betaFeedback

WorksMathematics

A computer-assisted bound Λ ≤ 0.107964910079 for the de Bruijn–Newman constant

Abstract

In the Polymath normalisation, a computer-assisted comparison gives Λ≤4276930720779424736615780984913961408125713216879677197516800<0.107964910079.\Lambda\le\frac{427693072077942473661578098491}{3961408125713216879677197516800}<0.107964910079. The initial squared zero height is bounded by 9/169/16, using OpenAI's ordinary-zeta 7/87/8 nonvanishing theorem formalised in Lean at the identified source revision. Finite signed translated-probe functionals, matched physical source estimates, and continuous-domain kernel inequalities supply a piecewise comparison for that height. A newly implemented exact rational consumer checks every one of the 221 010221\,010 arcs, their outward rounding and joins, and the complete final de Bruijn contraction. The fifteen-stage finite replay includes a direct reconstruction of all 6767 source chords from their primitive payments, with explicit binding of the main terms and the complete derivative functionals. Separate consumers verify their continuous kernel certificates and the actual physical and arithmetic charges. The analytic argument includes the corrected 11/1211/12 Gaussian remainder, finite-head localisation, attainment of a positive-time highest contact, and repeated complex contacts. The accompanying packet contains the mathematical sources, certificate data, checker programs, exact source identities, and replay instructions. The initial theorem is used at cited-theorem trust with authenticated external Lean-build evidence; a fresh local Lean build and an end-to-end formalisation of the present comparison are not claimed.

Result and proof structure

Let

Ht(z)=∫0∞etu2Φ(u)cos⁡(zu) du(1)H_t(z)=\int_0^\infty e^{tu^2}\Phi(u)\cos(zu)\,\mathrm{d}u \tag*{(1)}
Φ(u)=∑n=1∞(2π2n4e9u−3πn2e5u)e−πn2e4u.\Phi(u)=\sum_{n=1}^\infty\left(2\pi^2n^4e^{9u}-3\pi n^2e^{5u}\right)e^{-\pi n^2e^{4u}}.

The decay of Φ\Phi makes HtH_t entire for every real tt. These functions are even and real on the real axis, and satisfy

H0(z)=18ξ((1+iz)/2),∂tHt=−∂z2Ht,ξ(s)=12s(s−1)π−s/2Γ(s/2)ζ(s).(2)H_0(z)=\frac{1}{8}\xi((1+iz)/2),\qquad\partial_tH_t=-\partial_z^2H_t,\qquad\xi(s)=\frac{1}{2}s(s-1)\pi^{-s/2}\Gamma(s/2)\zeta(s). \tag*{(2)}

The de Bruijn–Newman constant Λ\Lambda is the real threshold for which all zeros of HtH_t are real exactly when t≥Λt\geq\Lambda [1, 2, 3].

Theorem 1.1. For the normalisation (1)–(2),

0≤Λ≤B∗:=4276930720779424736615780984913961408125713216879677197516800<1079649100791012.(3)0\leq\Lambda\leq B_*:=\frac{427693072077942473661578098491}{3961408125713216879677197516800}<\frac{107964910079}{10^{12}}. \tag*{(3)}

The upper bound is computer assisted and uses the ordinary-zeta 7/8 theorem [24] and the finite-height zero theorem of Platt and Trudgian [4]. The exact finite certificates and analytic estimates used to obtain the upper bound are specified below and supplied in the accompanying packet.

The lower inequality is the theorem of Rodgers and Tao [5]. The task here is to establish the upper inequality. The comparison does not use an assumption of the full Riemann hypothesis. The ordinary 7/8 input, including its source and external verification boundary, is stated separately so that a referee can check the imported theorem independently of the new comparison.

