{"id":"ab98330c-cb6d-447d-bd92-2dc4ab7c5b36","arxiv_id":"2607.02739","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":4,"one_line_summary":"The a priori bound on (d/dt)||u||_q^q for 3D Navier-Stokes is sharp up to a prefactor, as shown by Riemannian conjugate-gradient maximizers for several q>3.","lead":"Numerical optimization finds velocity fields that nearly saturate a classic a priori bound on how fast Lq norms of Navier-Stokes velocity can grow. The result indicates the bound cannot be fundamentally improved and that growth rates diverge as q approaches 3.","discovery_kind":"extension","skeptic_critique":{"model":"grok-4.5","headline":"Local maximizers from ABC continuation may not realize the true supremum of R_q as B\to∞, so the observed power-law saturation of (5) could be branch-specific rather than sharp.","rationale":"The Reader correctly isolates the local-vs-global and resolution issues as the weakest link. The paper’s evidence is clean for the branches it computes, the methodology is a careful extension of prior work, and the measured exponents match the analytic upper bound to within a few percent. That is enough for CONDITIONAL acceptance, but not for an unconditional claim that the bound is sharp. The concrete multi-start test above is the minimal check that would either confirm the ABC branch is representative or expose a higher-growth family. No stronger objection (e.g., an internal inconsistency in the derivation of R_q or a clear numerical artifact) is visible; the concern is precisely the one the Reader already flagged.","tokens_in":22720,"tokens_out":636,"duration_ms":6966,"concrete_test":"Restart Algorithm 1 from at least two qualitatively different initial fields (e.g., a pair of colliding vortex rings of the type that saturate the enstrophy bound, and a random divergence-free field with the same ||u||_q = B_init) for q=5 and B up to ~10^{2}; if any new branch yields a compensated rate R_5 / B^{10} that exceeds the ABC-branch prefactor by more than ~20 % or produces a fitted exponent >10.1, the global-supremum assumption fails and the sharpness claim must be weakened.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central claim (bound (5) is sharp up to a prefactor) rests on the numerical observation that R_q(eu_B) ~ C B^{q(q-1)/(q-3)} for the branches computed in §5 (Table 3, Figs. 4–5). Those branches are obtained exclusively by Riemannian CG continuation from the ABC eigenfunction that maximizes the small-data limit (§3, Alg. 1). The paper itself notes that the optimizers are only local (Problem 3 is non-convex), that they become increasingly localized and approach the edge of H^{3/2-1/q} regularity (Fig. 2b), and that for q=3 the same iteration diverges. Nothing rules out the existence of other families of fields (different topology, different localization scale, or fields outside the Hilbert embedding) that produce a strictly larger growth rate, possibly with a higher exponent. If such fields exist, the measured exponents would only lower-bound the true supremum and the claim that (5) “cannot be fundamentally improved” would not follow. Finite resolution (N≤1024^{3}) and the particular choice of retraction/transport further leave open the possibility that the observed scaling is an artifact of the continuation path rather than a property of the global maximizer.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.5","summary":"The paper investigates the sharpness of the a priori bound (5) on the instantaneous growth rate of the L^q norm of velocity for 3D Navier-Stokes flows (q>3), which is closely tied to the Ladyzhenskaya-Prodi-Serrin conditions. An objective functional R_q(u) for (d/dt)||u||_q^q is derived (Eq. (15)), and maximizers subject to fixed ||u||_q=B, divergence-free and zero-mean constraints are sought via a Riemannian conjugate-gradient method on the Hilbert manifold X_B subset H^{3/2-1/q} (Problem 3). Branches of local maximizers are continued from the small-data ABC eigenfunction (Section 3, Algorithm 1). Numerical results for q=4,5,6,9 show that R_q(eu_B) saturates the power q(q-1)/(q-3) as B\to∞ (Table 3, Figures 4–5), up to a numerical prefactor, while the q\to3 limit appears ill-posed. The optimizers become increasingly localized and approach the edge of the ambient Sobolev space.","tokens_in":23090,"tokens_out":1018,"duration_ms":8404,"significance":"If the numerical evidence is accepted as representative, the work supplies concrete computational support that bound (5) cannot be improved in the exponent, complementing earlier variational studies of enstrophy growth (Problem 1) and completing the picture summarized in Table 1. The derivation of R_q, the small-data analysis, and the Riemannian CG formulation are clean and reusable. The observation that the same states do not simultaneously saturate the enstrophy bound, and