Disciplines
Computer Sciences (30%); Mathematics (70%)
Keywords
MULTILEVEL PRECONDITIONING,
ALGEBRAIC MULTIGRID,
SCALABLE ALGORITHMS
Abstract
Erwin Schrödinger Fellowship J 1915 Algebraic Multilevel Methods Johannes KRAUS 08.05.2000
This project deals with the development and exploration of new algebraic multilevel preconditioners and algebraic
multigrid (AMG) methods for large sparse matrix problems in multiphysics applications. The major issues are the
development of effective (parallel) grid-coarsening strategies for unstructuredgrid problems, the development of
efficient smoothers, and an adaptation of certain methods to nonsymmetric problems. Special emphasis is put on
the elaboration of robust and scalable algorithms. Apart from theoretical investigations I also intend to implement
the derived algorithms on a multiprocessor supercomputer and to perform a scalability analysis. Some application
studies in the area of fluiddynamics, e.g., convection dominated flow problems, and elastomechanics, especially
thin-body elasticity problems, will round off the project work.