NeuralConvection 3D: Boussinesq-to-Compressible Natural Convection in a Cube for Fourier Neural Operators (Menglik, Shao, Hyde)
- Provider
- Nurshat Menglik, Alex Shao, David Hyde
- License
- CC-BY-4.0 (source)
- Size
- 24 files
- Format
- CSV (in sampled files)
- Stored at
- Hugging Face (https://huggingface.co/datasets/NurshatMenglik/NeuralConvection_3D_TrainTest)
- Access
- Free and public. No account with the data provider is needed.
- Category
- Physics
- Source
- https://arxiv.org/abs/2606.25259
About this dataset
Training and test tensors for the 3D Fourier Neural Operator (FNO) surrogate in "A Neural Surrogate Approach for Simulating Natural Convection Problems" (Menglik, Shao and Hyde, arXiv:2606.25259). Each pair maps a steady Boussinesq natural-convection (buoyancy-driven) field in the unit cube to the fully compressible field for the same randomized wall-temperature problem, both solved on one 24-cubed mesh. X.pt and Y.pt are float32 PyTorch tensors of shape (N, 5, 24, 24, 24) with channels u_x, u_y, u_z, temperature and pressure, axes (x, y, z) and gravity along -y. Splits: train (1,000 samples with 2 to 5 Voronoi cells of wall temperature per active wall, with a 900/100 train/validation split file), test_in_distribution (50), test_ood (50, 6 to 9 Voronoi cells per wall) and test_legacy (50 classical hot/cold side-wall cubes). Rayleigh number runs from 1e2 to 1e5, the range where this grid reproduces reference Nusselt numbers within one percent, with Prandtl number 0.71; both members share buoyancy (free-fall) units. per_sample.csv and MANIFEST.json record parameters and conventions; flagged_samples.json lists 24 low-Ra pairs with solver drift. Load with torch.load. Released under CC BY 4.0; cite arXiv:2606.25259.
How to load this dataset in Python
import pandas as pd
df = pd.read_csv("hf://datasets/NurshatMenglik/NeuralConvection_3D_TrainTest/test_in_distribution/per_sample.csv")
df.head()Data structure
File types in sampled files: CSV
Files
Example file: test_in_distribution/per_sample.csv
Analyze this dataset without downloading it
On Scigantic, NeuralConvection 3D: Boussinesq-to-Compressible Natural Convection in a Cube for Fourier Neural Operators (Menglik, Shao, Hyde) opens in a hosted JupyterLab notebook with the data already mounted, so you can read files where they are stored instead of copying them first. Each notebook includes an AI assistant that can see the mounted files and write and run analysis code. Open the dataset page to launch a notebook, or browse the full catalog.
This page is also available as Markdown and as JSON.
Related Physics datasets
- TIDE Strongly Rotating Turbulence (Coriolis Force): 3D Incompressible Navier-Stokes DNS at 256^3 (Dai et al.)
- TIDE Passive Scalar Mixing in Forced Turbulence: 3D Incompressible Navier-Stokes DNS at 256^3 (Dai et al.)
- WAKESET: 1,091 High-Reynolds-Number RANS Simulations of Turbulent Wakes behind an XLUUV
- The Well: PlanetSWE, Forced Rotating Shallow Water on a Sphere
- JetClass-II: 188-Class Large-Radius Jet Dataset for Jet Foundation Models (LHC Delphes Simulation)
- PDEBench 1D Compressible Navier-Stokes Solutions
- The Well: Turbulent Radiative Mixing Layer (3D)
- TIDE Freely Decaying Turbulence (Saffman Initial State): 3D Incompressible Navier-Stokes DNS at 256^3 (Dai et al.)