Neuroscientists have spent two decades testing whether the brain sits at a “critical point,” a phase transition between stable and unstable states that would give neural circuits maximum flexibility, dynamic range, and sensitivity to inputs. The accumulating evidence points to a more precise answer: the brain operates very close to that boundary but stays slightly on the stable side, in what researchers call a subcritical or quasicritical regime. That small offset turns out to be not a flaw but a feature, one that shifts depending on sleep stage, cognitive demand, and possibly age.
What Criticality Means for Neural Circuits
In physics, criticality describes a point where a system’s order parameter changes continuously but non-smoothly with a control parameter, producing scale-free fluctuations that span all sizes. For the brain, this translates into cascades of neural firing called “neuronal avalanches” that follow power-law statistics rather than clustering around a single typical size. The foundational demonstration came from recordings in cortical cultures, which showed that avalanche-size distributions obeyed a power law with a mean-field exponent of approximately 3/2, the signature expected from a critical branching process.
Why would the brain benefit from hovering near this boundary? At the critical point, even weak stimuli can propagate across large networks, the system responds efficiently to gentle perturbations, and information storage and transmission are theoretically optimized. Computational models describe this as an economically and ecologically optimal working point where the trivial resting state loses stability and the network becomes maximally responsive. The catch is that maintaining an exact critical state poses serious challenges: small perturbations or noisy inputs can push the system into runaway, seizure-like activity.
Avalanches Across Species and Brain States
The critical-branching signature first observed in cultured tissue was later confirmed in living, awake brains. Multielectrode array recordings from awake monkey cortex demonstrated that spontaneous cortical activity organized into neuronal avalanches with the same scale-invariant statistics, establishing cross-species and awake-brain relevance. Human evidence followed from noninvasive magnetoencephalography, which detected scale-invariant cascades in resting brain signals, providing a complementary measurement stream to invasive electrode studies.
The most telling data, however, came from intracranial depth recordings in humans across different sleep stages. That work showed the brain is “very close to the critical state” but remains in a subcritical regime that varies with consciousness. Slow-wave sleep brought the brain closest to the critical boundary. Wakefulness was slightly more subcritical, and REM sleep sat further still from the transition. The finding that criticality distance tracks with sleep architecture suggests the brain actively regulates how close it sits to the edge rather than passively drifting there.
Driven and Slightly Subcritical
If avalanche statistics look so much like criticality, what keeps the brain from tipping over? In vivo spike recordings offered a direct answer: the system operates in a driven subcritical state where ongoing sensory and internal inputs pull neural dynamics away from the exact critical point. The avalanche power laws still appear, but they coexist with measurable subcritical signatures, meaning the network can produce large cascades without the risk of uncontrolled runaway activity.
A separate line of work formalized this idea as “quasicritical” dynamics. Published in Physical Review Letters, that framework showed how external stimuli systematically displace brain dynamics from the critical point while preserving near-critical responsiveness. The quasicritical label captures the practical reality: the brain harvests many of the computational advantages of criticality without paying the full cost of instability.
Attention Tunes the Distance
The gap between the brain and the critical boundary is not fixed. Experimental work comparing resting-state and task-engaged brain activity found that focused cognitive tasks shift dynamics to a more subcritical regime with reduced dynamic range. In plain terms, when a person concentrates on a demanding task, the brain appears to sacrifice some of its broad sensitivity in favor of stable, noise-resistant processing. At rest, the system relaxes back toward criticality, regaining wide-ranging responsiveness.
This dynamic tuning aligns with a broader theoretical proposal: the distance to criticality may be varied on a moment-to-moment basis to match task requirements. A brain region optimized for low-noise sensory coding, for example, might perform poorly at long-range coordination, and vice versa. Different cognitive demands call for different positions along the subcritical-to-critical spectrum, which means no single fixed operating point would serve all purposes equally well.
Why the Brain Does Not Need Exact Criticality
One persistent question is how a biological system could stay near a critical point without fine-tuning. A theoretical answer comes from work on Griffiths phases in heterogeneous networks. Because the brain’s wiring is highly irregular, with hubs, modules, and long-range connections, the critical region is not a razor-thin line but a “stretched” critical-like regime that spans a range of control-parameter values. Network heterogeneity itself acts as a buffer, keeping the system in a broad near-critical zone without requiring precise parameter adjustment.
That said, the evidence is not without caveats. Methodological reviews have questioned whether apparent power-law distributions uniquely imply criticality, noting that finite-size effects, filtering choices, and limited recording durations can all bias the estimated exponents. Some studies that reanalyzed avalanche statistics with stricter model-comparison techniques have found that alternative heavy-tailed distributions sometimes fit as well as pure power laws. This has prompted a shift away from treating any single exponent value as a definitive hallmark and toward a more holistic view that combines avalanche distributions, temporal correlations, and responses to perturbations.
Within this broader picture, the emerging consensus is cautious but robust: cortical networks show multiple, converging signatures of operation near a phase transition, yet they are systematically nudged to the stable side by ongoing drive and regulatory mechanisms. Rather than a knife-edge, the brain appears to inhabit a flexible corridor of near-critical states. Sleep, anesthesia, and task engagement move it up and down that corridor, trading off sensitivity against stability as behavioral demands change.
This perspective also reframes long-standing clinical questions. If seizures represent pathological supercriticality, then understanding how healthy brains maintain a subcritical buffer could illuminate new therapeutic strategies. Conversely, disorders characterized by sluggish or fragmented neural dynamics might reflect an overly deep subcritical regime, where signals die out too quickly to support coherent cognition. Early work along these lines has begun to examine whether deviations from near-critical statistics can serve as biomarkers for altered consciousness or neuropsychiatric disease.
For now, the most striking lesson from criticality research is conceptual. The brain does not seem to chase a single, perfectly tuned operating point. Instead, it rides close to a phase transition, adjusting its distance from the edge as circumstances demand. That strategy, made possible by heterogeneous wiring and constant external drive, lets neural circuits capture much of criticality’s computational power while avoiding its most dangerous instabilities. In that sense, being almost critical (subcritical, driven, and quasicritical) may be exactly what makes the brain both resilient and adaptable.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
