Identification of Cu-N₂ Sites for Zn-Air Batteries in Harsh Electrolytes: Computer Virtual Screening, Machine Learning, and Practical Application

Nonprecious-metal catalysts possess great potential to replace noble metals in catalysis; however, selecting efficient candidates through experiments is time-consuming and costly. Herein, we employed a data-driven virtual screening (VS) method to discover new electrocatalysts. Specifically, we identified Cu-N₂ sites for Zn-air batteries (ZABs) in harsh electrolytes by combining VS and machine learning (ML) techniques. A thermodynamically stable and highly active Cu-N₂ Lewis acid site was pinpointed using molecular dynamics (MD) simulations and density functional theory (DFT) calculations: MD simulations filtered stable structures, while DFT calculations excluded those with high overpotentials by evaluating adsorption energies of oxygen intermediates using the “Volcano plots” theory. The ML model, trained in atomic types and geometries, predicted catalysts with DFT-level accuracy. The as-predicted Cu-N₂ Lewis acid site was experimentally synthesized in a hollow nitrogen-doped octahedron carbon framework (Cu-N₂@HNOC), with a high Cu loading of 13.1 wt %. A ZAB with Cu-N₂@HNOC as the cathode catalyst showed prolonged cycling stability and a high maximum power density of 78.1 mW/cm². Our strategy is applicable in the quest of valuable catalysts across a wide range of applications. With the accumulation and experimental validation of datasets to improve quality, this approach is expected to accurately predict promising electrocatalysts by integrating deep ML.