Finding the Next Superhard Material through Ensemble Learning.

Area of Science:

• Materials Science
• Computational Materials Science
• Machine Learning

Background:

• Discovering superhard materials is crucial for advanced technological applications.
• Predicting material properties from composition alone remains a significant challenge.
• Experimental synthesis and characterization of novel superhard materials are time-consuming and resource-intensive.

Purpose of the Study:

• To develop and validate an ensemble machine learning model for predicting load-dependent Vickers hardness.
• To screen a large database of compounds for potential superhard materials.
• To identify novel superhard materials, particularly in ternary borocarbide systems.

Main Methods:

• Extracted 1062 experimentally measured load-dependent Vickers hardness data from literature.
• Trained a supervised machine learning algorithm using boosting, achieving R² = 0.97.
• Screened over 66,000 compounds in crystal structure databases and analyzed ternary borocarbide phase spaces.

Main Results:

• The ensemble model accurately predicts Vickers hardness based solely on chemical composition.
• Identified 68 known materials with Vickers hardness ≥40 GPa at 0.5 N, and 10 at 5 N.
• Discovered over ten thermodynamically favorable ternary borocarbide compositions with hardness >40 GPa at 0.5 N.

Conclusions:

• Ensemble machine learning is a powerful tool for discovering superhard materials.
• The developed model significantly expands the known set of high-hardness compounds.
• This approach accelerates the identification of materials with outstanding mechanical properties.