pith. sign in

arxiv: 2604.26763 · v3 · pith:N67I6AWTnew · submitted 2026-04-29 · ⚛️ physics.plasm-ph

Exploring the link between coil non-planarity and magnetic surface geometry across a dataset of QI stellarators

Andrea Pavone , Sehyun Kwak , Felix Warmer This is my paper

Pith reviewed 2026-05-21 00:27 UTC · model grok-4.3

classification ⚛️ physics.plasm-ph
keywords coil non-planarityplasma boundaryprincipal curvaturesquasi-isodynamic stellaratorssurface geometrycoil complexitystellarator designmagnetic surface
0
0 comments X

The pith

The rate at which principal directions rotate across the plasma boundary is the strongest predictor of coil non-planarity in quasi-isodynamic stellarators.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Stellarator fusion devices require complex non-planar coils to confine plasma, and the link between plasma boundary shape and coil complexity remains a key design question. This work analyzes a large dataset of quasi-isodynamic stellarator boundaries by computing optimized filamentary coil sets and quantifying their non-planarity through metrics such as torsion and inclination angle. Surface geometry features extracted from the second fundamental form and principal curvatures are compared against these coil metrics using univariate correlations and multivariate regression. The principal-direction rotation rate stands out as the top single predictor, while a small collection of surface features supports accurate statistical forecasting of non-planarity. The results indicate that changes in principal curvatures across the boundary are the main driver of coil complexity within this class of configurations.

Core claim

In the examined dataset of quasi-isodynamic stellarators, the principal-direction rotation rate of the plasma boundary is the single best predictor of coil non-planarity. Filamentary coil configurations are obtained through constrained optimization, and coil-complexity metrics including torsion, SVD non-planarity score, and spectral width are defined alongside surface features such as the second fundamental form, principal curvatures, and their rotation rates. Univariate and multivariate statistical analyses establish that this rotation rate correlates most strongly with coil non-planarity, and a random forest model using up to four surface features reaches an R-squared value of 0.882 for预测.

What carries the argument

The principal-direction rotation rate, which measures how the directions of maximum and minimum curvature change across the plasma boundary surface.

If this is right

  • Coil non-planarity increases directly with higher principal-direction rotation rates on the plasma boundary.
  • Surface curvature properties dominate over magnetic axis properties as predictors of coil complexity.
  • A compact set of surface geometry features suffices to model required coil non-planarity with high accuracy.
  • Plasma boundary designs that limit changes in principal curvature directions can reduce the engineering demands placed on the coil set.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Optimization codes for stellarator boundaries could add explicit penalties on principal-direction rotation to target simpler coil realizations.
  • The identified surface-to-coil link may help explain historical differences in coil intricacy across different stellarator families.
  • Similar statistical mappings could be tested on non-quasi-isodynamic configurations to check whether surface geometry remains the dominant factor.
  • A controlled experiment that varies only the boundary curvature rotation while holding other parameters fixed would directly test the causal direction of the observed correlation.

Load-bearing premise

The filamentary coil configurations obtained from constrained optimization accurately represent the coil complexity truly required by each plasma boundary, and the dataset covers a representative range of quasi-isodynamic shapes.

What would settle it

A new plasma boundary engineered for low principal-direction rotation rate that nevertheless requires highly non-planar coils after the same optimization procedure would contradict the reported predictive relationship.

Figures

Figures reproduced from arXiv: 2604.26763 by Andrea Pavone, Felix Warmer, Sehyun Kwak.

