The reduced `p -cohomology in degree 1 and harmonic functions Antoine Gournay University of Warwick, May 21st 2015 A.Gournay (TU Dresden) 1 `p H & harmonic functions 2015/05/21 1 / 30 Why `p -cohomology? A ball packing in a manifold M is a [countable] set of closed balls in M so that any two balls intersect at most in a point. The incidence graph of a ball packing is a graph whose vertices are the balls and there is an edge if the ball touches. Theorem (Koebe 1936) A finite graph can be packed in R2 if and only if it is planar. Quasi-round packing: replace balls by generic domains, require there is a K so that the ration “outer radius / inner radius” is K . There is no obstruction for quasi-round packings of finite graphs in A.Gournay (TU Dresden) 1 `p H & harmonic functions 2015/05/21 R3. 2 / 30 Why `p -cohomology? Theorem (Koebe 1936) A finite graph can be packed in R2 if and only if it is planar. There is no obstruction for quasi-round packings of finite graphs in Theorem (Benjamini & Schramm 2009) R3. R If an infinite graph can be quasi-roundly packed in d either it is d-parabolic or it has non-trivial reduced `d -cohomology in degree 1. () fk k j g d-parabolic inf ∇f `p f has finite support and f (x0 ) = 1 = 0. “Easy” to understand, e.g. 2-parabolicity is recurrence. A Cayley graph is d-parabolic if and only if it has polynomial growth of degree d. f j g In general, dpar = inf d d-parabolic belongs to [disop ; dgr ] where disop is the isoperimetric dimension (see later) and dgr the minimal polynomial degree growth of balls. A.Gournay (TU Dresden) 1 `p H & harmonic functions 2015/05/21 3 / 30 What is `p -cohomology? In degree 1, the `p -cohomology of a graph G = (V ; E ) is defined via incidence operators between vertices and edges. Take E V V symmetric, and let ∇: V E f ∇f (x ; y ) = f (y ) f (x ) f ! Rg ! f ! Rg 7! In graphs of bounded valency, ∇ : `p (V ) ! p ` (E ) is a bounded operator. f ! R j ∇f 2 p (E )g. It is endowed with a semi-norm kf kD = k∇f k . (“semi-” ! constant The space of p-Dirichlet functions is Dp (G) = f : V p ` p ` functions). A.Gournay (TU Dresden) 1 `p H & harmonic functions 2015/05/21 4 / 30 What is `p -cohomology? ∇: fV ! Rg ! fE ! Rg f 7! ∇f (x y ) = f (y ) f (x ) ; kf kD p = k∇f k p ` Definition The reduced `p -cohomology in degree 1 of a graph is p ` H (G ) = 1 Theorem (Élek 1998, Pansu Im ∇ \ `p (E ) p ` ∇`p (V ) Dp (G) = p ` (V ) + cst Dp ?) Fix a bound on the geometry (valency, curvature and injectivity radius). Then the [reduced] `p -cohomology [in degree 1] is an invariant of quasi-isometry. A.Gournay (TU Dresden) 1 `p H & harmonic functions 2015/05/21 5 / 30 A simple (yet important) example. p ` H p (G) = Dp (G)=`p (V ) + RD p (G) Example: g: 0 0 1 1 v 1 1 1 1 gn : 0 0 1 n 1 n 2 n 1 n 0 0 v v v v v v gn finitely supported so ∇(g v v 2 v p ` v v v v v (V ) for any p. gn ) takes n times the value 1=n A.Gournay (TU Dresden) v =) kg 1 `p H & harmonic functions gn kD p = (n=np )1 p = n = 2015/05/21 1=p0 . 