next up previous contents
Next: Parallel preconditioners in ParPre Up: Introduction Previous: Introduction

Preconditioners

In many scientific applications, one needs to solve linear systems

Ax=b

with A square and non-singular, often symmetric, or positive definite, or an M-matrix. For a variety of reasons, one can choose to solve this system by an iterative method, rather than by some form of Gaussian elimination.

Iterative methods feature a preconditioning step, that is, the exact solution of a linear system

Cx=b

where C in some sense approximates A. In general, a more accurate choice of C will yield a more speedy solution of the iterative process, but such a choice will also be more costly to construct and to apply.

For general background information about iterative methods and preconditioners, see the `Templates ' book [2], and books by Axelsson [1], Hackbusch [7], and Saad [10]. More specifically for domain decomposition methods and Schwarz methods, see, e.g., [4,11].


Victor Eijkhout
7/27/1998