/teal-sea
teal-sea / zeta-labstate of record · compiled 28 Sep 2026 · revision e4945c4 · source

Library · hunts/lambda_dh_bounds/MISSION.md

MISSION: Two-sided bounds for the de Bruijn-Newman constant of the Davenport-Heilbronn function

1,531 words · 167 lines · source

Opened 2026-08-16. Nothing in this directory is a result until the case log in hunts/README.md says how it ended. Everything here obeys the probe discipline: the strongest words used are measured, observed and decided (an enclosure with an exact sign or an exact integer count); the reserved enclosure word belongs to zeta/rigor.py and appears nowhere in this directory.

Frame note added 2026-08-16 (GATE.md closure item (a)). Nothing in the preregistration below is altered; this note only says which normalization its numbers are in, because the mission did not say and a Lambda number means nothing without it. The deformation defined below sits at s = 1/2 + iz, which is Stopple's published frame (arXiv:1301.3158). de Bruijn as usually quoted, Newman, Rodgers-Tao, Polymath 15 and Dobner all sit at s = (1+iz)/2, where the same constant is four times larger. So the pre-registered prediction P5, "0.0575 < Lambda_DH <= 0.400634", reads "0.2300 < Lambda_DH <= 1.602537" there. Conversion table, derivation and per-row numerical checks: FRAME.md.

Post-registration note added 2026-08-16, after the gate closed (GATE.md, verdict YES). Nothing in the preregistration below is altered. A corollary of the decided floor was recognized only after the gate: in the wide frame the floor 0.2304 = 144/625 exceeds Polymath 15's unconditional Lambda_zeta <= 0.22, so Lambda_DH > Lambda_zeta unconditionally. This separation was not preregistered, was not a work package, and is recorded with its chain, its grade (cited plus decided) and its adversary-sanctioned phrasing in SEPARATION.md.

The question

hunts/flow_repair/ (closed, 2026-08-07) measured the backward-heat repair times of nine off-line quadruples of the Davenport-Heilbronn function and closed with the sentence "Nothing here bounds Lambda_DH from above". Its floor, Lambda_DH >= 0.0576518, is float grade: one route, no enclosures.

This hunt upgrades that loose end to theorem grade on both sides. In the normalisation flow_repair measured to be exact, with a_n the period-5 coefficients (1, kappa, -kappa, -1, 0),

Phi_DH(u) = 4 e^{3u/2} sum_{n>=1} n a_n exp(-pi n^2 e^{2u} / 5), H_t(z) = int_0^inf e^{t u^2} Phi_DH(u) cos(zu) du, H_0 = Xi_DH,

and Lambda_DH := inf{t : H_t has only real zeros}, which is a well-defined finite real number >= 0 by Dobner's theorem for the extended Selberg class (arXiv:2005.05142, Theorems 1 and 2; the half-line structure {t : all zeros real} = [Lambda_F, inf) is his Theorem 1).

What are the first quantitative two-sided bounds for Lambda_DH?

Positioning, so the claim is neither over- nor under-stated: existence, finiteness and Lambda_DH >= 0 are published (Dobner 2020, for all of S#). Strict positivity is an immediate corollary of Dobner's half-line structure plus any off-line zero (Davenport-Heilbronn 1936; computed by Spira 1994). What is absent from the literature, per the novelty sweep recorded in NOVELTY.md, is any number: no quantitative bound, upper or lower, for the de Bruijn-Newman constant of any RH-violating L-function. The deliverable is that number, from both sides.

Work packages

Independence note added 2026-08-16 (GATE.md closure item (e)). The preregistration below is not altered; this note says what WP1's phrase "two independent winding routes" turned out to be worth, because the word independent was never measured when it was written. Measured with harness/independence.py, the two routes share the whole evaluator: independence radius 9 of 12 declared layers at the granularity of independence_decl.py, 8 of 11 at the gate's slightly coarser one, the same declaration either way. Their agreement is evidence about the two schemes that turn H-balls into an integer and about nothing else, and route 2 ran at one t and one box. The independence the claim rests on comes from two other legs, both landed here as runnable scripts: crosscheck_dhflow_winding.py (radius 0, N = 1 at both t) and crosscheck_quadfree.py (radius 0, one reconvergent layer, a different kappa equation). Layer lists, the two repaired validation gaps and what is still not covered: INDEPENDENCE.md.

Pre-registered predictions

Scope

This hunt may write: hunts/lambda_dh_bounds/, figures/, one new docs/NN-*.md if the run earns it, a case-log entry in hunts/README.md, and a pinning test under tests/ (precedent: tests/test_frontier_math_clean_kill.py).

This hunt may not write: zeta/, ontology/, harness/, lean/, and may not promote its own claim into README.md, ROADMAP.md or HANDOFF.md as an established finding.

id: lambda_dh_bounds
question: What are the first quantitative two-sided bounds for the de Bruijn-Newman constant Lambda_DH of the Davenport-Heilbronn function?
frontier: lower 0.0576518 measured float-grade by flow_repair (nine quadruples, no enclosures), upper absent everywhere; Dobner 2020 gives existence, finiteness and Lambda_DH >= 0 with no numbers
proposed_attack: Arb ball argument-principle winding count at t1 = 0.0575 for the lower side; de Bruijn 1950 Theorem 13 with a directed-rounding zero-strip constant sigma_0 for the upper side
dead_routes:
  - reading an O(1) residue as an off-line zero without a completeness check (Hunt 2, withdrawn)
  - float-grade landing times as bounds (flow_repair already measured them; they decide nothing)
  - mp.diff on internally rounded functions (returns exactly 0, pinned by regression tests)
required_oracles:
  - Arb ball arithmetic winding counts that must enclose an exact integer
  - directed-rounding interval arithmetic on both in-tree backends for the strip constant
  - cross-route agreement between H_0 quadrature and zeta.epstein.completed_dh
  - published theorems cited with hypotheses checked (de Bruijn 1950, Dobner 2020)
kill_conditions:
  - the all-zeros-in-strip form of de Bruijn Theorem 13 cannot be confirmed from the original text or a reliable restatement, in which case the upper bound is withdrawn and the hunt reports the lower side alone
  - a winding enclosure fails to bracket an integer after the subdivision budget, in which case no lower-bound claim is made at that t
  - any enclosure that fails to shrink when working precision rises, which marks the instrument as the artifact
  - any source found to already bound Lambda_DH quantitatively, in which case the novelty claim is withdrawn and the delta restated against that source
agents_may:
  - build instruments inside this directory
  - run ball-arithmetic computations and record enclosures with their precisions
  - fetch and read the cited papers
  - write measured and decided values into RESULTS.md and results.json
agents_may_not:
  - modify zeta or ontology or harness or lean
  - use the reserved enclosure vocabulary of zeta/rigor.py anywhere in this directory
  - promote this hunt's claim into repo-level status files
  - treat a float measurement as a decided fact

Vocabulary contract

Measured: one float route. Observed: a pattern in measured data. Decided: an Arb or interval enclosure whose exact endpoints settle a sign or an integer count, stated with backend and precision. The composite headline takes the weakest grade of its steps and says so: the lower bound is decided modulo Dobner's Theorem 1; the upper bound is a cited theorem applied to a decided strip constant.