{"id":"08a3f147-adb9-4deb-b079-e991067a7e41","arxiv_id":"1908.06737","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"The spacing of paraorthogonal polynomial zeros is bounded below by a measure's Hausdorff continuity, and sparse Verblunsky coefficients can force clock spacing even for singular continuous measures.","lead":"This mathematics paper proves that the more continuous a measure is, in a precise Hausdorff-dimension sense, the more the zeros of its associated paraorthogonal polynomials repel each other; it also constructs singular continuous measures that still show perfect clock-like repulsion. A generalist might read it to see a clean bridge between measure regularity and eigenvalue-like spacing statistics.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The proof of Theorem 1.1(2) uses δ = 1/γ − ε, but the displayed ratio bound then has positive exponent and diverges; replacing minus by plus makes the subordinacy condition hold and the stated dimension bound follow.","rationale":"The central claims of the paper are plausible and largely well-supported. Theorem 2.3 and the sparse-coefficient construction in Theorem 3.1 are carefully argued, and Theorem 1.2's mechanism via sine-kernel asymptotics is standard and convincing. My main concern is localized to the proof of Theorem 1.1(2), where the exponent in the subordinacy ratio is algebraically wrong: with the printed δ = 1/γ − ε, the displayed bound gives n^{(1−δγ)/2+η} = n^{εγ/2+η}, which tends to infinity, not zero. This is not merely a typo in the final conclusion; it is the step that connects the zero-spacing assumption to the Hausdorff dimension bound. Fortunately, the argument is repairable by replacing the minus with a plus, and then using a countable intersection of the dimension-support sets to pass to the infimum. The reader's weakest assumption concerned the hypotheses of [26, Theorem 4.3.16] and the subordinacy theorems. For the first, the needed bound is actually a consequence of ∫ K_n(z,z)dµ = n and Borel–Cantelli, so no hidden regularity assumption is needed; the subordinacy theorems in [26, Chapter 10] are likewise stated for arbitrary probability measures. Therefore the reader's identified vulnerability is less serious than the δ-sign error, which is why I mark agreement as partial rather than full. Overall, the proof needs a correction but the theorem is likely valid, so the existing CONDITIONAL verdict should stand unchanged.","tokens_in":19802,"tokens_out":25334,"duration_ms":266538,"concrete_test":"Re-derive the exponent in the proof of Theorem 1.1(2). With δ = 1/γ − ε, substitute the two bounds and evaluate (1 − δγ)/2 + η; the result is εγ/2 + η > 0, so the displayed 'which again converges to zero' statement is false. Then repeat with δ = 1/γ + ε: choosing η < εγ/2 gives liminf ||φ·||_n / ||ψ·||_n^δ = 0 along the subsequence where ||ψ||_n² ≥ c n^γ, and inf_{ε>0} 2δ/(1+δ) = 2/(1+γ) by taking a countable intersection of the support sets. This algebraic check settles whether the proof of the dimension bound in Theorem 1.1(2) needs only a sign change or a deeper revision.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Part 2 of Theorem 1.1 is not established as written because of a sign error in the choice of δ. The proof reaches the estimates ||φ·(z)||_n ≤ C n^{1/2+η} and limsup n^{-γ} ||ψ·(z)||_n² > 0. For δ = 1/γ − ε with 0 < ε < γ, these bounds give, along the subsequence where ||ψ||_n² ≥ c n^γ,\n\n||φ·(z)||_n / ||ψ·(z)||_n^δ ≤ C' n^{1/2+η − δγ/2} = C' n^{εγ/2 + η},\n\nwhich diverges as n → ∞ rather than converging to 0. Thus the claimed liminf condition liminf ||φ||_n / ||ψ||_n^δ = 0, needed for [26, Theorems 10.8.5 and 10.8.7], is not obtained. Replacing the minus by a plus, δ = 1/γ + ε, makes the exponent η − εγ/2, which is negative for η < εγ/2, so the ratio tends to 0 along the subsequence. The final dimension bound then follows by taking δ_k ↓ 1/γ and intersecting the corresponding support sets, so the theorem is likely true, but the printed proof requires this correction. The reader's concern about hypotheses of the cited external results is less serious: the bound ||φ·(z)||_n² ≤ n^{1+2η} for µ-a.e. z follows directly from ∫ K_n(z,z) dµ = n and Borel–Cantelli, so Theorem 4.3.16 applies to all probability measures without extra regularity assumptions.