Deep Sleep (Slow-Wave Sleep)

Deep sleep, slow-wave sleep, and N3 all name the same thing. The AASM staging label is N3. The EEG signature is slow-wave sleep (SWS). "Deep sleep" is the consumer term your wearable shows on screen.

Formally, a sleep lab scores a window as N3 when at least 20 percent of it contains slow waves at 2 Hz or below with a peak-to-peak amplitude of 75 µV or more (the EEG threshold used to call a 30-second window deep). Those slow waves come from cortical neurons firing in synchronized up-and-down states across huge populations. Hippocampal sharp-wave ripples (fast 150 to 250 Hz bursts in the memory system) nest inside these slow oscillations, linked by thalamocortical sleep spindles. That triple coupling (slow oscillation up-state plus sleep spindle plus ripple) is the physical basis of memory consolidation.

Where SWS sits in the night. In healthy young adults, N3 takes up roughly 13 to 23 percent of total sleep time. It is heavily front-loaded: most N3 happens in the first two sleep cycles. The last cycle is usually REM-rich and N3-poor. So if you cut sleep short on the back end (5 hours instead of 8), you mostly lose REM. Cut it short on the front end, or fragment the first half with alcohol, and you mostly lose N3.

N3 is not the only "deep" sleep. N2 sits in the middle, with sleep spindles and K-complexes, and makes up about 45 to 55 percent of total sleep. It restores you too. The density and shape of N2 spindles are themselves biomarkers of how your brain works. Treating N3 as the only "good" sleep oversimplifies things. You need every stage.

So why does N3 get all the longevity attention? Three reasons. First, growth hormone release is tied to N3. Van Cauter et al. 1997 J Clin Invest showed that boosting SWS with a drug (gamma-hydroxybutyrate, in healthy young men) raised GH output in proportion. Roughly two-thirds of your daily GH is released during night sleep, linked to the first N3 episode. Second, the glymphatic literature (the brain's overnight waste-rinse system; Xie 2013, Fultz 2019) shows CSF oscillations coupled to NREM slow waves, with the strongest flow during slow-wave-rich sleep. Third, N3 is the stage that falls earliest and fastest with age (next section).

Sleep spindles deserve a mention. Spindle density and shape track with thinking performance and decline in Alzheimer's. Tweaking spindles is a real research target. Closed-loop auditory stimulation (pink-noise pulses timed to slow-wave up-states) boosts slow-wave activity, and in some studies, the memory consolidation that follows (Ngo et al. 2013 Neuron; Papalambros 2017 Front Hum Neurosci). The slow-wave boost replicates. The memory benefit is fragile.

Three mechanisms anchor the link between SWS and longevity.

1. Growth hormone and tissue upkeep. Van Cauter and colleagues spent two decades mapping how SWS relates to the GH axis (the brain-pituitary-liver loop that releases growth hormone). Their 2000 JAMA paper (Van Cauter et al.) showed that as N3 fell from about 19 percent of sleep in young adult men (ages 16 to 25) to about 3.4 percent by midlife (36 to 50), night-time GH release fell about 75 percent right alongside it. The 1997 J Clin Invest paper proved the link runs both ways: boosting SWS with gamma-hydroxybutyrate (sodium oxybate) raised GH in proportion. Adult GH is not the same as childhood-growth GH, but it shapes tissue repair, lipolysis (fat breakdown), IGF-1 production, and bone turnover. The SWS-GH link is one of the strongest mechanistic reasons to guard deep sleep as you age.

2. Glymphatic clearance and brain waste removal. The Xie 2013 Science paper showed that the space between brain cells in mice expands about 60 percent during sleep (or anesthesia), roughly doubling how fast injected amyloid-beta (the protein that clumps in Alzheimer's) gets cleared. Fultz et al. 2019 Science showed in humans that large NREM slow waves are followed by anticorrelated BOLD oscillations and big CSF flow pulses into the fourth ventricle. The strongest flow happens during slow-wave-rich sleep. The honest caveat: Miao et al. 2024 Nature Neuroscience used a different injection method and reported reduced clearance during sleep, so how much waste really gets cleared is still in dispute. What is settled: CSF dynamics change during sleep, and the change is biggest in SWS-rich periods.

3. Memory consolidation. Diekelmann & Born 2010 Nat Rev Neurosci and Rasch & Born 2013 Physiological Reviews pulled together the active systems consolidation model. During SWS, hippocampal sharp-wave ripples replay the firing patterns from your waking day (Wilson & McNaughton 1994 Science) and lock to thalamocortical sleep spindles, which nest within cortical slow oscillations. This triple coupling broadcasts hippocampal memory traces to the neocortex over and over, slowly filing them into long-term storage. Lose SWS, and your recall of facts suffers. Closed-loop auditory stimulation can boost slow-wave activity and, in some studies, memory consolidation. The memory effect is fragile and does not always replicate.

The dementia link. Mander et al. 2017 Neuron pulled together evidence that age-related SWS loss is mechanistically tied to thinning of medial prefrontal grey matter, lower spindle density, and steadily worse memory consolidation. Sabia et al. 2021 Nature Communications (Whitehall II, n=7,959) found that habitually sleeping 6 hours or less at age 50 came with a 22 percent higher risk of later dementia. Sticking with short sleep across ages 50, 60, and 70 pushed that to about 30 percent. The mechanism plausibly runs through SWS loss leading to worse clearance plus worse consolidation.

What SWS does NOT do. SWS is not a stand-in for total sleep time. Eight hours of light sleep is not the same as four hours with extra-deep SWS. Every stage pulls its weight. The popular line of "chase deep sleep, ignore the rest" is a marketing shortcut. The right framing: for a given total sleep time, more SWS is better. But don't trade away total sleep duration to chase a higher SWS percentage.

SWS is the sleep stage that vanishes earliest and fastest. It is not subtle.

