Nuclear Data Uncertainty Propagation in Neutron Transport Computations
Dr. Anil K. Prinja
Chemical and Nuclear Engineering Department
University of New Mexico, Albuquerque
The propagation of aleatoric uncertainties, or irreducible uncertainty in physical data and model parameters, in the simulation of complex physical phenomena, and the quantification of the resulting uncertainties in the output variables or the system response, is an essential ingredient of large-scale code design and system analysis. Uncertainly analysis provides a level of confidence in the output variables, for example through explicit confidence interval estimates, and promotes an assessment of the relative importance of parameters that are not known with adequate precision, due either to gaps in measurements or to simplifying assumptions in the underlying physics. In nuclear reactor physics applications, uncertainties in nuclear data, such as cross sections, average fission multiplicity, and prompt fission neutron spectra, when propagated through neutron transport codes lead to uncertainties in fluxes and key integral parameters such as the system criticality or k-eigenvalue, leakage rates, and various spectral indices.
Quantification of output variable uncertainty highlights deficiencies in nuclear data and models and helps direct further experimental and theoretical research to reduce uncertainties and improve margins. In this talk we present a stochastic method for propagating nuclear data uncertainty based on spectral expansions, using Karhunen-Loeve and Principal Component representation of input variable uncertainty and generalized Polynomial Chaos expansion of the output variable uncertainty. We show that existing computational methods and codes developed for application without uncertainty can be directly used within this framework, i.e., the approach is nonintrusive, and demonstrate its application to the computation of the mean, variance and probability distribution of the k-eigenvalue, the prompt fission neutron spectrum and other integral parameters as well as the scalar flux. The spectral method is shown to be convergent, efficient and capable of high accuracy at reasonable cost, in contrast to traditional random sampling or Monte Carlo approaches to uncertainty quantification.
Dr. Anil K. Prinja is currently Professor and Program Director of Nuclear Engineering in the Chemical and Nuclear Engineering Department at the University of New Mexico. He obtained his Ph.D. (1980) and B.Sc. in Nuclear Engineering from the University of London, UK, and was a Research Faculty at UCLA (1980-1987) prior to joining UNM. He has affiliations with Los Alamos National Laboratory, Imperial College, London, and Politecnico di Torino, Italy. Dr. Prinja’s research interests are in applied theory and numerical methods for neutral and charged particle transport, with specific thrusts in: reduced order physics models and numerical algorithms for high energy charged particle beams; stochastic methods for uncertainty quantification; and neutron branching processes in fissile systems.