Quantifying uncertainty in equation-of-state and opacity models

Materials properties models – opacity, equation-of-state (EOS), and others – underpin our understanding of condensed matter, high-energy-density and laboratory astrophysics experiments due to their critical role in hydrodynamics simulations. These models are built from a complex combination of first-principles simulations, physics-based approximations, and experimental data, making it difficult to quantify and represent uncertainty. Even state-of-the-art materials properties tables rarely provide uncertainty information, and even when available they usual neglect potentially important contributions like model form error, physical constraints, and systematics. How should the next generation of tabular models represent uncertainty? How can they provide a complete accounting of all uncertainty sources from all sources of information? How can those uncertainties be used to improve first-principles simulations and hydrocode predictions?



In this talk we will describe ongoing work to develop an uncertainty quantification framework for material properties models. Our approach aims to combine first-principles calculations, experiments, and hydrocode simulations in a unified framework to learn uncertainties from data and propagate them into tabular models and hydrocode outputs. We will discuss the key components of our framework, including physics-informed machine learning for EOS, uncertainty-aware tabular model formats, and uncertainty-enabled experimental design and interpretation. Finally, we will demonstrate the application of these components to new laser-driven shock experiments at the Omega laser facility.