Energy of flows on Z 2 percolation clusters

Similar documents
APPENDIX A Some Linear Algebra

More metrics on cartesian products

Appendix B. Criterion of Riemann-Stieltjes Integrability

Complete subgraphs in multipartite graphs

Lecture 10: May 6, 2013

Lectures - Week 4 Matrix norms, Conditioning, Vector Spaces, Linear Independence, Spanning sets and Basis, Null space and Range of a Matrix

Maximizing the number of nonnegative subsets

Linear, affine, and convex sets and hulls In the sequel, unless otherwise specified, X will denote a real vector space.

Affine transformations and convexity

BOUNDEDNESS OF THE RIESZ TRANSFORM WITH MATRIX A 2 WEIGHTS

Exercise Solutions to Real Analysis

FACTORIZATION IN KRULL MONOIDS WITH INFINITE CLASS GROUP

Finding Dense Subgraphs in G(n, 1/2)

THE CHINESE REMAINDER THEOREM. We should thank the Chinese for their wonderful remainder theorem. Glenn Stevens

20. Mon, Oct. 13 What we have done so far corresponds roughly to Chapters 2 & 3 of Lee. Now we turn to Chapter 4. The first idea is connectedness.

Anti-van der Waerden numbers of 3-term arithmetic progressions.

NP-Completeness : Proofs

MAT 578 Functional Analysis

2.3 Nilpotent endomorphisms

6. Stochastic processes (2)

6. Stochastic processes (2)

THE WEIGHTED WEAK TYPE INEQUALITY FOR THE STRONG MAXIMAL FUNCTION

STEINHAUS PROPERTY IN BANACH LATTICES

princeton univ. F 17 cos 521: Advanced Algorithm Design Lecture 7: LP Duality Lecturer: Matt Weinberg

PHYS 705: Classical Mechanics. Calculus of Variations II

3.1 Expectation of Functions of Several Random Variables. )' be a k-dimensional discrete or continuous random vector, with joint PMF p (, E X E X1 E X

Math 217 Fall 2013 Homework 2 Solutions

Problem Set 9 Solutions

n α j x j = 0 j=1 has a nontrivial solution. Here A is the n k matrix whose jth column is the vector for all t j=0

Ballot Paths Avoiding Depth Zero Patterns

Another converse of Jensen s inequality

Case A. P k = Ni ( 2L i k 1 ) + (# big cells) 10d 2 P k.

Introductory Cardinality Theory Alan Kaylor Cline

Bernoulli Numbers and Polynomials

Graph Reconstruction by Permutations

PROCEEDINGS OF THE AMERICAN MATHEMATICAL SOCIETY Volume 125, Number 7, July 1997, Pages 2119{2125 S (97) THE STRONG OPEN SET CONDITION

Errors for Linear Systems

Assortment Optimization under MNL

Caps and Colouring Steiner Triple Systems

The L(2, 1)-Labeling on -Product of Graphs

First day August 1, Problems and Solutions

Learning Theory: Lecture Notes

Section 8.3 Polar Form of Complex Numbers

find (x): given element x, return the canonical element of the set containing x;

Spectral Graph Theory and its Applications September 16, Lecture 5

PHYS 705: Classical Mechanics. Newtonian Mechanics

A CHARACTERIZATION OF ADDITIVE DERIVATIONS ON VON NEUMANN ALGEBRAS

College of Computer & Information Science Fall 2009 Northeastern University 20 October 2009

Remarks on the Properties of a Quasi-Fibonacci-like Polynomial Sequence

THERE ARE INFINITELY MANY FIBONACCI COMPOSITES WITH PRIME SUBSCRIPTS

PES 1120 Spring 2014, Spendier Lecture 6/Page 1

1 Matrix representations of canonical matrices

Stanford University CS359G: Graph Partitioning and Expanders Handout 4 Luca Trevisan January 13, 2011

Appendix for Causal Interaction in Factorial Experiments: Application to Conjoint Analysis

Every planar graph is 4-colourable a proof without computer

Lecture 4: Universal Hash Functions/Streaming Cont d

Affine and Riemannian Connections

Physics 5153 Classical Mechanics. Principle of Virtual Work-1

Finding Primitive Roots Pseudo-Deterministically

Lecture 3. Ax x i a i. i i

Lecture Notes Introduction to Cluster Algebra

U.C. Berkeley CS294: Beyond Worst-Case Analysis Luca Trevisan September 5, 2017

Geometry of Müntz Spaces

Some basic inequalities. Definition. Let V be a vector space over the complex numbers. An inner product is given by a function, V V C

