Speaker
Description
Zirconium-90 is the most abundant isotope in natural zirconium, a material that serves as fuel cladding and structural assembly component in the majority of operating nuclear reactors. Accurate neutron-induced cross-section data for Zr-90, together with reliable uncertainty and covariance information, are therefore directly relevant to reactor and burnup calculations.
We present an evaluation of Zr-90 neutron-induced cross-sections in the fast-energy range, carried out using the Nuclear Data Evaluation Pipeline of Uppsala (NEPU). NEPU is built around the TALYS nuclear reaction code [1] and a Bayesian statistical framework, in which a prior-aware Levenberg-Marquardt algorithm is used to locate the maximum a posteriori estimate of model parameters given experimental data from the EXFOR database. One methodological challenge is the presence of resonance structures in the measured cross-sections, at energies where TALYS operates in the statistical model regime and produces only a smooth mean curve. This results in residuals that are larger than the reported experimental uncertainties and, hence, if left unaccounted for, an underestimation of the uncertainty in the resulting TALYS model parameters. NEPU addresses this issue through a heteroscedastic Gaussian process (GP) that estimates an energy-dependent variation directly from the data distribution around the smooth model prediction [2, 3]. This variation is treated as an extra random uncertainty in the regression. Inconsistencies between datasets at the normalization level are handled separately through a marginal likelihood optimization procedure, which introduces additional systematic uncertainty components where the data require them.
In addition, several model parameters are allowed to become energy-dependent and are placed under GP priors with a physically motivated varying length scale. A further GP on the fit residual captures any remaining correlated model-data deviation and propagates it into the experimental covariance matrix.
In this work, the pipeline architecture, improved generalization and reproducibility, and results of the Zr-90 evaluations are presented.
References
[1] A. J. Koning and D. Rochman, Nucl. Data Sheets 113, 2841 (2012).
[2] G. Schnabel et al.,Nuc. Data Sheets 173, 239 (2021).
[3] A. Gook et al., EPJ Web Conf. 294, 04005 (2024).
| Session | Evaluation of Nuclear Data (theory, models, codes) |
|---|