Captivity of mean-field particle systems and the related exit problems
Résumé
The aim of this work is to establish the stability of mean-field system under non-convex confining potential. A mean-field system corresponds to a system of $N$ particles in weak interaction and confined by an exterior force. With our hypotheses, it is a Kolmogorov diffusion with potential $\Upsilon^N$. Exit time of these systems have been studied in details in the small-noise limit. Here, we will deal with the large-dimension limit with fixed noise. In one hand, we show that the meta-potential $\Upsilon^N$ admits a number of wells which tends to infinity when $N$ goes to infinity. In the other hand, by using the convergence of McKean-Vlasov processes in long-time and the propagation of chaos, we prove that there exist traps such that the diffusion can not escape from. Furthermore, the traps do not coincide with the wells of $\Upsilon^N$.
Domaines
Probabilités [math.PR]
Origine : Fichiers produits par l'(les) auteur(s)