Speaker
Description
Reliable nuclear data and well-characterized uncertainties are crucial for predictive reactor physics modeling, safety analysis, criticality assessment, and the design of advanced nuclear systems. Nuclear data uncertainties, arising from experimental measurement errors, theoretical model deficiencies, and evaluation procedures, often introduce uncertainty in key integral parameters such as the effective multiplication factor. Quantifying, reducing, and transparently propagating these uncertainties is therefore a cornerstone of next-generation reactor analysis.
We present a reproducible and automated nuclear data evaluation pipeline being developed at Uppsala University that addresses the uncertainty propagation in nuclear cross-section data through structured integration of experimental data, nuclear reaction modeling, and statistical inferences [1]. The pipeline is designed as a modular workflow to facilitate systematic evaluation of neutron-induced cross sections, focusing on the fast neutron energy range above the resolved resonance region. Reproducibility and ease of deployment are achieved via containerized execution, enabling consistent environments across computational platforms and facilitating collaborative development and system independent output.
The core concept of the pipeline combines the TALYS [2] nuclear model parameters with differential experimental data, primarily sourced from EXFOR nuclear database, to quantify uncertainty in cross-section predictions. A tailored Levenberg-Marquardt optimization algorithm is employed to calibrate non-linear model parameters against experimental observations [1], embedding sensitivity to both statistical and systematic data uncertainties within the calibration process. A particular emphasis of the pipeline lies in the explicit treatment of model defects, especially where smooth theoretical predictions diverge from unresolved resonance-like structures present in experimental data. This is addressed through the inclusion of heteroscedastic Gaussian process models that adaptively characterize energy-dependent deviations between model and experiment [3]. The result is evaluated cross-sections with integrated uncertainty descriptions that reflect both measurement limitations and theoretical modeling uncertainty.
In this contribution, we will outline the design and methodological development of the nuclear data evaluation pipeline, including data ingestion, uncertainty analysis, and covariance matrix generation. We illustrate its relevance to reactor physics applications by discussing how the evaluated data can be leveraged in uncertainty studies for generation IV reactor concepts.
References:
[1] G. Schnabel et al., Nuc. Data Sheets 173, 239 (2021).
[2] A. J. Koning and D. Rochman, Nucl. Data Sheets 113, 2841 (2012).
[3] A. Göök et al., EPJ Web Conf. 294, 04005 (2024).