First-Principles Surface Spectroscopy at the Air/Water and Graphene/Water interfaces

The water-graphene and water-air interfaces are among the most representative liquid-solid and liquid-gas interfaces, respectively. Experimental Sum-Frequency Generation (SFG) spectroscopy of these interfaces presents challenges in directly capturing the microscopic dynamics occurring at the surface. Computational simulations can complement these studies by providing insights at the atomic scale. However, accurately calculating SFG spectra requires fully converging both the dipole moment and the polarizability of the system, which needs nanosecond simulations and is difficult to achieve with conventional ab initio methods. By employing deep learning techniques, we demonstrate that machine learning models trained on datasets containing not only potential energy surfaces but also dipole moments and polarizabilities can potentially overcome these challenges. The training datasets are generated using electronic structure calculations with the SCAN functional under the influence of an external electric field, ensuring quantum mechanical accuracy. Through comparison with recent experimental signals, our findings uncover both microscopic similarities and nuanced differences between air/water and graphene/water interface.