Machine Learning Directed Search for Ultraincompressible, Superhard Materials.

Area of Science:

• Materials Science
• Computational Materials Science
• Solid State Chemistry

Background:

• Developing materials with exceptional mechanical properties, particularly high hardness, is a significant challenge.
• Predictive modeling can accelerate the discovery of novel functional inorganic materials.

Purpose of the Study:

• To develop a machine-learning model to predict elastic moduli as a proxy for material hardness.
• To identify and synthesize novel ultraincompressible and superhard inorganic compounds.

Main Methods:

• A support vector machine regression model was employed to screen 118,287 compounds from crystal structure databases.
• Ternary rhenium tungsten carbide and quaternary molybdenum tungsten borocarbide were synthesized at ambient pressure.
• High-pressure diamond anvil cell measurements and Vickers microhardness tests were conducted.

Main Results:

• Machine learning accurately predicted the bulk modulus of synthesized compounds with less than 10% error.
• Both synthesized compounds exhibited ultraincompressible behavior and extremely high hardness (>40 GPa).
• The identified materials surpassed the superhard threshold at low indentation loads.

Conclusions:

• Machine learning is an effective strategy for accelerating the discovery of advanced functional inorganic materials.
• The developed model successfully identified novel superhard compounds with exceptional mechanical properties.
• This work demonstrates a powerful approach for targeted materials design and synthesis.