Installation

mpifield needs Python 3.10 or newer. Install it from PyPI:

pip install mpifield

or, for development, from a checkout with uv:

git clone https://github.com/max-models/mpifield.git
cd mpifield
make install    # uv sync --extra dev, plus the pre-commit hooks

Optional extras

Extra	Installs	For
mpi	mpi4py	runs under mpiexec or srun; serial runs use maybempi’s stand-in and never start MPI
plot	matplotlib, plasma-plots	Field.plot
gpu	CuPy	the CUDA kernels, for fields and markers on the GPU
metal	MLX	the Metal kernels on Apple-silicon Macs
test	pytest, pytest-mpi, coverage	the tests
docs	the notebook runner, griffe	this site
dev	ruff, pyright, pre-commit, plus test and docs	development

pip install "mpifield[mpi,plot]"

The compiled kernels

The CPU kernels are Python files compiled with Pyccel the first time they run, with OpenMP, and cached on disk:

Variable	Default	Effect
MPIFIELD_KERNEL_CACHE	~/.cache/mpifield/kernels	where the compiled kernels are kept
MPIFIELD_OPENMP	1	0 builds them without OpenMP
OMP_NUM_THREADS	all cores	threads per process; under MPI, the cores per rank

Compiling needs a C compiler. Without one, run the NumPy versions: cunumpy.kernels.set_host_kernel_implementation("numpy").