Error estimation and adaptive computation for elliptic problems with randomly perturbed coefficients


Axel M\aa lqvist
Department of Mathematics
UCSD
Joint with Donald Estep and Simon Tavener

Abstract

We develop and analyze an efficient numerical method for computing the response of the solution of an elliptic problem with randomly perturbed coefficients. We use a variational analysis based on the adjoint operator to deal with the perturbations in data. To deal with perturbations in the diffusion coefficient, we construct a piecewise constant approximation to the random perturbation then use domain decomposition to decompose the problem into sub-problems on which the diffusion coefficient is constant. To compute local solutions of the sub-problems, we use the infinite series for the inverse of a perturbation of an invertible matrix to devise a fast way to compute the effects of variation in the parameter. Finally, we derive a posteriori error estimates that take into account all the sources of error and derive a new adaptive algorithm that provides a quantitative way to distribute computational resources between all of the sources.