Large-scale simulation of discrete dislocation dynamics based on dynamic heterogeneous parallel fast multipole method

Discrete dislocation dynamics (DDD) studies the plastic deformation of structural materials by directly simulating the collective evolution of a large number of dislocations, revealing the intrinsic physical correlation between material microstructure, dislocation microstructure, and plastic mechanical behavior. This paper proposes PFM-DDD, a novel parallel computation framework designed to combine DDD with fast multipole method (FMM) for simulating plastic deformation with over millions of degrees of freedom. PFM-DDD achieves large-scale fast simulation through three synergistic modules: (1) custom data structures linking FMM with DDD for efficient storage and retrieval of dislocation segments and FMM cells; (2) adaptive task partitioning to mitigate load imbalance issues in parallel simulations; (3) two-stage FMM heterogeneous parallel DDD design, including subdomain-level parallel FMM on CPUs and GPU-accelerated strategies. Experimental results demonstrate that PFM-DDD has good accuracy and exhibits superior computational performance compared to the state-of-the-art DDD simulation program ParaDiS.