{"id":"bf2c0126-8806-4eb0-9036-2802fea8598a","arxiv_id":"1908.11615","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A fast, open-source orbit-averaged neoclassical solver that matches DKES and EUTERPE results for stellarators and adds tangential magnetic drift and flux-surface potential effects.","lead":"KNOSOS is an open-source code that rapidly computes how fast particles leak energy from twisted stellarator fusion devices using orbit-averaged drift-kinetic equations. It benchmarks against established codes on W7-X, LHD, NCSX and TJ-II, and can run parameter scans in seconds instead of hours.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The benchmark suite cannot validate the radial-locality ordering of Eq. (23); the W7-X edge discrepancy may mark its boundary, so the conditional verdict stands.","rationale":"The reader identified the same outer-envelope assumption: the radially-local ordering in Section 2, specifically inequality (23), is what justifies dropping the ∂ψgb term from Eq. (9). I agree that this is the most load-bearing concern because it defines the domain in which the code solves the correct physics. The benchmarks are internally consistent: KNOSOS reproduces DKES and EUTERPE when the same simplified equations are solved, and the speed advantage is real. But the benchmarks cannot test the local approximation itself, since DKES is also local and the EUTERPE comparisons are restricted to cases without tangential magnetic drift or with adiabatic electrons. The paper's own Fig. 11 shows a quantitative disagreement at the W7-X edge that is attributed to a physics mechanism (plateau-regime contribution) excluded from KNOSOS's equations, which is exactly the kind of boundary the local ordering controls. The manuscript is honest about the limitation, and the conditional verdict is appropriate. My proposed test would settle whether the concern is practical or merely formal: if the ratio in inequality (23) remains small for the benchmarked devices, the local approximation is well founded there and the code can be used with confidence in that envelope; if not, the dropped term is responsible for real errors and the W7-X edge result is a warning rather than an isolated benchmark discrepancy. I see no basis to move the verdict; the concern reinforces the existing CONDITIONAL rating.","tokens_in":34335,"tokens_out":3654,"duration_ms":37767,"concrete_test":"For each benchmark case (W7-X high-mirror, LHD Rax=3.75 m, NCSX, and TJ-II at the same surfaces and collisionalities, plus the W7-X edge surface of Fig. 11), evaluate the two sides of inequality (23) directly: compute the trapped-particle integrals using KNOSOS's own bounce-averaged coefficients, estimate ∂ψgb by finite differences from KNOSOS solutions at ψ±Δψ, and form the flux-weighted ratio R = |∫(dl/|v||) vD·∇ψ ∂ψgb| / |∫(dl/|v||) vD·∇α ∂αgb| over the contributing λ interval. If max R exceeds ~0.1 in any flux-contributing orbit family, the dropped term is not negligible and the local equation is not the governing one for that case; cross-check with a radially-global code (EUTERPE or SFINCS) at the same point to see whether discrepancies grow with R.","verdict_should_be":"UNCHANGED","load_bearing_attack":"KNOSOS's central claim is carefully qualified ('when solving equivalent equations'), and the DKES and EUTERPE comparisons do support that claim. The load-bearing unvalidated premise is the radially-local ordering represented by inequality (23): the bounce-averaged radial-drift term ∫dl vD·∇ψ ∂ψgb is dropped from Eq. (9) under assumptions of closeness to omnigeneity and large aspect ratio. Nothing in the paper quantifies this inequality for the four devices benchmarked. TJ-II and NCSX are not close to omnigeneity, LHD is a heliotron, and only W7-X is an optimized helias. Because DKES itself solves a radially-local equation, agreement with DKES cannot validate the neglected radial derivative; both codes would fail together if the ordering breaks down. The Fig. 11 W7-X edge underestimation of φ1, attributed to EUTERPE plateau-regime contributions, is consistent with the local approximation's boundary being reached in that benchmark. The paper explicitly states that the local approach limits applicability, but since KNOSOS is positioned as a general fast tool 'as long as the radially local approach is valid', the outer envelope of the benchmarked regimes needs explicit testing.