{"id":"5385396a-9ad9-49d8-ad6f-face4bc646c7","arxiv_id":"1908.02896","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Two-dimensional wall-to-wall optimal heat transport is computed up to Peclet 1e5, showing Nu ~ Pe^0.54, with a conditional upper bound Nu <= C Pe^{6/11} under a separability ansatz.","lead":"Using computational optimization, this paper finds two-dimensional fluid flows that move heat between parallel walls as efficiently as possible, given a fixed amount of flow stirring. The computed flows show heat transport growing roughly like the 0.54 power of the stirring rate, and a conditional mathematical bound supports this scaling.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The Pe^{6/11} bound requires a single horizontal profile φ for velocity and temperature; the SVD evidence uses different profiles φ1 and φ2, so the link between the numerical scaling and the conditional bound is not established.","rationale":"The reader's CONDITIONAL verdict is appropriate. The numerical scaling has external support from the cited independent computations of Motoki et al. (2018b) and Ding & Kerswell (2019), and the paper clearly discloses that the analytic result is conditional. The most load-bearing unchecked step is not merely the presence of non-separable modes, which the SVD shows to be small, but the stronger requirement of the proof: a common φ in (5.3)-(5.5). Because the SVD's transport-capture statistic uses different φ1 and φ2, it does not bound the common-φ projection error. A projection-plus-simulation test would settle whether the conditional bound explains the numerics. If the test fails, the paper's two headline claims remain logically disconnected and the conditional upper bound, while mathematically true for its subclass, would not constrain the computed optimizers. The verdict stays CONDITIONAL rather than REJECT because the authors explicitly label the bound conditional and the numerical scaling is separately credible.","tokens_in":24908,"tokens_out":24191,"duration_ms":249982,"concrete_test":"Project each computed optimizer onto the common-φ separable manifold by minimizing, for a single periodic φ(x), the combined error ||ψ(z,x)−Ψ(z)φ(x)||² + ||ξ(z,x)−Ξ(z)φ(x)||² over Ψ, Ξ, φ, with φ normalized. Then solve the steady advection-diffusion equation (1.2) with the projected velocity field u_p = (−Ψ′φ, Ψφ′) and compute Nu(u_p) over Pe ∈ [10³, 10⁵]. If Nu(u_p) stays within a few percent of the full optimal Nu and the local slope d log(Nu−1)/d log Pe stays within ±0.02 of the reported value, the separable ansatz is validated as transport-negligible; if the projected transport drops significantly or the slope changes, the Pe^{6/11} bound cannot be invoked for the computed optimizers.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The load-bearing gap is the transfer of the numerical scaling to the conditional upper bound. Section 5 proves Nu ≲ Pe^{6/11} only within the exact separable ansatz (5.3)-(5.5), which uses one horizontal profile φ(x) for both the velocity (u1,u3) and the adjoint temperature ξ. The numerical evidence in §4.1 is rank-one, but it does not verify this common-φ form. The first singular profiles for ψ and ξ are different functions, called φ1 and φ2 in (4.6)-(4.7), and the reported 99% accuracy measure (N1−N2)/N1 ≤ 0.01 uses N2 = ⟨Ψ(∂xφ1)Ξφ2⟩ with those two distinct profiles. That measures how much transport is captured by the leading singular triad, not the error incurred by imposing φ1 = φ2 = φ. If the optimal flows' horizontal structures for velocity and temperature differ at the few-percent level, the analyzed separable subclass excludes them, and the Pe^{6/11} bound has no demonstrated bearing on the computed Nu ∼ Pe^{0.54}. The authors are explicit about this limitation, but it remains the key unproven assumption connecting the two headline claims.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper addresses the wall-to-wall optimal transport problem in two dimensions with no-slip isothermal boundaries. The authors develop pseudo-spectral gradient-ascent schemes based on the variational formulation of the Nusselt number, and compute steady locally maximizing flows for Peclet numbers up to approximately 10^5, reporting a transition from Nu-1 ~ Pe^2 to Nu ~ Pe^{0.54} and an optimal aspect-ratio scaling Gamma ~ Pe^{-0.37}. They observe that the computed fields are nearly rank-one via singular value decomposition, which motivates a separable ansatz. Under that ansatz, they derive a conditional upper bound Nu <= C Pe^{6/11} and discuss connections to the background method, Howard-Busse-Malkus theory, and a