Edge Isoperimetric Inequalities

Structure and Drive Paul A. Jensen Copyright July 20, 2003

THE CHVÁTAL-ERDŐS CONDITION AND 2-FACTORS WITH A SPECIFIED NUMBER OF COMPONENTS

CSCE 790S Background Results

THE ARIMOTO-BLAHUT ALGORITHM FOR COMPUTATION OF CHANNEL CAPACITY. William A. Pearlman. References: S. Arimoto - IEEE Trans. Inform. Thy., Jan.

MATH 829: Introduction to Data Mining and Analysis The EM algorithm (part 2)

REAL ANALYSIS I HOMEWORK 1

Random Walks on Digraphs

The Order Relation and Trace Inequalities for. Hermitian Operators

12 MATH 101A: ALGEBRA I, PART C: MULTILINEAR ALGEBRA. 4. Tensor product

A note on almost sure behavior of randomly weighted sums of φ-mixing random variables with φ-mixing weights

Self-complementing permutations of k-uniform hypergraphs

COMPLEX NUMBERS AND QUADRATIC EQUATIONS

where a is any ideal of R. Lemma 5.4. Let R be a ring. Then X = Spec R is a topological space Moreover the open sets

REGULAR POSITIVE TERNARY QUADRATIC FORMS. 1. Introduction

Economics 101. Lecture 4 - Equilibrium and Efficiency

Min Cut, Fast Cut, Polynomial Identities

Volume 18 Figure 1. Notation 1. Notation 2. Observation 1. Remark 1. Remark 2. Remark 3. Remark 4. Remark 5. Remark 6. Theorem A [2]. Theorem B [2].

U.C. Berkeley CS294: Spectral Methods and Expanders Handout 8 Luca Trevisan February 17, 2016

Uniqueness of Weak Solutions to the 3D Ginzburg- Landau Model for Superconductivity

STAT 309: MATHEMATICAL COMPUTATIONS I FALL 2018 LECTURE 16

Subset Topological Spaces and Kakutani s Theorem

The path of ants Dragos Crisan, Andrei Petridean, 11 th grade. Colegiul National "Emil Racovita", Cluj-Napoca

A permuted random walk exits faster

princeton univ. F 13 cos 521: Advanced Algorithm Design Lecture 3: Large deviations bounds and applications Lecturer: Sanjeev Arora

The Multiple Classical Linear Regression Model (CLRM): Specification and Assumptions. 1. Introduction

Genericity of Critical Types

Week 2. This week, we covered operations on sets and cardinality.

An Introduction to Morita Theory

One Dimension Again. Chapter Fourteen

SPECTRAL PROPERTIES OF IMAGE MEASURES UNDER THE INFINITE CONFLICT INTERACTION

arxiv: v1 [math.co] 1 Mar 2014

Modulo Magic Labeling in Digraphs

MATH 241B FUNCTIONAL ANALYSIS - NOTES EXAMPLES OF C ALGEBRAS

Supplement: Proofs and Technical Details for The Solution Path of the Generalized Lasso

The Geometry of Logit and Probit

Transcription:

Energy of flows on Z 2 percolaton clusters Chrstopher Hoffman 1,2 Abstract We show that f p > p c (Z 2 ), then the unque nfnte percolaton cluster supports a nonzero flow f wth fnte q energy for all q > 2. Ths extends the work of Grmmett, Kesten, and Zhang and Levn and Peres n dmensons d 3. As an applcaton of our technques we exhbt a graph that has transent percolaton clusters, but does not admt exponental ntersecton tals. Ths answers a queston asked by Benjamn, Pemantle, and Peres. Keywords : percolaton, energy, electrcal networks, exponental ntersecton tals. Subject classfcaton : Prmary: 60J45; Secondary: 60J10, 60J65, 60J15, 60K35. 1 Department of Mathematcs, Unversty of Maryland, College Park, MD, 20742 2 Research supported n part by an NSF postdoctoral fellowshp 1

