Treating Superhard Materials as Anomalies.

Area of Science:

• Materials Science
• Computational Materials Science
• Artificial Intelligence in Materials Discovery

Background:

• Superhard materials, defined by Vickers hardness ≥40 GPa, are exceptionally rare, constituting less than 0.1% of known materials.
• Discovering new superhard materials is crucial for various technological applications but remains a significant challenge due to their scarcity.

Purpose of the Study:

• To develop a novel unsupervised anomaly detection framework for identifying superhard materials.
• To leverage AI for discovering materials with rare and exceptional physical properties beyond superhardness.

Main Methods:

• An anomaly detection framework was created to encode and reconstruct material composition and crystal structure data without supervision.
• The model quantifies deviations from 'normal' material behavior to identify outliers with potential superhard properties.

Main Results:

• The framework successfully identified outlier materials exhibiting superhard characteristics.
• Analysis of discrepancies between encoded and decoded crystal structures provided fundamental insights into the origins of material hardness.

Conclusions:

• The unsupervised anomaly detection approach is effective for discovering scarce superhard materials.
• The methodology is generalizable for designing other outlier materials with unique and unexpected physical properties.