next up previous
Next: About this document ...

Eric de Sturler
Computing Preconditioners for Parameterized Matrices

Department of Mathematics
McBryde Hall
Rm 460
Virginia Tech
225 Stanger St
Blacksburg
VA 24061-0123
USA
sturler@vt.edu
Ming Li
Arielle Grim McNally

Parameterized matrices or linear systems arise in many applications. Often we need to solve such linear systems for a large number of parameters, for example, for a predetermined sweep of parameter values or while sampling from parameter space. Rather than computing a new preconditioner for each parameter or using a single preconditioner for all parameters, we can often update a high quality preconditioner computed for one system, at low cost, to be just as good for a subsequent system. The key idea in our approach is to consider updating a preconditioner as computing an approximate map from a new matrix to a previous matrix. We discuss a few approaches to updating preconditioners, their analysis, and several applications.





Copper Mountain 2014-02-23