1 Introducton Consder Bernoull bond percolaton on Z d wth parameter p. Recall that ths s the ndependent process on Z d whch retans an edge wth probablty p and deletes an edge wth probablty 1 p. For all d > 1, there exsts a crtcal parameter p c (Z d ) < 1 such that f p < p c, then a.s. there s no connected component wth nfntely many edges. If p > p c, then a.s. there s a unque connected component wth nfntely many edges [1]. Ths component s called the unque nfnte cluster. Kesten proved that p c (Z 2 ) = 1/2 [9]. The phase where p > p c s called supercrtcal Bernoull percolaton. We consder only the supercrtcal case. Let G = G(E, V ) be the graph wth vertex set V and edge set E. Consder each undrected edge on G as two drected edges, one n each drecton. Let vw be the drected edge from v to w. A flow f on G wth source v 0 s an nonnegatve edge functon such that the net flow out of any vertex v v 0 s zero: w f(v 0 w) w f(wv 0 ) = 0. The strength of a flow f from the orgn s the amount flowng from 0: w f(0w) w f(w0). A multsource flow f on G s the sum (or ntegral) of flows on G. The q energy of a flow (or multsource flow) f on G s E q (f) = e E f(e) q. The energy of a flow s the 2 energy of the flow. Note that f a multsource flow on G has fnte q energy then t s the sum (or ntegral) of flows wth fnte q energy. Grmmett, Kesten and Zhang proved that f d 3 and p > p c (Z d ), then smple random walk on the nfnte percolaton cluster, C (Z d, p), s a.s. transent [6]. They proved ths result by constructng a flow on the percolaton cluster wth fnte energy, whch s equvalent to transence of smple random walk on the cluster. Grmmett, Kesten, and Zhang showed that there exsts a tree n C (Z d, p) where the branches bfurcate at farly regular ntervals. Ths tree gves rse to a flow n a natural way. It s then easy to bound the energy of ths flow. Benjamn, Pemantle and Peres [3] gave an alternatve proof of Grmmett, Kesten and Zhang s result. They constructed unpredctable processes on Z. They used them to create a measure on the collecton of paths n Z d whch emanate from 0. For d 3 the measure µ they created has exponental ntersecton tals. That s, there exsts C and a θ < 1 such that for 2

all n µ µ { (ϕ, ψ) : ϕ ψ n } Cθ n, where ϕ ψ s the number of edges n the ntersecton of ϕ and ψ. We say that a graph G admts exponental ntersecton tals f there exsts a measure µ on paths on G wth exponental ntersecton tals. Then they proved that any graph that admts exponental ntersecton tal has transent percolaton clusters for some p < 1. They asked whether the converse s true. We show that t s not. Levn and Peres adapted the approach of Benjamn et al. to show that for d 3 these flows have fnte q energy a.s. for q > d/(d 1) [10]. Ths result s optmal snce Z d also supports flows of fnte q energy f and only f q > d/(d 1) [11]. In ths paper we wll use the method of Grmmett, Kesten, and Zhang to extend Levn and Peres s result to d = 2. Theorem 1.1 For every p > p c (Z 2 ) and q > 2 there exsts a flow on C (Z 2, p) wth fnte q energy a.s. Levn and Peres also proved a result usng a generalzaton of the energy of a flow. For any d 2 and α > 0, let H d,α (u) = u d/(d 1) /[log(1 + u 1 )] α for u > 0 and H d,α (0) = 0. Levn and Peres proved that f d 3 then C (Z d, p) supports a flow of fnte H d,α energy for any α > 2. (That s e H d,α (f(e)) <.) Hoffman and Mossel sharpened ths result by showng that C (Z d, p) supports a flow of fnte H d,α energy for d 3 and α > 1 [8]. Ths last result s optmal because there exsts a flow of fnte H d,α energy on Z d f and only f α > 1 [12]. We wll extend ths result of Levn and Peres to Z 2. Theorem 1.2 For every p > 1/2 and α > 2 there exsts a flow on C (Z 2, p) wth fnte H 2,α energy a.s. We are unable to prove a verson of Hoffman and Mossel s results for d = 2, but we conjecture that t s true. Conjecture 1.1 For every p > 1/2 and α > 1 there exsts a flow on C (Z 2, p) wth fnte H 2,α energy a.s. 3

