Anisotropic perturbations of the simple symmetric Christian Maes, Frank Redig
Transcription
Anisotropic perturbations of the simple symmetric Christian Maes, Frank Redig
Anisotropic perturbations of the simple symmetric exclusion process : long range correlations Christian Maes, Frank Redig To cite this version: Christian Maes, Frank Redig. Anisotropic perturbations of the simple symmetric exclusion process : long range correlations. Journal de Physique I, EDP Sciences, 1991, 1 (5), pp.669684. <10.1051/jp1:1991161>. <jpa-00246361> HAL Id: jpa-00246361 https://hal.archives-ouvertes.fr/jpa-00246361 Submitted on 1 Jan 1991 HAL is a multi-disciplinary open access archive for the deposit and dissemination of scientific research documents, whether they are published or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers. L’archive ouverte pluridisciplinaire HAL, est destin´ee au d´epˆot et `a la diffusion de documents scientifiques de niveau recherche, publi´es ou non, ´emanant des ´etablissements d’enseignement et de recherche fran¸cais ou ´etrangers, des laboratoires publics ou priv´es. J. Phys. (1991) J1 669-684 MAi 1991, PAGE 669 Classification Physics Abstracts 05.40 02.90 46.10 Anisotropic perturbations of the shnple symmetric correlations process : long range Christian (') (~) and Maes Instituut Frank Theoretische voor exclusion Redig ~) Fysica, Leuven, K. U. Celestijnenlaan 200D, B-3001 Leuven, Belgium f) Instelling Uriiversitaire (Received 25 Antwerpen, J990, October accepted ih Universiteitsplein final form 22 I, B-2610 Wilrijk, Belgium January J991) spin system in which nearest neighbor spins are exchanged at rates neighboring configuration. The system has the same symmetry of the on correlation lattice rotations. for lattice We set up a perturbation expansion for the functions exclusion Convergence is proven for time t around the simple symmetric at process. limit reproduces the usual high temperature expansion in the case of small t and the formal t co Abs«act. which Consider lattice a weakly depend except possibly the ~ detailed balance. then, generically, decays only like 1. If the the a system is isotropic, two power points function r~~, where r then has is the each the term direction spatial expansion dependence of in the separation and is a strictly local. If quadrupole field dm 2 is the not, and dimension. Introduction. have correlations decaying in the same the at high temperatures sense as potential. This well known fact from equilibrium statistical mechanics has been established methods by various such as the cluster expansion [1-3], or a Dobrushin type analysis [4-6], and is in last instance a consequence of the so called local Markov property of Gibbs interaction, given the configuration appropriate inside the states : for a finite range boundary layer surrounding a volume, the inside and outside are conditionally independent. It allows follow the interdependence of certain different regions in space via in to events intermediate points (or, «polymers») along which information is transported (and lost). These « explicitly connections constructed in the cluster expansion, starting from the are Boltzmann factor Z~ (pH), small, for interaction order p H. For example, to first exp an in p, for a lattice spin system, only the neighborhood of a site, as determined by H, can influence the spin there. probabilistic More methods the (quasi-) locality of the use conditional distributions estimate the dependence of a spin on its neighboorhood. This to explains in fact why we expect that local interactions generically (that is, away from critical points) give rise to a finite correlation length. Tuming to real non-equilibrium situations, we cannot expect such methods to be applicable by only considering the spatial structure. For a general lattice spin dynamics, we will not find a simple local mechanism by which to describe the mutual influence of spins in different Gibbs states interaction JOURNAL 670 PHYSIQUE DE N I 5 regions. The