{"id":"2a24a7e6-e073-4597-b3fc-416cb5099d3c","arxiv_id":"2601.08400","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":7,"one_line_summary":"A public database of over 800,000 near-axis quasi-isodynamic stellarator configurations with computed stability, transport, and coil-complexity proxies, plus statistical heuristics for design.","lead":"The authors used a fast near-axis model to generate a database of more than 800,000 quasi-isodynamic stellarator configurations, computing cheap proxies for stability, transport, and coil distance. The database gives designers and machine-learning tools a systematic map of this magnetic-confinement design space.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Maximum-J fractions drop the first-order buffer drift that the text says dominates near the axis; no radius crossover is given, so f_J-based heuristics may be unsupported.","rationale":"The reader's weakest assumption already flags the omission of first-order buffer terms in f_J and ε_eff; I agree that this class of omission is the most load-bearing issue. My stress-test sharpens it to the f_J calculation because Appendix D explicitly admits the dropped term dominates close to the axis and gives f_J ≈ 0.5 there, yet the main text uses only the omnigenous part. Since f_J is one of the four central physics measures and feeds the conclusion that maximum-J can be achieved at all N, a systematic bias in f_J would change a headline heuristic. The paper is otherwise internally consistent, and the database itself remains a valuable near-axis resource, so this concern does not justify rejection. It does justify retaining the reader's CONDITIONAL verdict: the authors should either quantify r* or recompute f_J with the full Eq. (D1). I therefore leave the verdict unchanged rather than moving it, but the verification should be a stated condition.","tokens_in":40406,"tokens_out":10899,"duration_ms":110573,"concrete_test":"Take a stratified sample of ≥1000 database configurations spanning N=2–6 and the top/bottom f_J deciles. At the paper's chosen reference radius r_ref, compute f_J via Eq. (3.6) using the full ω_α of Eq. (D1) (including ω^non-QI_α,-1 cos α / r) and compare with the main text's Eq. (3.8) approximation. Report the distribution of r* = |ω^non-QI_α,-1|/|ω_α,vac| and the resulting Δf_J. If the median |Δf_J| exceeds ~0.05, or r* > 0.1 r_ref for more than 10% of the sample, the maximum-J statistics and heuristics in Sec. 3.3 need revision.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Appendix D writes the full precession as ω_α = ω^non-QI_α,-1(λ) cos α / r + ω_α,vac(λ) + ..., then states that the 1/r term 'dominate[s] precession sufficiently close to the magnetic axis, expecting f_J ≈ 0.5 there.' The main text nevertheless evaluates f_J using Eq. (3.8), i.e. ω_α ≈ ω_α,vac, dropping this first-order buffer term. No estimate is given for the crossover radius r* = |ω^non-QI_α,-1|/|ω_α,vac|, nor is r* compared with the reference radius r_ref used in Eq. (3.6). Since f_J integrates with weight r dr from 0 to r_ref, the inner region contributes substantially unless r* ≪ r_ref. If that condition fails, the reported f_J maxima (Table 6: 0.86–0.91) are biased upward, and the Sec. 3.3 heuristics — e.g. that maximum-J is achievable at all N without excessive shaping — may be artifacts. This is an internal correctness risk, independent of whether near-axis fields survive global reconstruction.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript constructs a database of more than 800,000 near-axis quasi-isodynamic (QI) stellarator vacuum configurations, built from the authors' previously developed second-order near-axis construction with a restricted set of input parameters (axis torsion/curvature coefficients, elongation coefficients, mirror ratio, well-width parameter, and field period number). Each configuration is evaluated with near-axis measures: magnetic gradient scale L_∇B, critical aspect ratio A_mhd^c for a marginal vacuum well, maximum-J fraction f_J, effective ripple ε_eff, Shafranov shift sensitivity, and finite-orbit-width measures. The authors apply statistical tools (correlation, FSFS, PFI/cSAGE, clustering) to identify design heuristics, e.g. that low N favours coil compatibility and low ripple, high N favours high rotational transform and low Shafranov shift, and that maximum-J can be achieved at all N. The database is publicly released.","tokens_in":40770,"tokens_out":7092,"duration_ms":69712,"significance":"If the reported construction and diagnostics hold up, this is a valuable open resource: it maps a substantial, if deliberately restricted, part of near-axis QI design space and provides a large set of initial conditions for optimisation. The construction is internally careful — axis closure and first-order realness/periodicity are checked, and many measures are computed from