Closing the Loop for Oxide Catalyst Design

Multicomponent oxide ceramics hold great potential for applications in catalysis and electronics. By carefully doping these materials with various elements, their electronic and transport properties can be optimized for high-performance uses. This creates a vast and complex design space for selecting the best elements and their ratios to achieve desired properties.

In this study, we discuss how machine learning models, enhanced by electronic structure simulations, can predict properties and aid in the design of multicomponent oxides. Specifically, we focus on perovskite oxides, which can accommodate quaternary or more complex compositions, and how they can be engineered computationally.

We will cover the use of elemental and electronic descriptors to forecast chemical ordering in synthesized perovskite oxides, the application of deep learning models to predict atomic-level properties like magnetic moments and catalytic activity, the use of machine learning interatomic potentials to relax perovskite structures and generate surface phase diagrams, and the capability of equivariant models to capture symmetry-breaking relaxations and properties from idealized, unrelaxed prototypes