Predicting MXenes hybrid functional bandgaps for photocatalytic applications using deep neural networks

The accelerated discovery of efficient photocatalytic materials is critical for advancing sustainable hydrogen production technologies. Two-dimensional transition metal carbides and nitrides, known as MXenes, have emerged as promising candidates owing to their tunable band structures and diverse surface terminations. In present study, an artificial neural network (ANN) model was developed to predict hybrid functional bandgap ($(E_g^{PBE0}$) of 3679 MXene compounds, encompassing both Ti-based and non-Ti-based variants. The dataset integrates elemental, structural, and electronic features, along with engineered descriptors such as electronegativity differences (M–X and T–X), mean atomic mass, and total valence. The model was trained using a feed forward architecture with three hidden layers and optimized using the Adam optimizer with early stopping. Performance was assessed using standard regression metrics, yielding high accuracy with R² = 0.9884, MAE = 0.0217 eV, and RMSE = 0.0602 eV. To corroborate trends, Ti₃C₂T_x was synthesized and characterized (XRD/XPS/SEM/TEM/AFM/UV-Vis). This work highlights the effectiveness of using deep learning models trained on diverse MXene datasets for bandgap prediction and provides a reliable foundation for future integration with experimental validation and HER-specific screening.