Research map / Large language models / Agents, code and math
Editable Proof Sketch for Automated Theorem Proving
Zikai Xiao, Hanzheng Wang, Meng-Hao Guo, Shi-min Hu, Shing-Tung Yau
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add refutation-guided handling of unprovable goals (ablation) | SketchRefine (DeepSeek-V3.2 + DeepSeek-Prover-V2-7B) · FormalMath-Lite | pass rate: 69.2% -> 76.0% | Improved |
| add refutation-guided handling of unprovable goals (ablation) | SketchRefine (DeepSeek-V3.2) · MiniF2F-test | pass rate: 96.7% -> 97.5% | Improved |
| add refutation-guided handling of unprovable goals (ablation) | SketchRefine (Gemini) · MiniF2F-test | pass rate: 98.8% -> 99.6% | Improved |
| decompose subgoals in place (edit original sketch) instead of building a new sketch per subgoal (ablation) | SketchRefine · MiniF2F-test | pass rate vs token budget | Improved |
| use SketchRefine instead of Hilbert under matched token budget | SketchRefine · MiniF2F-test | pass rate vs token budget | Improved |
| enable Construction nodes in addition to Deduction nodes (ablation) | EditableSketch · MiniF2F-test | pass rate | Improved |
| enable Assumption/Conclusion node pairs in addition to Deduction nodes (ablation) | EditableSketch · MiniF2F-test | pass rate | Improved |
| add more irrelevant premises to subtheorems (ablation) | DeepSeek-Prover-V2-7B · MiniF2F-test subtheorems | prover pass rate | Hurt |
| use SketchRefine (DeepSeek-V3.2) instead of DeepSeek-Prover-V2-671B | SketchRefine · FormalMath-Lite | pass rate: 61.9% -> 76.0% | Improved |
Closest papers by meaning
- ImProver: Agent-Based Automated Proof Optimization (ICLR 2025 · similarity 0.87)
- FaithSieve: Fine-Grained Evaluation of Math Proofs with Faithful Formal Evidence (NeurIPS 2026 · similarity 0.86)
- Proving Theorems Recursively (NeurIPS 2024 · similarity 0.86)
- Subgoal-based Demonstration Learning for Formal Theorem Proving (ICML 2024 · similarity 0.86)
- APOLLO: Automated LLM and Lean Collaboration for Advanced Formal Reasoning (NeurIPS 2025 · similarity 0.86)
- Alchemy: Amplifying Theorem-Proving Capability Through Symbolic Mutation (ICLR 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
