Unlocking Lithium Superionic Conduction via Phonon Softness Descriptors: A High-Throughput Machine Learning Paradigm

Lithium superionic conductors (LISICONs) are pivotal for next-generation solid-state batteries, yet rational design remains challenged by poorly understood correlations between lattice dynamics (phonons) and ion transport. We unveil phonon softness as the unifying descriptor for rapid Li + conduction through a high-throughput computational framework combining ab initio calculations, machine learning potentials (fine-tuned MACE MPA model), and molecular dynamics across 1304 dynamically stable Li materials screened from an initial pool of 8578 candidates. We demonstrate that low-frequency phonons dominate Li + migration, with the Li + vibrational density of states (VDOS) center and Debye frequency exhibiting strong inverse correlations with diffusion coefficients. Phonon mode-resolved analysis reveals that delocalized, collective vibrations (<2 THz) lower activation barriers by promoting in-phase host lattice coupling. Crucially, global mechanical properties─bulk modulus, shear modulus, elastic modulus, and hardness─all inversely scale with Li + mobility, confirming that structural compliance facilitates reduced activation barriers for Li + hopping, consistent with prior observations that lattice softness enhances ionic migration. Using ab initio molecular dynamics simulations, we confirm superionic conduction in 25 Li structures that are screened from our workflow, including halide-based and nitrogen-containing materials. Our study establishes readily accessible lattice dynamics descriptors (VDOS center, Debye frequency) and mechanical metrics as universal screening tools, accelerating the discovery of high-performance solid electrolytes for energy-dense, safe batteries.