SUMMARY

In particle transport, hybrid methodologies are used to reduce the simulation time required for converged Monte Carlo statistics. For shielding simulations, deterministic adjoint solutions provide information for source biasing and particle splitting. With accurate adjoint solutions, hybrid methodologies reduce run times by orders of magnitude in problems that would otherwise converge in unreasonable amounts of time. However, some problems span areas that require prohibitive amounts of memory. Previous work has shown that mesh refinement and coarsening can reduce memory requirements, but savings have been limited by using structured Cartesian meshing.

The proposed work will develop an algorithm to drastically reduce the amount of memory required for biasing information during particle transport simulations. Information will be refined or coarsened based on Contributon theory, which provides an estimate of the flow of simulated particles. The method requires a novel data structure to be implemented that uses separate unstructured Cartesian mesh schemes by particle energy. Ideally, there will be minimal increase in simulation time from the use of a more complex data structure with small loss in biasing efficiency from mesh reduction.