A Minimal Agent for Automated Theorem Proving

Researchers propose a minimal baseline architecture for AI-based theorem proving that achieves competitive performance with state-of-the-art systems while using significantly simpler design. The open-source implementation demonstrates that iterative proof refinement approaches are more sample-efficient and cost-effective than single-shot generation methods.

• →A minimal agent baseline for automated theorem proving achieves competitive performance with much simpler architecture than existing systems.
• →The implementation features core capabilities including iterative proof refinement, library search, and context management.
• →Iterative approaches consistently outperform single-shot generation methods in sample efficiency and cost effectiveness.
• →The system has been released as open-source to serve as a reference for future research.
• →The baseline enables systematic comparison across different AI-based theorem prover architectures.