5D Library › HPA-Axis Coupled Oscillations, Cortisol Awakening Response Flattening and Burnout Recovery Kinetics · 中文
Chapter 6. The Variable That Moves in Days
How bright, and when
Of everything in this book only one quantity can be moved quickly, and that is the phase of the daily rhythm. Light moves it, and the two things that decide how much are how bright the light is and where in the phase-response curve it falls. Both are measured quantities, and both are routinely misstated.
Zeitzer and colleagues measured the dose-response of the human pacemaker to light and found it strikingly sensitive at low levels, with a substantial fraction of the maximum phase shift produced by ordinary indoor illumination and saturation well below the levels light boxes provide (Zeitzer et al. 2000). Khalsa and colleagues mapped the phase-response curve to single bright pulses, establishing that the same light advances or delays the clock depending on when it arrives relative to the body’s own temperature minimum (Khalsa et al. 2003).
Evidence
Published. Zeitzer and colleagues measured the sensitivity of the human circadian pacemaker to light and found half of the maximum phase-resetting response produced by illumination of the order of ordinary room light (Zeitzer et al. 2000). Khalsa and colleagues established a phase-response curve to single bright light pulses in humans, showing the sign of the shift depends on timing relative to the temperature minimum (Khalsa et al. 2003).
Calculation
# Chapter 6. p11: Pulling the clock back with light. A phase that has
# drifted late is corrected by light in the morning, which advances the
# clock, and made worse by light in the evening, which delays it. The
# size of a day's shift is the product of three things: how bright the
# light is, how long it lasts, and where in the phase-response curve it
# falls. The brightness term saturates, with half the maximum effect
# near the level below, which is the fact the marketing of very bright
# lamps does not mention. The response curve is taken as a sine that
# crosses zero at the body's temperature minimum. The first result is
# not the one the protocol expects: a person whose clock is three hours
# late has a temperature minimum at eight in the morning, so light at
# seven falls before it and delays them further. Light has to be timed
# against the person's own minimum, not against the clock, and the hour
# has to move earlier as the phase comes back.
import math
AMAX, I_HALF, I_POW = 1.30, 120.0, 0.5 # hours a day at saturation
D_HALF = 20.0 # minutes for half the effect
TMIN = 5.0 # temperature minimum, hours
START_ERROR = 3.0 # hours late at the start
PI2 = 6.283185307179586
SETTINGS = (
("10,000 lux, 30 min", 10000.0, 30.0),
("5,000 lux, 30 min", 5000.0, 30.0),
("1,000 lux, 30 min", 1000.0, 30.0),
("300 lux, 30 min", 300.0, 30.0),
("300 lux, 120 min", 300.0, 120.0),
("100 lux, 30 min", 100.0, 30.0),
)
LIGHT_AT = 7.0 # the fixed clock hour the usual protocol names
OFFSET = 2.0 # hours after the person's own minimum, adaptive
def brightness(lux):
return lux ** I_POW / (I_HALF ** I_POW + lux ** I_POW)
def duration(minutes):
return minutes / (D_HALF + minutes)
def prc(hours_after_tmin):
# positive means an advance; light just after the minimum advances
return math.sin(PI2 * (hours_after_tmin % 24.0) / 24.0)
def shift_fixed(lux, minutes, error):
# light at a fixed clock hour, against a minimum that has moved late
return (AMAX * brightness(lux) * duration(minutes)
* prc(LIGHT_AT - (TMIN + error)))
def shift_adaptive(lux, minutes, error):
# light placed OFFSET hours after the person's own minimum each day
return AMAX * brightness(lux) * duration(minutes) * prc(OFFSET)
if __name__ == "__main__":
print(" half the maximum effect is reached at {:.0f} lux"
.format(I_HALF))
print(" setting brightness duration fixed adaptive")
for label, lux, mins in SETTINGS:
print(" {:<22}{:<13.4f}{:<11.4f}{:<7.3f}{:.3f}"
.format(label, brightness(lux), duration(mins),
