Could You See a Single Photon?
Yes. That Is the Least Surprising Thing on This Page.
Put me in a dark room for as long as it takes, and then send exactly one photon at my eye. Not a dim flash, not a few. One. I want to know whether I would notice, and if not, how many it would take before I did.
The first wall grants the request, which is unusual for this section and is the reason the page exists. A single cell in your eye responds to a single photon, that response has been recorded with an electrode, and a person really can do slightly better than guessing about whether one arrived. The walls after that are about why you cannot use it. The eye is a detector running at body temperature, its own chemistry produces hundreds of thousands of false flashes every second, and everything else on this page is what a system has to do to get a real signal out of that.
What Physics Says, Wall by Wall
Granted. A rod fires on one photon, and a person can beat chance.
Start with the cell. A rod in the retina is a stack of about a thousand discs carrying a hundred million molecules of a pigment that changes shape when it absorbs light. One absorption twists one molecule, and that single event is amplified by a chain of chemistry into a measurable change in the current flowing through the cell. Baylor and his colleagues put an electrode on a single rod in 1979 and watched the individual responses arrive, one photon at a time, as discrete bumps.
Then the person. In 2016 a group built a source that emits exactly one photon at a time and asked dark-adapted volunteers, over thirty thousand trials, in which of two intervals it had been sent. The answer came out above chance. Not by much, at just over 51 percent where guessing gives 50, but with enough trials to be a real effect, and better still when the observer was asked how confident they felt.
So the honest answer to the request is yes. A photon carries about two and a half electronvolts, your eye catches it, and something in you knows. The rest of the page is about the distance between that and anything you would call seeing.
Your brain refuses to tell you about it
Between the rod and the report there is a filter, and it is deliberate. The visual system will not call a flash a flash on the strength of one absorption. It waits for several to land close together, in a small patch of retina and within about a tenth of a second, and only then passes it up as an event.
Hecht and his colleagues measured how many that takes in 1942, and the design is worth admiring: they worked out how many photons left the source, how many survived the eye’s own optics, and how many a rod would absorb, then asked people to say whether they had seen anything. Somewhere between five and fourteen absorptions were needed before an observer said yes sixty percent of the time, which is the criterion the paper used. Every replication since has landed in that region.
The picture below is that curve. The ruler is how many photons arrive at the front of your eye; the bar is how often you would say you saw something. About a tenth of what arrives is caught, and the coincidence rule does the rest. Slide it along and watch two things: the bar lies flat on the floor for the whole first decade, and then the whole climb from never to always happens inside a single one.
Because the detector is warm, and warm detectors lie
Here is why the filter exists. The pigment molecule does not only change shape when it absorbs light. It can also do it from ordinary heat, by chance, with no photon involved at all, and the cell has no way to tell the difference. The output is identical.
The rate is a beautiful pair of numbers. Any individual molecule of the pigment is astonishingly stable: it waits, on average, about five hundred years before misfiring. And there are a hundred million of them in each rod, so a rod produces a false flash roughly every three minutes. And there are around a hundred million rods, so your retina, sitting in a completely dark room, is generating something like seven hundred thousand spurious single-photon signals every second.
That is what one real photon has to compete with. A system that passed on every single absorption would not be more sensitive; it would be a snowstorm. Demanding that several arrive together in one place is not the visual system being slow. It is the only way to separate a signal from a noise floor that looks exactly like it.
How well it works is a question about size, and the picture below is that. The ruler is how much retina gets pooled into one judgment. Everything is spread over a hundred million rods, so the noise in a patch grows with the patch and the signal does not, and there is a place where they cross.
And most of the photons never arrive at all
There is a second tax before any of that. Of the photons that reach the front of your eye, only about half get through the cornea, lens and the fluid behind them without being absorbed or scattered somewhere useless. Of those that reach the retina, a good fraction pass between the rods or straight through one without being caught, because a rod is a small target and its pigment is not perfectly absorbing.
Multiply it out and roughly one photon in ten arriving at your eye produces a signal in a cell. That is not a poor figure, and it is much better than any camera managed until the 1980s, but it means the request has a hidden factor in it. Sending one photon at an eye is not the same as delivering one to a rod, and about ten have to be sent before one is caught.
This is the same problem every telescope has, and the eye is losing on one count
Nothing above is about biology. Any detector of faint light is judged on three numbers: what fraction of arriving photons it registers, how many counts it invents by itself, and how long it can add up before something moves. Astronomers have been optimizing those three since the photographic plate, and the eye can be scored on exactly the same card.
On the first the eye does respectably: one in ten, against about nine in ten for a good modern sensor. On the third it does badly: it adds up for about a tenth of a second and then starts again, where a telescope can stare at the same patch of sky for hours. But the count it loses worst is the middle one, and the reason is a single number. A telescope camera is cooled to a hundred degrees below zero, precisely so that its own molecules stop misfiring. Your retina runs at 37 degrees, because it is in your head.
