NeuralMag#

NeuralMag is a differentiable micromagnetic simulator built around a SymPy form compiler, designed specifically with inverse problems in mind. It runs on top of JAX or PyTorch and supports CPU and GPU execution out of the box.

What you get#

Nodal finite-difference discretization on a regular cuboid grid, plus a cell-centred FIC variant that reuses the same form compiler — see Discretization.
Built-in field contributions: external field, exchange, demagnetization, uniaxial / cubic anisotropy, interface and bulk DMI, and interlayer exchange.
A differentiable Landau–Lifshitz–Gilbert time integrator and a small set of loggers.
A purely functional state interface: material parameters, applied fields, even the geometry mask can be lambdas of other state attributes, which makes jax.grad / torch.autograd loops natural — see Dynamic Attributes, resolve and remap.

Where to go next#

New here? Read Introduction for the mental model and then walk through the Getting Started notebook.
Setting up a multi-material or topology-optimization problem? See Domains and Material Density and Dynamic Attributes, resolve and remap.
Curious about the math? Discretization covers the nodal finite-difference and FIC schemes; Form Compiler explains how the SymPy-based form compiler works under the hood.
Simulating extended or periodic systems? See Periodic boundary conditions for true and pseudo-periodic boundary conditions.
Looking for an API? See the Reference.
Looking for a working script? Browse the Examples.
Using an AI assistant (Claude, Cursor, ChatGPT)? See Using NeuralMag with AI assistants — NeuralMag ships machine-readable context so an LLM can help you write correct simulation code.

Download and install#

NeuralMag is a Python package and requires Python ≥ 3.8 (≥ 3.10 for the JAX backend). Install with one of

pip install "neuralmag[jax]"

pip install "neuralmag[torch]"

You can also install both backends and choose between them at runtime via the NM_BACKEND environment variable.

How to cite#

If you use NeuralMag in scientific work, please cite the accompanying paper:

C. Abert, F. Bruckner, A. Voronov, M. Lang, S. A. Pathak, S. Holt, R. Kraft, R. Allayarov, P. Flauger, S. Koraltan, T. Schrefl, A. Chumak, H. Fangohr, D. Suess, “NeuralMag: an open-source nodal finite-difference code for inverse micromagnetics”, npj Comput. Mater. 11, 193 (2025), doi:10.1038/s41524-025-01688-1.

BibTeX:

@article{Abert2025NeuralMag,
  author  = {Abert, Claas and Bruckner, Florian and Voronov, Andrii and
             Lang, Martin and Pathak, Swapneel Amit and Holt, Sam and
             Kraft, Roman and Allayarov, Rustam and Flauger, Paul and
             Koraltan, Sabri and Schrefl, Thomas and Chumak, Andrii and
             Fangohr, Hans and Suess, Dieter},
  title   = {{NeuralMag}: an open-source nodal finite-difference code for
             inverse micromagnetics},
  journal = {npj Computational Materials},
  volume  = {11},
  pages   = {193},
  year    = {2025},
  doi     = {10.1038/s41524-025-01688-1},
}

Contribute#

Contributions of any kind — bug fixes, new field terms, documentation improvements — are very welcome. NeuralMag is licensed under the MIT License; by contributing you agree to license your contribution under the same terms.