(607f) Active Learning of Multimetalic Single Atoms Catalysts

Single-atom catalysts (SACs) exhibit high activity for a wide range of sluggish reactions and allow performance tunability at atomic-level through the selection of central metals, ligand environments, and secondary metal sites. However, the design space with varying structures and compositions significantly hinders the fast and accurate identification of desired multimetallic SACs. In this work, we demonstrate a self-driving computational strategy for exploring binary metallic sites with varying combinations of 3d transition metals and different ligand environments, resulting in over 30,000 single atom sites for the electrochemical catalysis of the oxygen reduction and evolution reactions (ORR/OER). This approach is based on the density functional theory (DFT) calculations of binding energies and atomic descriptors as target properties and utilizes an equivariant graph neural network (GNN) as a surrogate model for predicting DFT labels directly from the atomic structure. The chemical environments learned by the GNN lead to capturing composition-structure-property relationships for the ORR/OER activity and selectivity. Active learning facilitates the investigation of the search space by balancing the exploration of unseen atomic structures with the exploitation of the active ones. The computational predictions of promising Co-Fe, Co-Co, and Co-Zn metal pairs are consistent with the state-of-the-art results of experimental measurements reported in the literature. The GNN-based analysis of multiple active sites on a single surface for target catalytic reaction can be extended to a broader class of multi-element high entropic materials systems.