spatial of non-Gibbsian stationary best be studied by structure states can embedding it in the spatio-temporal Simple examples for which readily process. one recognizes the advantages of connecting spatially separated spins via their history common interacting particle systems or probabilistic cellular [7-8]. In fact, such an automata are analysis naturally arises for general non-equilibrium phenomena. This may seem obvious but it has important If the is not Gibbsian, and the spatial stationary state consequences. correlations intermediated via events in the past, then the spatial decay must be thought off as will be related to the temporal correlations. In particular if the latter have a weak decay such for diffusive similar then behavior be expected for the stationary systems, as a can correlations. the starting point of the analysis in [9] to study the phenomenon of long was in non-equilibrium dynamics. Here we wish to spatial correlations conservative range perturbation expansion around an exactly solvable investigate in more detail the associated Z( on the d-dimensional consists of spins «(x) lattice gas dynamics. The system ± I, x e neighbor spins are exchanged with rates introduced in which lattice. A dynamics is nearest interested in the which depend weakly on the configuration in a finite neighborhood. We are This, in short, = behavior of the xe fl functions correlation «(x) for finite some A , Z( < stationary in the A To states. study them, make we symmetric simple so-called the an expansion exclusion around [7]. It can be the product viewed as state a which system of is invariant random for walkers potential only. Although the stationary states are trivial for this fluctuations correlations decay weakly. Typically, in the density of the system, the temporal walkers decay diffusively in time like t~ ~'~. This decay enters in the perturbation expansion as analysis considerably wl~ile this makes of the free the propagator. our more mass zero complicated, it has the interesting feature to produce, for generic perturbations, also a weak and Putting r~~ t as the usual space-time scaling for diffusive decay in space. systems intermediated diffusive decay, we end up assuming that the spatial decay is via the temporal bonstant Q such that («(0); «(r)) that there is Q (r(-~, with the prediction some jr for the stationary two points function. wl~ile, a priori, this scenario is completely co general, the details of the dynamics enter in the prefactor Q. We will argue that generically situations in which by symmetry Q »0. Those Q # 0 but there are special (non-generic) (overlapping) classes. be divided into two situations can The first important special case when this « self-organized criticality disappears by occurs arranging the dynamics in such a way that it satisfies the condition of detailed balance, and the appropriate Gibbs states become stationary. In that case we (in an unusual way) the recover usual high-temperature expansion. In the light of the discussion follows that emphasize we this is as expected and is independent of the (an-)isotropy of the that corresponding Hamiltonianthat we need is completely contained the in the detailed balance symmetry which interact via hard a core = - condition, Secondly, anisotropy the fact as that become will dynamics, of the the systems clear, All we an essential (necessary) ingredient in our analysis is leading to long range correlations based are arguments consider do not our possess the full symmetry of the the on lattice. that starting from the formal expansion for the stationary correlation isotropic have perturbation short correlations. This is must any range consistent with the results of [10] for similar dynamics, but where it is not excluded to have than particle per site. more one Finally, it may still happen that by some other symmetry Q 0 even though the model does described classes above. An example of this is presented at the end of not fit in one of the two still have a power law decay with the section 4 and it is found there that the correlations Moreover, functions, we will show fully " N POWER 5 dependence direction Q DECAY LAW of octopole an STOCHASTIC IN (instead field LATTICE of quadrupole the 671 GASES decay for the case 0). # The difference rigorous analysis of the (but, and model In of the details [9] lies in the of the find we the section weak) results concerning the anisotropy. We also add complete possibility of octopole-like decay. define we expansion. We is strictly finite, can the model. only show finite for all that times. 