Universal lower bound for community structure of sparse graphs

TL;DR

This paper establishes a universal lower bound on the modularity of sparse graphs, showing that many graphs with certain degree sequences inherently have high community structure, regardless of randomness.

Contribution

It introduces a new lower bound on graph modularity applicable to various degree sequences, extending understanding of community structure in sparse graphs.

Findings

Modularity of graphs with average degree  is ^{-1/2}

High modularity applies to graphs with power-law or near-regular degree sequences

High modularity can occur in graphs with degree sequences similar to Erds-Re9nyi graphs

Abstract

We prove new lower bounds on the modularity of graphs. Specifically, the modularity of a graph $G$ with average degree $\bar d$ is $\Omega(\bar{d}^{-1/2})$, under some mild assumptions on the degree sequence of $G$. The lower bound $\Omega(\bar{d}^{-1/2})$ applies, for instance, to graphs with a power-law degree sequence or a near-regular degree sequence. It has been suggested that the relatively high modularity of the Erd\H{o}s-R\'enyi random graph $G_{n,p}$ stems from the random fluctuations in its edge distribution, however our results imply high modularity for any graph with a degree sequence matching that typically found in $G_{n,p}$. The proof of the new lower bound relies on certain weight-balanced bisections with few cross-edges, which build on ideas of Alon [Combinatorics, Probability and Computing (1997)] and may be of independent interest.