A neural-network based framework of developing cross interaction in alloy embedded-atom method potentials:...

Area of Science:

• Materials Science
• Computational Materials Science
• Condensed Matter Physics

Background:

• Developing accurate interatomic potentials is crucial for molecular dynamics simulations at atomic scales.
• Combining established single-element potentials through cross-interaction (CI) is a key strategy for alloy embedded-atom method (EAM) potential development.

Purpose of the Study:

• To propose and validate a novel framework for constructing CI potential functions using neural network (NN) models.
• To develop a Zr-Nb alloy CI potential by leveraging existing Zr and Nb single-element potentials.
• To assess the capability of the developed potential in reproducing alloy properties and deformation behaviors.

Main Methods:

• A four-step framework involving extraction of characteristic points from single-element potentials, cubic spline interpolation for CI functions, accuracy evaluation against first-principle (FP) data, and NN-based optimization.
• Utilizing MA-III (Zr) and FPW (Nb) potentials as the base single-element potentials.
• Developing a Zr-Nb alloy CI potential using the proposed NN framework.

Main Results:

• The developed CI potential functions successfully integrated features from both Zr and Nb potentials.
• Calculated normalized energy-volume curves for Zr3Nb, ZrNb, and ZrNb3 phases showed good agreement with FP results.
• The new potential accurately reproduced the slip characteristics of pure Zr and Nb, surpassing other ab initio potentials.

Conclusions:

• The NN-based framework efficiently utilizes single-element potentials for alloy EAM potential development.
• The newly developed Zr-Nb alloy EAM potential provides a reliable description of complex deformation behaviors in Zr-Nb systems.
• This approach offers a convenient and effective method for creating accurate alloy potentials.