Research map / Large language models / Agents, code and math
Let's Explore Step by Step: Generating Provable Formal Statements with Deductive Exploration
Qi Liu, Kangjie Bao, Yue Yang, Xinhao Zheng, Renqiu Xia, 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 |
|---|---|---|---|
| add DExploration step-by-step exploration (vs one-shot Conjecturer-Prover) (ablation) | Goedel-Prover-V2-8B agent · Lean 4 statement synthesis | success rate (%): 23.28 -> 54.52 | Improved |
| add DExploration step-by-step exploration (ablation) | Goedel-Prover-V2-8B agent · Lean 4 statement synthesis | proven rate (%): 31.78 -> 64.18 | Improved |
| add DExploration step-by-step exploration (ablation) | Goedel-Prover-V2-8B agent · Lean 4 statement synthesis | contradictory rate (%): 36.34 -> 9.94 | Improved |
| add DExploration step-by-step exploration (ablation) | Goedel-Prover-V2-8B agent · Lean 4 statement synthesis | token cost per valid statement: 187K -> 8.8K | Improved |
| add DExploration step-by-step exploration (ablation) | Goedel-Prover-V2-8B agent · Lean 4 statement synthesis | valid statements: 1164 -> 2726 | Improved |
| add interleaved introducing and deducing from exploratory transformation (vs staged) (ablation) | DExplorer · Lean 4 statement synthesis | proven rate (%): 52.10 -> 64.18 | Improved |
| add interleaved introducing and deducing (vs staged) (ablation) | DExplorer · Lean 4 statement synthesis | valid statements: 2340 -> 2726 | Improved |
| add interleaved introducing and deducing (vs staged) (ablation) | DExplorer · Lean 4 statement synthesis | top-500 complexity: 1193 -> 1374 | Improved |
| add interleaved introducing and deducing (vs staged) (ablation) | DExplorer · Lean 4 statement synthesis | overall complexity: 470 -> 515 | Improved |
| add interleaved introducing and deducing (vs staged) (ablation) | DExplorer · Lean 4 statement synthesis | top-500 item difficulty (lower=harder): 0.09 -> 0.05 | Improved |
| add interleaved introducing and deducing (vs staged) (ablation) | DExplorer · Lean 4 statement synthesis | token cost per valid statement: 8800 -> 8841 | No clear effect |
| add DExploration (vs Conjecturer-Prover, same data and model) (ablation) | Goedel-Prover-V2-8B agent · Lean 4 statement synthesis | Rouge-L (lower=more diverse): 0.174 -> 0.173 | No clear effect |
Closest papers by meaning
- Subgoal-based Demonstration Learning for Formal Theorem Proving (ICML 2024 · similarity 0.85)
- Learning Formal Mathematics From Intrinsic Motivation (NeurIPS 2024 · similarity 0.85)
- APOLLO: Automated LLM and Lean Collaboration for Advanced Formal Reasoning (NeurIPS 2025 · similarity 0.85)
- MARGE: Improving Math Reasoning with Guided Exploration (ICML 2025 · similarity 0.85)
- RLAD: Training LLMs to Discover Abstractions for Solving Reasoning Problems (ICLR 2026 · similarity 0.85)
- 3D-Prover: Diversity Driven Theorem Proving With Determinantal Point Processes (NeurIPS 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.
