{"id":"d526ff81-ce35-48eb-812a-04fe27656288","arxiv_id":"2512.04574","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"Adding Vreman SGS dissipation to very high-order DGSEM improves under-resolved Taylor-Green simulations but is neutral or harmful when the scheme's own dissipation already covers the active wavenumbers.","lead":"This paper tests whether adding the Vreman subgrid-scale model helps high-order discontinuous Galerkin simulations of the Taylor-Green vortex. It finds that the model helps only when the flow is strongly under-resolved; in well-resolved cases it does not improve accuracy and can over-damp intermediate scales.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"P=7-only TGV evidence cannot support the abstract's lower- vs very-high-order generalization at matched DOF.","rationale":"The reader's weakest-assumption analysis identifies the same gap I see: the paper promises lower- vs very-high-order comparisons at matched DOF but does not deliver them, and the entire regime map rests on one flow, one mesh, and one polynomial order. This is the most load-bearing weakness because the paper's headline practical guidance is inherently comparative: it tells users that at similar DOF, lower-order schemes damp small scales more and that very-high-order schemes need different SGS treatment. Without lower-order TGV data, that guidance is an extrapolation. I do not see a more fundamental internal inconsistency: the P=7 results are internally coherent, the dissipation and spectral diagnostics are appropriate, and the conclusion that static SGS tuning cannot simultaneously serve transitional and turbulent regimes is well supported by the presented data. The concrete test I propose would settle the missing comparison directly and would be a reasonable addition to the paper. Since the reader already conditioned the verdict on exactly this issue, my stress-test does not move the verdict; it reinforces it.","tokens_in":17399,"tokens_out":4354,"duration_ms":44956,"concrete_test":"Run the same viscous and inviscid TGV cases at P=3 with 32^3 elements, giving the same total DOF per direction (128) as P=7 with 16^3 elements. Use the same Chandrasekhar split form, Roe flux, BR1 viscous discretization, and Vreman constants Cv = 0, 0.01, 0.07. Compare kinetic-energy dissipation rate and E(k) at t/tc = 9 against the same references. If the qualitative ordering (SGS neutral in well-resolved, beneficial in under-resolved, large Cv over-damps intermediate scales) persists at P=3, the abstract's matched-DOF claim is supported. If the ordering changes, the paper's conclusions must be restricted to P=7 and the abstract should be revised.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central conclusion is a regime map: SGS modeling is neutral or slightly harmful in well-resolved high-order DG, beneficial in under-resolved cases, and harmful when the model constant is too large. This map is explicitly framed in the abstract as emerging from comparisons of 'lower- and very high-order configurations at similar degrees of freedom.' However, the body contains no such comparison. All TGV simulations in Section 4 use P=7 on a fixed 16^3 mesh; the only P=3–7 data are the GPU-efficiency benchmark in Fig. 1, which measures throughput per DOF and has no bearing on spectral fidelity or dissipation. The extrapolation in Section 5 that 'this issue is less pronounced at lower orders, where the inherent numerical damping partially compensates for the deficiencies of the model' is therefore an assertion, not a result. The load-bearing condition for the paper's practical guidance is that the P=7 behavior is representative of very-high-order DG and that lower-order behavior at matched DOF follows the stated ordering. If that condition fails—for example, if lower-order schemes already damp the high-wavenumber range so strongly that Vreman is redundant or harmful even in under-resolved cases, or if the optimal Cv shifts with P—the abstract's generalization collapses, even though the P=7 results themselves may remain valid.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies how an explicit Vreman subgrid-scale model interacts with the inherent numerical dissipation of a high-order DGSEM solver for the Taylor–Green vortex. Using a fixed 16^3 mesh and polynomial order P=7, the authors compare split-form stabilization, Riemann solvers (central, Roe, LD-Roe, matrix dissipation, Lax–Friedrichs), and Vreman constants C_v=0.01 and 0.07 in both a viscous (Re=1600) and an inviscid (Re=∞) configuration. On the basis of kinetic-energy dissipation-rate histories and energy spectra at t/t_c=9, they conclude that the usefulness of explicit SGS modeling is regime-dependent: in well-resolved LES the implicit dissipation of split forms and Riemann solvers is sufficient, while under strongly under-resolved conditions a weak SGS contribution can remove excess high-wavenumber energy, but a large constant over-damps intermediate scales. The paper also includes a GPU-efficiency benchmark showing that P=7 approximately doubles throughput per DOF relative to P=3.","tokens_in":17673,"tokens_out":4612,"duration_ms":43642,"significance":"If the regime map is accepted, the paper provides useful practical guidance for choosing dissipation mechanisms in very-high-order DG turbulence simulations on GPU architectures. The systematic comparison of many split forms, Riemann