{"id":"04dd8c52-e7dc-4113-80a1-9279734184db","arxiv_id":"2412.14896","paper_version":2,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":8.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"The Fourier spectrum yields a continuum of Stein-Tomas type restriction estimates that improve known L2 ranges for several natural measures and sharp negative thresholds.","lead":"The authors use the Fourier spectrum, a family of dimensions that interpolates between the Fourier and Sobolev dimensions of a measure, to prove new L2 restriction estimates for Fourier transforms. The new bounds improve the classical Stein-Tomas range for the cone, the moment curve, and many fractal measures, and also give sharp ranges where restriction must fail.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Equation (3.2) is invalid: Young’s inequality is applied to the Fourier multiplier instead of the convolution kernel, so the dyadic decay underpinning Theorem 3.1 is not established.","rationale":"The Fourier-spectrum framework and the explicit spectral computations for the cone, moment curve, and fractal examples are valuable, and the reader is right that no issue was found with the statements of the applications themselves. However, the proof of the central theorem rests on a dyadic estimate that appears to conflate a Fourier multiplier with its inverse Fourier transform. The reader’s flagged assumption (3.3) is actually a trivial consequence of |\\hat\\mu|≤1 on the j≥0 dyadic pieces, so the load-bearing weakness is one step earlier: the Young-inequality estimate (3.2). For the circle, the claimed decay is contradicted by stationary phase, showing the error is real rather than a matter of convention. Because the same estimate is used in Theorem 3.2, Corollary 3.4, and the applications, the manuscript should not be accepted as is; a corrected proof of the dyadic estimate, or a replacement interpolation argument, is needed. This is a correctness gap, not a disagreement with the field’s consensus, and it does not impeach the spectral computations themselves.","tokens_in":23466,"tokens_out":30624,"duration_ms":258159,"concrete_test":"For d=2, take μ the normalized surface measure on S^1 and choose θ=1/4, s=3/4 (so dθ<s<dim^θ_Fμ). Compute K_j=F^{-1}(φ(2^{-j}·)\\hat\\mu) and its L^8 norm by stationary phase, or numerically at large j. The estimate (3.2) predicts ‖K_j‖_{L^8}≲2^{-j/8}, whereas the radial stationary-phase computation gives ‖K_j‖_{L^8}∼2^{7j/8}. If this growth is confirmed, (3.2) is false and the interpolation proof of Theorem 3.1 is invalid.","verdict_should_be":"UNVERDICTED","load_bearing_attack":"In §3.1.1 the proof defines \\hat\\mu_j(ξ)=φ(2^{-j}ξ)\\hat\\mu(ξ) and then asserts ‖\\hat\\mu_j∗f‖_{L^{4/θ}} ≤ ‖\\hat\\mu_j‖_{L^{2/θ}}‖f‖_{L^{4/(4−θ)}}. Young’s inequality needs the L^{2/θ} norm of the kernel K_j=(\\hat\\mu_j)ˇ, not of the Fourier multiplier \\hat\\mu_j. The displayed computation bounds \\hat\\mu_j in L^{2/θ} using J_{s,θ}, so the two objects are conflated. This is not merely notational: for μ the circle measure in R^2, |\\hat\\mu(ξ)|∼|ξ|^{-1/2}; the multiplier norm decays for θ<1/2, while K_j, by stationary phase, has amplitude ∼2^j on a 2^{-j}-neighbourhood of the unit circle, so ‖K_j‖_{L^{2/θ}}∼2^{j(1−θ/2)} grows. Thus (3.2) is false in a standard example. Since the Riesz–Thorin interpolation between (3.2) and (3.3) is the sole argument for Theorems 3.1, 3.2, Corollary 3.4, and the restriction ranges in §§7–9, the central claim is currently unsupported. The reader’s identified assumption (3.3) is not the bottleneck: for j≥0 it follows from the trivial multiplier bound ‖\\hat\\mu_j‖_∞≤1, which is stronger than the stated exponent.