{"id":"80d5917e-c38f-475f-9e83-5bf71d0c059f","arxiv_id":"2411.19562","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For any compact set Ω and any ε>0, there is a discrete set Λ of density at most (1+ε)|Ω| such that the exponential system on Λ is a frame for L2(Ω) with frame bounds depending only on ε.","lead":"The authors construct families of exponential functions that can reconstruct any signal on a given finite frequency set while using only slightly more samples than the theoretical minimum. This settles a question posed in 2013 and extends the construction to commutative group settings.","discovery_kind":"first_principles","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 2.4 applies Theorem 2.3 with weights s'_i that may vanish, so the support-count bound #{i:s_i≠0}≤⌈dn⌉ is not justified as written; the gap is load-bearing for the density estimate in Theorem 3.1 but likely fixable by restricting to the support of s'.","rationale":"The reader's weakest-assumption analysis correctly identifies the main soft spot: the proof of Theorem 2.4 applies Theorem 2.3 to weights that may be zero, and the support-control of the resulting sampling function is not stated. This is genuinely load-bearing for the cardinality bound and hence for the critical density claim, but it is a presentation gap rather than a mathematical error: restricting to the positive-weight support trivially repairs the argument without changing the bounds. I verified the surrounding steps of Theorem 3.1 and Theorem 4.5; the frame-bounds derivation, the density computation modulo the index calculation, and the lifting arguments are internally consistent. The LCA extension relies on substantial external results, notably the unpublished preprint [6], which adds verification risk but is not an internal inconsistency. The typo in the displayed formula for B(ε) in the proof of Theorem 4.4 (the minus sign in (1−ε/4)) is clearly a typo because the preceding text and the closing bound use the plus sign. Overall, the central claim is credible and the identified gap is fixable, so the conditional verdict is appropriate.","tokens_in":22222,"tokens_out":24194,"duration_ms":176267,"concrete_test":"Write out the proof of Theorem 2.4 with the index set I_+ = {i : s'_i > 0} in place of I for the application of Theorem 2.3, taking ai = s'_i for i ∈ I_+. Verify that the resulting sampling function π satisfies π(I') ⊆ I_+, that the conclusions (ii) and (iii) remain unchanged, and that #{i : s_i ≠ 0} ≤ |I_+| ≤ ⌈dn⌉. If this verification succeeds, the support-control gap is closed and the density estimate D(Λ) ≤ (1+ε)|Ω| in Theorem 3.1 is fully justified.","verdict_should_be":"UNCHANGED","load_bearing_attack":"In the proof of Theorem 2.4, Theorem 2.3 is applied with ai = s'_i. Theorem 2.3 explicitly assumes a sequence of positive numbers ai, but the s'_i produced by Theorem 2.1 can be zero. Moreover, Theorem 2.3 does not state that the sampling function π has image contained in the support of the weights. Consequently, as written, the proof does not rule out π(I') hitting indices with s'_i = 0, which would create nonzero s_i outside the original support and could break condition (i), #{i : s_i ≠ 0} ≤ ⌈dn⌉. This cardinality bound is exactly what controls D(Λ) ≤ (1+ε)|Ω| in Theorem 3.1 (via (3.7)). The gap is, however, easily closed: apply Theorem 2.3 only to I_+ = {i : s'_i > 0}. Since terms with s'_i = 0 do not contribute to T = Σ s'_i v_i v_i^*, the hypotheses hold on I_+ with positive weights, and the sampling function then has image in I_+, so support(s) ⊆ I_+ and #{i : s_i ≠ 0} ≤ |I_+| ≤ ⌈dn⌉. The same argument yields the LCA version, since Theorem 4.4 inherits the density bound from (4.14). A secondary, non-fatal slip occurs in Theorem 4.4's density computation: [H0 : Hm] is 2^{m(d+ℓ)}#F, not 2^{m(d+ℓ)}, but the final inequality DH(T) ≤ (1+ε)μ(Ω) still follows from (4.14) and the equality DH(T) = DH0(T0).","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proves that, for every compact set Omega in R of measure |Omega| and every epsilon>0, there exists a set Lambda of uniform density at most (1+epsilon)|Omega| such that the exponentials {e_lambda 1_Omega} form a frame for L^2(Omega) with frame bounds A(epsilon)|Omega| and B(epsilon)|Omega|, where A and B depend only on epsilon. This answers Open Problem 1 of Nitzan, Olevskii, and Ulanovskii. The proof combines the real-line sampling construction of those authors with a finite-dimensional sparsification theorem. The finite-dimensional part is new: the authors revisit the Batson-Spielman-Srivastava sparsification, use a result of the first author to control the nonzero weights from below, and derive an unweighted frame