Topology of Reasoning: Understanding Large Reasoning Models through Reasoning Graph Properties
Gouki Minegishi, Hiroki Furuta, Takeshi Kojima, Yusuke Iwasawa, Yutaka Matsuo
Abstract
Recent large-scale reasoning models have achieved state-of-the-art performance on challenging mathematical benchmarks, yet the internal mechanisms underlying their success remain poorly understood. In this work, we introduce the notion of a reasoning graph, extracted by clustering hidden-state representations at each reasoning step, and systematically analyze three key graph-theoretic properties: cyclicity, diameter, and small-world index, across multiple tasks (GSM8K, MATH500, AIME 2024). Our findings reveal that distilled reasoning models (e.g., DeepSeek-R1-Distill-Qwen-32B) exhibit significantly more recurrent cycles (about 5 per sample), substantially larger graph diameters, and pronounced small-world characteristics (about 6x) compared to their base counterparts. Notably, these structural advantages grow with task difficulty and model capacity, with cycle detection peaking at the 14B scale and exploration diameter maximized in the 32B variant, correlating positively with accuracy. Furthermore, we show that supervised fine-tuning on an improved dataset systematically expands reasoning graph diameters in tandem with performance gains, offering concrete guidelines for dataset design aimed at boosting reasoning capabilities. By bridging theoretical insights into reasoning graph structures with practical recommendations for data construction, our work advances both the interpretability and the efficacy of large reasoning models. Implementation available here: https://github.com/gouki510/Topology_of_Reasoning
Joy can read 8 pages of a book in 20 minutes. How many hours will it take her to read 120 pages?
In 1 hour, there are 3 sets of 20 minutes. 8 pages÷20 min=0.4 pages/minute. 120 pages÷0.4 pages/minute=360 minutes. 360 minutes÷60 minutes/hour=6 hours. The answer is 6 hours. In 1 hour, there are 3 sets of 20 minutes. 8 pages÷20 min=0.4 pages/minute. 120 pages÷0.4 pages/minute=360 minutes. 360 minutes÷60 minutes/hour=6 hours. Wait, let me verify the calculations. … 8 pages÷20 min=0.4 pages/minute. 120 pages÷0.4 pages/minute=300 minutes. 300 minutes÷60 minutes/hour=5 hours. The answer is 5 hours.
Ask about this paper
Your agent reads all of it.
Lune indexed this paper to the last equation, along with the top-tier papers that cite it. Ask a question and the answer quotes them.
Your agent calls
Luneget_paper_fulltext
Free to start. No credit card required.
Terminal
Install the CLIlune papers fulltext a2aa5ca5-d37e-47bb-ae95-1cd47b3294f8Cited by top-tier papers14
- The Geometry of Reasoning: Flowing Logics in Representation SpaceYufa Zhou, Yixiao Wang, Xunjian Yin, Shuyan Zhou et al.ICLR 2026 · 29 citations
- Attention Illuminates LLM Reasoning: The Uncovered Preplan-and-Anchor Rhythm Enables Fine-Grained Policy OptimizationYang Li, Zhichen Dong, Yuhan Sun, Weixun Wang et al.ICML 2026 · 25 citations
- Reasoning or Retrieval? A Study of Answer Attribution on Large Reasoning ModelsYuhui Wang, Changjiang Li, Guangke Chen, Jiacheng Liang et al.ICLR 2026 · 13 citations
- ReJump: A Tree-Jump Representation for Analyzing and Improving LLM ReasoningYuchen Zeng, Shuibai Zhang, Wonjun Kang, Shutong Wu et al.ICML 2026 · 5 citations
- ThoughtFold: Folding Reasoning Chains via Introspective Preference LearningZiyan Liu, Xueda Shen, Yuzhe Gu, Songyang Gao et al.ICML 2026 · 3 citations
Builds on18
- Direct Preference Optimization: Your Language Model is Secretly a Reward ModelRafael Rafailov, Archit Sharma, Eric Mitchell, Christopher D. Manning et al.NeurIPS 2023 · 10,924 citations
- Let's Verify Step by StepHunter Lightman, Vineet Kosaraju, Yuri Burda, Harrison Edwards et al.ICLR 2024 · 3,045 citations
- STaR: Bootstrapping Reasoning With ReasoningEric Zelikman, Yuhuai Wu, Jesse Mu, Noah D. GoodmanNeurIPS 2022 · 1,126 citations
- Sparse Autoencoders Find Highly Interpretable Features in Language ModelsRobert Huben, Hoagy Cunningham, Logan Riggs Smith, Aidan Ewart et al.ICLR 2024 · 1,072 citations
- Scaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth ApproachJonas Geiping, Sean McLeish, Neel Jain, John Kirchenbauer et al.NeurIPS 2025 · 431 citations
Related papers
- How Far Are We from Optimal Reasoning Efficiency?Jiaxuan Gao, Shu Yan, Qixin Tan, Lu Yang et al.NeurIPS 2025 · 12 citations
- Reasoning Structure of Large Language ModelsFrédéric Berdoz, Luca Lanzendörfer, Fabian Farestam, Roger WattenhoferICML 2026
- Structured Reasoning for LLMs: A Unified Framework for Efficiency and ExplainabilityYubo Dong, Hehe Fan, Linchao Zhu, Yi YangICLR 2026
- Mapping the Minds of LLMs: A Graph-Based Analysis of Reasoning LLMsZhen Xiong, Yujun Cai, Zhecheng Li, Yiwei WangEMNLP 2025
- Learning Graph Rationales to Compress Long Chains of Thought in Multimodal ReasoningYizhi Wang, Linan Yue, Deng-Bao Wang, Tong Wei et al.KDD 2026
