The Infinite Volume Limit of Dissipative Abelian Sandpiles C. Maes ∗ F. Redig † E. Saada ‡ 7th October 2002 Abstract: We construct the thermodynamic limit of the stationary measures of the Bak-Tang-Wiesenfeld sandpile model with a dissipative toppling matrix (sand grains may disappear at each toppling). We prove uniqueness and mixing properties of this measure and we obtain an infinite volume ergodic Markov process leaving it invariant. We show how to extend the Dhar formalism of the ‘abelian group of toppling operators’ toinfinitevolumeinordertoobtainacompactabeliangroupwithauniqueHaarmeasure representing the uniform distribution over the recurrent configurations that create finite avalanches1. 1 Introduction The abelian sandpile is a model on a lattice in which a discrete height-variable (e.g. representing the slope of a sandpile at that site) is associated to each lattice site. A configuration of height variables is called stable if the height at any site is less than a critical value γ, otherwise it is called unstable. The mechanism of stabilization of an unstable configuration is by a sequence of so-called topplings. In such a toppling, the site gives one (sand) grain to each of its neighboring sites which in their turn can become “unstable” and “topple” etc., until every site has again a subcritical height- value. An unstable site thus creates an “avalanche” involving possibly the toppling of many sites around it. The dynamics on stable configurations consists in adding at ∗Instituut voor Theoretische Fysica, Celestijnenlaan 200D, 3001 Heverlee, Belgium †Faculteit Wiskunde en Informatica, Technische Universiteit Eindhoven, Postbus 513, 5600 MB Eindhoven, The Netherlands ‡CNRS, UMR 6085, Laboratoire de math´ematiques Rapha¨el Salem, Universit´e de Rouen, site Col- bert, 76821 Mont-Saint-Aignan Cedex, France 1MSC 2000: Primary-82C22; secondary-60K35. Key-words: Sandpile dynamics, Nonlocal interactions, Interacting particle systems, Thermodynamic limit, Dissipative systems, Decay of correlations and Mass-gap. 1 randomlychosensitesandstabilizingtheconfiguration. Sincetheavalancheoftopplings can involve many sites (depending on the configuration) this dynamics is highly non- local. Therefore, defining infinite volume limits poses a non-trivial problem. Since its appearance in [1], the abelian sandpile model has been studied intensively, see for instance [4], [5], [8], [14], and also [12] for a mathematical review of the main properties of the model in finite volume. The main technical tool in the analysis of the model is the “abelian group” of toppling operators which can be identified with the set of recurrent configurations. Our aim is to define the model on infinite graphs. The main problem to overcome is the non-locality or extreme sensitivity to boundary-conditions of the dynamics. We have constructed in [9] and [10] the infinite volume standard sandpile process on the one-dimensional lattice and on homogeneous trees. In the present paper we focus on the thermodynamic limit for dissipative models. There, the infinite graph S is a subgraph of the regular lattice Zd and on each site the height has a critical value γ, an integer not less than the maximal number N of neighbors of a site in S. The finite volume rule now starts as follows: choose a site x at random from the volume V and add one grain to it. Suppose that x has N nearest neighbors and that the new height at x is γ+1. Then, it x topples by giving to each of its nearest neighbors one grain and dissipating γ−N grains x to