Figure 1
Figure 1. Figure 1: Comparison of tokamak and stellarator coil geometries. The plasma surface is view at source ↗
Figure 2
Figure 2. Figure 2: Schematic of the torus illustrating the toroidal angle view at source ↗
Figure 2
Figure 2. Figure 2: Representative plasma boundary–coil configurations from the three filter pools, [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Representative plasma boundary–coil configurations from the three filter pools, view at source ↗
Figure 3
Figure 3. Figure 3: Illustration of the Frenet–Serret frame and coil torsion. Planar coils (left) have [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Illustration of the Frenet–Serret frame and coil torsion. Planar coils (left) have view at source ↗
Figure 4
Figure 4. Figure 4: Illustration of the SVD non-planarity score. [PITH_FULL_IMAGE:figures/full_fig_p010_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Illustration of the SVD non-planarity score. view at source ↗
Figure 5
Figure 5. Figure 5: Two complementary non-planarity regimes from the dataset. [PITH_FULL_IMAGE:figures/full_fig_p011_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Two complementary non-planarity regimes from the dataset. view at source ↗
Figure 6
Figure 6. Figure 6: R–z projection of a coil showing the definition of the inboard-side inclination angle θinc. The pink dot marks the innermost midplane crossing (z = 0, minimum R). The coil tangent tˆ (green) and the vertical direction zˆ (gold) are shown at that point; their enclosed angle θ = 22.9 ◦ is the inclination angle for this coil. a flux surface and generating the field on the very same surface need be orthogonal … view at source ↗
Figure 7
Figure 7. Figure 7: R–z projection of a coil showing the definition of the inboard-side inclination angle θinc. The pink dot marks the innermost midplane crossing (z = 0, minimum R). The coil tangent tˆ (green) and the vertical direction zˆ (gold) are shown at that point; their enclosed angle θ = 22.9 ◦ is the inclination angle for this coil. directions, which is a geometric signature of strongly shaped, twisted plasma boundaries view at source ↗
Figure 7
Figure 7. Figure 7: Geometric illustration of the principal-direction rotation rate. The tangent [PITH_FULL_IMAGE:figures/full_fig_p013_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Geometric illustration of the principal-direction rotation rate. The tangent view at source ↗
Figure 8
Figure 8. Figure 8: Progressive illustration of coil trajectories in the [PITH_FULL_IMAGE:figures/full_fig_p014_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Progressive illustration of coil trajectories in the view at source ↗
Figure 9
Figure 9. Figure 9: Three-dimensional views of plasma boundaries with coil sets (white) for the [PITH_FULL_IMAGE:figures/full_fig_p016_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Three-dimensional views of plasma boundaries with coil sets (white) for the view at source ↗
Figure 10
Figure 10. Figure 10: Three-dimensional views for the highest- and lowest-twist configurations, [PITH_FULL_IMAGE:figures/full_fig_p017_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Three-dimensional views for the highest- and lowest-twist configurations, view at source ↗
Figure 11
Figure 11. Figure 11: Principal-direction rotation rate (pdrot) for the highest-pdrot (left) and [PITH_FULL_IMAGE:figures/full_fig_p018_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Principal-direction rotation rate (pdrot) for the highest-pdrot (left) and view at source ↗
Figure 12
Figure 12. Figure 12: Same five-row layout as Fig [PITH_FULL_IMAGE:figures/full_fig_p019_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Same five-row layout as Fig view at source ↗
Figure 13
Figure 13. Figure 13: Demeaned scatter plot of maximum SVD non-planarity vs. mean twist rate [PITH_FULL_IMAGE:figures/full_fig_p020_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Demeaned scatter plot of maximum SVD non-planarity vs. mean twist rate view at source ↗
Figure 14
Figure 14. Figure 14: Demeaned scatter plot of p95 coil torsion [PITH_FULL_IMAGE:figures/full_fig_p021_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Demeaned scatter plot of p95 coil torsion view at source ↗
Figure 15
Figure 15. Figure 15: Demeaned scatter plot of maximum inboard inclination angle vs. mean twist [PITH_FULL_IMAGE:figures/full_fig_p022_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Demeaned scatter plot of maximum inboard inclination angle vs. mean twist view at source ↗
Figure 16
Figure 16. Figure 16: ExtraTrees (ET) and OLS best-subset R2 vs. number of features k for the three coil targets (strict-zero filter, N ≈ 670). Each point is the globally optimal subset of size k found by exhaustive search. The ET–OLS gap quantifies the nonlinear contribution to predictability. setting (ρpartial = +0.17, p < 10−5 ). In the tolerant filter, |sin 2α| and pdrot both retain partial correlations of ρpartial ≈ +0.15… view at source ↗
Figure 17
Figure 17. Figure 17: Random Forest (RF) and OLS best-subset R2 vs. number of features k for the three coil complexity targets. Solid curves: RF; dashed curves: OLS. Each point is the best subset of size k found by exhaustive search over the surface-geometry feature set. zig-zag is directly visible in the coil footprints overlaid on the parameter-space panels of Figures 12 and 13: the projected coil curves on high-pdrot and hi… view at source ↗
Figure 17
Figure 17. Figure 17: Predicted vs. actual scatter plots for the full ET models (all [PITH_FULL_IMAGE:figures/full_fig_p024_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: Predicted vs. actual scatter plots for the full Random Forest models ( view at source ↗
Figure 18
Figure 18. Figure 18: Three-dimensional views of the highest- and lowest-geometric-torsion configura [PITH_FULL_IMAGE:figures/full_fig_p025_18.png] view at source ↗
Figure 19
Figure 19. Figure 19: Three-dimensional views of the highest- and lowest-geometric-torsion configura view at source ↗
Figure 20
Figure 20. Figure 20: Geometric torsion of the surface (related to the rate of change of the surface view at source ↗
read the original abstract

