{"id":"a9103f45-8801-4c73-8cad-3c501964a28d","arxiv_id":"2506.03459","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"At the JET pedestal-forming edge, linear stability changes from TEM/ITG modes to resistive modes with collisionality, and QuaLiKiz is verified only up to rho_tor=0.85 while TGLF-SAT2 is verified up to 0.9.","lead":"This paper maps the turbulent instabilities that drive the edge of JET tokamak plasmas just before they switch into high-confinement mode, using high-fidelity gyrokinetic simulations. It then tests two fast reduced models against those simulations and finds that one, QuaLiKiz, stops being reliable beyond 85 percent of the plasma radius while TGLF-SAT2 holds up to 90 percent, an important boundary for ITER scenario modeling.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 'agrees well' boundary at rho_tor=0.9 is not quantitatively supported: the paper's own Section IV.D.2 reports wrong dominant modes at low k_y, wrong propagation direction, and up to 30% growth-rate overprediction at that radius.","rationale":"The paper is a careful linear gyrokinetic verification study, and the QuaLiKiz inadequacy beyond rho_tor=0.85 is well supported by the s-alpha geometry benchmark and the collisionality scans. The absence of nonlinear saturation and E-cross-B shear is explicitly declared, so I do not treat that as a hidden flaw. My most load-bearing concern is the abstract's TGLF-SAT2 boundary claim at rho_tor=0.9, because the paper's own detailed comparison contains documented mode misidentifications, wrong propagation directions at low k_y, and up to 30% growth-rate overprediction at that radius. The reader's weakest_assumption focused on profile reconstruction uncertainty; that is real, but the paper's sensitivity scans partly bound it, and even accepting the inputs as given, the 'agrees well through 0.9' statement is not supported by a quantitative error criterion. Appendix E adds a second, independent fragility: the saturated flux output of TGLF-SAT2 changes by tens of percent depending on the Miller parameterization, which matters for the integrated-modeling recommendation even if the linear growth-rate spectra are robust. My proposed check would settle whether the 0.9 boundary is a reproducible quantitative statement or a generous qualitative summary. Since the reader already returned CONDITIONAL and flagged overstatement in the rationale, my concern does not move the verdict; it sharpens the condition under which the abstract should be revised.","tokens_in":52643,"tokens_out":6979,"duration_ms":77730,"concrete_test":"Using the data behind Figs. 23 and 34, compute a single error metric at rho_tor=0.90 over k_y rho_s in [0.1,0.5] for all seven discharges: fraction of wavenumbers where TGLF-SAT2 matches GENE Miller/Sugama in sign(omega), mean relative error in |gamma|, and mean relative error in q_i/q_e. Set a pre-registered tolerance, e.g. >=80% sign agreement and <30% mean growth-rate error. If the metric fails at rho_tor=0.90, the abstract must be revised to quantify the k_y range and error level rather than state unqualified agreement through 0.9. Separately, run TGLF-SAT2 at rho_tor=0.90 with both FLUSH and MEGPy Miller parameterizations and compare saturated heat fluxes; if the flux difference exceeds the quoted GENE agreement tolerance, the integrated-modeling recommendation should be conditional on the parameterization choice.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that TGLF-SAT2 'agrees well with linear spectra and quasilinear heat fluxes from GENE up to and including rho_tor=0.9' depends on a quantitative standard for 'agrees well' that the paper never defines. The evidence in Section IV.D.2 shows substantial disagreement at rho_tor=0.9: for k_y rho_s < 0.15 TGLF-SAT2 predicts different dominant instabilities than GENE for all discharges; for several low-triangularity discharges the dominant modes at k_y rho_s <= 0.3 have the correct growth rate but the wrong propagation direction; for high-density branch discharges TGLF-SAT2 overpredicts growth rates by up to 30% (attributed to the PAS collision model); and the q_i/q_e ratios for the l=1 ion-direction branches are significantly higher in GENE than in TGLF-SAT2. These are not small deviations at unimportant wavenumbers: k_y rho_s ~ 0.2 is the wavenumber used throughout the paper to represent transport-relevant contributions. In addition, Appendix E (Fig. 34) shows that the saturated heat and particle fluxes from TGLF-SAT2 change by 40-80% at rho_tor=0.85 and 20-60% at rho_tor=0.95 depending on whether the Miller equilibrium is parameterized with FLUSH or MEGPy, so the flux part of the recommendation is not parameterization-independent. The linear spectra themselves differ by only ~8% in growth-rate area between parameterizations, so the fragility is in the saturation/flux part that the abstract also endorses. Thus the abstract's boundary statement is an interpretation, not a measured conclusion.