README.md

---
pretty_name: CLT-IML Tokamak MHD Simulation Database
license: other
task_categories:
  - tabular-regression
tags:
  - plasma-physics
  - tokamak
  - magnetohydrodynamics
  - disruption-prediction
  - surrogate-modeling
  - scientific-machine-learning
  - tabular

---

# CLT-IML Tokamak MHD Simulation Database

> **Access and use.** This database is source-available for academic
> communication, inspection, and reproducibility assessment. It is not open
> data. Copyright (c) 2026 Zhejiang University. All rights reserved. Any use
> requires prior written permission from Zhejiang University or its duly
> authorized representative. See [Terms of access and use](#terms-of-access-and-use).

## Dataset summary

This dataset contains tabular scalar responses, sampled two-dimensional fields,
test fields, and reference visualizations derived from CLT tokamak
magnetohydrodynamic simulations. It supports the three surrogate-modeling tasks
implemented in the companion CLT-IML code repository:

1. regression from equilibrium/profile parameters to scalar simulation
   outcomes;
2. pressure-crash regression and thresholded disruption classification; and
3. two-dimensional perturbation-field reconstruction with a neural operator.

The data are numerical simulation products and are not experimental plasma
measurements. The CLT solver and complete raw outputs from the underlying
simulation campaigns are not part of this dataset.

## Associated manuscript

**Manuscript title:** *Millisecond prediction of tokamak disruptions using a
first-principles surrogate model* (manuscript in preparation)

**Authors:** W. Zhang, Y. J. Ma, S. Z. Cai, Z. W. Ma, Z. M. Sheng, and Y.
Zhang. W. Zhang and Y. J. Ma contributed equally. W. Zhang and Z. W. Ma are
the corresponding authors.

Affiliations:

1. Institute for Fusion Theory and Simulation, School of Physics, Zhejiang
   University, Hangzhou 310027, China.
2. College of Control Science and Engineering, Zhejiang University, Hangzhou
   310027, China.

## Repository structure

```text
.
├── Double_Tearing_Train_Database_Bisland_Ek.csv
├── Transferlearning_Database_Bisland_Database_Bisland.csv
├── pressure_crash_cls.csv
├── TMONet-test_by_p0.csv
├── Double_Tearing_Train_Database_by_p0.csv  # exact Task III train manifest
├── data_frame_by_p0/                         # dense field tables/reference plots
├── selected_B9_633/                          # sampled Task III training fields
└── TMON-test_by_p0/                          # Task III test fields/reference plots
```

The release includes the exact Task III training manifest used with the sampled
fields. Every `folder_name` in that manifest resolves to one CSV in
`selected_B9_633/` after `/` and `\` are replaced by `__`.

## Primary tables

| File                                                     | Rows | Purpose                              |
| -------------------------------------------------------- | ---: | ------------------------------------ |
| `Double_Tearing_Train_Database_Bisland_Ek.csv`           |  195 | Task I scalar-response data          |
| `Transferlearning_Database_Bisland_Database_Bisland.csv` |   24 | Target-domain transfer-learning data |
| `pressure_crash_cls.csv`                                 |  246 | Task II pressure-crash data          |
| `TMONet-test_by_p0.csv`                                  |   22 | Task III test-case manifest          |

Row counts exclude CSV headers.

## Data fields

The scalar tables use combinations of `p0`, `r1`, `r2`, `s1`, and `s2` as
inputs. Targets include inner/outer magnetic-island widths, linear growth rate,
kinetic energy, and central-pressure crash percentage. Field CSVs use spatial
coordinates (`X`, `Z` or `R`, `Z`) and `Value`; sampled training fields also
contain `Gradient`.

Coordinate shifts, axis swaps, normalization, logarithmic transforms, and
target-extraction rules are implemented in the companion code repository. See
its `docs/DATA_AND_MODELS.md` and extraction scripts before combining or
reprocessing tables.

## Intended uses

- reproducing the surrogate-model experiments in the associated manuscript;
- benchmarking scalar regression and neural-operator methods on CLT-derived
  simulation outputs;
- studying mappings between designed equilibrium parameters and MHD outcomes;
- validating or extending the supplied post-processing workflow.

## Limitations

- Results are specific to the simulated configurations, parameter ranges,
  normalization, and CLT setup used to generate the database.
- The data should not be interpreted as direct experimental measurements or as
  a validated real-time control system.
- Extrapolation outside the sampled parameter domain is not guaranteed to be
  physically meaningful.
- Exact numerical reproduction can depend on software versions, hardware, and
  nondeterministic GPU operations.

## Terms of access and use

Copyright (c) 2026 Zhejiang University. All rights reserved.

This database, including its tables, field data, manifests, visualizations,
metadata, and organization, is made publicly viewable for academic
communication, inspection, and reproducibility assessment. It is
**source-available**, but it is **not open data** and is not released under a
Creative Commons, Open Data Commons, or other open license. The
`license: other` value in the dataset-card metadata denotes these custom,
permission-required terms.

Public availability, downloading, or possession of a copy does not by itself
grant permission to use the database. Except to the extent expressly permitted
by applicable law, any use, reproduction, modification, adaptation, extraction,
creation or publication of derived data or results, model training or
evaluation, distribution, redistribution, sublicensing, deployment,
commercialization, or other exploitation of all or any substantial portion of
the database requires prior written permission from Zhejiang University or its
duly authorized representative. Rights not expressly granted in a written
permission or license agreement are reserved by Zhejiang University.

Permission to use the companion CLT-IML software does not automatically
authorize use of this database, and permission to use the database does not
automatically authorize use of the software. Requests for permission should be
directed to Zhejiang University or its duly authorized representative and
should identify the requester, institution, intended purpose, requested
materials, and planned form of use.

THE DATABASE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO WARRANTIES OF MERCHANTABILITY, FITNESS
FOR A PARTICULAR PURPOSE, ACCURACY, COMPLETENESS, AND NONINFRINGEMENT. TO THE
MAXIMUM EXTENT PERMITTED BY APPLICABLE LAW, ZHEJIANG UNIVERSITY AND THE
DATABASE CREATORS SHALL NOT BE LIABLE FOR ANY CLAIM, DAMAGES, OR OTHER
LIABILITY ARISING FROM, OUT OF, OR IN CONNECTION WITH THE DATABASE OR ITS USE.

## Citation

Authorized users should cite the associated manuscript and identify the exact
version or Hugging Face revision of the database used.