Stellarator fusion devices confine plasma by means of complex, non-planar electromagnetic coils. Understanding how the shape of the plasma boundary determines the required complexity of the coil set is a central open question in stellarator design, with direct implications for engineering feasibility and the prospects of building next-generation fusion power plants. In this work we address this question using a large data-driven study. Starting from the Constellaration dataset of quasi-isodynamic (QI) stellarator plasma boundaries, we compute a set of filamentary coil configurations using constrained optimisation within SIMSOPT, and define quantitative coil-complexity metrics (torsion, SVD non-planarity score, inboard-side inclination angle, spectral width) together with a rich set of surface and magnetic geometry features (second fundamental form, principal-direction rotation rate, surface curvatures, and magnetic axis properties). Univariate and multivariate statistical analyses, reveal a strong, central role of the surface geometry: the principal-direction rotation rate of the plasma boundary is the single best predictor of coil non-planarity, while a Random Forest model using up to four surface features achieves R2 = 0.882 for the same target. These results provide quantitative evidence that the rate of change of the principal curvatures cross the plasma boundary are the primary drivers of coil non-planarity in this dataset of quasi-isodynamic stellarators.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The paper performs a data-driven statistical study on the Constellaration dataset of quasi-isodynamic stellarator plasma boundaries. Filamentary coil sets are generated via constrained optimization in SIMSOPT for each boundary; quantitative coil non-planarity metrics (torsion, SVD score, inclination angle, spectral width) are then correlated with surface-geometry features (principal-direction rotation rate, second fundamental form, curvatures, magnetic-axis properties). Univariate analysis identifies the principal-direction rotation rate as the strongest single predictor of coil non-planarity, while a Random Forest regressor using at most four surface features reaches R² = 0.882.

Significance. If the reported correlations prove robust, the work supplies the first quantitative, dataset-scale evidence that the rate of change of principal curvatures across the plasma boundary is the dominant geometric driver of coil complexity in QI stellarators. This finding has direct implications for shape optimization aimed at engineering-feasible coil sets. The study’s strengths are its use of a sizable, publicly referenced configuration database, the definition of multiple independent non-planarity proxies, and the combination of univariate ranking with multivariate modeling.

