Skip to content

Erdős Problem #318

Editorial depth: Reviewed explanatory article — evidence qualified.
Snapshot classification: solved (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 318 studies when signs can cancel a finite sum of unit fractions. Let \(A\) be an infinite arithmetic progression of positive integers, and let \(f:A\to\{-1,1\}\) be nonconstant. Must there be a finite nonempty subset \(S\subset A\) such that

\[ \sum_{n\in S}\frac{f(n)}{n}=0? \]

The page asks two related questions as well: does the same assertion hold for every set of positive density, and does it hold when \(A\) is the set of squares greater than \(1\)? The sign function is allowed to be arbitrary, but the selected set must be finite and have no repeated denominator. Thus the challenge is a finite balancing problem that must work for every nonconstant two-colouring of the relevant infinite set.

The three clauses have different answers. The arithmetic-progression clause is affirmative. The positive-density clause is false in general. The square clause, with \(1\) removed, is affirmative according to the maintained page and Daniel Larsen's readable 2026 proof note. These are distinct claims; a result about one of them does not silently settle the others.

A small example

Here is an exact finite cancellation among squares greater than \(1\):

\[ \frac{1}{2^2}=\frac{1}{3^2}+\frac{1}{4^2}+\frac{1}{5^2}+\frac{1}{7^2}+\frac{1}{12^2}+\frac{1}{15^2}+\frac{1}{20^2}+\frac{1}{28^2}+\frac{1}{35^2}. \]

It can be reproduced with exact rational arithmetic by setting left = Fraction(1, 2**2) and right = sum(Fraction(1, k**2) for k in [3,4,5,7,12,15,20,28,35]); the assertion left == right == Fraction(1, 4) succeeds. With \(f(4)=1\) and \(f(k^2)=-1\) on the nine right-hand denominators, this gives a nonempty finite signed sum equal to zero. In a common denominator calculation, the left numerator is \(44100\) over \(176400\), and the nine right numerators add to the same \(44100\).

This is a useful picture of what the square case requires: a finite identity whose two sides can be assigned opposite signs. It does not prove the universal statement, because one successful colouring and one selected set say nothing about every possible nonconstant colouring. Nor does a finite search establish an infinite theorem.

What is known

The historical references on the maintained page are Erdős and Straus, “Solution to Problem 387” (1975, p. 183), and Sattler, “Solution to Problem 387” (1975, pp. 184–189). They establish the natural-number and odd-number cases, respectively. Sattler's 1982 paper on the arithmetical sequence, pp. 347–352, is credited with proving the arithmetic-progression case. The page also reports that Sattler announced a square result in two 1982 papers, but that the announced paper on squares never appeared.

The positive-density question has a simple obstruction. If \(A\) contains exactly one even number \(e\), put \(f(e)=-1\) and put \(f(n)=1\) on every other element. A finite sum omitting \(e\) is positive. A sum containing \(e\) would require a sum of reciprocals of odd denominators to equal \(1/e\) plus or minus an odd-denominator rational. Clearing an odd common denominator shows that an odd-denominator rational cannot equal the reciprocal of an even integer. Hence no zero occurs. This is the observation attributed to Erdős through Sattler's 1982 discussion.

For the final square clause, Larsen's source document, Sufficiently abundant numbers are pseudoperfect, proves a stronger-looking partition statement. Theorem 6 on p. 13 of the PDF says that every partition of the squares greater than \(1\) into two nonempty parts \(X\) and \(Y\) has nonempty finite \(X'\subset X\) and \(Y'\subset Y\) with equal reciprocal sums. Taking the two parts to be the positive and negative signs gives the required cancellation. The proof uses an Egyptian-fraction circle-method theorem, so the finite identity above is only an illustration of the interface, not a replacement for that argument.

GCL's current state

The original issue #1588 assigns a source and reusable-lemma audit. Its registry entry binds Problem 318 to that issue, records catalogue status solved, the class OUTSIDE_STRICT_OPEN_TRIAGE, and topics number theory and unit fractions. The issue's source packet identifies the frozen catalogue commit and the protected formal-conjectures commit; the preflight found no existing RESULT/1 comment. The original SOURCE RESULT/1 citation is authenticated original RESULT/1.

The frozen catalogue bytes are teorth/erdosproblems@536408716724a2402bdfa7c1d764bf83d64a8c89, Git blob 4e60263a80af6207aca494d2e1388e44fa372889. They report informal status solved, with status last updated 2026-04-04, formal status Lean last updated 2026-09-16, and statement formalized. The live page was observed last edited 1 April 2026 and currently reports the same three-part resolution. These are maintained external records, not GCL mathematical certification.

The pinned formal file is google-deepmind/formal-conjectures@85f863718beeec7b58a3a1926ee92e3472bc2020, blob d6facd3660c7bf33c949e876455eb586f5b71c10. It defines \(P_1\) using functions on all naturals whose range is contained in \(\{1,-1\}\), while nonconstancy and the selected set are restricted to \(A\setminus\{0\}\). This is close to the page's \(f:A\to\{-1,1\}\) formulation but not textually identical, especially when \(0\) is present. Every research theorem in the pinned file ends in by sorry; that file is a statement record, not a checked proof.

What remains

The strongest documentary conclusion is therefore: the arithmetic-progression and square assertions have named external sources, while the positive-density assertion has an external counterexample mechanism. A bounded successor should compare the exact Larsen theorem with the square branch of \(P_1\), record the zero exclusion and sign-extension convention, and preserve the source-level fact that Sattler's announced square paper is unavailable. It should not launch an exhaustive Lean build or turn a sorry declaration into a certification.

The main source ambiguity is historical rather than logical. Sattler's 1982 full texts were not readable through the available publisher route during this audit, so the claims attributed specifically to those papers remain bibliographic reports. Larsen's TeX source is readable and supplies the accessible primary proof document. Any later formal replay would still need to check the imported theorem dependencies and the semantic bridge from equal reciprocal sums to the exact \(P_1\) proposition.

Sources and claim boundaries

Catalogue and evidence provenance

  • Pinned catalogue source: teorth/erdosproblems at 536408716724.
  • GCL issue: #1588 — 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