(169bw) Proton Transport on Graphamine: A Deep-Learning Potential Study

The performance of hydrogen fuel cells relies critically on the conduction of protons. Conventional proton exchange membranes employ materials, such as Nafion, that only conduct protons when properly hydrated. If the relative humidity is either too low or too high, the fuel cell will cease to operate. This limitation highlights the need to develop new materials that can rapidly conduct protons without requiring hydration. We have designed a new material in silico that will conduct protons anhydrously. We call this material graphamine, as it is composed of graphane functionalized with amine groups. We use a combination of density functional theory (DFT) and machine learning to evaluate the performance of graphamine because utilizing DFT alone to assess new membrane materials is impractical due to restrictions in length and time scales. We have constructed a deep learning framework tailored for modeling graphamine, enabling us to fully characterize and evaluate proton conduction within this material. The trained deep-learning potentials (DPs) are computationally economical and have near-DFT accuracy. These DPs were trained with DFT data and were improvised using an active learning approach. We then use these DPs to calculate proton self-diffusion coefficients as a function of temperature and evaluate the overall barrier of proton diffusion. The DPs also allow us to carry out large-scale and longtime simulations necessary to identify the overall proton transport mechanisms.