Machine-learning interatomic potential for barium sulfide: From thermodynamic properties to crystal growth kinetics

Area of Science:

• Materials Science
• Computational Physics
• Chemical Engineering

Background:

• Accurate interatomic potentials are crucial for simulating material properties and processes.
• Classical potentials often struggle to capture complex behaviors in semiconductors like barium sulfide (BaS).
• Machine learning potentials offer a promising alternative for high-fidelity simulations.

Purpose of the Study:

• To develop and validate a neural network-based interatomic potential (DeePMD) for BaS.
• To evaluate the performance of DeePMD against classical potentials for thermodynamic and growth properties.
• To assess the utility of DeePMD in multiscale simulations of crystal growth.

Main Methods:

• First-principles simulations were used to train the DeePMD potential for solid and liquid BaS.
• Molecular dynamics simulations were employed to calculate bulk properties and interfacial energies.
• Crystal growth simulations and integration with a kinetic phase-field model were performed.

Main Results:

• DeePMD demonstrated improved predictions of BaS density and liquid structure compared to the classical Rino potential.
• Both potentials reproduced melting temperature and linear growth near melting.
• DeePMD predicted enhanced crystal growth velocities at lower temperatures (< 1800 K), suggesting potential for spontaneous nucleation.

Conclusions:

• Machine learning potentials like DeePMD are valuable tools for accurate multiscale simulations of crystal growth.
• DeePMD provides a more predictive model for BaS properties and growth kinetics than classical potentials.
• This work highlights the potential of AI-driven methods in materials science research.