Parallel Monte Carlo Particle Transport and the Quality of Random Number Generators: How Good is Good Enough?
It might be assumed that use of a ''high-quality'' random number generator (RNG), producing a sequence of ''pseudo random'' numbers with a ''long'' repetition period, is crucial for producing unbiased results in Monte Carlo particle transport simulations. While several theoretical and empirical tests have been devised to check the quality (randomness and period) of an RNG, for many applications it is not clear what level of RNG quality is required to produce unbiased results. This paper explores the issue of RNG quality in the context of parallel, Monte Carlo transport simulations in order to determine how ''good'' is ''good enough''. This study employs the MERCURY Monte Carlo code, which incorporates the CNPRNG library for the generation of pseudo-random numbers via linear congruential generator (LCG) algorithms. The paper outlines the usage of random numbers during parallel MERCURY simulations, and then describes the source and criticality transport simulations which comprise the empirical basis of this study. A series of calculations for each test problem in which the quality of the RNG (period of the LCG) is varied provides the empirical basis for determining the minimum repetition period which may be employed without producing a bias in the mean integrated results.
- Research Organization:
- Lawrence Livermore National Lab. (LLNL), Livermore, CA (United States)
- Sponsoring Organization:
- US Department of Energy (US)
- DOE Contract Number:
- W-7405-ENG-48
- OSTI ID:
- 15015930
- Report Number(s):
- UCRL-CONF-208668; TRN: US0501667
- Resource Relation:
- Conference: Presented at: Monte Carlo 2005, Chattanooga, TN (US), 04/17/2005--04/21/2005; Other Information: PBD: 7 Dec 2004
- Country of Publication:
- United States
- Language:
- English
Similar Records
Vectorization of Random Number Generation and Reproducibility of Concurrent Particle Transport Simulation
Random Number Generation for Petascale Quantum Monte Carlo