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proves two theorems about the spacing of zeros of paraorthogonal polynomials on the unit circle. Theorem 1.1 shows that if a probability measure mu gives zero weight to sets of Hausdorff dimension at most alpha, then for mu-almost every Theta the rescaled zero gap n^gamma (theta_0^(n)(Theta)-theta_{-1}^(n)(Theta)) tends to infinity for gamma = 2/alpha - 1; more generally, the set where liminf of this gap is finite is supported on a set of Hausdorff dimension at most 2/(1+gamma). Theorem 1.2 constructs purely singular continuous measures for which clock spacing n(theta_{j+1}^(n)(Theta)-theta_j^(n)(Theta)) tends to 2 pi at every point Theta, for every sequence of paraorthogonal polynomials. The proofs combine a transfer-matrix lower bound on gaps (Theorem 2.3), OPUC subordinacy theory, and a sparse Verblunsky-coefficient construction for which the Christoffel-Darboux kernel has sine-kernel asymptotics.","tokens_in":20142,"tokens_out":36914,"duration_ms":328516,"significance":"If the results are correct, they give a quantitative connection between measure continuity in the Hausdorff-dimension sense and zero repulsion for POPUC, which appears to be new, and they show that pure singularity of the measure does not preclude maximal repulsion. The transfer-matrix bound in Theorem 2.3 is a clean and useful OPUC analog of the Last-Simon bound, and the appendix provides a missing OPUC version of the Freud-Levin-Lubinsky theorem. The proofs are structured, use standard tools, and the main external inputs are cited precisely. I found no load-bearing error; the concern about a sign error in the choice of delta does not survive a careful reading of the formula as delta = 1/(gamma-epsilon).","major_comments":[],"minor_comments":[{"comment":"Please typeset the parameter in the proof of Theorem 1.1(2) as delta = 1/(gamma-epsilon). If a reader parses it as (1/gamma)-epsilon, the exponent (1-delta gamma)/2 is positive and the displayed liminf would not be zero; the dimension computation 2 delta/(1+delta) = 2/(1+gamma-epsilon) confirms the intended reading, but the ambiguity should be removed.","section":"Section 2"},{"comment":"In the display after (2.9), the vectors in the sum should be w'_{m-1} and w'_m (with primes), matching the statement in (2.7); the current display appears to omit the primes.","section":"Lemma 2.1"},{"comment":"The factor obtained from (3.2) should be (1+|v_{\\ell+1}|)/(1-|v_{\\ell+1}|), not (1+|v_{\\ell+1}|^2)/(1-|v_{\\ell+1}|^2). The error does not affect convergence because both factors tend to 1 as v_{\\ell+1}->0, but the displayed inequality is incorrect as written.","section":"Section 4.1"},{"comment":"To conclude Theorem 1.2 from [26, Theorem 12.5.2], the recursive sparseness condition should explicitly include N_{\\ell+1}/N_\\ell -> infinity, and the sequence v_\\ell should be chosen with sum |v_\\ell|^2 = infinity. These are compatible with the construction but should be stated.","section":"Section 4"},{"comment":"The proof of Theorem 1.1(2) relies on [26, Theorem 4.3.16] and [26, Theorems 10.8.5 and 10.8.7]; please state the hypotheses of these theorems or indicate explicitly that they apply to arbitrary probability measures on the unit circle, so that the application to purely singular continuous measures is transparent.","section":"Section 2"},{"comment":"The name 'Hurowitz' in the proof of Theorem 5.1 should be 'Hurwitz'.","section":"Appendix"}],"recommendation":"minor_revision","confidential_remarks":"The paper is within the scope of the journal and the central results appear correct. The requested changes are expository and local; I do not have concerns about novelty or citation practice."