conjectured duality gap. The paper is explicitly careful to label the analytic bound as conditional, and it notes that the computed finite-Pe scaling cannot persist for global optimizers as Pe tends to infinity in light of known lower bounds.","tokens_in":25181,"tokens_out":12159,"duration_ms":125485,"significance":"The paper is significant because it provides the most complete numerical map to date of locally optimal two-dimensional wall-to-wall transport in the no-slip case, identifies a clear finite-Pe scaling regime, and supplies a conditional analytic bound at the same exponent. The variational machinery connecting the wall-to-wall problem to the background method and Howard's functional is elegantly presented, and the polynomial example illustrating a possible duality gap is instructive. The authors are explicit that the analytic bound is conditional on the separable ansatz, and they correctly note that the computed Pe^{0.54} branch cannot represent global optimizers at asymptotically large Pe in light of the Tobasco-Doering lower bound. The main weaknesses are that the numerical evidence does not directly validate the common horizontal profile required by the ansatz, that the derivation of the bound contains an incorrect identity that must be repaired, and that the numerical scaling lacks rigorous error or resolution estimates.","major_comments":[{"comment":"The diagnostic (N1-N2)/N1 <= 0.01 reported in Section 4.1 uses two different horizontal profiles phi1 and phi2 in the leading-rank approximation (4.6)-(4.7), so it measures the fraction of transport carried by the leading singular triad rather than the error incurred by imposing the common-profile ansatz phi1 = phi2 = phi that is used in the bound of Section 5.1. Consequently the computed Nu ~ Pe^{0.54} branch is not demonstrated to lie within the separable class analyzed, and the claimed accord between the numerics and the Pe^{6/11} bound is not established. I ask the authors to add a direct test of the common-phi ansatz, for example by computing the transport of the projection of the optimal fields onto a single horizontal profile, or to soften the connection between the two results.","section":"Section 4.1 and Section 5.1, Eqs. (4.6)-(4.9) and (5.3)-(5.5)"},{"comment":"The identity <(u3 xi - <u3 xi>)^2> = integral (Psi Xi - 1)^2 dz is not correct when phi(x) is nonconstant: with the normalization <(phi')^2> = 1, horizontal averaging yields <(u3 xi - <u3 xi>)^2> = integral (Psi Xi)^2 <(phi')^4> dz - (integral Psi Xi dz)^2, which generally exceeds integral (Psi Xi - 1)^2 dz. The lower-bound argument still goes through if the equality is replaced by the inequality '>=' (justified by Jensen's inequality since <(phi')^4> >= <(phi')^2>^2 = 1), but the derivation as written is invalid and must be corrected.","section":"Section 5.1, Eq. (5.11)"},{"comment":"The central numerical scaling Nu ~ Pe^{0.54} is inferred from local logarithmic slopes at the largest computed Pe, yet the paper reports no error bars, resolution or convergence studies, or statement of data availability. Because the local slope oscillates (bottom-left panel of Figure 2), the single value 0.544 at the largest Pe is not by itself a robust estimate of the exponent over the claimed range, and the manuscript should provide a quantitative uncertainty estimate or additional resolution checks.","section":"Section 4, Figure 2"}],"minor_comments":[{"comment":"The abstract and several places describe the computed flows as 'maximizing' or 'optimal' without the qualifier 'locally'; since the optimization problem is non-convex and the method finds local maxima, please add 'locally' to these statements to avoid overstatement.","section":"Abstract and Section 1"},{"comment":"There is a typo in the keywords: 'variational methds' should be 'variational methods'.","section":"Keywords"},{"comment":"In the first paragraph, 'complimentary role' should be 'complementary role'.","section":"Introduction"},{"comment":"The inequalities in the derivation of the conditional bound omit all constants; since the result is an upper bound with an unspecified prefactor, a sentence stating that all constants are finite and depend only on the boundary conditions would clarify the logical status of the '>' signs.","section":"Section 5.1"},{"comment":"The overline notation for horizontal averages is introduced in Section 2.6 but is reused in Section 5.1 in expressions such as phi^2 and (phi'')^2 without an explicit reminder of the convention; please make the notation self-contained in Section 5.","section":"Section 2.6 and Section 5.1"}],"recommendation":"major_revision","confidential_remarks":"The paper is likely of genuine interest to JFM readers and is honestly written, with the conditional nature of the analytic bound clearly disclosed. The main obstacle is the gap between the SVD evidence and the common-phi separable ansatz; this is fixable by a targeted numerical test. The incorrect identity in Eq. (5.11) is also fixable and does not appear to sink the argument once corrected to an inequality. I would not reject the paper, but the requested revisions are substantive because they affect the connection between the two headline claims."