WorksMathematics
A narrower admissible 234 360 000-tuple and H_5 <= 4 891 341 734
Abstract
Let H_m = liminf (p_{n+m} - p_n). Kapila proved DHL[234360000,6] and deduced H_5 <= 5040082836, the diameter of the tuple formed by the 234360000 smallest primes larger than 234360000. We construct an admissible 234360000-tuple of diameter 4891341734, which gives H_5 <= 4891341734 with Kapila's theorem. The tuple consists of +-1 and of +-q for primes q in a symmetric window, with 1746831 primes removed; its admissibility is checked by two independent programs.
Result
For let , where is the -th prime. A set of integers is admissible if for every prime it misses at least one residue class modulo ; it suffices to check the primes . The statement asserts that every admissible -tuple has infinitely many translates containing at least primes, and it implies for every admissible -tuple . Following Maynard [2], the Polymath project [3] obtained , Stadlmann [6] improved this to , and Roeling [5] to . Kapila [1, Theorem 1.1] proved and obtained . This number is the diameter of the tuple of the smallest primes larger than [1 Table 3], which is admissible because none of its elements is divisible by a prime ; we recomputed this diameter exactly (ancillary file previous_record_width.py).
Theorem 1. There is an admissible -tuple of diameter .
Corollary 1. Assume , which is proved in [1]. Then .
The only new ingredient is the tuple: the bound relies on the statement of [1], which we use as stated.
The tuple
Let , and let be the set of the primes listed as successive differences in two ancillary files, to be read one after the other: removed_gaps_1.csv, then removed_gaps_2.csv; they all lie in . Put
Then , and is prime and lies in , so .
Lemma 1. is admissible if, for every prime with and , there is an integer with , , such that no satisfies .
Proof. If , no element of is divisible by , since the elements are and with prime and . If and , the only element of divisible by is $p itself, which is not in ; so the class $0$ is missed. Otherwise, by hypothesis no element of lies in the class mod , because or would give . For some class is missed trivially. □
How was chosen. We started from and treated the primes with in decreasing order: we searched the smallest with both classes mod empty, and if there was none we removed . Removing elements keeps the classes found earlier empty, so the final set satisfies the hypothesis of Lemma 1; consists of the elements of absolute value at most . This is a variant of the sieve of Hensley and Richards, which sieves a symmetric interval by the classes mod for increasing primes (see [4]); the difference is that for most primes we do not sieve the class but exhibit an empty pair of classes , so that these primes stay in the tuple. The removed primes are those for which the expected number of elements per class, about , is too large for an empty class to exist or to be found within the search budget; they are all smaller than .
Verification
verify_tuple.py(a C program embedded in a Python script, compiled with gcc) builds with a sieve of Eratosthenes on bytes, removes , checks and , and then, for every prime , scans the class directly (all multiples of up to ) and, if it is occupied, searches a pair with that is empty, again by direct scanning. It reports primes, the class missed for of them and a pair for the other , and the diameter . It needs about 3 GB of memory and three minutes.build_tuple.pyreproduces the construction and writes the list of the classes found for each prime (about 178 MB, not included).verify_classes_numpy.py, written independently in Python with numpy, rebuilds with its own sieve and checks this list: every prime has exactly one entry, every listed pair is empty, and the cardinality and the diameter are as stated.For the same construction gives a tuple of diameter $1,449,286$; both programs accept it, and a direct computation of all residues modulo every prime confirms its admissibility.
verify_tuple.pyrejects the tuple in which one removed prime is put back and another element removed (no class is then empty modulo 1021, by exhaustive search), andverify_classes_numpy.pyrejects a certificate in which one class is altered, both for and for $k = 234,360,000 in the second case.
Remark 1. The diameter is about . For the values of used in [1] for (2 355, 111 300, 5 129 500) the same construction gives the diameters 25 562, 1 449 286 and 90 027 092, larger than those of [1]; so we do not improve these bounds.
Acknowledgements and use of AI
The mathematics, the computations and the writing were produced with Claude (Anthropic) under the author’s direction. The bound depends on the statement DHL[234 360 000,6] of [1]. The result has not been examined by a professional mathematician.
Independent researcher, Meaux, France
URL: https://orcid.org/0009-0009-1526-1178
References
- [1]A. Kapila, Improved bounds on Hₘ for 2 ≤ m ≤ 5, Hexagon preprint, hexagon:2610.00419v1 (2026).
- [2]J. Maynard, Small gaps between primes, Ann. of Math. 181 (2015), 383–413.arxiv.org/abs/1311.4600
- [3]D. H. J. Polymath, Variants of the Selberg sieve, and bounded intervals containing many primes, Res. Math. Sci. 1 (2014), Art. 12; arXiv:1407.4897.arxiv.org/abs/1407.4897
- [4]A. V. Sutherland, New bounds on gaps between primes, lecture slides, https://math.mit.edu/~drew/PrimeGaps.pdf.
- [5]M. P. Roeling, NeuralCert: certified computational discovery of extremal mathematical constructions, arXiv:2609.30296 (2026).arxiv.org/abs/2609.30296
- [6]J. Stadlmann, On primes in arithmetic progressions and bounded gaps between many primes, Adv. Math. 468 (2025), Art. 110190; arXiv:2309.00425.arxiv.org/abs/2309.00425