--- title: Erdős Unit Distance Conjecture Update url: https://www.emergentmind.com/topics/erdos-unit-distance-conjecture type: topic --- # Erdős Unit Distance Conjecture Update The Erdős unit distance conjecture is a central problem in discrete and combinatorial geometry, concerning the maximal number of pairs of points separated by distance one among $n$ points in the Euclidean plane. Formulated by Paul Erdős in 1946, the conjecture predicted that this number, denoted $u(n)$, is bounded by $n^{1+o(1)}$ as $n\to\infty$, i.e., that it grows only very slightly faster than linearly for large $n$. Recent developments, culminating in the 2026 explicit counterexample and quantitative refinements, have dramatically altered this landscape, demonstrating that $u(n)$ can exceed $n^{1+\varepsilon}$ for a fixed positive $\varepsilon$, and thus falsifying Erdős's conjecture. ## 1. Formal Statement and Historical Context Let $P\subset\mathbb{R}^2$ be a set of $n$ points. The unit distance function is $$ u(n) = \max_{|P|=n} \#\left\{\{x, y\} \subset P : \|x - y\| = 1 \right\} $$ Erdős conjectured that for every $\epsilon > 0$, there exists $C_{\epsilon}$ such that $$ u(n) \leq C_{\epsilon} n^{1+\epsilon} $$ for all $n$, or equivalently $u(n) = n^{1+o(1)}$ as $n\to\infty$. Prior to 2026, the best upper bound, due to Spencer, Szemerédi, and Trotter, was $$ u(n) = O(n^{4/3}) $$ which arises from incidence theory between points and algebraic curves, specifically unit circles [2601.18831], [2507.15679]. The best lower bounds were of the form $n^{1 + O(1/\log\log n)}$, obtained via lattice constructions. ## 2. Disproof and Quantitative Refinements In 2026, a counterexample produced by OpenAI and independently verified by a team including Alon, Bloom, Gowers, Litt, Sawin, Shankar, Tsimerman, Wang, and Matchett Wood showed that $u(n)$ can exceed $n^{1 + c}$ for fixed $c > 0$ and all sufficiently large $n$ [2605.20695], [2605.20579]. The construction is deeply arithmetic, using Minkowski lattices arising from the ring of integers of high-degree number fields with controlled root discriminant, together with a pigeonhole principle on split primes to build large “unit shells.” The initial exponent $c$ was extremely small (on the order of $6.2 \times 10^{-38}$), but any such $c>0$ is enough to refute the conjecture [2605.20579]. Shortly thereafter, Will Sawin produced an explicit construction and quantitative lower bound, establishing that for a computable $\delta > 0$, $$ u(n) > n^{1+\delta} $$ for infinitely many (indeed, arbitrarily large) $n$, where currently $\delta$ can be taken as $0.014114\ldots$ [2605.20579], later increased to $0.01526\ldots$ via computational optimization [2606.03419]. This dramatic advance shattered the conjectural barrier $u(n) = n^{1+o(1)}$. ## 3. Explicit Construction and Certificate Methodology Sawin's explicit lower bound and subsequent optimizations rely on selecting algebraic number fields $K/F$ (typically CM extensions of totally real fields) with properties: - Small relative root-discriminant $\operatorname{rd}_{K/F}$. - Many primes of small norm splitting with controlled local data. The construction then employs a “certificate” comprised of: - A finite set of odd primes $T$. - A selected set $S_Q$ of rational primes. - Integer multiplicities $k(p)$ for $p \in S_Q$. - A real parameter $R > 1$ (encoded as a rational). For any explicit choice of such data obeying number-theoretic and tower constraints (notably, the Golod–Shafarevich inequality), an explicit formula yields $$ u(n) > n^{1 + \delta} $$ with $\delta = \delta(T, S_Q, k, R)$ computable via explicit arithmetic expressions [2606.03419]. Recent computational approaches have optimized these certificates using integer programming and evolutionary search heuristics. The best current values support $u(n) > n^{1.0152}$ for infinitely many $n$ [2606.03419]. ## 4. Structural and Algebraic Frameworks The qualitative barrier of $u(n)=O(n^{4/3})$ was historically explained combinatorially via point-circle incidence bounds, but this fails to distinguish the actual Euclidean metric structure from more general topological settings ("pseudo-circles"). Recent work, both before and after the counterexample, has focused on algebraic rigidity and the presence of rigid substructures in “dense” unit-distance