3 is it convergence explicit and devoted converges the small formal derivation the to for taking Still expansion, the the thorough and results for a specific t - times, or, co limit, of the when the discuss we points decay is like term we per term r~~, where r is the spatial separation, and dm 2. We emphasize the role of the anisotropy. This behavior is explicitly calculated first order expansion for a specific in the up to perturbation. In the Appendix we apply our formal expansion to the Kawasaki dynamics and give this expansion more significance by recoveringGibbs without using the standard formalismthe well known high temperature expansion for the equilibrium correlation in section function 4 the of presence long Section perturbation of the admittedly, role next perturbation perturbation with work our studied is correlations range expamion, in the stationary in the find and The measure. generically that two its functions. 2. The model. Z~ in which the rates We consider stochastic for spins «(x) time evolution ± I, a x e c(x, y, « ) at which the spin at site x and y, ix y I are exchanged depend weakly on the configuration at other sites in a finite neighboorhood of the bond (xy). For a finite volume Pf(«) to find configuration « e (- I, + ~ in A, A < Z( the probability say with periodic boundary conditions, is govemed by the master equation = = ) PI(") £ = «~Y is the where «'YJ Pf("~~ y, c(x, y, « ) PI(«)) (2.I) , (xy)e (xy) (c(x, A configuration obtained from @fter exchanging « the neighbor nearest sites : «~Y(z) (z) ~y) «(x) « if = z if = « z if = The process which on «i, t local m 0, in the infinite f(«), functions Lf(«) volume z = x z , z = x, = y. limit is (- I, +1) e « # ~ # y , (2.2) conveniently is defined by c(x,Y, «) ~f(«~Y~ defined its via generator -f(«)) L, (2.3) (.<y) The c(x, sum y, in «) formulation = (2.3) is over c~y, x, «) m is postponed nearest 0. Further until neighbor sites assumptions after dynamics. The simplest example corresponds the satisfy the condition rates ~0(~, have we to y, ) on " ye the introduced infinite an " x, Co (X, y, Z( rates some temperature W ~~ ix y( must well = be I, and imposed known Kawasaki of course but type dynamics, of their such where (2.4) 672 JOURNAL DE Clearly, all Bemoulli measures with this evolution. Finite temperature p time m satisfying 0, the condition cp (x, y, ) « « («~)~ m = M e [- I, dynamics Kawasaki detailed cp (x, y, = I spin average of PHYSIQUE ], I + have invariant are 5 for cp(x,y,«), rates balance (- p (H(«~Y~ exp ~Y~ H(«))) (2.5) , where H(«) £ m fl JA A is a local translation equilibrium whenever 0 = = from Hamiltonian, invariant that JA JA, and J~ ~ ~ Gibbs-measures pp for (2.6) mechanics ( WA ) as a convergent magnetic p = 0, there measures, h3~«(x) least how co(x, y, to are further a rates perturbed The = the express 2 c numbers say if diam (A ) dynamics, and the such The R. m know we functions correlation dpp(«) «(x) (2.7) course, always may we verified. As corresponding the add the in different to Gibbs called canonical local Hamiltonian. so a case discussion. in dynamics with generic perturbations ) « for real are large, = interested look Z~ finite AC of A is too measures to fl m p (2.6) expansion around p 0. Of (2.6) and (2.5) remains to parameter family of Gibbs one measures invariant for the dynamics. They are a which [11] for see are therefore rates temperature term is at magnetizations, We high field (JA, reversible