{"id":"31f55c46-5170-415b-9c15-2417505a951f","arxiv_id":"2507.07321","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Self-similar measures pushed to non-degenerate curves satisfy optimal L^p-flattening estimates for their Fourier transforms, implying quantitative dimension improvement under convolution.","lead":"For any self-similar fractal measure on a line, bending it along a non-degenerate curve in higher-dimensional space forces its Fourier transform to spread out evenly, so that its L^p mass grows only like R^ε. The paper shows this quantitatively and proves that convolving with such a curved measure improves the effective dimension of any other measure.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Unproved adaptation of [Kha23, Thm 6.23] is load-bearing: if (1.2) does not imply uniform subspace Frostman bounds, Lemma 3.3 and Proposition 3.1 collapse.","rationale":"The reader's weakest-assumption analysis identifies exactly the point where the argument is least secure, and my reading of Sections 2 and 3 confirms that the chain Theorem 2.4 → Proposition 2.3 → Lemma 3.3 → Proposition 3.1 is load-bearing. The rest of the proof is coherent: the self-affine realization of V_dμ (Lemma 2.10), the Cauchy-Schwarz averaging in Lemma 3.2, the Taylor reduction for general curves in Proposition 1.5, and the near-first-coordinate estimate in Section 4 all appear internally consistent apart from minor notational slips, such as the displayed formula for ζ·C1 in Lemma 3.2 which appears to write λ^k where the subsequent use of λζ·y suggests λ. Those slips are repairable and do not change the assessment. The decisive issue remains the unproved adaptation of a preprint theorem by the second author: if (1.2) does not imply uniform Frostman control on all proper affine subspaces, then the central theorem is unsupported. A conditional verdict is therefore appropriate, pending a precise proof or citation of Theorem 2.4.","tokens_in":21360,"tokens_out":24199,"duration_ms":257854,"concrete_test":"Take [Kha23, Theorem 6.23] and write out a self-contained derivation of Theorem 2.4 from it, checking every hypothesis of the source theorem for an arbitrary compactly supported probability measure satisfying (1.2) with no extra stationarity or self-similarity assumption. In particular, determine whether the source proof uses (1.2) only through the moment-sum estimate s_m(ν,2) ≪ 2^{-m(d−ε)} or whether it also uses additional structure such as a Markov partition, an entropy-growth estimate, or an invariant measure. If additional structure is used, trace whether it holds for the specific measures \tilde V_{d−1}μ appearing in Lemma 3.3; if it does not, the transfer fails and Proposition 3.1 must be reproved.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"Theorem 2.4 is asserted without proof; Remark 2.5 concedes that its statement differs from [Kha23, Theorem 6.23] and that the proof there is only 'written for measures satisfying (1.2)'. This theorem is the sole bridge from the inductive flattening hypothesis to the bad-frequency estimate in Proposition 3.1. Proposition 2.3 uses it to transfer uniform subspace Frostman bounds to discretizations; Lemma 3.3 applies the result to the measures γλ with ε ≍ ∥ζ∥^{-ϱ}; and the resulting bound γλ ∗ γˇλ({y : λζ·y ∈ I}) ≤ C∥λζ∥^{-β} is what turns Tsujii's exceptional set B into a negligible loss. Without that bound, the terms ∥ζ∥^{(ε−β)ϱ} and ∥ζ∥^{-ϱδ} in the final display of Proposition 3.1 are unjustified, so the pointwise decay of Proposition 1.5, and hence Theorem 1.3 and Theorem 1.1, have no proof. The paper does not supply a precise source statement, a proof, or a verification that all hypotheses of the source theorem are satisfied by the arbitrary compactly supported measures to which Theorem 2.4 is applied.