Erdős Problem #351
Editorial depth: Reviewed explanatory article — evidence qualified.
Snapshot classification: proved (not automatically the live result).
Subject areas: number theory, complete sequences.
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 351 asks whether a sequence built from a rational-coefficient polynomial is strongly complete. Let \(p\) be a non-constant polynomial with rational coefficients and positive leading coefficient, and form
The question is whether, after deleting any finite set \(B\) of values from \(A\), every sufficiently large integer can be written as a sum of distinct remaining values. “Sufficiently large” may depend on \(p\) and on \(B\). The quantifiers matter: this is an eventual assertion for every admissible polynomial and every finite deletion, not merely a statement that many integers have representations.
The live problem page now uses “integers,” not “rationals.” That correction matters. Representing every sufficiently large rational would be a much stronger and different property, called rational completeness in the discussion. The page was last edited 10 May 2026 and currently shows nine comments, no proof exposition, and no proof claim. This article records source evidence and boundaries; it does not decide accepted coverage.
A small example
Take \(p(n)=n\). The three distinct indices \(2,3,6\) satisfy
Therefore the corresponding values of \(A\) give the exact calculation
This is exact rational arithmetic, not a decimal approximation. A second check from Graham’s displayed examples is \(2+4+6+12=24\) and \(1/2+1/4+1/6+1/12=1\), giving a sum of values equal to \(25\).
These calculations illustrate the useful reduction: if distinct positive integers \(a_i\) have sum \(N\) and reciprocal sum \(1\), then the values \(a_i+1/a_i\) sum to \(N+1\). They do not establish an eventual theorem. A finite list neither covers all large integers nor shows that the indices can avoid an arbitrary finite deletion. The latter is exactly why the “for every lower bound on the \(a_i\)” clause in Graham’s theorem is important.
What is known
The primary source I read is R. L. Graham, “A theorem on partitions,” Journal of the Australian Mathematical Society 3 (1963), 435–441. On p. 435, Theorem 1 states that every integer exceeding 77 is a sum of distinct positive integers whose reciprocals sum to 1. On pp. 437–439, Theorem 2 states that for every integer \(m\) there is a threshold \(r(m)\) such that every integer \(N>r(m)\) has a representation by distinct positive integers \(a_i\) with \(a_i\ge m\) and \(\sum a_i=N\), \(\sum 1/a_i=1\). Substituting \(p(n)=n\) proves the strong-completeness statement for that special polynomial, including finite deletions.
The live page cites Erdős and Graham, Old and New Problems and Results in Combinatorial Number Theory (1980), p. 58, as the original catalogue source. It also cites Graham’s “Complete sequences of polynomial values,” Duke Mathematical Journal (1964), 275–285, for a related characterization of complete polynomial-value sequences. Those references contextualize the question but do not, from the material inspected here, supply a proof of the full strong statement for every rational polynomial.
For \(p(n)=n^2\), the live page reports a route through Graham’s theorem and Max A. Alekseyev’s 2019 paper “On partitions into squares of distinct integers whose reciprocals sum to 1,” pp. 213–221, cited on Problem 283. The report is plausible source evidence, but this audit did not independently read Alekseyev’s paper or verify the reduction. A discussion post also points to a short write-up; that is an external report, not GCL certification.
GCL's current state
The frozen catalogue packet is teorth/erdosproblems@536408716724a2402bdfa7c1d764bf83d64a8c89, with data/problems.yaml blob 4e60263a80af6207aca494d2e1388e44fa372889. Its entry for 351 records informal status proved, formal status Lean, statement metadata yes, and last update 12 May 2026. Those are catalogue fields, not a mathematical certificate.
The pinned google-deepmind/formal-conjectures@85f863718beeec7b58a3a1926ee92e3472bc2020 file FormalConjectures/ErdosProblems/351.lean has blob 55255d5fddf5b70e0768d17d6ca174e186a2af46. Its theorem quantifies over rational polynomials with positive natural degree and positive leading coefficient, and its imageSet includes the natural input \(0\); the file comments that \(1/0=0\) does not change validity. The theorem and the two variants for \(p(n)=n\) and \(p(n)=n^2\) are all closed by sorry. Thus the file captures a formal interface, not a checked proof.
An external plby/lean-proofs file at repository commit 8822f7ddef30fadbd92e1c6ab4ed897af356af5e and file blob 21a36199340d2b010142f3834ef5820e20d01713 supplies a non-sorry proof claim through an imported corollary_7_pos_leading; it reports ordinary Lean axioms. No Lean replay was performed here, and external source presence is not GCL mathematical certification. The Programme triage class remains OUTSIDE_STRICT_OPEN_TRIAGE.
What remains
The useful bounded successor is a source-semantic reconciliation, not an attempt to rebuild the whole proof. First, obtain and pin the complete proof source for the all-polynomial assertion, including the underlying result called corollary_7_pos_leading, and compare its definition of strong completeness with the page and FormalConjectures definitions. In particular, check natural-number zero, duplicate values, finite subsets versus finite index sets, and whether “positive leading coefficient” plus non-constant degree is exactly the hypothesis used.
Second, independently verify the \(n^2\) reduction against Alekseyev’s primary paper and record theorem and page locations. Acceptance should require a line-by-line semantic match and a reproducible exact-arithmetic fixture; falsification is any missing hypothesis, changed quantifier, or proof that handles only ordinary completeness. Stop after the pinned comparison, one exact finite fixture, and a written discrepancy ledger. Do not launch a broad Lean build or promote a catalogue label.
Sources and claim boundaries
The original assignment is MATH-PROGRAMME issue #1621. The maintained entry is Erdős Problem 351, with its discussion thread and history. The primary Graham PDF is A theorem on partitions. The pinned catalogue and formal file are linked by their exact commits above. The external Lean report is Erdos351.lean.
The exact calculation above was checked with rational arithmetic. Graham’s special-case theorem is an established primary result. The \(n^2\) route and the all-polynomial external Lean file are reports requiring verification. The catalogue and sorry-bearing formal declaration are status and interface evidence. GCL has supplied no mathematical certification here. Original SOURCE RESULT/1 citation: authenticated original RESULT/1.
Catalogue and evidence provenance
- Pinned catalogue source: teorth/erdosproblems at
536408716724. - GCL issue: #1621 — 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.