fluxes, and two Vreman constants against high-resolution reference data is a strength, as is the use of the publicly available HORSES3D solver, which supports reproducibility. However, the central generalization to lower polynomial orders at matched degrees of freedom is not supported by any lower-order TGV simulation, and the abstract's blanket statement that Vreman modeling does not improve accuracy in well-resolved cases is contradicted by the paper's own conclusions. The relevance of the findings to the broader practical guidance therefore rests on a single flow, a single mesh, a single polynomial order, and qualitative visual comparisons.","major_comments":[{"comment":"The abstract claims the study is 'comparing lower- and very high-order configurations at similar degrees of freedom,' and §5 states that 'this issue is less pronounced at lower orders, where the inherent numerical damping partially compensates for the deficiencies of the model.' These cross-order claims are not supported by any experiment in the paper: all TGV simulations use a single 16^3 mesh and P=7 (§3.1), and the only P=3–7 data are the GPU throughput measurements in Fig. 1, which concern computational efficiency, not spectral fidelity or dissipation. The load-bearing condition for the abstract's practical guidance is that P=7 behavior is representative of very-high-order DG and that lower-order behavior at matched DOF follows the stated ordering. Since that condition is never tested, the generalization is an assertion. Either lower-order TGV runs at matched DOF must be added, or th","section":"Abstract; §3.1; §5"},{"comment":"The abstract states, 'In the well-resolved cases considered, Vreman modeling does not improve accuracy because its active wavenumber range overlaps with the scheme's inherent dissipation.' This is internally inconsistent with the conclusions in §6, which recommend, for well-resolved LES in the turbulent regime, 'superior spectral fidelity is obtained using the split form and an LD-Roe flux, supplemented by an SGS model with a low constant (Cv = 0.01).' The same recommendation is supported by §4.1.5, where 'LD-Roe coupled with the Vreman model provides the best spectral fidelity.' These are not merely wording differences: one central message says Vreman is not helpful in well-resolved cases, while the other says the best well-resolved turbulent configuration uses Vreman. The abstract and conclusions need to be reconciled.","section":"Abstract vs §6; §4.1.5"},{"comment":"The accuracy assessments are made exclusively by visual inspection of dissipation-rate curves and energy spectra, without any quantitative error measure. This matters because the conclusions are finely graded: e.g., §4.1.3 claims that C_v=0.07 'over-dissipates energy at intermediate scales' while C_v=0.01 leaves the highest wavenumbers 'slightly under-dissipated'; §4.2.2 claims that C_v=0.01 'overestimates energy at intermediate wavenumbers.' Such claims are load-bearing for the proposed regime map and for the guidance in §6, but they rest on subjective comparison to reference spectra. The authors should provide quantitative metrics, for example relative L1/L2 errors of E(k) over defined wavenumber ranges or time-integrated dissipation-rate errors, so that 'best spectral fidelity' and 'over-dissipation' are defined operationally.","section":"§4, Figs. 3–10"}],"minor_comments":[{"comment":"The first sentence of §4.1.2 contains a typo: 'As in the reminder of the paper' should be 'As in the remainder of the paper.'","section":"§4.1.2"},{"comment":"The caption says 'The standard and Morinishi schemes are shown only in Fig. 3a, as they became unstable before t/t_c = 9. Results for all schemes are shown in both figures' — this is self-contradictory. If standard and Morinishi are omitted from Fig. 3b, the sentence should say 'Results for all stable schemes are shown in both figures.'","section":"Fig. 3 caption"},{"comment":"The GPU efficiency metric Time/(DOF×RHS) is introduced without a definition of the RHS evaluation context (e.g., which flux/split form). Since the comparison spans P=3–7, a one-sentence statement of the test problem used for the benchmark would improve clarity.","section":"§2.6; Fig. 1"},{"comment":"The standard versus split-form comparison uses Gauss nodes for the standard discretization and Gauss–Lobatto nodes for the split forms, so the effect of split-form stabilization is not isolated from the effect of nodal distribution. The text notes this, but a brief interpretive caution would be helpful for readers.","section":"§4.1.1"},{"comment":"The sentence 'Note that as only C_v = 0.01 and C_v = 0.07 were tested, the global optimum for this specific problem may lie within this range' is appropriate, but the bullet list should make explicit that the 'optimal' labels follow from only two constants, not from a true optimization over C_v.","section":"§6"}],"recommendation":"major_revision","confidential_remarks":"The mismatch between the abstract's cross-order claim and the actual P=7-only experiments is a significant framing issue; the authors need either to add the missing lower-order simulations or to substantially narrow the abstract and conclusions. The internal contradiction about the benefit of Vreman in well-resolved turbulent states is another point that must be fixed before publication. The core qualitative observation — that the value of explicit SGS dissipation depends on resolution and regime — is plausible and sufficiently supported by the P=7 data to merit further scrutiny, but the paper in its current form overclaims its generality."