Studies on parallel processing for coupled field problems

Abstract Some concepts for the parallelization on distributed memory computers of a large finite element package for the numerical simulation of coupled problems are discussed in this paper. Two different approaches, based on physical and spatial decomposition of the coupled problem, are considered. The parallel algorithms were implemented on an iPSC/2 Hypercube, and results for a test problem and an industrial application are discussed. Moreover, a concept for a block-level domain decomposer is presented that guarantees good load balancing and allows the use of individual meshes for each part of the coupled problem. Adaptive modifications of either mesh in the course of the computation can also be included in this concept without significant reduction of efficiency.