GreenLearning is a deep learning library based on Tensorflow for learning Green's functions associated with partial differential operators.

Additional datasets are available on Zenodo.

Exact and learned Green’s function of the Laplace operator..

Below is an example of the Green's function of a second-order differential operator with variable coefficients learned by GreenLearning.

Learned Green’s function of a second order ODE with variable coefficients.

See https://greenlearning.readthedocs.io/en/latest/guide/gallery.html for more examples.

The library is maintained by Nicolas Boullé. If you are interested in using it, do not hesitate to get in contact with him at nb690@cam.ac.uk.

Documentation: ReadTheDocs

• GreenLearning learns Green's functions and homogeneous solutions associated with scalar and systems of linearized partial differential equations in 1D and 2D with deep learning.
• Rational neural networks are implemented and used to increase the accuracy of the learned Green's functions.
• GreenLearning requires no hyperparameter tuning to successfully learn Green's functions.
• The neural networks can be created and trained easily with a few lines of code.
• It is simple to generate the training datasets with MATLAB scripts.

GreenLearning relies on the following Python libraries:

• TensorFlow >= 1.15.0
• Matplotlib
• NumPy
• SciPy

• For users, you can install the stable version with pip:

  pip install greenlearning
  

  or with conda:

  conda install -c conda-forge greenlearning
  

• For developers, you should clone the GitHub repository and install it manually on your machine::

  git clone https://github.com/NBoulle/greenlearning.git
  cd greenlearning
  pip install -e.
  

Please cite the following papers if you are using GreenLearning.

• About GreenLearning:

  @article{boulle2022data,
  title={Data-driven discovery of Green's functions with human-understandable deep learning},
  author={Boull{\'e}, Nicolas and Earls, Christopher J. and Townsend, Alex},
  journal={Scientific Reports},
  volume={12},
  pages={4824},
  year={2022},
  doi={10.1038/s41598-022-08745-5}
  }
  

• About Rational neural networks:

  @inproceedings{boulle2020rational,
  title={Rational neural networks},
  author={Boull{\'e}, Nicolas and Nakatsukasa, Yuji and Townsend, Alex},
  booktitle = {Advances in Neural Information Processing Systems},
  volume = {33},
  pages = {14243--14253},
  year={2020},
  url = {https://proceedings.neurips.cc/paper/2020/file/a3f390d88e4c41f2747bfa2f1b5f87db-Paper.pdf}
  }