{"id":"85213f63-bc78-4f07-8032-f300afe794d4","arxiv_id":"2509.02081","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For advection-diffusion on T^d there exist incompressible velocity fields and initial data whose L2 mass decays double exponentially in 2D, like e^{-Ct^2} in 3D, and superexponentially in 4D.","lead":"This paper constructs incompressible flows on a flat torus that make particular diffusing tracers decay faster than any exponential rate. It gives the first examples of superexponential dissipation on a compact domain and shows an earlier conjecture against such decay is false.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The constructed velocity field is not shown to be real: the ODE control theorem imposes v_{-k}=v_k, but real-valued v on T requires v_{-k}=\\overline{v_k}; as written, the PDE is solved with a complex drift.","rationale":"The reader identified Prop 2.5's omitted verification as the load-bearing gap. That step is indeed only asserted, but it is likely routine: the needed difference d_{k+1}-d_{1-k} is proportional to b·c/L, with b·c containing a positive term of size ~2p^2 and L ~ 2p, so Assumption 3.1(4) should hold for large p. I therefore do not see Prop 2.5 as the main risk to the central claim. The more serious issue is that the control theorem produces complex Fourier coefficients with palindromic symmetry, which does not correspond to a real-valued v. Since the main theorems assert real, incompressible velocity fields and the reality-preservation argument depends on u being real, the proof as written does not establish the central claim. However, the flaw appears fixable: imposing the correct Hermitian symmetry v_{-k}=\\overline{v_k} still leaves a solvable linear control problem, so the verdict remains CONDITIONAL rather than moving to REJECT. Because the reader already assigned CONDITIONAL, the verdict is unchanged, though the stated reason differs.","tokens_in":29378,"tokens_out":21511,"duration_ms":247150,"concrete_test":"Take a single k in Prop 3.5 with d_{k+1}\\neq d_{1-k} and generic nonzero z_{k+1}^0, z_{1-k}^0 (e.g. the d's from Prop 2.1 with r=10); compute a_k,b_k by the displayed B_k formula and evaluate \\hat v(-k)-\\overline{\\hat v(k)}. The paper's formula will give a nonzero value, confirming that v is not real. Then solve the same linear system with the second equation replaced by its conjugate (Hermitian symmetry); if Assumption 3.1(4) makes the determinant nonzero, the construction is repairable and the central claim can survive.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The proof of Theorem 3.2 (and Prop 3.5) constructs Fourier coefficients v_k^t satisfying v_{-k}^t = v_k^t with complex values: the controls a_k,b_k in Prop 3.5 are complex because B_k and the target errors are complex. In Corollaries 4.3–4.4 the paper then says 'v is R-valued as v_{-k}^t=v_k^t'. This is incorrect: a real-valued function on T has \\hat v(-k)=\\overline{\\hat v(k)}, not equality. Consequently the real-space vector field w_{a,b,v,L} constructed from this v is complex-valued, not an admissible real velocity. The reduction in §2.1—'if some complex solution decays super-exponentially, the real and imaginary parts are also solutions'—uses reality of u and therefore does not apply. Hence Theorems 1.1–1.3 are not established for real velocity fields as stated. The gap is repairable: re-solving Prop 3.5 with the Hermitian constraint v_{-k}=\\overline{v_k} gives a 2x2 system with determinant nonzero by Assumption 3.1(4), so the control step should still work; but the proof as written is invalid.