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proves a strong L^2-flattening estimate for the pushforwards of non-atomic self-similar measures on R to non-degenerate curves in R^d. Specifically, for every epsilon>0 there is p>1 such that the L^p norm of the Fourier transform of the pushed measure over the ball B(R) grows like O_epsilon(R^epsilon). The proof decomposes frequency space into two regions, proves a pointwise decay estimate away from the first coordinate via self-affineness of the moment curve and an inductive Fourier-flattening hypothesis, and proves an average decay estimate near the first coordinate via Tsujii's large-deviations theorem. A corollary gives a quantitative L^2-dimension improvement under convolution. The strategy is novel and largely self-contained except for a few external ingredients.","tokens_in":21597,"tokens_out":31216,"duration_ms":279636,"significance":"If correct, this is a major advance: it shows that all non-atomic self-similar measures, regardless of their Frostman exponent, become optimally flat when pushed to non-degenerate curves, improving on earlier results for Frostman measures on parabolas and other curves. The proof technique, combining dimension induction with the self-affine structure of moment-curve pushforwards, is original and likely to influence future work. The paper also makes explicit the role of L^p-flattening as a sufficient condition for quantitative convolution improvement. However, the proof relies on an unproved and slightly mis-cited external theorem, which is load-bearing; the needed repair is local and does not invalidate the overall strategy.","major_comments":[{"comment":"Theorem 2.4 is a load-bearing statement asserting that (1.2) implies uniform Frostman bounds on proper affine subspaces, yet it is not proved. Remark 2.5 concedes that the statement differs from [Kha23, Theorem 6.23] and only says that the proof there is 'written for measures satisfying (1.2)' without reproducing that proof or stating the reference theorem precisely. This theorem is used to prove Proposition 2.3, which in turn is the only bridge to the Fisherman bound in Lemma 3.3; if Theorem 2.4 fails, the bound on the bad-frequency set in Proposition 3.1 collapses. The authors must either prove Theorem 2.4 in the appendix or provide the exact statement of the cited theorem and a verifiable argument that its proof transfers to the hypothesis (1.2) as defined in this paper.","section":"Section 2.2, Theorem 2.4 and Remark 2.5"},{"comment":"Lemma 3.3 applies Proposition 2.3 to the measure \\tilde{\\nu}=\\widetilde{eV}_{d-1}\\mu, but Proposition 2.3 requires a measure of the form Q\\mu with Q(x)=(x,g(x)). The measure \\tilde{\\nu} is not of this form: it is the pushforward of \\mu by the curve (2x,3x^2,\\dots,d x^{d-1}) without a distinguished first coordinate. The intended conclusion is likely obtainable by a direct application of Theorem 2.4 plus the same Lipschitz-approximation argument used in Proposition 2.3, since that proof does not otherwise use the graph structure. However, as written, the invocation is unjustified. The authors should either state and prove a generalized version of Proposition 2.3 for arbitrary measures satisfying (1.2) and their discretizations, or spell out the direct argument in Lemma 3.3.","section":"Section 3.1, Lemma 3.3"},{"comment":"The displayed formula for ζ·C^1_{ω1,ω2} is incorrect. It reads ∑_{k=2}^d λ^k(t_{ω1}^{k-1}−t_{ω2}^{k-1})ζ_{k-1}, but by the structure of the first column of A_ω the correct expression is ∑_{k=2}^d k λ (t_{ω1}^{k-1}−t_{ω2}^{k-1})ζ_{k-1}, i.e., λ times the dot product of ζ with (2(t_{ω1}−t_{ω2}), 3(t_{ω1}^2−t_{ω2}^2), …). The subsequent integral expression with γλ∗γ̌λ evaluated at λζ·y appears to use the correct formula, but the erroneous identity as written is a technical error in a key derivation and must be corrected.","section":"Section 3.1, Lemma 3.2"}],"minor_comments":[{"comment":"The parameter ϱ in the conclusion is never defined in the statement; it appears to be a constant depending on the measure and the discretization, and its relation to γ and β should be made explicit.","section":"Section 2.2, Proposition 2.3"},{"comment":"The definition of F_i contains a typesetting ambiguity: the vector multiplied by A_i has a leading minus sign that is visually confusing. Please clarify the expression.","section":"Section 2.4, Lemma 2.10"},{"comment":"The phrase 'For the Frostman exponents0 of µ' is garbled; it should read 'For the Frostman exponent s0 of µ', and the notation s0 should be introduced at that point.","section":"Section 3.1, Lemma 3.4"},{"comment":"The parameter ε is used both as the argument of the Frostman bound and as the exponent in the covering estimate in the proof; this overloaded notation is confusing and should be