A sample-based spectral method approach for solving high-dimensional stochastic partial differential equations
Zhibao Zheng,H. Dai,Yuyin Wang,Wei Wang
2020-01-01
Abstract:In the past few decades, a growing interest has been devoted for the propagation of uncertainties through physical models governed by stochastic partial differential equations (SPDEs). Despite their success and applications, existing methods are mainly limited to low-dimensional stochastic problems due to the extreme computational costs. In this paper, on the basis of an universal construct of stochastic solutions, we explore an efficient strategy for solving high-dimensional linear and nonlinear SPDEs, where original SPDEs are transformed into deterministic PDEs and one-dimensional stochastic algebraic equations by use of Galerkin method. Deterministic PDEs are solved by existing pde technologies and one-dimensional stochastic algebraic equations are calculated by a sample-based numerical method. Since computational costs are almost insensitive to the stochastic dimensions, the proposed strategy beats the so-called Curse of Dimensionality with great success. Results on numerical examples in high dimensions, including the linear elliptic PDE ∗Corresponding author. Email address: zhibaozheng@hit.edu.cn (Zhibao Zheng) Preprint submitted to Journal August 11, 2020 ar X iv :1 90 5. 05 80 2v 2 [ m at h. N A ] 1 0 A ug 2 02 0 with stochastic coefficients, the nonlinear Burgers equation with stochastic force and the linear wave equation with stochastic initial values, demonstrate that the proposed strategy provides an efficient and unified framework for solving SPDEs, and is particularly appropriate for high-dimensional SPDEs of practical interests.