Analysing the statistics of group constants generated by serpent 2 monte carlo code

An important topic in Monte Carlo neutron transport calculations is to verify that the statistics of the calculated estimates are correct. Undersampling, non-converged fission source distribution and inter-cycle correlations may result in inaccurate results. In this paper, we study the effect of the number of neutron histories on the distributions of homogenized group constants and assembly discontinuity factors generated using Serpent 2 Monte Carlo code. We apply two normality tests and a so-called "drift-in-mean" test to the batch-wise distributions of selected parameters generated for two assembly types taken from the MIT BEAVRS benchmark. The results imply that in the tested cases the batch-wise estimates of the studied group constants can be regarded as normally distributed. We also show that undersampling is an issue with the calculated assembly discontinuity factors when the number of neutron histories is small.