Uncertainty Reduction and Quantification in Computational Thermodynamics

Uncertainty quantification is an important part of materials science and serves a role not only in assessing the accuracy of a given model, but also in the rational reduction of uncertainty via new models and experiments. In this contribution, recent advances and challenges related to uncertainty quantification for Calphad-based thermodynamic and kinetic models are discussed, with particular focus on approaches using Markov chain Monte Carlo sampling methods. General differentiable and probabilistic programming frameworks are identified as an enabling and rapidly-maturing new technology for scalable uncertainty quantification in complex physics-based models, including those for Calphad. Special challenges for uncertainty reduction and Bayesian design-of-experiment for improving Calphad-based models are discussed in light of recent progress in related fields.