{"success":true,"data":{"author":"Nurshat Menglik, Alex Shao, David Hyde","bucketAccessType":"public","bucketName":"NurshatMenglik/Paired_Boussinesq_Compressible_Dataset","bucketPrefix":"","category":"Physics","cloudProvider":"huggingface","createdAt":"2026-11-21T20:46:01.170Z","description":"Ten thousand matched pairs of steady 2D natural-convection (buoyancy-driven flow, hot and cold walls) simulations in the differentially heated unit-square cavity, released with \"A Neural Surrogate Approach for Simulating Natural Convection Problems\" (Menglik, Shao and Hyde, arXiv:2606.25259). Each sample solves one randomized wall-temperature problem twice, once with the Boussinesq approximation and once with the fully compressible model, so the pair differs only in the physics. Both solves use a monolithic, fully implicit Taylor-Hood mixed finite element method (P2 velocity, P1 scalars) in FEniCSx/DOLFINx. Mesh resolution is swept in four blocks of 2,500 samples (32, 64, 128 and 256 squared). Each sample folder (sample_000000 to sample_009999, about 112 GB) holds meta.json with generation parameters and steady-state records, NumPy arrays of temperature, velocity components, raw and normalized pressure and density, and matching node coordinates; arrays are flat float64 finite-element nodal values, not images, and manifest.csv indexes the set. This is version 2 (September 2026); the earlier release is kept under the tag v1.0-legacy. Load with numpy. Use it for multi-fidelity learning and Boussinesq-to-compressible correction. Released under CC BY 4.0; cite Menglik, Shao and Hyde, arXiv:2606.25259.","displayTitle":"Paired Boussinesq and Fully Compressible Natural Convection: 10,000 FEniCSx Cavity Simulations (Menglik, Shao, Hyde)","huggingFaceUrl":"https://huggingface.co/datasets/NurshatMenglik/Paired_Boussinesq_Compressible_Dataset","id":"3b06d068-5e1c-5b67-9cf6-8a1039e71f34","license":"CC-BY-4.0","licenseEvidenceUrl":"https://huggingface.co/datasets/NurshatMenglik/Paired_Boussinesq_Compressible_Dataset","objectCountEstimate":100000,"probeStatus":"ok","referenceUrl":"https://arxiv.org/abs/2606.25259","region":"us","schemaCard":{"accessHints":[{"language":"python-datasets","snippet":"from datasets import load_dataset\nds = load_dataset(\"NurshatMenglik/Paired_Boussinesq_Compressible_Dataset\")"},{"language":"python-hub","snippet":"from huggingface_hub import snapshot_download\nsnapshot_download(repo_id=\"NurshatMenglik/Paired_Boussinesq_Compressible_Dataset\", repo_type=\"dataset\")"}],"caveats":["HuggingFace metadata: author=NurshatMenglik, license:cc-by-4.0, size_categories:10K<n<100K, downloads=2153"],"format":"mixed","generatedAt":"2026-11-21T20:48:51.504Z","layout":{"example":"manifest.csv","pattern":"{repoId}/{filename}"},"mountPath":"/mnt/archive","sampleHeaders":[{"bytes":"id,folder,res,Ra,Ma,seed,n_left,n_right,n_top,n_bot,legacy_cavity,nodes_P1,nodes_P2,comp_status,comp_steps,bouss_status,bouss_steps,comp_residual_drop,comp_last_rel_change,bouss_last_rel_change,T_min,T_max,rho_min,rho_max,retry,Ra_canonical,sqrt_RaPr,bouss_scaling\r\n0,sample_000000,32,181738.28211307156,0.0035255649412159625,774243,3,1,2,1,0,1089,4225,steady,26,steady,18,5.623629905431012e-12,2.7751815945631513e-07,6.229175041601635e-07,0.48135806019393923,1.571051704895913,0.5827440712995308,1.9524576814364","encoding":"utf8","key":"manifest.csv","truncatedAt":512}],"sampleObjects":["manifest.csv","sample_000000/T_bouss.npy","sample_000000/T_comp.npy","sample_000000/coords_T_bouss.npy","sample_000000/coords_T_comp.npy"],"sidecars":[{"bytes":"---\nlicense: cc-by-4.0\npretty_name: Paired Boussinesq/Compressible Natural Convection (10k)\nsize_categories:\n- 10K<n<100K\ntags:\n- computational-fluid-dynamics\n- natural-convection\n- boussinesq\n- compressible-flow\n- finite-element-method\n- scientific-machine-learning\n---\n\n# Paired Boussinesq / Compressible Natural Convection — 10,000 Simulations (version 2)\n\n📄 **Paper:** [A Neural Surrogate Approach for Simulating Natural Convection\nProblems (arXiv:2606.25259)](https://arxiv.org/abs/2606.25259) — Nurshat Menglik,\nAlex Shao, David Hyde.\n\n> **Version 2 (September 2026) replaces the original release.