Machine learning potential as a guide for eutectic in ultra-refractory multicomponent ceramics.

Area of Science:

• Materials Science
• Computational Materials Science
• Physical Chemistry

Background:

• Eutectic point determination is crucial for materials development but challenging for ultra-refractory alloys (melting points > 3000 K) due to experimental costs and technical difficulties.
• Conventional methods are impractical for high-melting-point systems, hindering the study of advanced materials.

Purpose of the Study:

• To propose a novel AI-driven criterion for determining eutectic point concentrations in ultra-refractory alloys.
• To develop a computational approach that bypasses the limitations of experimental methods for high-temperature systems.

Main Methods:

• Development of a machine-learning interatomic potential using a neural network, achieving accuracy comparable to ab initio methods.
• Application of the novel criterion to the Ti-B-C system, a well-studied three-component refractory system.
• Algorithm designed to operate effectively in the liquid phase, not requiring solid-state crystalline structure information.

Main Results:

• The proposed AI criterion successfully predicts eutectic point concentrations in ultra-refractory alloys.
• The machine learning potential demonstrates high accuracy, comparable to established computational techniques.
• The method's efficacy was validated using the Ti-B-C system.

Conclusions:

• Advanced AI modeling provides a powerful and cost-effective alternative for predicting eutectic points in challenging ultra-refractory alloy systems.
• The developed approach enables accurate eutectic point estimation without reliance on solid-state structural data, broadening applicability.
• This work paves the way for accelerated discovery and design of novel high-performance refractory materials.