Machine Learning Prediction Models for Solid Electrolytes Based on Lattice Dynamics Properties.

Area of Science:

• Materials Science
• Computational Chemistry
• Solid-State Physics

Background:

• Machine learning accelerates materials design, but current predictors for solid electrolytes often lack dynamic properties.
• Static structural parameters may not fully capture the complexities of ionic transport mechanisms.

Purpose of the Study:

• To develop advanced machine learning models for predicting ionic conductivity in solid electrolytes.
• To incorporate dynamic properties, specifically phonon-related features, into predictive models.
• To identify novel superionic conductor materials through computational screening.

Main Methods:

• Compiled 14 phonon-related descriptors from first-principles calculations.
• Included 16 additional structural and electronic property descriptors.
• Developed and evaluated logistic regression and random forest regression models.

Main Results:

• Logistic regression models achieved 93% accuracy.
• Random forest regression model yielded a root-mean-square error of 1.179 for log(σ) and an R² of 0.710.
• Phonon-related features were identified as essential for accurate conductivity prediction.
• Screened 264 Li-containing materials, identifying 11 promising superionic conductor candidates.

Conclusions:

• Machine learning models incorporating dynamic phonon features significantly improve ionic conductivity prediction for solid electrolytes.
• The developed models offer a powerful tool for accelerated discovery of advanced solid-state materials.
• Identified 11 novel Li-containing materials with potential as superionic conductors.