Domain decomposition: parallel multilevel methods for elliptic partial differential equations
Domain decomposition: parallel multilevel methods for elliptic partial differential equations
| Hi-index | 0.00 |
This paper describes a new parallel algorithm for solving multiphysics problems. These kind of problems are very demanding in terms of CPU time and memory space, which are typically not available on a single processor. Using domain decomposition techniques it is possible to divide the original computational domain into subdomains, which need less memory and may be distributed onto the processors of a parallel computer. In this work, we introduce a hybrid domain decomposition, which uses nonoverlapping as well as overlapping partitions. The solution algorithm is a combination of the alternating Schwarz method and a block Gauss - Seidel scheme. This algorithm has been used to parallelize a Finite Element program able to calculate coupled thermomechanical problems, using a staggered solution strategy. Tests of the parallel algorithm with a realistic example show a rather good speed up.