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper constructs divergence-free velocity fields on the torus that force a particular smooth Fourier-mode solution of the advection-diffusion equation to decay superexponentially. In 2D a bounded velocity field gives double-exponential decay e^{-C^{-1}e^{C^{-1}t}}; in 3D a Lipschitz velocity field gives decay e^{-C^{-1}t^2}; in 4D a C^\\infty velocity field gives some superexponential rate. The mechanism is to move the Fourier mass from a mode a to a larger mode a+b in finite time using a control velocity field built from a one-dimensional ODE system, then iterate while using the energy identity to convert the growing wavenumber into accelerated decay.","tokens_in":29686,"tokens_out":28966,"duration_ms":304250,"significance":"If the proof were correct, the paper would settle a natural open question: on the compact torus, with uniformly bounded velocity, dissipation can be much faster than exponential, matching the double-exponential lower bound of Miles and Doering and disproving the conjecture that no faster-than-exponential decay is possible. The architecture is self-contained: the reduction to an ODE control problem, the dyadic Newton-type error removal, and the use of the energy identity are clearly laid out, and there are no fitted parameters. The explicit construction of different regularity regimes in dimensions 2, 3, and 4 is an interesting contribution in itself. However, I find several load-bearing gaps in the control-theoretic core, detailed below.","major_comments":[{"comment":"The claim that the constructed v is real-valued because v_{-k}=v_k is false. A real-valued function on the torus satisfies \\hat v(-k)=\\overline{\\hat v(k)}, not \\hat v(-k)=\\hat v(k). The controls a_k,b_k in Proposition 3.5 are complex (they solve a complex 2x2 system), so Theorem 3.2 and Proposition 3.3 produce complex coefficients. Hence w_{a,b,v,L} is complex-valued and the reduction in §2.1 to real/imaginary parts does not apply because the PDE is solved with a complex drift. This affects Theorems 1.1–1.3. The gap is repairable by imposing Hermitian symmetry v_{-k}=\\overline{v_k}; Assumption 3.1(4) should make the resulting control problem well-posed, but the proof as written is invalid.","section":"§4.1, Corollaries 4.3–4.4; §2.1"},{"comment":"Substituting the displayed definition a_t = -i2^8 z0|z1|/(z1|z0|) into the equation for z0 gives i a_t z_1 = +2^8 z0|z1|/|z0|, not the negative term appearing in the subsequent system. With d0=0 this term drives |z0| upward, so the claimed conclusion z0(1)=0 and the lower bound |z1(1)|≥1/96 do not follow. A mere sign change may not be enough: the single complex control must simultaneously produce the desired real-part effects on both z0 and z1, which imposes phase constraints that are not addressed. Proposition 3.4 is the first step of Theorem 3.2 and therefore is load-bearing for all later results.","section":"§3, Proposition 3.4"},{"comment":"The line 'using that v_k=v_{-k} to show the final term is 0' is incorrect. For complex coefficients and the standard ℓ2 inner product, \\sum_{k,j} v_{k-j} \\psi_j \\overline{\\psi_k} is not generally zero under v_{-k}=v_k; the term Re⟨i v*\\psi,\\psi⟩ vanishes only when v corresponds to a real-valued function, i.e. v_{-k}=\\overline{v_k}, after taking the real part. Since the controls are complex, the displayed estimate leading to (3.11) is not established. This invalidates the error-reduction bound used in Corollary 3.6 and Theorem 3.2.","section":"§3, Proposition 3.5, energy estimate"},{"comment":"The proof asserts that the verification of Item 3 of Corollary 4.4 (the condition d_{k+1}-d_{1-k}≥1 and the support condition |a+kb|≥|a+b|) is obtained by 'combining the arguments' from Propositions 2.3 and 2.4, but no explicit computation or lemma is supplied. Since Proposition 2.5 is needed for Theorem 1.3, this omitted verification should be written out, ideally as a lemma analogous to Lemma 4.5 for the downhill configuration.","section":"§4.2, Proposition 2.5"}],"minor_comments":[{"comment":"The use of Legendre's three-square theorem should explicitly state that one selects an integer n in [|(m,n,ℓ)|^2+1, |(m,n,ℓ)|^2+8] with n≡1 mod 4, then represents n as a sum of three squares.","section":"§4.2, proof of Proposition 2.3"},{"comment":"The phrase 'giving the e^{-C^{-1}t^2} decay of Theorem 1.3' should refer to Theorem 1.2.","section":"§2.1.6"},{"comment":"The conclusion from Proposition 2.7 is stated as immediate, but the