Skip to content

Erdős Problem #403

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

Erdős Problem 403 asks whether a power of two can be written as a sum of distinct factorials in infinitely many different ways. In the usual notation, the question is whether

\[ 2^m=a_1!+a_2!+\cdots+a_k! \]

has only finitely many solutions when the indices are strictly increasing, \(a_1<a_2<\cdots<a_k\). “Distinct” is important: an index may occur at most once. The historical statement, as printed by Erdős and Graham, does not spell out the domain of \(m\) or the lower bound on the \(a_i\) in the displayed question. The natural reading is nonnegative \(m\) and positive integer indices, but those conventions should be preserved as an explicit normalization rather than silently assumed.

The problem is more precise than the informal phrase “find a large power.” It asks for a global finiteness theorem over every finite set of positive indices. A claim that \(2^7\) is the largest solution is stronger: it also asserts that no larger power of two occurs, not merely that the set of solutions is finite.

A small example

The advertised maximal example is exact:

\[ 2^7=128=2!+3!+5!=2+6+120. \]

I checked this with integer arithmetic, and also enumerated every one of the \(2^7-1=127\) nonempty subsets of \({1,2,3,4,5,6,7}\). The complete list of powers of two obtained from those subsets was

\[ (m,S)=(0,\{1\}),(1,\{2\}),(3,\{2,3\}),(5,\{2,3,4\}),(7,\{2,3,5\}). \]

For reproducibility, the calculation used the exact values \(1!=1\), \(2!=2\), \(3!=6\), \(4!=24\), \(5!=120\), \(6!=720\), and \(7!=5040\), summing each subset and retaining a sum exactly when it was a positive power of two. This verifies the finite statement “among indices at most \(7\), these are the matches.” It does not prove that \(2^7\) is globally largest, because a solution using an index greater than \(7\) would not be tested; nor does it reproduce the historical proof.

What is known

The accessible primary source document is P. Erdős and R. L. Graham, Old and New Problems and Results in Combinatorial Number Theory (1980), printed pp. 79–80 in the scan linked below. It presents the Burr–Erdős question, records \(2^7=2!+3!+5!\) as the largest solution known there, and says that this was proved to be the largest solution by P. Frankl and independently by S. Lin. The same passage reports Lin’s stronger divisibility bound that the largest power of two dividing a sum of distinct factorials containing \(2\) is \(2^{254}\).

That source is evidence for the historical statement and attribution, not a readable copy of either resolving proof. The maintained page’s bibliography identifies [Li76] as S. Lin, “On two problems of Erdős concerning sums of distinct factorials,” Bell Laboratories internal memorandum, but displays the year as 1960; the 1980 monograph refers to Lin (76), and its Frankl entry is a personal communication. Thus the proof provenance and date/version are not yet source-locked at primary-proof level. The article does not turn those reports into an independently checked theorem.

The maintained Erdős Problems #403 page, read on 2026-10-13, currently says “PROVED,” repeats the \(2^7\) example, and says Frankl and Lin showed the answer is yes. Its /history/403 record identifies the current historical revision as 2025-10-20; the page’s rendered metadata separately says it was last edited 28 October 2025. The frozen catalogue snapshot teorth/erdosproblems@536408716724a2402bdfa7c1d764bf83d64a8c89, blob 4e60263a80af6207aca494d2e1388e44fa372889, records informal status proved, formal status Lean, and statement-formalized metadata yes, with catalogue updates dated 2026-06-21 and 2026-07-22. These are different evidence surfaces, not interchangeable certifications.

GCL's current state

The protected Programme registry maps problem 403 to original issue #1673, assignment ERDOS-CATALOGUE-0403-S01, with catalogue status proved, triage class OUTSIDE_STRICT_OPEN_TRIAGE, and a protected FormalConjectures file present. At readback the issue has no comments and no existing RESULT/1; the original source-audit citation is therefore authenticated original RESULT/1, not a future link invented here. The V2 ledger alone decides accepted coverage. This article is documentary source prose and does not certify the mathematics, accept the catalogue label, or authorize publication.

The pinned formal source is FormalConjectures/ErdosProblems/403.lean at commit 85f863718beeec7b58a3a1926ee92e3472bc2020, blob a4b8e6b288168e3b09ef87ed9c9c71805ca6d8c7. It encodes pairs \((m,S)\) with \(m\in\mathbb{N}\) and a finite set \(S\) of positive natural indices satisfying \(2^m=\sum_{a\in S}a!\), and states that this set is finite. Its proof is by sorry. It does not formalize the reported largest solution, Lin’s \(2^{254}\) divisibility result, or the equivalence between this encoding and every convention in the historical wording.

What remains

The next useful bounded step is to obtain a readable, provenance-preserving copy of Lin’s memorandum or the Frankl source behind the personal-communication attribution. A reviewer should transcribe the theorem and proof pages, record the exact document date and hash, and compare its quantifiers with the normalized statement above. The check should explicitly test positivity, whether \(m\) is natural or positive, distinctness of indices, and whether “largest solution” means largest exponent or a complete list.

If the primary proof remains inaccessible, the correct result is a documented source blocker, not a guessed bibliography or a Lean build. A finite exact-arithmetic replay can support examples and small bounds, but it cannot replace the global proof. The formal file can be used as a statement interface for future work, while its sorry remains an unverified formal gap.

Sources and claim boundaries

The external “proved” status and the finite check are reported as evidence at their stated scopes. Neither is GCL mathematical certification, and no claim is made here that the historical proof has been independently replayed.

Catalogue and evidence provenance

  • Pinned catalogue source: teorth/erdosproblems at 536408716724.
  • GCL issue: #1673 — protected issue/assignment identity; its labels and results are live, not reproduced automatically here.
  • GCL triage: OUTSIDE_STRICT_OPEN_TRIAGE.
  • Formal-statement coverage: A provider statement was included in the pinned 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