Spectral Theory for Networks with Attractive and Repulsive Interactions

TL;DR

This paper explores the spectral properties of signed graph Laplacians in networks with both attractive and repulsive interactions, revealing topological constraints on eigenvalues and decomposing dynamics into fixed and free components.

Contribution

It establishes topological bounds on eigenvalue counts of signed Laplacians and introduces a homology-based framework for understanding network dynamics.

Findings

Eigenvalue counts are constrained by the network's topology.

Homology groups determine the rigidity of the Laplacian's index.

Numerical studies illustrate the theoretical results on large random matrices.

Abstract

There is a wealth of applied problems that can be posed as a dynamical system defined on a network with both attractive and repulsive interactions. Some examples include: understanding synchronization properties of nonlinear oscillator;, the behavior of groups, or cliques, in social networks; the study of optimal convergence for consensus algorithm; and many other examples. Frequently the problems involve computing the index of a matrix, i.e. the number of positive and negative eigenvalues, and the dimension of the kernel. In this paper we consider one of the most common examples, where the matrix takes the form of a signed graph Laplacian. We show that the there are topological constraints on the index of the Laplacian matrix related to the dimension of a certain homology group. In certain situations, when the homology group is trivial, the index of the operator is rigid and is determined only by the topology of the network and is independent of the strengths of the interactions. In general these constraints give upper and lower bounds on the number of positive and negative eigenvalues, with the dimension of the homology group counting the number of eigenvalue crossings. The homology group also gives a natural decomposition of the dynamics into "fixed" degrees of freedom, whose index does not depend on the edge-weights, and an orthogonal set of "free" degrees of freedom, whose index changes as the edge weights change. We also present some numerical studies of this problem for large random matrices.