# Downforce Airfoils: 9,977 RANS CFD Cases of Inverted-Camber NACA Airfoils for Motorsport Aerodynamics

## Facts

- Provider: Ratio Labs (Haitz Saez de Ocariz Borde, Flavio Savarino et al.)
- License: CC-BY-4.0 (https://huggingface.co/datasets/ratiolabs/downforce-airfoils)
- Size: 10,037 files
- Format: Parquet, HDF5 (in sampled files)
- Stored at: Hugging Face, https://huggingface.co/datasets/ratiolabs/downforce-airfoils
- Access: public, readable without an account
- Category: Physics
- Source: https://arxiv.org/abs/2609.38213
- Page: https://scigantic.com/archive/637c0c51-8a12-50d4-8a42-51345f916b86
- Keywords: Physics, CFD, Parquet, HDF5, Hugging Face

## About

Steady incompressible 2D RANS simulations of downforce-generating, negative-camber NACA 4-digit airfoils, released by Ratio Labs with the Airfoil2Vec paper (arXiv:2609.38213) to cover the road-vehicle and motorsport regime (rear and front wings) that aircraft-oriented airfoil datasets leave out. 907 valid geometries (m from 1 to 8, p from 1 to 8, thickness 4 to 18 percent, sharp trailing edge) are each run at 11 angles of attack from -10 to 0 degrees, giving 9,977 converged cases at Re = 4.66e6 and Mach 0.2 with OpenFOAM simpleFoam and the k-omega SST turbulence model on structured C-type meshes of about 276k cells. Each case is one HDF5 file holding the per-cell volume field (coordinates, pressure, velocity, temperature, Stanton number), extra turbulence quantities, a per-panel surface field (Cp, Cf, wall shear), a 1,999-point spline contour and integrated Cl, Cd and Cm. Splits are train (860 geometries), test (47 held-out NACA geometries) and non_naca (S1223, E423, LS(1)-0413, Clark Y and FX 63-137 for geometry extrapolation); metadata Parquet files index every case. About 420 GB; read with h5py and filter cases with Hugging Face datasets. CC-BY-4.0; cite Saez de Ocariz Borde et al. 2026.

## Load the data

```python
from datasets import load_dataset
ds = load_dataset("ratiolabs/downforce-airfoils")
print(ds)
```

## Data structure

File types in sampled files: Parquet, HDF5

Example file: `metadata_non_naca.parquet`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/637c0c51-8a12-50d4-8a42-51345f916b86
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/637c0c51-8a12-50d4-8a42-51345f916b86
- Over MCP: `get_schema_card` with id `637c0c51-8a12-50d4-8a42-51345f916b86` at https://api.scigantic.com/mcp

## Related topics

- [Parquet datasets](https://scigantic.com/topics/parquet)
- [HDF5 datasets](https://scigantic.com/topics/hdf5)
- [Physics datasets](https://scigantic.com/topics/physics)
- [CFD datasets](https://scigantic.com/topics/cfd)
