Composition-Dependent Li-Ion Transport in Mixed-Metal ZIF-62 Glasses Revealed by Machine-Learning Molecular Dynamics.

Understanding ion transport in metal-organic frameworks requires resolving the interplay between framework dynamics, local disorder, and thermally activated hopping on extended time and length scales. Here, we develop a robust deep neural network (DNN) interatomic potential, trained and validated against density functional theory data using the DP-GEN framework, to investigate Li-ion transport in pure and mixed-metal ZIF-62 across a wide range of Co/(Co + Zn) ratios. Large-scale deep potential molecular dynamics simulations at 450 K reveal that Li-ion motion is governed by localized vibrational dynamics within metastable cages, punctuated by activated hopping events through the interconnected pore network. Van Hove correlation analysis reveals that intermediate metal substitution [Co/(Co + Zn) ≈ 0.25-0.50] optimizes the spatial extent of Li displacements while maintaining a stable local coordination environment. Jump-resolved analysis further demonstrates that the 0.50-ZIF-62 composition exhibits the lowest effective activation energy (∼0.093 eV) and the highest ionic conductivity (∼6.0 × 10-3 S cm-1), arising from a dynamically evolving, composition-dependent energy landscape. Correlated-motion analysis indicates that long-range diffusion proceeds predominantly via independent hopping rather than sustained collective migration. Together, these results establish a unified and dynamic picture of Li-ion transport in mixed-metal MOF electrolytes, demonstrating the predictive capability of machine-learning interatomic potentials for the rational design of high-conductivity solid-state ion conductors.