anon-closure-challenge-v2/closure-challenge-v2

Closure Challenge v2

Extended ML-RANS turbulence-modelling benchmark with composite scoring across 14 test cases.

This dataset is the lightweight integral-profile portion used by the scoring protocol — CSV reference data and k-ω SST baseline predictions sampled at the canonical evaluation points for each case. The full OpenFOAM cases (mesh, fields, run scripts) and DNS-native partitioned VTUs are hosted on a separate Hugging Face dataset, anon-closure-challenge-v2/closure-challenge-v2-cfd-cases.

NeurIPS 2026 Evaluations & Datasets track submission. Authors anonymised for double-blind review.

Code repository (scoring package, evaluation protocol, per-case attribution): https://anonymous.4open.science/r/closure-challenge-v2

Heavy CFD assets (full OpenFOAM cases, VTK volumes): anon-closure-challenge-v2/closure-challenge-v2-cfd-cases

What's in the test set

8 inherited v1 cases:

• 4 parametric periodic-hill geometries at Re=5600
• 3 square / rectangular duct configurations (AR=1, 3, 14.4 at Re_τ=180/360)
• NASA wall-mounted hump

6 new v2 cases:

Case	Reference type	Quantities of interest
NASA_2DZP	NASA TMR theory	C_f(x), u⁺(log y⁺)
NASA_2DN00	Ladson NASA TM 4074 + Gregory NPL R&M 3726	C_L(α), C_D(α), C_p(x/c), C_f(x/c)
NASA_ASJ	Bridges-Wernet ARN consensus PIV	U/U_jet centerline + 5 stations, ⟨u'v'⟩/U_jet² at 5 stations
ERCOFTAC_AhmedBody25	LDA wake + pressure taps (case082)	rear-surface C_p, integrated C_D vs canonical 0.285
NASA_FaithHill	PIV centerline + PSP + FISF	mean velocity, TKE, surface C_p, surface C_f
ERCOFTAC_WingBodyJunction	DNS 1-6	symmetry-plane velocity / TKE / R_xx + bottom-wall and wing-root C_p

Layout

data/
├── alpha_15_13929_4048/   v1 PHLL
├── alpha_15_13929_2024/
├── alpha_05_4071_4048/
├── alpha_05_4071_2024/
├── AR_1_Ret_360/          v1 DUCT
├── AR_3_Ret_360/
├── AR_14_Ret_180/
├── NASA_2DWMH/            v1 hump
├── NASA_2DZP/             v2 new
├── NASA_2DN00/
├── NASA_ASJ/
├── ERCOFTAC_AhmedBody25/
├── NASA_FaithHill/
├── ERCOFTAC_WingBodyJunction/
└── evaluation_points/     v1 case grids

All 14 are held-out test cases. This dataset contains no training data. The standardized training and validation sets (27 and 5 cases) are published in the companion heavy dataset under data/train/ and data/validation/: anon-closure-challenge-v2/closure-challenge-v2-cfd-cases.

For each v2-new case:

<case>/
├── baseline_komegasst/   k-ω SST baseline (.dat + .csv pairs)
├── highfidelity/         Reference data (.dat + .csv pairs)
└── plots/                Pre-rendered comparison figures

Reading the periodic-hill case names

The digits in alpha_<slope>_<Lx>_<Ly> are domain dimensions, not Reynolds numbers. alpha_05_4071_2024 has streamwise extent x = 0 .. 4.071 and wall-normal extent y = 0 .. 2.024; alpha_15_13929_4048 spans x = 0 .. 13.929 and y = 0 .. 4.048. The leading alpha value is the hill-slope parameter.

Quick start

import pandas as pd
df = pd.read_csv("data/NASA_2DZP/highfidelity/cf_as_function_of_x.csv")
print(df.columns.tolist())
# ['zone', 'x', 'skinfr', '5percenterror']

For end-to-end scoring, use the closure-challenge-v2 Python package and the notebooks/sample_eval.ipynb example, both available in the code repository.

License

This dataset aggregates material under two different licenses. See the LICENSE file at the repository root for the authoritative statement.

• CC-BY-4.0: the 12 non-ERCOFTAC cases (alpha_*, AR_*, NASA_*) and evaluation_points, derived from public-domain or CC-BY upstream sources (NASA Turbulence Modeling Resource, Vinuesa duct database, Xiao parametric periodic-hill database).
• CC-BY-NC-SA-4.0: ERCOFTAC_AhmedBody25 and ERCOFTAC_WingBodyJunction, curated extracts of ERCOFTAC kbwiki material (Classic Collection case082; DNS 1-6), inheriting the upstream non-commercial and share-alike conditions.
• Code: MIT (see LICENSE-CODE in the code repository).

Hugging Face has no per-directory license field, so this repository declares license: other and records the split in LICENSE. The configs block above groups the two license scopes for readers and machine consumers; it does not itself carry licensing force.

Per-case attribution is in SOURCES.md of the code repository.

Citation

@inproceedings{closure_challenge_v2_neurips26,
  title={The Closure Challenge: A Benchmark Task for Machine Learning in Turbulence Modeling},
  author={Anonymous},
  booktitle={NeurIPS Datasets and Benchmarks Track (under review)},
  year={2026}
}

Croissant metadata

A Croissant 1.0 + RAI metadata file is provided as croissant.json in this repository, suitable for the JoaquinVanschoren/croissant-checker validator.