This repository includes the codes to produce datasets and implement the DDP, DSMAG, and TL referenced in the

accompanying paper Data-driven subgrid-scale modeling of forced Burgers turbulence using deep learning with generalization to higher Reynolds numbers via transfer learning (https://doi.org/10.1063/5.0040286). The following links to a dataset that can be used with the given DSMAG and DDP codes, https://zenodo.org/record/4316338.

This code creates a DNS dataset. Parameters like the Reynolds number and resolution can be altered easily to create datasets to experiment with transfer learning.

This codes generate filtered and coarse grained variables for the training and a posteriori testing of DDP.

This code contains a function to take in the DNS dataset and then calculate the filtered variables and subgrid Pi terms.

This code contains a function to apply the box filter.

This code is an implementation of the Dynamic Smagorinsky LES.

This code trains and runs a posteriori prediction for DDP.

This code takes in a set of weights for the ANN used in DDP and retrains it for a different training regime.

Read more on [arXiv]

Read more on [PoF]

@article{subel2020data,
  title={Data-driven subgrid-scale modeling of forced Burgers turbulence using deep learning with generalization to higher Reynolds numbers via transfer learning},
  author={Subel, Adam and Chattopadhyay, Ashesh and Guan, Yifei and Hassanzadeh, Pedram},
  journal={arXiv e-prints},
  pages={arXiv--2012},
  year={2020}
}

@article{subel2021data,
  title={Data-driven subgrid-scale modeling of forced Burgers turbulence using deep learning with generalization to higher Reynolds numbers via transfer learning},
  author={Subel, Adam and Chattopadhyay, Ashesh and Guan, Yifei and Hassanzadeh, Pedram},
  journal={Physics of Fluids},
  volume={33},
  number={3},
  pages={031702},
  year={2021},
  publisher={AIP Publishing LLC}
}