The Bramble−Pasciak preconditioner for saddle point problems
Martin Stoll and Andy Wathen
Abstract
The Bramble-Pasciak Conjugate Gradient method is a well known tool to solve linear systems in saddle point form. A drawback of this method in order to ensure applicability of Conjugate Gradients is the need for scaling the preconditioner which typically involves the solution of an eigenvalue problem. Here, we introduce a modified preconditioner and inner product which without scaling enable the use of a MINRES variant and can be used for the simplified Lanczos process. Furthermore, the modified preconditioner and inner product can be combined with the original Bramble-Pasciak setup to give new preconditioners and inner products. We undermine the new methods by showing numerical experiments for Stokes problems.
Details
| Institution |
Oxford University Computing Laboratory |
| Month |
June |
| Number |
NA−07/13 |
| Year |
2007 |
Links
Related pages
|
People |