parameter-free near-axis formulas rather than fitted to targets. The statistical analysis is thorough and reproducible in principle. The main weakness is the f_J computation, which omits a first-order non-omnigenous term that the manuscript itself says dominates near the axis; this could bias the maximum-J heuristics. The abstract's 'stable' also overstates the marginal W=0 vacuum-well nature of the representative configurations.","major_comments":[{"comment":"The f_J computation uses only the O(r^0) omnigenous precession term ω_{α,vac} in Eq. (3.8), dropping the first-order non-QI term ω^{non-QI}_{α,-1}(λ) cos α / r shown in Eq. (D1). Appendix D itself states that this 1/r term 'dominate[s] precession sufficiently close to the magnetic axis, expecting f_J ≈ 0.5 there.' No estimate is given for the crossover radius r* = |ω^{non-QI}_{α,-1}|/|ω_{α,vac}|, nor is r* compared with the reference radius r_ref entering Eq. (3.6). Since Eq. (3.6) integrates with weight r dr from 0 to r_ref, the inner region contributes non-negligibly unless r* ≪ r_ref. If that condition is not met, the reported f_J maxima (Table 6: 0.86–0.91) are biased upward, and the conclusion in §3.3.2 and §4 that maximum-J is achievable at all N without excessive shaping may be an artifact. Please quantify r* for representative configurations across the database, or recompute f_J","section":"Appendix D, §3.3, Eq. (D1) and Eq. (3.6)-(3.8), Table 6"},{"comment":"The abstract and introduction describe the database as containing 'stable ... configurations', but the representative chosen for each first-order configuration is the marginal W=0 vacuum-well case (§2: 'marginal representative, W=0'; §3.2: 'marginally stable construction'). A vanishing vacuum well is a marginal condition, not a stable one, and A_mhd^c as defined in Eq. (3.3) is the aspect ratio at which the near-axis construction first breaks, not a demonstrated stability boundary. The 'stable' wording should be qualified as 'marginally stable (W=0)' in the abstract and wherever the database is summarised, to avoid overstating the physical content.","section":"Abstract, §1, §2, §3.2"}],"minor_comments":[{"comment":"The acronym SVR is used inconsistently: 'Statistical Vector Regression' in Appendix B.2, 'Statistical Virtual Regression' in the main text and later in Appendix B. Please use the standard 'Support Vector Regression' throughout.","section":"Appendix B and §3.1.2"},{"comment":"The integral limits 'Z 1/Bmin over 1/Bmax' are difficult to read; the notation should be cleaned up (likely ∫_{1/Bmax}^{1/Bmin}).","section":"Eq. (3.11)"},{"comment":"The superscript/subscript notation for maxima and minima (e.g. '0.43^0.12_0.27') is non-standard and hard to parse. Please define the notation explicitly in the caption or use a clearer format.","section":"Table 6"},{"comment":"The statement that the first-order buffer contribution ε_{eff}^{3/2,(0)} 'may be in practice ignored' would benefit from a quantitative justification — for example, citing the values in Table 6, which appear to be several orders of magnitude smaller than the ε_edge values used in the 'good' configuration criteria.","section":"§3.4.1"}],"recommendation":"major_revision","confidential_remarks":"The f_J issue is the main technical obstacle: the manuscript's own appendix admits the omitted 1/r term dominates near the axis, and no crossover estimate is provided. I would require a quantitative r* analysis or a revised f_J calculation before publication. The 'stable' wording in the abstract should also be corrected to 'marginally stable at W=0'. The database itself is a promising resource and the remainder of the analysis is careful, so this is fixable within a revision."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"First thing to know: this is a real resource. 800k+ near-axis QI vacuum fields, openly available, built on the authors' own second-order near-axis construction, with each configuration scored on a battery of physics measures. No comparable QI database exists; the quasisymmetric ones (Landreman, Giuliani et al.) are different beasts. If you work on stellarator design space or ML surrogates, this is the dataset you've been missing.\n\nWhat's good: the construction is careful. They check axis closure and realness/periodicity of first-order solutions, scan over a non-rectangular valid region, and publish the generation scripts. The physics measures are clearly defined from stated near-axis formulas, not fitted to outcomes. The statistical analysis is standard but thorough—linear/nonlinear correlation, FSFS, PFI/cSAGE, clustering—and the heuristics that come out (torsion at B_min sets coil-approximation scale; low integrated torsion helps MHD stability; low N is easier for coil/stability/ripple, high N helps Shafranov shift and orbit-width measures) are coherent with the theory. The 'good configuration' thresholds and the finding that no N=6 makes the cut are useful baselines.