a sink associated to the volume. We say that the site x is dissipative when γ > N x and the model is dissipative when this happens for a considerable fraction of sites. This condition can be rephrased in terms of the simple random walk on S with a sink associated to the dissipative sites: the model is dissipative when the Green’s function decays fast enough in the lattice distance, see (2.12) below for a precise formulation. Dissipative abelian sandpile models have appeared in the physics literature in [15] and [3], whereitwasarguedthatdissipationremovescriticality, thatis, correlationfunctions decay exponentially fast uniformly in the volume. From the point of view of defining the thermodynamic limit, the main simplification of dissipative models is that there is a stronger control of the non-locality: more precisely, the probability that a site y is influenced by addition on x decays exponentially fast (or at least in a summable way) in the distance between the sites. Hence avalanche clusters are almost surely finite. 1.1 Results Our three main results are: 1. The extension of the Dhar formalism to infinite volume sandpile dynamics. That includes the construction of a compact abelian group of recurrent configurations on which we can define addition (of sand) operations. 2. The construction of the thermodynamic limit of the finite volume stationary mea- sure with exponential decay of correlations in the case of “strong dissipativity”. 3. The construction of an infinite volume sandpile process which converges exponen- tially fast to its unique stationary measure. 2 1.2 Plan of the paper Thepaperisorganizedasfollows: inSection2werepeatsomeofthebasicresultsonthe abelian sandpile model in finite volume and we introduce the definition of dissipativity, with examples. In Section 3 we show how to extend the dynamics on infinite volume recurrent configurations and we recover the group structure of “addition of recurrent configurations.” In Section 4 we prove existence and ergodic properties of the infinite volume dynamics. Section 5 is devoted to the proof of exponential decay of correlations. 2 Finite volume model In this section we recall some definitions and properties of abelian sandpiles in finite volume. In [4], [5], [8], [14] and [12], the reader will find more details. The infinite graphs S on which we construct the dissipative abelian sandpile dy- namics are S = Zd, and “strips”, that is, S = Z × {1,...,‘}, for some integer ‘ > 1 (notice that ‘ = 1 corresponds to S = Zd with d = 1). Finite subsets of S will be denoted by V,W; we write S = {W ⊂ S : W finite}. We denote by ∂V the external boundary of V: all the sites in S \ V that have a nearest neighbor in V. Let N be the maximal number of neighbors of a site in S, e.g., N = 2d for S = Zd and N = 4 for S = Z × {1,...,‘},‘ ≥ 3. The state space of the process in infinite volume is Ω = {1,...,γ}S, with some integer γ ≥ N. We fix V ∈ S, a nearest neighbor connected subset of S. Then Ω = {1,...,γ}V is V the state space of the process in the finite volume V. We denote by N (x) the number V of nearest neighbors of x in V. A (infinite volume) height configuration η is a mapping from S to N = {1,2,...