Ths would requre a new approach as the Grmmett et al. method cannot be extended to reach ths concluson and the Benjamn et al. approach says nothng about graphs, lke Z 2, that do not admt exponental ntersecton tals. Häggström and Mossel used Benjamn, Pemantle and Peres s approach to extend the Grmmett, Kesten, and Zhang result n another way. They defned two classes of subgraphs of Z 3 and showed these subgraphs admt exponental ntersecton tals. Thus smple random walk on the nfnte percolaton cluster of those subgraphs s transent [7]. On certan subgraphs of Z 3 where Häggström and Mossel showed that smple random walk on the nfnte percolaton cluster s transent t s clear that the Grmmett, Kesten, and Zhang approach wll not work. We gve an example of a graph where the Grmmett et al. approach works but the Benjamn et al. approach does not. More specfcally we prove the followng. Theorem 1.3 There exsts a graph G whch has transent percolaton clusters but does not admt exponental ntersecton tals. To do ths we construct a graph G whch s the drect product of a tree and Z. The subset of G whch projects onto a branch of the tree s a subgraph of Z 2. On the percolaton cluster restrcted to that subset we construct a flow of fnte H 2,3 energy. Integratng these flows over all branches gves a multsoure flow of fnte energy on the percolaton cluster of G. Thus smple random walk on the nfnte percolaton cluster of G s transent. Then we show that ths graph does not support exponental ntersecton tals. 2 Flows on C (Z 2, p) We proceed usng the method of Grmmett, Kesten, and Zhang. We wll fnd a tree n C (Z 2, p). The branches of ths tree splt nto two at farly regular ntervals. Wth ths tree we wll assocate a flow. Snce we have control on how often the branches of the tree bfurcate we wll be able bound the q energy of the flow. Ths method allows us to show that there exst flows on C (Z 2, p) of fnte q energy for all q > 2. Frst we ntroduce some notaton. A path P s a sequence of open drected edges where the end of one edge s the begnnng of the next edge. A path P connects x and 4

y f x and y are endponts of edges n P. We wrte x y f there exsts an open path from x to y. We wrte x f x s part of the unque nfnte cluster. If x y then we let D(x, y) be the length of the shortest open path from x to y. We use the taxcab metrc on Z 2, (x, y), (x, y ) = x x + y y. Let β > 2, ρ 1, and b be constants whch wll be defned later. Let a = 20ρb. Let X k () = (β k, (β/2) k ) for all, 2 k 1 < 2 k 1, and k 1. Defne X k () =.5(X k+1 (2) + X k+1 (2 1)). The tree n the percolaton cluster that we wll construct wll have the followng propertes. For each and k there wll be one branch gong from near X k () to near X k (). Ths branch wll then bfurcate near X k+1 (2). One of the branches goes toward X k+1 (2), whle the other branch goes toward X k+1 (2 1). Defne L(u, v) to be the elements n Z 2 whch are wthn 2 of the lne segment jonng u and v. Let B(k) be all (x, y) Z 2 such that x + y k. Defne T k () = B(ak) + ( L(X k (), X k ()) L(X k+1 (2), X k+1 (2 1)) ), where + represents Mnkowsk addton. Lemma 2.1 There exsts a functon K(a) such that T k () T k (j) = for all j and k > K(a). We also have that T k () T k+1 (j) = for all j 2 or j 2 1 and k > K(a). Proof: If T k () T k (j) s not empty the there exsts a pont n L(X k (), X k ()) L(X k+1 (2), X k+1 (2 1)) whch s wthn 2ak of a pont n L(X k (j), X k (j)) L(X k+1 (2j), X k+1 (2j 1)). 5

Wthout loss of generalty ths mples ether X k (), X k (j) 2ak or X k+1 (2), X k+1 (2j 1) 2ak. Those two dstances are at least (β/2) k. So the frst condton s satsfed f (β/2) K(a) > 2aK(a). The second condton s also satsfed wth the same choce of K(a) for the same reason. Our next goal s to gve a suffcent condton for there to be an open path from a pont near X k () to a pont near X k+1 (2). Our condton wll also mply that the path les entrely nsde T k () and has length bounded by C X k (), X k+1 (2). Fnd y 1 = y 1 (k, ) through y t = y t (k, ) such that 1. y 1 = X k (), 2. y t X k () 1 3. 4bk y u, y u+1 8bk, for all 1 u t 1 4. y u L(X k (), X k ()), for all 1 u t 1 and 5. t β k+1 /4bk. Defne E u = E u (k, ) to be the event that 1. there exsts z 1 y u + B(bk) such that z 1, and 2. any two ponts z 2 (y u + B(bk)) and z 3 (y u+1 + B(bk)) whch are connected have D(z 2, z 3 ) ak/2. Ths mples that the shortest path connectng z 2 and z 3 les n y u + B(ak) T k (). If all the E u hold then there s a path from near X k () to near X k () nsde T k (). In a smlar manner we can fnd y 1 through y t such that 1. y 1 = X k+1 (2 1), 2. y t = X k+1 (2), 3. 4bk y u, y u+1 8bk, for all 1 u t 1 4. y u L(X k+1 (2 1), X k+1 (2)), for all 1 u t 1, 6

