{"id":"5e5d4aa6-a62b-4c6f-a82e-ccf76e6872a3","arxiv_id":"2505.16405","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Almost surely, the Fourier dimension of a Mandelbrot microcanonical cascade measure equals log_2(1/(E[W0^2]+E[W1^2])), settling the Mandelbrot-Kahane problem for this class.","lead":"The paper proves an exact formula for the Fourier dimension of Mandelbrot microcanonical cascade measures, a long-open problem in random fractal harmonic analysis. It also gives sharp Holder exponents for the Dubins-Freedman random homeomorphisms.","discovery_kind":"first_principles","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The lower-bound half of Theorem 1.1 is carried by an unproved imported lemma: Section 6 converts integer-frequency decay into full real-line Fourier dimension via [CHQW24a, Lemma 1.8], which is not reproduced or verified here.","rationale":"The central theorem is a complete, exact Fourier-dimension formula. The proof has two halves: the lower bound from Proposition 1.4 plus an imported integer-to-real transfer, and the upper bound from the dyadic CLT plus Lemma 6.1. I checked the internal machinery of both halves: the martingale-difference decomposition in Lemma 4.2 is algebraically consistent, the variance computations in Section 5 match after computing T(u,s,m) = 2(1-e^{i2πs2^{-m}})W̃_0, the Biggins condition in Lemma 5.6 follows from the strict monotonicity of K_W on [1,2], and the conditional CLT application is standard given the stated Lindeberg estimate. I found no internal contradiction or computational error that would invalidate the argument as written. The single load-bearing uncertainty is therefore the same one the reader flagged: the paper relies on two lemmas from the companion paper [CHQW24a] without reproducing them. Lemma 1.8 is the more delicate of the two, because Fourier decay on integers is not a priori equivalent to real-line Fourier dimension for a general measure, and the paper's own Section 4 only produces integer-frequency bounds. Proposition A.3, by contrast, is a standard conditional CLT and its failure here would be more surprising, but it is still an unchecked import. Since both imports are plausibly correct and the internal structure supports them, the appropriate disposition remains conditional: accept the result provisionally pending independent verification of the two companion lemmas, rather than rejecting or fully accepting. The reader's conditional verdict and moderate confidence are therefore appropriate, and my stress-test does not change that verdict.","tokens_in":23142,"tokens_out":32318,"duration_ms":270211,"concrete_test":"Independently prove or refute the integer-to-real transfer used in Section 6: for a probability measure μ on [0,1] with no atoms at 0 and 1 and with |μ̂(n)| ≤ C|n|^{-α}, show that |μ̂(ξ)| = O(|ξ|^{-α+δ}) for every δ>0, using the expansion e^{2πiθx} = Σ_k (e^{2πiθ}-1)e^{2πikx}/(2πi(θ-k)) on (0,1). If a counterexample with no atoms at the endpoints exists, test whether the cascade measure μ∞ provides one; then the lower bound in Theorem 1.1 fails. If the transfer holds with an ε-loss, re-run Section 6 with that loss and confirm it still yields dim_F(μ∞) ≥ D_F.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's own arguments establish, in Proposition 1.4, a weighted ℓ^q-summability bound for the integer Fourier coefficients n^{D_F/2-ε} bμ∞(n). From this, Section 6 immediately concludes dim_F(μ∞) ≥ D_F-2ε by citing [CHQW24a, Lemma 1.8 or Remark 1.2]. That transfer is not a trivial consequence of the displayed estimates: decay along the integers does not automatically imply decay along all real frequencies unless one controls the factor e^{2πiθx} for the fractional part θ of ξ, and endpoint atoms can spoil the natural Fourier-series expansion used for such transfers. If Lemma 1.8 carries hidden hypotheses—for example absolute continuity, aperiodicity, or a specific mixing structure from the canonical cascade setting—then the lower bound in Theorem 1.1 is not established by this paper. The second import, Proposition A.3 (conditional Lindeberg-Feller CLT), is used exactly where the dyadic-subsequence fluctuation is converted into a nonzero limiting law; if the conditional independence required there is with respect to F_∞ rather than F_n, the verification in Section 5.4 would have to be reworked. The internal computations checked here (Lemmas 4.2, 5.1, 5.4, and the entropic Lemma 5.6 via Biggins) appear consistent, so the concern is specifically the two black-box imports, matching the reader's weakest assumption.