} assigning to each site x a “number of sand grains” η(x) ≥ 1. If η ∈ Ω, it is called a stable configuration. Otherwise η is unstable. For η ∈ Ω, η is its restriction to V, V and for η,ζ ∈ Ω, η ζ denotes the configuration whose restriction to V (resp. Vc) V Vc coincides with η (resp. ζ ). V Vc The configuration space Ω is endowed with the product topology, making it into a compact metric space. A function f : Ω → R is local if there is a finite W ⊂ S such that η = ζ implies f(η) = f(ζ). The minimal (in the sense of set ordering) such W W W is called the dependence set of f, and is denoted by D . A local function can be seen f as a function on Ω for all W ⊃ D , and every function on Ω can be seen as a local W f W function on Ω. The set L of all local functions is uniformly dense in the set C(Ω) of all continuous functions on Ω. 2.1 The dynamics in finite volume The toppling matrix ∆ on S is defined by, for x,y ∈ S, ∆ = γ, xx ∆ = −1 if x and y are nearest neighbors, xy ∆ = 0 otherwise (2.1) xy 3 We denote by ∆V the restriction of ∆ to V ×V. A site x ∈ V is called a dissipative site in the volume V if X ∆ > 0. xy y∈V Thus if γ > N, every site is dissipative. If γ = N, the internal boundary sites of V (that is all the sites in V that have a nearest neighbor in S\V), are the only dissipative sites in V. To define the sandpile dynamics, we first introduce the toppling of a site x as the mapping T : NV → NV defined by x T (η)(y) = η(y)−∆V if η(x) > ∆V , x xy xx = η(y) otherwise. (2.2) In words, site x topples if and only if its height is strictly larger than ∆V = γ, by xx transferring −∆V ∈ {0,1} grains to site y 6= x and losing itself in total ∆V = γ grains. xy xx As a consequence, if the site is dissipative, then, upon toppling, some grains are lost. Toppling rules commute on unstable configurations, that is, for x,y ∈ V such that η(x) > γ = ∆V and η(y) > γ = ∆V : xx yy T (T (η)) = T (T (η)) x y y x For η ∈ NV, we say that ζ ∈ Ω arises from η by toppling if there exists a k-tuple V (x ,...,x ) of sites in V such that 1 k k Y ζ = ( T )(η) xi i=1 The toppling transformation is the mapping T : NV → Ω defined by the requirement V that T (η) arises from η by toppling. The fact that stabilization of an unstable configu- ration is always possible follows from the existence of dissipative sites. The fact that T is well-defined, that is, that the same final stable configuration is obtained irrespective of the order of the topplings, is a consequence of the commutation property, see [12] for a complete proof. For η ∈ NV and x ∈ V, let ηx denote the configuration obtained from η by adding one grain to site x, that is ηx(y) = η(y)+δ . The addition operator defined by x,y a : Ω → Ω ;η 7→ a η = T (ηx) (2.3) x,V V V x,V represents the effect of adding a grain to the stable configuration η and letting a stable configuration arise by toppling. Because T is well-defined, the composition of addition operators is commutative. We can now define a discrete time Markov chain {η : n ≥ 0} n on Ω by picking a point x ∈ V randomly at each discrete time step and applying the V addition operator a to the configuration. We define also a continuous time Markov x,V process {η : t ≥ 0} with infinitesimal generator t X L0,ϕf(η) = ϕ(x)[f(a η)−f(η)]; (2.4) V x,V x∈V this is a pure jump process on Ω , where ϕ : S → (0,∞) is the addition rate function. V 4 2.2 Recurrent configurations, invariant measure The Markov chain {η ,n ≥ 0} (or its continuous time version {η }) has a unique n t recurrent class R , and its stationary measure µ is the uniform measure on that class, V V that is, 1 X µ = δ . (2.5) V η |R | V η∈RV A configuration η ∈ Ω belongs to R if it passes the burning algorithm (see [4]), V V which is described as follows. Pick η ∈ Ω and erase the set E of all sites x ∈ V with a V 1 height strictly larger than the number of neighbors of that site in V, that