then are statistical I.e. diameter the (x) « xeA (x, Y, ) « satisfying (2.5) for not of the infinite process is i = qA dynamics, having temperature then z + any assumed (x, Y) have to We constant rates (2.8) «A , A where WA fl m «(x). (qA(x, y)) coefficients The taken be must so that the (2.8) are (2.13)). We rates xeA non-negative, also local that assume (A (, is odd. bounded and the rates Translation are of the configuration « (see (2.12) of qA(x, y) 0 whenever the number condition that is imposed by the functions even : invariance any a e I.e. if axes, Z~. Finally, (e~)~ are that assume we the unit ~A " the " a(X + + model in vectors ~0~ A(°, ea') elements = ~A(X, Y) for and is Z( (2.~) ~) ~, Y + symmetric reflection in A, over all d coordinate then ~A(°, ea') If £Y # £Y' , and qo~ A(0, ea) = qA(0, e«) « , d, i, = (2,io) , where @~Am ((xi,..,-x~,..,x~):x= (xi,..,x~,..,x~)eA). Let q~Bm~ £ j.<y) (q~xy~B(x,y)-q~AB(x,y)) (2.ii) N POWER 5 with A « the symmetric the » labels LAW of (xy). sites the difference STOCHASTIC IN DECAY between q~~ if 0 = GASES obtained and A~Y is sets, require further We LATTICE from 673 A by exchanging that (A diam ) AB (2.12) R, m and q~~ for certain measures. perturbation order In in mind model not oneself loose to dimensional two Hamiltonian « to There I. « y as y use in the is no an (2.13) B priori a expansion and « ) which we exchange the e~, + x for notations following simple example C(x, for and co ~ therefore try A, information on and parameter the set up stationary formal a theory. the is R constants We for all y w rates (- exp = will the most take up for are general again at a model the good it is end of keep to section 4. The 1, 2 given by = p~jH(«~'~~~~) H(«)j (2.14) » H(«) £ 2 = «(x [Ki «(x) + ei) K~ «(x) + «(x + e~)] (2.15) x pi, p Here of ~ detailed question, the 3. they and if small are (2.5). The happens balance what both equal to p, then (2.14) satisfies some problem this paper investigates might be if pi # p~. stationary correlations are main the to condition the by summarized expansion. The unperturbed process tool in its analysis is the corresponds to the simple symmetric exclusion process. self-duality [7]. In this case, the generator (2.3) is The Lo f(«)=( £ The main jf(«xY~-f(«)j. (3.i) (~y) If we define for finite any Z( A < function the D (A, ) « fl m check that Lo D is same both Lo D(A, « the it when acts £ = on (D(A~( A and « ) on m WA, then it is easy to I.e. «, (A, D (x) « A xe « )) (3.2) (<Y) We Z( therefore with consider the dual called so process A(t), m0, t on the finite subsets of generator Lo f (A ) = ~ £ f (A ~Y~ f (A (3. 3) , jxy) and let pi(A,B) collection interaction, of be random end corresponding transition probability, I.e. the probability that a starting from the sites in A and subject only to a hard core unless A B (, and set B at time t. Obviously, pi (A, B ) is zero the walkers up in the = formally, ~. j £ lm dt~Pi(A, xy o B) -Pi(A~l B)) = 3~,B (3.4) JOURNAL 674 is if A one and B = otherwise. zero Et where denotes (3,2) From Et(D(A, WA) PHYSIQUE DE derives one zPi(A, dP («) = B duality the D expectation in the simple symmetric the M I (B, « exclusion relation 5 : ), (3.5) process started Li let I from the p. measure Consider the now expectation with (with «rprocess perturbed respect to By integrating E"lD(A, Al « (3.6) " overs from s £pi(A, B ) p of the with commutes ~ljD(Ai-w 0 "s)I to ID (B, t and L Lo = + (with and « ) denote the and the Lo) generator D(A, since Lo D(Ai-s, ljLi " = Lo D(A, ) « "s)1 (3.6) get we )j « (2.3) L Arprocess generator product L). Lo generator with process the + s i £ ds + ~(A, pi ) E"[Lj B D (B, «~)] (3.7) , o B expectation in the perturbed with initial «i-process measure p. perturbed process with rates (2.8) has as initial state p the Bemoulli measure with magnetization («~)~ from (3.7), the time evolved spin-spin correlations 0. Then, satisfy the equation with E" the Suppose the = («~) 3~ = j ii ds + £ o (3.8) ~(A, B) v~(B) pi B with