6 / 30 A simple (yet important) example. p ` H p (G) = Dp (G)=`p (V ) + RD p (G) Example: g: 0 0 1 1 v 1 1 1 1 gn : 0 0 1 n 1 n 2 n 1 n 0 v 0 p H 1 (G). v v v v v =) kg gn kD v p D ! g if p A.Gournay (TU Dresden) v v = (n=np )1 p = n = p Then gn v > 1=p0 v v v v . 8 1. Thus 1 < p < ∞; [g ] = 0 1 v `p H & harmonic functions 2 ` 2015/05/21 6 / 30 A simple (yet important) example. p ` H p (G) = Dp (G)=`p (V ) + RD p (G) Example: g: 0 0 1 1 v 1 1 1 1 gn : 0 0 1 n 1 n 2 n 1 n 0 0 v v v v v v v v v fg A.Gournay (TU Dresden) v v v v v 2]1 ∞[ and 1H 1(G) ' R. Id q, the map p H 1 (G) ! q H 1 (G) is not always injective... In fact `p H 1 (G) = 0 if p Remark: If p < v ; ` ` ` 1 `p H & harmonic functions 2015/05/21 6 / 30 Ends 1 ` H (G) is intimately related to the ends of a graph. 1 Definition (Freudenthal, 193?) An end of a graph = (V ; E ) is a function from finite subsets of V to infinite ones, such that ξ(F ) is an infinite connected component of F c ; F ; F 0 V (finite), ξ(F ) ξ(F 0 ) = . 8 \ 6 ? Examples: A finite graph has 0 ends. Z2) has 1 end. The infinite line (a Cayley graph of Z) has 2 ends. Regular trees of even valency 3 (Cayley graphs of free groups) have ∞ The infinite grid (a Cayley graph of many ends. A.Gournay (TU Dresden) 1 `p H & harmonic functions 2015/05/21 7 / 30 Ends Lemma The number of ends is a quasi-isometry invariant. Theorem (Hopf, 1944) The number of ends of a Cayley graph is 0, 1, 2 or ∞. Idea: 3 ends =) ∞ ends Theorem (Stallings, 1971) Z [The Cayley graph of] a group has 2 ends iff it contains as a finite index subgroup. It has ∞ many ends iff it is a “non-trivial” amalgamated product or HNN extension. A.Gournay (TU Dresden) 1 `p H & harmonic functions 2015/05/21 8 / 30 Ends and `1 H 1 1 ` H (G) = D1 (G)=`1 (V ) + 1 Lemma (“well-known”) Rends G If G has finitely many ends, `1 H (G) = 1 Preliminary claim: D1 (G) A.Gournay (TU Dresden) RD ( ) 1 1 (G) . ∞ (V )...? ` 1 `p H & harmonic functions 2015/05/21 9 / 30 Ends and `1 H 1 1 ` H (G) = D1 (G)=`1 (V ) + 1 Lemma (“well-known”) Rends G If G has finitely many ends, `1 H (G) = 1 Preliminary claim: D1 (G) Hint: f (y ) RD ( ) 1 1 (G) . ∞ (V )...? ` f (x ) = ∑e2P ∇f (e) for P a path from x to y . A.Gournay (TU Dresden) 1 `p H & harmonic functions 2015/05/21 9 / 30 Ends and `1 H 1 1 ` H (G) = D1 (G)=`1 (V ) + 1 Lemma (“well-known”) Rends G If G has finitely many ends, `1 H (G) = 1 jf (y ) f (x ) RD ( ) 1 1 (G) . j = ∑ ∇f (e) k∇f k 1 (E ) : ` 2 e P Shows more: the `1 norm of ∇f tends to 0 outside larger and large balls. On the [infinite] connected components of Bnc (Bn vertex), f becomes uniformly constant as n ∞. ! A.Gournay (TU Dresden) 1 `p H & harmonic functions = balls centred at some 2015/05/21 9 / 30 Ends and `1 H 1 1 ` H (G) = D1 (G)=`1 (V ) + 1 Lemma (“well-known”) Rends G If G has finitely many ends, `1 H (G) = 1 jf (y ) f (x ) RD ( ) 1 1 (G) . j = ∑ ∇f (e) k∇f k 1 (E ) : ` 2 e P Shows more: the `1 norm of ∇f tends to 0 outside larger and large balls. On the [infinite] connected components of Bnc (Bn vertex), f becomes uniformly constant as n ∞. ! = balls centred at some 2 D1(G) on each end: βf (ξ) := nlim !