proof should record that the time increments T_n in Proposition 2.5 grow at most polynomially in |a_n| (indeed T_n=O(log(1/η_n))=O(|a_n|) in the construction) so that f(t)→∞ as t→∞. Without such a bound, the claimed superexponential rate is not justified.","section":"§2.3, proof of Theorem 1.3"},{"comment":"Definition 4.1 requires strict |a·b|<|a||b|, while Corollary 4.4 states the hypothesis with ≤. Please make the strict inequality explicit in Corollary 4.4 and confirm that the applications satisfy it.","section":"§4.1, Definition 4.1 and Corollary 4.4"}],"recommendation":"major_revision","confidential_remarks":"The central idea is promising and the overall architecture is coherent, but the current version has more than presentation-level issues: the reality of the constructed velocity field is not established, the control definition in Proposition 3.4 has a sign inconsistency, and the energy estimate in Proposition 3.5 relies on a false symmetry argument. These are repairable within the scope of the paper, so I recommend major revision rather than rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"First thing to know: this is the first construction of superexponential decay for advection–diffusion on the compact torus, including double exponential in 2D with a bounded incompressible flow. The core mechanism—moving a single Fourier mode from a to a+b in finite time, then using diffusion to clean up error by a dyadic Newton iteration—is genuinely new and mostly well executed. The 3D use of Legendre's three-square theorem to control gaps in the lattice norm is clever.\n\nThe real question is whether the constructed velocity is actually real. The ODE control theorem works with complex coefficients v_k satisfying v_{-k}=v_k. For a real function on T^d the Fourier coefficients must satisfy v_{-k}=overline{v_k}. The paper says at Corollaries 4.3–4.4 that v is R-valued because v_{-k}=v_k; that is the wrong condition. The controls in Proposition 3.5 are explicitly complex (a_k,b_k solve a complex linear system), so the real-space vector field w is complex-valued. That means the reduction in Section 2.1—take real and imaginary parts of a decaying complex solution—does not apply, because the equation with a complex drift is not reality-preserving. This gap touches Theorems 1.1–1.3 as stated. It is probably repairable: imposing the Hermitian constraint v_{-k}=overline{v_k} in the control problem still leaves a 2x2 system with nonzero determinant by Assumption 3.1(4), so the Newton step should survive. But that is a nontrivial rewrite, not a typo.\n\nThere is also a smaller soft spot the reader flagged: Proposition 2.5 in 4D verifies the key diffusion-coefficient hypothesis only by saying 'combine the arguments' from Propositions 2.3 and 2.4. That is a real omission; if it fails, Theorem 1.3 collapses. But the 2D and 3D theorems do not depend on it.\n\nWhat is good: the ODE control estimates in Propositions 3.4–3.6 are explicit and internally consistent; the S and M assumptions are checked rather than assumed; no fitted parameters. The paper is self-contained and the citation pattern looks appropriate; self-citations are background, not load-bearing.\n\nWho is this for: people working on enhanced dissipation, mixing, and optimal decay for passive scalars. It deserves a serious referee. My recommendation: send it to review, but ask for a revision that fixes the real-valuedness issue and writes out the 4D verification. If repaired, I would cite it.","headline":"A genuinely new construction of superexponential dissipation on the torus, but as written the velocity field is complex-valued and the 4D step is under-verified; worth refereeing after a fix.","tokens_in":30140,"tokens_out":6160,"would_cite":false,"duration_ms":64767,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["35B40","35Q35","76F25"],"pacs":[],"model":"deepseek-v4-flash","headline":"Bounded incompressible flows on the torus can drive chosen smooth solutions of the advection-diffusion equation to decay at a double-exponential rate — matching