graphs [2507.15679], [2601.18831]. Specifically, if a unit-distance graph exceeds a critical density, it must contain rigid bipartite subgraphs (usually $K_{3,3}$ or higher), whose Euclidean embedding varieties collapse dimensionally—this “algebraic collapse” rules out the existence of highly amorphous extremal structures that would support $n^{4/3}$ scaling for $u(n)$ in the plane. Symbolic elimination and Cayley–Menger variety computations formalize the impossibility of large flexible clusters for Euclidean distances, leading to new (though non-explicit) improvements over the combinatorial $O(n^{4/3})$ bound [2601.18831]. ## 5. Optimization, Asymptotics, and Open Directions ### Optimization of Certificates The explicit-certificate approach is now formulated as an integer nonlinear programming problem, where one seeks choices of $(T, S_Q, k, R)$ maximizing $\delta$ subject to number-theoretic feasibility. Methods used include: - Deterministic greedy heuristics based on prime-scoring and “budget” constraints. - Tailored integer evolution strategies (mutation, repair, discrete recombination) that search the space of parameters efficiently [2606.03419]. These pipelines have produced improved certificates, all passing rigorous arithmetic validation, and showing incremental but concrete increases in $\delta$.
Certificate/Method Achieved Exponent $\delta$ Key Features
Sawin (2026) 0.0141 Published explicit parameter set
Greedy Optimization 0.01517 Score-based deterministic selection
Integer ES 0.01526 Stochastic, mutation/repair strategy
Integer ES + Discrete Recomb. 0.01526 Two-parent recombination; current best
### Asymptotic Prospects While the concrete optimized exponents remain small ($1.0152$ in the best instance), the limit of the procedure—using only class group lower bounds—reaches an exponent of $1.24295$, meaning the method may not construct sets surpassing $n^{1.24295}$ unit distances without fundamentally novel ideas [2606.03419]. The density of values of $n$ for which such constructions exist is also of interest: current certificates guarantee infinitely many $n$ (arising from layers of the relevant field tower) but not all $n$. ## 6. Connections to Incidence Geometry, Rigidity Theory, and Distribution The upper-bound landscape is evolving. While the $O(n^{4/3})$ bound still holds generically, rigidity-induced scaling laws and analysis of unit-distance configuration varieties suggest sub-$n^{4/3}$ bounds specifically for Euclidean settings (as opposed to the more general topological or combinatorial models) [2601.18831]. This links the unit distance problem to rigidity theory, Cayley–Menger varieties, and algebraic geometry. Separately, when sets are equipped with strong “uniform distribution” properties or equidistribution in the torus, the conjectural bound $u(n)=O(n^{1+\varepsilon})$ becomes provable under additional hypotheses [2202.05359]. Well-distributed sets, or those with small diameter relative to distance, also admit improved upper bounds for distance incidences, notably for small distances or in higher dimensions [1709.08048]. ## 7. Future Prospects and Outstanding Problems The primary open directions concern: - Further optimization of certificate exponents, potentially by exploiting deeper analytic number theory (class group estimates, alternative field constructions). - Establishing explicit positive $\delta$ for upper bounds in a fully algebraic-geometric framework, potentially quantifying the “rigidity-induced” decrement below $n^{4/3}$. - Extending explicit constructions to denser sets of $n$ and better arithmetic densities. - Understanding the ultimate potential and limitations of number-theoretic versus geometric/rigid-analytic constructions. - Bridging the gap between explicit lower bound constructions and “generic” (non-adversarial) sets, possibly through analyses based on uniform distribution or energy minimization. The 2026 disproof and following research reframe the Erdős unit distance conjecture, establishing that $u(n)$ can be substantially superlinear for infinite families of $n$ and inviting a new synthesis of combinatorial, algebraic, and analytic tools in discrete geometry [2606.03419], [2605.20579], [2601.18831], [2507.15679]. Source: https://www.emergentmind.com/topics/erdos-unit-distance-conjecture