Research map / Large language models / Agents, code and math
Beyond Theorem Proving: Formulation, Framework and Benchmark for Formal Problem-Solving
Qi Liu, Xinhao Zheng, Renqiu Xia, Xingzhi Qi, Qinxiang Cao, Junchi Yan
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 |
|---|---|---|---|
| prove ground-truth answer statement instead of solving for the answer (Proven vs Solved) | InternLM2.5-StepProver · FormalMath500 | success rate: 47.55% -> 23.77% | Hurt |
| solve instead of prove | DeepSeekProver-V1.5 · MiniF2F-Solving | success rate: 53.60% -> 22.40% | Hurt |
| solve instead of prove | TheoremLlama · FormalMath500 | success rate: 4.39% -> 16.02% | Improved |
| use D-FPS (deductive framework) instead of FPS | DeepSeek-V3 / provers · FormalMath500 | Solved: 23.77% -> 15.50% | Hurt |
| use D-FPS instead of FPS | provers · FormalMath500 | NE-Submitted: 14.47% -> 0.00% | Improved |
| use D-FPS instead of FPS | provers · MiniF2F-Solving | NE-Submitted: 8.80% -> 0.00% | Improved |
| add informal reasoning to formal steps (Hybrid CoT instead of ICL) (ablation) | DeepSeek-V3 D-FPS · FormalMath500 | Solved: 13.70% -> 15.50% | Improved |
| add informal reasoning to formal steps (Hybrid CoT instead of ICL) (ablation) | DeepSeek-V3 D-FPS · MiniF2F-Solving | Solved: 21.87% -> 21.60% | No clear effect |
| add Hybrid CoT on PutnamBench-Solving (ablation) | DeepSeek-V3 D-FPS · PutnamBench-Solving | Solved: 0.00% -> 0.00% | No clear effect |
| use whole-proof generation instead of proof search | Provers · FormalMath500 | Solved: 23.77% -> 22.22% | No clear effect |
Closest papers by meaning
- Lean-STaR: Learning to Interleave Thinking and Proving (ICLR 2025 · similarity 0.89)
- miniF2F-Lean Revisited: Reviewing Limitations and Charting a Path Forward (NeurIPS 2025 · similarity 0.89)
- APOLLO: Automated LLM and Lean Collaboration for Advanced Formal Reasoning (NeurIPS 2025 · similarity 0.89)
- MUSTARD: Mastering Uniform Synthesis of Theorem and Proof Data (ICLR 2024 · similarity 0.88)
- FaithSieve: Fine-Grained Evaluation of Math Proofs with Faithful Formal Evidence (NeurIPS 2026 · similarity 0.88)
- FVEL: Interactive Formal Verification Environment with Large Language Models via Theorem Proving (NeurIPS 2024 · similarity 0.88)
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.
