https://harmonism.io/wheel-of-harmony/health/sleep/sleep-science
Sub-article of Sleep — Wheel of Health. The mechanism beneath the practice; the applied discipline lives in Sleep Protocols.
Sleep is not a uniform state. It cycles through distinct stages in approximately 90-minute ultradian rhythms, repeating four to six times per night.
NREM Sleep encompasses three stages. N1 is the light transitional phase from wakefulness. N2 is deeper light sleep, characterized by sleep spindles and K-complexes — neural signatures of memory processing. N3 is deep slow-wave sleep, the stage most critical for physical restoration, growth hormone secretion, immune system activation, and glymphatic toxin clearance. NREM dominates the early hours of the night.
REM Sleep is characterized by rapid eye movements, brain activity resembling wakefulness, temporary muscle atonia (paralysis preventing dream enactment), and vivid dreaming. REM is the stage of emotional processing, creative integration, and memory consolidation of procedural and associative knowledge. REM periods lengthen toward morning — losing the last hours of sleep disproportionately sacrifices REM.
The two systems work in complementary succession: NREM handles physical restoration and declarative memory consolidation; REM handles emotional regulation and creative synthesis. Both are non-negotiable.
Sleep timing is governed by two independent systems whose interaction determines when sleep comes, how deep it is, and when it ends. This is the two-process model (Borbély 1982; reappraised in Borbély, Daan, Wirz-Justice & Deboer, J Sleep Res 2016), and almost every practical question about sleep timing resolves into a question about which process is being addressed.
Process S — the homeostatic sleep drive. Adenosine accumulates in the brain throughout waking, binding to receptors and creating mounting sleep pressure. It dissipates during sleep. Process S is a debt that builds from the moment of waking and is discharged by sleeping. Caffeine works by blocking adenosine receptors — it does not eliminate the debt, only masks the signal. Caffeine has a half-life of approximately 6 hours and can linger for up to 12, meaning an afternoon coffee actively degrades that night’s sleep.
Process C — the circadian pacemaker. An endogenous rhythm of roughly 24 hours, generated by the suprachiasmatic nucleus (SCN) in the hypothalamus and synchronized to the environment primarily by light. Process C does not accumulate or discharge; it oscillates regardless of whether you have slept. It determines when in the day sleep is possible and restorative.
The practical consequence: sleep is good when the two processes align — high sleep pressure meeting the circadian window for sleep. Sleep is poor when they conflict, which is what shift work, jet lag, and irregular schedules produce. Duration addresses Process S. Timing and regularity address Process C. They are different problems and they need different interventions.
The SCN receives light input from specialized retinal ganglion cells — intrinsically photosensitive cells containing melanopsin, distinct from the rods and cones of image-forming vision. This is a separate visual channel whose only function is to tell the body what time it is.
Melatonin, secreted by the pineal gland in response to darkness, is the pacemaker’s output signal — the body’s internal announcement of night. It is not a sedative. It marks biological night rather than causing sleep, which is why measuring its onset (dim light melatonin onset, DLMO) is the standard method for reading a person’s circadian phase.
Two internal reference points matter for everything that follows. Core body temperature minimum (CBTmin) is the trough of the circadian temperature rhythm and the pivot of the entire light-response system. In healthy entrained young adults it falls roughly two hours before habitual wake time — derived from Khalsa et al. 2003, where the melatonin midpoint occurred 4.05 ± 0.78 h before habitual waking and approximately 2 h before the fitted CBTmin. The individual spread is substantial, and in older adults CBTmin sits later relative to wake time (Duffy et al. 1998). DLMO typically precedes habitual sleep onset by about two hours.
The single most useful and least known fact about circadian biology: light does not simply “set” the clock — its effect depends entirely on when it arrives, and the same light can shift the clock in opposite directions depending on the hour. The function describing this is the phase response curve (PRC).
