Parallel processing of random number generation for Monte Carlo turbulence simulation (Q1116628): Difference between revisions

In Monte Carlo computations on multi-processor computers used as a model for drift-wave turbulence in plasma physics, pseudo-random numbers generators with the following characteristics are requested: 1) numbers given to each processor should be disjoint sets; 2) the sets are reproducible run to run; and 3) the sets of numbers are the same as those obtained by using the same random number generator on computers with different numbers of processors. A simple method is given for generating reproducible sets of pseudo-random numbers with these characteristics. The method employs hopping through the sequence at large equal interval. The relation with the leapfrog method [\textit{K. O. Bowman} and \textit{M. T. Robinson}, Proc. Second. Conf. Hypercube Multiprocessors, SIAM 445 (1987)] is mentioned.