AI & Computingpreprint2026-08-23

A Level-Four Sum Rule and the Resolution of the Small-Gap Deficit of the Riemann Zeros: A High-Precision Computational Report

Open access0 citations

Abstract

This computational report presents three novel findings regarding the local spacing and invariant structure of the nontrivial zeros of the Riemann zeta function, derived from high-precision evaluations at heights $\gamma$ ranging from $1.2 \times 10^4$ to $10^8$. Working within the mathematical framework of the jet tower $L^{(n)}=\mathbb{C}[b]/(b^{n})$, we report the following results: A Level-Four Sum Rule: We identify a sum rule linking the level-three one-sum ($S_1$) and the level-four zero-sum ($X$), demonstrating that $\mathbb{E}[S_1^2] = \mathbb{E}[X] - \pi^2/3$. The data reveals that finite-height arithmetic corrections enter both levels identically and cancel out exactly. This combination represents the first quantity in the tower that captures three-point correlations, moving beyond Montgomery's regime. Resolution of the Small-Gap Deficit: The previously reported scarcity of small gaps between zeros is proven to be strictly a finite-height effect. By comparing against the exact sine-kernel value, we show that the deficit decays according to $1 - A/A_{sine} = c/L^p$ with an exponent of $p \approx 2$ (specifically $c=17.49$ at $p=2$), statistically excluding the naive $1/L$ correction model with $\Delta\chi^2=10.0$. A Local Lehmer Criterion: We introduce a functional that certifies a Lehmer pair (anomalously close zeros) using solely two evaluations of the level-four jet, without requiring knowledge of any other surrounding zero. Using this criterion, a new tight pair is discovered at $u = 21137307.8265$ with $\Lambda=0.00214$. Additionally, the report documents four critical numerical traps encountered in high-height zeta computations, including the exact $3/2$ tail index (infinite variance) of the variable $X$, and details the mitigation of phase noise using a truncated Riemann-Siegel formula. All computations were performed and validated in MATLAB.

// Source

View paper (DOI)Open access versionOpenAlexZenodo (CERN European Organization for Nuclear Research)Published 2026-08-23

Authors: Jorge Vicente Romero