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a Fourier-spectrum generalization of the Stein–Tomas restriction theorem for measures. Theorem 3.1 claims an L^q→L^2 extension bound for measures whose Fourier spectrum lies above the diagonal dθ, with an optimized threshold over θ. Theorem 3.2 extends this to Lorentz spaces, and Section 3.2 gives a partial converse (Theorem 3.6). The paper also computes the Fourier spectrum for the cone and moment curve and uses the main theorems to derive restriction ranges for these examples and for fractal measures. The negative-direction results and the explicit spectrum computations are self-contained and appear correct, but the proof of the central positive results rests on a claimed equivalence that is false.","tokens_in":23783,"tokens_out":38871,"duration_ms":262208,"significance":"If the main positive results were valid, they would provide a new, flexible tool for L^2 restriction theory by converting information about the Fourier spectrum into restriction ranges that often improve on Stein–Tomas. The explicit computations of the Fourier spectrum for the cone and the moment curve (Propositions 7.2 and 8.1) are valuable in their own right, as are the general negative results in Theorem 3.6 and Corollary 3.7. However, the central derivation is invalid because the proof establishes a different estimate than the theorem claims, and the asserted equivalence between these estimates is false. The paper is clearly written and well organized, and the referee verified that the Young-inequality step in Eq. (3.2) is correct as a bound for the convolution operator with the Fourier multiplier; the difficulty is not there but in the mismatch between the proved estimate and the stated restriction theorem.","major_comments":[{"comment":"The equivalence asserted between the L^2 restriction estimate (1.1) with p=2 and the convolution estimate (1.2) is false. The dual of R: L^{q'}→L^2(μ), R(f)=\\hat f|_μ, is the extension operator E: L^2(μ)→L^q, E(g)=\\widehat{g μ}, and \\|R\\|=\\|E\\|. The operator in (1.2) is T f = \\hat μ * f, and \\|T f\\|_q = \\|\\widehat{f^\\vee μ}\\|_q while \\|f\\|_{q'} = \\|\\widehat{f^\\vee}\\|_{q'}. Thus T measures the norm of the extension operator on the subspace of functions whose Fourier transform lies in L^{q'}, with respect to the L^{q'} norm of that Fourier transform, not on L^2(μ). A concrete counterexample is the surface measure σ on the sphere S^{d-1}: the Stein–Tomas L^2 restriction estimate holds for q ≥ 2(d+1)/(d-1), but \\|\\hat σ\\|_{(d+1)/(d-1)} = ∞, so (1.2) fails for that q (Young's inequality would require \\hat σ ∈ L^p with p=(d+1)/(d-1)). Consequently, proving (1.2) does not prove the restriction estimate (1.1) with p=2.","section":"Section 3.1.1, Proof of Theorem 3.1"},{"comment":"The proof explicitly seeks to establish the L^{q'}→L^q bound for the convolution operator f ↦ \\hat μ * f, which is (1.2). The dyadic decomposition in (3.1) is applied to this operator, and the interpolation in (3.2)–(3.4) uses the L^{4/(4-θ)}→L^{4/θ} estimate for that operator and the L^2→L^2 estimate for the same operator. No step uses the L^2(μ) norm of the function f that appears in the theorem's conclusion. In a genuine proof of the extension estimate \\|\\widehat{f μ}\\|_q ≲ \\|f\\|_{L^2(μ)}, one would decompose \\widehat{f μ} = \\hat f * \\hat μ and estimate each dyadic piece by \\|f\\|_{L^2(μ)} times a j-dependent factor. The proof here instead estimates \\hat μ_j * f by \\|f\\|_{q'}, so even if every displayed inequality is correct, the argument does not reach the stated L^2(μ)→L^q conclusion. This is a load-bearing gap, not a local issue.","section":"Section 4, Proposition 4.2 and Theorem 3.2"},{"comment":"The endpoint and Lorentz-space results inherit the same defect. The estimates (4.5) and (4.6) are exactly the bounds for the convolution operator \\hat μ_j * f from (3.2) and (3.3), and Bourgain's interpolation trick is applied to prove restricted weak-type bounds for that operator. The conclusion (4.3) is an L^{p,q}→L^{r,q} estimate for the convolution operator, not for the extension operator E(g)=\\widehat{g μ}. Since the paper's Theorem 3.2 is derived from Proposition 4.2 via the identity (4.4), which treats f as an L^{q'} function on R^d, the endpoint estimate does not follow from the proved Lorentz-space bounds for the desired