ML-FSM

This repository provides an implementation of the Freezing String Method (FSM) for double-ended transition state searches with internal coordinates interpolation with ML-based potentials.

Installation

Install via PyPI

To install the latest stable release:

pip install mlfsm

Install from source (development version)

If you want the latest development version:

git clone https://github.com/thegomeslab/ML-FSM.git
cd ML-FSM
pip install .

Perform a test run with default parameters and the EMT calculator:

python examples/fsm_example.py data/sharada/06_diels_alder/ --calculator emt

Note the EMT calculator should only be used for quick demonstrations/tests.

Tutorials

1. A Google Colab notebook example is available at:

examples/FSM_Colab_AIMNet2.ipynb

This notebook demonstrates how to run a full FSM calculation on a Diels–Alder reaction using the AIMNet2 potential.

3. Example Script

The script:

examples/fsm_example.py

is a comprehensive and flexible tool that supports most ML-FSM functionality.

Note: Users are responsible for installing their desired quantum chemistry backend, current calculators supported in fsm_example.py are AIMNet2, MACEOFF23, FAIR UMA, TensorNet, xTB, QChem.

3. Some ML potentials do not offer a native ASE interface. To accommodate these, a minimal example of custom ase.Calculator wrappers is provided in the examples/ folder. This serves as templates for integrating arbitrary NNPs with ML-FSM

Usage

For projects referencing algorithmic improvements to the FSM please cite:

Marks, J., & Gomes, J. (2025). Incorporation of Internal Coordinates Interpolation into the Freezing String Method. Journal of Chemical Theory and Computation. https://doi.org/10.1021/acs.jctc.5c01492

For projects using the FSM with ML-based potentials please cite:

Marks, J., & Gomes, J. (2025). Efficient Transition State Searches by Freezing String Method with Graph Neural Network Potentials. http://arxiv.org/abs/2501.06159

Contributing

Contributions are welcome! Please read the Contributing Guide before opening a pull request. We ask that you open an issue or reach out to the maintainer first to discuss proposed changes.

See the Changelog for a history of notable changes.

License

This project is licensed under the MIT License. See the LICENSE file for details.

Third-Party Licenses and Attribution

This project depends on several third-party open-source Python packages. These dependencies are not bundled with this repository and must be installed separately by the user.

Below is a list of direct dependencies and their respective licenses:

These licenses are all compatible with the MIT license under which this project is distributed. Please refer to each package’s own repository for the full license text.

Credits

This package was created with Cookiecutter and the jevandezande/pixi-cookiecutter project template.