Every salami has two ends (Q7006052)

This is a paper about connected locally finite graphs \(G=(V,E)\) with an edge weight function \(w: V\times V\rightarrow [0,\infty)\) which is symmetric and zero on the diagonal, with edges being the pairs \(x\), \(y\) with positive \(w(x,y)\) (we write \(x\sim y\) for an edge \(xy\)), and a volume function \(m: V\rightarrow [0,\infty)\). All graphs will be endowed with a path metric \(d:V^{2}\rightarrow [0,\infty)\) such that \(d(x,y)\) is given by \(\inf \{\sum_{i=1}^{n}d(x_{i},x_{i-1}): x=x_{0}\sim x_{1}\sim x_{2}\sim \ldots \sim x_{n}=y\}\). We always assume all balls of the metric \(d\) are finite, which is equivalent to \(d\) being complete by the discrete Hopf-Rinow theorem. We write a graph as \(G=(V,w,m)\) or \(G=(V,w,m,d)\) if emphasising the choice of path metric. The aim of the paper is to define a special class of such graphs called salamis and prove some structural results about them.\N\NSuch a graph is said to be a salami if it has at least two ends of infinite volume and non-negative Ollivier curvature. Here, a graph is said to have at least \(n\) ends if there is a finite set \(K\subset V\) with \(V\setminus K\) having at least \(n\) infinite connected components, and to have exactly \(n\) ends if it has at least \(n\) components but does not have at least \(n+1\) components. Very roughly, an end is a direction in which the graph extends to infinity.\N\NTo define Ollivier curvature, given a graph \((G,w,m)\) we can define a Laplace operator \(\Delta: \mathbb{R}^{V}\rightarrow \mathbb{R}^{V}\) given by \N\[\N\Delta f(x)=\frac{1}{m(x)}\sum_{y}w(x,y)(f(y)-f(x))\N\]\Nand then if \(R=d(x,y)>0\) we define the Ollivier curvature of \((G,w,m,d)\) to be \N\[\N\kappa(x,y)=\frac{1}{R}\inf_{f(x)-f(y)=R, f\in \text{Lip}(1)} \Delta f(x)-\Delta f(y).\N\]\NHere, \(\mathrm{Lip}(1)=\{f\in \mathbb{R}^{V}: f(y)-f(x)\leq d(x,y)\}\). Note these definitions depend on the choice of path metric. Perhaps more helpful for those new to these ideas is the intuition that positive Ollivier curvature \(\kappa(x,y)>0\) implies that the neighbourhoods of \(x\) and \(y\) are close or overlapping, whereas negative Ollivier curvature would suggest the neighbourhoods are further apart than typically.\N\NThe main result is that every salami has exactly two ends, that it is flat (i.e., \(\inf_{y\sim x}\kappa(x,y)=0\) for all \(x\in V\)), that it is recurrent (i.e., for every \(\epsilon>0\) there is \(f\in \mathbb{R}^{V}\), zero for all but finitely many \(x\), such that \(f(x_{0})=1\) for some fixed \(x_{0}\in V\) and \(-\langle f, \Delta f \rangle < \epsilon\) where the inner product \(\langle f,g \rangle=\sum_{v\in V}f(x)g(x)m(x)\)), and that there is a harmonic function (i.e., one for which \(\Delta f=0\)) with constant gradient. The authors give a nice summary of the proof after this result and note that non-trivially more can be said if \(\inf_{x\in V}m(x)>0\) or if there is \(\epsilon>0\) such that \(w(x,y)>\epsilon\) for all \(x\sim y\). In the latter case, further results are proved about the space of harmonic functions (i.e., those with \(\Delta f=0\)), for example, that the space of harmonic functions is finite-dimensional and that its subspace of subexponentially growing harmonic functions has dimension 2.