Spatiotemporal Hidden-State Dynamics as a Signature of Internal Reasoning in Large Language Models

TL;DR

This paper introduces StALT, a novel, training-free metric that captures the spatiotemporal dynamics of hidden states in large reasoning models, effectively distinguishing correct reasoning trajectories from incorrect ones.

Contribution

It formalizes the Spatiotemporal Amplitude of Latent Transition (StALT) as a new way to analyze hidden-state dynamics, providing a practical, label-free correctness signal for reasoning models.

Findings

StALT reliably separates correct from incorrect reasoning trajectories across models and benchmarks.

Successful reasoning exhibits broad temporal dynamics with localized layer concentration, detectable by StALT.

Manipulations affecting reasoning demand systematically alter the spatiotemporal amplitude, validating its link to internal reasoning.

Abstract

Large reasoning models (LRMs) generate extended solutions, yet it remains unclear whether these traces reflect substantive internal computation or merely verbosity and overthinking. Although recent hidden-state analyses suggest that internal representations carry correctness-related signals, their coarse aggregations may obscure the token and layer structure underlying reasoning computation. We investigate hidden-state transitions across decoding steps and layers, and identify a distinct spatiotemporal pattern in LRMs: successful trajectories exhibit broad temporal dynamics with localized layer-wise concentration, while this structure is weaker in non-reasoning models and knowledge-heavy domains. We formalize this characteristic as Spatiotemporal Amplitude of Latent Transition (StALT), a training-free trajectory statistic that summarizes temporal changes between adjacent tokens weighted by within-token layer saliency. Across diverse models and benchmarks, StALT reliably separates correct from incorrect trajectories in reasoning-intensive regimes, providing a competitive label-free correctness signal alongside strong output-space and length-based baselines. Intervention analyses further show that this spatiotemporal amplitude responds systematically to manipulations that increase or reduce the demand for internal reasoning, supporting its association with latent reasoning dynamics in LRMs. These findings provide empirical evidence that LRMs exhibit measurable hidden-state dynamics and offer a practical probe for understanding internal computation beyond output-based evaluation.