is, satisfying the inequality η(x) > N (x) V Iterate this procedure for the new volume V \ E , and so on. If at the end some 1 non-empty subset V is left, η satisfies, for all x ∈ V , f f η(x) ≤ N (x) V f The restriction η is then called a forbidden subconfiguration (fsc). If V is empty, the V f f configuration is called allowed. The set A of allowed configurations coincides with the V set of recurrent configurations, A = R (see [8], [12], [14]). V V A recurrent configuration is thus nothing but a configuration without forbidden subconfigurations. This extends to infinite volume: Definition 2.6 A configuration η ∈ Ω is called recurrent if for any V ∈ S, η ∈ R . V V The set R of all recurrent configurations forms a perfect (hence uncountable) subset of Ω. This means that R is closed (hence compact) and every element η ∈ R is the limit of a sequence η ∈ R,η 6= η. n n On the set R , the finite volume addition operators a can be inverted and they V x,V generate a finite abelian group. This group is characterized by the closure relation Y ∆V a xy = Id (2.7) y,V y∈V By the group property, the uniform measure µ is invariant under the action of a V x,V and of a−1 . x,V 2.3 Toppling numbers For x,y ∈ V and η ∈ Ω , let n (x,y,η) denote the number of topplings at site y by V V adding a grain at x, that is, the number of times we have to apply the operator T to y stabilize ηx in the volume V. We have the relation X η(y)+δ = a η(y)+ ∆V n (x,z,η) (2.8) x,y x,V yz V z∈V 5 Defining Z G (x,y) = µ (dη) n (x,y,η) (2.9) V V V one obtains, by integrating (2.8) over µ : V G (x,y) = (∆V)−1. (2.10) V xy In the limit V ↑ S, G converges to the Green’s function G of the simple random walk V on S with a sink associated to the dissipative sites (that is every site x is linked with γ −N (x) edges to a sink and the walk stops when it reaches the sink). By (2.9), the S probability that a site y topples by addition at x in volume V is bounded by G (x,y). V Definition 2.11 We say that the sandpile model is dissipative if X sup G(x,y) < +∞ (2.12) x∈S y∈S In our examples, if γ > 2d for Zd or γ ≥ 4 for strips, the Green’s function G(x,y) decays exponentially in the lattice distance between x and y and hence (2.1) defines a dissipative model. From now on, we restrict ourselves to these cases. Definition 2.13 For any integer n, let ν be a probability measure on Ω , with Wn Wn W ∈ S, W ↑ S. Then ν converges to a probability measure ν on Ω if for any n n Wn f ∈ L, Z Z lim fdν = fdν. n→∞ Wn We denote by I the set of all limit points of {µ : V ∈ S} in the sense of Definition V 2.13. By compactness of Ω, I is a non-empty compact convex set. Moreover, by (2.5) and Definition 2.6, any µ ∈ I concentrates on R (see [10]). 2.4 Untoppling numbers On the set R the addition operators a are invertible. The action of the inverse V x,V operator on a recurrent configuration can be defined recursively as follows, see [8]. Consider η ∈ R and x ∈ V. Remove one grain from η at site x. If the resulting V configuration is recurrent, it is a−1 η, otherwise it contains a forbidden subconfiguration x,V (fsc) in V ⊂ V. In that case “untopple” the sites in V . By untoppling of a site z we 1 1 mean that the sites are updated according to the rule η(y) → η(y)+∆ . Iterate this zy procedure until a recurrent configuration is obtained: the latter coincides with a−1 η. x,V As an example, consider a graph with just three sites a ∼ b ∼ c for γ = 2. The configuration 212 is recurrent. After removal of one grain at site c, we get 211, which contains the fsc 11. Untoppling site b gives 130, and untoppling site c gives 122, which is recurrent. Conversely, one verifies that addition at site c on 122 gives back the original configuration 212. 