Initial data and terminal data

By [24], a nontrivial zeta zero ρ\rho satisfies Re⁡ρ≤7/8\operatorname{Re}\rho\leq7/8. The functional equation also gives Re⁡ρ≥1/8\operatorname{Re}\rho\geq1/8. Under (2) its corresponding H0H_0 zero therefore has ∣Im⁡z∣≤3/4|\operatorname{Im}z|\leq3/4. The strict squared-height barrier starts at

Q0=916+10−12=5625000000011012.(4)Q_0=\frac{9}{16}+10^{-12}=\frac{562500000001}{10^{12}}. \tag*{(4)}

The strict cushion is retained in every rational computation. At the terminal comparison time the certified data are

T=22101204800=0.1079150390625,QT=790237796383053674715864379228162514264337593543950336.(5)T=\frac{22101}{204800}=0.1079150390625,\qquad Q_T=\frac{7902377963830536747158643}{79228162514264337593543950336}. \tag*{(5)}

Here QT>0Q_T>0. Consequently a further time QT/2Q_T/2 is required for de Bruijn’s strip contraction. Exact addition gives T+QT/2=B∗;T+Q_T/2=B_*; the displayed decimal in (3) is rounded upward. Stopping at TT alone would omit part of the proof.

From a finite functional to a global height barrier

At a simple nonreal highest zero x+iyx+iy, set q=y2q=y^2. A finite signed combination of logarithmic-derivative probes supplies a physical lower bound JJ; its entire distinguished conjugate-pair contribution is the explicitly defined function Opair(q)O_{\mathrm{pair}}(q). A kernel inequality then gives

q′≤−2−2q(J−Opair(q)).(6)q' \le-2-2q\left(J-O_{\mathrm{pair}}(q)\right). \tag*{(6)}

The term −2-2 is the attraction to the distinguished conjugate zero. The subtraction Opair(q)O_{\mathrm{pair}}(q) belongs to the whole signed functional and is paid exactly once. In particular, optimising a source floor alone does not optimise the contraction rate.

The proof has three distinct analytic obligations before one can integrate (6). First, the sources must apply to the actual function HtH_t at every declared probe and real coordinate, including arbitrarily large exterior coordinates. Second, the signed kernel inequalities must cover all real spacings and the full permitted heights of every other zero, including nonreal zeros. Third, the first-contact argument must handle compactness, time zero and repeated complex zeros. These obligations are proved in the following sections; the finite numerical histories check the subsequent scalar comparison.

Proof componentContent and verification boundary
Initial stripOrdinary-zeta 7/8 theorem; exact strip map and strict seed (4).
Finite headPublished finite RH height, full continuous wall, and analytic zero continuation.
Exterior sourceCorrected height-six approximation and signed finite, matched-source inequalities.
KernelRational continuous-domain certificates, whole own pair, and harmonic extension to other-zero heights.
Global comparisonSimple and repeated contacts, remote compactness, all closed time cells and upward numerical resets.
Exact flow221 010 rational arcs, independent checker, and full terminal QT/2Q_T/2 cleanup.

Table 1.

Edition and reproducibility

This edition supersedes referee edition 3 of 28 October 2026; the numerical plan, source table and comparison history are unchanged; Appendix B adds the direct source reconstruction.

Relation to earlier versions

The pinned public repository [9] states the ceiling 0.116397726408291. The present displayed ceiling is smaller by 0.008432816329291. This compares the stated numerical bounds; independent recognition of a world record is not asserted. An earlier research comparison gave 0.107936602217 under the narrower initial assertion Re⁡ρ≤0.8748867\operatorname{Re}\rho\le0.8748867. Xu and Sang [21] obtain that boundary under the analytic inputs stated there. Liu’s related refinements give 0.874957019420098946… [22] and 0.874886640739964… [23], with their stated analytic inputs. Since 0.8748867 << 7/8, the narrower assertion does not follow from the ordinary 7/8 theorem alone. The present argument therefore checks a new history from the literal seed (4); its input remains the ordinary 7/8 theorem.

The local native and confluent calculations retained in the research archive are not used in Theorem 1.1. Nor is an unproved uniform-current premise used. In particular, neither Λ<0.1\Lambda< 0.1 nor the Riemann or Lindelof hypothesis is a conclusion of this manuscript.

Related work and the external strip input

The heat-flow and zero-motion framework follows de Bruijn, Newman and Polymath [1, 2, 3]. Ki, Kim and Lee [17] proved that, at every positive heat parameter, all but finitely many zeros are real and simple, and obtained Λ<1/2\Lambda< 1/2. The finite real-zero input used here is the verification of Platt and Trudgian [4]; the nonnegative lower bound is due to Rodgers and Tao [5].

