Erdős Problem #404
Editorial depth: Reviewed explanatory article — evidence qualified.
Snapshot classification: open (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
Start with a fixed positive integer \(a\) and a prime \(p\). Choose any finite strictly increasing list of integers beginning with \(a\), and add their factorials:
The first question asks whether the powers of \(p\) dividing such sums have a finite ceiling. If a ceiling exists, call it \(f(a,p)\) and ask how it depends on \(a\) and \(p\). The second question asks for one prime \(p\) and one infinite increasing sequence \(a_1<a_2<\cdots\) for which the exponent of the highest power of \(p\) dividing each successive partial sum tends to infinity. These are valuation questions: they concern exact divisibility, not the size of the factorial sum as an ordinary integer.
The quantifiers matter. The list length \(n\) is free and is not the exponent of \(p\); the starting point \(a\) is fixed while later factorial indices may be arbitrarily large. The source also asks for distinct factorials, so repeated indices are not allowed.
A small example
Take \(a=1\) and \(p=5\), with the increasing list \((1,4)\). Exact arithmetic gives
and \(5^3=125\) does not divide \(25\). Thus this particular sum has valuation at the prime \(5\) exactly \(2\), and, if \(f(1,5)\) is finite, it proves the lower bound \(f(1,5)\geq 2\).
This calculation is reproducible without floating-point arithmetic: evaluate the two factorials as integers, add them, and divide by \(5\) repeatedly until the quotient is no longer divisible by \(5\). It does not prove that \(f(1,5)\) is finite, determine its value, or construct an infinite sequence whose partial-sum valuations diverge. It is only one finite witness.
What is known
The primary source located for this audit is P. Erdős and R. L. Graham, Old and New Problems and Results in Combinatorial Number Theory, Monographies de l’Enseignement Mathématique 28 (1980), available as a scan at the author-hosted PDF. On printed page 79 it asks whether there is a bound depending on the first index and the prime for the exponent in a sum of distinct factorials, and then asks about an infinite increasing sequence with unbounded successive valuations. The same page reports that S. Lin showed the largest power of \(2\) occurring in a sum of distinct factorials containing \(2\) is \(2^{254}\).
The maintained Erdős Problems #404 page presents the normalized form, records Lin as giving \(f(2,2)\leq254\), and marks the problem open. Its discussion includes later comments with computational witnesses and conjectural observations, including very large finite lower bounds for some odd-prime cases and a suggested finite subset-sum extension lemma. Those comments are useful leads, not independently checked theorems for this article. In particular, the public discussion's stronger wording \(f(2,2)=254\) is reported here as an external claim, not as GCL verification.
GCL's current state
The frozen catalogue packet is teorth/erdosproblems data/problems.yaml at commit 536408716724a2402bdfa7c1d764bf83d64a8c89, whose entry has Git blob 4e60263a80af6207aca494d2e1388e44fa372889. It records open status, last update 2025-08-31, no prize, and no formalization. The live catalogue blob is 97ae243a7fc056bb9c6413ce665b575eadd27ff3 at live commit 76848eddd25687dfc69bb26bec740c35c4a9c6b4; the #404 metadata still agrees with the frozen entry. The live page was edited on 29 September 2025, so its prose and discussion are newer evidence than the frozen metadata but do not silently replace that source lock.
No #404 statement file exists in the pinned FormalConjectures revision 85f863718beeec7b58a3a1926ee92e3472bc2020. The admitted FC-GDM-85F86371 provider snapshot is metadata-only for Programme use; its manifest has SHA-256 faa43f174717ef0126114b9bbd61bb9fdc827ce7eb9f64e56c1e63c7b15608db. Accordingly, there is no Lean theorem source, no by sorry statement to compare, and no formal replay to report.
What remains
The smallest useful next step is a source-and-certificate check for one fixed finite transition suggested in the discussion. For \(p=5\), test the normalized factorials from \(5!\) through \(9!\) modulo \(5\): after dividing each by \(5^{\nu_5(5!)}=5\), their residues are \(4,4,3,4,1\). Their subset sums cover every residue class modulo \(5\). A bounded task can verify this table and the exact normalization, then check that an already supplied witness below \(5\) extends to the corresponding divisibility level below \(10\).
Acceptance is an exact-arithmetic residue table, an explicit increasing witness, and a short proof of the finite extension implication. Falsification is any missing residue, incorrect valuation, repeated index, or failure of the claimed divisibility. The task stops after this single transition; it must not be presented as evidence of an infinite construction. A separate, more ambitious successor would need a pinned Lin memorandum or a pinned computational certificate before treating the \(f(2,2)\) report as established.
Sources and claim boundaries
The original assignment is MATH-PROGRAMME issue #1674. The original SOURCE RESULT/1 citation is authenticated original RESULT/1. The source audit used the frozen catalogue commit and blob above, the live #404 page and its discussion, and the Erdős–Graham scan (download SHA-256 0cbf0c32f0ab1e1c71db5121a88bac905bf976c4a6ab6bb6d7d9cf9ddd184ed3). The Lin memorandum is cited by the primary book and maintained page but was not obtained, so its detailed proof and exact bibliographic date remain unverified here.
Catalogue labels, web comments, finite computations, and a future formalization are different evidence classes. This article supplies no mathematical certification, does not decide accepted coverage, and does not treat an upstream label or an external Lean report as a GCL result. V2 ledger disposition remains the authority for accepted coverage.
Catalogue and evidence provenance
- Pinned catalogue source: teorth/erdosproblems at
536408716724. - GCL issue: #1674 — protected issue/assignment identity; its labels and results are live, not reproduced automatically here.
- GCL triage: STATEMENT_CAPTURE_REQUIRED.
- 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.