Deep-learning potentials for proton transport in double-sided graphanol

There is a need to develop new materials for proton exchange membranes that can operate at higher temperatures and low humidities. Designing and evaluating novel membrane materials using density functional theory (DFT) is infeasible because of length and time scale limitations. We developed a deep-learning potential (DP) to evaluate double-sided graphanol (DSG), which is a potential anhydrous proton conducting material. We trained our DP on DFT data using an active learning approach. Our DP is computationally efficient and has near-DFT accuracy. We analyzed DSG by computing phonon properties, thermal fluctuations, and self-diffusivity using our DP. Our results for DSG are compared with single-sided graphanol (SSG). We observed lower thermal fluctuations and similar proton self-diffusivity at 800 K in DSG compared to SSG. Our DP simulations show that structural differences in DSG compared to SSG do not impact proton transport. DSG is a promising membrane material deserving of synthetic efforts.