574 I. Benjamini, O. SchrammFor the proof of (2), the following lemma will be needed.3.4. Lemma Let f : G ◦ → Gbeaκ-quasim<strong>on</strong>omorphism between boundedvalence <strong>graphs</strong> G ◦ =(V ◦ ,E ◦ )<strong>and</strong> G =(V,E). Let m : E → (0, ∞) be a metric<strong>on</strong> G, <strong>and</strong> let m ◦ : E ◦ → [0, ∞) be defined bym ◦( {v ◦ , u ◦ } ) = d m(f (v ◦ ), f (u ◦ ) ) ,for any {v ◦ , u ◦ }∈E ◦ Then ‖m ◦ ‖≤C‖m‖, where C is a c<strong>on</strong>stant that depends<strong>on</strong>ly <strong>on</strong> κ <strong>and</strong> the maximal valence in G <strong>and</strong> G ◦ .Proof. For every edge e ◦ = {v ◦ , u ◦ } in G ◦ , let γ e ◦ be some path of combinatoriallength at most 2κ in G from f (v ◦ )tof(u ◦ ). Thenm ◦ (e ◦ ) 2 ≤ length m (γ e ◦) 2 =( ∑e∈γ e ◦m(e)) 2≤ 4κ 2 ∑e∈γ e ◦m(e) 2 .Therefore,‖m ◦ ‖ 2 ≤ 4κ 2 ∑ e∈Em(e) 2 |{e ◦ ∈ E ◦ : e ∈ γ e ◦}| .Since each of the paths γ e ◦ has length at most 2κ <strong>and</strong> ∣ ∣f −1 (v) ∣ ≤ κ for everyv ∈ V , it is clear that for any e ∈ E the cardinality of {e ◦ ∈ E ◦ : e ∈ γ e ◦} isbounded by a number that depends <strong>on</strong>ly <strong>on</strong> κ <strong>and</strong> the maximal valence in G ◦ .Hence, the lemma follows. ⊓⊔Proof of 3.1(2). Let m ◦ : E → [0, ∞) be as in Lemma 3.4, <strong>and</strong> let m1◦ : E →(0, ∞) satisfy m1 ◦(e) > m◦ (e) for every e ∈ E, while still m1 ◦ ∈ L2 (E). (Thereas<strong>on</strong> for using m1◦ rather than m◦ , is that, strictly speaking, m ◦ may not bea metric; it may happen that m ◦ (e) = 0 for some e ∈ E.) Note that f is ac<strong>on</strong>tracti<strong>on</strong> from the metric space (V ◦ , d m ◦1) to the metric space (V , d m ). Thestraightforward proof that m1 ◦ is a resolving metric for G◦ is again based <strong>on</strong> 3.4,<strong>and</strong> will be omitted. ⊓⊔Instability of the weak-Liouville propertyAs promised, we shall now provide a simple example showing that theweak-Liouville property is not invariant under rough isometries.Let G =(V,E) be a countable transient graph, let A ⊂ V , let m be the naturalmetric <strong>on</strong> G, <strong>and</strong> let m ′ be a metric bilipschitz to m. Suppose that with probability1 the r<strong>and</strong>om walk <strong>on</strong> (G, m) hits A infinitely often, but with probability1 the r<strong>and</strong>om walk <strong>on</strong> (G, m ′ ) hits A <strong>on</strong>ly finitely many times. T. Ly<strong>on</strong>s [19]observed that <strong>on</strong>e can find such A <strong>and</strong> m ′ when G is an infinite regular tree(which is not Z). For example, <strong>on</strong> the binary tree c<strong>on</strong>sisting of all finite sequences(ɛ 1 ,ɛ 2 ,...,ɛ n ) of 0’s <strong>and</strong> 1’s, where an edge appears between each(ɛ 1 ,ɛ 2 ,...,ɛ n−1 ) <strong>and</strong> (ɛ 1 ,ɛ 2 ,...,ɛ n ), <strong>on</strong>e can let A be the set of all (ɛ 1 ,ɛ 2 ,...,ɛ n )