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"KNOSOS is an open-source code that solves the radially local drift-kinetic equation for trapped particles together with the flux-surface quasineutrality equation in stellarator geometry, using bounce averages to reduce the equation to the variables (α, λ). The code retains the tangential magnetic drift and the flux-surface variation of the electrostatic potential φ1, solves the quasineutrality equation via a linear response-matrix method, and is benchmarked against DKES monoenergetic transport coefficients for W7-X, LHD, NCSX, and TJ-II, and against EUTERPE for φ1 in LHD and W7-X. The paper reports runtimes of seconds to minutes and claims that, when solving equivalent equations, KNOSOS reproduces DKES and EUTERPE calculations orders of magnitude faster.","tokens_in":34482,"tokens_out":7748,"duration_ms":74323,"significance":"If the benchmarks are taken at face value, KNOSOS is a significant practical tool for stellarator neoclassical transport: it makes low-collisionality calculations fast enough for parameter scans and optimization, includes two effects that DKES-style monoenergetic approaches drop (the tangential magnetic drift and the flux-surface electrostatic potential variation), and is released as open source. The response-matrix solution of quasineutrality is a clean and efficient idea, and the agreement with the independently developed DKES and EUTERPE codes in the simplified comparisons is strong evidence that the discretization and bounce-average implementation are sound. The paper is also careful to qualify its central claim with \"when solving equivalent equations\" and to acknowledge the radially local approximation, which is a genuine strength.","major_comments":[{"comment":"The radially local ordering that justifies dropping the ∂ψgb term is never quantified for any of the devices benchmarked in Section 4. Because DKES also solves a radially local equation, agreement with DKES cannot validate inequality (23); both codes would fail together if the ordering broke down. The paper should either compute the relative size of the omitted bounce-averaged radial-drift term versus the retained tangential term for the benchmarked cases, or explicitly state that the benchmarks validate only the numerical solution of the local equation and not the applicability of the local approximation itself. This matters because inequality (23) defines the code's domain of validity.","section":"Section 2, Eq. (23)"},{"comment":"The main physical novelty of KNOSOS, the tangential magnetic drift in Eq. (24), is not benchmarked against any independent code. In Section 4.2 the comparison is internal (Q versus \\hat Q, both computed by KNOSOS), and in Section 4.3 the EUTERPE comparisons use Eq. (52), which sets the tangential magnetic drift to zero; the right-hand columns of Figs. 10 and 11, which include this drift, have no independent reference. Please add a comparison with SFINCS or FORTEC-3D for at least one case where the tangential drift changes the result, or clearly label these predictions as not independently validated.","section":"Sections 4.2-4.3, Figs. 8-11"},{"comment":"The W7-X edge case shows a clear underestimation of φ1 by KNOSOS relative to EUTERPE. The attribution to EUTERPE's plateau-regime contribution is plausible, but it is not demonstrated, and if correct it indicates that the local trapped-particle equation (9) is not adequate at that radial position. The paper should quantify the plateau contribution, for example by varying the cutoff v0 in Eq. (58) or by comparing with a non-local model, and should state explicitly over what radial range the φ1 validation is reliable.","section":"Section 4.3, Fig. 11"},{"comment":"No grid-convergence study is reported. The benchmarks use fixed grids (Nα=32 and Nλ=64 or 128), and the text acknowledges that the lowest-collisionality TJ-II points in Fig. 6 (bottom right) are under-resolved and would have required a finer grid. Since the paper's speed claims (seconds per case) are only meaningful if the chosen grids are adequate, please add a convergence study over Nα, Nλ, and the Fourier truncation N for at least one representative configuration, and report the corresponding accuracy of the computed fluxes.","section":"Sections 3.4 and 4.1"}],"minor_comments":[{"comment":"The sentence \"This regime, which cannot be not described by a bounce-averaged drift-kinetic equation\" contains a typo and should read \"cannot be described\".","section":"Section 4.1, near Fig. 6"},{"comment":"The example in the text says that field lines are followed for \"6\" toroidal periods and refers to N/ι, but the rotational transform used in the example is not stated; please give the value of ι so the reader can verify the construction.","section":"Section 3.3, Fig. 2"},{"comment":"The stated number of Fourier coefficients, N = 2(2Nn+1)(Nm+1), appears inconsistent with the summation limits in Eq. (45), where -Nn<n<Nn and 0<m<Nm; please check