Research map / Large language models / Agents, code and math
Lean Finder: Semantic Search for Mathlib That Understands User Intents
Jialin Lu, Kye Emond, Kaiyu Yang, Swarat Chaudhuri, Weiran Sun, Wuyang Chen
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 |
|---|---|---|---|
| use Lean Finder instead of Lean Search | retrieval · Informalized statement | R@1: 64.2 vs 49.2 | Improved |
| use Lean Finder instead of Lean Search | retrieval · Synthetic user query | R@1: 54.4 vs 47.1 | Improved |
| use Lean Finder instead of Lean Search | retrieval · Augmented statement | R@1: 82.7 vs 59.2 | Improved |
| use Lean Finder instead of GPT-4o with web search (stem match) | retrieval · Synthetic user query | R@1: 54.4 vs 17.8 | Improved |
| use Lean Finder instead of Real Prover Search | retrieval · Augmented proof state | R@1: 24.6 vs 8.0 | Improved |
| use Lean Finder instead of Lean State Search | retrieval · Raw proof state | R@10: 40.0 vs 32.1 | Improved |
| use Lean Finder instead of Lean Search | retrieval · User study (128 real queries) | Top-3 rate (%): 81.6% vs 56.9% | Improved |
| use Lean Finder instead of GPT-4o | retrieval · User study (128 real queries) | Normalized Borda: 0.67 vs 0.40 | Improved |
| augment proof state with natural language description of the transition | Lean Finder · Proof state queries | R@1: 8.3 -> 24.6 | Improved |
| integrate Lean Finder into LLM provers as RAG | Goedel Prover, DeepSeek-Prover V1.5, REALProver · MiniF2F, ProofNet, PutnamBench, FATE-M | proof success | No clear effect |
Closest papers by meaning
- Lean Workbook: A large-scale Lean problem set formalized from natural language math problems (NeurIPS 2024 · similarity 0.89)
- Semantic Search over 9 Million Mathematical Theorems (NeurIPS 2026 · similarity 0.89)
- Herald: A Natural Language Annotated Lean 4 Dataset (ICLR 2025 · similarity 0.87)
- LeanAgent: Lifelong Learning for Formal Theorem Proving (ICLR 2025 · similarity 0.87)
- MA-LoT: Model-Collaboration Lean-based Long Chain-of-Thought Reasoning enhances Formal Theorem Proving (ICML 2025 · similarity 0.87)
- FaithSieve: Fine-Grained Evaluation of Math Proofs with Faithful Formal Evidence (NeurIPS 2026 · similarity 0.87)
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.
