Representation Alignment as a Bottleneck in LLM-Based Retrosynthesis Planning
On the map: LLM reasoning → Reasoning for agents, planning and code
Abstract (arXiv)
While LLMs show promise in general reasoning, symbolic planning in chemistry remains a bottleneck. Direct ''SMILES-to-PDDL'' attempts fail because they force models to juggle chemical analysis and planning-language structuring simultaneously. We hypothesize that this failure stems from a lack of intermediate abstractions rather than insufficient model capacity. By decomposing retrosynthesis into molecule mapping, reaction mapping, and PDDL generation, we achieve high success rates where end-to-end approaches fail. This provides evidence that a primary bottleneck lies in representation alignment rather than raw model capacity. Our structural analysis demonstrates that intermediate representations are essential in retrosynthesis planning, highlighting the importance of representation-centric design in future systems.
Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Reasoning for agents, planning and code, one of 17 directions in LLM reasoning. Builds on 21 of its 35 references that are in the map. First posted on arXiv 28 Sep 2026.
Builds on
21 of 35 references.
Closest ancestors · 2
Same direction, newest first
- 2023On the Planning Abilities of Large Language Models : A Critical Investigation25 May
- 2022PlanBench: An Extensible Benchmark for Evaluating Large Language Models on Planning and Reasoning about Change21 Jun
Other references · 19
- 2023GPT-4 Technical Reportcited by 18,132
- 2022Chain-of-Thought Prompting Elicits Reasoning in Large Language Modelscited by 12,302
- 2022ReAct: Synergizing Reasoning and Acting in Language Modelscited by 8,355
- 2023Toolformer: Language Models Can Teach Themselves to Use Toolscited by 4,072
- 2023Sparks of Artificial General Intelligence: Early experiments with GPT-4cited by 2,579
- 2022Do As I Can, Not As I Say: Grounding Language in Robotic Affordancescited by 2,302
- 2022Least-to-Most Prompting Enables Complex Reasoning in Large Language Modelscited by 1,397
- 2022Code as Policies: Language Model Programs for Embodied Controlcited by 1,172
- 2022Inner Monologue: Embodied Reasoning through Planning with Language Modelscited by 1,057
- 2022Program of Thoughts Prompting: Disentangling Computation from Reasoning for Numerical Reasoning Taskscited by 1,057
- 2022PAL: Program-aided Language Modelscited by 635
- 2023ChemCrow: Augmenting large-language models with chemistry toolscited by 609
7 more
- 2023LLM+P: Empowering Large Language Models with Optimal Planning Proficiencycited by 496
- 2018Molecular Transformer - A Model for Uncertainty-Calibrated Chemical Reaction Predictioncited by 225
- 2018PDDLStream: Integrating Symbolic Planners and Blackbox Samplers via Optimistic Adaptive Planningcited by 147
- 2021Learning Symbolic Operators for Task and Motion Planningcited by 57
- 2023BioPlanner: Automatic Evaluation of LLMs on Protocol Planning in Biologycited by 42
- 2022On Grounded Planning for Embodied Tasks with Language Modelscited by 33
- 2021Process-Level Representation of Scientific Protocols with Interactive Annotationcited by 16
Built on by
Not cited yet.
Similar, not cited
Close in wording, no citation either way
- 2025AOT*: Efficient Synthesis Planning via LLM-Empowered AND-OR Tree Search
- 2026RetroAgent: Harnessing LLMs to Search Over Structured Memory for Agentic Retrosynthesis Planning
- 2025Chemical reasoning in LLMs unlocks strategy-aware synthesis planning and reaction mechanism elucidation
- 2025Synthelite: Chemist-aligned and feasibility-aware synthesis planning with LLMs
- 2025LARC: Towards Human-level Constrained Retrosynthesis Planning through an Agentic Framework
- 2026Reinforced Reasoning for End-to-End Retrosynthetic Planning
- 2026Order Matters in Retrosynthesis: Structure-aware Generation via Reaction-Center-Guided Discrete Flow Matching
- 2025RetroDFM-R: Reasoning-Driven Retrosynthesis Prediction with Large Language Models via Reinforcement Learning