response varies may be similar . the total torque will then be calculated numerically . however , an asymptotic form of the torque density at the magnetic resonances can be obtained , and this is done in the next subsection . the torque density in the vicinity of a magnetic resonance is , from equation ( [ torquedens ] ) and with @xmath266 given by equation ( [ wmm ] ) : @xmath267 to get the torque exerted over a magnetic resonance , we integrate the torque density over the width @xmath268 of the resonance , i.e. from @xmath269 to @xmath270 . note that @xmath271 . equation ( [ wmm ] ) has been used to derive the above expression for the torque density . therefore , only if @xmath272 , and hence @xmath187 , is small compared to @xmath156 can this expression be used to calculate the torque over the resonance . when this condition is satisfied , the torque can be written as : @xmath273 with @xmath274\sim \pi/2 $ ] if @xmath275 is a few times @xmath187 . here we have assumed that the resonance were symmetrical around @xmath119 , so that the term proportional to @xmath276 in equation ( [ torquedensm2 ] ) gives no contribution . since the integral over the term proportional to @xmath277 converges ( does not depend on the width of the resonance ) , the torque we obtain can be considered as a point like contribution from the resonance , similar to the corotation torque in the nonmagnetic case . however , the resonance may not be symmetrical , in which case the term proportional to @xmath278 gives a global contribution , which does depend on the extent of the resonance . we will see below that the torque around the magnetic resonances is indeed dominated by this global contribution , and not by the point like torque . since @xmath279 , the torque derived above has opposite sign at the inner and outer magnetic resonances . to know the amplitude and the sign of @xmath280 we need to calculate @xmath281 this can not be done from the local analysis presented in section [ sec : magneticres ] , as only the solutions of the homogeneous equation determine the behaviour of the disc response at the magnetic resonances . the solutions obtained around the magnetic resonances are valid only for @xmath282 and therefore can not be matched to the wkb solutions calculated in section [ sec : wkb ] . the constants @xmath172 will then be calculated by matching the numerical and analytical solutions , and the asymptotic form of the torque density given by equation ( [ torquedensm ] ) will be used to obtain a precise value of the torque at the magnetic resonances . in order to calculate the tidal torque exerted by the planet on the disc , we have solved the second order linearized differential equation ( [ eq2 ] ) numerically . the numerical scheme we have used is similar to that presented by kp93 . the coefficients @xmath51 of the fourier decomposition of the potential ( see eq . [ [ psifourier ] ] ) are calculated using the recurrence formulae between the generalized laplace coefficients ( see the appendix of kp93 , which unfortunately contains several typos ) . equation ( [ eq2 ] ) is then integrated using the fifth order runge kutta procedure with adaptative stepsize control given by press et al . ( 1992 ) from the disc inner radius @xmath283 to its outer radius @xmath284 . note that kp93 started the integration at the corotation resonance and shot toward the boundaries . this procedure would not be appropriate here as in the presence of a magnetic field there is a resonance on each side of corotation . for a given value of @xmath8 , we first compute two independent solutions @xmath285 and @xmath286 of the homogeneous equation , and one particular solution @xmath287 of the inhomogeneous equation starting with random initial values . the general solution is then given by @xmath288 , where @xmath140 and @xmath289 are constants which depend on the boundary conditions . to calculate @xmath140 and @xmath289 , we use the wkb solution of the equation , which should be valid at @xmath283 and @xmath284 if the boundaries are far enough from corotation . then , at these locations , @xmath290 , where @xmath291 is given by equation ( [ wkbk ] ) . since the physical solutions are outgoing waves , we take @xmath292 . the group velocity is then positive ( negative ) outside ( inside ) corotation . these two boundary conditions yield two algebraic equations for @xmath140 and @xmath289 ( see eq . [ 24 ] of kp93 ) which can be solved to find these coefficients . to get better numerical accuracy , we repeat the procedure taking for the starting value of the particular solution @xmath287 the value of @xmath74 calculated at @xmath293 with the above coefficients @xmath140 and @xmath289 . this leads to @xmath287 converging toward the full solution @xmath74 . when recalculating @xmath140 and @xmath289 after one iteration we indeed verify that these two constants are very close to zero . this procedure prevents large cancellation between @xmath287 and the general solutions of the homogeneous equation . note that kp93 found the wkb approximation not to be accurate enough for the boundary conditions , and they accordingly took into account the amplitude as well as the phase variation of the wkb solution . we found that we could reproduce their results without having to go beyond the wkb approximation . once @xmath74 is known , @xmath79 and @xmath75 can be calculated from equations ( [ wm2 ] ) and ( [ vphim2 ] ) , @xmath294 , and the torque density @xmath295 is given by equation ( [ torquedens ] ) . to calculate the total torque @xmath296 exerted by the planet on the disc , we integrate @xmath295 over the disc surface ( see eq . [ [ torquer1r2 ] ] ) . the planet potential softening length is set to @xmath297 . note that in the presence of a magnetic field the results are insensitive to the value of @xmath42 provided it is small enough , since there is no singularity of the homogeneous equation at @xmath55 , where the potential is singular . the landau parameter is set to @xmath298 . as expected , the total torque does not depend on the value of @xmath187 provided it is small enough . for @xmath299 , we integrate the equation from @xmath300 to @xmath301 , whereas , for @xmath302 , we limit the range of integration from @xmath303 to @xmath304 . we assume that @xmath24 , @xmath26 and @xmath305 vary like power laws of @xmath67 . according to equations ( [ bs ] ) and ( [ cd ] ) , we then have @xmath306 , @xmath307 and @xmath308 . then @xmath309 and @xmath310 . note that @xmath311 and @xmath312 . we comment that the values of the parameters we chose below may require a departure from keplerian rotation that is second order in @xmath128 to ensure hydrostatic equilibrium . this is here neglected , as in ward ( 1986 , 1997 ) . to simplify the discussion , we define the dimensionless quantities @xmath313 , @xmath314 , @xmath315 and @xmath316 , where the subscript @xmath317 indicates that the quantity is taken at @xmath55 . the numerical results presented in this section correspond to @xmath318 . since @xmath319 ( see section [ sec : potential ] ) , then @xmath320 and @xmath321 . by setting @xmath322 , we have checked that we recover the kp93 results . in that case , the corotation torque is non zero if @xmath323 and corresponds to a discontinuity of the angular momentum flux at @xmath55 . in figure [ fig2 ] , we plot @xmath324 , the angular momentum flux @xmath325 at radius @xmath67 and the torque @xmath326 exerted by the planet on the disc between the radii @xmath37 and @xmath67 for @xmath322 , @xmath327 , @xmath328 ( @xmath329 ) and @xmath330 . for comparison , we also plot the angular momentum flux corresponding to @xmath331 in the corotation region . the parameters used here are the same as those used by kp93 in their figures 2 , 5 and 6 , so that a direct comparison can be made . for @xmath327 , the dimensionless corotation torque is about 102 , whereas the dimensionless total torque @xmath332 is about 22 , in very good agreement with the results displayed by kp93 . we first consider the case @xmath327 , @xmath334 ( @xmath329 ) , @xmath335 , @xmath336 and @xmath330 . in figures [ fig3 ] , [ fig4 ] and [ fig5 ] we plot @xmath324 , @xmath337 and @xmath338 versus radius . note that the rapid variation of @xmath339 at @xmath55 is due to the rapid variation of the potential there , not to a singularity of the homogeneous differential equation . we have matched @xmath79 obtained numerically with the analytical expression ( [ wmm ] ) in order to compute the constants @xmath172 . we find that @xmath340 whereas @xmath341 . according to equation ( [ torqueres ] ) , the point like torque exerted at the outer magnetic resonance is then negative whereas that exerted at the inner magnetic resonance is positive . this is in agreement with the curves shown in figure [ fig4 ] around resonances . figure [ fig5 ] shows the total torque computed using the numerical form of the torque density . as a check , we have also calculated @xmath338 by using the analytical expression ( [ torquedensm ] ) for the torque density around the resonances ( represented by the dashed lines in the lower panels of fig . [ fig4 ] ) . in general the two calculations are in good agreement with each other , although in some cases the discrepancy may reach about 50% ( the sign of the torque does not change however ) . in figure [ fig6 ] , we plot the total torque @xmath338 exerted by the planet on the disc versus @xmath8 for different values of @xmath141 , @xmath140 and @xmath138 and for both @xmath322 and @xmath333 . in the magnetic case , the value of @xmath88 at @xmath37 is taken to be 1 . the cumulative torque @xmath342 is obtained by summing up over @xmath8 . we get an estimate by assuming @xmath338 is constant in the intervals @xmath343 , @xmath344 etc ... and @xmath345 for @xmath346 . the values computed for @xmath322 are in very good agreement with kp93 . in this case , the cumulative torque exerted by the planet on the disc is positive , which means that the planet loses angular momentum and migrates inward . when the magnetic field is non zero though , @xmath342 may become negative , with a subsequent outward planet migration . this occurs when @xmath138 is negative enough , i.e. when @xmath88 increases fast enough with radius or , equivalently , when the magnetic field decreases fast enough with radius . for instance , when @xmath347 , @xmath348 for @xmath349 , i.e. @xmath350 or , equivalently , @xmath351 , and for @xmath352 , i.e. @xmath353 or , equivalently , @xmath354 . note that this variation needs only be _ local_. when @xmath138 goes from positive to negative values , the cumulative torque , initially positive , then becomes negative passing through zero . therefore migration is slowed down and then reversed . in figure [ fig6b ] , we plot the one sided torque exterted by the planet on the disc ( i.e. the torque exterted inside and outside its orbit ) versus @xmath8 for @xmath327 and @xmath355 , i.e. @xmath329 . these plots show that the one sided torque has a larger amplitude when a magnetic field is present . in figure [ fig7 ] we plot the torque exerted between @xmath283 and @xmath67 versus @xmath356 for @xmath330 , @xmath347 , @xmath357 and for @xmath358 and @xmath359 . this shows that the total torque becomes negative when @xmath138 is negative due to the contribution from the region inside the lindblad resonances . in table [ tab1 ] we have listed the value of the torque exerted on different radii intervals in the disc for @xmath330 , @xmath327 , @xmath355 ( @xmath329 ) , @xmath357 and different values of @xmath138 . the torque exerted between the inner lindblad resonance @xmath360 and corotation @xmath37 is negative , whereas that exerted between corotation and the outer lindblad resonance @xmath361 is positive . ( note that for @xmath330 the lindblad resonances concide with the outermost turning points ) . we have found above that the point like torque exerted at the inner magnetic resonance is positive whereas that exerted at the outer resonance is negative . therefore , the torque is not dominated by the point like contribution at the magnetic resonances , but is determined by the whole region around these resonances , where waves propagate for @xmath362 ( see also the discussion in section [ sec : torquereson ] ) . when @xmath138 decreases , @xmath88 increases outward faster , so that the magnetic field becomes less and less important outside the planet s orbit . therefore the torque exerted between @xmath37 and @xmath361 , which is positive and tends to be negligible in the absence of a magnetic field , becomes less and less important . in contrast , the magnetic field becomes more important inside the orbit , where it gives rise to a negative torque . the net effect of decreasing @xmath138 is then to decrease the total torque . if the magnetic field is strong enough , the total torque exerted on the disc is dominated by the region inside the lindblad resonances ( which is closer to the planet than the regions outside these resonances ) , and it can then become negative for small enough values of @xmath138 . figure [ fig8 ] shows the total torque exerted by the planet on the disc versus @xmath8 for @xmath327 , @xmath355 ( @xmath329 ) , @xmath349 , which correspond to @xmath350 , and different values of @xmath363 . as we can see , the cumulative torque @xmath342 is significantly decreased for values of @xmath363 as large as @xmath364 and becomes negative for values of @xmath363 between @xmath364 and 10 . the cumulative torque corresponding to @xmath365 becomes negative if , for the same parameters as above , @xmath141 is taken to be at least as large as unity , i.e. if @xmath24 increases with radius at least as fast as linearly . this is because when @xmath141 increases @xmath88 increases faster with radius , which favours the negative inner torque exerted near corotation . this effect can be seen in figure [ fig6 ] by comparing the upper and middle panels , which show that the total torque corresponding to @xmath331 is larger than that corresponding to @xmath327 for @xmath349 . note that @xmath366 tends to favour the outer lindblad resonance compared to the inner one , which results in a larger lindblad torque ( ward 1986 , 1997 ) . the fact that the total torque actually decreases when @xmath141 increases means that the variations of @xmath24 affect more significantly the torque exerted around the magnetic resonances than the lindblad torque . ( note that the variations of the torque displayed in figure [ fig6 ] when @xmath141 varies in the nonmagnetic case are mainly due to changes in the corotation torque ) . an increase of @xmath140 ( corresponding to @xmath26 or , equivalently , @xmath128 , increasing faster with radius ) also results in a smaller total torque , as can be seen by comparing the upper and lower panels of figure [ fig6 ] . when @xmath140 gets larger , @xmath88 increases faster with radius , which again lowers the torque exerted around the magnetic resonances . also the lindblad torque is reduced when @xmath140 gets larger ( ward 1986 , 1997 ) . figure [ fig9 ] shows the effect of varying the value of @xmath367 at @xmath55 for @xmath330 , @xmath327 , @xmath329 , @xmath357 and @xmath349 and 3 . the total torque exerted on the disc tends to zero as @xmath128 increases . this is because the resonances are further away from the planet when @xmath128 is larger , so that the coupling between the perturbation and the tidal potential is weaker . for @xmath349 , the torque is always negative and decreases as @xmath128 decreases . for @xmath368 , the torque is positive and maximum for some value of @xmath128 ( @xmath369 ) . it stays positive ( and becomes smaller ) as @xmath128 increases , whereas it becomes negative for smaller values of @xmath128 . this is because even though @xmath88 increases with radius in that case , the region between the inner lindblad resonance and corotation contributes to a larger magnitude of the torque than the region between corotation and the outer lindblad resonances . this is also the case for @xmath232 and @xmath370 , as shown in table [ tab1 ] . note that what determines the sign of the torque is the gradient of @xmath88 , not of the magnetic field itself , as only @xmath88 enters into equation ( [ eq2 ] ) . we have calculated the linear torque exerted by a planet on a circular orbit on a disc containing a toroidal magnetic field . in contrast to the nonmagnetic case , there is no singularity at the corotation radius , where the frequency of the perturbation matches the orbital frequency . however , there are two new singularities on both sides of corotation , where the frequency of the perturbation in a frame rotating with the fluid matches that of a slow mhd wave propagating along the field line . these so called _ magnetic resonances _ are closer to the planet than the lindblad resonances . in addition , on each side of the planet , there are two turning points located beyond the magnetic resonance . one of them concides with the lindblad resonance . for values of @xmath8 larger than some critical value @xmath371 , there is a third turning point between the magnetic resonance and corotation , on each side of the planet . like in the nonmagnetic case , waves propagate outside the outermost turning points . but here they also propagate inside the intermediate turning points when @xmath254 , or between the intermediate and innermost turning points when @xmath248 ( see figure [ fig1 ] ) . the singular modes excited at the magnetic resonances can therefore propagate . there is a significant torque exerted on the region of the disc inside the outermost turning points ( which concide with the lindblad resonances for the values of @xmath8 of interest ) . like the lindblad torque , this torque is negative inside the planet s orbit and positive outside its orbit . the whole region around corotation , not just a narrow zone around the magnetic resonances , contribute to this torque . in other words , the magnetic resonances contribute to a global , not a point - like , torque . since these resonances are closer to the planet than the lindblad resonances , they couple more strongly to the tidal potential . therefore , the torque exerted around the magnetic resonances dominate over the lindblad torque if the magnetic field is large enough . if in addition @xmath0 increases fast enough with radius , the outer magnetic resonance becomes less important ( it disappears altogether when there is no magnetic field outside the planet s orbit ) and the total torque is then negative , dominated by the inner magnetic resonance . this corresponds to a positive torque on the planet , which leads to outward migration . the amount by which @xmath88 has to increase outward for the total torque exerted on the disc to be negative depends mainly on the magnitude of @xmath88 . we have found that for @xmath372 at corotation , the cumulative torque ( obtained by summing up the contributions from all the values of @xmath8 ) exerted on the disc is negative when @xmath88 increases at least as fast as @xmath373 . if @xmath374 , the cumulative torque becomes negative for values of @xmath363 between @xmath375 and 10 , whereas it is negative for @xmath376 if @xmath377 . the migration timescales that correspond to the torques calculated above are rather short . the orbital decay timescale of a planet of mass @xmath378 at radius @xmath37 is @xmath379 , where @xmath380 is the cumulative torque exerted by the planet on the disc . remembering that @xmath381 , we get : @xmath382 in a standard disc model , @xmath383@xmath6 g @xmath384 at 1 au ( see , for instance , papaloizou & terquem 1999 ) . therefore , @xmath385@xmath5 yr for a one earth mass planet at 1 au in a nonmagnetic disc , as @xmath386 in that case ( see fig . [ fig6 ] ) . this is in agreement with ward ( 1986 , 1997 ) . in a magnetic disc , @xmath387 may become larger , leading to an even shorter migration timescale ( note that given the limited accuracy of the numerical scheme used , as indicated in section [ sec : casebne0 ] , only orders of magnitude for the migration timescale are obtained here ) . however , it is important to keep in mind that these timescales are _ local_. once the planet migrates outward out of the region where @xmath88 increases with radius , it may enter a region where @xmath88 behaves differently and then resume inward migration for instance . such a situation would be expected to occur in a turbulent magnetized disc in which the large scale field structure changes sufficiently slowly . unless the magnetic field is above equipartition , it is unstable by the magnetorotational instability ( balbus & hawley 1991 , 1998 and references therein ) , the saturated nonlinear outcome of which is mhd turbulence . in a turbulent magnetized disc , the main component of the field is toroidal , as shown by numerical simulations ( hawley , gammie & balbus 1995 ; brandenburg et al . global simulations have also shown that the turbulence saturates at a level corresponding to @xmath1 if there is no mean flux or to lower values of @xmath88 if there is a mean flux ( hawley 2001 ; steinacker & papaloizou 2002 ) . also , substantial spatial inhomogeneities are created by radial variations of the maxwell stress ( hawley 2001 ; steinacker & papaloizou 2002 ) , so that the field may display significant gradients . on the basis of the work presented here , we are led to speculate that in such a disc the planet would suffer alternately inward and outward migration , or even no migration at all . it would then oscillates back and forth in some region of the disc , or undergo some kind of diffusive migration , either outward or inward depending on the field gradients encountered . note however that in a turbulent disc the magnetic field may vary locally on timescales shorter than the timescales needed to establish the type of tidal response assumed in this paper . it has been pointed out that protoplanetary discs may be ionized enough for the magnetic field to couple to the matter only in their innermost and outermost parts ( gammie 1996 ; fromang , terquem & balbus 2002 ) . a planet forming at around one astronomical unit , where the ionization fraction is very low , would then migrate inward on the timescale calculated by ward ( 1986 , 1997 ) until it reaches smaller radii where the field is coupled to the matter . at this point , there would be a magnetic field inside the planet s orbit but not outside its orbit . the torque on the planet would then reverse , and outward migration would occur . however , as soon as the planet would re enter the nonmagnetic region , inward migration would resume . hence , the planet would stall at the border between the magnetic and nonmagnetic regions . inward migration in a magnetized disc may then either be very significantly slowed down , occur only on limited scales , or not occur at all . the planet would then be able to grow to become a terrestrial planet or the core of a giant planet . when the planet becomes massive enough ( about 10 earth masses ) , the interaction with the disc becomes nonlinear , with a gap being opened up around the planet s orbit . because a rather strong torque is exerted in the vicinity of the planet in the presence of a magnetic field , a gap should open up more easily and be cleaner in a turbulent magnetic disc than in a similar laminar viscous disc . this seems to be in agreement with the numerical simulations of a giant planet in a turbulent disc performed by nelson & papaloizou ( 2002 ) . once the gap is completely cleared out , the effect of the magnetic resonances disappear as they are inside the edges of the gap ( which concide with the lindblad resonances corresponding to @xmath388 ) . the magnetic field may however still have an effect on the disc planet interaction because of the turbulence it produces ( nelson & papaloizou 2002 ) . we have focused here on a planet with a circular orbit . a planet on an eccentric orbit in a nonmagnetic disc usually suffers eccentricity damping , as in that case the corotation resonances , which damp the eccentricity , dominate over the lindblad resonances , which excite it ( goldreich & tremaine 1980 ; papaloizou , nelson & masset 2001 ) . this picture may change once a magnetic field is introduced . the effect of a magnetic field on the eccentricity of a planet will be studied in another paper . it is a pleasure to thank steven balbus and john papaloizou for their advice , encouragement , and many valuable suggestions and discussions . i am grateful to john papaloizou for his comments on an early draft of this paper which led to significant improvements . llllllll @xmath138 & @xmath389 & @xmath390 & @xmath391 & @xmath392 & total torque @xmath338 + + 8 & -172 & -110 & 624 & 194 & 536 + 7 & -171 & -153 & 553 & 186 & 415 + 3 & -165 & -305 & 408 & 208 & 146 + 2 & -163 & -344 & 377 & 209 & 79 + 1 & -160 & -385 & 333 & 211 & -1 + 0 & -158 & -423 & 298 & 217 & -66 + -1 & -154 & -463 & 263 & 217 & -137 + -3 & -155 & -570 & 184 & 217 & -324 + -5 & -145 & -641 & 129 & 225 & -432 + -8 & -105 & -645 & 69 & 225 & -456 + & & & & & + & -126 & -333 & 310 & 170 & 21 +

we present a new method to probe atmospheric pressure on earthlike planets using ( o-o ) dimers in the near - infrared . we also show that dimer features could be the most readily detectable biosignatures for earthlike atmospheres , and may even be detectable in transit transmission with the james webb space telescope ( jwst ) . the absorption by dimers changes more rapidly with pressure and density than that of monomers , and can therefore provide additional information about atmospheric pressures . by comparing the absorption strengths of rotational and vibrational features to the absorption strengths of dimer features , we show that in some cases it may be possible to estimate the pressure at the reflecting surface of a planet . this method is demonstrated by using the o a band and the 1.06 m dimer feature , either in transmission or reflected spectra . it works best for planets around m dwarfs with atmospheric pressures between 0.1 and 10 bars , and for o volume mixing ratios above 50% of earth s present day level . furthermore , unlike observations of rayleigh scattering , this method can be used at wavelengths longer than 0.6 m , and is therefore potentially applicable , although challenging , to near - term planet characterization missions such as jwst . we have also performed detectability studies for jwst transit transmission spectroscopy and find that the 1.06 m and 1.27 m dimer features could be detectable ( snr ) for an earth - analog orbiting an m5v star at a distance of 5 pc . the detection of these features could provide a constraint on the atmospheric pressure of an exoplanet , and serve as biosignatures for oxygenic photosynthesis . we have calculated the required signal - to - noise ratios to detect and characterize o monomer and dimer features in direct imaging reflected spectra and find that signal - to - noise ratios greater than 10 at a spectral resolving power of r=100 would be required .

1312.2025

atmospheric pressure is a fundamental parameter for characterizing the environment and habitability of an extrasolar planet . water s stability on a planetary surface as a liquid depends on both the surface temperature and pressure . while the freezing point of water is not strongly dependent on pressure , pressure does affect water s boiling point and sublimation . thus , a reliable estimate of the surface pressure is an important part of the measurement suite required to determine the habitability of an exoplanet . despite the importance of atmospheric pressure , current proposed methods for measuring pressure using remote - sensing techniques that could be applicable to exoplanet atmospheres are challenging . the existing techniques include the use of rayleigh scattering @xcite or the widths of individual absorption lines @xcite or absorption bands @xcite . the presence and location of a blue rayleigh scattering tail in a spectrum can provide information about the existence and pressure of an atmosphere . however , strong blue absorbers in the atmosphere ( e.g. o@xmath3 , so@xmath0 , no@xmath0 and many others ) or surface features can mask this tail @xcite . furthermore , the rayleigh scattering tail is most prominent shortward of 0.6 @xmath1 m , below the short wavelength cutoff of the james webb space telescope ( jwst ) @xcite . lastly , planets around m dwarfs are likely to be the first to be characterized @xcite , and m dwarfs have relatively less visible - flux to rayleigh scatter than solar - type stars . the rayleigh tail would be more difficult to detect and characterize for planets orbiting stars of this stellar class . it is also possible to use the widths of absorption features to estimate pressure . pressure increases the widths of vibration rotation lines of gases . this method has been successful for the earth using high - resolution spectra of the o@xmath0 a band @xcite , for mars using co@xmath0 features near 2 @xmath1 m @xcite and the cloud tops of venus using the 1.6 @xmath1 m co@xmath0 band @xcite . this method provides unambiguous results when the spectral resolution is sufficiently high to resolve the profiles of individual spectral lines . it can also be used at lower spectral resolution , but requires prior knowledge of the mixing ratio of the absorbing gas . here we explore the feasibility of a new method to directly measure the pressure of an earth - like atmosphere that combines the absorption features of dimers with those of monomer vibration - rotation bands to yield estimates of the atmospheric pressures even when the mixing ratio of the monomer is uncertain . previous pressure estimates using dimer absorption have been made for the cloud tops of earth , but these techniques have required prior knowledge of the gas mixing ratio profile @xcite . dimers are bound or quasi - bound states between two molecules driven together by molecular interactions . for example , the o@xmath0-o@xmath0 or o@xmath4 dimer consists of two o@xmath0 molecules temporarily bound to each other by van der waals forces . this dimer has its own rotational and vibrational modes , and produces spectral features distinct from its constituent o@xmath0 monomers . additionally , absorption from dimer molecules is more sensitive to pressure than that of monomers . the optical depth ( how much absorption occurs ) for dimers and monomers can be expressed by the following equations : @xmath5 where @xmath6 and @xmath7 are the monomer and dimer differential optical depths , @xmath8 is the monomer cross section , @xmath9 is the number density of the gas , @xmath10 is the dimer cross section , @xmath11 is the pressure , @xmath12 is the temperature , and @xmath13 is the path length . while the monomer ( e.g. o@xmath0 ) optical depth is directly proportional to pressure , the optical depth of the dimer ( e.g. o@xmath0-o@xmath0 ) is dependent on the square of the density ( and hence square of the pressure ) . this difference in pressure dependence allows us to estimate atmospheric pressure by comparing the dimer and monomer absorption features for an oxygen - rich , earth - like atmosphere , the best combination of bands to use for pressure determination at near infrared wavelengths ( @xmath20.6 @xmath1 m ) would be the ( o@xmath0 ) a band at 0.76 @xmath1 m , and the 1.06 @xmath1 m o@xmath4 dimer band . the 0.76 @xmath1 m o@xmath0 a band is the strongest o@xmath0 feature in the visible - near - infrared spectral region , and is found in a relatively clean region of the spectrum between two water vapor bands . we have chosen the dimer feature at 1.06 @xmath1 m as the likely best option , due to its combination of band strength and its location in a relatively uncluttered region of the planetary spectrum . other o@xmath4 features overlap with water features ( dimer feature between 5.5 and 7 @xmath1 m ) or o@xmath0 vibration - rotation bands ( 0.63 @xmath1 m , 0.76 @xmath1 m and 1.27 @xmath1 m dimer features ) or are weaker than the 1.06 @xmath1 m dimer feature ( 0.477 @xmath1 m and 0.57 @xmath1 m dimer features ) . nevertheless , some other features , in particular the strong 1.27 @xmath1 m feature , could be used if the 1.06 @xmath1 m dimer feature is not detectable . in the proposed technique , the o@xmath0 0.76@xmath1 m ( monomer ) band is used to provide an estimate of the atmospheric concentration of o@xmath0 , and combined with the o@xmath4 1.06 @xmath1 m ( dimer ) band to constrain the atmospheric pressure . this method can be used with either transmission spectroscopy , or directly - detected reflection or emission spectra , and serves as a complement to pressure determination techniques such as rayleigh scattering , which only work in the visible . although the proof of concept is shown here with oxygen , this technique is not limited to the oxygen dimer in an earth - like atmosphere in the visible to near - infrared . the same technique is applicable to pairs of monomer and dimer absorption features across a wider range of planetary atmospheric composition and spectral wavelength range . in this paper we generated transit transmission and direct imaging reflected spectra for cloud and aerosol - free earthlike exoplanets . the models we have used to do this are described below . when an extrasolar planet transits or occults its host star , the planetary atmosphere is backlit and some of the star s light traverses the planet s atmosphere on limb trajectories . this transmitted light can be used to characterize the planet s atmosphere @xcite . this has been done for a number of jupiter- and neptune - sized planets ( e.g. @xcite ) and the super - earth / mini - neptune gj1214b ( e.g. @xcite ) the transmission spectroscopy model used here is based on smart ( spectral mapping atmospheric radiative transfer ) @xcite , which is a spectrum resolving ( line - by - line ) , multi - stream , multiple scattering radiative transfer model . we have modified smart to generate transit transmission spectra by combining the monochromatic absorption and scattering opacities for each atmospheric layer calculated by smart with the limb path lengths inherent in a transit transmission event . the transmission model included gas absorption , rayleigh scattering , interaction - induced absorption , extinction from clouds and aerosols , refraction and limb darkening . because multiple scattering was not included in the transit transmission component of the model , we have considered only cloud and aerosol - free atmospheres for this work . an important characteristic of the transit transmission spectroscopy model is the inclusion of refraction . as described in @xcite , refraction sets a fundamental limit on the range of pressures that can be probed during a transit , independent of absorption and scattering . light refracts as it passes through an atmosphere , with a larger refraction angle as higher pressures and densities are probed . for every planet - star system there will be a maximum tangent pressure in the planet s atmosphere that can be probed , because at greater pressures the light will be refracted by too large of an angle to be able to reach a distant observer during the transit . for each tangent height in each atmosphere , we calculated the total angle of refraction for a beam of light emitted from the host star using a modified version of the method described in @xcite . their method was developed for calculating refraction for astronomical observations on the earth given a tangent altitude , apparent zenith angle and an atmospheric density profile . we calculated the angle of refraction over a range of zenith angles to determine if a path exists to connect the host star to a distant observer via the planetary atmosphere . transit transmission spectroscopy can not probe the tangent altitudes at which no such path exists . the radius of the star , planet - star distance and composition of the planet s atmosphere determine the maximum tangent pressure . the radius of the star and planet - star distance control the apparent angular size of the star from the planet s perspective . the larger the angular size , the greater the range of pressures that can be probed . for an earth - analog orbiting a sun - like star the angular size of the star is @xmath140.5@xmath15 while for an earth - analog orbiting an m5 dwarf and receiving the same total flux , the angular size of the star is @xmath142@xmath15 . therefore , transit transmission spectroscopy can probe higher pressures , i.e. see deeper into the atmosphere , for the planet orbiting an m dwarf . figure [ fig : refraction ] shows this effect by comparing all possible paths at one tangent height for a planet around an m dwarf and a planet around a sun - like star , with each planet receiving the same total flux . the composition of the atmosphere determines the refractivity ( index of refraction - 1 ) of the atmosphere . atmospheres with greater refractivities will have lower maximum tangent pressures . the refractivity at stp ( standard temperature and pressure ) can vary from @xmath141.5 times the refractivity of air for co@xmath0 to slightly less than half the refractivity of air for h@xmath0 , when considering only the common bulk atmospheric gases in the solar system . therefore , in general it will be possible to probe higher pressures for an h@xmath0 atmosphere than a co@xmath0 or air atmosphere . the reflected spectra were generated using the standard version of smart , which can include multiple scattering from clouds and aerosols . however , to maintain consistency with the transmission spectrum model , clouds were not included in this study . the reflected spectra were generated assuming a surface with a constant albedo of 0.16 , which is the average albedo of the cloud - free earth @xcite . we also assumed the surface is a lambertian scatterer . other surface types could have introduced an error into any quantitative estimates in this paper , unless explicitly included in a retrieval attempt . figure [ fig : albedo ] shows the wavelength - dependence of a variety of surface types from the aster ( advanced spaceborne thermal emission and reflection radiometer ) spectral library @xcite and the usgs ( united states geological survey ) digital spectral library @xcite . surface albedos are nearly constant in the bands considered here , though , for example , snow fluctuates by @xmath1420% within the 1.06 @xmath1 m dimer band . we modeled a test case with smart with the snow surface to determine the error different surfaces can introduce . we found a difference of 15% between the seawater and snow cases when measuring the equivalent width ( defined in section [ sec : measurements ] ) of the 1.06 @xmath1 m dimer band . this difference was significantly less than the difference in measured equivalent widths for the cases considered in this paper . therefore , we have considered any discrepancies due to variations in surface albedo to be minimal for the present work . the cloud - free model atmospheres were generated by a one - dimensional ( altitude ) photochemical code with an extensive history in early earth @xcite , modern earth @xcite and exoplanet @xcite research . the planetary radius and surface gravity used were the radius of the earth ( 6371 km ) and surface gravity of the earth ( 9.87 m / s@xmath16 ) . the vertical grid consists of 200 plane - parallel layers that are each 0.5 km thick in altitude , in which radiative transfer , atmospheric transport , and photochemical production and loss are solved simultaneously , subject to upper ( stellar flux , atmospheric escape ) and lower ( volcanos and biology ) boundary conditions . the model calculates the mixing ratios of each species in each layer by solving the coupled mass - continuity / flux equations with the reverse euler method ( appropriate for stiff systems ) and a variable time - stepping algorithm . dimer concentrations were computed using estimations from quantum mechanical calculations . first , we fit the temperature dependence of the equilibrium constant for o@xmath0-dimer formation using the following equations : @xmath17 where p(o@xmath0)@xmath0 and po@xmath0 are the dimer and monomer partial pressures and @xmath12 is the temperature @xcite . the dimer mixing ratio was then computed as : @xmath18 where @xmath11 is the pressure in atmospheres . the choice of quantum parameters in our fit of equation [ eqn : kpt ] ensures that our calculation is a lower limit to the dimer concentrations , an assumption which matches modern atmospheric data well @xcite . the model atmospheres used in this study started with boundary conditions that reproduce earth s modern atmospheric chemistry . we then replaced the solar spectrum with the m dwarf spectrum described in [ sec : stellar - prop ] and decreased the surface albedo to 0.16 to account for cloud - free conditions . this `` modified earth around an m dwarf '' model was then perturbed to examine changes to both total pressure and oxygen concentrations . total atmospheric pressures of 0.1 , 0.5 , 1.0 , 3.0 , 5.0 and 10.0 bar were examined . at each of these total pressures , lower boundary conditions on o@xmath0 mixing ratios were set at 0.1 , 0.5 , 1.0 and 2.0 times earth s present level . this corresponds to oxygen mixing ratios from 2% to 42% , which is roughly the range of oxygen concentrations experienced throughout the past 2.5 gyr of earth s history @xcite . boundary conditions for all other species were held fixed at modern values . figure [ fig : profiles ] shows the pressure - temperature profile and volume mixing ratio profiles for the 1.0 times the present atmospheric level ( pal ) o@xmath0 cases for spectrally active gases in the wavelength region examined here . stable steady - state solutions were analyzed in all but the three cases with the highest o@xmath0 surface partial pressures . for those three cases , we extrapolated from our converged results at lower o@xmath0 mixing ratios by increasing o@xmath0 concentrations , scaling the dimer concentrations with the square of the o@xmath0 mixing ratio , and keeping all other gas mixing ratios constant . for all tests presented here , we assumed earth - like temperature and water vapor profiles , precluding the need for costly climate simulations . more specifically , we took the modern earth temperature profile as a function of altitude , and computed the corresponding pressure levels assuming hydrostatic equilibrium . this simplification provides a useful baseline , but limits the range of validity of these results somewhat because the gas absorption cross sections and number densities depend on these atmospheric properties . the impact of this assumption is assessed in section [ sec : sensitivity ] . for pressure / temperature regions corresponding to earth s mesosphere , we adopted a temperature of 180 k which likely overestimates the temperature in these regions . dimer absorption preferentially occurs in higher pressure regions and so is insensitive to assumed conditions in the tenuous upper atmosphere . our model grid was capped at 100 km altitude ensuring optically thin conditions for nearly all species , and made allowances for co@xmath0 and n@xmath0 photolysis above the upper boundary . we used the modern measured eddy - diffusion profile to simulate convective motion for all atmospheres regardless of pressure . while this simplification would affect the prediction of trace gas concentrations , all species analyzed here are either well - mixed ( co@xmath0 , o@xmath0 , n@xmath0 ) or short - lived ( dimers ) so are not sensitive to changes in turbulent mixing . we used the hitran 2008 database line lists @xcite to generate opacities in smart . the dimer absorption cross sections for o@xmath0 were taken from @xcite . in this paper we have assumed that the planet orbits an m5v star , or an m dwarf with stellar radius of 0.20 r@xmath19 and a luminosity of 0.0022 l@xmath19 @xcite . the planet was placed at a distance of 0.047 au , so that the total integrated incoming stellar flux was equal to the total flux the earth receives from the sun today . an m5 dwarf was chosen for these tests because there is a high probability that the terrestrial planets which will be the most easily characterized in the near future will be orbiting this class of star @xcite . additionally , transit transmission spectroscopy can probe pressures up to the @xmath141 bar level in an atmosphere for an earth analog around an m dwarf while for the earth around the sun , it can only probe pressures as great as @xmath140.2 bars . at these pressures and lower , the 1.06 @xmath1 m dimer feature is very weak , even at 200% pal o@xmath0 . thus , using earth - like atmospheres around an m dwarf instead of around the sun provides a better demonstration of the pressure - dependence of dimer features . to simulate the spectrum of the star , we used a phoenix nextgen synthetic spectrum @xcite for all wavelengths greater than @xmath14300 nm . for the shorter wavelengths we used the uv spectrum of ad leo @xcite . the phoenix spectrum was normalized so that the total integrated flux was equal to 1373 w / m@xmath16 , which is the integrated flux the earth receives from the sun . the ad leo spectrum from @xcite was left unchanged , as it was already normalized to equal the amount of flux a planet near the inner edge of the habitable zone would receive . we have used a spectral resolving power of 100 to provide relevance to the james webb space telescope ( jwst ) , which will provide new opportunities for characterizing the atmospheres of transiting exoplanets @xcite . once the jwst is launched , the near - infrared spectrograph ( nirspec ) will provide spectra with a spectral resolving power ( r=@xmath20 ) of @xmath21100 between 0.6 and 5.0 @xmath1 m in single prism mode @xcite . we examine the effect of varying the spectral resolution in sections [ sec : resolution1 ] and [ sec : resolution2 ] . to make a quantitative estimate of absorption strengths we measured equivalent widths for the reflected spectra and measured parts per million ( ppm ) differences in flux for the transmission spectra . equivalent widths were calculated using the following equation : @xmath22 where @xmath23 is the equivalent width , f@xmath24 is the flux at each wavelength @xmath25 , and f@xmath26 is the continuum flux at each wavelength . to obtain the equivalent widths , we first measured the area of the spectral band below the continuum . for each absorption band , we define the continuum by hand . for the o@xmath0 a band , the continuum was assumed to be linear with wavelength , and for the 1.06 @xmath1 m feature the continuum was assumed to be constant with wavelength because in several of the cases the continuum at the longer wavelengths was difficult to define due to h@xmath0o absorption . the equivalent width is the width , in units of wavelength , of a rectangle measured from the continuum to the level of zero flux with the same total area as the spectral band . we could not use this type of measurement for the transmission spectra because of the difficulty in defining the zero flux level . therefore , we quantified the absorption strengths for the transmission spectra by measuring the change in flux from the continuum to the point of greatest absorption within a band . we performed detectability studies @xcite for the model spectra , assuming the exoplanet - star system is at a distance of 5 pc . we calculated the expected signal - to - noise ratio ( snr@xmath27 ) using the jwst exposure time calculator ( etc ) . the jwst etc includes background noise from sky , dark , thermal and zodiacal sources along with read - out noise and photon noise . the estimates provided are expected to be within 20% of the mission requirements . we note that for all the cases considered , the noise is dominated by photon noise . we do not include noise from detector intrapixel variations @xcite , but in principle calibration time could be devoted to mapping the pixels , as has been done with spitzer space telescope infrared array camera @xcite . we also note that noise levels within @xmath1420% of the photon noise limit have been obtained for transit transmission spectra with hst in spatial scan mode , wherein the target star is trailed during each exposure by telescope motion perpendicular to the direction of dispersion @xcite . spatial scan mode is being considered for jwst , and so assuming photon - limited noise in our calculations , while optimistic , provides a reasonable estimate of the detectability of absorption features ( drake deming , private communication , october 13 , 2013 ) . we assumed that every possible transit is observed in jwst s 5 year mission lifetime , ignoring decreases in integration time due to non - zero impact parameters ( where the planet does not traverse the center of the stellar disk and therefore has a lower transit duration ) and limits on visibility based on the ecliptic latitude of the exoplanet ( see @xcite ) . for the earth - sun analog , this corresponds to a total integration time of @xmath142.3 * 10@xmath28 seconds , and for the earth - m5v analog , an integration time of @xmath1410@xmath29 seconds . we normalize the solar spectrum ( using the phoenix g2v model available on the site ) to a johnson v magnitude of 3.32 and the m5v spectrum to 2.1 * 10@xmath30 erg @xmath31 s@xmath32 @xmath32 at 1 @xmath1 m . at each wavelength within an absorption band we measured the signal as the magnitude of the difference from the continuum flux . the noise is expressed in parts per million ( ppm , 10@xmath29/snr@xmath27 ) at each wavelength . the final snr of the transit transmission spectrum is the square root of the sum of the squares of the snr at each wavelength in the absorption band divided by @xmath33 , which is included because the transit transmission spectrum must be calibrated against the out of transit spectrum of the star . we also performed detectability studies for the direct imaging reflected spectra that could be relevant to proposed direct imaging planet detection and characterization missions . for these calculations , we did not use an instrument simulator because the exact specifications for these missions are not currently defined . for each absorption band , we calculated two snrs , one for detecting the spectral feature ( snr@xmath34 ) and one for measuring the flux at the center of the band to a precision of 3@xmath8 ( snr@xmath35 ) . we calculated two snrs because obtaining information about pressure from a spectral feature requires more than detection ; it also requires a quantitative estimate of the strength of that spectral feature . to calculate snr@xmath34 , we divided the reflected flux by the stellar flux , defined a continuum and calculated the signal as the difference between the continuum and the normalized reflected flux . we assumed the noise is constant over the entire absorption band , and then calculated the noise level required to detect the spectral feature with a snr@xmath36 of 3 in the absorption band . the final snr@xmath34 is the mean of the continuum reflected flux level divided by the calculated noise . to calculate snr@xmath35 , we selected the wavelength within the band with the lowest radiance . we set the value of a second noise level as the lowest radiance divided by 3 . snr@xmath35 is the continuum flux level divided by the noise required to obtain a snr of 3 at the lowest radiance in the absorption band . we have generated transit transmission spectra and direct imaging spectra for atmospheres with o@xmath0 concentrations of 10% , 50% , 100% and 200% of the pal and pressures of 0.1 , 0.5 , 1 , 3 , 5 and 10 bars . figure [ fig : plottran ] shows the resulting transit transmission spectra for 100% pal o@xmath0 . transmission spectra of atmospheres with pressures @xmath371 bar are nearly identical for a given o@xmath0 concentration because refraction limits to depth of penetration to @xmath381 bar . the dimer feature is weak for pressures of 0.1 and 0.5 bars . figures [ fig : plottran-0.1 ] , [ fig : plottran-0.5 ] , and [ fig : plottran-2.0 ] show the resulting transit transmission spectra for o@xmath0 concentrations of 10% , 50% and 200% pal . the 1.06 @xmath1 m dimer feature is very weak at all pressures for o@xmath0 concentrations at 10% pal and weak for 50% pal o@xmath0 , while it is very strong for pressures above 0.1 bars at 200% pal o@xmath0 . figure [ fig : plotrad ] shows the reflected spectra for modern o@xmath0 mixing ratios in atmospheres with a range of total pressure from 0.1 to 10 bars . the 0.5 and 5.0 bar cases are omitted from the plots to increase clarity , but are still included in the equivalent width and snr calculations . the 1.06 @xmath1 m dimer feature is extremely weak for the present - day atmosphere , but in contrast to its behavior in the transmission spectra , it is a very prominent feature for the 3 , 5 and 10 bar atmospheres . figures [ fig : plotrad01 ] , [ fig : plotrad05 ] and [ fig : plotrad2 ] show the reflected spectra at pressures between 0.1 and 10 bars for 10% , 50% , and 200% pal of o@xmath0 , respectively . in the 10% pal case , the o@xmath0 dimer feature at 1.06 @xmath1 m is very weak because the total amount of o@xmath0 is very low , even for a 10 bar atmosphere . the 1.06 @xmath1 m dimer feature is stronger in the 3 , 5 and 10 bar cases for atmospheres with 50% pal o@xmath0 . finally for the 200% o@xmath0 atmospheres , the 1.06 @xmath1 m dimer feature is one of the strongest spectral features , even in the 1 bar atmosphere . figure [ fig : equivwidths ] shows the flux change for the transmission spectra and equivalent widths for reflected spectra at different pressures and o@xmath0 concentrations for the o@xmath0 a band and the 1.06 @xmath1 m dimer feature . for a given o@xmath0 concentration the transit transmission flux differences for the o@xmath0 a band are roughly constant for pressures @xmath371 bar due to refraction . the 1.06 @xmath1 m dimer feature flux differences increase slightly with pressure but are also constant for pressures @xmath371 bar for a given o@xmath0 concentration . the dimer feature does not appear in transmission for cases with 10% pal o@xmath0 . for the direct imaging ( reflected ) spectra , both the o@xmath0 a band and 1.06 @xmath1 m dimer feature equivalent widths increase with pressure and increased o@xmath0 concentrations . however , the dimer feature equivalent widths are much more sensitive to pressure . at higher pressures the dimer feature is strong except for cases with 10% pal o@xmath0 , in which the 1.06 @xmath1 m dimer feature is too weak to quantify . figures [ fig : pressure - tran - only]-[fig : refl - only ] show the relationships between the quantitative absorption measurements described above and atmospheric quantities including the o@xmath0 mixing ratio and the o@xmath0 partial pressure at the surface . the ppm flux difference measured in transit transmission for the o@xmath0 a band could be used to constrain the o@xmath0 mixing ratio , as shown in figure [ fig : pressure - tran - only]b . the o@xmath0 partial pressure at the surface can be estimated using the ratio between the 1.06 @xmath1 m dimer and o@xmath0 a band absorption measurements . these ratios are shown in figure [ fig : pressure - tran - only]a using ppm flux differences , and in figure [ fig : refl - only - toto2 ] using equivalent widths . the o@xmath0 mixing ratio and o@xmath0 partial pressure at the surface can be combined to provide a unique estimate of the surface pressure of the planet . a more detailed description of the pressure measurement technique is given in section [ sec : measure ] . to quantify the errors introduced by assuming the modern day temperature profile , we generated spectra to test the sensitivity of our models to changes in the temperature profile and in changes to the water vapor profiles . we compared our 1.0 bar , 1.0x pal o@xmath0 spectra to a spectrum generated using the same volume mixing ratio profiles but with an isothermal atmosphere at 250 k. we also compared our 1.0 bar , 1.0x pal o@xmath0 transit transmission spectrum to spectra generated with atmospheres with 0.1 and 10.0 times the h@xmath0o levels . we perform a similar comparison for the 1.0 bar , 2.0x pal o@xmath0 reflected spectrum . figure [ fig : iso - comp ] shows the sensitivity of the spectra to the temperature profile , with our earth - like profile and an isothermal approximation profile compared . both the transit transmission and reflected spectra show little sensitivity to the temperature profile . the 1.06 @xmath1 m dimer band also shows little sensitivity to the temperature profile , despite the dependence of the dimer optical depth on the square of the temperature because the isothermal temperature profile approximates the average temeprature of the troposphere , where the majority of dimer absorption occurs . figures [ fig : h2o - test-1 ] , [ fig : h2o - test-2 ] and [ fig : h2o - test-3 ] show the results for the h@xmath0o sensitivity tests . the o@xmath0 a band equivalent widths and ppm flux differences are not strongly affected by changes in the h@xmath0o profiles . however , changes in the h@xmath0o mixing ratios affect the continuum flux in the wings of the 1.06 micron dimer feature , complicating measurements of the equivalent width of this feature . for the transit transmission spectra , the change in the total ppm flux difference ( across the entire band ) is less than 20% between the 0.1 and 10.0x h@xmath0o cases . for the reflected spectra , the change in the equivalent width of the 1.06 @xmath1 m dimer feature is less than 20% . for the reflected spectra , the equivalent widths can be much greater than that for the 1.0 bar , 2.0x o@xmath0 case . for these greater equivalent widths , the effect of increasing or decreasing h@xmath0o levels will diminish as the difference from the continuum flux ( affected by h@xmath0o ) and the flux within the absorption band will increase . table [ tab : detect ] shows the snrs for observations by jwst for the o@xmath0 a band , 1.06 @xmath1 m feature and the 1.27 @xmath1 m feature for the range of pressures and o@xmath0 concentrations considered here for an earth analog at a distance of 5 pc . the snr are calculated assuming that every transit of an earth analog orbiting an m5v star is observed over jwst s 5 year mission lifetime . in transit transmission , the o@xmath0 a band snrs are no greater than 1.1 . the 1.06@xmath1 m dimer feature is detectable at a snr of @xmath23 for many of the 2.0x pal o@xmath0 cases . the 1.27 @xmath1 m feature is the most detectable o@xmath0 feature in this wavelength range , with snrs greater than 5 for many of the 1.0x pal o@xmath0 cases and greater than 7 for many of the 2.0x pal o@xmath0 cases . jwst will not be able to detect o@xmath0 species for earth - like exoplanets in secondary eclipse in the visible and near - infrared , as shown by the secondary eclipse snr levels . even for the highest pressure cases , the snrs are no greater than @xmath140.2 . table [ tab : tpf ] shows the snrs necessary to detect and characterize the o@xmath0 a band , 1.06 @xmath1 m band and the 1.27 @xmath1 m feature for the range of pressures and o@xmath0 concentrations considered here for the direct imaging reflected spectra at r=100 . the snrs in table [ tab : tpf ] would be relevant to a direct imaging characterization mission . while these snrs were calculated for an earth analog orbiting an m5v star , the results should be independent of stellar spectral type because we divided out the stellar flux in our calculations . the o@xmath0 a band , 1.06 @xmath1 m dimer feature and 1.27 @xmath1 m feature are detectable at an average snr@xmath34 of 14 , 9 and 14 , respectively , for the cases when the features are strong enough to identify in the model spectra . the average required snr@xmath35 to use the features for pressure estimation are 11 , 31 and 34 . we examined the effect of spectral resolving power on the detectability of spectral features by generating transit transmission spectra for the 1.0 bar , 1.0x pal o@xmath0 case with spectral resolving powers of 500 , 200 , 100 , 80 , 60 , 40 , 30 and 20 . figure [ fig : specres1 ] shows the spectra for each of these cases . we also measured the snr of each feature at each resolving power . the snrs were calculated assuming a noise profile equivalent to the jwst nirspec noise profile , but with the noise level at each wavelength divided by ( 100/r)@xmath16 , so that the noise at each wavelength decreased as the resolving power decreased . figure [ fig : specres2 ] shows how the total snr in each absorption band changes with resolving power . in general , the snrs drop off rapidly as resolving power decreases for r@xmath3860 . we also generated direct imaging reflected spectra for the 1.0 bar , 2.0x pal o@xmath0 case at resolving powers of 500 , 200 , 100 , 80 , 60 , 40 , 30 and 20 , as shown in figure [ fig : specres - imaging ] . figures [ fig : specres - imaging - snrd ] and [ fig : specres - imaging - snrp ] show how the snr for detection and precision vary with resolving power for the o@xmath0 a band , 1.06 @xmath1 m dimer feature and the 1.27 @xmath1 m feature . the snr required to detect each feature increases as resolving power decreases . at r=20 , the snrs are not shown because no spectral features could be identified , and at r=30 only the o@xmath0 a band was identified . the snr required to quantify the flux at the center of each spectral feature decreases as resolving power decreases , because the lowest radiance level increases as the spectral resolving power decreases . the 1.06 @xmath1 m dimer feature is prominent in transit transmission for atmospheres with @xmath3750% pal o@xmath0 and surface pressures @xmath370.5 bars . it is a prominent feature in the reflected spectra for atmospheres with @xmath3750% pal o@xmath0 and surface pressure @xmath373 bars . for the reflected spectra , the dimer feature equivalent width is highly dependent on surface pressure when compared to the o@xmath0 a band and therefore the dimer feature can be used to constrain pressure . here we discuss how to do this for the cases investigated here . figure [ fig : equivwidths ] confirms that dimer absorption features are more strongly dependent on pressure than monomer features . therefore dimer features can be combined with monomer features to determine pressure , even if the mixing ratio of the absorbing gas is not known . with only transit transmission spectroscopy , it is impossible to probe pressures over @xmath141 bar for the cases examined here , but it may be possible to constrain the o@xmath0 mixing ratio and set a lower bound for pressure . in reflected spectra , it is possible to determine the surface partial pressure of o@xmath0 and set a lower bound for pressure using the 1.06 @xmath1 m dimer feature as an on / off pressure gauge . with both a transit transmission spectrum and a reflected spectrum , it should be possible to determine total atmospheric surface pressure for an earth - like exoplanet . transmission spectroscopy provides only a lower bound on the atmospheric pressure because refraction provides a fundamental limit to which pressures can be probed using this technique . for the spectra presented here , a lower limit of @xmath141 bar can be set for the high pressure atmospheres . for a given o@xmath0 concentration a unique estimate of the pressure can be retrieved from the ratio between the ppm flux differences of the 1.06 @xmath1 m dimer feature and the o@xmath0 a band . figure [ fig : pressure - tran - only]a shows the relationship between this ratio and the total amount of o@xmath0 above 0.9 bars , which is the highest pressure that can be probed in this particular case . there is a clear trend between this ratio and the amount of o@xmath0 in the atmosphere . when combined with an o@xmath0 mixing ratio this relationship can provide a quantitative estimate of a lower level of the surface or cloud - top pressure . the o@xmath0 mixing ratio can be estimated from the flux difference of the o@xmath0 a band in transmission . figure [ fig : pressure - tran - only]b shows the relationship between the o@xmath0 mixing ratio ( which is constant throughout the atmosphere ) and the o@xmath0 a band flux difference . for pressure @xmath390.1 bars the o@xmath0 flux difference is roughly constant for a given o@xmath0 concentration , meaning that a measurement of the o@xmath0 a band flux difference should correlate with the o@xmath0 mixing ratio . reflected spectra alone can provide an estimate of the surface partial pressure of o@xmath0 by examination of the ratio of the 1.06 @xmath1 m dimer equivalent width and the o@xmath0 a band equivalent width . figure [ fig : refl - only - toto2 ] shows the relationship between this ratio and the surface partial pressure of o@xmath0 . the strength of the 1.06 @xmath1 m dimer feature could also be used as an on / off gauge to set a lower bound for pressure . determining total atmospheric pressure with only a reflected spectrum is difficult due to degeneracies between the o@xmath0 concentration and total atmospheric pressure , as shown in figure [ fig : refl - only ] . for large equivalent widths of the dimer feature , the pressure will certainly be above 1 bar . however , it is difficult to differentiate between atmospheres with the same o@xmath0 surface partial pressure . nevertheless , it appears that it is possible to set a lower limit on pressure by measuring the 1.06 @xmath1 m dimer feature equivalent width . for example , a 1.06 @xmath1 m dimer feature equivalent width greater than @xmath1410 nm would imply a surface pressure @xmath21 bar . if both a transit transmission spectrum and a reflected spectrum are available , it should be possible to directly measure the total atmospheric surface pressure in the absence of clouds . transit transmission spectroscopy can provide an estimate of the o@xmath0 mixing ratio as described previously . a reflected spectrum can theoretically probe to the reflecting surface and therefore can be used to constrain the o@xmath0 partial pressure at the surface . by combining the o@xmath0 mixing ratio and o@xmath0 partial pressure we can determine the total pressure at the reflecting surface , which could either be a reflective cloud layer or the physical surface of the planet . the methods described here could be used in the near future by the jwst nirspec instrument , which will potentially be able to characterize transiting planets between 0.6 and 5.0 @xmath1 m . the o@xmath0 a band will likely not be detectable for a nearby earth - analog with jwst . although this feature is strong in the spectrum , the sensitivity of nirspec is poor at shorter wavelengths . the 1.06 @xmath1 m dimer feature is detectable at the 3@xmath8 level in transit transmission for cases with 2.0x pal o@xmath0 and high surface pressures . thus , the detection of the 1.06 @xmath1 m dimer feature would imply a surface or cloud - top pressure greater than or equal to 1.0 bars . for cases in which the 1.06 @xmath1 m dimer feature is not detectable , the 1.27 @xmath1 m feature could be used to constrain the pressure . this feature is not as strongly dependent on pressure as the 1.06 @xmath1 m dimer feature , but it is more detectable in all cases explored here . tpf , darwin or a similar direct imaging mission will be required to characterize the reflected spectra of nearby earth analogs in the visible and near - infrared . the snr values for secondary eclipse using jwst are all less than 1 , so jwst will not be able to characterize the reflected spectra of earth - analogs in secondary eclipse . table [ tab : tpf ] shows the necessary snrs to detect and characterize spectral features for a direct imaging planet characterization mission . while the snrs were calculated for an earth analog orbiting an m5v star , the results should be largely independent of spectral type because we have divided the reflected flux by the stellar flux in our calculations . the required snr values suggest that a snr of @xmath210 would be necessary to detect and quantify the o@xmath0 a band , 1.06 @xmath1 m dimer feature and 1.27 @xmath1 m feature for a true earth analog . however , because continuum brightness changes with pressure , a different snr criteria would be necessary for higher pressure atmospheres . for example , the continuum brightness near the o@xmath0 a band is three times lower for the 10.0 bar cases than it is for the 0.1 bar cases . for most cases , a snr of @xmath27 would likely be sufficient to set a lower limit on the surface pressure using the 1.06 @xmath1 m dimer feature . clouds and aerosols will also affect the detectability of absorption features . in transit transmission , clouds can effectively mask the highest pressures of the atmospheres at which dimer absorption is most prominent . however , in partially cloudy atmospheres some of the paths will probe pressures as high as the maximum tangent pressure , and thus the planetary transmission spectrum could show evidence of dimer absorption . furthermore , absorption in the 1.27 @xmath1 m dimer band can be detected with snr@xmath21 in even some 0.1 bar cases , meaning that there could be a detectable dimer absorption signal for even a completely cloud - covered planet if the cloud deck pressure was @xmath370.1 bars . for reflected spectra , clouds will truncate paths before they reach the surface and limit the dimer absorption for those paths . however , for partially cloudy atmospheres , dimer absorption could be detectable in the paths that do reach the surface . additionally , because cloud albedos are typically greater than surface albedos , the presence of clouds will increase the continuum brightness levels and the brightness in the center of absorption bands . the increase in brightness has been shown to decrease the required snr to detect and characterize o@xmath0 monomer absorption in cloudy atmospheres when compared to cloud - free cases @xcite , though the effect of cloud albedo on the detectability of dimer absorption features has not been heretofore examined . therefore , while clouds will impact the detectability of dimer features , using dimers to determine pressure and as biosignatures may still be feasible for cloudy atmospheres . figures [ fig : specres2 ] shows the snrs for spectral features at varying resolving powers for transit transmission spectra of a 1.0 bar , 1.0x pal o@xmath0 atmosphere . the snrs for each band are greatest at the highest resolving powers , and then gradually decrease until r@xmath1460 or 80 , at which the snrs decrease strongly . this dramatic decrease with resolving power occurs because at the lowest resolving powers , the absorption bands are indistinguishable from the continuum . additionally , the highest flux levels in the continuum can not be resolved at lower spectral resolving powers , decreasing the total signal . this effect can be seen most easily for the spectra with r=20 , in which no absorption features can be identified . figures [ fig : specres - imaging - snrd ] and [ fig : specres - imaging - snrp ] show the snrs for the direct imaging reflected spectra . in contrast to figure [ fig : specres2 ] , these two figures show the required snr to detect and characterize an absorption band , not the snr that could be obtained with jwst . the required snr to detect spectral features increases as resolving power decreases . however , this effect would be mitigated because the expected noise at each wavelength should decrease as resolving power decreases . the snr required to quantify each absorption band decreases as resolving power decreases because the lowest radiance level increases . at r@xmath3840 , however , spectral features are very difficult to identify , making these resolving powers unsuitable for detecting and characterizing o@xmath0-related absorption features . in addition to their utility as pressure probes , the 1.06 @xmath1 m dimer feature and 1.27 @xmath1 m feature could potentially be detectable biosignatures for nearby earth - like planets . the o@xmath0 a band has long been considered the most viable o@xmath0 biosignature , but it is unlikely to be the most detectable biosignature for an earth - like planet in transit transmission . as initially described in @xcite , lunar eclipse observations show that the 1.06 @xmath1 m and 1.27 @xmath1 m dimer features are more detectable than o@xmath0 monomer features like the a band , which is corroborated by our model spectra and detectability calculations . the 1.27 @xmath1 m o@xmath0 feature has been examined as a potential biosignature for ground - based telescopes by @xcite , but to our knowledge detectability studies of neither the 1.06 @xmath1 m dimer feature nor the 1.27 @xmath1 m feature have been undertaken for jwst . our results show that the 1.27 @xmath1 m feature would be detectable with a snr of 5 for a cloud - free earth - analog at 5 pc . therefore , we conclude that o@xmath0 features , especially the 1.06 @xmath1 m dimer feature and the 1.27 @xmath1 m feature , could be detectable biosignatures for oxygenic photosynthesis with jwst . clouds and aerosols will make estimating pressure using dimer features more difficult . a direct imaging observation of a partially cloud - covered planet will be able to probe to the surface for a fraction of the paths , such that the dimer feature will be weaker than for a cloud - free planet . cloud and aerosol extinction can also be wavelength dependent , which may complicate using equivalent widths or ppm flux differences to determine pressure . nevertheless dimer features can still provide a lower bound for pressure if clouds and aerosols can not be explicitly included in the retrieval method . higher h@xmath0o or co@xmath0 abundances in an atmosphere could also make this method more challenging . higher h@xmath0o abundances will make it more difficult to define a continuum for the 1.06 @xmath1 m dimer feature , as shown in figures [ fig : h2o - test-2 ] and [ fig : h2o - test-3 ] . however , as discussed in section [ sec : sensitivity ] , the magnitude of this error should typically be less than 20% for most cases in which the 1.06 @xmath1 m dimer feature could be detectable . co@xmath0 has an absorption feature near 1.06 @xmath1 m @xcite , which could make using the 1.06 @xmath1 m dimer feature difficult . this could be overcome by modeling out absorption from h@xmath0o and co@xmath0 or by using other dimer features to supplement information from the 1.06 @xmath1 m dimer feature , such as the 1.27 @xmath1 m dimer feature . lastly , not knowing the mixing ratio of o@xmath0 will make estimating pressure difficult when only a reflected spectrum is available . this is similar to the problem in using the absorption widths of rotation - vibration features to constrain pressure . however , the 1.06 @xmath1 m dimer feature is more sensitive to pressure than a monomer feature , and therefore can provide a better estimate of pressure than a monomer feature alone . furthermore , monomer features can not be used as an on / off pressure gauge , while dimer features can . spectrally resolved observations of o@xmath0 dimer absorption can be combined with observations of the absorption by o@xmath0 vibration - rotation bands to provide independent constraints on the o@xmath0 concentration and the surface or cloud - top pressure in oxygenated atmospheres for planets around m dwarfs , and with low levels of co@xmath0 . even if a precise estimate for the pressure is not possible , the presence of dimer absorption is indicative of pressures greater than @xmath210.5 bars in transmission spectroscopy , and greater than @xmath211 bars in reflected spectra . we have shown that this method is feasible in the absence of clouds for oxygenated atmospheres . for transmission spectroscopy this method can be used to estimate the surface or cloud - top pressure for o@xmath0 concentrations @xmath39 100% pal and for pressures @xmath20.5 bars . for reflected spectra this method will work if the o@xmath0 partial pressure at the surface is @xmath20.3 bars . clouds will reduce the absorption strength of dimer features by limiting the paths that can probe the highest pressures in transit transmission , and by truncating paths before they reach the surface in the reflected spectra . jwst may be able to detect the 1.06 @xmath1 m dimer feature and the 1.27 @xmath1 m o@xmath0 feature for an earth analog orbiting an m dwarf in transit transmission . thus , while not all the techniques described here will be applicable to jwst observations , a lower bound on pressure could be set for an exoplanet using an o@xmath0 dimer feature . furthermore , we showed that a direct imaging mission operating in the visible and near - infrared like tpf would require a spectral resolving power of r@xmath240 , and preferably higher . at r=100 , a snr of @xmath230 would be required to not only detect o@xmath0 related absorption features , but to also provide an estimate of an exoplanet s atmospheric pressure . we thank drake deming for helpful dscussions on the detectability calculations , and the two anonymous reviewers for their thorough and helpful reviews that greatly improved the paper . this work was performed by the nasa astrobiology institute s virtual planetary laboratory , supported by the national aeronautics and space administration through the nasa astrobiology institute under cooperative agreement solicitation nnh05zda001c . this work has also been supported by a generous fellowship from the arcs seattle chapter and funding from the astrobiology program at the university of washington under an nsf igert award . some of the work described here was conducted at the jet propulsion laboratory , california institute of technology , under contract with nasa . this research has made use of nasa s astrophysics data system . no competing financial interests exist . belu , a. r. , selsis , f. , morales , j .- c . , ribas , i. , cossou , c. , rauer , h. 2011 . primary and secondary eclipse spectroscopy with jwst : exploring the exoplanet parameter space . astronomy and astrophysics 525 , a83 . carey , s. , and 12 colleagues 2012 . absolute photometric calibration of irac : lessons learned using nine years of flight data . society of photo - optical instrumentation engineers ( spie ) conference series 8442 , . catling , d. c. , claire , m. w. , zahnle , k. j. , quinn , r. c. , clark , b. c. , hecht , m. h. , kounaves , s. 2010 . atmospheric origins of perchlorate on mars and in the atacama . journal of geophysical research ( planets ) 115 , 0 . crow , c. a. , mcfadden , l. a. , robinson , t. , meadows , v. s. , livengood , t. a. , hewagama , t. , barry , r. k. , deming , l. d. , lisse , c. m. , wellnitz , d. 2011 . views from epoxi : colors in our solar system as an analog for extrasolar planets . the astrophysical journal 729 , 130 . deming , d. , and 11 colleagues 2009 . discovery and characterization of transiting super earths using an all - sky transit survey and follow - up by the james webb space telescope . publications of the astronomical society of the pacific 121 , 952 - 967 . deming , d. , and 20 colleagues 2013 . infrared transmission spectroscopy of the exoplanets hd 209458b and xo-1b using the wide field camera-3 on the hubble space telescope . the astrophysical journal 774 , 95 . forget , f. , spiga , a. , dolla , b. , vinatier , s. , melchiorri , r. , drossart , p. , gendrin , a. , bibring , j .- p . , langevin , y. , gondet , b. 2007 . remote sensing of surface pressure on mars with the mars express / omega spectrometer : 1 . retrieval method . journal of geophysical research ( planets ) 112 , 8 . garca muoz , a. , zapatero osorio , m. r. , barrena , r. , montas - rodrguez , p. , martn , e. l. , pall , e. 2012 . glancing views of the earth : from a lunar eclipse to an exoplanetary transit . the astrophysical journal 755 , 103 . ignatiev , n. i. , titov , d. v. , piccioni , g. , drossart , p. , markiewicz , w. j. , cottini , v. , roatsch , t. , almeida , m. , manoel , n. 2009 . altimetry of the venus cloud tops from the venus express observations . journal of geophysical research ( planets ) 114 , 0 . kasting , j. f. , w. a. traub , et al . , exoplanet characterization and the search for life , white paper for astronomy and astrophysics decadal survey , available at : http://sites.nationalacademies.org/bpa/bpa_050603#planetarysystems kawahara , h. , matsuo , t. , takami , m. , fujii , y. , kotani , t. , murakami , n. , tamura , m. , guyon , o. 2012 . can ground - based telescopes detect the oxygen 1.27 @xmath1 m absorption feature as a biomarker in exoplanets ? . the astrophysical journal 758 , 13 . khler , j. , melf , m. , posselt , w. , holota , w. , te plate , m. 2005 . optical design of the near - infrared spectrograph nirspec . society of photo - optical instrumentation engineers ( spie ) conference series 5962 , 563 - 574 . meadows , v. s. , crisp , d. 1996 . ground - based near - infrared observations of the venus nightside : the thermal structure and water abundance near the surface . journal of geophysical research 101 , 4595 - 4622 . pont , f. , knutson , h. , gilliland , r. l. , moutou , c. , charbonneau , d. 2008 . detection of atmospheric haze on an extrasolar planet : the 0.55 - 1.05 @xmath1 m transmission spectrum of hd 189733b with the hubble space telescope . monthly notices of the royal astronomical society 385 , 109 - 118 . rauer , h. , gebauer , s. , paris , p. v. , cabrera , j. , godolt , m. , grenfell , j. l. , belu , a. , selsis , f. , hedelt , p. , schreier , f. 2011 . potential biosignatures in super - earth atmospheres . i. spectral appearance of super - earths around m dwarfs . astronomy and astrophysics 529 , a8 . slanina , z. , uhlk , f. , de almeida , w. , hinchliffe , a. 1994 . a computational thermodynamic evaluation of the altitude profiles of ( n@xmath0)@xmath0 , n@xmath0-o@xmath0 and ( o@xmath0)@xmath0 in the earth s atmosphere . thermochimica acta 231 , 55 - 60 . spiga , a. , forget , f. , dolla , b. , vinatier , s. , melchiorri , r. , drossart , p. , gendrin , a. , bibring , j .- p . , langevin , y. , gondet , b. 2007 . remote sensing of surface pressure on mars with the mars express / omega spectrometer : 2 . meteorological maps . journal of geophysical research ( planets ) 112 , 8 . vidal - madjar , a. , lecavelier des etangs , a. , dsert , j .- m . , ballester , g. e. , ferlet , r. , hbrard , g. , mayor , m. 2003 . an extended upper atmosphere around the extrasolar planet hd209458b . nature 422 , 143 - 146 . wakeford , h. r. , and 15 colleagues 2013 . hst hot jupiter transmission spectral survey : detection of water in hat - p-1b from wfc3 near - ir spatial scan observations . monthly notices of the royal astronomical society 435 , 3481 - 3493 . .snrs for the o@xmath0 a band , the 1.06 @xmath1 m dimer feature and the 1.27 @xmath1 m feature for all the cases considered in both transit transmission and secondary eclipse . the calculations were done for an earth analog orbiting an m5v star at a distance of 5 pc . the total integration time was assumed to be 10@xmath29s , equal to the total amount of time spent in transit for this case over jwst s 5 year mission lifetime . the 1.06 @xmath1 m dimer feature and the 1.27 @xmath1 m feature could be detectable , allowing a lower limit for atmospheric pressure to be set . [ tab : detect ] [ cols="<,<,<,<,<,<,<,<",options="header " , ] monomer and dimer bands . for any absorption band , as long as the albedo does not vary widely within the band s wavelength range it should be possible to measure an accurate equivalent width for each feature . the maximum variation within a band is no more than @xmath1420% . therefore we do not consider surface albedo variations important for this work . [ fig : albedo],width=302 ] . the black dashed lines represent the surface pressures and top of atmosphere pressures . we use the modern earth temperature - altitude profile in all cases and calculate the pressures assuming hydrostatic equilibrium . the remainder of the atmosphere is n@xmath0 for all cases . [ fig : profiles],width=623 ] m dimer feature is strong only in the spectra corresponding to atmospheric pressures greater than @xmath140.5 bar . the spectra for atmospheres with pressure @xmath371 bar are nearly identical except for an offset because there is a fundamental limit on which heights in an atmosphere can be probed using transmission spectroscopy . for an earth - like atmosphere around the m5v used here , only the top 0.9 bars can be probed for all atmospheres at all wavelengths . this pressure corresponds to an altitude of @xmath141 km above the surface for a 1 bar atmosphere , and 19 km above the surface for a 10 bar atmosphere . the offset is due to the flux blocked by layers in the atmosphere with pressure greater than 0.9 bars.[fig : plottran],width=302 ] but with different total atmospheric pressures . in the reflected spectrum there is no fundamental limit to which pressures can be probed in an atmosphere . assuming a cloud - free case , it is possible to probe the surface layers of the atmosphere . the 1.06 @xmath1 m dimer feature is fairly weak in the present day earth s atmosphere , but it is a very strong feature in atmospheres with greater pressures . the 1.27 @xmath1 m feature is very strong in most of the spectra . [ fig : plotrad],width=302 ] but for an o@xmath0 concentration of 10% pal . for this amount of o@xmath0 , the 1.06 @xmath1 m dimer feature is very weak . however , the 1.27 @xmath1 m feature is still quite strong.[fig : plotrad01],width=302 ] a band and the 1.06 @xmath1 m dimer feature at various pressures and o@xmath0 concentrations . in transmission , for the earth - like planet orbiting an m5v star considered here , only the top 0.9 bars can be probed , meaning the dimer and a band equivalent widths are roughly constant with pressure above 1.0 bars for a given o@xmath0 concentration . in comparison to the reflected spectra , the dimer equivalent widths are extremely sensitive to pressure . [ fig : equivwidths],width=302 ] partial pressure at maximum tangent pressure vs. the ratio of the dimer feature and o@xmath0 a band flux difference ratio . the ratio can be used to determine the o@xmath0 partial pressure . * ( b ) * fo@xmath0 ( or o@xmath0 mixing ratio ) vs a band flux difference . the a band flux difference is roughly constant for a given o@xmath0 mixing ratio . [ fig : pressure - tran - only],width=623 ] vs the ratio of the 1.06 @xmath1 m dimer feature equivalent width to the o@xmath0 a band equivalent width for the reflected spectrum . there is a trend between this ratio and the total amount of o@xmath0 . this could be used as a way to estimate the pressure if a transmission spectrum is not available . [ fig : refl - only - toto2],width=302 ] ( blue ) and an isothermal atmosphere at 250k with the same pressure - composition profiles ( green ) . in both cases , the spectra for the calculated temperature - pressure profile and the isothermal profile are very similar , showing that the spectra are not strongly dependent on the temperature profile . , width=302 ] earth analog with h@xmath0o concentrations varying from 0.1 to 10.0x pal h@xmath0o . while h@xmath0o absorbs near the wings of the o@xmath0 a band the ppm flux difference and the snr do not change greatly as the h@xmath0o concentration changes . the continuum near the 1.06 @xmath1 m dimer feature is strongly affected by the increases in h@xmath0o . however , the magnitude of the change in the ppm flux difference over the entire band and the snr is less than 20% when comparing the 1.0x pal h@xmath0o case to either the 0.1 or 10.0x h@xmath0o cases . [ fig : h2o - test-1],width=302 ] m dimer feature in transit transmission with different amounts of h@xmath0o . the spectra have been artificially offset for ease of viewing . the black dashed lines are spectra that do not include any o@xmath0 dimer absorption and are included to help show the continuum flux level . the total ppm flux difference and snr for the dimer band vary by less than 20% with respect to the 1.0xh@xmath0o case.[fig : h2o - test-2 ] , width=302 ] m dimer feature in the direct beam reflected spectra with varying amounts of h@xmath0o for an earth analog with a 1.0 bar , 2.0x pal o@xmath0 atmosphere . there is an artificial offset for ease of viewing , and the black dashed lines are spectra that do not include any o@xmath0 dimer absorption . as in transit transmission , the snr of the dimer feature varies by less than 20% with respect to the 1.0x h@xmath0o case . additionally , the snrs for higher pressure cases should be less affected by changes in h@xmath0o concentrations because of greater absorption within the dimer band.[fig : h2o - test-3],width=302 ] case at various spectral resolving powers . the wavelengths for the o@xmath0 a band , the 1.06 @xmath1 m dimer band , and the 1.27 @xmath1 m band are highlighted . an artificial offset has been added to the spectra for ease of viewing . at the lowest resolving powers , it is difficult to identify spectral features . [ fig : specres1],width=302 ] . the snrs decrease as resolving power decreases because the absorption bands can no longer be resolved from the continuum . for r=30 and r=20 it is very difficult to identify any of the o@xmath0 spectral features . [ fig : specres2],width=302 ] a band , 1.06 @xmath1 m dimer feature and 1.27 @xmath1 m feature are highlighted . at resolving powers of r=30 and r=20 it is difficult to identify any o@xmath0 absorption features.[fig : specres - imaging ] , width=302 ] a band , 1.06 @xmath1 m dimer feature and the 1.27 @xmath1 m feature for direct imaging reflected spectra . as resolving power decreases , the required snr to detect each feature increases because the continuum level decreases , resulting in a lower overall signal in the absorption band . furthermore , at the lowest resolutions the absorption bands can not be resolved . thus , this decreases the measurable signal even more , leading to an increase in the required snr . [ fig : specres - imaging - snrd ] , width=302 ] in the center of the absorption band ( defined as the lowest flux level ) for the o@xmath0 a band , 1.06 @xmath1 m dimer feature and the 1.27 @xmath1 m feature . the snr required for this level of precision decreases as resolving power decreases because the flux is averaged over fewer wavelength bins , leading to an increase in the lowest flux . this decrease in required resolving power occurs for all three absorption features , though the trend is most apparent for the o@xmath0 a band , as it has the narrowest spectral shape of the three features considered here . [ fig : specres - imaging - snrp ] , width=302 ]

in this paper a length - conserving numerical scheme for a nonlinear fourth order system of partial differential algebraic equations arising in technical textile industry is studied . applying a semidiscretization in time , the resulting sequence of nonlinear elliptic systems with algebraic constraint is reformulated as constrained optimization problems in a hilbert space setting that admit a solution at each time level . stability and convergence of the scheme are proved . the numerical realization is performed by projected gradient methods on finite element spaces which determine the computational effort and approximation quality of the algorithm . simulation results are presented and discussed in view of the application of an elastic inextensible fiber motion . + ams - classification 35j74 ; 58j05 ; 65k10 ; 65m12 ; 65m20 ; 65m60 ; 74k10 + keywords numerical scheme ; stability ; convergence ; semidiscretization ; constrained optimization ; finite elements ; elastic fiber dynamics

