Machine Learning-Assisted Exploration of Intrinsically Spin-Ordered Two-Dimensional (2D) Nanomagnets

Area of Science:

• Condensed Matter Physics
• Materials Science
• Computational Materials Science

Background:

• Two-dimensional (2D) nanomagnets exhibit unique spin-ordering properties distinct from conventional materials.
• Machine learning (ML) offers a powerful approach to accelerate the discovery and characterization of novel 2D materials.
• Accurate prediction of material properties is crucial for identifying 2D magnetic materials for specific applications.

Purpose of the Study:

• To develop and apply ML-accelerated methods for estimating key properties of 2D magnetic materials.
• To predict the HSE bandgap and magnetoanisotropic energy (MAE) of intrinsic 2D magnetic materials.
• To enhance the efficiency and reliability of exploring the vast 2D material space.

Main Methods:

• Utilized supervised ML algorithms to derive predictive descriptors for 2D magnetic material properties.
• Employed feature selection scores to reduce complexity and improve model accuracy.
• Trained and evaluated various regression models (Linear Regression, Lasso, Decision Tree, Random Forest, XGBoost, SVM) using data from the C2DB database.

Main Results:

• The Random Forest model achieved a low root-mean-square error (RMSE) of 0.22 eV for predicting HSE band gaps.
• Linear Regression models provided excellent fits for magnetoanisotropic energy (MAE(x) and MAE(y)) with RMSEs of 0.25 meV and 0.22 meV, respectively.
• Demonstrated the effectiveness of ML in accurately predicting critical material properties.

Conclusions:

• Integrating interpretable ML models with density functional theory (DFT) provides a fast and dependable method for discovering 2D magnetic materials.
• This collaborative approach significantly speeds up material property analysis and expands the searchable material space.
• ML-driven material discovery is essential for advancing research in 2D magnetism and related technologies.