NurshatMenglik/Paired_Boussinesq_Compressible_Dataset

Paired Boussinesq / Compressible Natural Convection — 10,000 Simulations

📄 Paper: A Neural Surrogate Approach for Simulating Natural Convection Problems (arXiv:2606.25259) — Nurshat Menglik, Alex Shao, David Hyde.

10,000 matched pairs of 2D natural-convection simulations of the differentially heated square cavity. Every sample solves the same randomized problem twice — once with the cheaper Boussinesq model and once with the fully compressible model — so the two fields differ only in the physics, not in the geometry, boundary data, or discretization. This is the raw finite-element output of the data-generation pipeline used in the paper.

Both solves use a monolithic, fully implicit mixed finite element method (Taylor–Hood: [P2]^2 velocity, P1 scalars) on the unit square, implemented in FEniCSx/DOLFINx.

Contents

10,000 sample folders, sample_000000 … sample_009999 (~52 GB total). Each folder contains:

File	Description
meta.json	Generation parameters for this sample (see below)
T_bouss.npy, T_comp.npy	Temperature (P1 nodal values)
ux_bouss.npy, uy_bouss.npy	Boussinesq velocity components (P2 nodal values)
ux_comp.npy, uy_comp.npy	Compressible velocity components (P2 nodal values)
p_bouss.npy, p_comp.npy	Pressure, min–max normalized to [0, 1]
rho_comp.npy	Density (compressible solve only)
coords_*.npy	(n, 2) node coordinates matching each field above
paired_fields_2x4.png	2×4 visualization of the Boussinesq/compressible pair

All arrays are float64 and unstructured: they are flat finite-element nodal value arrays, not images. Node counts vary per sample because the mesh resolution is swept across the dataset (res in meta.json), and velocity lives on a richer P2 space than the P1 scalars — so always pair a field with its own coords_*.npy file rather than assuming a common grid.

meta.json records seed, Ra, Ma, res, dt, time, length, T_init, dT, wall_modes, and the per-wall amplitude fractions left_eps, right_eps, top_eps, bot_eps.

Boundary conditions

Each of the four walls carries a random Dirichlet temperature built from a truncated Fourier series (wall_modes modes, per-wall amplitude fractions), clamped into an admissible band around T_init. No adiabatic walls.

Loading

import numpy as np, json

T   = np.load("sample_000000/T_bouss.npy")        # (n,)
xy  = np.load("sample_000000/coords_T_bouss.npy") # (n, 2)
meta = json.load(open("sample_000000/meta.json"))

# interpolate onto a regular grid for ML use
from scipy.interpolate import griddata
gx, gy = np.meshgrid(np.linspace(0, 1, 64), np.linspace(0, 1, 64))
T_grid = griddata(xy, T, (gx, gy), method="linear")

If you want ready-to-train tensors on a uniform grid instead of raw FE output, use NeuralConvection_2D_TrainTest and NeuralConvection_3D_TrainTest.

Source

Generated with DataGeneration/genData.py (batch driver DataGeneration/generate_dataset.sh) from https://github.com/Nurshat317/NeuralConvection

License

Released under CC BY 4.0. You are free to share and adapt this data, including commercially, provided you give appropriate credit and indicate any changes. The requested form of credit is a citation to the paper below.

Citation

If you use this dataset in your research, please cite:

@article{menglik2026neural,
  title         = {A Neural Surrogate Approach for Simulating Natural Convection Problems},
  author        = {Menglik, Nurshat and Shao, Alex and Hyde, David},
  journal       = {arXiv preprint arXiv:2606.25259},
  year          = {2026},
  eprint        = {2606.25259},
  archivePrefix = {arXiv},
  primaryClass  = {physics.comp-ph},
  url           = {https://arxiv.org/abs/2606.25259}
}

• Paper: arXiv:2606.25259
• Code: github.com/Nurshat317/NeuralConvection