1203.3692

the numerical simulation and optimization of the dynamics of thin long elastic fibers are of great importance in the technical textile industry , for example in production processes of yarns or non - woven materials @xcite . in the slender - body theory a fiber can be asymptotically described by a parameterized , time - dependent curve that might represent the center - line . then the fiber dynamics can be modeled by a nonlinear partial differential algebraic system ( pdaes ) @xcite . the partial differential equation of fourth order for the curve has a wavelike character with elliptic regularization . it can be considered as a reformulation of the kirchhoff - love equations for an elastic beam @xcite . the nonlinearity enters the system by the attachment of an algebraic constraint that incorporates inextensibility into the model and turns the inner traction to an unknown of the system , i.e. lagrange multiplier . depending on the application , also nonlinear or even stochastic source terms might play a role , see @xcite and figure [ fig : intro ] . the studies of fiber lay - down processes and longtime behavior require a fast , accurate numerical treatment , cf . so far , the used approaches were mainly addressed to high - speed performance without any theoretical results on convergence or length conservation . this work aims at the development of an appropriate numerical scheme for the pdaes with focus on analytical and computational aspects . we propose a semi - discretization in time . following the concept of @xcite and employing a horizontal line method , we replace the transient problem by a sequence of elliptic systems that are handled in their weak formulation in terms of the lagrange formalism . the algebraic constraint is incorporated in the definition of the optimization domain such that we study the solvability of a constrained minimization problem in a hilbert space setting @xcite . we prove the existence of the minimizer and of the lagrange multiplier on each time level . stability estimates on the discrete solution and the lagrange multiplier result then in the convergence of the numerical scheme . in addition , we derive an error bound on the fiber elongation . numerically , we solve the optimization problems in finite element spaces by applying a projected gradient method with armijo step size rule . thereby , the finite dimensional approximation of the constraint and thus the choice of the projection mapping determine the efficiency and accuracy of the scheme . the paper is structured as follows . we start with the presentation of the model equations in section [ sec:2 ] . after the introduction of the numerical scheme we deal with its theoretical analysis regarding existence , stability and convergence in section [ sec:3 ] . the numerical realization is discussed in view of computational effort and approximation quality in section [ sec:4 ] . the simulation results are in accordance with the analytical investigations and satisfy the predicted error estimates . we conclude with a summary and an outlook . application : melt - spinning process of non - woven materials . from left to right : turbulent air flow in process ( photo by industrial partner ) , mean velocity flow field , turbulence effects on immersed fiber modeled by stochastic forces ( source terms ) in fiber pdaes , @xcite . ] due to its slender geometry , a fiber can be asymptotically represented by its arclength - parameterized time - dependent center - line @xmath0 according to the special cosserat theory @xcite , where @xmath1 , @xmath2 with fiber length @xmath3 and end time @xmath4 . since extension and shear are here negligibly small in comparison to bending , the dynamics of an homogenous fiber can be described by a nonlinear partial differential equation of fourth order with an algebraic constraint for the conservation of length @xcite [ eq:1 ] @xmath5 denotes the line weight . the dynamics is caused by acting outer and inner forces ( newton s law ) . the outer force densities @xmath6 might come for example from gravity and/or aerodynamics regarding the application . the inner force densities stem from traction @xmath7 and bending with bending stiffness @xmath8 . the inner traction @xmath9 might be interpreted as lagrange multiplier to the algebraic constraint that is expressed in the euclidian norm @xmath10 . in addition , torsion could be included , yielding an extra term @xmath11 in with @xmath12 , @xcite . its neglect ( @xmath13 ) is here justified by the set - up that is characterized by a free fiber ending . in particular , we consider a fiber fixed at one ending @xmath14 that is freely swinging . this set - up is clearly a simplification to real applications in technical textile industry , but it still contains the major mathematical difficulty , i.e. the partial differential - algebraic structure of the model equations . then , dirichlet and neumann boundary conditions for the fixed and stress - free fiber ending respectively as well as appropriate initial conditions close to an initial boundary value problem @xmath15 with unit vector @xmath16 prescribing the direction of gravity . the fiber model is a wavelike system with elliptic regularization . it is a reformulation of the kirchhoff - love equations for an elastic beam @xcite . the mathematical challenge lies in the treatment of the algebraic constraint of inextensibility . this paper aims at the derivation and investigation of a numerical scheme for the partial differential algebraic system . for simplicity , we restrict here to sufficiently smooth outer forces that are independent of the fiber position , @xmath17 , like gravity . however , note that in melt - spinning processes of non - woven materials , even stochastic forces from turbulent air flows become important @xcite , cf . figure [ fig : intro ] . for a study on stochastic extensible beam equations see e.g. @xcite . in this section we derive an implicit semi - discretization for the partial differential algebraic fiber system . following the concept of @xcite and employing a horizontal line method ( rothe method ) in time , we replace the transient problem by a sequence of elliptic systems that are handled in their weak formulation in terms of the lagrange formalism . hereby , we incorporate the algebraic constraint in the definition of the optimization domain such that we study the solvability of a constrained minimization problem . in particular , we show the existence of the minimizer and of the lagrange multiplier on each time level . stability estimates on the discrete solution and the lagrange multiplier result then in a convergence proof for the numerical scheme . let @xmath18 be given . we divide the time interval @xmath19 $ ] into @xmath20 subintervals by introducing the temporal mesh @xmath21 where @xmath22 is prescribed by the time step @xmath23 . instead of the taken equidistant discretization one can also think of a varying time step @xmath24 and define the maximal subinterval length @xmath25 . then the partition is assumed to satisfy @xmath26 as @xmath27 . using an implicit euler scheme , we discretize the fiber system as [ eq:2 ] @xmath28 , @xmath29 , @xmath30 and @xmath31 . the implicit time discretization requires consequently the recursive solving of nonlinear elliptic systems in one space dimension . the approximate solution to is then given by the linear interpolation @xmath32 , i.e. , @xmath33 , \quad k=1 , \ldots , n , \end{aligned}\ ] ] and correspondingly for @xmath34 , where @xmath35 . for functions defined on @xmath19 $ ] , in turn , a subindex @xmath36 corresponds to the value of the function at time @xmath37 . the discretized system can be identified as euler - lagrange equations corresponding to the appropriate lagrange functional such that we study its solvability as variational problem in the following . the norm of a banach space @xmath38 we denote by @xmath39 and the dual pairing with its dual space @xmath40 by @xmath41 . if @xmath38 even is a hilbert space , then its inner product we denote by @xmath42 . by @xmath43 , @xmath44 lebesgue measurable , @xmath45 , we denote the hilbert space of ( equivalence classes of ) square integrable functions on @xmath46 w.r.t . the lebesgue measure taking values in @xmath47 . the space @xmath48 of continuous functions on compact @xmath46 with values in @xmath38 we consider to be equipped with the norm of uniform convergence . we use the notation @xmath49 , @xmath50 open , @xmath51 , @xmath52 $ ] , for sobolev spaces as in @xcite . in case of @xmath53 , the hilbert space @xmath54 is abbreviated by @xmath55 . in the case @xmath56 we suppress the range of function spaces . in particular , we introduce the notation @xmath57 @xmath58 . of course , @xmath59 equipped with the norm of @xmath60 is a hilbert space . its dual space @xmath61 we denote by @xmath62 . recall that @xmath60 is embedded continuously and compactly in the hlder spaces @xmath63;{\mathbb r}^n)$ ] for @xmath64 , @xmath65 , @xmath66 , see e.g. @xcite . we always , via the riesz representation theorem , identify spaces of square integrable functions with their dual space and consider an embedding of sobolev spaces in the sense of gelfand triples with the space of square integrable functions as central space . then , we define the affine linear fiber space @xmath67 we consider the constraint of inextensibility as @xmath68 and express it in terms of @xmath69 moreover , we deduce the cost functionals @xmath70 @xmath71 with the second temporal difference @xmath72 by applying variational calculus on for @xmath73 . + * lagrange formalism . * _ for @xmath73 , let @xmath74 be the cost functional of and @xmath75 the constraint functional of . define the lagrange functional @xmath76 by _ @xmath77 _ then , the minimizer of the lagrange functional is a weak solution of the fiber system . _ + the minimizer of the lagrange functional satisfies the adjoint problem for all test functions @xmath78 and @xmath79 , i.e. [ eq : ad ] @xmath80 = 0 & = { } _ { { h}_{0,l}^1(\omega_l)}{\!\langle e_{k+1}({\mathbf v}),\eta\rangle}_{{h}^{-1}(\omega_l)}\\\label{eq : adj2 } \nabla_{\mathbf v } l_{k+1}({\mathbf v},\lambda ) [ \boldsymbol{\phi } ] = 0 & = j_{k+1}'({\mathbf v})[\boldsymbol { \phi } ] + { } _ { { h}_{0,l}^1(\omega_l)}{\!\langle e_{k+1}'({\mathbf v})[\boldsymbol{\phi}],\lambda\rangle}_{{h}^{-1}(\omega_l ) } \\ \nonumber & = 2 \big ( \omega \ , ( { \mathrm d}_{k+1}^2 { \mathbf v},\boldsymbol{\phi } ) _ { { l}^2(\omega_l ) } + b \ , ( \partial_{ss}{\mathbf v } , \partial_{ss } \boldsymbol{\phi})_{{l}^2(\omega_l ) } -({\mathbf f}_{k+1 } , \boldsymbol{\phi})_{{l}^2(\omega_l ) } \\\nonumber & \quad \quad + { } _ { { h}_{0,l}^1(\omega_l)}{\!\langle \partial_{s}{\mathbf r}_k \cdot\partial_{s}\boldsymbol{\phi},\lambda\rangle}_{{h}^{-1}(\omega_l)}\big).\end{aligned}\ ] ] presupposing sufficient regularity of the lagrange multiplier @xmath7 , the duality pairing @xmath81 coincides with @xmath82 in the sense of a gelfand triple . this yields the euler - lagrange equations corresponding to the fiber system . hence , the weak solvability of the elliptic fiber system can be formulated as + * constrained minimization problem * @xmath83 [ lem:1 ] for @xmath73 , the cost functional @xmath84 defined in is strictly convex , coercive and weakly lower semi - continuous . the minimization domain @xmath85 is closed and convex and , in particular , weakly closed . here and throughout the following proofs where is no danger of confusion , we suppress the indices indicating the time levels for a simpler notation . let @xmath86 , @xmath87 , @xmath88 , @xmath89 . then , the strict convexity of @xmath90 is concluded from @xmath91 since @xmath92 . due to the assumed boundary conditions a poincar@xmath93 inequality holds and we obtain @xmath94 for some @xmath95 . hence @xmath96 thus , @xmath97 , if @xmath98 for fixed @xmath99 , i.e. , @xmath90 is coercive . let @xmath100 be a sequence in @xmath101 that converges weakly to @xmath87 in @xmath102 , i.e. , @xmath103 for @xmath104 . then , in particular , @xmath105 and @xmath106 for @xmath104 . since the norm is lower semi - continuous w.r.t . weak convergence and the inner product with @xmath99 is continuous w.r.t . weak convergence , we obtain @xmath107 , i.e. , @xmath90 is weakly lower semi - continuous . the convexity of @xmath108 results from the affine linearity of @xmath109 and the convexity of the set @xmath110 . let @xmath86 , @xmath111 , @xmath112 $ ] , then @xmath113 , since @xmath114 since @xmath115 is closed , @xmath109 is continuous and the set @xmath110 is closed , also @xmath108 is closed . this , together with convexity , implies that @xmath108 is also weakly closed . [ theo : mini ] the constrained minimization problem has a unique solution @xmath116 on every time level . according to lemma [ lem:1 ] the cost functional @xmath90 on every time level is coercive and weakly lower semi - continuous and the domain @xmath108 is convex and weakly closed . these are the requirements for a general existence and uniqueness result for constrained minimization problems , see e.g. @xcite . we state the proof here for completeness . choose a minimizing sequence @xmath100 , @xmath117 , with @xmath118 for @xmath104 . then @xmath119 and @xmath100 is bounded in view of the coercivity of @xmath90 . hence , there exists a subset @xmath120 and @xmath121 such that @xmath122 for @xmath123 . since @xmath108 is weakly closed , @xmath124 . the weak lower semi - continuity of @xmath90 implies @xmath125 whence @xmath126 is a minimizer . since @xmath108 is convex , the strict convexity of @xmath90 on @xmath101 implies the uniqueness of the minimizer . assume @xmath127 to be two minimizers , @xmath88 with @xmath128 . then @xmath129 for @xmath89 . since @xmath130 for @xmath89 , this contradicts the assumption . note that the uniqueness of the minimizer is meaningless for the solvability statement of the fiber system , since the unique minimizer need not necessarily be the only solution in view of possibly existing saddle points . since @xmath131 and @xmath132 for all @xmath73 , the following lemma follows immediately by the cauchy schwarz inequality . [ le : increase ] the relation @xmath133 holds for all @xmath134 , @xmath135 . [ th : exlm ] for @xmath73 the linearized constraint functional @xmath136 is surjective . we have @xmath137 = 2 \partial_s{\mathbf r}_k \cdot \partial_s \boldsymbol{\phi } , \quad \boldsymbol{\phi}\in { h}^2_{0,l}(\omega_l).\end{aligned}\ ] ] let @xmath138 be arbitrary . set @xmath139 note that @xmath140)$ ] hence , together with lemma [ le : increase ] , we obtain @xmath141 with @xmath142 . furthermore @xmath143 and therefore @xmath144.\end{aligned}\ ] ] moreover @xmath145 with @xmath146 . finally , @xmath147 can be concluded from @xmath148 , @xmath149 , @xmath150 and @xmath151 using the chain rule . this shows @xmath152 . [ rm : inverse ] note that @xmath153 is not injective on @xmath154 . nevertheless we denote the mapping @xmath155 in by @xmath156 , because @xmath157 is the identity on @xmath158 . of course , @xmath159 by the inverse mapping theorem ( applied in the proper quotient space setting ) . [ th : exist ] for @xmath73 let @xmath160 be the minimizer of @xmath74 on @xmath85 , provided in theorem [ theo : mini ] , and @xmath161 then @xmath162 ( see remark [ rm : inverse ] ) , and @xmath163 are solving weakly the discrete fiber system , i.e. , [ eq : exist ] @xmath164 for all test functions @xmath165 . by definition , all elements from @xmath85 fulfill . furthermore , since @xmath166 minimizes @xmath74 on @xmath85 and @xmath167 = 0 \quad \mbox { for all } \boldsymbol{\phi}\in { h}^2_{0,l}(\omega_l;\mathbb{r}^3),\end{aligned}\ ] ] we have @xmath168\\ & = 2 \ , \big ( \omega \ , ( { \mathrm d}_{k+1}^2 { \mathbf r}_{k+1 } , \boldsymbol{\phi } ) _ { { l}^2(\omega_l ) } + b \ , ( \partial_{ss}{\mathbf r}_{k+1 } , \partial_{ss } \boldsymbol{\phi } ) _ { { l}^2(\omega_l ) } - ( { \mathbf f}_{k+1 } , \boldsymbol{\phi } ) _ { { l}^2(\omega_l ) } \\ & \quad \quad + { } _ { { h}_{0,l}^1(\omega_l)}{\!\langle \partial_{s}{\bf r}_k \cdot\partial_{s}\boldsymbol{\phi},\lambda_{k+1}\rangle}_{{h}^{-1}(\omega_l)}\big ) & & \mbox { for all } \boldsymbol{\phi}\in { h}^2_{0,l}(\omega_l;\mathbb{r}^3).\end{aligned}\ ] ] in the following we provide stability estimates for the discrete solution . for function spaces @xmath169 and @xmath170 on sets @xmath171 and @xmath171 , respectively , we define as usual @xmath172 , where @xmath173 , @xmath174 , @xmath175 ( algebraic tensor product ) . the sobolev space @xmath176 on @xmath177 then is defined as the completion of @xmath178 w.r.t . the metric associated to its inner product , see e.g. ( * ? ? ? ii.4 ) ( i.e. , it is the sobolev space of functions on @xmath179 which are twice weakly differentiable in the first variable and once weakly differentiable in the second variable and square integrable on @xmath179 together with their derivatives ) . correspondingly , we set @xmath180 , @xmath181 , to be the completion of @xmath182 and use the notation @xmath183 . in the case @xmath184 we suppress the index @xmath185 . for functions @xmath186 defined on @xmath187 we use the following notation for discrete derivatives : @xmath188 [ stabestir ] let @xmath189 be as in theorem [ th : exist ] , @xmath73 , and let @xmath190 be the corresponding linear interpolation . then there exists @xmath191 , independent of @xmath192 ( or , equivalently , the time discretization @xmath193 ) , such that @xmath194 since @xmath195 for all @xmath73 and @xmath196 is piecewise linear for all @xmath134 , we have @xmath190 . we know that @xmath197 for all @xmath165 . note that the first summand on the right - hand side is discretized in an explicit way . since @xmath198 , the special choice @xmath199 results in @xmath200 by applying the identity @xmath201 and the functional constraint @xmath202 . hence , we obtain @xmath203 summing up @xmath204 gives the following crucial relation @xmath205 ( note that @xmath206 , as well as , @xmath207 ) . we estimate the scalar product on the right - hand side by cauchy schwarz and young s inequality , i.e. @xmath208 , and find @xmath209 the discrete gronwall lemma implies @xmath210 together with @xmath211 yields the existence of @xmath212 , independent of @xmath192 , such that @xmath213 combining and gives finally the existence of @xmath214 , independent of @xmath192 , such that @xmath215 the inequalities , together with the poincar@xmath93 inequality guarantee the existence of the desired @xmath191 in , independent of @xmath192 . before providing the stability bound for the lagrange multiplier , we need to establish some bounds on the inverse of the linearized constraint functional . let @xmath159 be as in remark [ rm : inverse ] , @xmath73 . then there exists @xmath216 , independent of @xmath192 , such that @xmath217\|_{h^2(\omega_l ) } \le l \ , \|g\|_{h^1(\omega_l ) } , \quad \quad \quad \quad \|(e'_{k+1})^{-1}[g]\|_{l^2(\omega_l ) } \le l \ , \|g\|_{l^2(\omega_l ) } , \\ \left\|\big({\mathrm d}\big((e')^{-1}\big)^\tau\big)_{k+1}[g ] \right\|_{{l}^2(\omega_l ) } \le l \ , \|g\|_{h^1(\omega_l)},\end{aligned}\ ] ] for all @xmath218 . let @xmath219 . as consequence of the poincar@xmath93 inequality , lemma [ le : increase ] and , we find constants @xmath220 ( independent of @xmath192 ) such that @xmath221\|_{{h}^2(\omega_l ) } & \le l_1 \left\| \partial_s \left(\frac{g \,\partial_s { \mathbf r}_k}{|\partial_s{\mathbf r}_k|^2}\right ) \right\|_{{l}^2(\omega_l ) } \le l_1 \left(\|\partial_s g \,\partial_s { \mathbf r}_k \|_{{l}^2(\omega_l ) } + 3\| g \,\partial_{ss } { \mathbf r}_k \|_{{l}^2(\omega_l)}\right ) \\ & \le l_2 \left(\|\partial_s { \mathbf r}_k \|_{c^0(\overline{\omega_l } ) } + 3\|\partial_{ss } { \mathbf r}_k \|_{{l}^2(\omega_l)}\right ) \|g\|_{h^1(\omega_l ) } \le l_3 \|g\|_{h^1(\omega_l)}.\end{aligned}\ ] ] using cauchy schwarz inequality and lemma [ le : increase ] , we obtain @xmath222\|^2_{{l}^2(\omega_l ) } = \int_0^l \left|\int_u^l \frac{g \,\partial_s { \mathbf r}_k}{2|\partial_s{\mathbf r}_k|^2 } \,ds \right|^2 du \le \frac{3l^2}{4 } \int_0^l \frac{g^2 |\partial_s { \mathbf r}_k|^2}{|\partial_s{\mathbf r}_k|^4 } \,ds \le \frac{3l^2}{4 } \|g\|^2_{l^2(\omega_l)}.\ ] ] the estimation of the discrete derivative is a little more lengthly . integration by parts yields @xmath223 \big\|^2_{{l}^2(\omega_l ) } = \int_0^l \left|\int_u^l \frac{g \,\partial_s { \mathbf r}_{k}}{2\tau|\partial_s{\mathbf r}_{k}|^2 } - \frac{g \,\partial_s { \mathbf r}_{k-1}}{2\tau|\partial_s{\mathbf r}_{k-1}|^2 } \,ds \right|^2 du \\ & \le \int_0^l \left|\int_u^l \frac{g \,\partial_s({\mathbf r}_{k}-{\mathbf r}_{k-1 } ) } { \tau|\partial_s{\mathbf r}_{k}|^2 } \,ds \right|^2 du + \int_0^l \left|\int_u^l g \,\partial_s{\mathbf r}_{k-1 } \left(\frac{1}{\tau|\partial_s{\mathbf r}_{k}|^2 } - \frac{1}{\tau|\partial_s{\mathbf r}_{k-1}|^2}\right ) \,ds \right|^2 du \\ & \le 2\int_0^l \left|\int_u^l \frac{{\mathbf r}_{k}-{\bf r}_{k-1}}{\tau}\,\,\partial_s\!\left(\frac{g}{|\partial_s{\mathbf r}_{k}|^2}\right)\,ds \right|^2 du + 2\int_0^l \left| \frac{{\mathbf r}_{k}-{\mathbf r}_{k-1 } } { \tau } \frac{g}{|\partial_s{\mathbf r}_{k}|^2 } \right|^2 ds \\ & \quad + \int_0^l \left|\int_u^l \frac{g \,\partial_s{\mathbf r}_{k-1 } } { \tau|\partial_s{\mathbf r}_{k}|^2|\partial_s{\mathbf r}_{k-1}|^2 } \,\,\partial_s({\mathbf r}_{k}-{\mathbf r}_{k-1})\cdot\partial_s({\mathbf r}_{k}+{\mathbf r}_{k-1 } ) \,ds \right|^2 du \\ & \le l_4 \,\|g\|^2_{h^1(\omega_l ) } + 2\int_0^l \left|\int_u^l \frac{{\mathbf r}_{k } -{\mathbf r}_{k-1}}{\tau } \cdot \partial_s\!\left(\partial_s({\mathbf r}_{k}+{\mathbf r}_{k-1})\frac{g \,\partial_s{\mathbf r}_{k-1 } } { |\partial_s{\mathbf r}_{k}|^2|\partial_s{\mathbf r}_{k-1}|^2}\right ) \,ds \right|^2 du \\ & \quad + 2\int_0^l \left| \frac{{\mathbf r}_{k}-{\mathbf r}_{k-1}}{\tau}\cdot\partial_s({\mathbf r}_{k}+{\mathbf r}_{k-1 } ) \frac{g \,\partial_s{\mathbf r}_{k-1}}{|\partial_s{\mathbf r}_{k}|^2|\partial_s{\mathbf r}_{k-1}|^2 } \right|^2 ds\\ & \le l_5 \,\|g\|^2_{h^1(\omega_l)}\end{aligned}\ ] ] for some @xmath224 ( independent of @xmath192 ) . here , again , we used lemma [ le : increase ] and the estimates provided in proposition [ stabestir ] in several steps . let @xmath225 be as in , @xmath73 , and let @xmath226 be the corresponding linear interpolation . then @xmath227 and there exists @xmath228 , independent of @xmath192 , such that @xmath229 for all @xmath218 , @xmath230 . according to , the lagrange multiplier is given by @xmath231 let @xmath219 and @xmath232 . then , due to the imposed initial and boundary conditions @xmath233{0pt}{0pt}_{{{{h}^{1}_{0,l}(\omega_l)}}}{\!\left\langle g , \frac{t - t_{k}}{\tau } ( { \lambda}_{k+1}-{\lambda}_{k } ) + { \lambda}_{k } \right\rangle}_{{h}^{-1}(\omega_l ) } h(t ) \,dt \\\nonumber & = \sum_{k=1}^{n-1 } { } _ { { h}^{1}_{0,l}(\omega_l)}{\!\langle g,{\lambda}_{k+1 } \rangle}_{{h}^{-1}(\omega_l ) } h^{(-1)}_{k+1 } -{}_{{h}^{1}_{0,l}(\omega_l)}{\!\langle g,{\lambda}_{k } \rangle}_{{h}^{-1}(\omega_l ) } h^{(-1)}_k \\\nonumber & \quad - \left ( { } _ { { h}^{1}_{0,l}(\omega_l)}{\!\langle g,{\lambda}_{k+1 } \rangle}_{{h}^{-1}(\omega_l ) } -{}_{{h}^{1}_{0,l}(\omega_l)}{\!\langle g,{\lambda}_{k } \rangle}_{{h}^{-1}(\omega_l ) } \right ) \frac{1}{\tau}\int_{t_k}^{t_{k+1 } } h^{(-1)}(t ) \,dt \\ & = \tau \sum_{k=2}^{n-1 } { } _ { { h}^{1}_{0,l}(\omega_l)}{\!\langle g,{\lambda}_{k } \rangle}_{{h}^{-1}(\omega_l ) } \big({\rm d}^2h^{(-2)}\big)_{k+1 } - { } _ { { h}^{1}_{0,l}(\omega_l)}{\!\langle g,{\lambda}_{n } \rangle}_{{h}^{-1}(\omega_l ) } \big({\rm d}h^{(-2)}\big)_{n},\end{aligned}\ ] ] where @xmath234 is the primitive function of @xmath235 with @xmath236 , @xmath237 , @xmath238 . furthermore , @xmath239 \\ \nonumber & = - 2 \big(\omega \big(({\rm d}^2{\bf r}^\tau)_{k+1 } , ( e'_{k+1})^{-1}[g ] \big)_{{l}^2(\omega_l ) } + b \big(\partial_{ss}{\bf r}_{k+1 } , \partial_{ss}((e'_{k+1})^{-1}[g ] ) \big)_{{l}^2(\omega_l)}\\ & \quad \quad \quad - \big({\bf f}_{k+1 } , ( e'_{k+1})^{-1}[g ] \big)_{{l}^2(\omega_l)}\big).\end{aligned}\ ] ] using and , now we estimate @xmath240 term by term . first we consider @xmath241 \bigg)_{{l}^2(\omega_l ) } \big({\rm d}^2h^{(-2)}\big)_{k+1 } \bigg| \\ & \le \left|\tau \sum_{k=2}^{n-2 } \left ( ( { \rm d}{\bf r}^\tau)_{k } , \big({\rm d}\big((e')^{-1}\big)^\tau\big)_{k+1}[g ] \right)_{{l}^2(\omega_l ) } \big({\rm d}^2h^{(-2)}\big)_{k+1 } \right| \\ & \quad + \left|\tau \sum_{k=2}^{n-2 } \left ( ( { \rm d}{\bf r}^\tau)_{k } , ( e'_{k+1})^{-1}[g ] \right)_{{l}^2(\omega_l ) } \big({\rm d}^3h^{(-2)}\big)_{k+2 } \right| \\ & \quad + \left|\left ( ( { \rm d}{\bf r}^\tau)_{n-1 } , ( e'_{n-1})^{-1}[g ] \right)_{{l}^2(\omega_l ) } \big({\rm d}^2h^{(-2)}\big)_{n } \right| \\ & \le kl \|g\|_{h^1(\omega_l ) } \,\tau \sum_{k=2}^{n-1 } \left(\left|\big({\rm d}^2h^{(-2)}\big)_{k+1 } \right| + \left|\big({\rm d}^3h^{(-2)}\big)_{k+1 } \right| \right ) \\ & \le 5\sqrt{t}kl \|g\|_{h^1(\omega_l ) } \|h\|_{h^1(\omega_t ) } \end{aligned}\ ] ] where we used several times and . since @xmath242 , we have @xmath243 \right)_{{l}^2(\omega_l ) } \big({\mathrm d}h^{(-2)}\big)_{n } \right| \le c_1 \|g\|_{h^1(\omega_l ) } \|h\|_{h^1(\omega_t)}\ ] ] for some @xmath244 , independent of @xmath192 . using , and the continuity of @xmath245 , a derivation of an appropriate bound for the remaining four terms is straight forward . [ th : onconv ] there exists a sequence of discretizations @xmath246 , @xmath247 and @xmath248 such that [ eq : conv ] @xmath249 ; { l}^2(\omega_l ; { \mathbb r}^3)),\label{eq : conv1 } \\ \lim_{n\to \infty } { \mathbf r}^{\tau_n } & = { \mathbf r } \quad \mbox { weakly in } \quad { h}^{2,1}(\omega ; { \mathbb r}^3),\label{eq : conv2}\\ \lim_{n\to \infty } { \mathbf r}^{\tau_n } & = { \mathbf r } \quad \mbox { strongly in } \quad { l}^2(\omega ; { \mathbb r}^3),\label{eq : conv3 } \\ \lim_{n\to \infty } { \lambda}^{\tau_n } & = { \lambda } \quad \mbox { strongly in } \quad { h}^{-\beta}(\omega),\label{eq : conv4}\end{aligned}\ ] ] for all @xmath250 . furthermore , @xmath251 are weakly solving , i.e. , [ eq : existcont ] @xmath252 for all @xmath253 and all @xmath254 . furthermore , for all @xmath255 @xmath256 ; { \mathbb r}^3 ) , \\ \lim_{n\to \infty } \partial_s{\bf r}^{\tau_n}(t ) & = \partial_s{\bf r}(t ) \quad \mbox { in } \quad c^{0 , \gamma}([0,l ] ; { \mathbb r}^3),\end{aligned}\ ] ] for all @xmath257 , where @xmath258 $ ] is countable ( in the following we are choosing @xmath258 $ ] dense and let @xmath259 ) . moreover , @xmath126 has a ( unique ) continuous version ( denoted by the same symbol ) and @xmath260 \times [ 0,t],\end{aligned}\ ] ] and even @xmath261 ( i ) theorem [ th : onconv ] provides existence of a weak solution to the nonlinear fourth order partial differential system with algebraic constraint given in . taking into account the regularity of the weak solution , the dirichlet and neumann boundary conditions in are fulfilled as far as possible in this situation . \(ii ) under the assumption of uniqueness of a weak solution as in theorem [ th : onconv ] , one has convergence even for all discretizations @xmath246 . \(iii ) the pairing @xmath262 in has to be understood in the sense of a uniquely closable bilinear form . i.e. , if @xmath263 converges weakly to @xmath264 in @xmath265 , @xmath266 is a sequence in @xmath267 which converges strongly to @xmath268 in @xmath269 , @xmath270 exists and @xmath271 or @xmath272 , then @xmath273 . from the first two estimates in we can conclude that @xmath274 is uniformly ( in @xmath192 ) lipschitz continuous in @xmath275 ; { l}^2(\omega_l ; { \mathbb r}^3))$ ] and @xmath276\ } \subset \{{\mathbf v } \in { h}^2(\omega_l ; { \mathbb r}^3 ) \,|\ , \|\partial_{ss}{\mathbf v}\|_{{l}^{2}(\omega_l)}\le k \},\end{aligned}\ ] ] which is a relative compact subset of @xmath277 . thus , there exists sequence of discretizations @xmath246 and @xmath278 ; { l}^2(\omega_l ; { \mathbb r}^3))$ ] such that @xmath279 converges to @xmath126 in @xmath275 ; { l}^2(\omega_l ; { \mathbb r}^3))$ ] for @xmath280 . the third estimate in gives the existence of a subsequence @xmath246 ( denoted the same ) and @xmath281 such that @xmath279 converges weakly to @xmath282 in @xmath283 for @xmath280 . since convergence in @xmath275 ; { l}^2(\omega_l ; { \mathbb r}^3))$ ] implies strong convergence in @xmath284 as well as weak convergence in @xmath285 implies weak convergence in @xmath284 , we have @xmath286 . in particular , this shows - . from together with the fact that the embedding @xmath287 , @xmath288 , is hilbert schmidt for all @xmath289 , we obtain by the kernel theorem , see e.g. ( * ? ? ? 1,2.3 ) , that @xmath290 is uniformly ( in @xmath192 ) bounded in @xmath291 for all @xmath289 . since the embedding @xmath292 is compact for all @xmath293 , there exists a subsequence @xmath246 and @xmath248 such that @xmath294 converges strongly to @xmath295 in @xmath296 as @xmath280 for all @xmath250 . multiplying the linear interpolation of with a time - dependent test function and integrating w.r.t . time yields @xmath297 for all @xmath298 ( because @xmath35 , on @xmath299 $ ] we can assign to the function @xmath300 any value , for simplicity we choose zero ) . since @xmath301 converges weakly to @xmath302 and @xmath303 converges strongly to @xmath304 , both in @xmath305 , as @xmath280 , we have @xmath306 for all @xmath298 . furthermore , integration by parts yields @xmath307 by together with the boundary conditions imposed on @xmath308 now from it follows @xmath309 where in the last step we used that @xmath310 converges weakly to @xmath311 , @xmath312 converges strongly to @xmath313 and @xmath314 converges strongly to @xmath315 , all in @xmath316 as @xmath280 . combining , , we obtain @xmath317 for all @xmath298 . now we restrict ourself to @xmath253 . since @xmath318 converges weakly to @xmath319 in @xmath320 and @xmath321 are bounded functions , also @xmath322 converges weakly to @xmath323 in @xmath265 as @xmath280 . furthermore , @xmath324 converges strongly to @xmath7 , e.g. in @xmath269 , as @xmath280 . thus @xmath325 for all @xmath253 in the sense of a uniquely closable bilinear form , see remark [ rm : onconv](iii ) . hence , follows from together with . now , using , we obtain for all @xmath254 @xmath326 due to lemma [ le : increase ] . because @xmath327 converges strongly to @xmath126 and @xmath328 , @xmath301 converge weakly to @xmath329 , @xmath302 , respectively , in @xmath305 as @xmath280 , by an integration by parts together with we can conclude @xmath330 for all @xmath254 , i.e. , is shown . due to the second estimate in , for each @xmath331 $ ] there exists a subsequence @xmath246 ( depending on @xmath332 ) such that @xmath333 ; { \mathbb r}^3),\ ] ] for all @xmath334 . let @xmath258 $ ] be countable . then , by dropping to subsequences and taking the diagonal sequence , we obtain for all @xmath257 . here we choose @xmath258 $ ] dense with @xmath335 $ ] . from this , together with the first and second estimate in , we can conclude that @xmath336 has a ( unique ) continuous version on @xmath337 \times [ 0,t]$ ] ( which we denote by the same symbol ) . moreover , @xmath338 \times [ 0,t].\end{aligned}\ ] ] finally , together with implies . on every time level @xmath37 the semi - discretized fiber system corresponds to a constrained minimization problem in a hilbert space setting . in this section we solve the optimization problem numerically in a finite dimensional approximation space ( finite element space ) by applying a projected gradient method with armijo step size rule . the convergence of the method is ensured by the choice of a minimal projection . the temporal evolution of the fiber behavior is then realized by a recursive solving of the optimization problems where the solution @xmath339 is used as initial guess in the projected gradient method for @xmath340 . the numerical results coincide well with the previous theoretical investigations . based on the classical method of steepest descent the projected gradient method is of first order convergence . but the theoretical convergence result requires the choice of a minimal projection . however , in practice also other projection mappings might lead to satisfying results ; for details see e.g. @xcite . for the fiber system in the hilbert space setting the projected gradient method is given by algorithm [ alg : gradient ] . note that here and in the following we suppress the time index @xmath341 to facilitate the readability . [ alg : gradient ] + let @xmath342 be a minimal projection mapping onto the set @xmath343 . + define @xmath344 by @xmath345 via the cost functional @xmath90 . + initialize iteration counter @xmath346 , choose @xmath347 , set absolute and relative tolerances @xmath348 , @xmath349 . + while the stationarity measure satisfies @xmath350 , do : + - calculate @xmath351 by the projected armijo rule @xcite such that @xmath352 - set @xmath353 and update @xmath354 + end for the numerical realization we apply the projected gradient method to finite dimensional approximation spaces . therefore we introduce finite element spaces @xmath355 and @xmath356 . depending on the definition of @xmath357 we propose and investigate different projections @xmath358 . we indicate all quantities associated to the finite element discretization by the space index @xmath359 that should not be mistaken for the time index @xmath341 . considering the space interval @xmath337 $ ] , we introduce an equidistant grid @xmath360 with space step size @xmath361 . in addition , to simplify the notation we define @xmath362 and @xmath363 . certainly one can also think of an irregular grid with different @xmath364 , then the partition is assumed to satisfy @xmath365 as @xmath366 . as approximation space @xmath355 we choose the finite element space of piecewise cubic polynomials for function and first derivative . in particular @xmath367\\ ( 2 - 3\xi+\xi^3)/4 , & \xi=(2s-(s_{j+1}+s_{j}))/h , & s\in [ s_{j},s_{j+1}]\\ 0 & & else \end{array } \right.\\ \phi_j(s)&=\left\ { \begin{array}{l l l } h(-1-\xi+\xi^2+\xi^3)/8 , & \xi=(2s-(s_j+s_{j-1}))/h , & s\in [ s_{j-1},s_j]\\ h(\,\,\,1-\xi-\xi^2+\xi^3)/8 , & \xi=(2s-(s_{j+1}+s_{j}))/h , & s\in [ s_{j},s_{j+1}]\\ 0 & & else \end{array } \right.\end{aligned}\ ] ] @xmath368 form a node basis , i.e. @xmath369 then any function @xmath370 can be represented by @xmath371 via its coefficient tuple @xmath372 with @xmath373 . in the finite dimensional fiber space @xmath374 , the degrees of freedom reduce by @xmath375 due to the dirichlet boundary conditions given for @xmath376 that fix the coefficients @xmath377 and @xmath378 . identifying the function @xmath379 with its coefficient tuple @xmath380 , the cost functional of can be expressed as quadratic function in terms of the coefficients , @xmath381 the sparse symmetric matrices @xmath382 and @xmath383 have a block structure of @xmath384 , @xmath385 where each block itself consists of @xmath386 subblocks of size @xmath387 , e.g.@xmath388 with @xmath389 identity matrix , the subblocks are diagonal matrices , but they only contain non - zero entries for @xmath390 . the notation holds for all other blocks analogously . the time dependence enters @xmath391 via the temporally earlier fiber positions @xmath392 and the outer forces @xmath393 . the linear constraint @xmath394 for @xmath395 yields infinitely many independent conditions for the finite number of unknowns in the approximation space . to allow for more than the trivial solution @xmath396 , we must weaken the constraint and pose the condition only on @xmath397 space points , @xmath398 , that should be equidistantly distributed in @xmath399 . the choice of @xmath397 is a compromise between approximation quality and numerical realization ; for big @xmath397 the set @xmath357 shrinks to a single function that obviously fulfills the constraint on whole @xmath399 but does not allow for a dynamic fiber motion @xmath400 for all @xmath401 . for small @xmath397 we obtain flexibility but pay the prize of a lower approximation quality . prescribing the constraint at the spatial grid points ( @xmath402 ) is a first intuitive choice . it implies the following definition for @xmath357 : @xmath403 where @xmath404 represents the coefficient tuple to @xmath405 . [ rem : kh ] alternatively to , one can certainly use the finite element functions @xmath406 , @xmath407 to evaluate the linear constraint not only at the grid points @xmath408 , but also at interior cell points . imposing the linear constraint for example additionally on the cell midpoints @xmath409 for @xmath410 , we obtain @xmath411 and @xmath412 the restriction @xmath413 on the continuous solution in @xmath343 ( to which we refer as @xmath414-inequality ) is implicitly contained in the set @xmath357 by an appropriate choice of @xmath397 . the determination of the @xmath415-minimal projection mapping @xmath416 reduces to the solving of a linear system of equations . according to the coefficient tuples @xmath417 of @xmath418 satisfy @xmath419 with @xmath420 orthogonal complement to @xmath421 . let @xmath422 , @xmath423 be basis of @xmath420 and set @xmath424 with unknowns @xmath425 . then , the projection @xmath426 is uniquely determined by the coefficient vector @xmath427 being the solution of the linear system @xmath428 with @xmath429 and @xmath430 known due to the dirichlet boundary conditions . the basis vectors to the orthogonal complements are collected in a @xmath431-matrix that is blown up to @xmath432 by @xmath433-times row - wise replication . although , the symmetric block matrix @xmath434 has to be assembled and decomposed only once in the projected gradient method for every time step @xmath37 , the bottle - neck in the computation lies in the determination of the @xmath435 . it slows down the performance drastically . the @xmath415-seminorm induces a vector norm in @xmath436 . since in finite dimensions all norms are equivalent , one might think of using instead of the minimal projection with respect to the euclidian vector norm . the orthogonal projection is comparably easy and fast to compute : @xmath437 the dirichlet boundary conditions are consistently fulfilled . for other definitions of @xmath357 ( remark [ rem : kh ] ) corresponding projection mappings can be formulated straightforward ( see @xcite for ) . the numerical realization is performed with matlab version 7.0.1 on a intel core 2 processor , 2ghz . the results are scaled with si - units to allow for a dimensionless presentation . the numerical studies show the expected performance of the projected gradient method : the strongest decay of the cost functional @xmath391 happens within the first iterations , then it slows down until the tolerances of the break - up criterion are finally fulfilled . thereby , the projections @xmath358 and @xmath438 yield similar results : the deviation of the minimized function values of @xmath391 is of order @xmath439 . requiring the same number of iterations , the explicit projection scheme with @xmath438 is much faster than the implicit one with @xmath358 . as mentioned , the bottle - neck of the computation is the assembly of the projection matrix @xmath434 . being interested in the temporal evolution of the fiber and hence in an efficient computation of thousands of time levels via the projected gradient method we apply @xmath438 in the following simulations . fiber position at @xmath440 computed with different time steps @xmath441 , @xmath442 . acting force in case a ( left ) : @xmath443 and in case b ( right ) : @xmath444.,title="fig:",scaledwidth=55.0% ] fiber position at @xmath440 computed with different time steps @xmath441 , @xmath442 . acting force in case a ( left ) : @xmath443 and in case b ( right ) : @xmath444.,title="fig:",scaledwidth=55.0% ] & & + & & cpu - time [ s ] & @xmath445 & @xmath446 + @xmath447 & 1371.0 & 1371 & 86 & @xmath448 & @xmath449 + @xmath450 & 342.0 & 684 & 40 & @xmath451 & @xmath452 + @xmath453 & 84.3 & 337 & 19 & @xmath454 & @xmath455 + @xmath456 & 21.0 & 168 & 11 & @xmath457 & @xmath458 + @xmath459 & 3.3 & 52 & 7 & @xmath460 & @xmath461 + @xmath462 & 3.6 & 116 & 14 & @xmath461 & @xmath463 + @xmath464 & 3.8 & 244 & 34 & @xmath465 & @xmath466 + @xmath467 & 4.2 & 542 & 80 & 0 & @xmath468 + + @xmath446 + @xmath449 + @xmath469 + @xmath470 + @xmath471 + @xmath472 + @xmath473 + @xmath474 + @xmath475 + & & + & & cpu - time [ s ] & @xmath445 & @xmath446 + @xmath447 & 1374.0 & 1374 & 88 & @xmath476 & @xmath477 + @xmath450 & 356.0 & 712 & 41 & @xmath478 & @xmath479 + @xmath453 & 94.0 & 376 & 21 & @xmath480 & @xmath481 + @xmath456 & 24.0 & 192 & 12 & @xmath482 & @xmath452 + @xmath459 & 3.8 & 60 & 7 & @xmath483 & @xmath484 + @xmath462 & 4.1 & 132 & 16 & @xmath481 & @xmath485 + @xmath464 & 4.5 & 289 & 36 & @xmath486 & @xmath487 + @xmath467 & 5.1 & 648 & 92 & 0 & @xmath488 + + @xmath446 + @xmath477 + @xmath489 + @xmath490 + @xmath491 + @xmath492 + @xmath493 + @xmath494 + @xmath495 + + length change @xmath446 in dependence of time step for different sets @xmath357 : with @xmath402 is marked by @xmath496 and with @xmath411 by @xmath497 . the cases a ( solid line ) and b ( dashed - dotted line ) refer to the different applied outer forces , cf . figure [ fig:1 ] and table [ tab:1 ] . ] we consider a set - up with @xmath498 and @xmath499 for different exemplary force functions @xmath264 . the fiber parameters are @xmath500 , @xmath501 and @xmath502 . the space discretization is chosen to be @xmath503 . the projected gradient method is initialized with the solution of the previous time level , the tolerances are set to be @xmath504 , @xmath505 . accuracy and efficiency of the whole algorithm are strongly affected by the choice of the time step @xmath414 . to get a qualitative impression , we define a sequence of time steps @xmath506 , @xmath507 , @xmath508 and determine the fiber behavior for @xmath509 , see figure [ fig:1 ] and table [ tab:1 ] . in accordance to the theory , we clearly observe the convergence rate of the implicit euler time discretization as first order ; the approximation error is linear in @xmath414 which is emphasized by the marked values for @xmath510 , @xmath511 , @xmath512 in table [ tab:1 ] . considering the computational effort , the averaged number of iterations required in the projected gradient method reduces for smaller @xmath414 due to the better initial guess , whereas the total number of projected gradient runs increases . thus , moderate choices of @xmath414 ( here : @xmath513 ) yield the fewest number of iterations in total and hence the fastest computation . the fiber elongation @xmath514 is originated in lemma [ le : increase ] and reduces for smaller @xmath414 , certainly @xmath515 for @xmath516 . however , its actual magnitude is a consequence of the definition of @xmath357 . replacing the set of by the one of where the linear constraint is imposed at the double number of space points ( @xmath411 ) we obtain a better length conservation ( smaller @xmath446 ) ; in particular , @xmath446 decreases here linearly in @xmath414 , see figure [ fig:2 ] . it is worth to mention that the computational effort stays comparably the same when a respective projection @xmath438 for is used . the reason lies in the same underlying finite element space @xmath436 . fiber dynamics over time for different sets @xmath357 where the linear constraint is satisfied at spatial points @xmath517 , @xmath518 . from left to right : @xmath519 and @xmath402 ; @xmath520 and @xmath411 ; @xmath521 and @xmath522.,title="fig:",scaledwidth=40.0% ] fiber dynamics over time for different sets @xmath357 where the linear constraint is satisfied at spatial points @xmath517 , @xmath518 . from left to right : @xmath519 and @xmath402 ; @xmath520 and @xmath411 ; @xmath521 and @xmath522.,title="fig:",scaledwidth=40.0% ] fiber dynamics over time for different sets @xmath357 where the linear constraint is satisfied at spatial points @xmath517 , @xmath518 . from left to right : @xmath519 and @xmath402 ; @xmath520 and @xmath411 ; @xmath521 and @xmath522.,title="fig:",scaledwidth=40.0% ] elongation over time in dependence of the set @xmath357 , cf . figure [ fig:3 ] . the dashed line visualizes the theoretical bound @xmath523 . ] we are interested in using the numerical scheme for the simulation of a fiber dynamics . therefore , it is important that the elongation is restricted over time . theoretically , @xmath524,\end{aligned}\ ] ] holds according to lemma [ le : increase ] . the bound results from the @xmath414-inequality in the definition of @xmath343 . in our finite element formulation , the inequality constraint is incorporated in @xmath357 by the choice of @xmath397 ( remark [ rem : kh ] ) . this implicit treatment avoids the costly and complex optimization with inequality constraints , but only allows for an a posteriori check of . we demonstrate the impact of @xmath357 on the fiber behavior for an easy , meaningful scenario : in the stated set - up we apply an additional vertical force , i.e. @xmath525 with @xmath526 and @xmath527 of same order . figure [ fig:3 ] shows the fiber motion over time computed with @xmath528 for @xmath529 ( here : @xmath530 , @xmath531 ) . whereas @xmath357 with @xmath402 yields an obviously wrong , disturbing and even growing elongation , the results to @xmath357 with @xmath532 , @xmath533 look very similar and reasonable on the first glance . however , only the set @xmath357 where the linear constraint is prescribed at the @xmath534 space points @xmath521 fulfills the theoretical bound and provides the desired numerical solution , figure [ fig:4 ] . this stays true for the longtime behavior . note that the computational effort of the algorithm with `` euclidean projections '' is determined by time and space discretization ( and not by @xmath397 ) . the commercial software tool fidyst ( fiber dynamics simulation tool ) developed by the fraunhofer institute of industrial mathematics ( itwm ) in kaiserslautern , germany for the simulation of technical textile production processes computes the fiber dynamics by help of a full discretization : implicit euler method in time and finite difference method of higher order in space . the resulting nonlinear system is then solved by a newton method . in comparison to that approach our proposed scheme is computationally competitive . moreover , we provide a strict theoretical basis / justification , including convergence result and error bound . in the technical textile industry the dynamics of an elastic inextensible slender fiber is modeled by an nonlinear fourth order partial differential algebraic system . in this paper we have proposed a numerical scheme focusing on an accurate and efficient treatment of the algebraic constraint . ongoing work deals with the extension of analysis and numerics to the stochastic partial differential algebraic system @xcite arising for fibers immersed in turbulent air flows . here , a stochastic force ( source term ) of a white noise type is added in the fiber equation . the challenge lies again in the handling of the constraint . so far , the corresponding extensible equation with additive gaussian noise has been studied in @xcite . + acknowledgments this work has been supported by bundesministerium fr bildung und forschung , schwerpunkt `` mathematik fr innovationen in industrie and dienstleistungen '' , projekt 03ms606 . nm thanks her student daria frank for the preliminary numerical studies on projections . , _ spectral methods in infinite - dimensional analysis _ , vol . 12/1 of mathematical physics and applied mathematics , kluwer academic publishers , dordrecht , 1995 . translated from the 1988 russian original by p. v. malyshev and d. v. malyshev and revised by the authors .

recent angle - resolved photoemission spectroscopy ( arpes ) has identified that a finite - range frhlich electron - phonon interaction ( epi ) with c - axis polarized optical phonons is important in cuprate superconductors , in agreement with an earlier proposal by alexandrov and kornilovitch . the estimated unscreened epi is so strong that it could easily transform doped holes into mobile lattice bipolarons in narrow - band mott insulators such as cuprates . applying a continuous - time quantum monte - carlo algorithm ( ctqmc ) we compute the total energy , effective mass , pair radius , number of phonons and isotope exponent of lattice bipolarons in the region of parameters where any approximation might fail taking into account the coulomb repulsion and the finite - range epi . the effects of modifying the interaction range and different lattice geometries are discussed with regards to analytical strong - coupling / non - adiabatic results . we demonstrate that bipolarons can be simultaneously small and light , provided suitable conditions on the electron - phonon and electron - electron interaction are satisfied . such light small bipolarons are a necessary precursor to high - temperature bose - einstein condensation in solids . the light bipolaron mass is shown to be universal in systems made of triangular plaquettes , due to a novel crab - like motion . another surprising result is that the triplet - singlet exchange energy is of the first order in the hopping integral and triplet bipolarons are heavier than singlets in certain lattice structures at variance with intuitive expectations . finally , we identify a range of lattices where superlight small bipolarons may be formed , and give estimates for their masses in the anti - adiabatic approximation .