with cardinality close to the ambient dimension and explicit bounds. The paper also proves an LCA-group analogue, Theorem 1.2, with a reduction through elemental groups, a lifting lemma for frames, and Beurling/Leptin densities.","tokens_in":22597,"tokens_out":7074,"duration_ms":59759,"significance":"If the proof is completed as indicated below, the paper solves a recognized open problem: it gives exponential frames whose density is arbitrarily close to the critical density and whose frame bounds are proportional to the measure of the spectrum with constants independent of the spectrum. The real-line proof is well organized, and the finite-dimensional Lemmas 2.5 and 2.6 are clean and likely to be useful independently. The LCA extension is nontrivial and improves the earlier result of Agora, Antezana, and Cabrelli by adding quantitative frame bounds. The paper is explicit about the constants in Theorem 3.1 and gives a transparent route from finite-dimensional estimates to the continuous statements. The main caveat is the support-control issue in the proof of Theorem 2.4, which is load-bearing for the density estimate but admits a straightforward repair.","major_comments":[{"comment":"Theorem 2.3 is applied with a_i = s'_i, but Theorem 2.3 is stated only for positive numbers a_i, and it does not assert that the sampling function pi takes values in the support of the sequence (a_i). Since some s'_i produced by Theorem 2.1 may be zero, the proof as written does not rule out pi(I') hitting indices with s'_i = 0. If that happens, the resulting coefficients s_i can be non-zero at indices outside {i : s'_i != 0}, and the bound #{i : s_i != 0} <= ceil(dn) in condition (i) could fail. This cardinality bound is exactly what controls D(Lambda) <= (1+epsilon)|Omega| in Theorem 3.1 through (3.7). The gap is fixable: apply Theorem 2.3 to I_+ = {i : s'_i > 0}, since the zero-weight terms do not contribute to T = sum s'_i v_i v_i^*, and then the sampling function has image in I_+, so support(s) is contained in I_+ and #{i : s_i != 0} <= |I_+| <= ceil(dn). Please add this argument explicitly, both for the real-line case and for the LCA case that inherits the density bound from Theorem 4.5.","section":"§2, proof of Theorem 2.4"}],"minor_comments":[{"comment":"The index [H0 : Hm] is 2^{m(d+ell)} #F, not 2^{m(d+ell)}. Consequently the displayed equalities DH0(Hm) = 2^{-m(d+ell)} and DH(T) = q 2^{-m(d+ell)} should carry an extra factor (#F)^{-1}. The final estimate DH(T) <= (1+epsilon) mu(Omega) is unaffected because only the equality DH(T) = DH0(T0) is used, but the numerical claims as written are incorrect.","section":"§4.5, proof of Theorem 4.4"},{"comment":"In the definition of the matrix F, the entries should be indexed as e_{h_j}(lambda_i) rather than e_{h_i}(lambda_i); the later use of F_I(J) with entries e_{h_j}(lambda_i) confirms this is a typographical indexing slip.","section":"§4.3, proof of Theorem 4.5"},{"comment":"The notation c(epsilon) is introduced as c (1 - 1/sqrt(1+epsilon))^2, but the subsequent substitutions use c(epsilon/4); please clarify that the displayed definition is intended with the same epsilon/4 argument, or change the notation to avoid confusion.","section":"§3, proof of Theorem 3.1"},{"comment":"Theorem 2.3 is imported from the first author's preprint [6] and is load-bearing for the main theorem. If [6] is not yet published, please include either a self-contained proof of Theorem 2.3 or a precise pointer to a version of record, so that the dependence is verifiable by the reader.","section":"§2, Theorem 2.3"},{"comment":"The final inequality in Corollary 3.2 says 'for all in PW_Omega'; it should say 'for all f in PW_Omega'.","section":"§3, Corollary 3.2"}],"recommendation":"major_revision","confidential_remarks":"The reliance on the unpublished preprint [6] is worth monitoring. The support-control gap in Theorem 2.4 is a genuine technical issue but appears to be repairable in a few lines, so I do not regard it as fatal. The paper's contribution is substantial and does not reduce to any circular use of the cited results."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"What you should know: this paper solves the Nitzan–Olevskii–Ulanovskii open problem on exponential frames near critical density. The main theorem gives, for any compact Ω⊂R and any ε>0, a set Λ of uniform density ≤(1+ε)|Ω| so that the exponential system is a frame for L²(Ω) with frame bounds A(ε)|Ω| and B(ε)|Ω|. That is exactly the combination of near-critical density and controlled bounds that was missing. The LCA extension (Theorem 1.2) is a real additional result, not a corollary.