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper is a competent, narrowly-scoped numerical study of the Taylor-Green vortex at P=7 showing that Vreman SGS helps only in under-resolved regimes. The central result holds up for the conditions tested. The soft spot is the abstract: it promises comparisons of 'lower- and very high-order configurations at similar degrees of freedom,' but the only lower-order data is a GPU throughput benchmark that says nothing about spectral fidelity. That overstatement is worth flagging, but it doesn't sink the P=7 conclusions.\n\nWhat's new here: a systematic sweep of split forms, Riemann solvers, and two Vreman constants (0.01 and 0.07) in a DGSEM code, on both viscous and inviscid TGV. The spectral-difference plots are a nice diagnostic—they show how the SGS constant controls the wavenumber at which model dissipation becomes active. The qualitative regime map (well-resolved iLES fine, under-resolved needs weak SGS, strong Cv over-damps intermediate scales) is consistently supported by the dissipation-rate and spectrum figures for P=7.\n\nWhere it's soft, in proportion: all TGV simulations are P=7 on a single 16^3 mesh. The Section 5 claim that 'this issue is less pronounced at lower orders' is an assertion, not a result. Accuracy is assessed visually against reference spectra; no error norms or convergence studies are reported. Only two Cv values are tested, though the authors themselves acknowledge the optimum may lie between. No data or configuration files are shipped. None of these kill the paper, but they limit the practical guidance to the specific P=7 TGV setting.\n\nCitation pattern looks fine; the authors engage relevant prior work and state their limitations clearly. The writing is direct.\n\nWho gets value: people working on high-order DG LES, especially with HORSES3D or GPU-oriented solvers, will find a useful data point for choosing Vreman constants and Riemann solvers. I'd send this to peer review, but ask the authors to either add lower-order TGV runs or soften the abstract's generalization. As is, it's a conditional accept with minor revisions.","headline":"Careful P=7 TGV parameter study with a useful regime map; abstract overclaims a lower-order comparison that the body never makes.","tokens_in":18203,"tokens_out":1991,"would_cite":false,"duration_ms":21287,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["47.11.Fg","47.27.Ep"],"model":"deepseek-v4-flash","headline":"Explicit Vreman subgrid-scale modeling helps very-high-order implicit LES only when the flow is under-resolved; in well-resolved simulations the scheme's own split-form and Riemann-solver dissipation already suffices, and adding the model c","keywords":["discontinuous Galerkin spectral element method","implicit LES","Vreman subgrid-scale model","Taylor-Green vortex","split form","Riemann solver","spectral fidelity","GPU high-order computing"],"falsifier":"Run the same Re=1600 Taylor-Green vortex at a lower polynomial order (say P=3 or P=4) with the same total degrees of freedom and compare the spectral accuracy of iLES versus iLES+Vreman; the paper predicts SGS is neutral or harmful there, so a clear improvement would falsify the regime map. A second check is to scan Cv finely in 0.01–0.07 in the inviscid case: the paper's trade-off implies a non-monotone accuracy curve, so monotone improvement with Cv would contradict it.","tokens_in":17303,"feed_emoji":"🌊","tokens_out":11258,"duration_ms":94917,"temperature":0.7,"pith_summary":"This paper asks whether adding an explicit Vreman subgrid-scale model to a very high-order discontinuous Galerkin solver improves implicit large-eddy simulations. Using the Taylor-Green vortex at Reynolds 1600 and in the inviscid limit, it finds the answer is regime-dependent: when the mesh resolves the flow, split-form stabilization plus Riemann-solver dissipation is already enough, and the Vreman model is neutral or harmful; when resolution is too coarse, a weak Vreman term removes excess high-wavenumber energy and improves spectral accuracy. The paper identifies the configuration that performed best in each regime—Chandrasekhar split form with Roe flux and no SGS for transition, LD-Roe with Cv=0.01 for well-resolved turbulence, Roe with Cv=0.07 for under-resolved turbulence—and concludes that no static setting spans all regimes. The work matters because GPU-oriented computing favors very high polynomial orders, where tuning the balance between numerical and modeled dissipation becomes the main accuracy lever.","feed_headline":"Explicit subgrid models aid high-order LES only when under-resolved","feed_subtitle":"In well-resolved simulations the Vreman model is neutral or harmful; under coarse resolution it stops energy pile-up.","key_machinery":"The carrying mechanism is the Vreman eddy-viscosity SGS model with the element-based filter length Δ = V^(1/3)/(P+1), whose single constant Cv controls both the amount of added dissipation and the wavenumber at which that dissipation activates. It is placed inside a DGSEM discretization that already has two intrinsic dissipation sources: split-form stabilization (using the Chandrasekhar form for robustness) and Riemann solvers (Roe and its