the known lower bound and refuting the conjecture that superexponential decay i","keywords":["advection-diffusion equation","enhanced dissipation","superexponential decay","double exponential rate","passive scalar","incompressible flow","Fourier mode control","Batchelor scale"],"falsifier":"For the 4D 'downhill' move of Proposition 2.5, compute M = min_{k≠0,1} dk, S = Σk (1+dk)^{-1}, and Δk = dk+1 − d1−k from dk = |a+kb|²/L − A with a = (m,n,ℓ,−p−1), b = (x,y,z,p) − a; failure of M ≥ 2²⁶, S ≤ 6, or Δk ≥ 1 for any k would break the C∞ construction. A more direct check: numerically integrate the ODE system (3.1) with the controls of Theorem 3.2 and see whether the transfer from δk,0 to βδk,1 actually occurs.","tokens_in":29282,"feed_emoji":"🌊","tokens_out":13259,"duration_ms":133205,"temperature":0.7,"pith_summary":"The paper studies how fast a passively advected, diffusing scalar (a solution θ of ∂tθ = Δθ + u·∇θ on the torus, with u divergence-free) can be made to decay in L2 by choosing the stirring flow u. It proves that on the torus, faster-than-exponential decay is attainable: in two dimensions a merely bounded velocity field drives a carefully chosen smooth initial datum to vanish at the double-exponential rate e^{−C⁻¹e^{C⁻¹t}}, in three dimensions a Lipschitz flow gives e^{−C⁻¹t²}, and in four dimensions a C∞ flow gives some superexponential rate. The 2D rate is essentially optimal, matching a known double-exponential lower bound, and it disproves the standing conjecture that no superexponential decay is possible for time-dependent incompressible flows on compact domains. The construction works by shuttling all Fourier mass from one pure mode to a larger pure mode, then outward to infinity, so the scalar's gradient-to-mass ratio grows without bound and the instantaneous exponential decay rate accelerates.","feed_headline":"Bounded flows make scalars decay at double-exponential speed","feed_subtitle":"On the 2D torus, an incompressible flow drives one solution to vanish at rate e^{−C⁻¹e^{C⁻¹t}}, matching the lower bound.","key_machinery":"The carrying object is the infinite Fourier-ODE system żk = −dk zk + i Σj vj zk−j (equation (3.1)), with damping coefficients dk = |a+kb|²/L − A; it arises because flows of the form ut(x) = vt(b·x) keep a solution starting at mode fa supported on the line fa+kb. The paper's main technical theorem, Theorem 3.2, is a control result for this system: under Assumption 3.1 — a spectral gap M ≥ 2²⁶ between the two active modes and all others, total weight S ≤ 6, and adjacent damping gaps dk+1 − d1−k ≥ 1 — there is a coefficient field v with v−k = vk that steers δk,0 to βδk,1. The steering splits into two moves: an explicit amplifier (Proposition 3.4) that empties z0 into z1 in unit time, and a Newt","core_discovery":"On the paper's own terms, the central claim is Theorem 1.1: there exists C > 0, a divergence-free velocity field u ∈ L∞([0,∞)×T²), and nonzero smooth initial data such that the solution of the advection-diffusion equation satisfies ‖θt‖L² ≤ C e^{−C⁻¹e^{C⁻¹t}}‖θ0‖L² for all t. Theorem 1.2 upgrades the flow to Lipschitz regularity in 3D with decay e^{−C⁻¹t²}, and Theorem 1.3 gives a C∞ flow in 4D with a non-explicit superexponential rate. The proof reduces the PDE to an infinite linear ODE system in Fourier space, restricted to modes a + kb for a fixed transition vector b, and then solves a control problem: design the coefficients vj so that the solution passes from δk,0 to βδk,1 (Theorem 3.2)","pith_inferences":["Stress test: the construction feeds on data concentrated on a single Fourier mode, and the cleanup step assumes mass is already almost entirely on one mode; testing whether the double-exponential rate survives small broadband perturbations of the initial datum would show how far the mechanism reaches beyond the paper's special data.","Threshold question: since the 2D rate matches the lower bound up to constants, the natural next classification is which time-exponent functions are attainable for each regularity class of flows; the paper's