Machine-learning-accelerated discovery of single-atom catalysts based on bidirectional activation mechanism

Abstract

Single-atom catalysts (SACs) have provided new impetus to the field of catalysis because of their high activity, high selectivity, and theoretically full utilization of active atoms. However, the ambiguous activation mechanism prevents a clear understanding of the structure-activity relationship and results in a great challenge of rational design of SACs. Herein, by combining density functional theory (DFT) calculations with machine learning (ML), we explore 126 SACs to analyze and develop the structure-activity relationship for the electrocatalytic nitrogen reduction reaction (NRR). We first propose a bidirectional activation mechanism with a new descriptor for catalytic activity, which provides new insights for the rational design of SACs. More importantly, we establish a ML model for predicting the catalytic performance of NRR, validated by both DFT calculations and experimental works. The successful ML prediction in this work helps with the accelerated design and discovery of new catalysts by computational screening with high practical significance.