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the Fourier dimension of Mandelbrot microcanonical cascade measures on [0,1]. For a splitting vector (W_0,W_1) with positive entries, mean 1/2, and W_0 not identically 1/2, the authors define D_F = log_2(1/(E[W_0^2]+E[W_1^2])) and prove in Theorem 1.1 that, almost surely, dim_F(mu_infinity) = D_F. The proof combines an L^2(ell^q) vector-valued martingale estimate for the integer Fourier coefficients (Proposition 1.4), a dyadic-subsequence fluctuation theorem for the rescaled coefficients (Proposition 1.5), and a transfer lemma from integer-frequency decay to real-frequency Fourier dimension. The paper also establishes Frostman regularity bounds and sharp Holder exponents for the associated Dubins-Freedman random homeomorphism (Proposition 1.2 and Corollary 1.3).","tokens_in":23384,"tokens_out":20569,"duration_ms":171570,"significance":"If the proof is fully justified, the paper resolves the microcanonical version of the Mandelbrot-Kahane problem with an exact, parameter-free formula for the almost-sure Fourier dimension. The internal estimates are mostly careful and substantial: the entropy-type monotonicity lemma (Proposition 3.1) is elegant, the martingale type-2 arguments in Section 4 are coherent, and the branching random walk arguments in Section 5 are applied with the correct normalizations. The paper also gives sharp Frostman and Holder statements for the Dubins-Freedman homeomorphism. The main weakness is that two load-bearing results are imported from the authors' companion preprint [CHQW24a] without statements or proofs, so the present text alone does not certify the full theorem.","major_comments":[{"comment":"The lower bound dim_F(mu_infinity) >= D_F - 2*epsilon is obtained by citing [CHQW24a, Lemma 1.8 or Remark 1.2] after Proposition 1.4 gives decay of the integer Fourier coefficients. This transfer is load-bearing and is not a formal consequence of the displayed estimates: controlling |hat mu(n)| for integer n does not by itself control |hat mu(xi)| for real xi, and endpoint atoms or other hypotheses would matter. Please state the lemma explicitly, verify its hypotheses for mu_infinity (in particular non-atomicity at {0,1} and any regularity or aperiodicity condition), and either prove it or give the precise statement with a proof in an appendix.","section":"Section 6, proof of Theorem 1.1"},{"comment":"The conditional Lindeberg-Feller central limit theorem is imported as [CHQW24a, Proposition A.3] without statement. The array used here is conditioned on F_n, which varies with n, and the conclusion requires that the limiting complex Gaussian be independent of M_infty^(2). Please reproduce the proposition and confirm that the hypotheses checked in Lemmas 5.7 and 5.8 (the supremum of Y(u) tending to zero and the conditional Lindeberg condition (5.23)) are exactly the hypotheses of that CLT, and clarify whether the conditional variance is required to converge in probability or almost surely.","section":"Section 5.4, proof of Proposition 1.5"}],"minor_comments":[{"comment":"In the two displayed identities after 'Thus', the first identity writes E[(Re V_n)^2 | F_n] = (rho+varpi)/2 * M_infty^(2); the subscript should be n, not infinity, to match the second identity and Lemma 5.4.","section":"Section 5.4, paragraph after Lemma 5.4"},{"comment":"The displayed expectation appears to place the exponent 2/q outside the expectation, whereas the proof and the equality with the L^2(ell^q)-norm require E[ ( sum_s |s^{D_F/2-epsilon} hat mu_infinity(s)|^q )^{2/q} ]. Please correct the typesetting and specify that the sum is over nonzero integers, or use s >= 1 together with the symmetry of the Fourier coefficients.","section":"Proposition 1.4 and Eq. (4.1)"},{"comment":"The claim that inequality (3.2) fails for d >= 17 is presented without data or a reference. Since this remark is not used in the paper, either provide the numerical experiment or delete the claim.","section":"Section 3, remark after Proposition 3.1"},{"comment":"The interpolation step is described as 'standard complex interpolation', but the norm varies in both the L^p and ell^p components. Please add a sentence or a reference explaining why the complex interpolation of L^{p0}(ell^{p0}) and L^{p1}(ell^{p1}) is L^{p_theta}(ell^{p_theta}).","section":"Proof of Proposition 3.1"},{"comment":"There are minor typos, including 'determing' in the abstract, and the phrase 'determing their exact Fourier dimensions' should read 'determining'. The formatting of some displayed formulas is also broken, notably in Proposition 1.4 and Lemma 4.4, and should be corrected during production.