(521cl) Using Machine Learning to Model the Enantiospecific Decomposition of Tartaric Acid on Copper Surface Orientations

Enantiospecific heterogeneous catalysis is focused on understanding how the structure of a chiral catalytic surface favors one of the adsorbate enantiomers. Due to the chiral nature of the human enzymes, understanding these chiral interactions is important in the pharmaceutical industry. This work focuses on modeling the enantiospecific decomposition of Tartaric acid over a large range of copper surface orientations. The enantiospecificity is quantified by the difference in the half-times of the decomposition of the two Tartaric acid enantiomers. The data to build this model were collected using multiple Surface Structure Spread Single Crystals (S4C) that expose a continuous distribution of plane orientations across their surface. The goal is to use Machine Learning to gain more insights about the experimental observations and find the surface orientation with optimal enantiospecificity. The model built uses Generalized Coordination Numbers (GCN) descriptors which capture information about the coordination of the atoms on each surface orientation. A linear regression model was used to make predictions about surface orientations that were not tested experimentally, and it identified surfaces such as Cu (93,32,28) and Cu (81,59,47) to be highly enantiospecific. The model suggests regions near the (100)-(111) stepped symmetry plane, with short terraces, and atoms with a high density of atoms with GCN values near 7 contribute to highly enantioselective tartaric acid decomposition. This work offers insight into the origins of highly enantioselective decomposition kinetics as it relates to surface structure, and its prediction capability can be utilized to guide the search and discovery of highly enantioselective surface orientations more efficiently.