Prediction model of type and band gap for photocatalytic g-GaN-based van der Waals heterojunction of density functional theory and machine learning techniques

Applying suitable two-dimensional (2D) heterojunctions to photocatalytic water cracking reaction can obtain excellent catalytic performance. However, due to many candidate materials and complex interface effects, finding suitable heterojunctions combinations has become a challenge for the above applications. Based on about 1000 pieces of material data in the computational 2D material database, we adopted a simple energy band shift hypothesis to creatively build a machine learning prediction model of g-GaN based 2D Van der Waals (vdW) heterostructures' type with good performance, and carried out a first-principle calculation to verify the hypothesis. The results show that the band shift hypothesis is valid for the g-GaN based vdW heterojunctions combination without the participation of elements located in the first transitional period and this classification model with the area under curve (AUC) value of 0.93. In addition, we further built a regression prediction model for the band gap value of type II g-GaN based vdW heterojunctions in line with the band edge position of photocatalytic water-splitting reaction, with a mean absolute error (MAE) of 0.24 eV. This work establishes a machine learning screening process for g-GaN based vdW heterojunction applied in the field of photocatalysis, which greatly improves the research efficiency.