An accurate interatomic potential for the TiAlNb ternary alloy developed by deep neural network learning method

Area of Science:

• Materials Science
• Computational Materials Science
• Machine Learning in Materials

Background:

• Traditional atomistic force fields struggle with the complex phase diagram and bonding of TiAl alloys.
• Accurate simulation of TiAlNb ternary alloys requires advanced interatomic potentials.

Purpose of the Study:

• To develop a machine learning interatomic potential for the TiAlNb ternary alloy.
• To validate the potential's accuracy against first-principles calculations and experimental data.

Main Methods:

• Utilized a deep neural network method for interatomic potential development.
• Trained the model on a dataset derived from first-principles calculations, including various structural configurations.
• Validated the potential by comparing predicted bulk properties, surface energies, and fault energies with DFT values.

Main Results:

• The machine learning potential accurately reproduced bulk properties, surface energies, and fault energies.
• Successfully predicted the formation energy and stacking fault energy of Nb-doped γ-TiAl.
• Simulated tensile properties of γ-TiAl, showing good agreement with experimental results.

Conclusions:

• The developed deep neural network-based interatomic potential is suitable for TiAlNb alloys.
• The potential demonstrates applicability under practical conditions for materials simulations.
• This work advances the use of machine learning in predicting properties of complex alloys.