v~(B) and from (Lj m D (B, « (2,8), (2,11), Us(B) = = ,i («BXY- «B) lC(x, z I WC1~ qBc Y, « ~ ) (3.9) c Substituting (3.9) (3.8) into we iterate can obtain to («Al = z VI (t) V$(t) where WE AJ, (3. lo) , i and VI (t) ids z = Pi -s(A, B qBc Vt~ n » (3.I1) , B,c The first ~"A)1' question is to see when (3.10), (3.ll) defines a convergent expansion for N LAW POWER 5 PROPOSITION evolved If tyR < I, For fixed I. correlations. Proof: By locality, then time t, STOCHASTIC IN DECAY (3.10)-(3. II) 4bfines («~ ) is of the 675 GASES LATTICE convergent order y ~ a expansion for the 'i y time small. (2.12), see £ pj(A, ) B if 0 q~c A C m = + (3.12) R B by Therefore, induction (3.ll) from V( (t) 0 whenever A On hand, other the (2.13) from : [Vj(t)[ Assuming for that some from have (A] if «ty (3.14) «R. m1 n V(~~(s)[ we (3.13) nR m = y~~~ R~~~s~~~, « (3.15) (3.13) that i'~(~) ii ~ ~S(N~l') (l'~ ~ ~~ S~ ~) ~ l'~ ~~ (~.l~) ~~ 0 that so By (3.10) this 2. If PROPOSITION yRt for converges construction I. < satisfies series the £ sup A (3.10)-(3.ll) then Proof: and Note that perturbation (3.17) c(x, y, invariance requires «) and times t (3.17) co < , from follows be a not x m co. < finite, it to do we (q~~ B only for #1/2 all is done. the are we ondition converges the Since for equations (3.8). evolution consider number this in finite trictly finite of ere ( = the sense as above that that bonds case («~) estimates sinfilar any further. (3.19) V$ " n=o where V$ w 3 ~j, Vi m £ G (A, B ) q V)~ ~c n (3.20) m , s, c and £ s G (A, B ) f(B) i m lim itm 0 £ s p~(A, B ) f(B) ds. (3.21) 676 JOURNAL There of are for First, (3.19) q~c. that the series reflect the question is They generally, More (3.19). On meaningful at for is argue we coefficients the time, finite every result convergence indicate that no problems many course expansion (3,10) the in well we some respects. and whether the f(B) £ additional two due have that may of form present in the dynamics. for functions f the of Kc qBc the law conservation (3.21) exists special the to we but results 5 the control cannot we purely formal, co have termwise the limit = Since I t hand, defined symmetry 1ilnit is the other least N I (3.19)-(3.21). with only not is PHYSIQUE DE form (3.22) c If that assume we Kc coefficients the Q~(N, f) f(B) is a quadrupole « £ xl £ m is B: then, (3.22) in » =N £ (3.23), (3.24) f(B) but we £ =N co < f(B) is =N a I, = , 2,.., 1, N (3.23) d , B =N xeB £ 0 = f(B) x~ f(B) x~ x~> 0 = Q~(N, f) 3~, = (3.24) ~, xes probably sufficient proof for general no are have numbers s xe £ B; Properties G (A, B all = f(B) £ £ for that ; B B-18 such are Gf for show now Gf(A)m by defined be to We A. sets how it be done Gf((0,x) )m can s the in where case V(x) (0,x) A = be can found £ consists unique the as of at solution most of Gf ( (0, (Gf ( (0, x)Y~) sites. two (3.4) From equation the )) x (0, x) f = (3.25) , ~~ or, ~ z da (v(x) 3x,ov(ea)) + (3.26) (x), p = conditions V(x) with boundary 0 (x( In (3,26), A~g (x) g(x + e~) + as co, Laplacian in the direction a and p (x) g(x e~) 2 g(x) is the discrete f( (0, x) is a quadrupole. Using the properties (3,23), (3.24), we have that - - w = for some x~ (k) Here, it suffices defined solution A second measures. measures = bounded to note We are ~ at k 0. I cos k~ (k) X ~ This will be the explicitly in section 4, equation (4.7). expression in (3.26) allows to find a well derived = substituting that z « ~~ above V(x). argument meaningful for dk e'~~ (X) P certain tum yielding dynamics therefore reversible. In to the substance on the which to we Kawasaki Appendix we the have detailed dynamics derive expansion formal the defined usual (3.17)-(3.19) information in high (2.5) about for temperature is the which it is that stationary the Gibbs expansion for N POWER 5 LAW equilibrium correlations fully relying but measures DECAY dynamical syInmetry « on » simplify to of formal our (2.5) LATTICE using any properties expansion (3.19). without the STOCHASTIC IN each the We corresponding find