(532db) Classifying Zeolite Material Selection for O2 Sorption Pump Materials Using Machine Learning and a Density Functional Theory

Many chemical processes depend on having an environment that is low in oxygen partial pressure (P_O2 < 100 Pa); sorption pumps are a promising route to establishing that environment by either oxygen pumping or oxygen separation from an inert gas. Recently, a new class of materials has been theoretically designed for this process, selectively substituted zeolites with transition metals and metalloids. The substituted zeolites contain favorable binding energies promoting a tunable material for a wide range of system applications. In this work, we use ab initio calculations with machine learning (ML) to explore and expand the promising O₂ adsorbing zeolites as a class of materials for sorption-based oxygen pumping/separation. We investigate a wide range of zeolite structures (120 different IZA structures). When substituted with over 30 elements, we construct a database of ~130,000 different potentially O₂ adsorbing structures. We employ two separate ML classification networks for down selecting: 1) a go-no-go classification to determine if the zeolite will have exothermic chemisorption of O₂ and 2) a binned classification network partitioning across the range of binding energies, designed to classify a zeolite into small binding energy ranges. This binned classification aids in down selecting these structures into system specific ranges of desired binding energies. The go-no-go classification down selects the 130,000 zeolite structs to ~ 40,000 that will have exothermic chemisorption of O₂. From these 40,000 we classify them with the binned classification network and test, through ab initio calculations, 10 predicted zeolites promising O₂ adsorption across a range of binding energies. We find that the key characteristics are: 1) the substitutions must be able to adopt an oxidation state that is more positive than the cation it replaces, 2) the size of the pore into which the O₂ adsorbs to the wall must be large as prevent steric hindrances.