Erdős Problem #314
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
Start at an integer \(n \geq 1\) and add consecutive reciprocals:
Let \(m\) be the first index at which this running sum reaches or passes \(1\). The overshoot is
Because \(m\) is the first crossing, the preceding sum is strictly below \(1\) and the displayed quantity is nonnegative. Erdős and Graham asked whether infinitely many starting points come unusually close to the threshold, in the precise sense
The factor \(n^2\) is important: a typical final reciprocal is of order \(1/n\), so the question asks for much more accurate-than-usual crossings.
A small example
The calculation can be checked with exact fractions. Take \(n=8\). Summing through \(19\) gives
whereas adding the next term gives
Thus the minimal index is \(m=20\), the overshoot is \(\epsilon(8)=378871/77597520\), and
This verifies the definition, the strict minimality check, and the exact arithmetic for one finite input. It does not establish any limiting statement: one example, even a small overshoot, cannot prove that arbitrarily large \(n\) have the required behavior.
What is known
The primary source located for this audit is Jeck Lim and Stefan Steinerberger, “On differences of two harmonic numbers,” arXiv:2405.11354v3 (11 June 2024), especially Theorem 1 on printed p. 1. It states that for every \(c>0\) there are infinitely many positive integer pairs \((m,n)\) such that
For any such pair, move left to the first index at which the sum reaches \(1\). Monotonicity makes that first-crossing overshoot no larger than \(c/n^2\). Since \(c\) is arbitrary, this proves the affirmative answer to the original question. The argument uses asymptotics for harmonic sums and a selected subsequence of continued-fraction convergents to \(e\).
Their Theorem 2, on printed p. 2, gives a refined but differently normalized statement: for every \(\varepsilon>0\), infinitely many pairs satisfy
The paper expressly notes that this version does not impose that the sum exceed \(1\). It therefore should not be silently identified with the minimal-crossing function \(\epsilon(n)\), although it is relevant quantitative evidence. The stronger lower-growth expectation \(n^{2+\delta}\epsilon(n)\to\infty\) for every \(\delta>0\) remains presented as a belief, not as a proved theorem in the sources read here.
There is also a useful logical distinction between “infinitely many pairs” and “the least crossing index.” If a sum reaches \(1\) at some \(m\), then the first crossing exists because the partial sums increase by positive terms. Replacing \(m\) by that first crossing can only remove positive summands, so it preserves the lower bound \(1\) and decreases the overshoot. This is the small normalization argument that transfers Theorem 1 to the exact definition in the problem. It does not transfer the absolute-value estimate in Theorem 2 without an additional above-threshold argument. The primary paper’s own wording and the live page’s update preserve that distinction, even though catalogue summaries naturally compress both results under “proved.”
GCL's current state
The assignment’s frozen catalogue source is teorth/erdosproblems data/problems.yaml at commit 536408716724a2402bdfa7c1d764bf83d64a8c89, entry 314, Git blob 4e60263a80af6207aca494d2e1388e44fa372889. It records informal status “proved,” formal status “Lean,” and statement-formalized metadata “yes”; its informal status date is 2025-08-31 and its formalized metadata date is 2026-08-03. The packet classifies the item as OUTSIDE_STRICT_OPEN_TRIAGE.
The maintained Problem 314 page was last edited 23 January 2026. It reports the affirmative result and the logarithmic refinement, and links the published paper as Lim and Steinerberger, Mathematika 71 (2025), e70009, DOI 10.1112/mtk.70009. The current upstream FormalConjectures file at commit 6fbb54f24ccc2e64dcfaffc28c58950e377110d2 defines the block sum over Finset.Icc, wraps the liminf equality in answer(True), and ends the theorem with by sorry. It is a statement/formalization report, not a replayed proof or GCL certification. The issue’s pinned FC-GDM-85F86371 snapshot has no admitted statement file for this problem, so the newer upstream file is not treated as a frozen GCL dependency.
What remains
The main mathematical question is externally resolved, but a bounded successor can make the interfaces safer. First, compare the published version with arXiv v3 and record whether Theorem 1’s one-sided construction is explicitly converted to the minimal index in the final text. Second, formalize only the finite, elementary bridge: if a pair \((n,m)\) has sum at least \(1\) and at most \(1+c/n^2\), prove that the first crossing index has overshoot at most \(c/n^2\). This isolates the normalization step from the continued-fraction argument. A separate task could then audit the exact hypotheses and dependencies of the continued-fraction construction, without claiming a Lean proof until a checked artifact exists.
Sources and claim boundaries
The original assignment and its source packet are MATH-PROGRAMME issue #1584. The original Erdos–Graham reference is P. Erdős and R. Graham, Old and new problems and results in combinatorial number theory (1980), p. 41, cited by the primary paper and the maintained page. The primary proof source is arXiv:2405.11354v3; the retrieved PDF SHA-256 was 8f499a6db6d4239041668620ddad62622b157ee252003df8ba8f511f536bdac5.
The exact finite calculation above was independently evaluated with rational arithmetic during this audit. It is not a proof of the asymptotic theorem. Catalogue labels, the live page, the current FormalConjectures sorry statement, and informal forum reports are evidence with different scopes; none is a GCL mathematical certificate. V2 ledger rules decide accepted coverage. Citation of the original SOURCE RESULT/1 is authenticated original RESULT/1; no future comment or result link is inferred here.
Catalogue and evidence provenance
- Pinned catalogue source: teorth/erdosproblems at
536408716724. - GCL issue: #1584 — protected issue/assignment identity; its labels and results are live, not reproduced automatically here.
- GCL triage: OUTSIDE_STRICT_OPEN_TRIAGE.
- Formal-statement coverage: No statement was confirmed in GCL's pinned formal-provider snapshot. Presence does not mean the statement or proof has been checked.
- External mathematical status: consult current problem page and its dated bibliography.