diverge-flow 0.9.1

divERGe implements various ERG examples

DivERGe implements various ERG examples

DivERGe provides a versatile framework to set up (one,) two and three dimensional functional renormalization group (FRG/ERG) calculations under the static vertex approximation.

It implements three backends, the grid FRG, truncated unity FRG (TUFRG) and orbital space n-patch FRG.

For maximum performance, the code is written in C/C++ with extensions in CUDA (GPUs). It makes minimal use of other dependencies, only FFTW and LAPACK are required. MPI may be used if desired. DivERGe can be interfaced from C/C++ or python, with an existing python FFI wrapper. This wrapper is published in pypi, such that you can run

pip install diverge-flow

on a 64bit linux machine and directly use divERGe. For different architectures, compilation is additionally required (and putting the correct libdivERGe.so in your LD_LIBRARY_PATH). You can verify the .so file in use by calling diverge.info() from python. For any other language, you must write all the FFI wrappers yourself.

Documentation

https://frg.pages.rwth-aachen.de/diverge/

Download CPU release

Generic linux (amd64) builds (GLIBC>=2.17, this should be given almost anywhere to date) can be downloaded here. We recommend building from source for an optimized version on the HPC infrastructure to your availability.

Testing

We use a slightly modified version of Catch2 for testing. To check divERGe's health from python, run

import diverge
diverge.init(None, None)
diverge.run_tests()
diverge.finalize()

Citation

Please cite this paper when using divERGe for your work. You may use the following BibTex entry:

@Article{10.21468/SciPostPhysCodeb.26,
	title={{divERGe implements various Exact Renormalization Group examples}},
	author={Jonas B. Profe and Dante M. Kennes and Lennart Klebl},
	journal={SciPost Phys. Codebases},
	pages={26},
	year={2024},
	publisher={SciPost},
	doi={10.21468/SciPostPhysCodeb.26},
	url={https://scipost.org/10.21468/SciPostPhysCodeb.26},
}

License

divERGe is published under the GPLv3. The releases include differently licensed software (OpenBLAS, FFTW) in binary form.

Authors

Jonas B. Profe and Lennart Klebl, 2025.