Skip to content

Erdős Problem #391

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

For a positive integer \(N\), take the factorial \(N!\). We want to split it into exactly \(N\) positive integer factors, allowing repetition, and make the smallest factor as large as possible. Write the factors in nondecreasing order, \(a_1\leq\cdots\leq a_N\), and define \(t(N)\) to be the largest value of \(a_1\) for which

\[ N!=a_1a_2\cdots a_N. \]

The question asks how close \(t(N)/N\) gets to \(1/e\) as \(N\) grows. More precisely, the historical formulation asks whether \(t(N)/N\) tends to \(1/e\) and whether it is eventually below \(1/e\) by a correction of order \(1/\log N\) for infinitely many \(N\). The factors are not required to be distinct, and the statement has no upper bound on any individual factor. Those quantifiers matter: changing “exactly \(N\) factors”, allowing only prime powers, or imposing a size cap gives a different problem.

The elementary upper bound comes from multiplying the factors: if every factor is at least \(t\), then \(N!\geq t^N\). Stirling's formula therefore gives \(t(N)/N\leq 1/e+O(\log N/N)\). The interesting issue is whether the prime factors of \(N!\) can be rearranged into \(N\) factors that nearly attain this bound.

A small example

The case \(N=4\) can be checked without approximation. There is a valid factorization

\[ 4!=24=2\cdot2\cdot2\cdot3, \]

so \(t(4)\geq2\). If four factors were all at least \(3\), their product would be at least \(3^4=81>24\), so \(t(4)\leq2\). Hence \(t(4)=2\) exactly. This also illustrates why the finite value need not lie below \(N/e\): here \(2>4/e\).

The calculation verifies one instance of the definition and the basic product obstruction. It says nothing by itself about a limit, an asymptotic error term, or the validity of a proof for all sufficiently large \(N\). Those are quantified statements over an unbounded parameter and require the source theorem.

What is known

The primary proof source is Boris Alexeev, Evan Conway, Matthieu Rosenfeld, Andrew V. Sutherland, Terence Tao, Markus Uhr, and Kevin Ventullo, “Decomposing a factorial into large factors”, arXiv:2503.20170v4, revised 3 April 2026. Its introduction defines the same \(t(N)\) and states Theorem 1.3 on printed page 6. Part (iv) proves, for sufficiently large \(N\),

\[ \frac{t(N)}{N}=\frac{1}{e}-\frac{c_0}{\log N}+O\left(\frac{1}{\log^{1+c}N}\right), \]

for an absolute constant \(c>0\) and explicit \(c_0=0.30441901\ldots\). Thus the limit is \(1/e\), and the correction is substantially more precise than the earlier question required. The same theorem records effective finite bounds, including \(t(N)\leq N/e\) for \(N\neq1,2,4\) and \(t(N)\geq N/3\) for \(N\geq43632\), with the threshold stated as best possible.

The paper distinguishes established arguments from computation-assisted certificates. Its methods combine elementary prime-factor accounting, greedy constructions, linear and integer programming, rearrangement of small prime powers, and an asymptotic construction using numbers with no prime factor \(2\) or \(3\). The authors state that the certificates used for the theorem were checked in exact arithmetic, but that external report is not a GCL mathematical certification. The historical record is also important: Erdős, Selfridge, and Straus were reported to have had an earlier proof of the limiting statement, but the notes were lost; Erdős later described that assertion as only a conjecture. The current primary source is therefore the accessible proof of the registered statement.

GCL's current state

Original issue #1661 is the direct-editorial source-audit assignment for Erdős #391. Its frozen packet identifies the catalogue commit 536408716724a2402bdfa7c1d764bf83d64a8c89 and the exact data/problems.yaml blob 4e60263a80af6207aca494d2e1388e44fa372889. That snapshot records informal status proved and formal status Lean, with later snapshot dates 2026-08-24 and 2026-09-19.

The maintained live problem page, last edited 28 September 2025, has a different formal interface: its visible text says “Formalised statement? No”, while the surrounding status text also contains “PROVED (LEAN)”. This is a genuine metadata discrepancy, not evidence that the mathematical theorem is open. No statement file for #391 was found in the issue's named FC-GDM-85F86371 provider snapshot, so no GCL Lean theorem, sorry boundary, or replay result is claimed here. V2 ledger state alone decides accepted coverage. GCL's triage class remains OUTSIDE_STRICT_OPEN_TRIAGE; it does not certify the external proof or formalization.

What remains

The mathematical question is answered by the primary paper, but a bounded successor task remains useful: lock the exact statement against the source, then reconcile the catalogue's “Lean” metadata with the live page and the absent GCL snapshot file. The task should preserve the distinction between an external theorem, an external computation certificate, a Lean source that may contain sorry, and GCL mathematical certification.

A suitable reusable lemma is the finite factor-product obstruction used above: if \(N!\) is expressed as exactly \(N\) positive factors each at least \(u\), then \(u^N\leq N!\). A successor can formalize or independently check this elementary implication for natural-number inputs, but it must not present it as a formalization of Theorem 1.3. Acceptance should require an exact statement comparison, a recorded source revision, and a reproducible finite check; falsification is any quantifier mismatch or any claimed formal theorem that cannot be located at the pinned commit. Stop once those records are complete or the formal source remains unavailable after the bounded lookup.

Sources and claim boundaries

  • Original GCL issue #1661, including the assignment packet and required labels.
  • Frozen catalogue file, commit 536408716724a2402bdfa7c1d764bf83d64a8c89, blob 4e60263a80af6207aca494d2e1388e44fa372889.
  • Live Erdős Problems entry and its LaTeX/source view.
  • Primary paper, arXiv:2503.20170v4, Theorem 1.3, p. 6. The downloaded v4 PDF SHA-256 was acd5a6779da7390901637996d4ddd88c47db661c6bd3b91c1229c01f397be879; the source archive SHA-256 was 68bc40d2ebf6d6a5b366ce380e1f27b0888482e88ffb6ff9599df2d15d2f4053.
  • Historical references named by that paper: Erdős, “Some problems I presented or planned to present in my short talk” (1996), pp. 333–335; Erdős and Graham, Old and new problems and results in combinatorial number theory (1980); and Guy–Selfridge, “Factoring factorial \(n\)”, American Mathematical Monthly 105 (1998), 766–767.

The original SOURCE RESULT/1 citation is authenticated original RESULT/1. This article reports source evidence and boundaries only; it supplies no Lean replay, mathematical certification, publication decision, or claim that the catalogue metadata is accepted coverage.

Catalogue and evidence provenance

  • Pinned catalogue source: teorth/erdosproblems at 536408716724.
  • GCL issue: #1661 — 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