In the last two decades, millions of materials properties were curated from density functional theory and machine learning calculations, greatly accelerating catalyst design using computation. Despite these efforts in elucidating structure-property relationships, designing materials synthesis to obtain a structure of interest remains a challenge. This is particularly important in the case of catalysts, where the connection between atomistic models and catalytic performance is difficult to probe in experimental settings. In this talk, I will describe how combined advances in high-throughput simulation, machine learning, and literature extraction enable designing and accelerating catalyst synthesis. Using examples from two classes of heterogeneous catalysts â nanoporous materials and multi-metallic alloys â I will show how data-driven methods can guide the synthesis of materials with increasingly controlled active site distributions and stability. Finally, I will show how incorporating chemical theory in machine learning increases the robustness of models and enable increasingly data-efficient simulations for catalysts. These computational tools open numerous opportunities for accelerating catalyst discovery beyond screening.
Prepared by LLNL under Contract DE-AC52-07NA27344.
