Chinese AI solves decade-old math conjecture in 80 hours

Yesterday 08:20

By: Dakir Madiha

Chinese AI solves decade-old math conjecture in 80 hours

A research team at Peking University has developed an artificial intelligence system capable of solving and formally verifying a long-standing mathematical conjecture without human input, completing the task in roughly 80 hours.

The system addressed the Anderson conjecture, a problem in commutative algebra introduced in 2014 by Dan Anderson. The conjecture had remained unsolved for more than a decade.

Led by mathematician Dong Bin, the team designed a dual-agent AI framework that integrates natural language reasoning with formal proof verification. The system combines a reasoning module called Rethlas with a theorem-search engine known as Matlas, enabling it to explore solution strategies in a way that mirrors human mathematical thinking.

Researchers said the system synthesized decades of mathematical literature and produced a complete proof with minimal human oversight. The result was also automatically formalized, ensuring machine-level verification.

The AI completed the proof in about 80 hours of computation. Such work would typically require collaboration among specialists across multiple mathematical fields. The researchers noted that no human mathematical judgment was involved during the process, though expert input could improve efficiency.

The findings were опубликованы in a preprint on arXiv and have not yet undergone peer review. The team described the result as evidence of growing automation in mathematical research.

This development follows a broader trend. Systems such as GPT-5.4 have recently solved benchmark problems that resisted researchers for years. Meanwhile, the AI startup Axiom reported solving multiple previously intractable problems earlier this year.

The rapid pace of progress is drawing both interest and concern within the mathematical community, as AI tools take on increasingly complex research tasks.