{"id":"5218144f-8583-47e5-8e65-c12b4851ecf8","arxiv_id":"2602.20313","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"The de Bruijn–Newman kernel is certified not to be a Pólya frequency function of order 5, via an interval-arithmetic-enclosed negative Toeplitz determinant.","lead":"A certified computation shows the de Bruijn–Newman kernel, a function tied to the Riemann zeta story, fails a strict positivity condition at order 5. A negative 5×5 determinant is bracketed by rigorous interval arithmetic, though some broader claims in the posted text are withdrawn.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Certificate rests on Lemma 4.1 tail bound whose proof uses reversed inequalities; without a corrected bound the determinant enclosure is not proven.","rationale":"The reader's weakest assumption was the rigor of the interval/tail-bound pipeline. I agree that this is the critical zone, and I found a specific defect there: Lemma 4.1's proof uses false polynomial inequalities, so the truncation bound is not proved as printed. This is a repair-level issue rather than a refutation: the tail is dominated by the n=51 term, which is around 10^{-3540}, so a corrected bound should be easy to supply. I did not find a mathematical flaw in the PF5 reduction itself—a single negative admissible 5x5 Toeplitz minor is sufficient. The separate internal inconsistency around Theorem 1.4(ii)/Lemma 5.2 and the withdrawn global sign claims is real but does not affect the central counterexample. Therefore I keep the reader's CONDITIONAL verdict: the paper should not be accepted until Lemma 4.1's tail-bound proof is corrected and the determinant certificate is independently reproduced.","tokens_in":11224,"tokens_out":12288,"duration_ms":114122,"concrete_test":"Repair and verify Lemma 4.1 directly: for n=51+m, bound the tail term by P(51+m)e^{-π(51+m)^2} ≤ P(51)(1+m/51)^4 e^{-π(102m+m^2)} and sum/upper-bound the first 1000 terms with interval arithmetic (or use a ratio test). Check the total is below 10^{-70}. Independently, recompute the determinant enclosure with a different rigorous interval library such as Arb; if the enclosure remains strictly negative, Theorem 1.1 survives after the proof correction.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Theorem 1.1's negative PF5 certificate depends on a rigorous enclosure of the 5x5 Toeplitz determinant. That enclosure relies on Lemma 4.1's truncation bound, but the proof of Lemma 4.1 contains a concrete error: for n≥51 it asserts n^4 ≤ 51^4 e^{2 log(n/51)} and n^2 ≤ 51^2 e^{log(n/51)}, both of which are false for n>51 (e.g., at n=52 the first reads 7.31e6 ≤ 7.03e6). These inequalities are used to justify the geometric tail bound, so as written the truncation error is not rigorously controlled. Lemma 4.2 then widens Φ_N enclosures by ±10^{-70} using exactly this bound, and Proposition 4.3 derives the negative determinant from those enclosures. Thus the formal certificate has a gap in its stated proof, even if the underlying tail estimate happens to be true. The issue is separate from whether det(M) is actually negative: the determinant may well be negative, but the paper's claimed rigorous enclosure is not justified without repairing Lemma 4.1. This is the most load-bearing concern because the entire PF5 counterexample is a certified numerical statement; if the certificate is not rigorous, the theorem is not established by the text as written.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper claims a computational proof that the de Bruijn--Newman kernel K(u)=Phi(|u|) is not a Polya frequency function of order 5. The central certificate is a 5x5 Toeplitz minor at (u0,h)=(0.01,0.05) whose determinant is enclosed in [-1.8472496e-9,-1.8472225e-9] by 80-digit interval arithmetic, using a truncation of the theta series at N=50 and a tail bound. The paper also derives an exact algebraic formula for the leading coefficient C_r(u0) in the small-spacing expansion of D_r(u0,h), reports certified positivity of lower-order Toeplitz minors at the central configuration, presents additional counterexample configurations, and studies a Gaussian deformation threshold. The result is framed as a negative answer to PF_5 for this kernel and explicitly not a statement about the Riemann Hypothesis.","tokens_in":11479,"tokens_out":3731,"duration_ms":34626,"significance":"If the interval certificate is sound, the paper settles a natural total-positivity question for the de Bruijn--Newman kernel: K is not PF_5. The exact formula for C_r and the explicit Toeplitz reduction are useful structural observations, and the availability of reproducible code is a strength. The paper is also careful to distinguish the Toeplitz threshold from a global PF_4 statement