Doctoral Dissertations

Date of Award

12-2025

Degree Type

Dissertation

Degree Name

Doctor of Philosophy

Major

Data Science and Engineering

Major Professor

Vladimir Sobes

Committee Members

Lawrence Heilbronn, Denise Neudecker, Jesse Brown

Abstract

Accurate uncertainty quantification for neutron cross sections is essential for credible transport simulations and decision support. In the resolved resonance region, generalized least--squares (GLS) fits implemented in \textsc{SAMMY} produce evaluated cross sections and covariances that are widely used, yet the associated uncertainties can be underestimated when ideal assumptions are violated. Manual variance inflation is common but undermines reproducibility. This dissertation develops a reproducible, evaluator‑in‑the‑loop pipeline that measures and corrects those distortions. The method couples \textsc{SYNDAT} with a scripted $R$‑matrix fitter (AutoFit) to produce labeled benchmarks and empirical evaluated cross‑section covariance matrices (ECSCMs) as ground truth. Controlled studies show that, under ideal GLS conditions, \textsc{SAMMY} matches empirical covariances. Otherwise, the dominant errors arise from missing resonances and using single‑level Breit–Wigner where multilevel coupling matters; moderate mis‑scaling of correlated experimental terms is comparatively benign. We learn per‑energy variance corrections from AutoFit features calibrated to empirical coverage. Random forests with a log‑normal residual family best balance accuracy and conservatism on held‑out ladders. We then extend to full energy‑ and channel‑covariances via a block‑feature design and nearest‑PSD projection, validating correlations with rotated coverage. The result is a transparent, data‑driven workflow that delivers calibrated, traceable ECSCMs for the resolved‑resonance region and supports in‑the‑loop use, with a path to on‑the‑loop quality control across evaluation campaigns.

Files over 3MB may be slow to open. For best results, right-click and select "save as..."

Included in

Data Science Commons

Share

COinS