Large deviations and a Kramers' type law for self-stabilizing diffusions (Q939074)

The authors consider a \(d\)-dimensional diffusion process with a small (Brownian) noise term (of order \(\varepsilon\)) moving in a drift field which in some sense is close to a potential type drift field. In addition there is a drift towards the ``center of the crowd'' (self-stabilization); that is, the diffusion is subject to a drift that is described in terms of a certain potential function and the law of the diffusion at the same time. This interaction of the diffusion with its own law preserves the Markov nature of the diffusion but destroys time-homogeneity. The question of existence and uniqueness of a solution is quite delicate. A substantial part of the paper under review is dedicated to establishing the existence of a unique solution under some regularity assumptions (Theorem 2.13). The authors prove a large deviations principle (as \(\varepsilon\to0\)) for the paths of the process (Theorem 3.4). Classical diffusions in a potential type drift field obey the so-called Eyring-Kramers law for the exit time of a domain in the low-noise limit: The leading order of the exit time is the exponential of the potential well divided by \(\varepsilon\). The authors show (Theorems 4.2 and 4.3) that a similar law is in force for the self-stabilized diffusion. Here the exit distribution coincides with the exit distribution of a classical diffusion in a potential perturbed by a drift towards a unique stable attracting point in the domain.