∞ f (xn ) where xn 2 ξ(Bn ) Thence, one defines a value of f A.Gournay (TU Dresden) 1 `p H & harmonic functions 2015/05/21 9 / 30 Ends and `1 H 1 1 ` H (G) = D1 (G)=`1 (V ) + 1 Lemma (“well-known”) Rends G If G has finitely many ends, `1 H (G) = 1 jf (y ) f (x ) ( ) 1 1 (G) . j = ∑ ∇f (e) k∇f k 2 1 (E ) : ` e P 2 D1(G): Boundary value of f RD βf (ξ) := nlim !∞ f (xn ) where xn 2 ξ(Bn ) β is linear and continuous on D1 (G). Also: 1 ` (V ) Ker β =) β sends `1 (V ) + cst D1 (G) A.Gournay (TU Dresden) 1 to constant functions. `p H & harmonic functions 2015/05/21 9 / 30 Ends and `1 H 1 1 ` H (G) = D1 (G)=`1 (V ) + 1 Lemma (“well-known”) Rends G If G has finitely many ends, `1 H (G) = 1 RD ( ) 1 1 (G) . We have a [linear & continuous] map β which associates to g function on the ends. 1 ` (V ) + R D1 (G) is sent to constant functions. Remains to show that if βg is constant, then g A.Gournay (TU Dresden) 2 D1(G) a 1 `p H & harmonic functions 2 1 ` (V ) + R D1 (G) . 2015/05/21 9 / 30 Truncation Lemma 1 Definition Say g : V ! R takes only one value at infinity if there is a K 2 R so that for any ε > 0 one can find a Fε V finite so that g (Fεc ) ]K ε; K + ε[: Lemma (“maximum principle revisited”) If g 2 Dp (G) takes only one value at infinity then [g ] = 0 2 Proof: WLOG K gε ( x ) = k g (x ) εg (x )= g (x ) j gε is finitely supported, so k H 1 (G). = 0. Define gε as fε = g p ` k g fε Dp = ∇gε 0; so tends to 0. A.Gournay (TU Dresden) k p (E ) ` 2 p ` j if g (x ) < ε if g (x ) ε (V ) for any p. is pointwise bouded by ∇g and tends pointwise to 1 `p H & harmonic functions 2015/05/21 10 / 30 Aim ... ... to describe `p H 1 as an ideal boundary. 1 ` 1 H is related to the ends [“well-known”]. [Bourdon & Pajot 2003]: In the hyperbolic case, there a strong link between -cohomology in degree 1 and some [Besov] space of functions on the visual boundary; also pc = inf p `p H 1 = 0 is a lower bound on the conformal dimension. p ` f j 6 g A natural idea, is to try to look at the “values” of this function on the “Poisson boundary”. A.Gournay (TU Dresden) 1 `p H & harmonic functions 2015/05/21 11 / 30 Questions and answers Question (Gromov 1992) If G is the Cayley graph of an amenable group is ]1; ∞[? [In fact, in all degrees] p ` fg H 1 (G) = 0 for any p 2 Theorem (Gromov 1992, ...) If G is the Cayley graph of a virtually nilpotent group then any p ]1; ∞[. 2 p ` fg H 1 (G) = 0 for Holds for any group with [virtually] infinite center. [Tessera 2009] show this also holds for polycyclic groups and some more. A.Gournay (TU Dresden) 1 `p H & harmonic functions 2015/05/21 12 / 30 Isoperimetric profiles Definition 2Z 9 Let d 1 . A graph G satisfies a d-dimensional isoperimetric profiles (noted ISd ) if K > 0 such that, F V finite, 8 jF j1 d K j∂F j 1 9 It has a strong isoperimetric profile (noted ISω ) if K finite, F K ∂F j j j j Examples: Cayley graphs of > 0 such that, 8F V Zd satisfy ISd . A group is amenable iff its Cayley graph does not satisfy ISω . (Restatement of Følner) A.Gournay (TU Dresden) 1 `p H & harmonic functions 2015/05/21 13 / 30 Isoperimetric profiles Satisfying ISα (for α