Skip to content

Erdős Problem #298

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

Erdős Problem 298 asks whether a sufficiently thick set of positive integers must contain a finite Egyptian-fraction representation of \(1\). In precise terms, if \(A\subseteq\mathbb N\) has positive density, must there be a finite set \(S\subseteq A\) such that

\[ \sum_{n\in S}\frac{1}{n}=1? \]

Here “positive density” means that the limit

\[ \lim_{N\to\infty}\frac{|A\cap[1,N]|}{N}=d \]

exists with \(d>0\). The quantifier is important: the same finite set \(S\) is allowed to depend on the particular set \(A\), but it must be a set of distinct members of \(A\), not a list with repetitions and not an infinite series. The question is therefore not whether the reciprocal series over all of \(A\) sums to \(1\); it asks whether one finite subcollection has exactly that sum.

The current mathematical status is proved. Bloom’s Theorem 1.2 proves the stronger assertion that positive upper density is enough: it assumes only

\[ \limsup_{N\to\infty}\frac{|A\cap[1,N]|}{N}>0. \]

Every set with positive density has positive upper density, so Bloom’s theorem implies the registered question. The converse implication does not hold in general: a set can have positive upper density while its ordinary density fails to exist. That distinction explains why the newer theorem is stronger rather than merely a restatement.

A small example

The finite arithmetic is elementary, but it illustrates exactly what the witness must look like. Take \(A=\{2,3,6\}\). Then \(S=A\) is finite, all denominators are distinct, and

\[ \frac12+\frac13+\frac16=\frac{3+2+1}{6}=1. \]

This finite \(A\) is only an arithmetic toy, not itself a positive-density set; the density theorem says that every genuinely positive-density set contains some witness of this kind.

There is also a useful scaled identity. For every integer \(m\geq 1\),

\[ \frac{1}{2m}+\frac{1}{3m}+\frac{1}{6m}=\frac{1}{m}. \]

At \(m=1\) this gives the required sum. For a reproducible exact check, the following uses rational arithmetic rather than floating-point approximations; it verifies the scaled identity for the finite range \(1\leq m\leq1000\):

from fractions import Fraction

assert Fraction(1, 2) + Fraction(1, 3) + Fraction(1, 6) == Fraction(1, 1)
assert all(
    Fraction(1, 2*m) + Fraction(1, 3*m) + Fraction(1, 6*m) == Fraction(1, m)
    for m in range(1, 1001)
)

This check is independently reproducible and is reported with the exact audit evidence in RESULT/1 for issue #1568. It proves only these finite identities. It says nothing by itself about density, which is an asymptotic hypothesis quantified over an infinite set of integers.

What is known

The original source trail is Erdős and Graham’s Old and new problems and results in combinatorial number theory (1980), where the positive-density strengthening appears on printed p. 36, and Erdős’s Some of my forgotten problems in number theory (Hardy–Ramanujan Journal 15, 1992, pp. 34–50). The 1992 scan is an historical source for the problem, but the exact page containing Problem 298 was not established in the source audit; that bibliographic limitation is retained here rather than filled by inference.

Bloom’s paper, On a density conjecture about unit fractions, JEMS 27 (2025), 4563–4589, DOI 10.4171/JEMS/1456, gives the decisive published theorem. On printed p. 4564, Theorem 1.2 states: if \(A\subseteq\mathbb N\) has positive upper density, then there is a finite \(S\subseteq A\) with reciprocal sum \(1\). The paper also gives a quantitative finite-set theorem (Theorem 1.3) and explains how its technical proposition supplies the density result. Those details are evidence for the theorem’s proof, not a license to replace its hypotheses with a different statement.

The paper’s Appendix B, printed pp. 4585–4586, describes a Lean formalisation and displays the corresponding upper-density theorem. That report is useful provenance. It does not turn a source audit into a new mathematical certification, and it does not erase the distinction between the published human-readable theorem and a separately replayed formal artifact.

