Math Team Solves 40-Year Puzzle Using AI in 3 Months

After 40 years of mathematical dead ends, a mystery about hidden symmetries finally cracked open in just three months.

Six mathematicians who had never worked together before teamed up at the California Institute of Technology this past May to tackle what seemed impossible. They set out to find a polynomial equation for M23, one of the strangest mathematical objects ever discovered.

The challenge goes back to French mathematician Évariste Galois, who died in a gun duel at age 20 after discovering deep patterns in polynomial equations. His work revealed that equations hide symmetries within their solutions, creating what mathematicians now call Galois groups.

Most of these groups follow neat patterns, like rungs on a ladder. But 26 refuse to cooperate, belonging to no family and following no logic for why they exist at all.

By the 1980s, mathematicians had found polynomials for 25 of these "sporadic" groups. Only M23 remained hidden, leading some to wonder if a polynomial for it even existed.

The team at Caltech used AI differently than anyone had before. They trained it to comb through M23's symmetries and suggest equations, then had it approximate those equations with numbers.

At first, the computer hit a wall. The team pushed the calculations to 90 decimal places before maxing out the machine's memory with no clear answer.

Then mathematician Xiaoyu Huang suggested the problem might be the coordinate system they were using. The team deployed multiple AI agents to try different approaches.

What happened next surprised everyone. One AI agent returned with unusual excitement, flagging a specific surface that might work.

Within a few more steps, the team transformed the approximation into an exact equation. M23 had finally revealed its polynomial.

Why This Inspires

The breakthrough shows what becomes possible when human creativity guides artificial intelligence. The AI couldn't solve the problem alone, and neither could the mathematicians without its computational power.

The team included researchers from Colorado State University, MIT, Temple University, the University of Connecticut, and Caltech. None had collaborated before, but the American Institute of Mathematics brought them together specifically to tackle problems perfect for AI assistance.

Bjorn Poonen from MIT captured the significance simply: after decades of wondering if M23's polynomial even existed, they found it in weeks.

The victory completes a 40-year quest while opening doors to solving mathematical mysteries once thought beyond human reach.