(544ds) Using Data Science to Reduce Large Reaction Networks in Catalysis
Catalysts are active surfaces that accelerate chemical reactions. Catalytic processes produce fertilizer and fuels but consume 2% of the global energy supply and limit the efficiency of energy technologies like fuel cells. Traditionally, Density Functional Theory (DFT) provides theoretical guidance for experiments and new catalysts. However, due to the complexity of reaction system pathways, expensive quantum simulations makes the catalysts discovery slow. Machine learning techniques can be used to greatly accelerate material selections by exploiting similarities among surface intermediates, creating surrogate models to predict reaction kinetics and pathways, and generating new surface structures automatically. In this work, we demonstrate the prediction of several intermediates (*C, *H, *O, *CO, *CHO) across a wide range of intermetallics. We show that this is possible by connecting site features with each intermediate adsorption energy or by exploiting linear scaling relations between intermediates. This data generation is the first step creating predictive models and building an active learning workflow for mechanism determination.