scaomath/navier-stokes-dataset

Navier Stokes Dataset of Isotropic Turbulence in a periodic box

<!-- Provide a quick summary of the dataset. -->

The dataset for tensor-to-tensor or trajectory-to-trajectory neural operators, generated from Navier-Stokes equations to model the isotropic turbulence [1] such that the spectra satisfy the inverse cascade discovered by A.N. Kolmogorov [2].

[1]: McWilliams, J. C. (1984). The emergence of isolated coherent vortices in turbulent flow. Journal of Fluid Mechanics, 146, 21-43. [2]: Kolmogorov, A. N. (1941). The local structure of turbulence in incompressible viscous fluid for very large Reynolds Numbers. Dokl. Akad. Nauk SSSR, 30, 301.

Dataset Details

Dataset Description

<!-- Provide a longer summary of what this dataset is. -->

The dataset contains several cases of isotropic turbulence modeled by Navier-Stokes equations. The data are generated either by a pseudo-spectral solver with 4th-order Runge-Kutta for the convection term, or a higher order Finite Volume IMEX methods. The different initial conditions have different peak wavenumbers of O(1), and eventually their spectra all converge to the Kolmogorov inverse cascade.

• —Curated by: S. Cao
• —Funded by National Science Foundation: NSF award DMS-2309778
• —License: MIT license

Dataset Sources [optional]

<!-- Provide the basic links for the dataset. -->

• —Repository: https://github.com/scaomath/torch-cfd
• —Paper: [More Information Needed]
• —Demo:
• —The classical Kolmogorov inverse cascade with a solenoidal forcing and small drag.
• —The fast training using the data with a small number of vortices.
• —The fast converging to the inverse cascade.
• —Note: Due to the change of how torch.load() treat serialization, the data would not be loaded directly by calling torch.load. If you plan to use it, you can add the following code after the imports:

     torch.serialization.add_safe_globals([defaultdict])
     torch.serialization.add_safe_globals([list])

Dataset Structure

<!-- This section provides a description of the dataset fields, and additional information about the dataset structure such as criteria used to create the splits, relationships between data points, etc. -->

Each individual chunk of data is pickled in single-file format.

Dataset Creation

TO-DO

Citation

<!-- If there is a paper or blog post introducing the dataset, the APA and Bibtex information for that should go in this section. -->

@article{2024SpectralRefiner,
  title={Spectral-Refiner: Fine-Tuning of Accurate Spatiotemporal Neural Operator for Turbulent Flows},
  author={Shuhao Cao and Francesco Brarda and Ruipeng Li and Yuanzhe Xi},
  journal={arXiv preprint arXiv:2405.17211},
  year={2024},
  primaryClass={cs.LG}
}