Skip to content

Erdős Problem #401

Editorial depth: Reviewed explanatory article — evidence qualified.
Snapshot classification: proved (not automatically the live result).
Subject areas: number theory, factorials.

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

Let \(p_r\) be prime number \(r\), and form the integer

\[ N_{n,r}=n!\,2^n3^n\cdots p_r^n. \]

Erdős Problem 401 asks whether there is one function \(f\) that grows without bound with \(r\), while for every \(r\geq1\) there are infinitely many positive integers \(n\) for which one can find positive integers \(a_1,a_2\) satisfying

\[ a_1+a_2>n+f(r)\log n \]

and \(a_1!a_2!\) divides \(N_{n,r}\). In words, the two factorials must fit inside a number built from \(n!\) and the first \(r\) primes, but the two factorial indices together must exceed \(n\) by a margin that becomes larger, on the logarithmic scale, as more primes are allowed.

The original source does not spell out all quantifiers. The maintained Erdős Problems entry adopts the “infinitely many \(n\)” reading, and its discussion explains why. This article follows that registered formulation; it does not silently promote the stronger “all sufficiently large \(n\)” variant.

A small example

A finite calculation illustrates the divisibility mechanism. Take \(r=1\), so the extra prime factor is \(2^n\), and choose \(n=36\), \(a_1=8\), and \(a_2=34\). Then

\[ \frac{36!\,2^{36}}{8!\,34!} =\frac{35\cdot36}{8!}\,2^{36} =\frac{1260}{40320}\,2^{36} =2^{31}. \]

Thus \(8!34!\) divides \(36!2^{36}\) exactly, and \(8+34=42>36\). This is reproducible integer arithmetic: the quotient is the displayed integer \(2^{31}\), with no numerical approximation. It does not prove the problem. It treats only \(r=1\) and one \(n\), gives no function \(f\), and does not show that suitable \(n\) occur infinitely often. It also does not show that the six-unit margin is larger than the unknown quantity \(f(1)\log 36\).

What is known

The primary historical source is P. Erdős and R. L. Graham, Old and New Problems and Results in Combinatorial Number Theory (1980), p. 78. The scan is available from Ronald Graham’s papers; the relevant page literally asks for \(a_1+a_2>n+d_r\log n\) with \(d_r\to\infty\), and asks whether the displayed quotient is an integer. The downloaded PDF had SHA-256 0cbf0c32f0ab1e1c71db5121a88bac905bf976c4a6ab6bb6d7d9cf9ddd184ed3.

The maintained page records an affirmative resolution attributed to Barreto and Leeham, using an argument essentially parallel to their solution of Problem 729. Its discussion also records a disproof of the alternative “for all large \(n\)” interpretation: with \(n=p_{r+1}^k-1\), the divisibility condition is reported to force \(a_1+a_2\leq n+2p_{r+1}\). That is evidence about a distinct stronger variant, not a contradiction to the statement concerning infinitely many \(n\).

The public discussion links an informal GPT-5.2 Pro proof and a long Lean development. The hosted source describes a construction with \(n=2m\), \(a_1=m+k\), \(a_2=m\), a prime-by-prime valuation reduction, Kummer-style carry counts, and a probabilistic choice of \(m\) in intervals \([M,2M]\). These are reported proof materials. This audit did not replay the Lean build or certify the mathematics independently.

GCL's current state

The frozen source packet for issue #1671 identifies teorth/erdosproblems commit 536408716724a2402bdfa7c1d764bf83d64a8c89. Its data/problems.yaml entry has full-file Git blob 4e60263a80af6207aca494d2e1388e44fa372889, catalogue state proved, formal state Lean, and statement metadata formalized: yes; the recorded status update is 2026-01-11. These are catalogue records, not GCL mathematical certification.

At the live page read on 2026-10-13, the displayed formulation is the same, the page says “PROVED (LEAN),” links a formalized statement, reports 21 discussion comments, and reports no proof exposition or proof claim. The page was last edited 12 January 2026. Its status and the catalogue record are therefore consistent at the level of reported external metadata, while remaining separate from acceptance.

The admitted FC-GDM-85F86371 snapshot pins Google DeepMind’s formal-conjectures at commit 85f863718beeec7b58a3a1926ee92e3472bc2020, but that exact snapshot has no FormalConjectures/ErdosProblems/401.lean. A later file was added at commit 0cfdea64bb89881ce06f41c303e2009ec6c66ea5, blob 4d72ebef0968705010b562f62d8373afccfd2eb0; its theorem body is by sorry, so it is statement infrastructure, not a checked proof. The hosted proof source inspected here is Erdos401.lean at plby/lean-proofs commit 1d7b3f00780b85ed0462e79a1cd5650ee9055655, blob 0d94c74d806d14a4be6ecfe44a363eab771457b4; its own footer reports kernel axioms, but that external report was not replayed here.

What remains

The next useful bounded step is a semantic and expository audit of the hosted proof. It should align the source’s \(P_r\), prime indexing, positivity assumptions, and “infinitely many \(n\)” conclusion with the registered statement, then explain the valuation and probabilistic lemmas in ordinary mathematical prose. It should separately test the reported counterexample to the variant for all large \(n\). A successful compilation would be evidence about that exact source and toolchain, not a GCL certificate.

The V2 ledger alone decides accepted coverage. This article records source evidence and boundaries; it does not decide acceptance, publication, priority, or mathematical certification. Original SOURCE RESULT/1 citation: authenticated original RESULT/1.

Sources and claim boundaries

Catalogue and evidence provenance

  • Pinned catalogue source: teorth/erdosproblems at 536408716724.
  • GCL issue: #1671 — 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.

Back to catalogue · Compendium introduction