Real Variance Estimation in Monte Carlo Criticality Calculation Accelerated by p-CMFD Feedback Using Spectral Analysis Method

Abstract

Performance enhancement of the spectral analysis method (SAM) for evaluating the real variance of local tallies from the partial current-based coarse-mesh finite difference (p-CMFD) feedback is verified and explained. In the SAM, on successive Monte Carlo (MC) cycles, the real variance is obtained from the cyclewise samples instead of an explicit evaluation of covariance. However, if the cycle correlation is strong, there is a bias and variance trade-off in the evaluated true uncertainty. This study shows that the p-CMFD feedback reduces the cycle covariance and hence eliminates the trade-off. A one-dimensional slab reactor and a three-dimensional simplified BEAVRS benchmark problem are analyzed, and the real standard deviation of the local tally is estimated from the SAM and compared with that from the conventional multibatch method. It is shown that the SAM with p-CMFD feedback can accurately calculate the real uncertainty without changing the MC algorithm and incurring computation burden.