Raymond’s earlier time-zero construction and translated geometry are set out in the analytic companion [20]. The subsequent phase-correlated, translated-probe and retuned releases are [11, 12, 13]. The physical estimates and direct source reconstruction used here are stated in the present manuscript.

Planat [18], Section 8 reports a complete numerical reproduction of Raymond’s frozen correlated-probe release, recovering 0.149395073827688, and a head-pair extension to 0.149344833997286. His later classical-strip comparison [19], Sections 4–5 reports a stored-source replay of the retuning archive, including all eight supplied comparison histories, and independently checks the 57-kernel catalogue and 19 023 steps of its refined comparison. These replications concern the earlier releases and retain the analytic premises stated in the cited theorems.

The 2026 entries in Tao’s maintained catalogue [31] are the Mosaic Intelligence packages for 0.1965, 0.1875, 0.16 and 0.1595 [25, 26, 27, 28], and Kornaus’s 0.172422 package [16]. The catalogue marks these as self-archived certificate packages awaiting external review. Further computer-assisted preprints include Gomila’s 0.1787854 release [29] and Gordon’s bound below 0.158 [30]. These references provide comparison context.

Recent comparisons also require attention to the initial theorem. Hakimoglu-Brown [14] states the 0.13963 ceiling under the seven-eighths hypothesis. The height-dependent region of Bellotti, Trudgian and Yang [15] is a different input from a fixed global seven-eighths half-plane. The OpenAI preprint [24] supplies the ordinary 7/8 input used here; its pinned Lean declaration is identified in § 5.1.

Analytic foundations and the passage from source bounds to a global strip

This section fixes the normalisation and proves the analytic implications used to consume the finite certificates. The published inputs are the strip-contraction theorem of de Bruijn, the local zero-dynamics and positive-time tail theorems in [3], and the finite Riemann-hypothesis verification in [4]. The explicit wall at the numerical value of XX below is a computation in this work. In particular, the published finite-height theorem is not being asserted at every later heat time.

Normalisation, symmetry, and the initial strip

Put

ξ(s)=s(s−1)2π−s/2Γ(s/2)ζ(s),(7)\xi(s) = \frac{s(s-1)}{2}\pi^{-s/2}\Gamma(s/2)\zeta(s), \tag*{(7)}
Φ(u)=∑n=1∞(2π2n4e9u−3πn2e5u)e−πn2e4u,(8)\Phi(u) = \sum_{n=1}^{\infty}(2\pi^{2}n^{4}e^{9u}-3\pi n^{2}e^{5u})e^{-\pi n^{2}e^{4u}}, \tag*{(8)}
Ht(z)=∫0∞etu2Φ(u)cos⁡(zu) du.(9)H_t(z) = \int_{0}^{\infty}e^{tu^{2}}\Phi(u)\cos(zu)\,\mathrm{d}u. \tag*{(9)}

The apparent singularities in (7) are removed. Thus ξ\xi is entire, and the standard Fourier identity gives

H0(z)=18ξ(1+iz2),∂tHt=−∂z2Ht.(10)H_0(z)=\frac{1}{8}\xi\left(\frac{1+iz}{2}\right),\qquad\partial_t H_t=-\partial_z^2 H_t. \tag*{(10)}

The super-exponential decay of Φ\Phi justifies differentiation under the integral on compact subsets of the (t,z)(t,z) domain. In particular HtH_t is jointly analytic locally in tt and zz, and, for real tt, is an entire function of order one. It is real on the real axis and even:

Ht(z‾)=Ht(z)‾,Ht(−z)=Ht(z).(11)H_t(\overline{z})=\overline{H_t(z)}, \qquad H_t(-z)=H_t(z). \tag*{(11)}

All zero counts below include multiplicity. Every summand of (8) is positive when u≥0u\ge0, since 2πn2e4u−3>02\pi n^2 e^{4u}-3>0. Hence

Ht(iy)=∫0∞etu2Φ(u)cosh⁡(yu) du>0(t,y∈R).(12)H_t(iy)=\int_0^\infty e^{tu^2}\Phi(u)\cosh(yu)\,du>0 \qquad(t,y\in\mathbb{R}). \tag*{(12)}

In particular HtH_t is never the zero function and has no zeros on the imaginary axis.

