Machine-learning-assisted high-throughput screening for accelerating the discovery of diatomic catalysts in N2 electroreduction

Abstract

The nitrogen reduction reaction (NRR) under ambient conditions provides a sustainable pathway for green ammonia synthesis. Bimetallic catalysts supported on nitrogen-doped, graphene-like C₂N frameworks with dual-metal active sites, denoted as MAMB–C₂N, have garnered significant research interest owing to their remarkable catalytic activity and structural stability. However, conventional experimental techniques and density functional theory (DFT) calculations struggle to systematically evaluate their catalytic activity across all possible coordination environments. This study integrates DFT with six distinct machine learning (ML) methods, delivering accurate and efficient Gibbs free energy predictions for all three reaction pathways – distal, alternating, and enzymatic. Twenty-seven transition metal dimers were considered, and comprehensive evaluation of catalytic activity was conducted across 729 MAMB-C2N configurations. The identification of 33 highly active catalysts was achieved through a four-stage screening approach, followed by construction of structure–activity relationships between descriptors and ΔG using the SISSO algorithm. This study establishes an important framework for designing sophisticated reaction processes and systematically screening highly active bimetallic NRR electrocatalysts.