separated.","section":"Section 3.1, Lemma 3.3"},{"comment":"The phrase 'for all p ≫ε,µ 1' is nonstandard; it should say 'for all p sufficiently large depending on ε and µ'.","section":"Section 4, Lemma 4.4"}],"recommendation":"major_revision","confidential_remarks":"The paper's central argument is strategically sound and the result, if established, is significant. The main obstacle is the unproved Theorem 2.4, which is essential to the proof of Proposition 3.1. Since the reference [Kha23] is authored by one of the current authors, I recommend that the editor request either a proof of Theorem 2.4 within the paper or a precise statement and a detailed verification of the claimed transfer. The other issues (misapplication of Proposition 2.3 and the incorrect identity in Lemma 3.2) are repairable but also need attention."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's the honest take: this is a strong paper with a real theorem, but right now it rests on an unproved adaptation of a result from the second author's preprint, and that piece is load-bearing. I'd send it to a serious referee, but I'd ask the authors to supply the missing proof before I'd use it myself.\n\nWhat's new: Theorem 1.1 gives O(R^ε) L^p-flattening for pushforwards of non-atomic self-similar measures under non-degenerate curves in every dimension, with the bound uniform in the measure's dimension. This is a clean advance over Orponen, Dasu–Demeter, OPP, and Demeter–Wang, which were dimension-dependent and restricted to special curves. The proof strategy is genuinely new: push to moment curves, exploit the self-affine structure of the pushed measure (Lemma 2.10), use the induction hypothesis on projections to get uniform Frostman bounds, then apply Tsujii's large deviations estimate. The decomposition of the frequency ball into C_{R,ε} and E_{R,ε} is elegant, and Proposition 1.5's pointwise decay away from the first coordinate is a nice step. The exposition is clear and the literature review is careful.\n\nThe main weakness is Theorem 2.4. It asserts that (1.2) implies uniform Frostman bounds ν(W(ε)) ≤ C ε^β for all proper affine subspaces. The paper cites [Kha23, Theorem 6.23], but Remark 2.5 admits the statement differs from the reference and that the proof there is only 'written for measures satisfying (1.2)'. That is not a proof. This theorem is used at the critical step in Lemma 3.3 to control the bad-frequency set; without it the final line of Proposition 3.1 has no justification. I don't see an obvious counterexample—plausibly (1.2) does force that uniformity—but it needs a real proof or a precise statement with all hypotheses checked. This is fixable, but it is currently a gap, not a minor footnote.\n\nI also did not verify every dependency among ε, p, ϱ, and δ in Sections 3 and 4; the reader flagged those parameter bookkeeping details, and a careful referee should check them. Apart from that, the structure is coherent and I found no explicit error.\n\nWho is this for? Harmonic analysts and fractal geometers working on Fourier decay of self-similar measures. If Theorem 2.4 gets resolved, this will be a valuable paper. As it stands, it deserves serious peer review, but I would not cite it as a black box yet.","headline":"A genuinely new and important flattening theorem, but the proof currently leans on an unproved adaptation of a cited preprint result that is load-bearing and needs to be sorted out before the paper is relied upon.","tokens_in":22155,"tokens_out":7990,"would_cite":false,"duration_ms":92541,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["28A80","42B10"],"pacs":[],"model":"deepseek-v4-flash","headline":"For any non-atomic self-similar measure on the line, its pushforward to a non-degenerate curve in $\\mathbb{R}^d$ has Fourier transform whose $L^p$ mass on every large ball grows at most like $R^\\varepsilon$, for any $\\varepsilon>0$…","keywords":["L^2-flattening","self-similar measures","Fourier transform","moment curves","non-degenerate curves","Frostman exponents","self-affine measures"],"falsifier":"Read the proof of the cited Theorem 6.23 and check whether it holds verbatim for every compactly supported probability measure satisfying (1.2); if it requires an extra hypothesis not stated in Theorem 2.4, such as self-similarity or an additional