Large deviations theorems for empirical measures in Freidlin-Wentzell exit problems (Q2640219)

For each \(\epsilon >0\), let \(\{(X^{\epsilon}(t),P_ x):\) \(0\leq t\), \(x\in {\mathbb{R}}^ d\}\) be a strong Markov process whose infinitesimal generator is defined by a (typical) integro-differential equation. Let D be a bounded domain of \({\mathbb{R}}^ d\) and \(\tau^{\epsilon}_ D=\inf \{t>0:\;X^{\epsilon}(t)\not\in D\}\) be the exit time for \(X^{\epsilon}\) from D. Large deviations theorems are proved for the empirical measures \[ \mu^{\epsilon}(dy)=\int^{\tau^{\epsilon}_ D}_{0}\mathbf{1}_{\{dy\}}(x^{\epsilon}(s))ds/\tau^{\epsilon}_ D, \] when \(\epsilon\to 0\). The cumulant of \(X^{\epsilon}(t)\) and its Legendre transformation are assumed to have some regularity properties. Under these assumptions \(X^{\epsilon}(t)\) can be considered as small random perturbation of a dynamical system.