Collection of abstracts 14th
GAMM-Seminar Kiel on
|
We introduce an abstract model for linear algebra computations on distributed overlapping partitions of indices on different levels. Local refinement can be modeled by active sets of indices and the identification of indices on successive levels.
We show that the specification of the parallel linear algebra is important for the performance of the resulting algorithms. Furthermore, in the implementation most parallel modifications of the algorithms are hidden in the linear algebra module. In particular, nearly all other numerical algorithms such as assembling, smoothing and linear, nonlinear and time dependent solvers can be implemented without any parallel extensions. At least, the model can check the required consistency and indicates, where the model is violated.
Finally, we demonstrate that the specification of the parallel linear algebra yields parallel realizations of geometrical multigrid methods, algebraic multigrid methods and multigrid methods for mortar elements.
Mail to WebMaster |
[Wed Dec 17 10:55:09 MET 1997] |
Impressum |