The transformation between zeta zeros and H0H_0 zeros is worth keeping explicit. If z=x+iyz=x+iy, then

1+iz2=1−y2+ix2.(13)\frac{1+iz}{2}=\frac{1-y}{2}+i\frac{x}{2}. \tag*{(13)}

Thus the classical critical strip gives ∣y∣≤1|y|\le1 at t=0t=0. The ordinary seven-eighths theorem used here is the nonvanishing assertion

ζ(s)≠0(Re⁡s>78, s≠1)(14)\zeta(s)\ne0 \qquad\left(\operatorname{Re}s>\frac{7}{8},\ s\ne1\right) \tag*{(14)}

for the ordinary Riemann zeta function; the finite extension assigned to the pole by the formal library is irrelevant to this nontrivial-zero statement. Its precise source version and verification provenance are specified with the external theorem input [24]. By the functional equation, a nontrivial zero with real part below 1/81/8 would give one with real part above 7/87/8. Consequently all nontrivial zeta zeros have real parts in [1/8,7/8][1/8,7/8]. The factors defining ξ\xi add no zeros at s=0s=0 or s=1s=1, and its zeros are precisely the nontrivial zeta zeros. Equation (13) therefore gives the global initial strip

H0(x+iy)=0⟹∣y∣≤34,y2≤916.(15)H_0(x+iy)=0 \quad\Longrightarrow\quad|y|\le\frac{3}{4}, \qquad y^2\le\frac{9}{16}. \tag*{(15)}

The weaker seven-eighths input is used literally; a strip with boundary 0.87488670.8748867 is not inferred from it.

The following classical contraction theorem is used twice in the proof. It is [3 Theorem 3.2], quoting [1 Theorem 13].

Theorem 2.1 (de Bruijn strip contraction). Suppose t0∈Rt_0\in\mathbb{R} and a>0a>0, and every zero of Ht0H_{t_0} has imaginary part in [−a,a][-a,a]. Then for t≥t0t\ge t_0 every zero of HtH_t has imaginary part of modulus at most

max⁡{a2−2(t−t0),0}.\sqrt{\max\{a^2-2(t-t_0),0\}}.

In particular all zeros are real at every time t≥t0+a2/2t\ge t_0+a^2/2.

The initial use of this theorem and (15) yields

Ht(x+iy)=0⟹y2≤max⁡{916−2t,0},t≥0.(16)H_t(x+iy)=0 \quad\Longrightarrow\quad y^2\le\max\left\{\frac{9}{16}-2t,0\right\}, \qquad t\ge0. \tag*{(16)}

Only a very short initial segment of this elementary comparison is needed for strict startup. The quantitative improvement comes from the source and geometry inequalities, not from extrapolating the contraction theorem.

The stationary wall and finite-head theorem

The finite wall is at

X=6000000185827,N=690988,Tw=41250.(17)X = 6000000185827,\qquad N = 690988,\qquad T_{\mathrm{w}} = \frac{41}{250}. \tag*{(17)}

It is stationary in the zz coordinate. Define

L=log⁡X4π,vn=log⁡n,L = \log\frac{X}{4\pi},\qquad v_n = \log n,

and use the leading gamma factor

M0(s)=s(s−1)16π−s/22π(s/2)s/2−1/2e−s/2.(18)M_0(s) = \frac{s(s-1)}{16}\pi^{-s/2}\sqrt{2\pi}(s/2)^{s/2-1/2}e^{-s/2}. \tag*{(18)}

The powers use the logarithm analytic on the relevant upper or lower half-plane. Set

