中文/EN

XequiNet

XequiNet is a package of equivariant graph neural network for predicting properties of chemical molecules or periodical systems.

Install environments

git clone https://github.com/X1X1010/XequiNet.git
cd XequiNet
conda env create -f environment.yaml -n <env_name>

If the automatic installation fails, or if you want to install packages of other versions, you can install it manually, mainly for the following packages.

• PyTorch: Greatness speaks for itself.
• PyG: For constructing graph neural networks.
• torch-cluster: For constructing edge_index with cutoff radius.
• torch-scatter: For index selecting operations in GNN structure.
• e3nn: For constructing equivariant modules.
• pytorch-warmup: For convenient learning rate warmup.
• OmegaConf: A nice package for parsing configuration files
• LMDB: Conda / PyPI. Dataset format. Be careful with installation by conda.
• ASE: A nice package for atomistic simulation. We mainly use for read input files with various format.
• PySCF: For handling some quantum chemical computation.
• geomeTRIC: Geometry optimization engine used by PySCF.
• TBlite: Python interface of xTB. Note that there are two packages, tblite and tblite-python. Both are needed.
• tabulate: For elegant printouts.

Extra packages

• cuda-toolkit: Nvidia CUDA Toolkit. Please be super careful with the cuda version.
• i-Pi: For PIMD simulation.

Setups

Once the requirements are installed, running

conda activate <env_name>
pip install -e .

Usage

For detailed documentation, please refer to the XequiNet User Manual. We apologize that the documentation is currently only available in Chinese.

Simple documents can be viewed at docs for details.

Model availability

You can download our trained models through our Zenodo page.

Citation

@article{doi:10.1021/acs.jctc.4c01151,
    author = {Chen, Yicheng and Yan, Wenjie and Wang, Zhanfeng and Wu, Jianming and Xu, Xin},
    title = {Constructing Accurate and Efficient General-Purpose Atomistic Machine Learning Model with Transferable Accuracy for Quantum Chemistry},
    journal = {J. Chem. Theory Comput.},
    volume = {20},
    number = {21},
    pages = {9500-9511},
    year = {2024},
    doi = {10.1021/acs.jctc.4c01151},
}