Fusion Physics · proxima-fusion
constellaration
A dataset of quasi-isodynamic stellarator plasma boundaries and MHD equilibria for fusion energy optimization.
What constellaration is
ConStellaration is a scientific dataset containing boundary shapes, performance metrics, and ideal magneto-hydrodynamic (MHD) equilibria for quasi-isodynamic (QI) stellarators, curated by Proxima Fusion. Stellarators are magnetic confinement devices designed to achieve steady-state fusion energy without relying on large net plasma currents, which intrinsically protects them from current-driven plasma disruptions. Within stellarator research, quasi-isodynamic configurations represent an important approach for designing commercially viable fusion power plants.
Designing stellarator magnetic fields is a challenging high-dimensional, constrained optimisation problem that traditionally requires computationally expensive physics simulations and specialised domain knowledge. Community progress in applying modern data-driven and automated methods has been limited by a lack of standardised benchmarks and large-scale open datasets. ConStellaration addresses this challenge by providing hundreds of thousands of precomputed configurations to support optimization and machine learning research for fusion energy.
How it was made
Plasma boundaries in the dataset were generated by sampling diverse quasi-isodynamic magnetic fields and optimising surface shapes targeting key confinement properties. Boundaries were sampled using four complementary approaches: the DESC equilibrium solver, the Variational Moments Equilibrium Code (VMEC/VMEC++), quasi-poloidal (QP) initialisation, and near-axis expansion (NAE). The baseline subset contains boundary Fourier coefficients, sampling settings, target fields, and ideal MHD metrics computed under vacuum conditions.
To capture behaviour in operating plasma, the boundary geometries were subsequently evaluated under finite plasma pressure across five distinct conditions: 1%, 2%, 3%, 4%, and 5% volume-averaged beta. For each condition, ideal MHD equilibria and physics metrics—such as magnetic mirror ratios, rotational transform profiles, flux compression, elongation, and quasi-isodynamicity measures—were recomputed. Complete equilibrium states are stored as JSON-serialised representations of VMEC WOut files, which can be exported directly into standard VMEC2000 NetCDF output files.
Key facts
- Curator
- Proxima Fusion
- Confinement concept
- Quasi-isodynamic (QI) stellarator
- Plasma pressure conditions
- Vacuum, 1%, 2%, 3%, 4%, and 5% volume-averaged beta
- Boundary generation methods
- DESC, VMEC/VMEC++, quasi-poloidal initialisation, near-axis expansion
- Data modalities
- Tabular metrics, Fourier boundary coefficients, and JSON-encoded VMEC equilibria
- Distribution format
- Parquet files
What you can use it for
- Machine learning surrogate modelling
- Researchers can train neural networks on boundary Fourier coefficients to rapidly predict key equilibrium properties, such as magnetic mirror ratios and rotational transforms, bypassing slow numerical solvers.
- Stellarator shape optimisation benchmarking
- Computational researchers can evaluate black-box and gradient-based optimisation algorithms on realistic quasi-isodynamic objective functions using standard baseline data.
- Finite-beta equilibrium scaling studies
- Physicists can compare magnetic field metrics across vacuum and 1% to 5% volume-averaged beta conditions to examine how plasma pressure alters confinement and stability.
- Generative plasma surface modelling
- Data scientists can develop generative models conditioned on target physics metrics to propose viable 3D stellarator surface boundaries.
Explore the data
The schema, real rows, full-text search and a SQL filter — read in your browser, nothing downloaded.
Explore
Read the real rows without downloading anything
Reading rows…
LiveRead from proxima-fusion/constellaration at the moment you asked. Nothing is cached or stored — every row above came from that request.
Strengths and limits
Good at
- Covers a broad parameter space evaluated under vacuum and five distinct finite-beta plasma pressure conditions up to 5%.
- Generates boundaries across four established stellarator modelling methods: DESC, VMEC, quasi-poloidal initialisation, and near-axis expansion.
- Provides full JSON-encoded VMEC equilibrium files that convert directly into standard VMEC2000 NetCDF outputs alongside scalar metrics.
- Formats all data into Parquet tables that interface directly with standard computational frameworks such as PyTorch, Polars, and Dask.
Watch out for
- Focuses exclusively on quasi-isodynamic configurations and does not encompass other stellarator families like quasi-axisymmetric designs.
- Assumes stellarator symmetry for the boundary representations, excluding non-symmetric field perturbations.
- Contains failed simulation entries and recorded error flags where numerical sampling or equilibrium convergence did not succeed.
Questions
Can I look inside constellaration without downloading it?
Yes. This page reads its schema, real rows and per-column statistics in the browser, with full-text search and a SQL filter. Nothing is downloaded to your machine.
What licence is constellaration under?
Its authors declare mit. Read the terms on the original page before reusing the data.
What plasma conditions are represented in the dataset?
The dataset contains evaluations in a vacuum state as well as five finite-beta conditions representing 1%, 2%, 3%, 4%, and 5% volume-averaged plasma pressure.
How are the plasma boundary shapes parameterised?
Boundaries are parameterised by Fourier coefficients representing radial and vertical cylindrical coordinates (R and Z) for a given number of field periods.
Can the ideal MHD equilibria be converted to standard fusion physics formats?
Yes. The JSON representations of the VMEC output files can be exported to standard VMEC2000 NetCDF WOut files using the accompanying constellaration Python package.
Which physics metrics are recorded for each configuration?
Metrics include quasi-isodynamicity, rotational transform over field periods, axis and edge magnetic mirror ratios, vacuum well, flux compression in regions of bad curvature, elongation, aspect ratio, and triangularity.
How to cite it
Cadena, Santiago A and Merlo, Andrea and Laude, Emanuel and Bauer, Alexander and Agrawal, Atul and Pascu, Maria and Savtchouk, Marija and Guiraud, Enrico and Bonauer, Lukas and Hudson, Stuart and others. ConStellaration: A dataset of QI-like stellarator plasma boundaries and optimization benchmarks. arXiv preprint arXiv:2506.19583 (2025).
Written from constellaration's own dataset card. Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.