The definitive human PRC comes from Khalsa, Jewett, Cajochen & Czeisler, “A phase response curve to single bright light pulses in human subjects,” J Physiol 549(3):945–952, 2003 (PMC2342968). Subjects received a single 6.7-hour bright light pulse at systematically varied circadian phases. The findings, verbatim where they matter:
“Phase delays occurred when the light stimulus was centred prior to the critical phase at the core body temperature minimum, phase advances occurred when the light stimulus was centred after the critical phase, and no phase shift occurred at the critical phase.”
Peak-to-trough amplitude of the curve was 5.02 hours. The largest observed individual shifts were −3.60 h (delay) and +2.01 h (advance). The transition from delay to advance at CBTmin is abrupt; the return from advance to delay across the subjective day is gradual. And crucially:
“During the subjective day, no prolonged ‘dead zone’ of photic insensitivity was apparent.”
CBTmin is therefore the hinge of the whole system. Light before it delays the clock — pushing sleep later. Light after it advances the clock — pulling sleep earlier. Since CBTmin sits roughly two hours before habitual waking, light received on rising falls in the advance zone, and light received in the late evening falls in the delay zone. This is the mechanism beneath every piece of morning-light and evening-darkness advice, and it explains why the advice is not arbitrary and not symmetrical.
It also explains a failure mode invisible to folk sleep hygiene: a person who wakes drastically earlier than their biological phase receives light before CBTmin, which delays the clock and makes the problem worse. Forcing a 4 a.m. rise on a late chronotype is not discipline; it is a phase delay administered daily.
Intensity. Zeitzer, Dijk, Kronauer, Brown & Czeisler, J Physiol 526(3):695–702, 2000 (PMC2270041) established the dose-response: the half-maximal phase-shifting response falls at roughly 120 lux, with saturation near 550 lux. Their headline finding — half the maximal phase-delaying response to a single episode of evening bright light (~9,000 lux) is obtained with just over 1% of that light, ordinary dim room lighting of ~100 lux. Household evening lighting is not circadianly innocent. Melatonin suppression saturates even lower, near 200 lux.
Duration is strongly non-linear. St Hilaire et al., J Physiol 590(13):3035–45, 2012 found that a 1-hour bright light pulse produced a PRC with peak-to-trough amplitude of 2.20 h — “~40% of that for the 6.7 h PRC despite representing only 15% of the light exposure duration.” The first hour does most of the work. This is the strongest available justification for the morning-light practice as a short daily discipline rather than an extended one.
Realistic rate of change. Laboratory ceilings are not field results. Eastman, Gazda, Burgess, Crowley & Fogg, Sleep 28(1):33–44, 2005, using three days of advancing schedule plus 3.5 h of intermittent morning bright light (~5,000 lux), achieved median DLMO advances of only 1.4–1.9 h — roughly 0.5 to 0.65 hours of true phase advance per day. Advancing the schedule 2 h/day was no better than 1 h/day, because the behavioral schedule outran the clock. Advances are harder than delays. A person moving their rhythm two hours earlier should expect three to four days of consistent work, not a decision.
A zeitgeber (“time-giver”) is any environmental cue capable of entraining a biological rhythm. They are not equal, and the ranking has direct practical consequences.
Light is dominant, and by a wide margin. Mistlberger & Skene, Biol Rev 79(3):533–56, 2004, reviewing the evidence: “In humans, social zeitgebers appear weak by comparison with light. In temporal isolation or under weak light–dark cycles, humans may ignore social cues and free-run independently.” They add a methodological point that dissolves much confusion — apparent entrainment to fixed sleep–wake schedules in sighted people is generally photic entrainment in disguise, because scheduled behavior necessarily creates a light–dark cycle. The corroborating evidence is stark: Klerman et al., Am J Physiol 274(4):R991–6, 1998, found that among 15 totally blind subjects with no light input, only 9 maintained synchronization to the 24-hour day despite regular activity schedules.