5. t 5ρβ k+1 /bk, and 6. there exst u and u such that y u = y u. Defne E u = E u(k, ) to be the event that 1. there exsts z 1 y u + B(bk) such that z 1, and 2. any two ponts n z 2 (y u + B(bk)) and z 3 (y u+1 + B(bk)) whch are connected have D(z 2, z 3 ) ak/2. Ths mples that the shortest path connectng z 2 and z 3 les n y u + B(ak) T k (). Defne E k () = ( u E u (k, )) ( u E u(k, )). In ths next lemma we show that f E k () holds for one k and then there exst a path from near X k () to near X k+1 (2) and a path from near X k () to near X k+1 (2 1). We also show that f E k () holds for all k suffcently large and all then there exsts a tree wth the desred propertes n C (Z 2, p). Ths mples that there exsts a flow of fnte q energy for all q > 1 + log 2 β. Lemma 2.2 If there exsts a K such that E k () holds for all k K and then there exsts a flow on C (Z 2, p) wth fnte q energy for all q > 1 + log 2 β. Proof: It causes no loss of generalty to assume that K > K(a). We construct a multsource flow that has a source near each X K (). We wll show that the multsource flow has fnte energy. Thus there exsts a flow wth fnte energy. Snce E k () holds for each k K and t s possble to pck ponts p k () such that p k () X k () + B(bk) and p k (). For each k and t s possble to pck ponts z u y u + B(bk) such that z u. It s also possble to select z u y u + B(bk) such that z u. We can also requre that there exst u and u such that z u = z u, z 1 = p k (), z 1 = p k+1 (2 1), and z t = p k+1 (2). The second condton n the defnton of E u mples that z u z u+1. Snce all the E u and E u hold we can pece together these paths to form paths P k () and P k() such that 1. P k () whch connects p k () and p k+1 (2) 2. P k() whch connects p k () and p k+1 (2 1) 3. P k (), P k() T k (), and 7

4. P k (), P k() (ak/2)(β k+1 /4bk) = 5ρβ k+1 /2. The unon of the P k () and the P k() does not necessarly form a tree. However t s easy to remove branches so that t does form a tree. Instead of dong ths we use the P k () and the P k() to defne a flow drectly. For each edge e assgn t mass f(e) = k, 1 2 k ( IPk ()(e) + I P k ()(e) ). Lemma 2.1, the fact that K K(a), and condton 3 on the P k () mples that f e P k () P k() and k > K(a) then f(e) 4(2 k ). If f(e) q + f(e) q < e P k () e P k () then f(e) q < s fnte. Thus the followng calculaton shows f has fnte q energy. f(e) q + (4(2 k )) q 5ρβ k+1 e P k () e P k () f(e) q C2 kq β k C2 k(q log 2 β) 2 k+1 2C2 k(q log 2 β 1) <. Now we show that wth probablty one there exsts a K so that E k () holds for all k > K and all. To do ths we need to bound the probabltes of E u and the E u. Ths requres one two theorems. The frst follows from the work of Russo [13], Seymour and Welsh [14], and Kesten [9]. Theorem 2.1 Gven p >.5 there exsts C 1 and α 1 so that P(B(k) ) < C 1 2 α 1k. The second s due to Antal and Psztora [2]. Theorem 2.2 [2] Let p >.5. Then there exsts ρ = ρ(p) [1, ), and constants C 2, and δ > 0 such that for all y P(0 y, D(0, y) > ρ y ) < C 2 2 δ y. 8