shift_fixed(lux, mins, START_ERROR),
shift_adaptive(lux, mins, START_ERROR)))
print(" a fixed hour of seven makes a {:.1f} hour delay worse,"
.format(START_ERROR))
print(" because the minimum has moved to {:.0f} and the light now"
.format(TMIN + START_ERROR))
print(" falls before it rather than after it")
print(" days to bring the delay under 0.5 hours, light placed")
print(" {:.1f} hours after the person's own minimum each day:"
.format(OFFSET))
for label, lux, mins in SETTINGS:
err, days = START_ERROR, 0
while err > 0.5 and days < 200:
s = shift_adaptive(lux, mins, err)
if s <= 0.0:
break
err -= s
days += 1
print(" {:<22}{}".format(label, days if err <= 0.5
else "not within 200 days"))
a = brightness(10000.0) * duration(30.0)
b = brightness(300.0) * duration(30.0)
c = brightness(300.0) * duration(120.0)
print(" the very bright lamp is worth {:.4f} times the room light"
.format(a / b))
print(" at the same thirty minutes, but only {:.4f} times the same"
.format(a / c))
print(" room light left on for two hours")
print(" so brightness buys less than the lamps claim and time buys")
print(" more than people expect")
print(" light at each clock hour, with the delay still {:.1f} hours:"
.format(START_ERROR))
print(" clock hour shift")
for h in (5.0, 7.0, 9.0, 12.0, 17.0, 21.0, 23.0):
after = h - (TMIN + START_ERROR)
s = AMAX * brightness(10000.0) * duration(30.0) * prc(after)
print(" {:<13.0f}{:+.4f}".format(h, s))
best = max(((AMAX * brightness(10000.0) * duration(30.0)
* prc(h / 4.0 - (TMIN + START_ERROR)), h / 4.0)
for h in range(4 * 24)))
print(" the best clock hour for this person today is {:.2f}"
.format(best[1]))
print(" that is an artefact of the sine, which puts its maximum")
print(" {:.1f} hours after the minimum while measured curves peak"
.format(6.0))
print(" earlier, so the protocol above uses the conservative")
print(" offset of {:.1f} hours and not this one".format(OFFSET))
print(" evening light moves the clock the wrong way, so a bright")
print(" lamp used after work is worse than no lamp at all")
# half the maximum effect is reached at 120 lux
# setting brightness duration fixed adaptive
# 10,000 lux, 30 min 0.9013 0.6000 -0.182 0.351
# 5,000 lux, 30 min 0.8659 0.6000 -0.175 0.338
# 1,000 lux, 30 min 0.7427 0.6000 -0.150 0.290
# 300 lux, 30 min 0.6126 0.6000 -0.124 0.239
# 300 lux, 120 min 0.6126 0.8571 -0.177 0.341
# 100 lux, 30 min 0.4772 0.6000 -0.096 0.186
# a fixed hour of seven makes a 3.0 hour delay worse,
# because the minimum has moved to 8 and the light now
# falls before it rather than after it
# days to bring the delay under 0.5 hours, light placed
# 2.0 hours after the person's own minimum each day:
# 10,000 lux, 30 min 8
# 5,000 lux, 30 min 8
# 1,000 lux, 30 min 9
# 300 lux, 30 min 11
# 300 lux, 120 min 8
# 100 lux, 30 min 14
# the very bright lamp is worth 1.4713 times the room light
# at the same thirty minutes, but only 1.0299 times the same
# room light left on for two hours
# so brightness buys less than the lamps claim and time buys
# more than people expect
# light at each clock hour, with the delay still 3.0 hours:
# clock hour shift
# 5 -0.4971
# 7 -0.1819
# 9 +0.1819
# 12 +0.6088
# 17 +0.4971
# 21 -0.1819
# 23 -0.4971
# the best clock hour for this person today is 14.00
# that is an artefact of the sine, which puts its maximum
# 6.0 hours after the minimum while measured curves peak
# earlier, so the protocol above uses the conservative
# offset of 2.0 hours and not this one
# evening light moves the clock the wrong way, so a bright
# lamp used after work is worse than no lamp at allThe first result is the one the standard protocol gets wrong. A person whose clock is three hours late has a temperature minimum at eight in the morning. Light at seven therefore falls before the minimum, which is the delaying half of the curve, so the protocol of bright light at a fixed early hour makes that person later. The computed day-one shift at seven o’clock is negative for every intensity in the table.