and to separate certified computations from non-certified numerics. However, the proof of the central certificate depends on a tail bound whose stated proof is faulty, and the global sign claim for C_5 is internally contradicted by the abstract's withdrawal notice. These issues affect the rigor of the main theorem as written, although they appear repairable in scope.","major_comments":[{"comment":"The truncation bound is not proved as written. For n≥51 the proof asserts n^4 ≤ 51^4 e^{2 log(n/51)} and n^2 ≤ 51^2 e^{log(n/51)}. These are equivalent to (n/51)^4 ≤ (n/51)^2, which is false for n>51; at n=52 the first reads 7.31e6 ≤ 7.03e6. The geometric-tail estimate therefore does not follow from the stated inequalities. Since Lemma 4.2 widens the Phi_N enclosures by ±10^{-70} using exactly this bound, and Proposition 4.3 derives the negative determinant from those enclosures, the rigorous status of Theorem 1.1 is not established by the text as written. The determinant may well be negative, but the certificate needs a corrected tail bound.","section":"§4.1, Lemma 4.1"},{"comment":"The claim that C_5(u0)<0 for all u0 in (0,u0*) is not supported. Negativity on 31 grid points together with continuity does not rule out positive excursions between grid points. The 'bisection-certified zero at u0*' locates a sign change at one point but does not certify uniqueness of that zero. Moreover, the abstract explicitly states that Version 2 withdraws the certified global sign and unique-threshold claims for C_5 because the derivative-tail enclosure was unsound, yet Theorem 1.4(ii) and Lemma 5.2 still assert those claims. This internal contradiction must be resolved: either a correct global proof is supplied, or the statement should be downgraded to non-certified numerical evidence on a grid.","section":"§5.2, Lemma 5.2 and Theorem 1.4(ii)"},{"comment":"The abstract says 'Eight further configurations are certified by both an explicit Leibniz expansion and an independent interval determinant computation,' but §4.2 certifies only two further configurations by interval arithmetic, and the independent Leibniz-expansion check is not described for these configurations. Table 2 lists three certified configurations total. The count should be reconciled.","section":"Abstract / §4.2"}],"minor_comments":[{"comment":"The manuscript is labeled arXiv:2602.20313v1 but the abstract contains a 'Version 2 withdraws...' passage. Please clarify which version is under consideration, since the withdrawal statement concerns claims that are still printed as Theorem 1.4(ii).","section":"Abstract"},{"comment":"Table 2 appears to list 24 configurations as found by scanning, but only 8 rows are shown and only 3 are marked as certified. A note on the selection criterion and the status of the remaining rows would help.","section":"§4.2, Table 2"},{"comment":"The derivative magnitudes are said to be certified by 'direct differentiation,' but the bound on the tail for each derivative is referenced to Lemma 4.1, which has the gap noted above. Once Lemma 4.1 is repaired, this statement should be revisited.","section":"§5.1, Proposition 5.1"},{"comment":"Several numerical values are presented with excessive digits (e.g. enclosures of width about 1e-12, while 80-digit precision is used). This is not an error, but the paper could state the precision needed for each computation.","section":"General"}],"recommendation":"major_revision","confidential_remarks":"The central counterexample is promising and likely correct, but the proof as written has a genuine gap in the tail bound, and the paper contradicts itself about the status of the C_5 global sign claim. Both are load-bearing for the certified claims. I recommend major revision rather than rejection because the errors appear local and repairable within the manuscript's scope."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know two things before reading: the paper's central negative claim about the de Bruijn–Newman kernel is probably true, but the proof as written is not. The explicit 5×5 Toeplitz minor is a nice idea, and the algebraic reduction to C_r is genuinely useful. But Lemma 4.1's tail bound has reversed inequalities—the stated bounds fail for every n>51—so the interval certificate for det(M) is not actually established. On top of that, the body still asserts a global sign claim for C5 that the abstract explicitly withdraws. Both need fixing before this is a clean contribution.\n\nWhat is new and good: the paper gives a systematic way to reduce the PF_r condition to a two-parameter Toeplitz determinant family, and derives an exact formula for the leading coefficient C_r(u0). The explicit negative 5×5 minor at (0.01,0.05), if the certificate holds up, would be a real result: nobody has shown this kernel fails PF5 before. The use of interval arithmetic with an independent interval-LU check is good practice, and the code is linked. The reader's confidence in the algebraic formula and the Leibniz expansion seems justified.