The fourth room
Behind the eye
Light goes less than half a millimetre into the retina, and stops. It bends one small molecule, and George Wald, who found that out, said the rest himself: everything else that happens is a “dark” consequence of that one reaction. The brain never meets light. This room follows the dark consequences, and at each step asks the question on the front door: how much of what you see was in the light?
The fourth room · Behind the eyeChapter 17 of 21
A rainbow holds every wavelength there is between violet and red. Your eye reports each point of it to the brain with three numbers, and has thrown the rest away before the signal leaves the eye.
Chapter 17 · Behind the eye
Three Numbers
Why a screen needs only three colours, why two lamps that look alike are not alike, and what the world looks like with one number missing.
Look first
A cone cannot tell you the colour
In daylight you see with cones, and there are three kinds. Each kind responds to a broad stretch of the spectrum, more strongly in the middle of its range than at the ends. Move the light below along the spectrum and watch what each kind of cone reports.
Now look at one kind alone. All it sends is a single number: how much it caught. A dim light at its favourite wavelength and a brighter light off to one side give it exactly the same number. One kind of cone, by itself, is colour-blind.
Colour begins only when the brain compares the three. And three numbers are all it is ever told about the light at one point.
The three curves are the CIE’s cone fundamentals of 2006, measured at the front of the eye. The three bars are what one unit of light at the chosen wavelength gives each kind of cone.
Commit to a prediction
The yellow of this screen
In chapter 14 salt in a flame gave out light of one wavelength, 589 nanometres, and it was yellow. The patch below is the same yellow, as nearly as a screen can show it.
How much of the light leaving that patch is near 589 nanometres?
Two lights. Above: sodium’s line with a little white light added. Below: a model screen’s red and green bands. The bars are the signals of the three kinds of cone.
A screen has three lamps behind every point: one red, one green, one blue. To show sodium’s yellow it lights the red and the green and leaves the blue dark. On the model screen here, only 4% of that light lies within 10 nanometres of sodium’s line. Of the light it is matching, 94% is the line itself.
The two lights could hardly be less alike, and no eye with three kinds of cone can tell them apart, because each kind of cone gets the same total from both. Two lights like that are called metamers. Every colour on every screen is a metamer of something.
Newton said it first, in 1704: “the Rays to speak properly are not coloured. In them there is nothing else than a certain Power and Disposition to stir up a Sensation of this or that Colour.”
Try it with your own eyes
The left half of this panel is ruled in stripes of pure red and pure green. The right half is a flat yellow. Narrow the stripes, or step back from the screen, until the two halves match.
your eyes
Nothing on the left is yellow at any stripe width. When the stripes are too fine for your eye to separate, as in chapter 2, their light falls on the same cones and is added there. Under a strong magnifying glass the flat half is separate red and green lamps too.
Why the first guesses made sense
Yellow is not in the light
If you said nearly all, you were reasoning as language does. We say yellow light, as if yellow were a property the light carried with it, like its wavelength. For sodium’s line the two happen to go together, and nothing in daily life pulls them apart.
A screen pulls them apart. Yellow is what three numbers in a certain proportion look like, and there are endlessly many lights that give those three numbers.
The rule
Add up what each cone catches
For any light, multiply the power at each wavelength by how strongly a cone responds there, and add. Do it for the three kinds of cone. Those three sums are the light, as far as seeing is concerned. Two lights with the same sums look the same, whatever their spectra.
The sums add: mix two lights and each cone’s number is the sum of its numbers for the two. So matching a colour is three equations in three unknowns, and three lamps are enough to match almost anything. That is the whole principle of colour screens, colour printing and colour film.
James Clerk Maxwell worked this out in the 1850s with a box that mixed three colours of sunlight. He found that any colour of the spectrum could be matched by three, with one catch: sometimes one of the three had to be added to the other side.
How much of three lights (650, 530 and 460 nm) matches one unit of each wavelength of the spectrum. Where a curve goes below zero, that light must be added to the other side.
A case you have not seen
Two whites and a pot of paint
Two lamps. One gives a smooth white, every wavelength present, like the 6,500 kelvin glow of chapter 13. The other gives three narrow bands of red, green and blue, set so that its light gives the cones exactly the same three numbers. Shone on a white wall, the two lamps cannot be told apart.
Now paint a board with an ideal yellow paint, one that reflects light between 570 and 600 nanometres and absorbs the rest.
Does the board look the same under both lamps?
Left: the two lamps’ spectra, with the band the paint reflects. Each spectrum is drawn to its own height. Right: the white wall and the painted board under each lamp, with their relative brightness kept.
Under the smooth lamp the paint has plenty to reflect. Under the three bands it finds only the foot of the red band and almost nothing of the green. The board comes out a fifth as bright, and redder.
Two lights that match as lights need not match as lamps. The eye’s three numbers are taken after the paint has done its work, and the paint works on the whole spectrum, including everything the three numbers leave out.
