Graph Neural Network-Assisted Performance Regulation of Photocatalytic Nitrogen Reduction Reaction: An Insight from Machine Learning-Accelerated Atomistic Dynamics

To cross the formidable obstacle in the way of developing renewable energy using photocatalysis, exact control over the chemical reactivity of the nanomaterials, as well as the behavior of photogenerated charge carriers, we investigated real-time photocarrier dynamics and used graph neural networks (GNN) to accelerate screening of the nitrogen reduction reaction (NRR) mechanism on economical and ecofriendly 2D sulfur-defected gallium sulfide (2D V-GaS). Our density functional theory study revealed its thermal stability, optical properties, and favorable band alignment for the NRR. The best site over GaS sheets for the photocatalytic NRR was identified using GNN and MD data. Gibbs free energy calculations showed a downhill energy profile under a light-induced potential. A prolonged electron–hole recombination time of 4.06 ns indicates that photogenerated electrons have enough time to reach the active sites of the reaction. Therefore, our study suggests 2D V-GaS is a promising photocatalyst for sustainable and cost-effective NH3 production via NRR.