References

  1. [1]N. G. de Bruijn, The roots of trigonometric integrals, Duke Math. J. 17 (1950), no. 3, 197–226. doi:10.1215/S0012-7094-50-01720-0.DOI
  2. [2]C. M. Newman, Fourier transforms with only real zeros, Proc. Amer. Math. Soc. 61 (1976), no. 2, 245–251. doi:10.1090/S0002-9939-1976-0434982-5.DOI
  3. [3]D. H. J. Polymath, Effective approximation of heat flow evolution of the Riemann ξ function, and a new upper bound for the de Bruijn–Newman constant, Research in the Mathematical Sciences 6 (2019), article 31. doi:10.1007/s40687-019-0193-5. Numbered references use arXiv:1904.12438v2.arxiv.org/abs/1904.12438
  4. [4]D. J. Platt and T. S. Trudgian, The Riemann hypothesis is true up to 3 · 10¹², Bull. Lond. Math. Soc. 53 (2021), 792–797. doi:10.1112/blms.12460; arXiv:2004.09765. Theorem 1 states the exact verification height 3000175332800.DOI
  5. [5]B. Rodgers and T. Tao, The de Bruijn–Newman constant is non-negative, Forum of Mathematics, Pi 8 (2020), e6, 62 pp. doi:10.1017/fmp.2020.6; arXiv:1801.05914.DOI
  6. [6]OpenAI, lean/OAI/NumberTheory/DirichletL/Nonvanishing.lean, Lean source for the ordinary-zeta declaration OAI.riemannZeta_ne_zero_of_seven_eighths_lt_re, in the official openai/math repository, revision fd4aeeb2ee4fc729c18d98444fed42fd0529eeeb. https://github.com/openai/math/blob/fd4aeeb2ee4fc729c18d98444fed42fd0529eeeb/lean/OAI/NumberTheory/DirichletL/Nonvanishing.lean. The independent rebuild in [7] uses revision adc7f1241b42e322a6451854ab7e4b4c146bf78a; the relevant proof subtree and build configuration match the retained source pin.
  7. [7]tomoto0, Quasi-Riemann hypothesis 7/8 verification, source/build report and explicit ordinary-zeta challenge, https://github.com/tomoto0/quasi-riemann-hypothesis-7-8-verification, commit b83a73c6cb626a391e551baafdb7d0cceafaa55a. The external report is supplied as verification evidence; the present work did not repeat the full Lean build locally.
  8. [9]D. J. Raymond, Retuned translated probes and the bound for the de Bruijn–Newman constant, public manuscript and proof materials, https://github.com/Dale-Dug/de-bruijn-newman-0116398, commit 4694f9ab2e8bc88aeeae1949d2e690c2360d2eba. The pinned repository states Λ ≤ 0.116397726408291; its October 26 update adds the ordinary 7/8 formal-input evidence.
  9. [11]D. J. Raymond, Phase-correlated certificates and the bound Λ < 0.163297 for the de Bruijn–Newman constant, Zenodo preprint, 8 September 2026. doi:10.5281/zenodo.22626161.DOI
  10. [12]D. J. Raymond, Correlated translated probes and the bound Λ < 0.149396 for the de Bruijn–Newman constant, Zenodo preprint, 21 September 2026. doi:10.5281/zenodo.22867547.DOI
  11. [13]D. J. Raymond, Retuned translated probes and the conditional bound Λ < 0.116398 for the de Bruijn-Newman constant, Zenodo preprint, 10 October 2026. doi:10.5281/zenodo.23272725.DOI
  12. [14]Cetin Hakimoglu-Brown. A conditional bound Λ < 0.13963 for the de Bruijn-Newman constant. 8 October 2026. DOI: 10.48550/arXiv.2610.12803. arXiv:2610.12803v1.
  13. [15]Chiara Bellotti, Tim Trudgian, Andrew Yang. Zero-free regions inspired by work of Heath-Brown. 23 March 2026. DOI: 10.48550/arXiv.2603.21490. arXiv:2603.21490v1.DOI
  14. [16]Daniel Martin Kornaus. Collision Exclusion in the de Bruijn-Newman Heat Flow. 8 September 2026. DOI: 10.5281/zenodo.22658444. Zenodo preprint, record 22658444.DOI
  15. [17]H. Ki, Y.-O. Kim and J. Lee, On the de Bruijn-Newman constant, Advances in Mathematics 222 (2009), no. 1, 281–306. doi:10.1016/j.aim.2009.04.003.DOI
