The Spectral Geometry of the Mesh Matrices of Graphs

Abstract

The mesh matrix $Mesh(G,T_0)$ of a connected finite graph $G=(V(G),E(G))=(vertices, edges) \ of \ G$ of with respect to a choice of a spanning tree $T_0 \subset G$ is defined and studied. It was introduced by Trent \cite{Trent1,Trent2}. Its characteristic polynomial $det(X \cdot Id -Mesh(G,T_0))$ is shown to equal $\Sigma_{j=0}^{N} \ (-1)^j \ ST_{j}(G,T_0)\ (X-1)^{N-j} \ (\star)$ \ where $ST_j(G,T_0)$ is the number of spanning trees of $G$ meeting $E(G-T_0)$ in j edges and $N=|E(G-T_0)|$. As a consequence, there are Tutte-type deletion-contraction formulae for computing this polynomial. Additionally, $Mesh(G,T_0) -Id$ is of the special form $Y^t \cdot Y$; so the eigenvalues of the mesh matrix $Mesh(G,T_0)$ are all real and are furthermore be shown to be $\ge +1$. It is shown that $Y \cdot Y^t$, called the mesh Laplacian, is a generalization of the standard graph Kirchhoff Laplacian $\Delta(H)= Deg -Adj$ of a graph $H$.For example, $(\star)$ generalizes the all minors matrix tree theorem for graphs $H$ and gives a deletion-contraction formula for the characteristic polynomial of $\Delta(H)$. This generalization is explored in some detail. The smallest positive eigenvalue of the mesh Laplacian, a measure of flux, is estimated, thus extending the classical inequality for the Kirchoff Laplacian of graphs.