{"id":"a4d49004-70d6-4bf2-abc8-42eb0f235909","arxiv_id":"2608.00843","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"A Helmholtz-Leray projection with VMS stabilization on the predicted velocity only gives a three-step Navier-Stokes solver that is second-order accurate in time and less dissipative than monolithic VMS in the reported tests.","lead":"This paper builds a faster stabilized solver for the incompressible Navier-Stokes equations by splitting the coupled velocity-pressure problem into three sequential steps and applying variational multiscale stabilization only to the predicted velocity. The method matches classic benchmark flows, reduces excess dissipation seen in monolithic VMS, and lowers average solve time per step by 1.3x to 2.7x.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Assumption 3.1 already contains the O(h^q) spatial estimate in Proposition 1; the VMS predictor also drops a divergence term under a condition that fails for the intermediate velocity, so the combined error estimate is conditional.","rationale":"The reader's conditional verdict is appropriate. The paper honestly labels Proposition 1 as depending on stability and spatial-approximation assumptions, and the temporal convergence evidence is real. However, the spatial part of the estimate is assumed, not derived, and there is a concrete unaddressed inconsistency in carrying a divergence-free manipulation into a predictor that is not divergence-free. The proposed h-refinement test would settle whether Assumption 3.1 actually holds for the implemented scheme; until then, the combined O(dt^2+h^q) statement and the 'certified method' reading should remain conditional. I agree with the reader that Assumption 3.1 is the key weak point, but I add the specific dropped-term mechanism as the reason it may fail. The numerical benchmarks and the measured speedups are useful independent support for the method as a practical scheme, but they do not by themselves establish the formal spatial error bound.","tokens_in":25827,"tokens_out":7498,"duration_ms":80554,"concrete_test":"Perform a spatial-convergence study on the manufactured problem of Section 4.1: fix dt small enough that the temporal error is below the expected spatial error (e.g., dt = T/512 or smaller), refine uniform meshes h = 1/8, 1/16, 1/32, 1/64, and measure the L2-in-time velocity error in (61) at one representative Re. If the error does not decrease as C h^q (or stalls because the omitted divergence term dominates), Assumption 3.1 fails and the O(h^q) part of the headline estimate is unsupported. As a companion check, re-derive Algorithm 1's momentum equation retaining the dropped (div(u~_h+u~'))u~' term and compare its residual contribution against the retained SUPG terms on the same meshes.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central formal claim (Prop. 1, Eq. 61) states total velocity error O(dt^2 + h^q). The temporal part is on solid ground: BDF2 plus Shen's incremental-splitting estimate gives O(dt^2), and the manufactured-solution tests are consistent with that rate. The load-bearing weakness is the spatial part. Assumption 3.1 (Eq. 59) directly assumes the stabilized fractional subproblems are consistent at rate h^q, including the modeled fine scale and the pressure correction. Proposition 1's proof of the spatial term is then only a subtraction of the velocity-update equation; it does not analyze the nonlinear predictor or the pressure Poisson equation. Moreover, the derivation of the momentum predictor imports the monolithic fine-scale advection manipulation in which the term (div(u_h+u'))u' is dropped using div(u_h+u')=0. For the split scheme this condition does not hold—the intermediate velocity is exactly the quantity that fails to be solenoidal—and the corresponding term for u~ is absent from Algorithm 1. Nothing in Assumption 3.1 shows that this omitted term is O(h^q); if it is not, the O(h^q) component of (61) is unsupported and the equal-order stability claim for the predictor is not established.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a VMS-stabilized incremental projection method for the incompressible Navier-Stokes equations. The multiscale decomposition is applied only to the predicted (intermediate) velocity; the pressure and the corrected velocity are not decomposed. The resulting scheme replaces the monolithic saddle-point system by a nonlinear momentum predictor, a pressure Poisson equation, and a velocity projection, with SUPG-like and PSPG-like terms arising from the modeled fine scales. A formal error analysis decomposes the total velocity error into BDF2 time-discretization, projection-splitting, and spatial-VMS contributions and states a combined O(dt^2 + h^q) estimate under regularity, stability, and spatial-approximation