The bench
Maxwell’s colour box
Choose a wavelength. The bench works out how much of three lights matches it, and shows the three cone signals from each side.
Things to try
- Run the wavelength from red to violet with the three single lights. Where must a light be added to the other side?
- Switch to the screen’s three bands. Is there any single wavelength they can match without help?
- Choose the eye without M cones. Find two wavelengths, one red and one green, whose patches look alike.
What you can now say
Colour is not in the light. The eye reports each point of light with three numbers, one from each kind of cone, and lights with the same three numbers look the same whatever their spectra. That is why three lamps can make every colour on a screen.
What the three numbers leave out is gone before the signal leaves the eye. No later stage can get it back.
For the determined reader: the sums, the screen and the limits
What is computed
With P the power of the light at each wavelength, and l, m and s the three cone curves:
L = ∑ P(λ) l(λ), M = ∑ P(λ) m(λ), S = ∑ P(λ) s(λ)
The curves are the CIE’s cone fundamentals of 2006 for a small central field, which are those of Andrew Stockman and Lindsay Sharpe. They describe an average eye with normal colour vision, measured at the front of the eye, so they include the yellowing of the lens. Every match on this page is found by solving three equations, and the tests solve them again by a different route.
The screen, and the colours you are shown
The model screen is three bell-shaped bands. It is not a measurement of any screen. The colours painted on this page pass through a second, older table, the CIE’s observer of 1931, on which the standard for screens is built. The two tables agree to a few percent on average and differ most in the violet. And no screen can show a single wavelength at full strength, so the spectrum drawn here is paler than a real one.
Where this stops
Eyes differ. The lens yellows with age, the pigment at the centre of the retina varies from person to person, and the cone curves themselves come in several inherited versions. Two lights that match for you may not quite match for the person beside you. The three numbers also describe one small patch of the central retina in daylight. They say nothing yet about what the patch next to it is doing, which turns out to matter a great deal.
One more case
With one number missing
About one man in twelve, and one woman in two hundred, has two working kinds of cone where most people have three, or has two kinds so alike that they nearly count as one. For them each point of light is two numbers.
What does a rainbow look like to an eye with two kinds of cone?
The spectrum, and a reconstruction of it for three kinds of two-coned eye. Each lower strip gives the two remaining kinds of cone what the top strip gives them, as closely as a screen can show it.
With two numbers, all that can differ between two lights of equal brightness is the balance of one number against the other. So the spectrum runs from one hue, through a point where the balance is the same as for white, to one other hue. At that point a single wavelength looks colourless.
For an eye without L cones the colourless point comes out at 491 nanometres, and without M cones at 498. In 1936 Selig Hecht and Simon Shlaer measured it in 46 people with such eyes, and found averages of 496.5 and 504.3: the same order, the same few nanometres apart, and a little further toward the green. Reds, oranges, yellows and greens all fall on one side of the point, and differ only in brightness.
The two commonest kinds are inherited through the X chromosome, where the genes for the L and M pigments sit side by side. The two pigments are 96% identical. Men have one X, so one faulty copy is enough.
Three kinds of fibre
In 1802 Thomas Young, a London physician, proposed that the retina has just three kinds of receiver, each answering to a broad part of the spectrum. It was a guess, and it stayed one for half a century.
Maxwell made it a measurement. With spinning tops painted in sectors, and then with his colour box, he showed that three well-chosen lights would match any colour, and worked out how much of each. In 1861, to prove the point at a lecture, he had Thomas Sutton photograph a tartan ribbon three times, through red, green and blue filters, and projected the three pictures through the same filters on top of one another. It was the first colour photograph.
John Dalton, the chemist, described his own colour-blindness in 1794. “That part of the image which others call red,” he wrote of the spectrum, “appears to me little more than a shade, or defect of light.” He thought the fluid inside his eye must be tinted. In 1995 his preserved eye was examined, and showed that he had no M cones.
The three pigments were finally measured in the 1960s. George Wald and Paul Brown shone light through the centre of human and monkey retinas and bleached the pigments one at a time. Wald shared the Nobel Prize of 1967 with Ragnar Granit and Keffer Hartline, “for their discoveries concerning the primary physiological and chemical visual processes in the eye”. The genes followed in 1986.
Where this explanation stops
Three numbers are not a colour
This chapter says which lights look alike. It does not say what any of them looks like. Nothing in the three sums is yellow. Somewhere after the cones the three numbers are compared, and the results of those comparisons are what the brain works with: more L than M, or less; more S than both, or less.
Why one of those comparisons should be red, and feel the way red feels, is not answered by anything here. The room will come back to it at the end.
End of chapter 17
Added to your notebook
Colour is not in the light. The eye reports each point of light with three numbers, one from each kind of cone, and lights with the same three numbers look the same whatever their spectra. That is why three lamps can make every colour on a screen.
Three numbers from every cone, and there are millions of cones. How much of that reaches the brain?
Chapter 18 · The Edited Picture