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a solid, honest paper whose main result is explicitly conditional, and the gap the reader flagged is real: the numerics do not actually verify the exact separability ansatz used for the bound. Still worth sending out.\n\nWhat is new: the conditional upper bound Nu ≲ Pe^{6/11} within a single-profile separable ansatz, and the systematic 2D no-slip computations up to Pe ~ 2.5e5 showing local optima with Nu ~ Pe^0.54. The numerical scaling by itself was already reported by Souza and confirmed by Motoki et al. and Ding & Kerswell; the paper says so. The theory section connecting the wall-to-wall functional to the background method and the Howard-Busse-Malkus functional, plus the conjectured duality gap, is a nice framing and the polynomial example is helpful.\n\nThe main soft spot is exactly what the stress-test note says. Section 4.1's dominant-mode measure uses different horizontal profiles φ1 and φ2 for the streamfunction and ξ. The ansatz (5.3)-(5.5) requires one φ for both u and ξ. The 99% transport-capture number does not constrain the difference between φ1 and φ2, so the Pe^{6/11} bound has no demonstrated bearing on the computed optimizers. The authors are explicit that the bound is conditional, but that condition is stronger than the numerical evidence supports. This is the load-bearing gap between the two headline claims.\n\nOther weaknesses are minor and standard: no code, no data, no error bars; the optima are local, so the computed scaling is a lower bound on the true max, not an upper bound. The conditional proof itself looks correct within its class, and the numerics are plausible given the resolution and the multiple starting points used.\n\nWho should read it: anyone working on variational bounds for convection or the wall-to-wall problem. It is a useful contribution despite the gap, because it sharpens where the difficulty lies. I would accept it for peer review and let referees ask for data, error bars, and ideally a quantification of how different φ1 and φ2 actually are.","headline":"The paper's headline bound is honestly conditional, and the numerics don't quite verify the separability assumption the bound uses—still a worthwhile contribution for the convection-bounds crowd.","tokens_in":25706,"tokens_out":3041,"would_cite":true,"duration_ms":32544,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Two-dimensional wall-to-wall heat-transport optimizers switch from Nu - 1 ~ Pe^2 to Nu ~ Pe^0.54, with a separable-ansatz upper bound of Pe^{6/11}.","keywords":["wall-to-wall optimal transport","Nusselt number","Péclet number","gradient ascent","separable ansatz","Rayleigh-Bénard convection","variational bounds","no-slip boundary conditions"],"falsifier":"Run gradient ascent from many random initial conditions at $Pe\\approx10^5$ and compare the best Nusselt number with the value on the published $Nu$–$Pe$ curve; if any flow exceeds that curve by more than numerical discretization error, the claimed optimal scaling fails. Alternatively, project the computed optimizer onto the separable subspace at that $Pe$ and measure $(N_1-N_2)/N_1$: the paper reports this transport error is at most 1%, so an independent computation that finds a non-separable optimizer with $(N_1-N_2)/N_1$ substantially larger than 1% would show that the separability assumption is the active restriction.","tokens_in":24728,"feed_emoji":"🔥","tokens_out":13027,"duration_ms":126974,"temperature":0.7,"pith_summary":"This paper asks which incompressible flow between two parallel no-slip walls, at a fixed root-mean-square vorticity (the Péclet number $Pe$), transports the most heat. Using gradient ascent on a variational formulation, it computes two-dimensional steady optimizers up to $Pe\\approx10^5$ and finds that the Nusselt number $Nu$, the total heat transport divided by the conductive value, grows as $Nu-1\\sim Pe^2$ for small $Pe$ and then as $Nu\\sim Pe^{0.54}$ for $Pe>10^3$. The computed optimizing fields are nearly separable, products of a wall-normal and a wall-parallel profile, and the dominant mode carries about 99% of the transport. Restricting the variational problem to exactly separable fields yields a conditional upper bound $Nu\\lesssim Pe^{6/11}=Pe^{0.