For nearly 80 years, mathematicians have wrestled with a deceptively simple question: if you place n points on a plane, how many pairs can be exactly distance 1 apart? This is the planar unit distance problem, first posed by Paul Erdős in 1946. Despite its elementary description, the problem has remained one of the most challenging open questions in combinatorial geometry.
Now, an internal OpenAI model has disproved a longstanding conjecture related to this problem. The AI system found an infinite family of examples that achieve a polynomial improvement over previously known constructions, marking the first time a prominent open mathematical problem has been solved autonomously by AI.
Why this problem matters to mathematicians
The unit distance problem holds special significance in mathematics. Erdős himself offered a monetary prize for its resolution, and Princeton’s Noga Alon describes it as “one of Erdős’ favorite problems.” The 2005 book Research Problems in Discrete Geometry calls it “possibly the best known (and simplest to explain) problem in combinatorial geometry.”
Since Erdős’s original work, mathematicians widely believed that “square grid” constructions were essentially optimal for maximizing unit-distance pairs. These constructions achieve a growth rate of n^(1+C/loglog(n)) for a constant C, which is only slightly faster than linear growth. The prevailing conjecture suggested no construction could significantly improve on this rate.
The breakthrough and its surprising methods
The OpenAI model disproved this conjecture by constructing configurations with at least n^(1+δ) unit-distance pairs for some fixed exponent δ > 0. A forthcoming refinement by Princeton professor Will Sawin shows one can take δ = 0.014.
What makes this result particularly surprising is its use of sophisticated concepts from algebraic number theory. The proof employs:
- Infinite class field towers
- Golod-Shafarevich theory
- Extensions of Gaussian integers with richer symmetries
- Unique factorization properties in algebraic number fields
These advanced number theory concepts had never been connected to elementary geometric questions about distances in the Euclidean plane. The unexpected bridge between these mathematical areas may open new research directions.
Expert reactions and verification
A group of external mathematicians has verified the proof and written a companion paper explaining its significance. Fields medalist Tim Gowers calls the result “a milestone in AI mathematics.” Leading number theorist Arul Shankar notes: “This paper demonstrates that current AI models go beyond just helpers to human mathematicians – they are capable of having original ingenious ideas, and then carrying them out to fruition.”
Noga Alon emphasizes the result’s unexpected nature: “The fact that the correct answer is not n^(1+o(1)) is surprising, and the construction and its analysis apply fairly sophisticated tools from algebraic number theory in an elegant and clever way.”
How the AI discovered the proof
The breakthrough came from a general-purpose reasoning model, not a system specifically trained for mathematics. As part of testing whether advanced models can contribute to frontier research, OpenAI evaluated the system on a collection of Erdős problems. The model produced this proof without being targeted at the unit distance problem specifically.
The success demonstrates the depth of reasoning these AI systems now support. Mathematics provides a clear testbed for reasoning capabilities because problems are precise, proofs can be verified, and long arguments only work if reasoning holds together from start to finish.
Implications for mathematics and AI research
This result represents more than solving a specific conjecture. As mathematician Thomas Bloom notes in the companion paper: “This shows that there is a lot more that number theoretic constructions have to say about these sorts of questions than we suspected; moreover, that the number theory required can be very deep.”
The discovery may inspire algebraic number theorists to examine other open problems in discrete geometry. Bloom suggests broader implications: “The frontiers of knowledge are very spiky, and no doubt the coming months and years will see similar successes in many other areas of mathematics, where long-standing open problems are resolved by an AI revealing unexpected connections.”
What this means beyond mathematics
The capabilities demonstrated here extend beyond pure mathematics. If a model can maintain coherent complex arguments, connect ideas across distant knowledge areas, and produce work that survives expert scrutiny, these abilities prove valuable in biology, physics, materials science, engineering, and medicine.
These advances point toward more automated research systems that can help scientists explore more ideas and tackle harder technical questions. However, human expertise remains crucial. AI can help search, suggest, and verify, but people still choose which problems matter, interpret results, and decide what questions to pursue next.
As AI begins taking serious roles in research’s creative aspects, this breakthrough reinforces the importance of understanding AI development challenges, aligning highly intelligent systems, and shaping the future of human-AI collaboration.