Exercise is a genuine but secondary zeitgeber with its own curve. Youngstedt, Elliott & Kripke, “Human circadian phase-response curves for exercise,” J Physiol 597(8):2253–2268, 2019 (PMC6462487) — 99 participants, 60 minutes of moderate exercise on three consecutive days at one of eight clock times. Significant PRCs emerged, with phase advances peaking around 07:00 and again in the 13:00–16:00 window, and delays from roughly 19:00–22:00. The afternoon advance zone is the novel finding and has no counterpart in the light PRC. Magnitude: one hour of exercise produced approximately one-third the shift of three hours of bright light in a comparable protocol. Morning exercise therefore reinforces morning light; late-evening hard training works against both.
Exogenous melatonin shifts the clock, and its curve is nearly the inverse of light’s. Burgess, Revell & Eastman, J Physiol 586(2):639–47, 2008 — maximum advance 1.8 h, maximum delay 1.3 h, with advances produced by afternoon/early-evening dosing (2–4 h before DLMO for the 0.5 mg dose; Burgess et al., JCEM 2010) and a “dead zone of minimal phase shifts… around the first half of habitual sleep.” The operational conclusion is exact and widely ignored: melatonin taken at bedtime as a sleep aid has minimal phase-shifting effect. Its power is in the timing, not the dose — larger doses do not produce larger shifts.
Food entrains the peripheral clocks, not the pacemaker — and this is the most underappreciated result in the field. Every tissue has its own oscillator. In animals, restricted feeding shifts liver clock gene expression by up to 12 hours while leaving the SCN untouched (Damiola et al., Genes Dev 2000; Stokkan et al., Science 2001 — a 10-hour liver shift within 2 days). In humans, Wehrens et al., “Meal timing regulates the human circadian system,” Curr Biol 27(12):1768–1775, 2017 (PMC5483233) delayed meals by 5 hours for six days: plasma melatonin and cortisol — the SCN markers — did not shift at all, while the plasma glucose rhythm delayed by 5.69 ± 1.29 hours and adipose PER2 expression by roughly an hour.
This produces the field’s cleanest structural insight. Light times the brain; food times the body. Internal desynchrony — the SCN on one schedule and the metabolic organs on another — is a distinct pathology from sleep deprivation, invisible to any measure of sleep duration, and produced precisely by eating at hours that contradict the light schedule. It is the mechanism that makes late-night eating a circadian problem rather than merely a caloric one.
Temperature and pure social cues remain poorly quantified in humans. No human PRC for ambient temperature has been established, though the field evidence from pre-industrial societies (below) suggests the falling-temperature signal may be more important than the literature currently reflects. Stated as open.
| Zeitgeber | Target | Relative strength |
|---|---|---|
| Light | SCN (master pacemaker) | Dominant |
| Exogenous melatonin | SCN | Moderate; timing-critical, dose-insensitive |
| Exercise | SCN | Moderate; ~⅓ of equivalent bright light |
| Feeding time | Peripheral/metabolic clocks | Dominant for periphery, negligible for SCN |
| Social cues | SCN | Weak; largely mediated through light and arousal |
| Ambient temperature | Unclear | Unquantified in humans |
This section exists to settle a specific and consequential error, and the answer runs against a claim held in common by Ayurvedic popularizers, sleep-hygiene writing, and earlier versions of this article: that sleep obtained before midnight is intrinsically deeper because deep sleep belongs to those hours.
The experimental design that can answer this is forced desynchrony — placing subjects on a rest–activity cycle the circadian pacemaker cannot follow (28 hours, say), so that sleep episodes land at every circadian phase across the study. This dissociates clock time from time-since-sleep-onset, which ordinary life confounds completely.
Dijk & Czeisler, J Neurosci 15(5):3526–38, 1995 — 8 men, 33–36 days, 28-hour cycles. The finding, verbatim from the abstract:
“Slow-wave activity in non-REM sleep exhibited a low amplitude circadian modulation which did not parallel the circadian rhythm of sleep propensity.”
and
“Slow-wave activity decreased and sleep spindle activity increased in the course of all sleep episodes.”
All sleep episodes. Wherever in the 24-hour day the sleep was placed, slow-wave activity front-loaded and decayed. The concentration of N3 in the first third of the night is a homeostatic phenomenon — the discharge of accumulated adenosine pressure — not a property of the hours between 10 p.m. and 2 a.m.