the counting or define the limits to match the expression.","section":"Section 3.5, Eq. (45)"},{"comment":"A summary table listing the benchmark configurations, grid sizes, and runtimes would improve reproducibility; these values are currently scattered through the text and figure captions.","section":"Section 4"}],"recommendation":"major_revision","confidential_remarks":"For the editor: the paper is honest about its limitations and the qualified central claim is likely correct, but as a code paper it should validate its headline physics (the tangential magnetic drift) against an independent solver and delineate the radial-locality envelope more explicitly. I do not see circularity: DKES and EUTERPE are independent codes, and no fitted constants enter the benchmarks. The manuscript fits the journal's scope and the open-source release is a positive feature."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"KNOSOS is a real contribution: a fast orbit-averaged solver for neoclassical transport in stellarators, with the code open-source and benchmarks against DKES and EUTERPE on four different configurations. The theory was derived in earlier papers by the same group; what's new here is the numerical implementation, the grid construction, the response-matrix approach to quasineutrality, and the demonstration that it can be orders of magnitude faster.\n\nThe benchmarks are the strongest part. When solving the same simplified equations, KNOSOS matches DKES across W7-X, LHD, NCSX and TJ-II, including the 1/ν and √ν regimes, and matches EUTERPE for φ1 in the core. The timing claims are concrete: seconds per case where DKES-style databases take hours. The paper is honest about its limits—TJ-II low-collisionality points are under-resolved, and the W7-X edge φ1 comparison is not fully quantitative.\n\nThe main soft spot is the radially-local ordering behind Eq. (23). The paper drops the radial derivative of g under assumptions of closeness to omnigeneity and large aspect ratio, and the stress-test note is right: agreement with DKES cannot validate that dropping, since DKES is also radially local. TJ-II and NCSX are not close to omnigeneity, and the W7-X edge φ1 discrepancy may mark where the envelope breaks. The paper does qualify its claims with \"as long as the radially local approach is valid,\" so this is an unvalidated boundary rather than a contradiction. Still, the outer envelope of applicability is not benchmarked, and a reader who wants to use KNOSOS on a non-optimized device needs to know where that limit is.\n\nOther soft spots are minor: there is no grid-convergence study, and the treatment of some deep-trapped orbits involves models whose differences are said to be smaller than DKES error bars but are not shown. These are stated limitations, not hidden ones.\n\nThis paper deserves a serious referee. The code is open, the benchmarks are reproducible in principle, and the speedups are real for the regimes tested. I'd accept with minor revisions. For anyone working on stellarator neoclassical transport or design optimization, this is worth citing and worth having in the toolkit.","headline":"A solid, open-source code paper that delivers real speedups and honest benchmarks; the main caveat is that the radial-locality envelope of the method is not directly validated.","tokens_in":35131,"tokens_out":2516,"would_cite":true,"duration_ms":23034,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"KNOSOS computes stellarator neoclassical transport by orbit-averaging the drift-kinetic equation, reproducing benchmark fluxes orders of magnitude faster than standard local codes while retaining the tangential magnetic drift and…","keywords":["stellarator","neoclassical transport","drift-kinetic equation","orbit averaging","bounce-averaged coefficients","tangential magnetic drift","flux-surface electrostatic potential","quasineutrality"],"falsifier":"Take a strongly non-omnigenous stellarator configuration, such as one with large effective ripple and small aspect ratio, and compute the two bounce-averaged drift terms in inequality (23); if the radial-drift term is not much smaller than the tangential-drift term, then run KNOSOS and a radially global code with identical collision operators and profiles and compare the radial fluxes, because a significant disagreement beyond the DKES/EUTERPE benchmark error bars would show the local-ordering premise does not hold for that device.","tokens_in":34041,"feed_emoji":"🧲","tokens_out":4683,"duration_ms":46466,"temperature":0.7,"pith_summary":"This paper introduces KNOSOS, an open-source solver for neoclassical transport in low-collisionality stellarator plasmas. The central idea is to orbit-average