AbstractsTransportation

The accurate modeling of complex physical phenomena, such as radiation transport in a laboratory experiment or a nuclear reactor, challenges the limits of modern computing resources and helps drive the requirement for next-generation exascale computers. The efficient and accurate propagation of uncertainty through these models is an area of ongoing research. Uncertainties in material properties contribute to the uncertainty in quantities of interest (QoIs) for which a problem is solved, but quantifying the uncertainty in a QoI is often prohibitively expensive. In this research an integrated approach to uncertainty quantification (UQ) for radiation transport problems with uncertain nuclear data is introduced. A novel dimension reduction method is applied to the nuclear data characterizing cross-section uncertainty. An adjoint-based sensitivity analysis is performed to yield sensitivity co-efficients for the QoI with respect to the reduced-dimensional space. Finally, response surfaces are constructed for the QoI over the reduced-dimensional input space. These surfaces yield information about the distribution of the QoI over the original uncertain input space. This multi-step approach is applied to several radiation transport problems for which traditional UQ methods are prohibitively expensive. Advisors/Committee Members: Morel, Jim E (advisor), Ragusa, Jean C (advisor), Adams, Marvin L (committee member), Guermond, Jean-Luc (committee member).