LLG3D

LLG3D is a solver for the stochastic Landau-Lifshitz-Gilbert equation in 3D. It is written in Python and may run in parallel using MPI or OpenCL. Details about the model, the numerical method and the physical results can be found in Courtès et al. [1].

Usage

Usage

• Install
  • Install Numpy (sequential) solver
  • Install MPI solver
  • Install OpenCL solver
  • Install LLG3D in editable mode
• Execute
  • Basic execution
  • Domain geometry and boundary conditions
  • Execute in parallel
  • Execute from a python script
  • Execute on a SLURM cluster
  • Domain wall simulation
• Postprocess

Computing performance

Because 3D transient simulation are computationally intensive, performance is a key aspect of LLG3D. We provide various techniques to profile and check the performance of LLG3D simulations. The data provided can be used to situate the performance of the code relative to the results observed in your computing environment.

Performance analysis

• Profiling
  • Using the internal profiler
    • NumPy-based profiling
    • OpenCL-based profiling
  • Using cProfile
• Throughput measurements
  • Performance on an Apple M3 Max
  • Performance on a multi-core server with GPUs
  • Comparison of three GPU platforms
• Scaling measurement
  • Acceleration on an Apple M3 Max
  • Acceleration on a SLURM cluster
• Checking for performance regressions
  • Usage example

References

llg3d.main	Define a CLI for running LLG3D simulations.
llg3d.parameters	Parameters for the simulation.
llg3d.grid	Define the computational grid for the simulation.
llg3d.element	Define the chemical elements.
llg3d.solvers	Define various types of solvers.
llg3d.io	Input/Output functions.
llg3d.post	Post-processing tools.
llg3d.benchmarks	Benchmarks package.

Bibliography

[1]

Clémentine Courtès, Matthieu Boileau, Raphaël Côte, Paul Antoine Hervieux, and Giovanni Manfredi. Micromagnetic simulations of the size dependence of the Curie temperature in ferromagnetic nanowires and nanolayers. Journal of Magnetism and Magnetic Materials, 598:172040, May 2024. URL: https://hal.science/hal-04364178, doi:10.1016/j.jmmm.2024.172040.