Multistability of Phase-Locking in Equal-Frequency Kuramoto Models on Planar Graphs

TL;DR

This paper derives an algebraic upper bound on the number of stable fixed points in equal-frequency Kuramoto models on planar graphs, linking network topology, cycle lengths, and stability properties.

Contribution

It introduces a new algebraic upper bound for stable fixed points in Kuramoto models on planar networks, improving upon previous exponential bounds.

Findings

Derived an upper bound depending on cycle count and lengths

Identified topologies with larger angle differences exceeding π/2

Conjectured a universal upper bound for planar networks

Abstract

The number $\mathcal{N}$ of stable fixed points of locally coupled Kuramoto models depends on the topology of the network on which the model is defined. It has been shown that cycles in meshed networks play a crucial role in determining $\mathcal{N}$, because any two different stable fixed points differ by a collection of loop flows on those cycles. Since the number of different loop flows increases with the length of the cycle that carries them, one expects $\mathcal{N}$ to be larger in meshed networks with longer cycles. Simultaneously, the existence of more cycles in a network means more freedom to choose the location of loop flows differentiating between two stable fixed points. Therefore, $\mathcal{N}$ should also be larger in networks with more cycles. We derive an algebraic upper bound for the number of stable fixed points of the Kuramoto model with identical frequencies, under the assumption that angle differences between connected nodes do not exceed $\pi/2$. We obtain $\mathcal{N}\leq\prod_{k=1}^c\left[2\cdot{\rm Int}(n_k/4)+1\right]$, which depends both on the number $c$ of cycles and on the spectrum of their lengths $\{n_k\}$. We further identify network topologies carrying stable fixed points with angle differences larger than $\pi/2$, which leads us to conjecture an upper bound for the number of stable fixed points for Kuramoto models on any planar network. Compared to earlier approaches that give exponential upper bounds in the total number of vertices, our bounds are much lower and therefore much closer to the true number of stable fixed points.