Accelerated spin dynamics using deep learning corrections.

Area of Science:

• Physics
• Materials Science
• Computational Science

Background:

• Theoretical models accurately describe microscopic magnetic material behavior, enabling precise computer simulations like spin dynamics.
• Current spin dynamics simulations face efficiency limitations due to the time step barrier in solving many-body equations of motion.
• Small time steps ensure accuracy but are inefficient; large time steps lead to unrecoverable numerical errors.

Purpose of the Study:

• To develop a Deep Learning (DL) method for correcting numerical errors in large time step spin dynamics simulations.
• To enhance the efficiency and accuracy of simulating magnetic materials.
• To overcome the inherent limitations of small time steps in achieving efficient and accurate simulations.

Main Methods:

• A Deep Learning model was employed to calculate numerical errors associated with large time steps.
• Corrections were applied based on the computed errors to improve simulation accuracy.
• The method was validated on a 3D Ferromagnetic Heisenberg cubic lattice across various temperatures.

Main Results:

• The Deep Learning method successfully computed and corrected numerical errors in spin dynamics simulations.
• The approach achieved a 10-fold acceleration in simulation speed.
• High simulation accuracy was maintained despite using larger time steps.

Conclusions:

• Deep learning offers a viable solution to the time step barrier in spin dynamics simulations.
• This novel approach significantly enhances computational efficiency for magnetic material simulations.
• The method provides a pathway to more accurate and faster simulations of magnetic phenomena.