{"id":"9d59a36e-1eb4-42ea-9678-08c9753185ee","arxiv_id":"2607.16132","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Randomly advected Navier-Stokes with fast time and longer spatial correlations converges to an enhanced-diffusion Navier-Stokes equation, and in 2D its fluctuations converge to a Gaussian solution of a linearized Navier-Stokes equation with multiplicative white noise.","lead":"This paper proves that a fluid stirred by a rapidly varying random velocity field behaves at large scales like a Navier-Stokes fluid with a slightly larger viscosity, and that the random fluctuations around that behavior are Gaussian in two dimensions. It gives a rigorous derivation of the enhanced, eddy-type viscosity and of the noise intensity from the correlation structure of the stirring field.","discovery_kind":"first_principles","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorems 1.1/1.3 are proved under δ^{-1}∈N (§2.2, 'without loss of generality'); the stated generalization to all δ∈(0,1] is unsupported because m(δ^{-1}x) is not 2π-periodic for non-integer δ^{-1} and no approximation argument is given.","rationale":"I read the full text and followed the expansion, corrector hierarchy, semigroup construction, endpoint estimate, and final absorption. The argument is coherent: the parameter-free Green-Kubo formula is derived explicitly, the threshold N defined by ι>d/(2N) is exactly what makes the last-generation error terms vanish, and the quantitative enhanced-diffusion estimates close at R=δ^{-1}. The reader's weakest assumption identifies the real soft spot: the fixed-torus construction in §2.2 requires δ^{-1}∈N, and the 'without loss of generality' is unsupported. This is load-bearing because Theorem 1.3 is a statement about arbitrary δ→0 and the model itself is not defined for non-integer δ^{-1} without a periodization convention. Lemma 10.3 is marked 'standard and omitted'; while not ideal, it is not the central issue. I therefore agree with the reader's CONDITIONAL verdict and recommend no change.","tokens_in":61271,"tokens_out":27511,"duration_ms":224062,"concrete_test":"Set δ_n = 1/(n+1/2) and define the periodic kernel Kδ_n on T^d by periodic extension of δ_n^{-d/2}K(δ_n^{-1}·). Compute its Fourier coefficients (2π)^{-d}∫_{T^d} Kδ_n(x)e^{-ik·x}dx for a fixed frequency, e.g. k=(1,0,...,0). If they do not equal δ_n^{d/2}(F_{R^d}K)(δ_n k) for all n, then (2.3) and the symbol computations of Lemma 4.1 fail for arbitrary δ and the WLOG is not removable. If they do, the algebraic core of the proof extends verbatim to arbitrary δ and the gap is presentational rather than substantive.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 2.2 constructs m^δ_t(x)=m_t(δ^{-1}x) and observes this is 2π-periodic only when δ^{-1}∈N; it then says 'Let us keep this assumption from now on' and 'assume ... without loss of generality'. Theorems 1.1 and 1.3 quantify over all ε,δ∈(0,1], and Theorem 1.3 is a convergence statement over arbitrary sequences δ→0. For δ^{-1}∉N, the expression m(δ^{-1}x) is not a function on the fixed torus: periodic extension of the covariance Kδ*Kδ is a different Gaussian process unless δ^{-1} is integral. The proof's Fourier diagonalization (2.3), the Riemann-sum identities in §4/§11, and the corrector estimates all take place on T^d with eigenvalues δ^{d/2}(F_{R^d}K)(δk); whether these extend to the periodized process for non-integer δ is asserted, not shown. Thus the central fluctuation theorem is presently established only for a restricted parameter set; either the model must be redefined for arbitrary δ (e.g. via periodized covariance) with estimates verified, or the theorem must be restricted and an approximation argument supplied. The omitted proof of Lemma 10.3 is a lesser issue since the Galerkin passage is standard.