DFT-aided machine learning-based discovery of magnetism in Fe-based bimetallic chalcogenides

Area of Science:

• Materials Science
• Computational Materials Science
• Condensed Matter Physics

Background:

• The demand for low-cost magnetic materials is increasing due to technological advancements.
• Metal chalcogenides offer potential as abundant, alternative magnetic materials.
• Predicting magnetism in diverse chalcogenide configurations presents a significant challenge.

Purpose of the Study:

• To develop a predictive model for magnetic moments in hexagonal Fe-based bimetallic chalcogenides.
• To accurately forecast the magnetic properties based on material composition.
• To facilitate the discovery of novel magnetic materials through computational modeling.

Main Methods:

• Utilized a stacked generalization machine learning model.
• Trained the model on a dataset generated using first-principles density functional theory calculations.
• Developed a generalized algorithm for broader applicability.

Main Results:

• The machine learning model achieved high accuracy, with MSE, MAE, and R² values of 1.655 (µB)², 0.546 (µB), and 0.922, respectively, on an independent test set.
• Demonstrated accurate prediction of composition-dependent magnetism in bimetallic chalcogenides.
• Validated the model's universality for various concentrations of Ni, Co, Cr, or Mn.

Conclusions:

• The developed stacked generalization model accurately predicts magnetic moments in hexagonal Fe-based bimetallic chalcogenides.
• This approach accelerates the search for new magnetic materials with desired properties.
• The generalized algorithm enhances the model's utility for future material discovery.