Erdős Problem #299
Editorial depth: Reviewed explanatory article — evidence qualified.
Snapshot classification: disproved (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 299 asks whether a steadily advancing list of positive integers must contain a finite collection whose reciprocals add to exactly one. More formally, consider a sequence \(a_1<a_2<\cdots\) of positive integers whose successive gaps \(a_{i+1}-a_i\) are bounded (the meaning of \(O(1)\) here). Is it possible that no finite set of indices \(I\) satisfies \(\sum_{i\in I}1/a_i=1\)? The catalogue records the answer as no: every such bounded-gap sequence contains an Egyptian-fraction representation of one.
The quantifiers matter. The sequence is infinite, strictly increasing, and positive; the chosen indices form a finite set, so denominators are distinct. This is not asking for a prescribed number of terms, a consecutive block, or a representation using every element of the sequence. It asks only for one finite subset. The original issue is MATH-PROGRAMME #1569, whose source packet identifies the historical citation as Erdős and Graham, Old and New Problems and Results in Combinatorial Number Theory (1980), p. 36.
A small example
Here is an exact finite witness, using distinct denominators:
This calculation can be reproduced without floating-point arithmetic:
from fractions import Fraction
terms = [Fraction(1, 2), Fraction(1, 3), Fraction(1, 6)]
total = sum(terms, Fraction(0))
assert total == Fraction(1, 1)
print(total)
The verification returns the exact rational value 1. It illustrates the kind of finite subset sought by Problem 299, but it proves nothing about arbitrary sequences: the set \(\{2,3,6\}\) is finite and specially chosen, and it does not have to come from a bounded-gap infinite sequence. Even checking many initial terms would not establish the required universal statement.
What is known
Thomas F. Bloom’s paper On a density conjecture about unit fractions, arXiv:2112.03726v2 (18 pages, updated 2023-10-12), proves a stronger density theorem. Its Theorem 2 says that if \(A\subseteq\mathbb N\) has positive upper density, then some finite \(S\subseteq A\) satisfies \(\sum_{n\in S}1/n=1\). Positive upper density means that \(|A\cap[1,N]|/N\) has a positive limsup; ordinary positive density is a special case.
The reduction to #299 is elementary at the level of statements. If the gaps of \(a_i\) are eventually bounded by an integer \(B\), then the range \(A=\{a_i:i\geq1\}\) has positive lower, and hence positive upper, density up to a harmless finite initial segment. Bloom’s theorem therefore supplies a finite subset of \(A\), which is exactly a finite set of sequence terms with reciprocal sum one. The paper’s proof obtains the density result by first finding finite sums equal to \(1/d\) for controlled integers \(d\), then combining disjoint copies; its technical core uses sieve estimates and Fourier analysis.
Bloom’s article also reports a complete Lean formalisation by Bloom and Bhavik Mehta in the separate b-mehta/unit-fractions project. That is an external report about another repository. It is useful provenance, but it is not the same thing as a checked theorem in the pinned GCL source file.
GCL's current state
The frozen catalogue evidence supplied by the issue is teorth/erdosproblems commit 536408716724a2402bdfa7c1d764bf83d64a8c89, data/problems.yaml, entry number: "299", Git blob 4e60263a80af6207aca494d2e1388e44fa372889. At that revision, the informal status is disproved, last updated 2025-12-29; the status field says disproved (Lean), and formalisation is marked yes. The live Erdős Problems entry, read for this audit, likewise says that no such sequence exists by the positive solution to #298, cites Bloom, and reports no proof exposition or discussion comments. Its page says it was last edited 28 October 2025, so the frozen catalogue is the later dated status surface among the two.
The issue pins google-deepmind/formal-conjectures commit 85f863718beeec7b58a3a1926ee92e3472bc2020, file FormalConjectures/ErdosProblems/299.lean, blob f34451cf7304487949be75c3afb410693af407e0. Its theorem uses StrictMono, positive natural values, a real-valued eventual-big-O bound on successive differences, and the assertion that every finite index set has reciprocal sum different from one. The adjacent density variant assumes \(0\notin A\) and positive upper density. Both declarations end with by sorry; the metadata points to an external proof repository. Thus the formal source records a statement and provenance, not a GCL-checked proof or semantic-equivalence certification.
What remains
A useful bounded successor is a semantic interface audit. Start from the pinned Lean file and Bloom’s Theorem 2, normalize both into explicit quantifiers, and document the short implication from an eventual bounded-gap sequence to a positive-upper-density range. Check the natural-number indexing, the positivity condition, distinctness supplied by StrictMono, the interpretation of finite sums, and the real versus rational casts. The audit should also identify whether the formal density variant is sufficient for the reduction without importing an unrecorded convention about zero.
Acceptance would be a line-by-line statement correspondence with cited source locations and a small exact-arithmetic regression check. Falsification would be any missing hypothesis, type-level division mismatch, or failed implication from bounded gaps to positive upper density. The task should stop after that interface is resolved or a precise mismatch is recorded; it should not launch a full Lean build or claim to replay Bloom’s external certificate.
Sources and claim boundaries
Primary sources used here are Bloom’s arXiv record and v2 PDF, whose retrieved PDF SHA-256 is 16130b55cf4f1e33d6958954f08f334becf24c70886560d76665d33f78dccd09, and the historical Erdős–Graham book citation reproduced in the paper and catalogue. The catalogue and Formal Conjectures commits and blobs above are frozen evidence, while the live problem page is a dated status observation and may change.
This article treats Bloom’s published mathematical argument as the primary external proof source, the catalogue and live page as status evidence, and the pinned Lean file as a formal statement with sorry, not as a proof. No GCL mathematical certification, accepted-coverage decision, or publication decision is supplied here; the V2 ledger remains the authority for accepted coverage. The original SOURCE RESULT/1 citation is authenticated original RESULT/1. No future comment, hidden review, or independent verification is being implied.
Catalogue and evidence provenance
- Pinned catalogue source: teorth/erdosproblems at
536408716724. - GCL issue: #1569 — 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.