Skip to content

Erdős Problem #283

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

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.

Editorial authority: The protected GCL V2 admissions ledger alone determines publication status. This source-bounded explanation is not mathematical certification.

The question in plain language

Choose a polynomial \(p\) with integer coefficients and positive leading coefficient, and is not forced to have a common divisor larger than \(1\) on all positive inputs. The question asks whether every sufficiently large integer \(m\) can be made in two ways at once: as a sum of distinct positive denominators whose reciprocals add to \(1\), and as the sum of the corresponding values of \(p\). In symbols, does

\[ \forall p\;\Bigl[\bigl(\mathrm{lc}(p)>0\bigr)\land\neg\exists d\ge 2\;\forall n\ge 1\;(d\mid p(n))\Bigr]\Rightarrow \exists M\;\forall m\ge M\;\exists k\ge 1\;\exists 1\le n_1<\cdots<n_k: \sum_{i=1}^k\frac1{n_i}=1\ \land\ \sum_{i=1}^k p(n_i)=m. \]

The quantifiers matter. “Sufficiently large” means one threshold \(M\) works for every later integer, not merely infinitely many \(m\). The denominators are distinct because of the strict inequalities, and the same \(n_i\) must satisfy both equations. The divisor condition is a local obstruction: if one \(d>1\) divides every available value \(p(n)\), then sums of those values cannot cover every large integer.

The exact public statement is Erdős Problem #283. The original GCL assignment is issue #1553, and the source audit’s authenticated RESULT/1 comment is the evidence reviewed here.

A small example

Take the square polynomial \(p(n)=n^2\) and denominators \(2,4,6,12\). The reciprocal calculation is exact:

\[ \frac12+\frac14+\frac16+\frac1{12}=\frac6{12}+\frac3{12}+\frac2{12}+\frac1{12}=1. \]

The corresponding value is

\[ 2^2+4^2+6^2+12^2=4+16+36+144=200. \]

Thus \(200\) has the required representation for this particular polynomial. The following calculation uses rational arithmetic, so it does not rely on floating-point rounding:

from fractions import Fraction

A = [2, 4, 6, 12]
assert sum((Fraction(1, n) for n in A), Fraction(0)) == 1
square_sum = sum(n * n for n in A)
assert square_sum == 200
print("reciprocal sum = 1; square sum =", square_sum)

Its output is reciprocal sum = 1; square sum = 200. This is a checked finite witness, not evidence that all large integers work. It also illustrates why the two requirements cannot be checked independently: changing one denominator can preserve the reciprocal identity while changing the polynomial sum.

What is known

The fully general polynomial statement should not be described as a verified published theorem on the evidence presently available. Several important special cases are genuine.

R. L. Graham’s “A theorem on partitions,” Journal of the Australian Mathematical Society 3 (1963), 435–441, proves that every integer exceeding \(77\) is a sum of distinct positive integers with reciprocal sum \(1\); this is Theorem 1 on p. 435, with DOI 10.1017/S1446788700039045. It also proves the reciprocal-partition generalization for a positive rational target and a lower bound on the parts. This is the case \(p(n)=n\), not the arbitrary-polynomial problem.

For \(p(n)=n^2\), Max A. Alekseyev’s “On partitions into squares of distinct integers whose reciprocals sum to 1,” published in The Mathematics of Various Entertaining Subjects, vol. 3 (2019), pp. 213–221, proves that \(8542\) is the largest exception; hence every \(m>8542\) works. The accessible arXiv version records the theorem and publication data, with the published chapter linked by DOI 10.2307/j.ctvd58spj.18.

Cassels’s 1960 paper, “On the representation of integers as the sums of distinct summands taken from a fixed set,” Acta Scientiarum Mathematicarum 21, pp. 111–124, has an accessible repository record and scan. Its Theorem I begins on p. 111 and its proof runs through pp. 112–122. The source audit found a completeness theorem for sums of distinct elements of a set, but did not find in that paper the reciprocal-sum-preserving polynomial theorem attributed to it by the catalogue page. That is a source-fidelity exception, not a reason to discard Cassels’s actual theorem.

Wouter van Doorn’s arXiv:2502.02200, v2 (23 July 2025), Theorems 1–3, gives bounds for partitions with prescribed reciprocal sum and lower-bounded parts. It supports the linear reciprocal-partition background, but the claimed polynomial ranges such as \(p(n)=n+b\) for \(1\le b\le5000\) and \(p(n)=n^2+b\) for \(1\le b\le800\) were not independently verified against a primary proof here. Burr’s attribution in the formal catalogue likewise remains without an identified primary paper in this audit.