Hyperbolic 2 Z1 [fωg) is invariant under quasi-isometries. =) ISω =) ISd ; 8d. 8 6 ) ISω. For example, Cayley graphs of Z2 oα Z where But ISd ; d = α(1) = 21 11 . Theorem (Varopoulos 1985 + Gromov 1981 + Wolf 1968) has polynomial growth of degree d () Cay( ; S ) does not have ISd +1 . =) groups which are not virtually nilpotent have ISd for all d. A.Gournay (TU Dresden) 1 `p H & harmonic functions 2015/05/21 14 / 30 Harmonic functions (n ) (n ) Let Px be the measure defined by Px (y ) = the probability that a simple random walker starting at x ends up in y after n steps. This gives a kernel: P (n) g (x ) := A function g : V R (n ) g (y )dPx (y ). ! R is harmonic if P 1 g = g (mean value property). ( ) H (G) :=space of harmonic functions. Hb (G) := H (G) \ `∞ (V ) =space of bounded harmonic functions. A.Gournay (TU Dresden) 1 `p H & harmonic functions 2015/05/21 15 / 30 Boundary values Theorem (G., 2013) Assume G satisfies ISd . Let 1 Dp (G) H (G) such that [ 1p d 2 < = ! p < d =2. There is a linear map π : 2 Dp (G), then π(g ) 2 Dq (G) for all q d dp2p . if g 2 Dp (G) \ ∞ (V ), then π(g ) 2 ∞ (V ). [g ] = 0 2 p H 1 (G) () π(g ) is a constant function () g takes only one value at ∞. if g > ` ` ` This extends a result of Lohoué (1990) which requires ISω . A.Gournay (TU Dresden) 1 `p H & harmonic functions 2015/05/21 16 / 30 Truncation Lemma 2 Lemma (Holopainen & Soardi 1994) 2 6 2 Assume g Dp (G) is such that [g ] = 0 `p H 1 (G). Let gK be the function with values truncated in [ K ; K ]. Then for some K0 and any K > K0 , [gK ] = 0 p 1 ` H (G ). 6 2 Proof goes along the same lines as Truncation Lemma 1. Corollary fg To show `p H 1 (G) = 0 , it suffices to show [g ] = 0 for any g A.Gournay (TU Dresden) 1 `p H & harmonic functions 2 Dp (G) \ 2015/05/21 ∞ (V ). ` 17 / 30 Corollaries Theorem (G., 2013) Assume G satisfies ISd . Let 1 Dp (G) H (G) such that [ 1p d 2 < = ! p < d =2. There is a linear map π : 2 Dp (G), then π(g ) 2 Dq (G) for all q d dp2p . if g 2 Dp (G) \ ∞ (V ), then π(g ) 2 ∞ (V ). [g ] = 0 2 p H 1 (G) () π(g ) is a constant function () g takes only one value at ∞. if g > ` ` ` Corollary 1 fg Indeed, π(g ) = 0, 8g 2 Dp (G) \ fg If G satisfies ISd and Hb (G) = 0 , then `p H 1 (G) = 0 for all p ∞ (V ), ` 2 [1 ; d 2 [. so Theorem gives [g ] = 0. Truncation Lemma 2 allows to conclude. A.Gournay (TU Dresden) 1 `p H & harmonic functions 2015/05/21 18 / 30 Corollaries Theorem (G., 2013) Assume G satisfies ISd . Let 1 Dp (G) H (G) such that [ 1p d 2 < = ! p < d =2. There is a linear map π : 2 Dp (G), then π(g ) 2 Dq (G) for all q d dp2p . if g 2 Dp (G) \ ∞ (V ), then π(g ) 2 ∞ (V ). [g ] = 0 2 p H 1 (G) () π(g ) is a constant function () g takes only one value at ∞. if g > ` ` ` Corollary 2 If G is the Cayley graph of a group Then the identity map `p H 1 (G) ! Z which is not virtually- and 1 1 ` H (G ) is injective. q fg fg p < q < ∞. [Cheeger-Gromov 1992] show `2 H (G) = 0 for amenable groups = for amenable groups, `p H 1 (G) = 0 for all 1 < p 2. 