U.S.-France Cooperative Research: Robust Parallel Preconditioning Methods: Bridging the Gap Between Direct and Iterative Solvers

University Of Minnesota-Twin Cities, Minneapolis MN

Investigators

Abstract

0003274 Saad This three-year award for U.S.-France cooperative research involves a US team led by Yousef Saad at the University of Minnesota and French team led by Bernard Philippe of French National Institute for Research in Informatics and Applied Mathematics (INRIA) at Rennes, France. Other participants include Randall Branley, University of Indiana, Maria Sosonkina, University of Minnesota, and Esmond Ng, Lawrence Berkeley Laboratory. The French team is multi-institutional with researchers from INRIA's campuses in Lyon and Toulouse, from the CNRS and from IFREMER (the French public organization for research on the sea). The project addresses parallel iterative methods for solving large scale engineering and scientific problems on high performance computers. Many scientific and engineering applications give rise to very large linear systems that are ill-conditioned, and often highly indefinite. Existing iterative parallel algorithms developed for solution of large linear systems rarely take these characteristics into account, and as a result they often fail to converge. The project will explore new direct and iterative methods as a means for developing memory efficient and cost effective solutions for very large linear systems on high performance computers. This award represents the US side of a joint proposal to the NSF and the French National Institute for Research in Informatics and Applied Mathematics (INRIA). NSF will cover travel funds and living expenses for the US investigators. INRIA will support the visits of French researchers to the United States. The project takes advantage of complementary expertise in solvers in parallel computing. The methods will be applied to problems in geophysical prospecting and in electronic materials.