The follow-up (Dijk, Shanahan, Duffy, Ronda & Czeisler, J Physiol 505(3):851–8, 1997), analyzing 1.22 million power spectra across 146 sleep episodes, reports that NREM EEG activity from 0.25 to 11.5 Hz was primarily dependent on prior sleep time and only slightly affected by circadian phase — with the lowest values coinciding with the phase of melatonin secretion. If that reading holds, the small residual clock effect on slow-wave power runs opposite in sign to the folk claim. (Verification note: the 1995 abstract above was read directly at source; the 1997 sentence has been confirmed only at second hand and should be checked against the print article before being quoted as decisive.)
What genuinely does vary with circadian phase:
| Variable | Circadian behavior |
|---|---|
| REM sleep | Strong rhythm; crest shortly after CBTmin |
| Sleep spindle activity | Marked rhythm; crest at the start of the habitual sleep episode |
| Sleep latency | Minimum at CBTmin |
| Wakefulness within sleep | Lowest near CBTmin; peaks 15–17 h after |
| Slow-wave activity | Low-amplitude modulation only; dominated by prior sleep time |
The honest synthesis, which is not a debunking. Sleeping at the wrong circadian phase is genuinely worse — but the damage takes the form of fragmented, poorly consolidated, latency-disturbed, REM-mistimed sleep, not shallower sleep. Sleeping 04:00–12:00 is worse than 23:00–07:00, and the traditions that insist on early sleep are not wrong about the outcome. They are wrong about the mechanism, and only in the specific sense that the depth of the early night belongs to sleep pressure rather than to the clock. A further confound operates in ordinary life: earlier bedtimes correlate with more total sleep and less accumulated debt, which produces more N3 for reasons that have nothing to do with the hour.
Held as doctrine in Sleep: sleep aligned with the solar cycle is superior. That claim survives intact. What does not survive is the specific mechanistic justification that pre-midnight hours contain more deep sleep as a property of those hours.
The most significant development in sleep epidemiology since Walker’s synthesis is the finding that when you sleep, measured as consistency, carries information that how long you sleep does not.
The Sleep Regularity Index (SRI) — introduced in Phillips et al., Sci Rep 7:3216, 2017 — is the probability that a person is in the same state (asleep or awake) at any two moments 24 hours apart, averaged across a week of accelerometry. 100 is perfect regularity; 0 is random.
Windred, Burns, Lane, Saxena, Rutter, Cain & Phillips, “Sleep regularity is a stronger predictor of mortality risk than sleep duration,” SLEEP 47(1):zsad253, 2024 (Oxford Academic) — 60,977 UK Biobank participants with accelerometry, 6.3 years mean follow-up, 1,859 deaths. Fully adjusted hazard ratios against the lowest SRI quintile:
| SRI quintile | All-cause mortality |
|---|---|
| 20–40th | 0.80 [0.69–0.93] |
| 40–60th | 0.75 [0.64–0.88] |
| 60–80th | 0.72 [0.61–0.84] |
| 80–100th | 0.70 [0.59–0.83] |
Independently replicated in a larger overlapping sample: Cribb et al., eLife 12:RP88359, 2023 — 88,975 participants, 3,010 deaths, hazard ratio 1.53 [1.41–1.66] at the 5th SRI percentile relative to the median.
State the claim precisely, because the popular version overstates it. Duration was also protective in Windred, with a top-quintile hazard ratio of 0.76 — overlapping with regularity’s 0.70. The “regularity beats duration” finding is a claim about model fit: SRI models fit better by AIC, and adding duration to a regularity model did not improve fit, while the reverse was not true. The correct formulation is regularity carries information that duration does not, and once regularity is known, duration adds little — not that irregular sleep is dramatically more lethal than short sleep. The two papers also report median SRI values of 81 and 60 on the same cohort, meaning the index is pipeline-dependent and no target number should be quoted to anyone.
