A statistical analysis of results from Monte Carlo codes

Following analytical work by Brissenden and Garlick, in which they showed that the eigenvalue and eigenvalue variance estimates generated by conventional tracking in MONK would be negatively biased, we set up a numerical experiment to determine whether these systematic errors could be statisically detected. Results are given in Table 1. The eigenvalue bias could not be detected, but the eigenvalue variance bias was found to lower artificially the variance for the system under study by a factor of 2. Superhistory tracking in MONK6 removes this eigenvalue variance bias, and genuinely reduces the variance. An exploratory survey of KENO5 runs shows that if eigenvalue variance bias exists within this code, then it is small. The precision achieved by KENO5 and by superhistory tracking in MONK6 is about the same for a given number of neutron histories, for the particular system under study. (author)