Inequalities for critical exponents in d-dimensional sandpiles

arXiv:1602.06475v2 [math.PR] 28 Sep 2017

∗
Sandeep Bhupatiraju Jack Hanson† Antal A. Járai
Indiana University City College of NY University of Bath

August 29, 2018

Abstract
Consider the Abelian sandpile measure on Zd , d ≥ 2, obtained as the L → ∞ limit of the
stationary distribution of the sandpile on [−L, L]d ∩ Zd . When adding a grain of sand at the origin,
some region, called the avalanche cluster, topples during stabilization. We prove bounds on the
behaviour of various avalanche characteristics: the probability that a given vertex topples, the radius
of the toppled region, and the number of vertices toppled. Our results yield rigorous inequalities for
the relevant critical exponents. In d = 2, we show that for any 1 ≤ k < ∞, the last k waves of the
avalanche have an infinite volume limit, satisfying a power law upper bound on the tail of the radius
distribution.

Contents
1 Introduction 2

2 Definitions and background 8

2.1 Graphs and sandpiles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.2 Toppling numbers and waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.3 Spanning trees and the burning bijection . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.4 Intermediate configurations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.5 Bijection for intermediate configurations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.6 Random walk notation and basic facts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.7 Wilson’s algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.8 Wiring o to the boundary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

3 Toppling probability bounds in low dimensions 18

3.1 Preliminary setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.2 SRW and LERW steering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.2.1 Estimates for a single walk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.2.2 Separation lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.2.3 Estimate on Γz,L . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.3 Proof of Theorem 1.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
∗ The research of S.B. was supported by NSF grants DMS-1007244 and DMS-1612363.
† The research of J.H. was supported by an AMS-Simons travel grant and NSF grant DMS-1612921.

1
 3.3.1 Proof of Theorem 1.1 when d = 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.3.2 Proof of Theorem 1.1 when d = 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.3.3 Proof of Theorem 1.1 when d = 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

4 Low-dimensional radius bounds 27

5 The last k waves in 2D 28

6 High-dimensional radius bounds 37

6.1 Radius bounds on transitive unimodular graphs . . . . . . . . . . . . . . . . . . . . . . . . 37
6.2 Radius lower bound when d ≥ 5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
6.3 Radius upper bound when d ≥ 5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

7 Bounds on the size 40

7.1 Upper bounds on the size when d = 3, 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
7.2 Lower bounds on the size when 2 ≤ d ≤ 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
7.3 Proof of the moment bounds in d = 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
7.3.1 Proof of the first moment bounds . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
7.3.2 Proof of the second moment bounds . . . . . . . . . . . . . . . . . . . . . . . . . . 46
7.4 The size of the past in invariant forests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
7.5 Lower bounds: d ≥ 5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
7.6 Upper bounds: d ≥ 5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

1 Introduction
The Abelian sandpile model is a particle system defined in terms of simple local redistribution events,
called topplings, which give rise to non-local dynamical events called avalanches. The model has received
a lot of attention in the theoretical physics literature (see [9, 3]) due to its remarkable self-organized
critical state, conjectured to be characterized by power-law behavior of various quantities related to
avalanches. Starting with the seminal work of Dhar, much mathematical progress has been made toward
understanding this self-organized critical state. The surveys [33] and [14] collect some of this. However,
establishing power law behavior for many fundamental avalanche characteristics on Zd appears difficult
in general. The purpose of this paper is to establish new rigorous inequalities, which in high dimensions
come close to identifying the correct tail behavior, for these quantities.
Given a finite set V ⊂ Zd , a sandpile on V is a collection of indistinguishable particles, given by a
map η : V → {0, 1, . . . }. We say that η is stable, if η(x) < 2d for all x ∈ V . If η is unstable at x, that
is η(x) ≥ 2d, we say that x is allowed to topple. On toppling, x sends one particle along each edge
incident with it, resulting in the new sandpile


η(y) − 2d if y = x;
η ′ (y) = η(y) + 1 if y ∼ x, y ∈ V ; (1.1)


η(y) otherwise.

Particles sent to vertices in Zd \ V are lost. It is well-known [8] that given any sandpile η, carrying out
all possible topplings in any sequence, results in a uniquely defined stable sandpile η ◦ . The sandpile
Markov chain on V is the Markov chain with state space equal to the set of stable sandpiles on V , where

2
at each time step a particle is added at a uniformly chosen vertex of V , and the sandpile is stabilized, if
necessary. The unique stationary distribution [8] is denoted νV .
We will be interested in sandpiles on Zd , where “stable” and “toppling” are defined the same way as
for finite V . Athreya and Járai [2] proved that if V (L) = [−L, L]d ∩ Zd , d ≥ 2, then νL := νV (L) converges
weakly, as L → ∞, to a limit measure ν, called the sandpile measure. Let η : Zd → {0, . . . , 2d − 1} be a
sample configuration from the measure ν. Let us add a particle to η at the origin o, and let Av = Av(η)
denote the set of vertices that topple, called the avalanche cluster. The set of all topplings, with
multiplicity, is called the avalanche. In this paper we study various characteristics of avalanches.
The concept of waves, introduced by Ivashkevich, Ktitarev and Priezzhev [13] in the context of finite
graphs, will play an important role. Waves provide a decomposition of an avalanche into smaller sets
of topplings: WL,1 , . . . , WL,N ⊂ V (L); see Section 2 for precise definitions. In each wave, every vertex
topples at most once, and the union of the waves includes all topplings of the avalanche with the correct
multiplicity. The paper [13] analyzed the last wave WN,L in particular when d = 2.
Our first set of results concern the probability that a given vertex topples. Based on an analysis of
the last wave, we prove the following rigorous lower bounds on the toppling probability.

Theorem 1.1.
(i) Let d = 2. Then
ν(z ∈ Av) ≥ |z|−3/4+o(1) , as |z| → ∞.
(ii) Let d = 3. There exist constants ζ < 1/2 and c > 0 such that

ν(z ∈ Av) ≥ c |z|−2ζ−1 , ∀z ∈ Z3 .

(iii) Let d = 4. There exists a constant c > 0 such that

−1/3
ν(z ∈ Av) ≥ c |z|−2 (log |z|) , ∀z ∈ Z4 .

In Theorem 1.1 (ii), ζ can be taken to be any value such that a random walk in Z3 of length n does
not hit the loop-erasure of an independent random walk of length n with probability ≥ c′ n−ζ for some
c′ > 0. We prefer to write the bound (ii) in this form to emphasize the dependence on this exponent,
whose value is of interest in the theory of loop-erased walks. The exponent ζ is known to satisfy the
bound ζ < 1/2; see [24, Sections 10.3 and 11.5].
The rigorous upper bound ν(z ∈ Av) ≤ C|z|2−d , for some C = C(d), follows from Dhar’s formula
(see [15, Eqn. (3.5)]). In dimensions d ≥ 5, Járai, Redig and Saada [16, Section 6.2] proved that ν(z ∈
Av) ≥ c|z|2−d , for some c = c(d), also based on an analysis of the last wave. [16] introduced the critical
exponent θ to quantify the departure from Dhar’s formula, assuming that ν(z ∈ Av) ≈ |z|2−d−θ , as
|z| → ∞. This means that θ = 0 when d ≥ 5. Our Theorem 1.1 shows that if θ exists in the sense that
log(ν(z ∈ Av))/ log(|z|) exists as |z| → ∞ (“logarithmic equivalence”), then


≤ 3/4 when d = 2,
0≤θ <1 when d = 3,


=0 when d = 4.

In particular, Theorem 1.1(iii) establishes that θ = 0 when d = 4, with at most a logarithmic correction.
The reason behind the fact that θ = 0 for d ≥ 5 is that in these dimensions loop-erased walk and
independent simple random walk do not intersect with positive probability. The difference in behavior

3
when d ≥ 5 also shows up in our other results, and d = 4 is expected to be the upper critical dimension of
the model, in the sense that critical exponents are no longer expected to depend on dimension when d ≥ 5
[32]. We expect that θ is positive in dimensions two and three, in analogy with other statistical physics
models below the upper critical dimension. However, it seems difficult to get rigorous upper bounds
improving on Dhar’s formula, since any such bound would have to control all waves of the avalanche.
For the last wave, we have a precise characterization in terms of loop-erased walk; however, we lack a
convenient description of the joint distribution of all waves of the avalanche. For similar reasons, we do
not expect the bounds coming from last waves in low dimensions to be tight.
Our next set of results concern the radius of the toppled region. Let R = R(η) = sup{|z| : z ∈
Av(η)} be the radius of the avalanche. As we explain below, some of the following inequalities are easy
consequences of Theorem 1.1, while some others follow from known results on uniform spanning forests
of Zd .
Theorem 1.2.
(i) Let d = 2. Then,

r−3/4+o(1) ≤ ν(R ≥ r), as r → ∞.

(ii) Let d = 3. There are constants c > 0 and C such that with ζ as in Theorem 1.1 we have

cr−(2ζ+1) ≤ ν(R ≥ r) ≤ Cr−1/6 , ∀r ≥ 1.

(iii) Let d = 4. Then there exist constants c > 0 and C such that
−1/3
cr−2 (log r) ≤ ν(R ≥ r) ≤ Cr−1/4 , ∀r ≥ 1.

(iv) Let d ≥ 5. There is a constant c = c(d) > 0 such that

cr−2 ≤ ν(R ≥ r) ≤ r−2 (log r)3+o(1) , ∀r ≥ 1.

The lower bounds of Theorem 1.2 in dimensions 2, 3, and 4 follow from taking z = re1 in Theorem
1.1, where e1 = (1, 0, . . . , 0) ∈ Zd . In dimensions d ≥ 3, upper bounds can be derived from results of
Lyons, Morris and Schramm [25]. They analyzed, using the “conductance martingale” of Morris [30], the
wired uniform spanning forest measure WSF on transient graphs, including Zd for d ≥ 3, as well as a
related measure WSFo , obtained by “wiring o to infinity”. See the book [26] for detailed background on
wired spanning forests. Let To denote the component of o under WSFo . The proof of [25, Theorem 4.1]
shows that for d ≥ 3,
1 1
WSFo (diam(To ) > r) ≤ C(d)r− 2 + d . (1.2)
The measure WSFo can be related to waves in sandpiles; in particular, this was used by Járai and Redig
[15] to show that when d ≥ 3, avalanches are finite ν-a.s. We derive the upper bounds in Theorem
1.2(ii)–(iii) from (1.2).
Above the critical dimension, d ≥ 5, Priezzhev [32] gave heuristic arguments for the mean-field
behaviour ν(R ≥ r) ≈ r−2 . Both the lower bound ν(R ≥ r) ≥ ν(re1 ∈ Av) and the upper bound of (1.2)
can be sharpened to establish this rigorously, in the sense of logarithmic equivalence. On the other hand,
Theorem 1.2(ii)–(iii) establishes that, if a critical exponent α satisfying ν(R > r) ≈ r−α governs the tail
of R in low dimensions, then this α is different from the mean-field value 2.
We deduce the lower and upper bounds in Theorem 1.2(iv) from very general mass transport argu-
ments, stated in Theorem 1.3 below; see [26, Chapter 8] for background on mass transport. While the

4
main focus of this paper is sandpile models on Zd , we believe this result may be useful on other graphs and
for other models. The proof is in Section 6.1, and is independent of the rest of the paper. Let G = (V, E)
be a graph and let Γ ⊂ Aut(G) be a transitive subgroup of the group of automorphisms of G, under the
topology of pointwise convergence. It is well known that every closed subgroup of Aut(G) has a Borel
measure which is invariant under the left multiplication by γ ∈ Γ. The group Γ is called unimodular if
this measure is also invariant under right multiplication. In addition, we call the graph G unimodular
if Aut(G) has some unimodular transitive closed subgroup. In this setting the mass transport principle
states that for o ∈ V (G) andPa non-negative P function f : V × V → [0, ∞], which is invariant under the di-
agonal action of Γ, we have x∈V f (o, x) = x∈V f (x, o). Let d be a Γ-invariant metric on V , and write
diam(A; x) = sup{d(v, x) : v ∈ A}, and let diam(A) = diam(A; o). Write Dx (r) = {y ∈ V : d(y, x) ≤ r}.
We say that an infinite tree T has one end, if any two infinite self-avoiding paths in T have a finite
symmetric difference. Given x ∈ T , we denote by pastx the set of vertices y ∈ T such that the unique
infinite self-avoiding path in T starting at y contains x. By a percolation on (V, E), we mean a proba-
bility measure on subgraphs of (V, E). Given a percolation, we write Cx for the connected component of
x. When the percolation is supported on spanning forests, we write Tx for Cx .
Theorem 1.3. Let (V, E) be a graph with a transitive unimodular group of automorphisms Γ, and let
o ∈ V be a fixed vertex. Let µ be a Γ-invariant percolation on (V, E).
(i) If µ is supported on spanning forests with one-ended components, then

X µ(o ∈ pastx )2
µ diam(pasto ) > r ≥ h i.
x∈V :r<d(x,o)≤2r Eµ To ∩ Do (4r) 1o∈pastx

(ii) We have
X  

1diam(Cx ;x)>4r
µ diam(Co ) > 4r = µ(o ∈ Cx ) Eµ o ∈ Cx .
|Cx ∩ Dx (4r) \ Dx (r)|
x∈V :r<d(x,o)≤4r

(iii) Suppose that WSFo (|To | < ∞) = 1. Then

WSFo diam(To ) > 4r
X  
1diam(Tx ;x)>4r
= WSFx (o ∈ Tx ) EWSFx o ∈ Tx .
|Tx ∩ Dx (4r) \ Dx (r)|
x∈V :r<d(x,o)≤4r

Regarding the upper bound in Theorem 1.2 (iv), Lyons, Morris and Schramm state the result

WSF(diam(pasto ) > r) ≤ r−2 (log r)O(1) ; (1.3)

see [25, page 1710]. However, since a proof of (1.3) is not included in [25], and we need a sharpening of
(1.2) for our results, we deduce a diameter estimate for To under WSFo from Theorem 1.3(iii). (This
implies (1.3) due to a stochastic comparison; see [25, Lemma 3.2]). In order to deal with the fact that
WSFo is not translation invariant, we restrict attention to To , which is unimodular; see Section 6.1.
We do not have an upper bound on ν(R ≥ r) in d = 2, and it is an open problem whether ν(R < ∞) =
1. It follows from Theorem 1.2(i) that Eν R = ∞, when d = 2. It may be of independent interest that
a short proof of the weaker statement, that EνL R diverges, can be given without reference to spanning
trees or the burning bijection described in Section 2.3. We state this as a separate result.

5
Proposition 1.4. If d = 2, then limL→∞ EνL R = ∞.
Our last set of results concern the number of topplings in the avalanche. Let S denote the total number
of topplings in the avalanche (that is, elements of Av are counted with multiplicity). Recall that WN,L
denotes the last wave in the finite graph VL . Based on the fractal dimension of loop-erased walk and scaling
assumptions, Ivashkevich, Ktitarev and Priezzhev [13] derived the exponent νL (|WN,L | ≥ t) ≈ t−3/8 , in
the limit L → ∞. We prove a rigorous lower bound with the same exponent, which we also extend
to higher dimensions. Above the critical dimension, d ≥ 5, we also have an upper bound on the total
number of topplings with an exponent which is independent of d. The upper bounds of Theorem 1.2 on
the radius in d = 3, 4 provide upper bounds on the size of the avalanche cluster.
Since the size of the avalanche cluster could be measured in two different ways – namely, via |Av| and
via S – there are in principle two different possible critical exponents τS ′ and τS given (if they exist) by
ν(|Av| > t) ≈ t−τS′ , ν(S > t) ≈ t−τS . As in the preceding cases, our theorems give corresponding bounds
on the possible values of τS , τS ′ . These bounds, as well as the best current bounds on the exponents θ
and α described above, are summarized in Table 1.
Theorem 1.5.
(i) Let d = 2. Then
t−3/8+o(1) ≤ ν(|Av| ≥ t), as t → ∞.
(ii) Let d = 3. With ζ as in Theorem 1.1, and for some constants C and c > 0, we have

ct−(2ζ+1)/3 ≤ ν(|Av| ≥ t) ≤ Ct−1/18 , ∀t ≥ 1.

Moreover, ν(S ≥ t) ≤ Ct−1/19 , ∀t ≥ 1.

(iii) Let d = 4. There exists C and c > 0 such that

ct−1/2 (log t)−5/6 ≤ ν(|Av| ≥ t) ≤ Ct−1/16 , ∀t ≥ 1.

Moreover, ν(S ≥ t) ≤ Ct−1/17 , ∀t ≥ 1.

(iv) Let d ≥ 5. There exist c = c(d) > 0 such that

ct−1/2 ≤ ν(|Av| ≥ t) ≤ ν(S ≥ t) ≤ t−2/5+o(1) , ∀t ≥ 1.

We establish the lower bounds in dimensions d = 2, 3 by showing that once a vertex at distance t1/d
from o is in the last wave, at least ct other vertices in its neighbourhood will also be in the last wave. In
d = 4, ct is replaced by ct/ log t. Given this, parts (i)–(iii) of Theorem 1.5 can be deduced from parts
(i)–(iii) of Theorem 1.1. For the lower bound in d ≥ 5, in Theorem 1.5(iv), we use the following analogue
of Theorem 1.3(i). Write

e o (r) = {x ∈ To (r) : the path from o to x in To stays inside Do (r)}.

To (r) = To ∩ Do (r) and T

Theorem 1.6. Let (V, E) be a graph with a transitive unimodular group of automorphisms Γ. Let µ be
a Γ-invariant percolation on (V, E) supported on spanning forests with one-ended components. For all
t, r ≥ 1 we have
 2

X µ o ∈ past , | e
T o (r/2)| > t
x
µ |pasto | > t ≥ h i .
x:r<d(x,o)≤(3/2)r E |To (4r)| 1{o∈pastx }

6
 toppling radius avalanche avalanche
probability cluster size size
ν(x ∈ Av) ≈ |x|2−d−θ ν(R > r) ≈ r −α ν(|Av| > t) ≈ t−τS′ ν(S > t) ≈ t−τS

d=2 [0, 3/4] [0, 3/4] [0, 3/8] [0, 3/8]

d=3 [0, 1) [1/6, 2) [1/18, 2/3) [1/19, 2/3)

d=4 =0 [1/4, 2] [1/16, 1/2] [1/17, 1/2]

d≥5 =0 =2 [2/5, 1/2] [2/5, 1/2]

Table 1: The best known bounds on the critical exponents introduced in this introduction. Bounds are
expressed in interval form: e.g., 1/6 ≤ α < 2 when d = 3.

The proof of Theorem 1.6 is in Section 7.4 and does not rely on the rest of the paper.
We believe, as it has been argued by Priezzhev [32], that the exponent 1/2 is sharp in Theorem
1.5(iv). But it seems challenging to establish a matching upper bound for S or Av. The main difficulty
lies in extracting useful information on the dependence between the waves from the bijection with WSFo .
Instead of such an approach, we control the number of waves using our upper bound on the radius in
Theorem 1.2(iv), which allows us to use a union bound instead of estimating the dependence. This leads
to the upper bounds in Theorem 1.5(ii)–(iv).
All our results have analogues in large finite V (L), or indeed are derived therefrom. Passing to the
limit of Zd is not too difficult when d ≥ 3, due to the result of Járai and Redig [15, Theorem 3.11] showing
that ν(|S| < ∞) = 1. When d = 2, this is not known. We bypass this problem with a more technical
argument, that we believe is of interest in its own right. We show that for any 1 ≤ k < ∞, the last
k-waves (when they exist) have a finite limit as V (L) ↑ Zd . Recall that for η ∈ RL the waves occurring
during the stabilization of η + 1o are denoted WL,1 , . . . , WL,N . Waves can also be defined on Zd , denoted
W1 , W2 , . . . ; see Section 2.2. On Zd , the number of waves N may take the value infinity.
Let ηN −k+1 denote a random configuration on V (L) with law νL (· | N ≥ k). Given a configuration η
on V (L) such that η(o) = 2d, let W(η) denote the set of sites that can be toppled with every site toppling
at most once. We extend this also to configurations ξ on Zd such that ξ(o) ≥ 2d.
Theorem 1.7. Assume d = 2.
(i) For all k ≥ 1, the law of ηN −k+1 converges weakly to the law of a random configuration ξk in Z2 . Let
ρk denote the law of ξk . The law of WN −k+1 = W(ηN −k+1 ) converges to the law of Wk∗ := W(ξk ), that
is a.s. finite under ρk .
(ii) For all k ≥ 0 we have limL→∞ νL (N = k) = ν(N = k).
(iii) For every k ≥ 1, the joint law of WL,1 , . . . , WL,k under νL (· | N = k), converges weakly, as L → ∞,
to the law of W1 , . . . , Wk under ν(· | N = k), and under this conditioning we have ν-a.s. |Wℓ | < ∞,
ℓ = 1, . . . , k.
See Section 5 for more detailed statements. Our argument in fact gives a power law upper bound on
the radii of the last k waves, and leads to the following extension of Theorem 1.2(i). Let Rk = sup{|x| :

7
x ∈ Wk∗ }.
Theorem 1.8. Assume d = 2.
(i) There are constants α1 > α2 > · · · > 0 and C1 , C2 , . . . such that

ρk (Rk > r) ≤ Ck r−αk , ∀r ≥ 1, ∀k ≥ 1.

