Stochastic delay equations with hereditary drift: Estimates of the density (Q1589672)

The authors study a one-parameter family of stochastic differential delay equations of the form \[ X^\varepsilon(t) = \int_0^tH(s, X^\varepsilon_s)ds + \varepsilon\int_0^tg(s, X^\varepsilon(s - r)) dW(s),\quad t \geq 0,\tag \(*\) \] with deterministic initial condition \(X^\varepsilon(t) = \eta(t), t \in [-r, 0]\). Here \(r\) is a fixed time delay and \(X^\varepsilon_s\) denotes the path \(X^\varepsilon(u), u \leq s\). Stochastic differential equations of the type \((*)\) have previously been studied by the reviewer and S. Mohammed. They showed that under conditions that allow the diffusion coefficient \(g\) to degenerate on a hypersurface, the solution process \(X^\varepsilon\) admits smooth densities \(P^\varepsilon_t(y)\) for all \(t > 0\). The authors study the asymptotic behavior of the system \((*)\) as \(\varepsilon \to 0\), under similar conditions. They establish large deviation estimates of the form \[ \lim \sup_{\varepsilon \downarrow 0} \varepsilon ^2\log P^\varepsilon_t(y) \leq -d^2(y), \qquad \lim \inf_{\varepsilon \downarrow 0} \varepsilon ^2\log P^\varepsilon_t(y) \geq -\widetilde d^2(y) \] where \(d^2(y)\) and \(\widetilde d^2(y)\) are finite constants depending on \(t\) and \(y\). The authors' proof of these estimates uses the Malliavin calculus, in combination with ideas of Bell-Mohammed and Leandre.