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents a linear gyrokinetic stability analysis of seven JET-ILW L-mode discharges in the pedestal-forming region (rho_tor = 0.85, 0.90, 0.95), just before the L-H transition, using the GENE code. It characterizes the dominant instabilities, including TEMs, ITGs, unconventional l=1 ion-direction modes, and collisionality-dependent resistive/hybrid modes, and performs extensive sensitivity scans in gradients, collisionality, isotope mass, magnetic geometry, and beta. The second half of the paper benchmarks the reduced quasilinear models QuaLiKiz and TGLF-SAT2 against GENE, using staged model reductions to isolate the effect of geometry, collision operators, and electromagnetic effects. The main conclusions are that QuaLiKiz is not reliable beyond rho_tor = 0.85, while TGLF-SAT2 agrees well with linear spectra and quasilinear flux ratios up to and including rho_tor = 0.9.","tokens_in":52971,"tokens_out":7377,"duration_ms":76801,"significance":"If the central claims hold, this is a valuable verification dataset for reduced-order transport models in a region of growing importance for integrated modeling. The paper is carefully structured: convergence checks are reported, model reductions are explicit, and the comparison is performed on a common input-generation pipeline. The identification of unconventional l=1 ion modes and the mapping of collisionality regimes onto neoclassical banana/plateau/Pfirsch-Schlueter boundaries are useful physical results. The authors also provide open-source tools (GyroKit, MEGPy) and state that the simulation database will be made available, which strengthens reproducibility. The main weaknesses are that the abstract overstates the level of TGLF-SAT2 agreement relative to the paper's own quantitative results, and that the flux part of the recommendation is shown in Appendix E to be sensitive to the flux-surface parameterization choice.","major_comments":[{"comment":"The statement that TGLF-SAT2 'agrees well with linear spectra and the quasilinear heat fluxes from GENE up to and including rho_tor = 0.9' is stronger than the evidence reported in Section IV.D.2. At rho_tor = 0.9 the paper states that for k_y rho_s < 0.15 TGLF-SAT2 predicts different dominant instabilities than GENE for all discharges; for several low-triangularity discharges the dominant modes at k_y rho_s <= 0.3 have the correct growth rate but the wrong propagation direction; for high-density branch discharges TGLF-SAT2 overpredicts growth rates by up to 30% (attributed to the PAS collision model); and the q_i/q_e ratios for the l=1 ion-direction branches are significantly higher in GENE than in TGLF-SAT2. These are not small deviations at unimportant wavenumbers, since k_y rho_s ~ 0.2 is used throughout the paper as transport-relevant. The paper never defines a quantitative criterion for 'agrees well.' The abstract and Section V should either introduce a quantitative error metric and show that it is met, or be reworded to distinguish 'reasonable agreement on the dominant mode at many wavenumbers' from 'agreement in mode identity, propagation direction, and growth-rate magnitude.'","section":"Abstract; Section IV.D.2"},{"comment":"The recommendation in Section V that TGLF-SAT2 is suitable for integrated modeling in this region is not supported by the flux sensitivity documented in Appendix E. Figure 34 shows that TGLF-SAT2 saturated heat and particle fluxes change by 40-80% at rho_tor = 0.85 and by 20-60% at rho_tor = 0.95 depending on whether the Miller equilibrium parameterization is generated with FLUSH or MEGPy, while the area under the linear growth-rate spectrum differs by only ~8%. Since rho_tor = 0.85 is within the range for which the abstract recommends TGLF-SAT2, the 'quasilinear heat fluxes' part of the recommendation is not parameterization-independent. The paper should either restrict the agreement claim to the linear spectra and flux ratios, or explicitly address how the flux sensitivity affects the practical recommendation for integrated modeling.","section":"Appendix E; Section