Improvement in treatment of finest level for parallel multilevel fast multipole algorithm

A novel design project on parallel algorithms is presented based on message passing interface (MPI), which comes from the implement process of the multilevel fast multipole algorithm (MLFMA). The parallel octree based on cubic boxes is improved to reduce the communication of the finest level. Considering the main components of the MLFMA memory, a method that the correlational near interaction matrix element of integral equations is build by distributing to each compute node equally is employed to improve the parallel efficiency. This method could ensure the related communication is close to zero and save memory when computing independently in each node. The memory distribution of other physics parameters is optimized. The numerical results show that the parallel efficiency is obviously improved and an excellent precision solution with good workload balancing is gained.