A fluctuation theorem associated with Cauchy problems for stationary random operators (Q1103271)

The motivation of this paper comes from \textit{R. Z. Khas'minskij}'s work [Teor. Verojatn. Primen. 11, 240-259 (1966; Zbl 0168.160)] on the solution fluctuation of Cauchy problems on R d with non-linear random functions in general. Now, for the infinite-dimensional but linear case, let \(\{\) H p, \(p\in R\}\) be a family of abstract Sobolev spaces on a compact manifold without boundary and denote \(u^{\epsilon}(\omega,t)\) and u 0(t) the solutions of the abstract Cauchy problems \[ du(t)/dt=L(\omega,t/\epsilon)u(t);\quad u(0)=u_ 0\quad and\quad du(t)/dt=L u(t);\quad u(0)=u_ 0, \] respectively, where L is the mean operator of L(\(\omega\),t), \(u_ 0\in H^{p+3m+1}\). Then, under some assumptions for L(\(\omega\),t) where the strong mixing condition is the most important, it is proved that \[ (1)\quad_{0\leq t\leq T}E \| u^{\epsilon}(t)-u\quad 0(t)\|^ 2_ p\leq C\epsilon,\quad for\quad any\quad T>0, \] \[ (2)\quad X^{\epsilon}(\omega,t)=(u^{\epsilon}(\omega,t)-u\quad 0(t))/\sqrt{\epsilon} \] converges weakly on C([0,T]\(\to H\) p) and the limit distribution coincides with that of an \(H^{p+m}\)-valued process \(\{\) X 0(\(\omega\),t); \(0\leq t\leq T\}\) described by a given stochastic integral.