Installation

In a Python 3.10+ environment:

pip install "mpiarray[mpi]"   # with mpi4py, for runs under mpiexec
pip install mpiarray          # serial only

The mpi extra installs mpi4py, which needs an MPI library:

brew install open-mpi                              # macOS
sudo apt-get install libopenmpi-dev openmpi-bin    # Debian/Ubuntu

Without mpi4py, mpiarray uses the serial stand-in for mpi4py.MPI from maybempi (from maybempi import MPI): every array lives on one rank, and no MPI library is needed. This is enough for development, notebooks and tests of serial code. Under mpiexec without mpi4py, maybempi warns, and every process then computes the whole problem as rank 0. With mpi4py installed, a plain python script.py still uses the stand-in and never starts MPI; set MAYBEMPI=1 to force MPI or MAYBEMPI=0 to force the stand-in.

For GPU arrays, also install the CuPy package for your CUDA version (e.g. cupy-cuda12x); see cunumpy for selecting the backend.

From a checkout

make install          # uv sync --extra dev, plus the pre-commit hooks
# or
pip install -e ".[dev]"

Optional extras

Extra	Installs
mpi	mpi4py, for parallel runs
test	pytest and coverage
docs	the notebook runner and griffe for the API reference
dev	formatters and linters, plus the mpi, test and docs extras