IMPLEMENTATION OF PARTICLE-IN-CELL METHOD ON DISTRIBUTED-MEMORY MULTIPROCESSORS

V.A. Vshivkov, G.L. Dudnikova, M.A. Kraeva, V.E. Malyshkin
VANT. Ser.: Mat. Mod. Fiz. Proc 1997. Вып.1. С. 67.

      The particle-in-cell method (PIC) is commonly used for the calculations in collisionless plasma physics where a great importance is given to the interaction through electromagnetic fields.
      The PIC method represents plasma as a set of model particles with the motion trajectories being the characteristics of Vlasov equation.
      Since the evaluation of new velocities and coordinates of the particle does not directly depend on the other particles the problem is well suited to parallelization. However depending on the technique for data distribution (particles, and field values at the grid nodes) among the processors we deal either with nonuniform loading of processing elements (PEs) or high communications overhead. These problems appear in the case of nonuniform distribution of particles in the modeling space. Thus the parallel code textually depends on the law of particles distribution in the modeling space. In addition, the decision about the data and computation distribution among the processing elements depends on the pattern of particle expansion, their amount, grid size, the difference between sequential algorithms and of course on the computer system architecture, particularly on PE memory size.
      The work presents the development results, for the programming system intended for the implementation of various PIC method versions. The work is performed within the ASSY project (ASsembly SYstems). The project is oriented to the development of the metasystem supporting the development of problem-oriented programming systems. PIC method is one of the problems used to verify actual capabilities of ASSY technology.
      The work was carried out under the auspice of Russian Fundamental Research Foundation (project N 95-01- 01358) and the Commission of the European Communities (grant ITDC-203-822165).