Abstract

Lithium carbonate (Li₂CO₃) is an important material for high-temperature molten carbonate fuel cells, electrochemical carbon capture, and lithium-based energy-storage technologies. However, understanding its molten-state structure and ionic transport remains challenging because of the limited time and length scales accessible to ab initio molecular dynamics (AIMD) and the accuracy limitations of classical force fields. Here, we employ equivariant machine-learned interatomic potentials (MLIPs) trained on AIMD data to investigate the structure and transport properties of molten Li₂CO₃ over extended time and length scales. Our MLIP-driven simulations agree with the experimentally measured static structure factor and shear viscosity and accurately describe the characteristic short- and medium-range structural correlations of molten-Li₂CO₃. Importantly, we find that Li⁺ diffusion is governed by strongly correlated and cooperative ionic motion rather than independent uncorrelated motion. At 1000 K, Li⁺ transport is anisotropic and highly concerted because the ions remain constrained within persistent oxygen-centered coordination environments. With increasing temperatures, the faster rearrangement of the local oxygen environments facilitates more isotropic and less correlated Li⁺ diffusion, leading to a distinct change in the transport mechanism by 1400 K. Our results establish a direct connection between the temperature-dependent dynamics of the local oxygen environment, which contributes to the local structure of Li₂CO₃ and the macroscopic Li⁺ transport. Our findings provide fundamental insights into ionic conduction in molten alkali carbonates and demonstrate the utility of equivariant MLIPs for investigating molten salts and ionic liquids relevant for energy applications.