(ii) We have
ν(R ≥ r, N ≤ k) ≤ Ck r−αk , ∀r ≥ 1.
We will also use Theorem 1.7 to prove the following theorem.
Theorem 1.9. Suppose d = 2. Then Eν N = ∞.
This is a strengthening of the statement Eν R = ∞, due to a simple comparison proved in Lemma 2.3,
and therefore also of Proposition 1.4.
Organization of the paper. In Section 2, we give definitions and background on sandpiles, spanning
trees, and random walks; we also prove Proposition 1.4. In Section 2.2, we prove Theorem 1.1 modulo a
technical argument required for the two-dimensional case, which we defer to Section 5.
Sections 4, 5, and 6 are devoted to the various radius bounds above. In Section 4, we prove Theorem 1.2
(i) – (iii). Section 5 contains additional arguments for the two-dimensional case; here we prove Theorems
1.7, 1.8, and 1.9. In Section 6, we complete the proof of Theorem 1.2 by proving the high-dimensional
bounds (Theorem 1.2 (iv) and Theorem 1.3 ).
Section 7 contains the proofs of the size bounds above: Theorems 1.5 and 1.6.
A note on constants. All our constants will be positive and finite, and they may depend on the
dimension d. Other dependence will always be indicated. Constants denoted C and c may change from
line to line; those with index (such as c1 ) stay the same within the same proof.

2 Definitions and background

In this section, we provide definitions and collect useful facts about the basic objects we use: toppling
numbers, waves, spanning trees, bijections and Wilson’s algorithm.

2.1 Graphs and sandpiles

We will work with finite connected graphs of the form H = (U ∪{s}, F ), where s is a distinguished vertex,
called the sink. We allow multiple edges, so in general H is a multigraph, but we exclude loop-edges. If
U = V is a finite subset of Zd , we let GV = (V ∪ {s}, E) denote the wired subgraph induced by V –i.e.,
where all vertices in Zd \ V are identified to the single vertex s, and loops at s are removed. Of prime
importance will be the standard exhaustion V (L) = [−L, L]d ∩ Zd , L ≥ 1; the wired subgraph induced
by V (L) will be denoted GL . In general, a subscript L will be shorthand for subscript V (L). We write
x ∼ y to denote that vertices x and y of a graph (understood from context) are neighbours. Given a
graph H = (U ∪ {s}, F ), we write degH (x) for the degree of the vertex x ∈ U ∪ {s} in H. When V ⊂ Zd ,
we write degV (x) for the degree of vertex x ∈ V in the subgraph of Zd induced by V .

8
 Given H = (U ∪ {s}, F ), the discrete Dirichlet Laplacian ∆H is given by
(
degH (x) if x = y;
∆H (x, y) = x, y ∈ U, (2.1)
−axy if x 6= y;

where axy equals the number of edges connecting x and y. In particular, when V is a finite subset of Zd ,
we have ∆V given by 

2d if x = y;
∆V (x, y) = −1 if x ∼ y; x, y ∈ V. (2.2)


0 otherwise,
We denote the inverse matrix by gH = (∆H )−1 , gV = (∆V )−1 .
We write ∆ for the matrix defined as in (2.2), but with x, y ∈ Zd , and g = ∆−1 when d ≥ 3. Up to
a factor (2d)−1 , these matrices are the Green function of simple random walk. Namely, let (S(n))n≥0
denote a simple random walk in Zd , and let

σA = inf{n ≥ 0 : S(n) 6∈ A}.

Then 2d gV (·, ·) = GV (·, ·) and 2d g(·, ·) = G(·, ·), where we define

 
X
GV (x, y) := Ex  1S(n)=y  , x, y ∈ V ;
0≤n<σV
 
X
G(x, y) := Ex  1S(n)=y  , x, y ∈ Zd .
0≤n<∞

Whenever G appears with arguments, it refers to a Green function as defined above; when it appears
without arguments, it refers to graphs as in the notation introduced previously.
Let us fix a finite connected graph H = (U ∪ {s}, F ). A sandpile on H is a function η : U →
{0, 1, 2, . . . }. We say that η is unstable at x ∈ U , if η(x) ≥ degH (x). In this case x is allowed to
topple, which means that x sends one particle along each edge incident with it. Particles arriving at
s are lost. Toppling x has the effect of subtracting row ∆H (x, ·) from η(·). It is a basic property of
the model that if unstable vertices are toppled in any order until there are no such vertices, the stable
sandpile obtained is independent of the order chosen (called the Abelian property) [8]. Hence for any
sandpile η there is a well-defined stabilization
Q of η, denoted η ◦ . The sandpile Markov chain is defined
as follows. The state space is ΩH = x∈U {0, . . . , degH (x) − 1}. Given that the current state is η, a
single step is defined by choosing a vertex X ∈ U uniformly at random, and moving to state (η + 1X )◦ .
The maps ax : ΩH → ΩH defined by ax : η 7→ (η + 1x )◦ , x ∈ U , are called the addition operators.
It follows from the uniqueness of stabilization that ax ay = ay ax , x, y ∈ U . When it is necessary to
emphasize the graph H on which the operator ax is applied, we write ax,H for ax ; we also use ax,L for
ax,GL . We denote the set of recurrent states of the sandpile Markov chain by RH . It is known that the
unique stationary distribution is given by the uniform distribution on RH [8], and that each (restricted)
map ax : RH → RH preserves this measure.
In the special case when the graph arises from a finite V ⊂ Zd , we denote the state space by ΩV =
{0, . . . , 2d − 1}V , and the set of recurrent states by RV .

9
 Now, let η : Zd → {0, 1, 2, . . . } be a sandpile on Zd . If x is unstable in η, we define the toppling of
x using the matrix ∆, that is, η 7→ η(·) − ∆(x, ·). Let us call a finite or infinite sequence consisting of
topplings of unstable vertices exhaustive, if any vertex that is unstable at some point, is toppled at a
later time. It can be shown, similarly to the finite graph case, that for all x ∈ Zd , x topples the same
number of times (possibly infinity) in any exhaustive sequence.
We write ei for the unit vector in the i-th positive coordinate direction, | · | for the Euclidean norm,
and k · k for the ℓ∞ norm on Zd . We denote by o the origin in Zd , and we let
Vx (n) = {y ∈ Zd : ky − xk ≤ n}; Bx (n) = {y ∈ Zd : |y − x| ≤ n} ,
and write V (n) = Vo (n) and B(n) = Bo (n).
Given A, B ⊂ Zd , we let dist(A, B) denote their Euclidean distance, and write dist(x, B) when
A = {x}. If A ⊂ Zd , we let ∂A = {x ∈ Zd \ A : dist(x, A) = 1}. When considering A as a subset of GL ,
we will often use ∂A to denote the boundary restricted to GL — that is, excluding from ∂A any x ∈ / GL
— the meaning will be clear in context.
If z1 , z2 are two elements of Rd , let ang(z1 , z2 ) denote the angle between z1 and z2 . We will make use
of the “little o” notation: an = na+o(1) if limn [ log an / log n] = a.

2.2 Toppling numbers and waves

Given a sandpile η on a finite connected graph H = (U ∪ {s}, F ), we write n(x, y) = n(x, y; η) for
the toppling numbers, i.e., number of times y topples when stabilizing η + 1x . A useful ordering of
topplings, introduced by Ivashkevich, Ktitarev and Priezzhev [13], for stabilizing η + 1x is in terms of
waves, which we now define. If η + 1x is stable, there are no waves. Otherwise, topple x and carry
out any further topplings that are possible without toppling x a second time. It is easy to verify that
in doing so, every vertex in U topples at most once. We write W1,H = W1,H (x; η) for the set of sites
toppled so far; this is the first wave. If x is still unstable after the first wave, which happens if and
only if all neighbours of o are in W1,H , topple x a second time, and carry out any further topplings
that are possible without toppling x a third time. The set of vertices that topple in doing so is denoted
W2,H = W2,H (x; η); this is the second wave. We define further waves analogously.
When the graph arises from a finite V ⊂ Zd , we write W1,V , W2,V , . . . for the waves. We make similar
definitions for sandpiles on Zd . On Zd , the toppling numbers n(x, y; η) are possibly infinite. Waves
W1 , W2 , . . . are defined analogously to the finite case, and their number may be infinite. The following
lemma is straightforward to verify.
Lemma 2.1.
(i) Let H = (U ∪ {s}, F ) be a connected finite graph. The number of waves containing y equals n(x, y; η).
In particular, the number of waves is N = N (x; η) := n(x, x; η).
(ii) In the case of Zd , the number of waves containing y equals n(x, y; η) (possibly infinite). In particular,
the number of waves is N = N (x; η) := n(x, x; η).
The next lemma gives the expected number of topplings in an avalanche.
Lemma 2.2 (Dhar’s formula [8]). Let H = (U ∪ {s}, F ) be finite. We have
EνH n(x, y) = gH (x, y), x, y ∈ U.
For cubes in Zd , we have the following deterministic comparison between n(o, o; η) and the radius
R(η).

10
Lemma 2.3.
(i) Let η be any sandpile configuration on Zd , d ≥ 1. Then
n(o, o; η) ≤ R(η) .
(ii) The same statement holds when η is a sandpile configuration in GL for any L ≥ 1.
Proof. (i) In the proof below, it will be convenient to denote n[η] := n(o, o; η). We will also use the notion
of “restricted topplings”. For L > 0, let η|L denote the restriction of η to V (L); the restricted toppling
number nL [η] will denote the number of topplings occurring at o in the stabilization of η|L + 1o on GL .
Note that such a stabilization amounts to taking the configuration η + 1o on Zd and toppling only sites
x ∈ V (L) until all such x are stabilized, neglecting any sites in Zd \ V (L) that may become unstable. In
particular, for any stable sandpile configuration η on Zd , n[η] ≥ nL [η] for any L.
Assume η and R < ∞ are as in the statement of the lemma. We claim that the equality n[η] = nR [η]
holds. Indeed, by the above observation, nR is exactly the number of topplings at o required to stabilize
only the sites of V (R), but by assumption these are the only sites which need to be toppled to stabilize
η + 1o in all of Zd . Therefore, it suffices to show that nR [η] ≤ R.
For this, we will use a special “maximal” configuration φR from [11, Lemma 4.2]:
(
2d − 1, x ∈ V (R)
φR (x) =
2d − 2 otherwise.

Note that nR [η] ≤ nR [φR ], since η|R ≤ φR pointwise. Moreover, nR [φR ] ≤ n[φR ]. It is proven in [11,
Lemma 4.2], and not difficult to see by computing each wave, that n[φR ] = R. This completes the proof
of (i).
(ii) The above proof applies here as well, with only minor changes.
Proof of Proposition 1.4. By Dhar’s formula [8], we have
lim EνL n(o, o) = lim gVL (o, o) = ∞, when d = 2.
L→∞ L→∞

By Lemma 2.3(ii), this implies limL→∞ EνL R = ∞.

Remark 1. We do not see a simple way to deduce the statement Eν n(o, o) = ∞ from the simpler statement
that limL→∞ EνL n(o, o) = ∞, when d = 2. This requires ruling out the possibility that EνL n(o, o) is
dominated by rare events with many waves. Our proof of Eν n(o, o) = ∞ in Theorem 1.9 will build on
quite a few other results; in particular, results from Section 5.

2.3 Spanning trees and the burning bijection

In 1990, Dhar [8] introduced a method for checking whether a particular stable configuration η lies in RH ,
called the “burning algorithm”. Application of the burning algorithm provides a bijection ϕ between RH
and the set of all spanning trees of H, that we denote by TH ; see [28]. We briefly describe this bijection
here. In Section 2.5, we give a version of it for “waves” which will be necessary in our analysis.
Recall that axy denotes the number of edges between vertices x and y.
Lemma 2.4 (Burning algorithm [8], [12, Lemma 4.1]). Let η be a stable sandpile on a connected finite
graph H = (U ∪ {s}, F ). At each x ∈ U , add axs grains of sand, and stabilize. We have η ∈ RH if and
only if each vertex in U topples exactly once.

11
 Note that instead of adding sand as in the lemma, we may initiate the toppling process by placing
P
all x∈U axs = degH (s) grains at s, and toppling s first. Suppose we carry out any possible topplings
in parallel. We say that x burns at time k, if it is toppled in the k-th parallel toppling step, where we
regard s to have burnt at time 0.
The bijection is defined as follows. For each y ∈ U , fix an arbitrary ordering ≺y of the edges adjacent
to y. Given η ∈ RH , for each y ∈ U , we adjoin to the tree T an edge connecting y to a neighbour burnt
one time step before, chosen as follows. If Py is the number of edges joining y to neighbours burnt before
y, and Ay is the subset of such edges leading to sites burnt one step before y, then the burning rule
implies
η(y) = degH (y) − Py + i for some 0 ≤ i < |Ay | .
We add to T the i-th edge in Ay in the ordering ≺y .
The resulting graph T will be a spanning tree (the fact that it spans — i.e., that every site topples
in this procedure — is part of the content of Lemma 2.4), and we set ϕ(η) = T . The map ϕ : RH → TH
is usually referred to as the “burning bijection”, and the toppling procedure used to construct ϕ will be
referred to as the “burning procedure”.

2.4 Intermediate configurations

Now, we give a description of waves in terms of recurrent configurations on an auxiliary graph, as
introduced in [13]. As in the previous section, we describe this in an arbitrary finite connected graph
H = (U ∪ {s}, F ). Suppose we are interested in waves started by the addition of a particle at a fixed
vertex w ∈ U . Consider the graph H ′ obtained from H by adding the edge f ′ := {w, s}. For readability,
′ ′
we denote ∆H := ∆H ′ and RH := RH ′ . Let ax′ = ax,H
′
, x ∈ U denote the addition operators on H ′ .
◦
We reserve the notation η for stabilization on the original graph H. When we need to emphasize that
′
addition is applied on the graph derived from H, we prefer the ax,H notation.
′
The burning algorithm (Lemma 2.4) implies that RH ⊂ RH . The following lemma compares the sizes
of these two sets.
′
Lemma 2.5. For any finite connected graph H = (U ∪ {s}, F ), and w ∈ U , we have |RH | = (1 +
gH (w, w))|RH |.
Proof. Let 1w,w denote the U × U matrix whose only non-zero entry is a 1 at (w, w). We have
′ ′
|RH | = det(∆H ) = det(∆H + 1w,w ) = (1 + (∆H )−1 (w, w)) det(∆H )
(2.3)
= (1 + gH (w, w)) |RH |.

The following immediate corollary will be useful in controlling avalanches. A version of part (i) was
proved in [15, Lemma 7.5].
Corollary 2.6. Consider the sequence (VL )L .
(i) Suppose d ≥ 3. There exists a constant C(d) such that |RL′ | ≤ C(d) |RL | for all L ≥ 1.
(ii) Suppose d = 2. There exists a constant C such that |RL′ | ≤ C log L |RL | for all L ≥ 2.
Proof. Both statements follow from Lemma 2.5, the equality gL (o, o) = (2d)−1 GL (o, o), and known
properties of the Green function GL (o, o); see e.g. [24].

12
 ′
We now describe the interpretation of waves as elements of RH \ RH ; introduced in [13]. Let η ∈ RH ,
and suppose that η + 1w is unstable at w in H, i.e. η(w) = degH (w) − 1. Consider the waves occurring
in stabilizing η + 1w in H. Recall that N = N (η) denotes the number of waves. For 1 ≤ k ≤ N, let
ηk = aw′ ηk−1 , where η0 = η. It is straightforward to check that ηk is the configuration seen just before
the k-th wave is carried out, and aw η = (a′w )N +1 η. Note that the latter statement also holds, trivially,
when η + 1w is stable, in which case N = 0.
Definition 1. Let η ∈ RH be such that η +1w is unstable at w. We call the sequence α(η) := (η1 , . . . , ηN )
the intermediate configurations corresponding to η.
We record here the characterization of R′H \RH ; a similar statement was shown in [16] for a continuous
height model.
Lemma 2.7. The collection {α(η) : η ∈ RH , η(w) = degH (w) − 1} forms a partition of R′H \ RH .
Proof. Since ηk (w) = degH (w), k = 1, . . . , N , we have ηk ∈ R′H \ RH , k = 1, . . . , N . This and the
relation aw η = (a′w )N +1 η imply that α(η) has distinct entries (the order of a′w is at least N + 1). By
Dhar’s formula, the average number of waves per recurrent configuration is gH (w, w), so gH (w, w)|RH |
elements of R′H correspond to intermediate configurations.
It is similarly easy to check that η ∈ R′H , accounting for another |RH | elements of R′H . Comparing
this with Lemma 2.5, we see that every element of R′H is either an intermediate configuration or recurrent
on H, completing the proof.
Given η∗ ∈ R′H \ RH , we denote by W(η∗ ) the set of vertices that topple in the stabilization a′w,H (η∗ ).
It is immediate from this definition that if α(η) = (η1 , . . . , ηN ), then W(ηk ) is the k-th wave corresponding
to η, i.e. Wk,H (w; η). Of particular interest will be the last wave W(ηN ). The following corollary follows
directly from the definitions.
Corollary 2.8. An intermediate configuration η∗ ∈ R′H \ RH is a last wave if and only if there exists
y ∼ w, y ∈ U ∪ {s}, such that y 6∈ W(η∗ ).
We will also need the following lemma.
Lemma 2.9. We have
1
|RH | ≤ |{η∗ ∈ R′H : η∗ is a last wave}| ≤ |RH |.
degH (w)
Proof. The upper bound is obvious. To see the lower bound, we assign to η ∈ RH the last intermediate
configuration in the stabilization of η + (degH (w) − η(w))1w . This map is at most degH (w) to 1, proving
the lower bound.

2.5 Bijection for intermediate configurations

We now specialize to the set-up where H = GV , V ⊂ Zd finite, o ∈ V . In this section, we describe a
version of the burning bijection on G′V , that will allow us to control topplings occurring in a wave. To the
best of our knowledge, such a bijection was first introduced by Ivashkevich, Ktitarev and Priezzhev [13].
See also [32, 15, 16], where it played a key role. For many of our results, the variant in [13] would suffice.
However, a more careful choice of the burning process will be needed in Section 5, so we introduce here
the version we need. Our burning process is similar to burning processes introduced in [17] and [10].

13
 Let η∗ ∈ R′V \ RV . We define a pair of vertex-disjoint trees (To , Ts ) = ϕ′ (η∗ ), such that To ∪ Ts spans
GV . Send one grain of sand from s to o, resulting in 2d grains at o. We sequentially topple vertices in
the balls B(0) ∩ V, B(1) ∩ V, B(2) ∩ V, . . . , and build a tree rooted at o, similarly to the usual burning
bijection. The precise definitions of burnt and unburnt sets are as follows. We let
(0) (0)
Bt0 = {o} Ut0 = V ∪ {s} \ {o}
(0) (0)
Btk = ∅, k ≥ 1, Utk = V ∪ {s} \ {o}, k ≥ 1.

For r ≥ 1, inductively, we set

(r) (r−1) (r) (r)
Bt0 = ∪ℓ≥0 Btℓ Ut0 = V ∪ {s} \ Bt0
n o
(r) (r) (r) (r) (r)
Btk = x ∈ B(r) ∩ Utk−1 : η∗ (x) ≥ degUt(r) (x) Utk = Utk−1 \ Btk , k ≥ 1.
k−1

(r)
For each r ≥ 1 there exists a smallest index J = J(r) ≥ 1 such that BtJ = ∅, and there is a smallest
(R+1)
index R ≥ 1 such that J(R + 1) = 1. Then Bt0 = W(η∗ ) is the set of vertices toppled in the wave
represented by η∗ .
We complete the burning process by sending asx grains of sand from s to x for each x ∈ V , and follow
the usual burning rule. That is, we set:

f 0 = ∪r≥0 ∪ℓ≥0 Bt(r)

Bt f 0 = V \ Bt
Ut f0
ℓ
n o
f f
Btk = x ∈ Utk−1 : η∗ (x) ≥ degUt
f k−1 (x)
f k = Ut
Ut f k−1 \ Bt
fk , k ≥ 1.

