Parallelization of radiation transport on unstructured triangular grids with spatial decomposition a

Abstract

Parallel computing can be of critical importance to the deterministic solution of radiation transport equations, especially with respect to solving the transport process within highly variable media with complex geometries. We propose an approach that extends the Alternating Direction of Transport Sweeps (ADTS) method of M. Yavus and E.D. Larson to unstructured triangular grids, using the Recursive Spectral Bisection (RSB) algorithm to partition our domain and OpenMP to implement the parallelism on shared memory computers. Our results, using a neutron transport test case, show that the ADTS method with the RSB algorithm leads to a significant increase in the parallel convergence rate, resulting in improved parallel efficiency, and thus improved turnaround time. We emphasize that excellent parallel efficiency is possible using domain decomposition and an SPMD-OpenMP implementation.