L^2(μ) input.","section":"Section 7, Application to the cone"},{"comment":"The claimed improvement for the cone, q > (3d-4)/(d-2) for d ≥ 5, depends entirely on Theorem 3.1. Since the proof of Theorem 3.1 is invalid for the reasons above, the application is unsupported. The referee notes that Proposition 7.2, the explicit computation of dim_θ^F ν_{d-1}, may be correct and useful; however, its use in Section 7 relies on the unproved restriction theorem. The same caveat applies to the moment-curve example in Section 8 and the fractal examples in Section 9, which are all consequences of Theorems 3.1 or 3.2.","section":"Section 3.1.1, Eq. (3.2)"},{"comment":"The referee investigated the specific objection raised in the stress-test note regarding Young's inequality. That objection is not valid: in the operator f ↦ \\hat μ_j * f, the function being convolved is literally \\hat μ_j, so Young's inequality with norm \\|\\hat μ_j\\|_{L^{2/θ}} is a correct application. The problem lies not in this step but in the fact that the resulting estimate is for the wrong operator relative to the theorem statement.","section":"Section 3.1.1, Eq. (3.3)"},{"comment":"The estimate (3.3) quoted from [Moc00] is, for j ≥ 0, weaker than the trivial bound \\|\\hat μ_j * f\\|_2 ≤ \\|\\hat μ_j\\|_∞ \\|f\\|_2 ≤ \\|f\\|_2, since the exponent 2^{-j(α_0-d)} = 2^{j(d-α_0)} ≥ 1 for α_0 < d. Thus the Frostman hypothesis is not actually used to obtain a useful decay in this step; the interpolation weight in (3.4) comes entirely from the (3.2) side. This is not by itself an error, but it shows that the role of the Frostman exponent in the proof is different from what the text suggests.","section":"Section 1.1, Eq. (1.2)"}],"minor_comments":[{"comment":"Several references to 'Hambrook and /suppress Laba' are corrupted by a LaTeX artifact; the intended name is 'Hambrook and Łaba'.","section":"Throughout"},{"comment":"The name 'Mockenhaupt' is misspelled as 'Mochenhaupt' in the first paragraph.","section":"Section 1.2"},{"comment":"The notation B_k(0,r) for balls in R^k is used but not defined in the proof of Proposition 7.2; it would help to state it explicitly.","section":"Section 7.1"},{"comment":"The condition dim_F^θ μ > dθ should be read as 'for some θ' inside the infimum; the text is clear but a parenthetical remark would prevent confusion.","section":"Theorem 3.1"}],"recommendation":"reject","confidential_remarks":"The central flaw is the false equivalence in Section 1.1 between the L^2 restriction estimate and the convolution estimate (1.2). This is not a minor gap: the entire proof of Theorems 3.1 and 3.2, and consequently the applications in Sections 7–9, rests on that claimed equivalence. The negative-direction results (Theorem 3.6 and Corollary 3.7) and the explicit Fourier-spectrum computations for the cone and moment curve appear sound and could be the basis of a revised paper, but the main positive claims are not established. The referee suggests the authors consider reformulating the main theorem as a bound for the convolution operator \\hat μ * f, if that is what their method genuinely proves, and separately addressing the extension operator with the correct L^2(μ) input. The paper is well written and the spectrum computations are valuable, but as it stands it should not be accepted."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The stress-test note misfires. Equation (3.2) is fine. Young's inequality is applied to the convolution of the function \\hat\\mu_j with f, and \\hat\\mu_j is the correct kernel: the identity \\int|\\hat g|^2 d\\mu = \\langle g, \\hat\\mu * g\\rangle is exactly what makes the convolution formulation equivalent to the extension estimate. No inverse Fourier transform is needed. The circle-measure example in the stress-test computes the L^{2/\\theta} norm of the inverse transform of \\hat\\mu_j, which is a different object and irrelevant to (3.2). Mockenhaupt's own estimate (3.3) uses the same convention, so this part of the proof stands.