Adjoint-derived weight windowing is a hybrid deterministic/Monte Carlo method to simulate radiation transport. In adjoint-derived weight windowing, a deterministic adjoint solution is used to create weight windows for a Monte Carlo simulation. The intent of this work is to identify factors that reduce the Figure of Merit (FOM) of Monte Carlo simulations using adjoint derived weight windowing. The method
used in this study pairs Transpire's deterministic code Attila™ and MCNP5. Two computationally difficult source/detector problems of interest to nuclear nonproliferation are used as case studies to determine the factors that affect the FOM.
Test Case I is an active interrogation problem similar to many radiography problems. The model is used in two sets of trials: in the first, the quality of the deterministic adjoint solution is varied to observe the effect of adjoint solution quality on the FOM. In the second, the shielding density is varied to determine the effect of increased shielding on the FOM. Results from Test Case I suggest that weight windows that decrease monotonically along relevant paths from the source to the detector maximize the FOM. The results also suggest that weight windowing is susceptible to false convergence that could be avoided using a different hybrid method, such as the Local Importance Function Transform (LIFT). A more sophisticated method for generating weight windows relevant to the forward Monte Carlo simulation is described for future work.
Test Case II is a detailed model of a detector array passively interrogating a uranium hexafluoride cylinder. Test Case II is used to test the effect of appropriate source biasing on the FOM.
Results from Test Case II confirm prior work, that source biasing is important for problems in which the adjoint function varies widely in the source domain. Since spectral information from the detector is very useful for nonproliferation purposes, a new use of the forward weighted consistent adjoint driven importance sampling
(FW-CADIS) method is described to model the energy-dependent
flux in a region of interest. Properly modeling Test Case II also requires the use of rejection sampling
of the source position paired with source biasing, which currently cannot be used together in MCNP5. The new use for the FW-CADIS method and a method to allow the use of rejection sampling with source biasing are described for future work.