  16. [18]M. Planat, Collective Contraction in the de Bruijn-Newman Heat Flow: Off-Zero Logarithmic Derivatives and Certified Barrier Refinements, preprint, version 2, 7 October 2026, 54 pp.; first version 29 September 2026. arXiv:2609.37164v2; doi:10.48550/arXiv.2609.37164.DOI
  17. [19]M. Planat, A certified classical-strip comparison for the de Bruijn-Newman constant, preprint, 13 October 2026, 19 pp. arXiv:2610.18214v1; doi:10.48550/arXiv.2610.18214.DOI
  18. [20]D. J. Raymond, Analytic foundations for the time-zero upper-bound construction, version 1.0, 21 September 2026, 19 pp. Companion manuscript in the deposit [12], doi:10.5281/zenodo.22867547; supplied as Prior_Analytic_Foundations.pdf. The original public companion is also retained in https://github.com/Dale-Dug/de-bruijn-newman-0149396.
  19. [21]Y. Xu and H. Sang, Amplified moments and a refinement of the quasi-Riemann hypothesis, preprint, version 2, 15 October 2026; first version 11 October 2026. arXiv:2610.15291v2; doi:10.48550/arXiv.2610.15291.DOI
  20. [22]B. Liu, Improved zero-free half-planes for the quasi-Riemann hypothesis, preprint, version 2, 12 October 2026; first version 8 October 2026. arXiv:2610.12234v2; doi:10.48550/arXiv.2610.12234.DOI
  21. [23]B. Liu, Zero-free half-planes for Hecke L-functions over number fields, preprint, submitted 19 October 2026. arXiv:2610.26835v1; doi:10.48550/arXiv.2610.26835.DOI
  22. [24]OpenAI. The Quasi-Riemann Hypothesis: A Zero-Free Half-Plane Re(s) > 7/8. 30 September 2026. https://github.com/openai/math/blob/main/preprints/The-Quasi-Riemann-Hypothesis-September-30-2026/paper.pdf.
  23. [25]Mosaic Intelligence. A certified unconditional upper bound Λ ≤ 0.1965 for the de Bruijn-Newman constant. 2026. DOI: 10.5281/zenodo.20724170.
  24. [26]Mosaic Intelligence. A certified unconditional upper bound Λ ≤ 0.1875 for the de Bruijn-Newman constant. 2026. DOI: 10.5281/zenodo.21175533.DOI
  25. [27]Mosaic Intelligence. A certified unconditional upper bound Λ < 0.16 for the de Bruijn-Newman constant. 2026. https://github.com/463464q435q43/optimizationproblems/tree/6baae4c15dccec1e94ac011de399ec382fcea7b6/certificates/certified0160. Repository revision 6baae4c15dccec1e94ac011de399ec382fcea7b6.
  26. [28]Mosaic Intelligence. A certified unconditional upper bound Λ < 0.1595 for the de Bruijn–Newman constant. 2026. https://github.com/463464q435q43/optimizationproblems/tree/cc8107854fedd5a96d7eb695178f405568712f35/certificates/certified01595. Repository revision cc8107854fedd5a96d7eb695178f405568712f35.
  27. [29]Jude Gomila. A computer-assisted proof of the bound Λ ≤ 0.1787854 for the de Bruijn–Newman constant. 20 August 2026. https://github.com/judegomila/dbn-lambda-01787854-candidate-audit. Author’s primary release manuscript; not yet peer reviewed.
  28. [30]Stefan Gordon. A computer-assisted upper bound for the de Bruijn–Newman constant. 12 September 2026. DOI: 10.2139/ssrn.7068638. Preprint; author-hosted PDF dated September 2026. SSRN posting 14 October 2026, revised 15 October 2026.
  29. [31]T. Tao and contributors, de Bruijn–Newman constant, Optimization Problems and Constants, maintained catalogue, accessed 29 October 2026. https://teorth.github.io/optimizationproblems/constants/21a.html.

Paper details

vlnc:W_s4yg53v5
Version
v1 · latest
Published
Fresh — published within the last 30 days
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents