|
Organizers |
Acceleration Strategies for Restarted Minimal Residual Methods
by
Michael Eiermann
Institut fuer Angewandte Mathematik II, TU Bergakademie Freiberg, GERMANY
Several seemingly disparate techniques have been recently proposed for accelerating restarted Krylov subspace methods such as GMRES. The common purpose of these techniques is to mitigate the deterioration of convergence of these methods due to restarting by eliminating the influence of certain subspaces considered to most impede convergence.
We distinguish two fundamental strategies in existing work: One lies in identifying a subspace which slows convergence, approximating this space and eliminating its influence from the iteration process. The second fundamental strategy consists of identifying the essential orthogonality constraints by comparing angles between subspaces and maintaining orthogonality only against the most important subspace of a given dimension.
The main intent of this talk is to provide an abstract framework which permits a uniform presentation as well as a comparison of these methods.
Date received: March 14, 2001
Copyright © 2001 by the author(s). The author(s) of this document and the organizers of the conference have granted their consent to include this abstract in Atlas Mathematical Conference Abstracts. Document # cagm-28.