Composition-based machine learning for predicting and designing Mn⁴⁺-doped phosphors.

Area of Science:

• Materials Science
• Solid-State Chemistry
• Computational Materials Science

Background:

• Predicting optical properties of phosphors is crucial for lighting and display applications.
• Existing methods often require complex structural information, limiting rapid material discovery.
• Mn4+-doped phosphors are important for applications requiring narrow-band red emission.

Purpose of the Study:

• To develop a data-driven model for predicting excitation/emission wavelengths and crystal field energy levels in Mn4+-doped phosphors.
• To establish the largest experimental dataset for Mn4+-activated phosphors for model training.
• To enable inverse design of phosphors based on desired optical outputs.

Main Methods:

• Construction of a comprehensive experimental dataset of Mn4+-activated phosphors.
• Application of machine learning models, including K-Nearest Neighbors and Extra Trees Regressors.
• Validation of models on Eu3+-doped systems to assess generalization capabilities.

Main Results:

• Accurate prediction of excitation and emission wavelengths using elemental composition alone.
• K-Nearest Neighbors and Extra Trees Regressors showed highest accuracy for specific property predictions.
• Successful generalization to Eu3+-doped systems, demonstrating broad applicability.

Conclusions:

• A data-driven approach can accurately predict optical properties of phosphors without complex descriptors.
• This method facilitates efficient and interpretable discovery of novel luminescent materials.
• The developed models provide a foundation for theory-informed materials design.