V"}],"minor_comments":[{"comment":"The word 'collisonality' is misspelled as 'collisionality' in the abstract and in several places in the main text; please correct globally.","section":"Throughout"},{"comment":"Several unresolved placeholders appear: '5-20% ?', 'Z_eff ~ 1.2-1.4 ?', and 'see Table I?'; these should be completed with references or deleted.","section":"Section II.A; Section III.B.3; Appendix B"},{"comment":"The sentence 'This is correlated with #94114 having about , as can be seen in Table I.' is incomplete and needs to be finished or removed.","section":"Appendix B"},{"comment":"The phrase 'quasilinear heat fluxes from GENE' is imprecise; the comparison in Section IV is of heat flux ratios q_i/q_e and convective heat flux ratios, not saturated heat fluxes. Please reword to avoid confusion.","section":"Abstract; Section IV.A"},{"comment":"Notation is inconsistent in places, e.g., 'GENE (s-α eq., β= 0)' is missing a space before 'β' and uses different abbreviation styles; please unify the notation for equilibria and collision operators.","section":"Figure 22 and elsewhere"},{"comment":"The data availability statement says data are available 'upon reasonable request'; given the stated intention to share the simulation database, a repository link or DOI would be more suitable for reproducibility.","section":"Data Availability"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is technically solid and the dataset is valuable for the community. The main issue is that the abstract and summary overstate the level of TGLF-SAT2 agreement relative to the detailed results in Section IV.D.2, and the flux sensitivity in Appendix E undermines the practical recommendation without an explicit caveat. If the authors revise the claims to match the evidence and add a quantitative definition of 'agrees well,' I would support publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Gijs — you should know about this one. It is the most complete linear gyrokinetic verification dataset for the L-mode pedestal-forming region that we have: seven JET-ILW discharges, three radii, full collisionality and gradient scans, mode structures, isotope, geometry, and impurity sensitivities. The QuaLiKiz conclusion — inadequate beyond rho_tor=0.85 — is well supported. The TGLF-SAT2 result is real, but the abstract's 'agrees well up to and including rho_tor=0.9' is stronger than the evidence.\n\nWhat is good: the study is careful. The paper benchmarks the model reductions (s-alpha vs numerical equilibria, Miller, collision operators) far more thoroughly than most verification papers. The identification of odd-parity l=1 ion modes and hybrid resistive drift-wave branches at 0.95 is a genuinely useful addition, and the collisionality scans around the experimental values are a nice way to map regime boundaries. The authors also flag their own data limitations (ECE cutoff, reflectometry position, impurity extrapolation) clearly.\n\nWhere it is soft: Section IV.D.2 itself documents wrong dominant modes for ky rho_s < 0.15 for all discharges at 0.9, wrong propagation direction for several low-triangularity discharges, and up to 30% growth-rate overprediction on the high-density branch from the PAS collision model. Those are transport-relevant wavenumbers. And Appendix E shows the saturated TGLF-SAT2 fluxes change 40–80% depending on whether the Miller equilibrium is parameterized with FLUSH or MEGPy — the linear spectra barely move (~8%), so the flux recommendation is fragile. The paper does not define a quantitative standard for 'agrees well', so the boundary at 0.9 is an interpretation rather than a measured conclusion. The reconstructed profile uncertainties are acknowledged but not propagated through the mode identification; that is a caveat, not a fatal flaw.\n\nWho this is for: integrated-modelling practitioners who use QuaLiKiz or TGLF-SAT2 near the L-mode edge, and reduced-model developers. The dataset itself is valuable independent of the abstract's phrasing.