symm-flows

This repository implements symmetry-aware invertible residual networks (iResNets) for calculating vibrational spectra of molecules, as described in the associated publication.

Installation

1. Clone the repository:

   git clone https://github.com/robochimps/symm-flows.git
   cd symm-flows

2. Build the environment (recommended: use conda):

    conda env create -f environment.yml

3. Activate the environment:

    conda activate flows-symm

4. Install the pyhami dependency:

    pip install external_packages/pyhami-0.0.1-py3-none-any.whl

Usage

To compute the vibrational spectra of NH$_3$ run

python nh3.py {n}

from the repository root directory, where - n=0 initializes the parameters of the normalizing flow randomly and - n=1 loads previously installed parameters.

Results for H_2CO can be obtained analogously by running

python h2co.py {n}

Both scripts produce results using a standard iResNet and a symmetry-aware iResNet, allowing for direct comparison between the two approaches.

Citation

If you use this code in your research, please cite:

E. Vogt and Á. F. Corral, Symmtery-aware normalizing flows for computing vibrational spectra of molecules (submitted). arXiv: eprint: xxx.yyy (2026) DOI:10.48550/XXX

@Misc{Vogt:arXivXXX:YYY,
  title        = {Symmtery-aware normalizing flows for computing
>  vibrational spectra of molecules},
  author       = {Vogt, Emil and Corral Fernández, Álvaro and Saleh, Yahya},
  year         = 2026,
  archivePrefix= {arXiv},
  eprint       = {XXX.YYYYY},
  primaryClass = {phys-chem},
  howpublished = {preprint},
}

Contact

For questions or feedback, feel free to open an issue or reach out to the authors directly via emil.vogt@cfel.de.