cond-mat0702495

a growing number of observations point to the possibility that high-@xmath0 cuprate superconductors may not be conventional bardeen - cooper - schrieffer ( bcs ) superconductors @xcite , but rather derive from the bose - einstein condensation ( bec ) of real - space pairs , as proposed by mott and others @xcite . a possible fundamental origin of such strong departure of the cuprates from conventional bcs behaviour is the unscreened ( frhlich ) epi with a polaron shift , @xmath1 of the order of 1 ev ( la@xmath2cuo@xmath3 , @xmath4ev ) @xcite , routinely neglected in the hubbard @xmath5 and @xmath6 models @xcite . this interaction with @xmath7axis polarized optical phonons is virtually unscreened because the upper limit for the out - of - plane plasmon frequency ( @xmath8 @xmath9 @xcite ) is well below the characteristic frequency of optical phonons , @xmath10 400 - 1000 cm @xmath11 . since screening is poor , the magnetic interaction remains small compared with the frhlich epi at any doping of cuprates . in order to generate a convincing theory of high - temperature superconductivity , one must treat the coulomb repulsion and _ unscreened _ epi on an equal footing . when both interactions are strong compared with the kinetic energy of carriers , the so - called `` coulomb - frhlich '' model ( cfm ) predicts a ground state in the form of mobile , preformed , inter - site pairs dressed by lattice deformations ( i.e intersite bipolarons ) @xcite . the most compelling evidence for ( bi)polaronic carriers in novel superconductors is the discovery of a substantial isotope effect on the carrier mass @xcite predicted by the bipolaron theory @xcite . recent high resolution arpes @xcite provides another piece of evidence for a strong epi in cuprates between electrons and c - axis - polarised optical phonons @xcite . these , as well as recent tunnelling @xcite , earlier optical @xcite and neutron scattering @xcite experiments unambiguously show that lattice vibrations play a significant though unconventional role in high temperature superconductors . remarkably , earlier path - integral studies of large bipolarons in the continuous limit @xcite ) led to a double surprise : ( a ) the large bipolaron is only stable in a very limited sector of the parameter space ( coulomb repulsion versus the frhlich coupling constant ) ( b ) most traditional `` frhlich polaron '' materials ( alkali halides and the like ) lie completely outside ( and `` far '' from ) this bipolaron stability sector , but several high-@xmath12 superconductors lie very close and even inside this rather restricted area of stability in the parameter space . when the strong frhlich epi operates together with a shorter range deformation potential and molecular - type ( e.g. jahn - teller ) epis , it readily overcomes the coulomb repulsion at short distances of about the lattice constant , so that large ( continuous ) bipolarons become local ( lattice ) bipolarons in narrow bands @xcite . even at significant doping local pairs are not overlapped , so that a high critical temperature for bose - einstein condensation ( bec ) could be achieved , if they are sufficiently mobile . analysis of the site - local holstein - hubbard model has indicated that in order for the coulomb repulsion ( hubbard @xmath5 ) to be overcome by the induced attractive force between electrons , epi must be so large that the polaron ( and bipolaron ) masses must be huge , rendering the transition temperature minuscule @xcite . all is not lost , however , since the holstein interaction is the extreme short - range limit of a finite range epi . using the finite - range epi it is possible for electrons to pair between sites @xcite without requiring the electron - phonon induced attraction to be larger than the hubbard @xmath5 . moreover , the individual polarons are significantly lighter , so the mass of the pair has potential to be orders of magnitude smaller than in the holstein case . to put these arguments on a solid microscopic ground we simulate the cfm hamiltonian on a lattice using an advanced qmc technique for bipolarons and compare with analytic results in the strong coupling and anti - adiabatic limits . first , we introduce the model . the hamiltonian for the cfm is written as @xmath13 each ion has a displacement @xmath14 . sites labels are @xmath15 and @xmath16 for electrons and ions respectively . @xmath17 annihilate electrons . the phonons are einstein oscillators with frequency @xmath18 and mass @xmath19 . @xmath20 denote pairs of nearest neighbours , and @xmath21 ion momentum operators . the instantaneous interaction @xmath22 has on - site repulsion @xmath5 and nearest neighbour interaction @xmath23 ( if the electron - phonon coupling term is set to zero , one obtains the simple @xmath24 model ) . the force function is of the screened frhlich type , @xmath25^{3/2}}\exp\left(-\frac{|{\mathbf{m}}-{\mathbf{n}}|}{r_{sc}}\right)\ ] ] ( @xmath26 is a constant ) @xcite . we will also use a slightly different notation for the electron - phonon interaction term here , @xmath27 where @xmath28 is a dimensionless interaction proportional to the force , and @xmath29 create phonons at site @xmath30 . we set @xmath31 . such a model has a remarkable property . unlike the site local holstein model , there is attraction ( and potentially pairing ) even in the presence of very strong on - site coulomb repulsion . the model is justified in the presence of alternating planes of itinerant electrons and ions , where there is strong screening along the @xmath17-axis . there have been a number of studies discussing the masses of polarons and bipolarons with long range interaction @xcite . the polaron formed from the long - range frhlich interaction proposed in @xcite has been simulated in reference @xcite , demonstrating that the polaron mass may be significantly lighter than its holstein counterpart . this is due to the nature of polarons in the holstein case , which may be demonstrated nicely by examining the lang - firsov transformation . in that transformation , the operators in the hamiltonian are replaced in the following way , @xmath32 \\ d_j \rightarrow \tilde{d}_j = d_{j } + \sum_i g_{ij } n_i & \hspace{10 mm } & c_i \rightarrow \tilde{c}_i = c_i \exp[-\sum_j g_{ij } ( d^{\dagger}_j - d_j ) ] \label{eqn : lffour}\end{aligned}\ ] ] thus hopping processes in the holstein polaron ( where @xmath33 ) take place by a complete relaxation of the lattice on the initial site , a hop , and then a distortion on the target site . with a longer range interaction , the lattice is pre - distorted before the particle moves , leading to a much smoother process with a lower intermediate energy state . we have recently determined that long - range interactions lead to a reduction of the importance of geometry on the properties of the bipolaron , especially the mass , leading to very similar results on triangular and square lattice @xcite . we will discuss the crossover between frhlich and holstein polarons later in this article . the lang - firsov transformation is an exact canonical transformation , and leads to a transformed hamiltonian with a new transformed wavefunction @xmath34 . it is most instructive to consider the transformation of the atomic hamiltonian and the transformation of the hopping terms separately , since typically the lang - firsov transformation is the starting point for a series of perturbative analyses . when the hopping term is set to zero , the phonon portion of the cfm is written as follows : @xmath35 ( n.b . the index @xmath36 is now taken to contain a spin and a site index ) . applying the lang - firsov canonical transformation , and [ eqn : lffour ] shows that electron number operators are unchanged on transformation , so the coulomb part of the hamiltonian is also unchanged ] @xmath37\\ & = & - \sum_{ii ' } n_i n_{i ' } \sum_{j } \frac{f_{ij } f_{i'j}}{2 \omega^2 m } + \omega \sum_j \left(d^{\dagger}_j d_j + \frac{1}{2}\right ) \label{eqn : atomic_g } \label{eqn : shift}\end{aligned}\ ] ] where we reordered the summation , and noted that @xmath38 . this shows the remarkable property of the lang - firsov transformation , since the electron and phonon subsystems in the atomic limit are now completely decoupled . at this stage , it is convenient to introduce the following function , @xmath39=\sum_{\mathbf{m}}f_{\mathbf{m}}[\mathbf{r}(\tau)]f_{\mathbf{m}+\delta\mathbf{r}}[\mathbf{r}(\tau')]\ ] ] where the reason for adding an additional translation , @xmath40 , to the phonon sub - system will become apparent when the action is introduced in the next section . for the following discussion , @xmath41 . the @xmath42 function for the ladder systems investigated in this paper is shown in figure [ fig : phifunction ] , corresponding to a screened frhlich interaction with @xmath43 . we also define the dimensionless interaction parameter @xmath44 where @xmath45 is the magnitude of the energy of the tight - binding electron ( normally the half band - width @xmath46 ) . as we can see from equation [ eqn : shift ] , the energy shift when there is only one particle ( polaron shift for @xmath47 ) is @xmath48 . thus @xmath49 . substituting that definition into the atomic hamiltonian , one obtains : @xmath50 the reason for introducting the new functions @xmath42 can immediately be seen , since they appear in the hamiltonian as a ratio . thus , in combination with @xmath51 they give a universal definition of coupling in models with long - range hopping . substitution of equations ( [ eqn : lfone]-[eqn : lffour ] ) transforms the tight binding hamiltonian in the following way : @xmath52 where @xmath53 i.e. we can see that the electron - phonon interaction in the transformed hamiltonian is part of the hopping process , and that to some extent , one may regard the operators @xmath54 as creating polarons ( i.e. electrons and a phonon cloud at the same time ) . we now apply the following identity , @xmath55/2 } = e^{a+b } \label{eqn : exponentidentity}\ ] ] ( which is valid if @xmath56 $ ] commutes with both @xmath57 and @xmath58 ) with @xmath59 , which is always valid . therefore , @xmath60/2}$ ] . in equation [ eqn : sigma ] we may choose @xmath61 and @xmath62 . so , @xmath63 and @xmath64 = 0 $ ] since the commutator is a number , equation [ eqn : exponentidentity ] may be applied . the hopping operator becomes : @xmath65 now , the identity may be used again , making the grouping , @xmath66 and @xmath67 . thus @xmath64 = -\sum_{j } ( g_{ij } - g_{i'j})(g_{ij } - g_{i'j})$ ] , and @xmath68 e^ { \sum_j ( g_{ij } - g_{i'j } ) d^{\dagger}_j } e^ { - \sum_j ( g_{ij } - g_{i'j } ) d_j } \ ] ] this form is particularly useful , due to the order of the creation and annihilation operators . thus , when one carries out the perturbation theory , calculation is reduced to computing matrix elements of the form , @xmath69 . it s interesting to note that , when computed as an average over the atomic wavefunction , the hopping integral may be regarded as being modified by the epi as @xmath70 , where @xmath71 . such an approximation is valid in the anti - adiabatic limit , and will be revisited later in this paper . the band - narrowing factor was originally introduced by tyablikov using an equations of motion scheme @xcite . the ctqmc algorithm presented here is an extension of a similar path - integral method for simulating the polaron problem @xcite . an integration over phonon degrees of freedom following feynman leads to an effective action , which is a functional of two polaron paths in imaginary time which form the bipolaron and is given by the following double integral when @xmath72 @xcite , @xmath73 & = & \frac{z\lambda\bar{\omega}}{2\phi_0(0,0 ) } \int_0^{\bar\beta } \int_0^{\bar\beta } d \tau d \tau ' e^{-\bar{\omega } \bar\beta/2 } \sum_{ij}\phi_0[\mathbf{r}_i(\tau),\mathbf{r}_j(\tau ' ) ] \nonumber\\ & & \hspace{20mm}\times \left ( e^{\bar{\omega}(\bar\beta/2-|\tau-\tau'| ) } + e^{-\bar{\omega}(\bar\beta/2-|\tau-\tau'| ) } \right)\nonumber\\ & + & \frac{z\lambda\bar{\omega}}{\phi_0(0,0 ) } \int_0^{\bar\beta } \int_0^{\bar\beta } d \tau d \tau ' e^{- \bar{\omega } \tau } e^{-\bar{\omega } ( \bar\beta - \tau ' ) } \nonumber\\ & & \hspace{20mm}\times \sum_{ij}\left ( \phi_{\delta\mathbf{r}}[\mathbf{r}_i(\tau),\mathbf{r}_j(\tau ' ) ] - \phi_0[\mathbf{r}_i(\tau),\mathbf{r}_j(\tau')]\right)\nonumber\\ & - & \frac{1}{2}\int_0^{\beta}v(\mathbf{r}_1(\tau),\mathbf{r}_2(\tau))\,d\tau \ : . \label{eq : seven}\end{aligned}\ ] ] where the vector @xmath74 is the difference between the end points of one of the paths in the non - exchanged configuration ( here @xmath75 and @xmath76 ) . the indices @xmath77 and @xmath78 represent the fermion paths . @xmath79 is an instantaneous coulomb repulsion . the part of the action depending on @xmath80 arises because the entire phonon subsystem at @xmath81 must also be shifted when there is a shift in the electron sub - system between the start and end configurations . the definition of @xmath82 and other nomenclature for the ctqmc simulation of ladder systems are shown in figure [ fig : examplepath ] . from this starting point , the bipolaron is simulated using the metropolis monte - carlo ( mc ) method . the electron paths are continuous in time with hopping events ( or kinks ) introduced or removed from the path with each mc step . analytic integration is performed over sections of parallel paths . the ends of the two paths at @xmath83 and @xmath81 are related by an arbitrary translation , @xmath84 . in contrast to the one - particle case , the fixing of the end configurations limits the update procedure to inserting and removing pairs of kinks and antikinks . we constrain particles to opposite legs of the ladder , which corresponds to two species of charged particles . in such a system , there is no exchange between particles . exchange and singlet triplet splitting from quantum monte - carlo simulations is briefly discussed in this section , with an analytical discussion in section 5 . a full discussion of the qmc procedure for exchange is left to a future article . , and sites in the chain by @xmath85 . the path configuration at @xmath83 is identical to that at @xmath81 up to a shift @xmath86 ( i.e. the ends of the beginning and end of the paths are separated by @xmath87 ) . each path has a number of kinks @xmath88 and antikinks @xmath89,width=340 ] . [ fig : examplepath ] in the path integral qmc with two paths , it is necessary to make two kink operations simulataneously to ensure that the end configurations remain identical up to a translation . there exist two classes of update . first , kink / antikink pair addition / removal on a single path is useful , since it always maintains the same end configurations up to a change in the interpath displacement . second , kink pair additions on different paths are needed in the analysis of the bipolaron mass , where the @xmath81 end configuration must be exactly equal to the @xmath83 end configuration up to a parallel shift . within both subclasses , there are two specific update types that satisfy the imaginary - time boundary conditions , as follows : * two kinks of the same type @xmath90 are added to or removed from two different paths . * a kink - antikink pair is added to or removed from one of the two paths . an antikink to kink @xmath90 is a kink with the opposite direction @xmath91 . * a kink of type @xmath90 is inserted into one path , and another kink of the same type @xmath90 is removed from the same path ( kink shift ) . * a kink of type @xmath90 is added to one path , and an antikink @xmath91 is removed from the other path . an important property of the bipolaron system is that the type and time of added or removed kinks still does not define the new path unambiguously . indeed , imagine a kink of type @xmath90 being inserted on a single path at time @xmath92 . this could change the path in two different ways : either the path at times @xmath93 is shifted in the direction @xmath90 , or the path at times @xmath94 is shifted in the antidirection @xmath90 . we refer to the former change as a top shift , and the latter as bottom shift . for the single - path ( polaron ) problem the distinction between the top and bottom shifts was not important because they are identical up to a translation of the entire path . this argument does not hold for two paths , since the resulting interpath distance in a binary update changes with shift type , thus a choice of shifts is an important part of the monte carlo update process . we proceed to derive the update probabilities for the monte - carlo scheme set out above . there is considerable flexibility in choosing the probabilities for adding and removing kinks . we choose an equal weighting scheme for choosing kinks , shifts and paths as follows : * choose a kink type @xmath90 from the list of all possible kinks , with equal probability @xmath95 , where @xmath96 is the total number of kink types . @xmath97 cancels on both sides of all balance equations considered below . * anti - kink types are always determined from the kink type . * shift type ( top or bottom ) is chosen with equal probability @xmath98 _ independently _ for the two kinks . @xmath99 also cancels on both sides of the balance equation . * assign path a with equal probability @xmath100 from the two available paths . * assign path b as the other path . we will also choose kinks to add and remove according to the following weightings , although there are some specific rules in the updates below to deal with cases where there are no kinks or antikinks of the chosen type on a path . * the probability density for kink time selection when adding a kink is always @xmath101 , ( @xmath102 ) . * the probability for removing a kink of type @xmath90 from path a in a configuration @xmath103 is @xmath104 where @xmath105 is the number of kinks of type @xmath90 on the path a. we note that the following is not the only set of possible rules , however , we consider this to be the most transparent method of choosing kinks to insert and remove . [ [ secthreeone - update - type - i - additionremoval - of - two - kinks - to - or - from - different - paths ] ] [ sec : threeone ] update type ( i ) : addition / removal of two kinks to or from different paths + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + consider two configurations with two paths , @xmath103 and @xmath106 , where configuration @xmath106 has two more @xmath107 kinks than @xmath103 , one on the first path at time @xmath108 , and one on the second path at time @xmath109 . the balance equation is @xmath110 the relative weight of configurations @xmath103 and @xmath106 is @xmath111 . in order to approach the limit of continuous time , we rewrite the probability @xmath112 of selecting two kinks at @xmath108 and @xmath109 to add to different paths given a configuration @xmath103 as @xmath113 , where @xmath114 is the corresponding _ two - dimensional _ probability density the probability of selecting the two specific kinks @xmath108 and @xmath109 to remove from the resultant configuration ( @xmath106 ) to get back to @xmath103 is a finite number @xmath115 , which is the probability of removing those two kinks given configuration @xmath106 . substituting these results into the balance equation , one may cancel the @xmath116 . applying metropolis , update probabilities are obtained as follows : @xmath117 @xmath118 thus we have obtained update probabilities that do not depend on the time discretisation , and we can immediately take the limit of continuous time . a similar approach can be taken for any of the update types i to iv . we now demonstrate all steps in the derivation of the first update probability as an example . the rules and resulting probabilities are as follows : 1 . choose kink types , shifts and paths according to rules ai - av . 2 . if the initial configuration has _ at least _ one @xmath107 kink on path a _ and _ on path b , then removal of a pair is proposed with probability @xmath119 , and addition of a pair is proposed with probability @xmath120 . otherwise , only pair addition can be attempted and @xmath121 . 3 . if pair addition is selected , times are selected to insert one kink on path a , and another on path b with independent equal probability density ( rule bi ) . if pair removal is chosen , then one candidate kink is selected with independent equal probability from each of paths a and b in configuration @xmath106 ( rule bii ) . implementing these choices , one obtains @xmath122 from the combination of rules ( i ) , ( ii ) , ( iii ) . likewise , the combination of rules ( i ) , ( ii ) and ( iv ) specifies that @xmath123 . rule ( iii ) leads to @xmath124 if @xmath125 and @xmath126 and @xmath127 otherwise . since configuration @xmath106 always has sufficent kinks to make a removal , we always have @xmath128 leading to the following acceptance rules : @xmath129 @xmath130 please note which configuration the kink numbers apply to . * in the case of kink addition , initial configuration is @xmath103 and final configuration is @xmath106 . in the case of kink removal , the initial configuration is @xmath106 and the final configuration is @xmath103 . * [ [ secthreetwo - update - type - ii - additionremoval - of - a - kink - antikink - pair - to - one - path ] ] [ sec : threetwo ] update type ( ii ) : addition / removal of a kink - antikink pair to one path + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the general properties of this update type are similar to update type ( i ) : the addition of a pair at times @xmath108 and @xmath109 is characterized by a two - dimensional probability density @xmath131 while removal of a pair is charaterized by a finite number @xmath132 ( [ eq : seventeen])-([eq : eighteen ] ) still apply with @xmath133 in place of @xmath134 . since @xmath135 , the acceptance rules are idential , and only @xmath136 and @xmath137 differ . we use the following rules . 1 . choose kink types , shifts and paths according to rules ai - av . 2 . if the initial configuration has _ at least _ one @xmath107 kink and one antikink ( @xmath138 , @xmath139 ) , then removal of a pair is proposed with probability @xmath140 , and addition of a pair is proposed with probability @xmath141 . otherwise , addition of a pair is proposed with probability @xmath142 . if pair addition is selected , kink and antikink insertion times are selected with independent equal probability density for insertion on path a. ( rule bi ) 4 . if pair removal is chosen , then one candidate kink and one candidate antikink are selected with independent equal probability from path a in configuration @xmath106 ( rule bii ) . one computes the update probabilities as before : @xmath143 @xmath144 [ [ secthreethree - update - type - iii - addition - and - removal - of - a - kink - to - one - path - kink - shift ] ] [ sec : threethree ] update type ( iii ) : addition and removal of a kink to one path ( kink shift ) + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + this update type does not change the number of kinks , and hence does not change the kinetic energy of the system . we define a configuration @xmath103 , and a configuration @xmath106 which is identical to @xmath103 except that one kink has been shifted . to get from configuration @xmath103 to @xmath106 , a kink is removed from path a at time @xmath108 and is reinserted in the path at time @xmath109 . since @xmath103 and @xmath106 have equal total kink number , the ratio of statistical weights is @xmath145 . from detailed balance and metropolis : @xmath146 there is only one update rule , since we can get from @xmath106 and @xmath103 using exactly the same process as going from @xmath103 to @xmath106 . all attributes of the kinetic energy have dropped out from the equations . the acceptance rules are determined solely by the electron - phonon interaction , as expected for this update type . in many practical situations it is reasonable to choose the functions @xmath137 and @xmath136 independent of time @xmath147 . in this case the above expressions simplify significantly . in particular , consider the following set of update rules . 1 . choose kink types , shifts and paths according to the general rules ai - av . 2 . if the path has no @xmath107 kinks , @xmath148 , the update attempt is aborted . otherwise , @xmath149 , so propose an @xmath90 kink for removal with equal probability @xmath150 ( rule bii ) . 3 . propose time for new @xmath107 kink with constant probability density @xmath151 ( rule bi ) . these rules result in cancelation of @xmath152 and @xmath153 from the above equations , which reduce to @xmath154 [ [ secthreefour - update - type - iv - kink - addition - to - one - path - and - antikink - removal - from - the - other - path ] ] [ sec : threefour ] update type ( iv ) : kink addition to one path and antikink removal from the other path + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + in this update , a kink @xmath107 is added to one path and an antikink @xmath155 is removed from the other path . as a result , the @xmath156 end configuration shifts as a whole by @xmath107 . in a reciprocal process , a kink @xmath107 is removed from one path , an antikink @xmath155 is inserted in the other path , and the @xmath156 configuration shifts by @xmath155 . the ratio of weights is @xmath157 . the balance equation is satisfied by the following solution @xmath158 since we can obtain the inverse process by changing kink @xmath90 for its antikink , the update probability @xmath159 is not necessary since we always choose @xmath90 from all kink types . as the simplest implementation , the following rules are used , 1 . choose kink types , shifts and paths according to rules ai - av . 2 . if path b has no antikinks @xmath155 , @xmath160 , then the update attempt is aborted . otherwise , @xmath161 , so an antikink is proposed for removal from path b with equal probability ( rule bii ) . the time location for kink insertion on path a is proposed with equal probability density ( rule bi ) . with these assumptions , the acceptance probability takes the form : @xmath162 when our monte - carlo scheme has reached equilibrium , we make a series of measurements of physical properties . the ground state energy is : @xmath163 \ : , \label{eq : nine}\ ] ] where @xmath164 is the number of kinks of type @xmath165 , and angular brackets denote ensemble averaging . the number of phonons is given by : @xmath166 where the derivative is taken keeping @xmath167 constant . the polaron band energy spectrum can be computed from : @xmath168 where @xmath169 is the quasi momentum . by expanding this expression in small @xmath169 , the @xmath36-th component of the inverse effective mass is obtained as @xmath170 thus the inverse effective mass is the diffusion coefficient of the polaron path in the limit of the infinitely long `` diffusion time '' @xmath156 . the bipolaron radius is the average distance between paths , @xmath171 finally , the mass isotope coefficient , @xmath172 , is calculated as follows @xmath173 \ : . \label{eq : thirteen}\ ] ] there are certain aspects of good practice for quantum monte - carlo simulations that we adhere to here . as always for monte - carlo simulations a random number generator with sufficient period is used . measurements are performed every few steps to avoid unnecessary correlations in results ( the aim here is to spend no longer measuring than simulating , since time correlated results do not make a large contribution to more accurate measurement ) . careful blocking analysis with large blocking sizes @xmath174 is performed to determine accurate error bars . to avoid anomalous error bars caused by long time correlations , we compare error bars computed with two block sizes @xmath174 and @xmath175 . exchanges are significantly more complicated , with several possibilities for the exchange update involving inserting and removing kinks . in the exchanged configuration , there are an additional 4 update rules , and there is also an ambiguous configuration where both paths have the same start and end points , which leads to some small additional modifications . we defer a full discussion of exchange update rules to a later paper . on our ladder models , exchanges are not required , since electrons sit on opposite legs . a consequence of exchange is that singlet and triplet states are not degenerate . we can see the singlet - triplet mass difference as a consequence of interference between paths in the monte carlo simulation . we take a simplified one - dimensional example to illustrate the mechanism . consider a bipolaron of separation @xmath176 propagating from the sites @xmath177 at @xmath83 to @xmath178 at @xmath81 . we assume that the weight @xmath179 of a single - electron path is a monotonically decreasing function of the number @xmath180 of kinks in the path , and the paths with the smallest number of kinks dominate . this is likely to be valid in the strong - coupling limit . we can therefore write the weight of as path as a monotonically rapidly decreasing function @xmath181 , where @xmath182 is the distance between endpoints . we also neglect the interaction between paths . consider first the case of periodic boundary conditions @xmath183 . the weights of the singlet and triplet bipolaron paths are respectively the sum and the difference of the direct and exchange paths . @xmath184 here if @xmath185 , the singlet and triplet weights are dominated by the direct path and are nearly equal . now consider a twist larger than the bipolaron radius , @xmath186 . the singlet and triplet weights are dominated by the shortest paths : @xmath187 in this case the total number of kinks in either the direct or the exchange path is the same ( @xmath188 ) , so there will be substantial cancellation in the triplet case . thus the diffusion coefficient of the triplet bipolaron will be smaller , and the effective mass larger , than that of the singlet . we briefly discuss the lattice dependent features of the polaron problem here . for more detailed discussion of the polaron problem , the reader is directed towards our papers on this subject @xcite , and to the paper by kornilovitch in this issue @xcite . in this section , we specifically discuss the changes to the masses of polaron moving on trangular and square lattices as the screening length @xmath189 is varied . the polaron mass forms part of the argument later in this article . screened frhlich electron - phonon interactions were simulated by spencer _ et al . _ @xcite , who demonstrated a continuous crossover between the frhlich and holstein limits on the chain . in particular , the mass of the particle is found to be light down to quite small screening radii , consistent with results by bona and trugman @xcite for nearest - neighbour electron - phonon interactions . we have previously computed the properties of polarons on several bravais lattices , showing that the effects of the lattice type on the properties of the polaron are ` washed out ' by long range interactions @xcite . figure [ fig : polaronmass ] shows the effective mass of the dicrete frhlich polaron on square and triangular lattices . frhlich polarons are significantly lighter than their holstein counterparts , due to the long range interaction . we have also shown that the overriding factor for the properties of holstein polarons is the number of nearest neighbours in the lattice , and not the dimensionality @xcite . since we are interested in long - range interactions in this paper , then it suffices to note that the masses of frhlich polarons are of the same order of magnitude on all lattice types , and that they are extremely light @xcite . the properties of realistic screened interactions lie somewhere inbetween , but such polarons remain light down to quite small interaction ranges of the order of a lattice spacing @xcite . the properties of bipolarons are significantly more complicated than those of polarons . one expects bound holstein bipolarons in strongly correlated materials to be extremely heavy , since a very large attractive potential is needed to overcome the repulsion . this is not true for longer range interactions . according to work by bona and trugman @xcite , even bipolarons with nearest neighbour interactions are significantly more mobile than their holstein counterparts ( indeed , one normally expects the mass of a bipolaron to scale like the squared mass of the polaron , so polarons which are an order of magnitude lighter than holstein ones , may become bipolarons which are two orders of magnitude lighter ) . another extremely interesting proposition is the role of geometry on the bipolaron mass . when bipolarons are bound on nearest neighbour sites , and another degenerate state may be reached in a single hopping event , the leading correction to the atomic hamiltonian is first order in the hopping term , and not second order as one might expect for the holstein model , which leads to a bipolaron with a mass that is of similar order of magnitude to the polaron mass ( i.e. a superlight bipolaron ) @xcite . such systems may be realised on triangular lattices , or on lattices with large next - nearest neighbour hopping . we have recently extended our quantum monte - carlo algorithm to explore this type of bipolaron , leading to similar conclusions @xcite . in this paper , we discuss some of these extensions to the algorithm to look at two types of ladder system shown in figure [ fig : laddersystems ] . since the particles on the different legs of the ladder can not exchange the very weak - couping limit is not well bound , consisting of two large polarons . as such , the weak - coupling perturbation theory can be made about the unbound state . as will become apparent when we show the quantum monte - carlo results , the number of phonons is small for the weakly bound states , especially in the anti - adiabatic limit . the perturbation theory in the electron - phonon coupling term only excites single phonons , so if the number of phonons becomes too large , the theory fails . in the presence of strong on - site repulsion , the bipolaron is not bound at zero coupling . in general , if there is a bound state for @xmath190 , then this perturbation expansion fails . the expansion is written as follows : @xmath191 @xmath192 @xmath193 where @xmath194 is the number of momentum states . thus the ground state polaron energy ( at @xmath195 ) is @xmath196 , which defines a dimensionless coefficient @xmath197 . there is no general analytic solution for the second order perturbation theory , but values may be computed using numerical integration . the number of phonons , isotope exponent and inverse mass may also be written as , @xmath198 , @xmath199 and @xmath200 . the weak coupling limits for the polaron are discussed in references @xcite amoung others . @xmath51 is defined as before . aspects of the strong coupling limit are easy to compute from the path integral formalism . in the very strong coupling limit , the most common configurations are straight paths , since the action is proportional to @xmath51 and that configuration gives the smallest possible action . thus the strong coupling action reads : @xmath201 for a polaron , the total energy for a self - interacting strong - coupling system ( i.e. a straight path ) is @xmath202 , where @xmath45 is the half band width , @xmath203 . computing the action for the straight path configuration , one obtains the energy as @xmath204 . leading to , @xmath205\ ] ] where @xmath206 is a nearest neighbour vector between chains indicating the point of closest approach . this result is of little suprise , since there are 2 polarons in a bipolaron , with @xmath202 each , and 2 inter - polaron binding terms with @xmath207 . we may also compute the number of phonons associated with the bipolaron in a similar manner using equation [ eq : ten ] with @xmath208 , which we obtain as : @xmath209/\hbar\omega\ ] ] as we have discussed , the @xmath42 functions for the ladder systems with @xmath43 are plotted in fig . [ fig : phifunction ] . numerical values for @xmath210 , @xmath211 and @xmath212 are shown in table [ tab : strong ] , from which numerical values for the strong coupling behaviour are computed . we discuss the region of validity of these results when we compute the exact qmc results . in particular , if we plot appropriate ratios , such as the ratio @xmath213 , we expect saturation at strong coupling , consistent with table [ tab : strong ] . typically , it has been discussed that an expansion of the lang - firsov transformed hamiltonian in the polaron hopping can be used at strong coupling . the perturbation expansion to second order may be written as , @xmath214 the values for the strong coupling energy may be computed from the leading term of this expansion , and have the same form as the energy computed from the straight paths . this is not suprising , since the corrections to the energy are at least as small as @xmath215 . there is a subtle point associated with phonon numbers . by examination , the ground state of the phonon subsystem in the * transformed * hamiltonian is @xmath216 . in order to determine the total number of phonons in the true ground state of the atomic hamiltonian , one must transform back to the regular wavefunction . one may either transform the wavefunction , or the phonon part of the hamiltonian . transforming the phonon part of the hamiltonian is easy , and one obtains , @xmath217 thus @xmath218 the expectation value of the phonons occupation may be rewritten in terms of @xmath51 and @xmath42 , thus : @xmath219 ( note the minus sign in front of the equation - phonon number is positive ) . thus one determines that the total number of phonons in the bipolaron case is : @xmath220 ( n.b . a similar argument can be used for polarons . ) unfortunately , the computation of the mass is not so simple . the mass is very sensitive to the exact form of the renormalised hopping , and the leading order of the perturbation expansion varies with lattice type . for the superlight small bipolarons on triangular plaquettes discussed here , there is a leading term with order @xmath215 , and for rectangular systems , the leading order is @xmath221 . examination of the path integrals demonstrates that the mass is computed from a parallel shift . the first perturbation of the paths from the straight line in the atomic limit is the insertion of two parallel kinks ( one on each path ) . in order for the mass to be genuinely first order in the polaron hopping , we would require updates with only one kink . thus , there are significant contributions from the second order term , including a cancellation between the orders as @xmath222 . in fact , the expansion of the lang - firsov transformed hamiltonian to 1st order in the hopping turns out to correspond to the extreme anti - adiabatic limit , which we now discuss . .strong coupling behaviour of the staggered and rectangular lattices . [ cols="<,<,<,<,<,<",options="header " , ] the bipolaron dispersion can be evaluated analytically in the anti - adiabatic strong - coupling limit ( @xmath223 ) , in which case the problem can be reduced to that of a rigid dimer . in this limit the lang - firsov transformation@xcite eliminates the phonons from the coulomb - frhlich hamiltonian ( [ fh ] ) . in combination with an averaging over phonons consistent with the anti - adiabatic approximation ( if the phonon frequency is very large , there are no real phonons ) one obtains the following effective @xmath24 hamiltonian @xmath224 where the primed sum excludes self - interaction , and the renormalised on - site and inter - site interactions are @xmath225 and @xmath226 respectively . ( there is no retardation in the anti - adiabatic limit . ) now suppose the on - site repulsion @xmath227 is large and repulsive and the inter - site interaction has a well - defined minimum @xmath228 at some separation . let @xmath229 be the set of sites of @xmath230 at that separation from @xmath231 : @xmath232 and @xmath233 we shall call @xmath229 the `` neighbours '' of @xmath231 . in general , @xmath229 need not be the hopping neighbours @xmath234 . the low - energy sector for two electrons corresponds to dimer states ( bipolarons ) in which the electrons are on neighbouring sites ; the energies in this sector are @xmath235 . the energy cost of internal excitations of the bipolarons introduces a gap . we can now sharply distinguish two types of bipolaron motion : `` crab - like '' , in which the constituent polarons remain neighbours , and `` crawler '' , which requires virtual transitions out of the low - energy sector . the crab - like bipolaron bandwidth will be @xmath236 , while the crawler contributions will be @xmath237 . we shall for the present purposes project onto the low energy sector and hence ignore the resulting higher - order terms in @xmath238 , immobilising crawlers . for simplicity we drop the tildes from the notation and absorb the polaron shift @xmath239 into the chemical potential . if a lattice @xmath240 has @xmath241 sites with a mean number @xmath242 of neighbours per site , then the single - polaron hilbert space is @xmath243-dimensional , the two - polaron hilbert space is @xmath244-dimensional and the crab bipolaron hilbert space is just @xmath245-dimensional . we can further reduce to one singlet and three triplet spaces , each of dimensionality @xmath246 . the singlet bipolaron space is @xmath247 and the @xmath248 sectors of the triplet bipolaron space are latexmath:[\[\begin{aligned } \mathcal t _ 1&= & \mathrm{span } \left\ { \mathcal t _ 0&= & \mathrm{span } \left\ { \frac 1 { \sqrt 2}\left(|\mathbf n\uparrow \mathbf{n^\prime}\downarrow\rangle - |\mathbf{n^\prime}\uparrow \mathbf n\downarrow\rangle\right ) : \mathbf n\in \lambda , \mathbf{n^\prime}\in n_\mathbf n\right\ } \label{triplet0 } \\ \mathcal t _ { -1}&= & \mathrm{span } \left\ { this enables us to write the low - energy effective hamiltonian of the dimers in each sector as a tight - binding hamiltonian on the _ dimer lattice _ constructed in the following way : a node is placed on the line joining neighbours in the lattice @xmath240 . if @xmath250 and @xmath251 are both hopping neighbours of @xmath231 ( @xmath252 ) , then the dimer can hop from @xmath253 to @xmath254 . a bond is then drawn between the two nodes on the dimer lattice with hopping integral @xmath255 in the singlet sector and @xmath256 in the triplet sector . this sign change ensures the correct exchange symmetry for closed paths on odd - membered rings : as a dimer completes one cycle of an odd - membered ring its end - points are interchanged . this can lead to a dramatic difference between singlet and triplet bipolaron masses on a non - bipartite lattice , as we shall see . in the staggered ladder depicted in fig [ fig : laddersystems ] the neighbours of a site on one chain are the two adjacent sites on the opposite chain , while the hopping neighbours of a site are along the same chain . the corresponding dimer lattice is a one - dimensional chain with hopping @xmath257 and two sites per unit cell . as there are no exchange paths , the sign of the hopping can be gauged away and there is no singlet - triplet splitting . the polaron and bipolaron dispersions are therefore respectively @xmath258 the bipolaron effective mass is therefore four times that of the polaron ( as previously reported @xcite ) : @xmath259 this result is remarkable if one considers the standard strong coupling result for the mass of the bipolaron . in the rectangular ladder the requirement for a virtual internal excitation of the bipolaron in the crawler dynamics would lead to a hopping @xmath260 and hence mass @xmath261 . the staggered ladder with long range electron phonon attraction has two degenerate nearest neighbour bound states ( as summarised in figure 12 ) , so no intermediate state is required when the particle hops . it is clearly important that the electrons are bound one lattice spacing apart to take advantage of this effect , but this can easily be achieved in the presence of a strong site - local coulomb repulsion ( so that the energy is not at a minimum when both particles are on the same site ) . this spaced minimum is clearly extremely important when real lattices beyond the toy ladder models presented here are considered . some details of real lattices will be discussed later in this paper , but to briefly summarise , in order to obtain this special kind of bipolaron , ( a ) the lattice must have several degenerate near - neighbour bound states which can be transformed from one degenerate state to another via a single hop ( b ) strong coulomb repulsion is required to stop a unique single - site bound state from forming between the polarons , and ( c ) long range attraction is required so that the minimum in the potential function ( attraction + repulsion ) is at approximately one lattice spacing . this information is summarised in figures [ fig : pictorial ] and [ fig : effinteraction ] in the conclusion to this paper . we compute qmc results for bipolarons moving on staggered and rectangular ladders with period 1000 for a range of @xmath51 and @xmath18 , including the total energy , inverse bipolaron mass , bipolaron radius , mass isotope exponent and the number of phonons in the bipolaron cloud . figure [ fig : totalenergy ] shows the total energy of the bipolaron for ( a ) the rectangular ladder and ( b ) the staggered ladder . a slight change in gradient between weak and strong coupling is just discernable , demonstrating ( as we shall see in the coming figures ) that the staggered ladder reaches the strong coupling limit at significantly lower @xmath51 than the rectangular ladder . strong coupling results from the previous section agree well . we see that there are no significant differences between the total energy of the bipolarons on the staggered and rectangular ladders , although the strong coupling limit is reached for slightly lower @xmath51 in the case of the staggered ladder . if one is to reach a superconducting state via the bose - einstein condensation of bipolarons , there are two conditions . first , the bipolaron pair must be light , and second , the pairing radius must be small . we demonstrate the differences between the inverse masses on the staggered and rectangular ladders in figures [ fig : inversemass](a ) and ( b ) , which show the inverse mass of the bipolaron for a number of different @xmath51 and @xmath262 . there is more than an order of magnitude difference between the mass of the bipolaron on the staggered ladder , and that on the rectangular ladder over significant regions of the parameter space . in fact , the magnitude of the bipolaron mass turns out to be of similar size to that of the polaron mass over a wide range of the parameter space . the mass is inversely proportional to the transition temperature of the bec , so a small mass is essential to obtain a decent @xmath263 . we have demonstrated that one of the precursors for a bose - einstein condensate of pairs above the mk range may be met on the staggered ladder arrangement , but we also require small pairs , with non - overlapping wavefunctions . in figure [ fig : bipolaronsize ] we show how the the size of the bipolaron varies as @xmath51 and @xmath262 are varied . not only is the bipolaron on the staggered ladder lighter than that on the rectangular ladder , it is also has a significantly smaller radius than the bipolaron on the rectangular ladder , making it a much better prospect for bose - einstein condensation . since the result that the bipolaron mass is proportional to the polaron mass via a numerical value relies on the anti - adiabatic limit , where the phonon frequency is very large , so that phonons may not be excited , we investigate the number of phonons asociated with the bipolaron in figure [ fig : nophonons ] . the result is weighted by phonon frequency and electron phonon coupling , so that the strong coupling result can clearly be seen . again , we can see that the strong - coupling limit is reached at significantly lower @xmath51 on the staggered ladder than on the rectangular ladder . strong coupling results from the previous section agree well , with the ratio @xmath264 approaching the numerical value given in table [ tab : strong ] . the anti - adiabatic limit can clearly be identified as regions on the graph where the phonon number approaches zero ( i.e. the @xmath265 quadrant of the parameter space ) , consistent with results for the mass . we show the mass isotope exponent of the bipolaron in figure [ fig : iex ] . again , we can see that the strong coupling limit is achieved at significantly lower @xmath51 on the staggered ladder , compared with the rectangular ladder . the isotope exponent is also smaller on the staggered ladder , demonstrating a much smaller range of mass from weak to strong coupling . finally , in figure [ fig : path ] , we show example path configurations on ( a ) the rectangular ladder and ( b ) the staggered ladder . hopping events on different paths on the rectangular ladder are very closely correlated on the imaginary time axis . on the staggered ladder , there are two degenerate configurations , and paths are just as likely to sit on either of the two neighbouring sites , significantly reducing the correlation between kinks , and increasing the probability that kink pairs can be inserted . it is this that leads to significantly lighter bipolarons on the staggered ladder . on the ladder systems , the electrons were held on neighbouring legs of the ladders . this is partially consistent with a very strong local coulomb repulsion or hubbard @xmath5 . it is also possible to examine the effects of a very strong or even infinite hubbard @xmath5 on the masses of bipolarons bound via long range attraction on other low dimensional systems . as we have seen by examining the ladder systems , there are 3 requirements for superlight bipolarons ( 1 ) electrons are not allowed to bind on a single site ( 2 ) there are at least two degenerate configurations of electrons sitting on neighbouring sites and ( 3 ) there are single hopping events which transform one configuration into another degenerate configurations . then , hopping of the bipolaron is first order in the hopping of the polaron . condition ( 2 ) is satisifed by long range electron - phonon intereraction , and there are several tight binding lattices which also satisfy conditions ( 2 ) and ( 3 ) , and a strong site - local coulomb repulsion satisfies condition ( 1 ) . we discuss bipolarons on these lattices in the anti - adiabatic approximation in this section . the staggered ladder system discussed earlier in this paper can be considered to be made up of triangular plaquettes , so the first logical step to looking beyond the ladder systems is to analyse the physics of a single plaquette . if hopping is allowed between all the sites on the plaquette , then one may consider exchange effects , which lead to singlet and triplet bound states . to see exchange effects , we require lattices with odd - membered rings . as the simplest example , consider three sites @xmath266 , all of which are neighbours , with hopping @xmath267 . the polaron hamiltonian is @xmath268 the dimer lattice is constructed by placing a node at the centre of each bond . the singlet bipolaron hamiltonian is therefore @xmath269 with eigenvalues @xmath270 , and the triplet bipolaron hamiltonian is @xmath271 with eigenvalues @xmath272 . ( an alternative choice of basis would be the six - dimensional unsymmetrised @xmath273 basis @xmath274 . this would transform the triangle into a six - member ring . such a basis becomes unwieldy for typical lattices ) . clearly , the properties of the plaquette are defined by single polaron hopping events , since there are no eigenvalues with @xmath275 terms . it is this property which makes lattices which can be constructed from triangular plaquettes a good starting place when looking for small superlight bipolarons . let us consider a triangular lattice with nearest - neighbour hopping @xmath276 . in the anti - adiabatic approximation , this is replaced by @xmath215 . the polaron band structure ( ignoring polaron shift ) is @xmath277 , where we have defined @xmath278 @xmath279 being the lattice parameter . expanding near the @xmath280 point gives @xmath281 the polaron effective mass is therefore @xmath282 [ kagomefig ] by placing a node on each bond in the lattice , we see that the resulting dimer lattice is a kagome lattice . the dispersion is easily found as follows ( see also ref @xcite ) by diagonalising the secular matrix @xmath283 with @xmath284 the sign in ( [ hkagome ] ) is @xmath285 for the singlet and @xmath286 for the triplet . there are three bipolaron bands with no gaps : @xmath287 with sign @xmath286 for the singlet and @xmath285 for the triplet . the lowest singlet band is @xmath288 with effective mass @xmath289 we notice that this increases _ linearly _ with the polaron mass , indicating that crab - like bipolarons can be relatively light . the lowest triplet band is the _ flat band _ @xmath290 . this implies that a triplet crab - like bipolaron has _ infinite _ mass on a triangular lattice . once crawler motion is permitted , the effective mass is expected to be finite but proportional to @xmath291 . lattices with triangular components are not the only ones where the ability to move between degenerate paired states can be utilised to lead to small masses . for example , if we introduce next - nearest - neighbour hopping to a linear chain , then , in the anti - adiabatic approximation , we obtain a bipolaron mass , @xmath292 and a polaron mass @xmath293 where @xmath294 and @xmath295 . while this does not lead to an exact cancellation of the exponents , the bipolaron to polaron mass ratio is linear in @xmath296 , and the bipolaron is therefore expected to be light . another suggestion for bipolarons with light mass comes from the cuprate lattice . it has often been suggested that the tight - binding structure of the cuprates is a square lattice with nearest - neighbour hopping @xmath276 and next - nearest - neighbour hopping @xmath297 . in the presence of very stong coulomb repulsion @xmath5 , one may determine that the ground state of the lang - firsov transformed hamiltonian consists of nearest neighbour pairs with degeneracy four . if one applies the anti - adiabatic approximation to such a lattice , one determines that the hopping of such a bipolaron is first order in @xmath297 . we obtain a bipolaron mass , @xmath298 and a polaron mass @xmath293 where @xmath299 and @xmath300 . here @xmath301 is the rotation operator . again , while we do nt get an exact cancellation of the exponents , we obtain a resulting mass proportional to @xmath296 , and light bipolarons are likely . such pairing is an appealing prospect , especially since the hague has demonstrated the potential for d - wave superconductivity mediated by holstein - like electron - phonon interactions in the intermediate coupling limit @xcite . the bipolaronic extension @xcite of the bcs theory towards the strong interaction between electrons and ion vibrations proved that the cooper pairing in momentum space @xcite and the ogg - schafroth real - space pairing @xcite are two extreme limits of the same problem . for a very strong electron - phonon coupling , polarons become self - trapped on a single lattice site and bipolarons are on - site singlets . in the holstein model of the zero - range epi their mass appears only in the second order of polaron hopping @xcite , so that on - site bipolarons are very heavy . this estimate led some authors to the conclusion that the formation of itinerant small polarons and bipolarons in real materials is unlikely @xcite , and high - temperature bipolaronic superconductivity is impossible @xcite . in fact , we have demonstrated here that small but light bipolarons could exist for realistic values of the finite - range epi with high - frequency optical phonons in staggered ladder systems . small light bipolarons are an essential precursor to high - temperature superconductivity , since the bose - einstein condensate has transition temperature that is inversely proportional to mass , and wavefunctions may not overlap . such bipolarons are easily formed on lattices with triangular plaquettes in the presence of extremely large on - site coulomb repulsion , and persist to large epi . this conclusion is backed up by analytics in the anti - adiabatic approximation in the presence of large intersite coulomb attraction . another important conclusion is that the triplet - singlet exchange energy is of the first order in the hopping integral , and triplet bipolarons are heavier than singlets in certain lattice structures at variance with simple intuitive expectations . we summarise the types of lattices where light `` crab '' bipolarons may be formed in figure [ fig : pictorial ] , contrasting with the traditional holstein bipolarons ( bottom ) and describe the required effective interaction in figure [ fig : effinteraction ] demonstrating the underlying physics of such bipolarons . our ctqmc simulations lead us to believe that the following recipe is worth investigating to look for very high - temperature superconductivity : ( a ) the parent compound should be an ionic insulator with light ions to form high - frequency optical phonons , ( b ) the structure should be quasi two - dimensional to ensure poor screening of high - frequency c - axis polarized phonons , ( c ) a triangular lattice is required in combination with strong , on - site coulomb repulsion to form the small superlight crab bipolaron ( d ) moderate carrier densities are required to keep the system of small bipolarons close to the dilute regime . many of these conditions are already met in the cuprates . we would like to acknowledge epsrc ( grant numbers ep / c518365/1 and ep / d07777x/1 ) . while writing this article , conversations with serge aubry , peter edwards , janez bona , jozef devresse , holger fehske , yurii firsov , martin hohenadler , viktor kabanov , wolfgang von der linden , peter littlewood , dragan mihailovic , arndt simon , marshall stoneham , and other participants of the european science foundation workshop `` mott s physics in nanowires and quantum dots '' ( cambridge , uk , 31 july-2 august , 2006 ) were especially helpful . we acknowledge support from loughborough university , gonville and caius college , cambridge university , and the esf for this workshop . g. zhao and d. e. morris , phys . b * 51 * , 16487 ( 1995 ) ; g .- m . zhao , m. b. hunt , h. keller , and k. a. mller , nature * 385 * , 236 ( 1997 ) ; r. khasanov , d. g. eshchenko , h. luetkens , e. morenzoni , t. prokscha , a. suter , n. garifianov , m. mali , j. roos , k. conder , and h. keller phys . 92 * , 057602 ( 2004 ) . a. lanzara , p. v. bogdanov , x. j. zhou , s. a. kellar , d. l. feng , e. d. lu , t. yoshida , h. eisaki , a. fujimori , k. kishio , j. i. shimoyana , t. noda , s. uchida , z. hussain , z. x. shen , nature * 2001 * , 412 , 510 ; g. h. gweon , t. sasagawa , s. y. zhou , j. craf , h. takagi , d. h. lee , a. lanzara , nature * 2004 * , 430 , 187 . w. meevasana , n. j. c. ingle , d. h. lu , j. r. shi , f. baumberger , k. m. shen , w. s. lee , t. cuk , h. eisaki , t. p. devereaux , n. nagaosa , j. zaanen , and z .- x . shen , phys . lett . * 96 * , 157003 ( 2006 ) . d. mihailovic , c.m . foster , k. voss , and a.j . heeger , phys . rev . b*42 * , 7989 ( 1990 ) ; p. calvani , m. capizzi , s. lupi , p. maselli , a. paolone , p. roy , s.w . cheong , w. sadowski , and e. walker , solid state commun . * 91 * , 113 ( 1994 ) ; r. zamboni , g. ruani , a.j . pal , and c. taliani , solid st . commun . * 70 * , 813 ( 1989 ) . a. s. alexandrov and j. ranninger , phys . b * 23 * , 1796 ( 1981 ) , ibid * 24 * , 1164 ( 1981 ) . under certain conditions relatively mobile intersite bipolarons can be found even in the holstein - hubbard model , generally unfavourable for coherent tunnelling [ a. la magna and r. pucci , phys . b * 55 * , 14886 ( 1997 ) ; l. proville and s. aubry , eur . phys . j. b * 11 * , 41 ( 1999 ) ; j. bona , t. katrasnic , and s. a. trugman , phys . rev . lett . * 84 * , 3153 ( 2000 ) ; a. macridin g. a. sawatzky , and m. jarrell , phys . rev . b * 69 * , 245111 ( 2004 ) ] .

we have derived an x - ray luminosity function using parallax - based distance measurements of a set of 12 dwarf novae , consisting of _ suzaku _ , _ xmm - newton _ and _ asca _ observations . the shape of the x - ray luminosity function obtained is the most accurate to date , and the luminosities of our sample are concentrated between 10 erg s , lower than previous measurements of x - ray luminosity functions of dwarf novae . based on the integrated x - ray luminosity function , the sample becomes more incomplete below 3 10 erg s than it is above this luminosity limit , and the sample is dominated by x - ray bright dwarf novae . the total integrated luminosity within a radius of 200 pc is 1.48 10 erg s over the luminosity range of 1 10 erg s and the maximum luminosity of the sample ( 1.50 10 erg s ) . the total absolute lower limit for the normalised luminosity per solar mass is 1.81 10 erg s m which accounts for 16 per cent of the total x - ray emissivity of cvs as estimated by . [ firstpage ] cataclysmic variables stars : dwarf novae x - rays : stars x - rays : binaries stars : distances stars : luminosity function

1006.5932

cataclysmic variables , i.e. cvs consist of an accreting white dwarf primary and a late - type main sequence star , and accrete via roche lobe overflow . cvs can be divided into several subclasses of which so called dwarf novae ( dne ) are the most numerous subclass of cvs in our galaxy . in these systems , the white dwarf has a weak magnetic field strength ( b @xmath9 10@xmath10 g , * ? ? ? * ) compared to magnetic cvs , such as polars , and thus the formation of an accretion disc is possible . from time to time , the disc is seen to brighten by several magnitudes lasting from days to several weeks . this brightening of the disc , i.e. an outburst , is thought to be caused by disc instabilities , which are described in detail in @xcite . in quiescence , dne are sources of optical emission emanating from the accretion disc and the bright spot where the material from the secondary hits the edge of the disc . quiescent optical spectra of dne are characterized by strong balmer emission lines and weaker he i lines with some heavier elements . also , dne are sources of hard x - rays which are thought to originate from an optically thin boundary layer during quiescence . however , during an outburst hard x - rays are quenched as the boundary layer becomes optically thick and thus a source of soft x - rays and euv emission @xcite . at the time of the discovery of the galactic ridge x - ray emission ( grxe ) in 1982 @xcite , discrete point sources were thought to be the origin of the grxe emission . however , the origin has been debated since , but observational evidence gathered to date since the grxe discovery supports the view that the grxe is not due to diffuse origin but due to discrete point sources , such as cvs and other accreting binary systems ( see recent studies by e.g. * ? ? ? * ; * ? ? ? more supporting evidence was given by the recent _ chandra _ study carried out by @xcite who resolved over 80 per cent of the grxe into point sources in the 67 kev energy range during an ultra - deep 1 msec observation . based on _ exosat _ observations of the x - ray emission in the galactic plane , @xcite concluded that if the grxe is assumed to be originating from discrete point sources , the bulk of the emission observed must be due to a population of low luminosity x - ray sources ( l@xmath11 @xmath12 10@xmath13 erg s@xmath3 ) , such as cvs . subsequently , @xcite suggested that dne could significantly contribute to the grxe based on their study of an _ exosat _ medium energy ( me ) dn sample . according to this study , the space density of dne is sufficiently high to account for a significant fraction of the grxe . later on , @xcite resolved sources down to 3 @xmath4 10@xmath14 erg @xmath15 s@xmath3 in their _ chandra _ observation of the galactic ridge , equivalent to @xmath16 @xmath17 2.3 @xmath4 10@xmath2 erg s@xmath3 at 8 kpc , concluding that the number of resolved point sources above this level is insufficient for them to be the major contributor to the grxe . since these previous works have not completely resolved the contribution of cvs to the grxe , further studies are needed . as was noted by @xcite , unbiased and sensitive surveys with accurate distance measurements of cvs are needed . this way , accurate x - ray luminosity functions ( xlfs ) can be obtained , and the contribution to the grxe estimated more precisely . the motivation for our work was mainly given by the inaccuracies in the xlfs of galactic cv populations , such as those by @xcite and @xcite . baskill et al . derived an xlf using 34 _ asca _ observations of non - magnetic cvs ( including 23 dn observations ) . their sample lacked accurate distance measurements as only 10 sources had parallax - based distance measurements . furthermore , this sample was biased by high x - ray flux sources since _ asca _ was intended to be a spectroscopic mission and the sources in the studied sample were known to be x - ray bright . also , the _ asca _ study was purely archival without any sample selection ( e.g. the distance was not limited ) as baskill et al . chose all non - magnetic cv observations in the archive , they did not filter out sources which were in an outburst state , or restrict the study to one type of objects only . the xlf study by @xcite focused on building up an xlf in the 210 kev range combining the _ rxte _ slew survey ( xss ) and _ rosat _ all - sky survey ( rass ) observations of active binaries , cvs and young main sequence stars in the luminosity range @xmath0 10@xmath18 @xmath12 @xmath16 @xmath12 10@xmath19 erg s@xmath3 . however , uncertainties in the luminosities in this study were introduced by inadequate accuracies in the distances , for example , many of the intermediate polars ( ips ) in their sample had poorly known distances . only a few sources had parallax measurements from , e.g. , the hipparcos or tycho catalogues ( astrometric uncertainties @xmath0 1 mas ) and ground - based parallax measurements from @xcite . in addition , the rass luminosities had 50 per cent uncertainties in addition to statistical errors after conversion from the 0.12.4 kev to the 210 kev range . the primary aim of this paper is to derive the most accurate shape of the xlf to date by using a carefully selected sample of dne . in order to achieve this , we aim to minimise the biases seen in other published xlfs by using sample selection criteria described in section [ sourcesample ] . one of the criteria worth mentioning here is that we only use parallax - based distance estimates . the source sample used in this study does not represent a complete sample of dne within 200 pc : the sample is more of a fair sample which was not chosen based on the x - ray properties of the sources , and which represents typical dne within the solar neighbourhood . we will consider the possible effects of the optical selection , inherent in the parallax sample , on the xlf which we derive in section 6 ( discussion ) . the motivation for choosing a group of dne as the sample is based on observations of the galactic cv populations and previous cv population models . various authors , such as @xcite , @xcite and @xcite , have pointed out that binaries which are brighter in x - rays and show frequent outbursts , may not represent the true majority of galactic cv population . the discovery methods of cvs usually favour brighter systems , and thus fainter cvs with low mass accretion rates are likely to be under - represented . however , population models of cvs predict that the majority of cvs are short - period systems ( @xmath20 @xmath12 2.5 h ) and x - ray faint ( e.g. * ? ? ? * ; * ? ? ? these models are supported by observational evidence , e.g. @xcite showed that a sample of cvs with a total space density of 6 @xmath4 10@xmath21 pc@xmath22 was dominated by low mass accretion rate , and thus short period , systems . also , the sdss study by @xcite showed that orbital periods of intrinsically faint galactic cvs accumulated in the 8086 min range ; they found that 20 out of 30 sdss cvs in this period range showed characteristics which implied that they are low mass accretion rate wz sge type dne . as has been shown by these studies , less x - ray luminous objects ( such as dne ) dominated the studied volumes , and thus we choose to focus on dne in this paper . it is also worth mentioning the study by @xcite who carried out the _ rosat _ north ecliptic pole ( nep ) survey using a purely x - ray flux limited and a complete sample of 442 x - ray sources above a flux limit of @xmath0 10@xmath23 erg @xmath15 s@xmath3 in the 0.52.0 kev band ( only five systems were cvs ) . they concluded that if the overall space density of cvs is as high as 2 @xmath4 10@xmath24 pc@xmath22 , then the dominant cv population must be fainter than 2 @xmath4 10@xmath25 erg s@xmath3 . we have carried out x - ray spectral analysis of our sample of 13 sources and derived an xlf in the 210 kev band for 12 of them with reliable distance measurements based on those of by @xcite , @xcite and @xcite . by using sources with accurate distance measurements , we minimise the error on the luminosity . also , we have carried out timing analysis for 5 sources in the sample which were recently observed by _ suzaku_. at the time of writing this paper , the z cam type star kt per went into an outburst in january 2009 , and thus we also briefly report on the _ suzaku _ observations of kt per during the outburst in section [ sec : ktper ] . since we wanted to obtain accurate luminosities for the sources ( and thus an accurate shape for the luminosity function ) , the first step was to avoid selecting sources randomly from the archive ( see e.g. * ? ? ? * ) or selecting an x - ray flux limited source sample . the aim was to have a _ distance - limited _ sample . thus , sources were not selected based on their x - ray properties , but we chose only those dne which have accurately measured distances based on trigonometric parallax measurements within @xmath0 200 pc . note that by using all available distance measurement techniques , patterson ( priv . comm . ) estimates that currently there are 13 dne within 100 pc , and @xmath0 33 dne within 200 pc from the sun , of which the latter count is clearly incomplete . above the 200 pc limit , ground - based parallax technique does not give accurate and reliable distance measurements . by using trigonometric parallax - based distance measurements , we are more likely to avoid biases in the distance measurements which are present in the previous , published x - ray luminosity functions . due to the lack of ground - based parallax measurement programme for the southern hemisphere , our sample is limited to northern and equatorial objects . however , this selection should not introduce any biases in terms of the optical or x - ray luminosities in our sample . the distance measurements of the sources chosen for this work are based on astrometric parallaxes obtained by the _ hubble space telescope _ ( hst ) fine guidance sensors ( fgss ) @xcite , and trigonometric parallaxes obtained by the ground - based 2.4 m hiltner telescope at the mdm observatory on kitt peak , arizona @xcite and thorstensen ( in prep . ) . the first accurate astrometric parallaxes of dne ( ss cyg , ss aur and u gem ) were measured in 1999 using the fgss which can deliver high - precision parallaxes with sub - milliarcsecond uncertainties @xcite . trigonometric parallaxes derived by ground - based observations have uncertainties around 1 mas (= 10@xmath22 arcsec ) or less @xcite which is almost as good as the uncertainty on the fgs parallax measurements . the second selection criterion was to restrict the sample to sources which had been observed by x - ray imaging telescopes with ccds in the energy range 0.210 kev . once we had obtained a list of targets with parallax measurements , we then looked for archival data of pointed imaging x - ray observations of these targets in the energy range @xmath0 0.210 kev . if the chosen targets did not have previous x - ray imaging observations , we requested _ suzaku _ x - ray observations . finally , we wanted to constrain the sample to those sources which were in their quiescent states during the observations in order to avoid biases in the luminosities , and thus _ aavso _ light curves of the selected sources were inspected to confirm that the sources were in quiescence during the x - ray observations . the final source sample consists of 9 su uma ( including 2 wz sge systems ) , 3 u gem and 1 z cam type dne . the main characteristic which separates these classes of dne is the outburst behaviour : u gem type dne outburst mainly in timescales of every few weeks to every few months whereas su uma stars show normal , u gem type dn outbursts and , in addition , superoutbursts with superhumps ( variations in the light curves at a period of a few per cent longer than the orbital period ) in timescales of several months to years . the extreme cases , wz sge stars , only have superoutbursts with outburst timescales of decades without normal dn outbursts . the defining characteristic for z cam stars is standstills , i.e. , it is possible that after an outburst , they do not return to the minimum magnitude ( unlike u gem stars ) , but remain between the minimum and maximum magnitudes for 1040 days . [ cols="^,^,^,^,^,^ " , ] _ suzaku _ @xcite , originally _ astro - e2 _ , was launched in 2005 and is japan s 5th x - ray astronomy mission . in this paper , we will focus on the x - ray imaging spectrometer ( xis ) data . the xis consists of four sensors : xis0,1,2,3 of which three ( xis0,2,3 ) contain front - illuminated ( fi ) ccds , and xis1 contains a back - illuminated ( bi ) ccd . the xis0,2,3 are less sensitive to soft x - rays than xis1 due to the thin si and sio@xmath26 layers on the front side of the xis0,2,3 ccds . since november 9 , 2006 , the xis2 unit has not been available for observations . the _ suzaku _ background is low and hardly affected by soft proton flares often seen in _ xmm _ observations . the unfiltered event lists of ss cyg and v893 sco were reprocessed with xispi and screened in xselect with xisrepro since the pipeline version for these observations was older than v.2.1.6.15 which does not include correction for the time- and energy - dependent effects in energy scale calibration . for all the other _ suzaku _ observations , the observations had been processed by more recent pipeline versions and thus reprocessing was not necessary . pile - up was not a problem for our data since the source count rates were safely below the pile - up limit ( 12 ct s@xmath3 ) for point sources observed in the normal mode using full window . the _ suzaku _ data reduction described below was carried out in a similar manner for all the _ suzaku _ observations . the cleaned event lists were read into xselect in which x - ray spectra were extracted for each source . light curves were extracted for sw uma , bz uma , ss aur , asas j0025 , and vy aqr for timing analysis studies . to include 99 per cent of the flux and to obtain the most accurate flux calibration , the spectra and light curves were extracted using a source radius of 260 ( 250 pixels ) . the backgrounds were taken as an annulus centred on the source excluding the inner 4 source region . the outer radii of the background annuli were determined according to how close the calibration sources were to the target . the response matrix files ( rmfs ) and ancillary response files ( arfs ) were created and combined within xisresp v.1.10 . the xis0,2,3 source and background spectra and the corresponding response files were summed in addascaspec to create the total xis0,2,3 source and background spectra , and the total xis0,2,3 response file . for the light curves , the background areas were scaled to match the source areas , and the scaled background light curves were then subtracted from the source light curves in lcmath . _ xmm - newton _ @xcite is the cornerstone mission of the european space agency ( esa ) . it has been operating since 1999 with three x - ray cameras ( epic pn , mos1 and mos2 ) , the optical monitor ( om ) , and the reflection grating spectrometer ( rgs ) onboard . the x - ray cameras cover the energy range 0.212.0 kev . the data were obtained from the _ xmm - newton _ science archive ( xsa ) and the _ xmm - newton _ data were reduced and analysed in the standard manner using the _ xmm - newton _ science analysis system sas version 8.0.0 . each observation was checked for high background flares in the range 1012 kev using single pixel events ( pattern = = 0 ) . the high background flares were cut above 0.35 ct s@xmath3 for the mos data and above 0.40 ct s@xmath3 for the pn data . the source and background extraction regions were taken from circular extraction areas avoiding any contaminating background sources . the radii of the source regions were calculated by using the sas task region in order to derive source extraction radii which include @xmath0 90 per cent of the source flux for each source . the background extraction region ( r@xmath27 = 130 arcsec ) was taken from the same chip as the source extraction region , or from an adjacent chip in case of a crowded source chip . when extracting the x - ray spectra , only well - calibrated x - ray events were selected for all the sources , i.e. for the pn spectra single and double pixel events with pattern @xmath28 4 were chosen , and in order to reject bad pixels and events close to ccd gaps , flag = = 0 was used . for the mos , pattern @xmath28 12 and # xmmea_em were applied . for most of the observations , we analysed the pn observations only . since the total effective area of the two epic mos cameras is nearly equal to the effective area of the epic pn , the mos spectra would not add any significant information to the pn spectra . the only exception was u gem pn observation which had been obtained in small window mode . thus , we used the mos1 and 2 data which had been obtained in large and small window modes , respectively , and selected the backgrounds from the surrounding ccds . to form the total mos spectrum for u gem , the u gem mos1 and 2 spectra were summed in addascaspec . all the observations were checked in case of pile - up by using the _ xmm _ sas task epatplot . pile - up did not occur in any of the observations , but the source psfs of ht cas and t leo were contaminated by out - of - time ( oot ) events , introduced by these two sources . therefore , the background regions in the ht cas and t leo observations were taken from the adjacent chip in order to avoid the oot events . these oot events were removed from the source x - ray spectra according to the sas threads , i.e. the source spectra extracted from the oot event lists were subtracted from the source spectra extracted from the original event list . the _ advanced satellite for cosmology and astrophysics _ ( _ asca _ , * ? ? ? * ) was japan s fourth cosmic x - ray astronomy mission operating between february 1993 and july 2000 and was the first x - ray observatory which carried ccd cameras . the main science goal of _ asca _ was the x - ray spectroscopy of astrophysical plasmas . it carried four x - ray telescopes with two types of detectors located inside them : two ccd cameras , i.e. the solid - state imaging spectrometers ( sis0 and sis1 ) with spectral resolution of 2 per cent at 5.9 kev at launch , and two scintillation proportional counters , i.e. the gas imaging spectrometers ( gis2 and gis3 ) . the _ asca _ data reduction was performed in the standard manner by mostly using the standard screening values for the gis and sis instruments as described in nasa _ asca _ online manual . for both instruments , intervals outside the south atlantic anomaly ( saa ) were chosen , also including intervals when the attitude control was stable with the upper limit of the angular distance from the target set to ang_dist @xmath12 0.02 degrees . for the sis instruments , the bright earth angle of br_earth @xmath17 10 was applied excluding the data taken below the 10@xmath29 angle . times of high background were excluded when the pixl monitor count rate was 3@xmath30 above the mean of the observation . also , the background monitor count rate of rbm_cont @xmath12 500 was applied ( the standard screening value is rbm_cont @xmath12 100 ) . events which occurred before the first day - night transition and at least 32 seconds after the day - night transition , and also before the passage of the south atlantic anomaly ( saa ) and at least 32 s after the saa , were selected ( t_dy_nt < 0 @xmath31 t_dy_nt > 32 & & t_saa < 0 @xmath31 t_saa > 32 ) . the source extraction regions for the sis and gis were centred on the source . for the gis , a @xmath0 6 arcmin circular source extraction region was used for both z cam and wz sge , while the sis source extraction regions were smaller so that they could be safely fitted within the chip . thus , the source extraction radii for z cam were 4.4 arcmin ( sis0 ) and @xmath0 3.5 arcmin ( sis1 ) and for wz sge @xmath0 4 arcmin ( sis0 ) and @xmath0 3 arcmin ( sis1 ) . the background extraction region for the gis was taken centred on the detector excluding a circular region of @xmath0 8 arcmin centred on the source . the background extraction radii for gis2 and gis3 were 15.7 and 15.5 arcmin ( z cam ) , and 15.4 and 13.8 arcmin ( wz sge ) respectively . for the sis background , blank - sky background observations were used for z cam due to lack of space for a local background region on the ccds . for wz sge , the total area of the two active ccds excluding a 5.5 arcmin region around the source was used as the background extraction region . the ancillary response files ( arfs ) for the gis and sis spectra were created with ascaarf and the sis response matrix files ( rmfs ) with sisrmg . finally , the total sis and gis x - ray spectra were created by combining sis0 and sis1 , and the gis2 and gis3 spectra in addascaspec , respectively . since vy aqr , ss aur , bz uma , sw uma and asas j0025 have not been subject to previous , pointed , imaging x - ray observations before the _ suzaku _ observations , we looked for periodicities from the data of these sources . kt per has been observed by the _ einstein _ observatory @xcite , but no previous x - ray spectral or timing analysis studies have been carried out for it . in order to ensure that these objects are not intermediate polars ( ips ) and to look for orbital and spin modulation in the data , the power spectra were calculated by using a lomb - scargle periodogram @xcite which is used for period analysis of unevenly spaced data . when searching over the frequency range 0.000010.03 hz , no significant periodicities were seen at the 99 per cent confidence level . we carried out x - ray spectral analysis in order to study the underlying spectra of the source sample , and , ultimately , to calculate the fluxes and luminosities of the sources . to employ gaussian statistics , the x - ray spectra were binned at 20 ct bin@xmath3 with grppha and then fitted in xspec11 @xcite . in cvs , the power source of x - ray emission is known to be accretion onto the white dwarf . the accreted material is shock - heated to high temperatures ( @xmath32 @xmath0 1050 kev , * ? ? ? * ) , and this material has to cool before settling onto the white dwarf surface . thus , the cooling gas flow is assumed to consist of a range of temperatures which vary from the hot shock temperature @xmath32 to the temperature of the optically thin cooling material which eventually settles onto the surface of the white dwarf @xcite . thus , when fitting x - ray spectra of cvs , cooling flow spectral models should represent more physically correct picture of the cooling plasma , unlike single temperature spectral models . cooling flow models have successfully been applied to cv spectra in previous studies by e.g. @xcite and @xcite . in this view , the multi - temperature characteristic is our motivation for emphasizing the cooling flow model in the rest of this work . the differential emission measure dem / dt for an isobaric cooling flow can be described by @xcite @xmath33 where @xmath34 is the mass of a proton , @xmath35 the mean molecular weight ( @xmath0 0.6 ) , @xmath36(t , n ) total emissivity per volume in units of erg s@xmath3 cm@xmath22 , @xmath38 accretion rate , @xmath39 particle density , and @xmath40 the boltzmann constant . the source of the x - ray emission above the white dwarf illuminates the surface of the white dwarf and thus causes a reflection , which is seen as fe k@xmath41 iron fluorescence line at 6.4 kev @xcite . according to george & fabian , an infinite slab reflector subtending a total solid angle of @xmath42 = 2@xmath43 where the x - ray source is located right above the slab , produces an equivalent width of up to @xmath0 150 ev for the 6.4 kev fe k@xmath41 fluorescence line . the equivalent width of the 6.4 kev iron line depends on the total abundance of the reflector @xcite , the inclination angle between the surface of the reflector and the observer s line of sight , and the photon index of the spectrum of the x - ray emission source @xcite . ccc + name & ew(mekal ) & ew(mkcflow ) + & ev & ev + bz uma & 67@xmath44 & @xmath12 79 + + ht cas & @xmath12 81 & @xmath12 91 + + ss aur & 73@xmath45 & 86@xmath46 + + sw uma & 201@xmath47 & @xmath12 141 + + u gem & 50@xmath48 & 60@xmath49 + + t leo & 71@xmath50 & @xmath12 73 + + v893 sco & 46@xmath51 & 45@xmath52 + + vy aqr & @xmath12 156 & @xmath12 157 + + wz sge & @xmath12 140 & @xmath12 76 + + ss cyg & 75@xmath53 & 73@xmath54 + + z cam & 120@xmath44 & 164@xmath55 + + asas j0025 & @xmath12 220 & @xmath12 200 + [ ev ] even though we believe that the cooling flow -type multi - temperature model is the correct description of the physics of the cooling gas flow in cvs , previous works have often used single temperature plasma models . thus , in order to compare the effects of two different spectral models on the spectral fit parameters , we fitted the spectra with 1 ) a single temperature optically thin thermal plasma model ( ` mekal , ` * ? ? ? * ; * ? ? ? * ) and 2 ) a cooling flow model ( ` mkcflow ` ) which was originally developed to describe the cooling flows in clusters of galaxies @xcite , adding photoelectric absorption ( ` wabs ` , * ? ? ? * ) to both models . in order to investigate the equivalent width of the 6.4 kev iron emission line , a gaussian line was added at 6.4 kev with a line width fixed at @xmath30 = 10 ev . the spectral fits did not necessarily require the 6.4 kev line , e.g. , for ss aur the @xmath56/@xmath57 = 0.96/629 when a gaussian line at 6.4 kev was not included . cccccccc + name & @xmath58 & @xmath59 & cfrac & kt & ab & @xmath56/@xmath57 & p@xmath60 + & 10@xmath61 @xmath15 & 10@xmath61 @xmath15 & & kev & @xmath62 & & + bz uma & @xmath120.19 & & & 4.17@xmath63&0.51@xmath64 & 1.02/905 & 0.349 + + gw lib & 3.16@xmath65&&&1.62@xmath66&0.20@xmath67 & 1.02/8 & 0.414 + + ht cas & &15.36@xmath68 & 0.95@xmath69 & 6.43@xmath70 & 0.71@xmath71 & 1.07/259 & 0.207 + + ss aur & @xmath120.56 & &&6.35@xmath72&1.0@xmath73 & 1.06/628 & 0.127 + + sw uma & @xmath120.20 & &&2.77@xmath74&0.20@xmath75 & 1.23/470 & 5.36@xmath410@xmath24 + + u gem & 0.89@xmath76&&&0.78@xmath77&1.05@xmath78&1.34/401 & 5.80@xmath410@xmath21 + & & & & 6.85@xmath79 & & & + + t leo & 1.09@xmath80&&&3.55@xmath81&0.50@xmath82 & 1.40/631 & 8.36@xmath410@xmath83 + + v893 sco & &80.89@xmath84&0.86@xmath85&7.99@xmath86&0.76@xmath87 & 1.02/1936 & 0.245 + + vy aqr & @xmath121.64 & &&5.06@xmath88&0.66@xmath71 & 0.90/445 & 0.942 + + wz sge & 8.97@xmath89&&&4.88@xmath90&0.33@xmath91&0.84/409 & 0.993 + + ss cyg & 2.98@xmath92&&&10.44@xmath63&0.51@xmath93&1.24/2881 & 2.74@xmath410@xmath94 + + z cam & 28.21@xmath95 & 292.41@xmath96&0.35@xmath87&8.68@xmath97 & 1.0 & 1.10/769 & 0.03 + + asas & @xmath120.84 & & & 4.38@xmath98&0.56@xmath99 & 0.88/366 & 0.958 + j0025 & & & & & & & + [ mekal ] cccccccc + name & @xmath58 & @xmath59 & cfrac & kt@xmath100 & ab & @xmath56/@xmath57 & p@xmath60 + & 10@xmath61 @xmath15 & 10@xmath61 @xmath15 & & kev & @xmath62 & & + bz uma & @xmath120.87 & & & 13.71@xmath101 & 0.57@xmath102 & 0.88/904 & 0.994 + + gw lib & @xmath12 3.76 & &&6.96@xmath103&1.0 & 0.60/8 & 0.782 + + ht cas & &16.74@xmath104 & 0.92@xmath105 & 23.09@xmath106 & 0.78@xmath107 & 0.99/258 & 0.525 + + ss aur & 3.30@xmath108&& & 23.47@xmath109 & 1.0 & 0.95/628 & 0.832 + + sw uma & @xmath120.67 & & & 8.33@xmath110 & 0.41@xmath111 & 0.87/469 & 0.978 + + u gem & 0.76@xmath112&&&25.82@xmath113&1.04@xmath114 & 1.23/402 & 1.1@xmath410@xmath22 + + t leo & 0.68@xmath115&&&10.97@xmath116 & 0.50@xmath117 & 1.08/629 & 7.24@xmath410@xmath118 + + v893 sco & &103.71@xmath119&0.90@xmath85&19.32@xmath120&0.94@xmath121 & 0.94/1934 & 0.973 + + vy aqr & 1.10@xmath122&& & 16.47@xmath123 & 0.69@xmath124 & 0.86/444 & 0.984 + + wz sge & 11.58@xmath125&&&13.31@xmath126 & 0.23@xmath127 & 0.83/408 & 0.996 + + ss cyg & 2.84@xmath128&&&41.99@xmath129&0.61@xmath130 & 1.19/2883 & 4.80@xmath410@xmath131 + + z cam & 31.92@xmath132 & 180.28@xmath133&0.47@xmath134&25.76@xmath135 & 1.0 & 1.08/768 & 0.06 + + asas & @xmath136 & & & 14.43@xmath137 & 0.68@xmath138 & 0.81/366 & 0.996 + j0025 & & & & & & & + [ mkcflow ] the _ suzaku _ xis1 and xis0,2,3 spectra were fitted simultaneously for each source as well as the _ asca _ gis and sis spectra of z cam and wz sge with the models mentioned above . some data sets required additional components to improve the fits . three of the sources , ht cas , v893 sco and z cam , required partial covering absorption model , ` pcfabs ` , to reduce residuals in the low energy end ( between @xmath0 0.62 kev ) . to reduce residuals around 0.80 kev in the ss cyg spectrum , we added a gaussian line at 0.81 kev with a line width of 0.24 kev letting the line energy and width both to vary free . for u gem , single absorbed optically thin thermal plasma model yielded a @xmath139/@xmath57 = 2.23/403 . since the fit was not statistically satisfactory , we added a second optically thin thermal plasma component to improve the fit and obtained @xmath139/@xmath57 = 1.34/401 which was good enough for our analysis . in the spectral fitting , the parameters of the spectral models were tied between different instrument spectra but let to vary free , apart from the gaussian line energy at 6.4 kev and the line width @xmath30 which were fixed . in order to estimate the abundances , the abundance parameter of the models was let to vary free for most data sets . for those sources for which abundance was significantly higher than the solar value , it was fixed at 1.0 . an example of a source with a super - solar abundance is z cam for which the obtained abundance was 1.46@xmath140 @xmath141 with the partial covering + photoelectric absorption combined with the cooling flow model when the abundance was let to vary free . the measured equivalent widths of the 6.4 kev line for each source are given in table [ ev ] , and the results of the spectral fitting for the optically thin thermal plasma and the cooling flow models are given in table [ mekal ] and [ mkcflow ] , respectively . these results show that in general , better @xmath56/@xmath57 values are achieved with the cooling flow model . for example , the improvement with the cooling flow model was statistically significant for sw uma and t leo . [ spectra ] illustrates the x - ray spectra of the new _ suzaku _ xis observations , i.e. vy aqr , sw uma , bz uma , ss aur , and asas j0025 , which have been fitted with the cooling flow model absorbed by photoelectric absorption with an added 6.4 kev gaussian line component . most of the x - ray spectra show that the clearest , discrete emission feature seen in the spectra of our source sample is the iron fe k@xmath41 complex , except in gw lib , for which the signal - to - noise at @xmath0 6 kev is too low for a reliable measurement . since the studied sources are all within @xmath0 200 pc , i.e. , within the solar neighbourhood , the effect of interstellar absorption should be negligible . thus , high measured absorption columns would mainly be due to intrinsic absorption , associated with the sources . for most of the sources , the measured absorption columns were typically of the order of a few @xmath4 10@xmath61 @xmath15 , or even lower ( 10@xmath142 @xmath15 ) which indicate low intrinsic absorption . the highest intrinsic absorption columns are found in v893 sco , z cam and ht cas when compared to the rest of the source sample . all these three sources have partial covering absorbers @xmath59 with values of the order of 10@xmath143 10@xmath144 @xmath15 depending on the model . in addition to the partial covering absorber , z cam also has a simple absorption component with the highest @xmath58 value , @xmath58 @xmath0 3 @xmath4 10@xmath143 @xmath15 , within the source sample . originally , v893 sco was found to have high intrinsic absorption by @xcite , and has a partial x - ray eclipse , also discovered by their study . also , according to the best - fit model of @xcite , z cam had large amounts of absorption with @xmath145 = 9 @xmath4 10@xmath143 @xmath15 during the transition state . baskill et al . suggested that this absorption was associated with a clumpy disc wind . the measured shock temperatures @xmath32 seem to be correlated with the white dwarf masses ( fig . [ mass_temp ] ) as one would expect . in fig . [ mass_temp ] it has been assumed that the white dwarf mass of vy aqr is 0.8 m@xmath146 ( see table [ sources ] ) . asas j0025 is not included in fig . [ mass_temp ] since the mass estimate is currently unknown . ss cyg appears to be located in the upper right corner due to its high - mass white dwarf and thus high shock temperature . the shock temperatures , @xmath32 , in fig . [ mass_temp ] have been derived from the spectral fits of the cooling flow model for each source . the blue dashed line in fig . [ mass_temp ] represents the theoretical shock temperatures for given white dwarf masses . the radii , @xmath147 , of the given white dwarf masses , @xmath148 , were calculated assuming the mass - radius relation for cold , non - rotating and non - relativistic helium white dwarfs ( see * ? ? ? * ) @xmath149 where @xmath150 = @xmath151 -1 . subsequently , the theoretical shock temperatures , @xmath152 , for non - magnetic cvs were calculated according to eq . [ eq : tshock ] for optically thin gas @xcite @xmath153 where @xmath154 is the mass of a hydrogen atom , @xmath35 the mean molecular weight , and @xmath40 the boltzmann constant . as it appears from fig . [ mass_temp ] , sources with high shock temperatures and low luminosities are not seen . this is sensible since the x - ray luminosity is proportional to @xmath32 and the mass accretion rate , i.e. the normalization of the cooling flow model ( eq . [ eq : cool ] ) , thus we would expect to see high shock temperatures and high luminosities . also , due to this proportionality , we expect to see an anti - correlation between @xmath38 and @xmath32 which indeed is seen for example in sw uma ( fig . [ 2dim ] ) . , of the source sample with 90 per cent uncertainties for @xmath32 . the light blue dashed line represents the theoretical shock temperatures for given white dwarf masses . the figure does not include asas j0025 since a mass estimate does not currently exist for this source.,width=302 ] as it appears from fig . [ mass_temp ] , the white dwarf mass obtained for t leo by @xcite is only 0.35 m@xmath146 . this mass estimate may be unreliable , since as @xcite argue , a low white dwarf mass would not allow superhumps to develop . see also @xcite who refer to previous superhump studies which have shown that the mass ratio @xmath155 = @xmath156 has a key role in producing superhumps where @xmath157 @xmath0 0.3 , although this value has not been confirmed by observational evidence . we found that for most of the objects in the sample the obtained abundances were sub - solar with both models . in general , the abundances seem to be dependent on the spectral model : abundances are slightly lower when the spectra are fitted with the optically thin thermal plasma model . this is due to the single temperature characteristic of the optically thin thermal plasma model , i.e. it is likely that the abundances are underestimated because the best - fit temperature usually converges close to the peak of the 6.7 kev he - like fe k@xmath41 emissivity , whereas the cooling flow model consists of a range of temperatures outside the peak ( see * ? ? ? the 210 and bolometric 0.01100 kev fluxes and luminosities which were derived using the cooling flow model are given in table [ fluxes ] . this shows that most of the 210 kev x - ray luminosities are concentrated around 10@xmath1 erg s@xmath3 . this is also seen in fig . [ histo ] which shows a histogram of the x - ray luminosities of our sample . only one object , gw lib , stands out with a very low luminosity ( 4 @xmath4 10@xmath6 erg s@xmath3 ) . the measured luminosity of gw lib is consistent with the results obtained by @xcite . @xcite showed that gw lib was still an order of a magnitude brighter ( l @xmath0 10@xmath1 erg s@xmath3 ) in x - rays during _ swift _ observations two years after the 2007 outburst than in 2005 . but since the optical magnitude had not reached the quiescence level ( v @xmath0 18 ) in 2009 , we do not consider the _ swift _ 2009 x - ray luminosity as the quiescent luminosity . thus , the higher x - ray luminosity measured in the _ swift _ data does not affect the results of this paper . ccccc + source & f(210 kev ) & l(210 kev ) & f(0.01100 kev ) & l(0.01100 kev ) + & x 10@xmath131 erg @xmath15 s@xmath3 & @xmath4 10@xmath1 erg s@xmath3 & x 10@xmath131 erg @xmath15 s@xmath3 & @xmath4 10@xmath1 erg s@xmath3 + bz uma & 2.4@xmath158 & 14.9@xmath159 & 5.8 & 36.5 + + gw lib & 0.04@xmath77 & 0.05@xmath160 & 0.1 & 0.1 + + ht cas & 2.9@xmath161 & 6.1@xmath162 & 7.2 & 14.9 + + ss aur & 2.9@xmath163 & 9.6@xmath164 & 7.1 & 23.9 + + sw uma & 1.5@xmath165 & 4.9@xmath166 & 4.2 & 13.7 + + u gem & 6.9@xmath163 & 8.3@xmath167 & 17.1 & 20.6 + + t leo & 5.2@xmath163 & 6.4@xmath168 & 13.2 & 16.3 + + v893 sco & 17.3@xmath169 & 50.1@xmath170 & 45.7 & 133.0 + + vy aqr & 1.1@xmath171 & 1.3@xmath172 & 2.6 & 3.0 + + wz sge & 3.1@xmath173 & 0.7@xmath174 & 7.7 & 1.8 + + ss cyg & 45.7@xmath175 & 150.0@xmath176 & 131.7 & 433.0 + + z cam & 19.2@xmath177 & 61.6@xmath178 & 52.3 & 168.0 + + asas j0025 & 0.4@xmath179 & 1.6@xmath180 & 1.1 & 3.9 + [ fluxes ] one of the sources in our sample , ss aur , has previously been listed in the _ rxte _ all - sky slew survey catalog where it appears more luminous in x - rays than in our _ suzaku _ observation ( the _ rxte _ flux of ss aur in 210 kev is @xmath0 1.1 @xmath4 10@xmath83 erg @xmath15 s@xmath3 ) . we suspect that the higher flux in the _ rxte _ observation is due to other , bright sources in the field which overestimate the flux . e.g. , the _ rosat _ bright source catalogue lists a cluster of galaxies , abell 553 , which is 53 away from ss aur and has a webpimms estimated flux of @xmath0 9.3 @xmath4 10@xmath131 erg @xmath15 s@xmath3 in 210 kev ( bremsstrahlung kt = 4 kev , galactic n@xmath181 = 1.56 @xmath4 10@xmath143 @xmath15 as in * ? ? ? * ) . thus , the higher _ rxte _ flux of ss aur is very likely biased by the background sources and not reliable . we also analysed the _ suzaku _ outburst data of kt per obtained in january 2009 , and report the results here . kt per is a z cam type dwarf nova , and was seen as an x - ray source by the _ einstein _ satellite in 1979 @xcite . we employed the same models which were used for the source sample above , i.e. an absorbed optically thin thermal plasma model and an absorbed cooling flow model with an added 6.4 kev line . both models yielded acceptable fits : @xmath182 = 0.97/838 ( thermal plasma ) and @xmath182 = 0.95/837 ( cooling flow ) . [ ktper_fig ] shows the xis1 and the combined xis0,3 x - ray spectra of kt per which have been fitted with an absorbed cooling flow model with a 6.4 kev gaussian line . the spectral fit parameters for the optically thin thermal plasma and the cooling flow models with fluxes , luminosities and fit statistics are given in table [ ktper ] . the luminosities given in table [ ktper ] are calculated for the distance of 180@xmath183 pc @xcite . @xcite noted that cooling flow models are often a good representation of quiescent x - ray spectra of cvs ( see also * ? ? ? * ) , but not outburst spectra . baskill et al . applied the xspec multi - temperature model ` cevmkl ` to their _ asca _ spectra in order to fit a range of outburst and quiescent spectra with a single simple model . we also investigated how this multi - temperature model combined with photoelectric absorption and a 6.4 kev gaussian line would fit the outburst data of kt per , and obtained a statistically acceptable fit : @xmath182 = 0.96/836 , p@xmath184 = 0.819 . ccc + parameter & thermal plasma & cooling flow + n@xmath181 & 14.40@xmath185 & 15.70@xmath186 + @xmath4 10@xmath61 @xmath15 & & + + kt & 5.11@xmath187 & 12.60@xmath188 + ( kev ) + + abundance & 0.40@xmath117 & 0.42@xmath189 + + ew & 52@xmath190 & 45@xmath191 + ( ev ) & & + + flux(210 kev ) & 2.55 & 2.62 + @xmath4 10@xmath131 erg @xmath15 s@xmath3 & & + + flux(0.01100 kev ) & 5.60 & 6.19 + @xmath4 10@xmath131 erg @xmath15 s@xmath3 & & + + luminosity(210 kev ) & 1.0 & 1.03 + @xmath4 10@xmath2 erg s@xmath3 + + luminosity(0.01100 kev ) & 2.19 & 2.42 + @xmath4 10@xmath2 erg s@xmath3 & & + + @xmath192 & 0.97/838&0.95/837 + + p@xmath184 & 0.730&0.819 + [ ktper ] the method which is used for calculating the height of the x - ray emission source above the white dwarf , has been explained by @xcite for ss cyg . we have adopted the same method here in our work . as was explained in the beginning of section 5 , an equivalent width of up to @xmath0 150 ev can be expected for the fluorescent fe k@xmath41 line at 6.4 kev . in this work , we have assumed that the reflection originates from the white dwarf surface only , thus the reflection from the accretion disc is @xmath193/2@xmath43 = 0 . the equivalent width of 150 ev calculated by @xcite was assumed under the solar abundance conditions of @xcite where [ fe / h ] = 3.2 @xmath4 10@xmath194 . we have employed the abundances of @xcite which are the default abundance values built in the xspec cooling flow and optically thin thermal plasma emission models . for the solar abundances of anders & grevesse , the [ fe / h ] composition is 4.68 @xmath4 10@xmath194 . @xcite , who also employed the anders & grevesse abundances , corrected this abundance difference ( see their eq.3 ) using their measured iron abundance of 0.37 @xmath141 . for solar abundance , the observed equivalent width of the 6.4 kev line is @xmath195 where @xmath196 is the measured elemental abundance in solar units @xmath62 and @xmath197 the solid angle of the white dwarf viewed from the plasma of the boundary layer . in our sample , the observed equivalent widths ( table [ ev ] ) are mainly below 150 ev and this implies that the x - ray source is located at a height @xmath198 above the white dwarf surface . in the following , we use the values of @xmath199 and abundances calculated from the cooling flow model . as an example , the @xmath199 for ss aur is 86 ev and the abundance is 1.0 , thus eq . [ eq : ew ] gives @xmath197/2@xmath43 = 0.39 . if the x - ray source is point - like , the height @xmath198 of the x - ray source above the white dwarf of a radius @xmath200 is @xmath198 @xmath12 0.14 @xmath200 . as another example , we obtain @xmath197/2@xmath43 = 0.22 and @xmath198 @xmath12 0.64 @xmath200 for v893 sco ( @xmath199 = 45 ev , @xmath196 = 0.94 z@xmath184 ) . we have derived an x - ray luminosity function for 12 dwarf novae using archival _ suzaku _ , _ xmm - newton _ , and _ asca _ observations , and obtained new observations for bz uma , sw uma , vy aqr , ss aur , v893 sco and asas j0025 with _ suzaku _ as originally , they were not available in the archive . our results show that the 210 kev luminosities , presented in table [ fluxes ] , span a range between 4 @xmath4 10@xmath6 and 1.5 @xmath4 10@xmath5 erg s@xmath3 , and that most of the source luminosities in the sample are located within 10@xmath1 erg s@xmath3 , see fig . [ histo ] , whereas , the x - ray luminosities of the _ asca _ sample by @xcite were mainly concentrated on higher luminosities between 10@xmath2 and 10@xmath5 erg s@xmath3 . this difference is most likely due to the fact that we did not apply x - ray selection criteria to our sample . also , the objects observed by _ asca _ were known to be x - ray bright , thus the sample of baskill et al . is very likely biased by sources which are x - ray bright . in order to derive the integrated x - ray luminosity function ( xlf ) , n(@xmath17 l ) , for 12 sources within a distance of d = 200 pc , we assumed that the luminosity function is characterized by a power law n(@xmath17 l ) = k(l / l@xmath201)@xmath202 ( see fig . [ lum ] where the best - fit parameters @xmath41 = -0.64 and k = 2.39 @xmath4 10@xmath203 , corresponding to a threshold luminosity of l@xmath201 = 3 @xmath4 10@xmath1 erg s@xmath3 ) . the histogram illustrates the cumulative source distribution per pc@xmath204 in which a break is seen at @xmath0 3 @xmath4 10@xmath1 erg s@xmath3 . this can be due to two possible scenarios : 1 ) a single @xmath41 power law describes the luminosity function of dne , but the sample becomes more incomplete below @xmath0 3 @xmath4 10@xmath1 erg s@xmath3 than it is above this limit , or 2 ) the shape of the true xlf of dne is a broken power law with a break at around 3 @xmath4 10@xmath1 erg s@xmath3 . from these two scenarios , the first one is more likely since the sample contains only a few sources below @xmath0 10@xmath1 erg s@xmath3 . also , as was shown by , e.g. , the study of @xcite , more fainter cvs , such as wz sge types , are expected to exist . based on the obtained power law slope , the sample is dominated by the brighter dne : this is probably caused by the parallax measurement method which favours optically brighter dne which usually have high x - ray luminosities . l ) as a function of x - ray luminosity in log l in the 210 kev energy band . the error bar on the top right represents a typical error on the luminosities.,width=340 ] when calculating the total , integrated luminosity of the sample , we restricted the calculations to the distance of 200 pc , thus excluding bz uma . integrating between the luminosities of 1 @xmath4 10@xmath6 and the maximum luminosity of the sample ( @xmath205 = 1.50 @xmath4 10@xmath5 erg s@xmath3 ) , yields the total integrated luminosity of 1.48 @xmath4 10@xmath5 erg s@xmath3 , whereas the integrated luminosity between the threshold luminosity 3 @xmath4 10@xmath1 and @xmath205 is 1.15 @xmath4 10@xmath5 erg s@xmath3 . these two results show that there are uncertainties in the integrated luminosities , most likely caused by the small number of sources in the sample . in order to obtain more accurate value for the integrated luminosity , the power law slope ( @xmath41 = -0.64 ) should be better established . if the obtained slope is not far from the true power law slope of dne in the solar neighbourhood , estimating the integrated luminosity more accurately and constraining the bright luminosity end ( 10@xmath5 erg s@xmath3 ) requires more dne to be included in the sample . since the source density at @xmath0 10@xmath5 erg s@xmath3 is @xmath0 3 @xmath4 10@xmath206 pc@xmath22 according to fig . [ lum ] , we would need to survey within a volume of 1 @xmath4 10@xmath207 pc@xmath204 to find @xmath0 30 ss cyg -type dne and thus find a statistically significant constraint for the brighter luminosities in the sample . this volume would correspond to a distance of @xmath0 620 pc with a flux limit of @xmath0 3.2 @xmath4 10@xmath131 erg @xmath15 s@xmath3 . following this , we estimated how easy it would be to hide typical dn luminosities in the solar neighbourhood . assuming a typical dwarf nova with a 5 kev bremsstrahlung and a low galactic n@xmath181 = 1 @xmath4 10@xmath61 @xmath15 in webpimms yields a 210 kev flux of 5 @xmath4 10@xmath208 erg @xmath15 s@xmath3 , corresponding to the _ rosat _ pspc count rate of 0.04 ct s@xmath3 which is just below the detection limit ( 0.05 ct s@xmath3 ) of _ rosat _ pspc @xcite . thus , luminosities above 2.4 @xmath4 10@xmath1 erg s@xmath3 within 200 pc or above 6 @xmath4 10@xmath25 erg s@xmath3 within 100 pc should have been found by _ rosat _ and thus should be in the rass . however , given that sources with luminosities of 10@xmath1 erg s@xmath3 and below at a distance of 100 pc were too faint for the rass , and that our xlf peaks at @xmath0 10@xmath1 erg s@xmath3 , we conclude that there is no existing x - ray selected sample that we can use for this line of research . how far is the total luminosity of our sample from accounting for the total cv x - ray emissivity ? in order to estimate this , we calculated the absolute lower limit for the luminosity per cubic parsec volume ( l@xmath11/vol ) . for a distance of r = 200 pc , the volume v = 4/3 @xmath4 @xmath43 @xmath4 ( 200 pc)@xmath204 = 3.3 @xmath4 10@xmath209 pc@xmath204 , and the total summed luminosity l@xmath11 of the sample is 2.39 @xmath4 10@xmath5 erg s@xmath3 ( without bz uma ) . thus , the total absolute lower limit l@xmath11/volume = 7.24 @xmath4 10@xmath210 erg s@xmath3 pc@xmath22 . normalising this value to the local stellar mass density 0.04 m@xmath146 pc@xmath22 @xcite yields 1.81 @xmath4 10@xmath7 erg s@xmath3 m@xmath8 in the 210 kev range . for comparison , @xcite obtained ( 1.1 @xmath211 0.3 ) @xmath4 10@xmath212 erg s@xmath3 m@xmath8 ( 210 kev ) for the total cv x - ray emissivity per unit stellar mass . thus , our sample would account for @xmath0 16 per cent of this value . and finally , how much would our sample account for the grxe ? the galactic ridge x - ray emissivity estimated by @xcite in the 320 kev range was l@xmath11/m @xmath0 ( 3.5 @xmath211 0.5 ) @xmath4 10@xmath212 erg s@xmath3 m@xmath8 , meaning that our sample would account for 5 per cent of the galactic ridge x - ray emissivity . as we estimated the x - ray emissivity of all cvs within 200 pc , we used the exponential vertical density profile @xmath213 of cvs with a scale height for short period systems ( @xmath198 = 260 pc ) as in @xcite , where @xmath214 = @xmath215 sin @xmath216 is the perpendicular distance from the galactic plane and @xmath216 galactic latitude . integrating eq.[density ] over a sphere with a radius of 200 pc gives @xmath0 280 as the total number of dne within 200 pc . if the space density of dne follows the space density of cvs as in @xcite , i.e. , @xmath217 = 1.1@xmath218 @xmath4 10@xmath194 pc@xmath22 , and if a typical dn has an x - ray luminosity corresponding to the mean luminosity ( 2 @xmath4 10@xmath2 erg s@xmath3 ) of our sample of 11 sources ( bz uma excluded ) , then the 210 kev x - ray emissivity of all dne in the solar neighbourhood would be 5.5@xmath219 @xmath4 10@xmath212 erg s@xmath3 m@xmath8 ( these account for the uncertainty on the space density , assuming that this is the dominant source of uncertainty for the x - ray emissivity of dne ) . this would account for more than 100 per cent of the grxe emissivity . if dne were uniformly distributed in the solar neighbourhood , the x - ray emissivity would be overestimated also in this case ( by 2030 per cent ) . however , in both cases , one should remember that the calculated x - ray emissivity of all dne within 200 pc is likely overestimated by the brighter sources in our sample , thus the calculations give excess emission . in order to understand whether the x - ray luminosity and the various parameters ( inclination @xmath220 , orbital period @xmath20 , shock temperature @xmath32 and white dwarf mass @xmath221 ) are correlated , we carried out spearman s rank correlation test . plotting x - ray luminosity versus a few of these parameters ( @xmath220 and @xmath20 ) shows that gw lib seems to appear as an outlier compared to the rest of the sample ( fig . [ orb ] and [ incl ] ) . thus , to explore how the presence / absence of gw lib affects the test results , two test cases were used : 1 ) gw lib was included , and 2 ) gw lib was excluded from the rest of the sample . in addition , we investigated whether a correlation between the white dwarf masses @xmath221 and the shock temperatures @xmath32 ( fig . [ mass_temp ] ) exists , although in this case , gw lib seems to follow the rest of the sample , thus , carrying out test case 2 ) was not necessary . a strong correlation was found at the 99.95 per cent significance level ( 2.8@xmath30 ) between the x - ray luminosities and orbital periods ( fig . [ orb ] ) when gw lib is included in the sample . the correlation still holds when gw lib is excluded ( significance is 99.67 per cent ) . @xcite noted that there was a weak correlation between the x - ray luminosities and the orbital periods in their _ asca _ sample , concluding that the x - ray luminosity probably also correlates with long - term mean accretion rate . the x - ray luminosity and the inclination @xmath220 are not correlated in either case ( fig . [ incl ] ) . the correlation between these parameters was measured when the inclination of bz uma was set to 65@xmath29 , and altering the inclination between 60@xmath29 and 75@xmath29 did not affect the result . since no correlation was found , this result is in contrast with the discovery of anti - correlation between the emission measure and inclination found by @xcite . it is worth noting that the _ rosat _ bandpass was very narrow , covering only 0.12.4 kev where the softer x - ray emission ( and more luminous emission ) is probably intrinsically absorbed by the sources . in addition , an anti - correlation between the x - ray luminosity and inclination was also seen by @xcite in the _ asca _ sample , although , baskill et al . noted that the inclinations might be uncertain , and this can also be the case in our sample . finally , the white dwarf masses @xmath221 and the shock temperatures @xmath32 correlate with a significance of 98.5 per cent when the mass of vy aqr is 0.80 m@xmath146 , but becomes less significant ( 97.4 per cent ) if the mass is 0.55 m@xmath146 . of the rest of the parameters , i.e. the x - ray luminosity @xmath16 versus @xmath32 and @xmath221 , @xmath32 showed evidence of correlation with @xmath16 at a significance of 97.6 per cent when gw lib was included in the sample , but @xmath16 and @xmath221 had a much lower correlation significance ( 69 per cent ) when including gw lib . for the latter correlation test ( @xmath16 versus @xmath221 ) , the result was the same with both @xmath221 values for vy aqr . excluding gw lib decreased the significance to 91 per cent ( @xmath16 versus @xmath32 ) and to 63 per cent ( @xmath16 versus @xmath221 ) . we have analysed the x - ray spectra of 13 dwarf novae with accurate parallax - based distance estimates , and derived the most accurate shape for the x - ray luminosity function of dne in the 210 kev band to date due to accurate distance measurements and due to the fact that we did not use an x - ray selected sample . the derived x - ray luminosities are located between @xmath0 10@xmath610@xmath5 erg s@xmath3 , showing a peak at @xmath0 10@xmath1 erg s@xmath3 . thus , we have obtained peak luminosities which are lower compared to other previous studies of cv luminosity functions . the shape of the x - ray luminosity function of the source sample suggests that the two following scenarios are possible : 1 ) the sample can be described by a power law with a single @xmath41 slope , but the sample becomes more incomplete below @xmath0 3 @xmath4 10@xmath1 erg s@xmath3 than it is above this limit , or , 2 ) the shape of the real x - ray luminosity function of dwarf novae is a broken power law with a break at around 3 @xmath4 10@xmath1 erg s@xmath3 . the integrated luminosity between 1 @xmath4 10@xmath6 erg s@xmath3 and the maximum luminosity of the sample , 1.50 @xmath4 10@xmath5 erg s@xmath3 , is 1.48 @xmath4 10@xmath5 erg s@xmath3 . in order to better constrain the integrated luminosity and the slope of the x - ray luminosity function , more dwarf novae need to be included in the sample . thus , we suggest more future x - ray imaging observations of dwarf novae in the 210 kev band with accurate distance measurements . the total x - ray emissivity of the sample within a radius of 200 pc is 1.81 @xmath4 10@xmath7 erg s@xmath3 m@xmath222 ( 210 kev ) . this accounts for @xmath0 16 per cent of the total x - ray emissivity of cvs as estimated by @xcite , and @xmath0 5 per cent of the galactic ridge x - ray emissivity . the x - ray luminosities and the inclinations of our sample do not show anti - correlation which has been seen in other previous correlation studies , but a strong correlation is seen between the x - ray luminosities and the orbital periods . also , evidence for a correlation between the white dwarf masses and the shock temperatures exists . in the future , larger dwarf nova samples are needed in order to confirm these results . this research has made use of data obtained from the _ suzaku _ satellite , a collaborative mission between the space agencies of japan ( jaxa ) and the usa ( nasa ) . jo acknowledges support from stfc . part of this work is based on observations obtained with _ xmm - newton _ , an esa science mission with instruments and contributions directly funded by esa member states and the usa ( nasa ) . we thank the reviewer m. revnivtsev for his helpful comments on this paper .

