Research map / Large language models / Agents, code and math
LeanForPhysics: Comprehensive Reasoning Framework for University-level Physics in Lean4
Yuxin Li, Minghao LIU, Ruida WANG, JI WenZhao, Zhitao He, Rui Pan, Junming Huang, Tong Zhang, Yi R. Fung
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 PhysLib to context | Gemini-2.5-pro · LeanPhysBench Overall | pass@16: 6.67% -> 40.00% | Improved |
| add PhysLib to context | Claude-Sonnet-4 · LeanPhysBench Overall | pass@16: 3.50% -> 34.50% | Improved |
| add PhysLib to context | GPT-4o · LeanPhysBench Overall | pass@16: 2.50% -> 14.17% | Improved |
| add PhysLib to context | Goedel-Prover-V2-8B · LeanPhysBench Overall | pass@16: 10.00% -> 14.67% | Improved |
| add PhysLib to context | DeepSeek-Prover-V2-7B · LeanPhysBench Overall | pass@16: 11.17% -> 14.83% | Improved |
| add PhysLib to context | Kimina-Prover-8B · LeanPhysBench Overall | pass@16: 9.33% -> 13.50% | Improved |
| add PhysLib to context | Qwen3-8B · LeanPhysBench Overall | pass@16: 2.35% -> 6.52% | Improved |
| add PhysLib to context | DeepSeek-R1-8B · LeanPhysBench Overall | pass@16: 3.33% -> 5.83% | Improved |
| add PhysLib to context (average over models) | 8 LLMs · LeanPhysBench Overall | pass@16: 11.90% | Improved |
| add PhysLib to context | Gemini-2.5-pro · LeanPhysBench Comp-Hard | pass@16: 0.00% -> 0.00% | No clear effect |
| use expert Lean prover instead of general LLM (Goedel-Prover-V2-8B vs Gemini-2.5-pro) with PhysLib | expert provers · LeanPhysBench Overall | pass@16: 40.00% -> 14.67% | Hurt |
| use expert Lean prover instead of Claude on Comp-Hard | expert provers · LeanPhysBench Comp-Hard | pass@16: 0.00% -> 5.88% | Improved |
Closest papers by meaning
- UGPhysics: A Comprehensive Benchmark for Undergraduate Physics Reasoning with Large Language Models (ICML 2025 · similarity 0.88)
- MA-LoT: Model-Collaboration Lean-based Long Chain-of-Thought Reasoning enhances Formal Theorem Proving (ICML 2025 · similarity 0.87)
- Scaling Physical Reasoning with the PHYSICS Dataset (NeurIPS 2025 · similarity 0.87)
- LeanAgent: Lifelong Learning for Formal Theorem Proving (ICLR 2025 · similarity 0.85)
- Lean Workbook: A large-scale Lean problem set formalized from natural language math problems (NeurIPS 2024 · similarity 0.85)
- Lean-STaR: Learning to Interleave Thinking and Proving (ICLR 2025 · similarity 0.85)
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.
