Proton dynamics simulations of solid-acid electrolytes using active learning and equivariant neural network force fields.

Understanding the rate-limiting steps of proton conduction across various solid acid electrolytes and the mechanisms behind superprotonic phase transition is crucial for designing next generation energy fuel cells. Due to the high computational cost of ab-initio molecular dynamics, previous studies of proton dynamics were constrained to a hundred atoms within a few hundred picoseconds and thus led to a limited statistics of proton-hopping events. Previous work has used material-specific empirical force fields to study the phase transition of CsHSO4, but it cannot be generalized to understand the mechanism of other solid acid materials.

In this work, we develop machine learning interatomic force fields for CsH2PO4 and CsHSO4 superprotonic conductors combining ab-initio accuracy with scalability to large system sizes and nanosecond time scale. First, we use Bayesian active learning framework FLARE[1] to generate training data containing various atomic configurations. Then, we train an equivariant neural network potential Allegro[2] and deploy it in large-scale molecular dynamics. After demonstrating the fidelity of the machine learning potential by comparing with experimental diffusivities and activation energies, we study the dynamics of oxyanions and protons to identify the rate-limiting step for conduction.

[1] arXiv:2203.03824

[2] arXiv:2204.05249