Machine learning-guided discovery of thermodynamically stable single-atom catalysts on functionalized MXenes for enhanced oxygen reduction and evolution reactions

Abstract

The transition to sustainable energy systems necessitates the development of efficient, cost-effective, and stable electrocatalysts, particularly for the oxygen reduction reaction (ORR) and oxygen evolution reaction (OER). In this study, we investigate transition-metal-doped MXene surfaces (Ti₃C₂T₂, T = O, S, F, Cl, Se) as single-atom catalysts (SACs) for ORR applications, employing a combination of density functional theory (DFT) and machine learning (ML) approaches. A comprehensive dataset encompassing binding, cohesive, and formation energies was generated through DFT calculations and used to train various ML models. Among them, convolutional neural networks (CNNs) achieved the highest prediction accuracy, exhibiting the lowest RMSE and an R² value exceeding 0.96 on the testing data. SHAP analysis revealed that catalyst surface properties predominantly influence adsorption behavior. Thermodynamic screening identified multiple stable SAC configurations, notably Ni–Ti₃C₂S₂ and Cu–Ti₃C₂S₂, which demonstrated low overpotentials and favorable ORR/OER performance. Specifically, Ni–Ti₃C₂S₂ showed low overpotentials of 0.31 eV for the oxygen reduction reaction (ORR) and 0.40 eV for the oxygen evolution reaction (OER), while Cu–Ti₃C₂S₂ displayed overpotentials of 0.35 eV for the ORR and 0.74 eV for the OER. The study further establishes linear scaling relationships among adsorption energies of reaction intermediates, providing insights for rational catalyst design. These findings highlight the potential of ML-accelerated materials discovery to guide the development of next-generation bifunctional electrocatalysts.