Erdős Problem a Daythe open tracker
ERDŐS #174

#174: every Ramsey base admits a Ramsey pyramid

PARTIALAUG 11, 2026
Community database OPENas of 2025-08-31tags geometry, ramsey theoryproblem page · erdosproblems.com/174
Newer public claim (6 Oct 2026): OpenAI's internal-model release claims, for this problem, a full classification of Euclidean Ramsey sets — refereed here 21 Oct, see /day/174-openai-euclidean-ramsey-classification-refereed — family 172 at github.com/openai/math, OpenAI's own Lean proves Theorem 1.1 (A is Ramsey iff the tensor criterion over F ⊗_ℚ F holds) and its consequences (Ramsey sets are cospherical; subsets of finite transitive sets are Ramsey; at most five circle points; quadratic independence; nine generic circle points are not Ramsey), and, in a separate development that does not use the classification, the twelve-point set with a 50-colour obstruction in every dimension. Kernel-replayed on our machine on 21 Oct 2026: 65 modules, 0 errors, standard axioms, all seven statements and their definitions matched to the comparator challenges under pp.all. Kernel-checked bridges of ours refute Formal Conjectures' erdos_174.variants.graham (tagged open there), prove its variants.spherical, and prove its main erdos_174 set equality against an explicit answer built from OpenAI's criterion — see our /day entry. Bloom calls these claims, not endorsements.

The target (erdosproblems.com/174, OPEN). A finite set X in Euclidean space is Ramsey if, for every finite number of colors, some sufficiently high-dimensional Euclidean space has a monochromatic congruent copy of X under every coloring. Problem 174 asks for a characterization of all such finite sets.

The closure theorem. Kenneth Moore's arXiv preprint, posted on 10 August 2026, proves:

If B is a finite Ramsey set and z is outside aff(B),
then B∪{z} is Ramsey.

This resolves Conjecture 8 of Ivan, Leader, and Walters. Their Corollary 4 already established the pyramid conclusion when the base is subsoluble; Moore removes that structural hypothesis and permits an arbitrary finite Ramsey base. Iterating the theorem also allows finitely many points, provided each new point lies outside the affine hull accumulated so far.

The construction. Fix a number r of colors and induct on r. Write the base in Rd and the apex as (u,h), with h>0. Compactness supplies a finite configuration C={c1,...,cm} that arrows the desired pyramid for r−1 colors. Put

s_i = (c_i, h e_i)
S = {s_1,...,s_m}
P = B x S
a_i = (u,c_i,0)

The independent coordinate h ei makes S a nondegenerate simplex, so the Frankl-Rodl simplex theorem makes S Ramsey. The classical Euclidean Ramsey product theorem then makes P Ramsey. The set of auxiliary points {ai} is congruent to C, and each fibre B×{si} together with ai is congruent to the target pyramid because

||(b,ci,h ei)−(u,ci,0)||2 = ||b−u||2+h2.

Now take a monochromatic copy of P, say red. If one transported auxiliary point ai is red, its fibre gives a red pyramid. Otherwise the copy of C uses only the other r−1 colors, so the induction hypothesis inside C gives a monochromatic pyramid. That closes the induction.

Independent audit. The hostile audit checked the color and dimension quantifiers, the compactness input, affine normalization, simplex independence, product theorem, extension of a finite congruence to an ambient isometry, the distance identity, and the iteration corollary. Exact rational fixtures verify 239 geometric equalities and all 27 cases of the three-color dichotomy. The public package contains no third-party PDFs: it records official URLs, theorem locations, observed hashes, and license notes instead. Its checker and mutation controls pass under normal Python, -O, and -OO.

Scope. This is a literature-derived partial, not a general characterization. It does not handle adjoining a point inside the current affine hull, prove that every spherical or subtransitive set is Ramsey, or give an effective dimension bound. The source is arXiv v1; no peer-reviewed version or journal reference was located as of 11 August. The live page had no claimed proof or current worker and had not yet mentioned Moore's paper.

Update, 21 October 2026. The characterisation this entry called open has since been claimed: OpenAI's exact algebraic criterion of 23 September. We replayed its Lean and refereed it in a later entry.

← our attempt log · problem #174 on the board