Predicting miscibility in binary compounds: a machine learning and genetic algorithm study.

Area of Science:

• Materials Science
• Computational Materials Science
• Data Science

Background:

• Materials informatics and data science are crucial for advancing multi-component compound synthesis.
• Predicting miscibility in binary compounds is essential for designing new materials.
• Existing databases like Materials Project (MP) and Inorganic Crystal Structure Database (ICSD) provide valuable experimental data.

Purpose of the Study:

• To predict miscibility in binary compounds using atomic-level data and machine learning.
• To identify key factors influencing binary system miscibility.
• To discover novel, thermodynamically stable phases in binary systems.

Main Methods:

• Integration of experimental data from MP and ICSD databases for 2346 binary systems.
• Application of a random forest classification model for training and prediction.
• Utilizing advanced genetic algorithms for phase discovery in the Co-Eu system.

Main Results:

• Demonstrated the feasibility of predicting binary compound miscibility using machine learning.
• Identified significant factors affecting binary system miscibility.
• Discovered three novel, thermodynamically stable phases: CoEu₈, Co₃Eu₂, and CoEu.

Conclusions:

• Machine learning models trained on atomic data can accurately predict material miscibility.
• The study provides theoretical insights to guide experimental synthesis of binary and complex materials.
• The identified novel phases in the Co-Eu system open new avenues for materials research.