Lemma 2.3 There exsts an α > 0 such that for all k P(E u (k, )) > 1 C 3 2 αbk. Proof: By Theorem 2.1 the probablty that condton 1 n the defnton of E u (k, ) does not hold s less than C 1 2 α1bk. If condton 2 s not true then there exst two ponts that are connected but the dstance between them s large. Theorem 2.2 mples that P ( z 2 and z 3 such that D(z 2, z 3 ) > ak/2 = 10ρbk ρ z 2 z 3 ) < (bk) 2 C 2 2 2δbk. Thus P(E u (k, )) > 1 C 1 2 α1bk (bk) 2 C 2 2 2δbk > 1 C 3 2 αbk for an approprate choce of C 3 and α. We also get the same bound for P(E u). Lemma 2.4 For p >.5 there exsts b, C 4, and α > 1 such that for all k and P(E k ()) 1 C 4 2 α k. Proof: Choose b large enough so that α = αb log 2 β > 1. P(E k ()) 1 u P(E u (k, ) C ) u P(E u(k, ) C ) 1 5ρβ k+1 C 3 2 αbk 1 C 4 2 k log 2 β αbk 1 C 4 2 α k. Proof of Theorem 1.1: Gven p >.5 and q > 2 choose β so that β > 2 and q > 1 + log 2 β. The prevous lemma mples that lm n P(,k>n E k ()) = 1. Thus the Borel- Cantell lemma and Lemma 2.2 there exsts a flow of fnte q energy wth probablty 1. 9

3 A Refnement In ths secton we wll be concerned wth a generalzaton of the energy of a flow. For any functon H we defne the H energy of a flow f as E H (f) = e H(f(e)). We wll be workng wth a specal class of functons. For any α > 0, let H 2,α (u) = u 2 /[log(1 + u 1 )] α for u > 0 and H 2,α (0) = 0. The man result of ths secton s that for d = 2 there exsts a flow of fnte H 2,α energy on C (Z 2, p) for all α > 2. The proof s very smlar to the prevous secton. Let β > 1, ρ 1, and b be constants whch wll be defned later. Let a = 20ρb. Defne t k = 2 k k β. Let X k () = (t k, (t k )/2 k ) for 2 k and k 0. Then defne T k () as n secton 2. Lemma 3.1 For each β > 1 there exsts a functon K(a) such that T k () T k (j) = (1) for all j and k > K(a). We also have that T k () T k+1 (j) = (2) for all j 2 or j 2 1 and k > K(a). Proof: If T k () T k (j) s not empty the there exsts a pont n L(X k (), X k ()) L(X k+1 (2), X k+1 (2 1)) whch s wthn 2ak of a pont n L(X k (j), X k (j)) L(X k+1 (2j), X k+1 (2j 1)). Wthout loss of generalty ths mples ether X k (), X k (j) 2ak or X k+1 (2), X k+1 (2j 1) 2ak. Those two dstances are at least k β. So the frst condton s satsfed f K(a) β > 2aK(a). Ths can be acheved whenever β > 1. The second condton s also satsfed wth the same choce of K(a) for the same reason. 10

Pck the y 1,..., y t, y 1,..., y t as n secton 2. Ths can be done wth t t k+1 /4bk < 2 k 1 (k + 1) β /bk. Defne E u, E u, and E k () the same as n the prevous secton. Lemma 3.2 If E k () holds for all k K and all then there exsts a flow on C (Z 2, p) wth fnte H 2,α energy for all α > 1 + β. Proof: Agan we create a multsource flow wth fnte energy. We assume that K > K(a). Snce E k () holds for each k and t s possble to pck ponts p k () such that p k () X k () + B(bk) and p k (). We can also defne z u and z u as n the prevous secton. Snce E k () hold then there exsts paths P k () and P k() such that 1. P k () whch connects p k () and p k+1 (2) 2. P k() whch connects p k () and p k+1 (2 1) 3. P k (), P k() T k (), and 4. P k (), P k() (ak/2)(2 k 1 (k + 1) β /bk) < 5ρ2 k (k + 1) β. For each edge e assgn t mass f(e) = k, 1 2 k ( IPk ()(e) + I P k ()(e) ). Lemma 3.1 mples that f e P k () P k() and k > K(a) then If, e P k () f(e) 4(2 k ). (3) H 2,α (f(e)) + H 2,α (f(e)) < e P k () then H 2,α (f(e)) <. Thus the followng calculaton shows f has fnte H 2,α energy. e P k () H 2,α (f(e)) + H 2,α (f(e)) (4(2 k )) 2 e P k () [log(1 + (4(2 k )) 1 )] α 10ρ2k (k + 1) β 11 C(2 k )k β [ log(4(2 k ))] α

