Inorganic Crystal Structure Prototype Database Based on Unsupervised Learning of Local Atomic Environments.

Area of Science:

• Materials Science
• Crystallography
• Computational Materials Science

Background:

• Recognizing inorganic crystal structure prototypes is crucial for materials science research and design.
• Current databases primarily classify structures using crystallographic space groups.
• A novel approach is needed to enhance the understanding and discovery of inorganic materials.

Purpose of the Study:

• To develop a new method for constructing an inorganic crystal structure prototype database.
• To classify materials based on local atomic environments (LAEs) rather than solely on space groups.
• To accelerate the materials discovery process through a data-driven approach.

Main Methods:

• Employed unsupervised machine learning, specifically hierarchical clustering, on known inorganic crystal structure data.
• Utilized advanced structure fingerprints, including improved bond-orientational order parameters and smooth overlap of atomic positions, to represent LAEs.
• Developed the Structure Prototype Generator Infrastructure (SPGI) package for prototype generation.

Main Results:

• Constructed the LAE-based Inorganic Crystal Structure Prototype Database (LAE-ICSPD) containing 15,613 unique structure prototypes with defined stoichiometries.
• Successfully classified inorganic crystal structures based on local atomic environments, offering a distinct perspective from traditional space group classifications.
• The SPGI package provides a robust toolkit for generating and analyzing structure prototypes.

Conclusions:

• The LAE-ICSPD offers a comprehensive resource for materials science research.
• The developed methodology and tools facilitate a global investigation of inorganic materials.
• This data-driven approach significantly accelerates the discovery and design of new materials.