low-dissipation variant LD-Roe). The diagnostics—kinetic-energy dissipation rate over time and the energy spectrum at t/tc=9—expose which of the three dissipation sources dominates in each regime, and the spectral difference E_base(k) - E_SGS(k) locates th","core_discovery":"The central discovery is a regime map for dissipation in very high-order DG. In the well-resolved TGV at Re=1600 with P=7, the inherent dissipation from split forms and Riemann solvers matches the reference transitional dynamics, and the Vreman model's added viscosity acts in a wavenumber range that overlaps the scheme's own dissipation, so it does not improve accuracy and the larger constant (Cv=0.07) visibly over-damps intermediate scales. In the turbulent phase of that case, a low-dissipation Riemann flux (LD-Roe) combined with a weak Vreman constant (Cv=0.01) gives the best high-wavenumber spectrum. In the inviscid, strongly under-resolved TGV, the same weak model is insufficient: Roe wi","pith_inferences":["The paper does not test lower-order TGV flows at equal degrees of freedom despite announcing them; if a lower-order run showed Vreman improving a well-resolved LES, the regime map would not generalize. That comparison is the most direct untested extension.","If the constant-to-wavenumber relation holds generally, it yields a practical calibration rule: run the iLES baseline briefly, find the wavenumber where energy piles up, and choose Cv so the model's activation wavenumber sits just below it. The paper demonstrates the relation but does not codify the rule.","A natural way to get scale-selective dissipation without a global constant is p-adaptivity: locally lowering the polynomial order damps near-cutoff scales more strongly, which mimics the weak-SGS effect the paper found beneficial in under-resolved regions.","Because only Cv=0.01 and Cv=0.07 were scanned, the spectral evidence suggests an intermediate constant—or a wavenumber-dependent variant—might hit the sweet spot of removing the pile-up without flattening intermediate scales; the paper notes the optimum may lie between the two values but does not scan it."],"forward_implications":["In well-resolved very-high-order LES, explicit SGS modeling can be omitted: the split-form plus Roe configuration matches the transitional reference, and adding Vreman only shifts dissipation into scales the scheme already handles.","For the turbulent phase of a well-resolved simulation, the most accurate tested setup combines a low-dissipation Riemann solver with a weak Vreman constant (Cv=0.01), which resolves the energy pile-up without over-damping intermediate wavenumbers.","For strongly under-resolved flows, a larger Vreman constant (Cv=0.07) is needed to remove high-wavenumber energy, but it still over-damps the scales just below the cutoff, so the correct constant depends on how under-resolved the simulation is.","The same split form with the same SGS model can be the best or the worst choice depending on the flow regime, so a static numerical configuration cannot be optimal across a simulation that passes through laminar, transitional, and turbulent phases.","Scale-aware or adaptive dissipation—the paper points to spectral vanishing viscosity and data-driven tuning—becomes a necessary next step rather than an optional refinement."],"fun_headline_variants":["Vreman SGS helps only in under-resolved very high-order LES","High-order LES: SGS model beneficial only when under-resolved","Subgrid models neutral or harmful in well-resolved DGSEM","Regime map: SGS aid under-resolved, hurt well-resolved DG","Explicit SGS: only useful for under-resolved high-order DG"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The regime map is built solely on the P=7 Taylor-Green vortex on a 16^3 mesh with two Vreman constants; the abstract announces lower-order comparisons at equal degrees of freedom, but the only reported lower-order data is a GPU-efficiency benchmark, not a flow-accuracy comparison. If P=7 TGV does not represent other very-high-order DG set-ups, the practical guidance does not transfer.","fun_headline_variants_meta":{"raw":{"variants":["Vreman SGS helps only in under-resolved very high-order LES","High-order LES: SGS model beneficial only when under-resolved","Subgrid models neutral or harmful in well-resolved DGSEM","Regime map: SGS aid under-resolved, hurt well-resolved DG","Explicit SGS: only useful for under-resolved high-order DG"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000246,"raw_usage":{"total_tokens":1428,"prompt_tokens":845,"completion_tokens":583,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":589,"completion_tokens_details":{"reasoning_tokens":486}},"tokens_in":589,"tokens_out":583,"duration_ms":4978,"temperature":1.0,"reasoning_tokens":486,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T18:33:25.436842+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the same Re=1600 Taylor-Green vortex at a lower polynomial order (say P=3 or P=4) with the same total degrees of freedom and compare the spectral accuracy of iLES versus iLES+Vreman; the paper predicts SGS is neutral or harmful there, so a clear improvement would falsify the regime map. A second check is to scan Cv finely in 0.01–0.07 in the inviscid case: the paper's trade-off implies a non-monotone accuracy curve, so monotone improvement with Cv would contradict it.","supporting_citations":[],"review_version":1}