To install the latest release, choose the command matching your hardware:
| Hardware | Installation |
|---|---|
| CPU | pip install pyscfad |
| NVIDIA GPU (CUDA 12) | pip install "pyscfad[cuda12]" |
| NVIDIA GPU (CUDA 13) | pip install "pyscfad[cuda13]" |
To install the development version, run:
pip install git+https://github.com/fishjojo/pyscfad.git
The dependent C/C++ library pyscfadlib can be compiled from source following the instruction
here.
Prebuilt wheels are published for the following platforms:
| Platform | CPU | NVIDIA GPU |
|---|---|---|
| Linux, x86_64 | yes | yes |
| Linux, aarch64 | yes | yes |
| macOS, Apple silicon (arm64) | yes | n/a |
| macOS, Intel (x86_64) | no | n/a |
| Windows, x86_64 | no | no |
| Windows WSL2, x86_64 | yes | yes |
On platforms without a prebuilt wheel, pyscfadlib can still be compiled from source
(see the link above).
pyscfad depends on
jax>=0.9.1,<0.11,
pyscfadlib>=0.3,
pyscf>=2.3, and
pyscf-properties.
The following paper should be cited in publications utilizing the PySCFAD program package:
Differentiable quantum chemistry with PySCF for molecules and materials at the mean-field level and beyond, X. Zhang, G. K.-L. Chan, J. Chem. Phys., 157, 204801 (2022)