A Center Manifold Reduction Technique for a System of Randomly Coupled Oscillators

TL;DR

This paper introduces a novel technique to analyze the statistical properties of high-dimensional, nonhyperbolic dynamical systems with random connectivity, focusing on their center manifolds and reduced dynamics, exemplified on a network of oscillators.

Contribution

It develops a method to study the statistical characteristics of nonhyperbolic systems with random connections and constrains possible center manifold models in large-scale networks.

Findings

The technique reveals how statistical properties influence the shape of the center manifold.

It constrains the family of possible reduced models for large random networks.

Demonstrated on a network of randomly coupled damped oscillators.

Abstract

In dynamical systems theory, a fixed point of the dynamics is called nonhyperbolic if the linearization of the system around the fixed point has at least one eigenvalue with zero real part. The center manifold existence theorem guarantees the local existence of an invariant subspace of the dynamics, known as a center manifold, around such nonhyperbolic fixed points. A growing number of theoretical and experimental studies suggest that some neural systems utilize nonhyperbolic fixed points and corresponding center manifolds to display complex, nonlinear dynamics and to flexibly adapt to wide-ranging sensory input parameters. In this paper, we present a technique to study the statistical properties of high-dimensional, nonhyperbolic dynamical systems with random connectivity and examine which statistical properties determine both the shape of the center manifold and the corresponding reduced dynamics on it. This technique also gives us constraints on the family of center manifold models that could arise from a large-scale random network. We demonstrate this approach on an example network of randomly coupled damped oscillators.