Origin of the Enhanced Lithium-Ion Conductivity in Lithium-Site Supervalent-Doped Li7La3Zr2O12 Revealed by Machine Learning Molecular Dynamics Simulations

Abstract

Garnet-type Li7La3Zr2O12 (LLZO) is a promising solid electrolyte, but its highly conductive cubic phase is metastable at room temperature. We investigate lithium-ion migration mechanisms in LLZO using machine-learning molecular dynamics (MLMD), which achieves near-AIMD accuracy while enabling nanosecond-scale simulations at significantly reduced computational cost. Li-site supervalent doping (Fe, Ga, Al, B, and Zn) stabilizes the cubic phase by reducing the energy difference between tetragonal and cubic phases. In tetragonal LLZO, supervalent doping enhances Li-ion transport through vacancy-mediated hopping. In contrast, cubic LLZO possesses a disordered Li sublattice, making transport relatively insensitive to doping and susceptible to migration pathway disruption. Overall, Li-site supervalent doping improves conductivity by stabilizing the cubic phase and activating transport in tetragonal LLZO, rather than increasing intrinsic mobility in cubic LLZO. These results highlight MLMD as a powerful tool for probing ion transport and provide optimal doping strategies to introduce Li vacancies while preserving well-connected migration pathways.