STOC2023
Cheeger Inequalities for Directed Graphs and Hypergraphs using Reweighted Eigenvalues
Lap Chi Lau, Kam Chuen Tung, Robert Wang
1 citation
Abstract
We derive Cheeger inequalities for directed graphs and hypergraphs using the reweighted eigenvalue approach that was recently developed for vertex expansion in undirected graphs [OZ22, KLT22, JPV22]. The goal is to develop a new spectral theory for directed graphs and an alternative spectral theory for hypergraphs. The first main result is a Cheeger inequality relating the vertex expansion ψ(G) of a directed graph G to the vertex-capacitated maximum reweighted second eigenvalue λ v * 2 that where ∆ is the maximum degree of G. This provides a combinatorial characterization of the fastest mixing time of a directed graph by vertex expansion, and builds a new connection between reweighted eigenvalued, vertex expansion, and fastest mixing time for directed graphs. The second main result is a stronger Cheeger inequality relating the edge conductance φ(G) of a directed graph G to the edge-capacitated maximum reweighted second eigenvalue λ e * 2 that λ e * 2 φ(G) λ e * 2 • log