Placed instead two hours after the person’s own minimum, and re-timed each day as the phase comes back, the same light works: the three-hour delay closes in 8 days at ten thousand lux for thirty minutes, 9 at a thousand, 11 at three hundred, and 14 at a hundred.
The intensity comparison is the second correction. Ten thousand lux for thirty minutes is worth 1.4713 times three hundred lux for the same thirty minutes, which is a real but modest advantage, and only 1.0299 times the same three hundred lux left on for two hours. Three hundred lux for two hours matches the lamp, in 8 days. Brightness buys less than the lamps claim and time buys more than people expect.
The table of clock hours makes the risk explicit. For this person, light at nine gives an advance of 0.1819 hours, at noon 0.6088, at five in the afternoon 0.4971, and at nine in the evening a delay of 0.1819. A bright lamp used after work moves the clock the wrong way. The best clock hour on the pure sine comes out at 14.00, which is an artefact of the sine putting its maximum 6.0 hours after the minimum while measured curves peak earlier, so the protocol uses the conservative offset of 2.0 hours instead.
Seventy-two hours, then thirty days
With the phase understood, the two clocks can be run together: light placed against the person’s own minimum, and the load removed as Chapter 5 requires.
Calculation
# Chapter 6. p12: Seventy-two hours, and then thirty days. The two
# clocks of this book run at very different speeds, and a plan has to
# respect both. Phase can be moved in days, because light acts on it
# directly. Receptor density cannot, because it is limited by protein
# turnover. So the first three days can fix the timing and almost
# nothing else, and the month that follows is where the density comes
# back. The two are run together here, with the light placed against
# the person's own minimum as p11 requires and the load removed as p10
# requires, and the state is reported at the end of each phase.
import math
AMAX, I_HALF, I_POW, D_HALF = 1.30, 120.0, 0.5, 20.0
OFFSET, PI2 = 2.0, 6.283185307179586
LUX, MINUTES = 10000.0, 30.0
SYNTH, BASAL, EXTRA = 1.0, 1.0 / 28.0, 3.0 / 28.0
C_ON, C_OFF = 2.5, 1.0
START_PHASE, DAYS = 3.0, 30
def brightness(lux):
return lux ** I_POW / (I_HALF ** I_POW + lux ** I_POW)
def duration(minutes):
return minutes / (D_HALF + minutes)
def day_shift():
return (AMAX * brightness(LUX) * duration(MINUTES)
* math.sin(PI2 * OFFSET / 24.0))
def occupancy(c):
return c / (1.0 + c)
def steady(c):
return SYNTH / (BASAL + EXTRA * occupancy(c))
def step(r, c):
return r + (steady(c) - r) * (BASAL + EXTRA * occupancy(c))
def settled_weekday():
r = steady(C_OFF)
for d in range(399): # ends on a Sunday, as in p09
r = step(r, C_ON if (d % 7) < 5 else C_OFF)
return r
if __name__ == "__main__":
full = steady(C_OFF)
r = settled_weekday()
phase = START_PHASE
print(" starting phase error {:.2f} hours, receptor density {:.4f}"
.format(phase, r / full))
print(" the light is worth {:.4f} hours of advance a day"
.format(day_shift()))
print(" day phase error receptors receptors of rested")
for d in range(1, DAYS + 1):
phase = max(0.0, phase - day_shift())
r = step(r, C_OFF)
if d in (1, 2, 3, 7, 14, 21, 30):
print(" {:<6d}{:<14.3f}{:<12.4f}{:.4f}"
.format(d, phase, r, r / full))
print(" at the end of the first three days the phase error is")
phase3 = max(0.0, START_PHASE - 3.0 * day_shift())
print(" {:.3f} hours, down from {:.2f}".format(phase3, START_PHASE))
r3 = settled_weekday()
for _ in range(3):
r3 = step(r3, C_OFF)
print(" while the receptors have moved from {:.4f} to {:.4f}"
.format(settled_weekday() / full, r3 / full))
print(" of rested, which is {:.4f} of the gap"
.format((r3 - settled_weekday())
/ (full - settled_weekday())))
print(" so three days buys {:.4f} of the phase and {:.4f} of the"
.format((START_PHASE - phase3) / START_PHASE,
(r3 - settled_weekday()) / (full - settled_weekday())))