< C (2 k )k β [k 2] α 2k+1 C k β α Proof of Theorem 1.2: Gven p >.5 and α > 2 choose β so that β > 1 and α > 1 + β. Choose b so that Lemma 2.4 mples that lm n P(,k>n E k ()) = 1. Thus by Lemma 3.2 and the Borel-Cantell lemma there exsts a flow of fnte H 2,α energy wth probablty 1. 4 A graph wth transent percolaton clusters that do not admt exponental ntersecton tals In ths secton we wll construct a graph G. Usng the results of the prevous secton we are able to show that G has transent percolaton clusters. Then we wll show that G does not admt exponental ntersecton tals. Let b = 2 2 /5. Before we defne G we defne T = {(x, y) x N, y {0, 1} N, y = 0 for all such that b x} to be the tree whose branches splt at dstance b from the root. Thus a vertex (x, y) s dstance x from the root. The vertex (x, y) s connected to the vertex (x + 1, y). If x = b for some then the vertex (x, y) s also connected to the vertex (x + 1, ỹ), where ỹ = 1 and ỹ j = y j for all j. At dstance d from the root of T there are O(log d) 5 branches. Let G = {(x, y, z) (x, y) T, z Z, and z.6x}. Theorem 4.1 For all p > 1/2 smple random walk on the nfnte percolaton clusters on G s a.s. transent. Proof: We use the constructon n the prevous secton wth β = 1.5. To each y {0, 1} N corresponds G y = {(x, y, z) G y = y for all such that x < b and y = 0 else}, 12

a subset of G whch looks lke a wedge of Z 2. By the prevous secton we have wth postve probablty exhbted a flow, f y, on the nfnte percolaton cluster of G y. Ths flow has fnte H 2,3 energy. Defne the multsource flow F (e) = f y (e)dm, where m s (1/2, 1/2) product measure on {0, 1} N. Denote by T k () all branches (x, y, z) such that (x, z) T k (). Now we calculate the energy of F. e T k () F (e) 2 = ( e T k () = ( e T k () ) 2 f y (e)dm B e f y (e)dm where B e = (y y = y for all b < x) for e = (x, y, z). The last equalty s true because the defnton of f y the nequalty mples that f y (e) = 0 for all y / B e. Jensen s nequalty generates ( Snce m(b e ) < C(log e ) 5 e T k () X ) 2 ( ) fdm (m(x)) (f) 2 dm. X F (e) 2 e T k () e T k () C k 5 k 5 <. C C k 2 C f k 5 y (e) 2 dm B e C k 5 e T k () f y (e) 2 dm ) 2 f y (e) 2 dm (4(2 k )) 2 10ρ2 k (k + 1) 3 (4) Lne (4) s true by equaton (3) and condton 4 n the proof of Lemma 3.2. Thus f p >.5 nfnte percolaton clusters on G support flows of fnte energy and are transent. 13

Now we show that there s no measure on paths on G wth exponental ntersecton tals. We do ths as follows. Frst we defne a functon f(ψ, ϕ). We show that f µ s a measure on paths wth exponental ntersecton tals then fd(µ µ) s fnte. Then we show that for any measure on paths on G that the ntegral must be nfnte. Thus the graph G does not admt exponental ntersecton tals. It causes no loss of generalty to assume that µ s supported on transent paths. Let ψ ϕ be the number of edges that ψ and ϕ have n common. Let χ e (ψ, ϕ) be the event that e ψ ϕ. Defne f(ψ, ϕ) = If e ψ let l(e) be the largest number such that ψ l(e) = e. Let ψ be the path gven by edges ψ +1, ψ +2,... Notce that ψ ϕ =1 5. f e χ e (ψ, ϕ)( ψ ϕ ψ l(e) ϕ ) 5. (5) Lemma 4.1 If µ s a measure wth exponental ntersecton tals then fd(µ µ) <. Proof: Because µ has exponental ntersecton tals there exsts constants C and θ < 1 such that µ µ( ψ ϕ ) Cθ. Thus fd(µ µ) = = ψ ϕ =1 5 =1 5 d(µ µ) χ ψ ϕ d(µ µ) 5 µ µ( ψ ϕ ) 5 Cθ <. Ths next lemma s the man tool we wll use to show that fd(µ µ) = for any measure µ on paths n G. 14

