Installation

muNavierStokes is a small pure-Python package. Its one non-trivial dependency is µGrid with FFT support — and, for parallel or GPU runs, built with MPI/CUDA. Everything else (numpy, mpi4py) is standard.

Dependencies

package	needed for
muGrid (with FFTEngine; NetCDF for file I/O)	the solver and output
numpy	array math
mpi4py	MPI-parallel runs
matplotlib, netCDF4	the post-processing scripts

Installing muGrid

µGrid with the FFTEngine may not yet be on PyPI as a release wheel, so it is usually built from source. A minimal CPU build:

pip install -v "muGrid @ git+https://github.com/muSpectre/muGrid.git"

µGrid autodetects MPI and NetCDF from the system; install their development packages first to enable parallel runs and file I/O, e.g. on Debian/Ubuntu:

sudo apt-get install -y cmake ninja-build \
    openmpi-bin libopenmpi-dev libnetcdf-dev libpnetcdf-dev
CC=mpicc pip install --no-binary mpi4py mpi4py
pip install -v "muGrid @ git+https://github.com/muSpectre/muGrid.git"

For a GPU build, compile µGrid with CUDA enabled (see GPU).

Installing muNavierStokes

Use --no-deps against a source-built µGrid

pyproject.toml pins muGrid>=0.109.0. A git/source build of µGrid carries a setuptools-scm dev version that does not satisfy that pin, so a normal pip install -e . would try to pull a release wheel from PyPI and clobber your MPI/NetCDF-enabled build. Install without re-resolving dependencies:

# numpy / mpi4py / muGrid already present from the steps above
pip install --no-deps -e .
pip install pytest pytest-cov netCDF4   # optional: tests + post-processing

Verifying the install

python -c "import muGrid; print('mpi', muGrid.has_mpi, 'gpu', muGrid.has_gpu, 'netcdf', muGrid.has_netcdf)"
python -c "import muNavierStokes; print(muNavierStokes.__version__)"
pytest            # run the test suite

The suite includes GPU tests (tests/test_gpu.py) that skip automatically when no GPU/CuPy is present, so pytest is a no-op for those on CPU-only machines.