This paper develops a deep learning-based interatomic potential for the Li₇La₃Zr₂O₁₂ (LLZO) system using a diverse dataset from databases and first-principles simulations. A novel convergence criterion based on principal component analysis (PCA) coverage of training and test sets is proposed. The resulting potential accurately describes LLZO's structural and dynamical properties, including phase transitions, at significantly reduced computational cost compared to DFT. This efficient training strategy offers a promising simulation tool for accelerating solid-state battery design.
