The Absolute Limits of Night Vision

This famous experiment first performed by Selig Hecht, C. Shlaer, and M. H. Pirenne in 1942 demonstrated that the rhodopsin in our eyes can detect single photons, but that the brain will not register a response unless several are detected in a short period of time.

Before the experiment can be conducted, scotopic vision must be induced in the subject. Most of the time, we see in photopic mode which uses the light receptors called cones. The cones can see in vivid colors but need a lot of light to do so. Scotopic vision, however, uses the rod receptors, which cannot detect colors but require very little light and are distributed to give much better peripheral vision.

To induce scotopic vision, the subject must be subjected to very dark conditions for some time. First, and relatively quickly, the irises constrict, dilating the pupils and letting more available light into the eyes. Another change that occurs is chemical in nature, and takes much more time. Rhodopsin is the light-sensing molecule in the rods, and each time a rhodopsin molecule senses light, it breaks down into retinal and opsin. The retinal and opsin spontaneously recombine, but very slowly, with a "half-life" of about five minutes (much too slow for the rods to be of any use in daylight). Waiting for 30-40 minutes will allow about 99% of the rhodopsin to be available for sensing light.*

Next the subject must have their head in some kind of stabilizing contraption, so that a precise light source can be accurately aimed at one eye. The best place to aim the light source is 20° away from where the subject's eye is focused. The rods are actually most dense on this part of the eye, resulting in better peripheral night vision than direct night vision. (You can observe this effect by looking at a very dim star at an angle; it will seem to disappear if you look right at it) The light source should be aimed at a 10 arcminute wide spot (about 1/3 the apparent width of the moon from earth), meaning that there will be about 300 rods in that area connected to the same nerve ending.

Now it is time to actually send photons into the subject's eye. Of all the photons that enter the subjects eye, only about 10% of them will actually change a molecule of rhodopsin. Others are either reflected, do not hit a rod, or fail to cause the molecule to change. This uncertainty means that it is neither necessary nor practical to use single photon LEDs (only a recent invention). It can be done statistically with any very dim light source capable of putting out a few hundred photons in less than 100 milliseconds. The lamp should be set for a few hundred photons per pulse, and randomly flashed on or left off. Each time the subject should be asked if they saw it or not. (Note: The lamp must be completely silent, otherwise a subject with good hearing could skew the results) If the subject can get an accuracy better than guessing, then the lamp is turned down and the experiment repeated until the subject gets an accuracy of about 60%. This may sound very low, but because the lamp itself is not perfect, it may sometimes put out fewer photons than expected, so some of the inaccuracy is not the subject's fault.

In the end, the lamp only has to emit about 90 photons for the subject to see it. This corresponds to about 9 photons entering this small area of the eye. It is extremely unlikely that two of these photons hit the same molecule of rhodopsin, so the conclusion is that our eyes can detect individual photons, but any fewer than 9 in a short time will not trigger a brain response.

*Note: Rhodopsin also spontaneously splits into retinal and opsin, but with a half-life of about 6 years. Because there are about 4 million molecules of retinal on each rod, and 300 rods in each 10 arc-minute, it is entirely probable that 9 will decompose within 100 ms of each other. This is why even in complete darkness you can sometimes see random flashes of light.

Sources: (Background information) (Description of the experiment)