That comparison also explains something people notice and misattribute. Looking slightly to one side of a faint star makes it easier to see, and the usual explanation is that the center of your vision has no rods, which is true. The deeper reason is that away from the center the retina pools many rods into one channel, trading detail for the ability to catch a coincidence across a wider patch. It is the same trick as reading out a camera in larger blocks, and it costs the same thing.
And it is already about as good as anything could be
The last wall is the one that turns the page around. Suppose you fixed everything. Cool the retina, remove the optics, count every absorption. There is still a floor, and it is not in the eye at all. Light arrives in whole photons at random times, so a faint source does not deliver a steady trickle; it delivers a number that varies from moment to moment purely by chance. Ask for a hundred and you get ninety or a hundred and ten.
At the light levels a dark-adapted eye works in, that randomness is comparable to the signal being looked for. So a perfect detector would still be uncertain, and the margin between what the eye achieves and what physics permits is much smaller than the margin between the eye and a camera. Evolution did not build an adequate light sensor. It built one that ran out of room to improve because the light itself had become the limitation.
How close is close is not fully settled, and it is an awkward thing to measure, because the answer depends on assumptions about what the brain does with the counts it gets. The 2016 experiment hints that a person’s sense of their own confidence carries information their yes-or-no answer does not, which suggests the filter is discarding less than it appears to. That is an open question about a sense organ everyone has, which is a pleasant thing for one to be.
The Pushback
The first of these is the objection everybody makes on being told the first wall, and it has a clean answer.
Because detecting and seeing are different jobs, and reading needs the second one. To detect a flash you need enough photons in one place to beat the noise. To read you need to tell one small patch of page from the patch next to it, which means enough photons in each of them separately, and there are a great many patches.
The trick that buys sensitivity is the thing that takes reading away. Away from the center of vision, dozens or hundreds of rods report through a single channel, which is how a coincidence gets found at all; but everything those rods saw arrives as one number, so whatever detail was in it is gone. That is also why nothing has color at night. There are three kinds of cone and only one kind of rod, and one kind of detector cannot distinguish wavelength from brightness.
The honest answer is that most of the limit is chemistry rather than skill, so there is not much room. The false flashes in wall three do not care how hard anyone is concentrating, and the coincidence rule is wired into the retina rather than decided in the moment.
What does move is what gets read out of the same signal. Observers in the 2016 work did better on trials where they reported feeling confident, and better again shortly after a genuine photon had arrived, which suggests the machinery upstream carries more than a bare yes or no. So the trainable part is not the eye. It is what the rest of you is willing to do with a signal it has been throwing away because throwing it away is usually right.
Where the Request Was Right
It was right in a way that most confident claims about the senses are not. A single quantum of light, the smallest amount of light that can exist, is enough to produce a measurable event in a cell in your head, and enough for a person to do better than a coin. That is a genuinely startling fact about a piece of ordinary human anatomy, and it is not a figure of speech or a rounding of something less impressive.
The assumption worth taking out of it is that a limit must be a shortcoming. Everything in the walls above that stops you seeing a single photon is doing a job. The coincidence rule is a noise filter. The pooling that costs you detail is what makes the filter possible. Each of them is a trade made in the right direction for an animal that needs to see in a forest at night, and the cost is exactly the request on this page.
And the request had the scale right without knowing it. Ten photons is roughly what has to arrive; a handful is what has to be caught. Not thousands, not millions. The number a careful person would guess if they knew nothing about eyes and something about how small a photon is.
An Opinion, Dated
The walls above are physics. This box is the drafting model’s own bets, and they are scoreable.
Three in five that the single-photon detection result is independently replicated by a different group with a different source by 2035. It is a hard experiment and the effect is small, which is exactly the combination that sometimes fails to reproduce.
One in four that by 2040 there is an accepted account of how much information the visual system discards at the coincidence stage, precise enough to say how far the eye sits from the limit set by the light itself.
Nineteen in twenty that no training regime is ever shown to raise a person’s single-photon detection meaningfully above the published figure. The obstacle is the pigment misfiring, and nothing anybody practises changes that.
These intuitions were distilled from human physics writing and carry its fashions. Weigh them accordingly.
The Updated Map
Vision is usually pictured as a camera of moderate quality: good enough for the job, far behind the instruments. On the count that matters at the faint end that is backwards. The eye catches one photon in ten and reports on the smallest amount of light there is, and what stops it doing better is not poor engineering but a decision, taken in the retina, to throw away anything that could be its own chemistry misfiring. Every real detector makes that decision. A telescope camera makes it too, and buys its way out by being cooled to a hundred below. Yours cannot be cooled, so it filters instead, and the price of the filter is precisely the thing this page asked for. The strange part is not that you cannot see one photon. It is that you very nearly can, and that the reason you cannot is the same reason the instrument on the mountain has a refrigerator bolted to it.
