A Sampling-Based Approach to Nuclear Data Uncertainty Propagation Using SANDY

Quantifying nuclear data uncertainties is an essential component of modern reactor analysis: it underpins the Best Estimate Plus Uncertainty methodology and directly influences reactor safety, performance, and economic assessments. This article presents a robust, sampling-based framework for nuclear data uncertainty quantification (UQ) using the SAmpler of Nuclear Data and uncertaintY (Sandy) tool, demonstrated across multiple reactor analysis types, using both Monte Carlo and deterministic solvers. Using the Total Monte Carlo method, perturbed nuclear data are generated with Sandy and Njoy from evaluated libraries and propagated through three complementary case studies: (1) depletion calculations for a light-water reactor (LWR) pin-cell model (VVER-1000); (2) standalone criticality and cross-section calculations for a lead-cooled fast reactor (LFR) pin-cell model; and (3) coupled multi-physics calculations for the C5G7 LWR benchmark core using the nodal diffusion code Nem and the thermal-hydraulics code Ctf.A sample size of 500 perturbed datasets was found to be sufficient for convergence in all three studies. The VVER-1000 depletion study revealed anomalies in the ENDF/B-VIII.0 U-235 covariance matrix that produces erroneously large uncertainties, underscoring the importance of library quality verification. The LFR criticality study showed that ENDF/B-VII.1 produces the largest keff uncertainty among the three libraries tested, while ENDF/B-VII.1 also exhibits a pronounced flux-spectrum uncertainty spike in the ∼0.6–6 MeV range. The C5G7 multi-physics study yielded keff uncertainties near 450 pcm in both the standalone Serpent and the coupled Nem/Ctf calculations, with the relative pin power showing the highest sensitivity to nuclear data uncertainty (maximum 1.87%) and coolant density the lowest (maximum 0.25%). Taken together, these results demonstrate the sampling-based approach as a robust and powerful general-purpose tool for nuclear data UQ, applicable across diverse reactor and analysis types.