Sleeping on a problem

Why does a night's sleep so often solve a problem that hours of trying could not?

You work at something until midnight and get nowhere. You wake, and within a minute of opening your eyes the answer is simply there, as though it had been sitting on the bedside table. Nothing happened in between — you were unconscious. So either the solving occurred while you were not present for it, or the phrase "slept on it" is doing something more modest than it sounds.

Both possibilities are live, and it is worth taking the dull one seriously before reaching for the interesting one.

Start with the dull explanation, because it explains a great deal. When you have been at a problem for six hours, you are tired, and you are also stuck in a groove — you keep applying the approach that has already failed, because it is the one most available to you. Any break at all weakens that grip. So a night's sleep might do nothing special; it might just be the longest break you take, at the end of which you are rested and no longer primed by your own dead end. Notice this account predicts that a long walk should work too. It often does.

So the real question is not whether sleep helps, but whether sleep does something a walk of the same length could not. What would have to be true for that?

Something would have to happen during sleep that changes the memory itself. And there is decent evidence that it does. Newly formed memories are not filed away intact; they are reactivated offline, particularly during slow-wave sleep, in the hippocampus, and gradually integrated with what the neocortex already holds. This replay is fragmentary and it is interleaved with older material. That matters, because a process that repeatedly re-plays pieces of an experience alongside related knowledge is not a process that preserves detail. It is a process that would preferentially strengthen whatever is common across episodes — the regularity, the gist, the rule — and let the surface particulars fade.

If that is the mechanism, then sleep should not help you solve problems in general. It should specifically help with the kind of problem whose answer is a hidden regularity you have already met without noticing.

Which is close to what Wagner and colleagues tested. They gave people a tedious number-reduction task that had a shortcut buried in its structure — a rule that, once seen, collapses the whole thing. Nobody was told the shortcut existed. After training, one group slept, and two control groups stayed awake for the same interval, one by day and one overnight. Roughly 59% of the sleep group discovered the rule, against about 23% in each waking group. The overnight-wake control is the part that does the work: it holds time of day and circadian phase roughly constant, so the difference is hard to write off as tiredness alone.

Should you believe it flatly? Cautiously. The effect is real but modest, the sleep-and-insight literature is small, and it is exactly the kind of literature — striking result, appealing story, moderate sample sizes — where publication bias is a live concern. Replications of the specific number-reduction finding have been mixed. It is fair to say sleep favours the extraction of gist and structure; it is not fair to say sleep solves problems.

The two explanations are not rivals so much as layers, and it is worth being precise about which is doing what. Fatigue relief and set-breaking are almost certainly part of every "slept on it" experience, and they are sufficient on their own for many of them. Consolidation is the additional claim — that the memory was reorganised, not merely rested — and it is the part that requires evidence rather than intuition.

One misconception is worth heading off: nothing in this says you were working on the problem while asleep. There is no unconscious deliberation running the search you abandoned. What is claimed is much stranger and much more passive — that the material was replayed and re-filed for reasons having nothing to do with your goal, and that the reorganisation happened to make a structure findable that had not been findable before.

Which also explains the precondition everyone reports. Sleep helps only after you have done the hours. There must be something to consolidate.

• Targeted memory reactivation: cueing a specific memory with a sound or odour during slow-wave sleep, and what it does and does not improve.
• Why incubation effects in the waking literature are so much smaller than their reputation.

Every term the collection defines is gathered in the glossary.