Li-Yau inequality for unbounded Laplacian on graphs

DOI10.1016/J.AIM.2019.106822MaRDI QIDQ2335477zbMATH OpenOpenAlexFDO

Authors Chao Gong, Yong Lin, Shuang Liu, Shing-Tung Yau

Publication date 14 November 2019

Published in Advances in Mathematics (Search for Journal in Brave)

Full work available at URL https://arxiv.org/abs/1801.06021

complete weighted graph Dirichlet form eigenvalue estimate Harnack inequality heat kernel infinitesimal generator iterated gradient form Li-Yau inequality semigroup unbounded Laplacian

Mathematics Subject Classification ID

PDEs on graphs and networks (ramified or polygonal spaces) (35R02) Heat and other parabolic equation methods for PDEs on manifolds (58J35)

Abstract: In this paper, we derive Li-Yau inequality for unbounded Laplacian on complete weighted graphs with the assumption of the curvature-dimension inequality

C D E^{'} (n, K)

, which can be regarded as a notion of curvature on graphs. Furthermore, we obtain some applications of Li-Yau inequality, including Harnack inequality, heat kernel bounds and Cheng's eigenvalue estimate. These are first kind of results on this direction for unbounded Laplacian on graphs.

Review text

The authors prove a Li-Yau inequality for the unbounded Laplacian on a complete weighted graph. They consider an infinite graph \(G=(V,E)\) where \(V\) is the set of vertices and \(E\) that of edges. For a given measure on vertices defined as a function \(m:V\rightarrow \mathbb{R}^{+}\) and a weight on edges defined as a function \(\omega :E\rightarrow \mathbb{R}^{+}\) which is symmetric \(\omega _{xy}=\omega _{yx}\) for every edge \(\{x,y\}\), the authors define the space \(l_{m}^{p}\) of functions on \(V\) which are \(p\)-summable on \(V\) and the Dirichlet form \(Q\) with respect to \(l_{m}^{p}\) defined on \(D(Q)\times D(Q)\) as \(Q(f,g)=\frac{1}{2}\sum_{x,y\in V}\omega _{xy}(f(y)-f(x))(g(y)-g(x))\), where \(D(Q)\) is the completion of the space \( C_{0}(V)\) of finitely supported functions on \(V\) with respect to the \(Q\) -norm \(\left\Vert f\right\Vert _{Q}=\sqrt{\left\Vert f\right\Vert _{l_{m}^{2}}^{2}+\frac{1}{2}\sum_{x,y\in V}\omega _{xy}(f(y)-f(x))^{2}}\). They introduce the infinitesimal generator \(L\) of this Dirichlet form and the associated semigroup \(P_{t}f=e^{-tL}f\). In the case of a locally finite graph, the generator \(L\) coincides with the Laplacian \(\Delta \) defined on its domain \(D(L)=\{f\in l_{m}^{2}\mid Lf\in l_{m}^{2}\}\) through \(\Delta f(x)=\frac{1}{m(x)}\sum_{y\sim x}\omega _{xy}(f(y)-f(x))\), where \(y\sim x\) means that the vertices \(y\) and \(x\) are connected through an edge \(\{x,y\}\). The authors then introduce the gradient form \(\Gamma \), the iterated gradient form \(\Gamma _{2}\) and its modified version as \(\Gamma (f,g)(x)= \frac{1}{2m(x)}\sum_{y\sim x}\omega _{xy}(f(y)-f(x))(g(y)-g(x))(g(y)-g(x))\), \(\Gamma _{2}(f,g)(x)=(\Delta \Gamma (f,g)-\Gamma (f,\Delta g)-\Gamma (g,\Delta f)))(x)\) and \(\widetilde{\Gamma }_{2}(f)=\frac{1}{2}\Delta \Gamma (f)-\Gamma (f,\frac{\Delta (f^{2})}{2f})\), respectively. They say that a graph satisfies the \(CDE^{\prime }(n,k)\) condition if for any positive function \(f\in V^{\mathbb{R}}\), the set of real-valued functions on \(V\), \( \widetilde{\Gamma }_{2}(f)(x)\geq \frac{1}{n}f(x)^{2}(\Delta \log f)(x)^{2}+K\Gamma (f)(x)\) for every \(x\in V\). The first main result of the paper is a Li-Yau inequality. The authors assume that \(G=(V,E,m,\omega )\) is a complete graph with non-degenerate measure \(m\). They suppose that \(G\) satisfies \(CDE^{\prime }(n,K)\) for a real \(K\). They prove an upper bound for \(\frac{\Gamma (\sqrt{P_{t}f})}{P_{t}f}\) for every nonnegative \(f\in l_{m}^{p} \) with \(p\in \lbrack 1,\infty ]\) and \(b\geq 1\), in terms of \(\frac{\Delta P_{t}f}{P_{t}f}\) and \(t\). They then deduce the Harnack inequality \( P_{t}f(x)\leq P_{s}f(z)(\frac{s}{t})^{n}\exp (\frac{4m_{\max }d(x,z)^{2}}{ \omega _{\min }(s-t)})\) for every \(x,z\in V\), \(t<s\) and nonnegative \(f\in l_{m}^{p}\) with \(p\in \lbrack 1,\infty ]\). Here \(d(x,z)\) is the graph distance, \(m_{\max }=\sup_{x\in V}m(x)\) and \(\omega _{\min }=\inf_{x,y\in V,x\sim y}\omega _{xy}\). The authors then prove that \(P_{t}(\frac{\delta _{y} }{m(y)})(x)\leq \frac{C}{V(x,\sqrt{t})}\) for a constant \(C\) depending on \(n\). Under further assumptions of the graph \(G\), the authors finally prove an estimate on the greatest lower bound for the \(l^{2}\)-spectrum of the graph Laplacian (Cheng's eigenvalue): \(\lambda ^{\ast }\leq \frac{Kn}{2}\). For the proof of the Li-Yau inequality, the authors introduce the function \(\Phi (s)=-2\sum_{x\in V}\Gamma (\sqrt{P_{t-s}f+\epsilon })P_{s}\xi (x)m(x)\) for \( f,\xi \in C_{0}(V)\), \(\epsilon >0\) and \(s\in \lbrack 0,t]\). They prove that this function is differentiable in \(s\in (0,t)\) and they compute its derivative. They also prove a Green formula in the present context and properties of the semigroup \(P_{t}\), of the gradient form \(\Gamma \) and of the Laplacian. The proofs of the applications are obtained through direct computations.

Recommendations

Cites work

Cited in

(26)

This page was built for publication: Li-Yau inequality for unbounded Laplacian on graphs

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2335477)