Remarks on the disproof of the unit distance conjecture

Abstract: We present a short, digested, human-verified version of the recent OpenAI-generated counterexample to the Erdős unit distance conjecture, and a sequence of reflections on it. The argument relies crucially on ideas that may, at least in retrospect, be attributed to Ellenberg-Venkatesh, Golod-Shafarevich, and Hajir-Maire-Ramakrishna.

• The paper demonstrates an AI-verified construction that refutes Erdős's unit distance conjecture by exhibiting point sets with unit distances exceeding n^(1+ε).
• The method integrates analytic number theory, lattice expansion, and high-degree CM field techniques to derive superlinear lower bounds.
• Implications include cross-disciplinary insights and a reimagined role for AI in verifying complex mathematical proofs in combinatorial geometry.

Disproof of the Erdős Unit Distance Conjecture: Technical Summary and Reflections

Introduction and Background

The unit distance problem, formulated by Erdős in 1946, asks for the maximal number $U(n)$ of unit distances determined by $n$ points in the Euclidean plane. While elementary constructions, such as point lattices, yield lower bounds of the form $U(n) \geq n^{1 + \Omega(1/\log \log n)}$, and combinatorial geometry supplied an upper bound of $O(n^{4/3})$ (Spencer, Szemerédi, Trotter), the prevailing conjecture had been $U(n) \leq n^{1 + o(1)}$—that is, the unit distance conjecture. Despite considerable attention and numerous attempts, the conjecture remained unresolved, engendering a canonical challenge in discrete geometry with deep connections to algebraic and arithmetic geometry.

This paper, "Remarks on the disproof of the unit distance conjecture" (2605.20695), provides a formal, human-verified account of an AI-generated construction that refutes the unit distance conjecture. Specifically, it supplies a family of increasingly large point sets $P_i \subset \mathbb{R}^2$ such that the number of unit distances in $P_i$ satisfies $|\nu(P_i)| \geq |P_i|^{1 + \varepsilon}$ for some $\varepsilon > 0$, thus surpassing the conjectured $n^{1 + o(1)}$ upper bound.

Main Technical Content

The essence of the argument involves an overview of analytic number theory, the geometry of numbers, and infinite class field towers. The central technical result can be distilled as follows:

Theorem. There exists $n$0 and a sequence of finite point sets $n$1 with $n$2 such that the number of unit distance pairs in $n$3 satisfies $n$4.

The proof strategy deviates sharply from previous constructions by employing high-degree CM (complex multiplication) fields and exploiting the asymptotic behavior of lattices arising from infinite class field towers. The approach is underpinned by two key lemmas:

Lattice Expansion via Units of Norm One

The first lemma leverages the geometry of numbers to convert the existence of many algebraic integers with complex absolute value one into sets with numerous unit distance pairs. Explicitly, given a full-rank lattice $n$5 with many elements of coordinatewise modulus $n$6, one constructs polydiscs of bounded radius whose lattice translates, after projection, yield point sets in the plane exhibiting a superlinear number of unit distances. The quantitative bounds are critical: If $n$7 for some $n$8 (with $n$9 a bound on the skewness), then polynomial growth in $U(n) \geq n^{1 + \Omega(1/\log \log n)}$0 guarantees the superlinear exponent in the cardinality of unit distances.

Construction of Many Modulus-One Algebraic Units

The second lemma is a combinatorial argument on number fields, rooted in the pigeonhole principle, that exploits the combinatorics of principal ideal classes and class numbers. For a CM field $U(n) \geq n^{1 + \Omega(1/\log \log n)}$1 with many pairs of complex-conjugate primes lying over a fixed rational prime $U(n) \geq n^{1 + \Omega(1/\log \log n)}$2 (achieved via infinite class field towers with prescribed splitting conditions), a large supply of elements of $U(n) \geq n^{1 + \Omega(1/\log \log n)}$3 with absolute value $U(n) \geq n^{1 + \Omega(1/\log \log n)}$4 in every embedding is constructed. Quantitative lower bounds on the size of this set follow from controlling the class number and the discriminant, with explicit connections to the ideal-theoretic structure of $U(n) \geq n^{1 + \Omega(1/\log \log n)}$5.