(R+1)
We now define the bijection. If o 6= u ∈ V ∩ Bt0 , then there exists a unique pair (r, k) with r ≥ 1
(r)
and k ≥ 1 such that u ∈ Btk . Due to the definition of the burning rule, there exists at least one y ∼ x
(r)
such that y ∈ Btk−1 . We select an edge that connects u to one of these vertices, using the ordering ≺u ,
(r)
as in Section 2.3. Namely, if Pu is the number of edges joining u to neighbours in ∪ℓ<k Btℓ , and Au is
(r)
the subset of such edges leading to vertices in Btk−1 , then necessarily

η∗ (u) = 2d − Pu + i for some 0 ≤ i < |Au | .

We add to To the i-th edge in Au in the ordering ≺u .

(R+1) f 0 , there is a unique k ≥ 1 such that u ∈ Bt
f k , and there exists at least one y ∼ x
If u ∈ Ut0 = Ut
f
with y ∈ Btk−1 . We select an edge to one of these vertices, using the ordering ≺u as before. Namely, if
Pu is the number of edges joining u to neighbours in ∪ℓ<k Btf ℓ , and Au is the subset of such edges leading
f k−1 , then necessarily
to vertices in Bt

η∗ (u) = 2d − Pu + i for some 0 ≤ i < |Au | .

We add to Ts the i-th edge in Au in the ordering ≺u . Let ϕ′ (η∗ ) := (To , Ts ) denote the two components
spanning forest obtained by the above construction. Let us write TV,o for the set of all spanning forests
of GV rooted at {s, o}.
Lemma 2.10.
(i) The map ϕ′ is a bijection between R′V \ RV and TV,o .

14
(ii) For any η∗ ∈ R′V \ RV , the vertex set of To (η∗ ) equals W(η∗ ).
(iii) We have the following property:
If there is a path from o to a vertex x ∈ V in To = ϕ′ (η∗ ) that stays inside B(r), then starting
(2.4)
from η∗ + 1o there is a sequence of topplings in B(r) that topples x.
Proof. (i) Let η∗ 6= ηb∗ ∈ R′V \ RV . Tracing the burning process to the first time when a vertex with
η∗ (x) 6= ηb∗ (x) is encountered, we see that ϕ′ is injective on R′V \ RV . It follows from the definitions that
ϕ′ (R′V \ RV ) is a subset of the set of spanning forests of GV rooted at {o, s}. By the matrix-tree theorem
applied to G′V , the number of spanning forests of GV rooted at {o, s} equals det(∆′V ) − det(∆V ). This
also equals |R′V \ RV | [8], so statement (i) follows.
(R+1)
(ii) The burning process that was used to define Bt0 can be identified with topplings in the wave
(R+1)
corresponding to a′o,V (η∗ ). This implies that Bt0 = W(η∗ ), and this is the vertex set of To .
(iii) This again follows directly from the interpretation of the burning process in terms of topplings
in the wave.

2.6 Random walk notation and basic facts

Many of our techniques require a detailed analysis of spanning trees via Wilson’s algorithm. For this
reason, we will often have to consider collections of simple random walks (“SRW”) and loop-erased
random walks (“LERW”) on (subsets of) Zd .
We will denote by Sx = (Sx (0), Sx (1), . . .) an infinite simple random walk on Zd started at x, so that
Sx (0) = x. We will suppress the subscript when the choice of x is clear or when x = o. In general, Sx
and Sy will be assumed independent when x 6= y. When multiple independent walks from one site are
necessary, we write Sx , Sx′ (etc).
If A ⊂ Zd we define the standard stopping times
σA = inf{n ≥ 0 : Sx (n) ∈
/ A} ξA = inf{n ≥ 0 : Sx (n) ∈ A} , (2.5)
σ A = inf{n ≥ 1 : Sx (n) ∈
/ A} ξ A = inf{n ≥ 1 : Sx (n) ∈ A} . (2.6)
Note that we suppress any dependence of these stopping times on the starting point x; when we write
(for instance) Sx (ξA ), we are referring to the location of Sx at its first hitting time on A. When the
starting vertex may be ambiguous, we use subscripts on the symbol P; for instance,
Px (ξA < ξB ) = P(Sx [0, ξA ] ∩ B = ∅) .
We will abbreviate σV (n) to σn , and write ξo for ξ{o} .
The trace of a walk Sx between two times a < b (where b can be infinite) will be denoted
Sx [a, b] := {Sx (j) : a ≤ j ≤ b},
and similar notation will be used for other intervals–e.g., S(a, b) and so on. We will sometimes abuse
notation and treat Sx [a, b] as a sequence instead of an unordered set.
If A ⊂ Zd is a finite connected set and if x, y ∈ Zd , recall the Green function
X
GA (x, y) := P(Sx (j) = y) . (2.7)
0≤j<σA

As before, GV (n) is abbreviated Gn . We will use the following standard asymptotics for the Green
function inside a large ball and the probability of hitting o before exiting a large ball:

15
Theorem 2.11 (See [23], Prop. 1.5.9 and 1.6.7).
(i) If d = 2, we have uniformly in x ∈ B(n):
2

GB(n) (o, x) = [log n − log |x|] + O |x|−1 + 1/n
π

1 

Px ξo < σB(n) = log n − log |x| + O |x|−1 + 1/n .
log n
(ii) For d ≥ 3, there exist c1 = c1 (d), c2 = c2 (d) > 0 such that, uniformly in n and x ∈ B(n):
 

GB(n) (o, x) = c1 |x|2−d − n2−d + O |x|1−d

 
Px ξo < σB(n) = c2 |x|2−d − n2−d + O(|x|1−d ) .

We also note the following simple observation about G. If K1 ⊂ K2 and x, y ∈ K1 ,

GK1 (x, y) ≤ GK2 (x, y). (2.8)

We will need the following result, usually called the Beurling estimate, which gives an upper bound
on the probability that a path in Z2 is not hit by SRW.
Lemma 2.12 (Beurling estimate [19], [24, Section 6.8]). Consider Z2 . There is a constant C such that
the following bound holds, uniformly in x, n, and lattice paths α connecting o to ∂B(n):
 1/2
|x|
Px (σB(n) < ξα ) ≤ C .
n
Given a transient random walk Sx , we denote by LSx the loop-erasure of Sx , where loops are erased
in forward chronological order. A similar definition is made for LSx [a, b] := L(Sx [a, b]), etc.; note that
for a finite segment of a random walk, the loop erasure can be defined even in the case that the walk is
recurrent. See [24, Chapter 9] and [23, Chapter 7] for background on LERW; in particular, for properties
not detailed below.
If A ∋ x is a finite subset of Zd , let SbxA denote a finite loop-erased random walk killed at the boundary
of A:
SbxA = LSx [0, σA ] .
We will also make use of infinite loop-erased random walks Sbx on Zd . When d ≥ 3, the definition

Sbx := LSx

is unambiguous with probability 1, as noted in the last paragraph.

For d = 2, the loop-erasure of the infinite random walk Sx is not well-defined; in this case, Sbx is
defined by taking limits. It is known (see [23, Section 7.4]) that for any finite lattice path γ with |γ| = k,

lim P(SbxB(n) [0, k] = γ) =: P(Sbx [0, k] = γ)

n→∞

exists, and the extension of this to a measure on infinite paths gives a definition of the distribution of Sbx .
B(n)
The rate of convergence of Sbx to Sbx is well-controlled; see Lemma 3.4 below. We will refer to all of
b bA
the processes Sx , Sx as loop-erased random walks or LERW. We will also assume as usual (unless stated
otherwise) that a LERW is independent of any other walks appearing in a given statement.

16
 b σ
We define LERW stopping times ξ, b analogously to ξ and σ; for instance,

bK = inf{n ≥ 0 : Sbx (n) ∈

σ / K} .

As before, we will use x as a subscript on P when considering stopping times to indicate the starting
point. When the LERW is finite (i.e., we are considering SbxA ), we also will use the superscript A on P,
writing (for instance)


PA b b
x ξK1 < ξK2

to avoid confusing the set in which the LERW lives with the set it is hitting. When the LERW is infinite,
we omit superscripts altogether.
One result important for analyzing LERW is the following “Domain Markov Property” (DMP). This
roughly says that the terminal segment of a LERW can be built by starting a SRW at the tip of the
initial segment, conditioning it not to hit the initial segment, then erasing loops.
Lemma 2.13 (Domain Markov Property; see [24, Chapter 11]). Let SbxK be a loop-erased walk in K, and
let α be a finite path of length m such that
 
P SbxK [0, m] = α > 0 .

Then for all paths β,

   
P SbxK [m, σ
bK ] = β SbxK [0, m] = α = P LSα(m) [0, σK ] = β σK < ξ α .

Note that Lemma 2.13 is in fact much more general: it holds for infinite LERW (on Zd for d ≥ 3) and
finite LERW on graphs which are not necessarily subsets of Zd .

2.7 Wilson’s algorithm

Let H = (U ∪ {s}, F ) be a connected finite graph. Since νH is uniform on RH , the bijection ϕ maps it
to the uniform measure on spanning trees of H, which we denote by µH . In this section, we collect basic
facts about Wilson’s algorithm, that we will use to analyze µH . For an in-depth introduction to uniform
spanning trees (UST) see the book [26].
We denote a sample from µH by TH . We will usually identify a spanning tree of H with the set of
edges it contains, so TH ⊂ F . Pemantle [31] proved that for x, y ∈ U , the path in TH between x and y
is distributed as a LERW from x to y (i.e. as LSx [0, ξy ]). Wilson’s algorithm [35] provides a method for
constructing the full UST (a sample from µH ) from LERWs.
Wilson’s algorithm. Let v1 , . . . , vn be an enumeration of the vertices in U . We construct a random
sequence of tree subgraphs F0 ⊂ . . . ⊂ Fn . Let F0 have vertex set {s} and empty edge set. If Fi−1 has
been defined for some 1 ≤ i ≤ n, we let Svi be a simple random walk on H started at vi , and let ξV (Fi−1 )
be the first hitting time of the vertex set of Fi−1 by Svi . We set Fi = Fi−1 ∪ LSvi [0, ξV (Fi−1 ) ]–that is,
the edges in the loop-erasure of Svi [0, ξV (Fi−1 ) ] are added to Fi−1 . The output of the algorithm is Fn .
Wilson’s theorem [35] implies that Fn is uniform, i.e. distributed as TH .
The measure µL := µGL is known as the UST in V (L) with the wired boundary condition. For
studying the L → ∞ limit of sandpiles on GL , as well as sandpiles on Zd , it will be useful to consider the
weak limit limL→∞ µL =: WSF, called the wired spanning forest measure. Existence of the limit is
implicit in [31]; see [26] for an in-depth treatment. We denote a sample from WSF by T.

17
 It is well known that WSF concentrates on spanning forests of Zd all whose components are infinite.
Pemantle [31] showed that for d ≤ 4, T is a tree WSF-a.s, while for d ≥ 5, T has infinitely many
connected components WSF-a.s. This dichotomy is the underlying fact behind mean-field behaviour of
the sandpile model for d ≥ 5; which is reflected in some of our results and proofs. We write Tx for the
component of T containing x ∈ Zd .
It is possible to construct T more directly, using an appropriate extensions of Wilson’s algorithm. Let
v1 , v2 , . . . be an enumeration of Zd . When d ≥ 3, we set F1 = LSv1 [0, ∞). Then for i ≥ 2, we inductively
define Fi = Fi−1 ∪ LSvi [0, ξV (Fi−1 ) ], where the stopping time may be finite or infinite. See [6, 26] for a
proof that ∪i≥1 Fi has the distribution of T given by WSF. This is called Wilson’s method rooted at
infinity.
When d = 2, a method analogous to that in finite volume can be used. We set F1 = {v1 }, and for
i ≥ 2 we inductively define Fi = Fi−1 ∪ LSvi [0, ξV (Fi−1 ) ]; see [6, 26].
It will be important for our proofs that WSF-a.s. each component of T has one end, for all d ≥ 2.
This means that any two infinite self-avoiding paths lying in the same component of T have a finite
symmetric difference. For 2 ≤ d ≤ 4 this was proved by Pemantle [31]. For d ≥ 5 this was first shown by
Benjamini, Lyons, Peres and Schramm [6] (who generalized it to a much larger class of infinite graphs).

2.8 Wiring o to the boundary

Let µL,o be the uniform measure on TL,o . Under µL,o , we denote the components containing o and s,
respectively, by TL,o and TL,s , respectively.
Due to monotonicity, the weak limit limL→∞ µL,o =: WSFo exists; see [6, 26]. We also use T to
denote a sample from WSFo , and write Tx for its component containing x ∈ Zd . In particular, To is the
component containing o under the measure WSFo . It was shown by Lyons, Morris and Schramm [25] that
for d ≥ 3 (and more generally under a suitable isoperimetric condition), we have WSFo (|To | < ∞) = 1,
and also that this is equivalent to the one-end property of WSF. Járai and Redig [15] used the finiteness
of To to show that ν(S < ∞) = 1 when d ≥ 3.

3 Toppling probability bounds in low dimensions

In this section, we give a proof of our toppling probability lower bounds stated in Theorem 1.1. These will
follow from the following theorem, which gives a lower bound in terms of a non-intersection probability
between a loop-erased walk and a simple random walk, and a random walk hitting probability. Our
arguments also apply to d ≥ 5 with an identical statement, however, this case is already known from [16,
Section 6.2].
Recall that Sx and Sx′ are independent simple random walks from the site x.
Theorem 3.1. Assume 2 ≤ d ≤ 4. There is a constant c = c(d) > 0 such that, for all z ∈ Zd we have
 
ν(z ∈ Av) ≥ c P So′ [0, σ|z| ] ∩ Sbo (0, σ
b|z| ] = ∅ P(z ∈ So [0, ∞)) . (3.1)

Moreover, the right hand side of (3.1) is a lower bound on νL (z ∈ Av) for all L ≥ 4kzk.
In Sections 3.1–3.2, we state and prove preliminary results which are useful for establishing Theorem
3.1, and in Section 3.3 we use these to prove the theorem. In Section 3.3, we also give a corollary which
will be useful for proofs of later theorems.

18
3.1 Preliminary setup
Our strategy for proving Theorem 3.1 will be to work in large finite volume V (L). That is, given a
particular z ∈ Zd , we will choose some L0 sufficiently large, so that the probability νL (z ∈ Av) is close
to the claimed value for all L ≥ L0 . We will require L0 to be on the order of some large multiple of kzk.
The main idea of the proof is to show a lower bound for the probability that z is in the last wave of
the avalanche. By Corollary 2.8,
|{η ∈ RL : z ∈ Av}| ≥ |{η ∈ RL : η(o) = 2d − 1, z ∈ WN (η) }|
= |{η∗ ∈ R′L \ RL : z ∈ W(η∗ ), v ∈
/ W(η∗ ) for some v ∼ o}| .
Dividing by |RL |, using Lemma 2.9 and symmetry of V (L), for any fixed e ∼ o we get
νL (z ∈ Av) ≥ µL,o (z ∈ TL,o | v ∈
/ TL,o for some v ∼ o)
(3.2)
≥ (2d)−1 µL,o (z ∈ TL,o | e 6∈ TL,o )
uniformly in z, L.
We analyze the event
A(z, e) = {z ∈ TL,o, e 6∈ TL,o }.
Let us apply Wilson’s algorithm in the graph GL,o , starting with a walk Se from e, followed by a walk
Sz from z. This gives that for fixed e ∼ o, the occurrence of A(z, e) is equivalent to a LERW from e to
exit V (L) without hitting o, and a SRW from z to hit o before hitting the LERW. We will bound the
right-hand side of (3.2) from below by analyzing this random walk event. By time reversal, we will be
able to consider the SRW going from o to z. The lower bound then contains two factors: the LERW and
SRW avoiding each other near o, and the SRW subsequently hitting z. This leads to the two probabilities
in Theorem 3.1.
We begin by expressing the probability in (3.2) in terms of the random walk construction specified
above (Lemma 3.2). We then lower bound the probability of the resulting walk event by something
amenable to analysis by walk intersection techniques that we give in Section 3.2. Let π = LSe [0, σL ].
Lemma 3.2.
(i) We have
 
Sz
µL,o (A(z, e)) = P π ∩ Sz [0, ξo ] = ∅, ξoSz < σL . (3.3)

(ii) There are constants κ(d) > 0, d ≥ 2, such that as L → ∞, we have

(

κ(2)(log L)−1 when d = 2;
µL,o (e ∈
/ TL,o ) = P o ∈
/ Se [0, σV (L) ] ∼ . (3.4)
κ(d) when d ≥ 3.

Proof. Note that e ∈/ TL,o if and only if Se exits V (L) before hitting o. This implies the equality in (ii).
The asymptotics in (ii) for d = 2 follow from Theorem 2.11.
Given that the event e ∈ / TL,o has occurred, z will be in TL,o if and only if Sz hits o before exiting
V (L), and does so avoiding π. This implies statement (i).
In the sequel, we make use of the event Γz,L , defined as
n o
So
Γz,L = π ∩ Vz (kzk/10) = ∅, ξzSo < σ4kzk , π ∩ So [0, ξz ] = ∅ .

19
Lemma 3.3. For all L ≥ 100|z|, we have
(
  c P(Γz,L ) log |z|, d=2,
Sz
P π ∩ Sz [0, ξo ] = ∅, ξoSz < σL ≥ (3.5)
(2d)−1 P(Γz,L ), d>2.

Proof. Using reversibility of the random walk, we can rewrite the probability in the left hand side of (3.5)
as follows:
 
Sz
P π ∩ Sz [0, ξo ] = ∅, ξoSz < σL
  
Sz
= E P π ∩ Sz [0, ξo ] = ∅, ξoSz < σL π
 
GV (L)\π (z, z)  So So (3.6)
=E P π ∩ So [0, ξz ] = ∅, ξz < σL π
GV (L)\π (o, o)
 
GV (L)\π (z, z)  So So
≥ E 1π∩Vz (kzk/10)=∅ P π ∩ So [0, ξz ] = ∅, ξz < σL π .
GV (L)\π (o, o)
In the presence of the indicator, (2.8) implies GV (L)\π (z, z) ≥ GVz (kzk/10) (z, z). Since e ∈ π, we also have
GV (L)\π (o, o) ≤ GZd \{e} (o, o) ≤ 2d,
since after each visit to o, the random walk next hits e with probability (2d)−1 . It follows that the
right-hand side of (3.6) is at least
  
So
(2d)−1 GVz (kzk/10) (z, z) E 1π∩Vz (kzk/10)=∅ P π ∩ So [0, ξz ] = ∅, ξzSo < σL π
(3.7)
≥ (2d)−1 GVz (kzk/10) (z, z) P(Γz,L ).

Using Theorem 2.11 and (2.8), we have

(
c log |z| , d=2
GVz (kzk/10) (z, z) ≥
1, d>2,

Inserting this estimates into (3.7) completes the proof.

3.2 SRW and LERW steering

In order to control the spanning tree event in (3.2), we need to give a lower bound on the probability
of the event Γz,L of Lemma 3.3. This will be achieved by “steering” the two walks so that they become
well separated as they reach distance of order kzk, which allows us to arrange that π avoids the box
Vz (kzk/10), and it does not influence very much the probability that So hits z. In Section 3.2.1 we collect
results we need for a single walk. In Section 3.2.2 we prove the required separation estimate. In Section
3.3 we prove a lower bound on P(Γz,L ) and complete the proof of Theorem 3.1.

3.2.1 Estimates for a single walk

The first lemma we need, shown by Masson, holds for general d, and compares an infinite LERW to a
finite LERW, allowing us to control probabilities by restricting to finite balls. Masson stated this in the
case K ⊃ B(4n), however, his proof applies in the slightly more general case we use here.

20
Lemma 3.4. [29, Corollary 4.5] Let d ≥ 2 be arbitrary. For any δ > 0, we have

P(Sbo [0, σ
bn ] = α) ≍δ P(SboK [0, σ
bn ] = α),


for all α, all n ≥ 1/δ, and all K ⊃ V (1 + δ)n , where the constants implied by the ≍δ notation only
depend on δ and d.
We will need the following “Boundary Harnack inequality”, to control a LERW after it has reached
the boundary of a box. Estimates of this flavour were proved in [29, Proposition 3.5], [34, Proposition
6.1.1], [4] and [5, Section 3]. The variant we need here is a simplified version of [5, Lemma 3.8]. We define

Hn = {x ∈ Zd : x · e1 = n}, Hn+ = {x ∈ Zd : x · e1 ≥ n}, Hn− = {x ∈ Zd : x · e1 ≤ n} .

