Methods for Efficient Computation of Neutron Multiplicity Distributions and SNM Sample Characterization

This dissertation advances neutron multiplicity counting (NMC) by developing a hierarchy of computational methods that combine deterministic and stochastic techniques with modern machine learning for improved special nuclear material (SNM) characterization. A system state-updating Monte Carlo method is introduced in addition to a dynamic point kinetic model based on a backward Master equation (BME) formulation. These models, along with newly developed distribution reconstruction methods, enable more efficient computation of neutron count distributions in a fixed detector time-gate. NMC limitations associated with finite-size and neutron phase effects are also addressed by extending the BME model to account for time-gated neutron count distributions in full phase space with implementation of the first four moment in LANL's deterministic transport code, PARTISN. The model's ability to incorporate arbitrary geometries and energy dependencies advances NMC capabilities, allowing for rapid computation of high-fidelity time-dependent moments. Machine learning algorithms are also applied to further improve SNM characterization. These innovations offer more accurate and cost-effective solutions, advancing nonproliferation and nuclear safeguards research.