\\overline{54}}$, matching the computed scaling. This matters because upper bounds on wall-to-wall transport convert directly into upper bounds on heat transport in Rayleigh-Bénard convection through $Pe^2=Ra(Nu-1)$.","feed_headline":"Optimal 2D wall-to-wall heat transport scales as Pe^0.54","feed_subtitle":"Gradient-ascent flows and a separable-product bound agree on Nusselt scaling through Pe 10^5.","key_machinery":"The engine is the symmetrized variational functional $$S = \\langle |\\nabla\\eta|^2 + 2\\xi u\\cdot(\\hat z-\\nabla\\eta) - |\\nabla\\xi|^2 + \\mu($Pe^{2}$-|\\nabla u|^2)\\rangle,$$ obtained from the Nusselt functional by writing the temperature deviation as $\\theta=\\xi+\\eta$ and the adjoint variable as $\\phi=\\xi-\\eta$; its saddle points are the wall-to-wall optimizers. Gradient ascent on $S$—ascent in $\\xi$ and $u$, descent in $\\eta$—produces the computed flows. For the analytic bound, the key object is the Howard-type functional $$M[\\$\\sigma$,v;Pe] = \\frac{(\\langle v_3\\$\\sigma$\\rangle)^2}{$Pe^{{-2}}$\\langle|\\nabla\\$\\sigma$|^2\\rangle\\langle|\\nabla v|^2\\rangle + \\langle(v_3\\$\\sigma$-\\langle v_3\\$\\sigma$\\rangle)^2\\rangle},$$ which bounds $Nu-1$ from above. Under the separable ansatz, elementary inequalities—Cauchy-Schwarz, the interpolation bound $[(\\varphi'')^2][\\varphi^2]\\ge1$, and Howard's lemma—give a denominator bounded below by $\\delta + Pe^{-2}\\delta^{-8/3}$, whose minimum is $\\sim Pe^{-6/11}$. That minimization is what produces the conditional upper bound $Nu\\lesssim Pe^{6/11}$.","core_discovery":"The central claim is that in two dimensions with no-slip boundaries, the local maximizers of the wall-to-wall transport problem—flows maximizing convective heat transfer at prescribed enstrophy—switch scaling at $Pe\\sim10$, entering a regime where $Nu\\sim Pe^{0.54}$ that persists to at least $Pe\\approx2.5\\times10^5$. Numerically, the optimizers are nearly separable in their singular-value decomposition, with the first mode carrying more than 99% of $Nu-1$, and the horizontal profile resembles Jacobi elliptic functions rather than sinusoids. Under the explicit separable ansatz $u_1=-\\Psi'(z)\\varphi(x)$, $u_3=\\Psi(z)\\varphi'(x)$, $\\xi=\\Xi(z)\\varphi'(x)$, the Howard-type functional is bounded above by $Pe^{6/11}=Pe^{0.\\overline{54}}$, so the computation and the conditional analysis agree. The paper also notes that this cannot be the global asymptotic scaling because existing lower bounds give $\\max Nu > C\\,Pe^{2/3}/(\\log Pe)^{4/3}$ as $Pe\\to\\infty$; the $Pe^{0.54}$ behavior is therefore presented as the optimal scaling in the computed moderate-$Pe$ range.","pith_inferences":["A natural testable extension is to prove an energy estimate on the second singular mode's contribution to $Nu-1$; if that contribution can be bounded uniformly small up to $Pe\\sim10^5$, the conditional $Pe^{6/11}$ bound would become unconditional over that range.","The near-separability of the optimizers suggests a practical control conclusion the paper does not draw: in two-dimensional settings, actuating a single large-scale roll (the dominant mode) should recover roughly 99% of the optimal heat transport, so low-order controllers may suffice.","The oscillatory relaxation of the local scaling exponent hints at a sequence of modal transitions; one could search for additional 'branched' optimizers slightly beyond $Pe=10^5$ and predict the onset where the exponent moves from $0.54$ toward $2/3$.","The same separable-ansatz machinery could be applied to three-dimensional wall-to-wall problems; a non-separable ansatz that produced a bound below $Pe^{6/11}$ in two dimensions would overturn the conclusion that the separable class captures the extremal scaling."],"forward_implications":["For two-dimensional no-slip wall-to-wall flows, optimal transport follows $Nu\\sim Pe^{0.54}$ over $Pe\\in[10^3,10^5]$, which translates in Rayleigh-Bénard variables to the single-wavenumber upper-bound scaling $Nu\\lesssim Ra^{3/8}$.","The optimizing fields are nearly separable, with the leading singular mode accounting for at least 99% of $Nu-1$; their horizontal profiles are Jacobi-elliptic-like rather than single sinusoids.","The conditional $Pe^{6/11}$ bound cannot describe the global maximum at arbitrarily large $Pe$, since rigorous lower bounds grow like $Pe^{2/3}/(\\log Pe)^{4/3}$; the $Pe^{0.54}$ regime is a moderate-$Pe$ plateau that branching