|SUBJECT:||M.S. Thesis Presentation|
|TIME:||Tuesday, May 4, 2010, 11:00 a.m.|
|PLACE:||Boggs Building, 3-47|
|TITLE:||Entropy-based Diagnostic of Criticality Monte Carlo Simulation and Higher Eigenmode Acceleration Methodology|
|COMMITTEE:||Dr. Bojan Petrovic, Chair (NRE)
Dr. Farzad Rahnema (NRE)
Dr. Dingkang Zhang (NRE)
Because of the accuracy and ease of implementation, Monte Carlo methodology is widely used in analysis of nuclear systems. The obtained estimate of the multiplication factor (keff) or flux distribution is statistical by its nature. In criticality simulation of a nuclear critical system, whose basis is the power iteration method, the guessed source distribution initially is generally away from the converged fundamental one. Therefore, it is necessary to ensure that the convergence is achieved before data are accumulated. Discarding a larger amount of initial histories could reduce the risk of contaminating the results by non-converged data but increases the computational expense. This issue is amplified for large loosely coupled nuclear systems with low convergence rate. Since keff is a generation-based global value, frequently no explicit criterion is applied to the diagnostic of keff directly. As an alternative, a flux-based entropy check available in MCNP5 works well in many cases. However, when applied to a difficult storage fuel pool benchmark problem, it could not always detect the non-convergence of flux distribution. Preliminary evaluation indicates that it is due to collapsing local information into a single number. This thesis addresses this problem by two new developments. First, it aims to find a more reliable way to assess convergence by analyzing the local flux change. Second, it introduces an approach to simultaneously compute both the first and second eigenmodes. At the same time , by computing these eigenmodes, this approach could increase the convergence rate. Improvement in these two areas could have a significant impact on practicality of Monte Carlo criticality simulations.