A Machine-Learning-Assisted Crystalline Structure Prediction Framework To Accelerate Materials Discovery.

Area of Science:

• Materials Science
• Computational Chemistry
• Crystallography

Background:

• Traditional crystal structure prediction methods are computationally expensive due to the need for efficient structure sampling and accurate energy evaluation.
• Developing new materials relies heavily on predicting their crystal structures and properties.

Purpose of the Study:

• To develop an accelerated system for predicting new crystal structures.
• To reduce the computational cost associated with crystal structure prediction.

Main Methods:

• Developed a Machine-learning-Assisted CRYStalline Materials sAmpling sysTem (MAXMAT).
• Utilized PyXtal for efficient crystal structure generation.
• Employed M3GNET, a machine learning potential model, for rapid energy evaluation.

Main Results:

• MAXMAT successfully performed crystal structure searches on TiO2, MgAl2O4, and BaBOF3 systems, demonstrating accuracy and efficiency.
• Predicted novel nonlinear optical materials in LiZnGaS3 and CaBOF3 systems.
• Identified several thermodynamically synthesizable structures with high performance.

Conclusions:

• MAXMAT significantly accelerates the crystal structure prediction process.
• The developed system aids in the discovery of new materials, including potential nonlinear optical applications.
• This machine learning-assisted approach offers a cost-effective alternative to traditional methods.