this paper is a companion to the authors forthcoming work extending heegaard floer theory from closed 3-manifolds to compact 3-manifolds with two boundary components via quilted floer cohomology . we describe the first interesting case of this theory : the invariants of 3-manifolds bounding , regarded as modules over the fukaya category of the punctured 2-torus . we extract a short proof of exactness of the dehn surgery triangle in heegaard floer homology . we show that-structures on the graded algebra formed by the cohomology of two basic objects in the fukaya category of the punctured 2-torus are governed by just two parameters , extracted from the hochschild cohomology of . for the fukaya category itself , .

1102.3160

this article is an offshoot of the authors forthcoming work @xcite . in that paper we will combine a detailed geometric examination of the lagrangian correspondences between symmetric products of riemann surfaces , studied by the second author in @xcite , with the @xmath1 quilted floer theory of mau wehrheim woodward @xcite and the functoriality principle of @xcite ( see also @xcite ) . by doing so , we will extend the package of heegaard floer cohomology invariants @xcite from closed 3-manifolds to compact 3-manifolds with boundary . to be precise , we construct invariants for compact , oriented , connected 3-manifolds with precisely two boundary components , marked as ` incoming ' and ` outgoing ' . when these are both spherical , our invariants capture the heegaard floer cochains of the capped - off 3-manifold . we refer to auroux s work @xcite for the relationship of this theory to bordered heegaard floer theory @xcite . the format of our invariants is alarmingly abstract : they take the form of @xmath1-functors between @xmath1-categories associated with the boundary surfaces , satisfying a composition law under sewing of cobordisms . enthusiasts for extended tqft will approve of this formulation , but geometric topologists will want to know how to extract topological information from it . in this article , we examine the next - to - simplest case of the theory by applying it to manifolds @xmath5 with incoming boundary of genus @xmath6 ( which we cap off to form @xmath7 ) and outgoing boundary of genus @xmath8 . the relevant @xmath1-categories are certain versions of the fukaya category of a symplectic 2-torus @xmath9 with a distinguished point @xmath10 . the simplest version of the invariant for @xmath11 is * an @xmath1-module @xmath12 over the fukaya category + @xmath13 of exact , embedded curves in @xmath14 . * this module evaluates on each object @xmath15 ( which is a circle @xmath16 with an exactness constraint and certain decorations ) to give a cochain complex @xmath17 . this complex is quasi - isomorphic to the heegaard floer cochains @xmath18 , where @xmath19 is the solid torus in which the circle @xmath15 bounds a disc . the different objects @xmath15 correspond to different dehn fillings of @xmath11 . we illustrate our theory by showing how it leads to a proof of the dehn surgery exact triangle in heegaard floer cohomology @xcite ( theorem [ exacttri ] , corollary [ surgery ] ) . by working with the modules @xmath12 rather than the heegaard floer cochain complexes , one can legitimately work on the genus 1 boundary rather than on the @xmath20-fold symmetric product of a genus @xmath20 heegaard surface . this makes the proof technically straightforward and also makes the signs , on which the proof depends , transparent . to better understand the nature of our invariants , we need to understand the structure of @xmath13 . this category is closely related to the one studied in @xmath21 ( see @xcite for a deformation - theoretic approach to the latter category ) , but the differences are significant . it follows from the surgery exact triangle ( theorem [ exacttri ] ) that @xmath13 ( whose objects we declare to be exact , oriented lagrangians with _ non - trivial _ spin structures ) is generated by two objects curves @xmath22 and @xmath23 that meet transversely at a point , generating a full @xmath1-subcategory @xmath24 . that is , the inclusion @xmath25 induces a quasi - equivalence of the triangulated envelopes @xmath26 ( ` @xmath27 ' stands for twisted complexes ; see ( * ? ? ? * ( 3l ) ) ) . this in turn induces an equivalence of triangulated categories between the derived categories @xmath28 and @xmath29 ) . the @xmath1-structure of @xmath24 can be transferred to an @xmath1-structure on the cohomology algebra @xmath30 . this structure is governed by the hochschild cochain complex @xmath31 , with its bigrading and gerstenhaber bracket . we show that over a field @xmath32 in which 6 is invertible gauge - equivalence classes of @xmath1-structures in @xmath2 determine and are determined by two parameters , @xmath33 and @xmath34 . the ` moduli space ' @xmath35 is parametrized by the invariants @xmath3 of dg algebras associated with a pair of sheaves the structure sheaf and the skyscraper at @xmath36$]on the weierstrass curves @xmath37 ( prop . [ weierstrass ] ) for @xmath38 . the value of @xmath39 is non - zero ( theorem [ m6 ] ) . our expectation is that @xmath24 is quasi - isomorphic to the dg algebra for a _ nodal _ cubic . [ [ note . ] ] note . + + + + + the topics we have chosen do not rely on detailed information about our general theory . most of the theorems proved in this article are independent of it ; the sole exception , corollary [ surgery ] , invokes a general feature of the theory . the section on hochschild cohomology can be read on its own . we invoke the definition of the fukaya category from @xcite . we refer to that book for points of homological algebra , but recommend ( * ? ? ? * section 3 ) as a briefer alternative reference . a _ liouville domain _ is a compact , even - dimensional manifold @xmath40 with boundary , equipped with a 1-form @xmath41 such that @xmath42 is everywhere non - degenerate , and such that the liouville vector field @xmath43 , characterized by the equation @xmath44 , points outwards along the boundary . an _ exact lagrangian submanifold _ is a closed , embedded @xmath45-submanifold @xmath46 such that @xmath47 is an exact 1-form . the _ fukaya category _ @xmath48 of a liouville domain @xmath49 with @xmath50 , as constructed in @xcite , is an @xmath1-category , linear over a field @xmath32 . its objects are arbitrary exact lagrangian submanifolds equipped with spin - structures and gradings . for objects @xmath51 and @xmath52 of @xmath48 , the morphism - space @xmath53 is a floer cochain complex @xmath54 , where @xmath55 is the time-1 diffeomorphism of the hamiltonian vector field for a function @xmath56 , drawn from a look - up table . the first structure map @xmath57 is floer s coboundary map . the structure constants of the higher composition maps @xmath58 defining the @xmath1-structure are counts of inhomogeneous pseudo - holomorphic polygons . up to quasi - isomorphism , the whole structure is an invariant of @xmath59)$ ] , where @xmath60 $ ] is the class of @xmath41 modulo closed 1-forms supported in @xmath61 . the case of surfaces @xmath62 is special because one can use the riemann mapping theorem to describe moduli spaces of holomorphic polygons , and thereby describe some of the structure maps in combinatorial terms . we say ` some ' because when @xmath58 is applied to a sequence of curves that includes repetitions , the nave moduli spaces tend to be irregular , and perturbations are required . we next give a partial combinatorial formulation , following @xcite and @xcite . [ [ the-1-pointed-2-torus . ] ] the 1-pointed 2-torus . + + + + + + + + + + + + + + + + + + + + + + let @xmath9 be a closed riemann surface of genus 1 equipped with a basepoint @xmath10a complex elliptic curve . choose a hermitian metric in the holomorphic line bundle @xmath63 , and a unitary connection @xmath64 such that the curvature form @xmath65 is an area form , positive with respect to the complex orientation . choose also a tangent line field @xmath43 . write @xmath66 for the riemann surface @xmath67 . the restriction @xmath68 is tautologically trivialized ; in this trivialization , we can write @xmath69 , where @xmath70 . notice that @xmath41 restricts to a small loop encircling @xmath10 as a contact 1-form ; the liouville field points towards @xmath10 along this loop . define a _ marked exact lagrangian _ to be a pair @xmath71 , where @xmath72 is an oriented circle embedded in @xmath66 as an exact lagrangian submanifold ( that is , @xmath73 ) , and @xmath74 . a _ grading _ on a marked exact lagrangian is a homotopy from @xmath75 to @xmath76 inside the space of line fields @xmath77 . gradings which are homotopic rel endpoints are considered identical , which has the consequence that the gradings for @xmath72 form a @xmath78-torsor . non - contractible embedded curves in @xmath66 are determined , up to homotopy , by their slope , which is an arbitrary element of @xmath79 . in each homotopy class there is an exact representative ( take an initial representative @xmath80 and move it by an isotopy of flux @xmath81 ) , unique up to isotopy through exact lagrangians . [ [ objects . ] ] objects . + + + + + + + + the objects of our fukaya category @xmath13 will be marked , exact graded lagrangians . the ` hat ' is there firstly because this category is relevant to the ` hat ' version of heegaard floer cohomology , and secondly because @xmath13 is not quite the same as the usual exact fukaya category @xmath82 , but is a full subcategory of it . the distinction is the following . the basepoint @xmath83 determines a spin - structure on @xmath72 . a spin structure amounts to a double covering @xmath84 , and we declare this double covering to be trivial over @xmath85 , and to interchange the sheets at @xmath83 . _ we do not allow @xmath72 to carry the trivial spin structure . _ [ [ morphisms . ] ] morphisms . + + + + + + + + + + fix the base field @xmath32 . take a pair of objects @xmath86 , where @xmath87 and @xmath88 , and pick a hamiltonian function @xmath89 whose flow @xmath90 generates a family of curves @xmath91}$ ] , such that @xmath92 for all @xmath93 $ ] . because transversality is maintained , the combinatorial pattern of intersection points is unchanged for @xmath94 $ ] . the points @xmath95 and @xmath96 should not occur as intersection points @xmath97 for any @xmath94 $ ] . we define @xmath98 to be the floer cochain complex @xmath99 . in more detail , each point @xmath100 has a maslov index @xmath101 defined using the gradings @xcite . the floer complex is @xmath102 the differential @xmath103 is defined by linearity and @xmath104 here @xmath105 is the set of homotopy - classes of _ immersed bigons _ in @xmath66 . by an immersed bigon , we mean a homotopy class of smooth , orientation - preserving immersions @xmath106 such that @xmath107 and @xmath108 , where @xmath109 extends smoothly to a map @xmath110 with @xmath111 and @xmath112 . furthermore , the image of the map should have _ convex corners _ at @xmath113 and @xmath114 . the set @xmath105 is finite , and to each @xmath115 one can assign a sign @xmath116 . this is given by @xmath117 , where @xmath118 is @xmath6 if the upper boundary of @xmath109 traverses @xmath119 in the positively oriented direction , and @xmath8 otherwise ; and @xmath120 is the number of stars ( @xmath95 or @xmath96 ) encountered by @xmath109 on its boundary . one has @xmath121 , so @xmath99 is a cochain complex , and one can form the cohomology @xmath122 . when @xmath123 and @xmath119 are exact - isotopic , @xmath124 . [ [ structure - maps . ] ] structure maps . + + + + + + + + + + + + + + + the @xmath1 structure maps of @xmath13 , @xmath125\ ] ] are in general defined via solutions to an inhomogeneous cauchy riemann equation , but in certain cases one can obtain consistent and regular moduli spaces of genuine holomorphic maps into @xmath66 . we summarize what we need , referring to ( * ? ? * chapter 2 ) and @xcite for further details . write @xmath126 , and suppose that all the @xmath127 are drawn from a fixed , finite set @xmath128 of objects such that the curves underlying two distinct objects intersect transversely ( and not at the @xmath83-points ) and all triple intersections are empty . in that case , we can suppose that the functions @xmath129 are zero when @xmath130 . the most pleasant instance of ( [ ainf ] ) occurs when any two of the objects @xmath127 in the sequence are distinct elements of @xmath128 ( hence transverse as curves ) . in that case , consider intersection points @xmath131 for @xmath132 . one has @xmath133 where @xmath134 is a signed count of immersed polygons . to be precise , the relevant polygons are smooth immersions of @xmath135 into @xmath66 . the immersion must preserve orientation , must extend continuously to a map on @xmath136 sending @xmath137 to @xmath138 , and must map the boundary interval @xmath139 to @xmath140 . it must also have convex corners . what we count are the homotopy classes of such immersions . the sign attached to an immersion is @xmath141 . as before , @xmath120 is the number of stars on the boundary ; @xmath142 is @xmath143 if the polygon travels along @xmath144 in the negative direction with respect to the orientation , and @xmath6 otherwise ; and @xmath118 is the sum of degrees @xmath145 over those @xmath146 ( not @xmath147 ) such that the polygon travels along @xmath140 in the negative direction . now suppose that the sequence @xmath148 includes precisely one repetition , occurring between cyclically adjacent curves ( i.e. , @xmath149 for some @xmath150 ) with all other pairs distinct , hence transverse . in this case , we replace @xmath140 by @xmath151 , the image of @xmath140 under the time @xmath152 hamiltonian diffeomorphism @xmath153 for the function @xmath154 associated with the pair @xmath155 . one then applies the above recipe to the transverse sequence of objects with @xmath156 replaced by @xmath157 . as stated this does not quite make sense because after this replacement , the intersection points are between the wrong curves . however , when @xmath158 , one has a canonical isomorphism of cochain complexes @xmath159 arising from the bijection @xmath160 coming from the flow @xmath161 . one also has a canonical isomorphism @xmath162 , induced by @xmath163 . one uses these isomorphisms to make sense of this formulation . [ [ versions - including - the - basepoint . ] ] versions including the basepoint . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + as a variant on the construction of @xmath13 , one can construct the @xmath1-category @xmath164 with the same objects , and morphism spaces @xmath165 $ ] , where @xmath166 has degree 0 . to have degree 2 . the genus 1 case is anomalous . the stabilization isomorphism relating heegaard complexes computed in genera 1 and 2 is not degree - preserving , but it does respect a natural `` geometric grading '' , refining the @xmath167 grading , by the @xmath78-set of homotopy classes of oriented 2-plane fields on the 3-manifold . ] the structure maps are constructed in just the same way as before , except that the immersed polygons are now allowed to pass through @xmath10 , and @xmath134 counts such a polygon @xmath109 with a weight @xmath168 , where @xmath169 is the multiplicity of @xmath10 in @xmath114 . ( this procedure is familiar from heegaard floer theory ) . since the multiplicities are non - negative , one can construct @xmath1-categories @xmath170 and @xmath171 whose morphism spaces are @xmath172,\\ & \hom_-(x , y)=\hom_{\euf(t_0)}(x , y)\otimes \bbk[u^{-1},u]]/\bbk[[u]].\end{aligned}\ ] ] @xmath13 and the @xmath173-categories are all three determined by the @xmath174 version , together with its filtration by the subcategories with hom - spaces @xmath175 $ ] . there is a contravariant yoneda embedding of @xmath48 into the @xmath1-category @xmath176 of right @xmath48-modules , that is , @xmath1 functors from @xmath48 to the dg category of cochain complexes . the embedding maps an object @xmath15 of @xmath48 to the module @xmath177 which assigns to each object @xmath178 the cochain complex @xmath179 @xcite . there is also an intermediate category @xmath180 , the _ extended _ fukaya category , whose objects are finite sequences of lagrangian correspondences ( of a constrained kind ) between liouville domains , where the sequence begins at @xmath181 and ends at @xmath49 . the morphism - spaces are quilted floer cochain complexes , as in @xcite . making this precise is the substantial task of @xcite . one has embeddings @xmath182 factoring @xmath183 ; the first functor is an obvious inclusion , @xmath184 another yoneda embedding . the construction in @xcite will attach to a 3-manifold @xmath11 bounding @xmath9 an object @xmath185 in @xmath186 . it depends on additional choices , but the resulting module @xmath187 over @xmath188 does not ( up to isomorphism ) . the object @xmath185 also defines a filtered module @xmath189 over @xmath164 , hence modules @xmath190 and @xmath191 over @xmath170 and @xmath171 . in this article , we shall use only one property of these modules , to be proved in @xcite , which is as follows . let @xmath87 be an object in @xmath192 , and @xmath193 the closed 3-manifold obtained from @xmath11 by dehn - filling @xmath72 . ( @xmath194 ) _ the @xmath195$]-linear cochain complex @xmath196 is quasi - isomorphic to the completion at @xmath166 of the heegaard floer cochains @xmath197 defined in @xcite , compatibly with the natural filtrations of these complexes . _ in an @xmath1-category @xmath198 , one has a notion of an _ exact triangle _ $ ] : a trio of morphisms ( only their classes @xmath200 $ ] , @xmath201 $ ] and @xmath202 $ ] in @xmath203 matter ) which is transformed by the yoneda embedding @xmath204 into a triangle of @xmath198-modules @xmath205 $ ] , isomorphic in @xmath206 to the standard triangle @xmath207 $ ] associated with the mapping cone of @xmath22 ( * ? ? ? * ( 3f ) ) . + [ exacttriangle ] , @xmath208 and @xmath209 in the torus , with the basepoint @xmath10 placed in the hexagonal part of their complement . hamiltonian deformations of these curves , used to define endomorphisms in the fukaya category , are shown in color . right : immersed polygons in the torus viewed as embedded polygons in the universal cover . ] [ exacttri ] let @xmath210 , @xmath211 and @xmath212 be three objects in @xmath13 . suppose that any two of the underlying curves , @xmath213 and @xmath214 , say , intersect at a single point @xmath215 , transversely , and that the curves are oriented so that @xmath216 as pictured in figure 1 ( left ) . then one has an exact triangle in @xmath164 , @xmath217,\ ] ] where @xmath109 is euler s generating function for partitions , @xmath218^\times.\ ] ] this triangle remains exact in @xmath170 and in @xmath171 , and specializes ( by putting @xmath219 in the @xmath220 version ) to an exact triangle in @xmath13 . the exactness of ( [ ex tri ] ) in @xmath13 is a special case of seidel s exact triangle ( * ? ? ? * section 17j ) ( see also @xcite ) . the full statement is new , the @xmath109-factors in particular , but cf . @xcite . the curves depicted as bold black lines in figure 1 ( left ) represent @xmath80 , @xmath208 and @xmath209 ; the orientations are shown as arrowheads . the gradings will turn out to be immaterial . in @xmath164 , one has @xmath221 \cdot e_{ij } , \quad i\neq j.\ ] ] we choose floer data on @xmath127 such that @xmath222 , where @xmath223 is the object whose underlying curve @xmath224 is the small hamiltonian push - off of @xmath213 depicted in figure 1 ( left ) . let @xmath225 be the full subcategory of @xmath164 with objects @xmath51 , @xmath52 and @xmath226 ; notice that , with our choice of floer data , @xmath227 . one can recognize exact triangles using the criterion of ( * ? ? ? * lemma 3.7 ) , which is particularly straightforward since @xmath228 . to prove that ( [ ex tri ] ) is exact in @xmath225 ( or equally in @xmath164 ) , it is sufficient that @xmath229 where @xmath230 is the unique element in @xmath231 representing the identity in @xmath232 ; and that for each object @xmath233 of @xmath164 , the cochain complex @xmath234 \oplus \hom(z , x_0)[1 ] \oplus \hom(z , x_1)\ ] ] is acyclic with respect to the differential @xmath235.\ ] ] we compute @xmath236 by examining immersed triangles with convex corners . for the composite @xmath237 , there are two embedded triangles , shaded in the left - hand figure , neither of them intersecting @xmath10 , contributing a leading - order term @xmath238 for certain signs @xmath239 . there are further immersed triangles , whose lifts to the universal cover of @xmath9 are visible in figure 1 ( right ) . for each integer @xmath240 , there are exactly two immersed triangles whose sides wrap around @xmath80 more than @xmath241 times but less than @xmath242 times . these triangles pass through @xmath10 at @xmath243 points . hence @xmath244 for signs @xmath245 . we can also see that @xmath246 for further signs @xmath247 . the leading term @xmath248 ( resp . @xmath249 ) comes from the embedded quadrilateral shown in brown ( resp . green ) on the right of the figure . the higher @xmath250 ( resp . @xmath251 ) terms account for larger , immersed quadrilaterals , approximately similar to the two depicted . [ signs ] now we check the signs , using the recipe described earlier involving orientations and stars . in figure 2 ( which depicts the universal cover ) , the points @xmath252 are shown as colored stars . we read off from figure 2 that @xmath253 where @xmath254 is the sign coming from orientation - mismatches , and the power of @xmath152 is the ` star - sign ' @xmath255 . the @xmath256- and @xmath251-polygons have boundary orientations consistent with the given orientations of the @xmath213 , so @xmath257 . the @xmath258 and @xmath250-triangles have boundary orientations opposite to the given orientations . for a convex - cornered @xmath259-gon with corners @xmath260 ( outgoing ) and @xmath261 ( incoming ) , in their natural boundary order , one has the relation @xmath262 ; from this we see that @xmath263 and @xmath264 . consequently , @xmath265 , where @xmath266 one has @xmath267 : this specialization of jacobi s triple product identity can be found in his 1829 treatise _ fundamenta novae theoriae ellipticarum _ ( section 66 ) . we turn now to the verification that the complex ( [ sum of three ] ) has no cohomology . this is well - known ( cf . @xcite ) , so we shall be brief . given @xmath233 , move @xmath208 by a hamiltonian isotopy which does not change the combinatorics of the pattern of intersection points between @xmath208 and @xmath80 , nor between @xmath208 and @xmath209 , to a curve @xmath268 which lies very close to @xmath269 . the hamiltonian isotopy produces an isomorphism @xmath270 in the fukaya category , and the cohomology of ( [ sum of three ] ) is unaffected by replacing @xmath208 by @xmath268 via this isomorphism . there is a bijection @xmath271 which sends an intersection point of @xmath272 or of @xmath273 to a nearby point in @xmath274 . work , for now , in @xmath13 . there are finitely many holomorphic polygons in @xmath66 contributing to the matrix entries of the differential in ( [ sum of three ] ) , and up to deformation , these curves do not change as we let @xmath268 degenerate towards @xmath269 . some of these curves are ` small ' in the sense that their area goes to zero in this degeneration ; the others are ` large ' . there is a unique small triangle contributing to @xmath275 , and its contribution is @xmath276 . for @xmath277 , there are no small - area triangles contributing to @xmath278 , but there is a unique small - area quadrilateral contributing to @xmath279 : its contribution is @xmath280 . there are no small - area bigons contributing to the diagonal entries of @xmath281 . the homology of the ` small area ' part of the differential is zero , and by a filtration argument based on the symplectic action ( as in @xcite ) it follows that @xmath282 . another filtration argument , this time based on the filtration by powers of @xmath166 , shows that @xmath282 also in @xmath164 and the other variants . [ surgery ] let @xmath11 be a 3-manifold bounding the torus @xmath9 . let @xmath80 , @xmath208 and @xmath209 be simple closed curves in @xmath9 so that any two of them intersect transversely at a simple point , and oriented so that @xmath283 let @xmath284 be the 3-manifold obtained by dehn - filling @xmath285 . then there is an exact triangle of @xmath164-modules , @xmath286,\ ] ] and hence by the principle @xmath287 an exact triangle of completed heegaard floer cohomology groups @xmath288 . the same holds true in the @xmath220 , @xmath289 and ` hat ' versions of @xmath290 , @xmath291 and @xmath292 . define @xmath293 to be the smallest set of curves on @xmath66 , containing @xmath51 and @xmath52 , such that given any two curves @xmath294 , if @xmath295 is a third curve such that @xmath296 form a ` surgery triangle ' as in theorem [ exacttri ] then @xmath297 . it is easy to check that @xmath293 contains curves of arbitrary rational slope . hence : + the objects @xmath22 and @xmath23 generate the triangulated @xmath1-category of twisted complexes @xmath298 . they likewise generate @xmath299 , @xmath300 and @xmath301 . let @xmath24 denote the full subcategory of @xmath13 with objects @xmath22 and @xmath23 . we have just seen that the induced inclusion @xmath302 is a quasi - equivalence . let @xmath303 be the cohomology category of @xmath24 , a graded @xmath32-linear category . we sometimes regard @xmath2 ( more precisely , the direct sum of its hom - spaces ) as an associative , unital @xmath32-algebra with two distinguished idempotent elements whose sum is 1 . the normalized hochschild cochain complex of the unital algebra @xmath2 , @xmath304 , is the direct sum of subspaces @xmath305 , the @xmath32-vector space of degree - preserving linear maps @xmath306 $ ] which vanish on monomials one of whose factors is @xmath8 . besides being a cochain complex with an internal grading @xmath120 preserved by the differential @xmath307 , the hochschild complex has its gerstenhaber bracket which makes @xmath308 a dg lie algebra . in a precise sense , cochains @xmath309 which satisfy the maurer cartan equation @xmath310=0 $ ] parametrize @xmath1-structures with cohomology algebra @xmath2 ( when 2 is invertible ) ( * ? ? ? * ( 3a , e ) ) . it is therefore of interest to compute the hochschild cohomology @xmath311 . this we accomplish using the interpretation of hochschild cohomology as a right - derived functor for graded bimodule homomorphisms . [ [ structure - maps - in - eua . ] ] structure maps in @xmath24 . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + an @xmath1-category is _ minimal _ if @xmath312 . in @xmath24 , the self - homs @xmath313 and @xmath314 are defined as floer complexes @xmath315 and @xmath316 , where @xmath317 is the image of @xmath22 under a chosen hamiltonian diffeomorphism , and @xmath318 is the image of @xmath23 under a ( possibly different ) hamiltonian diffeomorphism . one can arrange that @xmath319 and @xmath320 are transverse intersections , each consisting of two points . by making such choices , we ensure that @xmath24 is minimal . there is a unique point @xmath321 . this point represents a class @xmath322 which generates @xmath323 . by assigning @xmath324 degree 1 , we are in effect pinning down the gradings of @xmath22 and @xmath23 up to a common shift of both @xmath22 and @xmath23 . the point @xmath113 also represents a class @xmath325 , generating @xmath326 . that @xmath327 has degree @xmath6 is an instance of floer - theoretic poincar duality : @xmath328 . one has @xmath329 for generators @xmath330 and @xmath331 of degree @xmath6 and @xmath332 and @xmath333 of degree @xmath8 . we specify the product @xmath334 by giving instead the structure of the cohomological category @xmath335 . this graded - linear category has hom - spaces @xmath336 ( which equals @xmath337 by minimality ) and composition @xmath338\circ [ y ] = ( -1)^{|y| } [ \mu^2_\eua(x , y)]$ ] . in @xmath2 , @xmath339 $ ] ( resp . $ ] ) is the identity morphism of @xmath341 ( resp . @xmath342 ) . the non - zero products that do not involve @xmath339 $ ] or @xmath340 $ ] are @xmath343[v_0]=[f_1 ] ; \quad [ v_0][u_1]=[e_1 ] .\ ] ] we view @xmath2 as an associative algebra by taking the direct sum of all four morphism spaces ; multiplication in this algebra is given by composition when that makes sense , and by zero in other cases . to remember the categorical structure , one regards @xmath2 as an @xmath344-bimodule , where @xmath345 is the semisimple ring @xmath346,[f_0 ] \}$ ] . for example , @xmath347 a [ e_0]$ ] . [ [ quiver - presentation . ] ] quiver presentation . + + + + + + + + + + + + + + + + + + + + if one views @xmath2 as a @xmath32-algebra , one can present it as a quotient of the path - algebra for the following quiver @xmath348 : @xmath349}[r ] & b\ar@/^/^{[v_0]}[l ] } \ ] ] the path - algebra @xmath350 is the unital , graded , associative @xmath351-algebra with a basis given by all composable sequences of arrows @xmath352 or @xmath353 ( written as @xmath354 $ ] and @xmath355 $ ] and given degrees 1 and 0 respectively ) , including the trivial sequences starting and ending at @xmath22 or at @xmath23 . multiplication is concatenation when this is results in a composable sequence , and is zero otherwise . the @xmath32-algebra @xmath350 is actually an @xmath344-bimodule : @xmath339 $ ] and @xmath340 $ ] act by composing ( on the left or right ) with the trivial sequence at @xmath22 or @xmath23 respectively . to obtain @xmath2 from @xmath350 , we divide by the ideal generated by the sequence @xmath356 and @xmath357 . this ideal is the third power @xmath358 of the ` arrow ideal ' @xmath359the ideal generated by all paths of positive length . there is a canonical isomorphism of graded @xmath32-algebras @xmath360 , which sends the arrow @xmath354 $ ] to the class @xmath354\in a$ ] , and the arrow @xmath361 $ ] to @xmath361\in a$ ] . this isomorphism respects the @xmath344-bimodule structure . + [ bigradedhh ] work over an arbitrary base field @xmath32 . \(a ) the bigraded hochschild cohomology @xmath362)\ ] ] satisfies the relation @xmath363 which amounts to 8-periodicity in @xmath118 . if @xmath364 then one has @xmath365 \(b ) for @xmath366 , @xmath367 is as follows : @xmath368 the table in ( b ) is the result of a computation involving rather delicate sign considerations . we are grateful to paul seidel , who has written a python 3.0 program to compute @xmath369 for small @xmath118 , over a field of small positive characteristic , via the normalized bar complex . we have used this program to check our result . [ [ graded - ext . ] ] graded ext . + + + + + + + + + + + the ext - modules in the statement of the theorem are defined , for graded @xmath370-bimodules @xmath371 , as right - derived functors : @xmath372 . here @xmath373 refers to degree - preserving homomorphisms of @xmath370-bimodules . concretely , one takes a projective resolution @xmath374 by graded @xmath370-bimodules ( the normalized bar resolution is one possibility ) . the resolution is a sequence of bimodules , @xmath375 , and the @xmath376 are degree - preserving bimodule maps . applying the hom - functor results in a cochain complex @xmath377 , whose maps @xmath378 are given by @xmath379 , and one has @xmath380 . shifting a bimodule @xmath371 changes its structure as a left @xmath2-module : the multiplication @xmath381\otimes a\to b[s]$ ] is @xmath382 , where @xmath383 is the result of multiplication @xmath384 . if @xmath385 is another graded bimodule , we write @xmath386 for @xmath387)}$ ] . skldberg @xcite finds a projective resolution of @xmath2 as a graded @xmath370-bimodule . to describe it , let @xmath388 be the subspace of the path - algebra spanned by paths of length @xmath389 , and let @xmath390 be the subspace spanned by paths of length @xmath391 . then @xmath392 is an @xmath344-bimodule , and is a rank 2 free @xmath32-module . ordered bases @xmath393 for @xmath392 can be given as follows : @xmath394,[f_0]\ } ; \\ & b_{4k } = \bbk \ { ( [ u_1][v_0])^{3k } , ( [ v_0 ] [ u_1])^{3k}\ } \quad ( k>0 ) ; \\ & b_{4k+1 } = \bbk \ { [ v_0 ] ( [ u_1][v_0])^{3k } , [ u_1]([v_0 ] [ u_1])^{3k}\ } ; \\ & b_{4k+2 } = \bbk \ { [ v_0 ] ( [ u_1][v_0])^{3k+1},[u_1 ] ( [ v_0 ] [ u_1])^{3k+1 } \ } ; \\ & b_{4k+3 } = \bbk \ { ( [ u_1][v_0])^{3k+2 } , ( [ v_0 ] [ u_1])^{3k+2 } \ } .\end{aligned}\ ] ] so @xmath395 $ ] , etc . the resolution is @xmath396 where @xmath397 @xmath398 , and the maps @xmath399 are as follows . write a typical monomial @xmath400 as @xmath401b$ ] . then @xmath402b ) & = a[c_{3j } | \dots | c_3]c_2c_1 b + ac_{3j } [ c_{3j-1 } | \dots | c_2 ] c_1 a \\ & + a c_{3j } c_{3j-1 } [ c_{3j-2 } | \dots | c_1 ] b ; \notag\\ p_{2j+1}(a[c_{3j+1 } | \dots | c_1]b ) & = ac_{3j+1}[c_{3j}| \dots | c_1 ] b - a [ c_{3j } | \dots | c_2 ] c_1 b .\end{aligned}\ ] ] it s easy to check that @xmath403 and that @xmath404 . it is shown in @xcite that this complex is a projective bimodule resolution of @xmath2 . observe that @xmath405 is generated , as an @xmath370-bimodule , by its subspace @xmath406 . one has a basis for @xmath407 given by @xmath408 and @xmath409 , and hence an isomorphism of graded @xmath32-vector spaces @xmath410.\ ] ] similarly , @xmath411 \oplus \bbk[-3k-1]\\ & g_{4k+2 } \cong \bbk[-3k-1 ] \oplus \bbk[-3k-2]\\ & g_{4k+3 } \cong \bbk^2[-3k-2].\end{aligned}\ ] ] in each case , the isomorphism comes from the ordered basis @xmath412 for @xmath413 . moreover , @xmath405 is the direct sum of its sub - bimodules generated by @xmath414 and @xmath214 . these summands are not free ; one has , for instance , an isomorphism @xmath415 , \quad a \beta_{4k } a'\mapsto af_0\otimes f_0 a'.\ ] ] restriction to @xmath413 gives an injective map of graded vector spaces @xmath416 and one has , for example , @xmath417 \oplus e_0ae_0 [ -3k].\ ] ] we now consider the maps @xmath418 . as a sample , consider @xmath419 . if @xmath420 , then @xmath421f_0 - f_0 [ ( v_0u_1)^{3k } ] v_0 f_0 \big)\\ & = \theta \big ( v_0 \beta_{4k } - \gamma_{4k}v_0 f_0 \big)\\ & = \left ( v_0 \theta ( \beta_{4j } ) - \theta(\gamma_{4j})v_0\right ) .\end{aligned}\ ] ] using this and similar calculations , we can describe the cochain complex @xmath422 in explicit terms . the segment beginning at @xmath423 reads as follows : + @xmath424 \oplus e_0ae_0[-3k ] \\ @v { p_{4k+1}^*}v{\left [ \begin{array}{cc } l(v ) & -r(v ) \\ -r(u ) & \eta l(u ) \\ \end{array}\right ] } v\\ e_0af_0[-3k ] \oplus f_0ae_0[-3k-1 ] \\ @v{p_{4k+2}^*}v{\left [ \begin{array}{cc } \eta l(e_1)+ r(f_1 ) & l(v)r(v ) \\ \eta l(u)r(u ) & -\eta l(f_1)+ r(e_1 ) \\ \end{array}\right]}v\\ e_0af_0[-3k-1 ] \oplus f_0 a e_0[-3k-2 ] \\ @v{p_{4k+3}^*}v { \left [ \begin{array}{cc } -\eta l(u ) & -r(v ) \\ -r(u ) & l(v ) \\ \end{array}\right]}v\\ f_0af_0[-3k-2 ] \oplus e_0ae_0[-3k-2 ] \\ @v{p_{4k+4}^*}v { \left [ \begin{array}{cc } \eta l(f_1)+ r(f_1 ) & \eta l(u ) r(v ) \\ l(v)r(u ) & \eta l(e_1)+r(e_1 ) \\ \end{array}\right]}v\\ f_0 af_0[-3k-3 ] \oplus e_0 ae_0 [ -3k-3 ] . \\ \end{cd}$ ] here @xmath425 and @xmath426 ( this is multiplication in @xmath2 , disregarding degree shifts ) ; and @xmath427 . these four adjacent maps depend on @xmath428 only through the alternating sign @xmath429 and the degree shifts . so we see that @xmath430 ) \cong h^r(\hom(p _ * , a[s-6])$ ] for @xmath431 ; and that in characteristic 2 , @xmath432 ) \cong h^r(\hom(p _ * , a[s-3])$ ] . with this explicit description of the complex , the cohomology groups can be determined by a routine computation which is left to interested readers to reproduce . [ [ parametrizing - a_infty - structures . ] ] parametrizing @xmath1-structures . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + if we avoid characteristics dividing @xmath433 , the structure of @xmath434 is simple enough that we can use our result to determine the gauge - equivalence classes of minimal @xmath1-structures on @xmath2 ( for gauge - equivalence , alias formal diffeomorphism , see @xcite . ) by homological perturbation theory , these classes correspond naturally to pairs of an @xmath1-algebra @xmath435 with an isomorphism @xmath436 , up to quasi - isomorphism . + [ classification ] assume @xmath437 . consider an arbitrary @xmath1-structure @xmath438 on @xmath2 , with @xmath439 and @xmath440 . 1 . the structure @xmath438 is equivalent , under the action of the group of gauge transformations , to an @xmath1-structure @xmath441 in which the structure maps @xmath442 , @xmath443 , @xmath444 and @xmath445 all vanish . 2 . for @xmath441 as in ( 1 ) , the products @xmath446 and @xmath447 are cocycles representing classes @xmath448 and @xmath449 respectively . these classes are invariants of @xmath438 . two @xmath1-structures which define the same classes @xmath450 are gauge - equivalent . 3 . every class in @xmath451 is realized as the invariant @xmath3 of some minimal @xmath1-structure on @xmath2 . hence gauge - equivalence classes of minimal @xmath1-structures on @xmath2 are canonically parametrized by this product of two 1-dimensional @xmath32-vector spaces , and the action of @xmath452 rescaling structure maps , defined by @xmath453 , corresponds to the action @xmath454 on @xmath451 . in general , if @xmath455 and @xmath456 are the structure maps for @xmath1-structures on @xmath2 , with @xmath457 and @xmath458 , and if @xmath459 for @xmath460 , then the difference @xmath461 is a cocycle : @xmath462 . if it is a coboundary , say @xmath463 , then the gauge transformation @xmath464 , with components @xmath465 given by @xmath466 , @xmath467 and @xmath468 for @xmath469 , has the property that @xmath470 for @xmath471 . since @xmath472 for @xmath473 by theorem [ bigradedhh ] , to prove gauge - equivalence of @xmath438 and @xmath441 it suffices to make them agree up to @xmath474 . by taking @xmath456 to be the trivial @xmath1-structure , and noting that @xmath472 for @xmath475 , we see that one can find a gauge transformation @xmath20 so that the @xmath1-structure @xmath476 has vanishing structure maps of these orders . one then has @xmath477 . we shall see presently that the class @xmath478\in hh^2(a , a)^{-4}$ ] is an invariant of @xmath24 . the @xmath1-equation , and the vanishing of @xmath479 for @xmath475 , imply that @xmath480 and @xmath481 are again cocycles , representing classes @xmath482 and @xmath483 . now @xmath484 , and this space vanishes by theorem [ bigradedhh ] ( b ) , so there is a gauge transformation @xmath485 such that @xmath486 while @xmath487 for @xmath488 . hence two @xmath1-structures which have the same class @xmath489 are gauge - equivalent up to order 7 . if in addition their @xmath483-classes agree then a suitable gauge transformation makes their structure maps agree up to order 8 , whereupon the structures are gauge - equivalent . there exists a minimal @xmath1-structure with prescribed values for @xmath3 . begin by declaring @xmath490 for @xmath491 . if @xmath492 has been chosen for @xmath460 , and the @xmath1 relations hold for all input - sequences of length @xmath493 , then the condition on @xmath58 for the @xmath1 relations to hold on sequences of length @xmath281 is that @xmath494}.\ ] ] here @xmath495 $ ] is the gerstenhaber lie bracket ( * ? ? ? * ( 3b ) ) . by applying the same identity for @xmath496 instead of @xmath281 , and the jacobi identity , one sees that the right - hand side defines a cocycle in @xmath497 , and to define @xmath58 it suffices to check that this cocycle is a coboundary . but @xmath498 except for @xmath499 . for those values of @xmath281 , we take @xmath58 to be a hochschild cocycle representing the given cohomology class ( so @xmath500 ) , which is legitimate since the gerstenhaber brackets on the right - hand side of ( [ maurer cartan ] ) are all zero in these cases . inductively , one can define @xmath58 for all orders @xmath281 . the remaining question is the independence of @xmath489 and @xmath483 from the gauge transformations used to define them . here we appeal to indirect arguments . the well - definedness of @xmath489 can be seen as a consequence of its interpretation as a differential in the length spectral sequence ( see point @xmath501 below ) , which is itself invariant under gauge transformations . we can see that both @xmath489 and @xmath483 are invariant by exhibiting a 2-parameter family of non - gauge - equivalent @xmath1-structures which realize all possible values of @xmath3 . ext - algebras on a family of weierstrass cubics will achieve this ( see @xmath502 below ) . [ [ the - length - spectral - sequence . ] ] the length spectral sequence . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + @xmath1-categories @xmath435 also have hochschild cohomology @xmath503 . this graded @xmath32-vector space is not bigraded , but the underlying bar complex @xmath504 has a natural filtration , the _ length filtration_. the resulting spectral sequence @xmath505 has @xmath506 , where @xmath507 ( * ? ? ? * ( 1f ) ) . in general , convergence of the spectral sequence is not guaranteed , but from theorem [ bigradedhh ] we obtain the following two corollaries : + for any @xmath1-structure @xmath438 on @xmath2 , with @xmath439 and @xmath508 , the length spectral sequence @xmath509 for the length filtration on @xmath510 degenerates by @xmath511 . one has @xmath512 . the longest possible differentials in the spectral sequence are of form @xmath513 . assume @xmath514 . let @xmath438 be a minimal @xmath1-structure on @xmath2 . then either @xmath438 is formal , or @xmath515 , or @xmath516 in the length spectral sequence for @xmath517 . the only possibly non - trivial differentials in the length spectral sequence are @xmath518 , @xmath519 and @xmath520 . one has @xmath521 if @xmath438 is not formal then @xmath522 and @xmath523 . in the spectral sequence , one has @xmath524 for @xmath525 , while @xmath518 is the gerstenhaber bracket @xmath526 $ ] ( * ? ? ? * ( 3.7 ) ) . if @xmath527 then @xmath528 and @xmath529 $ ] . the euler derivation @xmath530 , given by @xmath531 for homogeneous @xmath113 , represents a non - trivial class @xmath532\in hh^1(a , a)^0\cong \bbk$ ] and one has @xmath533 = ( k-2 ) m^k$ ] when @xmath524 for @xmath534 ( * ? ? ? * ( 3.8 ) ) . ( @xmath501 this explains why @xmath489 is well - defined . ) armed with this information , and observing that the identity natural transformation @xmath535 represents a non - trivial class in @xmath536 , the assertions are straightforward to check . [ [ non - formality . ] ] non - formality . + + + + + + + + + + + + + + one can compare this corollary to results and conjectures in symplectic geometry . a conjecture of seidel s @xcite states that , for certain liouville domains @xmath49 , the ` open - closed string map ' @xmath537 from symplectic cohomology is an isomorphism . according to @xcite , one has @xmath538 there is an isomorphism @xmath539 , so the conjecture predicts that @xmath24 is non - formal . this is true : + [ m6 ] when 6 is invertible , the @xmath1-structure @xmath24 arising from @xmath13 is not formal . indeed , @xmath540 . our proof of this is computational : we take a dg model for @xmath24 due to abouzaid @xcite , apply the homological perturbation lemma to transfer it to an @xmath1-structure on @xmath2 , then explicitly find a gauge transformation that kills @xmath443 and @xmath444 . the resulting @xmath541 can then be seen to represent a non - zero class @xmath39 . the details of our computer - assisted calculation are given in the appendix [ appendix ] . one can also consider the closed torus @xmath9 . there is again an open - closed string map , which is an isomorphism @xcite of graded @xmath542-algebras @xmath543 the base field ( on both sides ) is the complex novikov field @xmath544 , consisting of formal series @xmath545 , where @xmath546 is a function such that @xmath547 $ ] is finite for any @xmath548 . on the right - hand side , @xmath549 is the full subcategory of the fukaya category of @xmath9 with objects @xmath22 and @xmath23 . [ [ weierstrass - cubics . ] ] weierstrass cubics . + + + + + + + + + + + + + + + + + + + we draw the reader s attention to an algebro - geometric interpretation of @xmath550 . we have seen that when @xmath551 , gauge - equivalence classes of @xmath1-structures on @xmath2 are parametrized by the 2-dimensional vector space @xmath550 , on which the @xmath452-action rescaling the structure maps has weights @xmath552 . this is reminiscent of the family of cubic curves @xmath553 defined by the the weierstrass equation @xmath554 considered as a family over @xmath555 ( see e.g. ( * ? ? ? 2 ) ) . here @xmath556 are coordinates on @xmath555 , @xmath557 $ ] homogeneous coordinates on @xmath558 . each fiber in this family carries a non - zero differential @xmath559 , and therefore by serre duality a basis for @xmath560 . over the locus where the discriminant @xmath561 is non - zero , the fiber @xmath562 is an elliptic curve . when @xmath563 and @xmath564 , @xmath562 has a node ; and when @xmath565 , it has a cusp . the action of @xmath452 on the base @xmath566 , @xmath567 , is covered by an action on the family , @xmath568 . there are two distinguished objects in the dg category of coherent sheaves on @xmath562 : the structure sheaf @xmath569 , and the skyscraper sheaf @xmath570 at the point @xmath571 $ ] at infinity , which is always a regular point . the full subcategory @xmath572 generated by these two objects comes with an isomorphism @xmath573 . one has @xmath574 , but we can use the basis for @xmath560 to identify this exterior algebra with @xmath575 . this explains why @xmath573 is canonical . [ weierstrass ] the map @xmath576 , @xmath577 , is a linear bijection . the crucial observation is that one can rebuild the homogeneous coordinate ring @xmath578and hence the elliptic curve @xmath562from the dg category @xmath572 , or equally from its @xmath1 minimal model . one inductively expresses @xmath579 as a twisted complex in @xmath569 and @xmath570 via exact triangles @xmath580 $ ] and @xmath581 $ ] . the ring structure is then realized via composition in @xmath582 . one also recovers the basis of @xmath583 serre - dual to the differential @xmath559 . these data are sufficient to determine @xmath556 . this proves that @xmath3 are gauge - invariants of an @xmath1-structure ( @xmath502 ) , and also shows that @xmath49 is bijective . we next claim that @xmath49 is a polynomial map ( we thank paul seidel for technical assistance with this point ) . being polynomial and @xmath452-equivariant , @xmath49 is necessarily linear . the dga @xmath584 can be realized , over the base @xmath585 $ ] , as @xmath586 , where the cochain complex @xmath587 of @xmath588-modules @xmath587 is the ech complex for an open cover by two affines , with coefficients in the sheaf @xmath589 . one checks that @xmath587 is a complex of free @xmath588-modules . the cohomology @xmath590 is a free @xmath588-module , so one can write @xmath591 for a submodule @xmath592 projecting isomorphically to @xmath593 . since @xmath594 is a projective @xmath588-module , we can split @xmath595 and so write @xmath596 . the projection map @xmath597 and the inclusion @xmath598 are then homotopy - inverses ; there is a contraction of @xmath435 to @xmath599 . applying the homological perturbation lemma to this contraction , one obtains an @xmath1-structure on @xmath599 , over @xmath588 . theorem [ bigradedhh ] remains valid over @xmath588 ( a flat @xmath32-module ) . points 1 - 2 of theorem [ classification ] are also valid the proof involves no division and so one obtains a class @xmath600 . the latter module is isomorphic to @xmath601 $ ] by a canonical map sending @xmath3 to @xmath49 ; hence @xmath49 is polynomial . a natural follow - up will be to compare this result with polishchuk s calculation @xcite . we will explain elsewhere that @xmath25 has parameters @xmath602 which lie on the discriminant curve @xmath603 . since @xmath604 , it follows that @xmath82 is quasi - isomorphic to the category @xmath572 for a nodal cubic an instance of homological mirror symmetry . * acknowledgments : * yl thanks mit for generous hospitality during the writing of this paper . tp was partly supported by by nsf grant 1049313 . we both thank mohammed abouzaid , david ben - zvi , pavel etingof and paul seidel for sharing their ideas and expertise , and we also thank seidel for writing a computer program that provided a check on our calculations . let @xmath24 be the @xmath1-algebra given by the direct sum of the four hom - spaces in the full subcategory of @xmath82the exact fukaya category of a once - punctured 2-torus whose objects are the two oriented lagrangian branes . the underlying curves meet transversely at a single point , and their fundamental classes span @xmath605 . they are equipped with gradings such that all hom - spaces are supported in degrees 0 and 1 , and with non - trivial spin - structures . we showed that , over a field @xmath32 in which 6 is invertible , @xmath24 determines a primary deformation class @xmath606 , where @xmath2 is the @xmath32-algebra @xmath607 . this class is a quasi - isomorphism invariant of @xmath24 ; if it is non - zero , @xmath24 is necessarily non - formal , i.e. , not quasi - isomorphic to @xmath2 with the ` trivial ' @xmath1-structure ( the one with all structure - maps vanishing except the 2-fold product ) . we have verified , with the assistance of some simple computer programs written in python 3.0 , that @xmath608 in this appendix we document this calculation . consider a symplectic manifold with two embedded lagrangian submanifolds @xmath609 and @xmath610 intersecting transversely . let @xmath49 be a regular neighborhood of @xmath611 . abouzaid s article @xcite describes ( under certain assumptions ) a dg algebra quasi - isomorphic to the full subcategory of the exact fukaya category of @xmath49 whose objects are @xmath609 and @xmath610 equipped with gradings and relative spin structures . observe next that there is a direct sum splitting @xmath620 , where @xmath621 , @xmath622 , and @xmath623 . we take @xmath592 to be the direct sum of @xmath624 , @xmath625 , @xmath626 , and the subspace @xmath627 , while @xmath628 . the expression @xmath620 defines a projection map @xmath629 , split by the inclusion map @xmath630 is the inclusion ; and a nullhomotopy @xmath631 of @xmath632 . one has @xmath633 , and @xmath634 for the remaining basis - vectors @xmath22 . the homological perturbation lemma ( we use the conventions of ( * ? ? ? * prop . 1.12 ) ) determines a minimal @xmath1-structure @xmath435 on @xmath2 ( minimal means that @xmath640 ) and an @xmath1 quasi - isomorphism @xmath641 extending @xmath642 . this structure is defined by an explicit recursion : for @xmath643 , @xmath644 since the image of @xmath9 is @xmath648 , one has @xmath649 . but @xmath650 , so @xmath651 for @xmath646 . inspecting the multiplication table of @xmath198 , we notice that @xmath652 so in fact @xmath653 even for @xmath654 . by examining @xmath655 we find that the non - zero terms in @xmath656 are @xmath657 and that if @xmath658 for some sequence of basis vectors @xmath659 with @xmath646 then we must have @xmath660 . hence , by induction , @xmath661 all other outputs for @xmath662 being zero . now assume @xmath646 . the map @xmath663 is non - zero only on @xmath664 , and so if @xmath665 then @xmath666 . hence the only non - zero @xmath667 products are @xmath668 and by a straightforward induction , these are as claimed . a _ gauge transformation _ @xmath464 for @xmath435 ( also known as a formal diffeomorphism ) is a sequence of degree - preserving @xmath32-linear maps @xmath669 $ ] , starting with @xmath670 . gauge transformations form a group which acts on the left on minimal @xmath1-structures , defined by the following explicit recursive formulae : the sum in the first term is over all @xmath672 and partitions @xmath673 . we write down a gauge transformation @xmath464 such that @xmath674 . its terms @xmath675 with @xmath676 are zero except for @xmath677 , which is as follows : @xmath678 our first computer - aided check is that @xmath679 has @xmath680 . the fourth - order product @xmath681 then has the following non - zero terms : @xmath682 the structure @xmath688 has vanishing @xmath443 and @xmath444 . ( checking that @xmath444 vanishes involves enough equations that computer aid is very useful here . ) it is possible to kill @xmath445 as well , but we do not need to do so , because the sixth - order product @xmath541 already defines a hochschild cocycle @xmath689 . we have carried out a computer - aided calculation of @xmath541 in @xmath690 , and verified that the computer indeed outputs a cocycle . rather than tabulating the results , we shall instead pick out a small subset of the data and use it to verify that the class @xmath691\in hh^2(a , a)^{2 - 6}$ ] is non - zero . though computer - guided , this part of the argument is human - checkable . suppose that @xmath692 for some @xmath693 ( the reason for the normalization will become apparent ) , where @xmath307 is the hochschild coboundary operator . then , in @xmath690 , @xmath694 we check that @xmath695 hence @xmath696 similarly , @xmath697 so @xmath698 . next , @xmath699 so @xmath700 . the _ coup de grce _ is the self - contradictory equation @xmath701 this proves that @xmath4 .

we study the perturbative evolution of the static configurations , quasinormal modes and quasi normal ringing in the apostolatos - thorne cylindrical shell model . we consider first an expansion in harmonic modes and show that it provides a complete solution for the characteristic value problem for the finite perturbations of a static configuration . as a consequence of this completeness we obtain a proof of the stability of static solutions under this type of perturbations . the explicit expression for the mode expansion are then used to obtain numerical values for some of the quasi normal mode complex frequencies . some examples involving the numerical evaluation of the integral mode expansions are described and analyzed , and the quasi normal ringing displayed by the solutions is found to be in agreement with quasi normal modes found previously . going back to the full relativistic equations of motion we find their general linear form by expanding to first order about a static solution . we then show that the resulting set of coupled ordinary and partial differential equations for the dynamical variables of the system can be used to set an initial plus boundary values problem , and prove that there is an associated positive definite constant of the motion that puts absolute bounds on the dynamic variables of the system , establishing the stability of the motion of the shell under arbitrary , finite perturbations . we also show that the problem can be solved numerically , and provide some explicit examples that display the complete agreement between the purely numerical evolution and that obtained using the mode expansion , in particular regarding the quasi normal ringing that results in the evolution of the system . we also discuss the relation of the present work to some recent results on the same model that have appeared in the literature .