6 Call n−(x,y,η) the number of untopplings at site y by removing one grain from x and V from untoppling sites until a recurrent configuration is obtained. As in the previous section, one easily proves the relation Z n−(x,y,η)µ (dη) = G (x,y) (2.14) V V V 3 The group of addition operators in infinite volume In this section we show how to obtain the group of addition operators in the infinite volume limit. The assumption of dissipativity is crucial in order to obtain a compact abelian group in the thermodynamic limit. 3.1 Addition operator The finite volume addition operators a (cf. (2.3)) are defined on Ω via x,V a : Ω → Ω : η 7→ a η = (a η ) η . (3.1) x,V x,V x,V V V Vc (with some slight abuse of notation). Similarly, the inverses are defined on R via a−1 : R → Ω : η 7→ (a−1 η ) η (3.2) x,V x,V V V Vc Remark that if η ∈ R, then (a η) ∈ R for all W ⊂ V but a η is not necessarily x,V W W x,V an element of R. Definition 3.3 For η ∈ Ω, we say that the limit of the finite volume addition operators is defined on η if for every x ∈ S, there exists Λ ∈ S such that for any Λ ∈ S,Λ ⊃ Λ , 0 0 a η = a η; in that case, we write x,Λ x,Λ0 a η = a η x x,Λ0 Similarly, forη ∈ R, wesaythatthelimitofthefinitevolumeinverseadditionoperators is defined on η if for every x ∈ S, there exists Λ ∈ S such that for any Λ ∈ S,Λ ⊃ Λ , 0 0 a−1η = a−1 η; we write x,Λ x,Λ0 a−1η = a−1 η x x,Λ0 Remark that if η ∈ R and a is defined on η, then a η ∈ R. x x Lemma 3.4 Assume (2.12). For any µ ∈ I there exists a tail measurable subset Ω ⊂ Ω such that: 1. µ(Ω) = 1; 2. The limit of the finite volume addition operators and their inverses is defined on every η ∈ Ω. 7 Moreover, every µ ∈ I is invariant under the action of a and a−1 , that is, for all x x x ∈ S and f ∈ L Z Z Z f(a η)µ(dη) = f(a−1η)µ(dη) = f(η)µ(dη) (3.5) x x and a a−1 = a−1a = id on Ω. x x x x Proof. We prove the result for the addition operators, the analogue for the inverses is proved along the same lines by replacing “number of topplings” by “number of untop- plings”. Pick W ∈ S,W ↑ S such that µ → µ and x ∈ S. We have to prove that k k W k µ[∀Λ ∈ S,∃V ⊃ Λ : a η 6= a η] = 0 (3.6) 0 0 x,V x,Λ0 We enumerate S = {x : n ∈ N}, with V = {x ,...,x } such that V ↑ S, x ∈ ∂V . n n 1 n n n n−1 If a η 6= a η, then some boundary site of V has toppled under addition at x in x,V x,Vn n volume V. This implies that for every m such that V ⊃ V some external boundary m site of V topples upon addition at x in V . Therefore, the left hand side of (3.6) is n m bounded by µ(cid:2)∀n ∈ N,∃p ≥ n,∃y ∈ ∂V : n (x,y,η) ≥ 1(cid:3) n Vp and we have to estimate (cid:2) (cid:3) µ ∃p ≥ n,∃y ∈ ∂V : n (x,y,η) ≥ 1 (3.7) n Vp Since n (x,y,η) ≤ n (x,y,η), Vp Vp+1 (cid:0) (cid:1) µ ∃p ≥ n,∃y ∈ ∂V : n (x,y,η) ≥ 1 ≤ lim µ(∃y ∈ ∂V : n (x,y,η) ≥ 1) n Vp k→∞ n Vk Z X ≤ lim n (x,y,η)µ(dη) V k→∞ k y∈∂Vn Z X ≤ lim n (x,y,η)µ (dη) W W k→∞ k k y∈∂Vn X = G(x,y) y∈∂Vn which implies that (3.7) converges to zero as n tends to infinity, by condition (2.12). Finally, (3.5) follows easily from Definition 3.3, µ(Ω) = 1, f ∈ L, and the invariance of µ under the finite volume addition operators a and a−1 . V x,V x,V Notice that for η ∈ Ω, we can take the limit V ↑ S in (2.8) and write X η(y)+δ = a η(y)+ ∆ n (x,z,η) (3.8) x,y x yz S z∈S for any x,y ∈ S, where n (x,z,η), the number of topplings at site z ∈ S by adding a S P grain at x, satisfies