Erdős Problem #381
Editorial depth: Reviewed explanatory article — evidence qualified.
Snapshot classification: disproved (not automatically the live result).
Subject areas: number theory, divisors.
Read this status correctly
The status above is reproduced from GCL's immutable external-catalogue snapshot. It is not a new claim of current mathematical resolution, a source-semantic review, an independent proof check, or GCL certification. A published proof and a GCL replay are separate events.
The question in plain language
For a positive integer \(n\), let \(\tau(n)\) be its number of positive divisors. The integer \(n\) is highly composite when every smaller positive integer has strictly fewer divisors. These are record-holders for the divisor-count function. If \(Q(x)\) counts the highly composite numbers at most \(x\), Problem 381 asks whether, for every fixed integer \(k\geq 1\), there is a positive constant (depending on \(k\)) such that
for all sufficiently large \(x\). In words: does the supply of record-holders eventually outrun every fixed power of the logarithm?
The quantifiers matter. This is not merely the claim that there are infinitely many highly composite numbers, nor that \(Q(x)\) beats one particular power. It asks for one lower bound for each \(k\), with the exponent allowed to grow without bound. The source statement uses \([1,x]\) and strict comparison with every \(m<n\); this article keeps those conventions.
A small example
The divisor-count identity gives an exact finite check. If \(n=\prod p_i^{a_i}\), then \(\tau(n)=\prod(a_i+1)\). Direct enumeration of divisors for \(1\leq n\leq 100\) gives the record pairs
Thus the finite computation finds \(Q(100)=9\). In particular, \(60=2^2\cdot3\cdot5\) has \(\tau(60)=(2+1)(1+1)(1+1)=12\), while the largest divisor count among \(1,\ldots,59\) is \(10\), attained at \(48\). Therefore \(60\) is a highly composite number under the exact strict definition.
This check is reproducible with integer arithmetic: for each \(n\), test each integer \(d\) from \(1\) through \(\lfloor\sqrt n\rfloor\), count both \(d\) and \(n/d\) when distinct, and retain \(n\) exactly when its count exceeds the previous maximum. It confirms terminology and a finite initial segment only. It does not estimate \(Q(x)\) asymptotically, prove any theorem about all highly composite numbers, or decide the quantified statement by itself.
What is known
The maintained Erdős Problems entry, accessed 13 October 2026, states that the answer is no. It attributes a lower bound \(Q(x)\gg (\log x)^{1+c}\) for some \(c>0\) to Erdős and the negative resolution to Jean-Louis Nicolas. The entry cites [Er44], P. Erdős, “On highly composite numbers,” Journal of the London Mathematical Society 19 (1944), 130–133, and [Ni71], Jean-Louis Nicolas, “Répartition des nombres hautement composés de Ramanujan,” Canadian Journal of Mathematics 23 (1971), 116–130.
The primary Nicolas article is available through Cambridge Core, with DOI 10.4153/CJM-1971-012-6. I inspected the primary PDF. Its Theorem 4, printed p. 127, states a fixed-power upper bound of the form \(Q(X)=O((\log X)^{1+c})\), with the constant inherited from the preceding argument. Its Theorem 5 on the same page records the lower bound of Erdős type. The upper bound is enough to refute the Problem 381 question: choosing any fixed \(k>1+c\) makes the proposed lower bound incompatible with Nicolas’s upper bound for large \(X\).
The phrase \((\log x)^{O(1)}\) on the maintained page is a status-level summary, not a replacement for the paper’s precise theorem. The primary source supports “disproved” through a bounded exponent; it does not justify silently assigning a new numerical value of \(c\) or claiming that every detail of Nicolas’s proof has been formalized.
GCL's current state
The original MATH-PROGRAMME issue #1651 supplies a frozen catalogue packet. At teorth/erdosproblems commit 536408716724a2402bdfa7c1d764bf83d64a8c89, the number: "381" record has Git blob 4e60263a80af6207aca494d2e1388e44fa372889, informal state disproved, last update 2025-08-31, formal state unformalized, and formalized: no. The live page agrees on the mathematical status and currently says “Formalised statement? No”; its history exposes a 2025-10-20 statement revision, so the live page is identified separately rather than treated as the frozen packet.
The issue also pins the admitted FormalConjectures provider upstream to commit 85f863718beeec7b58a3a1926ee92e3472bc2020 and says no statement file is present. A direct check of the pinned ErdosProblems tree finds no 381.lean; the current checked upstream head 817d8a74e6bcd505ac8407388cc12ac3cfda3d0d also has no such file. The provider record is public metadata, not a mathematical certificate. No Lean theorem, sorry body, replay, or GCL certification is being asserted.
The programme’s existing triage class is OUTSIDE_STRICT_OPEN_TRIAGE. V2 ledger state, rather than this prose or catalogue metadata, decides accepted coverage. This article records source evidence and boundaries only.
What remains
The external resolution is clear enough for a source-locked editorial record, but a future formalization would still need a faithful interface. It would define highly composite numbers over positive naturals, define \(Q(x)\) as a finite count, and encode the exact quantifier “for every \(k\geq1\)” before formalizing either Nicolas’s upper bound or the negation of the original question. The asymptotic notation also needs an explicit constant-and-threshold convention.
A bounded successor task is therefore to formalize only the finite lemma illustrated above: prove the divisor-count formula, prove that \(60\) is highly composite, and compute the record list through \(100\) by a transparent finite argument. Acceptance should require exact agreement with the listed nine records; a mismatch in the definition of \([1,x]\), divisor counting, or strict record comparison falsifies the task. This is useful interface work, not a proof of the asymptotic theorem. A larger task would require a source-faithful translation and specialist mathematical review of Nicolas’s Theorem 4.
Sources and claim boundaries
The primary mathematical references are Erdős [Er44] and Nicolas [Ni71], with the latter providing the inspected negative-resolution theorem and page location. The maintained page is a secondary status surface; the frozen YAML record is a provenance lock; the FormalConjectures snapshot is an absence check. No claim is made here about later papers, hidden formal files, unpublished proof reports, or Lean compilation.
The original SOURCE RESULT/1 citation is authenticated original RESULT/1. No future issue comment, acceptance, publication, award, or certification is fabricated in this article. The finite calculation is independently reproducible but local in scope. Accepted coverage and any subsequent disposition remain governed by the programme’s V2 ledger and accountable coordination.
Catalogue and evidence provenance
- Pinned catalogue source: teorth/erdosproblems at
536408716724. - GCL issue: #1651 — protected issue/assignment identity; its labels and results are live, not reproduced automatically here.
- GCL triage: OUTSIDE_STRICT_OPEN_TRIAGE.
- Formal-statement coverage: No statement was confirmed in GCL's pinned formal-provider snapshot. Presence does not mean the statement or proof has been checked.
- External mathematical status: consult current problem page and its dated bibliography.