Research map / Large language models / Agents, code and math
Mathesis: Towards Formal Theorem Proving from Natural Languages
Yu Xuejun, Jianyuan Zhong, Zijin Feng, Pengyi Zhai, Roozbeh Yousefzadeh, Wei Chong Ng, Haoxiong Liu, Ziyi Shou, Jing Xiong, Yudong Zhou, Claudia Beth Ong, Austen Jeremy Sugiarto and 8 more
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 Hierarchical Preference Optimization (DPO on proof success) after GRPO (ablation) | Mathesis-Autoformalizer · Gaokao-Formal | LC+LSC success rate@6: 67% -> 71% | Improved |
| add Hierarchical Preference Optimization after GRPO (ablation) | Mathesis-Autoformalizer · Gaokao-Formal | LC+LSC success rate@1: 45% -> 50% | Improved |
| add Hierarchical Preference Optimization after GRPO (ablation) | Mathesis-Autoformalizer · MiniF2F-test | LC+LSC success rate@1: 69% -> 79% | Improved |
| add Hierarchical Preference Optimization after GRPO (ablation) | Mathesis-Autoformalizer + Mathesis-Prover-7B · Gaokao-Formal (end-to-end) | proof success rate: 15.6% -> 18.0% | Improved |
| add Hierarchical Preference Optimization after GRPO (ablation) | Mathesis-Autoformalizer + Mathesis-Prover-7B · MiniF2F-test (end-to-end) | proof success rate: 63.5% -> 64.3% | Improved |
| replace Kimina-Autoformalizer with Mathesis-Autoformalizer (GRPO with composite reward) | Autoformalizer · Gaokao-Formal | LC+LSC success rate@6: 49% -> 71% | Improved |
| replace LLM-as-a-Judge with LeanScorer for semantic checking | Semantic checker · Gaokao-Formal semantic checker | F1: 0.85 -> 0.92 | Improved |
| replace Re-informalization with LeanScorer | Semantic checker · Gaokao-Formal semantic checker | F1: 0.65 -> 0.92 | Improved |
| upgrade the prover to Mathesis-Prover with Mathesis-HPO statements | End-to-end pipeline · MiniF2F-test | proof success rate: 52.8% -> 64.3% | Improved |
| upgrade the autoformalizer to Mathesis-HPO with Mathesis-Prover | End-to-end pipeline · MiniF2F-test | proof success rate: 34.8% -> 64.3% | Improved |
| upgrade the autoformalizer to Mathesis-HPO with Mathesis-Prover | End-to-end pipeline · Gaokao-Formal | proof success rate: 7.2% -> 18.4% | Improved |
| upgrade the prover to Mathesis-Prover with Mathesis-HPO statements | End-to-end pipeline · Gaokao-Formal | proof success rate: 11.6% -> 18.0% | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Herald-7B — Herald: A Natural Language Annotated Lean 4 Dataset (ICLR 2025) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Autoformalize Mathematical Statements by Symbolic Equivalence and Semantic Consistency (NeurIPS 2024 · similarity 0.91)
- ProofAug: Efficient Neural Theorem Proving via Fine-grained Proof Structure Analysis (ICML 2025 · similarity 0.91)
- APOLLO: Automated LLM and Lean Collaboration for Advanced Formal Reasoning (NeurIPS 2025 · similarity 0.90)
- FaithSieve: Fine-Grained Evaluation of Math Proofs with Faithful Formal Evidence (NeurIPS 2026 · similarity 0.90)
- ImProver: Agent-Based Automated Proof Optimization (ICLR 2025 · similarity 0.90)
- Don't Trust: Verify -- Grounding LLM Quantitative Reasoning with Autoformalization (ICLR 2024 · similarity 0.90)
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.
