Installation

For Linux and macOS, we recommend installation via Mamba. Windows users should use the Docker instructions.

Install the released package with Mamba on Linux and macOS

A prebuilt conda package is available for Linux and macOS. First install Miniforge, which provides Conda and Mamba, then create a separate environment for edrixs:

mamba create --name edrixs_env -c conda-forge --strict-channel-priority \
    edrixs matplotlib
conda activate edrixs_env

edrixs will also run on Google Colaboratory, but is not installed there by default. Install Conda and then edrixs from within a notebook cell:

!pip install -q condacolab
import condacolab
condacolab.install()
!conda install -c conda-forge edrixs

Build the current source with Mamba on Linux or macOS

This is the recommended way to build the current source on Linux and macOS. See the macOS compiler note if the standard build fails on a recent macOS release.

Install Miniforge, then create and activate the build environment:

mamba create --name edrixs_env -c conda-forge --strict-channel-priority \
    python=3.14 "numpy>=2" scipy sympy matplotlib mpi4py \
    "arpack=*=mpi_openmpi*" openmpi gfortran \
    "libblas=*=*openblas" cmake ninja pip setuptools wheel \
    sphinx ipython numpydoc pillow sphinx-copybutton sphinx-gallery \
    sphinx_rtd_theme pytest
conda activate edrixs_env

Clone and install edrixs from the repository root:

git clone https://github.com/EDRIXS/edrixs.git
cd edrixs
python -m pip install --no-build-isolation --no-deps .

The --no-build-isolation option makes the extension use NumPy and the compiler toolchain from the activated environment. The --no-deps option prevents pip from replacing the compatible conda packages installed above.

Requirements

The Mamba command above installs compatible versions of all build and runtime requirements. The supported versions and required components are:

Python 3.10 or newer; Python 3.14 is used in the recommended environment

NumPy 1.26 or newer at runtime; NumPy 2 or newer is used to build the extension for compatibility with both NumPy 1.26 and 2.x

SciPy, SymPy, and Matplotlib

Sphinx, IPython, numpydoc, Pillow, sphinx-copybutton, sphinx-gallery, and sphinx-rtd-theme for building the documentation

CMake 3.17.3 or newer and Ninja

A Fortran compiler; the recommended environment currently uses gfortran 16

An MPI environment with Fortran and C compilers; OpenMPI 5 and its mpif90 and mpicc wrappers are used in the recommended environment

mpi4py 4 or newer, built with the same MPI implementation used to build edrixs

BLAS and LAPACK; OpenBLAS 0.3 is used in the recommended environment

ARPACK-NG 3.9 or newer, built with MPI support

Install with Docker on Windows

For Windows, we recommend using the maintained edrixs Docker image instead of building the Fortran extension natively. See edrixs and docker for the image, Docker Compose configuration, and usage instructions.

macOS compiler note

If compilation reports that libSystem.tbd is malformed or contains an unknown architecture such as arm64e.x1, use Apple Clang and Apple’s SDK-compatible linker while retaining gfortran, OpenMPI, ARPACK, and BLAS from the active Conda environment:

xcode-select -p || xcode-select --install
export PATH="$CONDA_PREFIX/bin:/usr/bin:/bin"
export CC=/usr/bin/clang
export CXX=/usr/bin/clang++
export FC="$CONDA_PREFIX/bin/gfortran"
export F77="$FC"
export OMPI_CC="$CC"
export OMPI_CXX="$CXX"
export OMPI_FC="$FC"
apple_bin="$(dirname "$(xcrun --find ld)")"
export FFLAGS="-B${apple_bin}"
export FCFLAGS="$FFLAGS"
export LDFLAGS="-B${apple_bin}"
python -m pip install \
    --no-build-isolation --no-deps .

When retrying after changing compilers or environments, first remove the generated CMake cache with cmake -E remove_directory build.