major comments (3)
  1. [§3.2] §3.2 (coil optimization procedure): the manuscript provides no information on the number of filamentary coils, current bounds, or convergence tolerances employed in the SIMSOPT optimizations. Because these choices directly determine the reported non-planarity scores, their possible correlation with the surface features under study must be quantified before the geometric interpretation can be regarded as load-bearing.
  2. [§4.3] §4.3 (Random Forest results): the R² = 0.882 value is presented without any description of train/test partitioning, k-fold cross-validation, or hyperparameter search. In the absence of these controls it is impossible to assess whether the four-feature model generalizes or simply overfits the Constellaration sample.
  3. [§4.1] §4.1 (univariate ranking): the claim that principal-direction rotation rate is the single best predictor is not accompanied by the full correlation matrix or by any multicollinearity diagnostic among the surface features. Without this information the relative importance ranking cannot be considered definitive.
minor comments (2)
  1. [Figure 2] Figure 2: the color scale for the SVD non-planarity score should be stated explicitly in the caption.
  2. [Table 1] Table 1: the definition of “spectral width” is referenced to an earlier work but the precise formula used here is not restated, which would aid reproducibility.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their detailed and constructive report. We address each major comment below and indicate the revisions we will make to improve the manuscript.

read point-by-point responses

  1. Referee: [§3.2] §3.2 (coil optimization procedure): the manuscript provides no information on the number of filamentary coils, current bounds, or convergence tolerances employed in the SIMSOPT optimizations. Because these choices directly determine the reported non-planarity scores, their possible correlation with the surface features under study must be quantified before the geometric interpretation can be regarded as load-bearing.

    Authors: We agree that explicit documentation of the optimization parameters is necessary for reproducibility and to exclude confounding effects. In the revised manuscript we will expand §3.2 with the precise number of filamentary coils per half-period, the current bounds imposed, and the convergence tolerances used in SIMSOPT. We will also add a short sensitivity study that checks for correlations between these numerical choices and the surface-geometry features; any statistically significant correlations will be reported and discussed. revision: yes

  2. Referee: [§4.3] §4.3 (Random Forest results): the R² = 0.882 value is presented without any description of train/test partitioning, k-fold cross-validation, or hyperparameter search. In the absence of these controls it is impossible to assess whether the four-feature model generalizes or simply overfits the Constellaration sample.

    Authors: We accept that the machine-learning section requires additional methodological detail. The revised §4.3 will specify the train/test split (80/20 with stratification on the principal-direction rotation rate), the 5-fold cross-validation scheme, and the grid-search procedure used for hyperparameter selection. We will also report the mean cross-validated R² together with the held-out test-set R² to demonstrate that the quoted performance is not an artifact of overfitting. revision: yes

  3. Referee: [§4.1] §4.1 (univariate ranking): the claim that principal-direction rotation rate is the single best predictor is not accompanied by the full correlation matrix or by any multicollinearity diagnostic among the surface features. Without this information the relative importance ranking cannot be considered definitive.

    Authors: We concur that the univariate analysis would be strengthened by the full correlation structure. In the revision we will include the complete Pearson correlation matrix among all surface features (as a table or supplementary figure) and compute variance-inflation factors to quantify multicollinearity. These diagnostics will be used to confirm that the principal-direction rotation rate remains the strongest predictor even after accounting for linear dependencies among the other variables. revision: yes

Circularity Check

0 steps flagged

No circularity: independent computations of surface features and optimized coil metrics linked by statistical models

full rationale

The paper computes surface geometry features (principal-direction rotation rate, curvatures, second fundamental form) directly from the plasma boundary and obtains coil non-planarity metrics (torsion, SVD score, inclination angle, spectral width) via separate constrained optimizations in SIMSOPT for each Constellaration QI configuration. Univariate predictors and a Random Forest model are then fitted to relate these two independently generated sets of quantities, yielding R²=0.882. No equation or claim reduces a reported result to a fitted parameter or input by construction, no load-bearing self-citation is invoked to justify a uniqueness theorem or ansatz, and the derivation chain consists of explicit numerical steps that remain falsifiable against external coil optimizations. The analysis is therefore self-contained.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The study rests on standard computational plasma-physics assumptions and the representativeness of the Constellaration dataset; no new physical entities are introduced and only routine modeling choices appear as free parameters.