The mechanism is photic, and it closes the loop with the PRC. Phillips et al. 2017 found that the most irregular quintile of 61 undergraduates had DLMO at 00:08 versus 21:32 for the most regular — a phase delay of roughly two and a half hours — together with lower-amplitude light exposure rhythms (102 vs 179 lux). Modelling indicated the phase difference was primarily attributable to their light exposure patterns. Irregularity harms because it produces a blunted, mistimed light signal. It is not a moral failure of discipline; it is a degraded zeitgeber.
Social jetlag — Roenneberg’s term for the discrepancy between biological and social time, formally the absolute difference between mid-sleep on free days and on work days (Wittmann et al., Chronobiol Int 2006; Roenneberg et al., Curr Biol 2012 for the obesity association). Treat it with the caution its own author recommends: the 2019 self-critical review (Roenneberg, Pilz, Zerbini & Winnebeck, Biology 8(3):54) concedes the original concept “may be too simplistic,” that it confounds circadian misalignment with sleep debt, and that “the jury is certainly still out.” Their reframing deserves emphasis because it inverts the usual counsel: the pathology is located in the constraints social clocks impose on workdays, not in weekend sleep-in, which is compensatory rather than causal.
Individual circadian phase varies enormously between people, substantially by genetics, and this is the fact that most sleep advice ignores.
Distribution. Roenneberg et al., Sleep Med Rev 11(6):429–38, 2007, from over 55,000 Munich ChronoType Questionnaire respondents (the database later exceeding 200,000): chronotype, measured as sleep-debt-corrected mid-sleep on free days, is near-normally distributed with a slight late skew and spans approximately 9.5 hours across the population. Only 1% begin sleep on free days at 10 p.m. or earlier; 8.2% fall asleep at 3 a.m. or later. Lateness peaks around age 20–21, then advances steadily with age; sex differences disappear around 50.
Heritability. Twin studies place it at 40–47% (Vink et al., Chronobiol Int 2001; Toomey et al., Chronobiol Int 2015, h² = 0.42). SNP-based heritability from the largest genome-wide analysis is 13.7% (Jones et al., “Genome-wide association analyses of chronotype in 697,828 individuals,” Nat Commun 10:343, 2019 — 351 genome-wide significant loci, implicating PER1, PER2, PER3, CRY1 and ARNTL/BMAL1). Both figures should be quoted together; either alone misleads. A rare dominant CRY1 variant lengthening circadian period, found at allele frequencies up to 0.6% in some populations, causes familial delayed sleep phase disorder (Patke et al., Cell 169(2):203–215, 2017).
Effect size, honestly. Among 85,760 accelerometry participants in Jones 2019, the top 5% versus bottom 5% of morning-preference polygenic score differed by roughly 25 minutes of sleep timing. Common genetic variation moves the average person by tens of minutes, not hours. The 9.5-hour population spread is therefore mostly environmental, developmental and age-related — which is good news, because it means chronotype is substantially movable. It is not, however, movable by exhortation, and not movable quickly: the ceiling is the 0.5–0.65 h/day advance rate established above.
Delayed sleep-wake phase disorder affects approximately 3.3% of adolescents and young adults in the best-powered studies (Sivertsen et al., BMC Public Health 2013, n=10,220; Sivertsen et al., Sleep Med 2021, n=50,054) — both Norwegian and self-report-based, which should be attached to the number when it is cited.
Mendelian randomization complicates the moral story. Jones 2019 found morning chronotype causally associated with better subjective wellbeing and lower schizophrenia risk, but no causal effect on BMI, type 2 diabetes, or fasting insulin — undercutting the observational literature’s implication that eveningness causes metabolic disease. The metabolic harm attributed to late chronotypes may be the harm of misalignment between a late clock and an early society, which is a different problem with a different remedy.
The doctrinal consequence. A universal wake time prescribed for everyone — 5 a.m., 4 a.m., whatever the number — is not a discipline but a category error. Across a 9.5-hour distribution, a single clock hour lands at radically different circadian phases, and for a late chronotype it lands before CBTmin, where morning light delays rather than advances. Correct practice is a wake time positioned relative to one’s own phase, held constant. See Sleep Protocols for the method.
