---
license: cc-by-4.0
tags:
- turbulence
- computational-fluid-dynamics
- direct-numerical-simulation
- navier-stokes
- scientific-machine-learning
pretty_name: 'TIDE — decay_saffman_v2'
arxiv: 2608.04222
---

# TIDE — `decay_saffman_v2`

**Paper:** [arXiv:2608.04222](https://arxiv.org/abs/2608.04222) | **Index / datasheet:** [ydai17/TIDE](https://huggingface.co/datasets/ydai17/TIDE) | **Code:** [GitHub](https://github.com/Dyloong1/TIDE-dataset-benchmark)


One configuration of **TIDE** (Turbulent Incompressible DNS Ensembles): a 256^3
fp64 direct numerical simulation of the incompressible Navier-Stokes equations,
shipped as independent realizations and released only after passing a fixed
acceptance standard of statistical gates and equation-level residual checks.

- **Index of all configurations:** https://huggingface.co/datasets/ydai17/TIDE
- **Code (solver, acceptance referee, benchmark):** https://github.com/Dyloong1/TIDE-dataset-benchmark
- **DOI record:** https://doi.org/10.5281/zenodo.21589490

```bash
huggingface-cli download ydai17/TIDE-decay_saffman_v2 --repo-type dataset --local-dir ./tide-data/corpus
```

```python
import zarr
z = zarr.open("./tide-data/corpus/decay_saffman_v2.zarr", mode="r")
u = z["u"][0]   # (3, 256, 256, 256) velocity of the first frame
```

Each store holds velocity `[N,3,256^3]`, pressure `[N,256^3]`, the optional
scalar/buoyancy channel, and per-frame `t`, `k_max_eta`, `seed`. Fields are
computed in fp64 and stored in fp32; frames are exported every 0.05 T_L.
Data under CC-BY-4.0.