蒙特卡罗区域分解并行计算中确保串并行结果一致的伪随机数应用  

Pseudo-Random Numbers for Identical Results on Varying Numbers of Processors in Domain Decomposed Particle Monte Carlo Simulations

作　　者：李刚[1,2] 张宝印[1,2] 邓力[1,2] 上官丹骅[1,2] 李瑞 马彦[1,2] 付元光 胡小利 LI Gang ZHANG Baoyin DENG Li SHANGGUAN Danhua LI Rui MA Yan FU Yuanguang HU Xiaoli(Institute of Applied Physics and Computational Mathematics, Beijing 100094, China CAEP Software Center for High Performance Numerical Simulation, Beijing 100088, China)

机构地区：[1]北京应用物理与计算数学研究所,北京100094 [2]中物院高性能数值模拟软件中心,北京100088

出　　处：《计算物理》2017年第1期67-72,共6页Chinese Journal of Computational Physics

基　　金：能源局重大专项子项(2015ZX06002008);国防科工局(C1520110002);国家磁约束核聚变能源研究专项(2015GB108002)资助项目

摘　　要：物理建模的精细化和三维模拟给蒙特卡罗粒子输运计算的规模成千上万倍的增加,甚至超过单核内存的最大规模,仅仅依靠传统的粒子并行蒙特卡罗计算无法实现对模型模拟,区域分解并行是可能的解决方法之一.然而区域分解带来了粒子在各区域间进行迁移,导致现有的伪随机数应用方式无法确保串行计算和并行计算的结果一致.针对这种现象,本文提出赋予粒子随机数属性和动态派生次级粒子随机数的技巧来确保区域分解并行计算的串并行结果一致.Detailed modeling and massive tallying of nuclear reactors lead to memory overload for a single core processor. It could not be calculated by Monte Carlo particle transport with particle parallelism only. Domain decomposition is one of solutions. Domain decomposition needs to interchange particles between processors, so that inherit technique of pseudo-random number could not make identical results between serial and parallel. Two techniques of pseudo-random number are described to obtain identical results on different numbers of domains in a Monte Carlo particle simulation code.

关 键 词：伪随机数 蒙特卡罗方法 区域分解并行 JCOGIN框架 JMCT程序 

分 类 号：O571.51[理学—粒子物理与原子核物理]