Machine-learning-assisted insight into spin ice Dy₂Ti₂O₇.

Area of Science:

• Condensed matter physics
• Materials science
• Computational physics

Background:

• Extracting models from complex experimental data, such as spin liquid formation in frustrated magnets, is challenging.
• Disorder, glass formation, and scattering data interpretation hinder understanding of magnetic materials like Dy2Ti2O7.

Purpose of the Study:

• To develop an automated capability for extracting model Hamiltonians from experimental data.
• To identify different magnetic regimes and improve the understanding of complex magnetic behaviors.

Main Methods:

• Training an autoencoder to learn a compressed representation of 3D diffuse scattering data across various spin Hamiltonians.
• Matching the autoencoder's output to experimental scattering and heat capacity data to find optimal Hamiltonians with confidence intervals.

Main Results:

• The optimal Hamiltonian accurately predicted temperature and field dependence of magnetic structure and magnetization.
• The method successfully identified glass formation and irreversibility in Dy2Ti2O7.
• The autoencoder categorized different magnetic behaviors and removed noise/artifacts from raw data.

Conclusions:

• The developed automated methodology effectively extracts model Hamiltonians from experimental data.
• This approach enhances the analysis of complex magnetic materials and scattering problems.
• The technique is broadly applicable to other materials and scientific investigations.