prompt logo

prompt

Free

Blueprint-driven Lean 4 theorem proving via AI

FreeFree tier
Inputs: text
Type
Open Source

About prompt

A structured prompt for guiding large language models to act as a Formal Theorem Proving Architect, specializing in solving mathematical theorems using the Lean 4 proof assistant. The prompt implements a blueprint-first methodology: it generates a dependency graph of definitions and lemmas, then proves each lemma in parallel with iterative refinement based on compiler feedback. Based on research achieving 99.2% pass@1 on MiniF2F-test and 75.6% on PutnamBench, this prompt enables automated, agentic theorem proving from natural language theorem statements and optional proof sketches.

Key Features

Blueprint-first decomposition generating a dependency graph of definitions and lemmas
Parallel theorem proving of each lemma in topological order with scope isolation
Iterative refinement using Lean 4 compiler feedback to repair broken proofs
Support for natural-language proof sketches as structural guides
High pass rates on benchmarks: 99.2% MiniF2F-test, 75.6% PutnamBench

Pros & Cons

Pros
  • Achieves state-of-the-art results on formal theorem proving benchmarks
  • Structured blueprint methodology reduces dead-end exploration
  • Parallel proving speeds up overall proof construction
  • Explicitly handles compiler errors with refinement loops
Cons
  • Requires access to a capable large language model (not a standalone tool)
  • Needs Lean 4 environment and compiler setup for full workflow
  • Complex prompt with many phases may confuse less capable models
  • Dependency graph must remain acyclic, limiting some proof strategies

Best For

Automated formal theorem proving in mathematicsGenerating Lean 4 code for mathematical proofsAssisting researchers in formalization of complex theoremsEducational tool for learning formal proof techniques