High-Throughput Screening of II-Type/Z-Type Photocatalytic g-GaN Heterojunctions Based on Machine Learning and Density Functional Theory for Photocatalytic Splitting of Water

The direct Z-type van der Waals (vdW) heterojunction is perceived as a promising photocatalyst, offering more efficient carrier migration processes compared with conventional II-type vdW heterojunctions. Nevertheless, the abundance of candidates and the expensive validation process pose hindrances to the further advancement of direct Z-type vdW heterojunctions. Leveraging a two-dimensional materials database, we amassed a set of approximately 100 samples of II/Z-type g-GaN-based vdW heterojunctions and innovatively proposed a fusion descriptor. Based on this, we developed a binary machine learning model and applied the first-principles nonadiabatic molecular dynamics for theoretical verification. The results indicate that the optimal model prediction accuracy of the fusion descriptor attains 95%, and the prediction outcomes are entirely in line with the theoretical calculation results. Meanwhile, based on the idea of high-throughput screening, we have constructed a complete screening framework for g-GaN-based vdW heterojunctions. This work provides assistance for the discovery and research of direct Z-type vdW heterojunctions with suitable band gaps for the photocatalytic splitting of water.