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the randomly advected incompressible Navier–Stokes system (1.1) on the torus, with advecting field m^{ε,δ}(t,x)=m(ε^{-2}t,δ^{-1}x), where m is a centered, divergence-free, stationary Ornstein–Uhlenbeck process with covariance (K*K)Id and K smooth, compactly supported, isotropic. In the subcritical regime ε=o(δ), Theorem 1.1 proves convergence, along subsequences in law, to a deterministic Navier–Stokes system with enhanced viscosity ν=∥K∥²_{L²}/16 in d=2 and ν=∥K∥²_{L²}/5 in d=3, via a two-step corrector expansion and stochastic compactness. In d=2, under the stronger assumption ε=o(δ^{1+ι}), Theorem 1.3 proves that after subtracting deterministic macroscopic corrections v^{ε,δ} and the deterministic limit u, the normalized fluctuations δ^{-d/2}(u^{ε,δ}-v^{ε,δ}-u) converge in probability in L²(0,T;H^{-β}) to a Gaussian field z solving a linearized Navier–Stokes equation driven by χ dW·∇u, with χ=(F_{R²}K)(0). The proof builds a Wiener-chaos corrector hierarchy, absorbs effective operators into a Fourier-multiplier semigroup, and uses critical endpoint estimates for the convective terms.","tokens_in":61635,"tokens_out":5871,"duration_ms":56159,"significance":"If the theorems hold in their stated generality, this is a substantial contribution to the mathematical theory of turbulent transport and stochastic homogenization for fluid equations. The paper derives, rather than postulates, the enhanced-diffusion coefficient via a Green–Kubo-type formula, and it identifies the Gaussian fluctuation law with an explicitly computed noise intensity. The Fourier-symbol computations in Sections 4 and 11 are explicit and transparent, the Wiener-chaos inversion of the Ornstein–Uhlenbeck generator is conceptually clear, and the quantitative estimates in Section 11 are concrete and falsifiable. These are genuine strengths. However, the periodization gap in Section 2.2 currently prevents the main theorems from covering the full parameter range stated in the abstract and in Theorems 1.1 and 1.3; this is a load-bearing issue, not merely a presentation defect.","major_comments":[{"comment":"The manuscript assumes δ^{-1}∈N 'without loss of generality' and keeps this assumption throughout, but Theorems 1.1 and 1.3 quantify over all ε,δ∈(0,1]. For δ^{-1}∉N, the expression m(δ^{-1}x) is not a well-defined function on the fixed torus T^d: translation by 2πk changes the argument by δ^{-1}2πk, which is not a period of m unless δ^{-1}∈N. The subsequent Fourier diagonalization (2.3), the symbol computations of Lemma 4.1, the Riemann-sum limits of Lemma 4.3, and the quantitative estimates of Section 11 all rely on the periodized kernel K_δ on T^d. For non-integral δ^{-1}, the periodized covariance differs from q(δ^{-1}(x-y)) by aliasing terms, and no argument is supplied that the asserted limits remain unchanged. Thus the central theorems are presently established only for sequences with reciprocal-integer spatial correlation length. The manuscript should either restrict the statemen","section":"§2.2, Eq. (2.3)"},{"comment":"The existence proof for the macroscopic correction v^{ε,δ} is explicitly omitted: the proof says 'We proceed via a Galerkin approximation and derive uniform energy estimates; the passage to the limit is standard and omitted.' Since v^{ε,δ} appears in the very statement of Theorem 1.3 and its construction is needed for the definition of the fluctuation variable, this omission is load-bearing. I do not doubt that a Galerkin passage can be supplied, but it should either be written out or a precise reference should be given. The uniqueness part is also only sketched after 'preliminary mollification'; this is acceptable if the Galerkin passage is supplied.","section":"§10.3, Lemma 10.3"}],"minor_comments":[{"comment":"The sentence 'assuming δ^{-1}∈N ... keep this assumption from now on' should be flagged as an assumption, not a WLOG reduction, until an approximation argument is provided. Also, the later phrase 'assume throughout and without loss of generality that ε≤1/2' is harmless but should be stated after the periodization issue is resolved.","section":"§2.2"},{"comment":"The notation '∼=' for 'equality up to combinatorial constants' is used repeatedly. It would improve readability to state once that all such constants depend only on k and ℓ and are ultimately absorbed into the implicit constants of the estimates.","section":"§6.1"},{"comment":"The norm notation such as ||·||_{L²L²H^{-β}} is unusual; presumably it means the L² norm in time of an L²H^{-β}-valued map. Please define this notation explicitly at first use.","section":"§12.1"},{"comment":"The terms 'blue term' and 'magenta terms' are used in the text but the manuscript does not contain colors. It would be clearer to replace these by labels such as 'singular drift term' and 'remaining generator terms', or to include the color convention in a footnote.","section":"§6.2"},{"comment":"In Lemma 11.5, the bound is stated for β>0 but the proof uses H^{-2-β} and the energy estimate; the case β>1 follows from the case β=1 by embedding, but it would be useful to state this explicitly.","section":"§11.3"}],"recommendation":"major_revision","confidential_remarks":"The paper is innovative and the technical machinery is impressive, but the gap between the stated theorems and the actual model in Section 2.2 is significant. The issue is fixable within the manuscript's scope — either by restricting the parameter set or by redefining the model for all δ and re-proving the relevant estimates — so I recommend major revision rather than rejection. The omitted Galerkin passage in Lemma 10.3 should also be completed. Once these points are addressed, the paper would be a strong candidate for publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nYou should know this paper is the first to get a law of large numbers and a central limit theorem for the full Navier–Stokes system under random advection, in a subcritical two-parameter scaling. The effective viscosity is explicit via a Green–Kubo formula, and the fluctuation limit is a linearized Navier–Stokes SPDE driven by multiplicative space-time white noise, with the noise intensity computed from the zero mode of the covariance kernel. The structure is genuinely new relative to passive scalar and regularity-structure results, and the proof is a serious multiscale/corrector argument based on Wiener chaos inversion of the Ornstein–Uhlenbeck generator. No parameters are fitted; the constants come from the kernel K. The paper is also honest about scope: the critical scaling ε=δ is open, and the CLT is restricted to d=2 with clearly stated reasons.\n\nWhere I would push back is the claimed generality in δ. The setup in Section 2.2 constructs m^δ_t(x)=m(δ^{-1}x), and observes this is periodic on the fixed torus only when δ^{-1}∈N. It says 'we keep this assumption' and later 'without loss of generality,' but Theorems 1.1 and 1.3 quantify over all δ∈(0,1]. The Fourier diagonalization, the Riemann-sum identities, and the corrector estimates all use eigenvalues δ^{d/2}(F_{R^d}K)(δk) on the fixed torus. For non-integer δ^{-1} the periodized covariance is a different process, and no approximation argument is supplied. This is not a contradiction in the proof, but it means the theorems as stated are established only for a restricted parameter set. I think it is fixable—either by redefining the model with a periodized covariance for all δ and verifying the estimates, or by restricting the theorems and adding an approximation. A referee should ask for it.\n\nA second, smaller issue: Lemma 10.3, giving existence and estimates for the macroscopic correction v^{ε,δ}, is asserted with the Galerkin passage omitted. That is standard but still an omission in a central element of the CLT argument.\n\nOverall: this is a substantial paper that deserves careful peer review. The core strategy is coherent and, outside the periodicity gap, the math looks sound. I would send it to a referee, preferably someone who can check the corrector hierarchy carefully.