Lean Refactor: Multi-Objective Controllable Proof Optimization via Agentic Strategy Search
Researchers have developed Lean Refactor, a new framework designed to optimize proofs generated by large language models (LLMs) in the Lean mathematical proof assistant. This system addresses key challenges such as proof length, compilation cost, and version compatibility, which are often in tension. By using a retrieval-augmented agentic approach with a curated database of refactoring strategies, Lean Refactor achieves significant compression rates and reduces compilation times, outperforming previous methods and demonstrating improved version transfer capabilities. AI
IMPACT Introduces a novel method for improving the efficiency and robustness of LLM-generated mathematical proofs, potentially accelerating formal verification efforts.