In this talk we discuss information theoretic uncertainty quantification methods for probabilistic models in materials, catalysis and related design problems. We first demonstrate how new, tight and scalable information inequalities can provide computable uncertainty quantification indices for observables of interest; these indices account for model-form uncertainty in a neighborhood of a baseline model defined via information divergences. We apply these tools in three problems arising in catalysis and energy research: a) fast sensitivity screening of complex reaction networks with thousands of parameters; b) quantifying the impact of multiple sources of uncertainty in mesoscale bayesian reaction networks that include electronic structure data, and c) the design of power generation devices under model uncertainty.
Copyright. All Rights Reserved.