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is part of a two-paper project and imports two key results from [CHQW24a], a companion that is not yet published as far as the text indicates. The internal estimates in the present paper appear sound, so I am not recommending rejection, but I would ask the editor to require that the companion results be stated precisely, or that the authors add appendices with the necessary proofs, before the main theorem can be considered verified."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"What you should know first: this paper almost certainly solves the Mandelbrot-Kahane problem for microcanonical cascade measures, giving the exact Fourier dimension D_F and sharp Holder exponents for Dubins-Freedman homeomorphisms. Those are real, new results, not repackaged earlier work. The canonical-cascade solution in the authors' companion paper is a different class, and the transfer to microcanonical cascades requires genuinely new arguments, most notably the entropy-type monotonicity lemma (Prop 3.1) which is clean and independently interesting.\n\nThe internal proof is largely coherent. The martingale-type estimates in Section 4 are carefully done; the fluctuation analysis in Section 5 is plausible; and I checked the internal computations for Lemmas 4.2, 5.1, 5.4, and 5.6, which look consistent. Lemma 5.6 via Biggins is a proper application. So the concern is not sloppy math in the parts that are actually shown.\n\nThe soft spot is exactly where the reader's report and the stress-test note point: two load-bearing black-box imports from [CHQW24a]. Section 6 converts the integer-frequency decay of Proposition 1.4 into a full real-line Fourier dimension lower bound via [CHQW24a, Lemma 1.8 or Remark 1.2]. That transfer is not a trivial consequence of the displayed estimates; decay along integers does not automatically imply decay at all real frequencies unless some regularity or non-atomicity is controlled. The paper does not reproduce or verify that lemma, so the lower-bound half of Theorem 1.1 is not self-contained. The second import, Proposition A.3's conditional Lindeberg-Feller CLT, is used to get the nonzero limiting law; I expect it is fine, but it is also not checkable from this text alone.\n\nThere are minor formal blemishes: the proof of Lemma 4.2 drops the s^alpha factor in a display, and the sharpness statement in Proposition 1.2 is awkward when gamma_-o is infinite. Neither appears to break the main theorem.\n\nWho is this for? Specialists in random measures, Fourier dimension, and branching random walks. A reader who wants a fully self-contained proof will be frustrated, but the result is important enough that the companion lemmas should be verified rather than assumed.\n\nRecommendation: send to peer review. The referee should have access to [CHQW24a] and be explicitly asked to check Lemma 1.8 and Proposition A.3 against the microcanonical setting. If they hold, this is publishable as a major result.","headline":"A serious, likely correct solution to a named 1976/1993 problem, but the proof leans on two unverified imports from the authors' companion paper; referee it with those imports checked.","tokens_in":23985,"tokens_out":2710,"would_cite":false,"duration_ms":25725,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60G57","42A61","46B09","60J80","60G46"],"pacs":[],"model":"deepseek-v4-flash","headline":"The Fourier dimension of every microcanonical cascade measure is almost surely a fixed constant set by the second moment of its splitting weights.","keywords":["Mandelbrot microcanonical cascades","Fourier dimension","vector-valued martingales","Hölder regularity of random homeomorphisms","branching random walks","Mandelbrot-Kahane problem","Frostman regularity","Dubins-Freedman random homeomorphisms"],"falsifier":"Simulate the cascade for the uniform splitting rule $W=(U,1-U)$ with $U$ uniform on $(0,1)$, where $D_F=\\log_2(3/2)\\approx0.585$, and estimate the decay exponent of $|\\widehat{\\mu}_\\infty(\\xi)|$ for large real frequencies $\\xi$; a statistically robust exponent differing from $0.585$ would refute the claimed almost-sure formula.","tokens_in":22890,"feed_emoji":"🌀","tokens_out":22224,"duration_ms":163255,"temperature":0.7,"pith_summary":"This paper solves the Mandelbrot-Kahane problem for microcanonical cascade measures, showing that their Fourier dimension is almost surely a deterministic constant $D_F$ rather than a random or distribution-dependent quantity. The value depends only on the second moments of the two splitting weights, so the result is a single closed formula for an entire family of random fractal measures. The Fourier dimension controls the high-frequency decay of the measure's Fourier transform, so knowing it exactly says how harmonically spread out these turbulence-inspired measures are. The same analysis yields sharp upper and lower Frostman exponents and, as a consequence, sharp Hölder exponents for the associated Dubins-Freedman random homeomorphism and its inverse.","feed_headline":"One formula fixes the Fourier dimension of random cascade measures","feed_subtitle":"The exact decay exponent is a fixed number set by the second moment of the splitting weights.","key_machinery":"The proof runs on two martingales. The first is the $\\ell^q$-valued martingale $M_n=(s^\\alpha \\widehat{\\mu}_n(s))_{s\\geq1}$ of weighted Fourier coefficients of the approximating measures; its $L^2(\\ell^q)$ bound is obtained by applying the martingale type-2 inequality for $\\ell^q$ ($q>2$) twice, once globally and once conditionally on the dyadic decomposition, reducing the problem to a geometric sum whose ratio is $(\\mathbb{E}[W_0^2]+\\mathbb{E}[W_1^2])2^{2\\alpha+2/q}<1$. The second is the non-negative martingale $M^{(2)}_n=(8\\mathbb{E}[W_0^2])^{-n}\\sum_{|u|=n}\\prod_{j=1}^n X(u|_j)^2$, whose limit is shown to be positive almost surely by combining the branching random walk martingale convergence theorem with a new entropy-type monotonicity of $K_V(p)=\\log((\\mathbb{E}[V_0^p+V_1^p])^{1/p})$ on $[1,2]$ for two-dimensional splitting vectors. These two mechanisms are linked by the identity $\\widehat{\\mu}_\\infty(2^n)=2^{-n}\\sum_{|u|=n}(\\prod_{j=1}^n X(u|_j))\\widehat{\\mu}^{(u)}_\\infty(1)$, which expresses dyadic Fourier coefficients as weighted sums of i.i.d. boundary copies and feeds the conditional central limit theorem that pins down the optimal exponent.","core_discovery":"The paper's central claim is the almost sure equality $\\dim_F(\\mu_\\infty)=D_F$, where $\\mu_\\infty$ is the Mandelbrot microcanonical cascade measure on $[0,1]$ and $D_F=\\log_2(1/(\\mathbb{E}[W_0^2]+\\mathbb{E}[W_1^2]))$, under the standing assumptions $W_0+W_1=1$ and $\\mathbb{E}[W_0]=\\mathbb{E}[W_1]=1/2$. The lower bound $\\dim_F(\\mu_\\infty)\\geq D_F$ follows from a vector-valued martingale estimate showing that $\\sum_{n\\geq1} (|n|^{D_F/2-\\varepsilon}|\\widehat{\\mu}_\\infty(n)|)^q$ has finite expectation for some $q>2$, which forces the Fourier coefficients to decay faster than $|n|^{-D_F/2+\\varepsilon}$ almost surely. The matching upper bound $\\dim_F(\\mu_\\infty)\\leq D_F$ comes from the dyadic self-similarity of the measure: the rescaled coefficients $2^{nD_F/2}\\widehat{\\mu}_\\infty(2^n)$ converge in distribution to the product of a non-degenerate complex Gaussian and the square root of a positive martingale limit, so no exponent above $D_F$ can hold. Along the way the paper proves the Frostman regularity exponents $\\gamma_o^+$ and $\\gamma_o^-$ are sharp, which yields sharp bi-Hölder regularity for the Dubins-Freedman random homeomorphism $F_\\infty$ and its inverse.","pith_inferences":["A natural extension would replace the binary tree by a $b$-ary tree; the same martingale arguments should give $D_F^{(b)}=\\log_b(1/(b\\mathbb{E}[W_0^2]))$ for $b$-ary microcanonical cascades, provided the two imported lemmas carry over.","The entropy-type monotonicity that underpins the positivity of the martingale limit is special to two-dimensional splitting vectors, and the paper notes the corresponding inequality fails in high dimension; exact Fourier-dimension formulas for cascades with three or more children will likely need a qualitatively different proof.","The