GCL's current state

The protected catalogue registry records problem 298 as proved, with topics “number theory” and “unit fractions,” and links it to issue #1568. The authenticated source audit by fyremael returned RESULT/1 with disposition PARTIAL, not because the mathematical status was found to conflict, but because two evidence tasks remained incomplete: the exact 1992 page was not located and the pinned external Lean project was not independently replayed.

The source packet pins the catalogue at teorth/erdosproblems commit 536408716724a2402bdfa7c1d764bf83d64a8c89, entry number: "298", Git blob 4e60263a80af6207aca494d2e1388e44fa372889, whose snapshot status is proved and whose last update is 2025-12-29. The fixed Programme registry is itself identified by Git blob 90f6499cfbaca8c8a9e1ac4c6beb468ca09127a5; its problem-298 row binds issue #1568 to issue-body SHA-256 0114a03459bb7e2fec44bda7b213884de5206bff50d0c2dc1429d3f169328ad8. The packet also pins google-deepmind/formal-conjectures at commit 85f863718beeec7b58a3a1926ee92e3472bc2020, file blob d5c9653d076dd3c493260fed888a0586fcd65a04. That Lean file includes the guarded positive-density statement and an upper-density variant, but both declarations end in by sorry. They are formal statement records, not completed proof certificates.

The cited external repository b-mehta/unit-fractions is pinned at commit 10ef71a300cf29e5f19beb2bbc723a035a0678de; its audited src/final_results.lean blob is 72e323563b9f4c6038c443931809c32c65b7b32c, and its source contains the theorem unit_fractions_upper_density. The audit found no placeholder tokens in that source, but no successful exact build replay or recorded check run was available. An external proof repository, self-reporting CI, or absence of sorry is not by itself mathematical certification.

What remains

There is no remaining open mathematical conjecture at the registered strength in the evidence reviewed here: Bloom’s published theorem settles it, and in fact settles the weaker upper-density hypothesis. What remains is evidence hygiene and semantic reconciliation. A successor audit could identify the exact page in the 1992 scan, replay the pinned Lean project with its specified Lean 3.42.1 and mathlib revision, and compare the formal theorem line by line with the natural-language statement.

That comparison must check several small but consequential details: whether the ambient naturals include \(0\); whether the finite witness is represented as a Finset; whether inclusion means every witness denominator lies in \(A\); whether the reciprocal is taken in \(\mathbb Q\) rather than by natural-number division; and whether upper_density A > 0 has exactly the intended limsup meaning. The external theorem does not exclude \(0\), while the pinned Formal Conjectures statement does. In Lean’s rationals, \(1/0=0\); removing \(0\) from a witness therefore preserves its sum, but this is a semantic observation to document, not an excuse to skip comparison.

The stopping point for that successor task is one exact, hash-checked replay and one documented semantic comparison. A successful compilation would establish what that pinned source compiles to. It would not independently certify Bloom’s mathematics, change the published theorem’s status, or authorize a claim beyond the source evidence.

Sources and claim boundaries

The primary statement and live status are in the maintained Erdős entry for #298, which was edited 31 March 2026. Its cited originals are the Erdős–Graham 1980 scan, printed p. 36, and Erdős 1992, pp. 34–50. The exact source-audit record is MATH-PROGRAMME issue #1568 and its exact authenticated RESULT/1 comment.

The fixed catalogue packet and formal comparison anchors are the pinned catalogue YAML, the pinned Formal Conjectures file, and the pinned external Lean source. The accepted mathematical claim in this manuscript is limited to the inspected published theorem and its direct quantifier implication. Catalogue labels, formalisation metadata, source hashes, finite calculations, and audit returns document provenance; they do not substitute for a proof.

This is documentary editorial prose, not a mathematical certification or an acceptance decision. The protected V2 ledger determines acceptance.

Catalogue and evidence provenance

  • Pinned catalogue source: teorth/erdosproblems at 536408716724.
  • GCL issue: #1568 — 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.

Back to catalogue · Compendium introduction