{"id":"0d338964-a636-4e3e-bdfd-9324a4357bbd","arxiv_id":"2412.00267","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":7,"one_line_summary":"A wireframe mesh of current-carrying segments, optimized by regularized least squares or a greedy loop-adding algorithm, enables sparse stellarator coil designs with arbitrary spatial constraints.","lead":"This paper introduces a computational framework for designing stellarator magnets, in which the design space is a mesh of wire segments and currents are optimized with two new methods, including a greedy discrete algorithm. The framework allows coils to be excluded from specific regions, such as ports, and yields sparse layouts that are easier to assemble.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The field-error objective fB does not penalize resonant error fields; Appendix C shows a slightly better fB with worse flux surfaces, so the reported metrics do not establish that GSCO outputs are useful starting points.","rationale":"After reviewing the wireframe formulation (Sec. II), RCLS (Sec. III), and GSCO (Sec. IV), the central claim hinges on the usefulness of the produced current distributions as starting points for coil design. The paper demonstrates low surface-averaged normal-field error, but Appendix C explicitly shows that a slightly lower such error can coexist with degraded flux surfaces due to a resonant perturbation. This is a direct threat to the claim because the stated purpose of the wireframe solutions is to serve as starting points for convenient coil designs; a starting point with hidden resonant errors may not be convenient. The paper acknowledges the issue and suggests repeating optimizations with varied geometry, but does not resolve it. In contrast, the lack of a benchmark against existing coil optimization methods is a weaker concern for the technical validity of the framework; it affects the novelty claim but not the internal correctness. We therefore focus on the metric-alignment issue. The proposed test directly checks whether the observed failure is due to the objective function (fixable) or the fixed-wireframe parameterization (structural). We agree with the reader's identification of the fixed-geometry sensitivity as a key limitation, though we emphasize the objective's insensitivity to resonant errors as the underlying mechanism. The verdict remains CONDITIONAL: the framework is promising and the examples are informative, but robustness to wireframe geometry and metric choice must be addressed before the strongest claims are taken at face value.","tokens_in":24666,"tokens_out":7290,"duration_ms":64628,"concrete_test":"Recompute the Appendix C comparison with the GSCO objective augmented by a penalty on the resonant Fourier harmonics of the normal field (e.g., the m/n components matching the rotational transform on the boundary), using the same two wireframes and hyperparameters. If the penalized objective produces a solution on the modified wireframe with flux surfaces comparable to the original wireframe, the concern is the objective's blindness to resonant errors rather than a structural limitation of the wireframe. If the resonant error persists, the fixed wireframe geometry itself is the limiting factor.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that GSCO yields useful starting points for convenient coil designs rests on the assumption that the optimized current distributions actually confine the target plasma. This assumption is not guaranteed by the objective being minimized. The objective fGSCO (Eq. 17) uses fB, the surface-averaged squared normal-field error (Eq. 10), plus a sparsity term. Appendix C provides a concrete counterexample: changing only the poloidal spacing of wireframe nodes yields a solution with a slightly better metric ⟨|B·n|/|B|⟩ = 2.18e-3 versus 2.44e-3, yet the flux surfaces are degraded by a resonant error that deforms the boundary. Because fB is an integral quantity, it can be insensitive to resonant Fourier harmonics of the error field. Thus the reported field-accuracy metrics do not by themselves establish that a GSCO solution is a viable starting point; the fixed wireframe geometry, chosen a priori, can determine whether the method finds a good solution. This is a load-bearing caveat to the claim of 'convenient coil designs,' and it is only partially mitigated by the paper's suggestion to repeat optimizations with varied wireframe geometry.