VALENCE AS STABILITY, AROUSAL AS EXPLORATION: A DYNAMICAL ACCOUNT OF AFFECTIVE BRAIN STATES

A dynamical account of brain states and transitions can characterise metastable activity patterns, offering a principled link between neural dynamics and emotion. However, many EEG emotion-recognition pipelines rely on opaque features or end-to-end deep learning, limiting interpretability. To address these challenges, we propose an energy landscape analysis (ELA) framework for EEG that conceptualises affective brain activity as trajectories across a multi-dimensional energy surface. From the inferred landscape, we quantified attractor structure, stability, and switching dynamics. By correlating these dynamical descriptors with subject-independent affect ratings, we reveal a fundamental dissociation in how the brain encodes emotion dimensions. Valence was primarily associated with local dynamics within basins of attraction. Specifically, valence ratings correlated significantly with intra-basin transition statistics (ρ = 0.481, p_adj < 0.001), suggesting that valence reflects the stability of local attractors or how effectively a specific neural pattern captures and holds the system. In contrast, arousal reflected global exploratory dynamics between basins. High arousal was marked by increased basin conditional entropy (ρ = 0.405, p_adj =0.0015) and inter-basin transitions (ρ = 0.297, p_adj = 0.0218). This indicates that arousal does not merely reflect a specific state, but rather the system's kinetic energy to overcome energy barriers and transition between diverse functional network configurations. In conclusion, we present a compact state-space view of emotion, defining valence as a constraint on local state persistence and arousal as a driver of global state-space exploration.