Cloud Studio is an ongoing cloud crushing research and visualization project by Kaan Boge. It studies how a dense cloud evolves in a hot wind using independent hydrodynamics codes. This repository contains the software, visualizations and analysis available so far. This is not a finished paper.
Explore the research & data · Open the 3D viewer · Browse movies & images · Download the evidence bundle
Website & viewer: MIT · License scope · Third party notices
Currently published: 174 historical viewer entries, 919 analysis files, and a velocity sensitivity summary containing 50 accepted controls, 25 pairs and 5,054 analyzed native states. These are separate collections, not one combined count of validated simulations. Counts describe the dated public catalog, not live simulation status.
| If you want to… | Open |
|---|---|
| Check current activity, measured runtimes and the planning inventory | Live simulation progress |
| Rotate, zoom and compare cloud structures | Interactive 3D viewer |
| Read the current scientific findings | All code velocity sensitivity summary |
| Find reports, scripts, checks and figures | Research hub · Analysis directory |
| Check how many frames a historical entry contains | Frame inventory CSV · JSON |
| Inspect the new Enzo level 5 preparation | Pinned plan and validation |
| Understand what is public versus retained locally | Data guide |
| See the contents of the published collection | Current contents |
| Reproduce or review the workflow | Methods and reproducibility |
The published velocity sensitivity study measures the effect of the initial velocity prescription at several resolutions. Each accepted within code pair uses the same native executable, grid or initial particle sampling, numerical settings and physical parameters. Only the chosen velocity law and its consistent kinetic energy change.
The pilot fixes χ = 100, Mach = 2, γ = 5/3, with the existing density edge and domain. AthenaPK's historical prescription was constant outer momentum; it was not the same tanh velocity prescription used by the other accepted pairs. Native boundary, tracer and SPH pressure caveats are retained, not hidden behind a common label.
One part of the study: the original two code pilot. Curves use the measured native outputs. See the updated all code report for the wider evidence; this image is not a code ranking or a convergence proof.
| Code / method | Execution in this project | Published sensitivity evidence |
|---|---|---|
| AthenaPK | CUDA/GPU | L3 to L4 |
| Athena++ | MPI/CPU | L3 to L4 |
| Athena 4.2 | MPI/CPU | L3 to L5 |
| Enzo | MPI/CPU | L3 to L4 · new L5 preparation |
| Enzo-E | Charm++/CPU | L3 to L4 |
| RAMSES | MPI/CPU | L3 to L4 |
| FLASH 4.8 | MPI/CPU | L3 to L4 |
| Flash-X | MPI/CPU | L3 to L4 |
| Arepo | MPI/CPU, moving mesh | L3 to L4 |
| GIZMO MFM | MPI/CPU, mesh free finite mass | L3 to L4 |
| GIZMO MFV | MPI/CPU, mesh free finite volume | Repaired L3 to L4 |
| Gadget-4 | MPI/CPU, SPH | L3 to L4 |
| Gasoline | CPU, SPH | Validation hold |
These are twelve code families and thirteen methods when GIZMO MFM and MFV are counted separately. Results are produced by each code's own solver, not by relabelling FLASH output.
| Project level | Streamwise × transverse × transverse | Initial elements per cloud radius |
|---|---|---|
| 1 | 16 × 8 × 8 | 0.8 |
| 2 | 32 × 16 × 16 | 1.6 |
| 3 | 64 × 32 × 32 | 3.2 |
| 4 | 128 × 64 × 64 | 6.4 |
| 5 | 256 × 128 × 128 | 12.8 |
| 6 | 512 × 256 × 256 | 25.6 |
The domain is 20 × 10 × 10 cloud radii. Level 6 is not 512³. For moving mesh and particle codes, this ladder describes the initial sampling; it does not guarantee identical evolved spatial resolution. Levels 1 to 2 do not resolve a cloud well enough for quantitative convergence conclusions.
Native code + pinned input
↓
3D output fields + logs + restart data, retained locally
↓
Native readers / yt → checked mass, time and field diagnostics
↓
matplotlib figures · extracted 3D surfaces · ffmpeg movies
↓
GitHub Pages + the existing streamed-mesh frames branch
The website displays derived results; uploading an arbitrary HDF5 file does not turn it into a movie. The viewer uses recorded frame timestamps; this catalog does not independently revalidate them against every original native file. A target of 101 snapshots is not a promise that every historical entry has 101 frames, identical intermediate times, or all cloud material inside the box.
research.html Research reports, data downloads and status guide
index.html Existing image/movie studio
viewer.html Existing interactive 3D viewer
analysis/ Custom analysis scripts, reports, diagnostics and figures
data/ Viewer manifests and downloadable research/frame catalogs
assets/ Published movies and images
docs/ Methods, data guide, current contents and update notes
downloads/ Portable research-evidence bundle
vendor/ Existing browser dependencies
The separate frames branch holds compressed derived mesh frames, not complete native fields. File inventory and SHA-256 checksums make the published analysis package inspectable. The download scope explains bundle exclusions.
No raw output was deleted for this publication. The local storage inventory is metadata, not a cloud backup. Compression tests restored every tested byte, but compression has not been enabled for the active Enzo pair.
- The historical viewer is not a certified matched code experiment. Shared grid labels and color scales do not remove differences in initial conditions, boundaries or numerical methods.
- Gasoline's field repeatability gate remains failed. Completed diagnostics are not accepted full controls.
- Earlier cooling labelled cases did not enable cooling. They are not radiative results.
- Existing frame tracking tests verify several implementation properties, but do not establish complete cloud material retention.
- Movies and quantized isosurfaces are not replacements for native pressure, energy, velocity or tracer fields.
See verification, tracking/cooling evidence, and current published contents.
This repository is maintained by KaanBoge. Please cite the particular report and commit used; CITATION.cff provides project metadata. No DOI or peer reviewed publication is claimed.
Cloud Studio's original website/viewer software and the specifically listed utilities and documentation are released under the MIT License, copyright 2026 Kaan Boge. MIT permits reuse, modification and commercial distribution while requiring preservation of the license and copyright notice. It is a software license, not an MIT institutional affiliation or a scientific certification.
The grant is scoped in LICENSING.md. Third party simulation codes, adapted native code files, research datasets and archived evidence are not blanket relicensed. Bundled three.js retains its authors' MIT notice. See third party notices and contribution guidance. Scientific references and recorded provenance remain intact.