Based The Jasmin Framework For Parallel Multi-layer Structure Of Fast Multipole Algorithm And Applied Research,

The fast analysis of the scattering characteristic of three-dimension structure has always been an intensive research in computational methods for electromagnetics. This paper focuses on the research of multilevel fast multipole algorithm (MLFMA), which is used to solve the complex three-dimensional electromagnetic scattering problems.Firstly, the paper introduces the integral equation method used in analysis of electromagnetic scattering problems of metal structures. Method of moment (MoM), which is used to solve the integral equation, is also studied. And, the principles and steps of fast multipole method (FMM) and multilevel fast multipole algorithm (MLFMA) based on MoM is detailed.Secondly, some optimization strategies for MLFMA are studied in this paper. By adopting modified multipole number method, the multipole number and number of spectrum direction in integration are decreased, as well as the CPU time and memory requirement in the solution, and the high accuracy is hold. The application in spectrum symmetry of aggregation terms reduces the memory requirement of aggregation and disaggregation terms by a factor of four for electrical field integral equation (EFIE), and two for combined field integral equation (CFIE). Interpolation technique, one of the most important components, is studied in this paper. The optimization of Lagrange interpolation and Gaussian interpolation is achieved and used to analyzes the root mean square (RMS) error of radar cross scattering (RCS).Thirdly, the study of precondition technique in MLFMA is implied in this paper. The block diagonal precondition (BDP) and sparse approximate inverse (SAI) precondition are studied. And, they are achieved in generalized minimum residual (GMRES) iteration solver to accelerate the convergence of the matrix equation. The iterative effects of the two different preconditions are compared with that the precondition is not applied.Then, in order to improve the ability of solution of large electrical size structure scattering problems, this paper has studied and developed the two kinds of parallel MLFMA. One is with OpenMP standard based on shared memory architecture. The other is with JASMIN frame based on distributed memory architecture. The JASMIN frame has an advantage of high portability and scalability, and a good load balancing. Although the infrastructure is structured and MLFMA is non-structured, the parallel procedure is accomplished overcoming the contradiction. C++/Fortran language and double precision data type are used to realize the procedure for high performance computing. The parallel efficiency of the two kinds of parallel MLFMA is analyzed in this paper, respectively.Finally, the parallel MLFMA based on JASMIN frame (JPMLFMA) is applied in electromagnetic scattering and radiation problems analysis in this paper. The numerical results of JPMLFMA are compared to analytical one, simulation ones of other software and MLFMA solutions based on OpenMP, which demonstrate the reliability and stability of the parallel algorithm.