Abstract. In this work we first focus on the Stochastic Galerkin approximation of the solution u of an elliptic stochastic PDE. We rely on sharp estimates for the decay of the coefficients of the spectral expansion of u on orthogonal polynomials to build a sequence of polynomial subspaces that features better convergence properties compared to standard polynomial subspaces such as Total Degree or Tensor Product.We consider then the Stochastic Collocation method, and use the previous estimates to introduce a new effective class of Sparse Grids, based on the idea of selecting a priori the most profitable hierarchical surpluses, that, again, features better convergence properties compared to standard Smolyak or tensor product grids.Key words: Uncertainty Quantification, PDEs with random data, elliptic equations, multivariate polynomial approximation, Best M -Terms approximation, Stochastic Galerkin methods, Smolyak approximation, Sparse grids, Stochastic Collocation methods.AMS Subject Classification: 41A10, 65C20, 65N12, 65N35
IntroductionMany works have been recently devoted to the analysis and the improvement of the Stochastic Galerkin and Collocation techniques for Uncertainty Quantification for PDEs with random input data. These methods are promising since they can exploit the possible regularity of the solution with respect to the stochastic parameters to achieve much faster convergence than sampling methods like Monte Carlo.Stochastic Galerkin and Collocation methods can be classified as parametric techniques, since both expand u, the solution of the PDE of interest, as a summation over suitable deterministic basis functions in probability space, typically polynomials or piecewise polynomials. Stochastic Galerkin is a projection technique over a set of orthogonal polynomials with respect to the probability measure at hand (see e.g. [1,12,15,20]), while Stochastic Collocation is a sum of Lagrangian interpolants over the probability space (see e.g. [2,10,22]
11The comparison between performances of these deterministic methods is a matter of study (see e.g.[3]). However, both suffer the so-called "Curse of Dimensionality": using naive projections/interpolations over tensor product polynomials spaces/tensor grids leads to computational costs that grow exponentially fast with the number of input random variables. In such a case, careful construction of approximation spaces/sparse grids is needed in order to retain accuracy while keeping computational work acceptably low.In a Stochastic Galerkin setting this requirement can be translated to the implementation of algorithms able to compute what is known as "Best M -Terms approximation". In other words, the method should be able to establish a-priori the set of the M most fruitful multivariate polynomials in the spectral approximation of the solution u, and to compute only those terms.Important contributions in the study of the Best M -Terms approximation have been given by Cohen, DeVore and Schwab: estimates on the decay of the coefficients of the spectral expansi...