Some results on the problem of exit from a domain (Q1198598)

The paper deals with the problem of exit from a domain of attraction of a stable equilibrium point in the presence of small noise. The motivation is to fill a gap between some results proposed by two different approaches. On one side, the large deviation approach gives rigorous results, but seems not refined enough to provide an insight or proof of certain facts which, in contrast, are conjectured by a second approach based on asymptotic expansions. The latter approach yields sharper informations, but only under assumptions on the quasi-potential which cannot be easily checked in practice. The paper restricts the attention to a two-dimensional diffusion model of the form \(dX_ t=f(X_ t)dt+\varepsilon BdW_ t\) under rather restrictive (but explicit) assumptions on \(f\) and \(B\); however it seems that the main ideas have a more general range of application, to be investigated. The specific question rigorously addressed by the paper, conjectured by the asymptotic expansion approach, is that the exit measure, which by large deviation theory concentrates around a point \(S\) on the boundary of the domain of attraction as \(\varepsilon\) goes to 0, in fact is skewed, in the sense that there exists an \(\varepsilon\)-dependent neighbourhood \(\Delta\) of \(S\) such that \(P(\text{exit in }\Delta)/|\Delta|\) goes to 0, so that the most probable exit point for nonzero \(\varepsilon\) is not \(S\) but is skewed aside.