(621du) Machine Learning Approaches to Design Catalysts for C1 Chemistry
Simultaneous carbon dioxide and methane conversion to chemicals and fuels shows great promise in mitigating the environmental impact of greenhouse gas emissions, but requires the activation of the molecules from their energetically stable ground states and the selective cleavage of thermodynamically stable bonds. We have most recently studied carbon dioxide activation and hydrogenation on a reduced ceria catalyst. Density functional theory calculations and micro kinetic modeling, with reaction rate coefficients calculated using transition state theory and a modified Brønsted-Evans-Polanyi relation, have been employed to calculate macroscopic reactor observables, including turnover frequencies, as a function of temperature, pressure, and feed ratio.
We now turn our attention to finding new catalysts, composed of inexpensive and/or earth-abundant elements, that possess superior performance in either carbon dioxide reduction or selective methane oxidation. "Big data" approaches are promising for inverse design of new materials with desired properties. We first present our efforts in automating computational catalysis, using Bayesian optimization to find adsorbate binding sites on surfaces, and then show how supervised machine learning may be used to identify new materials with strong redox behavior.