Domain decomposition method of stochastic PDEs: a two-level scalable preconditioner

For uncertainty quantification in many practical engineering problems, the stochastic finite element method (SFEM) may be computationally challenging. In SFEM, the size of the algebraic linear system grows rapidly with the spatial mesh resolution and the order of the stochastic dimension. In this paper, we describe a non-overlapping domain decomposition method, namely the iterative substructuring method to tackle the large-scale linear system arising in the SFEM. The SFEM is based on domain decomposition in the geometric space and a polynomial chaos expansion in the probabilistic space. In particular, a two-level scalable preconditioner is proposed for the iterative solver of the interface problem for the stochastic systems. The preconditioner is equipped with a coarse problem which globally connects the subdomains both in the geometric and probabilistic spaces via their corner nodes. This coarse problem propagates the information quickly across the subdomains leading to a scalable preconditioner. For numerical illustrations, a two-dimensional stochastic elliptic partial differential equation (SPDE) with spatially varying non-Gaussian random coefficients is considered. The numerical scalability of the the preconditioner is investigated with respect to the mesh size, subdomain size, fixed problem size per subdomain and order of polynomial chaos expansion. The numerical experiments are performed on a Linux cluster using MPI and PETSc parallel libraries.

[1]  C. Farhat,et al.  A scalable dual-primal domain decomposition method , 2000, Numer. Linear Algebra Appl..

[2]  Yousef Saad,et al.  Iterative methods for sparse linear systems , 2003 .

[3]  W. T. Martin,et al.  The Orthogonal Development of Non-Linear Functionals in Series of Fourier-Hermite Functionals , 1947 .

[4]  A. Sarkar,et al.  Domain decomposition of stochastic PDEs: Theoretical formulations , 2009 .

[5]  Nelson H. F. Beebe.  Numerical Linear Algebra with Applications and Numerical Linear Algebra with Applications , 1998 .

[6]  R. Ghanem,et al.  Stochastic Finite Elements: A Spectral Approach , 1990 .

[7]  Jacques Periaux,et al.  On Domain Decomposition Methods , 1988 .

[8]  Waad Subber,et al.  Domain Decomposition of Stochastic PDEs: A Novel Preconditioner and Its Parallel Performance , 2009, HPCS.

[9]  Tarek P. Mathew,et al.  Domain Decomposition Methods for the Numerical Solution of Partial Differential Equations , 2008, Lecture Notes in Computational Science and Engineering.

[10]  Jan Mandel,et al.  On the Equivalence of Primal and Dual Substructuring Preconditioners , 2008, 0802.4328.

[11]  T. Mathew,et al.  Domain Decomposition Methods for the Numerical Solution of Partial Differential Equations (Lecture Notes in Computational Science and Engineering) , 2008 .

[12]  Waad Subber,et al.  Primal and dual-primal iterative substructuring methods of stochastic PDEs , 2010 .

[13]  N. Wiener.  The Homogeneous Chaos , 1938 .

[14]  Barry F. Smith,et al.  Domain Decomposition: Parallel Multilevel Methods for Elliptic Partial Differential Equations , 1996 .

[15]  Andrea Toselli,et al.  Domain decomposition methods : algorithms and theory , 2005 .