1 ) A.Gournay (TU Dresden) 1 `p H & harmonic functions 2015/05/21 18 / 30 Corollaries Theorem (G., 2013) Assume G satisfies ISd . Let 1 Dp (G) H (G) such that [ 1p d 2 < = ! p < d =2. There is a linear map π : 2 Dp (G), then π(g ) 2 Dq (G) for all q d dp2p . if g 2 Dp (G) \ ∞ (V ), then π(g ) 2 ∞ (V ). [g ] = 0 2 p H 1 (G) () π(g ) is a constant function () g takes only one value at ∞. if g > ` ` ` Corollary 3 \ f g ) 8q d dp2p p H 1(G) = f0g p H 1(G) = f0g =) H (G) \ Dp (G) = fconstantsg If G has ISd and 1 p < d =2, then Hb (G) Dp (G) = constants = < + ;` : ` This uses that π is the identity on harmonic functions. A.Gournay (TU Dresden) 1 `p H & harmonic functions 2015/05/21 18 / 30 Corollaries Theorem (G., 2013) Assume G satisfies ISd . Let 1 Dp (G) H (G) such that [ 1 p<d =2 ! p < d =2. There is a linear map π : 2 Dp (G), then π(g ) 2 Dq (G) for all q d dp2p . if g 2 Dp (G) \ ∞ (V ), then π(g ) 2 ∞ (V ). [g ] = 0 2 p H 1 (G) () π(g ) is a constant function () g takes only one value at ∞. if g > ` ` ` Corollary 4 fg Assume G0 G is a connected spanning subgraph of G so that G0 has ISd and p 1 0 ` H (G ) = 0 for 1 p < d =2. `p H 1 (G) = 0 . fg If g 2 Dq (G) then g 2 Dq (G0 ). This implies g takes only one value at ∞ as seen on G0 . But this is also the case on G. A.Gournay (TU Dresden) 1 `p H & harmonic functions 2015/05/21 18 / 30 Corollaries Corollary 4 fg Assume G0 G is a connected spanning subgraph of G so that G0 has ISd and 0 p 1 p < d =2. `p H 1 (G) = 0 . ` H (G ) = 0 for 1 fg Corollary 5 6 fg Let L = and has at most two ends and H is infinite and has most two ends. Then the lamplighter L H has no harmonic function with gradient in `p (p < ∞). o [Thomassen 1978] shows that up to a quasi-isometry (actually bi-Lipschitz map of constant 6) both L and H contain a connected spanning subgraph which is a line, half-line or cycle. Hence they contain a of G0 = H 0 L0 where H 0 = Cn ; or and L0 = or . This G0 has Hb (G0 ) = 0 and ISd for all d. N Z N Z fg o ) gradient in ! These graphs have lots of bounded harmonic functions ( = ∞ ). ` Any Cayley graph has harmonic functions with gradient in `∞ (= Lipschitz). A.Gournay (TU Dresden) 1 `p H & harmonic functions 2015/05/21 18 / 30 Corollaries Corollary 4 fg Assume G0 G is a connected spanning subgraph of G so that G0 has ISd and p 1 0 ` H (G ) = 0 for 1 p < d =2. `p H 1 (G) = 0 . fg This can also be interpreted as a “forbidden subgraph” result: if H 1 (G) = 0 then there are no spanning connected subgraphs G0 with Hb (G0 ) = constants and ISd (for some d > 2p). p ` 6 fg f g Zd for d e.g. no spanning is a spanning line! Corollary 6 > 2p... Recall: [Thomassen 1978] very often there 6 fg Assume G has ISd and `p H 1 (G) = 0 for 1 p the Poisson boundary is quasi-isometry invariant < d =2. Then some part of Both ISd and `p H 1 are QI-invariant, so always get a non-trivial bounded harmonic function by looking at π(g ) where [g ] = 0 `p H 1 and g bounded. 