Voltage-Dependent Dynamics and Surface-Specific IR Spectra of Water at Gold Electrodes from Deep Neural Network-Assisted Ab Initio Calculations

Electrode-liquid interfaces under voltage biases demonstrate unique properties that govern their many energy conversion applications. Mean-field models and classical simulations of the electrical double layer (EDL) cannot describe molecular-level polarization or reactivity, necessitating controlled-potential density functional theory calculations to investigate such systems. Born-Oppenheimer ab initio molecular dynamics (AIMD) calculations of electrified Au(111) slabs interfaced with multiple monolayers of liquid water were performed using a hybrid explicit-implicit solvent approach. Fixed excess charges were localized on the Au slab and electrode potentials were computed on the fly to determine the system’s differential capacitance. The effects of modest positive and negative voltages on the structure, dynamics, and hydrogen bonding properties of interfacial water were elucidated and compared to available experiments. Additionally, the molecular dipoles of the water subsystem were computed at thousands of AIMD snapshots and further interpolated at each voltage using deep neural networks (DNNs). From these dipoles, the IR spectra specific to the interfacial water molecules versus applied voltage were obtained in good agreement with surface-enhanced IR absorption experiments. This work provides a framework for investigating species in the EDL and demonstrates the utility of machine learning for interrogating surface-specific vibrational spectroscopies of chemical species.