the radially local drift-kinetic equation, which collapses the problem from five dimensions to an equation in just two variables: the field-line label and the pitch angle. This reduction is what makes the code orders of magnitude faster than standard local codes. KNOSOS keeps effects that DKES-style monoenergetic calculations drop, namely the component of the magnetic drift tangent to flux surfaces and the flux-surface variation of the electrostatic potential, and it solves quasineutrality efficiently because the equation is linear in that potential variation. The paper demonstrates agreement with DKES and EUTERPE when the same equations are solved, in simulations that can be orders of magnitude faster.","feed_headline":"Stellarator transport solver runs orders of magnitude faster","feed_subtitle":"Orbit-averaged KNOSOS matches DKES and EUTERPE benchmarks while keeping drift terms other codes drop.","key_machinery":"The load-bearing mechanism is orbit-averaging of the drift-kinetic equation. Bounce-integrals along field lines—I_{vM,α}, I_{vE,α}, I_{vM,ψ}, I_{vE,ψ}, and I_ν—absorb all dependence on the arc-length coordinate, leaving a differential equation in only the field-line label α and the pitch-angle λ. The radial coordinate ψ and speed v become parameters. Divergences of these integrals near trapping bifurcations are removed and evaluated analytically, and fast Fourier-based evaluation of the magnetic field along straight field lines accelerates the coefficient calculation. The linearity in φ₁ allows a response-matrix solution of quasineutrality, in which the drift-kinetic equation is solved once for each Fourier basis element and the factorization is reused, giving a speed-up of roughly the number of basis elements.","core_discovery":"The central claim is that a rigorously orbit-averaged, radially local drift-kinetic equation can capture the neoclassical transport regimes relevant to stellarators—1/ν, √ν, and superbanana-plateau—while including physics that older local codes omit. Specifically, the tangential magnetic drift and the radial E×B drift caused by the flux-surface variation of the electrostatic potential φ₁ are retained, and quasineutrality is solved consistently because the equations are linear in φ₁. When the same simplifications as DKES are imposed, KNOSOS reproduces DKES monoenergetic transport coefficients across four very different stellarator configurations; when the full tangential-drift terms are kept, it matches EUTERPE calculations of φ₁ and shows that the tangential magnetic drift can change the radial energy flux by more than 50% in ion-root conditions and can qualitatively alter the amplitude and phase of φ₁.","pith_inferences":["This is an inference beyond the paper: because the speed gain comes from eliminating the arc-length and radial-derivative dependence, the same bounce-averaged response-matrix strategy could be extended to trace impurity species, yielding systematic impurity-flux scans across configurations and collisionalities.","The benchmarks cover devices that are relatively close to omnigeneity; a natural testable extension would be to quantify, in a strongly non-omnigenous configuration, how much the dropped ∂ψg term corrupts fluxes as inequality (23) breaks down.","The fast computation of √ν and superbanana-plateau transport suggests a geometry optimization target based on the variation of the second adiabatic invariant on the flux surface rather than the effective ripple alone; the paper points in this direction but does not develop such a figure of merit.","The response-matrix formulation of quasineutrality makes KNOSOS a natural building block for coupling neoclassical fluxes into time-dependent transport codes, since repeated evaluations of fluxes for changing profiles would reuse the same precomputed matrices."],"forward_implications":["Databases of monoenergetic transport coefficients for a stellarator configuration can be generated on the order of seconds to minutes rather than hours, enabling broad parameter scans.","The √ν and superbanana-plateau regimes, which the effective-ripple figure of merit ignores, can be computed fast enough to include them in stellarator optimization loops.","Consistent solutions of quasineutrality with the flux-surface potential φ₁ become practical for systematic impurity-transport studies that were previously computationally prohibitive.","KNOSOS can supply the radial electric field, the tangential electric field, and the full bulk-species distribution function as inputs to gyrokinetic turbulence calculations.","The retained tangential magnetic drift can change predicted energy fluxes by tens of percent in ion-root plasmas, so transport calculations