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces a 'wireframe' framework for stellarator coil design, in which the design space is a fixed mesh of straight current-carrying segments enclosing the plasma. Two optimization methods are developed: Regularized Constrained Least Squares (RCLS), a linear least-squares method with equality constraints and Tikhonov regularization, and Greedy Stellarator Coil Optimization (GSCO), an iterative greedy method that adds discrete current loops to the wireframe one at a time. The framework is demonstrated on the Precise QA equilibrium of Ref. [34], producing RCLS solutions with surface-averaged relative normal fields around 6e-4 and GSCO solutions that range from dense saddle-coil distributions to sparse modular-coil sets, including cases with spatial constraints for ports or toroidal sectors and a multi-current solution built from a sequence of GSCO runs. The appendices provide a derivation of the Biot-Savart formula for a straight segment and a derivation of the RCLS solution procedure via QR factorization.","tokens_in":24841,"tokens_out":6140,"duration_ms":57962,"significance":"If the results hold, the wireframe framework is a valuable addition to the stellarator coil design toolbox. Its spatially local parametrization makes it straightforward to enforce arbitrary spatial restrictions on the current distribution, which is more difficult with Fourier-based winding-surface or space-curve parameterizations. The GSCO algorithm is a fully discrete alternative to continuous coil optimization, capable of producing sparse and topologically flexible coil sets that could serve as starting points for further refinement. The paper is notable for shipping open-source implementations (in SIMSOPT) and preserving reproducibility via a data DOI. The formal derivations in Appendices A and B are correct and clearly presented. The main weakness is the reliance on the surface-averaged normal-field error as the primary accuracy metric, which Appendix C itself shows can be misleading for flux-surface quality; this affects the strength of the central claim that GSCO solutions are useful starting points for convenient coil designs.","major_comments":[{"comment":"The GSCO modular coil solutions in Sec. IV B are assessed only through the surface-averaged relative normal field ⟨|B·n|/|B|⟩, with no Poincaré sections or other independent checks of flux-surface integrity for the solutions in Fig. 9b-d. Appendix C demonstrates that this metric is not a reliable proxy for confinement: a solution with ⟨|B·n|/|B|⟩ = 2.18e-3 (modified wireframe) has worse flux surfaces than one with 2.44e-3 (original wireframe) because of a resonant error. Since the paper's central claim is that GSCO yields 'starting points for convenient coil designs,' the absence of independent flux-surface checks for the main GSCO examples is a load-bearing gap. I request that Poincaré sections (or equivalent measures of rotational-transform and island structure) be provided for the solutions in Fig. 9b-d, or that the claims be explicitly limited to the field-error metric.","section":"Section IV B, Fig. 9; Appendix C"},{"comment":"The paper's framing that the wireframe framework 'enables the attainment of sparse magnet solutions with arbitrary restrictions on where coils may be located' is stronger than what the evidence supports. Appendix C shows that the quality of GSCO solutions is highly sensitive to the fixed wireframe geometry; a small change in poloidal node spacing yields a solution with a slightly better fB but a resonant error that deforms the flux surfaces. This means the method does not by itself guarantee useful sparse solutions for arbitrary spatial constraints; it only does so for favorable a priori choices of the wireframe geometry. The suggestion to 'repeat optimizations with slight variations in wireframe geometry' is a workaround, not a resolution. I recommend that the abstract and introduction be qualified to state that GSCO can produce sparse solutions for these example cases, with the caveat that the fixed wireframe geometry is an essential constraint that must be chosen carefully.","section":"Introduction and Section V; Appendix C"}],"minor_comments":[{"comment":"In the second paragraph, 'encompasing' is a typo; it should be 'encompassing'.","section":"Introduction"},{"comment":"The color-scale labels in the right column of Fig. 9 appear to be cut off in the manuscript (e.g., 'mean: . × 10 4'); please ensure the full numerical values are visible in the final figures.","section":"Fig. 9"},{"comment":"The input parameter 'Ncells,min' is used in the pseudocode but is never defined in the text; please add a definition, e.g., in the caption or in the surrounding paragraph of Sec. IV D.","section":"Algorithm 2"},{"comment":"The text states that the solution in Fig. 12 was run with λS = 10^-7.5 T^2 m^2, but the figure caption does not list the