Entry point

The arXiv paper, in two pages

For a reader with ten minutes: the abstract, the one statement the paper claims as a theorem, and the table that tests it. Everything below is copied from the paper's source text (v1.6, 4 August 2026); the notation is transliterated to HTML and nothing is added or softened. The 98-page PDF is the version of record.

One thing this page is behind on, said here rather than discovered. A newer source, v1.8 of 4 September 2026, exists and is not yet built. It changes exactly the two things a ten-minute reader sees first: the title's third clause is translated out of this programme's private vocabulary, and the abstract is reframed to lead with the census and the measurements and to price the sum rule as what it is — an elementary consequence of Littlewood's lemma under two normalisations, written out because a census needs one. The mathematics is unchanged; no printed number moved, and every result statement is byte-identical to v1.6's. This page tracks the built form, so it will be regenerated from v1.8 when v1.8 is built — and it is flagged now because reading a title the paper no longer carries is exactly the kind of thing this page exists to prevent.

Title and disclosure

A Smoothing Identity, a Displacement Sum Rule, and Finite-Height Laws for the Value Register of the Riemann Zeta Function. Ondřej Dvořák, independent researcher, Děčín, Czech Republic.

Abstract

Let S_N(s) = Σ_n≤N n^−s and χ the factor of the functional equation ζ(s) = χ(s)ζ(1−s). Call the integer interval G_k = (x_k+1, x_k) between consecutive saddle abscissae x_ν = t/2πν the k-th packet, and the average of S_N over it the packet centroid ⟨S⟩_k. A smoothing identity (Theorem 1 with Lemma 2.2; machinery under two labelled classical inputs, no priority claimed) gives ζ − ⟨S⟩_k = χ(s)P_k(s) + χ(s)·O(t^−1/2+ε), P_k(s) = Σ_m≤k m^s−1, uniformly on compact σ-subsets of (0,1). Reflected, the zeros of ζ(s) − χ(s)P_k(s) are the zeros of the Hurwitz zeta ζ(w, k+1). For these we prove a Littlewood-type displacement sum rule (Proposition 2): for k ≥ 2 and T₁ above a threshold in k alone, evaluated here for k ≤ 3,

Σ_{vρ∈(T1,T2]} (u_ρ − ½) = ((T₂ − T₁)/4π)·log(k+1) + O_k(log T₂),

confirmed by a census whose pre-recorded deficits (four windows) close to ≤ 0.6 %. We report finite-height measurements with window, bar and instrument: 1-point displacements of ζ over seven windows in t ∈ [200, 100090], window means fit c = 1.023 ± 0.011; spacing of ζ(w,2) and ζ(w,3) zero ordinates against matched controls; exact coil-regime laws; an aperture-crop law for close zero pairs; a resonance register; a threshold ladder with closed-form constant; a conditional Kac–Rice crossing-density theorem at fixed σ > 1; and costings of two positivity routes (Li; restricted-support Weil), both negative. Weil and Li positivity are equivalences of the Riemann Hypothesis, not reductions of it. No result in this paper decides the location of any zero of ζ, and nothing here is evidence for or against the Riemann Hypothesis in either direction.

The one statement claimed as a theorem (§3)

Write F_k(w) = ζ(w) − S_k(w) = ζ(w, k+1), the Hurwitz zeta at integer parameter k+1. Its roots are ρ = u_ρ + i v_ρ, strip family and outer family together, and

D_k(T₁,T₂) := Σ_{Fk(ρ)=0, vρ∈(T1,T2]} (u_ρ − ½).

What is claimed, and where the boundary lies — the paper's own words. The case k = 1 is not claimed. The k = 1 statement is the classical displacement sum rule for the 1-points of ζ, available by Selberg's route and printed for the Hurwitz zeta by Steuding and by Garunkštis–Steuding; the 2^s(ζ−1) normalisation that produces it is likewise theirs. What is claimed is the law at k ≥ 2 — the value log(k+1) of the constant for ζ(·, k+1) at integer parameter k+1 ≥ 3, together with the proof route — and it is read throughout as "new as far as an admittedly incomplete search reaches".

How the proof goes, and how much of it is classical. Littlewood's lemma is applied to G(w) = (k+1)^wF_k(w) on a rectangle to the right of the line u = ½, and to H(w) = χ(w)F_k(1−w) on the same rectangle, whose zeros are the reflections of the roots to the left of it. On the line, log|G| = log|F_k| + ½ log(k+1) while |H| = |F_k| exactly, so the two line integrals of log|F_k| cancel identically and the whole displacement budget is the normalisation asymmetry between the two ends of the strip. The paper says of this (Remark 3.4) that the rule is "free": the method family is Littlewood's lemma plus a normalising factor, classical and attributed; what the paper supplies is the confinement of the outer family (Lemma 3.1), the left anchor from the functional equation (Lemma 3.2), the edge and counting estimates written out in full (Lemma 3.3), and the evaluated thresholds.

The table that tests it (§4.3)

The census was run in two stages. The deficits the second stage had to recover were computed from Proposition 2 and recorded before the patch census was run — the author's own internal record, with no external timestamp, which is why the paper does not call it pre-registration. The four deficits are deterministic consequences of the proposition given the frozen first-stage census, so they could not have been tuned to the outcome.

Residuals are taken against the two-decimal D_k shown; against the unrounded values (275.7945, 43.7124, 437.1239, 55.1589) they read −0.39, +0.07, −0.23, +0.35, moving no percentage at the printed precision. A fifth window, k = 1 on [1000, 1500], closed without any patch at −7.3 % of D₁ = 27.6, inside an assumed tolerance of ±O(log T) whose constant the paper does not evaluate; it is printed as found rather than as a success. A second, independent closure on the count of roots matches the density integral to under one root on the three windows where both sides print (4810.20 against 4810; 4487.54 against 4487; 310.35 against 310).

What the closure does and does not establish, in the paper's words. The match between the deficits recorded in advance and the census outcome is independent confirmation that the census recovers what Proposition 2 predicts, rather than a calibration artefact. The proposition's correctness rests on the proof of §3, not on this match; conversely the census is not a proof of anything the proof does not already give.

What the rest of the paper is

Measurements, each printed with the height window it was taken on, the bar it was tested against and the instrument that produced it (§§5–11); four elementary propositions and two lemmas claimed at the strength stated at their sites; a conditional theorem; and a positioning section (§13) that lists every object the programme reached independently and then found to be somebody else's. Theorem 1, the smoothing identity, enters as machinery and is not claimed: two of its three components are located in print and named.

The referee history

Thirteen blind referee rounds, each by a fresh model instance sealed to the draft alone, plus one cross-model control round; every printed constant re-derived centrally rather than accepted from a report. The last round returned no mathematical error in any proof and no printed constant that failed to reproduce. One reporting error — a fit-model label carried over wrongly from the source paper, invisible to referees sealed to the draft — was found afterwards by a non-blind cross-read against the source record and corrected at v1.6; no measured quantity moved.