\n\nRecommendation: yes, send to peer review. It deserves a serious referee. The authors need to tighten the abstract, define agreement metrics, and take the FLUSH/MEGPy sensitivity seriously enough to temper the flux part of the claim.","headline":"Careful, valuable verification of QuaLiKiz and TGLF-SAT2 at the L-mode edge, but the abstract oversells TGLF at rho_tor=0.9.","tokens_in":53594,"tokens_out":2871,"would_cite":true,"duration_ms":29595,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["52.35.Qz","52.55.Fa"],"model":"deepseek-v4-flash","headline":"The paper claims QuaLiKiz is inadequate beyond $\\rho_{\\mathrm{tor}}=0.85$ in JET's L-mode pedestal-forming region, while TGLF-SAT2 agrees with GENE linear spectra and quasilinear heat fluxes through $\\rho_{\\mathrm{tor}}=0.9$.","keywords":["gyrokinetic simulations","L-H transition","L-mode pedestal","reduced-order turbulence models","QuaLiKiz","TGLF","trapped electron modes","collisionality"],"falsifier":"A nonlinear GENE simulation at $\\rho_{\\mathrm{tor}}=0.9$ for a high-density discharge, with the experimental profiles and numerical equilibrium, would settle the inward-particle-flux claim: if the saturated particle flux is outward, the quasilinear transport prediction validated here fails in the regime where TGLF-SAT2 is recommended.","tokens_in":52431,"feed_emoji":"🧲","tokens_out":12255,"duration_ms":126678,"temperature":0.7,"pith_summary":"This paper tests whether two fast quasilinear transport models can be trusted in the outermost region of an L-mode tokamak plasma, the zone where the pedestal forms just before the transition to H-mode. Using linear gyrokinetic simulations with the GENE code for seven JET discharges at $\\rho_{\\mathrm{tor}}=0.85,0.9,0.95$, the authors map which microinstabilities dominate on the low- and high-density branches of the L-H power threshold. They find that QuaLiKiz becomes inadequate beyond $\\rho_{\\mathrm{tor}}=0.85$, whereas TGLF-SAT2 agrees with the linear spectra and quasilinear heat fluxes up to $\\rho_{\\mathrm{tor}}=0.9$. The practical stake is that integrated modeling codes need to know how far toward the separatrix they can extend the outer simulation boundary before the reduced turbulence model misrepresents the transport.","feed_headline":"QuaLiKiz turbulence model fails beyond 85% of the plasma radius","feed_subtitle":"JET edge tests show TGLF-SAT2 matches gyrokinetic spectra through 90% of the radius, guiding where fast transport models can be trusted.","key_machinery":"The central object is the local linear gyrokinetic spectrum computed with the GENE code, organized as a scan in normalized electron collisionality $\\nu_e^*$ at each radius. The paper identifies the experimental collisionality interval as a minimum-drive gap between collisionless ITG/TEM branches at low $\\nu_e^*$ and resistive branches at high $\\nu_e^*$, and uses that gap as the yardstick for what a reduced model must reproduce. The fidelity reduction then isolates each reduced model's assumptions: QuaLiKiz's electrostatic $s$-$\\alpha$ geometry and Krook collision operator, and TGLF-SAT2's gyrofluid equations, Miller geometry, and pitch-angle-scattering collision model.","core_discovery":"On its own terms, the paper establishes a boundary on the applicability of reduced-order quasilinear models in the L-mode pedestal-forming region: QuaLiKiz is reliable only where electron collisionality stays in the banana regime, roughly $\\rho_{\\mathrm{tor}}\\leq0.85$; TGLF-SAT2 remains reliable through $\\rho_{\\mathrm{tor}}=0.9$, including the high-density $\\ell=1$ ion modes with trapped-ion drive, but breaks down at $\\rho_{\\mathrm{tor}}=0.95$ where hybrid modes and resistive drift-wave branches with non-adiabatic passing electrons dominate. The evidence is a systematic linear gyrokinetic characterization: trapped-electron modes and ITG modes at inner radii, unconventional ballooning ion modes and collisionality-driven resistive branches near the edge, with the experimental collisionality sitting near a minimum of linear drive between branches. The same dataset shows that the experimentally observed dependencies of the L-H power threshold on density, isotope mass, shaping, and impurities are mirrored in the linear stability properties.","pith_inferences":["A practical switching rule for integrated modeling could be built from local $\\rho_{\\mathrm{tor}}$ and $\\nu_e^*$, choosing QuaLiKiz below $0.85$, TGLF-SAT2 through $0.9$, and higher-fidelity gyrokinetics beyond; the paper stops short of proposing such a rule.","Machine-learning surrogates of QuaLiKiz trained on JET-like edge data would inherit the failure beyond $\\rho_{\\mathrm{tor}}=0.85$; the GENE dataset assembled here offers a target for retraining on the edge regime.","A nonlinear simulation campaign at $\\rho_{\\mathrm{tor}}=0.9$ comparing saturated particle flux direction with the quasilinear prediction would test whether the inward-flux window survives saturation."],"forward_implications":["Integrated modeling of L-mode plasmas can extend the outer