[Paper Review] Three Formulations of the Kuramoto Model as a System of Polynomial Equations

Tianran Chen, Jakub Mareček+2 more|arXiv (Cornell University)|Mar 18, 2016

Nonlinear Dynamics and Pattern Formation42 references4 citations

TL;DR

This paper introduces and compares three polynomial formulations of the Kuramoto model's equilibrium equations, with a novel half-angle transformation that yields tighter bounds on the number of real equilibria using Bernstein-Kushnirenko-Khovanskii (BKK) theory. It demonstrates that this reformulation significantly improves performance in both numerical homotopy continuation and moment-based optimization methods, offering superior computational efficiency and tighter upper bounds on complex solutions compared to traditional formulations.

ABSTRACT

We compare three formulations of stationary equations of the Kuramoto model as systems of polynomial equations. In the comparison, we present bounds on the numbers of real equilibria based on the work of Bernstein, Kushnirenko, and Khovanskii, and performance of methods for the optimisation over the set of equilibria based on the work of Lasserre, both of which could be of independent interest.

Motivation & Objective

• To develop and compare multiple polynomial reformulations of the Kuramoto model's equilibrium equations for improved computational analysis.
• To derive tighter upper bounds on the number of real and complex equilibria using BKK theory, especially for sparse graph topologies.
• To evaluate and enhance the performance of optimization over equilibria using moment-based methods and numerical homotopy continuation.
• To demonstrate that the proposed half-angle formulation outperforms existing formulations in both solution counting and optimization efficiency.
• To provide constructive tools for solving and analyzing the Kuramoto model using algebraic geometry and polynomial system techniques.

Proposed method

• Reformulates the Kuramoto model's equilibrium equations into three distinct systems of polynomial equations using trigonometric identities and variable substitutions.
• Introduces a novel half-angle transformation involving $ \tan(\theta_i/2) $, which simplifies the system and reduces monomial degrees.
• Applies Bernstein-Kushnirenko-Khovanskii (BKK) theory to compute upper bounds on the number of complex solutions for each formulation.
• Employs numerical polynomial homotopy continuation (NPHC) via HOM4PS-3.0 and Bertini to compute actual solution counts and validate bounds.
• Uses the method of moments via Lasserre’s hierarchy to optimize over the set of equilibria, comparing performance across formulations.
• Constructs sparse semidefinite programming (SDP) relaxations using SparsePOP to evaluate computational complexity and solver efficiency.

Experimental results

Research questions

• RQ1 How do different polynomial reformulations of the Kuramoto model's equilibrium equations affect the upper bound on the number of complex solutions?
• RQ2 Can the proposed half-angle transformation yield tighter BKK-based bounds than traditional formulations, especially for sparse graphs?
• RQ3 How does the choice of reformulation impact the performance of numerical homotopy continuation and moment-based optimization methods?
• RQ4 To what extent are the BKK bounds tight, and do they match the actual number of complex solutions for generic parameter values?
• RQ5 Can the new formulation enable more efficient optimization over equilibria in power system and dynamical systems applications?

Key findings

• The proposed half-angle formulation yields significantly lower BKK bounds than traditional formulations for sparse graphs like path graphs, improving upon bi-homogeneous Bézout bounds.
• For the complete graph with inhomogeneous couplings, the BKK bound from the new formulation matches the known upper bound $ \binom{2(N-1)}{N-1} $, confirming tightness.
• Numerical homotopy continuation shows that the BKK bound is tight for generic natural frequencies and coupling strengths, as the number of complex solutions matches the bound exactly.
• The new formulation reduces the number of non-zero entries in the SDP constraint matrix (nnz.) by 10% compared to the standard formulation (F1), improving solver efficiency.
• The method of moments achieves better performance on the new formulation (F3) than on the traditional one (F1), due to lower-degree monomials and reduced SDP relaxation size.
• The half-angle transformation is shown to be effective across multiple benchmarks, offering a robust alternative to existing reformulations in both theoretical and computational contexts.