hyperparameters; adding them would improve reproducibility.","section":"Section IV C and Fig. 12"},{"comment":"The QR factorization is written for C^T, but in the text R is described as upper triangular; the conventions for the dimensions of R in Eq. (B1) appear inconsistent with the later use of forward substitution on R^T. Please clarify the indexing or state that the factorization is applied to C (not C^T) if that resolves the inconsistency.","section":"Appendix B, Eq. (B1)"}],"recommendation":"major_revision","confidential_remarks":"The paper is a solid methods contribution with clear derivations and a reproducible open-source implementation. The primary concern, which I share with the stress-test reviewer, is that the central claim about GSCO's practical value rests on a metric that Appendix C itself shows can fail to capture resonant flux-surface degradation. The requested additions (Poincaré checks for Fig. 9 and a qualified claim) are well within the manuscript's scope, so I do not recommend rejection. I also note that the citation and prior-work coverage is appropriate, and the fit to the journal is good."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Read this one if you work on stellarator coil design. The wireframe framework is a clean way to discretize the coil design space, and the greedy algorithm (GSCO) is genuinely new. The math is solid, the code and data are public, and the examples show something useful: you can enforce arbitrary spatial constraints, like leaving room for ports or confining all coils to toroidal sectors.\n\nThe paper does two things well. First, RCLS is a spatially localized relative of REGCOIL—the author says so—and it produces accurate fields on coarse grids. Second, GSCO is a discrete greedy loop-adding algorithm inspired by greedy permanent-magnet methods; it can reshape existing coils, add saddle coils, and produce sparse solutions with different current levels. The multistage example with external TF coils is a nice touch.\n\nThe main soft spot is the field-accuracy metric. Throughout, the paper quotes ⟨|B·n̂|/|B|⟩, a surface-averaged normal field error. That quantity is largely insensitive to resonant Fourier harmonics, and Appendix C shows exactly this: a small change in wireframe geometry gives a slightly better metric (2.18e-3 vs 2.44e-3) but visibly worse flux surfaces. The paper discloses this and suggests repeating optimizations with varied geometry, which is reasonable, but the abstract's claim that examples 'achieve high field accuracy' should be read with that caveat. The reported metrics don't guarantee confinement; the Poincaré plots do most of that work.\n\nThe second weakness is the lack of a baseline comparison. The introduction claims sparse solutions with spatial constraints 'could not be easily found using previous methods,' but no comparison to REGCOIL, FOCUS, or current-potential patches is presented. That claim may be true, but it isn't demonstrated. Adding a benchmark would strengthen the paper considerably.\n\nThe fixed wireframe geometry is also a structural limitation: the solution space is restricted to the chosen surface, and if that surface can't represent the needed currents, the method can't find them. The paper acknowledges this explicitly, so it's not a hidden flaw.\n\nWho this is for: fusion coil designers, especially those working on stellarator magnet optimization. It's a subfield methods paper, not a resolution of a long-standing physics question—but it's a useful one.\n\nMy recommendation: send it to peer review. A good referee should ask for a baseline comparison and a more prominent discussion of resonant-error sensitivity, but the core framework and algorithm are worth publishing.","headline":"Genuinely new wireframe and greedy algorithm for stellarator coil design, with solid math and public code; the main caveat is that the standard field-error metric can miss resonant errors, as Appendix C shows.","tokens_in":25431,"tokens_out":3711,"would_cite":true,"duration_ms":33525,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Sparse stellarator coil designs can be optimized on a fixed wireframe mesh, either by constrained least squares or by adding current loops one at a time.","keywords":["stellarator coil design","wireframe current distribution","greedy optimization","constrained least squares","sparse solutions","magnetic field shaping","spatial constraints"],"falsifier":"Re-run the RCLS example of Section III B on the same Precise QA equilibrium with the wireframe nodes moved from the conformal offset surface (~0.3 m away) to a sphere of radius 1.3 m centered on the plasma. If $\\langle |B\\cdot\\hat{n}|/|B| \\rangle$ stays below $10^{-3}$, the fixed wireframe geometry is not the limiting assumption; if it rises above $10^{-2}$ or the Poincaré cross-sections show broken flux surfaces, the method's usefulness depends on choosing a good wireframe surface rather than on the optimizer alone.","tokens_in":24351,"feed_emoji":"🧲","tokens_out":8233,"duration_ms":69519,"temperature":0.7,"pith_summary":"This paper introduces a way to design stellarator magnets by placing a fixed mesh of straight current-carrying segments around the plasma and optimizing only the current in each segment. The mesh, called a wireframe, makes it trivial to forbid currents in any chosen region, so coils can be routed around ports or confined to assembly-friendly sectors. The paper proposes two optimizers: a fast regularized constrained least-squares solver and a greedy algorithm that adds one small current loop at a time. Example solutions reach surface-averaged relative normal fields below $10^{-3}$, and the greedy solutions are sparse enough to serve as starting points for conventional smooth-coil refinement.","feed_headline":"Greedy optimizer designs sparse stellarator coils one loop at a time","feed_subtitle":"A fixed wireframe mesh turns coil design into a choice of segment currents, leaving room for ports and easy assembly.","key_machinery":"The central object is the wireframe: a fixed toroidal mesh of interconnected straight segments, each carrying a current $x_j$. Its field at plasma-boundary test points is linear in the segment currents (Biot-Savart, Eqs. 1-3), so the entire problem reduces to choosing a vector $\\mathbf{x}$ under linear equality constraints (current continuity, net poloidal/toroidal current, zero-current segments). RCLS solves this as a regularized constrained least-squares problem; GSCO instead adds a unit current loop around individual mesh cells one at a time, with interior shared segments canceling so that adjacent loops merge into a saddle coil or reshape an existing coil. The loop addition is what carries the discrete greedy mechanism, and the linearity of the field in $\\mathbf{x}$ is what makes the greedy one-step selection cheap to evaluate.","core_discovery":"On its own terms, the paper claims that a wireframe—a toroidal mesh of straight filamentary segments, with currents as the only decision variables—is a useful solution space for stellarator coil design because it combines the spatial locality of permanent-magnet arrays with the field-shaping power of surface current distributions. The central new result is that a fully discrete greedy procedure (GSCO) can build coil-like current paths by adding single-cell loops one at a time, selecting at each step the loop that most reduces the combined field-error and sparsity objective; because shared segment currents cancel between adjacent loops, the added loops assemble into saddle coils or reshape existing modular coils. With this procedure, and with the linear RCLS solver, the paper demonstrates solutions that achieve field accuracies around $\\langle|B\\cdot\\hat{n}|/|B|\\rangle\\sim 10^{-3}$ or better while obeying hard spatial constraints such as blocked ports or toroidal sectors.","pith_inferences":["The stop condition for GSCO—the best next loop cancels the previous loop—is a local-minimum signal that resembles matching-pursuit algorithms; one could analyze the gap between GSCO solutions and the unconstrained least-squares optimum using greedy suboptimality bounds, a question the paper leaves open.","Because the wireframe is independent of the optimizer, the same mesh could host a combined optimizer that co-optimizes node positions and currents, or one that uses triangular or volumetric cells; the paper notes this as a next step, and the sensitivity result in Appendix C makes node-position co-optimization a natural extension.","The resonant-error sensitivity identified in Appendix C suggests an immediate testable improvement: adding a penalty term for resonant Fourier components of the normal field to $f_{\\mathrm{GSCO}}$ could restore flux-surface quality without changing the wireframe geometry.","A wireframe loop is the current analogue of a permanent-magnet dipole block, so a hybrid design that places permanent magnets in some regions and wireframe currents in others could exploit both local parameterizations; this is not explored in the paper."],"forward_implications":["A fixed wireframe can turn coil design into a linear least-squares problem, so highly accurate vacuum fields for a given equilibrium can be produced in about 100 ms on a laptop, enabling fast scans over equilibria or port layouts.","The same wireframe can be initialized with planar