n (x,z,η) < +∞. z∈S S 8 Lemma 3.9 Assume (2.12). For any µ ∈ I there exists a tail measurable subset Ωo ⊂ Ω with µ(Ωo) = 1 such that for any V ∈ S and n ,x ∈ V integers, the product Q anx x x∈V x is well-defined, as the limit of Q anx as Λ → S, on every η ∈ Ωo. x∈V x,Λ Proof. We fix V ∈ S,x ∈ V,n a positive integer, and we prove that anx is well- x x defined on Ω (the case of negative n is similar and the extension to finite products it x straightforward). Following the same lines as in the preceding proof, we have to replace (3.6) by µ(cid:2)∀Λ ∈ S,∃Λ ⊃ Λ : anx η 6= anx η(cid:3) = 0 0 0 x,Λ x,Λ0 We denote by E (n ,x,z,η) the event that addition in V of n grains at x causes at Vp x p x least one toppling at z. As these events are increasing in p, we estimate (cid:0) (cid:1) X µ ∃p ≥ n,∃y ∈ ∂V : E (n ,x,y,η) ≥ 1 ≤ lim µ (E (n ,x,y,η)) n Vp x k→∞ Wk Wk x y∈∂Vn X ≤ n G(x,y) x y∈∂Vn where the last inequality is a consequence of (2.10) and (3.5). From this we deduce that for any V ∈ S, n = (n ,x ∈ V) ∈ ZV, the product Q anx is well-defined on a tail x x∈V x measurable set Ω(V,n) of µ-measure one. The set Ωo is then the countable intersection Ωo = ∩ Ω(V,n) V∈S,n∈ZV of tail measurable µ-measure one sets. The following proposition extends this to addition on infinite products. Proposition 3.10 Assume (2.12). Ifn = (n ,x ∈ S) ∈ ZS satisfiesP |n |G(0,x) < x x∈S x +∞, the product Q anx is well-defined on a set Ω(n) of µ-measure 1, for every µ ∈ I. x∈S x Proof. Take n ≥ 0 for every x ∈ S; the case of negative n is treated again by replacing x x “topplings” with “untopplings”. It suffices to show that for every Λ ∈ S 0 ! ! ! Y Y µ ∃V ,∀V ⊃ V ,∀y ∈ Λ : anxη (y) = anxη (y) = 1 0 0 0 x x x∈V x∈V0 or ! ! ! Y Y lim µ ∃V ⊃ V ,∃y ∈ Λ : anxη (y) 6= anxη (y) = 0 (3.11) 0 0 x x V0↑S x∈V x∈V0 The left hand side of (3.11) is bounded by the sum ! ! ! X Y Y µ ∃V ⊃ V : anxη (y) 6= anxη (y) (3.12) 0 x x y∈Λ0 x∈V x∈V0 9 If none of the external boundary points of Λ topples upon addition of n grains at 0 z z ∈ V \V to the configuration (cid:0)Q anxη(cid:1), we have that for all y ∈ Λ : 0 x∈V0 x 0 ! ! Y Y anxη (y) = anxη (y) x x x∈V x∈V0 Since µ is invariant under the a , see (3.5), the sum (3.12) is bounded from above by x X X X X X X µ(E (n ,z,x,η)) ≤ n G(z,x) S z z y∈Λ0|x−y|=1z∈V0c y∈Λ0|x−y|=1z∈V0c which implies (3.11) by the hypothesis on n. 3.2 Group structure Here we show that the product Q anx can be defined on any recurrent configuration, x∈S x provided we identify recurrent configurations which differ by a multiple of ∆. Given n ∈ ZS and η ∈ R, we consider the set A (η) = {ξ ∈ R : ∃m ∈ ZS,η +n = ξ +∆m} n Similarly, for subtraction, S (η) = {ξ ∈ R : ∃m ∈ ZS,η −n = ξ +∆m} n Fix n ∈ ZS so that X sup [|n |+2γ]G(y,x) = B < +∞ (3.13) x y∈S x∈S and let Ω = {η ∈ R : S (η) 6= ∅,A (η) 6= ∅} n n n be the set of recurrent configurations for which both addition and subtraction with n gives rise to a new recurrent configuration, modulo the toppling matrix applied to an integer function. Lemma 3.14 Ω = R. n Proof. We prove that Ω is closed. Let (η ) be a sequence in Ω which converges to n k k≥0 n η as k → ∞. For each k, there exist η± ∈ R and m± ∈ [−B,B]S such that k k η ±n = η± +∆m± (3.15) k k k Since R×[−B,B]S is compact, there exists a subsequence k → ∞ such that η± → η± i ki and m± → m±. Taking limits along this subsequence in (3.15) yields ki η ±n = η± +∆m±, that is, η ∈ Ω . Looking back at Proposition 3.10, Ω(n) ∩ R ⊂ Ω and Ω(n) is a n n µ-measure one (hence non-empty) tail set. Therefore it is dense and Ω = R. n 10