6 A.Gournay (TU Dresden) 1 `p H & harmonic functions 2 2015/05/21 18 / 30 The proof Theorem (G., 2013) Assume G satisfies ISd . Let 1 Dp (G) H (G) such that [ 1 p<d =2 ! p < d =2. There is a linear map π : 2 Dp (G), then π(g ) 2 Dq (G) for all q d dp2p . if g 2 Dp (G) \ ∞ (V ), then π(g ) 2 ∞ (V ). [g ] = 0 2 p H 1 (G) () π(g ) is a constant function () g takes only one value at ∞. if g > ` ` ` n π is naively defined: π(g ) = nlim !∞ P g. The “difficult” parts are: π is [well-]defined. π(g ) is a constant implies g takes only one value at ∞. A.Gournay (TU Dresden) 1 `p H & harmonic functions 2015/05/21 19 / 30 π defined, take one P (n) g (x ) Cauchy? ! P n g (x ) ( ) P (m) g (x ) = R (n) gdPx R (m) gdPx . Problem/Idea ? R gdξ R gdφ = hg j ξ φi =??? = h∇g j??i k∇g k p k??k p ` A.Gournay (TU Dresden) 1 `p H & harmonic functions ` 0 2015/05/21 20 / 30 Duality ∇ the adjoint of ∇: ∇: Note: fE ! Rg ! fV ! Rg f 7! ∇f (x ) = ∑y x f (y () (I = ∇ ∇ then g = 0 P )g ; x) ∑y x f (x ; y ) = 0. Definition ξ; φ finitely supported probability measures. A transport pattern from ξ to φ is a finitely supported function τξ;φ : E ! R so that ∇ τξ;φ = φ ξ. A.Gournay (TU Dresden) 1 `p H & harmonic functions 2015/05/21 21 / 30 π defined, take two P (n) g (x ) Cauchy? ! P n g (x ) ( ) P (m) g (x ) = R (n) gdPx R (m) gdPx . Problem/Idea ? R gdξ R gdφ = hg j ξ φi = hg j ∇ τφ ξ i = h∇g j τφ ξ i ; jj k∇g k kτφ ξk ; p ` A.Gournay (TU Dresden) 1 `p H & harmonic functions ; p0 ` 2015/05/21 22 / 30 π defined, take two P (n) g (x ) Cauchy? ! P n g (x ) ( ) P (m) g (x ) = R (n) gdPx R (m) gdPx . Problem/Idea ? P (m+k ) g (x ) P (n ) g ( x ) = hg j Px(m+k ) Px(m) i = hg j ∇ τP (m) P (m+k ) i x x = h∇g j τP (m) P (m+k ) i ; jj k∇g k kτP x p ` How to define τ (m) Px A.Gournay (TU Dresden) (m+k ) Px ; ; x (m) x (m+k ) Px ; k p0 ` ? 1 `p H & harmonic functions 2015/05/21 22 / 30 π defined, take two P (n) g (x ) Cauchy? ! P n g (x ) ( ) P (m) g (x ) = R (n) gdPx R (m) gdPx . Problem/Idea ? P (m+k ) g (x ) P (n ) g ( x ) = hg j Px(m+k ) Px(m) i = hg j ∇ τP (m) P (m+k ) i x x = h∇g j τP (m) P (m+k ) i ; jj k∇g k kτP x p ` How to define τ (m) Px (m+k ) Px ; ; x (m) x (m+k ) Px ; k p0 ` ? ∇ is linear so (a possible choice is): τP (m) ;P (m+k ) = x A.Gournay (TU Dresden) x 1 k 1 ∑ τP i =0 (m+i ) x (m++i +1) Px ; `p H & harmonic functions 2015/05/21 22 / 30 Simple but inefficient... How to define τ (m) Px (m+1) Px ; ? There is a very natural transport pattern: take a random step! kP Then τ (m) x (m+1) Px ; A.Gournay (TU Dresden) k K kPxm k p0 ( ) ` p0 ` 1 (where K depends on the valency). `p H & harmonic functions 2015/05/21 23 / 30 Transport: to infinity and beyond! jP m ( +k ) g (x ) P (n) g (x ) j k∇g k kτP p ` m +k 1 (m) x (n) So it suffices to check that ∑n0 Px (m+k ) Px ; k k∇g k ∑ kPxi k p ` p ` 0 i =m () p0 ` is in `p (V ). 0 Theorem (Varopoulos 1985) If G has ISd , then for some K > k (n) 0, Px k ∞ (V ) ` Kn d =2 . Thus kPxn kq ( ) q (V ) ` kPxn kq 1V kPxn k ( ) ( ) ∞( ) ` 1 (V ) ` K 0n d (q 1)=2 and ∑ P (n) converges in `q if q 0 < d =2. A.Gournay (TU Dresden) 1 `p H & harmonic functions 2015/05/21 24 / 30 π(g ) =cst ) constant at ∞ = Want to prove: π(g ) is a constant function = ) g is constant at infinity. WLOG the constant π(g ) is 0. Will prove: 8ε > 9 0, nε such that x 2 