The tuning-for-criticality hypothesis (Pearlmutter & Houghton, 2009) proposes that the essential function of sleep is to re-establish a margin of safety between the brain's operating point and the onset of runaway neuronal oscillations, which the recurrent circuitry built up during waking otherwise risks. Later work has confirmed the hypothesis's central empirical prediction at the network level — wakefulness drives cortical dynamics away from criticality and sleep restores them — but the mechanism of this restoration remains unspecified: it is not known what REM sleep and slow-wave sleep each causally contribute, how the brain monitors its proximity to the critical boundary, or how protective synaptic changes are implemented. The problem asks for a stage-resolved mechanistic account consistent with the standing observation that chronic pharmacological REM suppression (e.g., by monoamine oxidase inhibitors) produces no gross cognitive impairment and that cetaceans lack REM sleep almost entirely.

Background

During waking, learning strengthens recurrent excitatory connections in cortical networks. Such loops are what allow rapid responses and the prolonged retention of short-term memories, but the same recurrence means that small parameter changes can push a network into self-amplifying, runaway oscillations. Pearlmutter and Houghton (2009) therefore proposed that the critical function of sleep is 'tuning for criticality': while wake permits stimulus-dependent network changes only when a margin of safety exists around the current parameters, sleep re-establishes that margin by exposing the network to a variety of inputs, monitoring for erratic behavior, and adjusting parameters; when sleep is impossible, an emergency mechanism — tiredness — prevents runaway behavior at the expense of processing efficiency. Here 'criticality' refers to the regime near a dynamical phase transition at which balanced excitation and inhibition produce neuronal avalanches (cascades of activity whose sizes follow a power law) with a branching ratio near one, i.e., each active neuron on average activates about one other, a regime associated with maximal dynamic range and information transmission.

Several strands of later evidence support the framework's core prediction. In humans, sustained wakefulness degrades the signatures of critical dynamics — avalanche size distributions and long-range temporal correlations in the EEG — and sleep restores them (Meisel et al., 2013). Intracranial recordings show that avalanche statistics differ systematically across vigilance states — large avalanches in slow-wave sleep (SWS, the deepest stage of non-REM sleep dominated by slow, synchronized cortical oscillations), intermediate ones in wake, small ones in REM sleep (rapid-eye-movement sleep, the stage characterized by wake-like desynchronized activity, muscle atonia, and vivid dreaming) — and modeling suggests these differences can be produced by tiny global changes in synaptic strength (Priesemann et al., 2013). In rats, continuous recordings over 10–14 days demonstrated that normal waking experience progressively disrupts criticality in visual cortex, that sleep restores it, and that the deviation from criticality predicts future sleep/wake behavior better than slow-wave activity, the standard homeostatic marker (Xu et al., 2024). Avalanche statistics in humans are also modulated across the NREM–REM cycles of sleep architecture (Scarpetta et al., 2023).

Yet the implementation of the hypothesis is open. The source paper itself left unresolved how the division of labor between REM and SWS should be understood — REM might serve to search for near-critical loops via dream-like stimulation, SWS might implement protective plastic changes, or both might collaborate in unspecified ways — and the mechanism by which the brain monitors proximity to the critical boundary and implements protective synaptic changes is unknown. Two phenomena sharpen the question. First, monoamine oxidase (MAO) inhibitors, an antidepressant class, virtually abolish REM sleep in humans for months to years without apparent memory impairment, which challenges any theory assigning REM an obligatory function; pharmacological REM-suppression studies more generally argue against a vital psychological role for REM, and its function remains contested (Siegel, 2023). Second, cetaceans (dolphins and whales), which sleep unihemispherically (one cerebral hemisphere at a time), show almost no REM sleep despite large, highly encephalized brains, which any account assigning REM a protective tuning role must reconcile. Candidate mechanisms proposed so far remain in silico or in vitro: homeostatic adaptation rules that drive networks along a critical manifold (Sormunen et al., 2023), and noise-driven spontaneous activity combined with spike-timing-dependent plasticity (a learning rule that modifies synapses based on the relative timing of pre- and postsynaptic spikes), which restores criticality and excitation–inhibition balance in simulations and dissociated cultures (Ikeda et al., 2025).

