Machine learning guided O₂ reduction to H₂O₂ on BN co-doped γ-graphyne: insights from time-domain atomistic dynamics

Visible-light-driven production of hydrogen peroxide (H₂O₂) offers a promising and sustainable pathway. Metal-free carbon-based materials are attractive photocatalysts for the two-electron oxygen reduction reaction (2e^- ORR), but their performance is often limited by rapid electron-hole recombination and sluggish kinetics. To overcome these limitations, we have performed combined density functional theory (DFT) and nonadiabatic molecular dynamics (NAMD) simulations to investigate the photocatalytic conversion of O₂ to H₂O₂ on a BN co-doped γ-graphyne monolayer. Upon BN co-doping, the band gap of pristine γ-graphyne increases from 0.96 to 1.10 eV, while retaining favorable band edge positions and strong UV-Vis absorption. We herein employed pretrained Universal Model for Atoms (UMA) to identify the most favorable adsorption site on the BN co-doped γ-graphyne monolayer. Gibbs free energy calculations confirm the thermodynamic feasibility of the reaction under light irradiation. Importantly, the NAMD results reveal a significantly prolonged electron-hole recombination time (10.75 ns) as compared to the pristine system (4.17 ns), providing sufficient time for photogenerated electrons to effectively participate in the reduction of O₂ to H₂O₂. Overall, our results suggest that BN co-doped γ-graphyne can serve as a promising and efficient metal-free photocatalyst for converting O₂ to H₂O₂ under UV-vis light.