The Van Cauter trajectory. Van Cauter et al. 2000 JAMA studied 149 healthy men aged 16 to 83 with two nights of polysomnography. SWS fell from about 19 percent of the sleep period at ages 16 to 25 to about 3.4 percent at ages 36 to 50, losing roughly 38 minutes per decade through midlife. Then it leveled off. REM stayed fairly steady until midlife, then dropped about 10 minutes per decade after 50. The SWS collapse runs 25 years ahead of the REM collapse.

Night-time GH release fell about 75 percent from young adulthood to midlife, in step with the SWS loss. The evening cortisol nadir rose only after age 50. This is the most cited dataset on age and deep sleep in the field.

The Ohayon meta-analysis. Ohayon et al. 2004 Sleep pooled 65 studies covering ages 5 to 102 (n=3,577 healthy people). Same pattern: SWS drops steadily, with the sharpest fall between young adulthood and middle age. Women lose SWS more slowly than men.

Why does it happen? Mander et al. 2017 Neuron argues the loss is mechanistically tied to thinning of medial prefrontal grey matter (the brain region that generates the slow oscillations in the first place). Add lower spindle density (fewer fast spindles in N2), a circadian phase advance (going to bed earlier), and steady fragmentation, and you get the older-adult sleep profile: shorter total sleep, less N3, more N1, more awakenings.

Sex differences. Women generally hold onto more SWS than men across most of adulthood. Perimenopausal and postmenopausal women hit a distinct disruption: hot flashes and vasomotor symptoms fragment sleep, often worst in the first half of the night (which is SWS-rich). Falling estrogen changes sleep architecture on its own, separate from vasomotor symptoms.

What this means in practice. If you are 25, your SWS is plentiful and resilient. At 50, you are working with roughly a third of what you had at 25. Protecting what remains (through the levers below) gets more valuable the older you are. The flip side: the older you get, the harder it is to push SWS up a lot with any intervention. You can hold what you have or claw back modest amounts. You cannot recover what the cortex no longer generates.

Disease-related SWS loss. Several conditions speed up the age-related decline. Obstructive sleep apnea (repeated airway collapse during sleep) fragments sleep and blocks solid N3. Chronic insomnia and untreated depression cut SWS. Many drugs (Z-drugs, benzodiazepines, lipophilic beta-blockers, SSRIs) suppress N3 or distort its EEG signature. Treating these often brings a measurable SWS recovery.

If you cannot measure it, you cannot optimize it. But most consumer wearable "deep sleep" numbers are noisy. Here is what is real.

The gold standard. Polysomnography (PSG, the full in-lab sleep study with multi-lead EEG) scores N3 by AASM rules. Even between trained human scorers, window-by-window agreement is only about 83 percent (Cohen's kappa about 0.80; Arnal et al. 2020 Sleep). That is the ceiling. No algorithm can beat the inter-rater limit of PSG itself.

Wearable accuracy for deep sleep specifically. The Schyvens et al. 2025 SLEEP Advances validation against PSG (n=62) gives clean per-stage numbers. Deep sleep (N3) sensitivity:

• Apple Watch Series 8: ~51 percent
• Fitbit Sense / Charge 5: ~51 percent
• Whoop 4.0: ~70 percent (best for N3 among wrist devices in that cohort)
• Withings Scanwatch: ~67 percent
• Garmin Vivosmart 4: ~48 percent

Oura Gen3 with OSSA 2.0 is the best in class. Svensson et al. 2024 Sleep Medicine tested it against multi-night ambulatory PSG (421,045 epochs): overall accuracy 91.7 percent, Cohen's kappa (PABAK) 0.83, which matches human inter-rater agreement. N3 alone reached about 64 percent sensitivity. Funded by Oura Health.

How to read these numbers honestly. Consumer wearables guess "deep sleep" from heart rate variability patterns and movement, not from EEG. The guess lines up with PSG-measured N3 at the group level. It is noisy on any single night. One Oura or Whoop reading saying "only 30 minutes of deep sleep last night" could easily be 30 minutes off in either direction. Track 7-day rolling averages, not single nights. Use them for trends.

Where home EEG comes in. Muse S Athena (about $475) is the current best consumer option for real EEG-based sleep staging. It is a forehead 4-channel device with overall kappa about 0.76 in vendor-supported validations. N3 is harder to score from a forehead-only montage than from a full PSG hookup, but the device is a real EEG, not an HR/motion proxy.

Dreem 2/3 was the research-grade home EEG headband. Arnal et al. 2020 Sleep reported N3 kappa about 0.74, matching human inter-rater agreement. Dreem stopped consumer sales in 2021. The technology lives on under Beacon Biosignals for research. If you find a used Dreem 2 in good shape, it is still the best home EEG for measuring SWS.

Z-Machine Insight+ does sleep/wake reliably but is weaker at stage scoring than Dreem or Muse.

Closed-loop auditory stimulation (CLAS). Pink-noise pulses timed to slow-wave up-states can boost slow-wave activity in research settings (Ngo 2013; Papalambros 2017). The Philips SmartSleep DeepSleep Headband was the consumer device. Philips discontinued it in 2023. As of May 2026 there is no FDA-cleared, peer-reviewed consumer CLAS device on the market. Real CLAS needs live EEG phase detection, not just timed pink-noise loops. Anything sold as "AI-powered deep-sleep enhancement" without EEG and phase-locking is vapor.

Practical buying guide for SWS specifically:

• Best independent and currently buyable: Muse S Athena. Expect 75 to 85 percent accuracy at the stage level. Track trends.
• Best ring or wrist proxy: Oura Gen3 with OSSA 2.0 firmware. Use it for 7-day rolling averages, not single nights.
• If you find a used Dreem 2/3: the cleanest home N3 measurement going, but no current vendor support.
• Don't pay for "deep sleep enhancement" products without published peer-reviewed PSG validation.

The strongest SWS levers are behavioral, not pharmacological. Here are the ones with real evidence behind them.

