# The Well: Acoustic Scattering Through Mazes

## Facts

- Provider: Polymathic AI (The Well); Michael McCabe
- License: CC-BY-4.0 (https://arxiv.org/html/2412.00568)
- Size: 31 files
- Format: HDF5 (in sampled files)
- Stored at: Hugging Face, https://huggingface.co/datasets/polymathic-ai/acoustic_scattering_maze
- Access: public, readable without an account
- Category: Physics
- Source: https://arxiv.org/abs/2412.00568
- Page: https://scigantic.com/archive/923d2f7f-e6cb-5e2c-9833-a7d6f6841dcf
- Keywords: Physics, HDF5, Hugging Face

## About

2D simulations of the variable-coefficient linear acoustic equations, which describe a pressure wave propagating through materials with different density and sound speed. Pressure waves leave point sources and travel through maze-like domains of low-density paths and walls orders of magnitude denser, a machine learning challenge with parallels to optimal placement problems such as WiFi coverage in a building. Solved with Clawpack on a uniform grid over [-1, 1] by [-1, 1], open in y with reflective walls in x; each trajectory has 201 steps of 256 by 256 images over t from 0 to 4. Fields are pressure, material density, material sound speed and velocity. The datasheet lists 2,000 trajectories. Listing data/ (measured 2026-11-13) shows train/ (16 files, 256 GB), valid/ and test/ (2 files, 32 GB each), stored as chunked files of many trajectories; about 320 GB in the repository. Files are HDF5 in the common The Well format; open them with the_well package (WellDataModule with base path hf://datasets/polymathic-ai/) or directly with h5py. Part of The Well (Ohana et al., NeurIPS 2024 Datasets and Benchmarks, arXiv:2412.00568); released under CC-BY-4.0, cite The Well and the simulation's source papers.

## Load the data

```python
from huggingface_hub import hf_hub_download
path = hf_hub_download(repo_id="polymathic-ai/acoustic_scattering_maze", filename="data/test/acoustic_scattering_maze_chunk_18.hdf5", repo_type="dataset")
print(path)
```

## Data structure

File types in sampled files: HDF5

Example file: `data/test/acoustic_scattering_maze_chunk_18.hdf5`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/923d2f7f-e6cb-5e2c-9833-a7d6f6841dcf
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/923d2f7f-e6cb-5e2c-9833-a7d6f6841dcf
- Over MCP: `get_schema_card` with id `923d2f7f-e6cb-5e2c-9833-a7d6f6841dcf` at https://api.scigantic.com/mcp

## Related topics

- [HDF5 datasets](https://scigantic.com/topics/hdf5)
- [Physics datasets](https://scigantic.com/topics/physics)
