{"id":"e6e5b4bd-3204-444e-af3d-b493b9eb8a48","arxiv_id":"2411.13923","paper_version":3,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For every sub-critical 1D Gaussian multiplicative chaos measure, the Fourier dimension equals the correlation dimension: 1-γ² for small γ, (√2-γ)² for large γ.","lead":"Random fractal measures called Gaussian multiplicative chaos have Fourier coefficients that decay at a specific rate; this paper proves the exact rate for all sub-critical parameters, confirming a conjecture by Garban and Vargas. The result pins down the Fourier dimension of these measures and yields sharp Fourier restriction estimates.","discovery_kind":"first_principles","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Upper bound of Theorem 1.1 is outsourced to dim2(μγ,GMC)=Dγ; the paper never checks that Bertacco's L2-spectrum theorem covers the exact-log-kernel Bacry-Muzy GMC on [0,1], so the equality rests on an unverified external premise.","rationale":"Good-faith review of the lower-bound proof: I traced the martingale-type argument, the odd-even decomposition in Proposition 5.1, and the separation-of-variable estimate in Proposition 5.2. The exponent bookkeeping in Steps 7-12 is consistent: ‖vL‖^p E[R_L^p] and ‖wL‖^p E[Q_L^p] both decay like 2^{-kθ} times a summable 2^{-pL((1-τ)/2-1/q)} factor, and the condition q>4/(1-τ) makes the L-series converge. Lemma 4.2's variational computation for p,q with Θ>0 is correct. No internal contradiction found. The paper's own contribution—the lower bound—is technically sound as far as I can see. The equality in Theorem 1.1, however, requires the upper bound, and that upper bound is not self-contained: it is imported from Bertacco/LRV15. The paper cites these results but does not show that their domain of applicability includes the exact-log-kernel GMC on the interval produced by Bacry-Muzy white noise. This is exactly the weakest assumption flagged by the reader. I agree with that identification. Since the external results are published and standard, and the paper provides two routes (§6.2.1 and §6.2.2), I would not change the ACCEPT verdict, but the verification step in concrete_test is worth doing to close the gap.","tokens_in":26991,"tokens_out":36113,"duration_ms":325380,"concrete_test":"Check whether [Ber23, Thm 3.1] applies verbatim to the Bacry-Muzy GMC on [0,1] defined by (3.2)-(3.9), i.e., that the exact log kernel (1.2) is covered by the theorem's hypotheses (not just the perturbed kernel (1.4) on positive-type domains). If the hypotheses fail, re-derive the upper bound directly by computing the correlation dimension of the approximate measures μγ,m and taking the m→∞ limit; the identity dim2(μγ,GMC)=Dγ must be reproduced for all γ∈(0,√2).","verdict_should_be":"UNCHANGED","load_bearing_attack":"Lemma 6.2 (upper bound) is the only part of Theorem 1.1 not proved from the paper's own construction. It relies on the identity dim2(μγ,GMC)=Dγ, imported from Bertacco [Ber23, Thm 3.1] in §6.2.1, or equivalently on the LRV15 threshold ∫∫ μ μ/|t-s|^β<∞ a.s. iff β<Dγ used in §6.2.2. The paper does not verify that these external results apply to the GMC defined through the Bacry-Muzy white-noise decomposition (§3) with the exact covariance log(1/|t-s|), as opposed to perturbed kernels (1.4) on domains where the log kernel is positive type. If the correlation-dimension identity is inapplicable, only the lower bound dimF≥Dγ is established and equality in Theorem 1.1 fails. This is the most load-bearing premise: it is external, it is not re-derived, and the paper's substantial lower-bound argument cannot compensate for a missing upper bound.