Parallel algorithm for particle-grid dual discretization

LY Zhao and S Wang and Y Liu, COMPUTATIONAL MECHANICS, 71, 257-276 (2023).

DOI: 10.1007/s00466-022-02237-z

Particle-grid dual discretization methods employ one group of Lagrangian particles and one set of background grid, which demonstrate attractive properties because they inherit both the advantages of Lagrangian and Eulerian methods. An efficient parallel algorithm for particle-grid dual discretization methods is proposed in this paper. The feature that all physical quantities are carried by particles is considered, so the domain repartition and particle migration can be carried out easily on demand of dynamic load balancing. The background grid suitable for both regular and irregular domain decomposition is generated according to partition of particles, and the communication of grid nodal information is carefully designed. The material point method (MPM), the smoothed molecular dynamics (SMD) method and the concurrent multiscale method coupling molecular dynamics, SMD and MPM with this parallel algorithm are implemented in the large-scale atomic/molecular massively parallel simulator (LAMMPS) code. One weak scaling example and four strong scaling examples are computed to demonstrate nice parallel efficiency of the proposed parallel algorithm. The results also show that the irregular domain decomposition based on recursive coordinate bisection algorithm can achieve nice parallel efficiency for complex problems.

Return to Publications page