---
title: Disproof of Erdős Unit Distance Conjecture
url: https://www.emergentmind.com/topics/disproof-of-the-erdos-unit-distance-conjecture
type: topic
---

# Disproof of Erdős Unit Distance Conjecture

The Erdős unit distance conjecture, posed in 1946, asserts that for all $\epsilon>0$, there exists a constant $C_\epsilon$ such that any $n$-point set $P\subset \mathbb{R}^2$ determines at most $C_\epsilon n^{1+\epsilon}$ pairs of points at unit distance, or, equivalently, that the function $U(n)$—the maximal number of unit distances among $n$ planar points—satisfies $U(n)=n^{1+o(1)}$. This conjecture has shaped decades of combinatorial geometry, yet recent breakthroughs have established explicit counterexamples demonstrating that $U(n)$ can exceed $n^{1+\delta}$ for some absolute $\delta>0$ and infinitely many $n$, refuting the conjecture. These results synthesize techniques from discrete geometry, algebraic number theory, and the theory of infinite class field towers, revealing a deep and previously unexploited bridge between geometric and arithmetic phenomena [2605.20695], [2605.20579].

## 1. Formulation of the Erdős Unit Distance Conjecture and Its Disproof

The original conjecture is that for every $\epsilon>0$, there exists $C_\epsilon>0$ such that
\[
|\{ (x,y) \in P^2 : \|x-y\|=1 \}| \leq C_\epsilon n^{1+\epsilon}
\]
for any $n$-point set $P \subset \mathbb{R}^2$. Equivalently, $U(n)$, the maximal number of unit-distance pairs among $n$ points in the plane, grows as $n^{1+o(1)}$. Erdős's classical lower bound, given by placing points on an integer grid, yields $U(n) \gtrsim n^{1+\Omega(1/\log\log n)}$, and previously no construction substantially exceeded $n^{1+o(1)}$.

The work of Alon, Bloom, Gowers, Litt, Sawin, Shankar, Tsimerman, and Wang presents a human-verified proof of a counterexample originally generated by an OpenAI system, establishing:

**Theorem 1.1 (main counterexample):**  
There exists $\epsilon>0$ and an infinite sequence of finite planar point sets $P_1,P_2,\dots$ with $|P_i|\to\infty$ such that
\[
\nu(P_i) := |\{ (x,y)\in P_i^2 : \|x-y\|=1 \}| \geq |P_i|^{1+\epsilon}
\]
Thus, for infinitely many $n$, $U(n)\geq n^{1+\epsilon}$, disproving the conjecture $U(n)=n^{1+o(1)}$ [2605.20695].

Sawin constructs explicit sets with $n\to\infty$ points yielding more than $n^{1.014}$ unit-distance pairs, providing an explicit exponent [2605.20579].

## 2. Core Geometric and Number-Theoretic Lemmas

The proof architecture relies on transforming an arithmetic property of lattices in high-dimensional complex vector spaces into a geometric property of planar point sets.

Let $\Lambda\subset \mathbb{C}^f$ be a full-rank lattice, normalized so that all nonzero $x\in \Lambda$ have $|x_i|\geq \delta>0$ for some $i$ and the projection onto the first coordinate is injective. Let $U_\Lambda = \{u\in\Lambda : |u_i|=1 \ \forall \, 1\leq i\leq f\}$ denote the unit-norm lattice points. The geometric “window” plus “translate-and-count” lemma [2605.20695, Lemma 2.1] shows that for large $R$, one can obtain planar sets $P$ with
\[
|P| \leq \left( \frac{9R^2}{\delta^2} \right)^f, \quad
\nu(P) \geq \left(\frac{u\pi R^2}{4v\delta^2}\right)^f,
\]
where $u=|U_\Lambda|^{1/f}$ and $v$ is a constant depending on the lattice. If $u > 36v/\pi$, one obtains $\nu(P) \geq |P|^{1+\eta}$ for some fixed $\eta>0$ independent of $f$, and letting $f\to\infty$ produces infinitely many sets with superlinear unit-distance complexity.

## 3. Construction of Lattices with Many Unit-Norm Points

The lattice construction exploits properties of CM number fields $K$ (i.e., totally imaginary quadratic extensions of totally real number fields) and their rings of integers $\mathcal{O}_K.$ The approach generalizes the techniques of Ellenberg–Venkatesh in the context of $\ell$-torsion bounds in class groups:

- Fix a CM field $K$ of degree $2f$ and select $s$ distinct prime ideals $P_j$ of $\mathcal{O}_K$, none conjugate to another.
- Define $Q=\prod_j (P_j \overline{P}_j)^{k_j}$ for integers $k_j$, and set $U = \{ u\in Q^{-2} : |u|=1 \}$ where $|u|=1$ means $u$ has absolute value $1$ in every complex embedding.
- A pigeonhole argument in the class group yields $|U|\geq \prod_j (k_j+1) / h(K)$, with $h(K)$ the class number.