free parameters (1)
  • Random Forest feature count and hyperparameters
    Choice of up to four surface features and unspecified model hyperparameters that affect the reported R².
axioms (1)
  • domain assumption Filamentary coil solutions from constrained SIMSOPT optimization faithfully encode coil non-planarity for the given plasma boundaries.
    Invoked when coil-complexity metrics are computed from the optimized coils.

pith-pipeline@v0.9.0 · 5779 in / 1392 out tokens · 47377 ms · 2026-05-21T00:27:15.844243+00:00 · methodology

Review history (2 revisions) →

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

22 extracted references · 22 canonical work pages · 2 internal anchors

  1. [1]

    J. M. García-Regaña, I. Calvo, E. Sánchez, H. Thienpondt, J. L. Velasco, and J. A. Capitán. Reduced electrostatic turbulence in the quasi-isodynamic stellarator configuration CIEMAT-QI4.Nuclear Fusion, 65:016036, 2024

    work page 2024

  2. [2]

    Warmer et al

    F. Warmer et al. Overview of european efforts and advances in stellarator power plant studies.Fusion Engineering and Design, 203:114386, 2024

    work page 2024

  3. [3]

    A. G. Goodman, K. Camacho Mata, S. A. Henneberg, R. Jorge, M. Landreman, G. G. Plunk, H. M. Smith, R. J. J. Mackenbach, C. D. Beidler, and P. Helander. Constructing precisely quasi-isodynamic magnetic fields.Journal of Plasma Physics, 89(5):905890504, 2023. doi: 10.1017/S002237782300065X

    work page doi:10.1017/s002237782300065x 2023

  4. [4]

    A. G. Goodman, P. Xanthopoulos, G. G. Plunk, H. M. Smith, C. Nührenberg, C. D. Beidler, S. A. Henneberg, G. Roberg-Clark, M. Drevlak, and P. Helander. Quasi- isodynamic stellarators with low turbulence as fusion reactor candidates.PRX Energy, 3:023010, 2024. doi: 10.1103/PRXEnergy.3.023010. 32

    work page doi:10.1103/prxenergy.3.023010 2024

  5. [5]

    A. G. Goodman, G. G. Plunk, P. Xanthopoulos, M. Drevlak, J. Geiger, R. Davies, H. M. Smith, C. Nührenberg, C. D. Beidler, S. A. Henneberg, and P. Helander. A quasi-isodynamic stellarator configuration towards a fusion power plant.Journal of Plasma Physics, 91(6):E153, 2026. doi: 10.1017/S0022377825100974

    work page doi:10.1017/s0022377825100974 2026

  6. [6]

    Kappel, M

    J. Kappel, M. Landreman, and D. Malhotra. The magnetic gradient scale length explains why certain plasmas require close external magnetic coils.Plasma Physics and Controlled Fusion, 66(2):025018, 2024

    work page 2024

  7. [7]

    Hirshman, S

    J. Kappel, M. Landreman, P. Jurašić, and S. A. Henneberg. How does the magnetic gradient scale length influence complexity of filamentary coils in stellarators?arXiv preprint arXiv:2602.18974, 2026

    work page arXiv 2026

  8. [8]

    Paul, and William Dorland

    Arthur Carlton-Jones, Elizabeth J. Paul, and William Dorland. Computing the shape gradient of stellarator coil complexity with respect to the plasma boundary.Journal of Plasma Physics, 87(2):905870222, 2021. doi: 10.1017/S0022377821000386

    work page doi:10.1017/s0022377821000386 2021

  9. [9]

    S. A. Cadena, A. Merlo, E. Laude, A. Bauer, A. Agrawal, M. Pascu, M. Savtchouk, E. Guiraud, L. Bonauer, S. Hudson, and M. Kaiser. ConStellaration: A dataset of QI-like stellarator plasma boundaries and optimization benchmarks.Advances in Neural Information Processing Systems (NeurIPS), 2025. HuggingFace dataset https://huggingface.co/datasets/proxima-fusi...

