Erdős Problem #290
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: The protected V2 ledger determines acceptance. This documentary explanation is not mathematical certification.
The question in plain language
For positive integers \(a\leq B\), put the consecutive unit-fraction sum into lowest terms:
where \(u_{a,B}\) and \(v_{a,B}\) are coprime integers and \(v_{a,B}>0\). Erdős Problem #290 asks whether every starting point \(a\geq1\) is eventually followed by an endpoint at which the reduced denominator gets smaller when the last term is added:
The quantifiers matter. This is not merely the claim that some denominator drop occurs, nor that drops occur for infinitely many starting points. It asks for at least one later endpoint for each fixed \(a\). Van Doorn proves more: for every fixed \(a\), there are infinitely many such endpoints, and the first one is at most linear in \(a\).
There are two endpoint conventions in the evidence. The maintained Erdős Problems entry writes the two adjacent sums with endpoints \(b\) and \(b+1\). The pinned Formal Conjectures file, blob 55bc70e2663d52304004b50d0a40da6bc1411dcd, uses the pre-addition endpoint \(b\):
Thus its \(b\) corresponds to the question's \(B-1\). This is an index shift, not a different mathematical assertion. The original Erdős–Graham citation is Old and New Problems and Results in Combinatorial Number Theory (1980), p. 34; the authenticated source audit did not obtain a direct scan for page-level quotation, so the book wording is retained here as a bibliographic and source-reported boundary rather than silently reconstructed.
A small example
Take \(a=3\). Exact fraction reduction gives
Consequently,
So \(B=6\) is a witness for \(a=3\), while the formal pre-addition witness is \(b=5\). This is an exact integer-arithmetic calculation, not a decimal approximation. A reproducible Python 3 check is:
from fractions import Fraction
def h(a, b):
return sum((Fraction(1, n) for n in range(a, b + 1)), Fraction(0))
for b in (4, 5, 6):
q = h(3, b)
print(b, q, q.denominator)
assert h(3, 6).denominator < h(3, 5).denominator
It prints 4 7/12 12, 5 47/60 60, and 6 19/20 20. The authenticated RESULT/1 audit records the same exact check and reports that the explicit witness construction was checked for \(1\leq a\leq20\). A finite computation illustrates the statement; it cannot establish the universal quantifier over all positive \(a\).
What is known
The primary resolving source inspected for this article is Wouter van Doorn, “On the non-monotonicity of the denominator of generalized harmonic sums,” arXiv version 2, revised 23 July 2025. In its introduction, the paper defines \(u_{a,b}/v_{a,b}\) and states that in the original all-coefficients-one case there are infinitely many \(b>a\) with \(v_{a,b}<v_{a,b-1}\). Its Theorem 2 gives the explicit bound
where \(b(a)\) is the first successful endpoint. The same paper gives the simpler bound \(b(a)\leq6(a-1)\) for \(a>1\) and, in Theorem 8,
The theorem is stronger than the original existence question: it supplies infinitely many successful endpoints for every fixed \(a\), an effective upper bound, and asymptotic information about the first one. The source archive identified by the audit has SHA-256 3a838d971ea669b049f880e9c38c64dfd1f70482b63477d0c3d1bac063ee3904.
Accordingly, the mathematical status of the registered question is proved, in the sense that this inspected primary preprint gives an affirmative theorem. The live maintained entry and the pinned catalogue record agree with that status. This wording does not turn a catalogue label, a web exposition, or a formal source reference into GCL certification.
GCL's current state
The authenticated source audit is MATH-PROGRAMME issue #1560, read under GitHub identity fyremael; its exact RESULT/1 comment has disposition BOUNDED_TASK_DESIGNED. The issue remains open. The Programme registry entry binds problem 290 to issue 1560, records issue-body SHA-256 7c8a7ed622f3350681d0b7449f95ffd6cc19f505dea8126d9c7fd7d3d44a9df7, catalogue status proved, and triage class OUTSIDE_STRICT_OPEN_TRIAGE.