1206.5861

the general analysis of the evolution of material systems interacting with their own gravitational field is undoubtedly a very important and very difficult problem in general relativity . as usual in any physical theory , one can gain insight into the more difficult issues by considering idealized simplified models that , nevertheless , still preserve some of the relevant features of the general problem . one of these simplified examples was described some time ago by apostolatos and thorne @xcite . the material contents in this example is given by a cylindrical shell of counter rotating particles , and one is interested in the dynamics that results as a consequence of its interaction with its own gravitational field . as shown in @xcite , once some appropriate choice of coordinates is made , it is a straightforward matter to obtain a set of coupled equations for the evolution of the matter and field variables , which is described in some detail in the next section . in their original paper apostolatos and thorne were interested in general features of the evolution , in particular the possibility of naked singularities forming as the result of the collapse of these structures , but the detailed evolution of the shell in a situation close to one of its static configuration was only qualitatively mentioned , with no detailed results . this problem is interesting in particular because in the newtonian approximation the shell is either in a static configuration or performs periodic oscillations about this static configuration , but , once the radiative modes of the gravitational field are introduced , one expects these oscillations to be eventually damped as the results of the transfer of energy to the radiative modes . thus , in the fully relativistic description one expects to find the characteristic `` quasi normal ringing '' ( qnr ) , related to the `` quasi normal modes '' ( qnm)@xcite of the system . a recent analysis of the dynamics of the apostolatos and thorne model was given by hamity , et.al . we notice , however , that the approximations introduced in the explicit examples considered in @xcite are such that the radiative modes are effectively neglected , and , therefore , the results obtained are possibly relevant only close to the newtonian limit . a detailed analysis of the full relativistic equations in the linearized approximation was carried out by the present authors in @xcite , where we obtained the general solution for the harmonic modes with real frequencies , and considered the solutions obtained by general linear combinations of these modes . although in all the examples considered in @xcite we found a stable evolution under perturbations of ( essentially ) compact support , at least two questions remained unanswered . the first was the possible completeness of the mode expansion , and its relation to the initial value problem for the system . the second was the apparent lack of qnr in the example solutions . the main purpose of the present work is to give answers to those questions . we first review in sections 2 and 3 , mostly for completeness , the main features of the model . then , in section 4 , we review the derivation of the periodic solutions in the linearized regime , introducing a modified form , as compared with that given in @xcite , that allows us to establish , in section 5 , the completeness of expansion in the context of the characteristic value problem for the system , and , therefore , the completeness of the mode expansion . this completeness , in turn , as discussed in section 6 , implies the linear stability of static configurations under this type of perturbations , a result that differs from some conclusions reached in @xcite , and recent results obtained by kurita and nakao @xcite . next , in section 7 , we discuss the existence of qnms for the system . as shown there , these are related to the zeros of a complex expression that appears in the relation between the incoming and outgoing wave amplitudes as functions of the frequency . this expression has a complicated form involving bessel functions that makes it very difficult its analysis . for this reason we have not been able to obtain explicit expressions for the frequencies of the qnm , or even to ascertain whether there is finite number or infinite number of these frequencies . nevertheless , by resorting to numerical methods we found at least a couple of qnm corresponding to a range of parameters characterizing static configurations . the values obtained are also displayed in section 7 . in section 8 we consider the numerical evaluation of the integrals obtained in section 7 that provide the evolution of the system for arbitrary characteristic data . the examples chosen display clearly the quasi normal ringing associated to one of the qnm . the reason why this qnr was not seen in @xcite is discussed in section 12 , using the price - husain toy model as an illustration . in section 9 we consider again the full relativistic equations and assuming a general linear departure from a static configuration obtain the general linearized equations for the system . then , in section 10 we prove the existence of an associated positive definite constant of the motion that puts absolute bounds on the dynamic variables of the system , establishing the stability of the motion of the shell under arbitrary , finite perturbations . in section 11 we show that the corresponding set of coupled ordinary and partial differential equations for the relevant dynamical variables can be solved numerically as an initial plus boundary values problem , without resorting to expansions in terms of bessel or other functions . we provide a couple of examples , using the same shell parameters as in the mode expansion , but a different form for the incoming wave form . the results , that clearly display quasi normal ringing are in complete agreement with those of the mode expansion , and with the computed qnm frequencies , although the techniques used in both approaches are totally different , confirming their separate correctness . we end the paper with some comments and conclusions , as well as comments on related work by other authors , in particular to that of reference @xcite . the apostolatos - thorne model @xcite describes the dynamics of a self gravitating cylindrical shell of counter rotating particles . both the inner ( @xmath0 ) and outer ( @xmath1 ) regions of the shell are vacuum space times with a common boundary @xmath2 . the corresponding metrics may be written in the form , @xmath3 where the @xmath4 sign corresponds to the outside and @xmath5 to the inner regions . the functions @xmath6 , and @xmath7 depend only on @xmath8 and satisfy the equations : @xmath9 @xmath10\ ] ] the shell is located on the hypersurface @xmath2 given by @xmath11 , where @xmath12 is the proper time of an observer at rest on the shell . we may interpret @xmath13 as playing the role of a gravitational field whose static part is the analogue of the newtonian potential . the time dependent solutions of ( [ ateq2 ] ) represent gravitational waves ( einstein - rosen ) . equation ( [ ateq2 ] ) is the integrability condition of eqs . ( [ ateq3 ] ) . the coordinates @xmath14 and the metric function @xmath6 are continuous across the shell @xmath2 , while @xmath15 and the metric function @xmath7 are discontinuous . smoothness of the spacetime geometry on the axis @xmath16 requires that @xmath17 , @xmath6 finite at @xmath18 , and @xmath19 . the junction conditions of @xmath0 and @xmath1 through @xmath2 require the continuity of the metric and specify the jump of the extrinsic curvature @xmath20 compatible with the stress energy tensor on the shell . the induced metric on @xmath2 is given by @xmath21 here @xmath22 . the evolution of the shell is characterized by @xmath23 , which is the radial coordinate @xmath24 at the shell s location and @xmath12 the proper time of an observer at rest on @xmath2 . if we assume that the shells is made up of equal mass counter rotating particles , the einstein field equations on the shell may be put in the form , @xmath25 @xmath26 where the constants @xmath27 and @xmath28 are , respectively , the proper mass per unit killing length of the cylinder and the angular momentum per unit mass of the particles . the other quantities in ( [ ateq5],[ateq6 ] ) are given by , @xmath29 @xmath30 where a dot indicates a @xmath12 derivative , and we also have , @xmath31 \nonumber \\ & & + \frac { r^2 \psi^-_{,n } x^-}{r^2+e^{2 \psi_{\sigma}}j^2 } -\frac{\lambda r^2 x^-}{(r^2+e^{2 \psi_{\sigma}}j^2)^{3/2 } } + \frac{j^2 e^{2 \psi_{\sigma}}x^-x^+}{r(r^2+e^{2 \psi_{\sigma}}j^2)}\end{aligned}\ ] ] these equations together with ( [ ateq2],[ateq3 ] ) determine the evolution of the shell and of the gravitational field to which it is coupled . we briefly review the conditions for a static solutions , corresponding to a shell of constant radius @xmath32 . we restrict to the case of a regular interior . in this case the interior must be flat and we may take @xmath33 , @xmath34 , implying @xmath35 . for the exterior field the general static solution ( satisfying @xmath35 ) is of the form , @xmath36 then , @xmath37 since @xmath38 , @xmath39 , and @xmath40 , we find , @xmath41\ ] ] and , @xmath42 it is important to notice that this relation can be satisfied for real @xmath43 , @xmath28 and @xmath27 only if @xmath44 , with @xmath45 therefore , the system admits static solutions only if @xmath44 . it will be convenient to write @xmath46 . replacing in ( [ eq4a ] ) , we find , @xmath47 which we now consider as an equation for @xmath48 as a function of @xmath27 . it is now easy to check that there are no real solutions for @xmath48 if @xmath49 . for @xmath50 we have a double solution @xmath51 . the interesting region for our analysis is the range @xmath52 where we have _ two _ real and positive solutions for @xmath48 , one larger and one smaller than @xmath53 , which approach respectively @xmath54 and @xmath55 as @xmath56 . this implies that for appropriate @xmath27 and @xmath28 we have _ two _ static solutions . it was concluded in previous analysis @xcite that , at least from a perturbative point of view , one of these solutions ( the one with the larger @xmath43 ) would be stable and the other ( with the smaller @xmath43 ) would be unstable . this conclusion , as shown in the following sections , is _ not _ supported by the present analysis of the perturbative evolution that follows from initial data in the neighborhood of either static configuration . in the next section we derive the appropriate equations for a perturbative analysis , and then show how this construction can be used to solve the evolution equations given characteristic data . in this section we consider the problem of finding linearized periodic solutions for our system imposing the condition of a regular interior . this problem was already considered in @xcite , where the main idea was to obtain stationary periodic solutions , but left open the problem of relating these solutions to the initial value problem . here we will consider a slightly modified approach , where we express the solutions in @xmath57 in terms of incoming and outgoing wave solutions for the radiative part of @xmath58 . we first notice that if we impose regularity on the symmetry axis @xmath59 , and assume that the interior region is empty but may contain gravitational radiation , then @xmath60 admits periodic solutions of the form @xcite , @xmath61 where @xmath62 is a bessel function , and we consider @xmath63 as a quantity of first order in perturbation theory . then , restricting again to linearized order we may set , @xmath64 since , in this case , from ( [ ateq3 ] ) , @xmath65 is of second order , and , therefore , also to the appropriate order , we may also set , @xmath66 we assume a periodic perturbation around an equilibrium configuration characterized by @xmath43 and @xmath28 and therefore take , @xmath67 where we assume that @xmath68 is also of first order . we need to specify now a corresponding solution for @xmath58 . in our previous analysis , @xcite , we noticed that the general periodic perturbation for @xmath58 may be written as , @xmath69 e^{i \omega_+ t_+}\ ] ] where @xmath70 is a bessel function and @xmath71 and @xmath72 are first order quantities . to this order we have , @xmath73 e^{i \omega_+ t_+}\ ] ] where @xmath74 and @xmath75 are given by ( [ eq3a ] ) . a straightforward calculation @xcite shows that consistency of the equations at first order requires @xmath76 , @xmath77 and , @xmath78 replacing now in ( [ ateq5 ] ) , ( [ ateq6 ] ) , and ( [ ateq9 ] ) , and expanding to first order , we find a set of three linear independent equations for @xmath63 , @xmath71 , @xmath72 , and @xmath68 , for every choice of @xmath79 , @xmath43 and @xmath28 , and , therefore , three of these quantities can be solved in terms of the fourth . since all the evolution equations are linear , we may consider linear superpositions of these solutions to obtain more general ( non periodic ) solution . in @xcite , for simplicity , we chose @xmath68 as the independent variable , and obtained corresponding expressions for the other three amplitudes . assuming an arbitrary dependence of @xmath68 on @xmath79 , given by the function @xmath80 , @xmath23 is given by , @xmath81 while for @xmath82 and @xmath83 we obtain expression in the form of fourier transforms of @xmath80 multiplied by complicated expressions involving bessel functions of @xmath84 and @xmath85 . details are given in @xcite . we notice that although in principle , ( [ eq12da ] ) is completely general , and one has a definite @xmath80 for any possible @xmath23 , in practice , what one would like to solve is the problem of the evolution of the system given some initial condition , and because of the structure of the equations that resulted in @xcite , the best one could do was to assume some form for @xmath80 , and find the corresponding evolution of the system . carrying out this program it was found there that there are rather general evolutions , in the sense that they represent the interaction of the shell with an incoming gravitational pulse of rather arbitrary shape , for which the response of the shell is given precisely by @xmath80 . an important question here is actually how general these solutions of the problem are , namely , whether they include every possible evolution of the system , or on the contrary , they only represent a restricted set . since we had no proof of the completeness of the set of modes , this question is highly non trivial . further , the scheme developed in @xcite did no appear to show the quasi normal modes and quasi normal ringing expected in a system such as this , that has a simple newtonian limit , where the shell can execute periodic motions . one would expect that the presence of radiative modes in the general relativistic version should lead to damped oscillations , at least in some appropriate limit , and this was not immediately apparent in the discussion given in @xcite . as we shall show here , both the completeness problem and the question of the qnr can be answered by a simple change in the formalism presented in @xcite . this can be achieved as follows . the expression , @xmath86\ ] ] where @xmath87 and @xmath88 are constants , is a solution of ( [ ateq2 ] ) for @xmath58 . for large @xmath24 , using the asymptotic expansions for the bessel functions @xmath62 and @xmath70 , we find , @xmath89 and , therefore , ( [ eq5d ] ) represents a purely incoming wave solution of ( [ ateq2 ] ) . similarly , @xmath90\ ] ] where @xmath91 is a constant , is a solution of ( [ ateq2 ] ) , for @xmath58 , which , for large @xmath24 has the asymptotic behaviour , @xmath92 and , therefore , ( [ eq7d ] ) represents a purely outgoing wave solution of ( [ ateq2 ] ) . we notice now that we have , @xmath93 e^{i \omega_2 t_+}\ ] ] therefore , instead of ( [ eq4da ] ) we may write the full solution for @xmath94 in the form , @xmath95 where we have the identifications : @xmath96 in what follows we will consider @xmath87 and @xmath91 as first order quantities . to this order , from ( [ ateq3 ] ) , we have , @xmath97 where @xmath74 and @xmath75 are given by ( [ eq3a ] ) . the important point here is that we have now expressions for @xmath58 and @xmath98 where we have separate amplitudes for the incoming and the outgoing wave parts . notice that since ( [ eq9d ] ) is just a reordering of terms in ( [ eq4da ] ) , all the previous relations regarding @xmath99 , @xmath79 , and @xmath100 are valid . then replacing again the expressions for the dynamic variables , but using now ( [ eq9d ] ) and ( [ eq10d ] ) , in ( [ ateq5 ] ) , ( [ ateq6 ] ) , and ( [ ateq9 ] ) , and expanding to first order , we find this time a set of three linear independent equations for @xmath63 , @xmath87 , @xmath91 , and @xmath68 . in this paper we will be mainly interested in the evolution of an initially static shell , under the influence of a bounded incoming compact pulse . we choose therefore to solve these equations for @xmath63 , @xmath91 , and @xmath68 in terms of @xmath87 . the resulting explicit expressions are , unfortunately , rather long and difficult to read . nevertheless , they have in general the structure , @xmath101 where @xmath102 are bounded regular functions of @xmath103 and @xmath104 is given by , @xmath105 \nonumber \\ & & \left ( \left| \omega \right| j_{{0 } } \left ( \left| \omega_{{2 } } \right| r_{{0 } } \right ) - i\omega\,y_{{0 } } \left ( \left| \omega_{{2 } } \right| r_{{0 } } \right ) \right ) \nonumber \\ & & - \left [ \left ( { r_{{0}}}^{2 } \left ( { r_{{0}}}^{2}+{j}^{2 } \right ) ^{2}{\omega}^{2}-{j}^{2 } \left ( 2{r_{{0}}}^{2}+{j}^{2 } \right ) \right ) j_{{0 } } \left ( \left| \omega \right| r_{{0 } } \right ) \right . \nonumber \\ & & \left.+2{j}^{2 } \omega r_{{0 } } \left ( { r_{{0}}}^{2}+{j}^{2 } \right ) j_{{1 } } \left ( \left| \omega \right| r_{{0 } } \right ) \right ] \omega\ , \left ( 2\,{j}^{2}+{r _ { { 0}}}^{2 } \right ) ^{2 } \nonumber \\ & & \left ( \omega\,j_{{1 } } \left ( \left| \omega_{{2 } } \right| r _ { { 0 } } \right ) -i \left| \omega \right| y_{{1 } } \left ( \left| \omega _ { { 2 } } \right| r_{{0 } } \right ) \right)\end{aligned}\ ] ] where @xmath106 . in spite of its appearance it is not difficult to show that @xmath104 has no zeros for real @xmath79 . a zero of @xmath104 for real @xmath79 would imply that the imaginary part of @xmath107 vanishes . but this is given by , @xmath108 where we have used properties of the bessel functions to obtain the right hand side , and , therefore , the imaginary part is non vanishing for all @xmath79 . this result is crucial , because it means that if @xmath87 depends regularly on @xmath79 , the solutions obtained using ( [ 13d1 ] ) and ( [ 13d2 ] ) are well defined modes for all @xmath79 in @xmath109 . therefore , because of linearity , arbitrary linear combinations of these modes will also be solutions of the problem . before discussing the possible completeness of this set of modes we will write explicit forms for the expressions we have in mind . these are given by , @xmath110 d \omega\ ] ] where @xmath111 is an arbitrary complex function of @xmath79 which we will assume at least square integrable.then , taking into account ( [ 13d1 ] ) , we have for the remaining dynamical variables , @xmath112 d \omega \nonumber \\ \xi(\tau ) & = & \int_{-\infty}^{+\infty } \frac{h_3}{d } f(\omega ) e^{i \omega \tau } d \omega\end{aligned}\ ] ] we shall analyze the properties of this expansion in the next section , and show that they solve a well defined characteristic value problem for our system , and , therefore , in this sense , the mode expansion is also complete . we remark , for completeness , that ( [ 16d ] ) , for real @xmath79 , represents the most general solution for @xmath113 , such that both @xmath114 are bounded for large @xmath24 . consider again ( [ 16d ] ) . the problem that we have in mind is one where at some given time the shell and a large neighbourhood of the space time that includes the shell and the symmetry axis , are in a static state , but there is a finite gravitational pulse incoming from large @xmath24 . in other words , if @xmath43 corresponds to a static solution , then for some large time in the past and some @xmath115 , the solution in @xmath116 coincides with the static solution . for @xmath117 , on the other hand , there is region where @xmath113 is non vanishing . to make this more precise , we recall that for large @xmath24 an incoming wave solution for @xmath94 has the asymptotic form , @xmath118 where @xmath119 is an arbitrary function of ( essentially ) compact support . we may compare this with the asymptotic form for large @xmath24 of ( [ 16d ] ) , @xmath120 which can be inverted to give , @xmath121 therefore , we find a one to one relation between @xmath111 and characteristic data given for @xmath122 , @xmath123 . in this sense , we have proved the following result : _ the set of modes given by ( [ 13d1 ] ) and ( [ 13d2 ] ) , used in the construction of ( [ 16d ] ) and ( [ 17d ] ) is complete , and , in particular , ( [ 16d ] ) and ( [ 17d ] ) provide the evolution of the corresponding dynamical variables for arbitrary characteristic data given at @xmath122 , @xmath123 . _ in the next two sections we shall use these results to analyze the stability of the static solutions and then consider the presence of quasi normal modes ( qnm ) and their associated quasi normal ringing ( qnr ) . conceptually , we may say that a static configuration of the shell is stable if , given data arbitrarily close to that configuration , the system evolves towards that static configuration . otherwise we would say that the static configuration is unstable . but , as we have already indicated , we have explicit forms for the evolution of the system when it is perturbed from an initially static configuration by an incoming gravitational pulse of ( essentially ) compact support . each one of these pulses is uniquely characterized by a corresponding function @xmath111 . concentrating in particular on @xmath124 , we have , @xmath125 and , using ( [ 13d2 ] ) , we can check that for large @xmath79 , @xmath126 plus terms of order @xmath127 . as consequence , on account of ( [ 17d ] ) , and our assumptions on @xmath111 , @xmath124 is the fourier transform of a square integrable function for _ any _ @xmath28 and @xmath43 , but then , we must have @xmath128 for @xmath129 , and , therefore , _ all _ static configurations are stable . more explicitly , we have shown that all finite admissible gravitational pulses incoming from large @xmath24 can be described by an appropriate function @xmath111 , and that this function is in one - to - one correspondence with the shape of the pulse at large @xmath24 . we have also shown there is a unique evolution for the system corresponding to each function @xmath111 . this evolution is such that the dynamic variables are essentially given by the fourier transform of a square integrable functions and , therefore , they are themselves square integrable and must vanish in the limit @xmath130 . but this shows that all initially static configurations of the shell that are perturbed by an incoming pulse of the type described above , will eventually settle back to their original static configuration , and , in this sense , they are all _ stable . _ just for completeness , it should be clear that we have not found a self adjoint extension associated to the coupled partial plus ordinary system of differential equations of this problem , so that we could not state completeness of the mode expansion in the same way as when such an extension is possible . what we have found is that the characteristic value problem can indeed be solved by our mode expansion , since it reduces essentially to a fourier transform , and , that in this sense , the expansion is complete , and proves that the static configurations of the system are stable under a perturbation that has the form of a bounded , ( essentially ) compactly supported incoming pulse . but this is essentially all that is required to consider the system physically stable . it is still an open and interesting question to find the relation , if it exists , between the system of equations governing the evolution of the system and an associated self adjoint problem that would be more useful an the analysis of the initial value problem . nevertheless , as show in section 10 , for the general initial plus boundary value problem we have a conserved , positive definite constant of the motion , and therefore we have stable evolution for any finite initial data . + we have already seen that for a shell with given values of @xmath27 and @xmath28 there may be just one , two or no static solutions of the equations of motion . it has been suggested @xcite that in the case of two solutions only one is stable and the other is unstable . however , our present analysis does not show a qualitative difference in the behaviour of two static solutions under perturbations . there appears to be , nevertheless , a more subtle difference between the solutions when we analyze the evolution in more detail . this is related to the presence of quasi - normal modes , and is considered in the next section . in this section we analyze the presence of quasi normal modes and the associated quasi normal ringing in our system . very roughly , qnm and qnr appear in systems coupled to a field that admits a decomposition in periodic incoming and outgoing waves.(see e.g. @xcite for more details , and further references ) . generally qnm are associated to non trivial solutions where the incoming wave amplitude vanishes . this requires , generally , a complex value for the period , and therefore , in general , the corresponding outgoing and other amplitudes are unbounded in space and time , and , therefore , are not physical . in our problem , we have seen that for the periodic solutions , the dynamical variables other than @xmath111 are given by expressions of the form @xmath131 , and therefore , qnm , that is , nontrivial solutions with @xmath132 , can exist only for values of @xmath79 such that @xmath133 . we have already seen that @xmath104 is non vanishing for _ real _ @xmath79 , but it may vanish for _ complex _ values of @xmath79 . these zeros of @xmath104 introduce complex poles in the expressions for @xmath134 and the other field variables whose complex frequency does not depend on @xmath87 although the amplitude of their possible contributions does depend on @xmath87 . since @xmath134 , @xmath63 and @xmath91 are given by fourier transforms of these expressions containing complex poles , we may get contributions dominated by these poles . these are the qnr amplitudes associated to the qnm . therefore , to find the qnm we need to find the complex zeros of @xmath104 . in principle this is the simple matter of solving the equation @xmath133 for @xmath79 . unfortunately , if we consider the explicit expression for @xmath104 given by ( [ 13d2 ] ) , we notice that it contains bessel functions of different types and arguments depending on @xmath79 , so that only a numerical computation of the zeros appears as feasible , but even in this case the accurate computation of these zeros is a complicated task . in this paper we only attempt a rather rough computation of the zeros closest to the real @xmath79 axis in the complex @xmath79 plane , as these would correspond to the most noticeable qnr of the system . for this explicit computation we go back to ( [ 13d1 ] ) and notice that if define @xmath135 and @xmath136 , the equation @xmath133 may be written in the form , @xmath137 \nonumber \\ & & \left ( - \left| \sigma \right| j_{{0 } } \left ( \left| \sigma \right| \left ( 2\,{x}^{2}+1 \right ) ^{2 } \right ) + i\sigma\,y_{{0 } } \left ( \left| \sigma \right| \left ( 2\,{x}^{2}+1 \right ) ^{2 } \right ) \right ) \\ & & - \left ( \sigma\,j _ { { 1 } } \left ( \left| \sigma \right| \left ( 2\,{x}^{2}+1 \right ) ^{2 } \right ) -i \left| \sigma \right| y_{{1 } } \left ( \left| \sigma \right| \left ( 2\,{x}^{2}+1 \right ) ^{2 } \right ) \right ) \nonumber \\ & & \left [ \left ( { \sigma}^{2 } \left ( 1+{x}^{2 } \right ) ^{2}-{x}^{2 } \left ( { x}^{2}+2 \right ) \right)j _ { { 0 } } \left ( \left| \sigma \right| \right ) + 2 \sigma{x } ^{2 } \left ( 1+{x}^{2 } \right ) j_{{1 } } \left ( \left| \sigma \right| \right ) \right ] \sigma \left ( 2{x}^{2}+1 \right ) ^{2 } \nonumber\end{aligned}\ ] ] actually , since we are looking for zeroes near the real @xmath79 axis , we should make the replacements @xmath138 to look for zeros in the region @xmath139 , and @xmath140 in the region @xmath141 . in any case , the zeroes of @xmath104 satisfy the scale law , @xmath142 where @xmath143 is some complex function of its real argument . we searched for zeroes of @xmath104 using two procedures . in the first we noticed that if we look for zeroes in the @xmath144 region we may replace @xmath138 in ( [ eq01f ] ) , and after canceling some common factor of @xmath2 , we find a maximum power @xmath145 in the coefficients of the bessel functions . we formally solve for this factor and obtain , @xmath146 { x } ^{2 } \\ & & \left [ \left ( \left ( iy_{{1 } } \left ( \sigma\ , \left ( 2\,{x}^{2}+1 \right ) ^{2 } \right ) -j_{{1 } } \left ( \sigma\ , \left ( 2\,{x}^{2}+1 \right ) ^{2 } \right ) \right ) j_{{0 } } \left ( \sigma \right ) \right . \nonumber \\ & & \left . \left.+ \left ( j_{{0 } } \left ( \sigma\ , \left ( 2\,{x}^{2}+1 \right ) ^{2 } \right ) -iy _ { { 0 } } \left ( \sigma\ , \left ( 2\,{x}^{2}+1 \right ) ^{2 } \right ) \right ) j_{{1 } } \left ( \sigma \right ) \right ) \left ( 2\,{x}^{2}+1 \right ) ^{2 } \left ( 1+{x}^{2 } \right ) ^{2 } \right]^{-1 } .\nonumber\end{aligned}\ ] ] we can use this equation in an iterative scheme where we input a value for @xmath2 on the right hand side , and the cubic root of the ( complex ) number obtained is inserted again as an `` improved '' value for @xmath2 . we have numerically checked that this works very well for sufficiently small values of @xmath48 , say @xmath147 , but stops converging for larger values of @xmath48 . a different , in a way more direct , although not very accurate method that we have used is to consider again ( [ eq01f ] ) , fix a value of @xmath48 , and plot the real and the imaginary parts of ( [ eq01f ] ) , as functions of @xmath148 , for fixed @xmath149 . with this method one can easily visualize the possible common zeros of the curves and adjust , by trial an error , the best values of @xmath148 and @xmath149 . we have found by this method that there are complex zeros of @xmath104 at least in the range @xmath150 , and possibly for larger values of @xmath48 . we have found zeros both near @xmath144 and @xmath151 . some of these are shown in table i. + . the zeros of @xmath104 . [ cols="^,^,^",options="header " , ] the first thing to notice is that all these zeroes have positive imaginary part , as one would expect from the stability arguments of the previous section . we also notice that there are zeros both for @xmath144 and @xmath151 , although having different imaginary parts and different @xmath152 . in general the imaginary part of @xmath2 increases with @xmath48 , and the modes become very strongly damped for @xmath48 of the order or larger than one . all these results are in good agreement with the `` first class solutions '' of @xcite for @xmath153 , although they obtain a much larger set of solutions . regarding this point we remark that , unfortunately , the analytic structure in the complex @xmath79 plane of the integrands in ( [ 17d ] ) is very complicated , and the exact relation between these zeroes and the evolution of the dynamical variables is not easily established . to obtain more information on the subject of qnr for our system we carried out numerical integrations of @xmath124 , as given by ( [ 17d ] ) , for simple incoming @xmath87 , and found examples of qnr , with parameters close to those of the zeros of @xmath104 found here this is detailed in the next section . next , in section ix , we consider again the perturbation problem in general , and obtain a set of coupled equations directly for @xmath134 , @xmath60 and @xmath58 , without resorting to a mode expansion . this allows us to set up an initial value problem that can be solved fully numerically , without the use of the mode expansion or bessel functions . we have carried out explicit numerical integrations of ( [ 17d ] ) to obtain @xmath124 , with the choice , @xmath154 this corresponds to an asymptotically incoming pulse of the form , @xmath155 as a first choice we took @xmath156 and @xmath157 . this corresponds to @xmath158 and , therefore , is within the range considered `` stable '' in previous analysis . the resulting @xmath124 is displayed in fig . 1 . we can see the characteristic shape of a qnr , that is , a damped oscillation . as usual , we can get a better look at this shape by displaying , as in fig . 2 , @xmath159 as a function of @xmath12 ( thick line ) . we also display in fig . 2 an approximate fit using the function @xmath160 , where @xmath161 , @xmath162 , @xmath163 and @xmath164 , although in the actual figure , for clarity , we set @xmath165 to avoid superposition of the two graphs . these parameters are in good agreement with those corresponding to the qnm with @xmath139 for @xmath158 shown in table i. as a second example we took @xmath156 and @xmath166 . this corresponds to @xmath167 and , therefore , is outside the range considered `` stable '' in previous analysis . the resulting @xmath124 is displayed in fig . 3 . we can see again the characteristic shape of a qnr , but , in this case , as a strongly damped oscillation . this result is also in good agreement with the fact that the qnm for @xmath167 shown in table i corresponds to a strongly damped oscillation . we remark once again that the analytic structure of the integrand in ( [ 17d ] ) is far from simple , and that although several sets of complex qnm frequencies can be established rather accurately , their effect on the resulting amplitudes can not be easily established . nevertheless , the integrals in ( [ 17d ] ) are taken along the real @xmath79 axis , and therefore , they are insensitive to the manner in which the integrand might be extended to , e.g. , the upper complex-@xmath79 plane , where the poles corresponding to the qnm reside . in the next section we consider a different approach , where we linearize the equations of motion and obtain a set of coupled linear equations for the amplitudes , together with a set of boundary and matching conditions that make possible to establish ( and numerically solve ) a well defined initial data problem . we consider again the set of dynamical equations and matching conditions for our problem , assuming that we are close to a static solution , characterized by certain values of @xmath28 and @xmath43 . we therefore write , @xmath168 with @xmath169 satisfying the equations , @xmath170 and expand the remaining equations to first order in @xmath171 . using these expansion we first obtain , @xmath172 in these equations @xmath74 and @xmath75 are given by ( [ eq3a ] ) and ( [ eq4a ] ) , so that the zeroth order terms coincide with the static solution of section iii . from these , again to first order we find , @xmath173 the explicit form of the terms of order @xmath171 is not required in what follows . choosing appropriately some integration constants we set , @xmath174 next we replace in the matching conditions and expanding again to first order we obtain two equations for @xmath175 and @xmath176 on the boundary @xmath177 @xmath178 and an equation of motion for @xmath124 , @xmath179 for explicit numerical integration it turns out to be convenient to introduce instead of @xmath180 a new function @xmath181 defined by , @xmath182 and , therefore , satisfying the equation : @xmath183 this way all the dynamic variables involved evolve directly in terms of @xmath12 . this simplifies the numerical treatment of the problem as it eliminates the need of separate grid spacing for @xmath184 and @xmath185 . but , most important , using @xmath186 we may demonstrate the existence of a crucial constant of the motion , as is shown in the next section . as discussed in the previous section , we have a complete solution for characteristic data , and we can use this fact to prove stability regarding the evolution resulting from that type of data . this , nevertheless , leaves still open the question of the evolution of general initial data . in other words , the possibility of the existence of initial data that somehow is not registered as characteristic data , but such that it renders the motion unstable , in the sense that , i.e. @xmath124 can acquire arbitrarily large values starting from finite initial data . to answer this question we consider again ( [ eq05h ] ) , replace @xmath176 in terms of @xmath186 , and solve for @xmath187 and @xmath188 . we get , @xmath189 and , @xmath190 replacing now ( [ stab01 ] ) and ( [ stab02 ] ) in ( [ eq06h ] ) , upon multiplication by @xmath191 , and some rearrangement , we find , @xmath192 we notice that the first line ( [ stab03 ] ) has the form of a total @xmath12-derivative . we may put the whole expression in this form as follows . we first take the derivative of ( [ stab01 ] ) with respect to @xmath12 , solve for @xmath191 and replace it in the second line in ( [ stab03 ] ) . next , we derive ( [ stab02 ] ) with respect to @xmath12 , solve again for @xmath191 , and replace it in the third line in ( [ stab03 ] ) . after a new rearrangement we get , @xmath193 where , again , all derivatives of @xmath175 and @xmath186 are evaluated at @xmath177 . the first two lines in ( [ stab04 ] ) are total @xmath12-derivatives . to analyze the last line we notice that , on account of ( [ eq01 ha ] ) we have , @xmath194 dr & = & \left.\left ( r \frac{\partial \chi_1(\tau , r)}{\partial \tau}\frac{\partial \chi_1(\tau , r)}{\partial r}\right ) \right|_0^{r_0 } \nonumber \\ & = & \left . r_0 \frac{\partial \chi_1(\tau , r_0)}{\partial \tau}\;\frac{\partial \chi_1(\tau , r)}{\partial r } \right|_{r = r_0}\end{aligned}\ ] ] where in the last line we have assumed that @xmath195 satisfies the ( regularity ) boundary conditions that @xmath196 is finite , and that @xmath197 . similarly , from ( [ eq08h ] ) , we have , @xmath198 dr & = & \left . \left ( r \frac{\partial \chi_3(\tau , r)}{\partial \tau } \frac{\partial \chi_3(\tau , r)}{\partial r}\right ) \right|_{r_0}^{\infty } \nonumber \\ & = & -\left . r_0 \frac{\partial \chi_3(\tau , r_0)}{\partial \tau}\;\frac{\partial \chi_3(\tau , r)}{\partial r } \right|_{r = r_0}\end{aligned}\ ] ] where we have assumed that @xmath199 is such that the integral exists , and , therefore , there is no contribution for @xmath122 . comparing ( [ stab05 ] ) and ( [ stab06 ] ) with ( [ stab04 ] ) we find that , if @xmath200 corresponds to a solution of ( [ eq08h ] ) that is regular for @xmath59 , ( and , therefore , @xmath201 for @xmath202 ) , and @xmath199 is such that @xmath203 for @xmath122 , ( which must happen for the integral in ( [ stab06 ] ) to exist ) , then the quantity , @xmath204 dr \\ & & + { \frac { { r_{{0}}}^{8 } } { 2\left ( { r_{{0}}}^{2}+{j}^ { 2 } \right ) \left ( 2\,{j}^{2}+{r_{{0}}}^{2 } \right ) ^{2}{j}^{2 } } } \int_{r_0}^{\infty}\frac{r}{2}\left[\frac{(2 j^2+r_0 ^ 2)^4}{r_0 ^ 8}\left(\frac{\partial \chi_3}{\partial \tau}\right)^2 + \left(\frac{\partial \chi_3}{\partial r}\right)^2 \right ] dr \nonumber\end{aligned}\ ] ] is a _ constant of the motion _ , i.e. , @xmath205 . notice that @xmath206 is positive definite for any non trivial solution of the equations of motion . therefore , it provides an absolute bound for each term in ( [ stab07 ] ) . thus , if at any time @xmath12 all the terms in ( [ stab07 ] ) are finite , then they will remain finite for any evolution that satisfies the condition of regularity for @xmath59 , demonstrating the absolute stability of the system under any finite perturbation for which all the terms in ( [ stab07 ] ) exist . of course , a simple example is the case of an incoming wave of ( essentially ) compact support , for which we have already demonstrated the stability . we should further notice that the term , @xmath207 in the second line of ( [ stab07 ] ) allows for mild compensating divergences of @xmath208 , and @xmath209 , provided they are compatible with the existence of the integrals in ( [ stab07 ] ) . we notice nevertheless that , taking into account ( [ stab01 ] ) , and ( [ stab02 ] ) , we may also write @xmath210 in the alternative forms , @xmath211 and , since both @xmath210 and @xmath124 are bounded , we find that both @xmath187 and @xmath188 must also remain bounded , and this implies that @xmath212 is also bounded . on account of ( [ eq02h ] ) , this result implies that @xmath213 is also bounded to first order . we remark that @xmath65 is directly of second order and vanishes to first order . therefore , the full metric remains bounded under arbitrary perturbations . one can now check that the system ( [ eq01 ha ] ) , ( [ eq05h ] ) , ( [ eq06h ] ) , and ( [ eq08h ] ) , with appropriate replacements of @xmath176 by @xmath186 , can be used to set up and solve numerically the initial plus boundary value problem where , for some arbitrarily chosen @xmath214 we give arbitrary values to @xmath215 , @xmath216 , to @xmath217 and @xmath218 , in the interval @xmath219 , and to @xmath220 and @xmath221 , in the interval @xmath222 , subject to the constraints @xmath223 , and those resulting from ( [ eq05h ] ) . although this procedure is well defined , and leads to a unique evolution for appropriate initial data , at least in its cauchy domain , we must remark that the well - posedness of this type of problems , as regards sensitivity to small changes in the initial data , is still an open and interesting issue , outside the limits of the present research . the purpose of this exercise is mainly to display the evolution of some appropriately chosen initial data , in particular of @xmath124 , in the region where we may expect the presence ( or absence ) of quasi normal ringing . as we shall see , this behaviour , at least in the region analyzed , is in complete agreement with our previous results . the integration procedure we have chosen uses a simple finite difference method for updating @xmath175 in @xmath224 , and for @xmath186 in @xmath225 , where @xmath226 is some appropriately chosen outer boundary . the values of @xmath134 are updated using a simple leap frog scheme . once these updates are carried out , we use the matching conditions ( [ eq05h ] ) to update @xmath175 , and @xmath186 at @xmath177 . we will be mainly interested in the behaviour of @xmath134 as a function of @xmath12 . on this account we choose some appropriate value of @xmath227 and boundary condition for @xmath186 at @xmath227 , but carry out the integration only from @xmath228 to @xmath229 , to ensure , by causality , that no signal coming in from @xmath230 has time to reach @xmath177 and affect that behaviour . in particular , in all the examples below we set @xmath231 , while we chose @xmath232 for the first and second example , and @xmath233 for the other two examples . the first explicit example of the results of this numerical integration for @xmath124 is given in fig.4 and fig . the initial data used was : @xmath234 and we set @xmath156 , @xmath157 , corresponding to @xmath158 , and @xmath232 . in fig . 4 we display the evolution of @xmath124 as a function of @xmath12 . the graph corresponds clearly to an evolution dominated by qnr , although a more detailed analysis indicates the presence of a small non oscillating background at large @xmath12 . in fig 5 we have a plot of @xmath235 ( thick line ) in the region dominated by the damped oscillation mode . the thin line curve corresponds to an approximate fit with the function @xmath236 , although the actual plot in fig . 5 has been displaced up by 0.5 for clarity . we can see again the very clear signal of a damped oscillation , with a frequency and fall off in very good agreement with those of the example of fig . 2 , although we have used a different procedure and form for the incoming pulse . the agreement might probably improve using a more elaborate rather than our simple minded implementation of the numerical procedure . in any case , it was not the purpose of the authors to optimize it but rather to show that even with a simple implementation we can solve the initial value problem related to the perturbation treatment of the dynamics of our system close to a static configuration , and obtain results in perfect agreement of those extracted from the mode expansion . as a second full numerical example we considered again the initial data given ( [ eq09h ] ) , but setting @xmath156 and @xmath166 , that is @xmath167 . the results are shown in fig . 6 . in this case , in agreement with our analysis of the qnm , we find a very strongly damped oscillation , very much like that shown in fig 3 . for the same value of the parameters @xmath43 and @xmath28 , although the computational procedures applied in each case were completely different . we can also check that both the frequency and damping are in good agreement with the values given in table i for @xmath167 and positive real part for @xmath79 . as a final example , we present the numerically computed evolution of @xmath124 as a function of @xmath12 for the cases @xmath156 and @xmath237 , that is @xmath238 , given as the thin line curve in fig . 7 , and for @xmath156 and @xmath239 , that is @xmath240 , also given in fig . 7 as the thick line curve . in these examples the qnr is over damped and no trace of `` ringing '' is apparent in the time dependence of @xmath124 . in particular , the last case , @xmath240 , corresponds to @xmath241 in the notation of @xcite , where in what they call the first class solutions one has a pure imaginary qnm frequency , and , therefore , we would have only a purely exponentially decreasing qnr . these plots confirm also the stability of the corresponding static configurations . in this section we consider the problem of the apparent discrepancy between the evolutions obtained using the formalism developed in @xcite and those described here . we shall illustrate the reasons for this by considering the qnm toy model given by price and husain in @xcite , since , as will be seen , it bear a certain formal resemblance to the shell model discussed here . the price and husain model consists of two torsion strings , the first attached at @xmath242 and extending to @xmath243 , where it is attached to a second string that extends to @xmath244 . the strings have different characteristic constants , so that the respective equations of motion are , @xmath245 and , @xmath246 where @xmath247 is a constant . @xmath248 , and @xmath249 , represent the torsion angle respectively in the intervals @xmath250 , and @xmath251 . the boundary and matching conditions are , @xmath252 the general solutions of ( [ eeq01a ] ) and ( [ eeq01b ] ) , are , @xmath253 where @xmath254 is , in principle , an arbitrary function of @xmath255 , and we have included the boundary condition at @xmath242 , and , @xmath256 where @xmath257 , and @xmath258 are also functions of @xmath255 . one can now check that , imposing the matching conditions , and assuming that all the functions vanish for large values of their arguments , the general solution functions @xmath254 , @xmath257 , and @xmath258 satisfy the relations , @xmath259 \nonumber \\ g(t - k x ) & = & \frac{1}{2 k } \left [ ( k-1 ) f(t - kx+ka+a)-(k+1)f(t - kx+ka - a ) \right]\end{aligned}\ ] ] and , therefore , an arbitrary @xmath254 determines a complete solution of the problem . in particular , if we choose @xmath254 of compact support , we will also get @xmath87 and @xmath91 of compact support . this type of solution describes then an incoming pulse of compact support that approaches the junction at @xmath243 , excites a non vanishing @xmath260 for a finite time , and gives rise to an out going pulse , also of compact support . there is , in general , no `` ringing '' , as no quasi normal mode is excited . this is precisely the situation considered in @xcite , where we could fix the function @xmath124 arbitrarily , and , by choosing it of compact support we got also @xmath60 and @xmath58 of compact support , with no quasi normal ringing . but the price and husain model does display qnr . to see how this comes about we assume that the functions admit fourier transforms of the form , @xmath261 replacing in the system ( [ eeq08 ] ) we eventually find that we may write , @xmath262 } { { { \rm e}^{2\,i\omega a } } \left ( 1+k \right ) -k+1 } } d \omega\end{aligned}\ ] ] these expressions provide the solution for the problem of finding the evolution assuming that we are given an incoming pulse @xmath263 as the data for the problem , just as in the present shell problem . if @xmath264 has suitable analytic properties it may be possible to compute the integrals in ( [ eeq11 ] ) by extending the path of integration to the complex @xmath265 plane . we notice here that the denominators in ( [ eeq11 ] ) have simple poles at , @xmath266 where @xmath267 is an integer , including zero . notice that all @xmath268 have the same positive imaginary part , and , therefore , when it is possible to close the path of integration by adding an infinite semicircle in the upper half plane of @xmath265 , each one of these poles will contribute a factor of the form @xmath269 times an oscillating factor depending on @xmath267 and on the explicit form of @xmath264 . these are the characteristic features of the `` ring down '' associated with the quasinormal modes . in fact , if we go back to ( [ eeq08 ] ) , and write it in the form , @xmath270 \nonumber \\ g(t ) & = & \frac{1}{2 k } \left [ ( k-1 ) f(t+ka+a)-(k+1)f(t+ka - a ) \right]\end{aligned}\ ] ] and assume @xmath271 we get , @xmath272 \exp(i \omega ( t - k a ) ) \nonumber \\ g(t ) & = & \frac{1}{2 k } \left [ ( k-1 ) \exp(i \omega a)-(k+1 ) \exp(-i \omega a ) \right ] \exp(i \omega ( t+k a))\end{aligned}\ ] ] the condition @xmath273 corresponds to a purely out going wave for @xmath274 . this is achieved precisely for @xmath265 of the form ( [ eeq12 ] ) . taken literally , these type of solution correspond to both @xmath275 and @xmath276 that decrease exponentially for @xmath277 , but increase exponentially as @xmath278 . similarly , we have @xmath279 that grows exponentially for large @xmath48 . these solutions are therefore unphysical . what is then the relation between these solutions and the poles in ( [ eeq11 ] ) ? the crucial point is that the poles are effective only if the integration path can be closed on the upper half plane of @xmath265 . in general , for @xmath280 , this requires @xmath281 , and therefore , the exponential terms are not present for @xmath255 , and as a consequence @xmath15 , less than a certain lower bound . but , why do we see a ring down in one case and not in the other ? to understand what is happening here we go back to ( [ eeq13 ] ) , and assume that @xmath254 has a fourier transform @xmath282 . then , replacing in ( [ eeq08 ] ) , we find , @xmath283\ ] ] but this implies that for general @xmath282 , the transform @xmath284 will contain a factor that precisely cancels the denominators in ( [ eeq11 ] ) , and therefore there will be no ring down in the solution of the problem , unless , of course , @xmath285 itself contains the appropriate poles , as in ( [ eeq11 ] ) . the general conclusion is then that both ( [ eeq08 ] ) and ( [ eeq11 ] ) provide a complete solution of the problem and are , therefore , completely equivalent . clearly , for the purpose of making the qnr apparent it is simpler and , in a sense , more `` natural '' to use ( [ eeq11 ] ) and specify freely @xmath257 , i.e. , the incoming pulse shape , just as was done for the shell problem in the present work . in this paper we have considered again the perturbative analysis of the static configurations of the apostolatos and thorne shell model by modifying the formalism developed in @xcite . as a result we have been able to show the completeness of the mode expansion as regards the characteristic data problem for the perturbative dynamics of the shell . this , in turn , provides a simple and direct proof of the stability under bounded ( symmetry preserving ) perturbations of the general static configurations of the shell . we have also derived a set of coupled linear ordinary and partial differential equations that describe the general perturbative evolution of the shell . this set of equation can be used to set up an initial value problem for the shell that can be solved numerically , but , more importantly , one can prove the existence of a positive definite constant of the motion , that implies the stability of the motion resulting for an arbitrary perturbation . in several examples considered we find perfect agreement between the full numerical evolution and that obtained through the numerical integration of the mode expansion , although they are completely different in detail . at first sight our results seem to be in contradiction with those of kurita and nakao @xcite . we believe that there is no contradiction here , and that our computations and those in @xcite are perfectly compatible . the problem appears because kurita and nakao conclude that the existence of certain complex zeros ( or poles ) in the complex @xmath265 plane automatically imply a direct effect on the evolution of the shell , but this is not necessarily the case . the usual arguments for relating these poles to the evolution relay rather heavily on the possibility of extending integrals on the real @xmath265 axis to the complex @xmath265 plane , in such a way that one effectively picks up the residues of those poles , but this may not always be possible , and , even , it might happen that one can disregard those poles , when they exist , by extending the integration path in the opposite direction . in fact , since the mode expansion involves bessel functions in rather complex combinations , the matter of finding the extensions of appropriate function to the complex @xmath265 plane is a highly nontrivial undertaking . going back to our own derivations , we notice that although we found qnm with a negative real part of @xmath265 , we only find evidence of qnr for those with a positive real part for @xmath265 . we remark once again that the mode expansion does not require explicit consideration of the complex @xmath265 plane , as only the real @xmath265 axis is involved . we believe that although the work @xcite is valuable as regards the finding of the qnm of the system , since these by themselves are unphysical ( they diverge generally in some space or time direction ) , given the very complex nature of the functions and of their possible extensions to the complex plane , conclusions about the stability of the static configurations of the shell can only be reached by explicitly showing the relation between these modes and the general evolution of the shell . in fact , in view of our results concerning completeness of the mode expansion and its relation to stability , we can only conclude that all in principle unstable qnm must be suppressed , as well as possibly some of the stable ones . of course , explicit confirmation of this conclusion would require performing and analyzing appropriate extensions into the complex plane , but that is completely outside the scope of the present research . this work was supported in part by conicet ( argentina ) . 99 t. a. apostolatos , and k. s. thorne , phys . d * 46 * , 2435 ( 1992 ) see . e.g. nollert , class.quant.grav . * 16 * ( 1999 ) r159-r216 , for a recent introduction , review and references on this subject . v. h. hamity , m. a. ccere , and d. e. barraco , gen . . gravit . * 41 * , 2657 ( 2009 ) r. j. gleiser and m. a. ramirez , phys.rev . * d85 * ( 2012 ) 044026 we must warn the reader that the use of @xmath286 in the equations that follow applies only to expressions where @xmath79 is real , and is just a shorthand for @xmath79 , if @xmath287 , and for @xmath288 , if @xmath289 . this appears to be trivial for real @xmath79 , but is crucial for _ complex _ @xmath79 , because , as it should be clear , expressions such as ( [ eq1d ] ) , ( [ eq4da ] ) , or ( [ eq5d ] ) , with @xmath286 in the argument of the bessel functions , are _ not _ solutions of ( [ ateq2 ] ) for complex @xmath79 . we must therefore replace appropriately @xmath286 in any expression related to the solution of the evolution equations before we attempt to compute it for complex @xmath79 . this should also be taken into account in our discussion of quasi normal modes and in any other instance where we consider complex values of @xmath79 . y. kurita and k. i. nakao , _ dynamical instability in a relativistic cylindrical shell composed of counter rotating particles _ [ arxiv:1112.4252 [ gr - qc ] ] prog.theor.phys . * 128 * ( 2012 ) 191 - 211 r. h. price and v. husain , phys . letters , * 68 * , 1973 ( 1992 )

emerging sonography techniques often require increasing the number of transducer elements involved in the imaging process . consequently , larger amounts of data must be acquired and processed . the significant growth in the amounts of data affects both machinery size and power consumption . within the classical sampling framework , state of the art systems reduce processing rates by exploiting the bandpass bandwidth of the detected signals . it has been recently shown , that a much more significant sample - rate reduction may be obtained , by treating ultrasound signals within the finite rate of innovation framework . these ideas follow the spirit of xampling , which combines classic methods from sampling theory with recent developments in compressed sensing . applying such low - rate sampling schemes to individual transducer elements , which detect energy reflected from biological tissues , is limited by the noisy nature of the signals . this often results in erroneous parameter extraction , bringing forward the need to enhance the snr of the low - rate samples . in our work , we achieve snr enhancement , by beamforming the sub - nyquist samples obtained from multiple elements . we refer to this process as compressed beamforming " . applying it to cardiac ultrasound data , we successfully image macroscopic perturbations , while achieving a nearly eight - fold reduction in sample - rate , compared to standard techniques . array processing , beamforming , compressed sensing ( cs ) , finite rate of innovation ( fri ) , ultrasound , xampling

1202.6037

diagnostic sonography allows visualization of body tissues , by radiating them with acoustic energy pulses , which are transmitted from an array of transducer elements . the image typically comprises multiple scanlines , each constructed by integrating data collected by the transducers , following the transmission of an energy pulse along a narrow beam . as the pulse propagates , echoes are scattered by density and propagation - velocity perturbations in the tissue @xcite , and detected by the transducer elements . averaging the detected signals , after their alignment with appropriate time - varying delays , allows localization of the scattering structures , while improving the signal to noise ratio ( snr ) @xcite . the latter process is referred to as beamforming . performed digitally , beamforming requires that the analog signals , detected by the transducers , first be sampled . confined to classic nyquist - shannon sampling theorem @xcite , the sampling rate must be at least twice the bandwidth , in order to avoid aliasing . as imaging techniques develop , the amount of elements involved in each imaging cycle typically increases . consequently , the rates of data which need to be transmitted from the system front - end , and then processed by the beamformer , grow significantly . the growth in transmission and processing rates inevitably effects both machinery size and power consumption . consequently , in recent years there has been growing interest in reducing the amounts of data as close as possible to the system front - end . in fact , such reduction is already possible within the classical sampling framework : state of the art devices digitally downsample the data at the front - end , by exploiting the fact that the signal is modulated onto a carrier , so that the spectrum essentially occupies only a portion of its entire base - band bandwidth . the preliminary sample rate remains unchanged , since the demodulation is performed in the digital domain . nevertheless , a key to significant data compression lies beyond the classical sampling framework . indeed , the emerging compressive sensing ( cs ) framework @xcite states , that sparse signals may be accurately reconstructed from a surprisingly small amount of coefficients . complementary ideas rise from the finite rate of innovation ( fri ) framework @xcite , in which the signal is assumed to have a finite number of degrees of freedom per unit time . many classes of fri signals can be recovered from samples taken at the rate of innovation @xcite . for a detailed review of previously proposed fri methods , the reader is referred to @xcite . combining the latter notions with classical sampling methods , the developing xampling framework @xcite involves methods for fully capturing the information carried by an analog signal , by sampling it far below the nyquist - rate . following the spirit of xampling , al . proposed in @xcite , that ultrasound signals be described within the fri framework . explicitly , they assume that these signals , formed by scattering of a transmitted pulse from multiple reflectors , may be modeled by a relatively small number of pulses , all replicas of some known pulse shape . denoting the number of reflected pulses by @xmath0 , and the signal s finite temporal support by @xmath1 , the detected signal is completely defined by @xmath2 degrees of freedom , corresponding to the replicas unknown time delays and amplitudes . based on @xcite , the authors formulate the relationship between the signal s fourier series coefficients , calculated with respect to @xmath1 , and its unknown parameters , in the form of a spectral analysis problem . the latter may be solved using existing techniques , given a subset of fourier series coefficients , with a minimal cardinality of @xmath2 . the sampling scheme is thus reduced to the problem of extracting a small subset of the detected signal s frequency samples . two robust schemes are derived in @xcite , extracting such a set of coefficients from samples of the signal , taken at sub - nyquist rates . the system presented in @xcite employs a single processing channel , in which the analog signal is filtered by an appropriate sampling kernel and then sampled with a standard low - rate analog to digital converter ( adc ) . the method of @xcite employs multiple processing channels , each comprising a modulator and an integrator . these approaches were shown to be more robust than previous fri techniques and also allow for arbitrary pulse shapes . the initial motivation for our work stems from the need to translate the ultrasound xampling scheme proposed in @xcite , into one which achieves the final goal of reconstructing a two - dimensional ultrasound image , by integrating data sampled at multiple transducer elements . in conventional ultrasound imaging , such integration is achieved by the beamforming process . the question is how may we implement beamforming , using samples of the detected signals taken at sub - nyquist rates . a straightforward approach is to replace the nyquist - rate sampling mechanism , utilized in each receiver element , by an fri xampling scheme . having estimated the parametric representation of the signal detected in each individual element , we could reconstruct it digitally . the reconstructed signals can then be further processed via beamforming . however , the nature of ultrasound signals reflected from real tissues , makes such an approach impractical . this is mainly due to the detected signals poor snr , which results in erroneous parameter extraction by the xampling scheme , applied to each element independently . our approach is to generalize the fri xampling scheme proposed in @xcite , such that it integrates beamforming into the low - rate sampling process . the result is equivalent to that obtained by xampling the beamformed signal , which exhibits significantly better snr . furthermore , beamforming practically implies that the array of receivers is dynamically focused along a single scanline . consequently , the resulting signal depicts reflections originating in the intersection of the radiated medium with a vary narrow beam . such a signal better suits the fri model proposed in @xcite , which assumes the reflections to be caused by isolated , point - like scatterers . we refer to our scheme by the term compressed beamforming , as it transforms the beamforming operator into the compressed domain @xcite . applied to real cardiac ultrasound data obtained from a ge breadboard ultrasonic scanner , our approach successfully images macroscopic perturbations in the tissue while achieving a nearly eight - fold reduction in sampling rate , compared to standard imaging techniques . the paper is organized as follows : in section [ sec:02 ] , we summarize the general principles of beamforming in ultrasound imaging . in section [ sec:03 ] we outline the fri model and its contribution to sample rate reduction in the ultrasound context . we motivate compressed beamforming in section [ sec:04 ] , considering the nature of ultrasound signals reflected from biological tissues . beamforming and fri xampling are combined in section [ sec:05 ] , where we propose that the signal obtained by beamforming may be treated within the fri framework . following this observation , we derive our first compressed beamforming scheme , which operates on low - rate samples taken at the individual receivers . this approach is then further simplified in section [ sec:06 ] . in section [ sec:07 ] we focus on image reconstruction from the parametric representation obtained by either xampling scheme . in this context , we generalize the signal model proposed in @xcite , allowing additional unknown phase shifts of the detected pulses . we then discuss an alternative recovery approach , based on cs . simulations comparing the performance of several recovery methods are provided in section [ sec:08 ] . finally , experimental results obtained for cardiac ultrasound data are presented in section [ sec:09 ] . receivers are aligned along the @xmath3 axis . the origin is set at the position of the reference receiver , denoted @xmath4 . @xmath5 denotes the distance measured from the reference receiver to the @xmath6th receiver . the imaging cycle begins when an acoustic pulse is transmitted at direction @xmath7 . echoes are then reflected from perturbations in the radiated medium.,width=207 ] in this section , we describe a typical b - mode imaging cycle , focusing on the beamforming process , carried out during the reception phase . the latter constitutes a significant block in ultrasound imaging , and plays a major role in our proposed fri xampling scheme . consider the array depicted in fig . [ fig:01 ] , comprising @xmath8 transducer elements , aligned along the @xmath3 axis . denote by @xmath5 the distance from the @xmath6th element to the reference receiver @xmath4 , used as the origin , namely @xmath9 . the imaging cycle begins when , at time @xmath10 , the array transmits acoustic energy into the tissue . subsequently , the elements detect echoes , which originate in density and propagation - velocity perturbations , characterizing the radiated medium . denote by @xmath11 the signal detected by the @xmath6th receiver . the acoustic reciprocity theorem @xcite suggests , that we may use the signals detected by multiple transducer elements , in order to probe arbitrary coordinates for reflected energy . namely , by combining the detected signals with appropriate time delays , echoes scattered from a chosen coordinate will undergo constructive interference , whereas those originating off this coordinate will be attenuated , due to destructive interference . in practice , the array can not effectively radiate the entire medium simultaneously . instead , a pulse of energy is conducted along a relatively narrow beam , whose central axis forms an angle @xmath7 with the @xmath12 axis . focusing the energy pulse along such a beam is achieved by applying appropriate time delays to modulated acoustic pulses , transmitted from multiple array elements . rather than arbitrarily probing the radiated tissue , we are now forced to adjust the probed coordinate in time , in coordination with the propagation of the transmitted energy . this practically implies that , combining the detected signals with appropriate time - varying delays , we may obtain a signal , which depicts the intensity of the energy reflected from each point along the central transmission axis . throughout the rest of this section , we derive an explicit expression for creating this beamformed signal . assume that the energy pulse , transmitted at @xmath10 , propagates at velocity @xmath13 in the direction @xmath7 . at time @xmath14 , the pulse crosses the coordinate @xmath15 . consider a potential reflection , originating in this coordinate , and arriving at the @xmath6th element . the distance traveled by such a reflection is : @xmath16 the time in which the reflection crosses this distance is @xmath17 , so that it reaches the receiver element at time @xmath18 it is readily seen that @xmath19 . hence , in order to align the reflection detected in the @xmath6th receiver with the one detected in the reference receiver , we need to apply a delay to @xmath11 , such that the resulting signal , @xmath20 , satisfies @xmath21 . denoting @xmath22 , and using , we obtain the following distorted signal for @xmath14 : @xmath23 with @xmath24 . the aligned signals may now be averaged , resulting in the beamformed signal @xmath25 which exhibits enhanced snr compared to @xmath26 . furthermore , by its construction , @xmath27 represents , for every @xmath14 , the intensity which was measured when focusing the array to @xmath28 . therefore , it may eventually be translated into an intensity pattern , plotted along the corresponding ray . although defined over continuous time , ultrasound systems perform the process formulated in - in the digital domain , requiring that the analog signals @xmath11 first be sampled . confined to the classic nyquist - shannon sampling theorem , these systems sample the signals at twice their baseband bandwidth , in order to avoid aliasing . the detected signals typically occupy only a portion of their baseband bandwidth . exploiting this fact , some state of the art systems manage to reduce the amount of samples transmitted from the front - end , by down - sampling the data , after demodulation and low - pass filtering . however , since such operations are carried out digitally , the preliminary sampling - rate remains unchanged . to conclude this section , we evaluate the nominal number of samples needed to be taken from each active receiver element in order to obtain a single scanline using standard imaging techniques . consider an ultrasound system which images to a nominal depth of @xmath29 . the velocity at which the pulse propagates , @xmath13 , varies between @xmath30 ( fat ) to @xmath31 ( spleen ) @xcite . an average value of @xmath32 is assumed by scanners for processing purposes , such that the duration of the detected signal is @xmath33 . the signal s baseband bandwidth requires a nominal sampling rate of @xmath34 , resulting in an overall number of @xmath35 real - valued samples . assuming that the signal s passband bandwidth is only @xmath36 , the data sampled at nyquist - rate may be finally down - sampled to 1680 real - valued samples . these samples , taken from all active receivers , are now processed , according to - , in order to construct the beamformed signal . since standard imaging devices carry out beamforming by applying delay and sum operations to the sampled data , the amount of operations required for generating a single scanline is directly related to the sample rate . regardless of our computational power , physical constraints imply that the time required for constructing a single scanline is at least @xmath37 . this takes into account the round - trip time required for the transmitted pulse to penetrate the entire imaging depth , and for the resulting echoes to cross a similar distance back to the array . nevertheless , sufficient computational power may allow construction of several scanlines , within that same time interval , increasing the overall imaging rate . by using compressed beamforming , we aim at capturing significant information in the imaging plane , while reducing the sampling rate and consequently the processing rate . this , in turn , may improve the existing trade - off between imaging rates and both machinery size and power consumption . in a pioneer attempt to implement xampling methodology in the context of ultrasound imaging , @xcite suggests that the signal detected in each receiver element may be sampled at a rate far below nyquist , by modeling it as an fri signal . the authors propose that @xmath11 , detected in the @xmath6th element , be regarded as sum of a relatively small number of pulses , all replicas of some known pulse shape . explicitly : @xmath38 here @xmath0 is the number of scattering elements , distributed throughout the sector radiated by the transmitted pulse , @xmath39 denotes the time in which the reflection from the @xmath40th element arrived at the @xmath6th receiver , and @xmath41 denotes the reflection s amplitude , as detected by the @xmath6th receiver . finally , @xmath42 denotes the known pulse shape , regarded , in our work , by the term two - way pulse . the signal in is completely defined by @xmath2 real - valued parameters , @xmath43 . sampling fri signals was first treated by vetterli et . al . @xcite . their approach involves projecting the fri signal , characterized by @xmath2 degrees of freedom per unit time , onto a @xmath2-dimensional subspace , corresponding to a subset of its fourier series coefficients . having extracted @xmath2 frequency samples of the signal , spectral analysis techniques ( e.g. annihilating filter @xcite , matrix pencil @xcite ) may be applied , in order to extract the unknown signal parameters . applying this solution to the problem formulated in , @xcite formalizes the relationship between the ultrasound signal s fourier series coefficients to its unknown parameters , as a spectral analysis problem . let @xmath37 be the duration of @xmath11 . we can then expand @xmath11 in a fourier series , with coefficients @xmath44&=\frac{1}{t}\int_{0}^{t}{\varphi_m\left(t\right)e^{-i\frac{2\pi}{t}kt}dt}\\ & = \frac{1}{t}\int_{0}^{t}{\sum_{l=1}^l{a_{l , m}h\left(t - t_{l , m}\right)}e^{-i\frac{2\pi}{t}kt}dt}\\ & = \frac{1}{t}h\left(\frac{2\pi}{t}k\right)\sum_{l=1}^l{a_{l , m}e^{-i\frac{2\pi}{t}kt_{l , m } } } , \end{split}\end{aligned}\ ] ] where @xmath45 denotes the continuous time fourier transform ( ctft ) of @xmath42 . consider the sequence @xmath46 , comprising @xmath47 integers , and define the length-@xmath47 vector @xmath48 with @xmath49th element @xmath50 $ ] . then may be written in matrix form : @xmath51 where @xmath52 is a diagonal matrix with diagonal elements @xmath53 , @xmath54 contains @xmath55 as its @xmath56th element , and @xmath57 is the length @xmath0 vector , with elements @xmath41 . choosing @xmath58 such that @xmath59,we can express as : @xmath60 where @xmath61 . if the values @xmath58 are a sequence of consecutive indices , then @xmath54 takes on a vandermonde form , and has full column rank @xcite as long as @xmath62 and the time - delays are distinct , i.e. , @xmath63 , for all @xmath64 . the formulation derived in is a standard spectral analysis problem . as long as @xmath65 , it may be solved for the unknown parameters @xmath43 , using methods such as annihilating filter @xcite or matrix pencil @xcite . having obtained , the sampling scheme reduces to the problem of extracting @xmath47 frequency samples of @xmath11 , where @xmath66 . a single - channel xampling scheme , such as the one derived in @xcite , allows robust estimation of such coefficients from point - wise samples of the signal , after filtering it with an appropriate kernel . the estimation is performed by applying a linear transformation to @xmath67 complex - valued samples ( equivalently , @xmath68 real - valued samples ) of the filtered signal , requiring that @xmath69 . in this context , @xcite introduces the sum of sincs kernel , which satisfies the necessary constraints , and is additionally characterized by a finite temporal support . combining the requirements that @xmath66 and @xmath69 , the xampling scheme proposed in @xcite allows reconstruction of the signal detected in each receiver element from a minimal number of @xmath70 real - valued samples . considering the nominal figures derived in the previous section for standard beamforming , we conclude that , as long as @xmath71 , such a xampling method may indeed achieve a substantial rate reduction . applied to a single receiver element , the xampling scheme proposed in @xcite achieves good signal reconstruction for an actual ultrasound signal , reflected from a setup of phantom targets . in principle , we could apply this approach to each receiver element individually , resulting in a parametric representation for each of the signals @xmath72 . being able to digitally reconstruct the detected signals , we could then proceed with the standard beamforming process , outlined in section [ sec:02 ] , aimed at constructing the corresponding scanline . computational effort would have been reduced , by limiting the beamforming process to the support of the estimated pulses . in fact , we could possibly bypass the beamforming stage , by deriving a geometric model which maps the set of delays , @xmath73 , associated with the @xmath40th reflector , to its two - dimensional position @xmath74 . however , applying the proposed fri xampling scheme to signals reflected from biological tissues , we face two fundamental obstacles : low snr and proper interpretation of the estimated signal parameters , considering the profile of the transmitted beam . these two difficulties may be better understood by examining fig . [ fig:02 ] , which depicts traces acquired for cardiac images of a healthy consenting volunteer using a ge breadboard ultrasonic scanner . ( a ) ( b ) in the left plot ( a ) , are signals detected by 32 of 64 active array elements , following the transmission of a single pulse . the pulse was conducted along a narrow beam , forming an arbitrary angle @xmath7 with the @xmath12 axis . the right plot ( b ) depicts the signal obtained by applying beamforming to the detected signals , as outlined in section [ sec:02 ] . examining the individual traces , one notices the appearance of strong pulses , possibly overlapping , characterized by a typical shape , as proposed in . let us assume that we could indeed extract the delays and amplitudes of these pulses , by applying the proposed fri xampling scheme to each element . we suggested that beamforming could be bypassed , by deriving a geometric model for estimating the two - dimensional position of a scattering element , based on the delays of pulses associated with it , yet estimated in different receivers . in order to apply such a model , we must first be able to match corresponding pulses across the detected signals . however , referring to the practical case depicted in ( a ) , we notice that such a task is not at all trivial - the individual signals depict reflections , originating from the entire sector , radiated by the transmitted pulse . these reflections may , therefore , vary significantly across traces . in fact , some pulses , visible in several traces , are not at all apparent in other traces . in contrast , the beamformed signal , by its construction , depicts intensity of reflections originating from along the central transmission axis , while attenuating reflections originating off this axis . attempting to apply fri xampling to each receiver element individually , we encounter an even more fundamental obstacle , at the earlier stage of extracting the signal s parametric representation from its low - rate samples . the individual traces contain high levels of noise . the noisy components , especially noticeable in traces 54 and 64 , rise mainly from constructive and destructive interference of acoustic waves , reflected by dense , sub - wavelength scatterers in the tissue . the latter are typically manifested as granular texture in the ultrasound image , called speckle , after a similar effect in laser optics @xcite . the noisy components inherently induce erroneous results , when attempting to sample and reconstruct the fri components using the xampling approach . in extreme scenarios , where the noise masks the fri component , the extracted parameters will be meaningless , such that any attempt to cope with errors in the parametric domain will turn out useless . the motivation to our approach rises from the observation , that we may resolve the aforementioned obstacles by xampling the beamformed signal , @xmath27 , rather than the individual signals @xmath11 . whereas beamforming is a fundamental process in ultrasound imaging since its early days , our innovation regards its integration into the xampling process . we derive our compressed beamforming approach , beginning with conceptual xampling of the beamformed signal , using the scheme proposed in @xcite . we then show that an equivalent result may be obtained from low - rate samples of the individual signals @xmath11 . a necessary condition for implementing our approach is that @xmath27 , generated from @xmath72 satisfying , is also fri of similar form . examining fig . [ fig:02 ] we notice that @xmath27 exhibits a structure similar to that of the individual signals , comprising strong pulses of typical shape , which may overlap . in this case , there are several obvious advantages in xampling @xmath27 . first , since @xmath72 are averaged in @xmath27 ( after appropriate distortion , derived from the acoustic reciprocity theorem ) it naturally exhibits enhanced snr with respect to the individual signals . the attenuation of noise in the beamformed signal , compared to the individual signals , is apparent in fig . [ fig:02 ] , especially in the interval @xmath75 . second , @xmath27 is directly related to an individual scanline . this means that we are no longer bothered with the ambiguous problem of matching pulses across signals detected in different elements . finally , recall that the signal model derived in assumes isolated point - reflectors . such a model is better justified with respect to @xmath27 since , by narrowing the effective width of the imaging beam , we may indeed approximate its intersection with reflecting structures to be point - like . this effect is noticeable in fig . [ fig:02 ] where some pulses , visible in individual traces , appear attenuated in the beamformed signal . such pulses correspond to reflectors located off the central axis of the transmission beam . in the next section , we focus on justifying the assumption that @xmath27 may be treated within the fri framework . an additional challenge , implied in section [ sec:02 ] , regards the fact that @xmath27 does not exist in the analog domain - standard ultrasound devices generate it digitally , from samples of the signals detected in multiple receiver elements , taken at the nyquist - rate . our goal is , therefore , to derive a scheme , which manages to estimate the necessary samples of @xmath27 , from low - rate samples of filtered versions of @xmath72 . our approach is based on the assumption that the fri scheme , outlined in section [ sec:03 ] , may be applied to the beamformed signal @xmath27 , constructed according to - . the latter exhibits much better snr than signals detected in individual receiver elements . additionally , it depicts reflections originating from a sector much narrower than the one radiated by the transmission beam . its translation into a single scanline is therefore straightforward . in section [ sec:51 ] we prove that if the signals @xmath11 obey the fri model , then @xmath27 is * approximately * of the form : @xmath76 where @xmath77 denotes the time in which the reflection from the @xmath40th element arrived at the reference receiver , indexed @xmath4 . @xmath27 may thus be sampled using the xampling schemes derived in @xcite . in practice , we can not sample @xmath27 directly , since it does not exist in the analog domain . in second [ sec:52 ] we show how the desired low - rate samples of @xmath27 can be determined from samples of @xmath11 . throughout this section we apply three reasonable assumptions . first , we assume that @xmath78 . practically , such a constraint may be forced by appropriate apodization , as often performed in ultrasound imaging . namely , @xmath11 is combined in @xmath27 only for @xmath79 . as an example , for the breadboard ultrasonic scanner used in our experiments , the array comprised @xmath80 receiver elements , distanced @xmath81 apart . the proposed apodization implies that the receivers located farthest from the origin are combined in the beamformed signal for imaging depth greater than @xmath82 . second , we assume the two - way pulse , @xmath83 , to be compactly supported on the interval @xmath84 . finally , we assume that @xmath85 . the last assumption may also be forced by appropriate apodization . as an example , the nominal duration of the pulse acquired by the breadboard ultrasonic scanner used in our experiments was @xmath86 . in this case , echoes scattered from depth greater than @xmath87 already satisfy @xmath88 . suppose that @xmath11 can be written as in . applying the beamforming distortion , we get @xmath89 the resulting signal comprises @xmath0 pulses , which are distorted versions of the two - way pulse @xmath42 . suppose that some of the pulses originated in reflectors located off the central beam axis . beamforming implies that , once averaging the distorted signals according to , such pulses will be attenuated due to destructive interference . being interested in the structure of the beamformed signal @xmath27 , we are therefore concerned only with pulses which originated in reflectors located along the central beam . for convenience , we assume that all pulses in satisfy this property ( pulses which do not satisfy it , will vanish in @xmath27 ) . we may thus use @xmath90 , defined in , in order to express @xmath39 in terms of @xmath77 . substituting @xmath91 into @xmath90 , we get @xmath92 , so that becomes @xmath93 where we defined @xmath94 . applying our second assumption , the support of @xmath95 is defined by the requirement that @xmath96 using and , it is readily seen that @xmath95 is supported on @xmath97 , where @xmath98 further applying our assumption that @xmath78 , we obtain @xmath99 . we have thus proven that @xmath100 for @xmath101 . next , let us write any @xmath102 in @xmath103 as @xmath104 , where @xmath105 . then @xmath106 we now rely on our assumption that @xmath85 . since @xmath107 , we also have @xmath108 . the argument of @xmath109 in may therefore be approximated , to first order , as @xmath110 where @xmath111 up until this point , we assumed that @xmath78 . further assuming that @xmath112 , @xmath113 . replacing @xmath114 by @xmath115 , may therefore be written as @xmath116 combining with the fact that @xmath117 is zero outside @xmath118 , may be approximated as @xmath119 averaging the signals @xmath120 according to , we get : @xmath121 which is indeed the fri form . additionally , assuming that the support of @xmath11 is contained in @xmath1 , we show in the appendix that there exists @xmath122 , such that the support of @xmath27 is contained in @xmath123 and , additionally , @xmath124 . as @xmath125 grows towards @xmath77 , @xmath126 decreases , resulting in a larger distortion of the @xmath40th pulse . consequently , the approximation of @xmath127 as a sum of shifted replicas of the two - way pulse becomes less accurate . the xampling schemes used by @xcite rely on the projection of the detected signal onto a subspace of its fourier series coefficients . we therefore examine the dependency of the projection error on the distortion parameters , @xmath125 , @xmath77 and @xmath7 . in fig . [ fig:03 ] , we show projection errors calculated numerically , for a signal comprising a single pulse of duration @xmath128 . the pulse was simulated by modulating a gaussian envelope with carrier frequency @xmath129 . it was then shifted by multiple time delays , @xmath77 , where @xmath130 , and @xmath131 , corresponding to an imaging depth of @xmath132 . for each delay , we generated the signals @xmath11 , assuming that the reflector is positioned along the @xmath12 axis ( @xmath133 ) , and that the receiver elements are distributed @xmath81 apart , along the @xmath3 axis . we chose @xmath134 , such that the center ( reference ) receiver was indexed @xmath135 . the beamforming distortion was then applied to the simulated signals , based on . finally , the distorted signals were projected onto a subset of @xmath136 consecutive fourier series coefficients , taken within the essential spectrum of the two - way pulse . the coefficients extracted from the @xmath6th distorted signal were arranged into the length @xmath137 vector , @xmath138 . as implied by , no distortion is applied to the signal detected at the reference receiver . we therefore evaluate the projection error by calculating the snr defined as @xmath139 . the traces obtained for several values of @xmath140 are depicted in the figure . as @xmath77 grows , @xmath126 approaches 1 , and the approximation becomes more valid . as a result , the projection error decreases . for receivers located near the origin , such that @xmath141 , the error decreases very quickly . for instance , examining @xmath142 , the snr grows above @xmath143 for a reflection originating at distance greater than @xmath144 of the imaging depth . the snr improves more moderately for receivers located farther away from the origin . nevertheless , considering the receiver located farthest away from the origin , @xmath145 , the snr grows above @xmath146 for a reflection originating at distance greater than @xmath147 of the imaging depth . vs. pulse delay , @xmath77 , for several receiver elements . the elements are distributed @xmath81 apart , such that @xmath148 ( element farthest from array center ) and @xmath142 . zero error is obtained for the center element , @xmath149 , since no distortion is required in this case.,width=283 ] concluding this section , our empirical results indeed justify the approximation proposed in , where appropriate apodization may further improve this approximation . assuming to be valid , we may reconstruct the beamformed signal using the xampling schemes proposed in @xcite . an obvious problem is that @xmath27 does not exist in the analog domain , and therefore may not be xampled directly . we now propose a modified xampling scheme , which allows extraction of its necessary low - rate samples , by sampling filtered versions of @xmath11 at sub - nyquist rates . since the support of @xmath27 is contained in @xmath123 , where @xmath122 , we may define @xmath27 s fourier series with respect to the interval @xmath1 . denoting by @xmath150 the @xmath151th fourier series coefficient of @xmath152 , we have @xmath153 where @xmath154 is the indicator function , taking the value 1 for @xmath155 and 0 otherwise . plugging the indicator function in may seem unnecessary . however , once transforming into an operator applied directly to @xmath72 , it serves an important role in zeroing intervals , which are assumed zero according to , but , in any practical implementation , contain noise . substituting into , we can write @xmath156 where , from , @xmath157 and @xmath158 the process defined in - can be translated into a multi - channel xampling scheme , such as the one depicted in fig . [ fig:04 ] . each signal @xmath11 is multiplied by a bank of kernels @xmath159 defined by , and integrated over @xmath160 . this results in a vector @xmath161^t$ ] . the vectors @xmath162 are then averaged in @xmath163^t$ ] , which has the desired improved snr property , and provides a basis for extracting the @xmath2 parameters which define @xmath27 . since @xmath27 satisfies , we apply a similar derivation to that outlined in section [ sec:04 ] , yielding @xmath164 where @xmath165 is a diagonal matrix with @xmath49th diagonal element @xmath166 , @xmath167 contains @xmath168 as its @xmath56th element , and @xmath169 is the length @xmath0 vector , with elements @xmath170 . the matrix @xmath167 may be estimated by applying spectral analysis techniques , allowing for the vector of coefficients @xmath169 to be solved by a least squares approach @xcite . [ fig:11 ] illustrates the shape of the resulting kernels @xmath171 , setting @xmath133 and choosing two arbitrary values of @xmath151 . for each choice of @xmath151 we plot the kernels corresponding to 7 receiver elements , selected from an array comprising 64 elements , distanced @xmath172 apart . for @xmath131 and @xmath151 satisfying : ( a ) @xmath173 , ( b ) @xmath151=5 . we assume an array comprising @xmath174 elements , distanced @xmath172 apart , and plot 7 traces which were obtained for the elements indexed @xmath175.,width=264 ] ( a ) for @xmath131 and @xmath151 satisfying : ( a ) @xmath173 , ( b ) @xmath151=5 . we assume an array comprising @xmath174 elements , distanced @xmath172 apart , and plot 7 traces which were obtained for the elements indexed @xmath175.,width=264 ] ( b ) in the previous section , we developed a xampling approach to extract the fourier series coefficients of @xmath27 . however , the complexity of the resulting analog kernels , together with their dependency on @xmath7 , makes hardware implementation of the scheme depicted in fig . [ fig:04 ] complex . here , we take an additional step , which allows the approximation of @xmath176 , and consequently @xmath177 , from low - rate samples of @xmath11 , obtained in a much more straightforward manner . we begin by substituting @xmath11 of by its fourier series , calculated with respect to @xmath1 . denoting the @xmath178th fourier coefficient by @xmath179 $ ] , we get : @xmath180\frac{1}{t}\int_0^t{q_{j , m}(t;\theta)e^{-i\frac{2\pi}{t}\left(k_j - n\right ) t}dt}}\\ & = \sum_{n}{\phi_m\left[k_j - n\right]q_{j , m;\theta}\left[n\right ] } , \end{split}\end{aligned}\ ] ] where @xmath181 $ ] are the fourier series coefficients of @xmath182 , also defined on @xmath160 . let us replace the infinite summation of by its finite approximation : @xmath183q_{j , m;\theta}\left[n\right]}. \end{split}\end{aligned}\ ] ] the following proposition shows that this approximation can be made sufficiently tight . assume that @xmath184 . then , for any @xmath185 , and for any selection @xmath186 , there exist finite @xmath187 and @xmath188 such that @xmath189 . let @xmath190 be the space of square - summable sequences , with norm @xmath191 . let @xmath192\right\}_{n=-\infty}^{\infty}$ ] and @xmath193\right\}_{n=-\infty}^{\infty}$ ] . since @xmath11 is of finite energy , @xmath194 . we may calculate the @xmath190 norm of @xmath169 , based on the definition of @xmath195 in , resulting in @xmath196 . this implies that @xmath197 as well . let @xmath198 be the truncated sequence @xmath169 for @xmath199 and zero otherwise . we may then write the approximation error as : @xmath200 where @xmath201 is the inner product defined as @xmath202 . the last transition in is a result of cauchy - schwartz inequality . by definition of @xmath198 and @xmath169 , it is readily seen that @xmath203 . denoting @xmath204 , becomes @xmath205 since @xmath206 , @xmath207 can approach @xmath208 as close as we desire , by appropriate selection of @xmath209 and @xmath210 . for any @xmath185 , there exists @xmath211 , such that the right side of is smaller than @xmath212 . selecting @xmath209 and @xmath210 for which @xmath213 , results in @xmath214 , as required . furthermore , setting an upper bound on the energy of @xmath11 , and thereby on @xmath215 , @xmath209 and @xmath210 may be chosen off - line , subject to the decay properties of the sequence @xmath216\right\}_{n=-\infty}^{\infty}$ ] . using proposition 1 , we can compute @xmath217 as a good approximation to @xmath218 . we now show how @xmath219 can be obtained directly from the fourier series coefficients @xmath179 $ ] of each @xmath11 . we first evaluate @xmath209 and @xmath210 for a certain choice of @xmath6 and @xmath7 , such that @xmath218 may be approximated to the desired accuracy using . equivalently , we obtain the minimal subset of @xmath11 s fourier series coefficients , required for the approximation of @xmath218 . performing this for all @xmath220 , we obtain @xmath137 such subsets . denoting the union of these subsets by @xmath221 , we may now simultaneously compute @xmath222 from @xmath223\right\}_{n\in\kappa_m}$ ] by a linear transformation . define the length-@xmath47 vector @xmath138 , with @xmath40th element @xmath224 $ ] , and @xmath225 being the @xmath40th element in @xmath221 . using , we may write @xmath226 where @xmath227 is the length-@xmath137 vector with @xmath49th element @xmath219 , and @xmath228 is a @xmath229 matrix with elements @xmath230 & n_1\left(k_j\right ) \leq k_j - k_l \leq n_2\left(k_j\right ) \\ 0 & \mbox{otherwise } \end{array}\right . \end{split}.\end{aligned}\ ] ] notice , that we have omitted the dependency of @xmath209 and @xmath210 on @xmath212 , @xmath6 and @xmath7 , since , unlike @xmath151 , these remain constant throughout the construction of @xmath228 . the resulting xampling scheme is depicted in fig . [ fig:05 ] . based on @xcite , we propose a simple mechanism for obtaining the fourier coefficients in each individual element : a linear transformation , @xmath231 , is applied to point - wise samples of the signal , taken at a sub - nyquist rate , after filtering it with an appropriate kernel , @xmath232 , such as the sum of sincs . in this scheme , while we do need to extract larger number of samples at the output of each element , as @xmath233 , we avoid the use of complicated analog kernels as in section [ sec:52 ] . furthermore , as we show in section [ sec:09 ] , in an actual imaging scenario good approximation is obtained with just a small sampling overhead . so far we derived our approach for extracting the parameters @xmath234 which determine @xmath27 from sub - nyquist samples , taken at the individual receiver elements . in this section we focus on the reconstruction of @xmath27 from these parameters . once @xmath27 is constructed for multiple values of @xmath7 , a two - dimensional image may be formed , by applying standard post - processing techniques : first , @xmath27 s envelope is extracted using the hilbert transform @xcite ; logarithmic compression is then applied to each envelope , resulting in a corresponding scanline ; finally , all scanlines are interpolated onto a two - dimensional grid . having obtained the parametric representation of @xmath27 , the first two steps may be calculated only within the support of the recovered signal . in section [ sec:71 ] we describe the reconstruction of @xmath27 from its estimated parameters , while generalizing the model proposed in : we assume that the detected signals are additionally parametrized by unknown carrier phases of the reflected pulses , and show that the xampling approach allows estimation of these unknown phases . in section [ sec:72 ] we propose an alternative approach for reconstructing @xmath27 , using cs methodology . consider the signal defined in . modeling a signal of physical nature , it is obviously real - valued , implying that @xmath41 are real . consequently , by , @xmath235 must also be real - valued . however , when we apply spectral analysis techniques aimed at solving the system formulated in , there is generally no constraint that @xmath169 be real - valued . indeed , solving it for samples obtained using our proposed xampling schemes , the resulting coefficients are complex , with what appears to be random phases . in fact , a similar phenomenon is observed when solving for samples taken from the individual signals , @xmath11 , as proposed in @xcite . below we offer a physical interpretation of the random phases , by generalizing the model proposed in . the result is a closed - form solution for reconstructing the estimated signal , using the complex coefficients . when applied , a significant improvement is observed , comparing the envelope of the reconstructed signal , with that of the original signal . the ultrasonic pulse @xmath42 may be modeled by a baseband waveform , @xmath236 , modulated by a carrier at frequency @xmath237 : @xmath238 , where @xmath239 and @xmath240 is the phase of the carrier . the model proposed in , just like the one in , assumes the detected pulses to be exact replicas of @xmath42 . however , a more accurate assumption is that each reflected pulse undergoes a phase shift , based upon the relative complex impedances involved in its reflection @xcite . we thus propose to approximate the beamformed signal as : @xmath241 @xmath242 being an unknown phase . the @xmath49th fourier series coefficient of @xmath27 is now given by @xmath243 where @xmath244 is the ctft of @xmath236 and @xmath245 . let @xmath236 be approximated as a gaussian with variance @xmath246 and assume that @xmath247 . it is readily seen that @xmath248 we can then choose @xmath249 so that @xmath250 this allows to be approximated as @xmath251 and additionally @xmath252 combining and , we get @xmath253 where we define @xmath254 . denoting by @xmath255 the length @xmath137 vector , with @xmath150 as its @xmath49th element , the last result may be brought into the exact same matrix form written in . however , now we expect the solution to extract complex coefficients , of which phases correspond to the unknown phase shifts of the reflected pulses , @xmath256 . having obtained the complex coefficients , we may now reconstruct @xmath27 according to , and then proceed with standard post - processing techniques . the constraint imposed in is mild , considering nominal ultrasound parameters . assuming , for instance , @xmath131 , @xmath257 , and @xmath258 , we must choose @xmath259 . the requirement that @xmath165 of be invertible , already imposes a stronger constraint on @xmath151 , the @xmath49th fourier coefficient , since @xmath260 drops below @xmath261 for @xmath262 . throughout the previous sections , we addressed the problem of ultrasound signal reconstruction , within the fri framework . as shown in @xcite , for various fri problems , the relationship between the unknown signal parameters and its subset of fourier series coefficients takes the form of a spectral analysis problem . the latter is then typically solved by applying techniques such as annihilating filter @xcite or matrix pencil @xcite . in this section , we consider an alternative approach for reconstructing the signal defined in , based on cs methodology @xcite . assume that the time delays @xmath263 in are quantized with a @xmath264 quantization step , such that @xmath265 , @xmath266 . using , we may write the fourier series coefficients of @xmath27 as : @xmath267 let @xmath268 be the ratio @xmath269 . then may be expressed in the following matrix form : @xmath270 where @xmath165 is the @xmath271 diagonal matrix with @xmath260 as its @xmath49th diagonal element , and @xmath272 is a length @xmath268 vector , whose @xmath49th element equals @xmath235 for @xmath273 , and @xmath274 otherwise . finally , @xmath275 is a @xmath276 matrix , formed by taking the set @xmath277 of rows from an @xmath278 fft matrix . the formulation obtained in , is a classic cs problem , where our goal is to reconstruct the @xmath268-dimensional vector @xmath272 , known to be @xmath0-sparse , with @xmath279 , based on its projection onto a subset of @xmath137 orthogonal vectors , represented by the rows of @xmath280 . this problem may be solved by various cs methods , as long as the sensing matrix @xmath280 satisfies desired properties such as the restricted isometry property ( rip ) or coherence . in our case , @xmath280 is formed by choosing @xmath137 rows from the fourier basis . selecting these rows uniformly at random it may be shown that if @xmath281 for some positive constant @xmath282 , then @xmath280 obeys the rip with large probability @xcite . as readily seen from , the resolution of the grid , used for evaluating @xmath263 , directly effects the rip . recall that , by applying spectral analysis methods , one may reconstruct @xmath283 from a minimal number of @xmath2 samples , if it is indeed @xmath0-sparse . however , these samples must be carefully chosen . using matrix pencil , for instance , the sensing vectors must be consecutive . moreover , in any practical application , the measured data will be corrupted by noise , forcing us to use oversampling . in contrast , the bound proposed in regards random selection of the sensing vectors . additionally , applying the cs framework , we may effectively cope with the more general case , of reconstructing @xmath272 which is not necessarily @xmath0-sparse . in this section , we provide results obtained by applying three recovery algorithms to ultrasound signals which were simulated using the _ field ii _ program @xcite . the evaluation was performed based on multiple beamformed signals , each calculated along the @xmath12 axis ( @xmath133 ) for a random phantom realization . the phantom comprised @xmath0 strong reflectors , distributed along the @xmath12 axis , and multiple additional reflectors , distributed throughout the entire imaging medium . a measurement vector was obtained by projecting the beamformed signal onto a subset of its fourier series coefficients . finally , each algorithm was evaluated for its success in recovering the strong reflectors positions from the vector of measurements . the first two algorithms which were evaluated were matrix pencil @xcite and total least - squares approximation , enhanced by cadzow s iterated algorithm @xcite . both algorithms may be considered spectral analysis techniques . the third algorithm was orthogonal matching pursuit ( omp ) @xcite , which is a cs method . the simulation setup is depicted in fig . [ fig:06 ] . we created an aperture comprising 64 transducer elements , with central frequency @xmath284 . the width of each element , measured along the @xmath285 axis , was @xmath286 , and the height , measured along the @xmath287 axis , was @xmath288 . the elements were arranged along the @xmath3 axis , with a @xmath289 kerf . the transmitted pulse was simulated by exciting each element with two periods of a sinusoid at frequency @xmath237 , where the delays were adjusted such that the transmission focal point was at depth @xmath290 . additionally , hanning apodization was used during transmission , by applying an appropriate excitation power to each element . elements are aligned along the @xmath3 axis with a @xmath289 kerf . the width of each element is @xmath291 . speckle pattern is simulated by randomly distributing @xmath292 point reflectors within the box @xmath293 . additionally , @xmath294 point reflectors are aligned along the @xmath12 axis , also within the boundaries of the box . the pulse is transmitted along the @xmath12 axis , and the beamformed signal is constructed along the same line.,height=132 ] in each iteration , we constructed a random phantom , for which we simulated the beamformed signal . the phantom was constructed in two stages . we first created a speckle phantom , by drawing positions of @xmath292 point reflectors uniformly , at random , within the three - dimensional box @xmath295 . the corresponding amplitudes were also drawn randomly , with zero - mean and unit - variance normal distribution . we then generated a signal phantom , by drawing positions of @xmath294 point reflectors , @xmath296 , with @xmath297 and @xmath298 uniformly distributed in the interval @xmath299 . these reflectors were assigned identical amplitudes , which were adjusted according to the snr requirement , in the following manner : for each of the two phantoms , we simulated the beamformed signal , acquired along @xmath133 following pulse transmission in the same direction . denoting the beamformed signal obtained for the first ( speckle ) phantom by @xmath300 and that obtained for the second ( signal ) phantom by @xmath301 , we defined @xmath302 the amplitudes of the reflectors comprising the second phantom were modified , such that complied with the desired snr value . after this calibration , we combined the two phantoms into a single one , for which we generated an individual beamformed signal realization . the detected signals and the resulting beamformed signal were simulated at sampling rate @xmath303 . since the spectrum of the detected pulses decayed to @xmath304 at @xmath305 , this rate was far beyond nyquist . hanning apodization was used for constructing the beamformed signal , by applying appropriate weights to the detected signals . this type of apodization may be easily implemented with both our xampling schemes , by replacing the average in by a weighted one . [ fig:07 ] illustrates the method by which we simulated a realization of the noisy beamformed signal . this image was obtained by applying standard imaging techniques to an individual phantom . we are interested in recovering the strong reflections aligned along the @xmath12 axis . the corresponding beamformed signal was corrupted by speckles , originating in the multiple point reflectors scattered throughout the medium . the phantom was calibrated such that the snr of the beamformed signal along @xmath133 , defined in , was @xmath306 . strong reflectors aligned along the @xmath12 axis . @xmath292 point reflectors were distributed in the imaging plain , resulting in echoes which corrupted the detected signals . in the ultrasound image , these reflections are manifested in a speckle pattern . the phantom was calibrated such that the snr of the beamformed signal , calculated along @xmath307 , defined in , was @xmath306.,height=188 ] having generated the beamformed signal , we obtained a measurement vector , by projecting the signal onto a subset of its @xmath137 fourier series coefficients , where @xmath308 , and @xmath309 is the desired oversampling factor . for the spectral analysis techniques , we chose the coefficients consecutively , around @xmath310 . omp was tested using both this selection of coefficients , and a random selection , taken such that @xmath260 is above @xmath311 . with this selection , we obtain samples which are better spread in the frequency domain . we emphasize , that the coefficients were drawn once , for each choice of @xmath114 . an additional degree of freedom , using the omp method , regards the density of the reconstruction grid , determined by @xmath268 . we set @xmath312 , complying with a sampling frequency @xmath313 , of order typically used in imaging devices . recovery was evaluated based on the estimated time delays . these were compared to the delays associated with the known reflector positions , @xmath314 . at the @xmath315th iteration , we examined , for each algorithm , all possible matches between actual delays @xmath263 , and estimated delays @xmath316 . of all possible permutations ( a total number of @xmath317 ) , we selected the one for which the number of matches , achieving error smaller than the width of @xmath42 , was maximal . denoting this number by @xmath318 , @xmath319 corresponding to the evaluated method , we estimate the probability of recovery by the @xmath320th method as @xmath321 where @xmath322 is the total number of iterations , set to 500 in our simulation . we note that all reconstruction algorithms require that we first calculate @xmath260 . for this purpose , we simulated the signal beamformed along @xmath133 , for a phantom which comprised a single reflector at the transmission focal point @xmath323 . we used the detected signal , depicted in fig . [ fig:08 ] , for calculating @xmath260 . evaluated from the beamformed signal , calculated for a single reflector using _ field ii _ simulator . the reflector was positioned at the transmission focal point.,height=151 ] ( a ) ( b ) ( c ) ( d ) the simulation results obtained for multiple combinations of snr and oversampling factor are illustrated in fig . [ fig:09 ] . the calculated recovery probabilities are represented by gray - levels , where a common color - bar was used for all plots . for clarity , we plotted a line separating between probabilities lower than 0.85 and probabilities above 0.85 , and a line separating between probabilities lower than 0.97 and probabilities above 0.97 . of the two spectral analysis techniques , matrix pencil appears preferable , as it obtains high probability values over a wider range of snr and oversampling . both omp methods outperformed the spectral analysis ones , with an obvious advantage to random omp . an additional aspect which should be taken into consideration , when choosing the reconstruction method , regards the complexity of the xampling hardware . using the xampling scheme proposed in @xcite , random selection of fourier series coefficients will increase the hardware complexity : in such case , the sampling kernel , e.g. sos , must be specifically designed for the choice of coefficients . this is in contrast with the relatively simple kernel , applied for a consecutive choice of coefficients . on the other hand , the xampling scheme proposed in @xcite is practically invariant to the manner in which the coefficients are selected . ( a ) ( b ) ( c ) in this section , we examine results obtained by applying our xampling schemes , illustrated in figs . [ fig:04 ] and [ fig:05 ] , to raw rf data , acquired and stored for cardiac images of a healthy consenting volunteer . the acquisition was performed using a ge breadboard ultrasonic scanner of 64 acquisition channels . the transducer employed was a 64-element phased array probe , with @xmath324 central frequency , operating in second harmonic imaging mode : 3 half cycle pulses are transmitted at @xmath325 , resulting in a signal characterized by a rather narrow bandpass bandwidth , centered at @xmath325 . the corresponding second harmonic signal , centered at @xmath326 , is then acquired . the signal detected in each acquisition channel is amplified and digitized at a sampling - rate of @xmath327 . data from all channels were acquired along 120 beams , forming a @xmath328 sector , where imaging to a depth of @xmath329 , we have @xmath330 . the imaging results are illustrated in fig . [ fig:10 ] . the first image ( a ) was generated using the standard technique , applying beamforming to data first sampled at the nyquist - rate , and then down - sampled , exploiting its limited essential bandwidth . for a single scanline , sampling at @xmath331 , we acquire @xmath332 real - valued samples from each element , which are then down - sampled , to @xmath333 real - valued samples , used for beamforming . the resulting image is used as reference , where our goal is to reproduce the macroscopic reflectors observed in this image with our xampling schemes . we begin by applying the scheme illustrated in fig . [ fig:04 ] , utilizing the analog kernels defined in . modulation with the kernels is simulated digitally . assuming @xmath334 reflectors , and using two - fold oversampling , @xmath277 comprises @xmath335 consecutive indices . with such selection , the corresponding frequency samples practically cover the essential spectrum of @xmath42 . since each sample is complex , we get an eight - fold reduction in sample - rate . having estimated the fourier series coefficients of @xmath27 , we obtain its parametric representation by solving using omp . we then reconstruct @xmath27 according to , that is we apply phase shifts to the modulated pulses , based on the extracted coefficients phases . the resulting image ( b ) depicts the strong perturbations observed in ( a ) . moreover , isolated reflectors at the proximity of the array ( @xmath336 ) remain in focus . we next apply the approximated scheme , illustrated in fig . [ fig:05 ] : for every @xmath337 , @xmath338 and @xmath7 , we find @xmath209 and @xmath210 of such that @xmath339 . this process is performed numerically , off - line , based on our imaging setup . consequently , we construct @xmath340 off - line , according to . choosing this level of approximation , we end up with a seven - fold reduction in sample rate , where , for the construction of a single scanline , an average of @xmath341 complex samples must be taken from each element . we point out that in this scenario , the maximal number of samples , taken from certain elements , reaches @xmath342 for specific values of @xmath7 . thus , if a common rate is to be used for all sensors , for all values of @xmath7 , we may still achieve a six - fold reduction in sample rate . as before , we use omp in order to obtain @xmath27 s parametric representation , and reconstruct it based on our generalized fri model proposed in . the resulting image ( c ) appears very similar to ( b ) . table [ tab:01 ] gathers snr values , calculated for the beamformed signals estimated using both our xampling schemes , after envelope detection with the hilbert transform . the values were calculated with respect to the envelopes of the beamformed signals , obtained by standard imaging . explicitly , let @xmath343 denote the beamformed signal obtained by standard beamforming along the direction @xmath344 , @xmath345 , let @xmath346 denote the beamformed signal reconstructed from the parameters recovered by compressed beamforming along the same direction , and let @xmath347 denote the hilbert transform . for the set of @xmath348 scanlines , we defined the snr as @xmath349 this calculation was repeated when reconstructing the signals without the random phase assumption , proposed in section [ sec:71 ] . for the latter case , reconstruction of a real - valued @xmath346 , given complex coefficients , may be heuristically achieved by either ignoring the coefficients imaginary part , or by taking their modulus . it may be seen that , weighting over all @xmath350 beamformed signals , the random phase assumption achieves a relatively minor improvement ( @xmath351-@xmath352 ) compared to reconstruction using the modulus of the coefficients . however , when examining individual signals , we observed that , for certain values of @xmath344 , the improvement exceeded @xmath353 . .snr in [ db ] of @xmath27 obtained with the proposed xampling schemes and three reconstruction methods [ cols= " < , < , < " , ] [ tab:01 ] . we emphasize , that the calculated snr values provide a useful measure for quantitatively comparing the different xampling and reconstruction approaches . however , they are of smaller value when attempting to evaluate the overall performance of xampling , compared to standard imaging : recall that our scheme is aimed at reproducing only strong pulses , reflected from macroscopic reflectors . the reference signal , on the other hand , generated by standard technique , already contains the additional speckle component , caused by multiple microscopic perturbations . a possible approach for evaluating the overall performance of either xampling scheme , would be to examine its success rate in recovering strong reflections , detected by standard beamforming . for this purpose , we tracked the @xmath0 strongest local maxima in each beamformed signal . if the xampling scheme recovered a pulse within the range of @xmath354 from a certain maximum , we say that this maximum was successfully detected . certain pulses , detected by xampling , may match more than one maximum in the beamformed signal . in such case , we choose the one - to - one mapping which achieves smallest mse . applying this evaluation method to signals xampled using our approximated scheme , and reconstructed with the random phase assumption , we conclude that the reconstruction successfully retrieves @xmath355 of the significant maxima , with standard deviation of the error being approximately @xmath356 . in this work , we generalized the xampling method proposed in @xcite , to a scheme applied to an array of multiple receiving elements , allowing reconstruction of a two - dimensional ultrasound image . at the heart of this generalization was the proposal that the one - dimensional xampling method derived in @xcite be applied to signals obtained by beamforming . such signals exhibit enhanced snr , compared to the individual signals detected by the array elements . moreover , they depict reflections which originate in a much narrower sector , than that initially radiated by the transmitted pulse . a second key observation , which made our approach feasible , regarded the integration of the beamforming process into the filtering part of the xampling scheme . the first approach we purposed comprised multiple modulation and integration channels , utilizing analog kernels . we next showed that the parametric representation of the beamformed signal may be well approximated , from projections of the detected signals onto appropriate subsets of their fourier series coefficients . the contribution of our schemes regards both the reduction in sample rate , but additionally , the resulting reduction in the rate of data transmission from the system front - end to the processing unit . in particular , our second approach is significant even when preliminary sampling is performed at the nyquist - rate . in such a case , it allows a reduction in data transmission rate , by a relatively simple linear transformation , applied to the sampled data . an additional contribution of our work regards the method by which we reconstruct the ultrasound signal , assumed to obey a specific fri structure , from a subset of its frequency samples . rather than using traditional spectral analysis techniques , we formulate the relationship between the signal s samples to its unknown parameters as a cs problem . the latter may be efficiently solved using a greedy algorithm such as the omp . we show that , in our scenario , cs is generally comparable to spectral analysis methods , managing to achieve similar success rates with sample sets of equal cardinality . moreover , working in a noisy regime , cs typically outperformed spectral analysis methods , provided that the frequency samples were highly spread over the essential spectrum of the signal . using actual cardiac data , a relatively large number of reflectors was assumed . consequently , by simply choosing the fourier series coefficients consecutively , as in the spectral analysis techniques , we end up with the necessary wide distribution . however , as shown in our simulations , cs approach inherently allows a wide distribution of samples , even when the cardinality of the sample set is small , since we are not obliged to unique configurations of samples . a final observation discussed in our work , regards the generalization of the signal model proposed in @xcite , allowing additional , unknown phase shifts , of the detected pulses . we show that these shifts may be estimated by appropriate interpretation of the extracted coefficients , without changing the recovery method . combining the random phase assumption with our proposed xampling schemes and the cs recovery method , we construct two - dimensional ultrasound images , which well depict strong perturbations in the tissue , while achieving up to seven - fold reduction of sample rate , compared to standard imaging techniques . we assume @xmath42 to be supported on @xmath84 , and that the support of @xmath11 is contained in @xmath1 . the last assumption may be justified by the fact that the pulse is transmitted at @xmath10 , such that reflections may only be detected for @xmath357 . additionally , the penetration depth of the transmitted pulse allows us to set @xmath37 , such that all reflections arriving at @xmath358 are below the noise level . for all @xmath359 and @xmath360 : @xmath361 applying the relation @xmath92 , justified in section [ sec:51 ] , and using the fact that @xmath90 is non - decreasing for @xmath357 we conclude that @xmath362 @xmath363 being the inverse of @xmath90 . explicitly : @xmath364 assuming that @xmath365 , then , since is true for every @xmath366 , we may write : @xmath367 this allows us to set the following upper bound on the support of @xmath27 : @xmath368 once again , using the assumption that @xmath365 . from it is readily seen that @xmath369 , since we can always find @xmath125 with sign opposite to that of @xmath370 , such that : @xmath371 finally , by construction of @xmath372 we see that , for all @xmath360 , @xmath373 . t. l. szabo , `` diagnostics ultrasound imaging : inside out , '' in _ academic press series in biomedical engineering _ , j. bronzino , ed.1em plus 0.5em minus 0.4emelsevier academic press , 2004 , pp . ch . 7 , 10 . m. a. davenport , m. f. duarte , y. c. eldar , and g. kutyniok , `` introduction to compressed sensing , '' in _ compressed sensing : theory and applications _ , y. c. eldar and g. kutyniok , eds.1em plus 0.5em minus 0.4em cambridge university press , 2012 . j. uriguen , y. c. eldar , p. l. dragotti , and z. ben - haim , `` sampling at the rate of innovation : theory and applications , '' in _ compressed sensing : theory and applications _ , y. c. eldar and g. kutyniok , eds.1em plus 0.5em minus 0.4emcambridge university press , 2012 . m. mishali and y. c. eldar , `` xampling : compressed sensing for analog signals , '' in _ compressed sensing : theory and applications _ , y. c. eldar and g. kutyniok , eds.1em plus 0.5em minus 0.4emcambridge university press , 2012 . a. e. brown , `` rationale and summary of methods for determining ultrasonic properties of materials at lawrence livermore national laboratory , '' lawrence livermore national lab . , ca ( united states ) , tech . ucrl - id119958 , 1995 .