Lemma 4.2 There exsts C such that for any µ and e G E( ψ ϕ ψ l(e) ϕ e ψ ϕ) > C log e. Proof: Let E n (e) be the set of all edges at dstance n from e. Snce e = ϕ l(e) and µ s supported on transent paths then there exsts a k > l(e) so that ϕ k E n (e). We can rewrte ths as P(e ψ \ ψ l(e) e ψ) 1. e E n(e) The expected number of ntersectons of ψ \ ψ l(e) and ϕ n E n (e) s E( ψ ϕ E n (e) ψ l(e) ϕ E n (e) e ψ ϕ) P(e ψ \ ψ l(e) e ψ) 2 e E n(e) 1/ E n (e). The last nequalty s true because of the Cauchy-Schwartz nequalty. Let e = (x, y, z), e = (x, y, z ) and e E n (e). Defne e = x + z. Then.5 e < x e. If n e /10 then.3x < x < 1.2x. As any nterval (j, 4j) has at most one b there can be at most 3 possble values for z. For each x and z there are at most two possble values for y. Thus for any e G and n e /10 we have that E n (e) 12n. Puttng these two facts together we get that E( ψ ϕ ψ l(e) ϕ e ψ ϕ) > e /10 n=1 1/12n > C log e. Theorem 4.2 G does not admt exponental ntersecton tals. Proof: We argue by contradcton. If there were a measure µ wth exponental ntersecton tals then Lemma 4.1 says that fd(µ µ) <. The followng calculaton shows that ths ntegral must be nfnte. fd(µ µ) χ e (ψ, ϕ)( ψ ϕ ψ l(e) ϕ ) 5 d(µ µ) (6) e χ e (ψ, ϕ)e( ψ ϕ ψ l(e) ϕ e ψ ϕ) 5 d(µ µ) (7) e χ e (ψ, ϕ)(c log j) 5 d(µ µ) (8) e e =j 15

j j j j (C log j) 5 (µ µ)(e ψ, e ϕ) e =j (C log j) 5 µ(e ψ) 2 e =j (C log j) 5 1 C j(log j) 5 (9) C j Lne (6) comes from lne (5). Lne (7) follows from Jensen s nequalty. Lne (8) follows from Lemma 4.2. Lne (9) s true because the number of e such that e = s bounded by C (log ) 5. Thus µ does not have exponental ntersecton tals. Proof of Theorem 1.3: Theorem 1.3 s a combnaton of Theorems 4.1 and 4.2. References [1] M. Azenman, H. Kesten, C. Newman (1987) Unqueness of the nfnte cluster and contnuty of connectvty functons for short and long range percolaton. Comm. Math. Phys. 111, no. 4, 505 531. [2] P. Antal and A. Psztora (1996). On the chemcal dstance n supercrtcal Bernoull percolaton. Ann. Probab. 24 1036 1048. [3] I. Benjamn, R. Pemantle and Y. Peres (1998). Unpredctable paths and percolaton. Ann. Probab., to appear. [4] P. G. Doyle and E. J. Snell (1984). Random walks and electrcal networks. Carus Math. Monographs 22, Math. Assoc. Amer., Washngton, D. C. [5] G.R. Grmmett (1989). Percolaton. Sprnger, New York. [6] G. R. Grmmett, H. Kesten and Y. Zhang (1993). Random walk on the nfnte cluster of the percolaton model. Probab. Th. Rel. Felds 96, 33 44. [7] O. Häggström and E. Mossel (1998). Nearest-neghbor walks wth low predctablty profle and percolaton n 2 + ɛ dmensons. Ann. Probab., to appear. 16

[8] C. Hoffman and E. Mossel (1998). Energy of flows on percolaton clusters. to appear n Potental Analyss. [9] H. Kesten (1980). The crtcal probablty of bond percolaton on the square lattce equals 1/2. Comm. Math. Phys. 74 no. 1, 41 59. [10] D. Levn, Yuval Peres (1998). Energy and cutsets n nfnte percolaton clusters. Preprnt. [11] Lyons, T. (1983) A smple crteron for transence of a reversble Markov chan, Ann. Probab. 11, 393 402. [12] F. Y. Maeda (1977). A remark on the parabolc ndex of nfnte networks. Hroshma J. Math. 7, 147 152. [13] L. Russo (1978) A note on percolaton. Z. Wahrschenlchketstheore und Verw. Gebete 43 (1978), no. 1, 39 48. [14] P. Seymour, D. Welsh (1978). Percolaton probabltes on the square lattce. Advances n graph theory (Cambrdge Combnatoral Conf., Trnty College, Cambrdge, 1977). Ann. Dscrete Math. 3 227 245. 17