ADVANCES IN PARALLEL ADAPTIVE SIMULATION ON UNSTRUCTURED MESHES

Parallel simulations of time-dependent problems on unstructured meshes using h-adaptation require the cooperation of numerical methods, distributed mesh management and dynamic load balancing and migration. Parallel-adaptive solution schemes, e.g. for tracking a front, incorporate mesh changes preferably in regions of significant solution phenomena. This work focuses on advances in parallel adaptive computing on unstructured meshes. We present recent extensions to a parallel programming model, which allow a more efficient and stable realization of small mesh changes caused by grid adaptation. For reasons of algorithmic complexity processor local changes result only in incremental operations preserving a global and consistent view onto the distributed mesh. While tracking fronts with parallel grid adaptation processor local computation load varies significantly. Thus dynamic repartitioning of the already distributed computation load is needed. We present and compare two schemes PRCB and MCAR capable to do a rebalancing in the context of multiplicative multigrid as optimal-complexity solver inside the numerical scheme. Two applications from reservoir engineering demonstrate how these new capabilities are efficiently used in 2D and 3D parallel-adaptive simulations which require mesh adaptation and dynamic load balancing and migration during run-time.