Sleep & Consciousness
Why do we lose hours of our lives every night - and what happens inside the brain while we do?
โถ Run the interactive simulationWhy your brain needs to go offline every night!
Junior level โ plain language, no maths
Ever wondered what your brain is actually up to while you sleep? It turns out sleep is not "doing nothing" at all - your brain is flat out busy. While your body rests, it's cleaning itself, replaying memories, rebalancing its chemistry and running essential repairs that simply can't be done while you're awake.
Sleep rolls along in cycles of about 90 minutes. Each one moves through light sleep, then deep sleep (where the brain rolls out big, slow waves), and finally REM sleep - the dreaming stage. During REM your brain is almost as active as when you're awake, yet your muscles go limp so you can't act your dreams out, while your eyes flick back and forth beneath your lids (that's the "rapid eye movement").
Deep sleep is when the housekeeping happens. A network of channels opens up and cerebrospinal fluid washes through the brain, sluicing away toxic waste - including amyloid-beta, the very protein that clumps into plaques in Alzheimer's disease. This cleansing glymphatic system mostly runs at night. Think of sleep as your brain's nightly dishwasher: skip it for a few nights and the gunk starts to pile up.
Things worth knowing
- During sleep the brain's glymphatic system clears ~60% more metabolic waste - including amyloid-beta, the protein that accumulates in Alzheimer's disease.
- After 17โ19 hours without sleep, cognitive performance drops to the equivalent of a blood-alcohol level of 0.05% - legally drunk in most countries.
- Dolphins and whales sleep with only half their brain at a time (unihemispheric slow-wave sleep), keeping one eye open to watch for predators.
Sleep Architecture, EEG Oscillations, and the Glymphatic System
Student level โ the core equations
Sleep isn't one state but a structured tour of several, each with its own EEG signature. NREM descends from light N1 into N2 - flecked with 12โ15 Hz sleep spindles - and down to N3, deep slow-wave sleep, where great sub-2 Hz delta waves sweep the cortex. Then comes REM, whose almost waking-like EEG belies a body held in near-total paralysis. A night runs four to six of these ~90-minute cycles, front-loaded with deep sleep and back-loaded with REM - which is why your longest, most vivid dreams arrive just before you wake.
Two clocks decide when you sleep. A homeostatic pressure (Process S) builds the longer you stay awake as the molecule adenosine accumulates in the brain - the physical feeling of tiredness. Riding on top is the circadian rhythm (Process C), a ~24-hour cycle kept by a master clock in the hypothalamus and reset each day by light striking the eye. Your alertness is roughly the tug-of-war between them, \(W(t) = C(t) - S(t)\). Caffeine works by impersonating adenosine at its receptor without switching it on - it doesn't lower your sleep pressure, it merely masks the signal, which is why the tiredness comes flooding back the moment it wears off.
The most startling discovery is why deep sleep is non-negotiable. The glymphatic system (Nedergaard, 2013) is a plumbing network in which cerebrospinal fluid flows along the blood vessels and rinses through the brain tissue, carrying off metabolic waste - including the amyloid-ฮฒ and tau proteins tied to dementia. That flushing runs about 60% stronger in NREM sleep than in waking, driven by the slow arterial pulses of deep sleep. A single bad night measurably raises amyloid-ฮฒ in the human brain - a sobering clue to why chronic sleep loss and Alzheimer's travel together.
Key formulas
| Homeostatic pressure | \(S(t) \propto [\text{adenosine}]\) | rises awake, falls asleep |
|---|---|---|
| Circadian process | \(C(t):\ \sim\!24\text{-h oscillator}\) | |
| Alertness | \(W(t) = C(t) - S(t)\) | |
| Caffeine | \(\text{blocks adenosine receptors}\) | hides sleep pressure |
| Glymphatic flow | \(J_{\text{CSF}} \propto \text{AQP4} \times \text{pulse amplitude}\) | |
| NREM clearance | \(\Delta[\text{A}\beta] \approx -60\%\) | vs waking; Xie et al. 2013 |
Things worth knowing
- Caffeine works by blocking adenosine receptors - it doesn't reduce sleep pressure, just hides the signal. When it wears off, adenosine floods in and fatigue returns suddenly.
- One night of sleep deprivation increases amyloid-beta levels in the human brain by ~5%, as measured by PET scan (Shokri-Kojori et al., 2018).
