AI-Driven Discovery of Lithium and Sodium Superionic Conductors via Lattice Dynamics Descriptors

Speaker: Mr. Ogheneyoma Aghoghobia
Affiliation: University of South Carolina
Date: October 23, 2026

Abstract:

Discovering solid-state superionic conductors across vast chemical spaces remains a major  challenge for next-generation all-solid-state batteries. Herein, we propose an AI-driven approach  that uses lattice-dynamics descriptors to accelerate the discovery of Li- and Na-superionic  conductors. Fine-tuned machine-learned interatomic potentials, benchmarked against DFT,  provide reliable second- and third-order force constants, phonon dispersions, and vibrational  densities of states (VDOS). From these, we derive physically meaningful descriptors: phonon  mean-squared displacement (MSD), acoustic cutoff frequency, Debye frequency, and VDOS  center. Across 921 Na and 1,304 Li structures, high phonon MSD and low characteristic  frequencies strongly correlate with fast ion diffusion from MD simulations, reflecting soft, flexible  host frameworks. Mode resolved analysis shows that a few delocalized, low-frequency phonons  coupled in phase with the host lattice dominate ion migration. Using these descriptor-driven  screening criteria, we identify Na conductors with activation energies of 133–198 meV and 25  new Li superionic conductors confirmed by ab initio MD. The approach is far more efficient than ab initio MD while maintaining near-DFT accuracy.