A suitable domain decomposition for the adaptive mesh refinement method

Abstract

We applied the dynamic Domain Decomposition Method (DDM) to the local Adaptive Mesh Refinement method (AMR) and parallelize d the calculations. As a result, it became possible to equalize local increases in load and thus to obtain a high performance. Specifically, when 8PE's were used for the flow problem in jurisdiction with the difference, a 6.2 times speed improvement ratio was obtained. Furthermore, a limitation was added to the calculation area's shape since the calculation area allocated to each PE generally becomes complex in shape when using dynamic division. The complexity of communication was suppressed by applying this limitation and the coding complexity problem was avoided.