Erdős Problem #305
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.
The question in plain language
An Egyptian fraction writes a proper fraction as a sum of distinct reciprocals. For integers \(1\leq a<b\), let \(D(a,b)\) be the smallest possible value of the largest denominator in such a representation, and then set
The maximum is over every numerator, not just numerators coprime to \(b\). The question attached to Erdős Problem 305 asks whether there is an absolute constant and a function tending to zero such that, for all sufficiently large \(b\),
This is the precise meaning of saying that \(D(b)\) is at most a constant multiple of \(b(\log b)^{1+o(1)}\). It is an asymptotic question about how far one must search for denominators, uniformly over all fractions with denominator \(b\); it does not ask for a bounded number of summands.
The original issue is MATH-PROGRAMME issue #1575. Its frozen catalogue packet identifies the entry at teorth/erdosproblems@536408716724a2402bdfa7c1d764bf83d64a8c89, blob 4e60263a80af6207aca494d2e1388e44fa372889, and records the catalogue status as proved on 2025-08-31.
A small example
Take \(a=2\) and \(b=3\). The distinct-denominator identity
shows that \(D(2,3)\leq 6\). To check that the bound is exact, enumerate every subset of the possible denominators \(\{1,2,3,4,5\}\) with rational arithmetic. The reproducible check was:
from fractions import Fraction
from itertools import combinations
target = Fraction(2, 3)
hits = [s for r in range(6) for s in combinations(range(1, 6), r)
if sum((Fraction(1, n) for n in s), Fraction(0)) == target]
print(hits)
print(Fraction(1, 2) + Fraction(1, 6))
Its exact output was [] and 2/3. Therefore no representation has largest denominator at most \(5\), and \(D(2,3)=6\). This is a finite check of one small value, not evidence for the asymptotic bound. In particular, checking many denominators cannot establish a statement quantified over every sufficiently large \(b\).
What is known
The maintained live page Erdős Problem 305 gives the same definition and says PROVED. It cites Erdős and Graham, Old and new problems and results in combinatorial number theory (1980), p. 38, as the original source. Bleicher and Erdős, “Denominators of Egyptian fractions,” Journal of Number Theory 8 (1976), 157–168, established the earlier upper bound that is a constant multiple of \(b(\log b)^2\); their primary scan is available here. For prime \(b\), they also supplied a lower bound of order \(b\log b\).
Yokota’s “On a problem of Bleicher and Erdős,” Journal of Number Theory 30 (1988), 198–207, DOI 10.1016/0022-314X(88)90017-0, improved the upper bound to a factor of \((\log\log b)^4(\log\log\log b)^2\) beyond \(b\log b\).
The most direct primary evidence checked here is Yang P. Liu and Mehtaab Sawhney, “On further questions regarding unit fractions,” arXiv:2404.07113, version 1 submitted 10 April 2024. Their Theorem 1.5, PDF p. 3, states that for every \(1\leq a<b\) there are distinct integers \(1<n_1<\cdots<n_k\) with
The proof is on PDF p. 14. Since the displayed extra factors are \((\log b)^{o(1)}\), this implies the original target. The paper explicitly identifies a sharper factor \((\log\log b)^{1+o(1)}\) as an open direction; that refinement is not needed to settle Problem 305 as stated. The downloaded primary PDF has SHA-256 29f6ea8f73bb17a9084bec115af4b3d8015d9b6d5e3c803a3caa8f79a287d91b.
GCL's current state
The frozen packet says no Problem 305 statement file was present in the admitted FC-GDM-85F86371 provider snapshot. The live page now links a FormalConjectures file and labels the statement formalised, so this is a real source-time drift rather than a reason to rewrite the frozen packet.
For comparison only, the current file FormalConjectures/ErdosProblems/305.lean was read at exact main commit 817d8a74e6bcd505ac8407388cc12ac3cfda3d0d, Git blob 60ff6254e4e23e5237d43e1516dd7fcd2c9abbfc. Its D uses finite sets of nonzero natural denominators bounded by a cap, and Dmax ranges over \(1\leq a<b\), matching the intended distinct-positive-denominator interface. The main theorem and four variants end in by sorry; this is a formal statement/interface, not a completed GCL proof.
That file points to an external Lean source at plby/lean-proofs@8822f7ddef30fadbd92e1c6ab4ed897af356af5e, blob 7405ba2cc36cd7ac13b8273c0f5635432c7dc63f. It contains a fuller reported proof, but no Lean replay or certification is supplied here. The V2 ledger, rather than this article or catalogue metadata, decides accepted coverage.
What remains
The classical resolution is clear at the level of primary theorem scope: Liu–Sawhney’s theorem is uniform in every proper numerator and has the required denominator cap. A bounded follow-up should nevertheless formalize the semantic bridge carefully. Its inputs should be the exact current D, Dmax, and asymptotic statement, together with the Liu–Sawhney Theorem 1.5 citation and pinned source hash. Its target should be a checked implication from their bound, with a fixed interpretation of the implicit \(O(1)\) exponent and of “sufficiently large.”
Acceptance should require a source comparison of every quantifier and a clean build of the external Lean file under its declared toolchain. Falsification would be a mismatch in the numerator range, denominator positivity or distinctness, the asymptotic exponent, or the primary theorem’s hypotheses. Stop after those interfaces and the finite example are checked; do not spend this documentary task on an exhaustive proof replay. Improving the triple-logarithmic factor remains a separate mathematical question.
Sources and claim boundaries
This article uses the frozen catalogue commit and blob recorded above, the live page at erdosproblems.com/305, and the primary Bleicher–Erdős, Yokota, and Liu–Sawhney references. The Liu–Sawhney theorem is external mathematical evidence; it is not a GCL certification. The current FormalConjectures file is an unpinned comparison because the issue packet records no admitted FC snapshot for 305, and its sorry bodies are reported exactly as read.
The finite identity for \(2/3\) is exact and reproducible, but proves no general theorem. Catalogue labels, live-page labels, formal metadata, and an external Lean source are distinguished from established external proof text, and none is silently promoted. The original source-audit RESULT/1 citation is authenticated original RESULT/1. Accepted compendium coverage remains a V2-ledger decision, not a consequence of this documentary article.
Catalogue and evidence provenance
- Pinned catalogue source: teorth/erdosproblems at
536408716724. - GCL issue: #1575 — protected issue/assignment identity; its labels and results are live, not reproduced automatically here.
- GCL triage: OUTSIDE_STRICT_OPEN_TRIAGE.
- 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.