Skip to content

Erdős Problem #294

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: Publication status comes exclusively from the protected V2 ledger. This explanation does not constitute mathematical certification.

The question in plain language

An Egyptian-fraction representation of \(1\) is a finite sum of reciprocals of distinct positive integers. For a positive integer \(N\), consider representations whose denominators begin with a specified integer \(t\) and never exceed \(N\):

\[ 1=\frac{1}{n_1}+\frac{1}{n_2}+\cdots+\frac{1}{n_k}, \qquad t=n_1<n_2<\cdots<n_k\leq N. \]

Problem #294 asks about the first value of \(t\) for which this is impossible. More precisely, \(t(N)\) is the least positive integer \(t\) for which no such finite sequence exists. The quantifiers are worth spelling out: \(N\) is fixed first; \(t\) is then the least positive integer with the failure property; and the representation, when it exists, may have any finite length \(k\geq1\). The condition \(t=n_1\) says that \(t\) is the smallest denominator, not merely one denominator somewhere in the sum.

The answer is not expected to be a simple exact formula. The published result gives the scale of this first obstruction, up to powers of iterated logarithms:

\[ \frac{N}{(\log N)(\log\log N)^3(\log\log\log N)^{O(1)}}\ll t(N)\ll\frac{N}{\log N}. \]

Here \(f(N)\ll g(N)\) means that \(f(N)\) is at most a constant multiple of \(g(N)\) for all sufficiently large \(N\). Thus the statement locates \(t(N)\) near \(N/\log N\), while leaving a polylogarithmic gap in the lower bound. It does not assert that every finite \(N\) has a displayed asymptotic value.

A small example

Take \(N=6\). The denominator set is small enough that an exact enumeration can be done by considering every subset of \(\{t,t+1,\ldots,6\}\) containing \(t\) and summing with rational arithmetic. An independent Python replay using fractions.Fraction and itertools.combinations returned

\[ \{1\}\quad\text{and}\quad \{2,3,6\} \]

as the only representations of \(1\) with first denominator \(1\) or \(2\), respectively. Indeed,

\[ 1=\frac11, \qquad 1=\frac12+\frac13+\frac16. \]

There is no representation beginning with \(3\). After choosing \(1/3\), the required residual is \(2/3\), while even using every available larger denominator gives only

\[ \frac14+\frac15+\frac16=\frac{15+12+10}{60}=\frac{37}{60}<\frac23. \]

Any proper subcollection is smaller. For \(t\geq4\), even the sum of all available reciprocals is at most \(1/4+1/5+1/6=37/60<1\). Therefore the exact finite answer is \(t(6)=3\). This calculation checks the definition and its quantifiers; it is not evidence by itself for the asymptotic theorem.

What is known

The historical source is P. Erdős and R. L. Graham, Old and new problems and results in combinatorial number theory (1980). On printed page 35 of the cited scan, the quantity called \(k_1(n)\) is the least integer that does not occur as the first denominator \(x_1\) in a representation with \(x_1<\cdots<x_r\leq n\). That is the same first-denominator quantity as \(t(N)\) after renaming \(n\) as \(N\). The same page discusses nearby quantities concerning whether an integer occurs in any denominator position; those must not be conflated with #294.

The primary modern source is Yang P. Liu and Mehtaab Sawhney, “On further questions regarding unit fractions.” Their Theorem 1.6 (arXiv version 1, printed p. 3) states exactly the definition above and proves the two-sided bound. Lemma 4.1 (arXiv version 1, p. 13) supplies the smooth-denominator construction used in the lower-bound argument, and the proof of Theorem 1.6 continues on pp. 13–14. In outline, the upper bound chooses a prime \(t\) just above a constant multiple of \(N/\log N\) and uses a congruence obstruction; the lower bound repeatedly builds controlled reciprocal sums from smooth integers. The paper is published in International Mathematics Research Notices (2026), article rnaf382, DOI 10.1093/imrn/rnaf382. These are claims and proofs in the cited primary paper, not a new proof by this manuscript.

