(159f) Active Learning Exploration of Single Atom Active Sites for Oxygen Reduction and Evolution Reactions

Rational design of a single atom catalyst for target catalytic reaction is of great interest. Here, we present a high-throughput screening of local environments of M-N-C (M: 3d transition metals, N: boron, carbon, nitrogen, oxygen, and sulfur, and C: carbon) for oxygen reduction and evolution reaction. We explore the active metal sites with different ligand field interactions inducing the charge and spin states change. We apply the active learning for efficient data generation and equivariant graph convolutional network for surrogate model construction. Our results indicate that the local atomic environments of a single atom catalyst and the spin states can effectively tune the adsorption strength of reaction intermediates. Site-property of active sites electronic structure parameters i.e., atomic charge, coordinated ligands, spin state, d-band center energy, etc. Our strategy suggests an effective guidance of the atomic level engineering to activate the center metal for the highly active nonprecious metal single atom catalysts.