Skip to content

Erdős Problem #304

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

Problem 304 asks how many distinct unit fractions are needed to write a rational number as an Egyptian fraction. For integers \(1 \le a < b\), define \(N(a,b)\) to be the least \(k\) for which there are integers

\[ 1 < n_1 < n_2 < \cdots < n_k \]

such that

\[ \frac{a}{b}=\frac{1}{n_1}+\frac{1}{n_2}+\cdots+\frac{1}{n_k}. \]

The strict inequalities make the denominators distinct and put them in increasing order. The worst case at denominator \(b\) is

\[ N(b)=\max_{1\le a<b}N(a,b). \]

The open question is whether this worst-case number grows as slowly as a constant multiple of \(\log\log b\). Here \(\ll\) means “bounded above by a constant multiple of,” with an absolute constant independent of \(a\) and \(b\).

The wording matters. This is a uniform question over every numerator in the interval, not merely a question about one fraction or about the average number of terms. It also asks for distinct denominators greater than one.

A small example

The pinned formal source includes the test case \(N(2,15)=2\). Its two-term representation is directly checkable:

\[ \frac{1}{10}+\frac{1}{30}=\frac{3}{30}+\frac{1}{30}=\frac{4}{30}=\frac{2}{15}. \]

One term cannot work. If \(2/15=1/m\), then \(m=15/2\), which is not an integer. Thus two terms are sufficient and one is impossible, so the minimum is exactly two. This is an exact finite calculation, not a floating-point experiment; it checks one small instance only. It does not bound \(N(b)\) for general \(b\), settle the asymptotic question, or certify the surrounding Lean theorems.

What is known

The readable primary source is P. Erdős, “Az \(1/x_1+1/x_2+\cdots+1/x_n=a/b\) egyenlet egész számú megoldásairól,” Mat. Lapok 1 (1950), 192–210. The scan states the same definition of \(N(a,b)\). On printed page 195, Theorem 1 gives an absolute-constant upper bound of order \(\log b/\log\log b\) for every \(1\le a<b\). On printed page 196, Theorem 2 gives a lower bound of order \(\log\log b\) for the average and separately proves \(N(b-1,b)>\log\log b-1\). The paper’s final English summary also records these results and says that the \(\log\log b\) upper bound is expected, rather than proved.

The maintained problem entry attributes the stronger uniform upper bound

\[ N(b)\ll\sqrt{\log b} \]

to M. D. Vose, “Egyptian Fractions,” Bulletin of the London Mathematical Society 17 (1985), 21–24, DOI 10.1112/blms/17.1.21. The citation and result are also recorded in the pinned Formal Conjectures file. This audit confirms the paper’s bibliographic record and the attribution, but does not claim a line-by-line verification of Vose’s proof from an accessible full text.

There is a small literal indexing difference worth preserving. The current web entry displays the average over \(1\le a<b\); the scanned Erdős theorem writes the displayed sum through \(b-2\) and treats \(N(b-1,b)\) separately. Those formulations have the same coarse order for the recorded lower bound, but they should not be silently called identical. The formal definition of the envelope uses Finset.Ico 1 b, which includes \(1\) through \(b-1\).

GCL's current state

At the frozen catalogue revision teorth/erdosproblems commit 536408716724a2402bdfa7c1d764bf83d64a8c89, the exact number: "304" entry is marked open, with last update 2025-08-31, formal status unformalized, and statement-formalized metadata yes. The entry lists number theory, unit fractions, and OEIS references A097847 and A097849. The protected catalogue blob is 4e60263a80af6207aca494d2e1388e44fa372889.

The pinned Formal Conjectures commit 85f863718beeec7b58a3a1926ee92e3472bc2020 contains FormalConjectures/ErdosProblems/304.lean, blob 88ec512db9602d4638956426e466260e355eff5c. It defines the finite-set version of unit-fraction expressibility, the least cardinality, and the envelope over 1 <= a < b. It records Erdős’s two bounds and Vose’s bound as theorem statements, but those statements end in sorry; they are not completed formal proofs or GCL certification. The open conjecture is likewise posed as an asymptotic answer(sorry) proposition.

The live Erdős Problems entry was last edited 2025-12-29 and remains marked open. Its displayed text preserves the source question and historical bounds. V2 ledger status, rather than this article or any catalogue label alone, decides accepted coverage. The original SOURCE RESULT/1 citation is authenticated original RESULT/1.

What remains

A useful next bounded task is a source-and-semantics review of Vose’s 1985 argument. Its input should be the full paper, the frozen Erdős definition, and the pinned Lean definitions. The target is a proposition-by-proposition map: identify exactly which \(N(a,b)\) convention Vose uses, verify the quantifiers behind the \(\sqrt{\log b}\) bound, and record whether each step applies to the maximum over all \(1\le a<b\).

Acceptance should require page-level quotations or faithful paraphrases for the main theorem, a checked correspondence of variables and hypotheses, and an explicit resolution of the \(b-1\) versus \(b-2\) indexing issue. Falsification is any counterexample to the claimed convention match, missing hypothesis, or source passage that supports only an average bound. Stop when the source is inaccessible, the definitions cannot be aligned, or the bounded comparison is complete; do not turn this audit into a new proof campaign.

Sources and claim boundaries

The original assignment and its complete source packet are GitHub issue #1574. The primary Erdős scan is 1950-02.pdf. The Vose record is the DOI-linked journal page above. Frozen machine-readable evidence is the catalogue commit/blob and Formal Conjectures commit/blob identified above; live-page drift is reported rather than folded into the frozen status.

This article reports established statements from the cited primary scan, attributed reports from the maintained entry and pinned Lean source, and one independently checked finite identity. It does not claim that a sorry theorem has been proved, that Lean has been replayed, that the open question has been solved, or that GCL has supplied mathematical certification. The article is a source-bound editorial record; acceptance, coverage, and any later promotion remain governed by the applicable ledger and coordinator-controlled evidence.

Catalogue and evidence provenance

  • Pinned catalogue source: teorth/erdosproblems at 536408716724.
  • GCL issue: #1574 — protected issue/assignment identity; its labels and results are live, not reproduced automatically here.
  • GCL triage: PROTECTED_READY_BACKLOG.
  • 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