OpenAI Disrupts Discrete Geometry with an Unexpected Breakthrough

A Decades-Old Question Finally Resolved

For nearly eight decades, an intriguing question has captivated the minds of mathematicians: how many pairs of points can be exactly one unit apart when placing n points in a plane? This problem, known as the unit distance problem, was first formulated by the renowned mathematician Paul Erdős in 1946. Considered one of the most famous and simplest problems to state in combinatorial geometry, it has been extensively studied but never solved until recently. The book Research Problems in Discrete Geometry by Brass, Moser, and Pach, published in 2005, describes it as potentially the most well-known problem in this field. Noga Alon, a prominent figure in combinatorics at Princeton, refers to it as "Erdős's favorite problem." Erdős even offered a reward for its resolution.

A Revolutionary Breakthrough by OpenAI

A significant breakthrough has recently been made regarding this long-standing problem. Since Erdős's initial work, the prevailing belief was that "square grid" configurations were optimal for maximizing the number of pairs at unit distance. However, a model developed by OpenAI challenged this hypothesis, providing an infinite series of examples that polynomially improve upon this approach. The validity of this proof has been confirmed by a group of external mathematicians, who also drafted an explanatory document to contextualize the importance of this discovery.

A General Reasoning Model at Work

What makes this result particularly remarkable is the method by which it was obtained. The proof was generated by a general reasoning model, rather than a system specifically designed to solve mathematical problems or tackle the unit distance problem. As part of a broader initiative to assess the capabilities of advanced models to contribute to cutting-edge research, this model was tested on a series of problems posed by Erdős. It successfully produced a proof for this open problem.

A Major Step for Mathematics and Artificial Intelligence

This proof represents a significant advancement for both the mathematics and artificial intelligence communities. It is the first time a major open problem, central to a subfield of mathematics, has been autonomously solved by an AI. This also demonstrates the ability of these systems to support complex reasoning. Mathematics provides an ideal framework for testing reasoning: problems are well-defined, proofs can be verified, and a lengthy argument holds only if the reasoning is coherent from start to finish. The method used to solve the problem has also brought new and sophisticated ideas from algebraic number theory to a fundamental geometric question.

Enthusiastic Reactions from the Mathematical Community

The mathematical community has reacted enthusiastically to this discovery. Noga Alon expressed his admiration by stating, "This is one of Erdős's favorite problems. I believe it is fair to say that every mathematician working in combinatorial geometry has thought about this problem. The solution provided by OpenAI's internal model is, in my opinion, an outstanding achievement, solving a long-standing open problem."

Tim Gowers, a Fields Medalist, called this result a "milestone in AI mathematics." He added that if a human had submitted this paper to the Annals of Mathematics, he would have recommended its acceptance without hesitation. Arul Shankar, a renowned number theorist, highlighted the originality of the model's reasoning: "The model's CoT is deeply interesting. A significant majority of the reasoning attempts to construct a counterexample to the widely accepted upper bound, rather than trying to prove it."

Finally, Jacob Tsimerman expressed his approval by stating, "This is truly impressive work, and I would accept it for any journal without hesitation."

The Unit Distance Problem in Detail

To better understand the magnitude of this discovery, it is useful to delve into the problem itself. Let u(n) be the maximum number of pairs at unit distance among n points in the plane. Simple examples achieving a linear growth rate are easy to construct: by placing n points in a line, one obtains n-1 pairs, while a square grid produces about 2n². The best known construction so far, based on a resized square grid, even reached n^{1 + C / log log(n)} for some constant C.

For decades, it has been widely accepted that this rate was the best possible, and that no other construction could significantly surpass the square grid. Erdős conjectured an upper bound of n^{1+o(1)}. The new result refutes this conjecture. More specifically, for an infinite number of values of n, the proof proposes configurations of n points with at least n^{1+δ} pairs at unit distance, for some fixed exponent δ > 0.

The Contribution of Algebraic Number Theory

The proof begins with a familiar geometric idea and pushes it in an unexpected direction. Erdős's original lower bound can be understood through Gaussian integers: numbers of the form a + bi, where a and b are integers and i is the square root of -1. Gaussian integers extend ordinary integers and, like them, benefit from properties such as unique factorization into prime numbers.

The innovative argument replaces Gaussian integers with more complex generalizations from algebraic number theory, featuring richer symmetries capable of creating many unit-length differences. Tools such as infinite class field towers and Golod–Shafarevich theory demonstrate that the number fields necessary for the argument actually exist.

Implications for the Future of Mathematics

This result marks a turning point in the interaction between AI and mathematics. An AI system has autonomously solved a long-standing open problem at the heart of an active field. It also offers a promising glimpse into a new type of collaboration between AI and human mathematicians. In this case, the companion work of external mathematicians significantly enriched the original solution, highlighting the potential for future collaborations between humans and machines.