dxtb - Fully Differentiable Extended Tight-Binding

dxtb - Fully Differentiable Extended Tight-Binding#

This project provides a PyTorch-based fully differentiable implementation of the semi-empirical extended tight-binding (xTB) methods.

Introduction#

The xTB methods (GFNn-xTB) are a series of semi-empirical quantum chemical methods that provide a good balance between accuracy and computational cost. For more details and the original Fortran implementation, check out the GitHub repository and the documentation.

With dxtb, we provide a re-implementation of the xTB methods in PyTorch, which allows for automatic differentiation and seamless integration into machine learning frameworks.

If you use dxtb in your research, please cite the following paper:

    1. Friede, C. Hölzer, S. Ehlert, S. Grimme, dxtb – An Efficient and Fully Differentiable Framework for Extended Tight-Binding, J. Chem. Phys., 2024, 161, 062501. (DOI: 10.1063/5.0216715)

BibTeX

@article{dxtb.2024,
  title = {dxtb -- An Efficient and Fully Differentiable Framework for Extended Tight-Binding},
  author = {Friede, Marvin and H{\"o}lzer, Christian and Ehlert, Sebastian and Grimme, Stefan},
  journal = {The Journal of Chemical Physics},
  volume = {161},
  number = {6},
  pages = {062501},
  year = {2024},
  month = {08},
  issn = {0021-9606},
  doi = {10.1063/5.0216715},
  url = {https://doi.org/10.1063/5.0216715},
}
Contents