Heteroatom-tuned σ–π orbital reorganization enables efficient oxygen reduction and evolution of electrocatalysis beyond noble-metal catalysts

Abstract

Rigid coordination in single-atom catalysts (SACs) hampers the balanced adsorption–desorption of oxygen intermediates, constraining catalytic performance. Although conventional N-coordinated graphene-based systems exhibit strong ORR activity, their limited orbital flexibility suppresses OER kinetics and overall bifunctional efficiency. Rh-based SACs achieve exceptional bifunctional activity but remain impractical due to their high cost and scarcity, emphasizing the need for earth-abundant SACs with coordination-engineered transition-metal centers. Herein, density functional theory (DFT) calculations, combined with machine learning (ML), are employed to explore how second-shell heteroatom doping with boron (B) and phosphorus (P) can reconfigure orbital interactions in edge-anchored TM–N₄ graphene nanoribbons (EN₄G). Seventy-five SACs, comprising 26 transition metals (3d–5d) in pristine, B-doped, and P-doped environments, were systematically evaluated. B doping exerts only a minor electronic influence, resulting in marginal changes in both ORR and OER overpotentials and offering limited catalytic improvement over the pristine systems. In contrast, P doped Fe_EN4G, Zn_EN4G, and Cd_EN4G exhibit improved ORR activity but suffer from poor OER performance, reflecting their single-functional nature. Meanwhile, P–Co_EN4G and P–Rh_EN4G emerge as the most effective bifunctional catalysts with balanced overpotentials (η_ORR: 0.58 V and η_OER: 0.41 V and η_ORR: 0.65 V and η_OER: 0.30 V, respectively). Cumulative ICOHP analysis reveals that P doping weakens both TM–O and N–TM interactions (Fe, Co, and Rh), underscoring its role in tuning the local electronic environment. Electronic structure analysis reveals that P dopants reduce σ (d_z²–p_z) overlap that stabilizes TM–OH bonding and enhances ORR activity and the π* (d_xz and d_yz) orbital shifts below the Fermi level, limiting π*-backdonation for the OER. Furthermore, ML analysis of DFT-derived features identified d_xz orbital occupancy and Bader charge (q_TM) as dominant descriptors (≈80% cumulative importance). The random forest regression model (R² = 0.91) confirms that frontier-orbital occupation and charge redistribution dictate catalytic activity, offering a data-guided route for rational SAC design.