the primary goal of this paper is to develop and analyze some fully discrete finite element methods for a displacement - pressure model modeling swelling dynamics of polymer gels under mechanical constraints . in the model , the swelling dynamics is governed by the solvent permeation and the elastic interaction ; the permeation is described by a pressure equation for the solvent , and the elastic interaction is described by displacement equations for the solid network of the gel . the elasticity is of long range nature and gives effects for the solvent diffusion . it is the fluid - solid interaction in the gel network drives the system and makes the problem interesting and difficult . by introducing an elastic pressure " ( or volume change function " ) we first present a reformulation of the original model , we then propose a time - stepping scheme which decouples the pde system at each time step into two sub - problems , one of which is a generalized stokes problem for the displacement vector field ( of the solid network of the gel ) and another is a diffusion problem for a pseudo - pressure " field ( of the solvent of the gel ) . to make such a multiphysical approach feasible , it is vital to find admissible constraints to resolve the uniqueness issue for the generalized stokes problem and to construct a good " boundary condition for the diffusion equation so that it also becomes uniquely solvable . the key to the first difficulty is to discover certain conservation laws ( or conserved quantities ) for the pde solution of the original model , and the solution to the second difficulty is to use the generalized stokes problem to generate a boundary condition for the diffusion problem . this then lays down the theoretical foundation for one to utilize any convergent stokes solver ( and its code ) together with any convergent diffusion equation solver ( and its code ) to solve the polymer gel model . in the paper , the taylor - hood mixed finite element method combined with the continuous linear finite element method are chosen as an example to present the ideas and to demonstrate the viability of the proposed multiphysical approach . it is proved that , under a mesh constraint , both the proposed semi - discrete ( in space ) and fully discrete methods enjoy some discrete energy laws which mimic the differential energy law satisfied by the pde solution . optimal order error estimates in various norms are established for the numerical solutions of both the semi - discrete and fully discrete methods . numerical experiments are also presented to show the efficiency of the proposed approach and methods . gels , soft matters , poroelasticity , stokes equations , finite element methods , inf - sup condition , fully discrete schemes , error estimates . 65m12 , 65m15 , 65m60 ,

0903.4000

a gel is a soft poroelastic material which consists of a solid network and a colloidal solvent . the solid network spans the volume of the solvent medium . the solvent can permeate through the solid network and the permeation can be controlled by external forces . both by weight and volume , gels are mostly liquid in composition and thus exhibit densities similar to liquids . however , they have the structural coherence of a solid and can be deformed . a gel network can be composed of a wide variety of materials , including particles , polymers and proteins , which then gives different types gels such hydrogels , organogels and xerogels ( cf . gels have some fascinating properties , in particular , they display thixotropy which means that they become fluid when agitated , but resolidify when resting . in general , gels are apparently solid , jelly - like materials , they exhibit an important state of matter found in a wide variety of biomedical and chemical systems ( cf . @xcite and the references therein ) . this paper develops and analyzes some fully discrete finite element methods for a displacement - pressure model for polymer gels . the model , which was proposed by m. doi _ et al _ in @xcite , describes swelling dynamics of polymer gels ( under mechanical constraints ) . let @xmath0 be a bounded domain and denote the initial region occupied by the gel . let @xmath1 denote the displacement of the gel at the point @xmath2 in the space and at the time @xmath3 , @xmath4 and @xmath5 be the velocity and the pressure of the solvent at @xmath6 . following @xcite , the governing equation for the swelling dynamics of polymer gels are given by @xmath7 here @xmath8 is the friction constant associated with the motion of the polymer relative to the solvent , @xmath9 is the volume fraction of the polymer , @xmath10 denotes the @xmath11 identity matrix , and @xmath12 stands for the stress tensor of the gel network , which is given by a constitutive equation . in this paper , we use the following linearized form of the stress tensor : @xmath13 where @xmath14 and @xmath15 are respectively the bulk and shear modulus of the gel ( cf . @xcite ) . we remark that stands for the force balance , states darcy s law for the permeation of solvent through the gel network , and describes the incompressibility condition . in addition , if we introduce the total stress @xmath16 , then equation becomes @xmath17 . substituting into and into yield the following basic equations for swelling dynamics of polymer gels ( see @xcite ) @xmath18 which hold in the space - time domain @xmath19 for some given @xmath20 . to close the above system , we need to prescribe boundary and initial conditions . only one initial condition is required for the system , which is @xmath21 various sets of boundary conditions are possible and each of them describes a certain type mechanical condition and solvent permeation condition ( cf . @xcite ) . in this paper we consider the following set of boundary conditions @xmath22 where @xmath23 denotes the outward normal to @xmath24 . @xmath25 means that the mechanical force @xmath26 is applied on the boundary of the gel . since @xmath27 , hence , @xmath28 implies that the solvent can not permeate through the gel boundary . we also remark that the force function @xmath26 must satisfy the compatibility condition @xmath29 problem is interesting and difficult due to its multiphysical nature which describes the complicate fluid and solid interaction inside the gel network . it is numerically tricky to solve because it is difficult to design a good and workable time - stepping scheme . for example , one natural attempt would be at each time step first to solve a poisson problem for @xmath30 and then to solve a linear elasticity problem for @xmath31 . however , this strategy is difficult to realize because there is no good way to compute the source term @xmath32 for the poisson equation . in fact , the strategy even has a difficulty to start due to the fact that no initial condition is provided for the pressure @xmath30 . to overcome the difficulty , in this paper we shall use a reformulation of system , which is now introduced . define @xmath33 physically , @xmath34 measures the volume change of the solid network of the gel , and often called elastic pressure " or volume change function " . taking divergence on yields @xmath35 which and imply that @xmath34 satisfies the following diffusion equation @xmath36 however , the usefulness of the above diffusion equation is hampered by the lack of boundary condition for @xmath34 . we like to note that the above diffusion equation for @xmath34 was first noticed by m. doi @xcite , but it was not utilized before exactly because of the lack of boundary condition for @xmath34 . nevertheless , using the new variable @xmath34 we can rewrite as @xmath37 an immediate consequence of the above reformulation is that - implies @xmath38 satisfies the generalized stokes equations with @xmath34 being the source term at each time @xmath3 , and @xmath34 satisfies a diffusion equation and it interacts with @xmath38 only at the boundary @xmath24 . this is a key observation because it not only reveals the underlying physical process of swelling dynamics of the gel , but also gives the right " hint on how the problem should be solved numerically . this indeed motivates the main idea of this paper , that is , _ at each time step , we first solve the generalized stokes problem for @xmath39 , which in turn provides ( implicitly ) an updated boundary condition for @xmath34 , we then use this new boundary condition to solve the diffusion equation for @xmath34 . the process is repeated iteratively until the final time step is reached_. however , in order to make this idea work , there is one crucial issue needs to be addressed . that is , for a given @xmath34 the generalized stokes problem for @xmath39 is only unique up to additive constants . clearly , how to correctly enforce the uniqueness of the generalized stokes problem is the bottleneck of this approach . it is easy to understand that one can not use arbitrary constraints to fix @xmath39 because this will lead to bad or even divergent numerical schemes if the exact pde solution does not satisfy the constraints . instead , the constraints which can be used to fix @xmath39 should be those satisfied by the exact solution of the pde system . to the end , we need to discover some invariant ( or conserved ) quantities for the exact pde solution . it turns out that the situation is precisely what we anticipated and wanted . we are able to show that the exact pde solution @xmath40 satisfies the following identities ( see section [ sec-2 ] below for a proof ) : @xmath41 where @xmath42 denote the dimension of @xmath43 and @xmath44 obviously , the right - hand sides of and are constants . the right - hand side of is also a constant provided that @xmath26 is independent of @xmath3 , otherwise , it is a known function of @xmath3 . in this paper , we shall only consider the case that @xmath26 is independent of @xmath3 . it follows from and that @xmath45 it is clear now that and provide two natural conditions which can be used to uniquely determine the solution @xmath38 to the generalized stokes problem for a given source term @xmath34 . this then leads to the following time - discretization for problem : * algorithm 1 : * * set @xmath46 and @xmath47 . * for @xmath48 , do the following two steps + _ step 1 : _ solve for @xmath49 such that @xmath50 + _ step 2 : _ solve for @xmath51 such that @xmath52 where @xmath53 . we note that is the implicit euler scheme , which is chosen just for the ease of presentation , it can be replaced by other time - stepping schemes . provides a neumann boundary condition for @xmath51 . another subtle issue is the role which the initial value @xmath54 plays in the algorithm . seems @xmath54 is only needed to produce @xmath55 and there is no need to have @xmath56 in order to execute the algorithm . however , to ensure the stability and convergence of the algorithm , it turns out that @xmath56 not only needs to be provided but also must be carefully constructed when the algorithm is discretized ( see sections [ sec-3 ] and [ sec-4 ] ) . the above algorithm has a couple attractive features . first , it is easy to use . second , it allows one to make use of any available numerical methods ( finite element , finite difference , finite volume , spectral and discontinuous galerkin ) and computer codes for the stokes problem and the poisson problem to solve the gel swelling dynamics model . we remark that an almost same model as also arise from different applications in poroelasticity and soil mechanics and are known as boit s consolidation model ( cf . @xcite and the references therein ) . @xcite proposed and analyzed a standard finite element method which directly approximates @xmath57 under the ( restrictive ) divergence - free assumption on the initial condition @xmath54 . this paper consists of four additional sections . in section [ sec-2 ] , we first introduce notation used in this paper . we then present a pde analysis for the gel swelling dynamics model , which includes deriving a dissipative energy law , establishing existence and uniqueness , and in particular , proving the conservation laws stated in and . in section [ sec-3 ] , we first propose a semi - discrete ( in space ) finite element discretization for problem based on the multiphysical reformulation . the well - known taylor - hood mixed element and the @xmath58 conforming finite element are used as an example to present the ideas . it is proved that the solution of the semi - discrete method satisfies a discrete energy law which mimics the differential energy law enjoyed by the pde solution , and the semi - discrete numerical solution also satisfies the conservation laws and . we then derive optimal order error estimates in various norms for the semi - discrete numerical solution . in section [ sec-4 ] , fully discrete finite element methods are constructed by combining the time - stepping scheme of algorithm @xmath59 and the semi - discrete finite element methods of section [ sec-3 ] . the main results of this section include proving a fully discrete energy law for the numerical solution and establishing optimal order error estimates for the fully discrete finite element methods . finally , in section [ sec-5 ] , we present some numerical experiments to gauge the efficiency of the proposed approach and methods . the standard sobolev space notation is used in this paper , we refer to @xcite for their precise definitions . in particular , @xmath60 and @xmath61 denote respectively the standard @xmath62 and @xmath63 inner products . for any banach space @xmath64 , we let @xmath65^d$ ] and use @xmath66 to denote its dual space . in particular , we use @xmath67 and @xmath68 to denote the dual products on @xmath69 and @xmath70 , respectively . @xmath71 is a shorthand notation for @xmath72 . we also introduce the function spaces @xmath73 it is well known @xcite that the following so - called inf - sup condition holds in the space @xmath74 : @xmath75 throughout the paper , we assume @xmath76 be a bounded polygonal domain such that @xmath77 is an isomorphism ; see @xcite . in addition , @xmath78 is used to denote a generic positive constant which is independent of @xmath79 , and the mesh parameters @xmath80 and @xmath81 . we now give a definition of weak solutions to - . [ weak1 ] let @xmath82 , and @xmath83 . given @xmath84 , a tuple @xmath57 with @xmath85 is called a weak solution to , if there hold for almost every @xmath86 $ ] @xmath87 similarly , we define weak solutions to problem - , - . [ weak2 ] let @xmath82 , and @xmath83 . given @xmath84 , a triple @xmath88 with @xmath89 is called a weak solution to , - if there hold for almost every @xmath86 $ ] @xmath90 \(a ) clearly , @xmath91 gives back the pressure in the original formulation . what interesting is that both @xmath34 and @xmath92 are only @xmath93-functions in the spatial variable but their combination @xmath94 is an @xmath95-function . in other words , the new formulation provides an @xmath96 decomposition for the pressure @xmath30 . it turns out that this decomposition will have a significant numerical impact because it allows one to use low order ( hence cheap ) finite elements to approximate @xmath34 and @xmath92 but still to be able to approximate the pressure @xmath30 with high accuracy . \(b ) implicitly imposes the following boundary condition for @xmath34 : @xmath97 since problem consists of two linear equations , its solvability should follows easily if we can establish a priori energy estimates for its solutions . the following dissipative energy law just serves that purpose . [ lem2.1 ] every weak solution @xmath57 of problem satisfies the following energy law : @xmath98,\end{aligned}\ ] ] where @xmath99.\ ] ] moreover , @xmath100 we first consider the case @xmath101 . setting @xmath102 in and @xmath103 in yield @xmath104 & = \frac{d}{dt } \langle \bbf , \bu(t ) \rangle_{\small\rm dual } + \bigl(p(t ) , \div \bu_t(t ) \bigr ) .\end{aligned}\ ] ] follows from adding the above two equations and integrating the sum in @xmath3 over the interval @xmath105 for any @xmath106 $ ] . if @xmath57 is only a weak solution , then @xmath107 could not be used as a test function in . however , this difficulty can be easily overcome by using @xmath108 as a test function in , where @xmath109 denotes a mollification of @xmath31 through a symmetric mollifier ( cf . * chapter 7 ) ) , and by passing to the limit @xmath110 . finally , follows immediately from and . the proof is completed . lemma [ lem2.1 ] and theorem [ thm2.1 ] can be easily carried over to the reformulated problem , . the only difference is that the energy law now is replaced by the following equivalent energy law : @xmath111 } { l^2}^2\ , ds = j(0 ) \qquad\forall t\in ( 0,t],\end{aligned}\ ] ] and is replaced by @xmath112 where @xmath113.\ ] ] we also note that the weak solution to problem , is understood in the sense of definition [ weak2 ] . [ lem2.2 ] weak solutions of problem and problem , satisfy the conservation laws , and . and follows immediately from taking @xmath114 in , @xmath115 in both and followed by integrating in @xmath3 and appealing to the divergence theorem . to prove , taking @xmath116 in and using the identities @xmath117 and @xmath118 we get @xmath119 hence , @xmath120 where @xmath121 and @xmath122 are defined in and . so we obtain . similarly , follows by taking @xmath116 in . the proof is complete . with the help of the above two lemmas , we can show the solvability of problem . [ thm2.1 ] suppose @xmath82 , and @xmath83 . then there exists a unique weak solution to in the sense of definition [ weak1 ] , and there exists a unique solution to to , . the uniqueness is an immediate consequence of the energy laws and and the conservation laws and . the existence can be easily proved by using ( abstract ) galerkin method and the standard compactness argument ( cf . @xcite ) . the energy laws and provide the necessary uniform estimates for the galerkin approximate solutions . we omit the details because the derivation is standard . again , exploiting the linearity of the system , we have the following regularity results for the weak solution . [ regularity ] let @xmath123 be the weak solution of then there holds the following estimates : @xmath124^{\frac12}. \label{e2.11a } \\ \norm{tq_{tt}}{l^2(h^{-1 } ) } & \leq [ 2j(0)]^{\frac12}. \label{e2.11b } \\ \kappa \sqrt{\alpha } \norm{\sqrt{t } \del p}{l^2(l^2 ) } = \sqrt{\alpha } \norm{\sqrt{t } q_t}{l^2(l^2 ) } & \leq j(0)^{\frac12}.\label{e2.11c } \\ \kappa \sqrt{\alpha } \norm{t \del p}{l^\infty(l^2 ) } = \sqrt{\alpha } \norm{t q_t}{l^\infty(l^2 ) } & \leq [ 2j(0)]^{\frac12}.\label{e2.11d } \\ \norm{\nab\tp}{l^2(l^2 ) } + \norm{\nab\tp}{l^\infty(l^2 ) } + \alpha \sqrt{\kappa } \norm{\nab q}{l^2(l^2 ) } & \quad \label{e2.11e } \\ + \alpha \sqrt{\kappa } \norm{\sqrt{t}\nab q}{l^\infty(l^2 ) } & \leq c(c_{\tp } , j(0 ) ) . \no \\ \beta \norm{\del \bu}{l^\infty ( l^2 ) } = \norm{\nab\tp}{l^\infty(l^2 ) } & \leq c(c_{\tp } , j(0 ) ) . \label{e2.11f } \end{aligned}\ ] ] where @xmath125 denotes some positive constant which depends on @xmath126 and @xmath127 . differentiating each of equations , and with respect to @xmath3 yields ( note that @xmath26 is assumed to be independent of @xmath3 ) @xmath128 taking @xmath129 in , @xmath130 in , @xmath131 in , and adding the resulted equations we get @xmath132 integrating in @xmath3 over @xmath133 gives . alternatively , setting @xmath134 in , @xmath135 in ( after differentiating in @xmath3 one more time ) , @xmath136 in , and adding the resulted equations we have @xmath137 + t^2 \kappa \norm{\nab p_t}{l^2}^2 = t\beta\norm{\nab \bu_t}{l^2}^2 + t\norm{q_t}{l^2}^2 . \end{aligned}\ ] ] integrating in @xmath3 over @xmath133 and using yield . are immediate consequences of , , and . choose @xmath138 such that @xmath139 . then @xmath140 . by an equivalent version of the inf - sup condition ( cf . @xcite ) we have @xmath141.\end{aligned}\ ] ] hence , @xmath142,\ ] ] which together with and imply that ( recall @xmath143 ) @xmath144,\\ & \alpha \sqrt{\kappa } \norm{\nab q}{l^2(l^2 ) } \leq j(0)^{\frac12 } + \sqrt{\kappa t } \norm{\nab \phi_0}{l^2 } + \alpha_0^{-1}\sqrt{\kappa t}\bigl[\beta j(0)^{\frac12 } + \sqrt{d}\norm{\phi_0}{l^2}\bigr],\\ & \alpha \sqrt{\kappa } \norm{\sqrt{t}\nab q}{l^\infty(l^2 ) } \leq j(0)^{\frac12 } + \sqrt{\kappa t } \norm{\nab \phi_0}{l^2 } + \alpha_0^{-1 } \sqrt{\kappa t } \bigl [ \beta j(0)^{\frac12 } + \sqrt{d}\norm{\phi_0}{l^2}\bigr].\end{aligned}\ ] ] hence , holds . finally , follows immediately from the equation @xmath145 and . the proof is complete . the goal of this section is to present the ideas and specific semi - discrete finite element methods for discretizing the variational problem in space based on the multiphysical ( deformation and diffusion ) approach . that is , we shall approximate @xmath39 using a ( stable ) stokes solver and approximate @xmath34 by a convergent diffusion equation solver . the combination of the taylor - hood mixed finite element @xcite and the conforming @xmath58 finite element is chosen as a specific example to present the ideas . assume @xmath146 ( @xmath147 ) is a polygonal domain . let @xmath148 is a quasi - uniform triangulation or rectangular partition of @xmath76 with mesh size @xmath80 such that @xmath149 . also , let @xmath150 be a stable mixed finite element pair , that is , @xmath151 and @xmath152 satisfy the inf - sup condition @xmath153 a well - known example that satisfies is the following so - called taylor - hood element ( cf . @xcite ) : @xmath154 in the sequel , we shall only present the analysis for the taylor - hood element , but remark that the analysis can be readily extended to other stable combinations . on the other hand , constant pressure space is not recommended because that would result in no rate of convergence for the approximation of the pressure @xmath30 ( see section [ sec-3.3 ] ) . approximation space @xmath155 for @xmath34 variable can be chosen independently , any piecewise polynomial space is acceptable as long as @xmath156 . especially , @xmath157 can be chosen as a fully discontinuous piecewise polynomial space , although it is more convenient to choose @xmath155 to be a continuous ( resp . discontinuous ) space if @xmath158 is a continuous ( resp . discontinuous ) space . the most convenient and efficient choice is @xmath159 , which will be adopted in the remaining of this paper . we now ready to state our semi - discrete finite element method for problem . let @xmath160 . we seek @xmath161\to \bv_h\times m_h\times w_h$ ] and @xmath162 such that for all @xmath163 $ ] there hold @xmath164 where @xmath165 denotes the time derivative of @xmath166 . \(a ) we note that and defines @xmath167 and @xmath168 ( see their definitions below ) . it is easy to see that in general @xmath169 although @xmath170 . but it is not hard to enforce @xmath171 by defining @xmath172 slightly differently ( in the case of continuous @xmath158 ) or simply substituting by @xmath173 ( in the case of discontinuous @xmath158 ) , even such a modification is not necessary for the sake of convergence ( see section [ sec-3.3 ] ) . we also note that @xmath167 can be replaced by the @xmath93 projection @xmath174 , the only drawback " of using the @xmath93 projection is that it produces a larger error constant . \(b ) in the case that @xmath158 and/or @xmath155 are discontinuous spaces , since @xmath175 and/or @xmath176 , then the second term on the left - hand side of must be modified into a sum over all elements of the integrals defined on each element @xmath177 , and additional jump terms on element edges may also need to be introduced to ensure convergence . we conclude this subsection by citing a few well - known facts about the finite element functions . first , we recall the following inverse inequality for polynomial functions @xcite : @xmath178 second , for any @xmath179 , we define its @xmath93 projection @xmath180 as follows : @xmath181 it is well - known that the projection operator @xmath182 satisfies ( cf . @xcite ) , for any @xmath183 @xmath184 we remark that in the case @xmath176 , the second term on the left - hand side of the above inequality has to be replaced by the broken @xmath95-norm . finally , for any @xmath185 we define its elliptic projection @xmath186 by @xmath187 also , for any @xmath188 , we define its elliptic projection @xmath189 by @xmath190 it is well - known that the projection operators @xmath191 and @xmath192 satisfy ( cf . @xcite ) , for any @xmath193 , @xmath194 and @xmath195 @xmath196 in this subsection , we shall prove that the semi - discrete solution @xmath197 defined in the previous subsection satisfies an energy law similar to . in addition , they satisfy the same conservation laws as those enjoyed by their continuous counterparts @xmath198 ( see lemma [ lem2.2 ] ) . an immediate consequence of the stability and the conservation laws is the well - posedness of problem . moreover , the stability also serves as a step stone for us to establish convergence results in the next subsection . [ lem3.1a ] let @xmath197 be a solution of and set @xmath199 . then there holds the following energy law : @xmath200,\end{aligned}\ ] ] where @xmath201.\ ] ] if @xmath202 , then follows immediately from setting @xmath203 in , @xmath204 after differentiating with respect to @xmath3 , @xmath205 , and adding the resulted equations . on the other hand , if @xmath206 , then @xmath203 is not a valid test function . this technical difficulty can be easily overcome by smoothing @xmath207 in @xmath3 through a symmetric mollifier as described in the proof of lemma [ lem2.1 ] . [ lem3.2a ] every solution @xmath197 of satisfies the following conservation laws : @xmath208 @xmath25 follows from setting @xmath209 in and @xmath210 is . @xmath28 follows from taking @xmath211 in . @xmath212 derives from letting @xmath213 in . finally , @xmath214 follows from @xmath25 , @xmath212 , and the definition of @xmath215 . with the help of the above two lemmas , we can prove the solvability of problem . [ thm3.1a ] suppose @xmath82 , and @xmath83 . then has a unique solution . since can be written as an initial value problem for a system of linear odes , the existence of solutions follows immediately from the linear ode theory . to show the uniqueness , it suffices to prove that the problem only has the trivial solution if @xmath216 . for the zero sources , the energy law and the @xmath95-norm stability of the elliptic projection , @xmath217 immediately imply that @xmath218 , and @xmath34 are constant functions . since @xmath219 as @xmath216 , hence , @xmath218 , and @xmath34 must identically equal zero . the proof is complete . by exploiting the linearity of equations , we can prove certain energy estimates for the time derivatives of the solution @xmath220 . [ regularity_1 ] the solution @xmath220 of the semi - discrete method and @xmath199 satisfy the following estimates : @xmath221^{\frac12}. \label{eq3.16 } \\ \norm{t(q_h)_{tt}}{l^2(h^{-1 } ) } & \leq [ 2j_h(0)]^{\frac12}. \label{eq3.17 } \end{aligned}\ ] ] because the proofs of follow exactly the same lines of the proofs of their differential counterparts given in theorem [ regularity ] , we omit them . define the error functions @xmath222 and @xmath223 trivially , @xmath224 . subtracting each of from their respective counterparts in we obtain the following error equations : @xmath225 introduce the following decomposition of error functions @xmath226 trivially , @xmath227 . by the definition of @xmath228 and @xmath229 , the above error equations can be written as @xmath230 so @xmath231 and @xmath232 satisfy the same type of equations as @xmath233 and @xmath166 do , except the terms containing @xmath234 and @xmath235 on the right - hand sides of the equations . hence , @xmath231 and @xmath232 are expected to satisfy an equality similar to the energy law . to the end , taking @xmath236 ( after differentiating with respect to @xmath3 ) , @xmath237 in , and adding the resulted equations we obtain @xmath238 + \kappa \|\nab \phi_p\|_{l^2}^2 \\ & \hskip 0.5 in = \bigl(\lam_{\tp } , \div ( \bthetau)_t \bigr ) -\bigl(\div ( \blamu)_t , \theta_{\tp } \bigr ) + \bigl((\theta_q)_t , \psi_p-\lam_p \bigr ) \no \\ & \hskip 0.5 in = \frac{d}{dt } \big [ \bigl(\lam_{\tp } , \div \bthetau \bigr ) + \bigl(\theta_q , \psi_p-\lam_p \bigr ) \bigr ] - \bigl((\lam_{\tp})_t , \div \bthetau \bigr ) \no \\ & \hskip 1.9 in -\bigl(\theta_q , ( \psi_p-\lam_p)_t \bigr ) -\bigl(\div(\blamu)_t , \theta_{\tp } \bigr ) . \no \ ] ] integrating in @xmath3 and using schwarz inequality we get @xmath239 + \kappa \int_0^t \|\nab \phi_p(s)\|_{l^2}^2\ , ds \\ & \quad = \bigl(\lam_{\tp}(t ) , \div \bthetau(t ) \bigr ) + \bigl(\theta_q(t ) , \psi_p(t)-\lam_p(t ) \bigr ) -\int_0^t \bigl(\theta_q , ( \psi_p(s)-\lam_p(s))_t \bigr)\ , ds \\ & \qquad -\int_0^t \bigl [ \bigl((\lam_{\tp}(s))_s , \div \bthetau(s ) \bigr ) + \bigl(\div(\blamu(s))_s , \theta_{\tp}(t ) \bigr ) \bigr]\ , ds\\ & \quad \leq \frac{\beta}4 \norm{\nab \bthetau(t)}{l^2}^2 + \frac{1}{\beta } \norm{\lam_{\tp}(t)}{l^2}^2 + \frac{\alpha}4 \|\theta_q(t)\|_{l^2}^2 \\ & \qquad + \frac{1}{\alpha } \bigl[\norm{\psi_p(t)}{l^2}^2+\norm{\lam_p(t)}{l^2}^2\bigr ] + \frac{\epsilon_0 \kappa}2 \int_0^t \|\phi_p(s)\|_{l^2}^2\ , ds \\ & \qquad + \int_0^t \bigl [ \beta \norm{\nab\bthetau(s)}{l^2}^2 + \alpha \norm{\theta_q(s)}{l^2}^2 \bigr]\,ds + \frac{1}{\alpha } \int_0^t \bigl[\norm{(\psi_p(s ) -\lam_p(s))_s}{l^2}^2\bigr]\ , ds \\ & \qquad + c\int_0^t \bigl [ \frac{1}{\beta } \norm{(\lam_{\tp}(s))_s}{l^2}^2 + \bigl(\alpha+\frac{1}{\epsilon_0\kappa}\bigr ) \norm{\div(\blamu(s))_s}{l^2}^2 \bigr]\,ds . \ ] ] where we have used the facts that @xmath240 , @xmath241 , and @xmath242 to get the second inequality . @xmath243 is a positive constant to be chosen later . since @xmath244 , using poinca s inequality @xmath245 and choosing @xmath246 the above inequality can be written as @xmath247\,ds + \mathcal{a}(t;\bu,\tp , p , q ) , \no\end{aligned}\ ] ] where @xmath248\\ & \qquad + c\int_0^t \bigl[\frac{1}{\beta } \norm{(\lam_{\tp}(s))_s}{l^2}^2 + \bigl(\alpha+\frac{c_1}{\kappa}\bigr ) \norm{\div(\blamu(s))_s}{l^2}^2 \bigr]\,ds\\ & \qquad + \frac{4}{\alpha } \int_0^t \bigl[\norm{(\psi_p(s))_s}{l^2}^2 + \norm{(\lam_p(s))_s}{l^2}^2 \bigr ] \ , ds . \ ] ] from we deduce @xmath249 where @xmath250,\\ c_2(t;\bu,\tp , p , q ) & : = c\bigl [ \beta^{-1 } \norm{\tp}{l^\infty(h^2)}^2 + \beta^{-1 } \norm{\tp_t}{l^2(h^2)}^2 \\ & \qquad + ( \alpha + c_1 \kappa^{-1 } ) \norm{(\div\bu)_t}{l^2(h^2)}^2 + \alpha^{-1 } \bigl [ \norm{p}{l^\infty(h^2)}^2 + \norm{p_t}{l^2(h^2)}^2 \bigr].\end{aligned}\ ] ] it follows from an application of the gronwall s lemma to that @xmath251 + \kappa \int_0^t \|\nab \phi_p(s)\|_{l^2}^2\ , ds \leq \mathcal{a}(t;\bu,\tp , q)\,e^t.\end{aligned}\ ] ] an application of the triangle inequality yields the following main theorem of this section . [ theorem3.1 ] assume @xmath252 . let @xmath253,\\ \widehat{c}_2(t;\bu , q ) & : = c\bigl [ \beta \norm{\bu}{l^\infty(h^3)}^2 + \alpha \norm{q}{l^\infty(h^2)}^2 \bigr].\end{aligned}\ ] ] suppose that @xmath254 and @xmath255 , then there holds error estimate @xmath256 \\ & \qquad + \bigl ( \kappa \int^t_0\|\nab(p - p_h)(t)\|_{l^2}^2\,dt\bigr)^{\frac12 } \leq \bigl [ c_1(t;\bu,\tp , p , q)^{\frac12}\,e^{\frac{t}{2 } } + \widehat{c}_1(t;\bu , p , q)^{\frac12 } \bigr ] h , \no\end{aligned}\ ] ] moreover , if @xmath257 and @xmath258 , then there also holds @xmath259 \\ & \hskip 2.2 in \leq \bigl [ c_2(t;\bu,\tp , p , q)^{\frac12}\,e^{\frac{t}{2 } } + \widehat{c}_2(t;\bu , q)^{\frac12 } \bigr ] h^2 , \no\end{aligned}\ ] ] \(a ) we note that the above error estimates are optimal , and @xmath260 enjoys a superconvergence when the pde solution is regular . it is also interesting to remark that the initial errors at @xmath261 do not appear in the above error bounds . \(b ) in light of theorem [ regularity ] , the regularity assumptions of theorem [ theorem3.1 ] are valid if the domain @xmath43 and datum functions @xmath26 and @xmath54 are sufficient regular . in this section , we consider space - time discretization which combines the time - stepping scheme of algorithm 1 with the ( multiphysical ) spatial discretization developed in the previous section . we first prove that under the mesh constraint @xmath262 the fully discrete solution satisfies a discrete energy law which mimics the differential energy laws and . we then derive optimal order error estimates in various norms for the numerical solution , which , as expected , are of the first order in @xmath81 . using the time - stepping scheme of algorithm @xmath59 to discretize we get the following fully discrete finite element method for problem . * compute @xmath263 and @xmath264 by @xmath265 * for @xmath48 , do the following two steps + _ step 1 : _ solve for @xmath266 such that @xmath267 + _ step 2 : _ solve for @xmath268 such that @xmath269 some remarks need to be given before analyzing the algorithm . [ rem3.1 ] ( a ) at each time step , problem in _ step 1 _ of ( ii ) solves a generalized stokes problem with a slip boundary condition for @xmath31 . two conditions in are imposed to ensure the uniqueness of the solution . well - posedness of the stokes problem follows easily with help of the _ inf - sup _ condition . \(b ) @xmath270 is clearly well - defined in _ step 2 _ of ( ii ) . \(c ) the algorithm does not produce @xmath271 ( and @xmath272 ) , which is not needed to execute the algorithm . \(d ) reversing the order of _ step 1 _ and _ step 2 _ in ( ii ) of the above algorithm yields the following alternative algorithm . * compute @xmath263 and @xmath273 by @xmath274 * for @xmath48 , do the following two steps + _ step 1 : _ solve for @xmath268 such that @xmath275 + _ step 2 : _ solve for @xmath266 such that @xmath276 we note that algorithm 2 requires the starting values @xmath277 and @xmath278 , and the latter is generated by solving a generalized stokes problem at @xmath261 . in general , @xmath279 of algorithms 1 and 2 are different . later we will remark that algorithm 2 also provides a convergent scheme . the primary goal of this subsection is to derive a discrete energy law which mimics the pde energy law and the semi - discrete energy law . it turns out that such a discrete energy law only holds if @xmath80 and @xmath81 satisfy the mesh constraint @xmath280 . this mesh constraint is the cost of using the time delay decoupling strategy in . before discussing the stability of the fully discrete algorithm 1 , we first show that the constraints an are consistent with the equations , , and . [ lem3.1 ] let @xmath281 satisfy , , and . define @xmath282 for @xmath283 . then there hold @xmath284 where @xmath285 , and @xmath286 are defined by and taking @xmath287 in yields @xmath288 hence , @xmath289 so holds . immediately follows from setting @xmath290 in and using . as in the time - continuous case , follows from taking @xmath213 in and using the fact that @xmath291 and @xmath117 . finally , is an immediate consequence of the definition @xmath282 , , and . the proof is complete . next lemma establishes an identity which mimics the continuous energy law for the fully discrete solution of algorithm 1 . [ lemma3.1 ] let @xmath281 be a solution of . define @xmath292 for @xmath293 . then there holds the following identity : @xmath294 \\ & \hskip 2 in = \mathcal{j}_h^0-\kappa ( \del t)^2\sum_{n=0}^\ell \bigl ( d_t\nab \tp_h^{n+1 } , \nab p_h^{n+1 } \bigr ) , \nonumber \nonumber\end{aligned}\ ] ] where @xmath295 , \qquad\ell=0,1,2,\cdots\end{aligned}\ ] ] setting @xmath296 in gives @xmath297 applying the difference operator @xmath298 to and followed by taking the test function @xmath299 yield @xmath300 \nonumber \end{aligned}\ ] ] to get an expression for the first term on the right - hand side of , we set @xmath301 in ( after lowering the index from @xmath302 to @xmath303 in the equation ) and using the definition @xmath304 to get @xmath305 , \nab p_h^{n+1 } \bigr ) \\ & = -\kappa \norm{\nab p_h^{n+1}}{l^2}^2 + \kappa \del t \bigl(d_t\nab \tp_h^{n+1 } , \nab p_h^{n+1 } \bigr ) . \nonumber \end{aligned}\ ] ] combining we obtain @xmath306 + \kappa \norm{\nab p_h^{n+1}}{l^2}^2 \\ & \qquad\qquad + \frac{\del t}{2 } \bigl [ \beta \norm{d_t \nab\bu_h^{n+1}}{l^2}^2 + \alpha \norm{d_t q_h^n}{l^2}^2 \bigr ] = -\kappa \del t \bigl(d_t\nab \tp_h^{n+1 } , \nab p_h^{n+1 } \bigr ) \bigr ) . \nonumber\end{aligned}\ ] ] applying the summation operator @xmath307 for any @xmath308 yields @xmath309 \\ & \hskip 2 in = \mathcal{j}_h^0-\kappa ( \del t)^2\sum_{n=0}^\ell \bigl ( d_t\nab \tp_h^{n+1 } , \nab p_h^{n+1 } \bigr ) , \nonumber\end{aligned}\ ] ] where @xmath310 is defined by . hence , holds . the proof is completed . the extra term on the right - hand side of is due to the time lag in which is needed to decouple the original whole system into two sub - systems . this term is smaller if high order time - stepping schemes are used to replace the implicit euler scheme in algorithm @xmath59 . [ thm3.3 ] let @xmath281 be same as in lemma [ lemma3.1 ] . set @xmath311 . then there holds the following discrete energy inequality : @xmath312 \leq \mathcal{j}_h^0,\end{aligned}\ ] ] provided that @xmath313 . by schwarz inequality and the inverse inequality we have @xmath314\\ & \leq \frac{1}{\del t } \bigl [ c_0 ^ 2 h^{-2 } \norm{\tp_h^{n+1}-\tp_h^n}{l^2}^2 + \frac{1}2 \norm{\nab p_h^{n+1}}{l^2}^2 \bigr ] . \no\end{aligned}\ ] ] to bound the first term on the right - hand side , we appeal to the inf - sup condition ( note that @xmath315 ) and to get @xmath316 follows from combining , , and and choosing @xmath317 . the proof is complete . it can be shown that all the results obtained in this subsection for algorithm 1 also hold for algorithm 2 . the main differences are ( a ) the correct " discrete pressure @xmath318 for algorithm 2 is @xmath319 instead of @xmath292 ; ( b ) the correct " energy functional @xmath310 for algorithm 2 is @xmath320 , \qquad\ell=0,1,2,\cdots\ ] ] in this subsection we derive error estimates for the fully discrete algorithm @xmath59 . to the end , we only need to derive the time discretization errors by comparing the solution of with that of because the spatial discretization errors have already been derived in the previous section . introduce notations @xmath321 it is easy to check that @xmath322 [ lem3.2 ] let @xmath323 be generated by the fully discrete algorithm @xmath59 and @xmath324 , and @xmath325 be defined as above . set @xmath326 and @xmath311 . then there holds the following identity : @xmath327 \\ & \quad = \mathcal{e}_h^0-(\del t)^2 \sum_{n=0}^\ell\bigl[\kappa \bigl ( d_t \nab e_{\tp}^{n+1 } , \nab \he_p^{n+1}\bigr ) - \bigl(d_t^2 q_h(t_{n+1 } ) , e_{\tp}^{n+1 } \bigr ) \bigr ] \no\\ & \qquad\qquad + \del t\sum_{n=0}^\ell \bigl(r_h^{n},\he_p^{n+1 } \bigr ) , \no \end{aligned}\ ] ] where @xmath328,\quad \ell=0,1,2,\cdots , \\ r_h^{n } & : = -\frac{1}{\del t } \int_{t_{n-1}}^{t_{n } } ( s - t_{n-1})(q_h)_{tt}(s)\,ds . \label{e3.20a}\end{aligned}\ ] ] first , by taylor s formula and we have @xmath329 then , subtracting from , from , from , and from , from respectively , we get the following error equations : @xmath330 then follows from setting @xmath331 in , @xmath332 in ( after applying the difference operator @xmath298 to the equation ) , @xmath333 in ( after lowering the index from @xmath302 to @xmath303 in the equation ) , adding the resulted equations , and applying the summation operator @xmath334 to the sum . the proof is complete . from the above lemma we then obtain the following error estimate . [ thm3.2 ] let @xmath335 be a ( large ) positive integer and @xmath336 . let the error functions @xmath324 , and @xmath337 be same as in lemma [ lem3.2 ] . assume @xmath81 satisfies the mesh constraint @xmath338 . then there holds error estimate @xmath339 + \bigl ( \del t\sum_{n=0}^{n-1 } \kappa \norm{\nab \he_p^{n+1}}{l^2}^2 \bigr)^{\frac12 } \\ & \quad + \bigl[\sum_{n=1}^{n-1 } \bigl(\beta \norm{\nab(\be_{\bu}^{n+1}-\be_{\bu}^{n})}{l^2}^2 + \alpha\norm{e_q^{n}-e_q^{n-1}}{l^2}^2 \bigr ) \bigr]^{\frac12 } \leq c_3(t;q_h)^{\frac12 } \del t , \no\end{aligned}\ ] ] where @xmath340.\ ] ] to derive the desired error bound , we need to bound each term on the right - hand side of . before doing that , on noting that @xmath341 and @xmath342 and we rewrite as @xmath343 \no \\ & \qquad = -(\del t)^2 \sum_{n=0}^\ell\bigl[\kappa \bigl ( d_t \nab e_{\tp}^{n+1 } , \nab \he_p^{n+1}\bigr ) -\bigl(d_t^2 q_h(t_{n+1 } ) , e_{\tp}^{n+1 } \bigr ) \bigr ] \no\\ & \qquad\qquad + \del t\sum_{n=0}^\ell \bigl(r_h^{n},\he_p^{n+1 } \bigr ) . \no \end{aligned}\ ] ] we now estimate each term on the right - hand side of . first , by schwarz inequality , the inverse inequality , the inf - sup condition , and we have @xmath344 to bound the second term on the right - hand side of , we first use the summation by parts formula and @xmath345 to get @xmath346 we then bound each term on the right - hand side of as follows : @xmath347 @xmath348 finally , we bound the third term on the right - hand side of by @xmath349 where we have used the fact that @xmath350 substituting into and using the assumption that @xmath338 yields @xmath351 \\ & \quad + \del t \sum_{n=1}^\ell \bigl [ \kappa \norm{\nab \he_p^{n+1}}{l^2}^2 + \frac{\del t}{2 } \bigl ( \beta \norm{d_t \nab \be_{\bu}^{n+1}}{l^2}^2 + \alpha \norm{d_t e_q^{n}}{l^2}^2 \bigr ) \bigr ] \no \\ & \hskip 0.8 in \leq 16(\del t)^2 \bigl [ \beta_0^{-2}\beta \norm { ( q_h)_t}{l^2(l^2)}^2 + \kappa^{-1 } \norm { ( q_h)_{tt}}{l^2(h^{-1})}^2 \bigr],\end{aligned}\ ] ] which trivially implies . the proof is complete . [ thm4.2 ] the solution of the fully discrete algorithm 1 satisfies the following error estimates : @xmath352 \\ & \qquad + \bigl(\del t\sum_{n=0}^{n}\kappa\norm{\nab(p(t_n)-p_h^n)}{l^2}^2\bigr)^{\frac12 } \no \\ & \qquad\qquad \leq \bigl [ c_1(t;\bu,\tp , p , q)^{\frac12}\,e^{\frac{t}{2 } } + \widehat{c}_1(t;\bu , p , q)^{\frac12 } \bigr ] h + \widehat{c}_3(t;q_h)^{\frac12 } \del t , \no\end{aligned}\ ] ] provided that @xmath353 , @xmath354 , and @xmath338 . moreover , if , in addition , @xmath355 and @xmath258 , then there also holds @xmath356\\ & \hskip 1 in \leq \bigl [ c_2(t;\bu,\tp , p , q)^{\frac12}\,e^{\frac{t}{2 } } + \widehat{c}_2(t;\bu , q)^{\frac12 } \bigr ] h^2 + \widehat{c}_3(t;q_h)^{\frac12 } \del t. \no\end{aligned}\ ] ] the assertions follow easily from first using the triangle inequality on @xmath357 and then appealing to theorems [ thm3.2 ] and [ thm4.2 ] . \(a ) in light of theorems [ regularity ] and [ regularity_1 ] , the regularity assumptions of theorem [ thm4.2 ] are valid if the domain @xmath43 and datum functions @xmath26 and @xmath54 are sufficient regular . \(b ) it can be shown that all the results proved in this subsection for algorithm 1 still hold for algorithm 2 . the main differences are ( i ) the correct " discrete pressure @xmath318 for algorithm 2 is @xmath319 ; ( i ) the correct " error functional @xmath358 for algorithm 2 is @xmath359 , \qquad\ell=0,1,2,\cdots\ ] ] in this section we present some @xmath360-d numerical experiments to gauge the efficiency of the fully discrete finite element methods developed in this paper . three tests are performed on two different geometries . the gel used in all three tests is the ploy(n - isopropylacrylamide ) ( pnipa ) hydrogel ( cf . @xcite and the references therein ) . the material constants / parameters , which were reported in @xcite , are given as follows : @xmath361 two other material constants / parameters , which were not given in @xcite , are taken as follows in our numerical tests : @xmath362 in addition , we use the following initial condition in all our numerical tests : @xmath363 let @xmath364 . the external force is taken as @xmath365 where @xmath366 denotes the unit ( clockwise ) tangential vector on @xmath24 . note that the compatibility condition @xmath367 is trivially fulfilled . figure [ fig1a ] shows the computational domain , color plot of the force function @xmath368 , and the mesh on which the numerical solution is computed . the mesh consists of @xmath369 elements , and total number of degrees of freedom for the test is @xmath370 . @xmath371 is used in this test . figure [ fig1b ] displays snapshots of the computed solution at three time incidences . each graph contains color plot of the computed pressure @xmath292 and arrow plot of the computed displacement field @xmath372 . the three graphs on the first row are plotted on the computational domain @xmath364 , while three graphs on the second row are respectively deformed shape plots of the three graphs on the first row with @xmath373 times magnification , which shows the deformation of the square gel under the mechanical force @xmath26 on the boundary . as expected , the gel is slightly rotated clockwise and is slightly bent near the top and bottom edges . we also note that the expected conserved quantities are indeed conserved in the computation , their respective values are given as follows : @xmath374 and arrow plots of computed displacement @xmath372 at @xmath375 ( first row ) . deformed shape plots of the graphs on the first row ( second row ) . @xmath376.,title="fig : " ] and arrow plots of computed displacement @xmath372 at @xmath375 ( first row ) . deformed shape plots of the graphs on the first row ( second row ) . @xmath376.,title="fig : " ] and arrow plots of computed displacement @xmath372 at @xmath375 ( first row ) . deformed shape plots of the graphs on the first row ( second row ) . @xmath376.,title="fig : " ] and arrow plots of computed displacement @xmath372 at @xmath375 ( first row ) . deformed shape plots of the graphs on the first row ( second row ) . @xmath376.,title="fig : " ] and arrow plots of computed displacement @xmath372 at @xmath375 ( first row ) . deformed shape plots of the graphs on the first row ( second row ) . @xmath376.,title="fig : " ] and arrow plots of computed displacement @xmath372 at @xmath375 ( first row ) . deformed shape plots of the graphs on the first row ( second row ) . @xmath376.,title="fig : " ] this test is same as test 1 except @xmath377 is changed to @xmath378 so parallel forces of opposite directions are applied at the left and the right boundary of the square gel . clearly , the compatibility condition is satisfied . and arrow plots of computed displacement @xmath372 at @xmath375 ( first row ) . deformed shape plots of the graphs on the first row ( second row ) . @xmath376.,title="fig : " ] and arrow plots of computed displacement @xmath372 at @xmath375 ( first row ) . deformed shape plots of the graphs on the first row ( second row ) . @xmath376.,title="fig : " ] and arrow plots of computed displacement @xmath372 at @xmath375 ( first row ) . deformed shape plots of the graphs on the first row ( second row ) . @xmath376.,title="fig : " ] and arrow plots of computed displacement @xmath372 at @xmath375 ( first row ) . deformed shape plots of the graphs on the first row ( second row ) . @xmath376.,title="fig : " ] and arrow plots of computed displacement @xmath372 at @xmath375 ( first row ) . deformed shape plots of the graphs on the first row ( second row ) . @xmath376.,title="fig : " ] and arrow plots of computed displacement @xmath372 at @xmath375 ( first row ) . deformed shape plots of the graphs on the first row ( second row ) . @xmath376.,title="fig : " ] figure [ fig2b ] is the counterpart of figure [ fig1b ] for test 2 . since parallel forces of equal magnitude but opposite directions are applied on the left and the right boundary of the square gel , so the gel is squeezed horizontally . because of the incompressibility of gel , the total volume must be conserved . as a result , the deformation in the vertical direction is expected . the second row of figure [ fig2b ] precisely shows such a deformation ( with @xmath373 times magnification ) . it should be noted that the swelling dynamics of the gel goes super fast , it reaches the equilibrium in very short time . for the conserved quantities , @xmath122 and @xmath379 are same as those in test 1 , and the other two numbers are given by @xmath380 recall that @xmath381 and @xmath286 depend on the force function @xmath26 . same material parameters / constants and initial condition @xmath54 as in tests 1 and 2 are assumed . however , the computational domain is changed to the following one @xmath382 and the external force function @xmath26 is taken as @xmath383 as in test 2 , the above @xmath26 means that parallel forces of same magnitude but opposite directions are applied at the left half and the right half ellipse ( boundary ) , however , these two forces now collide at two points @xmath384 on the boundary . clearly , the compatibility condition is satisfied . and arrow plots of computed displacement @xmath372 at @xmath385 ( first row ) . deformed shape plots of the graphs on the first row ( second row ) . @xmath386.,title="fig : " ] and arrow plots of computed displacement @xmath372 at @xmath385 ( first row ) . deformed shape plots of the graphs on the first row ( second row ) . @xmath386.,title="fig : " ] and arrow plots of computed displacement @xmath372 at @xmath385 ( first row ) . deformed shape plots of the graphs on the first row ( second row ) . @xmath386.,title="fig : " ] and arrow plots of computed displacement @xmath372 at @xmath385 ( first row ) . deformed shape plots of the graphs on the first row ( second row ) . @xmath386.,title="fig : " ] and arrow plots of computed displacement @xmath372 at @xmath385 ( first row ) . deformed shape plots of the graphs on the first row ( second row ) . @xmath386.,title="fig : " ] and arrow plots of computed displacement @xmath372 at @xmath385 ( first row ) . deformed shape plots of the graphs on the first row ( second row ) . @xmath386.,title="fig : " ] figure [ fig3a ] is the counterpart of figures [ fig1a ] and [ fig2a ] for test 3 . the mesh consists of @xmath387 elements , and total number of degrees of freedom for the test is @xmath388 . a smaller time step @xmath389 is used in this test . figure [ fig3b ] is the counterpart of figures [ fig1b ] and [ fig2b ] for test 3 . the deformations displayed on the second row are magnified by @xmath390 times instead of @xmath373 times as in figures [ fig1b ] and [ fig2b ] . due to curve boundary , the applied parallel forces of same magnitude but opposite directions collide at the boundary points @xmath384 . it is expected that the gel should buckle around those two points , which is clearly seen in the plots on the second row of figure [ fig3b ] . we also note that because relatively bigger forces are applied on the gel , the swelling dynamics of the gel goes even faster , hence , reaches the equilibrium quicker . this is the main reason to use a smaller time step for the simulation . the first author would like to thank professor masao doi of tokyo university for introducing the gel swelling dynamic model to the author , and for his many stimulating discussions at i m a of university of minnesota , where they both were long - term visitors in fall 2004 .

the uninterrupted 7-day _ asca _ observations of the tev blazar mrk 421 in 1998 have clearly revealed that x ray flares occur repeatedly . in this paper , we present the results of the time - resolved spectral analysis of the combined data taken by _ asca _ , _ rxte _ , _ bepposax _ , and _ euve_. in this object and in many other tev blazars the precise measurement of the shape of the x ray spectrum , which reflects the high energy portion of the synchrotron component , is crucial in determining the high energy cutoff of the accelerated electrons in the jet . thanks to the simultaneous broadband coverage , we measured the 0.125 kev spectrum resolved on time scales as short as several hours , providing a great opportunity to investigate the detailed spectral evolution at the flares . by analyzing the time subdivided observations , we parameterize the evolution of the synchrotron peak , where the radiation power dominates , by fitting the combined spectra with a quadratic form ( where the flux at the energy obeys ) . in this case , we show that there is an overall trend that the peak energy and peak flux both increase or decrease together . the relation of the two parameters is best described as for the 1998 campaign . similar results were derived for the 1997 observation , while the relation gave a smaller index when included both 1997 and 1998 data . on the other hand , we show that this relation , and also the detailed spectral variations , differ from flare to flare within the 1998 campaign . we suggest that the observed features are consistent with the idea that flares are due to a appearance of a new spectral component . with the availability of the simultaneous tev data , we also show that there exists a clear correlation between the synchrotron peak flux and the tev flux .

