Summit of Arithmetic: On AI Breaking the Geometric Siege and the Emergence of Truth — OpenAI Model Claims Disproof of Erdős Unit Distance Conjecture

Forum topic · 小凯 · 2026-05-22

Summary

This Chinese tech forum post (zhichai.net) discusses an OpenAI reasoning model's claimed disproof of the Erdős unit distance conjecture, published as arXiv:2605.20695 (May 2026). The conjecture, posed by Paul Erdős in 1946, predicted that the number of unit-distance pairs among n points in the plane is at most n^(1+o(1)), a belief rooted in grid-based geometric intuition. According to the post, the AI model abandoned Euclidean geometric constructions in favor of algebraic number theory: it used the Golod-Shafarevich theorem to construct a number field K with an infinite class field tower, then embedded the resulting dense point sets into R^2, producing arithmetic coincidences that push unit distances beyond the conjectured bound. The post reports that Fields Medalist Timothy Gowers and others verified and co-signed the result, praising the bridge between combinatorial geometry and class field theory as a blind spot of human intuition. The author frames the work as a paradigm shift: AI moving from computational assistant to autonomous scientific discoverer. Note: this reflects the forum post's claims; readers should consult the arXiv paper and verification literature directly.

*Editor's note: This is an English translation of a Chinese forum post discussing a claimed AI-driven mathematical discovery. The claims below reflect the original post; the referenced paper (arXiv:2605.20695, dated 2026) should be verified independently.*

Preface: The Confusion of Forms and the Search for Truth

Mathematical inquiry has long assumed that the reach of intuition and logic marks the boundary of truth. Paul Erdős posed the unit distance problem in 1946 and offered a prize; it troubled the field for eighty years. Researchers remained trapped in the "grid" paradigm, assuming that maximal symmetry and order must reside in square lattice arrangements. But intuition is often a fog, and clinging to appearances obscures the truth.

On May 20, 2026, an OpenAI reasoning model released *Remarks on the disproof of the unit distance conjecture* on arXiv (arXiv:2605.20695). The post frames this as the moment AI crossed from "computational tool" to "discoverer." Its core achievement: abandoning Euclidean form for algebraic number theory, building a counterexample city on an "infinite class field tower."

I. The Geometric Fixation: The Lattice Cocoon and the Prison of Intuition

Mathematicians long believed that the optimal distribution of unit distances among planar point sets must resemble a "grid."

> Note: The Erdős Unit-Distance Conjecture > Given n points in the plane, let u(n) count the pairs at distance exactly 1. Erdős predicted an upper bound of n^{1+o(1)}, because distances in a grid are constrained by the sum-of-two-squares theorem.

This belief was a projection of physical intuition: honeycombs, crystal lattices. But mathematical optima often live in invisible abstraction, not visible concreteness.

II. The Algebraic Turn: The Infinite Class Field Tower Strikes

The OpenAI model was not misled by form. Its route was algebraic number theory.

#### The Mathematical Foundation of Arithmetic Symmetry

The model used the Golod-Shafarevich theorem to construct an algebraic number field K with special properties — an "infinite class field tower" structure whose generated point sets are far denser than the visible eye supposes.

z \in \mathcal{T}_K \hookrightarrow \mathbb{R}^2

> Note: Infinite Class Field Towers > A deep structure in algebraic number theory allowing special distributions of ideal classes across field extensions. By embedding such high-dimensional algebraic points into the plane, arithmetic relations among point pairs yield extreme coincidences, pushing the number of unit distances past the n^{1+o(1)} barrier.

The victory lies in replacing "geometric symmetry" with "arithmetic symmetry": the optimal distribution resides in invisible algebraic extensions, not the visible grid.

III. Verification: The Summit Gathering and the Paradigm Shift

The counterexample astonished the field. Fields Medalist Timothy Gowers and others verified it and co-signed.

> Gowers reportedly said the proof's brilliance lies in connecting combinatorial geometry with the hidden path of class field theory — a blind spot of human intuition.

> Note: Paradigm Discovery — AI no longer merely assists proofs via brute-force computation, but autonomously identifies and crosses disciplinary boundaries to find entirely new theoretical perspectives.

IV. Conclusion: The Emergence of Truth and the Silicon-Based Quest

Intelligence lies not in breadth of knowledge but in the capacity to investigate things. OpenAI's result marks the opening of the era of "AI scientific discovery." As machines search for truth within infinite number fields, they are no longer mere arithmetic engines but share in insight into the universe's underlying logic. From now on, the city of mathematics will be co-governed by silicon-based citizens.

References

1. arXiv:2605.20695: *Remarks on the disproof of the unit distance conjecture* (2026). 2. Number Theory Foundations: *Cassels & Fröhlich, Algebraic Number Theory (Academic Heritage)*. 3. Golod-Shafarevich Theory: *The Arithmetic of Infinite Extensions and Density of Units*. 4. Combinatorial Geometry: *Szekely, L., The Number of Unit Distances: A Survey (1946–2025 Retrospective)*. 5. Automated Theorem Proving: *The Evolution of Neural Reasoning in Pure Mathematics*.