that omit it may misestimate confinement even when thermal particles sit in the 1/ν regime."],"supporting_citations":[{"why":"Supplies the theory of the effect of tangential drifts on neoclassical transport in stellarators close to omnigeneity, which is the basis of the equations solved by KNOSOS.","marker":"[5]"},{"why":"Provides the rigorous derivation of the radially local drift-kinetic and quasineutrality equations, including the treatment of electrostatic potential variations on the flux surface.","marker":"[25]"},{"why":"Provides the ICNTS benchmarking framework and DKES monoenergetic transport coefficient databases that KNOSOS reproduces in Section 4.1.","marker":"[4]"},{"why":"Introduces the DKES code and its simplified drift-kinetic equation, which KNOSOS solves in DKES-like mode to compare transport coefficients.","marker":"[22]"},{"why":"Supplies the EUTERPE calculations of the flux-surface electrostatic potential that KNOSOS reproduces and extends by including the tangential magnetic drift.","marker":"[28]"},{"why":"Reports the preliminary KNOSOS calculations that first indicated the large tangential electric fields and qualitative φ₁ changes that the full version confirms.","marker":"[33]"}],"fun_headline_variants":["Orbit-averaged code accelerates stellarator transport calculations","Fast open-source solver for stellarator neoclassical transport","KNOSOS: fast stellarator transport with full drift physics","Orbit-averaging makes stellarator transport modeling faster","New code brings speed plus missing drift terms to stellarator transport"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument assumes the plasma is close enough to omnigeneity and large enough in aspect ratio that the bounce-averaged radial drift is much smaller than the tangential drift, so the radial-derivative term in the drift-kinetic equation can be dropped; if that inequality fails, KNOSOS solves a different equation that misses radially global effects.","fun_headline_variants_meta":{"raw":{"variants":["Orbit-averaged code accelerates stellarator transport calculations","Fast open-source solver for stellarator neoclassical transport","KNOSOS: fast stellarator transport with full drift physics","Orbit-averaging makes stellarator transport modeling faster","New code brings speed plus missing drift terms to stellarator transport"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000295,"raw_usage":{"total_tokens":1730,"prompt_tokens":979,"completion_tokens":751,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":595,"completion_tokens_details":{"reasoning_tokens":669}},"tokens_in":595,"tokens_out":751,"duration_ms":7428,"temperature":1.0,"reasoning_tokens":669,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T10:10:36.190858+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a strongly non-omnigenous stellarator configuration, such as one with large effective ripple and small aspect ratio, and compute the two bounce-averaged drift terms in inequality (23); if the radial-drift term is not much smaller than the tangential-drift term, then run KNOSOS and a radially global code with identical collision operators and profiles and compare the radial fluxes, because a significant disagreement beyond the DKES/EUTERPE benchmark error bars would show the local-ordering premise does not hold for that device.","supporting_citations":[{"cited_title":"Calvo, F","cited_arxiv_id":null,"evidence_quote":"Supplies the theory of the effect of tangential drifts on neoclassical transport in stellarators close to omnigeneity, which is the basis of the equations solved by KNOSOS."},{"cited_title":"Calvo, J","cited_arxiv_id":null,"evidence_quote":"Provides the rigorous derivation of the radially local drift-kinetic and quasineutrality equations, including the treatment of electrostatic potential variations on the flux surface."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the ICNTS benchmarking framework and DKES monoenergetic transport coefficient databases that KNOSOS reproduces in Section 4.1."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Introduces the DKES code and its simplified drift-kinetic equation, which KNOSOS solves in DKES-like mode to compare transport coefficients."},{"cited_title":"Garc ´ıa-Rega˜na, C","cited_arxiv_id":null,"evidence_quote":"Supplies the EUTERPE calculations of the flux-surface electrostatic potential that KNOSOS reproduces and extends by including the tangential magnetic drift."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Reports the preliminary KNOSOS calculations that first indicated the large tangential electric fields and qualitative φ₁ changes that the full version confirms."}],"review_version":1}