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's my read. The paper is a clean transfer of the OPRL machinery to OPUC. The genuinely new pieces are Theorem 1.1(2) — the first Hausdorff-dimension-based lower bound on POPUC zero spacing — and the construction in Theorem 1.2/Theorem 3.1, which gives singular continuous measures on the circle with universal clock behavior. The proof of the Last–Simon type bound (Theorem 2.3) is new and is a nice simplification of the existing strategy; the sparse-Verblunsky construction also avoids the CD formula, which is a genuine simplification over [2].\n\nI looked at the alleged sign error in the proof of Theorem 1.1(2). It does not hold up: the printed 'δ = 1/γ − ε' is ambiguous, but the displayed identity 2δ/(1+δ) = 2/(1+γ−ε) forces δ = 1/(γ−ε), which makes the exponent (1−δγ)/2 + η negative and the liminf ratio 0. So the subordinacy step goes through as written, modulo a typo in the typesetting of δ. Same for the Section 4.1 constant: the bound (3.2) gives |φ^{(ℓ)}|² ≤ ((1+|v|)/(1−|v|)) |φ^{(ℓ+1)}|², not the squared version displayed. That is a real typo, but harmless because the prefactor B_{ℓ+1} → 0 and the missing factor is bounded for large ℓ. Both should be corrected, but neither is load-bearing.\n\nThe reliance on external results is fair: [26, Thm 4.3.16] and Thms 10.8.5/10.8.7 are standard and apply to arbitrary probability measures on the circle; the growth bound also follows elementarily from the CD kernel trace, so I don't worry about hidden hypotheses. The citation pattern is honest, including the debt to [2] and the real-line results.\n\nMinor quibbles: the 'sufficiently sparse' condition in Theorem 3.1 is deliberately vague, though the recursive construction makes it clear; the remark about the unit-circle analog of Zlatoš' result is explicitly unsupported, which is fine. The paper does what it claims.\n\nBottom line: this deserves a serious referee. I'd recommend acceptance after minor revisions. It will be cited by people working on zero spacing of paraorthogonal polynomials and on universality for singular measures.","headline":"Solid OPUC transfer of OPRL zero-spacing results; the claimed sign error is a misreading, but two minor typos need fixing.","tokens_in":20705,"tokens_out":6208,"would_cite":true,"duration_ms":55828,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["42C05","28A78","47B36"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper proves that the Hausdorff dimension of a measure on the unit circle controls how fast the zeros of its paraorthogonal polynomials repel each other, and that purely singular continuous measures can still exhibit exact clock…","keywords":["paraorthogonal polynomials","unit circle","Hausdorff dimension","zero spacing","clock behavior","Christoffel-Darboux kernel","Verblunsky coefficients","transfer matrices"],"falsifier":"Construct a measure μ that is supported on a self-similar Cantor set of Hausdorff dimension α and gives zero weight to all sets of dimension at most α, compute the zeros of the paraorthogonal polynomials for a fixed β, and check whether $n^{{2/α−1}}$($θ_0^{{(n)}}$(Θ)−θ_{-1}^{(n)}(Θ)) diverges to infinity for μ-almost every Θ; if a point of positive μ-measure has a sequence of n with finite liminf of this quantity, Theorem 1.1(2) fails.","tokens_in":19567,"feed_emoji":"⭕","tokens_out":10203,"duration_ms":89653,"temperature":0.7,"pith_summary":"This paper establishes a link between how continuous a measure on the unit circle is, measured by Hausdorff dimension, and how strongly the zeros of the associated paraorthogonal polynomials repel each other. The main theorem says that if the measure gives zero weight to all sets of Hausdorff dimension at most α, then for almost every point on the circle the rescaled gap between the two zeros flanking that point, multiplied by $n^{{2/α−1}}$, tends to infinity. A second theorem shows the boundary is subtle: there exist purely singular continuous measures for which every choice of paraorthogonal