** All 10,000 pairs were\n> regenerated; see [What changed in version 2](#what-changed-in-version-2). The original\n> files remain available unchanged under the tag `v1.0-legacy`\n> (`snapshot_download(..., revision=\"v1.0-legacy\")`). New work should use version 2.\n\n10,000 *matched pairs* of steady 2D natural-convection simulations in the unit square\ncavity. Every sample solves the **same** randomized problem twice — once with the\nBoussinesq model and once with the fully compressible model — so the two members differ\nonly in the physics, not in the geometry, boundary data, or discretization. This is the\nraw finite-element output of the data-generation pipeline used in the paper.\n\nBoth solves use a monolithic, fully implicit mixed finite element method (Taylor–Hood:\n`[P2]^2` velocity, `P1` scalars) implemented in FEniCSx/DOLFINx.\n\n## Contents\n\n`manifest.csv` plus 10,000 sample folders `sample_000000` … `sample_009999` (~112 GB).\nThe mesh resolution is swept in four blocks of 2,500:\n\n| ids | mesh (`res`) |\n|---|---|\n| 0 – 2499 | 32 × 32 |\n| 2500 – 4999 | 64 × 64 |\n| 5000 – 7499 | 128 × 128 |\n| 7500 – 9999 | 256 × 256 |\n\nEach folder contains 23 files:\n\n| File | Description |\n|---|---|\n| `meta.json` | Generation parameters and the steady-state record of both solves |\n| `T_bouss.npy`, `T_comp.npy` | Temperature (P1 nodal values) |\n| `ux_bouss.npy`, `uy_bouss.npy` | Boussinesq velocity components (P2 nodal values) |\n| `ux_comp.npy`, `uy_comp.npy` | Compressible velocity components (P2 nodal values) |\n| `p_bouss.npy`, `p_comp.npy` | Pressure, min–max normalized to `[0, 1]` per sample |\n| `p_bouss_raw.npy`, `p_comp_raw.npy` | Pressure as solved (not normalized) |\n| `p_bouss_pert_raw.npy`, `p_comp_pert_raw.npy` | Raw pressure with its mean removed |\n| `rho_comp.npy` | Density (compressible solve only) |\n| `coords_*.npy` | `(n, 2)` node coordinates matching each field above |\n\nAll arrays are `float64` flat nodal arrays, not images. Always pair a field with its own\n`coords_*.npy`: the node ordering differs between samples, and because the solves ran in\nparallel the arrays can contain nodes on processor interfaces more than once (same\ncoordinates, same value). The nodes lie on a uniform grid — `(res+1)²` points for the P1\nfields and `(2·res+1)²` for the P2 velocities — so they can be placed on a regular array\nwithout interpolation (see Loading).\n\n## Parameters\n\n- Rayleigh number `Ra` log-uniform in `[1e2, 3e6]`, capped at `2e6` whenever the top or\n  bottom wall is conducting (median 1.5e4). Mach number `Ma` is tied to `Ra` by a log–log\n  map (`8.3e-5` … `1.4e-2`). `Pr = 0.71`, `γ = 1.4`, `Fr = 1.095`.\n- `T_init = 1`, `ΔT = 1.2`, so wall temperatures span `[0.4, 1.6]`. This is deliberately far\n  outside the Boussinesq regime (density typically varies by a factor of 3–4 across a sample, up to 5.3),\n  which is what makes the pair informative.\n- Each wall has a complexity dial `n` (`numT_left/right/top/bot` in `meta.json`). Left and\n  right walls: `n ∈ {1,2,3}`; top and bottom: `n ∈ {0,1,2}` with `n = 0` meaning adiabatic.\n  `n = 1` is a constant wall temperature; `n ≥ 2` is a smooth random trace built from\n  `n − 1` Fourier modes with `k^{-1.5}` decay, scaled to stay inside `[0.4, 1.6]` pointwise.\n  When both side walls have `n = 1` the sample is the classical differentially heated\n  cavity (hot left, cold right, adiabatic top and bottom): 1,159 samples\n  (`legacy_cavity = 1` in the manifest).\n- B","encoding":"utf8","key":"README.md","name":"README.md","size":10993,"truncated":true}],"sourceObjectKey":"manifest.csv","starterCell":"import pandas as pd\ndf = pd.read_csv(\"hf://datasets/NurshatMenglik/Paired_Boussinesq_Compressible_Dataset/manifest.csv\")\ndf.head()","starterSamplePath":"manifest.csv","totalObjects":100000},"summary":"Ten thousand matched pairs of steady 2D natural-convection (buoyancy-driven flow, hot and cold walls) simulations in the differentially heated unit-square cavity, released with \"A…","title":"Paired Boussinesq and Fully Compressible Natural Convection: 10,000 FEniCSx Cavity Simulations (Menglik, Shao, Hyde)","updatedAt":"2026-11-22T01:06:38.894Z","visibility":"public","websiteUrl":"https://huggingface.co/datasets/NurshatMenglik/Paired_Boussinesq_Compressible_Dataset"}}