Agentic Theorem Proving: Neurosymbolic Architectures for Formal Verification

ABSTRACT

Automated theorem proving requires a blend of creative intuition and rigorous logical deduction which eludes pure deep learning models. We propose a framework that marries the creative search capabilities of large models with the deterministic verification of formal proof assistants. Our system deploys an intuition module to propose proof steps and a verifier module to interface with the formal environment. This approach achieves record pass rates on complex mathematical benchmarks.