Size-Dependent Nucleation in Crystal Phase Transition from Machine Learning Metadynamics.

Area of Science:

• Computational Materials Science
• Solid-State Physics
• Machine Learning Applications

Background:

• Investigating solid-solid phase transitions is crucial for understanding material behavior under extreme conditions.
• Accurate simulation of phase transitions requires efficient and scalable computational methods.

Purpose of the Study:

• To develop and apply a novel framework combining machine learning potential (MLP) and metadynamics for studying solid-solid phase transitions.
• To elucidate the influence of system size on the atomistic mechanisms governing the B4-B1 phase transition in Gallium Nitride (GaN).

Main Methods:

• Development of a scalable MLP model using spectral descriptors and neural network regression for accurate energy surface interpolation.
• Application of the MLP model within a metadynamics framework to simulate the B4-B1 phase transition of GaN at 50 GPa.
• Analysis of phase transition mechanisms across various system sizes, from 128,000 atoms upwards.

Main Results:

• The MLP model accurately interpolates the energy surface where two phases coexist.
• A sequential change in the phase transition mechanism was observed, shifting from collective modes to nucleation and growth as system size increased.
• For systems ≤ 128,000 atoms, nucleation and growth followed a preferred direction; larger systems exhibited simultaneous nucleation at multiple sites.

Conclusions:

• The study highlights the critical role of system size in determining the atomistic pathways of solid-solid phase transitions.
• Simulations with larger system sizes are essential for capturing the statistical sampling required for accurate phase transition modeling.
• The developed MLP-metadynamics framework offers a powerful tool for investigating complex phase transformations in materials.