Skip to content

Erdős Problem #317

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.

The question in plain language

Erdős Problem #317 asks how close two different signed sums of the first \(n\) unit fractions can be. For each \(n\), choose coefficients \(δ_1,\ldots,δ_n\) from \({-1,0,1}\) and form

\[ S_n(δ)=Σ_{k=1}^n \frac{δ_k}{k}. \]

The first question asks whether there is one constant \(c>0\) such that, for every \(n\geq1\), at least one choice has a nonzero value satisfying

\[ 0<|S_n(δ)|<\frac{c}{2^n}. \]

The constant must not depend on \(n\). The zero value is deliberately excluded: it is usually easy to make a signed sum vanish, but the problem concerns a genuinely small nonzero sum.

The second question asks whether there is an \(N\) such that, for every \(n\geq N\), every nonzero such sum obeys

\[ |S_n(δ)|>\frac{1}{L_n}, \qquad L_n=\mathrm{lcm}(1,2,\ldots,n). \]

These are distinct asymptotic questions. The first seeks an upper bound of exponential scale \(2^{-n}\) for at least one choice; the second asks for an eventual strict lower bound for all nonzero choices.

A small example

Take \(n=4\) and coefficients \((δ_1,δ_2,δ_3,δ_4)=(0,1,-1,-1)\). Then exact rational arithmetic gives

\[ S_4=\frac12-\frac13-\frac14=-\frac1{12}. \]

Here \(L_4=\mathrm{lcm}(1,2,3,4)=12\), so

\[ |S_4|=\frac1{12}=\frac1{L_4}. \]

This verifies, without floating-point approximation, that the second question's strict inequality fails at \(n=4\). The calculation was independently checked with exact fractions and the assertions \(S_4=-1/12\), \(L_4=12\), and \(|S_4|=1/L_4\). It does not disprove an eventual statement, and it says nothing by itself about the first question's constant \(c\). It is also the explicit finite counterexample represented in the pinned formal source.

What is known

The primary source is 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. 42. It states the small-sum question in the signed-coefficient form, says that the authors do not know how to prove the proposed \(c/2^n\) bound, and records the lower bound \(|S_n|\geq1/L_n\) whenever \(S_n\ne0\). The lower bound follows directly because \(L_nS_n\) is an integer. The same page says that equality occurs for small \(n\), including the displayed \(n=4\) example, and describes eventual strictness as expected but unproved in that source.

The maintained Erdős Problems entry still presents both questions and records the problem as open. Its additional note reports a weaker upper bound, of the shape

\[ 2^{-n(\log\log\log n)^{1+o(1)}/\log n}, \]

attributed to arguments in the discussion by Vjekoslav Kovac and Wouter van Doorn, together with a heuristic about the likely order of magnitude. Those are useful research leads, but the page discussion is not itself a primary proof record. A January 2026 comment by Woett gives a related pigeonhole formulation using subset sums \(Q_n\) and a result cited there as Problem #320; that external report was not independently source-audited here.

GCL's current state

The source packet for issue #1587 pins the catalogue at teorth/erdosproblems commit 536408716724a2402bdfa7c1d764bf83d64a8c89, data/problems.yaml, entry blob 4e60263a80af6207aca494d2e1388e44fa372889. That entry records informal status open (last update 2025-08-31), formal status unformalized, and statement-formalized metadata yes. The live catalogue entry checked on 12 October 2026 has the same problem status fields; its page says it was last edited 6 January 2026.

The pinned provider is FormalConjectures commit 85f863718beeec7b58a3a1926ee92e3472bc2020, file FormalConjectures/ErdosProblems/317.lean, blob ab2b4b3a3ecce6a5f958b4544a3b08a454a1c411. Its main theorem uses \(δ:\mathrm{Fin}(n)\to\mathbb{Q}\) and denominator \(k+1\), which matches denominators \(1\) through \(n\) after reindexing. It also includes a second-question variant and the \(n=4\) counterexample. The principal theorem and the variant are declared with by sorry; the presence of a formal statement is therefore not a proof or a certification of semantic equivalence. The original SOURCE RESULT/1 citation is authenticated original RESULT/1. V2 ledger alone decides accepted coverage.

What remains

The smallest useful successor is a bounded source-and-finite-certificate review. Its input should be the printed p. 42 source, the pinned catalogue entry, the pinned Lean file, and the live page and discussion. First, the reviewer should write a line-by-line normalization of both quantifier patterns, including the nonzero condition, the eventual threshold, the indexing shift, and the exact type of \(c\). Second, an exact-arithmetic program should enumerate all coefficient vectors for \(1\leq n\leq12\), record the minimum nonzero absolute sum and the scaled quantity \(L_n|S_n|\), and emit a machine-readable certificate that can be independently checked. The finite output can test the strict inequality and expose indexing errors; it cannot decide either asymptotic question.

A separate source check is needed before treating the weaker bound as established: identify a primary paper or complete argument for the Kovac/van Doorn report, then compare its hypotheses and quantifiers with the first question. If that source cannot be located or does not prove the displayed bound, the report must remain explicitly unverified. No finite enumeration, formal statement containing sorry, or documentary article can supply the missing asymptotic theorem.

Sources and claim boundaries

This article reports what the pinned primary source, live page, formal source, and exact finite check support. It does not claim that either question is solved, that the later discussion is a verified theorem, or that GCL has supplied mathematical certification.

Catalogue and evidence provenance

  • Pinned catalogue source: teorth/erdosproblems at 536408716724.
  • GCL issue: #1587 — 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