A PDE approach to certain large deviation problems for systems of parabolic equations (Q1822831)

The authors investigate exponential decay properties as \(\epsilon\) \(\to 0\) for the solution of the linear parabolic system \[ u^{\epsilon}_{k,t}=(\epsilon /2)\sum^{n}_{i,j=1}a^ k_{ij}u^{\epsilon}_{k,x_ ix_ j}+\sum^{n}_{i=1}b^ k_ iu^{\epsilon}_{k,x_ i}+\epsilon^{- 1}\sum^{m}_{l=1}c_{kl}u_ l^{\epsilon}\quad in\quad {\mathbb{R}}^ n\times (0,\infty), \] \[ and\quad u_ k^{\epsilon}=g_ k\quad in\quad {\mathbb{R}}^ n\times \{t=0\}, \] where the \(g_ k: {\mathbb{R}}^ n\to {\mathbb{R}}\) have compact support, \(\sum^{m}_{l=1}c_{kl}=0\) and \((a^ k_{ij})\) is positive definite, \(k=1,...,m.\) Rather than using probabilistic arguments employing large deviations techniques by interpreting the system as the backward equation for a Markov process on \({\mathbb{R}}^ n\times \{1,...,m\}\) switching between the states 1 to m with rates \(\epsilon^{-1}c_{kl}\), they use PDE techniques to establish the exponential rate at which \(u_ k^{\epsilon}\) converges to zero off the support of the common limit u which is given as the solution of a transport equation. The authors provide a representation for the corresponding rate function I(x,t) using the fact that I solves a certain Hamilton-Jacobi PDE.