Convergence rates for finite elementapproximations of stochastic partial differential equations (Q4390916)

There exists an extensive literature on how to solve deterministic partial differential equations numerically. For stochastic problems of the same nature, the theory seems to be not so well developed. There is a problem of numerically calculating a combination of random variables called the Wick product. The main objective of this paper is to give a general framework to solve elliptic stochastic partial differential equations based on the finite element method. The authors' approach is a direct application of the Ritz-Galerkin (R-G) method put in a new setting. Namely, since the authors deal with stochastic equations, a new independent variable \(w\) comes in addition to the usual space variable \(x\). Due to this extra variable, new Hilbert spaces are called for to fit the framework to the R-G method: as the tensor product of a classical Sobolev space and a Hilbert space taking the stochastic regularity into account. A convergence rate estimate for a finite-dimensional approximation of generalized random variables is calculated. Also, the authors follow up with the Sobolev analogue of such a result. The authors apply these results to elliptic stochastic problems on a variational form. The authors' approach fits every elliptic problem that can be rewritten in a variational form.