The 3 p.m. slump usually isn't your willpower running out. It's one of your clocks loosening its grip right on schedule.
At 9 a.m. you blast through twenty emails. At 3 p.m. you read the same page of a report three times and still can't say what it's about. At 9 p.m., when you should be sleepy, your head is weirdly clear. The usual explanation is a metaphor: attention is a tank of fuel. Full in the morning, a little less with every use, empty by mid-afternoon.
It sounds right, but the evidence doesn't back it up well. Attention really is zero-sum in any given moment: look at this and you can't take in that. But the ups and downs across a day mostly aren't about "using it up." They come from two clocks stacked on top of each other, where you sit on an arousal curve, and a sleep debt you can't detect in yourself. Which one is at work changes what you should do about it.
Start with "right now." In front of you there are words on a screen, a bird flying past the window, the hum of the fridge. The neurons handling each of them are competing: whatever gets amplified gets seen; whatever doesn't is pushed down. This is called biased competition. The "bias" comes from your goals: a thought held in the prefrontal cortex like "I'm reading the report" gives the relevant signals an extra push. Prefrontal cortex So in any moment, attention really is a pie being divided: more here means less there (see Topic 2).
"Less and less over the day" is a different claim. In 1998 the American psychologist Roy Baumeister proposed ego depletion: self-control is like a muscle that tires with each use. People who first resisted eating cookies gave up sooner on a hard puzzle afterwards. The idea was hugely popular for more than a decade, and it spawned the claim that a glass of sugary drink could top you back up.
The trouble came with replication. In 2016, 23 labs and more than two thousand people repeated the experiment using one protocol, published before any data came in. This is called preregistration, and it stops researchers from picking the flattering results afterwards. The effect was close to zero. A second round in 2021, with 36 labs and 3,500 people, found nearly zero again. The large early effects most likely came from small samples and the habit of publishing only studies that "worked."
So is the feeling of getting worn down fake? No, it's real. But it may work less like a fuel gauge and more like an accountant. Another theory says the brain is constantly estimating whether carrying on with the current task is worth it. By sticking with it you give up other options (rest, something more fun), and that's the opportunity cost. As it rises, you get a nudge in the form of feeling tired or fed up. This fits a familiar scene: too tired to read another page of the report, you play a round of a game and suddenly perk up. Attention wasn't used up; the price of keeping it here just went up.
In a large language model, the attention mechanism gives every earlier position a weight each time it produces a word. A function called softmax normalizes those weights so they add up to exactly 1. More weight on one word means less on others, so it's zero-sum in the moment, just like biased competition. But the model doesn't "get tired" as it writes: the thousandth word gets the same computation as the first. There's no ego depletion. What really is limited is two kinds of budget. One is the context window, the total amount of text the model can see at once. The other is reasoning compute: some models can be told how long a hidden draft to write before answering, more for hard problems, less for easy ones. That second one is close to the opportunity-cost idea: the resource isn't fuel that runs dry, it's a budget you keep deciding where to spend.
The classic explanation for how alertness rises and falls over a day is the two-process model, proposed in 1982 by the Swiss sleep researcher Alexander Borbély. Two curves are playing tug-of-war in your head.
The first, Process S, is your stock of sleepiness. For every hour you're awake, brain cells burn energy and leave behind a molecule called adenosine. It slowly builds up, like sand in an hourglass, and the more of it there is, the stronger the pressure to sleep. Sleep clears it, and the next morning starts from a low point again. Caffeine works by blocking the sites where adenosine is received, so you stop feeling the sleepiness for a while. But the sand hasn't gone anywhere; when the caffeine wears off, the sleepiness comes back all at once.
The second, Process C, is your body clock. Its master switch is a small cluster of neurons in the hypothalamus called the suprachiasmatic nucleus. Hypothalamus It runs on a roughly 24-hour rhythm and resets itself to morning light. Through the day it sends out a growing "stay awake" signal that cancels out the mounting sleepiness. When this cluster was destroyed in monkeys, they didn't become more alert. Their daytime wakefulness broke into fragments, with many more naps. One of this clock's big jobs, in other words, is holding the day together.