GCL's current state

The authenticated original GCL issue #1564 records the direct-editorial source audit for #294. Its exact original RESULT/1 comment gives DISPOSITION: SOURCE_LOCK_READY, identifies Liu–Sawhney Theorem 1.6, and records that no external formal artifact was replayed. The return is therefore not PARTIAL or STATUS_CONFLICT; that exact status is preserved here rather than replaced by an inferred label. It also says that the catalogue status is externally proved but that the admitted formal snapshot has no problem-specific statement.

The pinned catalogue entry at teorth/erdosproblems commit 536408716724a2402bdfa7c1d764bf83d64a8c89 records informal status proved, formal status unformalized, and formalized: no. The entry is bound to issue #1564 by issue-body SHA-256 4cfc51d6dc0baa27600560c7040448ff283cedfbcb05447da2994f8d934d5736. For reproducibility, the local source registry file governance/erdos_catalogue/REGISTRY.json has SHA-256 9e3e8c8f22a1ef59c5c9fefb7d409d5404d018cb64ff13f4b8cb5baf28920609; the pinned catalogue entry has Git blob 4e60263a80af6207aca494d2e1388e44fa372889.

These fields are provenance, not certification. An external claim of a Lean proof, an author manuscript, or a successful CI run would still not be GCL mathematical certification unless it passed the applicable protected mathematical route and exact replay requirements. No such certification is claimed here, and no issue comment, publication, or admission is being made by this manuscript.

What remains

The mathematical source question is resolved at the level reported by the primary paper: #294 is the Erdős–Graham first-denominator problem, and Liu–Sawhney prove the displayed bounds. What remains for GCL is documentary and formal, not an invitation to silently strengthen the claim. A future formalization would need to encode the same quantifiers, strict denominator ordering, finite length, and upper bound \(N\), then be replayed in an identified Lean environment. The finite check \(t(6)=3\) is a useful statement-level test, but it is not a formal proof of Theorem 1.6.

The source audit also identifies a bounded successor: formalize the predicate for representations of \(1\), prove the finite \(N=6\) result with exact arithmetic, and preserve the concordance between #294 and the printed-page-35 \(k_1(n)\). That work must stop short of presenting the analytic asymptotic argument as GCL-certified unless the relevant mathematical certification route is actually completed.

Finally, editorial admission and mathematical proof status are separate. Editorial acceptance requires a distinct protected admission recording its exact manuscript hash, source checks, critical check, and rendered readback. Admission would control the documentary page; it would neither upgrade unformalized metadata nor turn an external theorem into GCL certification.

Sources and claim boundaries

  • P. Erdős and R. L. Graham, Old and new problems and results in combinatorial number theory, printed p. 35 in the cited scan, SHA-256 0cbf0c32f0ab1e1c71db5121a88bac905bf976c4a6ab6bb6d7d9cf9ddd184ed3.
  • Y. P. Liu and M. Sawhney, “On further questions regarding unit fractions,” arXiv:2404.07113v1, Theorem 1.6 p. 3, Lemma 4.1 p. 13, proof pp. 13–14; published as International Mathematics Research Notices (2026), rnaf382. The audited arXiv PDF SHA-256 is 29f6ea8f73bb17a9084bec115af4b3d8015d9b6d5e3c803a3caa8f79a287d91b.
  • Pinned catalogue bytes, entry 294, Git blob 4e60263a80af6207aca494d2e1388e44fa372889; and the authenticated GCL source audit, exact RESULT/1, SOURCE_LOCK_READY.
  • governance/erdos_compendium/EDITORIAL_PROTOCOL.md: source-first authoring, six-section contract, and the separate unaccepted-draft/admission boundary. This manuscript remains a source-qualified explanation and makes no claim of editorial admission, publication, novelty, priority, or GCL mathematical certification.

Catalogue and evidence provenance

  • Pinned catalogue source: teorth/erdosproblems at 536408716724.
  • GCL issue: #1564 — 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.

Back to catalogue · Compendium introduction