assumptions. Numerical experiments cover manufactured solutions, lid-driven cavity, flow past a cylinder, and the Taylor-Green vortex, and include comparisons with monolithic VMS and reference data, as well as wall-clock cost measurements.","tokens_in":26249,"tokens_out":8461,"duration_ms":106936,"significance":"If the central claims hold, the method is practically attractive: it enables equal-order velocity-pressure elements without the pressure fine scale, avoids the monolithic saddle point, and still supplies advective and pressure stabilization. The numerical verification is extensive and careful: the MMS tests confirm the expected second-order temporal rate across a wide Reynolds range, the cavity and cylinder benchmarks agree with established references, and the Taylor-Green study reports both accuracy and computational speedups. The temporal part of the error analysis is on solid ground, being based on the well-known incremental pressure-correction splitting. However, the spatial half of the headline estimate is not actually proved; it is an assumption that already contains the conclusion. The paper is transparent about its assumptions, but the formal result as stated is weaker than it appears.","major_comments":[{"comment":"The spatial part of the headline estimate is an assumption, not a derived result. Assumption 3.1 (Eq. (59)) assumes exactly the combined bound on the predicted-velocity error and the pressure-correction gradient error. Subtracting the continuous and discrete velocity-update equations and applying the triangle inequality (Eq. (60)) then yields the O(h^q) spatial error in Proposition 1. Thus the h^q term in (61) is Assumption 3.1 restated, not a proved consequence of the VMS stabilization or the fractional-step structure. The temporal O(dt^2) part is independently grounded, but the spatial rate remains a modeling hypothesis. Please either supply a proof (or a convincing derivation) of (59) for the nonlinear predictor and pressure Poisson equation under explicit regularity and SUPG/VMS hypotheses, or explicitly reframe Proposition 1 as a conditional statement in which the spatial rate is as","section":"§3.1.3, Assumption 3.1 and Eq. (60)-(61)"},{"comment":"The derivation of the SUPG-type fine-scale advection term uses the assumption that the total velocity is divergence-free: the text after Eq. (23) states that the term (div(u_h+u'))u' is dropped because 'we have assumed that the total velocity u = u_h + u' is divergence free.' This assumption is not valid for the intermediate velocity of the projection scheme. The paper itself notes (before Eq. (29)) that the intermediate velocity is not generally solenoidal. Nevertheless, the same expansion is used in (43a) and in Algorithm 1, and the corresponding term involving div(tilde u_h + tilde u') is absent. Assumption 3.1 does not state or justify that this omitted term is O(h^q). This is a consistency gap in the derivation of the predictor and in its propagation to the pressure equation. The authors should either include the omitted term in the fine-scale model, prove that it is of higher order","section":"§2.3 and §2.4.2, Algorithm 1, Step 1"},{"comment":"The Taylor-Green vortex DNS reference is cited as 'Wang et al., 2013 (512^3)' in Figures 12-15 and in the text. The cited reference [49] is 'High-order CFD methods: current status and perspective,' which does not appear to be a pseudo-spectral DNS dataset of the Taylor-Green vortex. As cited, this does not support the claim of agreement with a 512^3 pseudo-spectral DNS. If the data are taken from another source, please cite that source; if [49] does contain the DNS results used, please clarify the provenance in the text. This is important because the Taylor-Green comparison is a central validation claim for the method's accuracy.","section":"§4.4 and Reference [49]"}],"minor_comments":[{"comment":"Lemma 3.1 is referred to as 'Theorem 3.1' in the paragraph following Eq. (58) and in the proof of Proposition 1; Remark 3.2 refers to 'Theorem 3.2'. Please correct the cross-references.","section":"§3.1.2 and §3.2"},{"comment":"The explanation of the splitting defect is a bit terse. The statement that the pressure increment phi^n is O(dt) for the incremental scheme should be justified or accompanied by a citation; currently it is presented as self-evident.","section":"§3.1.2, Remark 3.2"},{"comment":"The boundary integral in the penultimate line of the momentum predictor is said to vanish when fine scales are neglected on the boundary (Remark 2.1). This is consistent, but it would help to explicitly mark the term as '(vanishing by assumption)' in Algorithm 1, since the equation as written appears to retain a boundary contribution.","section":"Algorithm 1, Step 1"},{"comment":"The legend entry 'VMS/monolithic, p' = 0' uses the prime symbol ambiguously; it should be clarified that this is the monolithic solver with the pressure fine scale set to zero.","section":"§4.3, Figures 10"}],"recommendation":"major_revision","confidential_remarks":"The paper has a useful numerical method and a solid experimental section, but the formal error analysis's spatial component is essentially an assumption. I would recommend requiring a substantial revision that either proves or explicitly downgrades the spatial-error claim. The questionable citation in the Taylor-Green section should also be resolved before acceptance."