inequality $\\gamma_o^+>D_F/2$ noted in the paper means these measures always have a Frostman regularity strictly better than half their Fourier dimension, so their geometric and harmonic scaling behaviours are genuinely different.","The sharp bi-Hölder control suggests the uniform-splitting Dubins-Freedman homeomorphisms may achieve the optimal rate for accelerating Fourier series convergence by a random change of variable, a direction the paper mentions but does not pursue."],"forward_implications":["The Mandelbrot-Kahane problem for microcanonical cascades is closed: the Fourier dimension is exactly $D_F$, almost surely, for every admissible splitting law.","Because $D_F$ depends only on $\\mathbb{E}[W_0^2]$, two microcanonical cascades with the same second moment of the splitting weights share the same Fourier dimension no matter how different their higher-order structure is.","The upper-bound argument shows that the correlation dimension and the Fourier dimension coincide almost surely for this family, since the paper cites the equality $D_2(\\mu_\\infty)=D_F$ as an alternative route to the same exponent.","For the uniform splitting rule $W=(U,1-U)$ with $U$ uniform on $(0,1)$, the formula reads $D_F=\\log_2(3/2)\\approx0.585$, giving a concrete numerical prediction for the Fourier decay of the corresponding random homeomorphism.","The sharp Hölder exponents $\\gamma_o^+$ and $(\\gamma_o^-)^{-1}$ for $F_\\infty$ and $F_\\infty^{-1}$ mean the bi-Hölder regularity of Dubins-Freedman random homeomorphisms is now exactly known for all admissible weight vectors."],"supporting_citations":[{"why":"Supplies the two imported proof tools: the lemma turning integer-frequency Fourier decay into the lower bound on the full Fourier dimension, and the conditional central limit theorem used to prove the upper bound.","marker":"[CHQW24a]"},{"why":"Provides the martingale type-2 inequalities for $\\ell^q$ that deliver the polynomial Fourier decay in Proposition 1.4.","marker":"[Pis16]"},{"why":"Supplies the branching random walk martingale convergence theorem used to show the limit $M^{(2)}_\\infty$ is almost surely positive and to compute the Frostman exponents.","marker":"[Shi95]"},{"why":"Gives the complex interpolation theorem used to prove the entropy-type monotonicity of $K_V(p)$ on $[1,2]$, which is needed for the strict inequality in the non-vanishing martingale proof.","marker":"[BL76]"}],"fun_headline_variants":["Exact Fourier dimension for microcanonical cascade measures","Mandelbrot-Kahane solved: cascade measures' Fourier dimension exact","Sharp Hölder regularity for Dubins-Freedman homeomorphisms","Cascade Fourier dimension equals a constant from second moment","Exact Fourier dimension solves Mandelbrot-Kahane problem"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof leans on two technical results imported from the authors' earlier companion paper — one that turns decay at integer frequencies into a lower bound on the full Fourier dimension, and one that supplies a central limit theorem for sums of random pieces — and the exact formula holds only if both apply to microcanonical cascades as stated.","fun_headline_variants_meta":{"raw":{"variants":["Exact Fourier dimension for microcanonical cascade measures","Mandelbrot-Kahane solved: cascade measures' Fourier dimension exact","Sharp Hölder regularity for Dubins-Freedman homeomorphisms","Cascade Fourier dimension equals a constant from second moment","Exact Fourier dimension solves Mandelbrot-Kahane problem"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000608,"raw_usage":{"total_tokens":2812,"prompt_tokens":905,"completion_tokens":1907,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":521,"completion_tokens_details":{"reasoning_tokens":1833}},"tokens_in":521,"tokens_out":1907,"duration_ms":12904,"temperature":1.0,"reasoning_tokens":1833,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T15:03:08.842879+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Simulate the cascade for the uniform splitting rule $W=(U,1-U)$ with $U$ uniform on $(0,1)$, where $D_F=\\log_2(3/2)\\approx0.585$, and estimate the decay exponent of $|\\widehat{\\mu}_\\infty(\\xi)|$ for large real frequencies $\\xi$; a statistically robust exponent differing from $0.585$ would refute the claimed almost-sure formula.","supporting_citations":[],"review_version":1}