\n\nBest,\n[Your name]","headline":"Solid, ambitious homogenization paper for randomly advected Navier-Stokes; the theorems as stated outrun the proof by a periodicity assumption on δ labeled WLOG but not justified.","tokens_in":62100,"tokens_out":3163,"would_cite":true,"duration_ms":28452,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["35Q30","60H15","35B27","60F05","76M50","76D05"],"pacs":[],"model":"deepseek-v4-flash","headline":"Fast small-scale random stirring makes fluid fluctuations Gaussian in two dimensions.","keywords":["random advection","Navier-Stokes equations","enhanced diffusion","law of large numbers","Gaussian fluctuations","Green-Kubo formula","homogenization","Ornstein-Uhlenbeck process"],"falsifier":"Run a two-dimensional direct numerical simulation of (1.1) with a fixed compactly supported isotropic kernel K, choose ε = o(δ^{1+ι}) with ε,δ small, solve for the deterministic corrections v^{ε,δ}, and measure the rescaled fluctuation δ^{-1}(u^{ε,δ} − v^{ε,δ} − u) in H^{−β}; if its space-time statistics deviate from the Gaussian solution of dz + [z·∇u + u·∇z + ∇p_z]dt = (1+ν)Δz dt + χ dW·∇u with ν = ‖K‖²_L²/16 and χ = (F_{R²}K)(0), the central claim is wrong. Equivalently, exhibiting a sequence δ → 0 with δ^{-1} not an integer along which the stated convergence fails would refute the 'without","tokens_in":61182,"feed_emoji":"🌀","tokens_out":4577,"duration_ms":44253,"temperature":0.7,"pith_summary":"The paper studies incompressible Navier–Stokes equations advected by a stationary, divergence-free random velocity field whose temporal and spatial correlation scales are ε² and δ, in the subcritical regime ε = o(δ). It proves two theorems. First, a law of large numbers in dimensions 2 and 3: as ε,δ → 0, the random solutions converge to a deterministic Navier–Stokes system in which the only trace of the stirring is an enhanced viscosity, explicitly ν = 1/16‖K‖²_L² in d = 2 and ν = 1/5‖K‖²_L² in d = 3, a Green–Kubo-type formula. Second, in two dimensions under the slightly stronger separation ε = o(δ^{1+ι}), the leading fluctuations around that deterministic limit, after subtracting deterministic macroscopic corrections, are Gaussian: they solve a stochastic linearized Navier–Stokes equation driven by multiplicative space-time white noise with intensity χ = (F_{R²}K)(0). If correct, the paper gives a complete two-level macroscopic description—deterministic averaged dynamics plus Gaussian fluctuations—derived from one fixed microscopic random field, with the transport noise inherited from the scaling rather than inserted by hand.","feed_headline":"Random advection yields a Gaussian limit for fluid fluctuations","feed_subtitle":"Two theorems pin down the effective diffusion and the noise that survives from a fast, small-scale stirring field.","key_machinery":"The argument is carried by a finite hierarchy of correctors indexed by binary sequences σ: digit 0 encodes an application of the transport term ε^{-1}P(m^{ε,δ}·∇·) and digit 1 an application of the Laplacian, each corrector carrying weight ε^{M+2L}φ_σ(u,m). Corrections are built by inverting the Ornstein–Uhlenbeck generator (−Mδ)^{-1}, which is explicit in Wiener chaos: each inversion contracts pairs of field factors against the covariance and lowers polynomial degree by two. Persistent expectations split as S_σ u + R_σ u; the leading operators S_σ are translation-invariant Fourier multipliers, so they can be absorbed into a semigroup generator of the form −λ Id + (1+ν)PΔ + Σ ε^{M+2L−1}S_σ,","core_discovery":"The central claim is Theorem 1.3: for d = 2 and ε = o(δ^{1+ι}) for some ι > 0, after subtracting deterministic macroscopic corrections v^{ε,δ} that solve a Navier–Stokes-type system with full quadratic self-interaction, the rescaled fluctuation δ^{-d/2}(u^{ε,δ} − v^{ε,δ} − u) converges in probability in L²(0,T;H^{−β}) for every β > 0 to a Gaussian field z solving dz + [z·∇u + u·∇z + ∇p_z]dt = (1+ν)Δz dt + χ dW·∇u, with enhanced viscosity ν = 1/16‖K‖²_L² and noise intensity χ = (F_{R²}K)(0), where W is a space-time white noise on the divergence-free mean-zero subspace. The companion law of large numbers (Theorem 1.1) identifies the deterministic limit in d = 2,3 as a Navier–Stokes system with","pith_inferences":["The proofs are carried out on a fixed torus under the standing assumption that δ^{-1} is an integer, so that m^{ε,δ} is exactly periodic; the paper declares this 'without loss of generality' but gives no argument covering arbitrary δ → 0, so the theorems as stated may strictly cover correlation lengths that are reciprocal integers.","The two-parameter family suggests a phase diagram in the (ε,δ) plane: subcritical temporal mixing produces a