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's my read. The new configuration is real: residual-based VMS applied only to the predicted velocity in an incremental Helmholtz-Leray projection, with the modeled fine scale entering all three subproblems. I haven't seen that exact combination before, and it is a sensible structural choice given that the splitting removes the saddle point. The numerical work is careful: MMS shows second-order velocity convergence across a wide Reynolds range, cavity and cylinder results match established references, and the Taylor-Green comparison against monolithic VMS—including the p'=0 ablation and the dissipation-gap diagnostic—is convincing. The 1.3x-2.7x per-step speedups on identical solver settings are plausible and practically useful. The citation pattern is sound: temporal splitting error is grounded in Shen and van Kan, and VMS ingredients trace to the standard Hughes/Bazilevs line.\n\nThe soft spots are in the formal section. Proposition 1's O(h^q) spatial term is Assumption 3.1 restated; the 'proof' of that part is just subtraction of the velocity-update equations. The paper is explicit the estimate holds under the assumption, so it is not deceptive, but 'formal error estimate' is overselling a conditional decomposition. More problematic is carrying the divergence-free assumption into the predictor: Section 2.3 drops (div(u_h+u'))u' using div(u_h+u')=0, which is valid for the monolithic total velocity but not for the intermediate velocity in the split scheme. Nothing in Assumption 3.1 shows the omitted term is O(h^q), so consistency of the momentum predictor is not fully established. The paper itself concedes at the end that a rigorous energy-stability analysis is still open; that caveat belongs in the abstract.\n\nNo code or data are shipped, which now matters for a methods paper; the benchmarks are not independently reproducible as is.\n\nWho's this for? CFD numericists in stabilized FEM and projection methods. They get a new scheme worth testing and a careful comparison. I'd send it to referees, but with the instruction that the formal claims be relabeled or proved and the divergence-free gap addressed before publication.","headline":"A genuinely new and useful VMS-projection combination with strong benchmarks, carrying a formal error estimate that is more assumed than proved.","tokens_in":26638,"tokens_out":3662,"would_cite":true,"duration_ms":39495,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["65M60","65M15","76D05","76M10"],"pacs":[],"model":"deepseek-v4-flash","headline":"Stabilizing only the predicted velocity yields a second-order-accurate, cheaper Navier-Stokes projection scheme.","keywords":["projection methods","variational multiscale stabilization","Navier-Stokes equations","pressure-correction schemes","finite element methods","error analysis","Taylor-Green vortex","equal-order elements"],"falsifier":"Run the manufactured-solution test with a very fine fixed mesh and successively halve the time step at Re=10^6: the measured velocity error at the final time should fall at about dt^2. If the measured slope drops below 2, or if refining the mesh at fixed dt produces no improvement because the error plateaus, then respectively the temporal or spatial claim in Proposition 1 fails.","tokens_in":25785,"feed_emoji":"🌊","tokens_out":10024,"duration_ms":107521,"temperature":0.7,"pith_summary":"The paper tries to establish that the expensive fully coupled form of variational multiscale (VMS) stabilization for incompressible flow can be replaced by a fractional-step projection scheme in which only the intermediate velocity is decomposed into coarse and fine scales. Because the projection step removes the velocity-pressure saddle point, the pressure and the final divergence-free velocity need no fine-scale stabilization; the missing advective stability is supplied by a residual-based model of the predicted velocity's unresolved part, which also feeds a PSPG-like term into the pressure Poisson equation. Under stated regularity and spatial-approximation assumptions, the authors derive a combined velocity error of order dt^2 + h^q for BDF2 time stepping, separating