\n\nSoft spots, in order of severity.\n\n1. The f_J calculation has an internal gap. Appendix D writes the full precession with a 1/r buffer term and says it dominates close to the axis, pushing f_J≈0.5 there. The main text computes f_J from Eq. (3.8), which drops that term, and gives no estimate of the radius where the second-order vacuum term takes over. Since the f_J integral weights r dr, the inner region contributes proportionally to (r*/r_ref)^2. If r* is not small relative to r_ref, the reported maxima 0.86–0.91 are biased upward and the Sec. 3.3 claim that maximum-J is achievable at all N without excessive shaping is unsupported. This is a missing quantitative estimate, not a dead end; the authors likely have the machinery to compute r*. But as written, that part of the paper is not fully specified.\n\n2. The abstract says 'stable' but the stability is only marginal W=0 vacuum-well (Mercier limit) within near-axis. The text is more careful. Wording fix.\n\n3. Everything is near-axis. The authors acknowledge this and frame the database as a starting point for global optimization and testbed, not a set of finite-β equilibria. That's a stated limitation, not a hidden one. The resource stands on that basis.\n\nWho's it for: stellarator theorists doing design-space exploration, and people building ML surrogates for QI configurations. The paper deserves peer review. The f_J issue should be raised in review but doesn't sink the database or the other heuristics.\n\nRecommendation: send it out, with a request for a crossover-radius estimate or a range-restricted f_J definition.","headline":"A genuinely new 800k-configuration QI near-axis database, carefully constructed and open; the f_J section has an internal consistency gap that needs addressing before the heuristics are trusted.","tokens_in":41171,"tokens_out":2740,"would_cite":true,"duration_ms":27495,"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":"This paper builds a database of over 800,000 quasi-isodynamic stellarator configurations and extracts design heuristics from it.","keywords":["quasi-isodynamic stellarators","near-axis expansion","stellarator database","stellarator design heuristics","effective ripple","MHD stability","maximum-J","magnetic gradient scale"],"falsifier":"Take a subset of the database's 'good' configurations (e.g., the N=2 figure-8) and reconstruct them as global equilibria at the near-axis-predicted finite aspect ratio using an independent equilibrium solver. If flux surfaces break before the predicted critical aspect ratio, or if the calculated effective ripple including the first-order buffer terms exceeds the near-axis threshold by more than a factor of two, the central claim would be undermined.","tokens_in":40298,"feed_emoji":"🧲","tokens_out":5888,"duration_ms":55920,"temperature":0.7,"pith_summary":"The paper constructs a first-of-its-kind database of more than 800,000 approximately quasi-isodynamic (QI) vacuum stellarator configurations using the near-axis expansion. Each configuration is defined by a handful of geometric inputs and evaluated with a broad set of physics measures: magnetic gradient scale, critical aspect ratio for MHD stability, maximum-J fraction, effective ripple, Shafranov shift sensitivity, and more. The authors analyze this dataset statistically to uncover heuristics about how field-period number, torsion, curvature, and elongation control these properties. The result is a systematic map of a wide region of QI design space, together with baseline configurations for future optimization.","feed_headline":"One database maps 800,000 quasi-isodynamic stellarator designs","feed_subtitle":"Field period number and torsion are the levers that control coil distance, stability, and transport across QI design space.","key_machinery":"The machinery is the near-axis expansion to second order for quasi-isodynamic fields, with a specific parameterization: axis curvature and torsion with two flattening points per field period, a flattened on-axis field minimum, and prescribed flux-surface elongation. This yields a low-dimensional input space (eight scalars) that is exhaustively scanned. Diagnostics include L∇B, A_mhd^c, f_J, epsilon_eff, Shafranov shift sensitivity, and q_eff. Statistical tools—linear and non-linear correlations, permutation feature importance, conditional Shapley values, forward sequential feature selection, and clustering—identify the dominant geometric drivers.","core_discovery":"The paper claims that the near-axis expansion now enables a systematic, exhaustive mapping of quasi-isodynamic stellarator design space. Its construction parameterizes each configuration by eight scalars (axis curvature and torsion, on-axis field shape, flux-surface elongation), yielding a database of over 800,000 vacuum, approximately QI fields. Statistical analysis yields heuristics: the magnetic gradient scale is set mainly by torsion at the field minimum; MHD stability favors low integrated torsion and elongation stretching; maximum-J can exceed 80% at any field period; and effective ripple is the most restrictive measure. 