An Integrated DFT–Machine Learning Framework for Screening Solid-State Electrolytes

The discovery of solid-state electrolytes (SSEs) is often hindered by time-consuming experimental methods that limit the systematic search of potential materials. This paper presents an integrated computational strategy that combines density functional theory (DFT) screening with machine learning (ML) validation to accelerate the assessment of potential materials for electric vehicle use. The strategy overcomes fundamental computational hurdles, such as the limitations of DFT calculations, validation, and the prediction of interfacial properties. Based on a set of 1,247 SSE materials, the model identified 12 potential candidates with predicted ionic conductivities above 1 mS cm−1. Of these, five were validated against experimental data with an average absolute error of 0.52 log units, while the remaining seven are new predictions. Promising candidates include the validated material Li7La3Zr1.8Ta0.2O12 (estimated: 3.8 mS cm−1, experimental: 3.2 mS cm−1) and the predicted material Li10.2GeP1.9S12 (estimated: 9.4 mS cm−1). The ML validation step resulted in cross-validated classification accuracies of 87.3% for conductivity and 84.1% for stability. The paper clearly points out the limitations of the approach, including the dependence on DFT functionals and approximations in interfacial impedance. While this approach reduces the screening time from years to months, experimental verification is still required for operational use