time-discretization, splitting, and spatial errors. Numerical comparisons indicate that omitting the pressure fine scale avoids the monolithic scheme's excess dissipation and drag overprediction at the cost of larger divergence error, while reducing per-step solve time by 1.3x to 2.7x in the 3D Taylor-Green test. If correct, this provides an equal-order, second-order-in-time alternative to monolithic stabilized solvers that is cheaper and less dissipative.","feed_headline":"Projection method keeps VMS accuracy at 1.3-2.7x lower cost","feed_subtitle":"Stabilizing only the predicted velocity removes the pressure fine scale's excess dissipation and keeps second-order time accuracy.","key_machinery":"The load-bearing mechanism is the residual-based fine-scale model of the predicted velocity, ~u' = -tau_m * ~r_m, where ~r_m is the strong residual of the momentum predictor and tau_m is an elementwise stabilization time scale. This object supplies advective SUPG-type stabilization in the predictor, a PSPG-type divergence term in the pressure Poisson equation, and a correction contribution in the velocity projection. Around it sits the incremental Helmholtz-Leray splitting: a nonlinear advection-diffusion predictor with the previous pressure, a pressure Poisson equation for the pressure correction, and an L2 projection onto weakly divergence-free velocities. BDF2 gives the O(dt^2) temporal b","core_discovery":"The central claim, Proposition 1, is that the BDF2 incremental Helmholtz-Leray projection method with residual-based VMS stabilization on the predicted velocity satisfies a discrete velocity error of order dt^2 + h^q. The multiscale decomposition is applied only to the predicted velocity, whose unresolved part is modeled by the momentum residual; neither pressure nor corrected velocity is decomposed. The modeled fine scale enters all three substeps, producing SUPG-like stabilization in the predictor, a PSPG-like term in the pressure Poisson equation, and no pressure-fine-scale/grad-div term. On equal-order elements the scheme matches reference data for manufactured solutions, lid-driven cavi","pith_inferences":["If the grad-div term is indeed the sole cause of the monolithic scheme's excess dissipation, then a one-parameter family interpolating between full VMS and p'=0 should trace a smooth tradeoff between divergence error and over-dissipation; this is directly testable in the monolithic solver and would sharpen the attribution beyond the binary ablation.","The same splitting logic should transfer to rotational pressure correction, giving improved pressure accuracy while keeping the VMS machinery unchanged; the paper notes the rotational variant but does not combine it with the stabilized projection.","Because all three substeps are elliptic or nonlinear advection-diffusion and avoid saddle-point linear algebra, the measured cost advantage with the chosen Krylov solver is likely a lower bound; multigrid or GPU solvers on the substeps could widen the speedup without changing the formulation.","The formal h^q component rests entirely on Assumption 3.1, so proving that a standard SUPG/PSPG stability estimate delivers the assumed spatial consistency would turn Proposition 1 into a fully derived a priori bound rather than a conditional one."],"forward_implications":["On equal-order elements, the scheme should show second-order temporal velocity convergence without any inf-sup-stable element pair, because pressure stability comes from the PSPG-like term rather than a pressure fine scale.","Monolithic VMS's pressure fine scale / grad-div term is the actionable source of excess modeled dissipation and cylinder-drag overprediction; removing it lets the projection scheme track DNS kinetic-energy decay on coarser meshes than the monolithic scheme.","The tradeoff is explicit: suppressing the pressure fine scale raises velocity-divergence error to about the level of monolithic VMS with p'=0, so users accepting this method accept larger pointwise divergence in exchange for better energy behavior and lower cost.","Because the coupled nonlinear saddle-point system becomes three sequential subproblems, the method reduces per-step wall-clock solve time by factors 1.3-2.7x in the 3D Taylor-Green tests under identical solver settings.","The total error estimate O(dt^2 + h^q) means time step and mesh should be refined together; refining only one leaves the other term dominating."],"supporting_citations":[{"why":"Supplies the underlying second-order incremental pressure-correction splitting that the method inherits and stabilizes.","marker":"[48]"},{"why":"Supplies the O(dt^2) splitting-error estimate for the continuous-in-space incremental scheme, used as the temporal splitting component of Proposition 1.","marker":"[41]"},{"why":"Supplies