thus expansion the in term 677 GASES how get to to Gibbs a the use strictly local function. Long 4. correlations. range Appendix the infinite perturbation expansion around in dynamics reproduces local high expansion the case of temperature temperature detailed satisfy condition. balance dynamics. Generic perturbations are not expected to such a investigate the behavior is of function predicted from each what the two points Here as we in the expansion, for generic perturbation. such Each V(o,~j, nml, term term a Z( in the expansion (3.19) satisfies the equation (3.26) with x e the In show we formal the that the strictly (x) P z m (Q j (Y, o, xi Qj ) z o, xi Y~ (Y, z (4. i ) ) (Yzi and Q~ (Y, Z) £ " ~A (Y, AB Z ) i'~ B translation The invariance Qi (Y, for any Z( as by and reflection Qi ) Z P(~) (4.I) consequence, a i (Q(0,-xj " + + a) e rewritten ~, that + Z (4.2) ) ~ (2.10), summetry be can a(Y " Qi (°, As (2.9), implies model, the of Qi " A (°, (4.3) e~ as Q)0,x+e~j(0,~a)+Q(0,xj (0,~a) (0,~a)~ ~Q)0,-x+e~j(0,~a))~ d £ A~ 3~ oQ)0 a The behavior functions Vii of ~j, Q( o,~j (0, e~) fixed for Q (0, o, xi (x( as ea £ = the qj order ~~ (0, ea ) VI I( correlation then m oj (4.4) ~ depends on the (n I )-th origin. Indeed, if (x( R, co - around A sets (0, e~) j e ' =1 (4.5) xi ~ First least holds consider the situation exponentially fast to for p (x). Taking where as zero V$j( ~j the ix - co. is That transforms, Fourier are k = 4a (k) " £ Q( 0, xi quasi-local in the sense certainly verified for n (kj,.. k k ~) e [, (0, ~a ~,.., they decay that I. Then, this at also = gr, gr ]~, (4.6) ) C'~~, x WC get > (k) = z P d (X) e'~ = z (I cos k~ ) Xa (k) (4.7) 678 JOURNAL PHYSIQUE DE N I 5 where e~'~~) 4a(k) (1 '~« analytic is " 0. k at (i One V)o, the j ~ neighbor nearest £ = ~~ ~ + k~) cos find first can = e'~~) 4a(- k) (i + i°'ear ~~ '~" ~~ ~~ functions correlation equal to (4.9) (T~ ~)~~ M~ , ~ i y Where the is T~ inverse the matrix 3~~ + of T~~ « given by T I (l _;~ Me k~) cos (4.10) ~ £ (2 gr) ~ (I k~,) cos and £ M~ Equation (3.26) order is then dk m (2 gr (1 k~) cos 2 (4. II) ia(k) £ (cos k~ £ Vj = 2 ix the if system analytic co at - least is o,xj e'~~ the n-th a (1 k~) g~(k) cos ~4 i~~ ~ £ = isotropic, k=0, at obtain to j ~ (I and exponentially the expansion, fast. the I«(k)=I(k) points two We thus all for = V(o,~j locality the .,d, then S~(k) decays to zero I, a function that see k~) cos x is V)o Xa(k) m £ f(k) 1) as S~(k) Hence, (k) X« ~ )~ by putting solved function structure e~'~~ correlation of the = as functions points functions, at order n, In particular this is true in assume we isotropic perturbation of an d I. This is infinite dynamics still gives rise to short spatial correlations in the temperature range S~(k) x~(k) equal, then is analytic stationary If however the state. not not at are correlations strictly local, anisotropic I, k 0. So, even if at order n I the are e-g- n correlations perturbations may lead to long range stationary order n. From (4.12), the two at points function then decays as order at f n I in is reproduced, least at for the two of the lattice. system has the full summetry consistent with [10], where it is also the case that an that = the = = V(o~j ~2 m£b] (X( (x( ~, ~ (4.13) -co, a for some constants b]. anisotropy is certainly not that example happen that x~(k) 2 S~(k) Note = Qj o,~j 2 e~ ) V)o J(x) produce to ~ J(x = sufficient + j. As an J(e~)13~,~~ this weak illustration, decay. suppose It may for that (4.14) 30, ~j , M POWER 5 J(x) strictly local, Then, x~(k) 4 with )(k). DECAY LAW )(k) symmetric, J(0) 4J(e~) and V(o,~j fact the reflection = isotropic a not. or I,.., in is anisotropy generic However, multiple This we rates be = c(x,x The energy « parametrized by the expansion of the detailed is (4.14) has Q[ As a 0,x) on be to " a such whether J(x) is dynamics, see (A24). simplest example is the Choose p~ close to zero, balance relations. The announced in (2.14). matter given by X« (k) [)(k) p 4 = pa iJ(X " (4.16) p~ all p, = (Al), or p~ + all not are Kj(3~ J(e~ )] and equal. In first in p. order + in P2) K~ + (-4 d with 2 (4.12) must K~(3~ nearest 3~ + neighbor coupling ~~) ~~ (4.18) , (4.9)-(4.ll), (-2gr+12(pi = (4.1?) 