These unit-norm elements translate into planar point-set constructions via the geometric lemma, provided that the parameters $\delta$ and $v$ (minimum norm, covolume) are controlled [2605.20695].

## 4. Infinite Class Field Towers and Prime Splitting

To maintain bounded discriminant and control parameter growth as $f\to\infty$, the construction lets $K$ run through the layers of an infinite class-field tower as in the Golod–Shafarevich theorem:

- Take $K=L(i)$ for totally real fields $L$ in a tower unramified outside a finite set $T$ of rational primes, with all splitting completely at some $p\notin T$.
- The Golod–Shafarevich theorem ensures the infinitude of such towers, and the fields can be chosen so that parameters (root discriminant, minimum norm) stay bounded.
- The “Frobenius-cutting” technique of Hajir–Maire–Ramakrishna enables the construction of towers with infinitely many split primes satisfying required congruence conditions, crucial for assembling many unit-norm elements.

This infinite-tower machinery allows taking $f\to\infty$ while keeping all technical requirements met [2605.20695].

## 5. Explicit Lower Bounds and the Optimization of Parameters

Sawin refines this framework to deliver explicit bounds:

- The exponent $\delta$ in $n^{1+\delta}$ unit-distance pairs is computable in terms of the degree $d$, root discriminant $\lambda$ of $K/F$, prime sets $T$ and $S_\mathbb{Q}$, assignments of prime multiplicities $k(p)$, and geometric window radius $R$.
- By optimizing these parameters with
  - $T = \{3,5,7,11,13,17,19,23,29,31,37,41,43\}$,
  - $S_\mathbb{Q}$ a set of 22 primes,
  - $k(2)=50, k(3)=31, k(5)=21, \dots, k(179)=6,$
  - $R = 72$,
one achieves $\delta\approx 0.014114$, so for infinitely many $n$ there exist planar sets with $\geq n^{1.014114}$ unit-distance pairs [2605.20579].

## 6. Methodological Significance and Verification

All combinatorial and number-theoretic lemmas used in the proofs are classical or elementary, including:
- The geometry-of-numbers window-counting argument
- The class-group pigeonhole principle
- Golod–Shafarevich theory for infinite towers
- Class number bounds in terms of discriminant

The proofs have been independently digested and rewritten by human experts, ensuring each step is rigorous and transparently justified. The innovation lies in applying the degree-growth flexibility of class field towers (traditionally not exploited in combinatorial geometry), in conjunction with explicit prime splitting and the arithmetic properties of number fields.

## 7. Reflections, Extensions, and Impact

These results mark the first disproof of the Erdős unit distance conjecture, dramatically changing the landscape of combinatorial geometry. The approach underscores a profound new connection between discrete geometry and high-degree arithmetic, with potential implications for related extremal problems such as distinct distances and higher-dimensional analogues.

Prominent reflections from the authors and other mathematicians include:
- The realization that the possibility of increasing the number field degree offers far more constructional flexibility than previously considered,
- The observation that prior research predominantly focused on upper bounds,
- The suggestion that the AI-generated proof required only short “hint-sequences” for human verification,
- The novelty of varying the underlying number field, rather than just selection of special primes,
- The sociological and citation implications for mathematics in the AI era.

A plausible implication is that this arithmetic-geometric paradigm could stimulate advances in related incidence problems, and that analytic improvement of class number estimates or refinements to local conditions may further push the exponent above $1.014$ [2605.20695], [2605.20579]. However, certain limitations (such as the construction only producing infinitely many $n$) and natural barriers (e.g., trivial lower bound $4/3$ for exponent with these methods) remain.

---

| Paper                                         | Main Contribution                                             | Exponent Achieved         |
|------------------------------------------------|--------------------------------------------------------------|---------------------------|
| "Remarks on the disproof of the unit distance conjecture" [2605.20695] | Human-verified, concise exposition of the AI-derived counterexample; lattice and number-field machinery; explicit mechanism for infinite $n$ | Existence of $\epsilon>0$  |
| "An explicit lower bound for the unit distance problem" [2605.20579] | Fully explicit parameter selection; detailed proof with class-number and discriminant computation | $\delta \approx 0.014114$ |

The disproof inaugurates a new era in additive combinatorics and discrete geometry, establishing a paradigm wherein infinite-dimensional number-theoretic constructions yield sharp incidence bounds unreachable by geometric extremal methods alone.

Source: https://www.emergentmind.com/topics/disproof-of-the-erdos-unit-distance-conjecture