Sleep restriction's odd SWS rebound. This is the most powerful lever, and the most counter-intuitive. The SWS system is homeostatic: staying awake longer builds up slow-wave "need," and when you finally sleep, SWS rebounds in both amount and delta-power density. It is the same principle that makes CBT-I's sleep restriction therapy work for chronic insomnia. You deliberately squeeze time in bed below your normal need, and high-efficiency, SWS-rich sleep follows. Borbely's Process S (the math model of sleep pressure building up while you are awake), set out in his 1982 paper, captures the dynamic: slow-wave activity per hour of sleep rises as sleep pressure builds.

For healthy adults without insomnia, a single night of 6-hour sleep gives you a noticeable SWS-rich recovery the next night. Chronic sleep restriction is not a strategy. It does not give you more SWS. It gives you a deeper-percentage SWS inside a smaller total budget, at the cost of REM and total sleep time. For people with chronic insomnia, structured sleep-restriction therapy (CBT-I) consolidates fragmented sleep and recovers SWS.

Aerobic exercise. Stutz et al. 2019 Sports Medicine meta-analysis of 23 evening-exercise studies found evening exercise raised SWS by +1.3 percentage points (p=0.041) versus control. Aerobic protocols showed the cleanest SWS signal in the wider meta-analytic literature (Kredlow 2015 J Behav Med). The mechanism is thermal: aerobic exercise raises core temperature 1 to 2 degrees C, and the overnight rebound deepens NREM. Hard resistance training also improves how you rate your sleep, but it does not move polysomnographic N3 as cleanly.

Practical: 3 to 4 sessions of moderate-vigorous aerobic exercise per week (30 to 60 min), ideally not ending within 60 minutes of bedtime. Evening exercise that finishes 1 hour or more before bed is a net positive for SWS.

Strict sleep timing. Fragmented or shifted sleep gives you fragmented SWS. Same bedtime and wake time every day, weekends included, anchors the circadian system and supports solid N3. Social jet lag (weekend sleep midpoint shifted an hour or more from weekdays) measurably hurts SWS the next weekday. Roenneberg's social-jet-lag work links it to metabolic disease. The SWS-fragmentation mechanism is plausible.

Stress and rumination management. Catecholamine and cortisol-driven sympathetic arousal (the stress-axis activation that keeps your nervous system in fight-or-flight) suppresses SWS. CBT-I components (stimulus control, cognitive restructuring) address this in chronic insomnia. For healthy adults, evening rituals that lower sympathetic tone (warm bath, slow paced breathing, dim light) support SWS by calming bedtime arousal.

Treat what is treatable. Untreated obstructive sleep apnea is the most common cause of severe SWS loss in adults. Apnea events trigger micro-arousals that abort SWS. Effective CPAP often brings a dramatic SWS rebound in the first weeks of treatment. Restless legs syndrome fragments deep sleep the same way. Iron repletion (ferritin below 75 ng/mL) and alpha-2-delta ligands restore it. If you have moderate-to-severe OSA you don't know about, no behavioral or supplement step in this guide will recover your SWS until the apnea is treated.

Temperature is the most evidence-backed non-drug lever for SWS that most people aren't using.

The mechanism. Falling asleep is gated by a dropping core body temperature, which you reach by actively shedding heat through your hands and feet dumping warmth into the air (distal vasodilation). Kräuchi et al. 2000 Am J Physiol showed the distal-proximal skin temperature gradient (DPG, how much warmer your hands and feet are than your torso) is the single best predictor of how fast you fall asleep. Better than core body temperature, better than melatonin, better than how sleepy you say you feel. Heat loss keeps going across the night, with the core minimum around 04:00 to 05:00.

SWS responds to skin-temperature tweaks specifically. Raymann et al. 2008 Brain used a thermosuit to shift skin temperature in 0.4 degrees C steps during polysomnography in older adults. Distal-warming plus proximal-warming shifts of just 0.4 degrees C doubled SWS from 8 to 14 percent and cut night-time awakenings (p<0.001). This is one of the most striking single-intervention SWS effects in the human polysomnography literature. The thermoregulatory window for SWS is genuinely narrow, and skin temperature is a high-precision lever.

Cool bedroom plus warm extremities: the best combination. Okamoto-Mizuno & Mizuno 2012 J Physiol Anthropol reviewed the thermal sleep literature: 16 to 19 degrees C (60 to 67 degrees F) under typical bedding is the consensus range for adults. Above 26 degrees C with humidity, both SWS and REM fall off. The right protocol is to keep the room air cool (so your skin has somewhere to dump heat from the periphery) while keeping your extremities themselves warm (warm bath pre-bed, socks if you run cold).

Hot bath protocol. Haghayegh et al. 2019 Sleep Medicine Reviews meta-analyzed 13 trials of pre-sleep passive body heating. Water at 40 to 42.5 degrees C for 10+ minutes, 1 to 2 hours before bed, cut sleep onset latency by 36 percent on average (Cohen's d about 1.01). The mechanism: warming the periphery triggers reflexive vasodilation; stepping out into a cool room dumps heat fast; core temperature falls over the next 60 to 90 minutes. The drop itself is the sleep signal. SWS effects in the meta-analysis trend positive but were not the main endpoint in most of the included studies.

Cooling mattresses and high-heat-capacity bedding. Herberger et al. 2024 Scientific Reports, an independent three-center crossover (n=72), tested a high-heat-capacity mattress against a standard one and found +7.5 minutes of N3 over a 7.5-hour night (p=0.0038) and a lower heart rate (-2.4 bpm). Modest but real, independently funded.

Manufacturer-funded studies of Eight Sleep Pod (Baron et al. 2024 Bioengineering) report bigger effects (+14 minutes deep sleep) but carry serious conflicts of interest and methodology limits. The honest synthesis: a high-heat-capacity cooling cover plausibly adds 5 to 15 minutes of N3 over a normal night for people who otherwise sleep warm. The marketing numbers from Eight Sleep and chiliPad run ahead of what independent data support.

