Population-based variance reduction for dynamic Monte Carlo
Dávid Légrády, Máté Halász, J. Kópházi, Balázs Molnár, Gábor Tolnai
Abstract
Dynamic Monte Carlo (DMC) simulation of realistic nuclear reactors requires powerful variance reduction methods for even a few seconds of real time calculations. State-of-the-art numerical methods deal with the dynamic nature of the problem via successive Monte Carlo transport and TH (thermal-hydraulic) runs in a time step by time step manner. Such halting of the sample population at the beginning of time steps also allows for a joint handling of samples in a variance reduction effort. A theoretical framework is given for the connection of weight distribution and tally variance by factorizing it into a population variance accumulated by previous time steps and the variance caused by the transport process in the last interval. A long term importance function is proposed for decreasing the main contribution of a power release tally variance. Novel techniques are shown and compared when using Russian Roulette and Splitting. A simple fast critical assembly and a detailed thermal reactor geometry are used for testing showing that a factor of at least two orders of magnitude is to be gained by a simple population comb targeting the average weight. Further improvement using importance and variance functions is less than a factor two.