Lemma 3.5. There exists c(d) > 0 such that the following holds. Let π ⊂ V (n/2) and x ∈ ∂V (n/2) ∩
Hn/2 . Let 1 ≤ m ≤ n/4, and L ≥ 4n. We have

m
Px S(σVx (m) ) ∈ Hn/2+m σL < ξ π ≥ c(d) .
n
Proof. Let π ′ = π ∩ Vx (m). It is shown in [5, Section 3] that

1

Px S(σVx (m) ) ∈ Hn/2+m , σVx (m) < ξ π′ ≥ Px σVx (m) < ξ π′ .
2d
This yields


Px S(σVx (m) ) ∈ Hn/2+m , σL < ξ π


≥ Px S(σVx (m) ) ∈ Hn/2+m , σVx (m) < ξ π′



min Pw σ2n < ξHn/2 min Pz σL < ξπ
w∈(∂Vx (m))∩Hn/2+m z∈∂V (2n)
1
m

≥ Px σVx (m) < ξ π′ c max Pz σL < ξπ
2d n z∈∂V (2n)

c(m/n)

m

≥ Px σV (2n) < ξ π max Pz σL < ξπ ≥ c(d) Px σL < ξ π .
2d z∈∂V (2n) n
Here we used a gambler’s ruin estimate and the Harnack principle in the second inequality.

3.2.2 Separation lemma

Separation lemmas for a loop-erased walk and a simple random walk appeared in [29] for d = 2 and in
[34] for d = 3. We give here a unified proof that works for all d ≥ 2. Let us write SbL,o = LSo [0, σL ].
Recall that Sx′ generally denotes a simple random walk independent of the walk Sx . We define An =
{SbL,o (0, σ
bn ] ∩ So′ [0, σn ] = ∅}, n ≤ L. Let
n

o
Dn = min dist SbL,o (b
σn ), So′ [0, σn ] , dist SbL,o [0, σ
bn ], So′ (σn ) .

Lemma 3.6 (Separation Lemma). Let d ≥ 2. There exists δ = δ(d) > 0 and c = c(d) > 0 such that
for all 1 ≤ n ≤ L/4 we have
P(Dn ≥ δn | An ) ≥ c.

21
 In the proof of the separation lemma, a basic step is to show that given that non-intersection occurred
to distance n/2, there is small probability that separation at distance n is bad.
Lemma 3.7. Let d ≥ 2. There exists a function r : (0, 1/2] → (0, ∞) with limδ→0 r(δ) = 0, and for all
δ ∈ (0, 1/2] there exists n0 (δ) such that we have we have


P An ∩ {Dn < δn} An/2 ≤ r(δ), n ≥ n0 (δ), 0 < δ ≤ 1/2.

Proof. We condition on π0 := SbL,o [0, σ bn/2 ] and So′ [0, σn/2 ], and denote x1 = SbL,o (b
σn/2 ), x2 = SbL,o (b
σn ),
′ ′
y1 = So (σn/2 ), y2 = So (σn ). We distinguish two cases:
(a) So′ [σn/2 , σn ] visits Bx2 (δn);
(b) SbL,o [b
σn/2 , σbn ] visits By2 (δn).
We bound the probabilities of the two cases separately, showing that each is bounded by a suitable r(δ).
Case (a). Let us further condition on x2 . Since Brownian motion in the cube {u ∈ Rd : kuk ≤ 1}
has continuous paths, and the path tends to its exit point, and the probability of any given exit point is
0, there exists r1 (δ), tending to 0, such that given any boundary point w, the Brownian path intersects
{u ∈ Rd : kuk ≤ 1, |u − w| ≤ 2δ} with probability ≤ r1 (δ). Hence the required bound follows from the
invariance principle.
Case (b). Let us condition on y2 . Due to the Domain Markov Property (Lemma 2.13), the path
SbL,o [b bL ] has the law of L(X[0, σL ]), where X has the law of Sx1 conditioned on the event {σL <
σn/2 , σ
ξ π0 }. On the event in Case (b), the path X[σ3n/4 , σL ] has to visit By2 (δn). Conditioning on the point
x′ = X(σ3n/4 ), the probability of this event is


′

Px′ σL < ξπ0 , ξBy2 (δn) < σL
P X[σ3n/4 , σL ] ∩ Bδn (y2 ) 6= ∅ X(σ3n/4 ) = x =
. (3.8)
Px′ σL < ξπ0

When d ≥ 3, the right hand side is at most

Px′ (ξBδn (y2 ) < ∞)

≤ Cδ d−2 .
Px′ ξBn/2 (o) = ∞

When d = 2, consider
 
h(w) := Pw σL < ξπ0 , ξBy2 (δn) < σL and w∗ := argmax h(w).
w∈V (2n)\Vy2 (n/8)

Using the Harnack principle, we have

   
h(w∗ ) ≤ Pw∗ ξBy2 (δn) < σ4n max Pv (σL < ξπ0 ) + Pw∗ σ4n < ξπ0 ∧ ξBy2 (δn) max h(w)
v∈∂V (4n) w∈∂V (2n)

≤ C (log 1/δ)−1 C Px′ (σL < ξπ0 ) + c1 h(w∗ ),

where c1 < 1. Therefore, we get

(1 − c1 )h(x′ ) ≤ (1 − c1 )h(w∗ ) ≤ C(log 1/δ)−1 Px′ (σL < ξπ0 ) .

This completes the proof.

22
 The next step is to show that given a “good” separation at distance n/2, the probability that the paths
can be “very well” separated at distance n is at least a constant. We denote Gn := σ(SboL [0, σ
bn ], So′ [0, σn ])
the information about the paths up to the exit from Vn .
Lemma 3.8. For any 0 < δ ≤ 1/2 there exists c(δ) > 0 and n1 (δ) such that for all n ≥ n1 (δ) and L ≥ 4n
the following hold.
(i) We have  

P An ∩ SboL (b −
bn ) ∈ H−n/2
σ3n/4 , σ +
, So′ (σ3n/4 , σn ) ∈ Hn/2 Gn/2 ≥ c(δ) (3.9)

everywhere on the event An/2 ∩ {Dn/2 ≥ δn/2)}.

(ii) Moreover, we may further require the event SboL [b bL ] ∩ Vn = ∅ in (3.9).
σ2n , σ

Proof. We condition on π0 = SboL [0, σbn/2 ] and So′ [0, σn/2 ]. Let us write x1 = SboL (b
σn/2 ) and y1 = So′ (σn/2 ).
Due to the conditioning in (3.9), we have |x1 − y1 | ≥ δ(n/2). We write X for a random walk starting at
x1 conditioned on the event {σL < ξ π0 }, so that SboL [b bL ] has the law of LX[0, σL ].
σn/2 , σ
An application of Lemma 3.5 yields that with probability ≥ cδ, the process X exits Vx1 (δn/8) on
the face furthest from V (n/2). Also, there is probability ≥ (2d)−1 that So′ [σn/2 , ∞) exits Vy1 (δn/8) on
the face furthest from V (n/2). Using the Harnack principle for X, and appropriate disjoint corridors of
width of order δn for the LERW and the SRW, respectively (see Figure 1 below), there is probability
≥ c(δ) that:
(a) So′ exits V (n) in Hn , with the appropriate portion in the required halfspace;
(b) X exits V (2n) in H−2n with X[σ3n/4 , σ2n ] ⊂ [−2n, −n/2] × [−3n/4, 3n/4]d−1;
(c) So′ [0, σn ] ∩ X[0, σn ] = ∅.
In order to further ensure that LX first exits V (n) at a point in H−n , and that An occurs, we show that
for w ∈ ∂V (2n) we have  
X
Pw σL < ξVX(n) ≥ c > 0. (3.10)

This is indeed sufficient, since the events in point (b) and (3.10) imply that the last visit of X to ∂V (n)
must occur at a point in H−n . In order to see (3.10), first note that the statement is clear for d ≥ 3,
since then



Pw σL < ξV (n) ≥ Pw ξV (n) = ∞ ≥ c ≥ c Pw (σL < ξπ0 .


When d = 2, let w∗ := argmax Pw σL < ξπ0 . Then we have
w∈∂V (2n)

Pw∗ σL < ξπ0 ≤ Pw∗ σL < ξV (n) + max Py σ2n < ξπ0 Pw∗ σL < ξπ0 .
y∈∂V (n)

The maximum over y is ≤ c2 < 1, which after rearranging gives

1

Pw∗ σL < ξπ0 ≤ Pw∗ σL < ξV (n) .
1 − c2
For other w ∈ ∂V (2n) we obtain the statement from the Harnack principle.
The construction we gave ensures both the event in (i) and (ii), and thus completes the proof of the
lemma.
We are ready to prove the Separation Lemma. The argument we give is inspired by [20].

23
 H3n/4

Figure 1: Depiction of the extension in disjoint corridors.

Each walk is assumed to exit Vn/2 at the center of one of
the corridors, which are distance ∼ δn apart. Note the
placement of the corridors ensures that terminal segments
of each walk remain in appropriate half-spaces as in (3.9).
V (n/2)

V (n)
Proof of Lemma 3.6. Let us write



f (n) = P An and g(n) = P An ∩ {Dn ≥ δn} .

Let δ > 0 that we will choose later. Let n ≥ max{n0 (δ), n1 (δ)}, where n0 and n1 are the constants from
Lemmas 3.7 and 3.8. Lemma 3.7 implies


f (n) = g(n) + f (n/2) P An ∩ {Dn < δn} An/2 ≤ g(n) + f (n/2) r(δ)
k−1
X (3.11)
≤ r(δ)ℓ g(n/2ℓ ) + r(δ)k f (n/2k ).
ℓ=0

Lemma 3.8 implies that on the event An/2ℓ ∩ {Dn/2ℓ ≥ δn/2ℓ }, we can extend both the loop-erased walk
and the random walk to opposite faces of ∂V (n/2ℓ−1 ), with probability at least c(δ). A gambler’s ruin
estimate then implies that, there is probability ≥ (c/2ℓ )2 that the walks reach ∂V (n) without intersecting.
This shows that g(n/2ℓ ) ≤ c(δ)22ℓ g(n). Substituting this into (3.11) yields
" k−1
#
X
f (n) ≤ g(n) 1 + c(δ) (4 r(δ))ℓ + r(δ)k f (n/2k ). (3.12)
ℓ=1

Choose δ > 0 so that 4 r(δ) < 1/2, and the smallest k such that max{n0 (δ0 ), n1 (δ)} ≤ n/2k . Then (3.12)
implies f (n) ≤ C(δ) g(n).

3.2.3 Estimate on Γz,L

In this section, we prove a lower bound on the probability of the event Γz,L , which yields a finite volume
analogue of Theorem 3.1. In the following, let


EsL (n) := P SboL (0, σ
bn ] ∩ So′ [0, σn ] = ∅ ,

which is an (expectation of an) escape probability for the walk So′ .

24
Lemma 3.9. Let d ≥ 2. There exists c = c(d) > 0 such that for any z ∈ Zd \ {o} and L > 4kzk we have
(


c EsL (kzk) Po ξ{z} < σ4kzk if d ≥ 3;
P Γz,L ≥
(3.13)
c (log L)−1 EsL (kzk) Po ξ{z} < σ4kzk if d = 2.

Proof. We may assume that kzk is sufficiently large. Without loss of generality, we also assume that the
first coordinate of z is positive and has maximal absolute value among all coordinates.
We first require the occurrence of the event

B(e) := π ∩ {o} = ∅ . (3.14)
We have (

c(log L)−1 when d = 2;
qL := P(B(e)) = Pe σL < ξo ≥ (3.15)
c when d ≥ 3.

Conditional on B(e), the law of π is the same as the law of SboL [1, σ
bL ] conditional on SboL (1) = e. Therefore,
we will express properties of π conditional on the event B(e) in terms of the properties of SboL [1, σ bL ]
conditional on SboL (1) = e.
Let us require the occurrence of the event

Akzk/4 ∩ Dkzk/4 ≥ δkzk/4 , (3.16)
where δ = δ0 is the constant chosen in Lemma 3.6. According to that lemma, the event in (3.16) has
unconditional probability ≥ cEsL (kzk/4) ≥ cEsL (kzk). Due to Zd -symmetry, the conditional probability
of (3.16) given SbL (1) = e is the same as the unconditional probability.
Let us further require the event in Lemma 3.8 with n = kzk/2, that is, that the paths extend disjointly
to opposite faces of V (kzk/2), with the random walk landing on Hkzk/2 , and the LERW landing on
H−kzk/2 . According to Lemma 3.8, this happens with conditional probability ≥ c. It follows from
Lemma 3.8, that there is probability bounded away from 0 that the LERW can be further extended to
−
land on H−8kzk , in such a way that π is contained in H3kzk/8 ∪ V (4kzk)c. Since So′ (σkzk/2 ) ∈ Hkzk/2 , the
conditional probability, given π and So′ [0, σkzk/2 ] that So′ hits z before ξπ ∧ σ4kzk is ≥ c |z|2−d when d ≥ 3,
and ≥ (log |z|)−1 when d = 2. Combining the estimates for each part of the construction yields:
(

c EsL (kzk) |z|2−d when d ≥ 3;
P Γz,L ≥ −1 L 1
c (log L) Es (kzk) log |z| when d = 2.

This completes the proof of the lemma when kzk is sufficiently large.
The following proposition summarizes the result of the finite L arguments we made in Sections 3.1–3.2.
Proposition 3.10. Let d ≥ 2. For any z ∈ Zd and all L ≥ 4kzk we have


νL (z ∈ Av) ≥ c EsL (kzk) GVz (kzk/10) (z, z) Po ξz < σ4kzk
(3.17)
≍ Eskzk (kzk) Po (ξz < ∞) .

Proof. Combining (3.2), Lemma 3.2, Lemma 3.3, and Lemma 3.9, we obtain the statement in both d ≥ 3
and d = 2. Lemma 3.4 implies that EsL (kzk) is comparable to Es4kzk (kzk). Masson [29, Lemma 5.1]
showed that the latter is comparable to Eskzk (kzk).

25
3.3 Proof of Theorem 1.1
In passing to the limit L → ∞, we use the following proposition. We prove only the d ≥ 3 case here; the
proof of the more technical d = 2 case is deferred to the end of Section 5; see Lemma 5.9 there.
Proposition 3.11.
Assume d ≥ 2. Then we have ν(z ∈ Av) = limL→∞ νL (z ∈ Av).
Proof of Proposition 3.11, d ≥ 3 case. Due to [15, Theorem 3.11] we have ν(|Av| < ∞) = 1. Therefore,
given ε > 0, we can find |z| < M < ∞ such that ν(Av ⊂ V (M )) > 1 − ε. Due to the weak convergence
νL → ν, there exists M < L0 < ∞ such that for all L ≥ L0 we have

|ν(z ∈ Av) − νL (z ∈ Av)| ≤ |ν(z ∈ Av) − ν(z ∈ Av, Av ⊂ V (M ))|

+ |ν(z ∈ Av, Av ⊂ V (M )) − νL (z ∈ Av, Av ⊂ V (M ))|
+ |νL (z ∈ Av, Av ⊂ V (M )) − νL (z ∈ Av)|
< 3ε.

This completes the proof.

We can now complete the proof of Theorem 3.1.
Proof of Theorem 3.1. Lemma 3.4 implies that EsL (kzk) in Proposition 3.11 is comparable to the avoid-
ance probability in the theorem, so the statement follows using Proposition 3.11.
For later use we state here a corollary of the construction.
Corollary 3.12. Let d ≥ 2. There exists a constant c = c(d) > 0 such that the following holds. For each
z ∈ Zd and L ≥ 4kzk and e ∼ o, with π denoting the path in TL,s connecting e to s, we have




/ TL,o ≥ c P So [0, σ|z| ] ∩ Sbo′ (0, σ
µL,o z ∈ TL,o, π ∩ Vz (kzk/10) = ∅ e ∈ b|z| ] = ∅ Po ξz < ∞ .

We now turn to the proofs of the explicit bounds for 2 ≤ d ≤ 4.

3.3.1 Proof of Theorem 1.1 when d = 2

By Theorem 3.1, Theorem 1.1(i) will hold once we know
 
b σ
P So′ [0, σn ] ∩ S(0, bn ] = ∅ ≥ n−3/4+o(1) . (3.18)

The exponent 3/4 + o(1) was first proved by Kenyon [18], who stated it for simple random walk in the
half plane. A proof for more general walks was given by Masson, who derived it from results on SLE2
[29, Theorem 5.7]. He established the analogue of (3.18) for a SRW and a finite LERW – via [29, Lemma
5.1], where the intersection probabilities for finite and infinite LERW are related. This implies Theorem
1.1(i).

26
3.3.2 Proof of Theorem 1.1 when d = 3
In this section, we complete the proof of the explicit lower bound in Theorem 1.1(ii) by showing that
P(So′ [0, σ(n)] ∩ Sbo (0, σ
b(n)] = ∅) ≥ cn−2ζ (this suffices, by the previously proven Theorem 3.1). Since Sbo
is the loop-erasure of S, it is enough to show


P So′ [0, σ(n)] ∩ S(0, ∞) = ∅ ≥ cn−2ζ . (3.19)

This is a simple adaptation of the results of [22]. Indeed, there exists a c1 > 0 such that (uniformly
in m)

P(So′ [0, m] ∩ S(0, ∞) = ∅) ≥ c1 m−ζ

by [22, discussion after (3)]. On the other hand, by [22, Lemma 4.7], there exists C2 , c2 > 0 such that
(uniformly in a, n)

P(So′ [0, σn ] ∩ S(0, ∞) = ∅, σn > an2 ) ≤ C2 exp(−a/c2 )n−2ζ .

Choosing a sufficiently large (relative to c1 , c2 ) and m = an2 completes the proof of (3.19). From this
Theorem 1.1(ii) follows.

3.3.3 Proof of Theorem 1.1 when d = 4

In four dimensions, the avoidance probability was determined by Lawler [21], who showed that

Es(n) ≍ (log n)−1/3 .

This and Theorem 3.1 yield Theorem 1.1(iii).

4 Low-dimensional radius bounds

In this section we prove the radius bounds stated in Theorem 1.2(i)–(iii) for dimensions 2 ≤ d ≤ 4. We
start with the lower bounds.
Proof of Theorem 1.2(i)–(iii), lower bounds. Observe that

ν (R ≥ r) ≥ ν(re1 ∈ Av) . (4.1)

Therefore, the claimed lower bounds follow immediately from Theorem 1.1(i)–(iii).
Proof of Theorem 1.2(ii)–(iii), upper bounds. Let us first consider a finite volume V (L). Recall that for
η ∈ RL we write α(η) = (η1 , . . . , ηN ) for the waves in the stabilization So (η + 1o ). Let us extend the
notation for the radius to waves and two-component spanning trees in the natural way:

R(η∗ ) = sup{|z| : z ∈ W(η∗ )}

R(To ) = sup{|z| : z ∈ To }

27
Under the bijection of Section 2.5 these two notions coincide. We have
 
η ∈ RL : R(η) > r = η ∈ RL : R(ηi ) > r for some 1 ≤ i ≤ N (η)

≤ η∗ ∈ R′L \ RL : R(η∗ ) > r .

Hence we get

|R′L \ RL |


νL R > r ≤ µL,o R(TL,o ) > r = gL (o, o) µL,o R(TL,o ) > r ,
|RL |
where the last equality uses Lemma 2.5.
Since {R > r} and {R(TL,o) > r} are both cylinder events, we can take the limit L → ∞ on both
sides to get


ν R > r ≤ C(d) WSFo R(TL,o ) > r . (4.2)
Lyons, Morris and Schramm [25] proved that


WSFo R(TL,o) > r ≤ C(d) r−βd , d ≥ 3,

with βd = 21 − 1d . Inserting this into (4.2) yields the upper bounds C r−1/6 and C r−1/4 in dimensions
d = 3 and d = 4, respectively.