the survey of projection methods and the non-rotational scheme's first-order pressure boundary-layer limitation invoked for the pressure estimate.","marker":"[24]"},{"why":"Supplies the variational multiscale decomposition and fine-scale modeling concept that the method adapts.","marker":"[26]"},{"why":"Supplies the residual-based VMS closure form and the simplifying assumptions about fine-scale time-derivative, viscous, and boundary terms used in Remark 2.1.","marker":"[4]"},{"why":"Supplies the SUPG stabilization term that the fine-scale contribution reduces to in the momentum predictor.","marker":"[9]"},{"why":"Supplies the PSPG pressure stabilization term that the fine-scale contribution reduces to in the pressure Poisson equation.","marker":"[27]"},{"why":"Supplies the pseudo-spectral DNS reference used to validate the Taylor-Green kinetic-energy and dissipation comparisons.","marker":"[49]"}],"fun_headline_variants":["Projection VMS: up to 2.7x faster, same second-order accuracy","Velocity-only fine scaling speeds VMS 1.3-2.7x","Cheaper Navier-Stokes: VMS projection without pressure fine scale","Split-step VMS: 2.7x speedup in Taylor-Green, accurate","No saddle-point: Helmholtz-Leray VMS runs faster, stays accurate"],"cache_read_input_tokens":2688,"weakest_assumption_plain":"The load-bearing premise is that the stabilized spatial discretization of the three fractional subproblems achieves the assumed error bound of order h^q; the proof does not derive this bound, so if the spatial accuracy assumption fails, the mesh part of the headline estimate collapses while the temporal O(dt^2) part remains.","fun_headline_variants_meta":{"raw":{"variants":["Projection VMS: up to 2.7x faster, same second-order accuracy","Velocity-only fine scaling speeds VMS 1.3-2.7x","Cheaper Navier-Stokes: VMS projection without pressure fine scale","Split-step VMS: 2.7x speedup in Taylor-Green, accurate","No saddle-point: Helmholtz-Leray VMS runs faster, stays accurate"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000382,"raw_usage":{"total_tokens":1905,"prompt_tokens":828,"completion_tokens":1077,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":572,"completion_tokens_details":{"reasoning_tokens":969}},"tokens_in":572,"tokens_out":1077,"duration_ms":10435,"temperature":1.0,"reasoning_tokens":969,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-05T00:11:31.366914+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the manufactured-solution test with a very fine fixed mesh and successively halve the time step at Re=10^6: the measured velocity error at the final time should fall at about dt^2. If the measured slope drops below 2, or if refining the mesh at fixed dt produces no improvement because the error plateaus, then respectively the temporal or spatial claim in Proposition 1 fails.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the underlying second-order incremental pressure-correction splitting that the method inherits and stabilizes."},{"cited_title":"On error estimates of the projection methods for the navier-stokes equations: second- order schemes.Mathematics of computation, 65(215):1039–1065, 1996","cited_arxiv_id":null,"evidence_quote":"Supplies the O(dt^2) splitting-error estimate for the continuous-in-space incremental scheme, used as the temporal splitting component of Proposition 1."},{"cited_title":"An overview of projection methods for incom- pressible flows.Computer Methods in Applied Mechanics and Engineering, 195(44–47):6011–6045, 2006","cited_arxiv_id":null,"evidence_quote":"Supplies the survey of projection methods and the non-rotational scheme's first-order pressure boundary-layer limitation invoked for the pressure estimate."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the variational multiscale decomposition and fine-scale modeling concept that the method adapts."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the residual-based VMS closure form and the simplifying assumptions about fine-scale time-derivative, viscous, and boundary terms used in Remark 2.1."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the SUPG stabilization term that the fine-scale contribution reduces to in the momentum predictor."},{"cited_title":"A new finite element formulation for computational fluid dynamics: V","cited_arxiv_id":null,"evidence_quote":"Supplies the PSPG pressure stabilization term that the fine-scale contribution reduces to in the pressure Poisson equation."},{"cited_title":"High-order cfd methods: current status and perspective.International Journal for Numerical Methods in Fluids, 72(8):811–845, 2013","cited_arxiv_id":null,"evidence_quote":"Supplies the pseudo-spectral DNS reference used to validate the Taylor-Green kinetic-energy and dissipation comparisons."}],"review_version":1}