The relationship between sleep duration and mortality is U-shaped, with a nadir at approximately 7 hours, and a long-sleep arm considerably steeper than the short-sleep arm.
Liu et al., “Sleep duration and risk of all-cause mortality: a flexible, non-linear, meta-regression of 40 prospective cohort studies,” Sleep Med Rev 32:28–36, 2017 — 40 cohorts, 2,200,425 participants, 271,507 deaths, restricted cubic spline with 7 hours as reference:
| Sleep duration | Relative risk [95% CI] |
|---|---|
| 4 h | 1.05 [1.02–1.07] |
| 5 h | 1.06 [1.03–1.09] |
| 6 h | 1.04 [1.03–1.06] |
| 7 h | 1.00 (reference) |
| 8 h | 1.03 [1.02–1.05] |
| 9 h | 1.13 [1.10–1.16] |
| 10 h | 1.25 [1.22–1.28] |
| 11 h | 1.38 [1.33–1.44] |
Corroborated by Cappuccio et al., Sleep 33(5):585–92, 2010 (1,382,999 participants: short sleep RR 1.12, long sleep RR 1.30) and by Svensson et al., JAMA Netw Open 4(9):e2122837, 2021 (Asia Cohort Consortium, 322,721 participants — “a sleep duration of 7 hours was the nadir for associations with all-cause, cardiovascular disease, and other-cause mortality in both men and women”).
Three things follow, and each corrects something in the popular synthesis. The nadir is 7 hours, not 8. The short-sleep hazards are small — relative risks of 1.04–1.06 at 5–6 hours — and in Liu’s analysis reached significance in women but not men. The long-sleep arm is three to five times steeper, which is almost certainly substantially reverse causation: long sleep is a known marker of undiagnosed illness, depression and frailty rather than a cause of death.
This does not license short sleep. Mortality is a crude endpoint that says nothing about the cognitive, emotional and metabolic costs documented below, which appear at durations well above where the mortality curve moves. It does mean the confident public messaging on short sleep outruns its effect sizes, and that a person lying awake in anxiety about not reaching eight hours is being harmed by the advice.
During deep NREM sleep, damaged tissues are rebuilt, muscles recover from exertion, and cells renew. Growth hormone secretion peaks during N3 slow-wave sleep, driving tissue repair, muscle recovery, and fat metabolism. Without adequate deep sleep, this hormonal cascade is truncated.
Sleep strengthens the innate immune system. During sleep, the body produces cytokines that fight infection, inflammation, and stress. Sleep deprivation can reduce natural killer cell activity by up to 70%, significantly increasing vulnerability to infection and potentially fostering cancer progression.
Sleep regulates cortisol, leptin and ghrelin (appetite hormones), and insulin sensitivity. Seven hours or less of sleep can raise cortisol by 10-20%. Short sleep disrupts leptin/ghrelin balance, driving increased hunger and weight gain. Serotonin — 90% of which is produced in the gut — depends on sleep for proper synthesis.
Note the interaction with meal timing established above: the metabolic consequences of poor sleep and the metabolic consequences of eating against the clock are related but separable, and a person sleeping adequately while eating at midnight has solved one problem and not the other.
The glymphatic system, active primarily during deep sleep, flushes metabolic waste products — including beta-amyloid proteins associated with Alzheimer’s disease — from the brain’s extracellular space. Acute sleep deprivation can elevate beta-amyloid levels by up to 30% in experimental models. Chronic restriction correlates with faster cognitive decline in observational studies. This mechanism makes sleep the brain’s primary detoxification process.
During sleep, the hippocampus replays neural patterns from the day, transferring experiences to long-term storage in the neocortex. Sleep also promotes synaptic pruning — selectively weakening unnecessary neural connections to optimize efficiency and prevent cognitive overload. A single night of poor sleep measurably impairs learning capacity.
