Robust domain decomposition algorithms for multiscale PDEs

I G Graham, R Scheichl

Research output: Contribution to journalArticlepeer-review

38 Citations (SciVal)

Abstract

In this article, we describe a new class of domain decomposition preconditioners suitable for solving elliptic PDEs in highly fractured or heterogeneous media, such as arise in groundwater flow or oil recovery applications. Our methods employ novel coarsening operators, which are adapted to the heterogeneity of the media. In contrast to standard methods (based on piecewise polynomial coarsening), the new methods can achieve robustness with respect to coefficient discontinuities even when these are not resolved by a coarse mesh. This Situation arises often in practical flow computation, in both the deterministic and (Monte-Carlo simulated) stochastic cases. An example of a suitable coarsener is provided by multiscale finite elements. In this article, we explore the linear algebraic aspects of the multiscale algorithm, showing that it involves a blend of both classical overlapping Schwarz methods and nonoverlapping Schur methods. We also extend the algorithm and the theory from its additive variant to obtain new hybrid and deflation variants. Finally, we give extensive numerical experiments on a range of heterogeneous media problems illustrating the properties of the methods. (c) 2007 Wiley Periodicals, Inc.
Original languageEnglish
Pages (from-to)859-878
Number of pages20
JournalNumerical Methods for Partial Differential Equations
Volume23
Issue number4
DOIs
Publication statusPublished - 2007

Bibliographical note

ID number: ISI:000247145900007

Fingerprint

Dive into the research topics of 'Robust domain decomposition algorithms for multiscale PDEs'. Together they form a unique fingerprint.

Cite this