\documentclass[11pt]{amsart} \usepackage[margin=1.1in]{geometry} \usepackage{amsmath,amssymb,amsthm} \usepackage[hidelinks]{hyperref} \newtheorem{theorem}{Theorem} \newtheorem{corollary}{Corollary} \newtheorem{lemma}{Lemma} \theoremstyle{remark} \newtheorem{remark}{Remark} \begin{document} \title{A narrower admissible $234\,360\,000$-tuple and $H_5\le 4\,891\,341\,734$} \author{David Niedbala Giraudin} \address{Independent researcher, Meaux, France} \urladdr{\href{https://orcid.org/0009-0009-1526-1178}{https://orcid.org/0009-0009-1526-1178}} \date{15 November 2026} \subjclass[2020]{11N05, 11N35, 11Y16} \keywords{gaps between primes, admissible tuples, Maynard--Tao sieve} \begin{abstract} Let $H_m=\liminf_{n\to\infty}(p_{n+m}-p_n)$. Kapila proved $\mathrm{DHL}[234\,360\,000,6]$ and deduced $H_5\le 5\,040\,082\,836$, the diameter of the tuple formed by the $234\,360\,000$ smallest primes larger than $234\,360\,000$. We construct an admissible $234\,360\,000$-tuple of diameter $4\,891\,341\,734$, which gives $H_5\le 4\,891\,341\,734$ with Kapila's theorem. The tuple consists of $\pm1$ and of $\pm q$ for primes $q$ in a symmetric window, with $1\,746\,831$ primes removed; its admissibility is checked by two independent programs. \end{abstract} \maketitle \section{Result} For $m\ge1$ let $H_m=\liminf_{n\to\infty}(p_{n+m}-p_n)$, where $p_n$ is the $n$-th prime. A set $\mathcal H$ of $k$ integers is \emph{admissible} if for every prime $p$ it misses at least one residue class modulo $p$; it suffices to check the primes $p\le k$. The statement $\mathrm{DHL}[k,m+1]$ asserts that every admissible $k$-tuple has infinitely many translates containing at least $m+1$ primes, and it implies $H_m\le\operatorname{diam}\mathcal H$ for every admissible $k$-tuple $\mathcal H$. Following Maynard \cite{May}, the Polymath project \cite{Pol14} obtained $H_5\le 80\,550\,202\,480$, Stadlmann \cite{Sta} improved this to $77\,510\,685\,234$, and Roeling \cite{Roe} to $14\,541\,349\,288$. Kapila \cite[Theorem~1.1]{Kap} proved $\mathrm{DHL}[234\,360\,000,6]$ and obtained $H_5\le5\,040\,082\,836$. This number is the diameter of the tuple of the $k=234\,360\,000$ smallest primes larger than $k$ \cite[Table~3]{Kap}, which is admissible because none of its elements is divisible by a prime $p\le k$; we recomputed this diameter exactly (ancillary file \texttt{previous\_record\_width.py}). \begin{theorem}\label{thm:tuple} There is an admissible $234\,360\,000$-tuple of diameter $4\,891\,341\,734$. \end{theorem} \begin{corollary} Assume $\mathrm{DHL}[234\,360\,000,6]$, which is proved in \cite{Kap}. Then $H_5\le 4\,891\,341\,734$. \end{corollary} The only new ingredient is the tuple: the bound relies on the statement $\mathrm{DHL}[234\,360\,000,6]$ of \cite{Kap}, which we use as stated. \section{The tuple} Let $P_0=50\,100$, $a=2\,445\,670\,867$ and let $R$ be the set of the $1\,746\,831$ primes listed as successive differences in two ancillary files, to be read one after the other: \texttt{removed\_gaps\_1.csv}, then \texttt{removed\_gaps\_2.csv}; they all lie in $(P_0,\,37\,847\,053]$. Put \[ A=\{1\}\cup\{q\ \text{prime}:\ P_0P_0$ and $p\notin R$, there is an integer $c$ with $1\le c\le 20\,000$, $2cP_0$. If $p>P_0$ and $p\in R$, the only element of $\{1\}\cup\{\text{primes}\}$ divisible by $p$ is $p$ itself, which is not in $A$; so the class $0$ is missed. Otherwise, by hypothesis no element $\pm m$ of $\mathcal H$ lies in the class $c\bmod p$, because $m\equiv c$ or $-m\equiv c$ would give $m\equiv\pm c$. For $p>234\,360\,000=|\mathcal H|$ some class is missed trivially. \end{proof} \emph{How $R$ was chosen.} We started from $\{1\}\cup\{q\ \text{prime}: P_0P_0$ in decreasing order: we searched the smallest $c\le 20\,000$ with both classes $\pm c\bmod p$ empty, and if there was none we removed $p$. Removing elements keeps the classes found earlier empty, so the final set satisfies the hypothesis of Lemma~\ref{lem:adm}; $\mathcal H$ consists of the elements of absolute value at most $a$. This is a variant of the sieve of Hensley and Richards, which sieves a symmetric interval $[-x,x]$ by the classes $0\bmod p$ for increasing primes $p$ (see \cite{Sut}); the difference is that for most primes $p\le k$ we do not sieve the class $0$ but exhibit an empty pair of classes $\pm c$, so that these primes $\pm p$ stay in the tuple. The removed primes are those for which the expected number of elements per class, about $|\mathcal H|/p$, is too large for an empty class to exist or to be found within the search budget; they are all smaller than $3.8\cdot10^{7}$. \section{Verification} \begin{itemize} \item \texttt{verify\_tuple.py} (a C program embedded in a Python script, compiled with \texttt{gcc}) builds $A$ with a sieve of Eratosthenes on bytes, removes $R$, checks $|\mathcal H|=234\,360\,000$ and $a\in A$, and then, for every prime $p\le 234\,360\,000$, scans the class $0$ directly (all multiples of $p$ up to $a$) and, if it is occupied, searches a pair $\pm c$ with $c\le20\,000$ that is empty, again by direct scanning. It reports $12\,868\,673$ primes, the class $0$ missed for $1\,751\,974$ of them and a pair $\pm c$ for the other $11\,116\,699$, and the diameter $4\,891\,341\,734$. It needs about $3$ GB of memory and three minutes. \item \texttt{build\_tuple.py} reproduces the construction and writes the list of the classes $c$ found for each prime (about $178$ MB, not included). \texttt{verify\_classes\_numpy.py}, written independently in Python with \texttt{numpy}, rebuilds $A$ with its own sieve and checks this list: every prime $P_0