Phase Transition in Silicon from Machine Learning Informed Metadynamics.

Area of Science:

• Computational Materials Science
• Condensed Matter Physics
• Materials Informatics

Background:

• Traditional computational methods like density functional theory (DFT) are computationally expensive for large systems and long timescales.
• Simulating reconstructive phase transitions in large systems necessitates efficient computational frameworks.

Purpose of the Study:

• To develop and demonstrate a computationally efficient framework for simulating phase transitions in large material systems.
• To accelerate the study of phase transition pathways and defect formation using machine learning potentials.

Main Methods:

• Integration of metadynamics simulation with a well-trained machine learning potential, specifically deep potential.
• Application of the developed method to simulate phase transitions in bulk silicon under high pressure.

Main Results:

• The deep potential-driven metadynamics approach significantly enhances computational efficiency for large-scale simulations.
• The study successfully revealed the transition path and formation of polycrystalline silicon under specific stress conditions.
• Demonstrated the effectiveness of the new method in studying complex material behaviors and defect development.

Conclusions:

• Machine learning potentials coupled with metadynamics offer a powerful and efficient approach for investigating phase transitions in large material systems.
• This integrated method provides valuable insights into material behavior, including grain and dislocation defect formation.
• The developed framework accelerates materials discovery and understanding of complex phenomena under extreme conditions.