Skip to content

Erdős Problem #288

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.

Editorial authority: The protected V2 admissions ledger alone determines publication status. This explanation is not independent mathematical certification.

The question in plain language

Take a consecutive block of positive integers, such as \([3,6]=\{3,4,5,6\}\), and add the reciprocals of its members. Write

\[ H(a,b)=\sum_{n=a}^{b}\frac{1}{n}, \]

where \(a\) and \(b\) are positive integers with \(a\le b\). Erdős Problem 288 asks whether there are only finitely many pairs of such intervals \(I_1=[a_1,b_1]\) and \(I_2=[a_2,b_2]\) for which

\[ H(a_1,b_1)+H(a_2,b_2)\in\mathbb{N}. \]

Here “only finitely many” means that the entire set of qualifying pairs is finite: there is some bound beyond which no new qualifying pair exists. The question is not asking us merely to find examples, or to decide all pairs below a chosen numerical cutoff. The target integer is positive because every reciprocal is positive. The intervals need not be disjoint, non-overlapping, adjacent, or different; no such restriction appears in the current two-interval formulation. Thus the quantifiers matter: \(I_1\) and \(I_2\) range independently over all nonempty consecutive blocks.

The source also records nearby questions, but they are not interchangeable. One asks whether \(H(a,b)+1/m\) is integral for only finitely many triples \((a,b,m)\); another asks about \(k\) intervals for every fixed \(k\). The universal statement for every \(k\) is stronger than the original two-interval problem, while a claim about the existence of some \(k>2\) is weaker. Keeping these readings separate prevents a result about a variant from being mistaken for a solution of Problem 288.

A small example

There are immediate examples if intervals may overlap: two copies of \([1,1]\) give \(1+1=2\). A more informative example uses one genuine block and one singleton:

\[ H(3,6)+H(20,20) =\frac13+\frac14+\frac15+\frac16+\frac1{20} =\frac{20+15+12+10+3}{60}=1. \]

This is an exact rational calculation, not a decimal approximation. I independently checked it with Python’s fractions.Fraction, which returned exactly \(1\). The calculation demonstrates why the problem is not the question “can such a pair exist?”; examples are easy to exhibit. The difficult assertion is that, after all examples are collected, only finitely many pairs remain.

The source audit also gives a useful one-interval diagnostic. Suppose \(H(a,b)=A/B\) is in lowest terms and

\[ H(a,b)+\frac1m=k \]

for an integer \(k\). Then \(A/B=(km-1)/m\). Since \(km-1\) is relatively prime to \(m\), the reduced denominator must be \(B=m\), and consequently \(B\) divides \(A+1\). This is a necessary-and-sufficient arithmetic test for a fixed block once its reduced numerator and denominator have been computed. It does not say that there are finitely many such blocks.

What is known

The primary source is the passage on printed page 34 of Paul Erdős and Ronald L. Graham, Old and New Problems and Results in Combinatorial Number Theory, Monographies de l’Enseignement Mathématique 28, Université de Genève, 1980, MR 0592420. There the authors define the same consecutive reciprocal sum, say that the sum of two such quantities is probably an integer only finitely often, and immediately explain that this may be difficult because even adding one reciprocal, \(H(a,b)+1/n\), is not known to be integral only finitely often. This is a problem statement and research outlook, not a theorem proving finiteness.

The pinned catalogue snapshot at teorth/erdosproblems@536408716724a2402bdfa7c1d764bf83d64a8c89 records Problem 288 as informal status open, with last update 2025-08-31; its formal-status field is unformalized, while separate metadata says the statement has been formalized. Those labels must be read literally: a captured formal statement is not a proved formal theorem. The catalogue entry’s Git blob is 4e60263a80af6207aca494d2e1388e44fa372889.

As finite evidence, the source audit enumerated every \(1\le a\le b\le1600\) in the singleton test. It found exactly the six tuples \((a,b,m,k)\) for which \(H(a,b)+1/m=k\):

\[ (1,1,1,2),\ (1,2,2,2),\ (2,2,2,1),\ (1,3,6,2),\ (2,3,6,1),\ (3,6,20,1). \]

My exact-rational rerun found the same set, with no extra tuple, and reproduced the source’s canonical-output SHA-256 7340b340c98419be46b50a0d46e9029b794c761edea658ee4d11f073b4d8507e. This establishes only a bounded census. It cannot exclude a later example and therefore cannot resolve the infinite problem.

GCL's current state

The authenticated source-audit record is original issue #1558, with exact RESULT/1 comment 6135726557. Its disposition is SOURCE_LOCK_READY, not proved, certified, or published. The protected catalogue registry identifies the problem with issue-body hash c650bde42441d8251af1baee3c313575afb2c8c0d5775b9b0f65c52b1a36052c; that registry identity is documentary provenance, not mathematical evidence.

The audit checked the 1980 scan (SHA-256 0cbf0c32f0ab1e1c71db5121a88bac905bf976c4a6ab6bb6d7d9cf9ddd184ed3), the live problem-page discussion, the pinned catalogue bytes, and the formal-conjectures snapshot google-deepmind/formal-conjectures@85f863718beeec7b58a3a1926ee92e3472bc2020. The corresponding Lean file has blob 738afcb5e13022836ae70dddad1f9841655d142a; its four theorem bodies end in by sorry, and the external artifact was not replayed in this work. A source file, an external author manuscript, or a claimed successful Lean/CI run is therefore not GCL mathematical certification. No proof status upgrade is asserted here.

There is one documentary wrinkle worth preserving. The RESULT/1 narrative says the issue had zero comments, although the exact linked RESULT/1 is itself a current issue comment. I treat the linked comment and API identity as authoritative for this article and do not use that stale sentence as a mathematical claim. The audit’s substantive disposition remains SOURCE_LOCK_READY; it is not PARTIAL or STATUS_CONFLICT.

This file itself remains a source-qualified documentary explanation. Protected publication or later editorial admission would be a separate documentary and rendering process, and neither event would prove Problem 288.

What remains

The missing mathematical result is a theorem establishing either finiteness of all qualifying two-interval pairs or infinitude of such pairs. The smallest clearly stated unresolved interface in the source audit is already the singleton variant: determine whether only finitely many \((a,b,m)\) satisfy \(H(a,b)+1/m\in\mathbb{Z}\). A solution of that variant would be relevant evidence, but it would still need to be related carefully to the two-interval statement; it should not be silently relabeled as a solution of the parent problem.

A bounded successor task can reproduce the singleton census through \(b=1600\) using exact rational arithmetic, apply the denominator test above, and check the six-tuple list and digest. Its falsification hooks are concrete: a missing row, an extra row, or a disagreement in the exact fractions. Its stopping rule is equally important: after the finite range is exhausted, report a bounded computation and stop. Extending the cutoff may generate evidence, but no finite search can by itself prove the required global finiteness claim.

The external formal statement can support notation and quantifier review, but it cannot be treated as a proof until an authorized, reproducible formal-certification route establishes that status. Likewise, website comments, a working preprint, heuristic estimates, and informal attributions may guide research without becoming verified theorems in this documentary article.

Sources and claim boundaries

  • Primary source: P. Erdős and R. L. Graham, Old and New Problems and Results in Combinatorial Number Theory, Monographies de l’Enseignement Mathématique 28 (1980), printed p. 34, scan, MR 0592420. The scan is the source for the definition and the cautious open-problem wording. No DOI is asserted for this monograph because none was confirmed in the checked bibliographic records.
  • Exact GCL evidence: issue #1558 and RESULT/1. The comment’s source-audit claims, including its finite census and limitations, are reproduced only with those limitations.
  • Source registry identity: Problem 288 at the pinned catalogue commit above, catalogue Git blob 4e60263a80af6207aca494d2e1388e44fa372889; protected GCL registry issue-body hash c650bde42441d8251af1baee3c313575afb2c8c0d5775b9b0f65c52b1a36052c.
  • Formal comparison source: FormalConjectures/ErdosProblems/288.lean, blob 738afcb5e13022836ae70dddad1f9841655d142a. It is an unreplayed statement artifact with unfinished theorem bodies, not certification.

The article claims only what these sources and the displayed exact checks support: Problem 288 is open in the pinned record; the primary passage is on printed p. 34; the example equals \(1\); and the finite singleton census was reproduced. It makes no claim of a proof, novelty, priority, publication, editorial acceptance, or GCL mathematical certification.

Catalogue and evidence provenance

  • Pinned catalogue source: teorth/erdosproblems at 536408716724.
  • GCL issue: #1558 — 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