Magnetic Hamiltonian parameter estimation using deep learning techniques.

Area of Science:

• Condensed Matter Physics
• Materials Science
• Computational Physics

Background:

• Understanding spin textures is crucial for spintronics applications.
• Extrapolating magnetic Hamiltonian parameters from experimental spin data is vital for linking theory and experiment.

Purpose of the Study:

• To demonstrate the capability of deep learning in quantifying magnetic Hamiltonian parameters directly from magnetic domain images.
• To establish a quantitative link between experimentally observed spin textures and fundamental magnetic interactions.

Main Methods:

• Generated synthetic magnetic domain configurations using the Monte Carlo method for training.
• Developed and trained a deep neural network to recognize and quantify Hamiltonian parameters from these configurations.
• Analyzed estimation errors using statistical methods to validate the network's performance.

Main Results:

• The deep neural network was successfully trained, demonstrating its ability to relate Hamiltonian parameters to magnetic structure characteristics.
• The trained network accurately estimated parameters from experimentally observed magnetic domain images.
• Results showed strong consistency with previously reported findings, validating the method's effectiveness.

Conclusions:

• Deep learning provides a powerful tool for quantifying magnetic Hamiltonian parameters from experimental data.
• This approach effectively bridges the gap between theoretical predictions and experimental observations in magnetism.
• The methodology holds potential for application across various scientific disciplines requiring the analysis of structural data.