The pinned catalogue snapshot is teorth/erdosproblems@536408716724a2402bdfa7c1d764bf83d64a8c89; its Problem 290 Git blob is 4e60263a80af6207aca494d2e1388e44fa372889. That packet reports informal status proved, formal status Lean, and statement metadata formalized: yes, with the dates recorded in the issue audit. These are fixed provenance records, not independent proof authority.
The pinned Formal Conjectures theorem has body by sorry. It is useful for checking the intended proposition and its endpoint convention, but it is not a proof. The linked external Lean source has commit 0dd69298bc5870970257f002ed376ac35aa8fa8c and blob 69434d891a60c3e8843b849141b5e4ee218191bc; it states the stronger witness bound \(a<B\leq6a\). The audit did not reproduce that file in Lean 4.24.0 with the pinned Mathlib commit or regenerate its axiom report. Therefore neither the sorry-based file, the external proof repository, nor self-reporting CI is treated as mathematical certification.
What remains
The original yes/no question has a primary affirmative resolution, but several documentary and verification tasks remain distinct from that result. First, a source-aware reader should keep the p. 34 Erdős–Graham citation separate from van Doorn’s later theorem: the authenticated audit retained the direct-scan limitation, so no unverified page quotation or stronger historical wording should be added.
Second, the endpoint adapter deserves explicit checking in any formal replay. The mathematical statement compares \(v_{a,B}\) with \(v_{a,B-1}\), whereas the pinned theorem writes harmonicDen a (b + 1) < harmonicDen a b. A replay must verify that Finset.Icc a b, positive-integer restrictions, rational denominator conventions, and the substitution \(B=b+1\) all match the intended statement. A type-checking theorem with sorry does not perform this comparison for us.
Third, the bounded formal successor task recorded by the audit is to replay the external source at exactly Lean v4.24.0 and Mathlib f897ebcf72cd16f89ab4577d0c826cd14afaafc7, regenerate the axiom report, and check the endpoint adapter and finite arithmetic. A clean replay would be evidence about that artifact, not a replacement for source reading or a grant of GCL certification. The stopping condition is one exact-environment replay or one precisely documented environment/source mismatch; it does not authorize upgrading the status of the pinned sorry theorem.
Sources and claim boundaries
- Live Erdős Problems #290 — maintained statement, example, bibliographic pointer
[ErGr80,p.34], and reported affirmative status. Its proof claims are not treated as independently verified. - P. Erdős and R. L. Graham, Old and New Problems and Results in Combinatorial Number Theory (1980), p. 34 — the original cited source. The authenticated #1560 audit records the direct page scan as unavailable for its review; no page-level quotation is asserted here.
- van Doorn, “On the non-monotonicity of the denominator of generalized harmonic sums,” arXiv:2411.03073v2 — inspected primary theorem text, especially Theorems 2 and 8; source-archive SHA-256 is
3a838d971ea669b049f880e9c38c64dfd1f70482b63477d0c3d1bac063ee3904. - Pinned catalogue packet — fixed entry and blob
4e60263a80af6207aca494d2e1388e44fa372889; metadata is provenance, not certification. - Pinned Formal Conjectures source — commit
85f863718beeec7b58a3a1926ee92e3472bc2020, blob55bc70e2663d52304004b50d0a40da6bc1411dcd, statement only because its proof isby sorry. - Original issue #1560 and the exact authenticated source comment — source audit, finite verification, semantic normalization, replay boundary, and
BOUNDED_TASK_DESIGNEDdisposition.
This manuscript therefore calls #290 proved on the strength of the inspected van Doorn theorem, while preserving the direct-source, endpoint-semantics, formal-replay, and certification boundaries. Protected editorial acceptance remains determined only by the V2 ledger.
Catalogue and evidence provenance
- Pinned catalogue source: teorth/erdosproblems at
536408716724. - GCL issue: #1560 — 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.