Prof. Dr. W. Hackbusch
Max-Planck-Institut für Mathematik
Universität Kiel
Abstract: The concept of hierarchical matrices is a further development of the panel clustering method. While in the panel clustering method the efficient matrix-vector multiplication is the main goal, the format of the hierarchical matrices is introduced to enable an (approximate) arithmetic of matrices including addition, multiplication and inversion of matrices. The subclass of hierarchical matrices is data-sparse, i.e., only a storage of the size O(n log n) is needed for nxn matrices. The addition is of the same size, while the matrix-matrix-multiplication and inversion requires O(n log**2 n) operations. The hierarchical matrices can be used to approximate discrete elliptic operators, in particular integral operators as they appear in the boundary element method.
Zeit: | Freitag, 18. Juni 1999, 14.15 Uhr |
Ort: | FU Berlin, Arnimallee 2-6, Raum 032 im EG |