Ergodicity of stochastic plates (Q1892256)

Let \(G \subset \mathbb{R}^ N\) be a bounded regular domain. Consider the operator \((- \Delta)^ s\) on \(L^ 2(G)\), \(s > N/2\), \(\Delta\) being the Laplacian with Dirichlet boundary conditions, and the SDE \[ dX_{\varepsilon, t} = \varepsilon(t) \alpha (\bullet, dt) - (- \Delta)^ s X_{\varepsilon,t} dt - V'(X_{\varepsilon, t}) dt,\tag{*} \] where \(\alpha\) is space-time white noise, \(V\) is of class \(C^ \infty_ b\), and whose solution is a \(C_ 0(G)\)-valued process \(X_{\varepsilon, t}\). Denote by \(\Pi_{\varepsilon}\) the Gaussian measure on \(L^ 2(G)\) with covariance operator \(\varepsilon^ 2 /2(- \Delta)^{-s}\), by \(\mu_ \varepsilon\) the Boltzmann-Gibbs measure with density (with respect to \(\Pi_ \varepsilon)\) \(c_ \varepsilon \text{exp}(-2\varepsilon^{-2} \int_ G V(\omega(x)) dx)\), \(c_ \varepsilon\) being the normalization constant, and by \(S\) the energy functional \[ S(\omega) = (1/2) \int_ G (((-\Delta)^{s/2} \omega(x))^ 2 + V(\omega(x))) dx. \] For fixed \(\varepsilon\) (not depending on \(t\)) the existence of solutions to \((*)\) is proved and an approximation by finite-dimensional processes is constructed. This approximation is used to show that the relative entropy satisfies \[ I(\text{law} (X_{\varepsilon,t}) \mid \mu_{\varepsilon}) \leq C \text{exp} (C \varepsilon^{-2} - \gamma_ \varepsilon t) \] for some constants \(C\), \(\gamma_ \varepsilon\) and that \(\lim_{\varepsilon \to 0} \varepsilon^ 2 \log \gamma_ \varepsilon = -2m_ S\) where \(m_ S\) is Hajek's constant for the energy functional \(S\). As an application an `approximate simulated annealing' for the process \(X_{\varepsilon (t),t}\) as \(t \to \infty\) is established, that is, for any natural \(J\) a process \(Y^ J_{\varepsilon (t),t}\) (closely related to \(X_{\varepsilon(t),t}\)) is constructed and it is shown that its finite- dimensional projections \(E_ J Y^ J_{\varepsilon (t),t}\) satisfy \(\text{dist}_{L^ 2 (G)} (E_ J Y^ J_{\varepsilon (t),t}\), \(E_ j \{S = \inf S\}) \to 0\) in probability as \(t \to \infty\).