Sleep allows nocturnal blood pressure dipping, facilitating vascular recovery and reducing chronic inflammation. Epidemiological data shows a 24% increase in heart attacks on the Monday following the spring daylight saving time shift, when populations lose just one hour of sleep — an experiment in acute circadian misalignment conducted on entire nations twice a year.
Sleep deprivation impairs prefrontal cortex control over amygdala reactivity, producing heightened anxiety, irritability, mood instability, and impaired decision-making. The cognitive impairment from sleep deprivation is comparable to alcohol intoxication.
Sleep is one of the primary factors in lifespan and healthspan. Adequate sleep reduces the incidence of cardiovascular disease, diabetes, cancer, and neurodegenerative conditions. It slows cellular aging by supporting repair and reducing chronic inflammation.
In Chinese medicine, sleep is understood as the primary mechanism for preserving Jing — the foundational essence inherited from one’s parents that constitutes the basis of vitality and longevity. Each night of quality sleep nourishes and protects Jing. Chronic sleep deprivation progressively depletes it, leading to premature aging, loss of vitality, and weakening of the body. Conversely, practices that deplete Jing without adequate sleep recovery (long-distance running, excessive stimulation, chronic overwork) accelerate the exhaustion of this irreplaceable resource.
The Jing frame and the regularity data converge on the same practical counsel from different directions, and the convergence is worth naming: what erodes Jing in the Chinese account is not a single short night but sustained disorder — the same thing the SRI measures. Sleep debt accumulates over years and decades. There is no shortcut to repaying it, only the slow, patient discipline of consistent restorative sleep and complementary practices.
Sleep deprivation effects are cumulative and systemic: anxiety, irritability, decreased concentration, inability to make critical decisions, major buildup of toxins, dramatic cortisol elevation, inflammatory cytokine surge. Chronic sleep deprivation contributes to cardiovascular disease, arthritis, diabetes, obesity, cancer, Alzheimer’s, depression, and anxiety disorders. Over 30% of adults in industrialized nations sleep fewer than 6 hours per night.
Sleep deprivation has been implicated in catastrophic events: the 1986 Chernobyl disaster, the 1989 Exxon Valdez oil spill, and over 6,000 fatal drowsy-driving crashes annually in the U.S. alone.
The closest substitutes for sleep — meditation, inversion therapy, Jing herbs — are valuable complements but not replacements. Sleep is irreplaceable.
The question matters doctrinally, because a widely repeated claim holds that consolidated eight-hour sleep is an industrial artifact and that waking in the night is ancestral and benign. The evidence is genuinely mixed and the honest position is more interesting than either popular version.
The historical case. A. Roger Ekirch (At Day’s Close: Night in Times Past, 2005; “Sleep We Have Lost,” American Historical Review 106, 2001) documented several hundred references across European sources to a “first sleep” and “second sleep” separated by an interval of wakefulness — “the watch” — spent in prayer, reflection on dreams, conversation, and lovemaking. The references are casual, as though describing something unremarkable. The pattern faded in the nineteenth century with domestic artificial lighting.
The experimental case. Wehr’s photoperiod work at NIMH placed subjects in 14 hours of darkness nightly. Sleep settled into two blocks separated by an interval of quiet wakefulness. The fully verified companion paper (Wehr et al., Am J Physiol 265(4):R846–57, 1993, n=15) established the broader finding — human endocrine night expands to fill the dark period: melatonin secretion duration 11.9 h under long nights versus 10.3 h under short, with parallel expansions in prolactin, low body temperature, and sleep duration (10.6 h vs 7.6 h). Humans are photoperiod-responsive animals.
The counter-case. Yetish et al., “Natural Sleep and Its Seasonal Variations in Three Pre-industrial Societies,” Curr Biol 25(21):2862–2868, 2015 — 94 participants, 1,165 days of actigraphy among the Hadza, Ju/’hoansi San, and Tsimane. Findings that unsettle several popular narratives at once: sleep duration was 5.7–7.1 hours, at or below industrial norms; sleep onset came on average 3.3 hours after sunset, not at dusk; waking was typically before sunrise; and explicitly, “the traditional groups do not regularly awaken for extended periods in the middle of the night.” Napping occurred on fewer than 7% of winter days. Neither the San nor the Tsimane has a word for insomnia, and 1.5–2.5% report regular sleep problems against 10–30% chronic insomnia in industrial societies.
