Peking University Team Achieves Breakthrough in AI-Assisted Mathematical Research, Solves Open Conjecture
Key Takeaways
- ▸First fully automated AI framework to independently solve and formally verify an open mathematical conjecture without human intervention
- ▸Dual-agent architecture (Rethlas + Archon) successfully combines natural-language reasoning with formal proof verification in Lean 4
- ▸Cross-domain knowledge retrieval capability enables AI to identify relevant techniques from unrelated mathematical fields, mimicking expert interdisciplinary collaboration
Summary
Researchers at Peking University's Beijing International Center for Mathematical Research have achieved a significant milestone in AI-assisted mathematics: an autonomous AI framework that solved an open problem in commutative algebra and formally verified the proof in approximately 19,000 lines of Lean 4 code. The breakthrough represents the first end-to-end automated pipeline where AI agents independently discovered and rigorously verified a solution to a previously unsolved research problem—specifically, a disproof of the Anderson Conjecture from commutative ring theory.
The achievement was led by Bin Dong and a team of four senior mathematicians including Ruochuan Liu, Liang Xiao, and Zaiwen Wen. The system employs two collaborating AI agents: Rethlas, which performs natural-language mathematical reasoning including literature search and proof construction, and Archon, which translates proofs into formally verified Lean 4 code. The infrastructure is supported by dual search engines—LeanSearch for semantic search over formalized theorems and Matlas covering tens of millions of mathematical statements in natural language.
A notable capability demonstrated in the Anderson Conjecture solution was the system's ability to cross-domain retrieval, locating a 2006 paper by Jensen on integral domain completions from an entirely different mathematical subfield that provided the key tool for constructing the counterexample—a kind of connection that traditionally requires expert collaboration across specialties.
- Supporting infrastructure (LeanSearch, Matalis) demonstrates practical scalability, with LeanSearch already adopted by the Lean community at 8,000+ daily API calls
Editorial Opinion
This represents a watershed moment for AI in mathematical research. Moving beyond pattern-matching and into rigorous, formally-verified mathematical discovery suggests that AI agents may soon become genuine collaborators in frontier mathematics rather than merely assistive tools. The ability to autonomously bridge disparate fields and discover novel connections hints at AI's potential to accelerate human mathematical progress—though the reliance on formal verification frameworks like Lean 4 means the impact will first be felt in areas where such formalization is feasible.