\n\nThat said, the paper is a genuine contribution. The idea of using the Fourier spectrum to interpolate between the Frostman L2 bound and a family of L^{q'} \\to L^q bounds is new and clean. Theorem 3.1 gives a continuum of Stein\\textendash{}Tomas type estimates; optimizing gives real improvements for the cone and moment curve, and the negative Theorem 3.6 recovers sharp failure thresholds in both examples. The explicit Fourier spectrum computations (cone, Proposition 7.2; moment curve, Proposition 8.1) are the most valuable part; the moment curve spectrum has multiple phase transitions, which is notable. The proofs are structurally sound. The Lorentz space endpoint section is dense, but the interpolation algebra checks out.\n\nSoft spots are minor. The proof leans on Mockenhaupt's (3.3), quoted without derivation; that is a standard cited theorem, but a referee might want a precise statement or a short derivation, since the endpoint thresholds all pass through it. There are a few expository slips and notation-density issues, but nothing load-bearing. The literature list is honest; the self-citations are for definitions and prior spectrum applications, not for the main results.\n\nWho is this for: people working in Fourier restriction for fractal measures, and geometric measure theorists who want the Fourier spectrum of natural examples like the cone and moment curve. It deserves a serious referee. I would accept it for peer review, and if the referee confirms the spectrum computations, it should be published.","headline":"The Fourier-spectrum framework is a real new tool for L2 restriction and the stress-test concern about Young's inequality does not hold up on reading; this deserves serious peer review.","tokens_in":24363,"tokens_out":9380,"would_cite":true,"duration_ms":76834,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["42B10","28A80","42B20","28A75","28A78"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper proves that the range of q for an L2-based Fourier restriction estimate is controlled by the whole Fourier spectrum of the measure, and optimizing over this spectrum improves the restriction range for the cone, the moment…","keywords":["Fourier restriction","Fourier spectrum","Stein-Tomas theorem","Frostman dimension","Sobolev dimension","Lorentz spaces","cone restriction","moment curve"],"falsifier":"Compute the true threshold of the $L^2 \\to L^q$ extension estimate for a measure whose Fourier spectrum is known explicitly, such as the cone in $\\mathbb{R}^5$: the paper predicts the estimate holds for all $q > 11/3$, so finding any $q \\geq 11/3$ where it fails would falsify the main theorem. More fundamentally, directly verifying the underlying dyadic estimate $\\|\\widehat{\\mu}_j * f\\|_{L^2} \\lesssim 2^{j(d-\\alpha)}\\|f\\|_{L^2}$ on an $\\alpha$-Frostman measure would test the load-bearing exponent on which all thresholds depend.","tokens_in":23255,"feed_emoji":"📐","tokens_out":11909,"duration_ms":88792,"temperature":0.7,"pith_summary":"This paper claims that the classical Stein-Tomas restriction theorem for measures on $\\mathbb{R}^d$ can be improved by replacing the single Fourier dimension with the entire Fourier spectrum, a family of dimensions that interpolates between the Fourier and Sobolev dimensions. The main theorem gives a continuum of $L^2 \\to L^q$ extension estimates parameterized by $\\theta \\in [0,1]$, and optimizing over $\\theta$ yields a range of $q$ that often beats the classical range. The same spectral information also gives a strengthened negative result: a range of $q$ for which no restriction estimate can hold, generalizing an earlier lower-bound observation. The authors compute the Fourier spectrum explicitly for the cone and the moment curve, where the new upper bound comes within a small additive constant of the sharp range and the new lower bound recovers the known necessary condition. For fractal measures convolved with a small Salem component, the improvement over the classical range can be large.","feed_headline":"Fourier spectrum widens Stein-Tomas