non-concentration condition, then Lemma 3.3 and Proposition 3.1 are unsupported.","tokens_in":21142,"feed_emoji":"📐","tokens_out":11951,"duration_ms":103434,"temperature":0.7,"pith_summary":"This paper proves that when a non-atomic self-similar measure on the real line is pushed forward along a sufficiently curved curve into $\\mathbb{R}^d$ ($d \\geq 2$), its Fourier transform becomes flat at all large scales: for every $\\varepsilon > 0$ there is an exponent $p > 1$ such that $\\|\\hat{\\nu}\\|^p_{L^p(B(R))} = O_\\varepsilon(R^\\varepsilon)$ for every $R > 0$. Equivalently, the moment sums of repeated self-convolutions of the pushed measure decay almost as fast as those of Lebesgue measure in dimension $d$, and convolution with the measure quantitatively improves $L^2$-dimension. The result is asserted for every non-atomic self-similar measure, with no assumption on overlaps, Hausdorff dimension, or the fine structure of the fractal support, and for arbitrary non-trapped analytic or $C^{d+1}$ non-degenerate curves. The only obstruction is the geometry of the curve: if the measure lives on a proper affine subspace, the flattening bound cannot hold.","feed_headline":"Curved self-similar measures flatten to any polynomial power","feed_subtitle":"Pushforwards onto non-degenerate curves have L^p Fourier mass O(R^ε); convolution improves L^2-dimension.","key_machinery":"The proof is carried by two structural facts. First, pushforward of a self-similar measure to the moment curve $V_d(x) = (x, x^2, \\ldots, x^d)$ is itself a self-affine measure (Lemma 2.10); this lets the Fourier coefficient $\\hat{\\nu}(\\xi)$ be written as an average of Fourier coefficients $\\hat{\\mu}(\\lambda \\zeta \\cdot y)$ of the original one-dimensional measure, with the averaging measures $\\gamma_\\lambda$ determined by the discrete cut-set approximations. Second, a large-deviations estimate for self-similar measures (Theorem 2.8) controls the bad frequencies of $\\hat{\\mu}$, while the inductive flattening hypothesis for $V_{d-1}\\mu$ yields uniform Frostman bounds for projections of the $\\gamma_\\lambda$ onto affine subspaces (Proposition 2.3), ensuring the averaged frequencies are well separated. The general curve case is reduced to the moment curve by Taylor expansion and a determinant non-vanishing argument for non-degenerate or non-trapped curves, and the full theorem follows by induction on the ambient dimension after decomposing frequency space into a region with dominant first coordinate and its complement.","core_discovery":"On its own terms, the paper's central assertion is Theorem 1.1: for a non-atomic self-similar measure $\\mu$ on $\\mathbb{R}$ and its pushforward $\\nu = Q\\mu$ to a non-trapped analytic or $C^{d+1}$ non-degenerate curve $Q$ in $\\mathbb{R}^d$, one has $\\|\\hat{\\nu}\\|^p_{L^p(B(R))} = O_\\varepsilon(R^\\varepsilon)$ for every $\\varepsilon > 0$ with a suitable $p > 1$. A sympathetic reading is that the Fourier mass of such measures is uniformly sub-polynomial at every large scale, and that this is a flattening phenomenon: in the equivalent moment-sum form, for every $\\varepsilon > 0$ there is $p \\in \\mathbb{N}$ with $s_m(\\nu^{*p}, 2) = O(2^{m(\\varepsilon - d)})$, meaning the $p$-fold self-convolution of the pushed measure behaves, down to the arbitrary small error $\\varepsilon$, like Lebesgue measure in dimension $d$. The paper claims this for all non-atomic self-similar measures on the line regardless of overlaps or Hausdorff dimension, and identifies the non-degeneracy of the curve as the only obstruction: measures supported on a proper affine subspace cannot satisfy the bound.","pith_inferences":["The stationarity of self-similar measures, rather than their Frostman dimension, is what defeats the general threshold of the parabola example discussed in the paper; a natural test is whether Ahlfors-David regular measures on the same curves also flatten, a question the paper leaves open.","The argument sidesteps the non-concentration condition that fails near tangent lines of the curve, so the same three-step structure—self-affine representation, projection Frostman bounds, large deviations—might yield flattening for pushforwards of other stationary measures such as self-conformal measures, where analogous large-deviations