5 The last k waves in 2D

In this section we prove that for any 1 ≤ k < ∞, as L → ∞, the last k waves on GL (when they exist)
have a weak limit. We introduce some notation for the last k waves. Recall that for η ∈ RL , we denote
by N = N (η) = nL (o, o) the number of times o topples during the stabilization of η + 1o . Recall from
Section 2.4, that given η ∈ RL , we write α(η) = (η1 , . . . , ηN ) for the set of intermediate configurations
(right before each wave). Let
ξ0L = ξ0L (η) := ao,L η
and whenever N (η) ≥ k, define
ξkL = ξkL (η) := ηN −k+1 .
Observe that when ξkL is defined, we have ξk−1
L
= a′o,L ξkL . Analogously to the finite graph case, a′o,Z2 is
the operator which adds a particle at the origin and stabilizes the resulting configuration in the graph
(Z2 )′ .
Theorem 5.1. Assume d = 2. We have:
(i) For every 1 ≤ k < ∞ the limit bk = limL→∞ νL (N ≥ k) exists.
(ii) For every 1 ≤ k < ∞, the law of ξkL under the measure νL (· | N ≥ k) converges weakly to the law ρk
of a configuration ξk .
(iii) The configuration ξk + 1o can be stabilized in (Z2 )′ with finitely many topplings ρk -a.s.
(iv) The transformation ξk 7→ a′o,Z2 ξk is measure-preserving between bk ρk and the restriction of bk−1 ρk−1
to the image. (Here b0 = 1, ρ0 := ν.)
(v) With ν-probability 1 on the event {N = k}, all k waves are finite, and we have ν(N = k) = bk − bk+1 .
Remark 2. It is not difficult to construct, for every 1 ≤ k < ∞, an explicit finite configuration around o
showing that lim inf L→∞ νL (N = k) > 0.

28
 Recall the bijection for intermediate configurations from Section 2.5. This bijection was between
η∗ ∈ R′L \ RL and the set of spanning forests of GL with two components To = To (η∗ ) and Ts = Ts (η∗ ),
where o ∈ To and s ∈ Ts . Recall the following property of the bijection from Section 2.5, rephrased for
the configurations ξkL .

If there is a path in To (ξkL (η)) from o to a vertex x that stays inside B(r), then starting from
(5.1)
ξkL + 1o there is a sequence of topplings in B(r) that topples x.

We write TL,o,k = To (ξkL ) and TL,s,k = Ts (ξkL ) for short. When x ∈ TL,o,k , we denote πL,k (x) the unique
self-avoiding path in TL,o,k from o to x.
Our control on the size of waves will be in terms of the following random variables.
L
Rin,k = sup{r ≥ 0 : B(r) ⊂ TL,o,k }, k ≥ 1;
L
Rout,k = inf{r ≥ 0 : TL,o,k ⊂ B(r)}, k ≥ 1; (5.2)
n  o
PkL L
= inf r ≥ 0 : πL,k (x) ⊂ B(r) for all x ∈ B Rin,(k−1) +1 , k ≥ 2.

All quantities in (5.2) are defined to be 0 when N < k. The following lemma states a basic inequality we
will need.
Lemma 5.2. We have:
L
(i) Rin,1 = 0.
L
(ii) Rin,k ≤ PkL , k ≥ 2.
Proof. (i) This follows directly from Corollary 2.9.
(ii) Write r = PkL for short, and assume for a proof by contradiction, that we had Rin,k L
≥ r + 1. This
implies that in the stabilization a′o,L (ξkL ) all vertices in B(r + 1) topple, and hence (ξk−1
L
)B(r) = (ξkL )B(r) .
L
Starting from the configuration ξk−1 + 1o , let us topple all sites in B(r) we can. The definition of PkL and
L
property (5.1) imply that all vertices in B(Rin,(k−1) + 1) topple. However, this contradicts the definition
L
of Rin,(k−1) .
In the following proposition we show that the in-radius of the k-th last wave is tight, with a power
law upper bound on the tail.
Proposition 5.3. There exist constants α′1 > α′2 > · · · > 0 and C1 , C2 , . . . such that
L

′
lim sup νL Rin,k > r ≤ Ck r−αk , ∀r ≥ 1, ∀k ≥ 1. (5.3)
L→∞

L
In particular, for all 1 ≤ k < ∞, the sequence {Rin,k }L≥1 is tight.
Proof. We prove the statement by induction on k. The case k = 1 holds trivially due to Lemma 5.2(i).
Assume k ≥ 2, and that (5.3) holds for k − 1. Let 1 ≤ r0 < ∞ be fixed, and find L0 = L0 (r0 ) < ∞ such
that for L ≥ L0 we have
L −α′
νL (Rin,(k−1) > r0 ) ≤ 2Ck−1 r0 k−1 . (5.4)
L L
It is sufficient to bound Rin,k when the event {Rin,(k−1) ≤ r0 } occurs, and due to Lemma 5.2(ii), it is
L L
enough to bound Pk on this event. In what follows, we assume the event {Rin,(k−1) ≤ r0 }.

29
 For ℓ ≥ 1 we are going to bound the probability that r0 2ℓ < PkL ≤ r0 2ℓ+1 . Due to Lemma 5.2(ii),
this event implies that (TL,o,k , TL,s,k ) belongs to the following event E(x, r0 , ℓ) for some x ∈ ∂B(r0 + 1):

E(x, r0 , ℓ) = (To , Ts ) ∈ TL,o : x ∈ To , π(x) visits B(r0 2ℓ )c and Ts ∩ ∂B(r0 2ℓ+1 ) 6= ∅ ,

where π(x) is the path in To from x to o. Therefore, using Corollary 2.6(ii), we have
  X {η ∈ R : r 2ℓ < P L (η) ≤ r 2ℓ+1 }
L L 0 k 0
νL Rin,(k−1) ≤ r0 , PkL > r0 2ℓ0 ≤
|RL |
ℓ≥ℓ0
X X (5.5)
≤ C (log L) µL,o (E(x, r0 , ℓ)).
ℓ≥ℓ0 x∈∂B(r0 +1)

We use Wilson’s algorithm to get an upper bound on the probability of E(x, r0 , ℓ). Let the first random
walk start at x. Let τ be the time of the last visit, before ξ{o} , to a vertex in ∂B(r0 2ℓ ). Let us condition
on the path Sx [0, τ ]. Let γ = LSx [0, τ ], and let γ0 be the initial segment of γ from x to the first visit of
γ to ∂B(r0 2ℓ ). The walk
S ′ (m) = Sx (τ + m), m = 0, . . . , ξo − τ ;
is a simple random walk on Z2 conditioned on ξo < ξ B(r0 2ℓ )c . On the event E(x, r0 , ℓ), S ′ cannot hit γ0 ,
so we bound the probability that S ′ hits B(r0 + 1) before γ0 .
The walk S ′ has to successively cross from ∂B(r0 2q ) to B(r0 2q−1 ) for q = ℓ − 1, . . . , 1. During each
crossing, it has a fixed constant probability of hitting γ0 , since this holds for simple random walk, and
the Harnack principle [24] then implies it holds for S ′ . Hence the probability that S ′ reaches B(r0 + 1)
before hitting γ0 is less than (1 − c1 )ℓ−1 for some 0 < c1 < 1. This bounds from above the probability
that x ∈ TL,o and π(x) visits B(r0 2ℓ )c . Assuming that this event occurs, we now bound the conditional
probability that TL,s contains a vertex y
∈ ∂B(r0 2ℓ+1 ). For this, continue Wilson’s algorithm with a walk
Sy0 starting at any y0 ∈ ∂B (r0 2ℓ+1 )4 , followed by walks starting at y1 , . . . , yM , where the latter is an
enumeration of all vertices in ∂B(r0 2ℓ+1 ). Let Fj denote the tree generated by the walks Sx , Sy0 , . . . , Syj .
Denote n S o
y Syj
Ej = σL j < ξFj−1 , j = 0, 1, . . . , M,

where F−1 := π(x). Then on the event {x ∈ TL,o, π(x) ∩ B(r0 2ℓ )c 6= ∅} we have

XM  
µL,o TL,s ∩ ∂B(r0 2ℓ+1 ) 6= ∅ F−1 ≤ E P(Ej | Fj−1 )1E0c . . . 1Ej−1
c .
j=0

Application of Theorem 2.11(ii) yields that the j = 0 term is at most C log (r0 2ℓ+1 )4 / log L when L >
(r0 2ℓ+1 )4 . For 1 ≤ j ≤ M , using Beurling’s estimate (Lemma 2.12), on the event E0c ∩ . . . Ej−1
c
we have


P(Ej | Fj−1 ) ≤ Pyj σB((r0 2ℓ+1 )4 ) < ξF0 max Pw (σL < ξo )
w∈∂B((r0 2ℓ+1 )4 )

≤ C(r0 2ℓ+1 )−1 (log r0 2ℓ+1 )/ log L.

Since M ≤ Cr0 2ℓ+1 , putting the j = 0 and 1 ≤ j ≤ M cases together we get that the sum over 0 ≤ j ≤ M
is bounded by C (log r0 2ℓ+1 )/ log L. Together with the earlier bound on π(x) leaving Bo (r0 2ℓ ) this gives

(1 − c1 )ℓ
P E(x, r0 , ℓ) ≤ C (log r0 2ℓ+1 ) .
log L

30
Substituting into (5.5), and summing over ℓ ≥ ℓ0 implies, for L sufficiently large that
L


νL Rin,(k−1) ≤ r0 , PkL > r0 2ℓ0 ≤ Cr0 (log r0 2ℓ0 ) (1 − c1 )ℓ0 . (5.6)
′
We apply (5.6) with 2ℓ0 = r0β , for some β ′ > 0. The expressions in the right hand sides of (5.6) and (5.4)
are of equal order (up to logarithms), when

log 2
β ′ = −(1 + α′k−1 ) .
log(1 − c1 )
(k)
Since Rin ≤ P (k) , the bounds (5.4) and (5.6) imply, for all large enough L, that
′ ′ −α′k−1
L
νL (Rin,k > r01+β ) ≤ νL (Rin,k−1
L L
> r0 ) + νL (Rin,k−1 ≤ r0 , P (k) > r01+β ) ≤ Ck (log r0 ) r0 .

Hence we get (5.3) for k with a choice of 0 < α′k < α′k−1 /(1 + β ′ ).
We next prove that the out-radius of the k-th last wave is also tight, and satisfies a power law upper
bound. We are going to need the following lemma.
Lemma 5.4. There exist constants C such that the following holds. Let 1 ≤ r < r′ < L, and let
K ⊂ V (L) ∪ {s} be a connected set of edges that contains a path connecting B(r) to s. We have

µL,o (TL,o 6⊂ B(r′ ) | K ⊂ TL,s ) ≤ C r3/2 (r′ )−1/2 .

Proof. Let us use Wilson’s algorithm in the contracted graph GL /K, that is, the edges in K are already
present at the start of the algorithm. We let walks start at {x1 , . . . , xM } = ∂B(r). If TL,o 6⊂ B(r′ ),
then at least one of these walks has to reach ∂B(r′ ) before hitting K. Beurling’s estimate (Lemma 2.12)
implies that for each xj , this has probability at most C (r′ /r)−1/2 . Since M = O(r), the statement
follows.
Proposition 5.5. There exist constants α1 > α2 > · · · > 0 and C1 , C2 , . . . such that
L


lim sup νL Rout,k > r ≤ Ck r−αk , ∀r ≥ 1, ∀k ≥ 1. (5.7)
L→∞

L
In particular, for all 1 ≤ k < ∞ the sequence {Rout,k }L≥1 is tight.
Proof. Fix 1 ≤ k < ∞, and 1 ≤ r0 < ∞. From Proposition 5.3 we have that there exists L0 = L0 (r0 ) < ∞
such that for all L ≥ L0 we have
L −α′
νL (Rin,k > r0 ) ≤ 2 Ck r0 k (5.8)
L
Assume the event {Rin,k ≤ r0 }, which implies that TL,s,k ∩ ∂B(r0 ) 6= ∅. We bound the probability that
L 1+β
Rout,k > r0 , where the parameter β > 0 will be chosen at the end.
Similarly to (5.5), we have:
  X  
L
νL Rin,k L
≤ r0 , Rout,k > r01+β ≤ C (log L) µL,o x ∈ TL,s , TL,o 6⊂ B(r01+β ) . (5.9)
x∈∂B(r0 )

31
Let K be the set of edges in TL,s on the path from x to s. Conditioning on the value K = K, the right
hand side of (5.9) equals
X X  
C (log L) µL,o (x ∈ TL,s ) µL,o (K = K | x ∈ TL,s ) µL,o To 6⊂ B(r01+β ) K = K . (5.10)
x∈∂B(r0 ) K

We have µL,o (x ∈ TL,s ) ≤ C(log r0 )/(log L). Applying Lemma 5.4 to the conditional probability in (5.10)
gives that the expression in (5.10) is at most
3/2 −(1+β)/2
X X 2−β/2
C (log r0 ) r0 r0 µL,o (K = K | x ∈ TL,s ) = C (log r0 ) r0 . (5.11)
x∈∂B(r0 ) K

Choose β so that 2 − β/2 < −α′k , so that (5.11) together with (5.8) gives
−α′k
L
νL (Rout,k > r01+β ) ≤ Ck r0 .
This implies the statement of the proposition with αk = α′k /(1 + β).
The next proposition shows that µL,o (A ⊂ TL,s , B ⊂ TL,o } for fixed finite fixed sets of vertices A and
B, satisfies a certain normalization as L → ∞. Tightness of the in-radius established in Proposition 5.3
will allow us to apply this proposition, and subsequently prove Theorem 5.1. We introduce the notation
qL := µL,o (e 6∈ TL,o ), where e ∼ o. Due to symmetry, qL does not depend on e. In fact, since qL is the
escape probability of random walk from o, we have qL = GL (o, o)−1 . We remark that for the square grid,
the quantity GL (o, o) has an explicit formula:
L
X ′
TL+1 (2 − cos θℓ )
GL (o, o) = , (5.12)
TL+1 (2 − cos θℓ )
ℓ=0

where θℓ = 2π(2ℓ + 1)/(4L + 4), and TL+1 is the degree L + 1 Tchebyshev polynomial. The formula (5.12)
can be derived via contour integration. However, we will not need it, and we omit the proof.
Proposition 5.6. Assume d = 2. Let A, B ⊂ Z2 be disjoint, non-empty finite sets, with o ∈ B. Then
−1
the limit pA,B := limL→∞ qL µL,o (A ⊂ TL,s , B ⊂ TL,o ) exists.
We first need the following lemma. In its statement, a(x) is the potential kernel for simple random
walk on Z2 ; see [24].
Lemma 5.7. Fix x ∈ A.
(i) We have limL→∞ qL −1
µL,o (x 6∈ TL,o) = a(x)
a(e) , where e ∼ o.
(ii) Conditional on x 6∈ TL,o , the law of the path from x to s has a weak limit, as L → ∞.
Proof. (i) We use Wilson’s algorithm in the graph GL,o with a walk starting at x. Considering the limit
of the bounded martingale {a(Sx (t ∧ σV (L) ∧ ξ{o} ))}t≥0 , and using Lemma 3.2 we have



−1 P o 6∈ Sx [0, σV (L) ] a(x)(log L)−1 + o (log L)−1 a(x)
qL µL,o (x 6∈ TL,o ) =
=
= + o(1).
P o 6∈ Se [0, σV (L) ] −1
a(e)(log L) + o (log L) −1 a(e)

(ii) Let Sxh,L denote a random walk conditioned on the event {σL < ξo }, started from x. The path in
TL,s from x to s, conditional on x 6∈ TL,o has the law of LSxh,L [0, σL ]. As L → ∞, Sxh,L converges weakly
to a transient process Sxh (the h-transform of random walk by the potential kernel a(·)). This implies
that LSxh,L [0, σL ] converges weakly to LSxh [0, ∞); see [24, Exercise 11.2].

32
Proof of Proposition 5.6. Let A = {x1 , . . . , xp }, (p ≥ 1) and B = {o, w1 , . . . , wq }, (q ≥ 0). We use
Wilson’s algorithm in the graph GL,o . We start with the vertex x1 , followed by the vertices x2 , . . . , xp ,
followed by the vertices in B. For the rest of the vertices we use an ordering such that their Euclidean
norms form a non-decreasing sequence. Due to Lemma 5.7(i), the probability that the first walk hits s
before o is asymptotic to qL a(x1 )/a(e), as L → ∞. Assuming this event happens, let πxL1 denote the
loop-erasure of the walk starting at x1 , and write πx1 for its weak limit under the conditioning, whose
existence is guaranteed by Lemma 5.7(ii). The probability that the walks starting in (A \ {x1 }) ∪ B
terminate before exiting a ball B(r) of large radius r goes to 1 as r → ∞, and L > r, uniformly in the
path πxL1 . Since these walks determine the event {A ⊂ TL,s , B ⊂ TL,o }, statement (i) follows.
In the proof of Theorem 5.1 we are going to need the following quantitative bound from [10] on the
rate of convergence of νL to ν.
Theorem 5.8 (Theorem 4.1 of [10]). Let d = 2. There exist constants 0 < α < β and C such that if E
is any cylinder event depending only on the configuration in B(ℓ), then

|νL (E) − ν(E)| ≤ Cℓβ L−α .

Proof of Theorem 5.1. (i)–(ii) We showed in Proposition 5.5 that for any fixed 1 ≤ k < ∞, the sequence
(k) (k)
Rout = RL,out , L ≥ 1, is tight. Therefore, we have

lim lim sup νL (N ≥ k, WN −i+1 6⊂ B(ℓ) for some 1 ≤ i ≤ k) = 0. (5.13)

ℓ→∞ L→∞

We establish weak convergence of ξkL . Fix ε > 0, and let ℓ and L0 be such that for all L ≥ L0 the
probability appearing in (5.13) is ≤ ε. Let ζ ∈ R′B(ℓ) be a configuration with the following properties:

(a) (a′o,B(ℓ) )j (ζ) ∈ R′B(ℓ) \ RB(ℓ) for j = 1, . . . , k − 1 and (a′o,B(ℓ) )k (ζ) ∈ RB(ℓ) .
(b) In the stabilization (a′o,B(ℓ) )k (ζ) none of the boundary vertices of B(ℓ) topple;

In other words, ζ is an intermediate configuration in B(ℓ), corresponding to a k-th last wave such that
all of the last k waves stay inside B(ℓ). We first show that


lim νL (ξkL )B(ℓ) = ζ N ≥ k
L→∞

exists for any ζ satisfying (a)–(c).

First observe that the properties of ζ imply that for any η∗ ∈ R′L \ RL , if (η∗ )B(ℓ) = ζ then η∗ is an
intermediate configuration, in the graph GL , corresponding to a k-th last wave, and the last k waves all
stay in B(ℓ). In particular, using that the k-th last wave stays inside B(ℓ), property (5.1) of the bijection
implies that ζ determines a unique pair (A0 , B0 ), with A0 ∪ B0 = B(ℓ), such that whenever η∗ ∈ R′L \ RL
and η∗ |B(ℓ) = ζ, then we have V (To (η∗ )) = B0 . Therefore, using Lemma 2.5 we can write

{η∗ ∈ R′L \ RL : η∗ |B(ℓ) = ζ}

νL N ≥ k, ξkL |B(ℓ) = ζ =
RL (5.14)


= gL (o, o) νL,o η∗ : η∗ |B(ℓ) = ζ .

For any η∗ appearing in the right hand side of (5.14), let η = η∗ |VL \B0 . Due to the burning process, the
conditional distribution of η, given the event {η∗ |B0 = ζ|B0 , V (To (η∗ )) = B0 }, equals that of a recurrent

33
sandpile in the subgraph GB L of GL induced by the set of vertices V (L) ∪ {s} \ B0 (i.e. with closed
0

boundary condition at B0 ). Hence the last expression in (5.14) equals

−1



qL νL,o η∗ : η∗ |B0 = ζB0 , V (To (η∗ )) = B0 νGB0 η : η|A0 = ζ|A0 . (5.15)
L

Since the wired spanning forest in the subgraph of Z2 induced by Z2 \ B0 is one-ended, we can apply [17,
Theorem 3] to deduce that the last factor in (5.15) has a limit as L → ∞. The first factor equals
1
q −1 µL,o (V (TL,o ) = B0 ) , (5.16)
|TB0 | L
where |TB0 | is the number of spanning trees in the graph induced by B0 . Due to Proposition 5.6, the
product of the second and third factors in (5.16) approaches pA0 ,B0 , as L → ∞. This implies the existence
of the limit

lim νL N ≥ k, (ξkL )|B(ℓ) = ζ =: ck (ζ).
L→∞

Summing over all ζ satisfying (a)–(c), and using the choice of ℓ made after (5.13), we get

lim sup νL (N ≥ k) − lim inf νL (N ≥ k) ≤ ε.