Sauna. Direct sleep RCT evidence is thin. The Putkonen & Elomaa 1976 Finnish study (n=5) reported +70 percent SWS in the first 2 hours after an evening sauna. That has not been replicated in a modern RCT. Mechanistically, sauna does what a hot bath does: raises core temperature about 1 to 1.5 degrees C and forces a strong cooling rebound. The cardiovascular benefit is well-documented (Laukkanen Finnish cohort, BMC Medicine 2018). Treat sauna as a hot-bath analog with stronger heart upside and weaker direct sleep evidence.

Practical SWS-targeted thermal protocol:

• Bedroom air: 17 to 20 degrees C (62 to 68 degrees F). Cool enough that you'd want a duvet.
• Warm extremities: warm bath or shower 60 to 120 minutes pre-bed, or wear bed socks if your feet run cold.
• Pre-bed hot bath: 40 to 43 degrees C, 10 to 15 minutes, 60 to 120 minutes before lights-out, then into the cool dark bedroom.
• Optional: high-heat-capacity mattress topper or cooling cover. Expect about 5 to 15 min more N3 over baseline, not the marketing figure.

Most "sleep supplements" do not deepen SWS. They cut the time it takes to fall asleep, lower anxiety, improve how you rate your sleep, or shift your circadian timing. Those are real benefits. They are not the same as raising N3 on polysomnography. This is the central dishonesty in the consumer sleep-supplement category.

The framework. Sort sleep supplements by what they actually do, based on PSG-grade human evidence.

Tier 1: Closest to a real SWS lever (and even here, the evidence is thinner than people claim):

• Magnesium bisglycinate plus glycine (combined or separate). Glycine at 3 g has Yamadera 2007 PSG evidence of faster latency to SWS, with no change in SWS amount. Bisglycinate delivers magnesium plus glycine. The glycine load depends on dose. See next section.
• Lemon verbena extract plus zinc, the only consumer combination with any human trial measuring something called "deep sleep." The measurement is Fitbit, not PSG. See its own section.

Tier 2: Onset and subjective quality, not SWS depth:

• L-theanine 200 to 400 mg: alpha-wave promotion, anxiolysis (calming without sedation), faster onset. No PSG evidence for SWS deepening.
• Saffron extract (affron) 14 to 28 mg/day for 4 to 6 weeks: better subjective sleep quality and mood. The effect builds over weeks. No PSG SWS evidence.
• Melatonin low-dose (0.3 to 0.5 mg) timed to DLMO (your dim-light melatonin onset, when your own melatonin starts rising): a phase-shifter, useful for jet lag and DSWPD. Faster onset by about 7 minutes in meta-analysis. NOT a depth enhancer. High doses (5 to 10 mg) are not more effective and leave you groggy in the morning.

Tier 3: Mechanism only, no human sleep RCT:

• Taurine on its own: the receptor mechanism is plausible, but no primary-endpoint human sleep RCT exists.
• Most multi-ingredient "sleep formulas": 8 to 12 ingredients at doses too low to work; effectively expensive multivitamins.

Tier 4: Actively counterproductive for SWS:

• Diphenhydramine, doxylamine (OTC anti-histamine sleep aids): anticholinergic load, dementia risk in older adults (AGS Beers criteria recommend against use in adults 65 and over), tolerance builds within days.
• Z-drugs and benzodiazepines: suppress N3 amount and alter slow-wave-activity spectral traits; tolerance, dependence, falls in older adults, observational mortality signals.
• Alcohol: suppresses first-half N3 and produces second-half fragmentation (Ebrahim 2013 meta).
• High-dose melatonin (5 to 10 mg): no better than 0.5 to 1 mg, brings a hangover and vivid dreams.

The big honest claim. No oral supplement has been shown in published polysomnography RCTs to substantially deepen SWS in healthy adults. Glycine slightly shortens latency to N3 without changing the amount. Magnesium in deficient populations shortens onset latency. Lemon-verbena trials use Fitbit, not EEG. Melatonin doesn't change N3 in any meaningful way. L-theanine works on alpha waves and anxiolysis, not slow waves.

If a supplement product promises to "increase your deep sleep," ask the seller for the polysomnography study. The answer is almost always that the measurement was wearable-derived (Oura, Fitbit, Eight Sleep Pod) or a subjective questionnaire (PSQI, LSEQ). Those measurements aren't invalid. They are just not the same as PSG-measured N3 amount or delta-power density.

The right place to deepen your sleep is behavioral and environmental, through the levers in the previous two sections. Supplements support onset and quality, which are real and worthwhile targets. They do not, in current human trial evidence, substantially raise SWS depth.

This is the supplement combination with the most defensible (if still modest) SWS-supportive evidence.

Glycine has the cleaner story. Yamadera et al. 2007 Sleep & Biological Rhythms, a small RCT in volunteers with chronic poor sleep, gave 3 g of glycine 30 minutes before bed. Polysomnography showed shorter latency to sleep onset AND shorter latency to SWS versus placebo, plus better subjective sleep quality and less daytime sleepiness. Important detail: sleep architecture did not change. Glycine got you to N3 faster. It did not increase the amount of N3.

Bannai et al. 2012 Frontiers in Neurology took this further: 3 g glycine plus 25 percent sleep restriction over three nights cut VAS fatigue (p=0.022) and improved PVT reaction time. Kawai et al. 2015 Neuropsychopharmacology mapped the mechanism: glycine acts through NMDA receptor co-agonism in the suprachiasmatic nucleus shell (the brain's master circadian clock; not glycine receptors, because strychnine didn't block the effect but AP5 did), triggering skin vasodilation in the periphery and a core body temperature drop that gates faster NREM onset. SCN ablation wiped out the effect in animals.

The Yamadera trial was funded by Ajinomoto (a glycine maker) and the N was small. Caveat the industry funding. The mechanism is biochemically clean and reproducible. Effective dose: 3 g, 30 minutes before bed. Safety: average dietary plus endogenous glycine intake tops 11 g/day in adults (Norwegian VKM 2016 risk assessment), so the 3 g supplemental dose sits well inside normal physiological exposure. Studied up to 9 g/day in healthy adults without issue.