The public page also cites an original source as [ErGr80, p. 32]. That book/page was not available in the retrieved source packet, so the original attribution and any claim resting only on that page remain unchecked. “Unavailable” here means unavailable to this audit, not disproved.

GCL's current state

The pinned source packet records catalogue status proved, formal status Lean, and formalized: yes, all updated 10 May 2026. It pins teorth/erdosproblems commit 536408716724a2402bdfa7c1d764bf83d64a8c89, entry blob 4e60263a80af6207aca494d2e1388e44fa372889, and google-deepmind/formal-conjectures commit 85f863718beeec7b58a3a1926ee92e3472bc2020, file blob 70b9dec34150a41ff06efa2a8756db3bcc0ea05a. The pinned Lean file uses \(p\colon\mathbb Z[X]\), an eventual integer variable, and a sentinel-indexed finite function to encode positive distinct denominators. Every theorem body in that file, including erdos_283, ends in by sorry.

The live FormalConjectures file has drifted to blob c9e6b5fdb077d6b662b7a439d11d987d22fc85a7: it changes the main domain to rational-coefficient, integer-valued polynomials, changes several casts and quantifiers, and adds proof links. That live file must not silently replace the pinned packet.

The authenticated source return is explicitly DISPOSITION: PARTIAL. It reports a stronger external Lean artifact, Proof_flat.lean at commit 286f856aa3fc08957b80950fd18a45aab8d045ea, claiming arbitrary positive rational reciprocal target and lower denominator bound. The artifact was not independently built, replayed, or checked with #print axioms here. It is therefore an unverified external formal-proof claim, not a GCL certification. The catalogue label, issue labels, and this manuscript are documentary evidence only; no protected GCL MathCert result has been located.

This is the mandatory PARTIAL / STATUS_CONFLICT exception: a catalogue “proved (Lean)” label coexists with an incomplete source audit, unavailable or mismatched primary support for some attributions, and an unverified external proof artifact.

What remains

The decisive task is a theorem-level reconciliation: build the external artifact at its exact commit, inspect its actual imported environment and axioms, and compare its hypotheses one by one with the pinned \(\mathbb Z[X]\) statement. In particular, verify positivity, distinctness, denominator indexing, the fixed-divisor condition, and the bridge from “there exists a threshold” to “every later integer.” Until that happens, the strongest safe conclusion is special-case publication plus an external claim of a stronger formal proof.

Related formal reports must not be used to fill this gap. For #276, a weak gcd-seed condition is not automatically equivalent to the stronger pinned Lean condition

\[ \forall d>1\;\exists k\;\gcd(d,a_k)=1. \]

That equivalence remains unresolved. For #285, the Lean Fin k.succ indexing shifts by one relative to f(k) and the reported file includes sorry; it is not a clean proof witness. For #284, a Croot claim of infinitely many suitable \(k\) does not automatically supply the bridge to every sufficiently large \(k\) required by an eventual statement; that step needs scrutiny. These neighboring discrepancies are explicitly excluded from the #283 proof status.

Sources and claim boundaries

  • Original issue #1553 and its authenticated RESULT/1: exact assignment, PARTIAL disposition, source discrepancies, and external-proof boundary.
  • Pinned catalogue file, commit 536408716724a2402bdfa7c1d764bf83d64a8c89, blob 4e60263a80af6207aca494d2e1388e44fa372889: snapshot metadata, not proof authority.
  • Pinned FormalConjectures file, blob 70b9dec34150a41ff06efa2a8756db3bcc0ea05a: formal statement shell with sorry, not certification.
  • Graham, Alekseyev, Cassels, and van Doorn: published or public primary records as identified above. The [ErGr80, p. 32] source was unavailable to this audit, and the Burr primary paper was not identified; both remain unavailable rather than silently promoted.
  • The external Lean artifact is claimed but unverified. GCL has not certified the arbitrary-polynomial theorem. This draft is an educational source reconciliation, and its unaccepted status remains subject to the protected V2 editorial ledger.

Catalogue and evidence provenance

  • Pinned catalogue source: teorth/erdosproblems at 536408716724.
  • GCL issue: #1553 — 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