- Body core temperature must drop ~0.5ยฐC to initiate sleep - this is why a cool bedroom helps you fall asleep faster, and why a warm bath paradoxically also works (it lowers core temperature by pulling heat to the skin).
Thalamocortical Oscillations, Integrated Information Theory, and the Neural Correlates of Consciousness
Scholar level โ full mathematical depth
01Where the rhythms of sleep are made
The signature waves of sleep are generated by a loop between the cortex and the thalamus. In N2/N3 the thalamic reticular nucleus paces sleep spindles - 12โ15 Hz bursts that wax and wane - through alternating inhibition and rebound calcium bursting in relay neurons. Beneath them runs the cortical slow oscillation below 1 Hz, the whole cortex tolling between up states of near-waking activity and down states of collective silence. Sleep, far from being off, is the brain playing a small set of exquisitely coordinated rhythms.
02Why we might sleep at all: synaptic homeostasis
Those slow oscillations may solve a problem that waking creates. The synaptic homeostasis hypothesis (Tononi & Cirelli) argues that learning all day drives a net strengthening of synapses that can't continue indefinitely - it costs energy and space and saturates the network. Deep sleep, on this view, gently downscales synapses across the board, preserving the relative pattern learned while restoring capacity. Tellingly, the amount of slow-wave activity on a given night scales with how much you learned that day.
03The brainstem switch behind dreaming
REM sleep is orchestrated from the brainstem. Cholinergic nuclei there fire up the thalamus and cortex, producing an EEG almost indistinguishable from waking, even as glutamatergic circuits drive a spinal inhibition that leaves the body atonic - paralysed, so you don't enact your dreams. When that safeguard fails, in REM sleep behaviour disorder, people physically act out their dreams, a vivid demonstration of just how thin the line between dreaming and doing really is.
04The hard problem's foothold: neural correlates
Sleep and anaesthesia matter to the deepest question in neuroscience because they toggle consciousness on and off, offering a handle on the neural correlates of consciousness - the minimal machinery sufficient for experience. What is it about an active cortex in waking or REM that yields felt experience, while the very same cortex in deep sleep does not, despite plenty of neural firing? Isolating that difference is the empirical route into a problem long thought purely philosophical.
05Two theories at war
Two frameworks dominate. Global Workspace Theory holds that a stimulus becomes conscious when it is broadcast widely across fronto-parietal networks - an all-or-none "ignition" that makes information globally available. Integrated Information Theory takes the opposite tack, defining consciousness as \(\Phi\), the amount of irreducible integrated information a system generates, and makes the provocative claim that any system with \(\Phi > 0\) has some glimmer of experience - while a purely feedforward network, however vast, has essentially none.
06Putting a number on awareness
Strikingly, this has become measurable at the bedside. Zap the cortex with a magnetic pulse and record the echo: a conscious brain answers with a complex, widely reverberating response, an unconscious one with a simple local blip. Compressing that echo yields the perturbational complexity index, which tracks consciousness across sleep, anaesthesia and brain injury more reliably than any behaviour. It can now flag awareness in unresponsive, "vegetative" patients who cannot move or speak - a rough consciousness-meter, born from the study of sleep.
Key formulas
| Integrated information | \(\Phi = \min_{\text{partition}} \text{KL}\!\left(p \,\|\, p_{\text{part}}\right)\) | |
|---|---|---|
| Conscious if | \(\Phi > 0 \text{ and irreducible}\) | |
| Global ignition | \(\text{late } (>300\,\text{ms}) \text{ fronto-parietal activation}\) | |
| Complexity index | \(\text{PCI} = \dfrac{\text{LZ complexity}(\text{TMS-EEG})}{\text{signal entropy}}\) | |
| Spindle frequency | \(f_{\text{spindle}} \approx 12\text{โ}15\ \text{Hz}\) | |
| Synaptic homeostasis | \(\text{SWA} \propto \textstyle\sum w_{ij}\) | Tononi & Cirelli |
Things worth knowing
- TMS-EEG perturbational complexity index (PCI) can determine with ~90% accuracy whether a locked-in or vegetative-state patient is consciously aware - without any behavioural response.
- IIT predicts that a simple feedforward neural network, however large, has ฮฆ โ 0 and is therefore not conscious - a claim that directly challenges deep learning models of mind.
- Slow-wave sleep homeostasis: the EEG slow-wave activity (SWA) after sleep deprivation is precisely proportional to the amount of learning done while awake - a signature of synaptic downscaling.