polymathic-ai/shear_flow

How To Load from HuggingFace Hub

1. Be sure to have the_well installed (pip install the_well)
2. Use the WellDataModule to retrieve data as follows:

from the_well.data import WellDataModule

# The following line may take a couple of minutes to instantiate the datamodule
datamodule = WellDataModule(
    "hf://datasets/polymathic-ai/",
    "shear_flow",
)
train_dataloader = datamodule.train_dataloader()

for batch in dataloader:
    # Process training batch
    ...

Periodic shear flow

Warning: this Hugging Face copy is v1.0 of the dataset, which has an error in its split. About 80% of its validation and test trajectories are exact copies of training trajectories (same PDE parameters and same initial condition), so validation and test results obtained on it are not reliable.

A corrected version (v2.0), with train, validation and test built from disjoint initial conditions, is available from Flatiron with:

the-well-download --base-path path/to/base --dataset shear_flow

If you already downloaded shear_flow before October 2026, delete your local copy first: the download resumes existing files and would otherwise keep the old ones. This Hugging Face copy will be updated to v2.0 soon.

One line description of the data: 2D periodic incompressible shear flow.

Longer description of the data: Shear flow are a type of fluid characterized by the continuous deformation of adjacent fluid layers sliding past each other with different velocities. This phenomenon is commonly observed in various natural and engineered systems, such as rivers, atmospheric boundary layers, and industrial processes involving fluid transport. The dataset explores a 2D periodic shearflow governed by incompressible Navier-Stokes equation.

Associated paper: Paper 1, Paper 2, Paper 3.

Data generated by: Rudy Morel, CCM, Flatiron Institute.

Code or software used to generate the data: Github repository, based on the software Dedalus.

Equation:

While we solve equations in the frequency domain, the original time-domain problem is

where $\mathrm{\Delta }=\mathrm{\nabla }\cdot \mathrm{\nabla }\backslash Delta\; =\; \backslash nabla\; \backslash cdot\; \backslash nabla$ is the spatial Laplacian, $u=(u_x,u_y)u\; =\; (u\_x,u\_y)$ is the (horizontal and vertical) velocity, $ss$ is the tracer, and $pp$ is the pressure, with the additional constraints $\int p=0\backslash int\; p\; =\; 0$ (pressure gauge).

These PDEs are parameterized by the Reynolds and Schmidt numbers through $\nu \backslash nu$ and $DD$.

Dataset	FNO	TFNO	Unet	CNextU-net
shear_flow	0.1567	$\mathbf{0.1348}\backslash mathbf\{0.1348\}$	0.5910	0.2037

Table: VRMSE metrics on test sets (lower is better). Best results are shown in bold. VRMSE is scaled such that predicting the mean value of the target field results in a score of 1.

About the data

Dimension of discretized data: 200 time-steps of 128 $\times \backslash times$ 256 images.

Fields available in the data: tracer (scalar field), pressure (scalar field), velocity (vector field).

Number of simulations: 1120 (28 PDE parameters $\times \backslash times$ 40 initial conditions).

Size of the ensemble of all simulations: 114.7 GB.

Grid type: uniform, cartesian coordinates.

Initial conditions: the shear field $u_{1}xu\_1x$ is composed of $n_{\text{shear}}n\_\backslash text\{shear\}$ shears uniformly spaced along the $zz$ direction. Each shear is implemented with a tanh (hyperbolic tangent) $\text{tanh}(5\frac{y-y_k}{n_{\text{shear}}w})\backslash text\{tanh\}(5\backslash frac\{y-y\_k\}\{n\_\backslash text\{shear\}w\})$ where $z_{k}z\_k$ is the vertical position of the shear and $ww$ is a width factor. The velocity field $u_{y}u\_y$ is composed of sinusoids along the $xx$ direction located at the shear. These sinusoids have an exponential decay away from the shear in the $yy$ direction $\text{sin}(n_{\text{blobs}}\pi x)e^{\frac{25}{w^2}\mathrm{\mid }y-y_k{\mathrm{\mid }}^2}\backslash text\{sin\}(n\_\backslash text\{blobs\}\backslash pi\; x)\backslash ,e^\{\backslash frac\{25\}\{w^2\}|y-y\_k|^2\}$. The tracer matches the shear at initialization. The pressure is initialized to zero. The initial condition is thus indexed by $n_{\text{shear}},n_{\text{blobs}},wn\_\backslash text\{shear\},n\_\backslash text\{blobs\},w$.

Boundary conditions: periodic.

Simulation time-step: 0.1 (simulation time unit).

Total time range ( $t_{min}t\_\{min\}$ to $t_{max}t\_\{max\}$): $t_{min}=0t\_\{min\}\; =\; 0$, $t_{max}=20t\_\{max\}\; =\; 20$.

Spatial domain size: $0\le x\le 10\backslash leq\; x\; \backslash leq\; 1$ horizontally, and $-1\le y\le 1-1\; \backslash leq\; y\; \backslash leq\; 1$ vertically.

Set of coefficients or non-dimensional parameters evaluated: $\text{Reynolds}\in [1e4,5e4,1e5,5e5],\text{Schmidt}\in [0.1,0.2,0.5,1.0,2.0,5.0,10.0]\backslash text\{Reynolds\}\backslash in[1e4,\; 5e4,\; 1e5,\; 5e5],\; \backslash text\{Schmidt\}\backslash in[0.1,\; 0.2,\; 0.5,\; 1.0,\; 2.0,\; 5.0,\; 10.0]$. For initial conditions $n_{\text{shear}}\in [2,4]n\_\backslash text\{shear\}\backslash in[2,4]$, $n_{\text{blobs}}\in [2,3,4,5]n\_\backslash text\{blobs\}\backslash in[2,3,4,5]$, $w\in [0.25,0.5,1.0,2.0,4.0]w\backslash in[0.25,\; 0.5,\; 1.0,\; 2.0,\; 4.0]$.

Approximate time to generate the data: per input parameter: $\sim 1500s\backslash sim\; 1500s$, total: $\sim 5\backslash sim\; 5$ hours.

Hardware used to generate the data and precision used for generating the data: 7 nodes of 64 CPU cores each with 32 tasks running in parallel on each node, in single precision.

What is interesting and challenging about the data:

Shear flow are non-linear phenomena arrising in fluid mechanics and turbulence. Predicting the behavior of the shear flow under different Reynolds and Schmidt numbers is essential for a number of applications in aerodynamics, automotive, biomedical. Furthermore, such flow are unstable at large Reynolds number.

Please cite these associated papers if you use this data in your research:

@article{burns2020dedalus,
  title={Dedalus: A flexible framework for numerical simulations with spectral methods},
  author={Burns, Keaton J and Vasil, Geoffrey M and Oishi, Jeffrey S and Lecoanet, Daniel and Brown, Benjamin P},
  journal={Physical Review Research},
  volume={2},
  number={2},
  pages={023068},
  year={2020},
  publisher={APS}
}