570 I. Benjamini, O. Schramm2. Notati<strong>on</strong>s <strong>and</strong> terminologyLet G =(V,E) be a graph. For c<strong>on</strong>venience, we usually <strong>on</strong>ly c<strong>on</strong>sider <strong>graphs</strong>with no loops or multiple edges (but the results do apply to multi<strong>graphs</strong>). The setof vertices incident with an edge e will be denoted ∂e; this is always a subset ofV that c<strong>on</strong>tains two vertices. We sometimes use {v, u} to denote the edge withendpoints v, u.Initially the graph G is unoriented, but for notati<strong>on</strong>al reas<strong>on</strong>s we also c<strong>on</strong>siderdirected edges. When {v, u} ∈E, we use [v, u] to denote the directed edge fromv to u. The set of all direceted edges will usually be denoted −→ E ; −→ E = {[v, u] :{v, u} ∈E}.The <strong>graphs</strong> we shall c<strong>on</strong>sider will be c<strong>on</strong>nected <strong>and</strong> locally finite. The lattermeans that the number of edges incident with any particular vertex is finite.Given any vertex v ∈ V , the collecti<strong>on</strong> of all edges of the form [v, u] whichare in −→ E will be denoted −→ E (v). The valence, ordegree, of a vertex v is just thecardinality of −→ E (v). G has bounded valence, if there is a finite upper bound forthe degrees of its vertices.Let f : V → R be some functi<strong>on</strong>. Then df is the functi<strong>on</strong> df : −→ E → Rdefined bydf ([v, u]) = f (u) − f (v).We also define the gradient of f to be equal to df ,∇f (e) =df (e).(The reas<strong>on</strong> for the multiplicity of notati<strong>on</strong> should become clear when we introducethe gradient with respect to a metric <strong>on</strong> G.)A functi<strong>on</strong> j : −→ E → R is a flow <strong>on</strong> G if it satisfiesj ([u,v]) = −j ([v, u])for every {v, u} ∈E. For example, for any f : V → R, df is a flow. Thedivergence of a flow j is the functi<strong>on</strong> div j : V → R, defined bydiv j (v) =∑j(e).If div j = 0, then j is divergence free.For an f : V → R we sete∈ −→ E(v)△f = div ∇f ,then △f : V → R is known as the discrete laplacian of f .If△f = 0, then f isharm<strong>on</strong>ic, while if △f = 0 <strong>on</strong> a subset V ′ ⊂ V , we say that f is harm<strong>on</strong>ic in V ′ .Equivalently, f is harm<strong>on</strong>ic iff its value at any v ∈ V is equal to the average ofthe values at the neighbors of v.
<str<strong>on</strong>g>Harm<strong>on</strong>ic</str<strong>on</strong>g> <str<strong>on</strong>g>functi<strong>on</strong>s</str<strong>on</strong>g> 571For a flow j <strong>and</strong> an e ∈ E we let |j (e)| denote |j (v)−j (u)|, where ∂e = {v, u}.The norm of a flow j is defined by‖j ‖ 2 = 1 ∑j (e) 2 = ∑ |j (e)| 2 .2−→ e∈Ee∈ EThe collecti<strong>on</strong> of all flows with finite norm is then a Hilbert space with thisnorm. The Dirichlet energy of a functi<strong>on</strong> f : V → R is defined byD(f )=‖df ‖ 2 .A Dirichlet functi<strong>on</strong> is an f : V → R with D(f ) < ∞. The space of allDirichlet <str<strong>on</strong>g>functi<strong>on</strong>s</str<strong>on</strong>g> <strong>on</strong> G is denoted D(G).The simple r<strong>and</strong>om walk <strong>on</strong> a locally finite graph G =(V,E) starting at avertex v 0 is the Markov process (v(1),v(2),...)<strong>on</strong>V such that v(1) = v 0 <strong>and</strong>the transiti<strong>on</strong> probability from a vertex v to a vertex u is the inverse of thecardinality of −→ E (v). A c<strong>on</strong>nected graph G is said to be transient, if there is apositive probability that a r<strong>and</strong>om walk that starts at a vertex v 0 will never visitv 0 again. It is easy to see that this does not depend <strong>on</strong> the initial vertex v 0 .An<strong>on</strong>-transient graph is recurrent.A metric m <strong>on</strong> a graph G =(V,E) is a positive functi<strong>on</strong> m : E → (0, ∞). Ther<strong>and</strong>om walk <strong>on</strong> (G, m) is the Markov process where the transiti<strong>on</strong> probabilityfrom v to u is equal to c(v, u)/c(v), where c(v, u) =m({v, u}) −1 , <strong>and</strong> c(v) isthe sum of c(v, u) over all neighbors u of v.The gradient of a functi<strong>on</strong> f : V → R with respect to a metric m is definedby∇ m f (e) =df (e)/m(e).f is said to be harm<strong>on</strong>ic <strong>on</strong> (G, m) if△ m f = div ∇ m f is zero. It is clear that fis harm<strong>on</strong>ic <strong>on</strong> (G, m) iff for every v ∈ V , f (v) is equal to the expected valueof f (u), where u is the state of the r<strong>and</strong>om walk <strong>on</strong> (G, m) that starts at v after<strong>on</strong>e step.The natural metric <strong>on</strong> G is the metric where each edge get weight 1. Inthe absence of another metric, all metric related noti<strong>on</strong>s are assumed to be withrespect to the natural metric. It is easy to check that if m is the natural metricthen a r<strong>and</strong>om walk <strong>on</strong> (G, m) is the same as a simple r<strong>and</strong>om walk <strong>on</strong> G, <strong>and</strong>the harm<strong>on</strong>ic <str<strong>on</strong>g>functi<strong>on</strong>s</str<strong>on</strong>g> <strong>on</strong> (G, m) are the harm<strong>on</strong>ic <str<strong>on</strong>g>functi<strong>on</strong>s</str<strong>on</strong>g> <strong>on</strong> G.Two metrics m, m ′ are mutually bilipschitz, if the ratios m/m ′ <strong>and</strong> m ′ /m arebounded.Let G =(V,E) be a c<strong>on</strong>nected, locally finite graph, <strong>and</strong> let m be a metric<strong>on</strong> G. The m-length of a path γ in G is the sum of m(e) over all edges in γ,length m (γ) = ∑ e∈γm(e).We define the m-distance d m (v, u) between any two vertices v, u ∈ V to be theinfimum of the m-lengths of paths c<strong>on</strong>necting v <strong>and</strong> u. Then (V , d m ) is a metricspace.
- Page 1: Invent. math. 126, 565-587 (1996)c
- Page 4 and 5: 568 I. Benjamini, O. Schramm1.4. De
- Page 8 and 9: 572 I. Benjamini, O. Schramm3. Stab
- Page 10 and 11: 574 I. Benjamini, O. SchrammFor the
- Page 12 and 13: 576 I. Benjamini, O. Schramm4.3. Le
- Page 14 and 15: 578 I. Benjamini, O. SchrammThe con
- Page 16 and 17: 580 I. Benjamini, O. Schrammlog : C
- Page 18 and 19: 582 I. Benjamini, O. SchrammProof (
- Page 20 and 21: 584 I. Benjamini, O. Schrammg : |G|
- Page 22 and 23: 586 I. Benjamini, O. SchrammLet C b