\n\nCredit where earned: the finite-dimensional sparsification lemmas (Lemma 2.5 and 2.6) are carefully shown and check out. Combining Batson–Spielman–Srivastava with the selector theorem to force lower bounds on the weights is a genuinely useful device. The real-line proof in Section 3 is detailed, and the density computation via sampling of the discrete Fourier matrix is sound.\n\nWhere the soft spots are. The proof of Theorem 2.4 applies Theorem 2.3 with weights a_i = s'_i, but those weights can be zero and Theorem 2.3 assumes positive weights. Worse, nothing forces the sampling function π to avoid indices with s'_i = 0, so terms could appear in the final sum that were not in the target T. That would break condition (i), #{i : s_i ≠ 0} ≤ ⌈dn⌉, which is load-bearing for the density estimate D(Λ) ≤ (1+ε)|Ω| in Theorem 3.1. This is a genuine gap as written. It is, however, easily fixed: apply Theorem 2.3 to I_+ = {i : s'_i > 0}; the zero-weight terms do not contribute to T, the weights are positive on I_+, and the sampling function then has image in I_+, so the count follows. I would want a referee to require this correction before publication.\n\nSeparately, in Theorem 4.4's proof, the index [H0 : Hm] is stated as 2^{m(d+ℓ)} but it is actually 2^{m(d+ℓ)}#F. The density inequality still follows from (4.14), so this is a typo, but it should be fixed.\n\nThe use of Bownik's unpublished preprint [6] is acceptable; it is an independent result used as a black box. I see no circularity and no manufactured evidence.\n\nBottom line: the core result is significant and the proof is fundamentally sound. The support-control gap is a patch, not a hole. I would send this to a serious referee and would expect it to be accepted after a modest revision.","headline":"Strong and important paper that solves a known open problem; one fixable gap in the key sparsification lemma and a minor density typo.","tokens_in":23127,"tokens_out":6789,"would_cite":true,"duration_ms":50753,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["42A65","42C30","43A70"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper constructs exponential frames for L²(Ω) with density (1+ε)|Ω| and ε-only frame bounds.","keywords":["exponential frames","critical density","frame bounds","uniform density","Paley-Wiener spaces","locally compact abelian groups","Parseval frames","frame sparsification"],"falsifier":"Consider a Parseval frame containing two identical vectors. The sparsification step may assign a zero weight to one copy; if the selector invoked in the proof of Theorem 2.4 can then sample that zero-weight copy, the number of nonzero coefficients need not stay below $\\lceil d n\\rceil$, and the density estimate in Theorem 3.1 loses its justification. Exhibiting such a sampling event, or proving that the selector never leaves the support of the initial weights, would settle whether the construction stands as written.","tokens_in":22012,"feed_emoji":"📐","tokens_out":17714,"duration_ms":130702,"temperature":0.7,"pith_summary":"For every $\\varepsilon>0$ and every compact set $\\Omega\\subseteq\\mathbb{R}$, the paper proves that a discrete set $\\Lambda$ exists with uniform density $D(\\Lambda)\\le(1+\\varepsilon)|\\Omega|$ such that the system $\\{e^{2\\pi i\\lambda t}\\mathbf{1}_\\Omega\\}_{\\lambda\\in\\Lambda}$ is a frame for $L^2(\\Omega)$ with bounds $A(\\varepsilon)|\\Omega|$ and $B(\\varepsilon)|\\Omega|$, where the constants depend only on $\\varepsilon$ and not on $\\Omega$. This is the first construction to achieve near-critical density and spectrum-scale frame bounds simultaneously, answering an open problem in the sampling literature. The same statement holds for exponential frames on second countable locally compact abelian groups, with bounds given by the dual measure of the spectrum. A reader should care because it fixes the exact trade-off between sampling density and stable reconstruction guarantees for bandlimited signals.","feed_headline":"Exponential frames exist arbitrarily close to critical density","feed_subtitle":"For every compact spectrum and ε>0, sampling at density (1+ε)|Ω| gives frame bounds that depend only on ε.","key_machinery":"The load-bearing mechanism is the controlled sparsification of finite Parseval frames, packaged as Lemma 2.6. Start with an $m\\times n$ matrix $M$ whose rows are equal-norm vectors forming a Parseval frame with $\\|v_i\\|^2=n/m$. The lemma outputs a row subset $J$ with $\\#J\\le\\lceil(1+\\varepsilon)n\\rceil$ such that the submatrix $M(J)$ satisfies explicit two-sided frame estimates with constants scaling like $(1-1/\\sqrt{1+\\varepsilon})^2$ and $(1-1/\\sqrt{1+\\varepsilon})^{-4}$. It is proved by combining a known sparsification result that produces weighted near-minimal-cardinality