Data Mining and Graph Network Deep Learning for Band Gap Prediction in Crystalline Borate Materials

Area of Science:

• Materials Science
• Computational Chemistry
• Solid-State Physics

Background:

• Crystalline borates are crucial functional materials for photocatalysis and lasers.
• Accurate band gap determination is challenging due to computational costs and accuracy limitations of first-principles methods.
• Machine learning (ML) shows promise for materials property prediction but requires high-quality datasets.

Purpose of the Study:

• To develop a high-accuracy ML model for predicting crystalline borate band gaps.
• To overcome limitations of traditional computational methods in materials design.
• To facilitate the discovery of novel borate materials with desired band gaps.

Main Methods:

• Constructed an experimental inorganic borate database using natural language processing and domain knowledge.
• Employed graph network deep learning for band gap prediction.
• Validated the ML model's accuracy against experimental measurements and a newly synthesized material.

Main Results:

• Achieved accurate prediction of borate band gaps across visible-light to deep-ultraviolet (DUV) regions.
• The ML model successfully identified most investigated DUV borates in a screening test.
• Demonstrated the model's extrapolative ability with a novel borate crystal, Ag3B6O10NO3.

Conclusions:

• Developed a cost-effective, high-quality ML model for predicting borate band gaps.
• The ML model aids in efficient material screening and design for photocatalysis and laser applications.
• Implemented a web application for convenient use in materials engineering for targeted band gap discovery.