A new integrative theory published in Progress in Neurobiology proposes that REM sleep functions as a biological simulation engine, allowing the brain to test and refine synaptic updates before committing them to long-term storage.
Sleep scientists have long recognized that rapid eye movement (REM) sleep supports emotional regulation, memory consolidation, and creativity, but a unified explanation for why the brain enters this distinctive state has remained elusive. Researchers at G. d’Annunzio University in Chieti-Pescara and the University of Bologna now offer a framework that reconciles electrophysiological, metabolic, and phenomenological dimensions of sleep into a single model: REM sleep as a “dummy model” of the external world.
The Framework. Stefano T. Censi, Alberto Granzotto, and Stefano L. Sensi divide the sleep cycle into two complementary phases. Non-REM (NREM) sleep serves as the critical period for synaptic reorganization and glymphatic clearance, during which neural circuits undergo structural updates based on prior waking experience. REM sleep then acts as an internally generated testing ground, simulating reality so the brain can evaluate whether these updates are functional.
“We conceptualize REM sleep as a state in which the brain constructs a provisional dummy model of the external world,” the authors write, drawing an analogy to the simulated environments engineers use to test software before deployment.
The framework distinguishes between phasic and tonic REM states and highlights the involvement of subcortical networks, including the Papez circuit and claustrum, in orchestrating state transitions. It integrates cerebrovascular, metabolic, and glymphatic dynamics within a multiscale systems approach.
Key Predictions. The model generates three testable predictions. First, prediction errors produced during REM simulation drive selective synaptic consolidation in the subsequent NREM episode through hippocampal sharp-wave ripple-mediated feedback. Second, awakening occurs when the global prediction error across the dummy-model network falls below a biological threshold. Third, total sleep duration should be proportional to the complexity of novel experience acquired during prior wakefulness.
Why It Matters. Current theories of REM sleep remain fragmented, with separate models for electrophysiological patterns, metabolic clearance, and the phenomenology of dreaming. This framework bridges these domains, offering a coherent explanation for why REM and NREM sleep alternate in ~90-minute cycles. If the model’s predictions hold, it could reshape how researchers approach disorders involving disrupted REM sleep and how they interpret the role of dreaming in cognitive health.
Limits. As a theoretical framework, the model synthesizes existing evidence but has not been empirically tested as a whole. The authors acknowledge that the specific neural mechanisms linking REM simulation to subsequent NREM consolidation require direct experimental validation.
Bottom Line. The dummy-model framework offers a unifying theory that positions REM sleep as an active testing ground for brain updates, rather than a passive recovery state. Its predictions about prediction-error-driven consolidation and sleep duration scaling with experience are experimentally accessible and could guide future sleep research.
Source: Censi ST, Granzotto A, Sensi SL. “REM Sleep as a Dummy-Model of the World: A Theoretical Framework.” Progress in Neurobiology. Published online June 19, 2026. DOI: 10.1016/j.pneurobio.2026.102939