Band-Edge Prediction of 2D Covalent Organic Frameworks from Molecular Precursor via Machine Learning

The band-edge positions of two-dimensional (2D) covalent organic frameworks (COFs) play a crucial role in their applications in photocatalysts and nanoelectronics. However, massive amounts of 2D COFs with targeted band-edge positions from high-level first-principles calculations based on their composition are time-consuming due to the diversity and complexity of unit cell structures. Here, we report a strategy to predict the band-edge positions of 2D COFs by combining first-principles calculations with machine learning (ML). The root-mean-square error (RMSE) of the predicted valence band maximum (VBM) and conduction band minimum (CBM) between ML prediction and first-principles calculated values at the Perdew-Burke-Ernzerhof (PBE) level are 0.229 and 0.247 eV in test data set, respectively. In addition, a linear relationship is established between the PBE results and the HSE06 results with RMSE values of 0.089 and 0.042 eV for VBMs and CBMs in the test data set. Finally, a workflow is developed to determine the band-edge positions of the 2D COFs.