deterministic diffusive limit plus Gaussian fluctuations, while the critical line ε = δ may host qualitatively different, possibly non-Gaussian, fluctuations analogous to anomalous scaling in the Kraichnan passive-scalar model.","A numerical test is readily conceivable: simulate the randomly advected Navier–Stokes system for a prescribed compactly supported isotropic K and small ε,δ, measure the empirical enhanced viscosity and the fluctuation covariance, and compare against ν = ‖K‖²_L²/16 and χ = (F_{R²}K)(0).","The d = 3 analysis stops at the law of large numbers; the fluctuation result fails for three distinct reasons named in the paper, one of which is the divergence of a lattice sum, suggesting that a genuinely three-dimensional fluctuation theory would require a different averaging mechanism or a different observable."],"forward_implications":["The effective viscosity is explicitly computable from the stirring kernel K, and the dimension-dependent constants arise purely from the Leray projection acting on the isotropic covariance.","In dimensions 2 and 3, any family of weak solutions satisfying the energy inequality converges (along subsequences) to the same deterministic enhanced-diffusion Navier–Stokes system, so the macroscopic limit does not depend on how solutions are constructed.","In two dimensions, randomness does survive below critical scaling, but only at the fluctuation level: after the δ^{d/2} normalization, the limit is Gaussian and obeys a linearized Navier–Stokes equation around the deterministic background flow.","The multiplicative noise in the fluctuation limit has intensity χ = (F_{R²}K)(0), so only the zero-frequency spatial component of the stirring kernel contributes to the surviving randomness.","The critical case ε = δ is explicitly left open; the paper expects a different form of enhanced diffusion there, combining the noise correlation function with the Green kernel of the Stokes operator."],"fun_headline_variants":["Random advection gives Gaussian limit for fluid fluctuations","Subcritical stirring: deterministic limit plus Gaussian noise","Enhanced diffusion from random advection, with Gaussian limit","Fast small-scale stirring yields Gaussian fluctuation law","Randomly advected fluids: Gaussian limit after subtracting mean"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The proofs run on a fixed torus under the assumption that δ^{-1} is an integer so that the rescaled stirring field is exactly periodic; the paper calls this 'without loss of generality' but supplies no argument that covers arbitrary δ → 0, so the theorems as stated may only cover reciprocal-integer spatial correlation lengths.","fun_headline_variants_meta":{"raw":{"variants":["Random advection gives Gaussian limit for fluid fluctuations","Subcritical stirring: deterministic limit plus Gaussian noise","Enhanced diffusion from random advection, with Gaussian limit","Fast small-scale stirring yields Gaussian fluctuation law","Randomly advected fluids: Gaussian limit after subtracting mean"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000158,"raw_usage":{"total_tokens":1103,"prompt_tokens":825,"completion_tokens":278,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":569,"completion_tokens_details":{"reasoning_tokens":203}},"tokens_in":569,"tokens_out":278,"duration_ms":3550,"temperature":1.0,"reasoning_tokens":203,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T21:15:19.704347+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run a two-dimensional direct numerical simulation of (1.1) with a fixed compactly supported isotropic kernel K, choose ε = o(δ^{1+ι}) with ε,δ small, solve for the deterministic corrections v^{ε,δ}, and measure the rescaled fluctuation δ^{-1}(u^{ε,δ} − v^{ε,δ} − u) in H^{−β}; if its space-time statistics deviate from the Gaussian solution of dz + [z·∇u + u·∇z + ∇p_z]dt = (1+ν)Δz dt + χ dW·∇u with ν = ‖K‖²_L²/16 and χ = (F_{R²}K)(0), the central claim is wrong. Equivalently, exhibiting a sequence δ → 0 with δ^{-1} not an integer along which the stated convergence fails would refute the 'without","supporting_citations":[],"review_version":1}