Yetish’s proposed zeitgeber is temperature, not light: sleep consistently spanned the period of falling ambient temperature and ended near the daily temperature nadir, and differences in wake time between groups tracked temperature rather than photoperiod. “The daily cycle of temperature change, largely eliminated from modern sleep environments, may be a potent natural regulator of sleep.” This is a substantial hypothesis with thin experimental support and it belongs in the open column.
The historiographic caution. Boyce, “Have we lost sleep? A reconsideration of segmented sleep in early modern England,” Medical History 67(2):91–108, 2023, searched Early English Books Online and found “second sleep” appearing only fifteen times in the entire 1475–1700 corpus, with only five instances containing both terms together — arguing that “first sleep” is polysemous and can denote simply the initial, deepest phase of an unbroken night. Ekirch has conceded that people “did not sleep according to the same timetable” and that his claim is one of predominance rather than universality.
Synthesis. All three streams are compatible if the bifurcation is understood as a response to long dark periods rather than a species-typical pattern. Equatorial peoples never experience 14-hour nights; northern Europeans experienced them annually. Wehr’s chamber reproduced the northern winter and reproduced the pattern. The defensible claims: segmented sleep is real, documented, and reliably inducible by extended darkness; it is absent in the best-measured pre-industrial societies; and the assertion that it is the natural human pattern outruns all three evidence streams. What follows practically is modest but real — a person who wakes in the night during a long winter is not necessarily malfunctioning, and the anxiety about the waking probably costs more than the waking.
Why We Sleep: Unlocking the Power of Sleep and Dreams — Matthew Walker (2017, Scribner). The most comprehensive modern synthesis on sleep science for a general audience. Walker is a British-American professor of neuroscience and psychology at UC Berkeley, where he founded the Center for Human Sleep Science; formerly assistant professor of psychiatry at Harvard Medical School. Over 100 peer-reviewed papers. New York Times bestseller, 1M+ copies, 30+ languages. Won the 2020 Carl Sagan Prize for Science Popularization.
The book is structured in four parts: sleep’s evolutionary history and neurophysiology; the vital roles sleep plays in brain function, physical restoration, and dreaming; the wide-ranging harms of insufficient sleep; and practical recommendations for optimizing sleep hygiene. Walker debunks the myth that alcohol or caffeine serve as legitimate sleep tools, critiques the cultural devaluation of sleep in favor of productivity, and advocates CBT-I (Cognitive Behavioral Therapy for Insomnia) as the non-pharmacological alternative to sleeping pills (70-80% insomnia reduction, no adverse effects).
Criticisms to note: Alexey Guzey’s essay “Matthew Walker’s Why We Sleep Is Riddled with Scientific and Factual Errors” (2019) — self-published and not peer-reviewed, but its central charges are checkable and, against the literature assembled above, largely correct. Walker’s claim that “the shorter your sleep, the shorter your life span” describes a linear relationship that the meta-analytic data does not show; the relationship is U-shaped with a 7-hour nadir and a steeper long-sleep arm. Walker’s assertion that the WHO “declared a sleep loss epidemic” appears to be unsupported. His adaptation of a figure from Milewski et al. 2014 removed the 5-hour column, which showed lower injury rates than 6 hours. The claim that under-seven-hour sleep more than doubles cancer risk is contradicted by a 2018 meta-analysis of 1,550,524 participants. Impossible statistics (“400–600 percent fewer errors”) appear in both the book and associated papers.
The directional argument — that industrial society is systematically sleep-deprived and that this is consequential — stands. The specific magnitudes require independent verification, and the rigid eight-hour prescription has produced documented iatrogenic harm in people who developed anxiety-driven insomnia trying to force it. This is why Sleep Protocols treats regularity and phase as the primary targets and duration as a range to be discovered rather than a number to be hit.