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper establishes the exact almost-sure Fourier dimension of the standard subcritical Gaussian multiplicative chaos measure on the unit interval, confirming the Garban-Vargas conjecture. Theorem 1.1 states that for every γ∈(0,√2), dim_F(μ_{γ,GMC})=D_γ, where D_γ=1−γ² for γ<√2/2 and D_γ=(√2−γ)² for γ≥√2/2. The lower bound is proved in detail by a vector-valued martingale method: weighted Fourier coefficients are arranged into ℓ^q-valued martingales, Pisier's martingale type p inequalities are applied twice, and a dyadic-discrete-time approximation together with an Abel summation step yields a separation-of-variable estimate that gives the required uniform L^p(ℓ^q) bound for every τ<D_γ. The upper bound is obtained from the classical inequality dim_F≤dim_2 combined with the known L²-spectrum/correlation-dimension identity dim_2(μ_{γ,GMC})=D_γ, cited from Bertacco and from Lacoin-Rhodes-Vargas. The paper also states corollaries on upper Frostman regularity and on a Fourier restriction estimate.","tokens_in":27157,"tokens_out":20752,"duration_ms":183231,"significance":"If correct, this is a substantial result: it resolves the Garban-Vargas conjecture and provides an exact formula for the Fourier dimension of 1D GMC. The lower-bound proof is the main novelty and is internally consistent; I checked the exponent bookkeeping in Steps 7–12 of §5.5 and the variational calculation in Lemma 4.2, and the threshold D_γ emerges correctly from the condition Θ(γ,τ,p,q)>0. The method, based on a random Fourier decoupling estimate and a separation-of-variable estimate, is likely to generalize to other multiplicative chaos models, as the authors indicate. The upper bound is not proved from first principles but is explicitly reduced to published L²-spectrum results; this is a legitimate strategy, though the verification of the hypotheses of those external results should be made more explicit. The paper is clearly written and the main proof is fully detailed.","major_comments":[],"minor_comments":[{"comment":"The upper bound in Lemma 6.2 is load-bearing and depends entirely on the external identity dim_2(μ_{γ,GMC})=D_γ from [Ber23, Theorem 3.1] or on the [LRV15] energy criterion. The manuscript states in §6.2.1 that these results cover perturbed log-kernels of the form (1.4), and the exact kernel (1.2) is the special case g≡0, but the verification that the Bacry-Muzy construction of §3 satisfies all hypotheses (including positive definiteness in d=1) is left implicit. Please add one explicit sentence in §6.2.1 and §6.2.2 stating precisely which cited theorem applies to the exact-kernel GMC on [0,1] and why.","section":"§6.2"},{"comment":"The cardinality of D_{k-1} is 2^{k-1}, not 2^k, so the display in §5.3 should be ∑_{k≥2} 2^{k-1}·2^{-(k-1)(1+Θ)}; the convergence is unaffected, but the displayed exponent is off by a harmless factor.","section":"§5.3"},{"comment":"The displayed equality E[(∑ n^{τq/2}|μ̂_{γ,GMC}(n)|^q)^{p/q}] = sup_m E[‖M_m‖^p_{ℓ^q}] should be an inequality ≤ sup_m, obtained by Fatou's lemma after the pointwise convergence of Fourier coefficients; the subsequent conclusion is unaffected.","section":"§6.1, Lemma 6.1"},{"comment":"The derivation of Corollary 1.2 is omitted as routine. Since Corollary 1.3 relies on the upper Frostman regularity statement, please either include a short derivation or give the precise argument in [CHQW24, Corollary 1.5] that is being invoked.","section":"§1.3, Corollary 1.2"},{"comment":"There are several typographical and formatting issues: 'refered' should be 'referred', 'halp-plane' should be 'half-plane', 'Garban-V arga' appears with a spurious space in the abstract, and some spacing in displays is irregular. A careful copyedit would improve readability.","section":"Throughout"}],"recommendation":"minor_revision","confidential_remarks":"The only substantive risk is the dependence of the upper bound on external L²-spectrum results. I am not convinced this is a gap, because the manuscript explicitly cites these results as covering the perturbed-kernel class that contains the exact log-kernel GMC; however, making the applicability check explicit would remove any ambiguity. The lower-bound proof is original and appears sound. I see no issues with novelty or citation practice."