Answers are grounded in this paper and cite their sources.
For the primes p ≡ a (mod q) the explicit formula makes the weighted series∑p log p p−1/2e−iω log p, minus its main term, a line spectrum: one line of equal heightat every zero 12 + iγ of every L(s, χ) modulo q, ζ included, with the phase χ(a). We computethese spectra for the primes in [10^6, 10^10] and in [10^8, 10^11] (4.5 × 10^8 and 4.1 × 10^9 primes)for q = 3, 5, 7 and every residue class, and compare them with the 2117 zeros of ζ and of thenine primitive characters modulo 3, 5, 7 up to height 300, which we compute independentlyand verify against the LMFDB: the agreement is 0.2% in rms over 14 < ω < 280, and thecharacter transform over the residue classes isolates each L-function. We then split eachseries by the gap class of the prime, the residue modulo 2q2 of the gap to the next primeof the same residue class, which in two previous papers we showed to be governed by asieved Hardy–Littlewood model S29 and to respond to the local density of primes withcomputable elasticities εs. We prove that in the sieve model a density perturbation of anywavelength long compared with the mean gap is transmitted into gap class s with the gain(1 + εs)ps, with an explicit frequency response εs(ω) whose real part departs from εs only atsecond order in ω/ρ (Theorem 3.1); hence, if the primes follow the model, the zero-drivenfluctuation enters the spectrum of every gap class with that gain, independently of thefrequency γ and of the L-function that drives it. We also prove a bound showing that thiscannot be seen line by line: in a single gap class the signal-to-noise ratio of one zero is atmost (1 + εs)√2ps/((1 − ps)φ(q)) < 1 at any height, because the multinomial noise of theclass split and the contribution of a zero scale alike as √x; the test is the regression ofthe class spectrum on the total spectrum over bands of frequency containing thousands oflines. With the predictions frozen and deposited with SHA-256 digests before each dataset was read, the measured elasticities agree with S29 band by band from γ = 10 to 3000(χ2 = 88 and 95 on 120 in the two runs; Cramér-in-class 668 and 278), pooled over all linesto ±0.02–0.05 (χ2 = 7.6 and 10.6 on 15; Cramér 587 and 193), and are the same on thelines of ζ and on those of each non-principal L(s, χ). The zeros of the L-functions thereforereach the finest arithmetic statistic of the primes we know of, the class of the gap to thenext prime, through the single scalar predicted by the sieve. Nothing here bears on the realpart of the zeros, and we say why.
No takes yet. Share an insight, caveat, or question.
Cristoni et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: