Skip to content

Erdős Problem #355

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 355 asks whether one can choose an infinite, rapidly growing sequence of positive integers

\[ a_1<a_2<a_3<\cdots \]

with a fixed growth factor greater than \(1\), while still making finite sums of distinct reciprocals unusually complete. More precisely, if there is a number \(\lambda>1\) such that \(a_{n+1}/a_n\geq\lambda\) for every \(n\), does there exist a non-empty open interval \((u,v)\) such that every rational number in that interval is a sum of finitely many distinct terms \(1/a_n\)?

This is a question about Egyptian fractions: sums such as \(1/4+1/6+1/12\). The tension is that lacunarity makes the available reciprocals sparse, whereas representing every rational in an interval requires a great deal of arithmetic flexibility. The older formulation in Bleicher and Erdős uses a strict inequality \(a_{n+1}/a_n>c>1\). For an existence question this agrees with the displayed \(\geq\lambda\) normalization: one may move slightly below or above the chosen constant without changing the underlying existential issue.

The original issue for this article is MATH-PROGRAMME issue #1625, assignment ERDOS-CATALOGUE-0355-S01. The issue’s source packet identifies the problem as proved, but that catalogue metadata is evidence to audit, not GCL mathematical certification.

A small example

The primary paper gives an explicit beginning of a possible construction with lacunarity constant \(3/2\):

\[ 1,2,4,6,12,24,48,72,120,180,360,720,\ldots \]

The first five terms provide a useful exact finite check. Put all fractions over the common denominator \(12\). The reciprocals correspond to integer weights

\[ 12,6,3,2,1. \]

Taking subset sums of the same weights in increasing order gives the following exact interval calculation. The weight \(1\) gives all integers from \(0\) through \(1\); adding \(2\) gives \(0\) through \(3\); adding \(3\) gives \(0\) through \(6\); adding \(6\) gives \(0\) through \(12\); and adding \(12\) gives \(0\) through \(24\). At each step the new weight is at most one more than the previous maximum, so there is no gap. Thus the five reciprocals represent exactly the grid points \(k/12\) for \(0\leq k\leq24\). For example, \(7/12=1/2+1/12\).

This is reproducible exact integer arithmetic, not a floating-point experiment. It verifies the finite-prefix claim recorded by the authors: the first five terms cover the fractions in \([0,2]\) whose reduced denominators divide \(12\). It does not prove that all rationals in \([0,2]\) are represented, and it says nothing by itself about the infinitely continued sequence. The infinite argument must arrange divisibility milestones and enough intervening terms at every stage.

What is known

The primary resolution is Wouter van Doorn and Vjekoslav Kovač, “Lacunary sequences whose reciprocal sums represent all rational numbers in an interval,” arXiv:2509.24971v3, especially Theorem 1 and Proposition 8; the work is also published in Acta Arithmetica 223 (2026), 275–295, DOI 10.4064/aa251001-13-1. The arXiv record identifies v3 as the 3 December 2025 revision, while the online-first publication supplies the later journal record.

Theorem 1(a) proves the stronger statement that for every \(\lambda\in(1,2)\) there is a sequence of positive integers with lacunarity constant \(\lambda\) whose finite reciprocal sums contain every rational in \([0,2]\). The construction is organized in blocks. Distinguished block endpoints are divisible by all previous terms and, over time, by every positive integer. A finite-sum filling lemma then turns the controlled ratio bound into complete rational coverage at each compatible denominator.

The same theorem says more: the sequence may have consecutive ratios tending to \(2\), and every rational in \((0,2]\) can have infinitely many finite representations. The boundary is sharp in the relevant sense. Theorem 1(c), proved through the paper’s necessary-condition argument, says that no sequence with lacunarity constant \(2\) has finite reciprocal sums containing all rationals from a non-empty open interval. Thus the answer to the original existence question is yes, with the admissible uniform constants below \(2\), while the exact endpoint \(2\) is excluded.

The maintained Erdős Problems entry reports this same conclusion and says it was last edited 18 November 2025. Its displayed references include Bleicher–Erdős and Erdős–Graham, and it links the van Doorn–Kovač work. The page is a useful maintained index, but the primary paper is the evidence for the theorem’s scope and proof.

GCL's current state

The frozen catalogue evidence is teorth/erdosproblems commit 536408716724a2402bdfa7c1d764bf83d64a8c89, entry 355, blob 4e60263a80af6207aca494d2e1388e44fa372889. It records informal state proved, formal status Lean, and last update 2026-02-02, with tags number theory and unit fractions. Those fields are catalogue metadata and do not replace the primary source.

The pinned formal provider is google-deepmind/formal-conjectures commit 85f863718beeec7b58a3a1926ee92e3472bc2020, blob 26e0bf9301cae6cfaa0f4feddfcb4381145c0c44. Its theorem has the expected existential shape, but its body is by sorry; it is a statement/formalization interface, not a checked formal proof. The quantifiers also deserve care: the Lean predicate uses an indexed sequence, IsLacunary, an open real interval, and finite subsets whose elements lie in the range. This is close to the paper’s formulation, but presence of the file alone does not establish semantic equivalence or proof completion.

An additional public Woett/Lean-files file describes a simplified construction and contains a proof script, but it is explicitly named obsolete and was not replayed here. No Lean build or GCL mathematical certification is claimed. V2 ledger state alone decides accepted coverage.

What remains

The substantive theorem is externally resolved, so the useful bounded follow-up is semantic and documentary rather than a search for a new solution. A reviewer can compare the paper’s Theorem 1(a) and Proposition 8 line by line with the pinned Lean statement: index origin, positivity, strict increase, the ratio inequality, whether the interval is open or closed, whether \(0\) is included, and whether finite subsets are represented through indices or values. The endpoint distinction matters: the paper proves coverage of \([0,2]\cap\mathbb{Q}\) for every fixed \(\lambda<2\), but separately excludes sequences with lacunarity constant \(2\).

A second bounded task is to transcribe the paper’s first four explicit construction blocks and verify the finite-prefix denominator claims with exact rational arithmetic, retaining the script and its input sequence as evidence. Acceptance should require matching the published list and all claimed finite grid points; falsification should be an omitted point, an incorrect ratio bound, or a mismatch in the divisibility milestone. Stop once the stated finite claims and the formal statement comparison are recorded; do not infer a formal proof from successful finite checks.

Sources and claim boundaries

Primary mathematical evidence: van Doorn–Kovač, arXiv:2509.24971v3, Theorem 1, Proposition 8, and the journal DOI above. Original problem context: Erdős Problems #355, with the original references listed there. GCL task context: issue #1625.

The catalogue status, live webpage, and formal-conjecture file are reported with their dates and pinned identifiers so that later drift is visible. The formal file is by-sorry. The finite calculation above is independently checkable but is not an infinite theorem. This article supplies no mathematical certification, publication decision, accepted-coverage decision, or claim that a Lean proof has been compiled.

Original SOURCE RESULT/1 citation: authenticated original RESULT/1.

Catalogue and evidence provenance

  • Pinned catalogue source: teorth/erdosproblems at 536408716724.
  • GCL issue: #1625 — 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