Computational Investigation of the Effect of Nitrogen Dopants and Oxygen Vacancies on the Energetics of Lithium Lanthanum Titanates

Doping is a powerful strategy for improving the ionic conductivity of ceramic-type solid-state ion conductors. Compared to cation doping, anion doping is much less studied but has been shown to improve Li-ion transport in perovskite lithium lanthanum titanates (LLTOs). In this work, the structure and energetics of nitrogen-doped LLTOs were studied using first-principles density-functional theory calculations. The calculations found high energy cost for nitrogen doping, which decreases with the introduction of oxygen vacancies or with the formation of nitrogen-nitrogen and nitrogen-oxygen dimers. Dimer formation reflects potentially significant structural distortions. Six machine-learning models, including four descriptor-based models (multiple linear regression, random forest, support-vector machine, and XGBoost) and two graph-based neural network models (CGCNN and MEGNet), were evaluated for predicting the energy of both unoptimized and DFT-optimized LLTO structures. XGBoost and MEGNet were found to be the best-performing models from the two categories, both exhibiting correlation coefficients larger than 0.99. SHAP analysis shows that oxygen vacancies prefer to form near La3+ ions, while close proximity of Li+ and vacancies have a destabilizing effect. The latter suggests that thermodynamically Li+ may be repelled from the oxygen-vacancy centers and thus unable to directly benefit from the potential advantages of vacancies for Li+ ion transport.