The application of these lemmas involves careful choice of the arithmetic setting: the construction proceeds by fixing a set of rational primes $U(n) \geq n^{1 + \Omega(1/\log \log n)}$6 and an additional split prime $U(n) \geq n^{1 + \Omega(1/\log \log n)}$7, then considering the infinite tower of number fields unramified outside $U(n) \geq n^{1 + \Omega(1/\log \log n)}$8 in which $U(n) \geq n^{1 + \Omega(1/\log \log n)}$9 splits completely. The underlying lattice in $O(n^{4/3})$0 is built from the ring of integers of these fields, after scaling by suitable denominators to handle the units with modulus $O(n^{4/3})$1.

A critical numerical consequence is that the explicit exponent surpassing $O(n^{4/3})$2 can be realized using concrete choices of primes (for example, $O(n^{4/3})$3 and $O(n^{4/3})$4), with the proof giving detailed estimates for all relevant parameters.

Contradiction to Conventional Beliefs and Numerical Thresholds

The paper's contradictory claim is that $O(n^{4/3})$5 is false, overturning the conventional consensus in combinatorial geometry. The construction shows that for certain highly structured, non-classical configurations derived from algebraic number theory, one can exceed any given exponent $O(n^{4/3})$6. The lower bound achieved by the construction is (for explicit parameter choices) $O(n^{4/3})$7 for some explicit, positive $O(n^{4/3})$8—a sharp improvement over the largest previous lower bounds.

Historical and Theoretical Context

The approach is inspired by previous applications of infinite class field towers—especially those of Golod and Shafarevich—and connects closely to work in analytic number theory, such as the construction of dense sphere packings, the study of large class numbers, and coding theory. However, the key novelty is the shift in viewpoint: instead of varying the set of allowed primes in a fixed field, the construction considers a tower of fields of increasing degree, with a fixed set of ramified places and infinitely many split primes. This perspective enables the bypassing of analytical obstructions inherited from the geometry of classical lattices.

Further, technical insights from Ellenberg–Venkatesh and Hajir–Maire–Ramakrishna concerning Frobenius-splitting in class field towers are foundational to the execution of the argument.

Critical Reflections and Implications

The expository sections feature substantive commentary from multiple leading mathematicians, highlighting several key implications:

• Cross-disciplinary Synergy: The success of the approach underscores the power of importing high-dimensional number-theoretical machinery into ostensibly low-dimensional combinatorial geometry.
• AI-Driven Discovery: The fact that an AI (specifically, an internal OpenAI model) produced the initial complete argument, which was then human-simplified and verified, marks a significant milestone for automated mathematical discovery. The AI's perseverance in pursuing counterexamples over proofs—contrary to typical human intuition—was essential.
• Potential for Extensions and Limits: While the construction definitively settles the unit distance question in the plane, analogous questions such as the distinct distances problem and unit distances in higher dimensions remain challenging due to deeper analytic obstacles related to the behavior of representations of integers by quadratic forms over number fields.
• Best Practice and Community Adaptation: The episode prompts reflection on academic practice, including attribution norms, the synergistic roles of AI and human researchers, and the planning required to maintain rigor and understanding as AI plays an increasingly substantial role in mathematical research.

Conclusion

This work provides a succinct, human-digestible exposition of the AI-generated disproof of the unit distance conjecture (2605.20695). By leveraging infinite class field towers and lattice constructions in high-degree CM fields, it constructs explicit families of planar point sets with more than $O(n^{4/3})$9 unit distances, decisively answering a decades-old question. The theoretical implications extend beyond discrete geometry, highlighting a surge in the interplay between algebraic number theory and extremal combinatorics, and signaling the pragmatic and philosophical issues introduced by AI-led mathematical research. Future progress will likely involve further application of deep arithmetic methods to classical geometric and combinatorial problems, and a broader rethinking of research strategies in light of the new capabilities demonstrated by generative AI.