Magnesium has a thinner story. Mah & Pitre 2021 BMC Complementary Medicine and Therapies meta-analyzed three RCTs (n=151) of oral magnesium in older adults with insomnia: sleep onset latency cut by 17.4 minutes (95% CI -27.27 to -7.44, p=0.0006); total sleep time +16 min, not significant. A single included trial reported about a 6-minute net SWS increase versus placebo, within measurement noise, not pooled, not significant on its own. The GRADE evidence quality across all outcomes was low to very-low. The authors' verdict: "the quality of literature is substandard for physicians to make well-informed recommendations."

Abbasi et al. 2012 J Res Med Sci, the most-cited magnesium-insomnia trial, randomized 46 elderly insomniacs to 500 mg elemental Mg (as oxide) twice daily or placebo for 8 weeks. Subjective sleep measures improved (Insomnia Severity Index p=0.006, sleep onset latency p=0.02). Biomarkers shifted favorably (melatonin up, cortisol down). No PSG was done. The population was magnesium-insufficient at baseline, so you cannot generalize to magnesium-replete healthy adults.

The 'magnesium bisglycinate = two active molecules' framing. Elegant in theory, partial in the numbers. Magnesium bisglycinate (Mg(C2H4NO2)2, MW about 172) is about 14 percent magnesium and about 86 percent glycine by mass. At a 400 mg elemental magnesium dose, you take in about 2,860 mg of total bisglycinate, which is roughly 2.5 g of glycine. That is 83 percent of the studied 3 g glycine sleep dose.

At a more typical 200 mg elemental Mg dose, you get only about 1.24 g of glycine, under half the studied dose. The framing only fully works at the high-dose end. If your magnesium-bisglycinate label says "1,000 mg providing 140 mg elemental magnesium," the glycine load is about 860 mg, far below the Yamadera dose.

Pragmatic conclusion for SWS-targeted use:

• Take magnesium bisglycinate 200 to 400 mg elemental in the evening (gentle on the gut, decent absorption, EU UL 250 mg supplemental; at 400 mg you are above the EU UL and approaching the NIH UL of 350 mg).
• Dose glycine separately at 3 g about 30 minutes before bed if you specifically want the glycine onset effect.
• Don't assume the bisglycinate carrier covers the glycine sleep dose at typical labeled magnesium amounts.

Magnesium L-threonate (Magtein). Marketed as superior for sleep on the strength of one rat study showing better brain magnesium uptake (Slutsky 2010 Neuron) and one industry-funded Oura-ring human trial (Hausenblas 2024 Sleep Medicine X, AIDP-funded). Premium pricing, no PSG evidence, a premium that independent data do not match. Skip.

Magnesium taurate is a defined Mg salt (about 9 percent elemental Mg, bound to taurine). Both molecules have plausible CNS effects (taurine = partial GABA-A plus glycine receptor agonist; magnesium = NMDA antagonist). For a heart-leaning user it is reasonable; for SWS-targeted use, bisglycinate is the cleaner choice.

Honest evidence summary:

• Glycine 3 g shortens latency to SWS, doesn't increase the amount: Moderate.
• Magnesium cuts sleep onset latency in deficient/elderly populations: Moderate.
• Magnesium raises polysomnographic N3 amount in replete healthy adults: Weak / Unproven.
• Bisglycinate co-delivers therapeutic glycine at the user's daily Mg dose: Only at the upper end (400 mg elemental Mg).

The lemon-verbena-plus-zinc combination is the only consumer formulation marketed specifically for "deep sleep" that points to any human-trial data with a stage-classification endpoint. The honest reading: that data is Fitbit-derived, not polysomnography, and the combination itself has never been tested as a single intervention in a published peer-reviewed PSG RCT.

What has actually been sold. The lemon verbena extract behind almost all the consumer "deep sleep" claims is RelaxPLX (also marketed as PLX) from Monteloeder S.L. (Alicante, Spain), a subsidiary of SUANNUTRA. Standardized to 28 percent or more total phenylpropanoids and 24 percent or more verbascoside. Clinical dose 400 mg/day. It is not the Frutarom/IFF or PLT Health Solutions ingredient. Those companies sell different lemon-verbena positions. Robuvit (Horphag) is unrelated. It is oak wood, not lemon verbena.

The two key human trials of the lemon-verbena extract alone:

Martinez-Rodriguez et al. 2022 Nutrients. RelaxPLX 400 mg/day, no zinc, n=40, 8 weeks. Outcomes: PSQI, perceived stress, blood cortisol, Fitbit Charge 2. No PSG. PSQI dropped 12.2 percent at 8 weeks (p<0.05); cortisol -15.6 percent; perceived stress -10.7 percent. The widely-cited "increase in deep sleep" is Fitbit-derived "deep sleep" from consumer accelerometry and heart rate, not polysomnographic N3. Fitbit's stage-classification accuracy for N3 runs roughly 40 to 60 percent sensitivity versus PSG in independent validations. Industry-funded (Monteloeder sponsor).

Perez-Pinero et al. 2024 Nutrients. Aloysia citrodora extract, 90 days, n=71. Outcomes: PSQI, VAS, actigraphy, nocturnal melatonin. No PSG. Subjective measures improved; nocturnal melatonin +14 percent. Three authors are Monteloeder employees.

Zinc-only sleep evidence. Cherasse & Urade 2017 Int J Mol Sci review covers zinc's NMDA modulation, glycine receptor potentiation, and GPR39 zinc-sensing receptor. Cherasse et al. 2015 Mol Nutr Food Res showed zinc-yeast extract raised NREM in mice by 20 to 30 minutes per hour for about 6 hours post-dose, without changing EEG spectral power (physiological-looking sleep, not benzodiazepine-like). Mice, not humans. Human evidence for zinc alone in sleep is limited to subjective sleep quality in elderly subjects on multi-ingredient formulations.