that the q=3 problem appears ill-posed, are of independent interest for the conditional-regularity program. Strengths include transparent appendices (A–C), explicit power-law fits, and careful spectral monitoring.","major_comments":[{"comment":"The central claim that bound (5) “cannot be fundamentally improved” rests on local maximizers obtained exclusively by continuation from the ABC flow (Algorithm 1, Section 5). Problem 3 is non-convex; nothing rules out other branches (different topology or localization) that could produce a strictly larger growth rate or a higher exponent. The paper should either (i) attempt restarts from qualitatively different initial data at large B, or (ii) rephrase the claim as “sharp along the computed branches / lower bound on the true supremum.” Without this, the global-sharpness language in the abstract and Section 6 overreaches the evidence.","section":null},{"comment":"Figures 4–5 and Table 3 report power-law saturation, yet the maximizers approach the edge of H^{3/2-1/q} (Fig. 2b) and become highly localized while resolution is capped at N=1024^{3}. No systematic resolution study or extrapolation of the fitted exponents eα versus N is provided. A short convergence check (e.g., recompute a large-B point at two resolutions and report the change in R_q and eα) is needed to confirm that the observed scaling is not an artifact of under-resolution or of the particular filter length ℓ=0.1.","section":null}],"minor_comments":[{"comment":"Table 1 and the abstract state that the bound is sharp “up to a numerical prefactor,” yet the measured prefactors eC in Table 3 are extremely small (10^{-15}–10^{-4}). A brief remark on the practical size of the constant would help readers assess how close the bound is to being saturated in absolute terms.","section":null},{"comment":"Section 5.4 (q\to3) reports diverging exponents with “relatively large uncertainties.” The fitting procedure and the range of B used for those fits should be stated more precisely so that the claimed divergence can be reproduced.","section":null},{"comment":"Typographical slips: “formatioon” (keywords), “ealier” (Section 6), and occasional missing spaces around math operators. A light copy-edit pass would suffice.","section":null},{"comment":"Figure 6 captions refer to “normalized” fields but do not specify the normalization; a short clarification would improve readability.","section":null}],"recommendation":"minor_revision","confidential_remarks":"The methodological lineage (Riemannian CG on constraint manifolds for NS extremals) is solid and the paper fits the journal’s scope. The main risk is over-statement of global sharpness from a single continuation branch; once the language is tempered or additional restarts are shown, the contribution is publishable. No concerns about novelty disclosure or citation pattern."},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"The new piece is clean: they take the classical a-priori bound on (d/dt)||u||_q^q, turn the right-hand side into a constrained optimization problem on the divergence-free manifold, and produce concrete fields that realize the same power of ||u||_q as B\to∞ for several q>3. That is the first numerical saturation evidence for this particular estimate, and it sits usefully next to the earlier enstrophy work by the same group.\n\nWhat they do well is the technical execution. The objective functional is derived carefully (Appendices A and C), the small-data ABC analysis is exact, and the Riemannian CG continuation is standard but carefully implemented. The power-law fits in Table 3 and the compensated plots in Figure 5 line up with the analytic exponent to within a few percent; the spectra stay resolved until the edge of the Hilbert space. They also show that these maximizers do not saturate the enstrophy bound, which is a useful negative observation.\n\nThe soft spot is exactly the one the stress-test flags: everything is local maximizers obtained by continuation from the ABC flow. The problem is non-convex, the fields localize and approach the regularity edge, and nothing rules out a different family with a higher growth rate. Finite resolution (N≤1024^{3}) and the particular retraction leave a residual doubt that the observed scaling is path-dependent. That is a genuine limitation, but it is the same limitation that has always attached to this variational program; it does not make the measured exponents meaningless. They still give a rigorous lower bound on the best constant and show that the analytic exponent is attainable.