Automatic Variance Reduction for Monte Carlo Transport
Gregory Davidson, University of Michigan
Since the study of computational particle transport began, two different solution methodologies have emerged. The first is a stochastic method, Monte Carlo, which is based on averaging the results of a large number of random particle histories. The second method is deterministic, which involves numerically solving the linear Boltzmann transport equation via discretization. These techniques are fundamentally different and have been developed independently.
Recently there have been efforts to develop hybrid methods, which draw on the different strengths of Monte Carlo and deterministic transport methods. When using Monte Carlo to solve difficult problems, variance reduction techniques must be used in order to obtain the solution in a timely manner. Unfortunately, current Monte Carlo variance reduction methods require the user to have knowledge of the solution of the problem. We are investigating hybrid techniques to provide automatic variance reduction using deterministic transport methods.
Abstract Author(s): Greg Davidson, Dr. Edward Larsen