Abstract Formulation of Mean-Field Models and Propagation of Chaos
Abstract
We formulate a unified abstract framework for mean-field particle systems and their macroscopic limits from the perspective of differential equations and functional analysis.
Combining semigroup theory with optimal transport, we study measure-valued evolutions directly through their generators, allowing different classes of mean-field models to be treated within a common structure.
Specifically, the framework involves (1) formulating mean-field models abstractly using semigroups and generators; (2) establishing well-posedness of the corresponding nonlinear mean-field evolution equations in a class of solutions stable under approximation; and (3) proving quantitative propagation of chaos for the associated $N$-particle systems.
The central tool is the transport derivative, which measures the infinitesimal variation of transport costs between Markov evolutions.
We establish transport-stability criteria expressed through local generator inequalities and show how these criteria underpin both nonlinear well-posedness and particle approximation.
When the generator's dependence on the law is controlled by averaged Hilbert-valued features, we obtain an $O(N^{-1})$ sampling contribution to the quadratic Wasserstein estimate through a variance argument.
As an application, we obtain quantitative propagation of chaos for Lévy-type mean-field systems with state- and law-dependent drift, diffusion, and jump kernels under intrinsic transport conditions.
The admissible jump kernels extend beyond those generated by multiplicative Lévy noise, demonstrating the scope of the abstract framework.