estimates are available.","A likely extension is to higher-dimensional bases: the same induction should give $L^2$-flattening for pushforwards of self-similar measures on $\\mathbb{R}^k$ to non-degenerate curves in $\\mathbb{R}^{k+d}$, with the moment curve $V_{k+d}$ playing the role of $V_d$."],"forward_implications":["For every $\\varepsilon > 0$, the $L^p$ mass of the Fourier transform of such a curved self-similar measure over the ball $B(R)$ grows at most like $R^\\varepsilon$, a super-polynomial flattening valid in every dimension $d \\geq 2$.","The moment sums of the $p$-fold self-convolution $\\nu^{*p}$ satisfy $s_m(\\nu^{*p}, 2) = O(2^{m(\\varepsilon - d)})$, so repeated convolution brings the measure arbitrarily close to Lebesgue-like uniformity in $d$ dimensions.","Convolution with $\\nu$ is quantitatively $L^2$-dimension improving: if a probability measure $\\theta$ has dyadic energy above a certain threshold, then $s_m(\\theta * \\nu) \\leq 2^{-\\eta m} s_m(\\theta, 2)$ with $\\eta > 0$ independent of $\\theta$.","For every $q \\in [2, \\infty]$, the $L^q$-dimension of the iterated self-convolutions $\\nu^{*n}$ converges to $d$ as $n \\to \\infty$.","The non-degeneracy and non-trapped hypotheses are necessary: any measure supported on a proper affine subspace has $|\\hat{\\nu}(\\xi)| \\approx 1$ for frequencies nearly orthogonal to that subspace, so the flattening bound cannot hold."],"supporting_citations":[{"why":"Supplies the uniform Frostman bound on affine subspaces for measures satisfying (1.2), adapted in Theorem 2.4 and used in Lemma 3.3.","marker":"[Kha23, Theorem 6.23]"},{"why":"Gives the large-deviations estimate for the Fourier transform of non-atomic self-similar measures, stated as Theorem 2.8 and used in Corollary 2.9.","marker":"[Tsu15]"},{"why":"Yields the self-affine iterated function system for the moment-curve pushforward of a self-similar set, used in Lemma 2.10.","marker":"[FK18, Lemma 3.1]"},{"why":"Provides upper Frostman (Hölder) regularity of non-atomic self-similar measures, used in Proposition 2.6 and Lemma 3.4.","marker":"[FL09, Proposition 2.2]"},{"why":"Supplies the argument converting the flattening estimate into convergence of $L^q$-dimensions and the quantitative convolution improvement of Corollary 1.2.","marker":"[MS18]"},{"why":"Provides the Lipschitz slow-variation detail used to derive the covering statement in Corollary 2.9 from the large-deviations estimate.","marker":"[ACWW25, Corollary 2.5]"}],"fun_headline_variants":["Convolving with curved self-similar measures improves L^2-dimension","For any ε, some p>1 makes L^p Fourier mass O(R^ε) on curves","Flattening on curves: pushforward self-similar measures have sub-polynomial Fourier mass","Self-similar measures on non-degenerate curves have O(R^ε) L^p Fourier mass"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof relies on an unproved transfer: Remark 2.5 states that a Frostman bound from the cited reference applies to every measure satisfying (1.2), even though the published statement differs; if that transfer is incorrect, the moment-curve case collapses.","fun_headline_variants_meta":{"raw":{"variants":["Convolving with curved self-similar measures improves L^2-dimension","For any ε, some p>1 makes L^p Fourier mass O(R^ε) on curves","Flattening on curves: pushforward self-similar measures have sub-polynomial Fourier mass","Self-similar measures on non-degenerate curves have O(R^ε) L^p Fourier mass"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.002529,"raw_usage":{"total_tokens":9693,"prompt_tokens":944,"completion_tokens":8749,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":560,"completion_tokens_details":{"reasoning_tokens":8653}},"tokens_in":560,"tokens_out":8749,"duration_ms":66299,"temperature":1.0,"reasoning_tokens":8653,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T18:44:10.214913+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Read the proof of the cited Theorem 6.23 and check whether it holds verbatim for every compactly supported probability measure satisfying (1.2); if it requires an extra hypothesis not stated in Theorem 2.4, such as self-similarity or an additional non-concentration condition, then Lemma 3.3 and Proposition 3.1 are unsupported.","supporting_citations":[],"review_version":1}