Machine Learning to Predict Quasicrystals from Chemical Compositions.

Area of Science:

• Solid-state materials science
• Crystallography
• Materials informatics

Background:

• Quasicrystals represent a distinct class of solid-state materials, characterized by long-range order without periodicity.
• Over 100 stable quasicrystals have been discovered, but the discovery rate has slowed due to a lack of guiding synthesis principles.
• Traditional methods for discovering new quasicrystals are limited, necessitating novel approaches.

Purpose of the Study:

• To accelerate the discovery of new quasicrystals using a machine-learning (ML) workflow.
• To develop a predictive model for classifying solid-state materials into quasicrystals, approximant crystals, and ordinary crystals.
• To identify interpretable empirical equations governing stable quasicrystal formation.

Main Methods:

• A machine-learning model was trained using the chemical compositions of known stable quasicrystals, approximant crystals, and ordinary crystals.
• The model performed a three-class classification task to predict material phases.
• The model's predictions were validated against observed phase diagrams of ternary aluminum systems.

Main Results:

• The machine-learning workflow demonstrated superior predictive power compared to traditional methods.
• The phase prediction task achieved an overall accuracy of approximately 0.728.
• Analysis of the ML model revealed nontrivial, human-interpretable empirical equations for stable quasicrystal formation.

Conclusions:

• Machine learning offers a powerful and efficient tool for accelerating the discovery of new quasicrystals.
• The developed ML model can accurately predict quasicrystal formation, aiding materials scientists.
• The identified empirical equations provide valuable insights into the fundamental principles of quasicrystal stability.