Keywords: numerical methods, linear algebra, parallel processing, distributed memory , eigenvalue problems, scheduling
Start Date: 1 October 94 / Duration: 36 months
[ participants / contact]
To carry out research on the implementation of numerical linear algebraic computations using parallel distributed-memory architectures. We focus on eigenvalue-eigenvector computations of importance for industrial applications such as structural analysis, seismic analysis and molecular dynamics. The goal is to provide results on the granularity and scheduling of jobs for optimum exploitation of a range of new distributed architectures. In order to extend to other groups in Tunisia and Algeria, these groups are led by researchers who have already visited the LIP or the FP Mons.
A French parallel computer was installed and inaugurated in September 1996 in Oujda. This has been used for parallel eigenvalue computations using the 2-face Jacobi method, the Jacobi method exploiting symmetry of the problem, parallel eigenvalue computations with the Jacobi one sided method, new methods for the parallel eigenvalue computation using Yau and Lu sequential ideas, and complexity studies for different parallel algorithms on different models of parallel machine.
The results of these works were presented in scientific conferences. They have led to one PhD thesis and a number of research reports. A first workshop was organised in August 1996 with another collaboration project involving the CNRS-LIP and the NSF-University of Tennessee. Further cooperation will start between this group and the ITDC project involving the LIP and the FP Mons with partners from South America.
Allée d'Italie 46
69364 Lyon, F
ENS Lyon, F
Faculté Polytechnique de Mons, B
Université Mohamed I, Oujda, MA
Mr. B. Tourancheau
Tel: +33 72 728 000
Fax: +33 72 728 080
NUMLINALG - KIT108, May 1997
please address enquiries to the ESPRIT Information Desk
html version of synopsis by Nick Cook