\n\nThe soft spots, in proportion. First, Lemma 4.1 is load-bearing and wrong as stated. For n≥51, the proof uses n^4 ≤ 51^4 e^{2log(n/51)} and n^2 ≤ 51^2 e^{log(n/51)}. Both are false for n>51; at n=52 the first gives 7.31e6 ≤ 7.03e6. So the geometric tail bound is not proven, and the ±10^-70 widening in Lemma 4.2 has no basis. The determinant enclosure in Proposition 4.3 therefore isn't a proof as written. The tail may well be small, but that needs a corrected bound.\n\nSecond, the paper has an internal inconsistency: the abstract says version 2 withdraws the certified global sign and unique-threshold claims for C5, but Theorem 1.4(ii) and Lemma 5.2 still claim C5(u0)<0 for all u0 in (0,u*0). The proof there uses 31 grid points plus continuity, which doesn't establish a continuum result without a derivative bound or monotonicity. Those sections need to be rewritten to match the abstract, or the claims removed.\n\nThird, and minor: the 'certified' status of the extra configurations is a bit uneven—Table 2 marks many rows with '—', and only three are actually interval-certified. That's fine, but the text should be clearer about which claims are proven and which are scan-level numerics.\n\nWho is this for? People working on total positivity of special kernels, and anyone interested in rigorous computational proofs in analytic number theory. It deserves a serious referee, because the question is real and the method is promising, but the current version needs a corrected Lemma 4.1 and a consistent treatment of the C5 claims before it can be accepted.","headline":"Plausible and interesting claim, but the certificate has a gap in Lemma 4.1 and the paper contradicts itself on the C5 global claims; needs repair before it is a proof.","tokens_in":12013,"tokens_out":3986,"would_cite":false,"duration_ms":33709,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["15A15","65G40"],"pacs":[],"model":"deepseek-v4-flash","headline":"A certified 5×5 Toeplitz determinant shows the de Bruijn–Newman kernel is not Pólya frequency of order 5.","keywords":["Pólya frequency function","total positivity","de Bruijn–Newman kernel","Toeplitz determinant","interval arithmetic","certified computation","theta function series","Riemann zeta function"],"falsifier":"Recompute det[K(0.01+(i−j)·0.05)] with an independent certified method—say, a different interval library, higher truncation N, or exact rational arithmetic on high-precision approximations—and check whether the resulting enclosure is still strictly negative. Any certified evaluation that contains zero would invalidate the counterexample. Alternatively, verify empirically that some entry c_n lies outside the claimed enclosure in Lemma 4.2; a single violated entry enclosure would break the certificate.","tokens_in":11057,"feed_emoji":"🧮","tokens_out":6026,"duration_ms":48899,"temperature":0.7,"pith_summary":"This paper proves that the de Bruijn–Newman kernel K(u)=Φ(|u|), which arises in the study of the Riemann zeta function, is not a Pólya frequency function of order 5. The proof is a certified computation: a 5×5 Toeplitz matrix formed from K at nine points has determinant enclosed between −1.8472496×10⁻⁹ and −1.8472225×10⁻⁹, so it is strictly negative. Since Pólya frequency functions require all such minors to be nonnegative, this single negative determinant settles the question. The paper also shows the same configuration has positive minors of orders 2, 3, and 4, so the failure is localized to order 5, and identifies an algebraic mechanism: the leading small-spacing coefficient C₅(u₀) is negative for small u₀. Why it matters: total positivity of this kernel would have constrained the zero distribution of the Riemann ξ-function; the failure at order 5 blocks that route, while leaving order 4 open.","feed_headline":"Pólya frequency order 5 fails for de Bruijn–Newman kernel","feed_subtitle":"A certified 5×5 Toeplitz determinant comes out negative, blocking a total-positivity route to the Riemann ξ-function.","key_machinery":"The argument rests on three objects: (1) the Toeplitz determinant family D_r(u₀,h)=det[K(u₀+(i−j)h)] which reduces the 2r-dimensional PF_r condition to two parameters; (2) the small-h expansion giving the leading coefficient C_r(u₀) as a Vandermonde-weighted sum of Taylor coefficients of K at u₀, which isolates the order at which negativity appears; and (3) a rigorous numerical certificate combining a proved truncation bound for the theta series (tail <10⁻⁷⁰ for N=50 on [0,0.21]) with directed-rounding interval arithmetic at 80-digit precision, yielding airtight enclosures of the determinant. The generalized Vandermonde factors W(k₀,…,k_{r−1}) are the mechanism that makes the r=5 coefficient","core_discovery":"The paper's central claim is Theorem 1.1: for K(u)=Φ(|u|), the 5×5 Toeplitz matrix M with entries K(0.01+(i−j)·0.05) has determinant −1.847236...