LeanMarathon: Toward Reliable AI Co-Mathematicians through Long-Horizon Lean Autoformalization
Researchers introduced LeanMarathon, a multi-agent AI system designed to help mathematicians formalize and prove complex theorems in the Lean proof assistant. It uses four contract-scoped agents to construct, audit, prove, and repair an evolving blueprint that serves as a formal proof skeleton, natural-language proof graph, and shared system of record, addressing issues like statement drift, tangled dependencies, and context decay.
Researchers from ArXiv cs.AI released LeanMarathon, a multi-agent AI system designed to assist mathematicians in formalizing and proving complex theorems in the Lean proof assistant. The system uses four contract-scoped agents that work together to construct, audit, prove, and repair an evolving blueprint. This blueprint serves simultaneously as a formal proof skeleton, natural-language proof graph, and shared system of record. LeanMarathon addresses common issues in long-horizon autoformalization, such as statement drift, tangled dependencies, context decay, and local repairs that corrupt distant work.
This development could significantly speed up mathematical research by reducing the time and effort required to formalize and verify complex proofs. For mathematicians, this means they can focus more on creative problem-solving and less on tedious proof-checking. It also makes advanced mathematical research more accessible to a broader audience, as the AI can help bridge the gap between natural language and formal mathematical language.
If you're a mathematician or just curious about how AI can assist in mathematical research, you can explore the LeanMarathon system by checking out the paper on ArXiv at https://arxiv.org/abs/2606.05400. The paper provides detailed information on how the system works and its potential applications in mathematical research.