Bulk and Point Defect Properties in α-Zr: Uncertainty Quantification on a Semi-Empirical Potential †
Résumé
Modelling studies of irradiation defects in α-Zr, such as point defects and their multiple clusters, often use semi-empirical potentials because of their higher computational efficiency as compared to ab initio approaches. Such potentials rely on a fixed number of parameters that need to be fitted to a reference dataset (ab initio and/or experimental), and their reliability is closely related to the uncertainty associated with their parameters, coming from both data inconsistency and model approximations. In this work, parametric uncertainties are quantified on a Second Moment Approximation (SMA) potential, focusing on bulk and point defect properties in α-Zr. A surrogate model, based on polynomial chaos expansion, is first built for properties of interest computed from atomistics, and simultaneously allows us to analytically compute the sensitivity indices of the observed properties to the potential parameters. This additional information is then used to select a limited number of material properties for the Bayesian inference. The posterior probability distributions of the parameters are estimated through two Markov Chain Monte Carlo (MCMC) sampling algorithms. The estimated posteriors of the model parameters are finally used to estimate materials properties (not used for the inference): in any case, most of the properties are closer to the reference ab initio and experimental data than those obtained from the original potential.
Origine : Fichiers éditeurs autorisés sur une archive ouverte
Licence : CC BY - Paternité
Licence : CC BY - Paternité