Installation

Python version support

Officially Python 3.9, 3.10, 3.11, 3.12, and 3.13.

Installing from PyPI

pyscfad can be installed via pip from PyPI.

pip install pyscfad

Installing from source

The source code of pyscfad can be obtained by cloning the Github repository.

cd $HOME
git clone https://github.com/fishjojo/pyscfad.git

The main part of pyscfad is pure Python. One can simply add the top directory of pyscfad to the environment variable PYTHONPATH.

export PYTHONPATH=$HOME/pyscfad:$PYTHONPATH

Alternatively, one can install pyscfad locally via pip by running the following command at the top directory of pyscfad.

cd $HOME/pyscfad
pip install -e .

Installing pyscfadlib

pyscfadlib is the C extension to pyscfad that provides efficient gradient backpropagation implementations. Similarly, one can install pyscfadlib locally via pip.

cd $HOME/pyscfad/pyscfadlib
pip install -e .

Or one can manually compile the C code, and then add pyscfadlib to PYTHONPATH.

cd $HOME/pyscfad/pyscfadlib/pyscfadlib
mkdir build
cd build
cmake ..
make
export PYTHONPATH=$HOME/pyscfad/pyscfadlib:$PYTHONPATH

Note

For Mac with ARM64 architectures, one needs to set the environment variable CMAKE_OSX_ARCHITECTURES=arm64.

Dependencies

pyscfad requires the following dependencies.

Package	supported versions
numpy	>=1.17
scipy	>=1.7
h5py	>=2.7
jax	>=0.4.14,<=0.4.35
jaxlib	>=0.4.14,<=0.4.35
pyscf	>=2.3.0
pyscfadlib	>=0.1.4

Note

Since jax version 0.4.36, the tracing machinery has been modified to eliminate data-dependent tracing, which conflicts with pyscfad’s flexibility. A comprehensive update to ensure compatibility with later jax versions may be introduced in the future pyscfad 0.2 release.