对高功率微波三维电磁PIC数值模拟中采用束流发射方式的粒子随机生成规律进行了研究.提出了一种基于共享与分布随机数的并行粒子发射算法,以满足多个处理机对单个随机数产生器产生的随机数序列的共享和分布需求,即共享部分随机数,而分别使用其他随机数.这些共享和分布需求在程序运行前是非确定性的.将该并行算法用于一典型高功率微波源器件中模拟,数值实验表明并行程序可以较好地吻合串行程序得到的粒子相空间物理图像.
As an initial and boundary conditions, the particle injection procedure strongly affects the overall accuracy of PIC(Particle-In-Cell) computation in three-dimensional fully electromagnetic simulations for high power microwave sources, where the random numbers are used unpredictable, which make it difficult to parallelize it. This paper investigates how the random numbers are used in this procedure, and finds it need share some random numbers from a single random number generator while distributing others on parallel computer. Which random numbers will be shared is unknown before the application runs. We provide a parallel randomized algorithm for particle injection to satisfy the requirements of sharing and distributing characteristic. Running the application using the parallel algorithm on hundreds of processors, the particle phase-space result agrees very well with the result of the serial program.