Dual-based a Posteriori error estimate for stochastic finite element methods

Lionel Mathelin, Olivier Le Maître

Research output: Contribution to journalArticlepeer-review

41 Scopus citations

Abstract

We present an a posteriori error estimation for the numerical solution of a stochastic variational problem arising in the context of parametric uncertainties. The discretization of the stochastic variational problem uses standard finite elements in space and piecewise continuous orthogonal polynomials in the stochastic domain. The a posteriori methodology is derived by measuring the error as the functional difference between the continuous and discrete solutions. This functional difference is approximated using the discrete solution of the primal stochastic problem and two discrete adjoint solutions (on two imbricated spaces) of the associated dual stochastic problem. The dual problem being linear, the error estimation results in a limited computational overhead. With this error estimate, different adaptive refinement strategies of the approximation space can be thought of: applied to the spatial and/or stochastic approximations, by increasing the approximation order or using a finer mesh. In order to investigate the efficiency of different refinement strategies, various tests are performed on the uncertain Burgers' equation. The lack of appropriate anisotropic error estimator is particularly underlined.

Original languageEnglish (US)
Pages (from-to)83-115
Number of pages33
JournalCommunications in Applied Mathematics and Computational Science
Volume2
Issue number1
DOIs
StatePublished - 2007
Externally publishedYes

Keywords

  • Error analysis
  • Refinement scheme
  • Stochastic finite element method
  • Uncertainty quantification

ASJC Scopus subject areas

  • Computer Science Applications
  • Computational Theory and Mathematics
  • Applied Mathematics

Fingerprint

Dive into the research topics of 'Dual-based a Posteriori error estimate for stochastic finite element methods'. Together they form a unique fingerprint.

Cite this