AI-driven study of carbon-dioxide activation on semiconductor oxides.

We have developed a strategy for a rational design of catalytic materials using subgroup discovery (SGD) – an artificial-intelligence method that identifies statistically exceptional subgroups in a dataset. With that, we identify features of catalyst materials (“catalysts’ genes”) that correlate with mechanisms promoting or hindering the activation of carbon dioxide (CO₂), towards a chemical conversion of CO₂ to fuels or other useful chemicals. Our training set consists of high-throughput first-principles calculations of CO₂ adsorption on the surfaces of a broad family of oxides. We demonstrate that the decrease of OCO-angle, previously proposed as the indicator of activation, is insufficient to account for the good catalytic performance of experimentally characterized oxides. Instead, SGD analysis shows that these surfaces consistently exhibit combinations of “genes” resulting in a strong elongation of a C-O bond due to binding of one O atom in CO₂ molecule to a surface cation. The same combinations of “genes” also minimize the OCO-angle, but under the constraint that the Sabatier principle is satisfied. Based on these findings, we propose a set of new promising oxide-based catalyst materials for CO₂ conversion, and a recipe to find more. – A. Mazheika et.al. ArXiv:1912.06515.