PySCF ships new features in a second distribution, and its short description is a list of author names
Python module for quantum chemistry. Hermes, Kevin Koh, Peter Koval, Susi Lehtola, Zhendong Li, Junzi Liu, Narbe Mardirossian, James D.
At a glance
- What is it?
- A quantum chemistry framework in Python, licensed Apache-2.0 and stable at 2.14.0. The installation page quietly splits the project in two: the stable line on PyPI, and the recent work published separately, plus a set of extensions that are mostly thin pointers at packages other people maintain.
- Who is it for?
- Install from PyPI and you get a deliberately conservative core; the faster-moving parts arrive through pyscf-forge or through extras that pull in packages maintained elsewhere. Decide which of those you actually need before installing, because the geomopt extra alone brings three optimizers.
- Can I use it commercially?
- Yes. Apache-2.0 is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
- Is it still maintained?
- Yes. The repository received new commits within the last day.
- What is it written in?
- Mainly Python, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on November 22, 2026, and from our analysis. They are not legal advice.
Star history
- pyscf/pyscf1,726 ★ · +81
| Repository | 2026-08-04 | 2026-11-22 |
|---|---|---|
| pyscf/pyscf | 1.6k | 1.7k |
81 Stars in the last 110 days, measured from the daily GitHub snapshot. Compare with other projects
Editorial analysis
New features live in a different distribution from the stable one
The installation section is four commands long, and the second one is not a variant of the first.
pip install pyscf
pip install pyscf-forge
pip install pyscf[all]
pip install pyscf[dispersion]The first installs the stable release. The second is a separate distribution, and the sentence above it says why: new features developed in recent years are available in the pyscf-forge package. So the split is deliberate and stated plainly, with the stable package staying where it is while active development lands somewhere else. The third and fourth commands cover the extension modules, which the page describes as maintained as extensions of PySCF rather than as part of it: dispersion, dmrgscf, fciqmc, icmpspt, properties, semiempirical, and shciscf, with a pointer to the wider organisation for the rest. Whether that means a module exists in the core, in forge, in an extension, or in all three is something a reader has to work out per module rather than from one page.
Most extras are pointers at packages other people maintain
The optional dependency groups in the packaging metadata read like an index of neighbouring projects rather than a feature list.
forge = ["pyscf-forge"]
geomopt = ["pyberny>=0.6.2", "geometric>=0.9.7.2", "pyscf-qsdopt"]
doci = ["pyscf-doci"]
properties = ["pyscf-properties"]
semiempirical = ['pyscf-semiempirical']
cppe = ["cppe"]
pyqmc = ["pyqmc"]
bse = ["basis-set-exchange"]Most entries name a distribution rather than a library, so installing an extra pulls in a project with its own release schedule and its own maintainer. One group is denser than the others: geomopt asks for two independent geometry optimizers plus a third package, so a user who wants one optimiser ends up with three codebases on disk. The list also stops partway through its final entry, which begins with the letters for dispersion, so the complete set of groups is longer than what is visible here. Two floor constraints appear in that group and nowhere else in the extras, a minimum on the Berny optimiser and a minimum on the Geometric one, which suggests those two were the ones with known compatibility floors.
PySCF does not implement density functionals and tells you to cite the substitute
One sentence in the citation section is the most consequential line on the page for anyone publishing DFT work. PySCF does not implement density functionals; it employs external libraries to handle their evaluation, and those libraries should be cited as well in any publication that used PySCF for density functional calculations. Two substitutes are named with full references: Libxc, described in a 2018 SoftwareX paper, and XCFun, described in a 2010 Journal of Chemical Theory and Computers paper. That turns a version question into a citation question: the functionals your calculation used came from one of those libraries, and the paper names the library rather than the framework. It also means the backend can change underneath a project without the framework version changing at all, since picking a functional and picking Libxc are two separate decisions.
The citation requirement is the largest part of the page
The base package asks to be cited through a single paper, and that paper has close to a hundred authors: The Python simulations of chemistry framework, 10 years of an open-source quantum chemistry project, published in the Journal of Chemical Physics, volume 165, article 102502, in 2026. The author list runs across many lines and includes a mix of long-standing contributors and recent ones. That framing is worth noting because it positions the project as a decade-old collective effort rather than a package with a maintainer and a roadmap. The repository's own short description reflects the same crowd and less care: the summary field reads Python module for quantum chemistry and then continues into a list of personal names, Hermes, Kevin Koh, Peter Koval, Susi Lehtola, Zhendong Li, Junzi Liu, Narbe Mardirossian, and a truncated James D., which is an author list that ran into the description and stopped mid-name.