Put the two curves together and the shape of the day appears. From about 1–2 p.m. to 3–4 p.m., the clock's alerting signal pauses briefly while sleepiness has been piling up for half a day, which gives you the post-lunch dip. Lunch isn't the whole story: the dip shows up even in experiments where people skip lunch, and a big meal just makes it more obvious. Then, two or three hours before bedtime, the alerting signal hits its daily peak. That window is actually the hardest time to fall asleep, sometimes called the "forbidden zone" for sleep. It's why you can feel oddly wired at 9 p.m.
Everyone's clock is also set a bit differently. This is your chronotype: some people are natural larks, others night owls. Genes play a part, the whole curve shifts later in adolescence, and it moves earlier again in old age. A series of studies found a synchrony effect: tasks that need focus and resistance to distraction go better at your own peak time. Off-peak, the first thing to slip is often the ability to hold back irrelevant thoughts. Interestingly, as the previous topic showed, that's exactly what makes off-peak hours better for insight.
During the Second World War, the Royal Air Force noticed a strange problem: the later into a shift, the more often radar operators missed submarine signals on their screens. The psychologist Norman Mackworth built a clock test. A pointer ticked forward one step at a time and occasionally jumped two; your job was to spot the jumps. Within half an hour, misses rose clearly, then kept slowly getting worse. This is the vigilance decrement: the more monotonous a watch-keeping task, and the rarer its targets, the less time you can hold up.
Why do we slip? One key player sits in the brainstem: a small cluster of neurons called the locus coeruleus. Brainstem arousal system There are only tens of thousands of them, yet their fibers reach almost the entire brain, releasing norepinephrine, a chemical messenger that tunes how alert you are. Neurotransmitters Its output acts like a volume knob for how sensitive the whole cortex is to incoming signals.
The American neuroscientists Gary Aston-Jones and Jonathan Cohen pulled this together into an inverted U. When the locus coeruleus's baseline firing is too low, you're groggy and nothing grabs you. At a moderate level, it stays quiet most of the time and then fires a crisp burst the moment a target appears, strongly amplifying that one signal. That's peak focus. When baseline firing is too high, it's firing all the time and everything gets amplified, so everything seems important and you end up restless and scanning around. Another researcher, Amy Arnsten, found the same shape in the prefrontal cortex: a moderate amount of norepinephrine helps it keep hold of the current thought, while too much, as under acute stress, knocks it offline. Prefrontal cortex
Now map it onto a day. Just after waking and in the post-lunch dip, you're on the left side of the curve. After a deadline crunch, three coffees in a row or a stressful meeting, you may have overshot to the right. "Can't focus" on the left needs a lift; on the right it needs calming down. The first calls for a short walk or some daylight, the second for slow breathing or one fewer coffee. And long, monotonous monitoring slowly slides you toward the left, which is one source of the vigilance decrement.
Everything so far has been about a single day. There's a sneakier layer underneath: sleep loss that builds up over several days. In 2003, Hans Van Dongen's team at the University of Pennsylvania split healthy young adults into groups allowed 4, 6 or 8 hours in bed each night for two straight weeks, plus a group kept awake for three nights in a row. Every day they measured attention with a very simple test: a counter lights up on a screen at random moments and you press a button as soon as you see it. Any response slower than half a second counts as a "lapse."
The results sting. The 8-hour group barely changed. In the 6-hour group, lapses climbed day after day, and after two weeks they were close to the level of people who had stayed up for one to two whole nights. The 4-hour group was worse still. Yet when asked to rate how sleepy they felt, the 6-hour group's ratings rose a little over the first few days and then hardly rose at all. Performance kept sliding while people felt they had "adjusted." When you're short on sleep, your sense of your own state is exactly the gauge you can trust least.
The good news is that some fixes really do take just minutes. In a 2011 experiment, people did a monotonous 50-minute watch-keeping task. One group was briefly switched to a different task twice along the way, and their performance barely dropped, while the group that watched non-stop declined as usual. A short "change of channel" seems to let attention re-lock onto the goal. A review pooling twenty-odd studies also found that micro-breaks of ten minutes or less reliably reduce fatigue and boost energy. Their effect on work performance is smaller, and clearer for routine tasks than for demanding mental work. A few minutes of brisk walking or stair climbing also gives attention and reaction speed a small lift for a short while afterwards.
Keep the two apart, though. Micro-breaks and short bouts of movement fix vigilance decrement and your position on the curve within a day. Sleep debt built up over days can only be paid back with sleep. Coffee can block the adenosine signal, but it can't flatten that climbing line in Van Dongen's experiment.