Parallelization of Large Scale Adaptive Finite Element Computations

摘要

The paper discusses general principles for parallelization of finite element codes and presents a model architecture for computational kernels of parallel adaptive programs. The model is based on message passing and domain decomposition. Parallel adaptivity, load balancing and multi-level (multigrid) solvers for systems of linear equations are included into the model. Numerical examples, computed using a prototype implementation of the proposed architecture, illustrate capabilities of the approach.