range","feed_subtitle":"A continuum of dimensions replaces one exponent, improving bounds on the cone, moment curve, and fractals.","key_machinery":"The central object is the Fourier spectrum of a measure: for $\\theta \\in (0,1]$, $\\dim_F^\\theta \\mu = \\sup\\{s : J_{s,\\theta}(\\mu)<\\infty\\}$, where $$J_{s,\\$\\theta$}(\\mu)=\\Big(\\int |\\widehat{\\mu}(\\xi)|^{2/\\$\\theta$} |\\xi|^{s/\\$\\theta$-d}\\,d\\xi\\Big)^\\$\\theta$,$$ with the $\\theta=0$ case defined by a sup norm. This one-parameter family interpolates between the Fourier dimension at $\\theta=0$ and the Sobolev dimension at $\\theta=1$, and it is concave and continuous for compactly supported measures. The proof of the main theorem decomposes the measure into dyadic frequency pieces $\\widehat{\\mu}_j$, estimates each piece in $L^2$ using the Frostman condition and in $L^{4/\\theta}$ using the finiteness of $J_{s,\\theta}(\\mu)$, and interpolates between the two bounds. Optimizing the interpolation parameter over the spectrum produces the continuum of thresholds; the endpoint result uses Lorentz spaces and a real-interpolation lemma for sums of dyadic operators.","core_discovery":"The central discovery is that the rate of decay of the Fourier transform of a measure has a whole continuum of useful meanings. For each $\\theta \\in [0,1]$, the Fourier spectrum $\\dim_F^\\theta \\mu$ measures decay in a weighted $L^{2/\\theta}$ sense, with $\\theta=0$ recovering the Fourier dimension and $\\theta=1$ recovering the Sobolev dimension. Theorem 3.1 states that if $\\mu$ is a compactly supported Borel measure on $\\mathbb{R}^d$ with Frostman dimension $\\alpha$, then $\\|\\widehat{f\\mu}\\|_{L^q(\\mathbb{R}^d)} \\lesssim \\|f\\|_{L^2(\\mu)}$ for all $q$ larger than $$2 + 2 \\inf_{\\$\\theta$\\in[0,1],\\, \\dim_F^\\$\\theta$\\mu>d\\$\\theta$} \\frac{(d-\\$\\alpha$)(2-\\$\\theta$)}{\\dim_F^\\$\\theta$\\mu-\\$\\alpha$\\$\\theta$}.$$ The proof interpolates the standard dyadic $L^2$ estimate, whose decay exponent is set by the Frostman dimension, against a new $L^{4/\\theta}$ estimate derived directly from the finiteness of the $(s,\\theta)$-energy $J_{s,\\theta}(\\mu)$; optimizing the interpolation parameter over the spectrum produces the continuum of thresholds. An endpoint version is proved through Lorentz spaces and a real-interpolation trick for sums of dyadic operators. The paper also proves a partial converse: if $J_{d\\theta,\\theta}(\\mu)=\\infty$ for some $\\theta$, then the restriction estimate fails for the constant function at $q=2/\\theta$, giving a negative range in terms of the spectrum that matches the necessary condition for the cone and moment curve.","pith_inferences":["Not stated in the paper: the same interpolation mechanism should also yield $L^p$ restriction estimates for $p\\neq 2$ by interpolating the Fourier-spectrum estimates against the trivial $L^1\\to L^\\infty$ bound, at the cost of a more complicated threshold.","Not stated in the paper: because Theorem 3.6 recovers the sharp necessary condition for the cone and moment curve, the Fourier spectrum may be the right structural object for identifying the true restriction threshold for fractal measures, not just an upper-bound tool.","A testable extension: one could search for measures whose Fourier spectra have more than the $d-2$ phase transitions seen on the moment curve and check whether the restriction threshold changes at each breakpoint.","The explicit cone and moment-curve spectra suggest that the Fourier spectrum of a curved surface is piecewise linear with breakpoints marking where different parts of the surface dominate the Fourier transform; testing other surfaces of revolution could reveal whether the optimal restriction threshold is always achieved at such a phase transition."],"forward_implications":["For any measure satisfying the hypotheses, the new range contains the classical Stein-Tomas range and is sometimes strictly larger: at $\\theta=0$ the formula reduces to the classical range, while optimizing over $\\theta$ can improve