Skip to content

Erdős Problem #327

Editorial depth: Reviewed explanatory article — evidence qualified.
Snapshot classification: open (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.

The question in plain language

Erdős and Graham ask about large subsets of the first \(N\) positive integers whose pairs avoid a divisibility pattern. The first question asks whether a set \(A \subseteq \{1,\ldots,N\}\) can be substantially denser than the odd numbers while satisfying, for every two distinct \(a,b \in A\), the condition \(a+b \nmid ab\). “Substantially more” is an asymptotic question: the issue is whether the largest possible density can exceed the baseline density of the odd numbers by a fixed positive amount, not whether an isolated small \(N\) example exists.

The second question changes the condition to \(a+b \nmid 2ab\) and asks whether every such set must have \(|A|=o(N)\). Equivalently, it asks whether the density \(|A|/N\) must tend to zero along every sequence of admissible sets as \(N\) grows. These are related but different questions; the factor \(2\) changes the forbidden pairs and neither question automatically settles the other.

The unit-fraction language explains the connection. Since

\[ \frac{1}{a}+\frac{1}{b}=\frac{a+b}{ab}, \]

this sum is a unit fraction exactly when \(a+b\) divides \(ab\). Likewise, the average of the two unit fractions is

\[ \frac{1/a+1/b}{2}=\frac{a+b}{2ab}, \]

which is a unit fraction exactly when \(a+b\) divides \(2ab\).

A small example

Take \(A=\{1,3,5\} \subseteq \{1,\ldots,5\}\). An exact integer check of its three pairs gives

\[ 4\nmid 3, \qquad 6\nmid 5, \qquad 8\nmid 15, \]

for the pairs \((1,3)\), \((1,5)\), and \((3,5)\) respectively. Thus this particular set avoids the first forbidden relation. The corresponding averages are, exactly,

\[ \frac{1+1/3}{2}=\frac{2}{3},\qquad \frac{1+1/5}{2}=\frac{3}{5},\qquad \frac{1/3+1/5}{2}=\frac{4}{15}, \]

and none is a unit fraction either. The familiar larger example is the set of all odd integers: an odd product \(ab\) cannot be divisible by the even number \(a+b\), so the first condition always holds. This gives roughly half of the integers, but it says nothing about whether a positive-density improvement exists. A useful contrast is \((a,b)=(2,6)\): \(2+6=8\) does not divide \(2\cdot6=12\), but it does divide \(2\cdot2\cdot6=24\), so the pair is allowed by the first condition and forbidden by the second. Finite checks illustrate the definitions; they do not establish an asymptotic bound.

What is known

The frozen catalogue packet records Problem 327 as open, with informal status last updated 2025-08-31, and as unformalized. The maintained problem page likewise presents both questions under an open heading. It reports a result attributed to Wouter van Doorn: if \(|A| \geq (25/28+o(1))N\), then the first forbidden pair with \(a+b \mid ab\) must occur. This is an upper-bound mechanism for the first question, not a determination of the optimal density. The odd-integer construction remains the elementary lower bound of about \(N/2\).

For the second question, Will Sawin’s paper gives a materially stronger dated external report. Theorem 1 states that there is a constant \(c>0\) and, for all sufficiently large \(N\), a set \(A_N \subseteq \{1,\ldots,N\}\) with \(|A_N|>cN\) such that distinct \(a,b \in A_N\) always satisfy \(a+b \nmid 2ab\). Thus the proposed \(o(N)\) conclusion is false if Sawin’s proof is accepted as a mathematical result. The theorem is precise: its construction orders possible witnesses by the number of prime factors counted with multiplicity, and its density conclusion is existential with an unspecified positive constant. It does not solve the first question, and it does not provide a numerical best constant.

GCL's current state

The protected source packet is the teorth/erdosproblems catalogue at commit 536408716724a2402bdfa7c1d764bf83d64a8c89, with the exact data/problems.yaml blob 4e60263a80af6207aca494d2e1388e44fa372889; its entry for number: "327" says open, unformalized, and tagged number theory and unit fractions. The programme registry classifies the item as STATEMENT_CAPTURE_REQUIRED. There is no Problem 327 statement file in the admitted FC-GDM-85F86371 provider snapshot, whose upstream FormalConjectures revision is 85f863718beeec7b58a3a1926ee92e3472bc2020. Consequently, no pinned Lean theorem, sorry-bearing source, replay result, or formal semantic comparison is available for this problem.

The live page’s unchanged open label and its omission of Sawin’s 2026 preprint show source/status drift, not a reason to invent a local resolution. The V2 ledger alone decides accepted coverage. This article therefore records the external theorem as a source report requiring adjudication, while retaining the frozen catalogue status and the programme’s statement-capture prerequisite. No GCL mathematical certification is supplied here.

What remains

The smallest unresolved mathematical target is the first question: determine whether the maximum size of a set with no pair satisfying \(a+b \mid ab\) is asymptotically strictly larger than \(N/2\), or prove a matching obstruction. A bounded successor could first normalize the extremal function \(m_1(N)\), reproduce the odd-set lower bound, and independently check the reported \((25/28+o(1))N\) upper-bound argument with all hypotheses and the source location recorded. Acceptance would require a source-locked statement, a verified finite test suite for the divisibility predicate, and either a correctly scoped proof of one explicit asymptotic inequality or a counterexample family with a stated density. Falsification would be any omitted quantifier, a pair violating the claimed construction, or an argument whose error term is used outside its stated range.

The second question needs a separate source reconciliation: Sawin’s Theorem 1 should be compared line by line with the catalogue wording and then routed for mathematical review. In particular, “positive density exists” must not be silently upgraded to a best density, a resolution of both questions, or GCL certification. A formalization effort should wait for a captured normalized statement and a decision about whether the target is the original open pair-avoidance problem or Sawin’s external counterexample theorem.

Sources and claim boundaries

The original assignment and source-audit boundary are GitHub issue #1597. The frozen catalogue evidence is the pinned problems.yaml blob, and the maintained entry is erdosproblems.com/327. The original reference is P. Erdős and R. L. Graham, Old and New Problems and Results in Combinatorial Number Theory, Monographies de L’Enseignement Mathématique 28, Geneva, 1980, cited by later work at p. 37; a source scan is available here, with SHA-256 0cbf0c32f0ab1e1c71db5121a88bac905bf976c4a6ab6bb6d7d9cf9ddd184ed3.

The primary modern source read for this audit is Will Sawin, “Sets of unit fractions without two members whose average is a unit fraction”, arXiv v1 dated 2026-07-16, especially Theorem 1 and pages 1–2; the downloaded PDF SHA-256 is c0a8f58b5a514d2dd88a4dc72657ad7bc5266e95120e5b1bcdb902bb7d3e4a45. Its theorem is an external proof report, not a GCL result. The original SOURCE RESULT/1 citation is authenticated original RESULT/1; no future comment URL is fabricated. All claims above remain bounded by these dated sources, exact hashes, and the distinction between external mathematics, finite verification, source metadata, and GCL adjudication.

Catalogue and evidence provenance

  • Pinned catalogue source: teorth/erdosproblems at 536408716724.
  • GCL issue: #1597 — 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.

Back to catalogue · Compendium introduction