Information Criteria and Higher Eigenmode Estimation in Monte Carlo Calculations

Recently developed Monte Carlo methods of estimating the dominance ratio (DR) rely on autoregressive (AR) fittings of a computed time series. This time series is obtained by applying a projection vector to the fission source distribution of the problem. The AR fitting order necessary to accurately extract the mode corresponding to DR is dependent on the number of fission source bins used. This makes it necessary to examine the convergence of DR as the AR fitting order increases. Therefore, we have investigated if the AR fitting order determined by information criteria can be reliably used to estimate DR. Two information criteria have been investigated: Improved Akaike Information Criteria (AICc) and Minimum Descriptive Length Criteria (MDL). These criteria appear to work well when applied to computations with fine bin structure where the projection vector is applied.