<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> 575such that ∑ nj =1 ɛ j > n/3, <strong>and</strong> let m ′( (ɛ 1 ,...,ɛ n−1 ),(ɛ 1 ,...,ɛ n ) ) =1+cɛ n , wherec > 0 is a sufficiently large c<strong>on</strong>stant.We now c<strong>on</strong>struct a new graph H whose vertex set is the disjoint uni<strong>on</strong> of V<strong>and</strong> Z 4 . Let X be the set {(x, 0, 0, 0) : x ∈ Z} ⊂Z 4 , <strong>and</strong> let φ : A → Z 4 be anyinjective map from A into X . Let the edges of H c<strong>on</strong>sist of the edges in G, theedges in Z 4 , <strong>and</strong> all edges of the form [a,φ(a)] with a ∈ A. Extend the metricsm, m ′ to H by letting m(e) =m ′ (e) = 1 for any edge of H that is not in E.3.5. Theorem (H , m) is weak-Liouville, but (H , m ′ ) carries n<strong>on</strong>-c<strong>on</strong>stant,bounded harm<strong>on</strong>ic <str<strong>on</strong>g>functi<strong>on</strong>s</str<strong>on</strong>g>.Sketch of proof. With probability 1, the r<strong>and</strong>om walk <strong>on</strong> (H , m) will be in Z 4infinitely often. But there is some c<strong>on</strong>stant c > 0, such that the probability thata r<strong>and</strong>om walk <strong>on</strong> Z 4 that starts at any vertex in X will never reach X again isc. Hence the r<strong>and</strong>om walk <strong>on</strong> (H , m) will be absorbed in Z 4 ; that is, it will bein V <strong>on</strong>ly finitely many times. Since we may couple [17] the r<strong>and</strong>om walk <strong>on</strong>(H , m) with the r<strong>and</strong>om walk <strong>on</strong> Z 4 , it follows that (H , m) is weak-Liouville.On the other h<strong>and</strong>, let f (v) be the probability that the r<strong>and</strong>om walk <strong>on</strong> (H , m ′ )that starts at a vertex v will be absorbed in Z 4 . Then f is a n<strong>on</strong>-c<strong>on</strong>stant, boundedharm<strong>on</strong>ic functi<strong>on</strong> <strong>on</strong> (H , m ′ ).4. <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> <strong>on</strong> <strong>planar</strong> <strong>and</strong> <strong>almost</strong> <strong>planar</strong> <strong>graphs</strong>4.1. Theorem Any <strong>planar</strong> bounded degree graph G =(V,E)has a resolvingmetric.The proof will use the circle packing theorem, which we state shortly. Supposethat P =(P v :v∈V) is an indexed packing of disks in the plane. This justmeans that V is some set, to each v corresp<strong>on</strong>ds a closed disk P v ⊂ R 2 , <strong>and</strong>the interiors of the disks are disjoint. Let G be the graph with vertices V suchthat there is an edge joining v <strong>and</strong> u iff P v <strong>and</strong> P u are tangent. Then G is thec<strong>on</strong>tacts graph of P. (There are no multiple edges in G). It is easy to see thatG is <strong>planar</strong>, the circle packing theorem provides a c<strong>on</strong>verse:4.2. Circle Packing Theorem Let G =(V,E)be a finite <strong>planar</strong> graph with noloops or multiple edges, then there is a disk packing P =(P v :v∈V)in R 2 withc<strong>on</strong>tacts graph G.This theorem was first proved by Koebe [16]. Recently, at least 7 other proofshave been found; some of the more accesible <strong>on</strong>es can be found in [20], [6],[4].We shall also need the following lemma.
- Page 1: Invent. math. 126, 565-587 (1996)c
- Page 4 and 5: 568 I. Benjamini, O. Schramm1.4. De
- Page 6 and 7: 570 I. Benjamini, O. Schramm2. Nota
- Page 8 and 9: 572 I. Benjamini, O. Schramm3. Stab
- 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