---
license: other
task_categories:
- time-series-forecasting
- unconditional-image-generation
pretty_name: HAWINS Bubble FM HDF5 v2 FP64
tags:
- radiation-hydrodynamics
- flow-matching
- hdf5
- multi-fidelity
- hawins
---

# HAWINS Bubble FM HDF5 v2 FP64

This dataset contains multi-fidelity HAWINS `bubble_shock2` radiation-hydrodynamics
trajectories for conditional flow-matching residual training.

## Contents

- `manifest.csv`: one row per parameter case, with split, file path, physical
  parameters, runtime, and step counts.
- `dataset_card.json`: machine-readable dataset metadata.
- `cases/case_*.h5`: one HDF5 file per parameter case.
- `visual_report/`: lightweight figures and summary tables for inspection.

## Simulation Setup

- Problem: `bubble_shock2`
- Storage dtype: `float64`
- Channels: `rho`, `mom_x`, `mom_y`, `energy_e`, `energy_i`, `energy_r`
- Parameters: `bubble_rho_ratio`, `bubble_pressure_ratio`, `bubble_radius`,
  `shock_distance_ratio`, `shock_velocity_scale`
- Snapshots per trajectory: 9
- Final time: `0.001`
- HAWINS commit recorded in metadata: `8610eb39a5d919af7922b01a7ffc7d772f0103a1`

## Fidelity Levels

| Fidelity | Grid |
| --- | --- |
| L0 | 128 x 48 |
| L1 | 256 x 96 |
| L2 | 800 x 288 |

The manifest records which fidelities are available for each case.

## HDF5 Layout

Each case file stores data under `fidelities/{L0,L1,L2}` when available:

- `traj/U`: trajectory array with shape `(time, channel, y, x)`
- `traj/t`: physical snapshot times
- `traj/tau`: normalized snapshot times
- `grid/x`, `grid/y`: grid coordinates
- `runtime`: per-fidelity runtime metadata
- `params/theta`: normalized/ordered parameter vector

## Usage

Use `manifest.csv` to select cases, then load the corresponding HDF5 file.

```python
import h5py

with h5py.File("cases/case_000000.h5", "r") as f:
    u_l0 = f["fidelities/L0/traj/U"][:]
    theta = f["params/theta"][:]
```

## Notes

The dataset is intended for scientific ML experiments on residual learning,
super-resolution, and multi-fidelity flow matching. Generated visualization
artifacts are included only for inspection; model training should use
`manifest.csv` and the HDF5 case files.