print(" density: the rescue is a timing intervention, not a repair")
print(" by day {} the phase error is {:.3f} hours and the density"
.format(DAYS, phase))
print(" is {:.4f} of rested, {:.4f} of the gap closed"
.format(r / full, (r - settled_weekday())
/ (full - settled_weekday())))
print(" days for the phase alone: {}"
.format(int(math.ceil(START_PHASE / day_shift()))))
print(" days for the density to close 0.90 of its gap:")
rr, dd = settled_weekday(), 0
while (rr - settled_weekday()) / (full - settled_weekday()) < 0.90 \
and dd < 400:
rr = step(rr, C_OFF)
dd += 1
print(" {}".format(dd))
print(" the ratio between the two timescales is {:.1f}"
.format(dd / (START_PHASE / day_shift())))
# starting phase error 3.00 hours, receptor density 0.8598
# the light is worth 0.3515 hours of advance a day
# day phase error receptors receptors of rested
# 1 2.649 9.7695 0.8723
# 2 2.297 9.8972 0.8837
# 3 1.946 10.0136 0.8941
# 7 0.540 10.3838 0.9271
# 14 0.000 10.7759 0.9621
# 21 0.000 10.9796 0.9803
# 30 0.000 11.1050 0.9915
# at the end of the first three days the phase error is
# 1.946 hours, down from 3.00
# while the receptors have moved from 0.8598 to 0.8941
# of rested, which is 0.2447 of the gap
# so three days buys 0.3515 of the phase and 0.2447 of the
# density: the rescue is a timing intervention, not a repair
# by day 30 the phase error is 0.000 hours and the density
# is 0.9915 of rested, 0.9395 of the gap closed
# days for the phase alone: 9
# days for the density to close 0.90 of its gap:
# 25
# the ratio between the two timescales is 2.9The light is worth 0.3515 hours of advance a day. After three days the phase error is down to 1.946 hours from 3.00, while receptor density has moved from 0.8598 to 0.8941 of rested, which is 0.2447 of its gap. So three days buys 0.3515 of the phase and 0.2447 of the density. The seventy-two hour rescue is a timing intervention; it is not a repair, and it should not be sold as one.
By day thirty the phase error is zero and the density is 0.9915 of rested, with 0.9395 of its gap closed. The phase alone needs 9 days and the density needs 25 days to close 0.90 of its gap, a ratio of 2.9 between the two timescales. That ratio is the reason a plan has to have two phases rather than one: the first three days fix what light can fix, and the month that follows is where the slow variable comes back.
Hypothesis · not established
The author’s interpretation (not yet tested): because light has to be placed relative to the individual’s temperature minimum rather than the clock, protocols specifying a fixed hour will show a bimodal response, helping the phase-advanced and harming the phase-delayed, and the published average effect of morning light therapy understates both. What would show it wrong: if, in a trial stratified by baseline phase, the response to fixed-hour morning light were uniform across strata rather than reversing in sign for the delayed group, the timing argument would be wrong.
Summary
Light at a fixed early hour makes a three-hour delay worse, because the temperature minimum has moved with the phase; placed two hours after the person’s own minimum it closes the delay in 8 to 14 days depending on intensity. Ten thousand lux for thirty minutes is worth 1.4713 times three hundred lux for thirty minutes but only 1.0299 times three hundred lux for two hours. Three days of the full plan buy 0.3515 of the phase and 0.2447 of the receptor gap, and the two timescales differ by 2.9 times. Chapter 7 asks what the classics said about the same two variables.
References
Fan, Y. (2026). HPA-Axis Coupled Oscillations, Cortisol Awakening Response Flattening and Burnout Recovery Kinetics: Glucocorticoid Receptor Downregulation, Ultradian Pulsatility and Allostatic Load Reset [下丘脑垂体肾上腺轴耦合振荡、皮质醇觉醒反应扁平化与倦怠恢复动力学] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.24090252 · Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).