Skip to content

Erdős Problem #654

Editorial depth: Reviewed explanatory article — evidence qualified.
Snapshot classification: open (not automatically the live result).
Subject areas: geometry, distances.

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

For each point in a set of \(n\) planar points, stand at that point and count the different distances to the other \(n-1\) points. Equal distances are counted only once. Suppose no four of the points lie on one circle. Must there be a point from which almost all the other points are at different distances?

Write

\[ r_X(p)=\bigl|\{\lVert p-q\rVert:q\in X,\ q\ne p\}\bigr|, \qquad f(n)=\min_{\substack{|X|=n\\\text{no four concyclic}}} \max_{p\in X}r_X(p). \]

Erdős Problem #654 asks for the size of \(f(n)\) and, more specifically, whether \(f(n)>(1-o(1))n\). A second, weaker proposed bound is \(f(n)>(1/3+c)n\) for some fixed \(c>0\) and all sufficiently large \(n\).

As of October 2026, these are not equally unresolved questions: a published 2026 construction disproves the nearly \(n\) assertion, while the maintained catalogue still regards the fixed improvement over one third as open. A related general-position question also requires that no three points be collinear. That additional restriction changes what constructions are permitted.

A small example

Take four points on the coordinate axes:

\[ X=\{(0,3),(0,-3),(2,0),(-2,0)\}. \]

From \((0,3)\) the three distances to the other vertices are

\[ 6,\quad \sqrt{13},\quad \sqrt{13}. \]

From \((2,0)\) they are \(4,\sqrt{13},\sqrt{13}\). By symmetry the other two vertices behave the same way. Thus every vertex has exactly two distinct distance values, despite having three neighbours.

These four points are not on one circle. Indeed, a circle with equation

\[ x^2+y^2+Ax+By+C=0 \]

passing through both vertical-axis points would require \(C=-9\), whereas passing through both horizontal-axis points would require \(C=-4\). That is impossible.

Consequently \(f(4)\leq 2\). This is a small model of the published construction. A finite example by itself says nothing decisive about the question for every sufficiently large \(n\).

What is known

The elementary lower bound. Fix any admissible \(X\) and any point \(p\in X\). Each positive-radius circle centred at \(p\) contains at most three other members of \(X\): if it contained four, those four would be concyclic. Since there are \(n-1\) other points,

\[ r_X(p)\geq \left\lceil\frac{n-1}{3}\right\rceil. \]

In particular, \(f(n)\geq(n-1)/3\). This simple counting argument does not rule out configurations where many equal-distance triples occur at most centres. Showing that the geometry of the different centres prevents that near-saturation is the unresolved direction.

The published upper-bound counterexample. In Section 3.1 of Feng and collaborators (2026), arXiv:2601.22401v3, an Aletheia-assisted construction places points on the two coordinate axes. For an integer \(m\geq10\), set \(n=4m\) and \(K=\{10,\ldots,m+9\}\). Use

\[ P=\{(0,\pm3^k):k\in K\}, \qquad Q=\{(\pm2^j,0):j\in K\}, \qquad X=P\cup Q. \]

There are exactly \(2m\) points on each axis. Why can no four lie on one circle? A circle can meet each axis in at most two points; any four-point circle would therefore have two points \((0,y_1),(0,y_2)\) on the vertical axis and two \((x_1,0),(x_2,0)\) on the horizontal axis. Substitution into its equation gives

\[ y_1y_2=C=x_1x_2. \]

Taking absolute values then yields

\[ 3^{k_1+k_2}=2^{j_1+j_2}, \]

which is impossible for positive integer exponents. Thus the no-four-concyclic hypothesis really is satisfied. Notice that merely saying the points lie on two axes would not be enough; the prime-power choice prevents circles meeting each axis twice.