3x,0) correspondingly f~ (k) = + W-2 ~~j pa J(~a)(3x,e~ + ~~ ~~ (2 W-2 K~) W 12 ar ~~ Kj + + gr ~~ +18 gr gr ~~ ~~ + ~~((4gr-18+~~)Ki+ K~) mvi ar V)o J(X)i 3~ + ~, integrals + (4.19) V~. m the f~(k) 4 = p ~[Kj ki cos K~ + k~] cos + 2 V~, K~ p a ~ explicitly V)o~j (4.16) is the then to up ~ the the ea) '' are (4.15) «~y) functions (°, ~a) 0,x) doing after y) «(x) J(x )o~j =(pi+P2)Kj+~~ Therefore, no transform Fourier = whenever a £ m = find, J(x), = balance J(x) we 2 having detailed J(x) as in (4.14). If exchange rates (2.5), (2.14) replaced for n I by local ~( consequence, depend and » H(«) Now, =0, 679 GASES LATTICE w( (i -(paiH(«~'~+~a) H(«)1) «) ea, + satisfy not exchange model the let does which temperature for scenario of [9] d, and STOCHASTIC IN determined. ~fi-4 p £ =2p~J(x)+ transforming (4.12) gives Fourier Inverse K (4.20) 1, 2 = V ~ A~I(x)+ ~ «=1,2 ~ + ~ 2 ~ [Kj hi (I (x ej ) + I (x + ei )) + K~ hi (I(x e~) + I (x + e~))] (4.21) where 1(x) « ~°~ dk (2 gr )~ l ~ (1/2) [cos kj ~ + cos k~] (4.22) 680 JOURNAL is the potential (4.21) that = derived be can pi if simple kernel for p~ p, = then (x( as walk o,~j the on pJ(x). 2 pi If = N I plane. # p~, We immediately verify then the asymptotic 5 from behavior from jxj xii ? (xj x])2 + ~ ~~~~~~ ~~~~~~ Hence, random V) PHYSIQUE DE ~~ ' ~ ~~'~~~ " co, - °, Xl ~ (~2 g~ xi) f~(0) ) (xl ( ii (0) i ~'( l ~~'~~~ ~ ~2)2 2 and (4.24) Kj is = K, m then (P P~)(Ki i = typical the K~ ~)-/ i~(0) ii (0) + K~) decay predicted in (4.14). points function still decays as field-like quadrupole (4.25) is zero but the two (4.25) Note a that power : if in this case, "2fl21°(3x,el~ ~x,-el~~x,e2~~x,-e2) fl2) l°hi~(X) fl2) l°hl~x,0 ~(fll + (PI xl + xl 6 xl xi 12 m--(pi-p~)K (x(-co, (4+X~)~ i')0,x) correlations octopole field. We thus get always long range that pi # p~. induction hypotheses on the V(jj, n m I must be Summarizing, we find that the correct is described that their decay, as (x( slower than in (4.14). flu (k) is no longer not co, derivation of (4.13) remains meaningful since the necessarily analytic at k 0 but the corresponding f~ (k) are then still bounded k Generic anisotropy then gives rise to 0. near has the for this long distance (4.26) , '~ behavior of an provided model temperature two - = = the decay (4.14) law power also for the order next in expansion. the Acknowledgement. We are indebted to Jan Naudts for a careful reading of the manuscript, and for useful criticism. Appendix. Derivation We choose of the the high rates expansion temperature c(x, y, «) c(x, of the y, « ) = da Kawasaki dynandcs. form ~P (p (H(«x~ H(«))) (Al) +(- z) (A2) with e~~'~ *(z) = +(z), + (z) = M 5 LAW POWER We P(0) that assume can DECAY IN STOCHASTIC P(z) that and (z) I = f cases, around for as Metropolis the = bi z~ o, (A3) even P(z) where rates 0 z = P Other analytic is 681 GASES LATTICE ~' '~, can be satisfy the following e~ = treated uniform via approximation. then We ~P ~~)(0) that have a~, m derivatives the 4l(z) of at 0 z relations = : if then (~ ( i )~ ~n ~n " (A4) k and ~~ £ (~ l)~ (A4) (A2) is immediate an (~) an-k that (A5) is (_ equivalent i )n and even £(z) is (~ ~~ e~~ ~P(- z). As = 0 (2~5) Sn Sm , (A5), for note we that from ~ 0 m z ~ ~ (A6) 1<2 (m~ 1/2 z i (z ("f ~n ~ this But n. « (- I) (Z~)iz ~ d~ k ~n o mm ~f hi to ( d~ k all f for an-k (A3) and ak so ~P(z) of consequence ~~ i " n-m (z 2 k obvious is m ~~~ ~ 1<2 z the i (z 2 m (n 0 because i + k function m (z 2 i + n-m (A8) 2 odd. Since they we appear want in to the obtain rates an expansion and ~~ (X, Y) qic m in p, have we the separate to different orders in p as define ~ ~i " z (qixy~c (x, IR.AAf=Ar~l I fl qi~c y) ~A)Y) (~A~ f (x, y)) = , 1, 2, (2~~) (Rio) xy) so that c(x, Y, « ) i + = ~£ £ q( (x, i The expansion (3.19) now the has y) «~ l. (Al i) A form ( («A)~ " V$ (A12) n=o with V$ =~ m ~£ £ B,c and V$ = 3~j. G (A, B ) qjB~ VI, n m I (Al 3) JOURNAL 682 We by computing start first the order Vj £ = PHYSIQUE DE M I 5 term G (A, £ G ) B q j~ s = (A, £ ) B that so (J~ (A14) J~xy) (~y) s by (3.4) Lo V( Lo JA. Using the boundary that V( =J~. We will repeatedly use V( condition this 0 - = conclude argument (A diam as in - derivation the co, we of the following. PROPOSITION 3 ~/ V$ ~' It is that clear we decay exponential thus get a Proof.