Abstract
Smoothed particle hydrodynamics with adaptive spatial resolution (SPH-ASR) suffers from load imbalance, communication redundancy, and inefficient neighboring particle search in parallel frameworks due to the non-uniform particle distribution. This work proposes a new tree-based parallel model with message passing interface (Tree-MPI model) for the SPH-ASR method to handle these challenges. A recursive binary-tree domain decomposition is implemented, which can adaptively allocate the hardware resource and chooses the split dimension during subdomain partitioning. In the hierarchical data structure, a load balancer based on synchronization times of processes is introduced, along with a novel data communication approach that leverages adaptive boundary cells. Besides, a parallel neighboring particle search is proposed by constructing multi-background cells in a local nested tree to address problems caused by variable smoothing lengths. Five benchmark cases concerning free surface, multiphase interface, and fluid-structure interface are studied. The results imply the generality, excellent parallel efficiency, and optimized memory usage of the Tree-MPI model of SPH-ASR for different complex flows. In the industrial problem with 100 million particles, it achieves a maximum parallel efficiency of 90% on 1000 cores. This indicates that the present model overcomes the limitation in large-scale SPH simulations through both algorithmic and hardware optimizations.