simulation boundary to $\\rho_{\\mathrm{tor}}=0.9$ when using TGLF-SAT2, provided the local collisionality stays below the plateau-Pfirsch-Schlüter boundary.","QuaLiKiz should not be used to predict pedestal-forming transport beyond $\\rho_{\\mathrm{tor}}=0.85$; within the banana regime at that radius it still captures the dominant TEM and ITG branches.","At $\\rho_{\\mathrm{tor}}=0.95$, neither reduced model reproduces the hybrid and resistive modes, so integrated modeling must keep its boundary inside this radius or supplement the reduced model with higher-fidelity physics.","The quasilinear particle flux direction is tied to the same collisionality window: inward flux appears only where the linear drive is minimized, which matters for predicting density-profile evolution during pedestal build-up."],"supporting_citations":[{"why":"supplies the seven JET-ILW NBI-heated deuterium discharges and the L-H threshold power scan that define the low- and high-density branches.","marker":"[53]"},{"why":"provides the local linear gyrokinetic solver GENE whose spectra are the reference for the whole study.","marker":"[54,55]"},{"why":"defines the QuaLiKiz reduced quasilinear model and its flux and particle transport formulation being assessed.","marker":"[16-18]"},{"why":"defines the TGLF gyrofluid transport model and its collision treatment being assessed.","marker":"[19,20]"},{"why":"supplies the SAT2 saturation rule and the modified linear solver that constitute the TGLF variant benchmarked here.","marker":"[22,23]"},{"why":"provides the Gaussian process regression profile fitting that generates the local gradients used as simulation input.","marker":"[57]"},{"why":"establishes the earlier edge-stability picture connecting collisionality and Zeff to the L-H transition that this study extends.","marker":"[40]"},{"why":"supplies prior edge gyrokinetic results on isotope mass and passing-electron dynamics used to interpret the outermost-radius modes.","marker":"[41]"},{"why":"supports the quasilinear assumption for L-mode edge turbulence that underpins the reduced-model comparison.","marker":"[44]"}],"fun_headline_variants":["QuaLiKiz edge fails at 85% plasma radius","TGLF-SAT2 trustworthy to 90% radius, not beyond","Reduced models split at rho=0.85: QuaLiKiz out, TGLF in","JET edge: QuaLiKiz fails, TGLF-SAT2 holds to rho=0.9"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the fitted local gradients, collisionalities, and reconstructed magnetic equilibria are accurate enough that the seven discharges really sit in the reported instability regimes; the paper acknowledges that reflectometry radial-position uncertainty was not propagated, ECE data were cut below 800 eV, and impurity profiles were extrapolated for two discharges, so a bias in any of these inputs would shift the mode identifications and therefore the reduced-model verdicts.","fun_headline_variants_meta":{"raw":{"variants":["QuaLiKiz edge fails at 85% plasma radius","TGLF-SAT2 trustworthy to 90% radius, not beyond","Reduced models split at rho=0.85: QuaLiKiz out, TGLF in","JET edge: QuaLiKiz fails, TGLF-SAT2 holds to rho=0.9"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00034,"raw_usage":{"total_tokens":1990,"prompt_tokens":1176,"completion_tokens":814,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":792,"completion_tokens_details":{"reasoning_tokens":715}},"tokens_in":792,"tokens_out":814,"duration_ms":8169,"temperature":1.0,"reasoning_tokens":715,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T11:02:05.503801+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A nonlinear GENE simulation at $\\rho_{\\mathrm{tor}}=0.9$ for a high-density discharge, with the experimental profiles and numerical equilibrium, would settle the inward-particle-flux claim: if the saturated particle flux is outward, the quasilinear transport prediction validated here fails in the regime where TGLF-SAT2 is recommended.","supporting_citations":[{"cited_title":"Vincenzi, et al","cited_arxiv_id":null,"evidence_quote":"supplies the seven JET-ILW NBI-heated deuterium discharges and the L-H threshold power scan that define the low- and high-density branches."},{"cited_title":"Ho, et al","cited_arxiv_id":null,"evidence_quote":"provides the Gaussian process regression profile fitting that generates the local gradients used as simulation input."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"supports the quasilinear assumption for L-mode edge turbulence that underpins the reduced-model comparison."}],"review_version":1}