- By induction, computed was £ (A, G above. flJ~, p~~ i ~ 0,.., gives calculation (n~~ ~ ~ =1 oYl A n~ ~~~ An r , = with (A16) lwkwn-1. jjJ~ £ I £ ~ ~ (g~i(Ai, x m small, fl r=1 straightforward Aj, for for k AAk=A '~ where, expansion temperature ~ A qic _~ VC~ B r n high £ AIR.. I (Al 5) J~ that assume ~' = A AA A. correlations. pk V(=~ k ~ fl ~ At converging spin-spin of £ = f ~~ n-f )] ~ ~~ ~~~' ( Al 7) '~ ~~~ ~ n, gy~ (Aj, A ~) (A, G m , (YA.. A AA (YAA~~ j A.. AA (A18) ~) Hence, ~~ p ~ )n n ~~ n ~~ Al, + =~ z (, an ~l ~~ ~A ~) ~~~~ ~~ ~~ () £ -k nr~l fl p £ (gin JA, ~ n A V$ pn = ~ p (- l)~ £ a~ ~' (~) fl A.. AA A n r ~~ m! (n m )I I g im (A xyj =~ i (~ ~'~ an-k pn n AAn"Ar=1 n)) j, ) A ~ , ~n-kl(l~l~) JA + q ~' [I (-1)~ an] I n fl JA ~ AIR..AAn"Ar=1 (A20) JA ~ AI A that fl n £ n g10 (Al, n) , -1 ' AjA q A I conclude to ~' = n r ~o ((-1)~ £ (A5) and n ~'m~j ~ ~~ X (A4) (Al, , Aj, use ~ jay> ~ we ~ ~ n AI, and ~~~ ~~ ~~ ~ ~~ ~ M POWER 5 local The character of condition. balance of each implies It DECAY LAW the (c(x, using (Al I), for or, each n order £ £ k o n Vi terms for that STOCHASTIC IN each y, (K~ ). some separately, )(«Axy- WA)) « must we detailed the have that 0 (A21) ~A ~1°A'Y (2~22) = ~~ h~(X, y)) V~ (~~~Y~B(X, Y) " B Assume whenever 0 = case 0, 1. = condition KA -1 £ £ n " k o 0 - of the as B is set (A.22) has diam (A) Then the - co. larger unique (A22) fact, put In = Q~ (X, Y) diameter the for k = i'~, (2~2~) VixY, (3.48) ~~ h~(X, Y) B that so Qi(X, Y) QixY(X, Y) and s from I, m inductively that Vi than kR, 0 w n I, as we know is certainly the solution K~ V$ by imposing the boundary implies that V$ whenever diam (A) mnR. 0 for (xy) 683 directly understood be can bond GASES LATTICE (x(I),..,x let I, I, r = (r)) such be x(I) that e of sequence a Vi " neighbor nearest A, x(s) J A for s = , 2,.. I, r I)), (x(s)x(s+ sites diam (A~(~~~~~~) and m nR. , Then Vi VI V$x~i)x~2)+ V(x(1)x~2) = Q (x~i)x<2)(x(I ), x(2)) = + gives solution the V$x<i)x<3) + + Q $x<i)x~r)(x(r + Q I V$x~i)x<r-i) (x(I ), x(2) ), x(r)) + V$x<i)x<r) Q (x<1)x~3)(x(I), x(3)) Q(x~i)x<r- i)(x(r Q (x~i)x<2) (A24) ), x(r)) I (A22). of References [1] [2] RUELLE [3] PARK D., GALLAVOTTI Phys. Y. 7 [4] GEORGii and (1968) M., The (1982) H. Mechanics, Statistical G. MiRACLE-SOLE Rigorous S., Results Correlation Benjamin, QV. A. Functions of a Inc., New York, 1969). System, Comm. Math. Lattice 274-288. Cluster Expansion for Classical and Quantum Lattice Systems, Gruyter Studies J. Stat. Phys. 27 553-576. O., Gibbs Measures and Phase Transitions, De in Mathematics its Applications, (1988). [5] KUNSCH H., Decay of Comm. [6] [8] Correlations 84 (1982) under Dobrushin's Uniqueness Condition and 207-222. SCHLOSMAN S.B., Analytic Constructive Completely Interactions: Phys. 46 (1987) 983-1014. LiGGETT T. M., Interacting Particle Systems (Springer-Verlag, New York, 1985). of LEBowiTz J.L., MAES C, and SPEER E. R., Statistical Mechanics Probabilistic Cellular Automata, J. Stat. Phys. 59 (1990) l17-169. DOBRUSHiN R.L. Description, [7] Phys. Math. and J. Stat. 684 [9] JOURNAL P.L., GARRIDO Conservative LEBOWITz J.L., Dynamics, Rutgers MAES PHYSIQUE DE C, and University SPOHN M I H., Long Range Preprint (1990) Phys. Rev. A 5 Correlations for (1990) 1954- 42 l968. [10] Correlations for REDIG F., Long Range Spatial University preprint (1990). Scale Dynamics of Interacting Particles, SPOHN H., Large Preprint (1989). Mfinchen MAES C. and Anisotropic Zero Range Processes, Leuven ill] Part B: Stochastic Lattice Gases,