L→∞ L→∞

But since the left hand side does not depend on ℓ, we have that the limit limL→∞ νL (N ≥ k) =: bk exists,
proving statement (i). It follows that

ck (ζ)

lim νL (ξkL )B(ℓ) = ζ N ≥ k = =: ρk ξk : (ξk )|B(ℓ) = ζ .
L→∞ bk
Statement (ii) follows immediately from this.
(iii) Observe that the proof of parts (i)–(ii) shows that up to a set of measure 0, the support of ρk can
be partitioned into a countable disjoint union of cylinder sets, such that on each element of the partition,
the stabilization (a′o,Z2 )k (ξk ) takes place within a finite set B(ℓ).
(iv) The countable partition into cylinder sets has the further property that the map ξk 7→ a′o,Z2 (ξk )
is measure preserving on each cylinder set of the partition. Hence the claim follows.
(v) Let ε > 0 be fixed. Let NB(ℓ) denote the number of times o topples if all topplings in B(ℓ) are
carried out, but no site in B(ℓ)c is allowed to topple. Due to the ν-a.s. convergence NB(ℓ) ↑ N , there
exists 1 ≤ ℓ < ∞ such that ν({N = k}∆{NB(ℓ) = k}) < ε, where ∆ denotes symmetric difference. Let
Fℓ,k = {NB(ℓ) = k and some boundary vertex of B(ℓ) topples}.
c
Since {N = k} ∩ Fℓ,k is a cylinder event, we have
c c
ν(N = k) ≥ ν(N = k, Fℓ,k ) = lim νL (N = k, Fℓ,k ) ≥ lim νL (N = k) − ε(ℓ, k),
L→∞ L→∞

where ε(ℓ, k) → 0 as ℓ → ∞, due to (5.13). It follows that bk − bk+1 = limL→∞ νL (N = k) ≤ ν(N = k).
For an inequality in the other direction, we write:
c c
ν(N = k) ≤ ν(N = k, Fℓ,k ) + ν(Fℓ,k ) = ν(NB(ℓ) = k, Fℓ,k ) + ν(Fℓ,k )
c
= lim νL (NB(ℓ) = k, Fℓ,k ) + lim νL (Fℓ,k )
L→∞ L→∞
≤ lim νL (N = k) + lim sup νL (N > k, Fℓ,k ) + lim νL (N = k, Fℓ,k ).
L→∞ L→∞ L→∞

34
Due to (5.13), the third term on the right hand side is at most ε(ℓ, k) → 0 as ℓ → ∞. Therefore, it is
enough to show that
lim lim sup νL (N > k, Fℓ,k ) = 0. (5.17)
ℓ→∞ L→∞
i
We fix 0 < δ < α/β (where α, β are the constants from Theorem 5.8). Let r(i) = Lδ/ρ , i = 0, . . . , k,
where the constant 1 < ρ < ∞ will be chosen later. Recall we denote by η1 , . . . , ηk ∈ R′L \ RL the first k
waves corresponding to η ∈ RL . We define the events
 
the i-th wave W(ηi ) does not topple any vertices in
H(i) = , i = 1, . . . , k.
B(r(i)) after it has reached ∂B(r(i − 1))

Recalling property (5.1) of the bijection, an argument similar to the one made in Proposition 5.3 yields
−i −i
νL (H(i)c ) ≤ C (log L) r(i − 1)−1/4 r(i)9/4 ≤ C (log L) L2δρ L−δρ (ρ−1)/4
. (5.18)

We choose ρ > 9, so that the right hand side of (5.18) goes to 0 as L → ∞. Therefore, in order to prove
(5.17), it is enough to bound


νL (N > k, Fℓ,k , H(1) ∩ · · · ∩ H(k)) ≤ νL N > k, NB(ℓ) = k, H(1) ∩ · · · ∩ H(k) . (5.19)

Suppose now that we are given a configuration η ∈ RL . Let us carry out the first wave up to ∂B(r(0)),
that is, stop the first wave when a vertex of B(r(0))c would need to be toppled, if any. Then carry out the
second wave up to ∂B(r(1)), the third wave up to ∂B(r(2)), etc. Let F ′ denote the event that during the
k-th “partial wave” defined this way, all neighbours of the origin topple. The event in the right hand side
of (5.19) implies the event F ′ . This is because the event {N > k}, in the presence of H(1) ∩ · · · ∩ H(k),
implies that the origin can be toppled a k + 1-st time after the first k partial waves.
Observe that F ′ is measurable with respect to the pile inside B(r(0)), and r(0) = Lδ . Hence, using
Theorem 5.8, and ν({N = k}∆{NB(ℓ) = k}) ≤ ε, the right hand side of (5.19) is at most



νL F ′ , NB(ℓ) = k ≤ ν F ′ , NB(ℓ) = k + CLδβ L−α ≤ ν(F ′ , N = k) + ε + CL−α+βδ
= ε + CL−α+βδ ,

where the last equality follows from the fact that F ′ ⊂ {N > k}. Due to the choice of δ, the second term
goes to 0, as L → ∞. Since ε is arbitrary, we obtain statement (v) of the Theorem.
We can now complete the proof of Theorem 1.7.
Proof of Theorem 1.7. (i) This follows immediately from Theorem 5.1(ii)–(iii).
(ii) This follows from Theorem 5.1(i),(v).
(iii) This follows from Theorem 5.1(ii),(v), since on the event {N = k} we can approximate by cylinder
events on which no vertex topples outside a fixed ball.
The following lemma completes the proof of the d = 2 case of Proposition 3.11.
Lemma 5.9. We have
ν(z ∈ Av) = lim νL (z ∈ Av).
L→∞

35
Proof. For a sufficiently large number k = k(z), we have the deterministic implication

N >k ⇒ z ∈ Av,

both in Z2 , and in VL for L sufficiently large. With such k, we have

k
X
ν(z ∈ Av) = ν(N ≥ k + 1) + ν(N = ℓ, z ∈ Av)
ℓ=1
k
X
= lim νL (N ≥ k + 1) + lim νL (N = ℓ, z ∈ Av) = lim νL (z ∈ Av).
L→∞ L→∞ L→∞
ℓ=1

In the second equality, we applied Theorem 5.1(v) to the first term. In the second term, a.s. finiteness
of the last ℓ waves allows us to approximate {N = ℓ, z ∈ Av} by a cylinder event, and the equality
follows.
Proof of Theorem 1.8. (i) The bound follows immediately from the Proposition 5.5.
(ii) Since on the event {N ≤ k} we have R ≤ max{R1 , . . . , Rk }, this also follows Proposition 5.5.
We now prove Theorem 1.9. The idea of the argument is that, if f (x) := Eν n(o, x) were finite, then
by invariance of ν under ao , f would have to be a bounded harmonic function, hence constant. This is in
contradiction with the structure of the avalanche. We first give two short lemmas on which this argument
will be based.
Lemma 5.10. Assuming ν(N < ∞) = 1, we have ν(R < ∞) = 1.
Proof. This follows easily from Theorem 1.8.
Let ao denote the operation on stable sandpiles on Z2 which maps η to (η + 1o )◦ if a finite stabilization
is possible (i.e., if S < ∞). Then the preceding lemma implies if ν(N < ∞) = 1, there exists a set Ω
with ν(Ω) = 1 such that ao is defined on Ω. Given such an Ω, the next lemma shows that, similarly to
νL , the infinite-volume measure ν is invariant under ao .
Lemma 5.11. Assuming ν(N < ∞) = 1, ν is invariant under ao . That is, for any ν-integrable function
f, Z Z
f (ao η) ν(dη) = f (η) ν(dη) .

Proof. The argument of [15, Prop. 3.14] carries over exactly. There a similar statement is proved in the
case d ≥ 3, but the argument requires only almost sure finiteness of avalanches.
Proof of Theorem 1.9. If ν(N = ∞) > 0, there is nothing to prove. We thus assume that ν(N < ∞) = 1.
We first note that, under this assumption, the infinite-volume addition operators are well-defined, since
n(o, x) ≤ n(o, o).
The invariance statement above makes possible a version of the argument underlying Dhar’s formula
(Lemma 2.2). Assume for the sake of contradiction that Eν N < ∞. In particular, 0 ≤ En(o, x) ≤
En(o, o) < ∞ for all x ∈ Z2 . Since N < ∞, Lemma 5.10 above shows we can (almost surely) write
X
(η + 1o )◦ = η + 1o − n(o, z)∆(z, ·) ,
z∈Z2

36
where the sum above has finitely many nonzero terms and ∆ is the graph Laplacian on Z2 . In particular,
taking expectations and using the invariance above (which implies Eν η(x) = Eν (η + 1o )◦ (x) for all x)
gives X
Eν n(o, z)∆(z, x) = 1o (x) for all x ∈ Z2 . (5.20)
z∈Z2

In other words, (5.20) says that f (x) = Eν n(o, x) is harmonic away from o and has Laplacian 1 at o.
Since f is bounded, recurrence of random walk implies that f is constant. Since n(o, x) ≤ n(o, o) for all
x, we have ν(n(o, x) = n(o, o) for all x ∈ Z2 ) = 1. However, if all vertices topple, the avalanche is infinite,
a contradiction.

6 High-dimensional radius bounds

In this section we prove the bounds for the radius of the avalanche for d ≥ 5. We prove Theorem 1.3 in
Section 6.1. We use the results of Section 6.1 in Sections 6.2 and 6.3 to prove the lower and upper bounds
on the radius.

6.1 Radius bounds on transitive unimodular graphs

See [26, Chapter 8] for background on unimodularity and mass transport.
Proof of Theorem 1.3. (i) We denote Ax (a, b) = Dx (b) \ Dx (a). Consider the following mass transport.
When diam(pastx ) > r, let x send unit mass distributed equally among all vertices y ∈ pastx ∩ Ax (r, 2r).
Let us write sent(x) and get(x),
respectively, for the amount sent and received by x, respectively. Then
E[sent(o)] = µ diam(pasto ) > r . On the other hand, using Jensen’s inequality, we have
X  

1
E[get(o)] = µ o ∈ pastx E o ∈ pastx
|pastx ∩ Ax (r, 2r)|
x∈Ao (r,2r)

2
X µ o ∈ pastx
≥  .
x∈Ao (r,2r) E pastx ∩ Ax (r, 2r) 1o∈past
x

Since pastx ∩ Ax (r, 2r) ⊂ To ∩ Do (4r), the statement follows.

unit mass distributed equally among vertices y ∈ Cx ∩ Ax (r, 4r).
(ii) When diam(Cx ) > 4r, let x send
Then E[sent(o)] = µ diam(Co ) > 4r . On the other hand,
X  

1diam(Cx ;x)>4r
E[get(o)] = µ o ∈ Cx E o ∈ Cx .
|Cx ∩ Ax (r, 4r)|
x∈Ao (r,4r)

and the statement follows.

(iii) Although WSFo is not automorphism invariant, essentially the same mass transport can be used
as in (ii), due to the fact that the rooted random network (To , o) is unimodular, in the sense of [7, 1].
Let T ⊂ V be a finite tree, and x, y ∈ T . Define the function
(
1y∈T |T ∩ Ax (r, 4r)|−1 when diam(T ; x) > 4r;
f (T, x, y) =
0 otherwise.

37
Let x send the following mass to y:
X
F (x, y) = WSFx (Tx = T ) f (T, x, y).
T ∋x,y

Let γ ∈ Γ. It is clear that f (γT, γx, γy) = f (T, x, y), and shifting Wilson’s algorithm by γ shows that
WSFx (Tx = T ) = WSFγx (Tγx = γT ). Therefore, F is invariant under the diagonal action of Γ, and
hence
X X
WSFo (diam(To ) > 4r) = F (o, x) = F (x, o)
x∈V x∈V
X X WSFx (Tx = T ) (6.1)
= .
|T ∩ Ax (r, 4r)|
x∈Ao (r,4r) T ∋x,o
diam(T ;x)>4r

This yields the statement.

Remark 3. Part (iii) of Theorem 1.3 can also be deduced from part (ii) via the following limiting procedure,
the details of which we omit. (This construction was our initial approach to the radius upper bound.)
Let ω ⊂ V be independent site percolation on V with density 0 < p ≪ 1. Conditional on ω, let WSFω
denote the measure on spanning forests of (V, E) where each vertex in ω is “wired to infinity”, in analogy
with WSFo . The measure obtained by averaging WSFω over ω is automorphism invariant. When x ∈ ω,
let x send or not send mass according to the same rule as in part (ii), otherwise let x send no mass. Now
let p ↓ 0. Conditional on o ∈ ω we have WSFω ⇒ WSFo , and the mass transport in part (iii) can be
recovered in this limit.

6.2 Radius lower bound when d ≥ 5

We prove the lower bound using the result of the previous section.
Proof of Theorem 1.2(iv), lower bound. We begin with some terminology. As in [25], let the ‘past of a
vertex x’ in a spanning tree T of GL , be the union of the connected components of T \ {x}, which do not
contain s. We denote this object by pastx (T ). Using the characterization of last waves from Section 2.4
and the bijection for intermediate configurations from Lemma 2.10 we have the following observation:
 
η ∈ RL : R(η) > r ≥ (To , Ts ) ∈ TL,o : R(To ) > r, e ∈/ To for some e ∼ o . (6.2)
Now, consider the map Ψ : TL → TL,o , which modifies a spanning tree T by deleting the unique edge
{o, e} such that e is in the future of o and {o, e} belongs to T . The following properties of the map Ψ
are immediate.
Lemma 6.1. Let U ⊂ TL . Then |Ψ(U )| ≤ |U | ≤ 2d |Ψ(U )|.
We use the lemma with U = {T ∈ TL : R(pasto (T )) > r}, for which we have

Ψ(U ) ⊂ (To , Ts ) ∈ TL,o : R(To ) > r, e ∈
/ To for some e ∼ o .
Therefore equation (6.2) gives that
 
η ∈ RL : R(η) > r |Ψ(U )| 1 |U | 1 T ∈ TL : R(pasto (T )) > r
≥ ≥ ≥ .
|RL | |RL | 2d |RL | 2d |TL |

38
This equation holds for any fixed r, and all L. Using the observation that for fixed r the events at both
ends are cylinder events, and taking the limit as L ր ∞, we have,

1

ν R>r ≥ WSF R(pasto (T)) > r . (6.3)
2d
The proof is completed by applying Theorem 1.3(i). Using the fact that in dimensions d ≥ 5 there is
probability at least c that two independent simple random walks starting at x do not intersect, we deduce
that WSF(x ∈ pasto ) ≥ c|x|2−d . On the other hand, Wilson’s algorithm gives
  X X 
E |To ∩ Bo (4r)| 1o∈pastx ≤ G(o, v) G(y, v) G(v, x) + G(o, x) G(x, v) G(y, v)
y∈Bo (4r) v∈Zd

≤ C r6−d .

Therefore, Theorem 1.3(i) implies WSF(R(pasto ) > r) ≥ cr−2 .

6.3 Radius upper bound when d ≥ 5

In this section we prove the upper bound in Theorem 1.2(iv). Taking d to be ℓ∞ distance, Theorem
1.3(iii) yields

WSFo (diam(To ) > 4r)

X  
1diam(Tx ;x)>4r
= WSFx (o ∈ Tx ) EWSFx x ∈ To
|Tx ∩ Vx (4r) \ Vx (r)|
x∈Zd :r<kxk∞ ≤4r (6.4)
X  
1diam(Tx ;x)>4r
≤ C r2−d EWSFx x ∈ To ,
|Bx |
x∈Zd :r<kxk∞ ≤4r

where we wrote Bx = Tx ∩ Vx (4r) \ Vx (r), and used WSFx (o ∈ Tx ) ≤ G(o, x) ≤ C r2−d . We show that
for δ > 0 there exists C1 = C1 (δ) such that the expectation in the right hand side is bounded above by
C1 (log r)3+δ r−4 . This implies the required upper bound on the tail of the diameter. We are going to
use the following theorem of [5] on the lower tail of the volume of WSF components. Given D ⊂ Zd ,
write WSFDc for the wired spanning forest measure on the contracted graph Zd /Dc .
Theorem 6.2. [5] Let x, y ∈ Zd be such that ky − xk∞ = 4r. Let D ⊂ Zd be such y ∈ ∂D, and
Vx (4r) ⊂ D. Let x ↔ y denote the event that in WSFDc the path from x to Dc reaches Dc via an edge
incident to y. There exist constants C, c > 0 independent of D and r, such that for all λ > 0 we have
 
WSFDc |Tx ∩ Vx (2r) \ Vx (r)| < λr4 x ↔ y ≤ C exp(−cλ−1/3 ).

We use Theorem 6.2 to establish the following regularity estimate. Fix a positive constant δ > 0.
Let us call Tx bad, if diam(Tx ) > 4r, but |Tx ∩ Vx (2r) \ Vx (r)| < (log r)−3−δ r4 . We show the following
lemma.
Lemma 6.3. Let x ∈ Vo (4r) \ Vo (r). We have

WSFx (Tx is bad) ≤ C exp(−c(log r)1+δ/3 ).

39
Proof. If diam(Tx ) > 4r, then there exists y with ky − xk∞ = 4r such that y ∈ Tx . Therefore,
X
WSFx (Tx is bad) ≤ WSFx (y ∈ Tx ) WSFx (Tx is bad | y ∈ Tx ).
y:ky−xk∞ =4r

The term WSFx (y ∈ Tx ) ≤ C r2−d , and there are O(rd−1 ) terms. On the other hand, we have
WSFx (Tx = T | y ∈ Tx ) = WSFy (Tx = T | x ∈ Ty ). This follows from the fact that LERW from
y to x has the same distribution as LERW from x to y, and hence we can use these walks in the first
step of Wilson’s algorithm on the two sides. See [24, Corollary 11.2.2] for ”reversibility” of LERW. Hence
Theorem 6.2 with D = Zd \ {y} implies that

WSFx (Tx is bad | y ∈ Tx ) = WSFy (Tx is bad | x ∈ Ty ) ≤ C exp(−c (log r)1+δ/3 ).

The statement of the lemma follows.

Lemma 6.4. We have " #
1diam(Tx ;x)>4r (log r)3+δ
EWSFx o ∈ Tx ≤ C .
|Bx | r4

Proof. When Tx is bad, we use that |Bx | ≥ 1, and hence due to Lemma 6.3 we have
" #
1diam(Tx ;x)>4r 1Tx is bad
EWSFx o ∈ Tx ≤ WSFx (Tx is bad, o ∈ Tx ) C rd−2
|Bx |
≤ WSFx (Tx is bad) C rd−2
≤ C exp(−c(log r)1+δ/3 )rd−2 = o(r−4 ).

Therefore it is enough to consider the contribution when Tx is not bad. When this occurs, we have
|Bx | ≥ (log r)−3−δ r4 , and hence
" #
1diam(Tx ;x)>4r 1Tx is not bad (log r)3+δ
EWSFx o ∈ Tx ≤ .
|Bx | r4

This proves the claim.

We can now complete the proof of the upper bound in Theorem 1.2(iv). Lemma 6.4 implies that the
right hand side of (6.4) is at most

C rd r2−d (log r)3+δ r−4 = C (log r)3+δ r−2 .

This completes the proof.

7 Bounds on the size

We now prove Theorem 1.5. Sections 7.1, 7.2 and 7.3 consider low dimensions, and Sections 7.4, 7.5 and
7.6 the case d ≥ 5.

40
7.1 Upper bounds on the size when d = 3, 4
Proof of Theorem 1.5(ii)–(iii), upper bounds. The claimed upper bounds on ν(|Av| ≥ t) follow immedi-
ately from Theorem 1.2, via the trivial estimate ν(|Av| ≥ t) ≤ ν(R > c(d) t1/d ). In order to obtain an
upper bound on the tail of S, fix some k ≥ 1. Recalling that N denotes the number of waves, we have

ν(S > t) = ν(S > t, N > k) + ν(S > t, N ≤ k). (7.1)

Recalling that d ≥ 3, we can upper bound the first term in the expression above by

ν(N > k) ≤ Eν N k −1 ≤ C k −1 .

Next, writing S j for the size of the j-th wave, if S > t and N ≤ k, then we have S j > t/k for some
1 ≤ j ≤ N . Hence,

ν(S > t) ≤ C k −1 + ν(S j > t/k, for some j ≤ N )

≤ C k −1 + C ν(R > c(d)(t/k)1/d ).

Due to Theorem 1.2(ii)–(iii), we get the bounds:

(
C k −1 + C t−1/18 k 1/18 when d = 3;
ν(S > t) ≤
C k −1 + C t−1/16 k 1/16 when d = 4.

Optimizing the choice of k yields:

(
C t−1/19 when d = 3;
ν(S > t) ≤
C t−1/17 when d = 4.

