AI Cracks an Erdős Conjecture for the First Time: A Historic Breakthrough on an Unsolved Math Problem

AI Solves an Unsolved Math Problem: A Historic Moment

At the intersection of mathematics and artificial intelligence, a milestone event is unfolding — AI has, for the first time in any meaningful sense, solved a well-known unsolved mathematical problem. This isn't some obscure fringe question, but rather what has been called "the most famous problem in combinatorial geometry" — an Erdős conjecture.

The mathematicians involved in this research were stunned. As one researcher described it: "When I saw the initial version of the model's output, I could barely believe it. I spent a long time reading it over and over, trying to figure out what was actually happening. My reaction was — you can't be serious, this sounds too good to be true."

The Heart of the Erdős Conjecture: Points in the Plane

The problem itself concerns points in the plane — a completely elementary geometric question. Yet its solution requires profoundly deep tools from algebraic number theory. Previously, the mathematical community generally believed that existing construction methods were essentially optimal, but AI proved that these constructions could in fact be dramatically improved.

Mathematical background of the Erdős conjecture: Paul Erdős (1913–1996) was one of the most prolific mathematicians of the 20th century, publishing over 1,500 papers and collaborating with more than 500 co-authors. He was renowned for posing numerous conjectures that were elegantly simple to state yet extraordinarily difficult to solve, often offering prize money — ranging from $25 to $10,000 — as incentives, with the amounts reflecting his subjective assessment of each problem's difficulty. The Erdős Unit Distance Problem in combinatorial geometry is one such example: given n points placed in the plane, what is the maximum number of pairs of points that can be exactly distance 1 apart? The upper bound for this problem has never been precisely determined. Erdős himself conjectured that the maximum number of pairs is on the order of n^(1+c/log log n). Despite the extreme simplicity of its statement, the problem touches on deep mathematical branches including algebraic number theory, graph theory, and harmonic analysis, and has attracted the sustained efforts of countless top mathematicians for over half a century.

This is precisely what makes the breakthrough so elegant: the problem statement is simple enough for anyone to understand, yet the path to a solution is unfathomably deep. Notably, the proof path relies heavily on tools from algebraic number theory — a field that studies algebraic integers, field extensions, and ideal decompositions. It serves as a core pillar of the "algebraic method," which encodes geometric or combinatorial problems as polynomial equations and then leverages algebraic tools to derive counting bounds. When Guth and Katz resolved the Erdős distinct distances problem in 2015, they did so by combining algebraic geometry with polynomial methods. AI's breakthrough here similarly depends on these deep tools, but the key difference is its ability to systematically search through an enormously vast space of algebraic constructions to find specific configurations that human intuition would struggle to discover. The research team applied the model to several problems posed by Erdős — problems that had captivated countless mathematicians for decades — and remarkably, the model produced a solution to one of the most important among them.

Why Couldn't Humans Complete This Proof?

A key question is: why couldn't humans carry out this method? The answer lies in the extreme complexity of the proof process.

As one researcher explained: "A human cannot execute this method. It's too fine-grained — there are too many decisions to make at every single step of the proof." It's like navigating a vast maze where each junction presents dozens of choices, and only one very specific path leads to the exit.

Understanding the "high-dimensional decision space": In the context of mathematical proofs, a high-dimensional decision space means that at each step of constructing a proof, one faces a large number of possible reasoning directions, lemma choices, and algebraic transformations. These choices combine to form an exponentially growing search tree. Human mathematicians rely on intuition, experience, and aesthetic judgment to "prune" — rapidly eliminating paths that "don't look right." But when every local choice at each step appears reasonable and only a global perspective can determine whether a path is promising, human working memory and attentional resources become the bottleneck. Modern large language models, combined with reinforcement learning (such as the "chain-of-thought" training paradigm used by OpenAI's o-series models) and Monte Carlo Tree Search (MCTS), can perform massive parallel exploration within these search trees while using reward signals to learn which intermediate steps are more likely to lead to valid proofs.

AI's advantage lies in its ability to explore all these possibilities far more comprehensively. Within this vast decision space, AI systematically searches, evaluates, backtracks, and ultimately finds a path to the solution. This isn't brute-force enumeration but rather an intelligent exploration — AI demonstrates an extraordinary ability to navigate high-dimensional decision spaces, essentially "outsourcing" mathematical intuition to statistical patterns and surpassing human cognitive limits in scale.

The Far-Reaching Impact of AI's Mathematical Breakthrough on Science

The significance of this breakthrough extends far beyond the Erdős conjecture itself. Several participants admitted that the result "kept them up for several nights." One researcher stated: "I always expected AI would do something like this, but this result dramatically shortened my expected timeline."

What does this mean? The researchers offered clear assessments:

• This is the first genuine mathematical breakthrough caused by AI — not auxiliary computation, but the independent discovery of a proof path that humans had failed to find
• This demonstrates that AI can achieve original breakthroughs in scientific domains — not limited to mathematics, but potentially extending to engineering, physics, biology, and medicine
• **This heralds a "golden age" of mathematics