lettuce: A Python repository from rawbby

GPU-accelerated Lattice Boltzmann Simulations in Python

Lettuce is a Computational Fluid Dynamics framework based on the lattice Boltzmann method (LBM).

GPU-Accelerated Computation: Utilizes PyTorch for high performance and efficient GPU utilization.
Rapid Prototyping: Supports both 2D and 3D simulations for quick and reliable analysis.
Advanced Techniques: Integrates neural networks and automatic differentiation to enhance LBM.
Optimized Performance: Includes custom PyTorch extensions for native CUDA kernels.

Resources

Documentation
Presentation at CFDML2021 - Paper | Preprint | Slides | Video | Code

Getting Started

To find some very simple examples of how to use lettuce, please have a look at the examples. These will guide you through lettuce's main features. Please ensure you have Jupyter installed to run the Jupyter notebooks.

The simplest example is:

https://github.com/lettucecfd/lettuce/blob/ef9830b59c6ad50e02fdf0ce24cc47ea520aa354/examples/00_simplest_TGV.py#L6-L21

Installation

Install the anaconda or miniconda package manager from https://www.anaconda.org

Clone this repository from github and change to it:

git clone https://github.com/lettucecfd/lettuce
cd lettuce

Update conda, create a new conda environment and activate it:

conda update conda
conda create -n lettuce
conda activate lettuce

Follow the recommendations at https://pytorch.org/get-started/locally/ to install pytorch based on your GPU's CUDA version. You may need to install the nvidia toolkit. You may follow the instructions at https://developer.nvidia.com/cuda-downloads. You may need to check the compatibility of your NVIDIA driver with the desired CUDA version: https://docs.nvidia.com/deploy/cuda-compatibility/. To get your CUDA version, run:
```
nvcc --version
```

For CUDA 12.1 (if supported by your GPU) use:

conda install pytorch pytorch-cuda=12.1 -c pytorch -c nvidia

Install the remaining dependencies. _This will not include CUDA-support unless you installed pytorch-cuda!_:
```
conda activate lettuce
conda install --file requirements.txt -c pytorch -c nvidia -c conda-forge
```
NOTE: Sometimes, the installation does not properly set the CUDA_HOME variable. In this case, you may install cudatoolkit and set CUDA_HOME to the package directory (it should containa /bin/nvcc/ subdirectory!).
If you want to only use lettuce, run the installation (from the lettuce base directory!). Ideally, use [PEP 517](https://peps.python.org/pep-0517/):
```
pip install --use-pep517 .
```
If you are a developer, add the changeable-installation-flag (-e):
```
pip install --use-pep517 -e .
```
Check out the convergence order, running on CPU:
```
lettuce --no-cuda convergence
```
Run the test cases:
```
pytest tests
```
For running a CUDA-driven LBM simulation on one GPU omit the --no-cuda. If CUDA is not found, make sure that cuda drivers are installed and compatible with the installed cudatoolkit (see conda install command above).
Check out the performance, running on GPU:
```
lettuce benchmark
```

Citation

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

@inproceedings{bedrunka2021lettuce,
  title={Lettuce: PyTorch-Based Lattice Boltzmann Framework},
  author={Bedrunka, Mario Christopher and Wilde, Dominik and Kliemank, Martin and Reith, Dirk and Foysi, Holger and Kr{\"a}mer, Andreas},
  booktitle={High Performance Computing: ISC High Performance Digital 2021 International Workshops, Frankfurt am Main, Germany, June 24--July 2, 2021, Revised Selected Papers},
  pages={40},
  organization={Springer Nature}
}

Credits

We use the following third-party packages:

This package was created with Cookiecutter and the audreyr/cookiecutter-pypackage project template.

License

Free software: MIT license, as found in the LICENSE file.

rawbby/lettuce