Small random perturbations of random evolution equations (Q443813)

The presented work treats large deviations of stochastic differential equations (SDE), parameterized by another stochastic process. It extends previously published results to a more general parametrization. The general setup is the SDE \[ dX_t^\varepsilon = \sqrt{\varepsilon} \sigma_{\nu_{\varepsilon}(t)}(X_t^\varepsilon)dW_t + b_{\nu_{\varepsilon}(t)}(X_t^\varepsilon)dt, \quad X_0^\varepsilon=x. \] The text derives a large deviation principle (LDP) for the laws of the solution \(X^\varepsilon\) on \(C[0,1]\) for small noise level \(\varepsilon\to0\). The solution is parameterized by a stochastic process \(\nu_\varepsilon=\left( \nu_{\varepsilon}(t) \right)_{0\leq t\leq 1}\). Two situations are considered. In the first case, (see \textit{C. Bezuidenhout} [Ann. Probab. 15, 646--658 (1987; Zbl 0622.60033)]), \(\nu_{\varepsilon}\) is a stochastic process independent of \(W\) and satisfying an LDP itself, with a good rate function \(I\). In contrast to the previous work, the diffusion coefficient \(\sigma\) is now allowed to depend on \(\nu_\varepsilon\). The rate functional is shown to be basically the sum of the two involved rate functionals, \( \lambda(g) = \inf_{(\phi,f)}\left\{ S(f)+I(\phi) \right\} \), subject to the fact that \(g\) solves an ODE with noise \(\dot f\) which is parametrized by \(\phi\) and \(S\) is the classical rate functional \(S(f)=\frac{1}{2}\|f\|^2\) given by the Cameron-Martin norm. In the second case (see [\textit{A. Eizenberg} and \textit{M. Freidlin}, Ann. Prob. 21, No. 2, 1015--1044 (1993; Zbl 0776.60037)]), \(\nu_{\varepsilon}\) is a finite state jump process with right-continuous trajectories. These results are generalized to non-constant, \(\nu_\varepsilon\)-dependent \(\sigma_{\nu_\varepsilon}\). The situation is assessed by potential theoretic methods, linking the LDP to a set of partial differential equations indexed by the (finite) state space of the process \(\nu_{\varepsilon}\). They discover that in the small noise limit, the process will leave a dissipative domain \(D\) at times when \(\nu_\varepsilon\) is in the state that minimizes the barrier provided by \(b_{\nu_\varepsilon}\) on \(\partial D\).