Constrained DFT-based magnetic machine-learning potentials for magnetic alloys: a case study of Fe-Al.

Area of Science:

• Materials Science
• Computational Materials Science
• Condensed Matter Physics

Background:

• Accurate modeling of magnetic materials is crucial for developing new technologies.
• Traditional methods struggle to capture the behavior of materials with non-equilibrium magnetic moments.
• Machine learning offers a promising avenue for developing more sophisticated interatomic potentials.

Purpose of the Study:

• To develop a novel machine-learning interatomic potential for multi-component magnetic materials.
• To incorporate magnetic moments as degrees of freedom alongside atomic properties.
• To enable predictions of material properties in excited magnetic states.

Main Methods:

• Developed a machine-learning interatomic potential considering magnetic moments, atomic positions, types, and lattice vectors.
• Created a training dataset using constrained density-functional theory (cDFT) for configurations with non-equilibrium magnetic moments.
• Trained and verified the potential on body-centered cubic (bcc) iron-aluminum (Fe-Al) alloys with varying compositions and atomic arrangements.

Main Results:

• The machine-learning potentials accurately predicted formation energies, lattice parameters, and total magnetic moments for Fe-Al systems.
• Calculated properties showed good agreement with results obtained from DFT.
• The model qualitatively reproduced experimentally observed anomalous volume-composition dependence in Fe-Al alloys.

Conclusions:

• The proposed machine-learning interatomic potential effectively captures the behavior of magnetic materials, including those in excited states.
• This approach provides a reliable tool for predicting properties of complex magnetic materials.
• The findings pave the way for accelerated discovery and design of novel magnetic materials.