Lithium and sodium ion diffusion in LiFePO₄, LLZO, and NASICON: A molecular dynamics and machine learning study

Abstract

Next-generation lithium- and sodium-ion battery development relies on solid-state electrolytes, offering enhanced safety, thermal stability, and high energy density. This research uses molecular dynamics (MD) simulations and machine learning (ML) to study ion diffusion in LiFePO₄, Li₇La₃Zr₂O₁₂ (LLZO), and Na₃Zr₂Si₂PO₁₂ (NASICON). MD simulations calculated 300 K diffusion coefficients (D) of 9.18 × 10⁻¹¹ m²/s for LiFePO₄, 4.00 × 10⁻¹² m²/s for LLZO, and 6.77 × 10⁻¹¹ m²/s for NASICON, with activation energies of 0.34 eV, 0.35 eV, and 0.31 eV, aligning with experimental ranges, though validation is limited and less accurate for LLZO due to a 2-order magnitude deviation. The ML model, trained on OBELiX data with temperature augmentation, systematically underpredicts diffusion coefficients (e.g., 3.84 × 10⁻¹¹ m²/s for LiFePO₄ vs. 9.18 × 10⁻¹¹ m²/s MD), likely due to overestimated ion densities. Despite a high R² of 0.996, the model indicates opportunities for further refinement. Our comparative evaluation demonstrates that sodium ion movement through NASICON frameworks exhibits similar characteristics to lithium-ion mobility within both olivine and garnet crystal structures. Our research results expand the current understanding of ion mobility pathways and provide numerical reference points that can guide future material refinement approaches and data-driven computational design of advanced solid electrolyte battery technologies.