    work page 2025

  10. [10]

    S. A. Cadena, A. Merlo, E. Laude, A. Bauer, A. Agrawal, M. Pascu, M. Savtchouk, E. Guiraud, L. Bonauer, S. Hudson, and M. Kaiser. ConStellaration: A dataset of QI-like stellarator plasma boundaries and optimization benchmarks.arXiv preprint arXiv:2506.19583, 2025

    work page arXiv 2025

  11. [11]

    Landreman, B

    M. Landreman, B. Medasani, F. Wechsung, A. Giuliani, R. Jorge, and C. Zhu. SIMSOPT: A flexible framework for stellarator optimization.Journal of Open Source Software, 6(65):3525, 2021

    work page 2021

  12. [12]

    Gil, Weiping Li, Julianne Stratton, Alan A

    Pedro F. Gil, Weiping Li, Julianne Stratton, Alan A. Kaptanoglu, and Eve V. Stenson. Efficient computation of stellarator coils with an augmented Lagrangian optimization method.Physical Review E, Apr 2026. doi: 10.1103/838f-vbq3

    work page doi:10.1103/838f-vbq3 2026

  13. [13]

    P. Merkel. Solution of stellarator boundary value problems with external currents. Nuclear Fusion, 27(5):867, may 1987. doi: 10.1088/0029-5515/27/5/018. URL https://doi.org/10.1088/0029-5515/27/5/018

    work page doi:10.1088/0029-5515/27/5/018 1987

  14. [14]

    Estimating coil features from an equilibrium

    Eduardo Rodríguez and Wrick Sengupta. Estimating coil features from an equilibrium. arXiv preprint arXiv:2604.12339, 2026. doi: 10.48550/arXiv.2604.12339

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2604.12339 2026

  15. [15]

    Rodríguez, G.G

    E. Rodríguez, G.G. Plunk, and R. Jorge. Near-axis description of stellarator-symmetric quasi-isodynamic stellarators to second order.Journal of Plasma Physics, 91(2):E59,

    work page

  16. [16]

    doi: 10.1017/S0022377825000157

    work page doi:10.1017/s0022377825000157

  17. [17]

    Near-axis measures of quasi-isodynamic configurations.Journal of Plasma Physics, 91(4):E112, 2025

    Eduardo Rodríguez and Gabriel Plunk. Near-axis measures of quasi-isodynamic configurations.Journal of Plasma Physics, 91(4):E112, 2025. doi: 10.1017/ S0022377825100627. 33

    work page 2025

  18. [18]

    Theproofandmeasurementofassociationbetweentwothings.American Journal of Psychology, 15:72–101, 1904

    C.Spearman. Theproofandmeasurementofassociationbetweentwothings.American Journal of Psychology, 15:72–101, 1904

    work page 1904

  19. [19]

    Geurts, D

    P. Geurts, D. Ernst, and L. Wehenkel. Extremely randomized trees.Machine Learning, 63(1):3–42, 2006

    work page 2006

  20. [20]

    Back to the figure-8 stellarator.Plasma Physics and Controlled Fusion, 67(3):035025, feb

    G G Plunk, M Drevlak, E Rodríguez, R Babin, A Goodman, and F Hindenlang. Back to the figure-8 stellarator.Plasma Physics and Controlled Fusion, 67(3):035025, feb

    work page

  21. [21]

    URLhttps://doi.org/10.1088/1361-6587/ adb64b

    doi: 10.1088/1361-6587/adb64b. URLhttps://doi.org/10.1088/1361-6587/ adb64b

    work page doi:10.1088/1361-6587/adb64b

  22. [22]

    Eduardo Rodriguez and Gabriel G. Plunk. Near-axis quasi-isodynamic database, 2026. URLhttps://arxiv.org/abs/2601.08400. 34

    work page internal anchor Pith review Pith/arXiv arXiv 2026