Erdős Problem #307
Editorial depth: Reviewed explanatory article — evidence qualified.
Snapshot classification: verifiable (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 Problem 307 asks whether there are two finite sets of primes \(P\) and \(Q\) for which the two reciprocal sums multiply to \(1\):
The sets are finite, so this is a concrete existential question. A proposed pair can be checked exactly: test that every listed denominator is prime, add the rational numbers, and compare the resulting product with \(1\). That finite-witness feature is the reason the maintained catalogue calls the status “verifiable”. It does not say that a witness exists, that one is likely to be found, or that the question is decidable in a chosen formal system. It also does not say that the problem is independent of that system.
The wording matters. The target uses primes, not arbitrary positive integers, and the members of a set are distinct because a set has no repetitions. The two sets are not required to have been presented as a partition in the original question; however, an elementary valuation argument shows that any genuine solution would make them disjoint. That consequence is not an extra hypothesis.
A small example
The maintained page records a useful warning example for a weakened problem. If “prime” is replaced by “pairwise relatively coprime”, and \(1\) is allowed, take \(P=\{1,5\}\) and \(Q=\{2,3\}\). Direct exact arithmetic gives
The calculation was independently checked with rational arithmetic; no decimal approximation is involved. A second recorded example is
These examples establish only the weakened statement. The denominator \(1\) is not prime, so neither example answers Problem 307. They are nevertheless helpful because they show exactly where the prime requirement enters and why a catalogue status must not silently turn a variant into the registered problem.
There is also an exact size check behind the standard lower bound. Summing reciprocals of the first 58 primes gives a rational number strictly below \(2\), while summing the first 59, through \(277\), gives a rational number strictly above \(2\). Therefore any set of distinct primes with reciprocal sum greater than \(2\) has at least 59 members: among all prime sets containing \(k\) elements, the first \(k\) primes give the largest sum.
What is known
The original source is E. J. Barbeau’s “Computer challenge corner: Problem 477: A brute force program”, Journal of Recreational Mathematics 9 (1976/77), page 30. The question was subsequently recorded on page 38 of Erdős and Graham’s Old and New Problems and Results in Combinatorial Number Theory (1980). The accessible source discussion says that the prime version was not known, and suggests the pairwise-coprime variant. The maintained Erdős Problems page still presents the prime question as open and reports no proof claim in its discussion.
One elementary consequence is especially robust. For a finite prime set \(S\), write \(D_S=\prod_{r\in S}r\) and \(N_S=\sum_{r\in S}D_S/r\). No prime in \(S\) divides \(N_S\), because modulo a fixed \(p\in S\) only the term \(D_S/p\) survives. Thus \(N_S/D_S\) is already in lowest terms. If the two reciprocal sums multiply to \(1\), lowest terms force \(N_P=D_Q\) and \(N_Q=D_P\), and then \(P\) and \(Q\) must be disjoint. Consequently their union has reciprocal sum equal to the sum of two positive numbers whose product is \(1\), hence at least \(2\). The exact 58/59 calculation then gives \(|P\cup Q|\geq59\).
This is a finite obstruction, not a solution. It rules out small supports but neither constructs a pair beyond the bound nor proves that no pair exists. A 2026 external repository by Vico Bonfioli reports a stronger product barrier and supplies a Lean development, but those are external reports for this article’s purposes; they have not been replayed here and do not constitute GCL mathematical certification.
GCL's current state
The frozen catalogue record is teorth/erdosproblems commit 536408716724a2402bdfa7c1d764bf83d64a8c89, with the entry’s Git blob 4e60263a80af6207aca494d2e1388e44fa372889. It records informal status verifiable, last updated 2025-09-09, formal status unformalized, and a formalized statement. The live page agrees on the question and current “verifiable” presentation, but it is a moving source and was read separately.
The pinned formal source is FormalConjectures commit 85f863718beeec7b58a3a1926ee92e3472bc2020, file 307.lean, blob aac9e4fef823e1f2ee124520558b877187274761. Its main theorem has the expected finite-set quantifiers over Finset ℕ, primality hypotheses, and equality in ℚ; its proof is by sorry. The variant with pairwise coprime sets is separately formalized and proved, but it is not the prime problem. The formal file also contains a barrier theorem marked as solved while its local proof remains sorry, so metadata and theorem-source status must be kept distinct.
What remains
The central unresolved claim is still existential: either exhibit finite prime sets satisfying the exact equation, or prove that no such sets exist. A sensible bounded successor is source-and-statement verification: obtain a readable scan of Barbeau’s page 30, compare it character by character with the page-38 formulation and the pinned Lean quantifiers, then check the 59-prime lemma from exact integer arithmetic. Acceptance would require the scan, stable bibliographic identity, a normalized statement, and a machine-readable record of any discrepancy. Falsification would be a changed quantifier, a condition absent from the formal statement, or a verified counterexample. Stop after the primary source is captured and reconciled, or after the bounded search finds no accessible scan; do not infer a resolution from non-discovery.
Any stronger barrier, arithmetic-derivative reduction, or claimed proof must be treated as a separate mathematical review. In particular, a Lean theorem containing sorry is a statement placeholder, not a checked proof, while an external repository’s claim of a sorry-free build remains an unverified report until an authorized review replays it. “Verifiable” means that a positive answer has a finite certificate; it is not a synonym for proved, true, decidable, falsifiable in both directions, or independent.
Sources and claim boundaries
Primary and maintained sources used here are Barbeau’s 1976/77 journal reference as identified in the source bibliography, Erdős and Graham’s 1980 source document at page 38 (PDF), the live catalogue entry, and the two pinned Git records above. The external barrier report is ElVec1o/erdos307 tag v1.0.0, whose Closed.lean blob is 5319381638adc46c8b7045fba0e943346d418940; it is cited only as an unverified external report.
The original issue is MATH-PROGRAMME #1577. The original SOURCE RESULT/1 citation is authenticated original RESULT/1; no future comment or result link is fabricated here. This article reports source evidence, finite checks, and explicit boundaries only. V2 ledger state, not this prose alone, decides accepted coverage, and nothing here supplies GCL mathematical certification, publication authority, or a claim that Problem 307 has been solved.
Catalogue and evidence provenance
- Pinned catalogue source: teorth/erdosproblems at
536408716724. - GCL issue: #1577 — 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.