astro-ph0310592

the uninterrupted 7-day _ asca _ observation of the tev blazar mrk 421 in 1998 has revealed that day - scale x ray flares seen in previous observations were probably unresolved superpositions of many smaller flares @xcite . the nearly continuous observation allowed not only the possibility to track the individual flares entirely from the rise to decay , but it also enabled quantitative statistical tests of the time series by employing the power spectrum or the structure function @xcite . the main characteristic of blazars is their high flux observed from radio to @xmath7rays coupled with strong variability and strong polarization . these properties are now successfully explained by the scenario where blazars are active galactic nuclei ( agn ) , possessing jets aligned close to the line of sight , and accordingly the doppler - boosted non - thermal emission from the jet dominates other emission components ( e.g. blandford & knigl 1979 ; urry & padovani 1995 ) . this is what makes blazars critical in understanding jets in agn . the broadband spectra of blazars consist of two peaks , one in the radio to optical uv range ( and in some cases , reaching to the x ray band ) , and the other in the hard x ray to @xmath7ray region . the high polarization of the radio to optical emission suggests that the lower energy peak is produced via the synchrotron process by relativistic electrons in the jet . the higher energy peak is believed to be due to compton up - scattering of seed photons by the same population of relativistic electrons . several possibilities exist for the source of the seed photons ; these can be the synchrotron photons internal to the jet @xcite , but also external , such as from the broad emission line clouds @xcite or from the accretion disk @xcite . the blazars with peak synchrotron output in the x ray range also emit strongly in the @xmath7ray energies , and the brightest of those have been detected in the tev range with ground - based cherenkov arrays . these are the so - called `` tev blazars . '' in tev blazars , variability of the synchrotron flux is measured to be the strongest and most rapid in the x ray band , and thus it provides the best opportunity to study the electrons that are accelerated to the highest energies . in particular , the synchrotron peak is a very important observable in two aspects : first because the flux at the peak represents the total emitted power from the blazar , and secondly because the peak frequency reflects the maximum energy of radiating particles gained in the acceleration process . mrk 421 is among the closest known blazars , at redshift of 0.031 , and was the first blazar ( and also the first extragalactic source ) discovered to be a tev emitter . it was first detected as a weak source by egret @xcite , and 9 months later , _ whipple_detected a clear signal from this object between 0.5 and 1.5 tev @xcite . flux variability on various time scales has been observed , including a very short flare with a duration of @xmath81 hour @xcite . ever since , mrk 421 has been repeatedly confirmed to be a tev source by various ground - based telescopes . it has also been one of the most studied blazars , and have been a target of several multi - wavelength campaigns @xcite . the multi - wavelength campaign of mrk 421 in 1998 was one of the first opportunities to observe a blazar in the tev range using several telescopes located in different locations in the world , so as to have as continuous coverage as possible @xcite . observations in other frequencies included x ray observations by _ asca _ , _ rxte _ @xcite , and _ bepposa@xmath0x _ ( maraschi et al . 1999 ; fossati et al . 2000a , b ) , euv observations by _ euve _ , optical observation with bvri filters organized by the webt collaboration ( http://www.to.astro.it/blazars/webt/ ) , and radio observations at the metshovi radio observatory . in this paper , we present the results of the spectral analysis of x ray and euv data during the 1998 april campaign . in particular , we estimate the location of the synchrotron peak in the @xmath9 spectrum . so far , quantitative studies of the variation of the synchrotron peak have been conducted only for the two brightest blazars , mrk 421 and mrk 501 . the largest variation of the synchrotron peak energy was observed in mrk 501 by _ bepposa@xmath0x _ , where the peak energy shifted @xmath82 orders of magnitude @xcite . collecting data from different epochs ( separated by as long as years ) , @xcite have shown the relation of the form @xmath10 with @xmath112 . a similar analysis was done by @xcite for the mrk 421 data obtained by _ bepposa@xmath0x _ in 1997 and 1998 , which showed a relation of @xmath12 with n=0.55 . in this paper we describe a continuous 7-day variation of the synchrotron peak , which allows us to investigate the dynamical change of the synchrotron spectrum during the flares . we first describe the observations and data analysis in 2 . the results of spectral analysis are described in 3 ; a summary of the results and a discussion of our findings with the emphasis on the temporal evolution of the spectrum are presented in 4 . the week - long multi - frequency campaign of mrk 421 was carried out in april 1998 @xcite . the goal of this paper is to study the evolution of the high energy synchrotron spectrum as a function of time during the campaign , and also the correlation with the tev flux . this requires the best possible spectral coverage in the x ray range , for which we assembled all available data collected during the campaign , including : pointings by x ray observatories _ asca _ , _ rxte _ , and _ bepposa@xmath0x _ , as well as the euv observations by _ mrk 421 was observed with _ asca _ for 7 days continuously ( except for the earth occultation ) during 1998 april 23.97 30.8 ut ( pi : t. takahashi ) . the _ asca _ detectors two siss ( solid - state imaging spectrometers ; burke et al . 1991 ; yamashita et al . 1997 ) and two giss ( gas imaging spectrometers ; ohashi et al . 1996 ) were in operation . we refer to @xcite to the details of the observation modes and data reduction . the obtained spectra were rebinned such that all bins have the same statistics . since late 1994 , the efficiency of both sis detectors below @xmath81 kev has been decreasing over time . the problem is believed to be caused by the increased residual dark current level , and also by the decrease in the charge transfer efficiency , although effort is still underway to model the effects . while it has been reported that it may be possible to parameterize the degradation by an additional absorption column as a function of time @xcite , since our observation concerns the precise shape of the continuum spectrum , we will use only the data obtained by the gis for spectral analysis . ( note that the x ray spectra in figure 2 of @xcite show the data obtained by sis . for this , there is a too strong decrease of inferred flux at the lower energy , which is due to instrumental effects described above . ) the _ rxte _ observations ( pi : g. madejski ) were mainly coordinated to coincide with the _ asca _ observation period , but were conducted mostly only during the low - background orbits . the _ rxte _ extends the _ asca _ bandpass to a higher energy with the instruments pca ( proportional counter array ; 260 kev ) and hexte ( high energy x ray timing experiment ; 15200 kev ) . however , because both the flux from the source and the sensitivities of the detectors decrease towards higher energies , in the analysis below we use only the data from the pca , in the energy range of 325 kev . all data reduction was performed using the heasoft software packages , based on the rex script , provided by the _ rxte _ guest observer facility ( gof ) . the screening criteria excluded the data when the elevation angle from the earth s limb was lower than 10 , and the data acquired during the saa passages . since the pcus are activated during the saa passage , we used the data only after when the background level dropped to the quiescent level and became stable , which is typically 30 minutes after the peak of a saa passage . for the best signal - to - noise , we selected events only from the top xenon layer , and excluded events from other gas layers . for the maximum consistency among the observations , we used only the three detectors which were turned on for all the observations ( known as pcu0 , pcu1 , and pcu2 ) . since the pca is not an imaging detector , the blank - sky observations must be used to estimate the background spectrum . for this , we used the background models provided by the _ rxte _ gof to evaluate the background during each observation . we note that the residual background contamination of the data within the energy bandpass selected by us was always significantly lower than the source counts . the mean net _ rxte _ pca count rate from mrk 421 in the 10 - 20 kev range for 3 pcus as above was at least @xmath13 counts s@xmath14 while the total residuals from blank - sky in this energy range @xmath15 the fluctuations of the cosmic x - ray background were lower than 0.2 counts s@xmath14 ( see heasarc web page regarding the _ rxte _ pca background ) , providing the confidence that an imprecise background subtraction did not affect our results . the _ bepposa@xmath0x _ observations ( pi : l. chiappetti ) started 3 days before the _ asca _ observations ( maraschi et al . 1999 ; fossati et al . 2000a , b ) . we use the data obtained by the lecs ( low energy concentrator spectrometer ; 0.110 kev ) , which extends the bandpass to lower energy , and also the mecs ( medium energy concentrator spectrometer ; 1.310 kev ) . the data analysis was based on the linearized event files , together with the appropriate background event files and response matrices , provided by the _ bepposa@xmath0x _ science data center ( sdc ) . data reduction was performed using the standard heasoft software , xselect . we extracted the source photons from a circular region centered at the source with a radius of 8 arcmin for both lecs and mecs . for the background photons , we constructed the spectra by extracting the photons from the same detector regions using blank - sky observations . each spectrum was rebinned using the grouping templates provided by the _ bepposa@xmath0x _ sdc , and we limited the energy range to 0.124.0 kev for lecs , and 1.6510.5 kev for mecs , as suggested in the @xcite . the _ euve _ observations covered the entire campaign of mrk 421 from april 19.4 to may 1.2 ( pi : t. takahashi ) . the source photons were extracted in a 12 arcmin aperture . in order to estimate the absolute flux in the euv range , we normalized it to be consistent with the simultaneous data obtained by the other observatories . given that the euv spectrum is difficult to determine within the rather limited bandpass , we used only the observed integrated count rate ( cf . marshall , fruscione & carone 1995 ) . for this , we used the period where the observations by all observatories overlapped ( mjd 50927.03 - 50927.34 ; [ sec : combine ] ) . during this period , the average _ euve _ count rate was [email protected] cts / s . at the same time , the 0.130.18 kev flux derived from the combined lecs , mecs , gis , and pca spectrum gives @xmath17 erg @xmath18 s@xmath14 , where the error represents the statistical error . an additional uncertainty arises from the fact that the difference in the spectral shape is not taken in to account . this can be estimated from the results in @xcite . taking @xmath19 @xmath18 for mrk 421 , assuming that the spectrum index varies from 0.5 to 1.0 , the flux density would vary from 13631394 @xmath20jy for 1 cts / s in the _ euve _ range . this gives an additional 2.5% error as a systematic error . in the following analysis , we use the above value for scaling the observed _ euve _ count rate to flux . thus , @xmath21 where @xmath22 is the _ euve _ count rate . during april 24 , observations by all 3 x ray satellites , _ asca _ , _ bepposa@xmath0x _ , _ rxte _ , and also _ euve _ happened to exactly overlap continuously for 26.5 ksec ( mjd 50927.03 - 50927.34 ) . this provides us with high quality data over a very broad energy coverage , spanning more than 2 decades , from 0.12 to 25 kev . the net exposure for each instrument during the overlapping 26.5 ks period were 9.1 , 13.3 , 3.6 , and 5.5 ksec for gis , pca , lecs , mecs , respectively . we first fitted the spectrum for each instrument with a power - law model , with a low energy absorption fixed to the galactic value , @xmath23 @xmath18 @xcite . the results are shown in the top half of table [ tbl : fit ] , suggesting that the fit is not adequate for any of the individual instruments . this certainly suggests that a more complex model , rather than just a simple power - law function , is required to describe the spectrum . the shape of the synchrotron spectrum is determined from the energy distribution of the radiating electrons , which cuts off at the high energy representing the acceleration limit . as described in the introduction , the high energy end of the synchrotron spectrum in tev blazars is located in the x ray range ; this is most likely the reason that the high - quality x ray data can not be described by a single power - law . given the curved spectrum for each instrument , we fitted the combined x ray spectrum with various functions to model the shape of the observed curvature . we note that our aim is not to develop the shape of the curvature from first principles , but rather , to reproduce the observed data . for this reason , we considered the following models for comparison : ( 1 ) a broken power - law , ( 2 ) a curved power - law , ( 3 ) a power - law with an exponential cutoff , and ( 4 ) a quadratic function in @xmath24-@xmath25@xmath9 plane . the curved power - law that we define below is similar to the broken power - law model , but the change in the spectral index is not discrete . the form attempted by us is @xmath26 . the form for the exponentially cutoff power law is @xmath27 , and the quadratic function is described as @xmath28 . since there are slight differences in the cross - calibration between the instruments , we allow the normalization to vary for each instrument . for all results below , the flux will be normalized to the gis value , unless otherwise noted . the @xmath29 is fixed to the galactic value . by fitting the spectrum with the models above , we first found that the power - law with an exponential cutoff is not a good representation of the data : the model cutoff is too sharp at high energies . the broken power - law model is also not acceptable , and the curved power - law is significantly better . we also find that the quadratic function fits the spectrum surprisingly well . the fitted spectrum and the residuals to the quadratic model are shown in figure [ fig : spec ] , where all lecs , mecs , gis , and pca spectra are shown to be well described with this single function . the results of the spectral fits by the 4 models are summarized in in the bottom half of table [ tbl : fit ] . the _ euve _ observation covered the whole campaign ( april 19.4 to may 1.2 ) , and thus completely overlapped our 7-day _ asca _ observation . this adds a data point at 0.13 - 0.18 kev in addition to the _ asca _ spectrum in the 0.77 kev range , providing us the best opportunity to study the continuous spectral variability . in fact this is the longest continuous monitoring of this object so far spanning over the 0.137 kev range . we divided the whole observation into shorter segments with each having a duration of 10 ksec . in order to combine the gis and _ euve _ data , the gis count rate spectra must first be converted to the source flux spectra as the _ euve _ count rate data converts directly into flux ( 2.4 ) . for this , we first fit the gis spectrum with a curved power - law model with galactic absorption . this gives the source spectrum in the gis energy range . note that this procedure depends on the selected model , but long as it describes the spectrum well , this should introduce the smallest bias . we then combine this with the _ euve _ data , which gives the combined source spectrum . in fitting the combined _ euve_/gis spectra , we first started with a single power - law model . in similarity to the results in 3.1 , none of the 56 spectra could be fit with a single power - law function , with their average @xmath30 of 7.7 , implying the curvature of the spectra for all data sets . given that a quadratic model fits the spectrum equally well as the curved power - law model , but with fewer parameters ( see 3.1 ) , we will assume the quadratic model for all data sets hereafter . in fact , because of the energy gap between the _ euve _ and gis , the fit with a curved power - law model resulted in having multiple minima in the @xmath31 plane ( since the parameters are correlated ) , and the exact value of the best - fit parameters depended highly on the initial value . we checked whether the fit was adequate for all data segments . the @xmath30 for each fit ( shown in the bottom panel in figure [ fig : lcpeak ] ) ranged from 0.71 to 1.45 . the number of degrees of freedom ranges from 60 to 187 , which gave @xmath325% for 53 out of the total of 56 data points . for the other 3 data points ( corresponding to times 230240 , 270280 , and 450460 ksec ) , we found several line or edge - like structures in the spectra , which resulted in the increase of @xmath30 . although it is not clear whether this is due to some systematic effect or if it is a true property of the source , since our aim is to model the continuum spectrum correctly , we excluded the energy ranges where those features are present . this improved the fit significantly , which is shown as the crosses . with this , all the 56 spectra were well described with the quadratic function . the time evolution of the derived parameters in the fit is shown in the top 3 panels in figure [ fig : lcpeak ] ; the peak flux , peak energy , and the curvature parameter @xmath33 ( see definition in table [ tbl : fit ] ) . the peak energy is observed to shift between 0.5 kev and 1.2 kev , in which a general trend is found such that the peak energy is higher when the peak flux is higher . this is demonstrated as the filled circles in figure [ fig : fpep - all ] . the fit results are listed in table [ tbl : lcpeak ] . we do note that the actual validity of using a quadratic function in the entire 0.125 kev range was checked in detail ( including an examination of trends in residuals ) only for the epoch when the overlapping _ bepposa@xmath0x _ , _ asca _ , and _ rxte _ observations were available . since we measured only the _ euve _ flux below 0.7 kev in the combined _ euve_/gis fit , the derived peak location can depend on the model function we assume . for instance if @xmath34 function was allowed , the peak can be located at any energy within the gap . for this , we emphasize that the derived peak location is the estimated peak , assuming that the spectrum gap can be extrapolated with the same quadratic function , and also that the error bar describes only the statistical error . we also analyzed the _ bepposa@xmath0x _ data during the campaign . furthermore , in order to make a comparison with data obtained from another campaign , we also analyzed the _ bepposa@xmath0x _ data obtained in 1997 ( pi : g. vacanti , l. chiappetti ) . we divided the observation into segments of the same 10 ks duration , and fit the combined lecs / mecs spectra with a quadratic function . bepposa@xmath0x _ data , we also fit the spectra with a curved power - law model for comparison . the derived peak flux and peak energy from both models are listed in table [ tbl : lcpeak_sax ] . in fitting the combined lecs / mecs spectra , the normalizations are allowed to vary independently , and the values in table [ tbl : lcpeak_sax ] are normalized to the mecs2 value . the fitting results show that although the spectral fit parameter is one less in the quadratic function , both models equally represent the observed spectra . in contrast to the combined _ euve_/gis data , there is no energy gap in the _ bepposa@xmath0x _ data , which provides support that our assumed quadratic form is appropriate in estimating the peak location in the _ euve_/gis case . the derived peak flux and peak energy using both functions are similar , but we also note that the quadratic function gives a slightly lower peak energy than the curved power - law function . the derived peak flux and peak energy assuming the quadratic model for the _ bepposa@xmath0x _ observations are shown in figure [ fig : fpep - all ] as the open circles ( for 1998 ) and squares ( for 1997 ) . note that the _ bepposa@xmath0x _ data points in figure [ fig : fpep - all ] are multiplied by a normalization factor of 1.07 , which we adopted from the combined fit in 3.1 it can be seen that the results during the 1998 campaign lie on a similar slope as the results derived from the _ euve_/gis data during the same campaign . the relation of peak flux @xmath5 and peak energy @xmath4 for the _ euve_/gis data set can be best described with a power - law function of @xmath35 , where @[email protected] ( or @[email protected] in form @xmath38 ; see description below ) . by including also the _ bepposa@xmath0x _ data points during 1998 , the index becomes @[email protected] ( @[email protected] ) . it can be seen that the data from 1997 lie on a somewhat different slope . the relation derived from the 1997 data _ alone _ gives @[email protected] ( @[email protected] ) , slightly steeper than the index derived from the 1998 campaign . fitting all data together we obtain @[email protected] ( @[email protected] ) . all errors correspond to a 1@xmath39 error . a similar analysis was performed by @xcite for the 1997 and 1998 _ bepposa@xmath0x _ data . @xcite note a tight relationship between the measured integrated 0.1 - 10 kev flux @xmath40 and the peak energy @xmath4 such that @xmath38 with @[email protected] , similar to the index derived from our results including all data sets . for a direct comparison , we fit simultaneously both 1997 and 1998 for _ bepposa@xmath0x _ data points only . this gave @[email protected] ( @[email protected] ) . the estimated @xmath37 is slightly smaller than reported in @xcite , which might be affected by different subdividing of the observations . nonetheless , it is worth noting that the slope derived from the combined data appears be softer ( smaller @xmath36 or @xmath37 ) than each epoch considered alone . furthermore , the relation appeared to also vary between individual flares during the 1998 campaign . in some flares the peak energy clearly increases together with flux , while the correlation is weaker in others . this can be seen clearly in figure [ fig : fpep - each ] , which shows the evolution of the synchrotron peak in each flare . the peak energy has a trend to shift to the higher energy and then return to the initial value together with the flux , but there is no common slope . in the previous section , we described the spectral variations as the evolution of the location of the peak in the @xmath9 spectrum . here , we take advantage of the wide energy range coverage afforded by the use of by _ euve _ , _ asca _ , and _ rxte _ data to investigate spectral variations in more detail . given that the _ rxte _ observations happen mostly during the flare minima rather than at their peaks ( see figure 1 in takahashi et al . 2000 ) , here we focus on the beginning ( rising portion ) of the flares . we divided the observations into 10 ksec segments , similar to the previous section . for each segment , the gis and pca spectra are fitted simultaneously with a curved power - law function with an additional constant factor for the normalization . the derived source spectra in the gis and pca range are combined with the _ euve _ data point , which gives the total source spectrum in the energy range of @xmath80.125 kev . in the top panels of figures [ fig : vfvratio8][fig : vfvratio34 ] , we show the evolution of the @xmath9 spectrum derived in this manner for the 2nd , 4th , and 5th flare ( starting from 80 ksec , 260 ksec , and 340 ksec , respectively , as defined in figure [ fig : lcpeak ] ) . note that the spectra are further rebinned for this figure . for each flare , we also plot a subsequent spectrum , but those only include the _ euve _ and gis data . to see the evolution more clearly , we calculated the ratio of the rebinned spectra by dividing these `` subsequent '' spectra by the spectrum measured at the onset of a given flare ( integrated over 8090 , 260270 , 340350 ksec , respectively ) . this is shown in the bottom panels of figures [ fig : vfvratio8][fig : vfvratio34 ] . interestingly , these three flares all showed different features . it can be seen that flare 2 starts with a hardening at the very highest energy range with a concave shape in the ratio . as time progresses , the whole energy range becomes dominated by a new spectral slope . in contrast to this , in flare 4 the lower energy photons appear to precede the total flux rise . flare 5 was similar to flare 2 , but all energy bands increase together , with a larger amplitude at higher energies . we also studied the spectral variation for other flares , but a detailed investigation was rather difficult without a denser temporal coverage by _ rxte_. since x ray and tev energies both reflect the electrons accelerated to the highest energies in tev blazars , simultaneous x ray and tev observations provide an important tool to probe the emission region of the jet ( e.g. , krawczynski et al . 2002 , and references therein ) . on the other hand , since blazars are highly variable on short time scales , exactly simultaneous observations are essential . thanks to the relatively high brightness of the source during the 1998 campaign , it was possible to generate a significant measurement for the separate tev pointings every night ( aharonian et al . 1999 ; takahashi et al . 2000 ) . here we provide the x ray spectrum from the exactly simultaneous time regions with the tev observations . for each tev observation by _ @xcite and _ whipple _ @xcite , we collected all existing x ray data during the tev exposure , from the _ asca _ , _ rxte _ , and _ euve _ observations ( the _ cat _ observations mostly overlap with the _ hegra _ observations ) . in particular , all _ whipple _ pointings during the _ asca _ observations happened to coincide with the _ rxte _ observations , providing an x ray spectrum ranging from 0.125 kev , simultaneous with the tev observation . we generated the combined x ray spectrum using the same procedures as described in the previous sections . we then fitted the x ray spectrum for each period assuming a quadratic function . the x ray spectra during the _ hegra _ observations ( consisting of the _ euve _ and _ asca _ data ) are shown in figure [ fig : vfv_x](a ) . the spectrum for each observation is shown as different symbols , and the dashed line shows the modeled quadratic function derived via the fit . the x ray spectra during the _ whipple _ observations , which consist of the _ euve _ , _ asca _ and _ rxte _ data are shown in figure [ fig : vfv_x](b ) . note that the spectra are further rebinned in the figures . while each integration time is as short as several hours , it can be seen that the curved shape of the high energy peak of the synchrotron spectrum is clearly resolved , and it is evident that the peak is located in the x ray range . the results of fits to each x ray spectrum concurrent with the tev pointings are summarized in table [ tbl : xtev ] . we also list the simultaneous tev fluxes , in which the _ hegra _ data were taken by @xcite . in figure [ fig : xtev ] , we show the correlation of the derived location of the synchrotron peak flux with the tev flux . since the sensitive energy ranges of the two tev telescopes are slightly different , we used the integrated flux in crab units . a good correlation can be seen , showing that the tev flux is higher when the synchrotron peak flux is higher . a best fit to the relation with a power - law function gives @xmath41 with @[email protected] . we have performed detailed analysis of the combined _ euve _ , _ asca _ , _ bepposa@xmath0x _ , and _ rxte _ data collected during the long look campaign of mrk 421 in 1998 , resulting in the measurement of the time - resolved spectrum in a broad energy range of 0.125 kev in segments as short as several ksec . these are among the highest quality spectra with regard to both photon statistics and energy coverage so far for any blazar , providing the precise spectral shape at the highest energy end of the synchrotron spectrum . we have shown that both curved power - law function and quadratic function in @xmath24@xmath25@xmath9 space reproduce the combined _ euve _ , _ bepposa@xmath0x _ , _ asca _ , and _ rxte _ spectrum . for the epochs where _ bepposa@xmath0x _ data do not exist , we assumed that the spectral gap between @xmath42 and @xmath43 can be extrapolated with a quadratic function , and we showed that the energy and the flux at the synchrotron peak in the @xmath9 spectrum are correlated , showing an overall trend of a higher synchrotron peak energy for higher peak flux , but the details of this relation differed from flare to flare . in particular , within the three flare beginnings , one flare started from a rise in the higher energy while another started from a rise in the lower energy . the relative amplitude of the rise in different energies also differed from flare to flare . an interesting result was shown from the 2nd flare , where it was indicated that this flare started with a hardening at the higher energy . this feature is difficult to describe by a simple cooling or acceleration of a single electron distribution ( see e.g. kataoka 2000 ) , and thus this concave curvature indicates that two different spectral components were likely to co - exist at the beginning of the flare . we remark that such a behavior is observed in only one flare , but the importance is that we observed a flare that could not be explained by a single electron distribution . this suggests an appearance of a new component , which generated the observed flare . recently , many theoretical studies has been performed to model the energy dependence , or time lags observed in the day - scale flares in blazars . one approach considers models with a homogeneous emission region @xcite . for instance , @xcite have suggested that a sudden increase in the maximum energy of the accelerated electrons could result in the x ray and tev flare observed in mrk 421 . similar results were shown by @xcite , in reproducing the soft - lag observed in pks 2155304 . however , most of these models assume a change in some parameter in the emission region ( such as the magnetic field , maximum energy of electrons , number density of electrons , etc . ) . the observational results for mrk 421 described above lead us to suggest a different scenario . an alternative scenario has each flare forming as a result of a separate electron distribution . this was suggested in modeling the high state sed of mrk 501 @xcite . they applied the one - zone homogeneous synchrotron self - compton model to the sed from april 1997 , and concluded that a single electron distribution is insufficient to reproduce the observed synchrotron spectrum . this scenario could work for instance when the jet is emitted intermittently from the central engine , and several separate regions generate different emission components . another viable scenario can be provided by an internal shock model , where the light curve results from a superposition of many flares due to the collisions of shells , which may occur when a faster shell catches up to a slower shell ( ghisellini 2001 , spada et al . 2001 ; sikora et al . 2001 ) in fact , @xcite has shown from simulations that the internal shock model can naturally explain various variability properties observed in tev blazars . one observational fact is that the x ray flares always appear to lie on top of an underlying offset - like component . in the internal shock model , flares can be considered as arising from collisions of shells that had the largest difference in the initial velocity . all other collisions generate smaller amplitude , longer flares , which pile up to generate the offset component . in this case , it was shown that the flares will tend to have a higher synchrotron peak energy as compared with that of the offset component . assuming that flares are due to an emergence of a new component with a higher synchrotron peak energy than the pre - existing component , the synchrotron peak energy in the observed spectrum would appear as shifting to a higher energy . it is interesting to note that our observations are consistent with this , with the overall trend of the synchrotron peak energy being higher during flares . furthermore in this case , the spectral evolution would depend on the relation of the new and pre - existing spectra . this would naturally explain the different spectral evolution among different flares . finally , we remark again that due to the data gap between the _ euve _ and _ asca _ data in our observation , the results concerning the spectrum peak are somewhat tentative . because of the rapid variability of the source in short time scales , the effective area of the instrument is most important . long - look observations by the new observatories such as @xmath44-@xmath45 should provide data allowing a more detailed analysis of the flaring mechanism , leading to the dynamics of the accelerated electrons in blazar jets . we thank the anonymous referee for numbers of constructive comments to improve the paper . support for this work was provided by the fellowship of japan society for promotion of science for young scientists , and by the department of energy contract to slac no . de - ac3 - 76sf00515 . we also acknowledge the nasa chandra grants via sao grant no . go1 - 2113x . aharonian , f. a. et al . 1999 , , 350 , 757 burke , b. e. , et al . 1991 , ieee trans . ed-38 , 1069 chiaberge , m. , & ghisellini , g. 1999 , , 306 , 551 dermer , c. d. , schlickeiser , r. , & mastichiadis , a. 1992 , , 256 , l27 dermer , c. d. , & schlickeiser , r. 1993 , , 416 , 458 elvis , m. , wilkes , b. j. , & lockman , f. j. 1989 , , 97 , 777 fossati , g. et al . 2000a , , 541 , 153 fossati , g. et al . 2000b , , 541 , 166 gaidos , j. a. , et al . 1996 , nature , 383 , 319 ghisellini , g. , & maraschi , l. 1989 , , 340 , 181 ghisellini , g. 2001 , asp conf . ser . 227 : blazar demographics and physics , 85 jones , t. w. , odell , s. l. , & stein , w. a. 1974 , , 188 , 353 kataoka , j. , et al . 1999 , , 514 , 138 kataoka , j. , takahashi , t. , makino , f. , madejski , g. m. , tashiro , m. , urry , c. m. , & kubo , h. 2000 , , 528 , 243 kataoka , j. , ph.d thesis , university of tokyo kataoka , j. et al . 2001 , , 560 , 659 krawczynski , h. , coppi , p. s. , & aharonian , f. 2002 , , 336 , 721 li , h. , & kusunose , m. 2000 , , 536 , 729 lin , y. c. et al . 1992 , , 401 , l61 macomb , d. j. , et al . 1995 , , 449 , l99 maraschi , l. et al . 1999 , , 526 , l8 marshall , h. l. , fruscione , a. , carone , t. e. 1995 , 439 , 90 mastichiadis , a. & kirk , j. g. 1997 , , 320 , 19 ohashi , t. , et al . 1996 , , 48 , 157 petry , d. et al . 1996 , , 311 , l13 pian , e. , et al . 1998 , , 492 , l17 piron , f. et al . 2001 , , 374 , 895 punch , m. , et al . 1992 , nature , 358 , 477 sikora , m. , begelman , m. c. , & rees , m. j. 1994 , , 421 , 153 sikora , m. , ba . zejowski , m. , begelman , m. c. , & moderski , r. 2001 , , 561 , 1154 spada , m. , ghisellini , g. , lazzati , d. , & celotti , a. 2001 , , 325 , 1559 takahashi , t. , et al . 2000 , apj , 542 l105 tanihata , c. et al . 2001 , , 563 , 569 tanihata , c. , takahashi , t. , kataoka , j. , & madejski , g. m. 2003 , , 584 , 153 tavecchio , f. et al . 2001 , , 554 , 725 fiore , f. , guainazzi , m. , grandi , p. 1999 , _ cookbook for _ bepposa@xmath0x _ nfi spectral analysis , ftp://www.sdc.asi.it/pub/sax/doc/software_docs/saxabc_v1.2.ps.gz_ urry , c. m. , & padovani , p. 1995 , , 107 , 803 yamashita , a. , et al . 1997 , ieee trans nucl . , 44 , 847 yaqoob , t. , & _ asca _ team , _ asca _ gof calibration memo [ asca - cal-00 - 06 - 01.v1.0(06/05/00 ) ] ll ccc ccc c lecs ( 0.12 - 4 ) & pl & [email protected] & -&-&-&-&- & 2.85 ( 83 ) + mecs ( 1.65 - 10.5 ) & pl & [email protected] & -&-&-&-&- & 1.28 ( 193 ) + gis ( 0.7 - 7 ) & pl & [email protected] & -&-&-&-&- & 1.29 ( 124 ) + pca ( 3 - 25 ) & pl & [email protected] & -&-&-&-&- & 2.63 ( 50 ) + lecs , mecs , gis , pca & broken pl & [email protected] & [email protected] & [email protected] & -&-&- & 1.86 ( 451 ) + lecs , mecs , gis , pca & curved pl & [email protected] & [email protected] & [email protected] & -&-&- & 1.08 ( 451 ) + lecs , mecs , gis , pca & cutoff pl & [email protected] & - & [email protected] & -&-&- & 1.74 ( 452 ) + lecs , mecs , gis , pca & quadratic & -&-&- & [email protected] & 217@xmath163 & [email protected] & 1.07 ( 452 ) + c| cccl 1 & [email protected] & 230@xmath162 & [email protected] & 1.06 ( 66 ) + 2 & [email protected] & 244@xmath163 & [email protected] & 0.95 ( 85 ) + 3 & [email protected] & 266@xmath162 & [email protected] & 0.95 ( 122 ) + 4 & [email protected] & 280@xmath162 & [email protected] & 1.22 ( 106 ) + 5 & [email protected] & 331@xmath162 & [email protected] & 1.08 ( 116 ) + 6 & [email protected] & 372@xmath162 & [email protected] & 0.97 ( 137 ) + 7 & [email protected] & 377@xmath165 & [email protected] & 0.89 ( 145 ) + 8 & [email protected] & 323@xmath164 & [email protected] & 1.20 ( 87 ) + 9 & [email protected] & 277@xmath163 & [email protected] & 0.80 ( 82 ) + 10 & [email protected] & 270@xmath162 & [email protected] & 0.92 ( 79 ) + 11 & [email protected] & 278@xmath162 & [email protected] & 1.10 ( 137 ) + 12 & [email protected] & 307@xmath162 & [email protected] & 1.04 ( 126 ) + 13 & [email protected] & 337@xmath162 & [email protected] & 0.97 ( 125 ) + 14 & [email protected] & 358@xmath162 & [email protected] & 1.01 ( 154 ) + 15 & [email protected] & 324@xmath163 & [email protected] & 0.73 ( 97 ) + 16 & [email protected] & 303@xmath162 & [email protected] & 0.89 ( 87 ) + 18 & [email protected] & 333@xmath162 & [email protected] & 0.93 ( 98 ) + 19 & [email protected] & 364@xmath162 & [email protected] & 0.92 ( 173 ) + 20 & [email protected] & 346@xmath162 & [email protected] & 1.16 ( 144 ) + 21 & [email protected] & 337@xmath162 & [email protected] & 1.04 ( 132 ) + 22 & [email protected] & 391@xmath162 & [email protected] & 1.14 ( 169 ) + 23 & [email protected] & 458@xmath162 & [email protected] & 0.97 ( 184 ) + 24 & [email protected] & 423@xmath163 & [email protected] & 1.21 ( 119 ) + 25 & [email protected] & 379@xmath163 & [email protected] & 0.85 ( 113 ) + 26 & [email protected] & 366@xmath162 & [email protected] & 0.88 ( 95 ) + 27 & [email protected] & 336@xmath162 & [email protected] & 1.14 ( 153 ) + 28 & [email protected] & 397@xmath163 & [email protected] & 1.09 ( 146 ) + 29 & [email protected] & 404@xmath163 & [email protected] & 0.85 ( 125 ) + 30 & [email protected] & 423@xmath162 & [email protected] & 1.08 ( 176 ) + 31 & [email protected] & 442@xmath162 & [email protected] & 0.88 ( 181 ) + 32 & [email protected] & 439@xmath163 & [email protected] & 0.81 ( 124 ) + 33 & [email protected] & 459@xmath163 & [email protected] & 1.07 ( 136 ) + 34 & [email protected] & 390@xmath163 & [email protected] & 0.97 ( 93 ) + 35 & [email protected] & 350@xmath162 & [email protected] & 1.02 ( 135 ) + 36 & [email protected] & 365@xmath162 & [email protected] & 0.78 ( 136 ) + 37 & [email protected] & 388@xmath162 & [email protected] & 0.96 ( 165 ) + 38 & [email protected] & 413@xmath162 & [email protected] & 0.97 ( 187 ) + 39 & [email protected] & 515@xmath162 & [email protected] & 1.07 ( 182 ) + 40 & [email protected] & 469@xmath162 & [email protected] & 0.96 ( 146 ) + 41 & [email protected] & 423@xmath162 & [email protected] & 1.11 ( 153 ) + 42 & [email protected] & 348@xmath162 & [email protected] & 0.70 ( 92 ) + 43 & [email protected] & 352@xmath162 & [email protected] & 0.98 ( 131 ) + 44 & [email protected] & 369@xmath163 & [email protected] & 1.23 ( 84 ) + 45 & [email protected] & 343@xmath162 & [email protected] & 0.98 ( 97 ) + 46 & [email protected] & 293@xmath161 & [email protected] & 0.96 ( 126 ) + 47 & [email protected] & 308@xmath162 & [email protected] & 1.09 ( 138 ) + 48 & [email protected] & 311@xmath162 & [email protected] & 1.04 ( 130 ) + 49 & [email protected] & 378@xmath163 & [email protected] & 1.13 ( 80 ) + 50 & [email protected] & 476@xmath162 & [email protected] & 1.04 ( 132 ) + 51 & [email protected] & 443@xmath162 & [email protected] & 0.79 ( 144 ) + 52 & [email protected] & 416@xmath163 & [email protected] & 0.86 ( 112 ) + 53 & [email protected] & 429@xmath162 & [email protected] & 1.04 ( 171 ) + 54 & [email protected] & 365@xmath162 & [email protected] & 1.13 ( 180 ) + 55 & [email protected] & 341@xmath162 & [email protected] & 1.02 ( 152 ) + 56 & [email protected] & 372@xmath162 & [email protected] & 1.10 ( 157 ) + 57 & [email protected] & 396@xmath162 & [email protected] & 1.00 ( 166 ) + ll |cccc | ccc 1997.4.29 & 1 & [email protected] & 137@xmath163 & [email protected] & 0.99 ( 126 ) & [email protected] & 146@xmath166 & 0.96 ( 125 ) + & 2 & [email protected] & 124@xmath162 & [email protected] & 1.21 ( 126 ) & [email protected] & 125@xmath164 & 1.23 ( 125 ) + & 3 & [email protected] & 131@xmath162 & [email protected] & 1.16 ( 126 ) & [email protected] & 134@xmath163 & 1.15 ( 125 ) + 1997.4.30 & 1 & [email protected] & 132@xmath164 & [email protected] & 1.24 ( 126 ) & [email protected] & 127@xmath165 & 1.25 ( 125 ) + & 2 & [email protected] & 127@xmath163 & [email protected] & 0.88 ( 126 ) & [email protected] & 132@xmath165 & 0.85 ( 125 ) + & 3 & [email protected] & 124@xmath163 & [email protected] & 0.85 ( 126 ) & [email protected] & 123@xmath164 & 0.84 ( 125 ) + & 4 & [email protected] & 129@xmath162 & [email protected] & 0.81 ( 126 ) & [email protected] & 130@xmath162 & 0.79 ( 125 ) + 1997.5.1 & 1 & [email protected] & 129@xmath164 & [email protected] & 1.11 ( 126 ) & [email protected] & 131@xmath166 & 1.11 ( 125 ) + & 2 & [email protected] & 144@xmath163 & [email protected] & 1.03 ( 126 ) & [email protected] & 146@xmath164 & 1.03 ( 125 ) + & 3 & [email protected] & 178@xmath164 & [email protected] & 0.84 ( 126 ) & [email protected] & 170@xmath165 & 0.85 ( 125 ) + & 4 & [email protected] & 143@xmath162 & [email protected] & 1.03 ( 126 ) & [email protected] & 145@xmath162 & 1.01 ( 125 ) + 1997.5.2 & 1 & [email protected] & 171@xmath163 & [email protected] & 0.86 ( 126 ) & [email protected] & 174@xmath164 & 0.85 ( 125 ) + 1997.5.3 & 1 & [email protected] & 120@xmath164 & [email protected] & 0.97 ( 126 ) & [email protected] & 116@xmath164 & 0.96 ( 125 ) + & 2 & [email protected] & 123@xmath163 & [email protected] & 1.13 ( 126 ) & [email protected] & 120@xmath164 & 1.14 ( 125 ) + 1997.5.4 & 1 & [email protected] & 105@xmath164 & [email protected] & 1.23 ( 126 ) & [email protected] & 104@xmath165 & 1.25 ( 125 ) + & 2 & [email protected] & 103@xmath163 & [email protected] & 1.23 ( 126 ) & [email protected] & 103@xmath164 & 1.24 ( 125 ) + 1997.5.5 & 1 & [email protected] & 127@xmath164 & [email protected] & 1.24 ( 126 ) & [email protected] & 131@xmath165 & 1.24 ( 125 ) + & 2 & [email protected] & 127@xmath163 & [email protected] & 1.24 ( 126 ) & [email protected] & 129@xmath164 & 1.25 ( 125 ) + 1998.4.21- & 1 & [email protected] & 341@xmath168 & [email protected] & 1.04 ( 126 ) & [email protected] & 357@xmath1613 & 1.01 ( 125 ) + & 2 & [email protected] & 404@xmath166 & [email protected] & 1.12 ( 126 ) & [email protected] & 397@xmath167 & 1.06 ( 125 ) + & 3 & [email protected] & 378@xmath165 & [email protected] & 1.03 ( 126 ) & [email protected] & 379@xmath166 & 1.04 ( 125 ) + & 4 & [email protected] & 337@xmath165 & [email protected] & 0.93 ( 126 ) & [email protected] & 337@xmath167 & 0.94 ( 125 ) + & 5 & [email protected] & 300@xmath165 & [email protected] & 0.90 ( 126 ) & [email protected] & 310@xmath167 & 0.86 ( 125 ) + & 6 & [email protected] & 274@xmath164 & [email protected] & 1.31 ( 126 ) & [email protected] & 274@xmath165 & 1.33 ( 125 ) + & 7 & [email protected] & 239@xmath164 & [email protected] & 1.04 ( 126 ) & [email protected] & 235@xmath165 & 1.02 ( 125 ) + & 8 & [email protected] & 236@xmath165 & [email protected] & 0.95 ( 126 ) & [email protected] & 235@xmath166 & 0.96 ( 125 ) + 1998.4.23- & 1 & [email protected] & 309@xmath168 & [email protected] & 1.11 ( 126 ) & [email protected] & 287@xmath168 & 1.06 ( 125 ) + & 2 & [email protected] & 257@xmath166 & [email protected] & 1.07 ( 126 ) & [email protected] & 257@xmath168 & 1.08 ( 125 ) + & 3 & [email protected] & 247@xmath165 & [email protected] & 0.90 ( 126 ) & [email protected] & 254@xmath167 & 0.88 ( 125 ) + & 4 & [email protected] & 248@xmath166 & [email protected] & 1.04 ( 126 ) & [email protected] & 239@xmath167 & 1.01 ( 125 ) + & 5 & [email protected] & 253@xmath165 & [email protected] & 0.86 ( 126 ) & [email protected] & 251@xmath167 & 0.87 ( 125 ) + & 6 & [email protected] & 233@xmath165 & [email protected] & 1.17 ( 126 ) & [email protected] & 227@xmath166 & 1.18 ( 125 ) + & 7 & [email protected] & 220@xmath165 & [email protected] & 1.34 ( 126 ) & [email protected] & 218@xmath166 & 1.36 ( 125 ) + & 8 & [email protected] & 219@xmath167 & [email protected] & 1.07 ( 126 ) & [email protected] & 213@xmath168 & 1.08 ( 125 ) + & 9 & [email protected] & 217@xmath167 & [email protected] & 1.04 ( 126 ) & [email protected] & 203@xmath167 & 1.01 ( 125 ) + & 10 & [email protected] & 203@xmath165 & [email protected] & 1.14 ( 126 ) & [email protected] & 194@xmath166 & 1.14 ( 125 ) + lc | cc| c c cc 50926.8814 & 11.0 & hegra & [email protected] & [email protected] & 226@xmath166 & [email protected] & 0.75(33 ) + 50927.1390 & 13.7&_whipple_&[email protected] & [email protected] & 218@xmath162 & [email protected] & 0.95(216 ) + 50927.8839 & 10.5 & hegra & [email protected] & [email protected] & 381@xmath163 & [email protected] & 1.18(210 ) + 50928.2273 & 5.3 & _ whipple_&[email protected] & [email protected] & 269@xmath163 & [email protected] & 0.95(142 ) + 50928.8824 & 10.5 & hegra & [email protected] & [email protected] & 316@xmath162 & [email protected] & 0.90(221 ) + 50929.8848 & 10.1 & hegra & [email protected] & [email protected] & 424@xmath162 & [email protected] & 1.13(205 ) + 50930.1516 & 12.2&_whipple_&[email protected] & [email protected] & 364@xmath162 & [email protected] & 0.99(274 ) + 50930.8845 & 9.8 & hegra & [email protected] & [email protected] & 472@xmath164 & [email protected] & 1.15(244 ) + 50931.1467 & 10.3&_whipple_&[email protected] & [email protected] & 367@xmath163 & [email protected] & 0.94(229 ) + 50932.2041 & 6.9 & _ whipple_&[email protected] & [email protected] & 367@xmath162 & [email protected] & 1.31(239 ) + 50933.2358 & 3.4 & _ whipple_&[email protected] & [email protected] & 454@xmath162 & [email protected] & 1.21(238 ) +

in this paper , we first briefly review some recent results on the distribution of the maximal eigenvalue of a random matrix drawn from gaussian ensembles . next we focus on the gaussian unitary ensemble ( gue ) and by suitably adapting a method of orthogonal polynomials developed by gross and matytsin in the context of yang - mills theory in two dimensions , we provide a rather simple derivation of the tracy - widom law for gue . our derivation is based on the elementary asymptotic scaling analysis of a pair of coupled nonlinear recursion relations . as an added bonus , this method also allows us to compute the precise subleading terms describing the right large deviation tail of the maximal eigenvalue distribution . in the yang - mills language , these subleading terms correspond to non - perturbative ( in expansion ) corrections to the two - dimensional partition function in the so called ` weak ' coupling regime .

1102.0738

quite a long time ago , wigner @xcite introduced random matrices in the context of nuclear physics . he suggested that the highly - excited energy levels of complex nuclei can locally be well represented by the eigenvalues of a large random matrix . a big nucleus is a rather complex system composed of many strongly interacting quantum particles and it is practically impossible to describe its spectral properties via first principle calculations . the idea of wigner was to model the spectral properties of the complex hamiltonian of such a big nucleus by those of a large random matrix preserving the same symmetry . this was a very successful approach in nuclear physics . since then , the random matrix theory ( rmt ) has gone beyond nuclear physics and has found a wide number of applications in various fields of physics and mathematics including quantum chaos , disordered systems , string theory and even number theory @xcite . a case of special interest is the one of gaussian random matrices ( originally introduced by wigner himself ) where the entries of the matrix are gaussian random variables . depending on the symmetry of the problem , dyson distinguished three classes for the matrix @xmath2 @xcite : @xmath3 the gaussian orthogonal ensemble ( goe ) : @xmath2 is real symmetric . @xmath3 the gaussian unitary ensemble ( gue ) : @xmath2 is complex hermitian . @xmath3 the gaussian symplectic ensemble ( gse ) : @xmath2 is quaternionic hermitian . let us write @xmath4 the adjoint of @xmath2 , i.e. the transpose of @xmath2 for the goe , the complex conjugate transpose for the gue and the quaternionic conjugate transpose for the gse . a gaussian random matrix is a @xmath5 self - adjoint matrix @xmath2 , i.e. @xmath6 distributed according to the law @xmath7 where , for convenience , we have chosen the prefactor @xmath8 of the @xmath9 to be @xmath10 for the goe , @xmath11 for the gue and @xmath12 for the gse . for instance , for the gue we have @xmath11 and @xmath13 as @xmath14 . this means that @xmath2 is a @xmath5 complex hermitian matrix with entries @xmath15 and @xmath16 for @xmath17 that are independent ( real ) random variables distributed according to the same centered gaussian law with variance @xmath18 and the @xmath19 are ( real ) independent gaussian variables with mean @xmath20 and variance @xmath21 . in case of gse , there are @xmath22 eigenvalues , each of them two - fold degenerate and tr in for @xmath12 is defined so that only one of the two fold degenerate eigenvalues in @xmath2 is counted . + self - adjoint matrices can be diagonalized and have real eigenvalues . the joint distribution of eigenvalues of the gaussian ensemble is well known @xcite @xmath23 where @xmath24 is a normalization constant such that @xmath25 ( it depends on @xmath8 ) and the power @xmath8 of the vandermonde term is called the dyson index @xmath26 depending on the ensemble ( resp . goe , gue or gse ) . note that we have chosen the prefactor of @xmath9 term in to be the same as the dyson index @xmath8 just for convenience . this prefactor is not very important as it can be absorbed by rescaling the matrix entries by a constant factor . in contrast , the value of the dyson index @xmath10 , @xmath27 or @xmath28 , characterizing the power of the vandermonde term , plays a crucial role . the normalization constant @xmath24 can be computed using selberg s integral @xcite : @xmath29 $ ] . because of the presence of the vandermonde determinant @xmath30 in eq . , the eigenvalues are strongly correlated random variables , they repel each other . in this paper , our focus is on the statistical properties of the extreme ( maximal ) eigenvalue @xmath31 . had the vandermonde term been not there in the joint distribution , the joint distribution would factorize and the eigenvalues would thus be completely independent random variables , each with a gaussian distribution . for such independent and identically distributed random variables @xmath32 , the extreme value statistics is well understood @xcite and the distribution of the maximum , properly shifted and scaled , belongs to one of the three universality classes gumbel , frechet or weibull ( for large @xmath33 ) depending on the tail of the distribution of individual @xmath34 s . however , in the case of random matrix theory , the eigenvalues @xmath34 s are strongly correlated variables . for strongly correlated random variables there is no general theory for the distribution of the maximum . in case of gaussian random matrices , where the joint distribution is explicitly known , much progress has been made in understanding the distribution of @xmath35 following the seminal work by tracy and widom @xcite . this then provides a very useful solvable model for the extreme value distribution in a strongly correlated system and hence is of special interest . let us first summarize some known properties of the random variable @xmath35 . its average value can be easily obtained from the right edge of the well known wigner semi - circle describing the average density of eigenvalues . for a gaussian random matrix of large size @xmath33 , the average density of eigenvalues ( normalized to unity ) @xmath36 has a semi - circular shape on a finite support @xmath37 $ ] called the wigner semi - circle @xcite : @xmath38 the quantity @xmath39 represents the average fraction of eigenvalues that lie within the small interval @xmath40 $ ] . therefore , eq . means that the eigenvalues of a gaussian random matrix lie on average within the finite interval @xmath37 $ ] . note also that one can rewrite , using the joint distribution in @xmath41 hence the average density of states @xmath42 can also be interpreted as the marginal distribution of one of the eigenvalues ( say the first one ) . thus , the marginal distribution also has the shape of a semi - circle . figure [ fig : denstw ] shows the average density @xmath42 ( @xmath43 here ) . + it then follows that the average value of the maximal eigenvalue @xmath35 is given for large @xmath33 by the upper bound of the density support : @xmath44 however , @xmath35 fluctuates around this average value from one realization to another and has a distribution around its mean value @xmath45 ( see fig . [ fig : denstw ] with @xmath43 ) . what is the full probability distribution of @xmath35 ? from the joint distribution of eigenvalues in eq . ( [ eq : jpdfev ] ) , it is easy to write down formally the cumulative distribution function ( cdf ) of @xmath35 as a multiple integral @xmath46 which can be interpreted as a partition function of a coulomb gas in presence of a hard wall at the location @xmath47 ( see the discussion in section 2 ) . the question is how does @xmath48 behave for large @xmath33 ? it turns out that the fluctuations of @xmath35 around its mean @xmath45 have two scales for large @xmath33 . while typical fluctuations scale as @xmath49 , large fluctuations scale as @xmath50 and their probability distributions are described by different functional forms ( see fig . [ fig : denstw ] with @xmath43 ) . as a function of @xmath51 ( blue dashed line ) . the density has a semi - circular shape ( `` wigner semi - circle '' ) and a finite support @xmath52 $ ] . the maximal eigenvalue has mean value @xmath53 for large @xmath33 and its distribution close to the mean value , over a scale of @xmath54 has the tracy - widom form ( red solid line ) . however , over a scale @xmath55 the distribution has large deviation tails shown by solid green ( left large deviations ) and solid blue ( right large deviations ) lines.,width=453 ] * typical fluctuations : * from an asymptotic analysis of the mutilple integral in eq . ( [ multint ] ) , forrester @xcite , followed tracy and widom @xcite deduced that for large @xmath33 , _ small and typical _ fluctuations of the maximal eigenvalue around its mean value @xmath45 are of order @xmath54 and can be written as @xmath56 where @xmath57 ( for goe and gue ) and @xmath58 ( gse ) and @xmath59 is a random variable characterizing the typical fluctuations . tracy and widom @xcite proved that for large @xmath33 , the distribution of @xmath59 is independent of @xmath33 : @xmath60 . the function @xmath61 depends explicitly on @xmath8 and is called the tracy - widom distribution . for example , for @xmath11 @xcite , @xmath62 \label{f2tw}\ ] ] where @xmath63 satisfies the special case of @xmath64 of the painlev ii equation @xmath65 for @xmath64 , the solution only requires the right tail boundary condition for its unique specification : @xmath66 as @xmath67 , where @xmath68 is the airy function that satisfies the differential equation @xmath69 and vanishes as , @xmath70 as @xmath71 . this solution of the special case @xmath64 of the painlev - ii equation is called the hastings - mcleod solution @xcite . for @xmath11 and @xmath12 , one has @xcite @xmath72^{1/2}\ , \exp\left[\frac{1}{2}\int_x^{\infty } q(z ) dz\right ] \label{f1tw } \\ f_4(x)&= & \left[f_2(x)\right]^{1/2}\ , \cosh\left[\frac{1}{2}\int_x^{\infty } q(z ) dz\right ] . \label{f4tw}\end{aligned}\ ] ] note that @xmath73 is the cumulative probability of the scaled random variable @xmath59 and hence it approaches to @xmath74 as @xmath75 and vanishes to @xmath20 as @xmath76 . the corresponding probability density function ( pdf ) @xmath77 vanishes as @xmath78 in an asymmetric fashion @xmath79 \quad { \rm as}\quad x\to -\infty \label{lefttw } \\ & \sim & \exp\left[- \frac{2\beta}{3}\ , x^{3/2}\right ] \quad { \rm as}\quad x\to \infty \label{righttw}\end{aligned}\ ] ] over the last decade or so , the tracy - widom distribution has appeared in a wide variety of problems ranging from statistical physics and probability theory all the way to growth models and biological sequence matching problems ( for reviews see @xcite ) . these include the longest increasing subsequence or the ulam problem @xcite , a wide variety of ( 1 + 1)-dimensional growth models @xcite , directed polymer in random medium @xcite and the continuum kardar - parisi - zhang equation @xcite , bernoulli matching problem between two random sequences @xcite , nonintersecting brownian motions ( see e.g. @xcite and references therein ) . this distribution has also been measured in a variety of recent experiments , e.g. , in the height distribution of fronts generated in paper burning experiment @xcite , in turbulent liquid crystals @xcite and more recently in coupled fiber laser systems @xcite . * large deviations : * tracy - widom distribution describes the probability of _ typical _ fluctuations of @xmath35 around its mean ( on a scale of @xmath49 ) , but not the _ atypical _ large fluctuations , i.e. , fluctuations of order @xmath80 around the mean value @xmath81 . questions regarding such large / rare fluctuations do arise in various contexts @xcite and have recently been computed @xcite to dominant order for large @xmath33 . as a summary , the probability density of @xmath35 , @xmath82 $ ] , is given for large @xmath33 by : @xmath83 where @xmath61 is the tracy - widom distribution and where @xmath84 and @xmath85 are respectively the left and right large deviation functions describing the tails of the distribution of @xmath35 . the rate function @xmath86 was explicitly computed in @xcite , while @xmath87 was computed in @xcite , both by simple physical methods exploiting the coulomb gas analogy . a more complicated , albeit mathematically rigorous , derivation of @xmath87 in the context of spin glass models can be found in @xcite . these rate functions read @xmath88+\frac{\ln 3}{2 } , \;\ ; \textrm{for $ z<\sqrt{2}$}\nonumber\\ \psi_{+}(z)&=&\frac{z \sqrt{z^2 - 2}}{2 } + \ln\left[\frac{z-\sqrt{z^2 - 2}}{\sqrt{2}}\right],\ ; \;\ ; \;\;\ ; \textrm{for $ z>\sqrt{2}$.}\end{aligned}\ ] ] note that in ref . @xcite , the function @xmath89 was expressed in terms of a complicated hypergeometric function , which however can be reduced to a simple algebraic function as presented above in eq . ( [ eq : psi_-+ ] ) . note also that while @xmath90 depends explicitly on @xmath8 , the rate functions @xmath86 and @xmath89 are independent of @xmath8 . these rate functions only give the dominant order for large @xmath33 in the exponential . in other words , the precise meaning of @xmath91 is that for large @xmath33 : @xmath92 for @xmath93 and @xmath94 for @xmath95 . when @xmath96 approaches @xmath97 ( from below or above ) it is easy to see that the rate functions vanish respectively as @xmath98 note that the physics of the left tail @xcite is very different from the physics of the right tail @xcite . in the former case , the semi - circular charge density of the coulomb gas is _ pushed _ by the hard wall ( @xmath93 ) leading to a reorganization of all the @xmath33 charges that gives rise to an energy difference of @xmath99 @xcite . in contrast , for the right tail @xmath100 , the dominant fluctuations are caused by _ pulling _ a single charge away ( to the right ) from the wigner sea leading to an energy difference of @xmath101 @xcite . the different behaviour of the probability distribution for @xmath93 and @xmath100 leads to a ` phase transition ' strictly in the @xmath102 limit at the critical point @xmath103 in the following sense . indeed , if one scales @xmath51 by @xmath104 and takes the @xmath102 limit , one obtains @xmath105 note that since @xmath106 can be interpreted as a partition function of a coulomb gas ( see eq . ( [ multint ] ) ) , its logarithm has the interpretation of a free energy . since @xmath107 as @xmath108 from below , the @xmath109-rd derivative of the free energy is discontinuous at the critical point @xmath103 . hence , this can be interpreted as a _ third order _ phase transition . however , for finite but large @xmath33 , it follows from that the behavior to the left of @xmath103 smoothly crosses over to the behaviour to the right as one varies @xmath96 through its critical point @xmath103 and the tracy - widom distribution in around the critical point is precisely this crossover function . indeed , if one zooms in close to the mean value @xmath81 by setting @xmath110 ( for @xmath111 ) in the rate functions @xmath112 and @xmath113 in , one expects to recover , by taking large @xmath33 limit , respectively the left and the right tail of the tracy - widom distribution . with this scaling , and using , one finds @xmath114 and thus @xmath115 , which indeed matches the dominant order in the far right tail of the tracy - widom distribution for @xmath111 in . similarly for the left tail ( @xmath116 ) , using , one finds @xmath117 , thus @xmath118 which matches the left tail of the tracy - widom distribution in . more recently , higher order corrections for large @xmath33 have been computed for the left tail of the distribution @xcite using methods developed in the context of matrix models . note that in @xcite a different notation for @xmath8 was used : @xmath119 ( goe ) , @xmath10 ( gue ) and @xmath11 ( gse ) . to avoid confusion , we present below the results in terms of the standard dyson index @xmath120 . @xmath121 where @xmath122 with @xmath86 given in eq . ( dominant order ) . the subleading terms are given by @xcite @xmath123 \label{phi1}\end{aligned}\ ] ] and @xmath124 and ( see eq . ( 4 - 35 ) in ref . @xcite ) @xmath125+\left(\frac{3}{8}-\frac{1}{4 \beta } -\frac{\beta } { 16}\right ) \ln \left [ -2 z+\sqrt{6+z^2 } \right]\nonumber \\ & + & \left(\frac{1}{2 } -\frac{1}{3\beta } -\frac{\beta } { 12}\right ) \ln \left [ z+\sqrt{6+z^2}\right ] \nonumber \\ & + & \left(\frac{-4}{3}+\frac{4}{3 \beta } + \frac{\beta } { 3}\right ) \ln \left[\sqrt{-2 z+\sqrt{6+z^2}}+\sqrt{3}\left(z^2 + 6\right)^{1/4 } \right]\nonumber \\ & + & \frac{5}{3}\left(1-\frac{1}{\beta } -\frac{\beta } { 4}\right)\ln \left[-z+2 \sqrt{6+z^2}+\sqrt{3}\left(z^2 + 6\right)^{1/4 } \sqrt{\sqrt{6+z^2}-2 z } \right]\nonumber \\ & -&\ln \left[\left(-18 + z^2\right)z+\left(6+z^2\right)^{3/2}\right]- \frac{\ln \beta } { 2}-\kappa_{\beta } \label{phi0}\end{aligned}\ ] ] where @xmath126 is a complicated function of @xmath8 . for @xmath127 integer , it reduces to @xcite @xmath128 for instance , for the gue ( @xmath11 ) , we find @xmath129 . for @xmath111 and @xmath28 , the expression in eq . matches the left asymptotics of the tracy - widom distribution , i.e. the asymptotic behaviour of @xmath130 for @xmath131 , see ref . . however , for the right tail of the distribution of @xmath35 , the corrections to dominant order for large @xmath33 are , to our knowledge , not known until now . in fact , one of the results of this paper is to compute these right tail corrections for the gue ( @xmath11 ) . both left and right large deviations are plotted in fig . [ fig : largedev ] for the gue . the left tail is described by @xmath132 in eq . , the right tail is described by our result given in eq . . another result of this paper concerns a simpler and pedestrian derivation of the tracy - widom distribution for the gue case . the original derivation of the tracy - widom law for the distribution of typical fluctuations of @xmath35 @xcite is somewhat complex as it requires a rather sophisticated and nontrivial asymptotic analysis of the fredholm determinant involving airy kernel @xcite . since this distribution appears in so many different contexts , it is quite natural to ask if there is any other simpler ( more elementary ) derivation of the tracy - widom distribution . in this paper , we provide such a derivation for the gue case . our method is based on a suitable modification of a technique of orthogonal polynomials developed by gross and matytsin @xcite in the context of two - dimensional yang - mills theory . in fact , the partition function of the continuum two - dimensional pure yang - mills theory on a sphere ( with gauge group @xmath133 ) can be written ( up to a prefactor ) as a discrete multiple sum over integers @xcite @xmath134 where @xmath135 is the area of the sphere . in the @xmath102 limit , the free energy @xmath136 , as a function of @xmath135 , undergoes a 3rd order phase transition known as the douglas - kazakov transition @xcite at the critical value @xmath137 . for @xmath138 , the system is in the ` strong ' coupling phase while for @xmath139 , it is in the ` weak ' coupling phase . for finite but large @xmath33 , there is a crossover between the two phases as one passes through the vicinity of the critical point . in the so called double scaling limit ( where @xmath140 , @xmath102 but keeping the product @xmath141 fixed ) , the singular part of the free energy satisfies a painlev ii equation @xcite . gross and matytsin ( see also @xcite ) used a method based on orthogonal polynomials to analyse the partition sum in the double scaling limit , as well as in the weak coupling regime ( @xmath139 ) where they were able to compute non - perturbative ( in @xmath1 expansion ) corrections to the free energy . actually , a similar @xmath109-rd order phase transition from a weak to strong coupling phase in the @xmath102 limit was originally noticed in the lattice formulation ( with wilson action ) of the two dimensional @xmath133 gauge theory @xcite and in the vicinity of the transition point the singular part of the free energy was shown to satisfy a painlev ii equation @xcite . note that similar calculations involving the asymptotic analysis of partition functions using orthogonal polynomials were used extensively in the early 90 s to study the double scaling limit of the so called one - matrix model ( for a recent review and developments , see e.g. ref . @xcite ) . in our case , for the distribution of @xmath35 , we need to analyse the asymptotic large @xmath33 behaviour of a multiple _ indefinite _ integral in eq . ( [ multint ] ) , as opposed to the discrete sum in eq . ( [ ympf ] ) . however , we show that one can suitably modify the orthogonal polynomial method of gross and matytsin to analyse the multiple integral in eq . ( [ multint ] ) in the limit of large @xmath33 . in fact , we find a similar third order phase transition ( in the @xmath102 limit ) in the largest eigenvalue distribution @xmath48 as a function of @xmath47 at the critical point @xmath142 . the regime of left large deviation of @xmath48 ( @xmath143 ) is similar to the ` strong ' coupling regime @xmath144 of the yang - mills theory , while the right large deviation tail of @xmath48 ( @xmath145 ) is similar to the ` weak ' coupling regime @xmath146 of the yang - mills theory . for finite but large @xmath33 , the crossover function across the critical point that connects the left and right large deviation tails is precisely the tracy - widom distribution . thus the tracy - widom distribution corresponds precisely to the double scaling limit of the yang - mills theory and one finds the same painlev ii equation . a similar 3rd order phase transition was also found recently in a model of non - intersecting brownian motions by establishing an exact correspondence between the reunion probability in the brownian motion model and the partition function in the @xmath27-d yang - mills theory on a sphere @xcite . the advantage of this orthogonal polynomial method to analyse the maximum eigenvalue distribution is twofold : ( i ) one gets the tracy - widom distribution in a simple elementary way ( basically one carries out a scaling analysis of a pair of nonlinear recursion relations near the critical point and shows that the scaling function satisfies a painlev ii differential equation ) and ( ii ) as an added bonus , we also obtain precise subleading corrections to the leading right large deviation tail @xmath147 . the subleading corrections , in the yang - mills language , correspond to the non - perturbative corrections in the weak coupling regime as derived by gross and matytsin @xcite . more precisely we show that @xmath148 where @xmath87 is given in eq . ( [ eq : psi_-+ ] ) . note that only the leading behaviour @xmath149 $ ] was known before @xcite , but the subleading corrections are , to our knowledge , new results . we also verify that our expression matches the precise right asymptotics of the tracy - widom distribution . figure [ fig : largedev ] shows the distribution of @xmath35 for the gue : close to the mean value it is described by the tracy - widom distribution , whereas the tails are described by the large deviations . the right tail ( right large deviation ) is given by our result in eq . . together with the subleading terms in the left tail in eq . ( [ eq : phigaetan ] ) , our new result in eq . ( [ eq : righttailintro ] ) then provides a rather complete picture of the tail behaviors of the distribution of @xmath35 on both sides of the mean @xmath45 . associated to the distribution @xmath150 of the maximal eigenvalue of a random matrix from the gue for large @xmath33 . close to the mean value @xmath103 , the distribution is a tracy - widom law ( red line ) , it describes the small typical fluctuations around the mean value . atypical large flutuations are described by the large deviations : the left large deviation in green ( @xmath93 ) , the right deviation in blue ( @xmath100).,width=453 ] the rest of the paper is organized as follows . in section [ sec : intro ] , we start with some general notations and scaling remarks for the gue . in section [ sec : orthpol ] , we explain the method of orthogonal polynomials on a semi - infinite interval and derive some basic recursion relations . in section [ sec : largedev ] , we compute the right tail of the distribution of @xmath35 ( dominant order and corrections for the gue ) : it describes atypical large fluctuations of @xmath35 to the right of its mean value . in section [ sec : tw ] , using results of the previous sections and basic scaling remarks , we derive the tracy - widom law ( with @xmath11 for the gue ) that describes small typical fluctuations close to the mean value . in the rest of the paper we focus only on gaussian random matrices @xmath2 drawn from the gue ( @xmath11 ) . they are hermitian random matrices @xmath6 such that @xmath151 where we have introduced an additional parameter @xmath152 for the purpose of certain mathematical manipulations that will be clear later . setting @xmath43 at the end of the calculation , we will recover the usual gue . with the additional parameter @xmath153 , the joint distribution of the eigenvalues of @xmath2 is given by ( see eq . ): @xmath154 where @xmath155 $ ] is the normalization constant . the vandermonde determinant appears with a power @xmath27 as we consider the gue ( @xmath11 ) . this determinant indeed comes from a jacobian for the change of variables from the entries of the matrix to its eigenvalues . the power is related to the number of real degrees of freedom of an element of the matrix , which is @xmath27 for complex entries , i.e. , for gue ( it is @xmath74 for real entries goe and @xmath28 for quaternion entries gse ) . as we will see later , this power @xmath27 is crucial for the method of orthogonal polynomials to work . the technique of gross and matytsin @xcite that we adapt here also works only for the gue @xmath11 case . + given the joint distribution of eigenvalues in eq . ( [ eq : jointpdfev ] ) , it is easy to write down the cumulative distribution of the maximal eigenvalue @xmath35 @xmath156=\frac{z_n \left(y,\alpha\right)}{z_n(\infty,\alpha)}\ ] ] where the partition function @xmath157 is given by the multiple indefinite integral @xmath158 the normalization @xmath159 is actually related to @xmath160 in eq . as @xmath161 . note that , by the trivial rescaling @xmath162 in , it follows from that @xmath163 thus @xmath164 and @xmath153 always appear in a single scaling combination @xmath165 . we will henceforth focus on the large @xmath33 limit . for fixed @xmath153 , one can easily figure out from the joint pdf in eq . ( [ eq : jointpdfev ] ) how a typical eigenvalue @xmath166 scales with @xmath33 for large @xmath33 . let us rewrite the joint distribution of eigenvalues in eq . as @xmath167 which can then be interpreted as a boltzmann weight @xmath168 $ ] , with effective energy @xmath169 . the eigenvalues can thus be seen as the positions of @xmath33 charges of a 2d coulomb gas ( but restricted to be on the real line ) which repel each other via a logarithmic coulomb potential ( coming from the vandermonde determinant in eq . ) in addition , the charges are subjected to an external confining parabolic potential . for large @xmath33 , the first term of the energy is of order @xmath170 , whereas the second is of order @xmath171 because of the double sum . balancing the two terms @xmath172 gives the scaling of a typical eigenvalue for large @xmath33 : @xmath173 . for large @xmath33 , the eigenvalues are close to each other and they can be described by a continuous charge density ( normalized to unity ) @xmath174 . the average density of states for large @xmath33 can be obtained by evaluating the full partition function @xmath175 ( the denominator in eq . ( [ eq : cdflmax ] ) ) via a saddle point method . the saddle point density is the density that minimizes the effective energy ( see the book of mehta @xcite ) @xmath176 ( in its continuous version ) . this gives the well - known semi - circle law ( which is exactly the same as in eq . for @xmath43 ) : @xmath177 the density is plotted in figure [ fig : denstw ] . + the average value of @xmath35 is again given for large @xmath33 by the upper bound of the density support ( see fig . [ fig : denstw ] ) : @xmath178 for @xmath43 , this evidently reduces to the usual expression for @xmath179 . the typical scaling for large @xmath33 is thus @xmath180 . hence , we will use @xmath181 , where @xmath96 is of order one . in this section , we introduce the method of orthogonal polynomials to evaluate the partition function in eq . ( [ eq : zn ] ) . as mentioned in the introduction , to evaluate this multiple indefinite integral we will adapt the method developed by gross and matytsin @xcite to enumerate the partition sum in the two - dimensional yang - mills theory . orthogonal polynomials are very useful to handle the square vandermonde determinant in the distribution of the eigenvalues in eq . . a vandermonde determinant can indeed be written as @xmath182 where @xmath183 is any polynomial of degree @xmath184 with leading coefficient one . the idea is to choose well these polynomials @xmath185 in order to simplify the computation of the integral . we define an operation on pairs of polynomials as follows : @xmath186 we consider a family @xmath187 of orthogonal polynomials with respect to the operation defined above , i.e. with weight @xmath188 on the interval @xmath189-\infty , y]$ ] . without any loss of generalization , we define the polynomial @xmath190 of degree @xmath191 such that the coefficient of @xmath192 term is always fixed to be @xmath74 , i.e. , @xmath193 . these polynomials satisfy the orthogonality property : @xmath194 for all @xmath195 . we also write @xmath196 . thus , @xmath197 note that @xmath190 and @xmath198 are implicitly functions of @xmath153 and @xmath164 , i.e. @xmath199 and @xmath200 . in particular , we can easily compute by hand the first few @xmath190 s for fixed @xmath201 and @xmath164 , but the expressions become rather complex as @xmath191 increases and it is hard to find a closed form expression for @xmath190 for every @xmath191 ( except for the limiting case @xmath202 ) : @xmath203}\\ p_2(\lambda)&=&\lambda^2+\frac{-2 y \sqrt{\alpha } -e^{y^2 \alpha } \sqrt{\pi } \left(1 + 2 y^2 \alpha \right ) \left(1+\text{erf}\left[y \sqrt{\alpha } \right]\right ) } { \sqrt{\alpha } \left(2 + 2 e^{y^2 \alpha } \sqrt{\pi } y \sqrt{\alpha } \left(1+\text{erf}\left[y \sqrt{\alpha } \right]\right ) -e^{2 y^2 \alpha } \pi \left(1+\text{erf}\left[y \sqrt{\alpha } \right]\right)^2\right ) } \ ; \lambda+\\ & & \;\;\;\;\;\;\;\;\;+ \frac{-4 - 4 e^{y^2 \alpha } \sqrt{\pi } y \sqrt{\alpha } \left(1+\text{erf}\left[y \sqrt{\alpha } \right]\right ) + e^{2 y^2 \alpha } \pi \left(1+\text{erf}\left[y \sqrt{\alpha } \right]\right)^2 } { 4 \alpha + 4 e^{y^2 \alpha } \sqrt{\pi } y \alpha ^{3/2 } \left(1+\text{erf } \left[y \sqrt{\alpha } \right]\right)-2 e^{2 y^2 \alpha } \pi \alpha \left(1+\text{erf}\left[y \sqrt{\alpha } \right]\right)^2}\end{aligned}\ ] ] thus we get for instance @xmath204 $ ] + and @xmath205 -\frac{2 e^{-2 \alpha y^2}}{\sqrt{\pi}\left[1+{\rm erf}(y \sqrt{\alpha})\right ] } } { 4 \alpha^{3/2}}$ ] , etc . in the limit @xmath202 , we recover the hermite polynomials : @xmath206 , @xmath207 , @xmath208 and @xmath209 , @xmath210 , @xmath211 . the partition function @xmath212 in eq . can be written as a function of the @xmath198 s . by combination of rows , the vandermonde determinant in the joint distribution of the eigenvalues can indeed be written @xmath213 then , the partition function can be expressed as @xmath214_{i , j}\\ & = & \det\left(\langle p_{i-1}|p_{j-1}\rangle\right)_{i , j } = \det\left(\delta_{i , j } h_{i-1}\right)_{i , j}=\prod_{i=0}^{n-1 } h_i\end{aligned}\ ] ] where in going from the second to the third line we have used the cauchy - binet formula @xcite . note that this step works only for @xmath11 . therefore the partition function reduces to : @xmath215 thus the integral @xmath212 is now expressed as a product of the coefficients @xmath198 . the goal of next subsection is to find recursion relations for the @xmath198 s in order to compute them and subsequently analyse their product @xmath157 in eq . ( [ eq : znh ] ) in the large @xmath33 limit . in general , for orthogonal polynomials ( with any reasonable weight function ) , one can write a recursion relation of the form : @xmath216 where @xmath217 and @xmath218 are real coefficients . this relation comes from the fact that @xmath219 and that @xmath220 for any polynomial @xmath221 of degree strictly smaller than @xmath191 . the coefficients @xmath217 and @xmath218 are functions of @xmath153 and @xmath164 , i.e. @xmath222 and @xmath223 . + let us first demonstrate that the coefficients @xmath218 and @xmath217 can be expressed in terms of @xmath198 s . from eq . , we get : @xmath224 . on the other hand , we have @xmath225 . therefore @xmath226 , thus @xmath227 . + from eq . again , we also get : @xmath228 . by definition we have @xmath229 . after integrating by part we find : @xmath230 . the last equality follows from the definition of @xmath198 . as @xmath231 , we have indeed @xmath232 . however , @xmath233 as @xmath234 is a polynomial of degree strictly smaller than @xmath191 ( as @xmath235 ) . therefore @xmath236 . + combining these relations , @xmath218 and @xmath217 are then given by : @xmath237 iterating the recursion relation @xmath238 starting from @xmath239 , we can write @xmath198 in terms of @xmath240 s : @xmath241 substituting this result in eq . ( [ eq : znh ] ) , we see that the partition function @xmath157 can be entirely expressed in terms of a product over the @xmath218 s . thus , if we can determine @xmath218 s , we can evaluate the partition function explicitly . + thus the next task is to determine the @xmath218 s . to do this , we will first derive a set of coupled recursion relations for @xmath218 s and @xmath217 s . we have @xmath242 where we have used the fact that @xmath243 ( which follows from the observation that @xmath244 is a polynomial of degree strictly less than @xmath191 and hence orthogonality dictates that it is identically zero ) . on the other hand , using eq . we find : @xmath245 . therefore , @xmath246 we can eliminate @xmath198 from the relations and and get a pair of coupled nonlinear recursion relations for @xmath218 and @xmath217 . using eq . for @xmath191 and @xmath247 , and as @xmath227 , we find @xmath248 . , we also find @xmath249 . finally , we then get our desired recursion relations : @xmath250 it is easy to show by induction that the two relations and together with the initial conditions @xmath251 , @xmath252 and @xmath253 uniquely determine @xmath218 and @xmath217 . the additional initial condition @xmath254 is enough to determine @xmath198 as given in eq . . + by definition , @xmath255 is a polynomial of degree @xmath20 with dominant coefficient @xmath74 . thus @xmath256 . therefore @xmath257 . we also have @xmath258 as the recursion relation in eq . must reduce for @xmath259 to @xmath260 , i.e. , @xmath261 . moreover we get from eq . @xmath262 $ ] . we now have @xmath251 , @xmath253 and @xmath254 , we can thus determine @xmath252 from eq . for @xmath259 : @xmath263 . thus , the initial conditions can be summarized as : @xmath264 as @xmath202 , we can explicitly compute the functions @xmath217 , @xmath218 and @xmath198 . as mentioned above , in the limit @xmath202 the polynomials @xmath265 are indeed simply the hermite polynomials . hence everything can then be computed explicitly in this case . we have @xmath266 and @xmath267 . then , by recurrence it is easy to show that : @xmath268 finally , using eq . we get @xmath269 and thus ( see eq . ) @xmath270 which could also have been computed directly using selberg s integral . we recover the normalization @xmath271 in eq . . in the previous section we derived a pair of coupled recursion equations and with initial conditions given in eq . that determines uniquely @xmath218 , @xmath217 and thus @xmath198 and subsequently @xmath157 via . however , these equations are hard to solve explicitly for general @xmath191 and @xmath164 ( apart from the case @xmath272 as shown in the previous subsection ) . in this section , we derive an approximate solution for @xmath157 and hence for the cdf , in the large @xmath33 limit where @xmath33 is the number of eigenvalues of the matrix @xmath2 and we will see that this solution for cdf is valid only for @xmath273 , i.e. , it only describes the right tail of the probability distribution . we have seen in section 2 that for large @xmath33 and fixed @xmath153 , the maximal eigenvalue typically scales as @xmath274 . we are going to work on this scale , hence in the definition of the maximal eigenvalue cdf in and , we will set @xmath275 with @xmath276 . we will then work in the limit of large @xmath33 with @xmath96 fixed . with this scaling , the operation @xmath277 defined in eq . for polynomials depends on @xmath33 ( since the upper limit of integration in is now @xmath278 ) . the coefficients @xmath218 , @xmath217 and @xmath198 , for a given @xmath191 , are also now implicitly functions of @xmath33 . we can make an expansion of these parameters for large @xmath33 and fixed @xmath191 . the dominant order will be given by the @xmath272 case ( as in previous subsection ) as @xmath279 as @xmath102 for fixed @xmath96 . in this section , we want to determine the first correction to this dominant term . however the partition function @xmath157 and thus the cumulative distribution ( cdf ) of @xmath35 is a product of all the @xmath198 s for @xmath280 . our expansion will thus give us the behaviour of the cdf of @xmath35 for large @xmath33 only if we can show that it is valid not only for fixed @xmath191 but also for @xmath191 of order up to @xmath33 . this constraint of validity will be discussed later . we will see that this expansion is actually valid only on the right of the mean value , i.e. for @xmath281 or equivalently @xmath282 . this method allows us to describe the right tail of the large deviation of the distribution of @xmath283 , i.e. @xmath284 with @xmath285 and @xmath286 . let us start by expanding the initial conditions for large @xmath33 . with the scaling @xmath287 with @xmath276 , equations become ( for @xmath288 ) for large @xmath33 : @xmath289 the dominant term for large @xmath33 corresponds to the limit @xmath202 ( see previous section ) : @xmath290 . therefore , ignoring the exponentially small correction for large @xmath33 leads to @xmath291 and @xmath292 . we want to determine the first correction for large @xmath33 . let us make the following ansatz : @xmath293 where @xmath294 and @xmath295 are expected to be polynomials of @xmath296 . this will be valid as long as @xmath297 where @xmath275 . the initial conditions in eq . give : @xmath298 let us replace @xmath218 and @xmath217 in the recursion equation and by the ansatz in eq . . we see that @xmath299 and @xmath300 are actually negligible in the equation giving @xmath301 , as they are exponentially smaller than the @xmath240 s . we thus get recursion relations for the coefficients @xmath302 and @xmath303 : @xmath304 @xmath198 is the product of the @xmath240 , it can thus be expressed in terms of @xmath305 . as @xmath157 is a function of the @xmath198 s , it is thus a function of @xmath254 and the @xmath218 s for @xmath306 , we can use the ansatz only if @xmath307 for all @xmath308 . in this case we can write : @xmath309\nonumber \\ & \approx & \ln h_n(\infty,\alpha ) + \left [ -\frac{1}{2 y \sqrt{\pi } } + \sum_{k=1}^n \frac{2 \alpha c_k(y,\alpha)}{k } \right]\ : e^{-n \alpha z^2 } \end{aligned}\ ] ] where @xmath287 , and @xmath310 is thus given by @xmath311\ : e^{-n \alpha z^2 } \ ] ] the partition function only depends on the @xmath305 s . we want now to solve the recursion relation for the @xmath305 s in eq . . we do not need to determine the @xmath312 s . let us define @xmath314 and @xmath315 such that : @xmath316 @xmath314 is large for large @xmath33 ( proportional to @xmath33 ) . @xmath317 depends only on @xmath318 . this can easily be shown by recurrence with initial condition @xmath319 ( as @xmath320 is given by eq . ) . the recursion for the @xmath313 s becomes @xmath321 with initial condition @xmath322 and @xmath319 . by recurrence again , it is easy to show that @xmath317 is a polynomial of @xmath314 of degree @xmath323 , with leading coefficient @xmath324 . let us consider the generating function of the @xmath325 : @xmath326 the @xmath317 are obtained from @xmath327 by a contour integration : @xmath328 where @xmath329 is a contour in the complex plane that encircles the origin @xmath330 in such a way that all singularities of @xmath331 ( as a function of @xmath332 for fixed @xmath314 ) are outside the contour . from eq . and the definition of @xmath327 , we deduce that @xmath331 satisfies the following partial differential equation : @xmath333 f\ ] ] this equation together with the requirement that @xmath334 as @xmath335 ( as @xmath336 ) determines uniquely @xmath331 . we find : @xmath337 @xmath317 is given by the contour integral in eq . where the contour @xmath329 encircles @xmath330 in such a way that @xmath338 and @xmath339 are outside of the contour . let us compute @xmath317 with @xmath340 for large @xmath33 , fixed @xmath96 and @xmath341 with fixed @xmath342 . we have @xmath343 where @xmath344 is of order one for large @xmath33 when @xmath341 with @xmath345 of order one . for fixed @xmath346 and for large @xmath33 , the contour integral can thus be computed using a saddle point method . the integral will be dominated by the neighbourhood of @xmath347 such that : @xmath348 there exists a real solution for @xmath347 iff @xmath349 ie @xmath350 . we want this condition to be satisfied for all @xmath308 , therefore we must have @xmath351 , i.e. @xmath352 . our method can only describe the regime @xmath352 , ie @xmath281 , which corresponds to the right tail of the distribution @xmath353 ( region where @xmath35 is above its mean value ) . let us call the critical point @xmath354 , i.e. @xmath355 . for @xmath356 , there are two real solutions @xmath357 $ ] . the contour @xmath329 must encircle @xmath20 but not @xmath74 and @xmath358 , therefore we impose @xmath359 . this implies @xmath360 = \frac{n \alpha z^2}{2 n}\left [ 1-\sqrt{1-\frac{2 n}{n \alpha z^2 } } \right]^2 \nonumber \\ & = & -1+\frac{\xi } { n}\left [ 1 - \sqrt{1-\frac{2 n}{\xi } } \right ] = \frac{\xi}{2 n}\left [ 1-\sqrt{1-\frac{2 n}{\xi } } \right]^2\end{aligned}\ ] ] thus @xmath361 -2 c \ln\left\{\sqrt{\frac { \alpha z^2}{2 c}}-\sqrt{\frac { \alpha z^2}{2 c}-1}\right\}\ ] ] the saddle point gives for large @xmath33 : @xmath362 where @xmath363 in this subsection we have found , as given in eq . , the expression of @xmath317 and thus the solution @xmath364 ( with @xmath365 ) of the recursion relation for large @xmath33 and @xmath341 with fixed @xmath342 . we have also shown that the validity of our approximation is the regime @xmath356 with @xmath366 . we want to compute for large @xmath33 and for @xmath356 ( with the scaling @xmath275 for large @xmath33 ) the cdf @xmath367 . using eq . and the definition of @xmath315 in eq . , we get : @xmath368 therefore we need to compute @xmath369 for large @xmath33 and @xmath340 ( with fixed @xmath153 and @xmath96 ) . for that purpose , we do not actually need to use the approximate expression of @xmath315 for @xmath191 of order @xmath33 that we derived in the previous subsection . we can use the formal expression of @xmath315 as a contour integral @xmath370 and compute the sum over @xmath191 before computing the integral by saddle point method . in particular we have : @xmath371 the function @xmath372 has indeed no singularity at the origin , its integral is thus zero . on the other hand , we have : @xmath373 where @xmath374 is a hypergeometric function : @xmath375 with @xmath376 . for large @xmath33 , we find : @xmath377 therefore we get for large @xmath33 @xmath378 equivalently we can write : @xmath379 where @xmath380 ( see ) . for large @xmath33 , the saddle point method thus gives @xmath381 where @xmath347 is given in eq . with @xmath382 : @xmath383 = \frac { \alpha z^2}{2 } \left [ 1-\sqrt{1-\frac{2 } { \alpha z^2 } } \right]^2\ ] ] thus we find @xmath384 with @xmath385 -2 \ln\left\{\sqrt{\frac { \alpha z^2}{2 } } -\sqrt{\frac { \alpha z^2}{2 } -1}\right\}\ ] ] therefore @xmath386 as @xmath387 for @xmath388 , i.e. @xmath389 , the first term in the parenthesis can be neglected for large @xmath33 : @xmath390 with @xmath287 . therefore we get the expression of the right tail of the cdf of @xmath35 : @xmath391 where the rate function @xmath392 is given by : @xmath393 + \ln\left\{\sqrt{\frac { \alpha z^2}{2 } } -\sqrt{\frac { \alpha z^2}{2 } -1}\right\ } \label{psi+z}\ ] ] we have thus found @xmath394 deriving with respect to @xmath395 we get an equivalent of the probability density function of @xmath35 for large @xmath33 : @xmath396 we thus recover the dominant order for large @xmath33 ( given by @xmath89 ) that was derived in @xcite by a coloumb gas method , but in addition this method also provides us with the first correction term to the dominant order . + let us now see how this precise right tail large deviation behavior in behaves when @xmath397 from the right . using the leading expansion of @xmath87 around @xmath103 in and setting @xmath398 , i.e. , @xmath399 one finds from @xmath400 on the other hand , using the boundary condition @xmath401 as @xmath71 in the definition of the tracy - widom function , one can easily derive the precise leading asymptotics of its right tail , @xmath402 as @xmath403 . thus , our right large deviation function for small argument ( when the fluctuation of @xmath35 to the right of its mean value @xmath404 is of @xmath54 ) in , matches smoothly with the precise right tail of the tracy - widom distribution @xmath405 . in this section , we provide an elementary derivation of the tracy - widom law for the gue based on simple scaling analysis of the recursion relations derived in section 2 in the vicinity of the critical point @xmath406 . this derivation , in our opinion , is mathematically simpler than the original derivation by tracy and widom @xcite as it avoids the sophisticated asymptotic analysis of fredholm determinants . the derivation of the painlev ii equation from the scaling analysis of recursion relations that we follow here is similar in spirit ( though rather different in details ) to the analysis of the partition function in the two dimensional yang - mills theory on a sphere by gross and matytsin @xcite . let us recall that the tracy - widom distribution @xmath405 is defined as @xmath407 where @xmath408 satisfies the painlev ii equation with the boundary condition @xmath409 from eq . , it follows that @xmath410 . + we want to show that for large @xmath33 the probabality of small typical fluctuations of @xmath35 around its mean value @xmath411 are described by the tracy - widom distribution . for this , we need to first estimate how do these typical fluctuations scale with @xmath33 for large @xmath33 . in the vicinity of the mean @xmath412 , let us then write @xmath413 where @xmath414 is the scale of the typical fluctuation and the random variable @xmath332 has an @xmath33 independent distribution for large @xmath33 . evidently the exponent @xmath415 ( so that the fluctuation is less than the mean ) whose precise value is yet to be determined . note also that since @xmath35 always appears in the distribution @xmath416 in the scaling combination @xmath165 ( see eq . ) , we have chosen the prefactor of the fluctuation term as @xmath417 which then ensures that the random variable @xmath332 describing the typical fluctuation is also independent of @xmath153 . one way to estimate the exponent @xmath418 is from the right large deviation tail computed in in the previous section . the right tail in describes the probability of large fluctuations of @xmath80 to the right of the mean . assuming that the right tail behaviour continues to hold even for fluctuations smaller than @xmath104 , we substitute @xmath419 in . this gives @xmath420 valid for @xmath421 , @xmath332 large . assuming that this continues to hold even for not so large @xmath332 ( so that it even captures the tail of the distribution of typical small fluctuations ) , we expect that in terms of this rescaled variable @xmath332 , the tail of the distribution in is independent of @xmath33 for large @xmath33 . clearly , for this to happen the power of @xmath33 must be zero both inside the exponential as well as in the prefactor in , indicating that @xmath422 , thus @xmath423 . hence , the correct scaling describing typical fluctuations , for large @xmath33 , is given by @xmath424 where @xmath332 has an @xmath33 independent distribution that we now have to compute and show that it is given by the tracy - widom function @xmath405 . the meaning of _ double scaling limit _ is now clear . it simply says the following . consider the cdf @xmath425 or rather its logarithm ( for convenience ) @xmath426 . in general , it is a function of two variables @xmath164 and @xmath33 . however , in the vicinity of the mean @xmath427 , if one takes the limit @xmath428 and @xmath102 , but keeping the scaling combination @xmath429 fixed , this function of two variables collapses into a function of the single scaled variable @xmath332 @xmath430 and our job is to to show that this scaling function @xmath431 where @xmath405 is the tracy - widom function defined in . in other words , we want to show that @xmath432 where @xmath408 satisfies the painlev ii equation . our starting point is the definition of the cdf in . from eq . it is easy to see that the partition function @xmath157 satisfies the recursion @xmath433 taking logarithm and using the definition in we get @xmath434 in the double scaling limit , we will now substitute the anticipated scaling form in for the logarithm of the cdf on the left hand side of . but we need to first evaluate @xmath435 . replacing @xmath33 by @xmath436 in and expanding for large @xmath33 , with @xmath437 fixed , we get @xmath438 substituting this result in we get for the left hand side @xmath439 from eq . and , we get for large @xmath33 @xmath440 as @xmath441 ( see eq . ) . this suggests that in this scaling limit , @xmath442 must scale as @xmath443 . more precisely , this leads us to the following large @xmath33 expansion of @xmath444 in the double scaling limit @xmath445 where @xmath446 and @xmath447 , @xmath448 etc . describing the higher order scaling corrections . thus , if we can now determine the first subleading scaling function @xmath449 in the expansion of @xmath450 , then we can determine @xmath451 by integrating @xmath449 twice . so , our next task is to determine @xmath449 by analysing the recursion relations and ( setting @xmath382 ) in the double scaling limit . we now know , from , how @xmath452 behaves in the scaling limit with the scaling combination @xmath453 fixed . in order to analyse the recursion relations and , we also need to know how @xmath454 behaves in this scaling limit . in order to match the leading @xmath33 behavior of @xmath452 with @xmath332 fixed in , it is not difficult to see that to leading order for large @xmath33 , @xmath454 must have the following scaling behaviour @xmath455 where @xmath456 is the leading order scaling function . let us first evaluate the difference @xmath457 that appears in . replacing @xmath33 by @xmath458 in , expanding for large @xmath33 , we get @xmath459 it rests to evaluate the partial derivative @xmath460 in . from the definition of the scaling variable @xmath461 , it follows , using chain rule , @xmath462 finally substituting and in ( with @xmath382 ) we get , @xmath463 matching the leading order @xmath464 term gives a relation between @xmath456 and @xmath449 : @xmath465 , i.e. , @xmath466 with @xmath467 a constant . from and the fact that when @xmath468 , @xmath75 , it follows that both the scaling functions @xmath449 and @xmath456 must vanish as @xmath75 . thus the constant @xmath469 and we have , for all @xmath332 , @xmath470 having determined the relation @xmath471 , we need one more relation between these two scaling functions in order to determine them individually . this will now be done by substituting the scaling solutions for @xmath452 ( given in ) and @xmath454 ( given in ) into the remaining recursion relation . to analyse ( seeting @xmath382 ) , we need to evaluate the derivative @xmath472 . from the definition of the scaling variable , @xmath461 , it follows , that @xmath473 . we then use the chain rule and to express @xmath474+ ... \ ] ] again replacing @xmath33 by @xmath436 in and expanding for large @xmath33 , keeping @xmath461 fixed , we get @xmath475+ ... \ ] ] and similarly from @xmath476 substituting these results in and matching the leading order term ( @xmath477 ) , we get the desired second relation between @xmath449 and @xmath456 @xmath478 eliminating @xmath456 by using @xmath479 we finally get a single closed equation for @xmath449 @xmath480 let us write @xmath481 eq . then becomes an equation for @xmath482 : @xmath483 let @xmath484 . then becomes @xmath485 which can simply be integrated to give @xmath486 where @xmath135 is an arbitrary constant . hence we have @xmath487 from the boundary condition @xmath488 as @xmath75 ( which follows from ) , it follows using @xmath489 that @xmath490 as @xmath75 . taking @xmath75 in then fixes the value of the constant @xmath491 . finally , from , we have @xmath492 where @xmath482 satisfies the painlev ii equation @xmath493 to fix the boundary condition for @xmath482 , we again invoke the matching with the right large deviation tail in . taking logarithm of with @xmath494 and using @xmath495 we find that @xmath496 hence @xmath497 and consequently as @xmath75 @xmath498 finally integrating @xmath499 twice and using the appropriate boundary condition as @xmath75 , we get @xmath500 where @xmath482 satisfies the painlev ii equation with the boundary condition . comparing to , we have thus shown that the scaling function @xmath501 where @xmath405 is the tracy - widom function ( @xmath11 ) . this then constitutes our derivation for the tracy - widom distribution for the gue ( @xmath11 ) . + using recursion relations that we have derived for orthogonal polynomials on a semi - infinite interval , we have shown that the large @xmath33 asymptotics of the distribution of the maximal eigenvalue of a gaussian random matrix ( from the gue ) is described in the double scaling regime by the painlev ii equation . similar recursion relations for other orthogonal polynomials leading to different kinds of painlev equations have also been established in a number of papers , see @xcite and references therein , in particular @xcite . in this paper , we have provided a rather simple and pedestrian derivation of the tracy - widom law for the distribution of the largest eigenvalue of a gaussian unitary random matrix . this was done by suitably adapting a method of orthogonal polynomials developed by gross and matytsin @xcite in the context of two dimensional yang - mills theory . our derivation requires just elementary asymptotic scaling analysis of a pair of coupled nonlinear recursion relations . strictly in the @xmath102 limit , there is a @xmath109-rd order phase transition in the form of the probability distribution of @xmath35 as @xmath35 crosses its mean value from left to right . for finite but large @xmath33 , the two regions are connected by a smooth crossover function and the shape of this crossover function is precisely the tracy - widom distribution that describes the ` typical ' small fluctuations of @xmath35 around its mean . the ` atypical ' large fluctuations to the left and right of the mean are described by large deviation tails that correspond to the two ` phases ' across this phase transition . in qualitative analogy to the two - dimensional yang - mills theory , the left ( left large deviation ) and the right ( right large deviation ) phases correspond respectively to the ` strong ' and ` weak ' coupling phases of the two - dimensional qcd . apart from the simple derivation of the tracy - widom gue law , we were also able to compute the precise right large deviation tail of the maximal eigenvalue distribution that is not described by the tracy - widom distribution . in the language of qcd , this right tail corrections are similar to the non - perturbative ( in @xmath1 expansion ) corrections to the qcd partition function in @xmath27-d @xcite . one drawback of our method is that it works only for the gue case ( with dyson index @xmath11 ) . it would be challenging to see if this method can be extended / generalized to derive the tracy - widom law for the other two gaussian ensembles , namely the goe ( @xmath10 ) and the gse ( @xmath12 ) . in addition , this method for @xmath11 should be useful to compute the largest eigenvalue distribution for other non - gaussian matrix ensembles , such as the laguerre ( wishart matrices ) or the jacobi ensembles . s. n. majumdar , random matrices , the ulam problem , directed polymers and growth models , and sequence matching " in _ complex systems _ ( les houches lecture notes ed . by j .- bouchaud , m. mzard , and j. dalibard ) ( elsevier , amsterdam ) ( 2007 ) 179 - 216 . g. amir , i. corwin and j. quastel , `` probability distribution of the free energy of the continuum directed random polymer in 1 + 1 dimensions '' , _ communications on pure and applied mathematics _ , * 64 * , 466 - 537 ( 2011 ) . m. crescimanno , s. g. naculich and h. j. schnitzer , `` evaluation of the free energy of two - dimensional yang - mills theory '' , _ phys . d _ , * 54 * , issue 2 , 1809 - 1813 ( 1996 ) , ` arxiv : hep - th/9601104 ` ( 1996 ) .