7.2 Lower bounds on the size when 2 ≤ d ≤ 4

We fix z = r e1 , where r = t1/d , and write u = kzk/10. Recall that in the course of the proof of Lemma
3.3 in Section 3, we showed that for L ≥ 100 kzk, and e a neighbour of the origin, and with π = LSe [0, σL ],
we have (
  c P(Γz,L ) log |z| when d = 2;
Sz Sz
P π ∩ Vz (u) = ∅, ξo < ξπ ≥ (7.2)
(2d)−1 P(Γz,L ) when d = 3, 4.
Let us write Fz,L for the event in the left hand side of (7.2). Our goal will be to show that conditional
on the event Fz,L (when z is in the last wave), a large number of vertices in Vz (u) are also in the last
wave, with probability bounded away from 0. We will use a second moment argument to prove this, that
is based on Proposition 7.1 below.
Let u′ = (1 − ε) u, where we are going to choose 0 < ε < 1/4 later. Consider the following partial
cycle popping in GL , defined in three stages; see [35] and [26] for background on cycle popping. In the
first stage, reveal the LERW π started at e and ending at s. In the second stage, reveal a LERW started
at z, ending on hitting π ∪ {o}. In the third stage, pop all cycles that are entirely contained in Vz (u) that

41
can be popped. Condition on the event Fz,L , that is measurable with respect to the result of the first two
stages. Let π ′ = LSz [0, ξo ], and let π ′ (u′ ) be the portion of π ′ from z to the first exit from Vz (u′ ). Let

I(x) = {stage three reveals a path from x to π ′ ∩ Vz (u)}, x ∈ Vz (u/2).

For each x ∈ Vz (u/2) for which I(x) occurs, let p(x) ∈ π ′ ∩ Vz (u) be the point where the revealed path
first meets π ′ ∩ Vz (u). For technical reasons (that are only required for our argument when d = 4), we
also define:
n o
S Sp(x)
J(x) = ξπ′p(x)
(u )
′ < σVz (u) .

Let X
Y = Yu,ε = 1I(x) 1J(x) .
x∈Vz (u/2)

The following proposition states bounds on the first and second moments of Y .
Proposition 7.1.
There exist 0 < ε < 1/4, c1 > 0 and C such that the following hold.
(i) When d = 2, we have

E(Yu,ε | Fz,L ) ≥ c1 u2 and 2

E(Yu,ε |, Fz,L ) ≤ C u4 .

(ii) When d = 3, we have

E(Yu,ε | Fz,L ) ≥ c1 u3 and 2

E(Yu,ε | Fz,L ) ≤ C u6 .

(iii) When d = 4, we have

E(Yu,ε | Fz,L ) ≥ c1 u4 (log u)−1 and 2

E(Yu,ε |, Fz,L ) ≤ C u8 (log u)−2 .

Proof of Theorem 1.5(i)–(iii); lower bounds; assuming Proposition 7.1.

Since on the event Fz,L , we have π ′ ⊂ TL,o, we have |TL,o | ≥ Y . Hence for any t′ > 0 we have

µL,o (z ∈ TL,o , e 6∈ TL,o, |TL,o | ≥ t′ )

νL (|Av| ≥ t′ ) ≥ (2d)−1
µL,o (e 6∈ TL,o )
(7.3)
P(Fz,L )
≥ (2d)−1 P(Y ≥ t′ | Fz,L ).
Pe (σL < ξo )

Let us choose t′ = (1/2) c1 u2 in d = 2; t′ = (1/2) c1 u3 in d = 3, and t′ = (1/2) c1 u4 (log u)−1 in d = 4.

The Paley-Zygmund inequality implies that

1 c21
P(Y ≥ t′ | Fz,L ) ≥ . (7.4)
4C
Letting L → ∞, we obtain the required lower bounds from (7.3), (7.4), (7.2), Lemma 3.9, and the
dimension-dependent estimates in Sections 3.3.1–3.3.3.

42
Proof of Proposition 7.1(i),(ii).
The upper bounds are immediate from Yu,ε ≤ |Vz (u/2)|.
For the lower bounds, using the strong Markov property of Sx at time ξπ′ , when Sx is at the point
p(x), we have



P(I(x) ∩ J(x) | Fz,L ) = Px ξπ′ ≤ ξπ′ (u′ ) < σVz (u) = Px ξπ′ (u′ ) < σVz (u) .

Let π ′′ := LSz [0, σVz (u) ], and let π ′′ (u′ ) be the portion of π ′′ up to its first exit from Vz (u′ ). Due to
Lemma 3.4, there exists c2 = c2 (ε), such that the distribution of π ′ (u′ ) is bounded below by c2 times the
distribution of π ′′ (u′ ). This implies



Px ξπ′ (u′ ) < σVz (u) ≥ c2 Px ξπ′′ (u′ ) < σVz (u) . (7.5)

We lower bound the probability in the right hand side of (7.5) separately in d = 2, 3.
When d = 2, we have


Px ξπ′′ (u′ ) < σVz (u) ≥ Px (Sx completes a loop around z before exiting Vz (u′ )) ≥ c
with some constant c > 0. This follows from the invariance principle; see for example [24, Exercise 3.4].
Summing over x ∈ Vz (u/2) yields the required lower bound.
When d = 3, we have




Px ξπ′′ (u′ ) < σVz (u) ≥ Px ξπ′′ < σVz (u) − Px σVz (u′ ) ≤ ξπ′′ \π′′ (u′ ) < σVz (u) . (7.6)

A result of Lyons, Peres and Schramm [27, Lemma 1.2] states that for two independent copies of the
same transient Markov chain, the probability for one path to intersect the loop-erasure of the other is at
least a universal constant c3 > 0 times the probability that the Markov chain paths themselves intersect.
Applying this to the random walks Sz [0, σVz (u) ] and Sx [0, σVz (u) ], we have



Px ξπ′′ < σVz (u) ≥ c3 P Sx [0, σVz (u) ] ∩ Sz [0, σVz (u) ] 6= ∅ . (7.7)
The right hand side in (7.7) is bounded below by a constant c3 > 0, independent of u. This can be seen
by arguments due to Lawler; by adapting the proof of [23, Theorem 3.3.2].
It remains to bound the negative term in (7.6). For this we write



Px σVz (u′ ) ≤ ξπ′′ \π′′ (u′ ) < σVz (u) ≤ P Sx [σVz (u′ ) , σVz (u) ] ∩ Sz [σVz (u′ ) , σVz (u) ] 6= ∅ . (7.8)
Consider independent Brownian motions, starting at x/u and z/u. Since the paths are continuous, and
with probability 1 they exit the cube (z/u) + [−1, 1]3 at different points, the invariance principle implies
that the probability in the right hand side of (7.8) goes to 0 uniformly in u ≥ (1/ε), as ε → 0. Therefore,
we can fix ε > 0 such that the right hand side of (7.8) is at most c3 /2, uniformly in u. With such a choice
of ε, the first moment is bounded below by c u3 .

7.3 Proof of the moment bounds in d = 4

7.3.1 Proof of the first moment bounds
We begin with the first moment lower bound – that is, the lower bound in Proposition 7.1(iii). The line
of reasoning leading to (7.6) and (7.7) above holds also for d = 4, and we take these as our starting point.
We next establish an appropriate analogue of the constant lower bound given above for (7.6). We restrict
to x ∈
/ Vz (u/4) for simplicity.

43
Lemma 7.2. There is a c > 0 such that, uniformly in z and x ∈ Vz (u/2) \ Vz (u/4),


P Sx [0, σVz (u) ] ∩ Sz [0, σVz (u) ] 6= ∅ ≥ c/ log u .
Proof. Fix such an x, and consider the number of intersections
σVz (u) σVz (u)
X X
Jx := 1Sx (k)=Sz (ℓ) . (7.9)
k=0 ℓ=0

Taking expectations and using Theorem 2.11 gives a c > 0 such that EJx ≥ c.
On the other hand, EJx2 is of order at most log u. By a computation similar to [23, Theorem 3.3.2;
lower bounds], we have
X  
EJx2 ≤ 2 G(x, y1 )G(z, y1 )G(y1 , y2 )2 + G(x, y1 )G(z, y2 )G(y1 , y2 )2 .
y1 , y2 ∈Vz (u)

Each term above gives a contribution of order log u; we discuss in detail only the first term. By summing
first over y2 and using Theorem 2.11(ii) (taking the n → ∞ limit in this theorem), we get an upper bound
of order X
log u G(x, y1 )G(z, y1 ) .
y1 ∈Vz (u)

For each y1 , either |x − y1 | or |z − y1 | is at least u/8 since |x − z| is at least u/4. Thus either G(x, y1 ) or
G(z, y1 ) is at most Cu−2 . Summing the other factor over y1 gives a factor of order u2 , giving the required
upper bound.
Using the second moment method and noting that Sx and Sz intersect if Jx > 0 completes the
proof.
It remains to control the negative term of (7.6). This will be accomplished using the following lemma:
Lemma 7.3. There exists a constant C3 > 0 such that, uniformly in 0 < ε < 1/2 and u large (how large
depends on ε), and uniformly in y ∈ Vz (u) \ Vz (u′ ),


P Sy [0, σVz (u) ] ∩ Sz [0, σVz (u) ] 6= ∅ ≤ C3 ε/ log u .
Proof of Proposition 7.1(iii), lower bound; assuming Lemma 7.3. Note that



Px σVz (u′ ) ≤ ξπ′′ < σVz (u) ≤ P Sx [σVz (u′ ) , σVz (u) ] ∩ Sz [0, σVz (u) ] 6= ∅ (7.10)


≤ sup P Sy [0, σVz (u) ] ∩ Sz [0, σVz (u) ] 6= ∅ .
y∈∂Vz (u′ )

The above, combined with Lemma 7.3, allows the choice of an appropriately small ε to give a uniform
lower bound of cu4 / log u for the right-hand side of (7.6), completing the proof of the first moment of
Prop. 7.1(iii).
We turn to the proof of Lemma 7.3, which is an adaptation of the proof of [23, Theorem 3.3.2; d = 4
upper bound]. Let u′′ = (1 + ε)u. To avoid complications introduced by intersections near the boundary,
consider the extended number of intersections
σVx (u′′ ) σVz (u′′ )
X X
Jx′ := 1Sx (k)=Sz (ℓ) , x ∈ Vz (u) .
k=0 ℓ=0

44
Lemma 7.4. There is C such that, uniformly in ε < 1/2 and z, and in x ∈ Vz (u) \ Vz (u′ ), we have

EJx′ ≤ C ε .

Proof. We have
X
EJx′ ≤ G(z, y) GVz (u′′ ) (x, y)
y∈Vz (u′′ )
X X
= G(z, y) GVz (u′′ ) (x, y) + G(z, y) GVz (u′′ ) (x, y) . (7.11)
y∈Vx (u/10) y∈Vz (u′′ )\Vx (u/10)

Let y ∈ Vz (u′′ ), and write kx − yk = r. A gambler’s ruin estimate yields

n εu o
GVz (u′′ ) (x, y) ≤ Px σVx (r/2) < σVz (u′′ ) sup G(a, y) ≤ C min , 1 r−2 . (7.12)
a∈Vx (r/2) r

Consider the first term of (7.11). Using kz − yk ≥ u/2 and (7.12), this term is at most
 
εu 
X  u/10
X
C  1 εu 3 1 
r3 · 2 + · r · 2 ≤ Cε ,
u2 r=1 r r=εu
r r
P
The second term, using (7.12) again, is bounded by C (εu/u3 ) y∈Vz (u′′ ) G(y, z) ≤ Cε.

Recall the definition of Jx from (7.9). We show that, conditional on {Jx > 0}, the expectation of Jx′
is at least c log u. This gives the desired upper bound for P(Jx > 0).
Lemma 7.5. We can find r > 0 such that, uniformly in 0 < ε < 1/2, u such that εu > u1/2 , and
x ∈ ∂Vz (u′ ) such that x is at least distance u/10 from all but one face of Vz (u), we have

E[Jx′ | Jx > 0] ≥ r log u . (7.13)

Proof. We follow a similar argument to the proofs of [24, Proposition 10.1.1] and [23, Theorem 3.3.2;
d = 4 upper bound]. On {Jx > 0}, there is a lexicographically first intersection in Vz (u). Specifically,
we can define ℓ1 := inf{j : Sx (j) ∈ Sz [0, σVz (u) ] ∩ Vz (u)} and ℓ2 := inf{j : Sz (j) = Sx (ℓ1 )} and note
that each ℓi is smaller than σVz (u) . Using the strong Markov property of Sx at time ℓ1 , conditionally on
Sz [0, σVz (u) ] and Sx [0, ℓ1 ] the expected value of Jx′ is bounded below by
Sz
σV (u′′ )
  X
z

E Jx′ | Sz [0, σVz (u) ], Sx [0, ℓ1 ] ≥ GVz (u′′ ) (Sz (ℓ2 ), Sz (i))
i=ℓ2
σVz (√u) σV (√u)
X d
X
≥c G(z, Sz (i)) = c G(o, So (i)) .
i=0 i=0

Here we have used the fact that GVz (u′′ ) (a, Sz (ℓ2 )) ≥ cG(a, Sz (ℓ2 )) for a ∈ VSz (ℓ2 ) (u1/2 ) along with
d
translation invariance, and = denotes equality in distribution.
The conclusion of Lemma 7.5 follows immediately from the above, using the following proposition
(along with an a priori power law lower bound for P(Jx > 0)):

45
Proposition 7.6 ([24, Lemma 10.1.2]). For every α > 0, there exist c, r such that for all n sufficiently
large,  
σX
n −1

P G(o, So (j)) ≤ r log n ≤ cn−α .

j=0

Proof of Lemma 7.3. Comparing Lemma 7.5, and Lemma 7.4, the claim nearly follows, except that y ∈
∂Vz (u′ ) in (7.10) may be closer than distance u/10 to more than one face of Vz (u). However, for such y
we can replace Vz (u) by a larger box V ′ ⊃ Vz (u), whose diameter is still of order u, in such a way that
y is at distance εu from the boundary of V ′ , and y is bounded away from the corners of V ′ . Since in V ′
the intersection probability is larger than in Vz (u), the claim follows.
The above completes the proof of the lower bound in Proposition 7.1(iii).

7.3.2 Proof of the second moment bounds

Here we prove the second moment bound in Proposition 7.1 (iii). We first introduce some notation. We
Q4
will work with dyadic cubes i=1 [ai 2k , (ai + 1)2k ) ∩ Z4 , where a1 , a2 , a3 , a4 ∈ Z, and k ≥ 1. We say
that such a cube is of scale k. If Q is dyadic cube, we denote by Q′ , respectively, Q′′ , the cubes that are
concentric with Q and have twice, respectively, four times, the side-length. Given v ∈ Z4 , we denote by
Q(v; k) the unique dyadic cube of scale k containing v.
The following inequality recasts the statement of Lemma 7.3. Let Q be a dyadic cube of scale k,
p ∈ Q, q ∈ Q′′ , such that dist∞ (q, ∂Q′′ ) = ε2k . There is a constant C such that
Cε
P (Sq [0, σQ′′ ] ∩ Sp [0, ∞) 6= ∅) ≤ . (7.14)
k
Fix x, y ∈ Bz (u/2) and assume that I(x) ∩ J(x) ∩ I(y) ∩ J(y) occurs. We distinguish the following two
cases:
(I) the paths from x and y to π ′ do not meet;
(II) the paths from x and y to π ′ meet at a vertex q(x, y) 6∈ π ′ .
In Case (I), using that the events J(x) and J(y) occur, let v, w ∈ π ′ (u′ ), respectively, be the points where
Sp(x) and Sp(y) , respectively, first hit π ′ (u′ ). (Note that v and w may coincide.) It follows that
 
S
P(Case (I)) ≤ P ∃v, w ∈ π ′ (u′ ) : ξvSx < σVSzx(u) , ξw
Sy
< σVzy(u) .

Due to Lemma 3.4, there exists C1 = C1 (ε), such that the distribution of π ′ (u′ ) is bounded above by C1
times the distribution of π ′′ (u′ ). Therefore, we have
 
S
P(Case (I)) ≤ C1 P ∃v, w ∈ π ′′ (u′ ) : ξvSx < σVSzx(u) , ξw
Sy
< σVzy(u) .

In Case (II), we let v = q(x, y), and noting p(x) = p(y), let w be the point where Sp(x) first hits π ′ (u′ ).
Again bounding above by π ′′ (u′ ), it follows that
 
S
Sx
P(Case (II)) ≤ C1 P ∃w ∈ π ′′ (u′ ), v ∈ Vz (u) : ξvSx < ξw < σVSzx(u) , ξvSy < σVzy(u) .

46
We bound the probabilities of Cases (I) and (II) separately. The idea of the bound is not to sum over
v and w, but rather, sum over the choice of suitable dyadic cubes that v and w fall into, and use the
bound (7.14) for the probability of random walk intersections. Throughout, we write K for the integer
such that 2K−1 < u ≤ 2K .
Case (I). We may assume without loss of generality that the walk Sz generating π ′′ hits v before w,
as the other case follows by a symmetric argument. For convenience, we assume that kz − vk, kv − wk,
kx − vk, ky − wk are all at least 32. At the end of the proof we comment on how to handle the remaining
configurations of points. We define the following dyadic scales and cubes:

kv := max k ≥ 1 : 2k+4 ≤ min{kv − zk∞ , kv − wk∞ , kv − xk∞ }

kw := max k ≥ 1 : 2k+4 ≤ min{kw − vk∞ , kw − yk∞ }

kvw := max k ≥ 1 : 2k+4 ≤ kv − wk∞
Q(v) := Q(v; kv ) Q(w) = Q(w; kw ).

We also let
 
kz := max k ≥ 1 : 2k+4 ≤ kz − vk∞ kx := max k ≥ 1 : 2k+4 ≤ kx − vk∞
 (7.15)
ky := max k ≥ 1 : 2k+4 ≤ ky − wk∞ .

A sketch of the argument is as follows: the walks Sz and Sx both have to hit Q′ (v), and then they
intersect at a point of Q(v). Following the intersection, the walk Sz has to hit Q′ (w), and so does the
walk Sy . These two walks then intersect at a point of Q(w). Breaking up the paths into pieces, the
various hitting and intersection events will give us the estimate:

2
2
2
2
2kv 2kv
1 2kw 2kw 1
C 2 2 log 2kv kw − vk2 2 log 2kw . (7.16)
kz k
(2 ) (2 )
x k
∞ (2 )
y

We need to sum this estimate over the choices of x and y, and the choices of the boxes Q(v) and Q(w).
In the summation we will need to distinguish a number of sub-cases according to the relative sizes of the
scales kv , kw , kz , kx , ky .
We first establish the bound in (7.16). This is provided by the following lemma.

Lemma 7.7. (Probability bound for Case (I)) Let R1 and R2 be dyadic boxes of scales k1 and k2 ,
and let x, y ∈ Bz (u/2) be points such that:
(i) R1′′ and R2′′ are disjoint;
(ii) dist(z, R1′′ ), dist(x, R1′′ ) ≥ 2k1 ;
(iii) dist(y, R2′′ ) ≥ 2k2 .
Define kx′ , ky′ , kz′ by the formulas (7.15) where Q(v) and Q(w) are replaced by R1 and R2 , respectively.
Then

2 k
2
2
2
2k1 2 1 1 2k2 2k2 1
P [∃v ∈ R1 , ∃w ∈ R2 s.t. Case (I)] ≤ C ′ 2 ′ 2 ′
2
. (7.17)
(2kz ) (2kx ) k1 dist∞ (R1 , R2 )2 2ky k2

Proof. We need to be careful about the event when the walk Sz first hits R1′ , leaves R1′′ and returns,
before intersecting the path of Sx . The following definitions take care of this possibility by introducing

47
the variables ℓ1 and ℓ2 that count crossings from ∂R1′′ to ∂R1′ and from ∂R2′′ to ∂R2′ , respectively. The
definitions are somewhat tedious to write down; however, estimating the resulting probabilities is then
straightforward using the strong Markov property. Given ℓ1 , ℓ2 ≥ 0, let
Sz ′′ ′
Tℓ1 = inf{n ≥ ξR ′ : Sz [ξR′ , n] has made at least ℓ1 crossings from ∂R1 to R1 }
1
1

σℓ1 ,R′′1 = inf{n ≥ Tℓ1 : Sz (n) 6∈ R1′′ }

ξℓ1 ,R′2 = inf{n ≥ σℓ1 ,R′′1 : Sz ∈ R2′ }
Tℓ1 ,ℓ2 = inf{n ≥ ξℓ1 ,R′2 : Sz [ξℓ1 ,R′2 , n] has made at least ℓ2 crossings from ∂R2′′ to R2′ }
σℓ1 ,ℓ2 ,R′′2 = inf{n ≥ Tℓ1 ,ℓ2 : Sz (n) 6∈ R2′′ }.

