Ionic Transport Predicting: A Machine Learning Dive into K ₃ HCh (Ch: S, Se, and Te) for Next-Generation Solid State Batteries

In this study, we investigate the ionic transport behavior of potassium-based antiperovskites (K3HCh, Ch: S, Se, and Te) through a comprehensive machine learning framework. The workflow included 809 Inorganic electrolyte compounds, feature generation from elemental properties, selection of the top 50 features, and training and testing of several predictive models, and the XGBoost model achieves a mean absolute error (MAE) of approximately 0.45. To validate the model, we applied it to lithium-based antiperovskites (Li3HCh, Ch = S, Se, Te) at both high and low temperatures. The predicted ionic conductivities via the XGBoost model showed minimum errors of 7.34%, 2.65%, and 1.58% for Li3HS, Li3HSe, and Li3HTe, respectively, compared to experimental values. For potassium analogues, the predicted ionic conductivities at 100 °C were 7.64 × 10 –8, 3.5 × 10–8, and 1.9 × 10–7 S·cm–1 for K3HS, K3HSe, and K3HTe, respectively. In addition, Nudged elastic band (NEB) calculations confirm that K3HTe possesses a low activation energy of 0.38 eV, highlighting its potential for efficient ion transport.