Uniform-in-time propagation of chaos for Second-Order Consensus-Based Optimization

TL;DR

This paper proves uniform-in-time propagation of chaos for second-order consensus-based optimization, introducing a hypocoercive coupling framework to handle the challenges posed by velocity-only stochastic forcing.

Contribution

It establishes the first quantitative uniform-in-time propagation of chaos results for second-order CBO dynamics using a novel hypocoercive coupling approach.

Findings

Achieves exponential decay of centered moments.

Provides classical Monte Carlo rate for propagation of chaos.

Yields faster convergence rate O(J^{-q}) for system stability.

Abstract

We study second-order Consensus-Based Optimization (CBO), a derivative-free global optimization algorithm in which both the consensus drift and the multiplicative exploratory noise act on the particle velocities. We establish the first quantitative uniform-in-time propagation of chaos results for the second-order CBO dynamics, together with an almost uniform-in-time stability estimate for the microscopic particle system. The main difficulty is that in the second-order model, both the consensus mechanism and the stochastic forcing act only on the velocity variable, while the position evolves via transport. As a result, no direct coercive mechanism is available on the spatial component, and, combined with the shift-invariant nature of the consensus interaction, a standard synchronous-coupling argument in the Euclidean phase-space distance cannot be closed uniformly in time. To overcome this difficulty, we develop a hypocoercive coupling framework together with a suitable shifting of variables. Passing to centered internal variables decouples the fluctuation dynamics from the empirical mean and yields exponential decay of centered moments, which is the key ingredient for uniform-in-time control. Combined with uniform-in-time raw moment bounds, concentration inequalities, stability estimates for the weighted mean, and a Monte Carlo estimate, we achieve the classical Monte Carlo rate for propagation of chaos uniformly in time. In addition, the system-to-system stability estimate avoids the mean-field sampling error and yields the faster convergence rate O(J^{-q}).