On the event in the left hand side of (7.17), the following events occur for some integers ℓ1 , ℓ2 ≥ 0:
z
Sz
(i) ξR ′ < ∞ (v) ξℓS1 ,R′2 < ∞
1
Sx S
(ii) ξR 1
<∞ (vi) ξRy2 < ∞
(iii) T ℓ1 < ∞ (vii) Tℓ1 ,ℓ2 < ∞
(iv) Sz [Tℓ1 , σℓ1 ,R′′1 ] ∩ Sx [ξR1 , ∞) 6= ∅ (viii) Sz [Tℓ1 ,ℓ2 , σℓ1 ,ℓ2 ,R′′2 ] ∩ Sy [ξR2 , ∞) 6= ∅

We bound the probability that (i)–(viii) occur, with each estimate conditional on the previous ones.
′ ′
The probability of (i)–(ii) is bounded by C (2k1 /2kz )2 (2k1 /2kx )2 , since d = 4. Using the strong Markov
property of Sz at times ξ R1 , T1 , . . . , Tℓ1 −1 , we have that (iii) occurs with conditional probability ≤ cℓ11
′

with some 0 < c1 < 1. Since Sz (Tℓ1 ) ∈ ∂R1′ and Sx (ξR1 ) ∈ ∂R1′′ are at distance of order 2k1 from each
other, the conditional probability of (iv) is bounded by C/(log 2k1 ) = C ′ /k1 . The probability of (v)-(vi)
′
is bounded by C (2k2 /dist∞ (R1 , R2 ))2 (2k2 /2ky )2 . The probability of (vii) is bounded by cℓ12 . Finally, the
probability of (viii) is bounded by C/k2 , again due to (7.14). Multiplying the bounds and summing over
1 ≤ ℓ1 , ℓ2 < ∞ yields the lemma.
We continue with the bound for Case (I). We break up Case (I) into the following sub-cases (that
partially overlap, but together cover all possibilities):

(I-1) kv < kz , kx and kw < ky ; (I-4) kv = kz ≤ kx and kw = ky ;

(I-2) kv < kz , kx and kw = ky ; (I-5) kv = kx ≤ kz and kw < ky ;
(I-3) kv = kz ≤ kx and kw < ky ; (I-6) kv = kx ≤ kz and kw = ky .

Fixing the scales kv , kw , kz , kx , ky , we bound the number of choices of x, y and the dyadic boxes containing
v and w in each case separately, and apply Lemma 7.7. Then we sum over the scales allowed in each
sub-case. A depiction of case (I-1) may be found in Figure 2.
Sub-case (I-1). The number of choices for Q(v) is of order 24kz /24kv . The number of choices for x is
of order 24kx . Given Q(v), the number of choices for Q(w) is O(1) (note that kw = kv ), and the number
of choices for y is of order 24ky . Mutiplying these bound together, and applying Lemma 7.7 to Sub-case
(I-1), we get the estimate:

24kz 4kx 4ky 22kv 22kv 1 22kv 22kv 1 2kz 2kx 2ky 2
2kv
2 2 = 2 2 2 .
24kv 22kz 22kx kv 22kv 22ky kv kv2

48
 x
Figure 2: Depiction of sub-case (I-1). The walk Sz is depicted
as a solid line; walks from x and y are dashed. Sx and Sz
intersect at v, which is surrounded by boxes Q(v), Q′ (v), and
v Q′′ (v). Walks Sz and Sy intersect at w (boxes not shown).
z Note that in this case, the factor limiting the size of Q(v) is the
proximity of the vertex w. After intersection, Sz terminates
w upon intersecting the sink s (i.e., the boundary of the large
square).

Summing this bound for fixed kv over kx , ky , kz such that kv < kx , ky , kz ≤ K, and then over 1 ≤ kv ≤ K,
we get
K
8
X
2kv 2K u8
2K 3 2
2 ≤ C = C .
kv2 K2 (log u)2
kv =1

Sub-case (I-2). The number of choices for Q(v) is of order 24kz /24kv . The number of choices for x
is of order 24kx . Given Q(v), the number of choices for Q(w) is of order 24kv /24kw , and the number of
choices for y is of order 24kw . Lemma 7.7 now gives:

24kz 4kx 24kv 4kw 22kv 22kv 1 22kw 1 2kz 2kx 2

2kv 2kw
2
4k
2 4k
2 2k 2k 2k
= 2 2 .
2 v 2 w 2 z 2 x kv 2 v kw kv kw
We sum over kv < kx , kz ≤ K for fixed 1 ≤ kw ≤ kv , then over 1 ≤ kw ≤ kv ≤ K. This yields
Cu8 /(log u)2 and completes the estimate in Sub-case (I-2).
The other four sub-cases are handled similarly, and we only state the number of choices, the probability
estimate, and the range of summations over the scales.

Choices Probability Summed over

2kv 2kw
2 1 2 1 kx such that kx ≥ kv ; ky such that ky > kw ; and
I-3 24kx 24ky
2 2kx kv 2 2ky kw kv , kw such that 1 ≤ kv ≤ kw ≤ K

24kvw 4kw 22kv 1 22kw 1 kx such that kx ≥ kv ; kvw such that kvw ≥ kw ; and
I-4 24kx 2
24kw 22kx kv 22kvw kw kv , kw such that 1 ≤ kv , kw ≤ K

24kz 4kv 4ky 22kv 1 22kw 1 kz such that kz ≥ kv ; ky such that ky > kw ; and
I-5 2 2
24kv 22kz kv 22ky kw kv , kw such that 1 ≤ kv ≤ kw ≤ K

24kz 4kv 24kvw 4kw 22kv 1 22kw 1 kz such that kz ≥ kv ; kvw such that kvw ≥ kw ; and
I-6 2 2
24kv 24kw 22kz kv 22kvw kw kv , kw such that 1 ≤ kv , kw ≤ K

49
 Case (II). We will use notation similar to Case (I), but with somewhat different meaning. Let

kv := max k ≥ 1 : 2k+4 ≤ min{kv − xk∞ , kv − yk∞ , kv − wk∞ }

kw := max k ≥ 1 : 2k+4 ≤ min{kw − vk∞ , kw − zk∞ }

kvw := max k ≥ 1 : 2k+4 ≤ kv − wk∞
Q(v) := Q(v; kv ) Q(w) = Q(w; kw ).

We also let

kz := max k ≥ 1 : 2k+4 ≤ kz − wk∞

kx := max k ≥ 1 : 2k+4 ≤ kx − vk∞ (7.18)

ky := max k ≥ 1 : 2k+4 ≤ ky − vk∞ .

The following lemma provides the probability bound in Case (II), and is proved similarly to Lemma 7.7.
Lemma 7.8. (Probability bound for Case (II)) Let R1 and R2 be dyadic boxes of scales k1 and k2 ,
and let x, y ∈ Bz (u/2) be points such that:
(i) R1′′ and R2′′ are disjoint;
(ii) dist(x, R1′′ ), dist(y, R1′′ ) ≥ 2k1 ;
(iii) dist(z, R2′′ ) ≥ 2k2 .
Define kx′ , ky′ , kz′ by the formulas (7.18) where Q(v) and Q(w) are replaced by R1 and R2 , respectively.
Then

2 k
2
2
2
2k1 2 1 1 2k2 2k2 1
P [∃v ∈ R1 , ∃w ∈ R2 s.t. Case (II)] ≤ C ′ 2 ′
2 2 ′ 2
. (7.19)
(2 x ) 2 y k1 dist∞ (R1 , R2 ) (2 z ) k2
k k k

We now list the sub-cases to be considered. These are:

(II-1) kv < kx , ky and kw < kz ; (II-3) kv = kx ≤ ky and kw < kz ;

(II-2) kv < kx , ky and kw = kz ; (II-4) kv = kx ≤ ky and kw = kz .

Interchanging the roles of x and y in (II-3) and (II-4) yields the remaining configurations not covered.

Choices Probability Summed over

4kz 2kv 2kv 2kv
2 2 2 1 2 1 kx , ky , kz > kv for fixed kv ;
II-1 24kx 24ky 2k
24kv 2 2kx 2 y kv 2 2kz kv and then over 1 ≤ kv ≤ K

22kv 22kv 1 22kw 1 kx , ky > kv for fixed kv , kw ;

II-2 24kx 24ky
22kx 22ky kv 22kv kw and then over 1 ≤ kw ≤ kv ≤ K

24kz 24kw 4kv 4ky 22kv 1 22kw 1 ky > kv and kz > kw for fixed kv , kw ;
II-3 2 2
24kw 24kv 22ky kv 22kz kw and then over 1 ≤ kv ≤ kw ≤ K

24kvw 4kv 4ky 22kv 1 22kw 1 ky > kv and kvw > kw for fixed kv , kw ;
II-4 2 2
24kv 22ky kv 22kvw kw and then over 1 ≤ kv , kw ≤ K

50
 This completes the analysis of Case (II).
It remains to comment on configurations where one of the ℓ∞ distances is < 32. In these cases, we
can replace the box Q(v) and/or Q(w) by the point v and/or w itself, and omit the random variables Tℓ1 ,
etc. The combinatorial bounds, as well as the probability bounds still hold with kv = 1 and/or kw = 1,
and this completes the proof of the upper bound in Proposition 7.1 (iii).

7.4 The size of the past in invariant forests

Proof of Theorem 1.6. Recall that Ax (a, b) = Dx (b) \ Dx (a). When |pastx ∩ Dx (2r)| > t, let x send mass

1 distributed equally among the vertices z ∈ pastx ∩ Ax (r, (3/2)r). Then E[sent(o)] ≤ µ |pasto | > t .
On the other hand, using Jensen’s inequality, we have
" # " #
X 1{o∈pastx } 1|pastx ∩Dx (2r)|>t X 1{o∈pastx } 1|Te o (r/2)|>t
E[get(o)] ≥ E ≥ E
|pastx ∩ Ax (r, (3/2)r)| |To (4r)|
x∈Ao (r,(3/2)r) x∈Ao (r,(3/2)r)


X µ o ∈ pastx , |T e o (r/2)| > t
≥ h i
e
x∈Ao (r,(3/2)r) E |To (4r)| o ∈ pastx , |To (r/2)| > t


X µ o ∈ pastx , |T e o (r/2)| > t 2
≥ h i .
x∈Ao (r,(3/2)r) E |To (4r)| 1{o∈pastx }
(7.20)

This completes the proof.

7.5 Lower bounds: d ≥ 5

We now apply Theorem 1.6 to the measure WSF.
Proof of Theorem 1.5(iv); lower bound. Let t = δ r4 , where δ > 0 will be chosen in course of the proof.
Wilson’s algorithm yields:
  X X
E |To (4r)|1{o∈pastx } ≤ WSF(o ∈ pastu , u ∈ pastx , y ∈ pastu )
y∈Bo (4r) u∈Zd

+ WSF(o ∈ pastx , x ∈ pastu , y ∈ pastu )
X X  (7.21)
≤ G(o, u) G(u, x) G(y, u) + G(o, x) G(x, u) G(y, u)
y∈Bo (4r) u∈Zd

≤ Cr6−d .

51
Write futurex = {y ∈ Zd : x ∈ pasty }. Using Cauchy-Schwarz we have
X
e o (r/2)| > t)2
WSF(o ∈ pastx , |T
x∈Ao (r,(3/2)r)
 2
X
≥ cr−d  e o (r/2)| > t)
WSF(o ∈ pastx , |T (7.22)
x∈Ao (r,(3/2)r)
h i2
= c r−d E |futureo ∩ Ao (r, (3/2)r)| 1|Te 0 (r/2)|>t .
 
e o (r/2)| ≥ c r4 , due to a result of Pemantle [31, Lemma 3.1], and using Wilson’s algorithm,
We have E |T
we have   X
e o (r/2)|2 ≤
E |T G(o, w) G(w, u) G(w, y) ≤ C r8 .
u,y,w∈Bo (r/2)

This yields WSF(|T e o (r/2)| > t) ≥ c1 > 0 for some c1 > 0 and sufficiently small δ. Fix such δ. We get
a lower bound on |futureo ∩ Ao (r, (3/2)r)| by considering the number of loop-free points of the random
walk So generating the path from o to ∞. A result of Lawler says that with probability 1, the fraction
of loop-free points is asymptotically a positive constant in d ≥ 5; see [23, Section 7.7]. Therefore, we
can find ε > 0 small enough, such that WSF(|futureo ∩ Ao (r, (3/2)r)| ≥ εr2 ) ≥ 1 − c1 /2. With these
choices of δ and ε, the right hand side of (7.22) is at least r−d ((c1 /2) ε r2 )2 =
c2 r4−d . Substituting this
and (7.21) into the bound given by Theorem 1.6 we obtain WSF |pasto | ≥ t ≥ c r−2 = c′ t−1/2 . As in
Section 6.2, we have
1
ν(|Av| > t) ≥ WSF(|pasto | > t),
2d
and the claim of the theorem follows.

7.6 Upper bounds: d ≥ 5

Unlike the case of the radius, an upper bound on the size of waves does not yield directly an upper bound
on the size of the avalanche. However, we can still get a power law upper bound, that we prove in this
section.
Proof of theorem 1.5(iv); upper bound. Fix some k ≥ 1. Recalling that N denotes the number of waves,
we have
ν(S > t) = ν(S > t, N > k) + ν(S > t, N ≤ k). (7.23)
From Lemma 2.3 we have N ≤ R, and we can upper bound the first term in the expression above by
ν(R > k). Next, writing S j for the size of the j-th wave, if S > t and N ≤ k, then we have S j > t/k for
some 1 ≤ j ≤ N . Hence,
ν(S > t) ≤ ν(R > k) + ν(S j > t/k, for some j ≤ N )
≤ C WSFo (diam(To ) > k) + C WSFo (|To | > t/k)
≤ k −2+o(1) + C WSFo (|To | > t/k), (7.24)
where we used Theorem 1.2(iv) for the first term. We now control the second term on the right in
equation (7.24).

52
Lemma 7.9. WSFo (|To | > t) ≤ t−1/2+o(1) .
Proof. We note that an application of Theorem 1.2(iv) yields

WSFo (|To | > t) ≤ WSFo (diam(To ) > t1/4 ) + WSFo (|B(t1/4 ) ∩ To | > t)
≤ (t1/4 )−2+o(1) + WSFo (|B(t1/4 ) ∩ To | > t). (7.25)

Due to Wilson’s algorithm, we have WSFo (x ∈ To ) ≤ C |x|2−d , and hence we have

EWSFo |B(t1/4 ) ∩ To | C (t1/4 )2

WSFo (|B(t1/4 ) ∩ To | > t) ≤ ≤ = C t−1/2 .
t t

Using the above lemma and optimizing the number of waves k in (7.24), we get ν(S > t) ≤ t−2/5+o(1) ,
thereby completing the proof of the upper bound in Theorem 1.5(iv).

Acknowledgements: The authors thank Russ Lyons for helpful discussions. J.H. and A.A.J. also
thank the organizers of the 2014 Bath Summer School, where some of this work was initiated.
J.H. thanks Michael Damron for postdoctoral advising and encouragement.
S.B. thanks Russ Lyons for support, encouragement and advice.

References
[1] D. J. Aldous and R. Lyons (2007). Processes on unimodular random networks. Electron. J. Probab.
12 no. 54, 1454–1508 (electronic).
[2] S. R. Athreya and A. A. Járai (2004). Infinite volume limit for the stationary distribution of Abelian
sandpile models. Commun. Math. Phys. 249 197–213.
[3] P. Bak, C. Tang and K. Wiesenfeld (1988). Self-organized criticality, Phys. Rev. A 38 364–374.
[4] M. T. Barlow and D. Karli (2015). Some remarks on uniform boundary Harnack Principles. Preprint.
http://arxiv.org/abs/1507.04115
[5] M. T. Barlow and A. A. Járai (2016). Geometry of uniform spanning forest components in high
dimensions. Preprint. Available at http://arxiv.org/abs/1602.01505
[6] I. Benjamini, R. Lyons, Y. Peres and O. Schramm (2001). Uniform spanning forests. Ann. Probab.
29, 1–65.
[7] I. Benjamini and O. Schramm (2001). Recurrence of distributional limits of finite planar graphs.
Electron. J. Probab. 6, no. 23, 13 pp. (electronic).
[8] D. Dhar (1990). Self-organized critical state of sandpile automaton models. Phys. Rev. Lett. 64,
1613–1616.
[9] D. Dhar (2006). Theoretical studies of self-organized criticality. Phys. A: 369, 29-70.

53
[10] S. L. Gamlin and A. A. Járai (2014). Anchored burning bijections on finite and infinite graphs,
Electron. J. Probab. 19, no. 117, 23 pp.
[11] A. Fey-den Boer and F. Redig (2008). Limiting shapes for deterministic centrally seeded growth
models. J. Stat. Phys. 130, 579–597.
[12] A. E. Holroyd, L. Levine, K. Mészáros, Y. Peres, J. Propp and D. B. Wilson (2008). Chip-firing and
rotor-routing on directed graphs. In and out of equilibrium. 2, Progr. Probab. 60, Birkhäuser, Basel,
331–364.
[13] E. V. Ivashkevich, D. V. Ktitarev and V. B. Priezzhev (1994). Waves of topplings in an Abelian
sandpile. Physica A: Statistical Mechanics and its Applications, 209, 347–360.
[14] A. A. Járai (2014). Sandpile models. Preprint. http://arxiv.org/abs/1401.0354
[15] A. A. Járai and F. Redig (2008). Infinite volume limit of the Abelian sandpile model in dimensions
d ≥ 3. Probab. Theory Related Fields 141, 181–212.
[16] A. A. Járai, F. Redig and E. Saada (2015). Approaching criticality via the zero-dissipation limit in
the abelian avalanche model. J. Stat. Phys. 159, 1369–1407.
[17] A. A. Járai and N. Werning (2014). Minimal configurations and sandpile measures. J. Theoret.
Probab. 27, 153–167.
[18] R. Kenyon (2000). The asymptotic determinant of the discrete Laplacian. Acta Math. 185, 239–286.
[19] H. Kesten (1987). Hitting probabilities of random walks on Zd . Stochastic Process. Appl. 25, 165–
184.
[20] H. Kesten (1987). Scaling relations for 2D-percolation. Comm. Math. Phys. 109, 109–156.
[21] G. F. Lawler (1995). The logarithmic correction for loop-erased walk in four dimensions. Proceedings
of the Conference in Honor of Jean-Pierre Kahane (Orsay, 1993), J. Fourier Anal. Appl. Special Issue,
347–361.
[22] G. F. Lawler (1996). Cut times for simple random walk. Electron. J. Probab. 1, Paper no. 13.

[23] G. F. Lawler (1996). Intersections of Random Walks. Birkhäuser, Boston.

[24] G. F. Lawler and V. Limic (2010). Random walk: a modern introduction. Cambridge studies in
advanced mathematics, Vol. 123, Cambridge University Press, Cambridge.
[25] R. Lyons, B. J. Morris and O. Schramm (2008). Ends in uniform spanning forests. Electron. J.
Probab. 13, Paper no. 58.
[26] R. Lyons and Y. Peres (2016). Probability on Trees and Networks. Cambridge University Press,
Available at http://pages.iu.edu/~rdlyons/
[27] R. Lyons, Y. Peres and O. Schramm (2003). Markov chain intersections and the loop-erased walk.
Ann. Inst. H. Poincaré Probab. Statist. 39, 779–791.

54
[28] S.N. Majumdar and D. Dhar (1992). Equivalence between the Abelian sandpile model and the q → 0
limit of the Potts model. Physica A: Statistical Mechanics and its Applications 185, 129–145.
[29] R. Masson (2009). The growth exponent for planar loop-erased random walk. Electron. J. Probab.
14, Paper no. 36.
[30] B. Morris (2003). The components of the wired spanning forest are recurrent. Probab. Theory Related
Fields 125, no. 2, 259–265.
[31] R. Pemantle (1991). Choosing a spanning tree for the integer lattice uniformly. Ann. Probab. 19,
1559–1574.
[32] V. B. Priezzhev (2000). The upper critical dimension of the Abelian sandpile model. J. Statist. Phys.
98, 667–684.
[33] F. Redig (2006). Mathematical aspects of the abelian sandpile model, in Mathematical statistical
physics, 657–729, Elsevier B. V., Amsterdam.
[34] D. Shiraishi (2014). Growth exponent for loop-erased random walk in three dimensions. Preprint.
arXiv:1310.1682v6.
[35] D. B. Wilson (1996). Generating random spanning trees more quickly than the cover time, in Pro-
ceedings of the Twenty-eighth Annual ACM Symposium on the Theory of Computing (Philadelphia,
PA, 1996), 296–303, ACM, New York.
[36] W. Woess (2000). Random walks on infinite graphs and groups. Cambridge Tracts in Mathematics
138, Cambridge University Press, Cambridge.

55