arXiv 2610.20459 · cs.CL · 2026-10-14
MolLean: A Process-Verifiable Language for Chemical Reasoning
Jiatong Li, Yatao Bian
Affiliation unverified
One-line takeaway
MolLean enables verifiable chemical reasoning with traceable, executable molecular state transitions and refinement.
MolLean通过可验证的分子状态流程实现化学推理,支持全过程追踪与修正,在多个任务中显著优于传统链式思维,尤其在反应任务中表现更优。
Key points
MolLean introduces a process-verifiable language for chemical reasoning, grounding assertions in explicit molecular states. Its Lean-based kernel validates state transitions and property claims, while a typed witness protocol separates structural checks from external measurements. MolLean-MCP enables interactive, feedback-driven reasoning by exposing state evolution during generation. Experiments on 8,800 examples across 12 tasks show that whole-trace refinement outperforms direct chain-of-thought, with MolLean-MCP enhancing performance in reaction tasks.
- MolLean grounds chemical reasoning in explicit molecular states for verifiability.
- A layered Lean kernel validates state transitions and molecular assertions step-by-step.
- Whole-trace refinement outperforms direct chain-ofthought on all 12 tasks.
- MolLean-MCP enables interactive feedback, improving reaction task performance.
Written only from the official arXiv abstract by an open-source model, then checked automatically (every number and name must appear in the abstract). The paper is authoritative.
Read on arXiv PDF Hugging Face discussion Code
Topics & disciplines
More on Reasoning
- LEV: Thinking Fast and Deep with Looped Decision Models ▲337
- Thinking Inertia: LLMs Keep Thinking When Told Not To ▲241
- Rethinking Latent Visual Reasoning: Grounding Latent Reasoning in Visual Evidence ▲224
- Questioning the Questions: Sustaining Self-Evolution in Reasoning Models ▲85
- Improving Test-Time Scaling with Adaptive Looped Transformers ▲62
More on AI for Science
- Scaling Parameter-Efficient Fine-Tuning for Brain Encoding 2026-10-19 · ▲247
- Into the Unknown: Benchmarking AI agents on open scientific research problems 2026-10-13 · ▲144
- EvoDuet: Bilevel Co-Evolution of Web Searching and Task Solving for Scientific Discovery 2026-09-30 · ▲112
- Predictive Credit: Measuring What Scientific Explanations Add to Experimental Forecasts 2026-09-29 · ▲103
- X-Ragent: Tool-Augmented Reinforcement Learning for Evidence-Grounded Chest X-Ray Reasoning 2026-10-12 · ▲79
About this data
Metadata from the arXiv API (verified); upvotes from Hugging Face Daily Papers; citations from Semantic Scholar. Last refreshed 2026-10-14.
The takeaway is written by AIMap from the abstract; the abstract itself is not reproduced — read the original.