Accurate space-group prediction from composition.

Area of Science:

• Crystallography and Materials Science
• Computational Chemistry
• Machine Learning

Background:

• Predicting crystal symmetry from chemical composition alone is a significant challenge in materials science.
• Existing crystallographic databases have limitations in data quantity and distribution, impacting the predictive power of machine learning models.
• Accurate prediction of crystal structure is crucial for understanding material properties and discovering new compounds.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting crystal symmetry directly from chemical composition.
• To overcome the limitations of existing databases by compiling and utilizing a comprehensive dataset of crystallographic information.
• To provide accessible tools for predicting crystal system, Bravais lattice, point group, and space group.

Main Methods:

• Compilation of virtually all available crystallographic information.
• Training and testing of multiple machine learning models, including composition-driven random-forest classification.
• Utilizing a large set of chemical and structural descriptors for model training.

Main Results:

• Composition-driven random-forest classification demonstrated the highest predictive performance.
• The models achieved notable accuracy in predicting crystal system, Bravais lattice, point group, and space group.
• The developed models significantly outperform predictions based on popular crystallographic databases alone.

Conclusions:

• Machine learning, particularly random-forest classification with comprehensive descriptors, offers a powerful approach for predicting crystal symmetry from chemical composition.
• The publicly available software (COSY) provides an accessible tool for researchers to predict crystallographic properties of inorganic compounds.
• This work advances the field of materials discovery by enabling more efficient prediction of crystal structures.