📝 SummarXiv

Abstract

We develop a mean-field theory for large, non-exchangeable particle (agent) systems where the states and interaction weights co-evolve in a coupled system of SDEs. A first main result is a generalization of the propagation of chaos. The weight adaptation in the SDEs makes the classical approach of using a static probability space as the continuum limit for the agent labels inadequate. We address this by introducing a label space endowed with a filtration that captures the stochasticity. While this yields a well-posed McKean-Vlasov SDE, its limit is not easily described by a Vlasov-type PDE. This difficulty, in turn, motivates the introduction of a unified metric that naturally combines the Wasserstein distance for states and the cut norm for weights. In this metric space, we establish the convergence of finite systems to their continuum limit. This result, analogous to the classical convergence of empirical measures, is a subtle consequence of a deep result from dense graph theory, namely the Sampling Lemma.