Machine Learning Prediction of Ionic Conductivity and Binding Energy in Doped Li₂ZrCl₆ Halides for Solid-State...

All-solid-state batteries are attracting significant attention, particularly the development of solid electrolytes. Among potential candidates, ternary halide compounds have emerged as promising materials for fast ion conduction. In this study, we integrate supervised machine learning (ML) models with density functional theory (DFT) calculations to accelerate the discovery of fast-ion-conducting halides. A database of halide compounds was constructed from the literature and merged with the existing Liverpool inorganic solid-state electrolyte database to predict ionic conductivity near practical operating temperatures. In addition, a database of over 12,000 DFT calculations was employed for the prediction of the binding energy. The model demonstrated excellent predictive performance for both properties (MAE = 0.6012, R² = 0.8062 for conductivity; MAE = 0.0314, R² = 0.9957 for binding energy). Using the trained model with interpretable elemental features, including Element Fraction, Element Property descriptors, and the number of atoms, we explored divalent and trivalent substitutions in the Li_2+2×xZr_1-xMe²⁺_xCl₆ and Li_2+xZr_1-xMe³⁺_xCl₆ systems (where Me = Mg, Ca, Mn, Zn, Al, Sc, Ga, In, Y, etc.). Finally, dopant elements such as Ti³⁺and Sn²⁺ are proposed to enhance the ionic conductivity and structural stability of the halide family. The results highlight the potential of machine learning to efficiently identify new compositions for solid-state battery electrolytes, illustrating the importance of efficient machine-learning approaches using easily obtainable descriptors to speed up the development of new solid-state electrolytes.