frames with a selector result that bounds the size of the nonzero weights from below, which in turn forces an unweighted frame with the same cardinality and controlled bounds. In the real-variable proof, the matrix is the $m\\times n$ submatrix of the $m\\times m$ discrete Fourier matrix whose columns correspond to the grid cells covering $\\Omega$; the selected rows index $\\Lambda = \\bigcup_{j\\in J}(j+m\\mathbb{Z})$, and the fact that $\\{e_{ml}\\}_{l\\in\\mathbb{Z}}$ is an orthogonal basis of $L^2(0,1/m)$ converts the matrix inequalities into frame inequalities on $L^2(\\Omega)$. The density identity $D(\\Lambda)=\\#J\\cdot D(m\\mathbb{Z})$ then gives the near-critical bound. For locally compact abelian groups the same lemma is applied to the character matrix of a finite quotient, and a lifting property for frames on compact subgroups carries the result from elemental quotients to compactly generated dual groups and then to the general case.","core_discovery":"The paper's central claim is Theorem 1.1: for every $\\varepsilon>0$ and compact $\\Omega\\subseteq\\mathbb{R}$, there exists $\\Lambda\\subseteq\\mathbb{R}$ of uniform density $D(\\Lambda)\\le(1+\\varepsilon)|\\Omega|$ such that\n$$A(\\varepsilon)|\\$\\Omega$|\\,\\|f\\|^2 \\le \\sum_{\\$\\lambda$\\in\\Lambda}|\\langle f,e_\\$\\lambda$\\rangle|^2 \\le B(\\varepsilon)|\\$\\Omega$|\\,\\|f\\|^2$$\nfor all $f\\in L^2(\\Omega)$, with $e_\\lambda(t)=e^{2\\pi i\\lambda t}$ and constants $A(\\varepsilon),B(\\varepsilon)$ depending only on $\\varepsilon$. If $\\Omega$ lies in an interval of length $d$, the frame can be chosen inside the lattice $d^{-1}\\mathbb{Z}$. The engine is a finite-dimensional statement: any $m\\times n$ matrix that is a submatrix of an orthonormal matrix and has equal row norms contains a selection of at most $\\lceil(1+\\varepsilon)n\\rceil$ rows that forms a frame for $\\mathbb{C}^n$ with lower constant of order $(1-1/\\sqrt{1+\\varepsilon})^2$ and upper constant of order $(1-1/\\sqrt{1+\\varepsilon})^{-4}$. Applying this selection to the discrete Fourier matrix on a fine grid covering $\\Omega$ yields an exponential frame on a union of lattice cosets; the orthogonal basis of each grid cell converts the finite-dimensional inequalities into the desired integral inequalities over $\\Omega$. The same finite block is used on elemental groups, then lifted to compactly generated dual groups via a frame lifting property for compact subgroups, and finally extended to all second countable locally compact abelian groups through the open subgroup generated by the spectrum.","pith_inferences":["Inference: the support-control gap in Theorem 2.4 is likely patchable by restricting the selector's sampling function to the support of the initial weights; if so, the main theorem and its constants would survive unchanged.","Inference: the same finite sparsification block could plausibly be applied to higher-dimensional spectra by using multidimensional discrete Fourier matrices, yielding near-critical exponential frames with explicit bounds for box-like spectra in $\\mathbb{R}^d$.","Inference: the explicit $\\varepsilon$-dependence suggests a concrete algorithm—choose the grid scale so that $\\lceil(1+\\varepsilon/4)n\\rceil\\le(1+\\varepsilon)n$, solve the small discrete sparsification problem, then lift to the periodic exponential frame—so the proof can be turned into a finite computation once the absolute constants are made explicit.","Inference: the LCA route through the open subgroup generated by the spectrum suggests that only the geometry of the group near the spectrum matters, so a similar statement may hold in amenable nonabelian settings once the corresponding density and lifting machinery exists."],"forward_implications":["Every compact spectrum $\\Omega\\subseteq\\mathbb{R}$ admits a sampling set with density within a factor $1+\\varepsilon$ of the critical density $|\\Omega|$, with sampling constants depending only on $\\varepsilon$; the density/frame-bound trade-off is therefore universal rather than spectrum-dependent.","When $\\Omega$ is contained in an interval of length $d$, the frame can be chosen periodic, $\\Lambda\\subseteq d^{-1}\\mathbb{Z}$, so the construction yields lattice-based sampling sets with a regular underlying structure.","Weak-limit arguments extend the frame conclusion to unbounded spectra, although that passage does not preserve the near-critical-density property, as the paper notes.","On every second countable locally compact abelian group, the same near-critical-density frame exists with bounds $A(\\varepsilon)\\mu_{\\widehat{G}}(\\Omega)$ and $B(\\varepsilon)\\mu_{\\widehat{G}}(\\Omega)$, improving prior LCA constructions by making