Installation¶

PySRF is built with the meson-python backend and ships a Cython extension (_bsum) that accelerates the inner solver. The backend compiles this extension for you during install, so you need a working C compiler (gcc or clang) on your system.

Requirements¶

• Python 3.10 or newer (development targets 3.12.4).
• A C compiler for building the Cython extension.

PySRF is not yet published on PyPI, so install it from source.

Install from source¶

git clone https://github.com/florianmahner/pysrf.git
cd pysrf
pip install .

Or install directly from GitHub without cloning first:

pip install "git+https://github.com/florianmahner/pysrf.git"

In both cases the meson-python backend compiles the _bsum Cython extension automatically. The runtime dependencies (numpy, scipy, scikit-learn, pandas, joblib, tqdm) are installed for you.

Editable install for development¶

For an editable install that compiles the extension into a separate build directory and keeps the source tree clean:

git clone https://github.com/florianmahner/pysrf.git
cd pysrf
pip install -e . --no-build-isolation

See the Development page for the full developer workflow, test commands, and contribution guidelines.

Verify the installation¶

import pysrf

print(pysrf.__version__)

To check whether the fast Cython backend is active:

from pysrf._bsum import BACKEND

print(BACKEND)  # 'cython' if compiled, 'python' otherwise

Performance: multi-threaded BLAS¶

The SRF solver relies on BLAS routines (via scipy) that benefit from multi-threading. Set OMP_NUM_THREADS before importing pysrf to enable parallel linear algebra:

export OMP_NUM_THREADS=4
python my_script.py

Or at the top of a script, before any imports:

import os

os.environ["OMP_NUM_THREADS"] = "4"

from pysrf import SRF

With verbose=1, SRF logs the active thread count at fit time.

Next steps¶

Head to the Quick start to fit your first SRF model.