Scientific computing: Matrix partitioning and aggregation-based multigrid methods


Abstract : AGMG is a software code for the solution of linear systems that implements an aggregation-based algebraic multigrid method. The multigrid method is an iterative method that solves a linear system of equations by using a hierarchy of progressively smaller systems. An aggregation-based multigrid method builds this hierarchy by grouping unknowns into aggregates. AGMG works well in parallel but supposes that the system matrix has already been partitioned; that is, distributed on the different processors involved in the solution process. The aim of the work is to develop a parallel version that, starting from a globally defined matrix, performs itself this partitioning and distribute accordingly the computation between the available processors. In this view, an original approach will be developed, which uses the aggregation process to guide the partitioning. This approach will be then compared to those implemented in standard software. In a second stage, the partitioning of arbitrary graphs will be considered.
Promoteur/Supervisor : Prof. Notay Yvan
Email : ynotay@ulb.ac.be
Site Web/Web site : http://homepages.ulb.ac.be/~ynotay/
Centre de recherche/Research center : Métrologie Nucléaire (CP 165/84)
Faculté/Faculty : Brussels School of Engineering (Faculty of Applied Sciences)/Faculté des Sciences appliquées - école polytechnique
Ecole doctorale/Graduate Colleges : Engineering/Sciences de l'ingénieur
Ecole doctorale thématique/Graduate School (French Only):



Retour à la liste principale/Back to main list