Hugoniot

Variational free energy method source code for hydrogen Hugoniot

Features

• Pretrain and train models for warm dense hydrogen or deuterium
• Compute equation of state for hydrogen in warm dense matter region
• Visualize results and compute Hugoniot curves

Installation

Clone this repository and install the dependencies:

git clone https://github.com/fermiflow/Hugoniot.git
cd Hugoniot
pip install -r requirements.txt

Install hqc

cd hqc
pip install -e .

Install cfgmanager

cd cfgmanager
pip install -e .

Then you can train your own hydrogen model.

Usage

Pretrain

• Use pretrainflow.py to pretrain nucleus flow model.
• Change config_path and config_name in pretrainflow.py l.31 to change the config file to load.
• Change the parameters in config file, like conf/pretrain/flow/config.yaml, or override values in the loaded config from the command line to pretrain your model (we use hydra to manage our config files):

python pretrainflow.py num=16 rs=1.86 T=10000 batchsize=256

• Use folder parameter to specify the directory to save data and checkpoint files.

Train

• Use main.py to train all of the three neural networks.
• Change config_path and config_name in main.py l.36 to change the config file to load.
• Change the parameters in config file, like conf/train/config.yaml, or override values in the loaded config from the command line to pretrain your model.:

python main.py \
      num=16 \
      rs=1.86 \
      T=10000 \
      batchsize=256 \
      load_pretrain.flow=/your/pretrain/checkpoint/path/epoch_001000.pkl

• Use load_pretrain.flow parameter to specify the pretrained flow to load (if None, it will train from uniform distribution).
• Use load parameter to specify the checkpoint file to load and continue training.
• Use folder parameter to specify the directory to save data and checkpoint files.

Data

Dataset including equation of states, training parameters, network checkpoints, nucleus and electron snapshots are published on Hugging Face

Citation

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

@article{8zn5-6dnt,
  title = {Deep variational free energy calculation of hydrogen Hugoniot},
  author = {Li, Zihang, Xie, Hao, Dong, Xinyang, and Wang, Lei},
  journal = {Phys. Rev. Lett.},
  pages = {--},
  year = {2026},
  month = {Jan},
  publisher = {American Physical Society},
  doi = {10.1103/8zn5-6dnt},
  url = {https://link.aps.org/doi/10.1103/8zn5-6dnt}
}

Contributing

Contributions are welcome! 🎉

If you’d like to contribute, open a Pull Request (PR).

License

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