At a point \((0,\pm3^{k_0})\in P\), the other \(2m-1\) points on the vertical axis contribute at most \(2m-1\) different distances. The distances to \(Q\) have the form

\[ \sqrt{3^{2k_0}+2^{2j}}\qquad(j\in K). \]

The two signs of each horizontal-axis coordinate produce the same distance, so the \(2m\) cross-axis points contribute exactly \(m\) values. We only need an upper bound on the union, hence

\[ r_X(p)\leq(2m-1)+m=3m-1. \]

The analogous estimate holds for a point on \(Q\). It follows that

\[ f(4m)\leq 3m-1<\frac34(4m)=\frac{3n}{4} \qquad(m\geq10). \]

This infinite family disproves any lower bound \((1-o(1))n\), because such a bound would eventually exceed \(3n/4\). The argument is explained here in full but is a published result, not a new GCL discovery.

What it does not prove. The inequality \(f(4m)<3n/4\) is consistent with \(f(n)>(1/3+c)n\) for sufficiently small positive \(c\). Moreover, the construction has many collinear triples; it cannot decide the separate question that assumes no three collinear.

The maintained source discussion distinguishes the versions and keeps the residual question labelled OPEN. It attributes the stronger proposed general-position hypothesis to earlier Erdős and Erdős–Pach sources, reported as [Er87b, p.168], [ErPa90, p.267], and [Er97e, p.530]. Those original page passages were not independently inspected for this draft.

GCL's current state

The original authenticated RESULT/1 source return has historical disposition PARTIAL. This status and the original research exceptions are preserved, not silently converted into a theorem verdict. The maintained Erdős source and source-specific mathematical definitions were reviewed, and this article is editorially reviewed as a source-exception-annotated explanation under EC-BULK-005-R10, effective only after the exact-hash protected admission merges.

Source interpretation and unresolved mathematical qualification: A published 2026 near-n conjecture counterexample is distinct from the open one-third-plus-constant target and general-position refinements.

The explanation is accepted for publication of established background, examples, and explicitly attributed current claims. It does not establish or formally certify the disputed original theorem, external model-assisted proof, historical interpretation, or any unreplayed Lean kernel file. Original source-fidelity and proof-review research holds remain.

What remains

The immediate source-fidelity obligation is to inspect the original Erdős passages cited as [Er87b, p.168], [ErPa90, p.267] and [Er97e, p.530] and distinguish the exact hypotheses and claims. The publication is not an adequate substitute for checking what the earlier authors actually wrote.

The mathematical problem then is whether there is a fixed \(c>0\) and \(n_0\) so that every admissible set of \(n\geq n_0\) points has some \(p\) with at least \((1/3+c)n\) distinct distances, particularly under the stronger no-three-collinear condition. Such a proof must use compatibility of distances across different centres, not merely count three points per circle. Alternatively, a genuine counterexample must satisfy every stated geometric restriction for arbitrarily large \(n\).

Before editorial admission, the issue-bound partial return must receive critical source review, the draft must pass the relevant math and MkDocs checks, the consolidated cohort must be protected-published and the public mathematics must be inspected in a browser. A later V2 exact-hash admission is separate.

Sources and claim boundaries

Authority boundary: independently checked elementary counting and finite geometry, and an explicitly attributed published counterexample. No claim of new mathematics, resolution of the \((1/3+c)n\) question, complete historical source lock, protected merge, MathJax browser verification, formal proof certification or V2 editorial admission.

Qualified editorial critical review: R10 source-exception dossier. The source-semantic exception is visible and its original research hold remains. Publication classification: REVIEWED EXPLANATORY ARTICLE WITH UNRESOLVED SOURCE/PROOF QUALIFICATIONS. Independent theorem proof or Lean certification: NOT CLAIMED.

Catalogue and evidence provenance

  • Pinned catalogue source: teorth/erdosproblems at 536408716724.
  • GCL issue: #1925 — protected issue/assignment identity; its labels and results are live, not reproduced automatically here.
  • GCL triage: STATEMENT_CAPTURE_REQUIRED.
  • 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.

Back to catalogue · Compendium introduction