polynomials still has exact clock spacing, with adjacent zeros separated by 2π/n. If true, these results make the smoothness of the measure a quantitative driver of zero repulsion, while showing that singularity alone cannot force irregular spacing.","feed_headline":"Hausdorff dimension sets a floor on zero repulsion","feed_subtitle":"Smoother measures force larger zero gaps; singular continuous measures can still have exact clock spacing.","key_machinery":"The load-bearing identity is the transfer-matrix gap lower bound (Theorem 2.3): for two consecutive zeros z'=$e^{{iθ_{-1}}$^{(n)}} and z''=$e^{{iθ_0^{(n)}}$} flanking z=$e^{{iΘ}}$, |$θ_0^{{(n)}}$−θ_{-1}^{(n)}| ≥ (Σ_{k=0}^{n−1}||T_k(z)||²)^{-1}, where T_k(z) are the 2×2 transfer matrices from the Szegő recurrence. This converts the question of zero spacing into a question about growth of transfer-matrix norms, to which subordinacy theory applies. For the construction, the central object is a sequence of Verblunsky coefficients that is sparse, with non-zero entries separated by rapidly growing gaps, and decaying to zero; the proof shows the Christoffel–Darboux kernel of the resulting measure converges to the sine kernel uniformly in the angle, and the appendix's unit-circle Freud–Levin–Lubinsky theorem converts that convergence into clock spacing.","core_discovery":"The central discovery is a quantitative lower bound on local zero spacing for paraorthogonal polynomials on the unit circle, controlled by the Hausdorff dimension of the underlying measure. Concretely, for γ>1 the set of points where liminf n^γ($θ_0^{{(n)}}$(Θ)−θ_{-1}^{(n)}(Θ))<∞ has μ-measure supported on a set of Hausdorff dimension at most 2/(1+γ). Therefore if μ gives zero weight to sets of dimension at most α, then for μ-almost every Θ the gaps satisfy $n^{{2/α−1}}$($θ_0^{{(n)}}$(Θ)−θ_{-1}^{(n)}(Θ))→∞. The paper also constructs purely singular continuous measures whose Verblunsky coefficients are sparse and decaying, and proves that their Christoffel–Darboux kernels have sine-kernel asymptotics uniformly in the angle, which forces n(θ_{j+1}^{(n)}(Θ)−$θ_j^{{(n)}}$(Θ))→2π for every Θ and every j.","pith_inferences":["The paper leaves open whether the dimension bound 2/(1+γ) is sharp; a natural test is to construct a measure whose support has Hausdorff dimension exactly 2/(1+γ) and check whether the liminf of the rescaled gap is finite on a positive-measure set.","The sparse-Verblunsky construction suggests a quantitative trade-off: if the gaps between nonzero coefficients grow only polynomially instead of rapidly, one might expect a slower approach to sine-kernel asymptotics, and computing that rate would give a finer picture of the singularity–repulsion boundary.","The remark that the real-line example is absolutely continuous with respect to every h_α with α<1 invites the same question for the circle measures; if true, these singular continuous measures are as continuous as possible while still singular, reinforcing the message that pure singularity does not constrain spacing."],"forward_implications":["For any measure giving zero weight to sets of Hausdorff dimension at most α, the rescaled gap n^{2/α−1}(θ_0^{(n)}(Θ)−θ_{-1}^{(n)}(Θ)) tends to infinity for μ-almost every Θ.","For the absolutely continuous part of any measure, the weaker bound limsup n(θ_0^{(n)}(Θ)−θ_{-1}^{(n)}(Θ)) > 0 holds at almost every point.","There exist purely singular continuous measures on the circle such that, for every choice of paraorthogonal polynomials, n(θ_{j+1}^{(n)}(Θ)−θ_j^{(n)}(Θ))→2π at every point Θ.","Singularity of the underlying measure therefore does not rule out maximal repulsion; the constructed measures satisfy the strongest possible clock behavior.","The appendix supplies a unit-circle version of the Freud–Levin–Lubinsky principle: sine-kernel asymptotics of the Christoffel–Darboux kernel imply clock spacing for every corresponding paraorthogonal sequence."],"supporting_citations":[{"why":"Supplies the Szegő recurrence, transfer-matrix