The combination itself. A targeted literature search returns no peer-reviewed RCT testing the lemon-verbena-plus-zinc combination as a single intervention against placebo with PSG endpoints. The consumer products that combine RelaxPLX with zinc bisglycinate (and often other botanicals) have not been published as a tested formulation.

Names sometimes mentioned in marketing (Buguet, Carrillo-Vico, Funes, Marhuenda) show up in the wider nutraceutical or sleep literature but not on a published lemon-verbena-plus-zinc PSG SWS trial. If a sponsor cites those names for slow-wave-sleep evidence on this combination, the citation is misleading.

Mouse PSG for lemon verbena alone. Choi et al. 2025 Int J Mol Sci tested lemon verbena extract in C57BL/6 mice at 40 to 160 mg/kg (allometrically about 230 to 900 mg human equivalent, overlapping the 400 mg RelaxPLX trial dose). NREM rose 110 to 194 percent; REM +73 percent. Mechanism: upregulated adenosine A1 receptor mRNA and altered GABA-A subunit composition (more alpha2, more beta2, less gamma2, distinct from benzodiazepines). Real mouse PSG. Not the same as a human PSG trial.

Honest verdict. The lemon-verbena-plus-zinc combination is a mechanistically reasonable, low-risk, modestly-studied subjective-sleep stack with rodent PSG backing for the lemon-verbena half. It is NOT a validated "slow-wave sleep booster." Calling it "the only combination on the list with direct slow-wave sleep human data" overstates the evidence by at least one tier.

Where it sits in the hierarchy: roughly level with magnesium glycinate, ashwagandha, or saffron for sleep quality. Plausible mechanism, modest subjective improvement, industry-funded human trials, no PSG SWS confirmation. Worth a look as part of a broader stack if you are targeting onset and quality. Not worth paying premium for "deep sleep enhancement."

Safety. Zinc EFSA UL is 25 mg/day total (NIH UL 40 mg). Long-term supplementation above 25 mg risks copper deficiency, sideroblastic anemia, neutropenia, and neurological symptoms. Keep total daily zinc (food plus supplement) at or below 25 mg unless tested-deficient. Lemon verbena (per EMA monograph) is a traditional herbal medicinal product with no significant safety signals at trial doses. Avoid in pregnancy and breastfeeding.

The recommended editorial framing. "Lemon verbena standardized to verbascoside, combined with a low dose of zinc, has a plausible biological story: verbascoside binds GABA-A and modulates adenosine A1; zinc modulates NMDA and glycine receptors. Two small industry-funded trials of the lemon verbena extract alone show modest improvements in subjective sleep quality and cortisol. There is no human polysomnography study of this combination, and the 'deep sleep' figure widely cited from the 2022 trial is a Fitbit-derived estimate, not slow-wave sleep measured on EEG. Mouse PSG data are real but rodent. Reasonable low-risk option, not a validated SWS booster."

These three are the most common ingredients in consumer "sleep formulas." They work on falling asleep, anxiety, and subjective quality. They do not work on N3 depth. The distinction matters.

L-theanine (200 to 400 mg, 30 to 60 minutes before bed). Nobre et al. 2008 Asia Pac J Clin Nutr established the mechanism: a single 50 mg dose of L-theanine raised frontal alpha-wave (8 to 12 Hz) activity (the EEG signature of relaxed wakefulness) on EEG. Alpha is the relaxed-but-awake signature, not a deep-sleep one.

Hidese et al. 2019 Nutrients: n=30, crossover, 200 mg/day for 4 weeks in healthy stressed adults. PSQI total improved (p=0.013); latency, disturbance, and medication use all improved. Trait anxiety p=0.006. Subjective measures only, no PSG. Lyon et al. 2011 Altern Med Rev: 98 boys aged 8 to 12 with ADHD, 400 mg/day for 6 weeks. Actigraphy showed better sleep percentage and efficiency. Sleep latency did not change. Williams et al. 2020 Plant Foods Hum Nutr systematic review of 9 RCTs concluded 200 to 400 mg/day cuts stress and anxiety under stressful conditions, which is the proximal mechanism for any downstream sleep benefit.

The honest framing: L-theanine is anxiolytic. It promotes alpha waves on EEG. It improves subjective sleep quality. No published L-theanine RCT has measured PSG-defined SWS or delta-power density and shown deepening. Best target group: people whose insomnia is driven by rumination and bedtime arousal. No EU health claim.

Saffron extract (affron, Crocus sativus stigma extract). The mechanism is serotonergic. Crocin and safranal show SSRI-like activity in preclinical work. The antidepressant signal is the strongest clinical finding for saffron as a class. The sleep benefit is plausibly downstream of better mood and lower anxiety rather than a direct hypnotic effect.

Lopresti et al. 2020 J Clin Sleep Med: n=55 adults with self-reported poor sleep, 28 mg/day affron for 4 weeks. Insomnia Severity Index improved 15.75 to 11.74 (Cohen's d=1.07) versus placebo 14.74 to 13.46 (d=0.31, group-by-time p=0.017). Restorative Sleep Questionnaire d=0.72 versus 0.10. Funded by Pharmactive Biotech (the maker of affron). Pachikian et al. 2021 Nutrients: n=66, 15.5 mg/day saffron extract for 6 weeks. Actigraphy and PSQI improved. Funded by Comercial Quimica Masso.

All positive saffron sleep trials are industry-funded and use subjective scales or actigraphy. The effect builds over 4 to 6 weeks. No PSG SWS deepening evidence. Best target group: subjective sleep quality where mood is a contributor. No EU health claim.

Taurine as a standalone sleep aid. The mechanism is preclinical: partial GABA-A agonist, glycine receptor agonist. A targeted literature search returns no RCT of standalone taurine supplementation with primary sleep endpoints in humans. Existing "taurine + sleep" data involve caffeine-plus-taurine energy-drink studies on simulated tasks, or hepatology/cardiology trials that did not measure sleep architecture. Taurine has a clean mitochondrial story (see our mitochondria guide) and a phase III RCT in MELAS. The sleep case is mechanistic only.