'Good' configurations exist for N=1-5 but not N=6.","pith_inferences":["If the database's near-axis heuristics survive global reconstruction, they imply a design principle not explicitly stated in the paper: the torsion at the field minimum, rather than average torsion, is the master parameter for coil-plasma distance. This could be used as a fast analytic proxy for coil complexity in future stellarator codes.","A natural extension is to use the database to train a surrogate model that predicts epsilon_eff from the eight input scalars, enabling inverse design by gradient-based optimization directly in near-axis parameter space.","The observed clustering into 'flat' and 'crown' families suggests distinct local minima in the design landscape; a testable hypothesis is whether these families are topologically distinct in terms of axis self-linking number or writhe, which might explain why optimization tends to converge to one family or the other.","Because the database is limited to half-helicity, flattening class (2,3), and second order near-axis, the absence of 'good' N=6 configurations is not a fundamental bound; extending the parameterization (e.g., more Fourier harmonics in curvature/torsion or a different helicity) could plausibly produce good high-N configurations, and the database's structure can guide where to search."],"forward_implications":["If the database is accurate, it provides thousands of baseline QI configurations that can serve as initial conditions for full-scale stellarator optimization, reducing the sensitivity of QI optimization to initial guesses.","The heuristics indicate that lower field period numbers are systematically better for coil compatibility, MHD-stability shaping, and low neoclassical transport, while higher N improves beta resilience and zonal-flow support—quantifying the trade-off behind the choice of intermediate N in reactor designs.","The figure-8-like configurations at N=2 stand out as the most compact stable configurations in the database, with aspect ratios around 2.3, offering a promising starting point for ultra-compact QI designs.","Maximum-J behaviour can be achieved approximately at all field period numbers in a vacuum, with f_J exceeding 80% for the best cases, suggesting that turbulence-suppressing configurations need not be limited to low N.","The effective ripple is the most restrictive 'goodness' criterion; no N=6 configurations satisfy all three thresholds, implying that transport-optimized high-N configurations require more flexible shaping than the current parameterization allows."],"fun_headline_variants":["Near-axis expansion yields 800K quasi-isodynamic designs","800K stellarator design database opens QI exploration","Systematic map of 800,000 quasi-isodynamic stellarators","Landscape of QI designs from near-axis parameters","Database of 800K configurations maps QI design space"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The entire database assumes that a vacuum field's near-axis second-order expansion accurately represents a real stellarator equilibrium at finite aspect ratio, and in particular that first-order 'buffer' deviations from omnigeneity can be ignored when computing the effective ripple; if these approximations fail, the rankings and heuristics could change.","fun_headline_variants_meta":{"raw":{"variants":["Near-axis expansion yields 800K quasi-isodynamic designs","800K stellarator design database opens QI exploration","Systematic map of 800,000 quasi-isodynamic stellarators","Landscape of QI designs from near-axis parameters","Database of 800K configurations maps QI design space"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000724,"raw_usage":{"total_tokens":3060,"prompt_tokens":701,"completion_tokens":2359,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":445,"completion_tokens_details":{"reasoning_tokens":2278}},"tokens_in":445,"tokens_out":2359,"duration_ms":16461,"temperature":1.0,"reasoning_tokens":2278,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T10:49:36.147661+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a subset of the database's 'good' configurations (e.g., the N=2 figure-8) and reconstruct them as global equilibria at the near-axis-predicted finite aspect ratio using an independent equilibrium solver. If flux surfaces break before the predicted critical aspect ratio, or if the calculated effective ripple including the first-order buffer terms exceeds the near-axis threshold by more than a factor of two, the central claim would be undermined.","supporting_citations":[],"review_version":2}