\n\nThis is for people who care about conditional regularity and about how sharp the classical estimates actually are. The math is standard, the numerics are reproducible in principle, and the citation pattern is honest. I would send it to referees; the claim is important enough inside the field and the evidence is strong enough to deserve a careful look. I would cite the scaling measurements myself.","headline":"Solid numerical evidence that the Robinson–Sadowski Lq-growth bound is saturated in the exponent by concrete local maximizers; the local-vs-global caveat is real but does not erase the result.","tokens_in":23649,"tokens_out":509,"would_cite":true,"duration_ms":6482,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["35Q30","49M41","65N35","76D05"],"pacs":[],"model":"grok-4.5","headline":"A classic upper bound on how fast Navier-Stokes velocity norms can grow is sharp up to a numerical constant.","keywords":["Navier-Stokes equations","Ladyzhenskaya-Prodi-Serrin conditions","Lebesgue norms","a priori bounds","sharpness","Riemannian conjugate gradient","singularity formation"],"falsifier":"An independent maximisation of the same objective, started from a qualitatively different family of initial fields or run at substantially higher resolution, that produces a strictly slower growth rate whose scaling is weaker than the predicted power of the Lq norm.","tokens_in":23635,"feed_emoji":"≈","tokens_out":624,"duration_ms":5641,"temperature":0.7,"pith_summary":"The paper asks whether a standard a priori estimate that controls the instantaneous growth of Lq norms of the velocity in three-dimensional Navier-Stokes flows can be improved. That estimate is intimately tied to the Ladyzhenskaya-Prodi-Serrin conditions that guarantee smoothness. By solving a constrained variational problem that maximises the growth rate for fixed Lq norm, the authors construct families of velocity fields whose growth realises the same power of the norm that appears on the right-hand side of the bound. The conclusion is that the bound cannot be strengthened in its scaling; only the prefactor might still be refined. The same computations also indicate that the growth rate becomes unbounded as the exponent q approaches the critical value 3.","feed_headline":"Navier-Stokes growth bound is sharp up to a constant","feed_subtitle":"Optimised velocity fields realise the same power of the Lq norm that the classical estimate allows.","key_machinery":"A Riemannian conjugate-gradient method on the Hilbert manifold of divergence-free, zero-mean fields with fixed Lq norm, maximising the objective functional Rq that expresses the instantaneous growth rate after elimination of pressure via the Poisson equation.","core_discovery":"Numerical maximisers of the instantaneous rate of growth of the Lq norm of velocity, obtained for several q>3 and for norms spanning several orders of magnitude, saturate the power-law upper bound d/dt ||u||_q^q ≤ C ||u||_q^{q(q-1)/(q-3)} as the norm tends to infinity. The bound is therefore sharp up to a numerical prefactor and cannot be fundamentally improved.","pith_inferences":[],"forward_implications":[],"fun_headline_variants":["Numerical maximisers show Navier-Stokes Lq growth bound is sharp","Optimised velocity fields saturate the classical Lq growth bound","Maximisers confirm NS bound on ||u||_q rate cannot be improved","Instantaneous Lq maximisers saturate power-law growth estimate","Bound on velocity Lq growth in Navier-Stokes is sharp up to factor"],"cache_read_input_tokens":16512,"weakest_assumption_plain":"That the local maximisers found by continuing from the small-data ABC flow remain representative of the true global maximum growth rate at large norms, and are not artefacts of finite resolution or of the particular geometric operations used on the constraint manifold.","fun_headline_variants_meta":{"raw":{"variants":["Numerical maximisers show Navier-Stokes Lq growth bound is sharp","Optimised velocity fields saturate the classical Lq growth bound","Maximisers confirm NS bound on ||u||_q rate cannot be improved","Instantaneous Lq maximisers saturate power-law growth estimate","Bound on velocity Lq growth in Navier-Stokes is sharp up to factor"]},"model":"grok-4.5","effort":"low","cost_usd":0.00432,"raw_usage":{"total_tokens":1276,"prompt_tokens":786,"num_sources_used":0,"completion_tokens":97,"cost_in_usd_ticks":43200000,"prompt_tokens_details":{"text_tokens":786,"audio_tokens":0,"image_tokens":0,"cached_tokens":128},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":393,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":786,"tokens_out":97,"duration_ms":5327,"temperature":1.0,"reasoning_tokens":393,"cache_read_input_tokens":128,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-12T07:21:47.035854+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"An independent maximisation of the same objective, started from a qualitatively different family of initial fields or run at substantially higher resolution, that produces a strictly slower growth rate whose scaling is weaker than the predicted power of the Lq norm.","supporting_citations":[],"review_version":1}