Machine learning accelerated nitrogen electrofixation on dual-atom catalysts

Abstract

Atomically dispersed catalysts offer the merits of theoretically maximum atomic utilization and adjustable electronic properties. While the current research focuses on single-atom catalysts, new opportunities would be opened by constructing dual-atom catalysts via adding a second single-atom site. This strategy can enhance their applications in many catalytic reactions. Nevertheless, the design of dual-atom catalysts can be influenced by many factors (e.g., elemental compositions and atomic positions), which exponentially increase the complexity of catalyst optimization. As exemplified by nitrogen fixation, here we employ machine learning (ML) to accelerate the screening of potential dual-atom catalysts. Through ML-driven predictions, we identified a CrNi/MoSe₂ catalytic system with an ultralow limiting potential of −0.45 eV. Furthermore, density functional theory (DFT) computations were conducted to unravel the underlying mechanisms. Moreover, experiments have verified that the synthesized CrNi/MoSe₂ has high electrochemical nitrogen reduction reaction performance. Our findings would provide a robust framework for exploring new catalysts in energy conversion systems, paving the way for future catalyst design.