three-rung ladder is plausibly part of a complete hierarchy.","Transferability: the mode-to-mode control is a general recipe — a spectral-gap inequality plus unequal damping on paired modes — so analogues for fractional dissipation, advection–reaction systems, or anisotropic diffusion are plausible testbeds.","Practical check: the constants (e.g., M ≥ 2²⁶) are far beyond numerical simulation of the PDE, but the reduced ODE system (3.1) can be integrated directly at moderate parameters, offering a cheap test of the claimed δk,0 → βδk,1 transfer."],"forward_implications":["If correct, Theorem 1.1 closes the gap between the best known decay (exponential) and the best known obstruction (double exponential) on the 2D torus: superexponential decay is attainable with only L∞-bounded velocity, and the rate is essentially optimal.","The rate–regularity ladder (L∞ → double exponential, Lipschitz → e^{−C⁻¹t²}, C∞ → qualitative superexponential) shows flow smoothness is a genuine cost: smoother mixers get worse guaranteed decay.","For the exhibited solutions, the dissipation rate ‖∇θt‖L²/‖θt‖L² grows without bound in time, so the scalar repeatedly outruns the Batchelor-scale saturation believed to cap dissipation rates — though only for specially prepared, mode-concentrated data.","Since the PDE preserves real signals, taking real and imaginary parts of the complex solutions yields real-valued smooth data with the same superexponential decay rates."],"supporting_citations":[{"why":"Supplies the double-exponential lower bound that the 2D construction matches up to constants, and the conjecture that superexponential decay is impossible, which the paper refutes.","marker":"[MD18]"},{"why":"Provides the unique-continuation argument that [MD18] adapts to obtain the double-exponential lower bound for time-inhomogeneous bounded flows.","marker":"[Poo96]"},{"why":"Documents the discrete-time pulsed-diffusion setting (toral automorphisms such as Arnold's cat map) with double-exponential decay, the model that the paper's mode-to-mode transfer strategy makes continuous-time.","marker":"[FI19]"}],"fun_headline_variants":["2D torus: bounded flow gives double-exponential decay","Bounded incompressible flows: superexponential scalar dissipation","In 2D, a bounded flow drives a solution to double-exponential decay","Superexponential decay via bounded flows on tori"],"cache_read_input_tokens":2688,"weakest_assumption_plain":"The construction rests on the quantitative hypotheses of Assumption 3.1 — a spectral gap M ≥ 2²⁶ separating the two active Fourier modes from all others, a bounded total weight S ≤ 6 of inactive modes, and adjacent-mode damping gaps of at least 1 — which are verified explicitly for the 2D and 3D steps but merely asserted for the 4D 'downhill' step that the qualitative C∞ result depends on.","fun_headline_variants_meta":{"raw":{"variants":["2D torus: bounded flow gives double-exponential decay","Bounded incompressible flows: superexponential scalar dissipation","In 2D, a bounded flow drives a solution to double-exponential decay","Superexponential decay via bounded flows on tori"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000658,"raw_usage":{"total_tokens":2840,"prompt_tokens":732,"completion_tokens":2108,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":476,"completion_tokens_details":{"reasoning_tokens":2035}},"tokens_in":476,"tokens_out":2108,"duration_ms":22648,"temperature":1.0,"reasoning_tokens":2035,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-05T12:04:13.674623+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"For the 4D 'downhill' move of Proposition 2.5, compute M = min_{k≠0,1} dk, S = Σk (1+dk)^{-1}, and Δk = dk+1 − d1−k from dk = |a+kb|²/L − A with a = (m,n,ℓ,−p−1), b = (x,y,z,p) − a; failure of M ≥ 2²⁶, S ≤ 6, or Δk ≥ 1 for any k would break the C∞ construction. A more direct check: numerically integrate the ODE system (3.1) with the controls of Theorem 3.2 and see whether the transfer from δk,0 to βδk,1 actually occurs.","supporting_citations":[],"review_version":1}