The fractional parts of gamma ln p/2pi beyond the first order: two primes on a torus, the explicit formula with a smooth window, a comb of resonances in the variance of sum x^(i gamma), and the sign of chi_D(p) for Dirichlet L-functions - a reproduci
Sixth study of the series (study I: doi:10.5281/zenodo.23815328; study II: doi:10.5281/zenodo.23928017; study III: doi:10.5281/zenodo.23991165; study IV: doi:10.5281/zenodo.24094318; study V: doi:10.5281/zenodo.24256378). At finite height the fractional parts {gamma ln p/2pi} of the ordinates of the Riemann zeros are not uniformly distributed: by Landau's formula their density has a dip at 0 of relative depth about 2w ln p/sqrt p (Ford and Zaharescu). Study V checked this first-order effect for single primes; this study takes it further in two parts, each with rules recorded before the computation. Part I: (A) Two primes at once: on 2,001,052 zeros of zeta the density on the torus ({gamma ln p/2pi}, {gamma ln q/2pi}) is the sum 1 + r_p + r_q, as the theorem of Ford, Meng and Zaharescu implies; the amplitude of a mixed term is between -0.036 and 0.041 for six pairs (errors 0.037-0.046), 21-27 errors away from the product of independent phases. (B) With a Gaussian window the explicit formula holds for every harmonic x^(i gamma): zero side and prime side agree to 8.5e-10 of the number of zeros on 162 points, the precision of the zero table, against 1.6e-4 to 8.6e-2 for a sharp window. (C) Second order: the variance V(x) of the windowed sum over 52,498 windows at t ~ 1e8 and 3e10 follows the diagonal sum over prime powers for ln x below ln(T/2pi) and the two-point function of Bogomolny and Keating above (18 numbers, largest |z| 1.68). Near x = T/2pi it shows a comb of narrow resonances at x = (T/2pi) A/B, the off-diagonal spikes of the form factor of study II; at x = T/2pi itself GUE without arithmetic is 18.6 and 46.1 errors away from the data. (D) Rational multiples alpha = (b/q) ln p/2pi give a density of period 1/q (smallest p 0.14), and on 446,224 newly computed zeros of eight quadratic Dirichlet L-functions up to height 36,000 the dip carries the sign of chi_D(p): agreement in all 32 combinations of D and p (smallest p 0.22), the opposite sign excluded (chi2 >= 1,518 for 50 bins). Part II, a supplementary measurement: (E) on 626,717 further zeros of the same eight L-functions between heights 36,000 and 80,000, computed after the rules were fixed, the sign of chi_D(p) holds again in all 32 combinations (smallest p 0.27; opposite sign chi2 >= 1,621), the first Fourier coefficient equals -w chi_D(p) ln p/sqrt p with the w of the new band to 2.1e-5 without a free parameter, the explicit formula with a Gaussian window holds to 1.1e-11 of the number of zeros on 640 points, and near x = |D|T/2pi the variance of the harmonics of a single L-function has the comb predicted by the two-point function of Bogomolny and Keating for Dirichlet L-functions (2013), with the peaks rearranged by the primes dividing D (mean z^2 over 201 points 0.08-0.24, against 48.5-78.5 for the formula of zeta with only the density changed, D = -3, -4, 8). (F-H, descriptive) The comb of zeta is decomposed prime by prime and followed at the window widths H = 10, 20 and 40; in the corner of four primes the number of zeros follows the sum of the densities and not the product; the structure factor of the zeros consists of resolved peaks at ln p^m below the resolution boundary tau* = ln sigma/L and of the form factor of GUE above it; and the number variance stays at Berry's saturation value up to l = 1e4, with small residuals that the two-point function of Bogomolny and Keating removes. All twelve hypotheses (seven in Part I, five in Part II) are SUPPORTED. The formulas tested in A, B and D and in the first three hypotheses of E are theorems or direct consequences of them; the two-point functions tested in C and in the comb of E are conjectures. What is added is data at these heights with null models and errors, zeros that did not exist when the rules were written, and a resolved picture of the comb. Two independent recomputations (one for each part) that share no code with the main scripts are included. Agreement is numerical support, not a proof; nothing here bears on the Riemann hypothesis. Article in English and Czech (PDF, DOCX, LaTeX, Markdown), 14 figures, every number of the article in two JSON files, the rules recorded beforehand with their dated addenda, the manifests with the SHA-256 of every window of zeros of the L-functions, and the scripts as they were run. Text, figures and data CC BY 4.0, code MIT.
Authors
- David Hess
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-19
- DOI
- https://doi.org/10.5281/zenodo.24381945
- Primary Topic
- Analytic Number Theory Research
- Type
- preprint