Acceleration of Monte Carlo shielding calculations with an automated variance reduction technique and parallel processing

This thesis describes the acceleration of Monte Carlo (MC) shielding calculations with an automatic variance reduction technique and parallel processing. The applicability of the analog MC method is limited by its computational expense. Thus, biasing techniques, which require a great deal of experience, time, and effort, are employed to make reactor shielding calculations feasible. To overcome this difficulty, a method has been developed for using the SN adjoint function for variance reduction of MC calculations through source biasing and consistent transport biasing with the weight window technique. This method is implemented into the standard production MC code MCNP and is applied to a realistic calculation, namely the reactor cavity dosimetry calculation. The computational effectiveness of the method, as demonstrated through the increase in calculational efficiency, is demonstrated and quantified. Important issues associated with this method and its efficient use are addressed and analyzed. Additional benefits in terms of the reduction in time and effort required of the user are difficult to quantify, but are possibly as important as the computational efficiency. The limitation of this approach is the requirement for an SN adjoint solution, which requires the user to be knowledgeable in both MC and deterministic methods; which is not often the case. Therefore, algorithms have been developed and implemented into MCNP for automatically generating input files for SN adjoint calculations directly from the MCNP problem description, including mesh generation and material cross-section preparation. This automation eliminates the tedious process of manually generating these files and requires very little experience on the part of the user. For additional acceleration, MCNP has been adapted to parallel processing with the Message Passing Interface (MPI). Parallel performance of two reactor application problems was examined. Practical use of parallel computing (with 16 processors) is shown to decrease the amount of time required for large/complex transport calculations by more than an order of magnitude. The combination of these acceleration approaches can result in dramatic performance increases and substantially reduce the current requirements for analylst's experience and time, and thus, greatly increase the applicability and reliability of the MC method for large shielding applications.