Convergence Analysis and Criterion for Data Assimilation with Sensitivities from Monte Carlo Neutron Transport Codes

Sensitivity coefficients calculated with Monte Carlo neutron transport codes are subject to statistical fluctuations. The fluctuations affect parameters that are calculated with the sensitivity coefficients. The convergence study presented here describes the effects that statistically uncertain sensitivities have on first-order perturbation theory, uncertainty quan-tification, and data assimilation. The results show that for data assimilation, posterior nuclear data were remarkably uninfluenced by fluctuations in sensitivity mean values and by sensitivity uncertainties. Posterior calculated values computed with first-order perturbation theory showed larger dependence on sensitivity mean-value convergence and small uncertainty arising from the sensitivities' uncertainties. A convergence criterion is proposed for stopping simulations once the sensitivity means are sufficiently converged and their uncertainties are sufficiently small. Employing this criterion economizes computational resources by preventing an excess of particle histories from being used once convergence is achieved. The criterion's advantage is that it circumvents the need to set up the full data assimilation procedure, but is still applicable to data assimilation results.