Thermally Averaged Magnetic Anisotropy Tensors via Machine Learning Based on Gaussian Moments

Area of Science:

• Computational chemistry
• Materials science
• Quantum mechanics

Background:

• Accurate modeling of molecular tensorial quantities is crucial for understanding material properties.
• Magnetic anisotropy tensor is key to magnetic behavior but computationally intensive to predict.
• Existing methods may lack accuracy or generalization for complex molecular systems.

Purpose of the Study:

• To introduce a novel machine learning method for predicting molecular magnetic anisotropy tensors.
• To demonstrate the accuracy and generalization capabilities of the proposed approach.
• To explore the application of this method in studying dynamic magnetic phenomena and spin-phonon relaxation.

Main Methods:

• Utilizing a Gaussian moment neural network approach for molecular tensorial quantity modeling.
• Training and validating the model on magnetic anisotropy tensor data.
• Integrating the model with machine-learned interatomic potentials based on Gaussian moments.

Main Results:

• Achieved high prediction accuracy for magnetic anisotropy tensors (0.3-0.4 cm^-1).
• Demonstrated excellent generalization capability for unseen molecular configurations.
• Successfully applied the method to study dynamic magnetic anisotropy behavior.

Conclusions:

• The Gaussian moment neural network provides an accurate and generalizable method for modeling magnetic anisotropy tensors.
• This approach enables the study of dynamic spin-phonon relaxation processes.
• Offers a powerful computational tool for advancing magnetic materials research.