PH.D Olaf Schenk
Department of Computer Science, Uni Basel
Combinatorial Approaches to the Solution of Highly Indefinite Symmetric Matrices
Abstract: We will analyze the use of combinatorial algorithms for the solution of symmetric indefinite matrices. In particular, we will use symmetric weighted graph matchings to build efficient preconditioning algorithms for highly indefinite systems. These combinatorial approaches have been recently used in sparse direct solvers as an interesting alternative to pivoting. This technique results in a factorization for symmetric indefinite matrices that is as scalable as the highly scalable Cholesky method. In this talk we will apply these combinatorial methods for the preconditioning within iterative methods. Various test cases are considered to demonstrate the stabilizing effectiveness and the efficiency of the method e.g. in the large-scale interior-point optimizations package IPOPT and from eigenvalue problems in computational physics. Our numerical examples reveal that these combinatorial solvers can accelerate the solution of symmetric indefinite systems by several orders of magnitude.| Zeit: | Freitag, 21. Oktober 2005, 16.15 Uhr (Kaffee/Tee um 15.30) |
| Ort: | FU Berlin, Arnimallee 2-6, Raum 032 im EG |