Challenges of Near Critical-Fixed Source Monte Carlo Simulations of CANDU-6 Reactor: Bundle Power Tally Bias and Error Autocorrelation from Exceeding Random-Number Stride

Abstract

This study quantifies the effect of exceeding random number stride (RN) in Monte Carlo simulations of subcritical, multiplying systems. We observed all fixed-neutron-source simulations of an accelerator-driven photoneutron source (ADS) coupled to a CANDU-6 full core model in different subcritical states readily exceeded the default RN stride up to a factor of 104. Bundle tally results from stride exceedance cases are biased by up to 4% relative to clean reference results using extended stride (no stride exceedance) due to correlated particle histories when the stride is exceeded, and the exact same RNs are used in the thermal diffusion random walk of the neutrons through the optically thick and repeating heavy water lattice. Error analysis of local bundle tallies clearly shows bias indicated by both a statistically significant number of relative errors exceeding three-sigma statistical uncertainty intervals of the reference solutions and strong autocorrelation of the relative errors correlated to the spatial relationship of bundle powers in the 3D core. Closer inspection on bundle-power variance of simulation where all histories exceeded the stride reveal there is no variance underestimation when stride exceedance occurs. This result is clearly contrary to the widely reported argument that Monte Carlo codes are insensitive to the stride exceedance and stride exceedance may cause underestimation of the variance solely based on the early study by Hendricks [1]. Monte Carlo practitioners should monitor RN usage statistics for problems where excessive RNs use may occur (i.e. subcritical multiplication, highly diffusing media, aggressive particle splitting, etc.) and reinvestigate the impact of stride exceedance in production radiation transport codes.