given positive integers and where , we give a minimum-degree condition that ensures a perfect matching in a-uniform hypergraph . this condition is best possible and improves on work of pikhurko who gave an asymptotically exact result . our approach makes use of the absorbing method , and builds on work in , where we proved the result for divisible by .

1210.7359

a central question in graph theory is to establish conditions that ensure a ( hyper)graph @xmath5 contains some spanning ( hyper)graph @xmath6 . of course , it is desirable to fully characterize those ( hyper)graphs @xmath5 that contain a spanning copy of a given ( hyper)graph @xmath6 . theorem @xcite characterizes those graphs with a perfect matching . ( a _ perfect matching _ in a ( hyper)graph @xmath5 is a collection of vertex - disjoint edges of @xmath5 which cover the vertex set @xmath7 of @xmath5 . ) however , for some ( hyper)graphs @xmath6 it is unlikely that such a characterization exists . indeed , for many ( hyper)graphs @xmath6 the decision problem of whether a ( hyper)graph @xmath5 contains @xmath6 is np - complete . for example , in contrast to the graph case , the decision problem whether a @xmath3-uniform hypergraph contains a perfect matching is np - complete for @xmath0 ( see @xcite ) . thus , it is desirable to find sufficient conditions that ensure a perfect matching in a @xmath3-uniform hypergraph . given a @xmath3-uniform hypergraph @xmath5 with an @xmath1-element vertex set @xmath8 ( where @xmath9 ) we define @xmath10 to be the number of edges containing @xmath8 . minimum @xmath1-degree @xmath11 _ of @xmath5 is the minimum of @xmath10 over all @xmath1-element sets @xmath8 of vertices in @xmath5 . clearly @xmath12 is the number of edges in @xmath5 . we also refer to @xmath13 as the _ minimum vertex degree _ of @xmath5 and @xmath14 the _ minimum codegree _ of @xmath5 . over the last few years there has been a strong focus in establishing minimum @xmath1-degree thresholds that force a perfect matching in a @xmath3-uniform hypergraph . see @xcite for a survey on matchings ( and hamilton cycles ) in hypergraphs . in particular , rdl , ruciski and szemerdi @xcite determined the minimum codegree threshold that ensures a perfect matching in a @xmath3-uniform hypergraph on @xmath15 vertices for all @xmath16 . the threshold is @xmath17 , where @xmath18 depends on the values of @xmath15 and @xmath3 . this improved bounds given in @xcite . less is known about minimum vertex degree thresholds that force a perfect matching . one of the earliest results on perfect matchings was given by daykin and hggkvist @xcite , who showed that a @xmath3-uniform hypergraph @xmath5 on @xmath15 vertices contains a perfect matching provided that @xmath19 . hn , person and schacht @xcite determined , asymptotically , the minimum vertex degree that forces a perfect matching in a @xmath20-uniform hypergraph . khn , osthus and treglown @xcite and independently khan @xcite made this result exact . khan @xcite has also determined the exact minimum vertex degree threshold for @xmath4-uniform hypergraphs . for @xmath21 , the precise minimum vertex degree threshold which ensures a perfect matching in a @xmath3-uniform hypergraph is not known . the situation for @xmath1-degrees where @xmath22 is also still open . hn , person and schacht @xcite provided conditions on @xmath23 that ensure a perfect matching in the case when @xmath24 . these bounds were subsequently lowered by markstrm and ruciski @xcite . alon et al . @xcite gave a connection between the minimum @xmath1-degree that forces a perfect matching in a @xmath3-uniform hypergraph and the minimum @xmath1-degree that forces a _ perfect fractional matching_. as a consequence of this result they determined , asymptotically , the minimum @xmath1-degree which forces a perfect matching in a @xmath3-uniform hypergraph for the following values of @xmath25 : @xmath26 , @xmath27 , @xmath28 , @xmath29 , and @xmath30 . pikhurko @xcite showed that if @xmath31 and @xmath5 is a @xmath3-uniform hypergraph whose order @xmath15 is divisible by @xmath3 then @xmath5 has a perfect matching provided that @xmath32 . this result is best possible up to the @xmath33-term ( see the constructions in @xmath34 below ) . in this paper we make pikhurko s result exact . in order to state our main result , we need some more definitions . fix a set @xmath35 of @xmath15 vertices . given a partition of @xmath35 into non - empty sets @xmath36 , let @xmath37 ( @xmath38 ) denote the family of all @xmath3-element subsets of @xmath35 that intersect @xmath39 in an odd ( even ) number of vertices . ( notice that the ordering of the vertex classes @xmath40 is important . ) when it is clear from the context , we write , for example , @xmath41 . define @xmath42 to be the @xmath3-uniform hypergraph with vertex set @xmath43 and edge set @xmath44 . note that the complement @xmath45 of @xmath46 has edge set @xmath47 . suppose @xmath48 such that @xmath3 divides @xmath15 . define @xmath34 to be the collection of the following hypergraphs : @xmath34 contains all hypergraphs @xmath49 where @xmath50 is odd . further , if @xmath51 is odd then @xmath34 also contains all hypergraphs @xmath52 where @xmath50 is even ; if @xmath51 is even then @xmath34 also contains all hypergraphs @xmath52 where @xmath50 is odd . it is easy to see that no hypergraph in @xmath34 contains a perfect matching . indeed , first assume that @xmath50 is even and @xmath51 is odd . since every edge of @xmath53 intersects @xmath39 in an odd number of vertices , one can not cover @xmath39 with an odd number of disjoint odd sets . similarly @xmath54 does not contain a perfect matching if @xmath50 is odd and @xmath51 is even . finally , if @xmath50 is odd then since every edge of @xmath55 intersects @xmath39 in an even number of vertices , @xmath56 does not contain a perfect matching . given @xmath57 such that @xmath58 define @xmath59 to be the maximum of the minimum @xmath1-degrees among all the hypergraphs in @xmath34 . for example , it is not hard to see that @xmath60 the following is our main result . [ mainthm ] let @xmath61 such that @xmath16 and @xmath62 . then there exists an @xmath63 such that the following holds . suppose @xmath5 is a @xmath3-uniform hypergraph on @xmath64 vertices where @xmath3 divides @xmath15 . if @xmath65 then @xmath5 contains a perfect matching . in @xcite , we proved theorem [ mainthm ] in the case when @xmath4 divides @xmath3 . independently to this , czygrinow and kamat @xcite proved theorem [ mainthm ] in the case when @xmath66 and @xmath67 . to prove theorem [ mainthm ] we use several ideas and results from @xcite . in particular , the so - called ` extremal ' case of theorem [ mainthm ] was proved in @xcite for _ all _ values of @xmath3 . however , in some parts of the proof of the ` non - extremal ' case we use a very different approach to that in @xcite . we discuss this in more detail in section [ presec ] . as explained before , the minimum @xmath1-degree condition in theorem [ mainthm ] is best possible . theorem [ mainthm ] and together give the aforementioned result of rdl , ruciski and szemerdi @xcite . in general , the precise value of @xmath59 is unknown because it is not known what value of @xmath50 maximizes the minimum @xmath1-degree of @xmath68 ( or @xmath69 ) . ( see @xcite for a discussion on this . ) however , in @xcite we gave a tight upper bound on @xmath70 . given a set @xmath71 and @xmath72 , we write @xmath73 for the set of all @xmath74-element subsets ( @xmath74-subsets , for short ) of @xmath71 . given a set @xmath8 and an element @xmath75 , we often write @xmath76 as @xmath77 and @xmath78 as @xmath79 . let @xmath80 . a @xmath3-uniform hypergraph @xmath5 consists of a set of _ vertices _ @xmath7 and a set of _ edges _ @xmath81 . so in the case when @xmath82 , @xmath83 . ( the notion of a @xmath84-uniform hypergraph will be used in section [ sectionnon ] . ) let @xmath61 . suppose @xmath85 is a @xmath3-uniform hypergraph . let @xmath86 be an @xmath1-subset of @xmath7 . often we write it as @xmath87 ( i.e. we drop the brackets and commas ) , or simply @xmath88 . given @xmath89 , we write @xmath90 or @xmath91 to denote the _ neighborhood of @xmath88 _ , that is , the family of those @xmath92-subsets of @xmath7 which , together with @xmath88 , form an edge in @xmath5 . then @xmath93 . given a vertex @xmath94 , we define @xmath95 and @xmath96 analogously . we denote the _ complement of @xmath5 _ by @xmath97 . that is , @xmath98 . given a set @xmath99 , @xmath100 $ ] denotes the @xmath3-uniform subhypergraph of @xmath5 _ induced by @xmath39 _ , namely , @xmath100 : = ( a , e(h)\cap \binom{a}{k})$ ] . given @xmath101 , we define @xmath102 : = ( v(h ) , b)$ ] . let @xmath40 be sets and let @xmath103 be a positive real . let @xmath104 denote the _ symmetric difference _ of @xmath39 and @xmath105 . we write @xmath106 if @xmath107 . let @xmath108 . suppose that @xmath5 and @xmath109 are @xmath3-uniform hypegraphs on @xmath15 vertices . we say that @xmath5 is _ @xmath110-close to @xmath109 _ , and write @xmath111 , if @xmath5 becomes a copy of @xmath109 after adding and deleting at most @xmath112 edges . more precisely , @xmath5 is @xmath110-close to @xmath109 if there is an isomorphic copy @xmath113 of @xmath5 such that @xmath114 and @xmath115 . given a graph @xmath116 , @xmath117 and @xmath118 , we denote by @xmath119 the number of vertices @xmath120 such that @xmath121 . given disjoint @xmath122 we let @xmath123 denote the number of edges in @xmath116 with one endpoint in @xmath39 and one endpoint in @xmath105 . further , we let @xmath124 denote the complete bipartite graph with vertex classes @xmath39 and @xmath105 . we will often write @xmath125 to mean that we can choose the constants @xmath126 from right to left . more precisely , there are increasing functions @xmath127 and @xmath128 such that , given @xmath129 , whenever we choose some @xmath130 and @xmath131 , all calculations needed in our proof are valid . hierarchies with more constants are defined in the obvious way . throughout the paper we omit floors and ceilings whenever this does not affect the argument . suppose that @xmath133 such that @xmath134 . let @xmath40 be a partition of a set of @xmath15 vertices . recall that @xmath42 is the @xmath3-uniform hypergraph with vertex set @xmath135 and edge set @xmath44 , and its complement @xmath56 has edge set @xmath47 . ( note that @xmath136 has edge set @xmath39 and @xmath137 has edge set @xmath105 . ) when @xmath138 and @xmath139 , we simply denote @xmath42 by @xmath140 , and @xmath56 by @xmath141 . following the ideas of rdl , ruciski and szemerdi @xcite , we define _ absorbing sets _ as follows : given a @xmath3-uniform hypergraph @xmath5 , a set @xmath142 is called an _ absorbing set for @xmath143 _ , if both @xmath144 $ ] and @xmath145 $ ] contain perfect matchings . in this case , if the matching covering @xmath8 is @xmath146 , we also say _ @xmath146 absorbs @xmath147_. when considering @xmath1-degree together with @xmath148-degree for some @xmath149 , the following proposition is very useful ( the proof is a standard counting argument , which we omit ) . [ prop : deg ] let @xmath150 and @xmath5 be a @xmath3-uniform hypergraph . if @xmath151 for some @xmath152 , then @xmath153 . the following two results are applied in section [ lastlemma ] . given an @xmath74-uniform hypergraph @xmath154 with two distinct vertices @xmath155 , define @xmath156 as the family of @xmath157-subsets @xmath158 such that @xmath159 and either @xmath160 and @xmath161 , or @xmath162 and @xmath163 . note that @xmath164 , where @xmath165 is the complement of @xmath6 . [ lem : kk ] given any @xmath166 and @xmath167 there exists an @xmath63 such that the following holds . let @xmath168 be an @xmath74-uniform hypergraph on @xmath64 vertices with edge density @xmath169 $ ] . then @xmath170 in particular , there exists two vertices @xmath171 such that @xmath172 . if @xmath173 then the second assertion is trivial . if @xmath174 , the second assertion follows by an averaging argument : if the first assertion holds , then there exist two vertices @xmath175 and @xmath176 in @xmath35 such that latexmath:[\[\begin{aligned } since @xmath15 is sufficiently large . we prove the first assertion by double counting and the kruskal katona theorem . let @xmath103 denote the left - hand side of , that is , the number of @xmath157-subsets @xmath178 that contains two labeled vertices @xmath179 such that exactly one of @xmath180 and @xmath181 is in @xmath182 . for @xmath183 , let @xmath184 denote the number of @xmath157-subsets of @xmath35 that span exactly @xmath185 edges of @xmath6 . it is easy to see that @xmath186 thus , @xmath187 a version of the kruskal katona theorem by lovsz @xcite states that given a family @xmath188 of @xmath3-element sets , if @xmath189 for some real number @xmath75 , then the size of its shadow @xmath190 is greater than @xmath191 . this implies that if an @xmath74-uniform hypergraph has at most @xmath192 edges , then @xmath193 , the number of the @xmath157-cliques in the hypergraph , is at most @xmath194 . since @xmath195 , we derive that @xmath196 . substituting this into , we have that @xmath197 since @xmath198 $ ] , we have that @xmath199 using the fact that @xmath200 , this minimum is at least @xmath201 . as @xmath15 is sufficiently large , this gives that @xmath202 [ evensum ] for @xmath166 , @xmath203 and @xmath204 , @xmath205 throughout the proof we assume that @xmath206 . we observe that @xmath207 since @xmath208 @xmath209 we have @xmath210 the two desired equalities follow immediately . most of the paper is devoted to proving the following result . [ nonexthm1 ] let @xmath211 and @xmath61 such that @xmath16 and @xmath2 . then there exist @xmath212 and @xmath63 such that the following holds suppose that @xmath5 is a @xmath3-uniform hypergraph on @xmath64 vertices . if @xmath213 then @xmath5 is @xmath110-close to @xmath140 or @xmath141 , or @xmath5 contains a matching @xmath146 of size @xmath214 that absorbs any set @xmath215 such that @xmath216 with @xmath217 . we prove theorem [ nonexthm1 ] in section [ sectionnon ] . once theorem [ nonexthm1 ] is proven , we can derive theorem [ mainthm ] in the same way as described in @xcite . for completeness , we include the proof here . * theorem 4.1)[extthm ] given @xmath218 , there exist @xmath219 and @xmath220 such that the following holds . suppose that @xmath5 is a @xmath3-uniform hypergraph on @xmath221 vertices such that @xmath15 is divisible by @xmath3 . if @xmath222 and @xmath5 is @xmath110-close to @xmath140 or @xmath141 , then @xmath5 contains a perfect matching . theorem [ extthm ] ensures a perfect matching when our hypergraph @xmath5 is ` close ' to one of the ` extremal ' hypergraphs @xmath140 and @xmath141 . when @xmath5 is non - extremal we will apply the following result of markstrm and ruciski @xcite to ensure an ` almost ' perfect matching in @xmath5 . * lemma 2)[marklemma ] for each integer @xmath16 , every @xmath223 and every @xmath224 there exists an @xmath63 such that the following holds . suppose that @xmath5 is a @xmath3-uniform hypergraph on @xmath225 vertices such that @xmath226 then @xmath5 contains a matching covering all but at most @xmath227 vertices . in @xcite , markstrm and ruciski only stated theorem [ marklemma ] for @xmath228 . in fact , their proof works for all values of @xmath1 such that @xmath223 . in the case when @xmath229 , we need a result of rdl , ruciski and szemerdi ( * ? ? * fact 2.1 ) : suppose @xmath5 is a @xmath3-uniform hypergraph on @xmath15 vertices . if @xmath230 , then @xmath5 contains a matching covering all but at most @xmath231 vertices in @xmath5 . * proof of theorem [ mainthm ] . * let @xmath232 be as in theorem [ extthm ] and @xmath233 be as in theorem [ nonexthm1 ] . that is , @xmath234 assume that @xmath235 . suppose that @xmath15 is sufficiently large and @xmath3 divides @xmath15 . consider any @xmath3-uniform hypergraph @xmath5 on @xmath15 vertices such that @xmath236 by the definition of @xmath237 , there exists a @xmath3-uniform hypergraph @xmath238 with @xmath239 . clearly @xmath240 . thus , by proposition [ prop : deg ] , @xmath241 . consequently @xmath242 . theorem [ nonexthm1 ] implies that either @xmath5 is @xmath110-close to @xmath140 or @xmath141 or @xmath5 contains a matching @xmath146 of size @xmath214 that absorbs any set @xmath215 such that @xmath216 with @xmath217 . in the former case theorem [ extthm ] implies that @xmath5 contains a perfect matching . in the latter case set @xmath243 $ ] and @xmath244 . since @xmath245 , @xmath246 and @xmath15 is sufficiently large , @xmath247 therefore , if @xmath248 , theorem [ marklemma ] implies that @xmath109 contains a matching @xmath249 covering all but at most @xmath250 vertices in @xmath109 . if @xmath251 , then since @xmath252 , fact 2.1 from @xcite implies that @xmath109 contains a matching @xmath249 covering all but at most @xmath231 vertices in @xmath109 . in both cases set @xmath253 then @xmath254 . by definition of @xmath146 , there is a matching @xmath255 in @xmath5 which covers @xmath256 . hence , @xmath257 is a perfect matching of @xmath5 , as desired . in @xcite we proved theorem [ nonexthm1 ] in the case when @xmath3 is divisible by @xmath4 . in this section we give an overview of the proof of theorem [ nonexthm1 ] and explain how our method differs to that used in @xcite . let @xmath80 be divisible by @xmath4 . consider a @xmath3-uniform hypergraph @xmath5 as in theorem [ nonexthm1 ] . define the graph @xmath258 with vertex set @xmath259 in which two vertices @xmath260 are adjacent if and only if @xmath261 . set @xmath262 . in @xcite the proof splits into two main steps . * step 1 : * we prove that @xmath258 or @xmath263 is ` close ' to @xmath264 or @xmath5 contains the matching @xmath146 as desired in theorem [ nonexthm1 ] . * step 2 : * if @xmath258 or @xmath263 is ` close ' to @xmath264 then we prove that @xmath5 is @xmath110-close to @xmath140 or @xmath141 . notice that we can not adopt quite the same approach as above to prove theorem [ nonexthm1 ] for _ all _ values of @xmath3 . indeed , to define @xmath258 we require that @xmath3 is even . moreover , the proof of step 2 in @xcite uses that @xmath3 is divisible by @xmath4 : since @xmath258 or @xmath263 is ` close ' to @xmath264 , we obtain a ` natural ' partition @xmath265 of @xmath266 where @xmath267 . consider a complete @xmath268-uniform hypergraph @xmath269 whose vertex set is @xmath7 . thus , @xmath270 . hence , we can view the partition @xmath265 of @xmath259 as a @xmath271-coloring of @xmath272 . we then apply the hypergraph removal lemma ( see e.g. @xcite ) together with a result of keevash and sudakov @xcite to obtain structure in @xmath269 . ( we show that @xmath273 $ ] or @xmath274 $ ] is ` close ' to @xmath275 . ) this structure in @xmath269 together with the fact that @xmath258 or @xmath263 is ` close ' to @xmath264 implies that @xmath5 is @xmath110-close to @xmath140 or @xmath141 . crucially , the result of keevash and sudakov concerns hypergraphs of even uniformity . thus , we require that @xmath269 has even uniformity and hence , that @xmath3 is divisible by @xmath4 . let @xmath5 be a @xmath3-uniform hypergraph on @xmath15 vertices as in theorem [ nonexthm1 ] . to prove theorem [ nonexthm1 ] we introduce a bipartite analog of @xmath258 . set @xmath276 , @xmath277 , @xmath278 and @xmath279 . further , let @xmath280 and @xmath281 . define the bipartite graph @xmath116 as follows : @xmath116 has vertex classes @xmath282 and @xmath283 . two vertices @xmath284 and @xmath285 are adjacent in @xmath116 if and only if @xmath286 . the proof again splits into two main parts . * step 1 : * we prove that @xmath116 is ` close ' to the disjoint union of two copies of @xmath287 or @xmath5 contains the matching @xmath146 as desired in theorem [ nonexthm1 ] . * step 2 : * if @xmath116 is ` close ' to the disjoint union of two copies of @xmath287 then we prove that @xmath5 is @xmath110-close to @xmath140 or @xmath141 . step 1 can be proved using a similar approach to the corresponding step in @xcite . step 2 , however , is tackled in a different way . indeed , we do not consider an auxiliary hypergraph @xmath269 as in @xcite . instead , we obtain structure in @xmath5 through direct arguments on the graph @xmath116 . in this section we prove theorem [ nonexthm1 ] . let @xmath167 and @xmath61 such that @xmath16 and @xmath2 . given a @xmath3-uniform hypergraph @xmath5 on @xmath15 vertices such that @xmath288 , by proposition [ prop : deg ] , we have @xmath289 where @xmath276 . hence , in order to prove theorem [ nonexthm1 ] it suffices to prove the following result . [ thm : extabs ] given any @xmath211 and integer @xmath290 , there exist @xmath212 and @xmath63 such that the following holds . set @xmath276 . suppose that @xmath5 is a @xmath3-uniform hypergraph on @xmath64 vertices . if @xmath291 then @xmath5 is @xmath110-close to @xmath140 or @xmath141 , or @xmath5 contains a matching @xmath146 of size @xmath214 that absorbs any set @xmath215 such that @xmath216 with @xmath217 . theorem [ thm : extabs ] immediately follows from lemmas [ lem : abs][lem : gh ] . the following lemma from @xcite states that in order to find the absorbing set described in theorem [ thm : extabs ] , it suffices to prove that there are at least @xmath292 absorbing @xmath293-sets for every fixed @xmath3-set from @xmath7 . * lemma 5.2 ) [ lem : abs ] given @xmath294 and an integer @xmath295 , there exists an @xmath63 such that the following holds . consider a @xmath3-uniform hypergraph @xmath5 on @xmath64 vertices . suppose that any @xmath3-set of vertices @xmath296 can be absorbed by at least @xmath292 @xmath293-sets of vertices from @xmath7 . then @xmath5 contains a matching @xmath146 of size @xmath297 that absorbs any set @xmath298 such that @xmath299 and @xmath300 . throughout this section we will use the following notation . let @xmath290 be an integer and set @xmath301 and @xmath302 . given a @xmath3-uniform hypergraph @xmath5 , define @xmath278 and @xmath279 . set @xmath280 and @xmath281 . given a @xmath3-uniform hypergraph @xmath5 , we define the bipartite graph @xmath303 as follows : @xmath303 has vertex classes @xmath282 and @xmath283 . two vertices @xmath284 and @xmath285 are adjacent in @xmath303 if and only if @xmath286 . when it is clear from the context , we will often refer to @xmath303 as @xmath116 . let @xmath290 and @xmath15 be positive integers . denote by @xmath304 the bipartite graph with vertex classes @xmath71 and @xmath305 of sizes @xmath306 and @xmath307 respectively which satisfies the following properties : * @xmath308 is a partition of @xmath71 such that @xmath309 and @xmath310 ; * @xmath311 is a partition of @xmath305 such that @xmath312 and @xmath313 ; * @xmath314 $ ] and @xmath315 $ ] are complete bipartite graphs . further , there are no other edges in @xmath316 . [ lem : g ] given any @xmath317 and an integer @xmath0 , there exist @xmath318 , and @xmath63 such that the following holds . suppose that @xmath5 is a @xmath3-uniform hypergraph on @xmath64 vertices so that @xmath319 set @xmath320 . then at least one of the following assertions holds . * @xmath321 ; in other words , @xmath116 becomes a copy of @xmath304 after adding or deleting at most @xmath322 edges . * there are at least @xmath292 absorbing @xmath293-sets in @xmath323 for every @xmath3-subset of @xmath7 . [ lem : gh ] given any @xmath211 and integer @xmath290 , there exist @xmath317 and @xmath63 such that the following holds . suppose that @xmath5 is a @xmath3-uniform hypergraph on @xmath64 vertices . suppose further that @xmath320 satisfies @xmath321 . then @xmath5 is @xmath110-close to @xmath140 or @xmath141 . the rest of the section is devoted to the proof of lemmas [ lem : g ] and [ lem : gh ] . given @xmath324 , we choose additional constants @xmath325 such that @xmath326 without loss of generality we may assume that @xmath327 . we also assume that @xmath15 is sufficiently large . we have that @xmath328 and so by proposition [ prop : deg ] , @xmath329 let @xmath143 be a @xmath3-set . it is easy to see that if @xmath147 has at least @xmath330 absorbing @xmath3-sets then @xmath147 has at least @xmath292 absorbing @xmath293-sets . indeed , let @xmath331 be an absorbing @xmath3-set for @xmath147 . then @xmath332 is an absorbing @xmath293-set for @xmath147 for any edge @xmath333 . note that @xmath334 thus , as @xmath15 is sufficiently large , there are at least @xmath335 edges in @xmath336 . since an absorbing @xmath293-set may be counted at most @xmath337 times when counting the number of @xmath338 , there are at least @xmath339 absorbing @xmath293-sets for @xmath147 . therefore , in order to prove lemma [ lem : g ] , it suffices to prove the following two claims . [ clm : ab ] if either of the following cases holds , then we can find @xmath330 absorbing @xmath3-sets or @xmath340 absorbing @xmath293-sets for every @xmath3-set @xmath341 . case ( a ) : : for any @xmath74-tuple @xmath342 , there are at least @xmath343 @xmath74-tuples @xmath344 such that @xmath345 . case ( b ) : : @xmath346 . [ clm : abg ] if neither case ( a ) or case ( b ) holds , then @xmath321 . * proof of claim [ clm : ab ] . * we argue in a similar way to the proof of claim 5.5 in @xcite . given a @xmath3-set @xmath347 , we will consider two types of absorbing sets for @xmath147 : absorbing @xmath3-sets : : these consist of a single edge @xmath348 with the property that both @xmath349 and @xmath350 are edges of @xmath5 . absorbing @xmath293-sets : : these consist of distinct vertices @xmath351 , @xmath352 , @xmath353 , @xmath354 such that @xmath355 , @xmath356 and @xmath357 are edges in @xmath5 . furthermore , @xmath349 and @xmath350 are also edges of @xmath5 ( see figure 1 ) . [ picture ] -set and ( ii ) absorbing @xmath293-set in the case when @xmath358 . ] write @xmath359 and @xmath360 . for any two ( not necessarily disjoint ) @xmath74-tuples @xmath361 , @xmath344 we call @xmath361 a _ good @xmath74-tuple for @xmath362 _ if @xmath363 . we first observe that @xmath147 has at least @xmath364 absorbing @xmath3-sets if there are @xmath365 indeed , assume that ( [ eq : good ] ) holds . there are at most @xmath366 @xmath74-tuples in @xmath367 that contain at least one element from @xmath368 . therefore , there are at least @xmath369 @xmath74-tuples in @xmath370 that are good for @xmath371 and disjoint from @xmath371 . let us pick such an @xmath74-tuple @xmath372 . thus , @xmath373 . we pick @xmath374 such that @xmath375 is disjoint from @xmath376 . note that there are at least @xmath377 choices for @xmath375 . notice that the @xmath3-set @xmath378 is an absorbing set for @xmath147 since @xmath379 , @xmath349 and @xmath350 are edges in @xmath5 . since an absorbing @xmath3-set may be counted @xmath380 times , this argument implies that there are at least @xmath381 absorbing @xmath3-sets for @xmath147 . now assume that case ( a ) holds . this implies that there are at least @xmath343 good @xmath74-tuples for @xmath371 . by ( [ minr ] ) , @xmath382 . so there are at least @xmath383 @xmath74-tuples in @xmath370 that are good for @xmath371 . thus , holds and consequently @xmath147 has at least @xmath330 absorbing @xmath3-sets . next assume case ( b ) holds . for any two ( not necessarily disjoint ) @xmath384-tuples @xmath361 , @xmath385 we call @xmath361 a _ good @xmath384-tuple for @xmath362 _ if @xmath386 . by arguing in an identical fashion as before , note that @xmath147 has at least @xmath364 absorbing @xmath3-sets if there are @xmath387 let @xmath388 . so by assumption , @xmath389 . note that every @xmath390 is good for arbitrary @xmath391 since @xmath392 , @xmath393 and therefore @xmath394 . thus , if @xmath395 then ( [ eq : good2 ] ) is satisfied . therefore , we may assume that @xmath396 . we also assume that fails ( otherwise we are done ) . thus , less than @xmath397 @xmath74-tuples in @xmath398 are good for @xmath399 and consequently , at least @xmath400 @xmath74-tuples @xmath401 satisfy @xmath402 . we pick such an @xmath74-tuple @xmath403 that is disjoint from @xmath371 ; there are at least @xmath404 @xmath74-tuples with this property . since @xmath405 it follows that @xmath406 now pick any @xmath407 that is disjoint from @xmath147 . ( note there are at least @xmath408 choices for @xmath409 . ) next pick an @xmath384-tuple @xmath410 such that @xmath375 is disjoint from @xmath376 , @xmath403 and @xmath409 . ( there are at least @xmath411 choices for @xmath375 here . ) by the definition of @xmath412 , there are at least @xmath413 @xmath74-sets in @xmath414 . we pick @xmath415 such that @xmath416 is disjoint from @xmath376 , @xmath371 and @xmath403 . ( there are at least @xmath417 choices for @xmath416 here . ) let @xmath8 denote the @xmath293-set consisting of the vertices contained in @xmath403 , @xmath375 , @xmath409 and @xmath416 . by the choice of @xmath403 , @xmath375 , @xmath409 and @xmath416 , @xmath8 is an absorbing @xmath293-set for @xmath147 . in summary , there are at least @xmath418 choices for @xmath403 , at least @xmath419 choices for @xmath409 , at least @xmath420 choices for @xmath375 and at least @xmath421 choices for @xmath416 . since each absorbing @xmath293-set may be counted @xmath422 times , there are at least @xmath423 absorbing @xmath293-sets for @xmath147 , as desired . * proof of claim [ clm : abg ] . * note that by ( [ minr ] ) and ( [ minr ] ) , @xmath424 further by assumption , the following conditions hold . ( i ) : : there exists a vertex @xmath425 such that at most @xmath426 vertices @xmath427 satisfy @xmath428 . ( ii ) : : @xmath429 . let @xmath430 and @xmath431 . then @xmath432 and @xmath433 . we also need an upper bound on @xmath434 . fix @xmath435 . since @xmath436 , we have @xmath437 which gives @xmath438 . by the definition of @xmath439 , we have @xmath440 . thus , ( [ gmin ] ) implies that @xmath441 where @xmath442 . next we show that @xmath443 is very small , where @xmath444 . [ clm : acb ] @xmath445 . assume for a contradiction that the claim is false . set @xmath446 . by assumption @xmath447 which gives that @xmath448 . by , as @xmath449 , we derive that @xmath450 let @xmath451 . we claim that @xmath452 . indeed , consider @xmath453 . if @xmath454 , then @xmath455 , contradicting . let @xmath456 . we have that @xmath457 . since @xmath458 and @xmath459 , we derive that @xmath460 . for every @xmath461 , we have @xmath462 ( the penultimate inequality follows since @xmath463 . ) this is a contradiction to ( ii ) . recall that @xmath464 . thus , by ( [ gmin ] ) @xmath465 ( the last inequality follows since @xmath466 . ) therefore , as @xmath467 , we have @xmath468 . pick a set @xmath469 of size @xmath470 such that @xmath471 is maximized . similarly , pick a set @xmath472 of size @xmath473 such that @xmath474 is maximized . set @xmath475 and @xmath476 . [ targets ] the following conditions hold : * @xmath477 ; * @xmath478 ; * @xmath479 . since @xmath480 , we have @xmath481 . further , @xmath482 and so @xmath483 . hence , @xmath484 ( the penultimate inequality follows as @xmath485 . ) note that @xmath486 as @xmath487 and @xmath488 as @xmath489 . thus , @xmath490 ( the penultimate inequality follows as @xmath491 . ) since @xmath468 , @xmath492 . further , @xmath493 and so @xmath494 . thus , by claim [ clm : acb ] , @xmath495 since @xmath496 , @xmath497 . further , @xmath491 and so @xmath498 . hence , @xmath499 claim [ targets ] immediately implies that @xmath500 , as desired . define constants @xmath501 and @xmath63 so that @xmath502 let @xmath5 and @xmath116 be as in the statement of the lemma . throughout this section , when it is clear from the context , we will write @xmath35 for the vertex set @xmath7 and @xmath182 for the edge set @xmath503 . note that @xmath504 and @xmath505 as @xmath290 . since @xmath116 is a bipartite graph with vertex classes @xmath282 and @xmath283 , and @xmath506 , there exists a partition @xmath507 of @xmath282 and a partition @xmath508 of @xmath283 so that * @xmath509 , @xmath510 , @xmath511 , and @xmath512 ; * @xmath513 , in other words , @xmath116 becomes the disjoint union of two complete bipartite graphs @xmath514 and @xmath515 after adding or removing at most @xmath516 edges . throughout the proof , we assume that @xmath517 . we call a @xmath3-subset @xmath8 of @xmath35 _ bad _ if there are two partitions @xmath518 and @xmath519 of @xmath8 such that both of the following conditions hold : 1 . @xmath520 , namely , either @xmath521 and @xmath522 or @xmath523 and @xmath524 ; 2 . @xmath525 , namely , either @xmath526 and @xmath527 or @xmath528 and @xmath529 . [ badclaim ] at most @xmath530 @xmath3-subsets of @xmath35 are bad . let @xmath531 be a bad @xmath3-set and let @xmath532 and @xmath533 be partitions of @xmath8 as in the definition of a bad @xmath3-set . if @xmath534 , then @xmath535 ; otherwise @xmath536 . in either case @xmath8 gives rise to an edge from @xmath537 . furthermore , two different bad @xmath3-sets give two different edges of @xmath537 . since @xmath538 , the number of bad @xmath3-sets is at most @xmath516 . viewing @xmath539 as the colors of @xmath74-subsets of @xmath35 , we define the color function @xmath540 with @xmath541 if @xmath542 . similarly we define @xmath543 such that @xmath544 if @xmath545 . given two distinct vertices @xmath546 , we define two symmetric functions functions as the families of @xmath547-sets or @xmath548-sets instead , e.g. , @xmath549 . however , when @xmath550 ( thus @xmath551 ) , this new definition defines @xmath552 for all @xmath546 , while our present definition of @xmath553 returns @xmath554 when @xmath555 and @xmath556 otherwise . ] @xmath557 similarly , we define @xmath558 note that the definition of @xmath559 is _ different _ to the definition of @xmath560 stated in section [ usesec ] . thus , when referring to the latter parameter we never omit the subscript . note that @xmath561 all but at most @xmath562 pairs of vertices @xmath546 satisfy the following two properties . ( i ) : : @xmath563 if and only if @xmath564 . ( ii ) : : @xmath565 if and only if @xmath566 . suppose for a contradiction that the claim is false . consider two vertices @xmath567 such that ( i ) fails . suppose that @xmath568 but @xmath569 ( the other case can be proven analogously ) . by , we have @xmath570 . we claim that @xmath571 is a bad @xmath3-set for all @xmath572 and @xmath573 such that @xmath574 . indeed , @xmath572 implies that one of @xmath575 and @xmath576 is in @xmath577 and the other is in @xmath578 , and @xmath579 implies that @xmath580 and @xmath581 are both in @xmath582 for some @xmath583 . if @xmath574 , then precisely one of the two pairs @xmath584 and @xmath585 is in @xmath586 . thus , by definition , the @xmath3-set @xmath571 is bad . next consider a pair of vertices @xmath155 that fails ( ii ) . suppose that @xmath587 but @xmath588 ( again , the other case can be proven analogously ) . by , @xmath589 . a similar argument to before yields that @xmath590 is a bad @xmath3-set for all @xmath591 and @xmath592 such that @xmath593 . note that given any @xmath548-set , at most @xmath594 @xmath547-subsets of @xmath35 are _ not _ disjoint from this set . thus , given any @xmath595-set @xmath147 that contains @xmath596 and @xmath175 , at most @xmath594 @xmath157-sets @xmath331 that contains @xmath596 and @xmath175 satisfy @xmath597 . further , note that @xmath598 . therefore , when considering all possible pairs of vertices @xmath155 that fail ( i ) or ( ii ) , we obtain at least @xmath599 bad @xmath3-sets as a @xmath3-set may be counted at most @xmath600 times . this contradicts claim [ badclaim ] . we call an ( unordered ) pair of vertices @xmath155 _ consistent _ if @xmath601 satisfy both ( i ) and ( ii ) from claim [ clm : x = y ] . thus , all but at most @xmath562 pairs of vertices from @xmath35 are consistent . we call two vertices @xmath155 _ similar _ if @xmath602 or @xmath603 . [ clm : sim ] less than @xmath604 pairs of vertices @xmath155 are consistent but not similar . let @xmath567 be consistent but not similar . thus , @xmath605 and @xmath606 . since @xmath607 are consistent , @xmath608 implies that @xmath609 . as seen in the proof of claim [ clm : x = y ] , @xmath571 is a bad @xmath3-set for all @xmath572 and @xmath573 such that @xmath574 . thus , if there are at least @xmath604 pairs of vertices @xmath155 that are not similar but are consistent , then the number of bad @xmath3-subsets of @xmath35 is at least @xmath610 contradicting claim [ badclaim ] . let @xmath611 be a vertex such that at least @xmath612 vertices of @xmath35 are both consistent and similar to @xmath613 : such a vertex @xmath613 exists because otherwise at least @xmath614 pairs of vertices are not consistent or are not similar , contradicting claim [ clm : x = y ] or claim [ clm : sim ] . let @xmath615 be the set of vertices in @xmath35 that are not consistent or not similar to @xmath613 . ( note that @xmath616 . ) the choice of @xmath613 implies that @xmath617 . define @xmath618 note that @xmath619 otherwise by , it implies that @xmath620 , a contradiction . [ ab ] the following properties hold . * @xmath621 and @xmath622 for all pairs of distinct vertices @xmath623 such that @xmath624 or @xmath625 ; * @xmath626 and @xmath627 for all @xmath628 and @xmath629 . let @xmath630 denote the set of @xmath547-subsets of @xmath35 that contain @xmath613 and let @xmath631 denote the set of @xmath548-subsets of @xmath35 that contain @xmath613 . consider distinct @xmath632 and suppose @xmath633 for some @xmath547-subset @xmath634 of @xmath635 . then either @xmath636 or @xmath637 . in the latter case , as @xmath638 and @xmath639 have different colors , one of them has the same color as @xmath640 and the other has a different color to @xmath640 . thus , either ( i ) @xmath641 and @xmath642 or ( ii ) @xmath643 and @xmath644 . this implies that latexmath:[\[\label{eq : ddd } for all distinct @xmath623 . an analogous argument implies that latexmath:[\[\label{eq : ddd2 } for all distinct @xmath623 . consider any distinct @xmath624 . by the definition of @xmath647 and , we have @xmath648 as @xmath649 and @xmath650 are both consistent pairs , the fact that @xmath624 implies that @xmath651 and @xmath652 . thus , by we have that @xmath653 . next consider any distinct @xmath625 . by the definition of @xmath654 and , we have @xmath655 we can show that @xmath656 as before . consider any @xmath657 and @xmath629 . suppose that @xmath658 is an @xmath547-set such that @xmath659 . so @xmath638 and @xmath639 have the same color . hence , @xmath640 has either a different color to @xmath638 or the same color as @xmath639 . thus , @xmath660 where the second inequality follows by the definitions of @xmath647 and @xmath654 . an analogous argument gives that @xmath661 . once we have obtained more information we will prove that @xmath662 and @xmath663 are close to @xmath664 . however , to prove claim [ clm : ai ] we first require the following weaker lower bounds on @xmath662 and @xmath663 . [ claim : v12 ] @xmath665 . suppose that @xmath666 for some @xmath667 . let us consider @xmath668 by claim [ ab ] ( a ) , the first term is at most @xmath669 . the second term is at most @xmath670 . together this gives that @xmath671 . on the other hand , the definition of @xmath304 yields @xmath672 . applying lemma [ lem : kk ] to the @xmath74-uniform hypergraph @xmath673 , we derive that @xmath674 ( note that @xmath675 for any distinct @xmath175 and @xmath176 ) . together with the upper bound for @xmath103 , it follows that @xmath676 since @xmath677 . this implies that @xmath678 . since @xmath679 and @xmath617 , we derive that @xmath680 ( again using @xmath677 ) . an analogous argument implies that @xmath681 . given two disjoint subsets @xmath682 and two integers @xmath683 , we call an @xmath684-subset @xmath531 an @xmath685-set , and write @xmath686 , if @xmath687 and @xmath688 . let @xmath689 and @xmath690 for @xmath691 . set @xmath692 and @xmath693 . [ clm : ai ] @xmath694 1 . for all @xmath695 , at least @xmath696 @xmath697-sets are in @xmath698 , where @xmath699 and @xmath700 ; 2 . for all @xmath701 , at least @xmath702 @xmath703-sets are in @xmath704 , where @xmath705 and @xmath706 . we only prove the first assertion as the proof for the second is analogous ( even in the case when @xmath551 ) . we proceed by induction on @xmath185 . we first apply lemma [ lem : kk ] to the @xmath74-uniform hypergraph @xmath6 with vertex set @xmath647 and edge set @xmath707 . if @xmath6 has edge density @xmath708 $ ] , then there are vertices @xmath624 such that @xmath709 . however , by claim [ ab ] ( a ) , @xmath710 . since @xmath711 and @xmath712 by claim [ claim : v12 ] , we have @xmath713 a contradiction . thus , @xmath714 or @xmath715 . if @xmath714 , then at least @xmath716 @xmath74-subsets of @xmath647 are in @xmath577 and we set @xmath717 . if @xmath715 , then at most @xmath718 @xmath74-subsets of @xmath647 are in @xmath577 . thus , at least @xmath716 @xmath74-subsets of @xmath647 are in @xmath578 and we set @xmath719 . for the induction step , we actually prove that there are vertices @xmath720 such that for each @xmath721 , at least @xmath696 @xmath722-sets are in @xmath723 , where @xmath724 if @xmath185 is even and @xmath725 if @xmath185 is odd . suppose this assertion holds for some @xmath726 . by an averaging argument , there exists a vertex @xmath727 and at least @xmath728 @xmath547-sets @xmath729 such that @xmath730 . this implies that @xmath731 since @xmath732 , given any @xmath733 , we have @xmath734 by claim [ ab ] ( b ) . thus , @xmath735 let @xmath736 be the number of @xmath737-sets in @xmath738 . we thus have @xmath739 it is easy to see that @xmath740 for @xmath741 . claim [ claim : v12 ] states that @xmath742 . consequently , @xmath743 for all @xmath741 . therefore , @xmath744 as desired . set @xmath745 , @xmath746 and @xmath747 . [ clm : n2 ] @xmath748 . we have that @xmath749 since @xmath750 . let @xmath751 . it suffices to show that @xmath752 because this implies that @xmath753 and @xmath754 . without loss of generality , assume that @xmath755 in the statement of claim [ clm : ai ] . thus , @xmath756 hence , by proposition [ evensum ] , @xmath757 , which implies that @xmath758 if @xmath759 , then @xmath760 since @xmath761 . consequently , @xmath762 . this contradicts the assumption @xmath763 . thus , @xmath764 if @xmath765 , then implies that @xmath766 and we are done . otherwise assume that @xmath767 . if @xmath74 is even , then implies that @xmath768 and so @xmath769 . if @xmath74 is odd , then we apply claim [ clm : ai ] and proposition [ evensum ] obtaining that @xmath770 if @xmath771 , then we obtain a contradiction as before because @xmath772 . hence , @xmath773 and consequently @xmath769 , as required . by claim [ clm : n2 ] , there exists a partition @xmath774 of @xmath35 such that @xmath775 , @xmath776 and @xmath777 for each @xmath778 . [ clm : x1b ] the hypergraph @xmath779 is @xmath780-close to @xmath781 or @xmath782 , and the hypergraph @xmath783 is @xmath780-close to @xmath784 or @xmath785 . by claim [ clm : ai ] , at least @xmath786 @xmath74-subsets of @xmath787 are in @xmath788 and have an even number of vertices in @xmath654 , and at least @xmath789 @xmath74-subsets of @xmath787 are in @xmath790 and have an odd number of vertices in @xmath654 . suppose that @xmath74 is odd . set @xmath791 . by definition , @xmath792 thus implies that @xmath793 on the other hand , letting @xmath794 , we have @xmath795 we thus derive that @xmath796 . note that at most @xmath797 vertices of @xmath35 are not in @xmath798 and each edge in @xmath799 must contain such a vertex . hence , @xmath800 therefore , @xmath801 since @xmath802 , we conclude that @xmath779 is @xmath780-close to @xmath803 or @xmath804 . the case when @xmath74 is even is analogous . moreover , analogous arguments show that @xmath783 is @xmath780-close to @xmath805 or @xmath806 . @xmath5 is @xmath110-close to @xmath807 or @xmath808 . since we always consider the partition @xmath774 of @xmath35 , we write , for example , @xmath140 instead of @xmath809 $ ] throughout the proof . we call a set @xmath531 _ even _ if @xmath810 is even . otherwise we say that @xmath8 is _ odd_. thus , @xmath811 consists of all odd @xmath74-sets , and @xmath812 consists of all even @xmath384-sets . for convenience , we use @xmath813 to denote the hypergraphs @xmath814 for @xmath815 . there are four possible cases : 1 . @xmath816 and @xmath817 , 2 . @xmath816 and @xmath818 , 3 . @xmath819 and @xmath817 , 4 . @xmath819 and @xmath818 . we claim that @xmath820 under cases 2 and 3 , and @xmath821 under cases 1 and 4 . below we show that @xmath820 under case 2 . ( the other cases are analogous . ) assume that @xmath816 and @xmath818 . our goal is to show that @xmath822 . first we show that @xmath823 . consider any @xmath3-set @xmath147 from @xmath824 . since @xmath825 ( and thus @xmath826 is even ) , @xmath147 can be partitioned into ( i ) an even @xmath74-set @xmath827 and an even @xmath384-set @xmath399 or ( ii ) an odd @xmath74-set @xmath827 and an odd @xmath384-set @xmath399 . as @xmath828 , in both cases we have that @xmath829 . thus , @xmath830 where @xmath831 is the set of all disjoint @xmath832 , @xmath833 such that @xmath834 and @xmath835 are even and @xmath836 and @xmath837 is the set of all disjoint @xmath832 , @xmath833 such that @xmath834 and @xmath835 are odd and @xmath836 . since @xmath838 ( and thus @xmath839 ) , there are at most @xmath840 pairs @xmath841 such that @xmath834 is not in @xmath578 . since @xmath842 , there are at most @xmath843 pairs @xmath841 such that @xmath835 is not in @xmath844 . by the structure of @xmath116 , at most @xmath845 pairs @xmath846 , @xmath847 are such that @xmath836 . together , this implies that @xmath848 . since @xmath838 , there are at most @xmath840 pairs @xmath849 such that @xmath834 is not in @xmath577 . since @xmath850 , there are at most @xmath843 pairs @xmath849 such that @xmath835 is not in @xmath851 . by the structure of @xmath116 , at most @xmath845 pairs @xmath852 , @xmath853 are such that @xmath836 . together , this implies that @xmath854 . so indeed , @xmath855 . next we show that @xmath856 . consider any @xmath3-set @xmath857 from @xmath858 . since @xmath859 ( and thus @xmath826 is odd ) , we can partition @xmath147 into ( i ) an even @xmath74-set @xmath827 and an odd @xmath384-set @xmath399 or ( ii ) an odd @xmath74-set @xmath827 and an even @xmath384-set @xmath399 . as @xmath860 , in both cases we have that @xmath861 . thus , @xmath862 where @xmath863 is the set of all disjoint @xmath832 , @xmath833 such that @xmath834 is even , @xmath835 is odd and @xmath864 and @xmath865 is the set of all disjoint @xmath832 , @xmath833 such that @xmath834 is odd , @xmath835 is even and @xmath864 . since @xmath839 , there are at most @xmath840 pairs @xmath866 such that @xmath834 is not in @xmath578 . since @xmath850 , there are at most @xmath843 pairs @xmath866 such that @xmath835 is not in @xmath851 . by the structure of @xmath116 , at most @xmath845 pairs @xmath846 , @xmath853 are such that @xmath867 . together , this implies that @xmath868 . since @xmath869 , there are at most @xmath840 pairs @xmath870 such that @xmath834 is not in @xmath577 . since @xmath842 , there are at most @xmath843 pairs @xmath870 such that @xmath835 is not in @xmath844 . by the structure of @xmath116 , at most @xmath845 pairs @xmath852 , @xmath847 are such that @xmath867 . together , this implies that @xmath871 . so indeed , @xmath856 . therefore @xmath822 , as desired . this thus completes the proof of lemma [ lem : gh ] . we thank oleg pikhurko for a helpful comment on this problem . 10 n. alon , p. frankl , h. huang , v. rdl , a. ruciski and b. sudakov , large matchings in uniform hypergraphs and the conjectures of erds and samuels , _ j. combin . theory ser . * 119 * ( 2012 ) , 12001215 . a. czygrinow , v. kamat , tight co - degree condition for perfect matchings in 4-graphs , _ electron . j. combin . _ * 19 * ( 2012 ) , no . 2 , p20 . m.r . garey and d.s . johnson , computers and intractability , freeman , 1979 . gowers , hypergraph regularity and the multidimensional szemerdi theorem , _ ann . of math . _ * 166 * ( 2007 ) , 897946 . h. hn , y. person and m. schacht , on perfect matchings in uniform hypergraphs with large minimum vertex degree , _ siam j. discrete math . _ * 23 * ( 2009 ) , 732748 . karp , reducibility among combinatorial problems , complexity of computer computations ( proc . , ibm thomas j. watson res . center , yorktown heights , n.y . , 1972 ) , pp . 85103 , plenum , new york , 1972 . p. keevash and b. sudakov , on a hypergraph turn problem of frankl , _ combinatorica _ * 25 * ( 2005 ) , 673706 . i. khan , perfect matching in @xmath20 uniform hypergraphs with large vertex degree , arxiv:1101.5830 . i. khan , perfect matchings in @xmath4-uniform hypergraphs , arxiv:1101.5675 . d. khn and d. osthus , matchings in hypergraphs of large minimum degree , _ j. graph theory _ * 51 * ( 2006 ) , 269280 . d. khn , d. osthus and a. treglown , matchings in @xmath20-uniform hypergraphs , arxiv:1009.1298 . l. lovsz , combinatorial problems and exercises , north - holland , amsterdam , 1979 . k. markstrm and a. ruciski , perfect matchings ( and hamilton cycles ) in hypergraphs with large degrees , _ european j. combin . _ * 32 * ( 2011 ) , 677687 . o. pikhurko , perfect matchings and @xmath872-tilings in hypergraphs of large codegree , _ graphs combin . _ * 24 * ( 2008 ) , 391404 . v. rdl and a. ruciski , dirac - type questions for hypergraphs a survey ( or more problems for endre to solve ) , _ an irregular mind ( szemerdi is 70 ) _ , bolyai soc . studies * 21 * ( 2010 ) , 130 . v. rdl , a. ruciski and e. szemerdi , a dirac - type theorem for 3-uniform hypergraphs , _ combin . comput . _ * 15 * ( 2006 ) , 229251 . v. rdl , a. ruciski and e. szemerdi , perfect matchings in uniform hypergraphs with large minimum degree , _ european j. combin . _ * 27 * ( 2006 ) , 13331349 . v. rdl , a. ruciski and e. szemerdi , perfect matchings in large uniform hypergraphs with large minimum collective degree , _ j. combin . theory ser . * 116 * ( 2009 ) , 613636 . v. rdl and j. skokan , applications of the regularity lemma for uniform hypergraphs , _ random structures and algorithms _ * 28 * ( 2006 ) , 180194 . a. treglown and y. zhao , exact minimum degree thresholds for perfect matchings in uniform hypergraphs , _ j. combin . theory a _ 119 ( 2012 ) , 15001522 . tutte , the factorisation of linear graphs , _ j. london math . soc _ * 22 * ( 1947 ) , 107111 .