magnum.np: a PyTorch based GPU enhanced finite difference micromagnetic simulation framework for high level...

Area of Science:

• Computational physics
• Materials science
• Software engineering

Background:

• Micromagnetic simulations are crucial for understanding magnetic materials.
• Existing codes can be complex and lack flexibility for novel algorithm development.
• PyTorch offers powerful tensor computation and automatic differentiation capabilities.

Purpose of the Study:

• To introduce magnum.np, a novel micromagnetic finite-difference library.
• To leverage PyTorch for enhanced maintainability, extensibility, and performance.
• To enable efficient simulations and facilitate the investigation of new modeling approaches and inverse problems.

Main Methods:

• Developed magnum.np as a micromagnetic finite-difference library using PyTorch.
• Utilized PyTorch's tensor operations and automatic differentiation (autograd).
• Benchmarked performance against state-of-the-art codes like mumax3.

Main Results:

• Magnum.np demonstrates competitive performance compared to mumax3.
• The library allows for efficient execution on GPUs and potentially TPUs.
• PyTorch's autograd feature enables handling of inverse problems.

Conclusions:

• Magnum.np provides a flexible and efficient platform for micromagnetic simulations.
• The PyTorch-based approach accelerates the implementation of new functionalities.
• This library opens new avenues for research in micromagnetics and inverse problem analysis.