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Bottom line: this is a strong paper. It proves the exact Fourier dimension D_gamma for all subcritical 1D GMC, resolving the Garban-Vargas conjecture. The lower bound is the real work, and it is new: for gamma >= 1/sqrt(2) only the Rajchman property and weaker bounds were known. The l^q-valued martingale machinery is adapted from CHQW24 with substantial new ingredients — Bacry-Muzy white-noise decomposition, odd-even interval splitting, dyadic-discrete-time approximation, Abel summation — and the exponent bookkeeping checks out. The threshold D_gamma emerges from a variational condition rather than being fitted, and the proof is self-contained given Kahane's theory and Pisier's inequalities. That is a serious contribution.\n\nThe upper bound is another matter, but not a fatal one. Lemma 6.2 imports dim_2(mu) = D_gamma from Bertacco, or equivalently the LRV15 energy criterion. The stress-test note claims the paper never verifies those results cover the exact log-kernel on [0,1]. On reading, the paper explicitly says Bertacco treated the more general perturbed kernel (1.4); the exact kernel is the g=0 case, so it is covered. The authors could have added one sentence saying so — that is a minor exposition gap, not a mathematical flaw. The alternative proof via LRV15's iff criterion is also standard. I do not think the stress-test concern holds up.\n\nOther soft spots are small. Corollary 1.2 is stated with proof omitted as routine; that is acceptable for a paper this length, though a sketch would help. The paper is long and dense, and the most valuable part (Section 5) demands close reading. But I found no circularity and no sign of fitting: the lower bound is derived from first principles, and the upper bound merely assembles known results.\n\nThis paper is for researchers in random measures, harmonic analysis, and fractal geometry. It deserves a serious referee. I would recommend acceptance after a minor revision where the authors verify explicitly that the external L2-spectrum results apply to the exact-log-kernel Bacry-Muzy GMC on the interval — a small patch to a solid result.","headline":"This paper settles the Garban-Vargas conjecture for 1D GMC with a genuinely new lower-bound technique, and the stress-test worry about the upper bound is minor rather than fatal; send it to review.","tokens_in":27787,"tokens_out":2697,"would_cite":true,"duration_ms":24145,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60G57","42A61","46B09","60G46"],"pacs":[],"model":"deepseek-v4-flash","headline":"For every sub-critical $\\gamma\\in(0,\\sqrt2)$, the Fourier dimension of the standard Gaussian multiplicative chaos measure on the unit interval is almost surely $1-\\gamma^2$ for $\\gamma<\\sqrt2/2$ and $(\\sqrt2-\\gamma)^2$ for…","keywords":["Gaussian multiplicative chaos","Fourier dimension","correlation dimension","vector-valued martingale method","martingale type p","white-noise decomposition","dyadic decomposition","Fourier decay"],"falsifier":"Compute the correlation dimension of the exact-log-kernel GMC measure on $[0,1]$ directly from the Riesz-energy definition, for example numerically for $\\gamma=0.8$: the double integral $\\int\\int \\mu(dt)\\mu(ds)/|t-s|^s$ should diverge for every $s>D_{0.8}$ and converge for every $s<D_{0.8}$. A deviation from this threshold at any sub-critical $\\gamma$ would falsify the upper bound and hence the equality.","tokens_in":26721,"feed_emoji":"📐","tokens_out":12268,"duration_ms":103305,"temperature":0.7,"pith_summary":"This paper resolves a conjecture about the harmonic analysis of Gaussian multiplicative chaos: for every sub-critical parameter $\\gamma\\in(0,\\sqrt2)$, the random measure on the unit interval has an exact Fourier dimension $D_\\gamma$, equal to $1-\\gamma^2$ when $\\gamma<\\sqrt2/2$ and to $(\\sqrt2-\\gamma)^2$ when $\\gamma\\ge\\sqrt2/2$. That means the Fourier coefficients decay polynomially at every exponent below $D_\\gamma$ and fail to decay at any exponent above it, almost surely. The result matters because Fourier dimension is the sharp measure of how a measure spreads in frequency space, and for continuous multiplicative chaos only non-matching bounds were known before. The equality also confirms that, for this model, Fourier dimension and correlation dimension coincide, which is the qualitative content of the conjecture.","feed_headline":"Exact Fourier dimension found for 1D multiplicative chaos","feed_subtitle":"A proof pins the decay rate of Fourier coefficients to the correlation dimension for every sub-critical parameter.","key_machinery":"The key machinery is an $\\ell^q$-valued martingale indexed by the Fourier coefficients of the approximating measures, $M_m = (n^{\\tau/2}\\widehat{\\mu}_{\\gamma,m}(n))_{n\\ge 1}$, with $1<p<2$ and $q>4/(1-\\tau)$ chosen so that a certain positive-exponent condition holds. The proof establishes a uniform $L^p(\\ell^q)$ bound for this martingale by a two-stage localization: first, martingale type $p$ inequalities for $\\ell^q$ reduce the bound to sums over dyadic intervals; second, a separation-of-variable estimate bounds each dyadic contribution by deterministic weight sequences times explicit random variables. The separation step combines a white-noise decomposition of the log-correlated field, an odd-even decomposition of dyadic intervals to create conditional independence, a quantitative modulus of continuity for the weight processes, Abel summation to turn oscillatory integrals into differences of adjacent weights, and a discrete product rule. These ingredients convert the problem into checking geometric series that converge exactly when $\\tau<D_\\gamma$.","core_discovery":"The paper's central claim is the almost-sure equality $\\dim_F(\\mu_{\\gamma,\\mathrm{GMC}}) = D_\\gamma$, with $D_\\gamma = 1-\\gamma^2$ for $0<\\gamma<\\sqrt2/2$ and $D_\\gamma = (\\sqrt2-\\gamma)^2$ for $\\sqrt2/2\\le\\gamma<\\sqrt2$. The hard direction is the lower bound $\\dim_F\\ge D_\\gamma$: the paper forms the $\\ell^q$-valued martingale $(n^{\\tau/2}\\widehat{\\mu}_{\\gamma,m}(n))_{n\\ge1}$ from the Fourier coefficients of the approximating measures, proves a uniform $L^p(\\ell^q)$ bound by localizing onto dyadic intervals and applying martingale type $p$ inequalities for $\\ell^q$, and then passes to the limit to obtain $|\\widehat{\\mu}_{\\gamma,\\mathrm{GMC}}(n)|^2 = O(n^{-\\tau})$ for every $\\tau<D_\\gamma$. The upper bound $\\dim_F\\le D_\\gamma$ is short: it combines the classical inequality $\\dim_F\\le\\dim_2$ with the already established correlation dimension $\\dim_2(\\mu_{\\gamma,\\mathrm{GMC}})=D_\\gamma$. The new content is the lower bound; the theorem is that the two dimensions agree.","pith_inferences":["A purely internal proof of the correlation-dimension identity would make the theorem self-contained; the paper's own lower bound already gives the reverse inequality.","The same $\\ell^q$-martingale scheme should transfer to the unit-circle GMC by swapping in a hyperbolic-disk white-noise decomposition, as the paper notes; a direct consequence, if carried out, is an exact Fourier dimension for that model as well.","Because the paper shows bounded continuous perturbations of the log-kernel can change Fourier dimension, it is natural to conjecture that the equality survives exactly for perturbations in some smoothness class; identifying the minimal class would decide which higher-dimensional chaos models admit the same sharp result.","The sharp decay threshold suggests that the normalized coefficients $n^{D_\\gamma/2}\\widehat{\\mu}_{\\gamma,\\mathrm{GMC}}(n)$ fluctuate on the scale of their mean for all sub-critical $\\gamma$, not only in the small-$\\gamma$ range where