Modeling Proton Transport in Platinum-based Fuel Cells using Machine-Learning Interatomic Potentials

Accurately describing proton transport has been a long-standing problem in understanding the performance of fuel cells. Quantum simulations are needed to represent the hopping of protons, but are unable to simulate the system sizes required for the study of transport in polymers found in fuel cells. In the present study, we use the MACE machine-learned interatomic potential framework to simulate fuel cell systems containing platinum, Nafion, water and hydronium, predicting molecular geometry, system structure and proton diffusion. We find excellent accuracy in predicting intramolecular geometry, such as bond distances and bond angles, as well overall system structure characterized by radial distribution functions. Finally, our potential captures the Grotthuss proton hopping mechanism and shows the expected trend of increasing proton diffusion with increasing system water content.