poloidal loops and reshaped by GSCO into modular coils, saddle coils, or mixtures of both, so the optimizer, not the parameterization, decides the coil topology.","Spatial restrictions are enforced by setting selected segment currents to zero, which means designs can reserve space for ports, maintenance access, or toroidal-sector assembly without changing the optimization algorithm.","GSCO output paths contain sharp corners and filamentary junctions, but they give a spline- or space-curve optimizer a concrete starting point with the coil count and rough geometry already determined.","A multistage variant of GSCO adds coils at successively halved current levels, allowing designs whose coils carry different currents rather than forcing a single current value."],"supporting_citations":[{"why":"Supplies the regularized linear least-squares current-potential formulation that RCLS extends to spatially local segment currents.","marker":"[2]"},{"why":"Introduces the winding-surface/current-potential method whose toroidal topology the wireframe generalizes.","marker":"[5]"},{"why":"Provides the precise quasisymmetry equilibrium used in all numerical examples in the paper.","marker":"[34]"},{"why":"The greedy dipole-array optimizers that add one magnetic source at a time; GSCO adapts their strategy to current loops.","marker":"[24–27]"},{"why":"Introduces current potential patches, the surface-current analogue to which a wireframe loop is a discrete counterpart.","marker":"[28]"},{"why":"Shows how small resonant field errors can destroy flux surfaces, the sensitivity mechanism behind the Appendix C geometry comparison.","marker":"[52]"}],"fun_headline_variants":["Wireframe mesh turns coil design into a greedy loop hunt","Discrete coil optimizer picks loops one by one for fusion magnets","Sparse stellarator coils from a wireframe: greedy wins","Greedy loop addition designs assembly-friendly stellarator coils","Wireframe coils: discrete optimization, easy assembly, high accuracy"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"All optimizations keep the wireframe geometry fixed, and the achievable field quality depends strongly on that preselected mesh; Appendix C shows a small change in poloidal node spacing creates a resonant error that deforms flux surfaces even when the averaged field error is slightly better.","fun_headline_variants_meta":{"raw":{"variants":["Wireframe mesh turns coil design into a greedy loop hunt","Discrete coil optimizer picks loops one by one for fusion magnets","Sparse stellarator coils from a wireframe: greedy wins","Greedy loop addition designs assembly-friendly stellarator coils","Wireframe coils: discrete optimization, easy assembly, high accuracy"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00017,"raw_usage":{"total_tokens":1254,"prompt_tokens":916,"completion_tokens":338,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":532,"completion_tokens_details":{"reasoning_tokens":254}},"tokens_in":532,"tokens_out":338,"duration_ms":3487,"temperature":1.0,"reasoning_tokens":254,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T05:32:50.721012+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Re-run the RCLS example of Section III B on the same Precise QA equilibrium with the wireframe nodes moved from the conformal offset surface (~0.3 m away) to a sphere of radius 1.3 m centered on the plasma. If $\\langle |B\\cdot\\hat{n}|/|B| \\rangle$ stays below $10^{-3}$, the fixed wireframe geometry is not the limiting assumption; if it rises above $10^{-2}$ or the Poincaré cross-sections show broken flux surfaces, the method's usefulness depends on choosing a good wireframe surface rather than on the optimizer alone.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the regularized linear least-squares current-potential formulation that RCLS extends to spatially local segment currents."},{"cited_title":"Imbert-G ´erard, E","cited_arxiv_id":null,"evidence_quote":"Introduces the winding-surface/current-potential method whose toroidal topology the wireframe generalizes."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the precise quasisymmetry equilibrium used in all numerical examples in the paper."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Introduces current potential patches, the surface-current analogue to which a wireframe loop is a discrete counterpart."},{"cited_title":"Frerichs, FLARE: field line analysis and reconstruction for 3D boundary plasma modeling (2024)","cited_arxiv_id":null,"evidence_quote":"Shows how small resonant field errors can destroy flux surfaces, the sensitivity mechanism behind the Appendix C geometry comparison."}],"review_version":1}