Physical simulations compute criticality levels in neutronically active configurations of materials. These simulations are used, for example, to design power reactors, set safety limits for material processing, estimate fundamental neutron cross sections, and determine what new experiments would best reduce relevant application uncertainties. This talk surveys several topics we have encountered in projects to quantify predictive uncertainties for neutronics simulations that are sensitive to the functional inputs that describe how neutrons interact with atomic nuclei.
Topics covered include (i) computer experiment design over a collection of input functions; (ii) functional dimension reduction with regularized sliced inverse regression; (iii) scalar on function surrogate modeling and sensitivity analysis with Bayesian MARS; and (iv) effective display of the functional co-variation modes that drive output uncertainties. We have found this collection of methods to be useful in quantifying uncertainties for national security applications.
|