a central limit theorem is already known."],"forward_implications":["For every sub-critical $\\gamma$, the Fourier dimension of the standard 1D GMC is known exactly, so the decay rate of its Fourier coefficients is no longer an open question.","For small parameters, the result upgrades the previously known upper bound and the nonexplicit lower bound to the sharp equality $1-\\gamma^2$.","For large sub-critical parameters, it gives the first exact value $(\\sqrt2-\\gamma)^2$, in the regime where the Fourier dimension is strictly smaller than the Hausdorff dimension.","Almost-sure $\\alpha$-upper Frostman regularity holds for every $\\alpha < D_\\gamma/2$, and Fourier restriction estimates hold with the stated range of exponents.","The same method, with one white-noise decomposition replaced by a hyperbolic-disk decomposition, yields the identical theorem for the GMC on the unit circle."],"supporting_citations":[{"why":"It supplies the vector-valued martingale method for Mandelbrot cascades that this paper adapts to GMC, including the idea of an $\\ell^q$-valued martingale.","marker":"[CHQW24]"},{"why":"It provides the L^q-spectrum computation from which the correlation dimension $D_\\gamma$ is read off; this is the external input for the upper bound.","marker":"[Ber23]"},{"why":"It gives the Riesz-energy divergence criterion used as an alternative route to the upper bound and for the zero-one law argument.","marker":"[LRV15]"},{"why":"It contains the martingale type $p$ inequalities for $\\ell^q$ that underlie the localization step.","marker":"[Pis16]"},{"why":"It introduces the white-noise decomposition of log-correlated fields used to build the approximating GMC measures.","marker":"[BM03]"},{"why":"It formulates the conjecture, proves the Rajchman property, and supplies the small-parameter central limit theorem used for one upper-bound argument.","marker":"[GV23]"},{"why":"It establishes basic properties of the lognormal multiplicative chaos construction, including the covariance computation used throughout.","marker":"[BKN+15]"}],"fun_headline_variants":["Exact Fourier dimension for 1D multiplicative chaos","Garban-Vargas conjecture proved for 1D GMC","Martingale method yields exact Fourier dimension","Fourier dimension of GMC pinned down exactly","Fourier dimension equals correlation dimension for subcritical GMC"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The upper bound assumes, as an imported result, that the two-point correlation dimension of this random measure is almost surely $D_\\gamma$; if that identity fails for the exact log-kernel on $[0,1]$, the equality could fail even though the paper's own lower-bound proof remains valid.","fun_headline_variants_meta":{"raw":{"variants":["Exact Fourier dimension for 1D multiplicative chaos","Garban-Vargas conjecture proved for 1D GMC","Martingale method yields exact Fourier dimension","Fourier dimension of GMC pinned down exactly","Fourier dimension equals correlation dimension for subcritical GMC"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001621,"raw_usage":{"total_tokens":6416,"prompt_tokens":878,"completion_tokens":5538,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":494,"completion_tokens_details":{"reasoning_tokens":5460}},"tokens_in":494,"tokens_out":5538,"duration_ms":33188,"temperature":1.0,"reasoning_tokens":5460,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T15:44:50.875591+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the correlation dimension of the exact-log-kernel GMC measure on $[0,1]$ directly from the Riesz-energy definition, for example numerically for $\\gamma=0.8$: the double integral $\\int\\int \\mu(dt)\\mu(ds)/|t-s|^s$ should diverge for every $s>D_{0.8}$ and converge for every $s<D_{0.8}$. A deviation from this threshold at any sub-critical $\\gamma$ would falsify the upper bound and hence the equality.","supporting_citations":[],"review_version":1}