The version is parsed out of a source file at build time
Version numbering has one home and the build reads it by hand. The packaging metadata declares the version dynamic, and the setup script opens the package's own init file, scans its lines for the one that starts with the version attribute, splits on whichever quote character that line uses, and returns what sits between them.
def get_version():
topdir = os.path.abspath(os.path.join(__file__, '..'))
with open(os.path.join(topdir, 'pyscf', '__init__.py'), 'r') as f:
for line in f.readlines():
if line.startswith('__version__'):
delim = '"' if '"' in line else "'"
return line.split(delim)[1]
raise ValueError("Version string not found")A missing or reformatted version line stops the build with that ValueError rather than producing a package with a blank version. The same script still handles the platform naming for macOS, including universal2 builds and the host platform variable used by cibuildwheel, so it is doing real work and not just holding the version. It sits next to a modern packaging file rather than replacing it, and the build requirements cap CMake below version 4, which is the kind of ceiling that exists because a newer release dropped compatibility older builds rely on.
Numpy excludes two releases and psutil is Windows only
The core dependency list is five entries and each one says something.
dependencies = [
'numpy>=1.13,!=1.16,!=1.17',
'scipy>=1.6.0',
'h5py>=2.7',
'setuptools',
"psutil; sys_platform == 'win32'"
]The numpy floor is 1.13 with two specific releases excluded outright, which tells you those two broke something rather than being merely untested. The psutil marker is scoped to Windows, so a Windows install gets an extra package that a Linux or macOS install never sees. The Python requirement is 3.7 or newer while the classifier list runs from 3.7 through 3.14, so the declared support window is wide. The operating system classifiers tell a narrower story: POSIX, Unix, and MacOS are listed and Windows is not, even though the dependency list has a Windows branch in it. Neither the page nor the metadata offers a Windows installation route beyond that marker.
An agent-recipes directory sits at the root with no explanation
The repository root holds a directory named agent-recipes that the front page never mentions, and a second one named doc_legacy, which implies the current documentation lives somewhere else. The examples tree is the clearest map of what the framework actually covers, with one directory per method: coupled cluster, density fitting, density functional theory, configuration interaction, gapped equations, atomic orbital to molecular orbital transformations, gradients, basis sets, geometry optimisation, Hessians, local orbitals, many-body perturbation, molecular dynamics, periodic systems, and an mpi directory for distributed runs, plus a numbered 0-readme file, an advanced folder, a benchmark folder, and a ci folder. Configuration for four different tools sits side by side, a flake8 file, a ruff file, a yapf style file, and coverage settings. The page itself is pinned to the July stable line, 2.14.0, while the last commit on the default branch is dated 2026-09-28, so the front page and the branch have drifted apart by two months of work.
Editorial conclusion
Install from PyPI and you get a deliberately conservative core; the faster-moving parts arrive through pyscf-forge or through extras that pull in packages maintained elsewhere. Decide which of those you actually need before installing, because the geomopt extra alone brings three optimizers. Two things to check in any paper that uses it: PySCF does not implement density functionals itself, so the external library that evaluates them has to be cited too, and the base package wants its own ten-year retrospective paper cited alongside whatever method paper you used. Finally, the framework supports Python 3.7 upward and lists only POSIX, Unix, and MacOS as supported systems, so a Windows install is on you.
Frequently asked questions
How to install pyscf?
pip install pyscf gives the stable release. Extension modules install with pip install pyscf[all] for the whole set or pip install pyscf[dispersion] for one, and features developed in recent years are published separately as pyscf-forge.
what is pyscf
PySCF is the Python-based Simulations of Chemistry Framework, distributed on PyPI under the name pyscf and licensed Apache-2.0. The stable line on the page is 2.14.0, and the documentation lives at pyscf.org.
is pyscf open source
Yes, under the Apache License 2.0, with the license text at the repository root. One caveat the project states about itself: PySCF does not implement density functionals, it calls external libraries such as libxc or XCFun, and a paper using them has to cite those libraries as well.
How to install pyscf on windows
The packaging metadata requires Python 3.7 or newer and lists POSIX, Unix, and MacOS as the supported operating systems, with no Windows entry, although the dependency list does carry a Windows-only marker for psutil. The build script handles macOS universal2 builds and the cibuildwheel host platform.
Official sources
Add this badge to your README
If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.
[](https://hysenlabs.com/projects/pyscf-pyscf)Star history chart for your README
[](https://hysenlabs.com/projects/pyscf-pyscf)Was this useful?