Demystifying Reasoning Dynamics with Mutual Information: Thinking Tokens are Information Peaks in LLM Reasoning

2025-06-03

AI Summary

This work investigates the causes and implications of mutual information (MI) surges—abrupt increases in MI between intermediate representations and the correct answer—during inference in large reasoning models (LRMs). Method: We identify that such MI peaks concentrate sharply at “thinking” tokens (e.g., “Hmm”, “Therefore”) and propose a lightweight thinking-token identification and regulation framework. It integrates MI quantification, intermediate-layer representation tracking, and token-level information dynamics modeling to guide decoding paths. Contribution/Results: We establish, for the first time, a strong statistical association between thinking tokens and MI peaks, demonstrating their decisive role in reasoning success. Our method significantly improves accuracy across diverse LRMs on mathematical and logical reasoning benchmarks. Empirical analysis confirms that thinking tokens serve as critical information carriers and that error probability decreases systematically with rising MI—validating MI as a reliable indicator of reasoning fidelity.

Technology Category

• Knowledge Representation and Reasoning: Reasoning with Beliefs
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning
• Reasoning under Uncertainty: Other Foundations of Reasoning under Uncertainty

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Large reasoning models (LRMs) have demonstrated impressive capabilities in complex problem-solving, yet their internal reasoning mechanisms remain poorly understood. In this paper, we investigate the reasoning trajectories of LRMs from an information-theoretic perspective. By tracking how mutual information (MI) between intermediate representations and the correct answer evolves during LRM reasoning, we observe an interesting MI peaks phenomenon: the MI at specific generative steps exhibits a sudden and significant increase during LRM's reasoning process. We theoretically analyze such phenomenon and show that as MI increases, the probability of model's prediction error decreases. Furthermore, these MI peaks often correspond to tokens expressing reflection or transition, such as ``Hmm'', ``Wait'' and ``Therefore,'' which we term as the thinking tokens. We then demonstrate that these thinking tokens are crucial for LRM's reasoning performance, while other tokens has minimal impacts. Building on these analyses, we propose two simple yet effective methods to improve LRM's reasoning performance, by delicately leveraging these thinking tokens. Overall, our work provides novel insights into the reasoning mechanisms of LRMs and offers practical ways to improve their reasoning capabilities. The code is available at https://github.com/ChnQ/MI-Peaks.

Problem

• Understanding internal reasoning mechanisms in large reasoning models
• Identifying mutual information peaks linked to critical thinking tokens
• Improving model reasoning performance using thinking tokens

Innovation

• Track mutual information peaks in reasoning
• Identify thinking tokens like Hmm, Wait
• Enhance reasoning via thinking tokens