Massive Monte Carlo simulations-guided interpretable learning of two-dimensional Curie temperature

Area of Science:

• Condensed Matter Physics
• Materials Science
• Computational Physics

Background:

• Monte Carlo (MC) simulations are standard for estimating Curie temperatures (T_C) in 2D Heisenberg magnets.
• Existing methods face challenges with the complexity of spin Hamiltonians and anisotropy.

Purpose of the Study:

• Develop accurate, data-driven models for predicting T_C in diverse 2D magnetic materials.
• Explore alternatives to traditional MC simulations for faster material screening.

Main Methods:

• Trained deep neural networks on a dataset of ~250,000 materials, considering up to four nearest neighbors and single-ion anisotropy.
• Employed a bisection-based MC technique for efficient data generation.
• Combined learning-from-data with data-from-learning to ensure uniform T_C distribution.

Main Results:

• Achieved testing R² scores near 0.99, demonstrating high accuracy of the data-driven models.
• Ensured a uniform data distribution across a wide T_C range (10–1,000 K).
• Confirmed model interpretability through feature analysis.

Conclusions:

• Data-driven models provide a highly accurate and efficient alternative to MC simulations for predicting T_C in 2D magnets.
• The developed approach facilitates accelerated discovery of novel 2D magnetic materials.
• First-principles calculations can accurately estimate T_C without empirical corrections.