structures should eventually replace.","The optimal aspect ratio shrinks with enstrophy budget, with $\\Gamma\\sim Pe^{-0.37}$ in the nonlinear regime, so optimal cells narrow as $Pe$ increases.","The gradient-ascent framework transfers to other geometries and boundary conditions wherever Poisson and Stokes solvers are available."],"supporting_citations":[{"why":"Defines the wall-to-wall optimal transport problem that this paper maximizes.","marker":"Hassanzadeh et al. (2014)"},{"why":"Supplies the single-wavenumber variational functional and the lemma used to derive the Pe^{6/11} bound.","marker":"Howard (1963)"},{"why":"Provides the background-method comparison and the form of Howard's lemma used in the conditional bound.","marker":"Doering & Constantin (1996)"},{"why":"Gives the exact variational characterization of Nu-1 and the Pe^{2/3} lower bound that situates the computed scaling.","marker":"Tobasco & Doering (2017)"},{"why":"Extends the variational characterization and lower-bound construction relevant to the asymptotic discussion.","marker":"Doering & Tobasco (2019)"},{"why":"Reports three-dimensional optimal transport reaching ~Pe^{2/3}, providing the dimensional contrast that sharpens the two-dimensional result.","marker":"Motoki et al. (2018a)"},{"why":"Independently confirms the two-dimensional no-slip scaling reported here.","marker":"Ding & Kerswell (2019)"}],"fun_headline_variants":["2D wall-to-wall heat transport hits Pe^0.54 scaling","Optimal 2D flows: Nusselt scales as Pe^0.54","Gradient ascent reveals Pe^0.54 transport bound","Separable ansatz confirms Pe^0.54 heat transport","Wall-to-wall optimal transport: Nu ~ Pe^0.54"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the optimizing fields are exactly, or negligibly close to, separable products of wall-normal and wall-parallel profiles, $u_1=-\\Psi'(z)\\varphi(x)$, $u_3=\\Psi(z)\\varphi'(x)$, $\\xi=\\Xi(z)\\varphi'(x)$; the paper provides no proof that non-separable parts contribute negligibly to transport over the computed $Pe$ range.","fun_headline_variants_meta":{"raw":{"variants":["2D wall-to-wall heat transport hits Pe^0.54 scaling","Optimal 2D flows: Nusselt scales as Pe^0.54","Gradient ascent reveals Pe^0.54 transport bound","Separable ansatz confirms Pe^0.54 heat transport","Wall-to-wall optimal transport: Nu ~ Pe^0.54"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.0007,"raw_usage":{"total_tokens":3209,"prompt_tokens":1039,"completion_tokens":2170,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":655,"completion_tokens_details":{"reasoning_tokens":2077}},"tokens_in":655,"tokens_out":2170,"duration_ms":15530,"temperature":1.0,"reasoning_tokens":2077,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T14:30:53.994185+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run gradient ascent from many random initial conditions at $Pe\\approx10^5$ and compare the best Nusselt number with the value on the published $Nu$–$Pe$ curve; if any flow exceeds that curve by more than numerical discretization error, the claimed optimal scaling fails. Alternatively, project the computed optimizer onto the separable subspace at that $Pe$ and measure $(N_1-N_2)/N_1$: the paper reports this transport error is at most 1%, so an independent computation that finds a non-separable optimizer with $(N_1-N_2)/N_1$ substantially larger than 1% would show that the separability assumption is the active restriction.","supporting_citations":[{"cited_title":", Chini, G","cited_arxiv_id":null,"evidence_quote":"Defines the wall-to-wall optimal transport problem that this paper maximizes."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the single-wavenumber variational functional and the lemma used to derive the Pe^{6/11} bound."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the background-method comparison and the form of Howard's lemma used in the conditional bound."},{"cited_title":"2017 Optimal wall-to-wall transport by incompressible flows","cited_arxiv_id":null,"evidence_quote":"Gives the exact variational characterization of Nu-1 and the Pe^{2/3} lower bound that situates the computed scaling."},{"cited_title":"& Tobasco, Ian 2019 On the optimal design of wall-to-wall heat transport","cited_arxiv_id":null,"evidence_quote":"Extends the variational characterization and lower-bound construction relevant to the asymptotic discussion."},{"cited_title":"2019 Exhausting the background approach for bounding the heat transport in rayleigh-b\\'enard convection","cited_arxiv_id":null,"evidence_quote":"Independently confirms the two-dimensional no-slip scaling reported here."}],"review_version":1}