If a multi-ingredient "sleep formula" includes taurine alongside glycine and magnesium, you cannot pin the benefit on taurine specifically. Calling taurine a "deep sleep" ingredient extrapolates from receptor pharmacology and animal data, not from human PSG evidence. No EU health claim.

Melatonin is explicitly NOT a depth supplement. Worth saying loudly, because the assumption is everywhere. Melatonin is a chronobiotic: it signals darkness and shifts circadian phase through MT1 and MT2 receptors. The Ferracioli-Oda 2013 PLoS ONE meta-analysis (19 studies, n=1,683): sleep onset latency cut by about 7 minutes; total sleep time +8 minutes; subjective sleep quality SMD 0.22. No PSG evidence for an increase in N3 amount or delta-power density. The Comai 2024 J Pineal Res review reinforces that melatonin receptor pharmacology is chronobiotic, not depth-modifying.

High doses (5 to 10 mg) are no more effective for sleep quality (Ferracioli-Oda shows no dose-response for quality). They produce supraphysiologic plasma levels that linger into the morning, driving grogginess and vivid dreams. The 10 mg gummies common in US retail are pharmacologically excessive.

The real value of melatonin is narrow: jet lag (especially eastward), delayed sleep-wake phase disorder, shift work, and blind circadian rhythm disorders. The EU has two authorised claims (jet lag at 0.5 mg or more; reduced sleep onset time at 1 mg). In Germany, products above 1 mg drift toward medicinal-product classification (BfR 17 September 2024 statement; OLG Koblenz 9 U 1947/22 May 2023 softened 1 mg or below). EMA's Circadin (2 mg prolonged release) is approved for primary insomnia in adults 55 and over.

The pattern. None of L-theanine, saffron, taurine, or melatonin has PSG evidence for deepening SWS in humans. They help onset, quality, mood, and circadian alignment. All real and worthwhile, but distinct from deepening N3.

Some substances actively suppress SWS. If you want to protect deep sleep, these are the first things to remove.

Alcohol. Ebrahim et al. 2013 Alcoholism: Clinical and Experimental Research meta-review of decades of polysomnography studies: alcohol shortens sleep onset latency and consolidates the first half of the night with deeper NREM at high doses. In the second half, sleep gets fragmented and arousals multiply as blood alcohol falls. REM is suppressed in the first half at moderate-to-high doses and rebounds vivid and fragmented in the second half. Total-night REM percentage drops across most studies. SWS in the first half often rises acutely (the sedative effect), but the net architectural cost is high. Fragmentation in the second half, lower next-morning HRV, sympathetic activation for 24 to 48 hours after even one or two drinks. Obstructive sleep apnea gets worse with alcohol via reduced upper-airway dilator muscle tone.

There is no "safe" alcohol dose for SWS protection. If you drink, finish at least 3 hours before bed, stay within about 10 g (one drink) on a sleep night, and avoid alcohol entirely if you have diagnosed OSA. Wearable HRV trackers will show the effect for days.

Z-drugs and benzodiazepines (zolpidem, zopiclone, eszopiclone, zaleplon, classic benzodiazepines). These positive allosteric modulators at GABA-A receptors cut sleep onset latency and raise total sleep time. They also suppress SWS in many polysomnography studies and alter slow-wave-activity spectral traits. Tolerance to the sedative effect builds within weeks. Next-day cognitive impairment, falls, and complex sleep behaviors (sleepwalking, sleep-driving) prompted FDA boxed warnings on Z-drugs in 2019. The Kripke et al. 2012 BMJ Open observational cohort reported all-cause mortality hazard ratios of 3.6 to 5.3 across rising dose tertiles versus matched non-users. The design is observational and confounding-vulnerable, but the signal has not been retired.

If you are on a Z-drug or benzodiazepine and want to taper, do it under medical supervision over 4 to 8+ weeks. Rebound insomnia is severe. Replace with CBT-I in parallel.

Dual orexin receptor antagonists (DORAs) are the alternative. Suvorexant, lemborexant, and daridorexant block orexin-driven wakefulness (orexin is the brain's wake-promoting neuropeptide system) rather than amplifying GABA. The Mignot et al. 2022 Lancet Neurology phase 3 trials of daridorexant (1,854 patients, two trials) showed efficacy on objective WASO and latency at 1 and 3 months. A pooled post-hoc architecture analysis (Hudgens et al. 2024 Sleep) confirmed preserved REM and SWS percentage versus baseline, unlike Z-drugs. If pharmacotherapy is needed for chronic insomnia, DORAs are the evidence-favored class. Daridorexant is NHS-recommended in the UK (NICE TA922, 2023) after CBT-I.

High-dose melatonin (5 to 10 mg). As covered earlier: pharmacologically excessive, no more effective than 0.3 to 1 mg for sleep quality, produces morning grogginess and vivid dreams. In Germany, products above 1 mg approach medicinal-product classification.

Diphenhydramine and doxylamine (OTC "sleep aids"). Anti-histamine sedatives. Anticholinergic load. Tolerance builds within a few nights. The AGS Beers Criteria recommend against use in adults 65 and over due to fall risk, confusion, and a link to dementia in longitudinal cohort data (Gray et al. 2015 JAMA Internal Medicine: the strongest anticholinergic exposure tertile carries elevated dementia risk over 10-year follow-up). The cost is high. The SWS benefit is zero.

Late nicotine and late caffeine. Both cut N3 amount. Caffeine at 6 hours pre-bed cuts total sleep time by over an hour (Drake 2013). Nicotine fragments and shortens deep sleep. Cut both hours before bed.