the frame bounds depend on the spectrum.","The explicit constants $A(\\varepsilon)=c(1-1/\\sqrt{1+\\varepsilon/4})^2$ and $B(\\varepsilon)=c'(1-1/\\sqrt{1+\\varepsilon/4})^{-4}$ show that the frame bounds degrade in a controlled power-law way as $\\varepsilon\\to0$."],"supporting_citations":[{"why":"It establishes the necessary density condition that any exponential frame for $L^2(\\Omega)$ must have lower density at least $|\\Omega|$, setting the critical-density benchmark the construction targets.","marker":"[14]"},{"why":"It states the open problem of combining near-critical density with spectrum-scale frame bounds and supplies the lower-bound construction the paper adapts.","marker":"[17]"},{"why":"It supplies the finite sparsification theorem for Parseval frames that yields weighted frames with cardinality close to the ambient dimension.","marker":"[5]"},{"why":"It supplies the selector result for scalable frames that gives the lower control on nonzero weights used to turn weighted frames into unweighted ones.","marker":"[6]"},{"why":"It provides the necessary density conditions and the lifting framework that carry the argument to locally compact abelian groups.","marker":"[12]"},{"why":"It gives prior near-critical-density frames on LCA groups and introduces the quasi-dyadic cube technique that the group proof reuses.","marker":"[1]"},{"why":"It shows how weak-limit methods extend such frames to unbounded spectra, a separate route that does not preserve near-critical density.","marker":"[18]"}],"fun_headline_variants":["Exponential frames at density (1+ε)|Ω|","Frame bounds independent of spectrum size","Near-critical density frames with ε-only bounds","Exponential frames: density (1+ε)|Ω| works"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"In the proof of Theorem 2.4, the load-bearing premise is that a certain selection step—stated for strictly positive weights—still keeps the number of selected entries below $\\lceil d n\\rceil$ when some of the weights are zero; the paper does not justify this support-control property, and the near-critical density inequality $D(\\Lambda)\\le(1+\\varepsilon)|\\Omega|$ rests on it.","fun_headline_variants_meta":{"raw":{"variants":["Exponential frames at density (1+ε)|Ω|","Frame bounds independent of spectrum size","Near-critical density frames with ε-only bounds","Exponential frames: density (1+ε)|Ω| works"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000436,"raw_usage":{"total_tokens":2288,"prompt_tokens":1086,"completion_tokens":1202,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":702,"completion_tokens_details":{"reasoning_tokens":1140}},"tokens_in":702,"tokens_out":1202,"duration_ms":10514,"temperature":1.0,"reasoning_tokens":1140,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T10:04:28.672656+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Consider a Parseval frame containing two identical vectors. The sparsification step may assign a zero weight to one copy; if the selector invoked in the proof of Theorem 2.4 can then sample that zero-weight copy, the number of nonzero coefficients need not stay below $\\lceil d n\\rceil$, and the density estimate in Theorem 3.1 loses its justification. Exhibiting such a sampling event, or proving that the selector never leaves the support of the initial weights, would settle whether the construction stands as written.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"It establishes the necessary density condition that any exponential frame for $L^2(\\Omega)$ must have lower density at least $|\\Omega|$, setting the critical-density benchmark the construction targets."},{"cited_title":"Nitzan, A","cited_arxiv_id":null,"evidence_quote":"It states the open problem of combining near-critical density with spectrum-scale frame bounds and supplies the lower-bound construction the paper adapts."},{"cited_title":"Batson, D","cited_arxiv_id":null,"evidence_quote":"It supplies the finite sparsification theorem for Parseval frames that yields weighted frames with cardinality close to the ambient dimension."},{"cited_title":"Gr¨ ochenig, G","cited_arxiv_id":null,"evidence_quote":"It provides the necessary density conditions and the lifting framework that carry the argument to locally compact abelian groups."},{"cited_title":"Agora, J","cited_arxiv_id":null,"evidence_quote":"It gives prior near-critical-density frames on LCA groups and introduces the quasi-dyadic cube technique that the group proof reuses."},{"cited_title":"Nitzan, A","cited_arxiv_id":null,"evidence_quote":"It shows how weak-limit methods extend such frames to unbounded spectra, a separate route that does not preserve near-critical density."}],"review_version":1}