PARA'04 State-of-the-Art
in Scientific Computing
June 20-23, 2004 (Home page)

Updated: 24 April 2004

Algorithms and Software Components for Multimethod SuperSolvers

Padma Raghavan
The Pennsylvania State University
USA

Abstract:

Many large-scale applications require the numerical solution of nonlinear models based on partial-differential equations. For such applications, the dominant cost is often from the solution of nonsymmetric sparse linear systems within an inexact Newton method. We consider the development of `multimethod solvers' that can combine several precondiitioned iterative solution methods to reduce execution time, provide a robust solution etc. We will provide an overview of our methods to construct `adaptive' and `composite' solvers. We will discuss their `component' implementations and report on their performance in computational fluid dynamics applications.

Home page


2004-04-24