setup, and the growth and subordinacy theorems used to turn gap bounds into dimension bounds.","marker":"[26]"},{"why":"Supplies the original a priori gap bound that the paper adapts to the unit circle as Theorem 2.3.","marker":"[13]"},{"why":"Provides the sparse-perturbation construction and sine-kernel asymptotics for singular measures on the real line that Theorem 1.2 transplants to the circle.","marker":"[2]"},{"why":"Proves the identity H_n^{(β)}(z)=c(z−z_0)K_n(z,z_0) linking paraorthogonal zeros to Christoffel–Darboux kernel zeros.","marker":"[28]"},{"why":"Supplies the absolutely-continuous-spectrum and transfer-matrix result used in the real-line counterpart of part 1.","marker":"[12]"},{"why":"Supplies the power-law subordinacy dimension estimate used in the real-line counterpart of part 2.","marker":"[8]"}],"fun_headline_variants":["Fractal dimension sets floor on zero repulsion","Clock spacing survives singular continuous measures","Zero gaps tied to measure's fractality","Singular measures still show equidistribution of zeros"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The dimension part of the main theorem assumes that two previously established results about how quickly certain recurrence solutions grow and what that growth implies about set size apply to every probability measure on the unit circle, without the paper restating their hypotheses.","fun_headline_variants_meta":{"raw":{"variants":["Fractal dimension sets floor on zero repulsion","Clock spacing survives singular continuous measures","Zero gaps tied to measure's fractality","Singular measures still show equidistribution of zeros"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000455,"raw_usage":{"total_tokens":2207,"prompt_tokens":788,"completion_tokens":1419,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":404,"completion_tokens_details":{"reasoning_tokens":1363}},"tokens_in":404,"tokens_out":1419,"duration_ms":11507,"temperature":1.0,"reasoning_tokens":1363,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T12:37:01.454125+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Construct a measure μ that is supported on a self-similar Cantor set of Hausdorff dimension α and gives zero weight to all sets of dimension at most α, compute the zeros of the paraorthogonal polynomials for a fixed β, and check whether $n^{{2/α−1}}$($θ_0^{{(n)}}$(Θ)−θ_{-1}^{(n)}(Θ)) diverges to infinity for μ-almost every Θ; if a point of positive μ-measure has a sequence of n with finite liminf of this quantity, Theorem 1.1(2) fails.","supporting_citations":[{"cited_title":"Simon, Orthogonal Polynomials on the Unit Circle , American Math- ematical Society Colloquium Publications 54, American Mathematical Society, Providence, RI, 2009","cited_arxiv_id":null,"evidence_quote":"Supplies the Szegő recurrence, transfer-matrix setup, and the growth and subordinacy theorems used to turn gap bounds into dimension bounds."},{"cited_title":"Last and B","cited_arxiv_id":null,"evidence_quote":"Supplies the original a priori gap bound that the paper adapts to the unit circle as Theorem 2.3."},{"cited_title":"Breuer, Sine kernel asymptotics for a class of singular measures , J","cited_arxiv_id":null,"evidence_quote":"Provides the sparse-perturbation construction and sine-kernel asymptotics for singular measures on the real line that Theorem 1.2 transplants to the circle."},{"cited_title":"Wong, First and second kind paraorthogonal polynomials and their zeros, J","cited_arxiv_id":null,"evidence_quote":"Proves the identity H_n^{(β)}(z)=c(z−z_0)K_n(z,z_0) linking paraorthogonal zeros to Christoffel–Darboux kernel zeros."},{"cited_title":"Last and B","cited_arxiv_id":null,"evidence_quote":"Supplies the absolutely-continuous-spectrum and transfer-matrix result used in the real-line counterpart of part 1."},{"cited_title":"Jitomirskaya and Y","cited_arxiv_id":null,"evidence_quote":"Supplies the power-law subordinacy dimension estimate used in the real-line counterpart of part 2."}],"review_version":1}