Remarks on the disproof of the unit distance conjecture

1. Introduction

• There exist point sets P_i in R2 with |P_i| tending to infinity and at least |P_i|^(1+ε) unit distances for some ε > 0.This contradicts Erdős’s conjectured n^(1+o(1)) upper bound.Source: §1 · p. 1, §1.1 · p. 1

1.1. History of the problem.

• The conjecture arose from Erdős’s grid construction, while the best known upper bound was O(n^(4/3)) and the conjectured bound was n^(1+o(1)).The unit-distance graph cannot contain K_2,3, yielding the classical O(n^(3/2)) bound before later improvements.Source: §1.1 · p. 1

1.2. Sketch of the proof.

• The proof constructs a large set U of magnitude-1 algebraic numbers in a number field K with bounded denominator D, and, assuming K is CM, realizes many as differences x − y from a bounded window W of D−1O_K.Constructing the bounded window requires embedding the ring of algebraic integers O_K as a lattice inside K ⊗ R.Source: §1.2 · pp. 1–2
• Infinite Golod–Shafarevich class-field towers keep root discriminants bounded while field degrees grow, and split primes help overcome class-number and discriminant losses.The resulting point sets are indexed by fields K with [K : Q] tending to infinity.Source: §1.2 · p. 2

1.3. Context for the proof.

• The novel feature is letting [K : Q] grow, yielding infinitely many totally real fields F where r2,F(4D^2) grows exponentially in [F : Q].Fixed-field bounds remain only O_{F,ε}(D^ε), while related constructions and split-prime methods have substantial prior literature [1],,, [20].Source: §1.3 · pp. 2–3

1.4. Organization of the paper.

• The paper first proves Theorem 1.1 and then collects reflections on the proof and the AI solution.The proof’s split-prime ingredient also relies on constructions already appearing in the literature.Source: §1.2 · p. 2, §1.3 · p. 3, §1.4 · p. 3

2.1. Statements of Main Lemmas.

• The two main lemmas are identified as the key novelty, with Theorem 1.1 deduced from them.The first lemma is geometric; the second constructs the needed algebraic elements.Source: §2.1 · p. 3
• If a lattice Λ has at least u^f unit-modulus boundary points, translating and projecting a bounded window produces a planar point set with many unit-distance pairs.The construction uses 2|U_Λ||Λ ∩ B_R−1| pairs among at most |Λ ∩ B_R| points.Source: §2.1 · p. 3
• The geometric lemma succeeds when u > 36v, while u, v, and δ remain constant as the lattice dimension f tends to infinity.Under these conditions, |P| tends to infinity and the logarithmic growth ratio stays above 1, proving Theorem 1.1.Source: §2.1 · p. 4
• Golod–Shafarevich towers provide CM fields of growing degree and bounded root discriminant, with split primes supplying the prime ideals needed by Lemma 2.2.The argument may use a single fixed split rational prime; stronger towers with infinitely many completely split primes were already known.Source: §2.1 · pp. 4–6
• The explicit parameters satisfy v = r/2, δ ≥ 101^(-2⌈18r^3/π⌉), and u = ⌈18r^3/π⌉r^(-2), with r = 2 · 3 · 5 · 7 · 11 · 13 · 17; moreover, u > 36v/π.Source: §2.1 · p. 6
• The original AI argument was more subtle than necessary: Proposition 2.3 supplies an infinite bounded-root-discriminant tower with infinitely many completely split primes q ≡ 1 mod 4.This makes it possible to take the number of split primes arbitrarily large relative to the ramified-prime parameter.Source: §2.1 · p. 6

2.2. Proofs of Main Lemmas.

• The proof of the geometric lemma averages translations, uses injective coordinate projection, and bounds lattice-point counts through separation and Minkowski-sum volume estimates.The algebraic lemma uses the pigeonhole principle on ideal classes to complete the construction.Source: §2.2 · pp. 6–7

2.1. Statements of Main Lemmas.

• Lemma 2.2 constructs many elements of absolute value 1 from prime ideals in a CM field, using pigeonholing among ideal classes and ratios of ideals in the same class.Its current form is useful only for CM fields, because the totally real case is vacuous and non-CM unit groups behave differently.Source: §2.1 · p. 4, §2.2 · pp. 6–7

• The result is described as a milestone in AI mathematics, but Gowers limits the claim by noting that AI’s advantage may depend on problem styles, proof complexity, and future efficiency gains.He suggests AI may excel at encyclopaedic connections and persistent search, while acknowledging that this assessment is speculative and could change.Source: §5 · pp. 9, 12
• The proposed “Kolmogorov complexity modulo experts” measures proof difficulty through hint length, reconstruction time, surprisal, and the cost of exploring possible hint sequences.The measure is intended to compare the complexity of proofs humans can find with what AI can find.Source: §5 · p. 10
• The first hints—look for a counterexample and generalize the best-known construction—lead from Erdős’s grid to Gaussian integers and then to number fields.The construction seeks integers represented as sums of two squares in many ways, then replaces Gaussian integers with integers in a totally real field and L(i).Source: §5 · pp. 10–11
• The disproof was especially informative because the conjecture’s falsity had low subjective probability, so the counterexample produced a large update in how the problem was perceived.Gowers contrasts this with a proof requiring a long sequence of surprising ideas, such as Guth and Katz’s proof of the distinct-distances conjecture.Source: §5 · pp. 9–11
• A further hint is to use number fields of increasing degree and bounded-norm prime ideals, but Gowers says the remaining calculations would still require substantial expert work.He emphasizes that the necessary hint length depends on the expert, motivation, and the time–hint tradeoff.Source: §5 · p. 11
• The solution to the unit-distance problem is described as a milestone in AI mathematics, with no previous AI-generated proof coming close to it.The passage says that, had a human written and submitted the paper to the Annals of Mathematics, the writer would have recommended acceptance without hesitation.Source: §5 · p. 12