×10⁻⁹, rigorously enclosed in [−1.8472496×10⁻⁹, −1.8472225×10⁻⁹]. Because a Pólya frequency function of order 5 must have all such determinants nonnegative, this is a direct counterexample. The authors further establish that the Toeplitz minors of orders 2–4 at the same configuration are positive, derive an exact formula for the leading coefficient C_r(u₀) of the small-h expansion of D_r(u₀,h), and prove that C₅(u₀) is negative on (0,u₀*) with u₀*≈0.031139763615 while C₂,C₃,C₄,C₆,C₇ are positive there. They also compute non-certified Gaussian-deform","pith_inferences":["The certification recipe here—proved tail bound plus directed-rounding interval arithmetic—could be reused to test PF thresholds for other kernels defined by theta-type series, not just Φ.","The location of the negative determinant at order 5 suggests that total positivity of Φ(|u|) may be obstructed by the rapid alternating growth of its even Taylor coefficients; a fully symbolic proof of C₅<0 would clarify whether this is a generic phenomenon or a special resonance.","Because λ*₅ depends on the configuration, testing whether a global PF₅ statement holds for some fixed t>0 would require a completely different, non-Toeplitz argument; one could start by checking whether the negative minor persists under the Gaussian deformation for t close to 6–12.","The non-symmetric Toeplitz matrices used here (u₀≠0) may probe the kernel's odd-order structure in a way that symmetric configurations cannot; scanning larger families of (u₀,h) might reveal whether order-5 is the only failing order."],"forward_implications":["The kernel K is not PF₅, so any total-positivity argument for the Riemann ξ-function based on PF order ≥5 cannot work.","At (0.01,0.05), the Toeplitz PF threshold is exactly 5: D₂,D₃,D₄>0 and D₅<0, all certified.","The sign pattern of C_r suggests a resonance localized to order 5: only C₅ is negative, and only for small u₀; C₆ and C₇ are positive at u₀=0.01.","The Gaussian-deformation threshold λ*₅(u₀,h) is configuration-dependent, so there is no single 'healing time' that makes the kernel PF₅ globally; the deformed kernel's PF₅ status remains open.","Whether K is PF₄ in the full sense (all admissible configurations, not just Toeplitz) is left open and is Problem 1 of the paper."],"fun_headline_variants":["Certified negative 5×5 Toeplitz minor defeats PF5 claim","de Bruijn–Newman kernel: Pólya frequency order 5 fails","Kernel not PF5: explicit counterexample with interval proof","No Pólya frequency order 5 for de Bruijn–Newman kernel","5×5 determinant negative: de Bruijn–Newman not PF5"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The certified negative determinant depends on two things being airtight: the tail bound showing the truncated theta series has error below 10⁻⁷⁰ on [0,0.21], and the directed rounding of every arithmetic operation in the 120-term Leibniz expansion. If either leaks, the enclosure could in principle contain zero, and the PF₅ counterexample would lose its proof.","fun_headline_variants_meta":{"raw":{"variants":["Certified negative 5×5 Toeplitz minor defeats PF5 claim","de Bruijn–Newman kernel: Pólya frequency order 5 fails","Kernel not PF5: explicit counterexample with interval proof","No Pólya frequency order 5 for de Bruijn–Newman kernel","5×5 determinant negative: de Bruijn–Newman not PF5"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000222,"raw_usage":{"total_tokens":1375,"prompt_tokens":910,"completion_tokens":465,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":654,"completion_tokens_details":{"reasoning_tokens":364}},"tokens_in":654,"tokens_out":465,"duration_ms":4429,"temperature":1.0,"reasoning_tokens":364,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-02T21:21:21.274255+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Recompute det[K(0.01+(i−j)·0.05)] with an independent certified method—say, a different interval library, higher truncation N, or exact rational arithmetic on high-precision approximations—and check whether the resulting enclosure is still strictly negative. Any certified evaluation that contains zero would invalidate the counterexample. Alternatively, verify empirically that some entry c_n lies outside the claimed enclosure in Lemma 4.2; a single violated entry enclosure would break the certificate.","supporting_citations":[],"review_version":1}