B3n =) jg (x )j = ε < K ε. Make a well-chosen splitting of the scalar product: h j P (n) g (x ) A.Gournay (TU Dresden) g (x ) = ∇g τ 1 `p H & harmonic functions (n) δx ;Px i 2015/05/21 25 / 30 π(g ) =cst from π(g ) ) constant at ∞ = 0, to prove: for some K 0 8ε 0, 9nε such that x 2 B3n =) jg (x )j > > = ε < K ε. Given ε > 0, define nε to be so that ∇g Bncε < ε k k n sup ∑i n kPx k x 2V ( ) ε P (n ) g ( x ) p0 < ` ε g (x ) = h∇g j τδ P (n) i x x = h∇g j τδ P (n) ijBnε + h∇g j τδ ; x; x (n) x ;Px ijB c nε k k 2nd term: ∇g is small, and there is an absolute bound on τ: ∑i 0 P (i ) 1st term: ∇g is not small, but the τ will be small: can replace τ (n) by τ (n ) (n) because it takes at least nε steps to go δx ;Px P ε ;Px from x to Bnε . A.Gournay (TU Dresden) 1 `p H & harmonic functions 2015/05/21 26 / 30 π(g ) =cst from π(g ) ) constant at ∞ = 0, to prove: for some K 0 8ε 0, 9nε such that x 2 B3n =) jg (x )j > > = ε < K ε. Given ε > 0, define nε to be so that ∇g Bncε < ε k k n sup ∑i n kPx k x 2V ( ) p0 < ` ε ε Massage 2nd term: jP n g (x ) ( ) j jh∇g j τδ jh∇g j τδ jh∇g j τδ jh∇g j τδ g (x ) (n) x ;Px (n) x ;Px (n) x ;Px (n) x ;Px A.Gournay (TU Dresden) 1 ijB ijB ijB ijB nε nε nε nε j + jh∇g j τδ P ijB j j + k∇g k B kτδ P k B i j+ ε K1 ∑i 0 kPx k j+ ε K2 `p H & harmonic functions (n) x; x p( c ) nε ` c nε (n) x; x p0 ( c ) nε ` () p0 ` 2015/05/21 27 / 30 ) constant at ∞ π(g ) =cst from π(g ) = 0, to prove: for some K 0 8ε 0, 9nε such that x 2 B3n =) jg (x )j > > = ε < K ε. Given ε > 0, define nε to be so that ∇g Bncε < ε k k n sup ∑i n kPx k x 2V ( ) p0 < ` ε ε Massage 1st term: jP n g (x ) ( ) j jh∇g j τδ P ijB j jh∇g j τP P ijB j k∇g k kτP P k k∇g k ∑i n kPxi k k∇g k ε g (x ) +εK2 +εK2 (nε ) (n) nε x x (n) (nε ) p (B ) +εK2 nε (n) nε x; x ; p ` ; x p ` p ` A.Gournay (TU Dresden) 1 `p H & harmonic functions ε x ` () 0 p0 ` +εK2 +εK2 2015/05/21 28 / 30 π(g ) =cst from π(g ) ) constant at ∞ = 0, to prove: for some K 0 8ε 0, 9nε such that x 2 B3n =) jg (x )j > > = ε < K ε. Given ε > 0, define nε to be so that ∇g Bncε < ε k k n sup ∑i n kPx k x 2V Up to now: 8x 2 B3n , ( ) p0 < ` ε = ε ε jP n g (x ) ( ) Letting n ! ∞: j K3 ε g (x ) jg (x )j K3ε as claimed. A.Gournay (TU Dresden) 1 `p H & harmonic functions 2015/05/21 29 / 30 Some questions Q1: Is there an amenable group with a non-constant (bounded or not) harmonic functions whose gradient is in c0 ? Q2: If G is the Cayley graph of an amenable group (with Hb (G) = constants , what is the maximal d so that G has a connected spanning subgraph G0 with ISd and Hb (G0 ) = constants ? 6 f g f g (n) Q3.a: Is there an explicit and more efficient transport pattern from δx to Px ? Q3.b: In an amenable Cayley graph, is there a Følner sequence and an explicit (not too unefficient) transport pattern from δe to 1lFn ? Q.4: For two infinite connected sets A and B let nA;B (`) = number of mutually disjoint paths of length ` from A to B : Estimates for the growth (in `) of nA;B (`) in an amenable Cayley graph (of exponential growth) A.Gournay (TU Dresden) 1 `p H & harmonic functions 2015/05/21 30 / 30