Nightly cannabis use. Acute THC cuts sleep onset latency and may briefly raise N3 at low doses. Tolerance to the sleep-promoting effects builds within weeks, and withdrawal causes rebound insomnia and vivid dreams (REM rebound). The acute THC/CBD trial by Suni et al. 2023 Sleep reported a significant 8 percent drop in REM and a 66-minute REM-latency increase. Nightly THC for sleep is a tolerance trap. CBD alone, at anxiolytic doses, does not acutely disrupt sleep architecture in healthy adults (Linares et al. 2018 Front Pharmacol).

Late evening meals, especially large or carb-heavy ones, blunt the night-time core-temperature drop that gates SWS and raise overnight glucose spikes. Lopes et al. 2019 J Clin Sleep Med in OSA patients: late dinner timing raised AHI (beta=1.28 events/h) and cut REM duration. Eat 3 hours or more before bed, especially if you have or suspect OSA.

The honest hierarchy of SWS protection. Remove alcohol, treat OSA, taper Z-drugs/benzodiazepines, cut late caffeine and nicotine, eat earlier, drop anticholinergic sleep aids, use melatonin only at low doses for circadian indications. None of those are supplements. They are subtractions. SWS-protective subtractions do more for you than supplement additions.

A 30-day deep-sleep protocol, ordered by evidence-to-effort ratio. Pick the top of the list first.

Days 1-7: Measure baseline and remove SWS-suppressors.

• Start a 7-day sleep diary. Track bedtime, wake time, alcohol, caffeine cutoff time, perceived quality 1 to 5.
• If you have a wearable, log nightly: total sleep time, sleep efficiency, deep-sleep estimate.
• Cut alcohol within 3 hours of bedtime. If you can stop drinking entirely for the 30-day protocol, do.
• Cut caffeine after noon if you are unsure of your CYP1A2 phenotype (the liver enzyme that decides whether you are a fast or slow caffeine metabolizer).
• Move the last meal to 3 hours or more before bed.
• Drop diphenhydramine, doxylamine, and OTC anti-histamine sleep aids.
• If you take Z-drugs or benzodiazepines nightly, talk to a clinician about tapering. Not for this protocol, but as a longer-term project.

Days 8-14: Add the thermal and light protocol.

• Bedroom temperature 17 to 20 degrees C (62 to 68 degrees F). Cooler than you probably keep it. Get the room genuinely cool. Use a duvet for warmth.
• Hot bath 40 to 43 degrees C for 10 to 15 minutes, 60 to 120 minutes before lights-out. This is the Haghayegh 2019 protocol: about 36 percent reduction in sleep onset latency on average, via distal vasodilation triggering a core-temperature drop after you exit. Even a hot shower works at a smaller effect size.
• If your feet run cold, wear bed socks. Warm extremities are the signal that sleep is coming.
• Morning outdoor light within 30 to 60 minutes of waking, 10 to 30 minutes, no sunglasses. 1,000+ lux at the eye minimum. Cloudy days still work. This anchors the circadian system.
• Evening dim light, 10 melanopic EDI or below, in the 3 hours before bed. Dim the room overall. Phones face-down or in another room, ideally.

Days 15-21: Add aerobic exercise.

• 3 to 4 sessions of moderate-vigorous aerobic exercise per week, 30 to 60 minutes per session, RPE 5 to 6 ("can talk but not sing"). Zone 2 is the practical target.
• Add 1 to 2 HIIT sessions per week (e.g. 10 x 1 minute at about 90 percent HRmax with 1 minute recovery). HIIT delivers fast mitochondrial signals and contributes to SWS.
• Finish all training 60 minutes or more before bed. Within 60 minutes, hard exercise delays onset. Outside that window it is a net positive for SWS (Stutz 2019).
• Morning aerobic exercise also pulls circadian phase forward about 0.6 h per session.

Days 22-28: Add supplements (only if the rest is in place). These add modest benefit. They do not substitute for the prior steps.

• Magnesium bisglycinate 200 to 400 mg elemental (typical evening dose; EFSA UL 250 mg supplemental, NIH UL 350 mg). At 400 mg you are approaching the NIH UL. Don't go higher without medical advice.
• Glycine 3 g about 30 minutes before bed (the dose used in the Yamadera trial). Sweet taste, mixes well with water or tea. Avoid combining with clozapine.
• Optional L-theanine 200 mg. Some users add this if bedtime rumination is the main issue.
• Avoid: high-dose melatonin (over 1 mg) unless you have a circadian indication (jet lag, DSWPD).

Talk to your doctor before starting any of these, especially if you take medication.

Days 29-30: Reassess.

• Compare week 1 (baseline) to week 4 sleep diary and wearable trends. Look at 7-day rolling averages, not single nights.
• Sleep efficiency should be 85 percent or higher.
• Subjective sleep quality should have improved.
• If your wearable shows more deep sleep, treat it as a directional trend, not a precise measurement.
• If you have made no progress, the issue is probably medical. Screen for OSA with STOP-BANG. If 3 points or more, get a home sleep apnea test.

When to see a sleep clinician.

• Loud snoring plus witnessed apneas plus daytime sleepiness, even with this protocol in place, means undiagnosed OSA. Get tested.
• Insomnia lasting 3 months or more despite the protocol: do structured CBT-I (somnio via DiGA in Germany, Sleepio in the UK, or a behavioral sleep medicine clinician).
• Acting out dreams: see a neurologist for REM sleep behavior disorder evaluation. This is the single most urgent referral in the sleep space.
• Restless legs syndrome pattern: check ferritin and transferrin saturation. The AASM 2024 guideline favors IV iron at ferritin below 75.

The principle. Slow-wave sleep is what falls fastest with age. Most of what you can do to protect it is subtractive: don't drink alcohol within 3 hours of bed, don't take supplements that suppress SWS, don't keep the bedroom warm, don't soak yourself in bright evening light, treat sleep apnea aggressively if you have it. The additive steps (cold room, warm bath, aerobic exercise, glycine, magnesium bisglycinate) deliver smaller, real benefits on top of the subtractive foundation. Anyone selling you a product as a one-step solution for deep sleep is overselling the additive side while ignoring the subtractive one.