Homogeneous nucleation of undercooled Al-Ni melts via a machine-learned interaction potential

Area of Science:

• Materials Science
• Computational Materials Science
• Chemical Engineering

Background:

• Homogeneous nucleation is crucial for material solidification and microstructure, but atomistic simulations are limited by scale and accuracy.
• Accurately describing interatomic interactions, especially in alloys with chemical order, is computationally demanding.
• Ab initio simulations cannot reach the large scales required to observe rare nucleation events.

Purpose of the Study:

• To develop a machine learning potential for binary Al-Ni alloys to overcome limitations in simulating nucleation.
• To apply this potential in large-scale molecular dynamics simulations to study homogeneous nucleation processes.
• To investigate the nucleation pathways of Al-Ni alloys and pure Ni and compare them with existing models.

Main Methods:

• A high-dimensional neural network potential was constructed for binary Al-Ni alloys.
• The potential was rigorously validated against experimental data, including diffusion, viscosity, scattering, and melting temperature.
• Large-scale molecular dynamics simulations were performed using the validated machine learning potential.

Main Results:

• Pure Ni was observed to nucleate in a single step into an fcc crystal phase, contrasting with previous simulations.
• The Al-Ni alloy at equiatomic composition showed a single-step nucleation pathway towards a B2 structure.
• The study highlights the significant impact of atomic interaction potentials on simulated nucleation pathways.

Conclusions:

• Machine learning potentials offer a powerful approach to accurately and efficiently simulate complex materials phenomena like nucleation.
• The distinct nucleation pathways observed underscore the sensitivity of solidification processes to interatomic forces.
• Findings provide insights into the solidification behavior of Al-Ni alloys, relevant for industrial applications.