Density-of-states similarity descriptor for unsupervised learning from materials data

Area of Science:

• Materials Science
• Computational Chemistry
• Condensed Matter Physics

Background:

• Large materials databases like the Computational 2D Materials Database (C2DB) contain vast amounts of data.
• Efficient methods are needed to analyze and categorize materials based on their properties.
• Understanding electronic structure is key to predicting material behavior.

Purpose of the Study:

• To develop a novel materials descriptor based on electronic density-of-states (DOS).
• To apply this descriptor to group similar materials within the C2DB.
• To enable automated exploratory and confirmatory data analysis of materials.

Main Methods:

• Development of a materials descriptor utilizing electronic density-of-states (DOS).
• Application of a clustering algorithm to group materials with similar electronic structures.
• Introduction of supplementary descriptors for crystal structure, composition, and electronic configuration analysis.

Main Results:

• Identification of material clusters with similar electronic structures in the C2DB.
• Characterization of clusters revealing commonalities like isoelectronic nature and crystal symmetry.
• Discovery of outlier materials whose similarities are not explained by conventional descriptors.

Conclusions:

• The developed DOS-based descriptor effectively groups materials with similar electronic properties.
• The approach facilitates the identification of structure-property relationships and material outliers.
• This methodology offers a powerful tool for exploring and understanding large materials datasets.