Problem Statement

Within the tuning-for-criticality account of sleep — under which sleep re-establishes the brain's margin of safety against runaway recurrent dynamics, with waking experience pushing networks toward supercritical behavior and sleep pulling them back — determine the stage-resolved implementation of this homeostatic process. Specifically: (1) what do REM sleep and slow-wave (non-REM) sleep each causally contribute to detecting and correcting excursion toward runaway oscillations — in particular, is the source's proposal correct that REM acts as a search process that probes the network with dream-like activation to find near-critical loops while SWS implements protective synaptic changes, or do the stages collaborate in some other division of labor; (2) by what mechanism does the brain monitor its proximity to the critical boundary and implement the protective synaptic adjustments during sleep; and (3) how is the account reconciled with the observations that chronic pharmacological REM suppression by monoamine oxidase inhibitors produces no gross cognitive impairment and that cetaceans, which sleep unihemispherically, show almost no REM sleep? These sub-questions serve one objective: a mechanistic, experimentally supported specification of how sleep's tuning-for-criticality function is implemented and staged, in an animal model or humans, such that the account is consistent with the known phenomenology of REM suppression across species.

References

  1. Pearlmutter BA, Houghton CJ. A New Hypothesis for Sleep: Tuning for Criticality. Neural Computation 21(6):1622–1641, 2009. DOI 10.1162/neco.2009.05-08-787. https://doi.org/10.1162/neco.2009.05-08-787
  2. Pearlmutter BA, Houghton CJ. Dreams, mnemonics, and tuning for criticality. Behavioral and Brain Sciences 36(6):625–626, 2013. DOI 10.1017/S0140525X13001404. https://doi.org/10.1017/S0140525X13001404
  3. Meisel C, Olbrich E, Shriki O, Achermann P. Fading Signatures of Critical Brain Dynamics during Sustained Wakefulness in Humans. The Journal of Neuroscience 33(44):17363–17372, 2013. DOI 10.1523/JNEUROSCI.1516-13.2013. https://doi.org/10.1523/JNEUROSCI.1516-13.2013
  4. Priesemann V, Valderrama M, Wibral M, Le Van Quyen M. Neuronal Avalanches Differ from Wakefulness to Deep Sleep – Evidence from Intracranial Depth Recordings in Humans. PLoS Computational Biology 9(3):e1002985, 2013. DOI 10.1371/journal.pcbi.1002985. https://doi.org/10.1371/journal.pcbi.1002985
  5. Xu Y, Schneider A, Wessel R, Hengen KB. Sleep restores an optimal computational regime in cortical networks. Nature Neuroscience 27(2):328–338, 2024. DOI 10.1038/s41593-023-01536-9. https://doi.org/10.1038/s41593-023-01536-9
  6. Scarpetta S, Morisi N, Mutti C, Azzi N, Trippi I, Ciliento R, Apicella I, Messuti G, Angiolelli M, Lombardi F, Parrino L, Vaudano AE. Criticality of neuronal avalanches in human sleep and their relationship with sleep macro- and micro-architecture. iScience 26(10):107840, 2023. DOI 10.1016/j.isci.2023.107840. https://doi.org/10.1016/j.isci.2023.107840
  7. Siegel JM. REM sleep function: Mythology vs. reality. Revue Neurologique 179(7):643–648, 2023. DOI 10.1016/j.neurol.2023.08.002. https://doi.org/10.1016/j.neurol.2023.08.002
  8. Steyn-Ross DA, Steyn-Ross ML. Brain and Nonlinear Dynamics: Slow-Wave Sleep Regulates to the Edge of Chaos. In: The Physics of the Mind and Brain Disorders, Springer Series in Cognitive and Neural Systems, pp. 213–232, 2017. DOI 10.1007/978-3-319-29674-6_10. https://doi.org/10.1007/978-3-319-29674-6_10
  9. Sormunen S, Gross T, Saramäki J. Critical Drift in a Neuro-Inspired Adaptive Network. Physical Review Letters 130(18):188401, 2023. DOI 10.1103/PhysRevLett.130.188401. https://doi.org/10.1103/PhysRevLett.130.188401
  10. Ikeda N, Akita D, Takahashi H. Emergent functions of noise-driven spontaneous activity: homeostatic maintenance of criticality and memory consolidation. Frontiers in Neural Circuits 19:1585087, 2025. DOI 10.3389/fncir.2025.1585087. https://doi.org/10.3389/fncir.2025.1585087
  11. Algom I, Shriki O. The ConCrit framework: Critical brain dynamics as a unifying mechanistic framework for theories of consciousness. Neuroscience & Biobehavioral Reviews 180:106483, 2026. DOI 10.1016/j.neubiorev.2025.106483. https://doi.org/10.1016/j.neubiorev.2025.106483
  12. Wang L, Ying S, Fan L, Qin Y, Liu T, Yao D. Disrupted near-critical dynamics and fragmented sleep in insomnia: evidence from neural avalanche analysis of EEG. Physiological Measurement 47(2):025003, 2026. DOI 10.1088/1361-6579/ae273e. https://doi.org/10.1088/1361-6579/ae273e