it.","For the cone in $\\mathbb{R}^d$ with $d\\geq 5$, the paper obtains the extension estimate for $q > (3d-4)/(d-2)$, beating the Stein-Tomas range $q>4$; for $d=3,4$ it recovers the known sharp ranges.","For the moment curve, the paper obtains the range $q > d^2+d+2$, within 2 of the sharp exponent $d^2+d$, and Theorem 3.6 gives failure for $q < (d^2+d+2)/2$, matching the known necessary condition.","For multifractal Cantor measures convolved with a small Salem component, Theorem 3.1 gives a restriction range far better than Stein-Tomas; for $p=0.6$ and $\\varepsilon=0.067$, the new threshold is $q>7.99$ versus $q>29.95$.","Corollary 6.1 gives a Sobolev-dimension version of the restriction theorem: if $\\dim_S\\mu<d$ and the Fourier dimension is positive, then the extension estimate holds for $q > 4 + 4(d-\\dim_S\\mu)/\\dim_F\\mu$."],"supporting_citations":[{"why":"Supplies the dyadic $L^2$ estimate (3.3) and the Stein-Tomas proof framework whose interpolation the main theorem extends.","marker":"[Moc00]"},{"why":"Provides the Lorentz-space restriction estimates and the auxiliary restricted weak-type estimate used for the endpoint version.","marker":"[BS11]"},{"why":"Introduces the Fourier spectrum and establishes its concavity and continuity, which the optimization over $\\theta$ relies on.","marker":"[Fra24]"},{"why":"Contributes the lower-bound observation generalized by Theorem 3.6 and used for the negative-range comparisons.","marker":"[H/suppress L13]"},{"why":"Supplies the Bessel-function and stationary-phase estimates used to compute the Fourier spectrum of the cone.","marker":"[Mat15]"},{"why":"Gives the average decay estimates for the moment curve that yield its Fourier spectrum in Proposition 8.1.","marker":"[BGGIST07]"},{"why":"Provides the Lorentz-space interpolation theorem used to prove Proposition 4.2 and the endpoint estimate.","marker":"[Gra14]"},{"why":"Gives the lower bound on the Fourier spectrum of a convolution with a Salem measure, used in the fractal examples.","marker":"[Fra24, Theorem 6.1]"},{"why":"Shows that many uniformly non-concentrated measures have maximal initial growth of the Fourier spectrum, used in Section 9.2.","marker":"[Kha23+]"}],"fun_headline_variants":["Continuum of dimensions sharpens restriction bounds","Fourier spectrum redefines Stein-Tomas theorem","New restriction theorem from spectral interpolation","One spectrum, many exponents: improving Stein-Tomas"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole argument rests on a known estimate that says the high-frequency pieces of a measure decay at a very specific rate set by its Frostman dimension; if that rate were even slightly different, every threshold in the paper would move.","fun_headline_variants_meta":{"raw":{"variants":["Continuum of dimensions sharpens restriction bounds","Fourier spectrum redefines Stein-Tomas theorem","New restriction theorem from spectral interpolation","One spectrum, many exponents: improving Stein-Tomas"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000204,"raw_usage":{"total_tokens":1453,"prompt_tokens":1073,"completion_tokens":380,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":689,"completion_tokens_details":{"reasoning_tokens":330}},"tokens_in":689,"tokens_out":380,"duration_ms":3262,"temperature":1.0,"reasoning_tokens":330,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T11:50:14.901283+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the true threshold of the $L^2 \\to L^q$ extension estimate for a measure whose Fourier spectrum is known explicitly, such as the cone in $\\mathbb{R}^5$: the paper predicts the estimate holds for all $q > 11/3$, so finding any $q \\geq 11/3$ where it fails would falsify the main theorem. More fundamentally, directly verifying the underlying dyadic estimate $\\|\\widehat{\\mu}_j * f\\|_{L^2} \\lesssim 2^{j(d-\\alpha)}\\|f\\|_{L^2}$ on an $\\alpha$-Frostman measure would test the load-bearing exponent on which all thresholds depend.","supporting_citations":[],"review_version":1}