NeuralMag: an open-source nodal finite-difference code for inverse micromagnetics.

Area of Science:

• Computational physics
• Materials science
• Machine learning applications

Background:

• Micromagnetic simulations are crucial for understanding magnetic materials.
• Existing simulation codes can be computationally intensive and lack flexibility.
• The need for efficient and adaptable tools for micromagnetic research is growing.

Purpose of the Study:

• To introduce NeuralMag, an open-source Python library for micromagnetic simulations.
• To leverage machine learning frameworks for enhanced computational performance.
• To provide a flexible and accurate simulation tool for scientific research.

Main Methods:

• Utilized PyTorch and JAX for efficient tensor operations on diverse hardware (CPUs, GPUs, TPUs).
• Implemented a novel nodal finite-difference discretization scheme for improved accuracy.
• Integrated automatic differentiation capabilities for solving inverse problems.

Main Results:

• NeuralMag demonstrates competitive performance against state-of-the-art simulation codes.
• The nodal finite-difference scheme enhances accuracy without increasing computational cost.
• The library's Python interface and backend integration offer significant flexibility.

Conclusions:

• NeuralMag provides a high-performance, flexible, and accurate open-source solution for micromagnetic simulations.
• Its machine learning foundation and novel discretization scheme advance the field.
• The library is well-suited for complex problems, including time-dependent inverse problems.