Finite difference schemes for stochastic partial differential equations in Sobolev spaces (Q496118)

The authors study spatial finite differences schemes for parabolic stochastic partial differential equations (SPDEs), possibly degenerate, with an initial condition. More precisely, they consider \[ \begin{multlined} du_t(x)=[D_i(a_t^{ij}(x)D_ju_t(x)) + b_t^{i}(x)D_i u_t(x)+c_t(x)u_t(x)+f_t(x)]dt \\ + [\mu_t^{ir}D_i u_t(x) + \nu_t^{r}(x)u_t(x)+g^r(x)] dw_t^r \end{multlined} \] for \((t,x)\in[0,T]\times\mathbb R^d\) with initial condition \(u_0(x)=\psi(x)\), \(x\in\mathbb R^d\). Error estimation in supreme norms are usually proved via Sobolev's embedding, requiring smoothness conditions on the coefficients of the equation. The main aim of this paper is to overcome these ``unnecessary conditions''. The authors give \(W_p^m\)-norm estimates assuring that the initial conditions in \(W_p^m\) and the free terms are \(W_p^m\) valued processes. Asymptotic expansion of the error, allowing acceleration of the approximation by Richardson's method, are given. Following ideas of \textit{N. V. Krylov} [Appl. Math. Optim. 52, No. 3, 365--399 (2005; Zbl 1087.65100)] to estimate the solution of finite difference schemes, the authors consider in the whole space rather than on a grid. The estimates obtained for the corresponding Sobolev norms on the whole space allow to estimate their supremum on a grid. For the finite difference approximations not only their convergence is proved, but also power series expansion in the mesh size is obtained. This allows to accelerate the rate of convergence using the Richardson extrapolation.