Hold a magnifying glass up to a switched-on phone and the image dissolves into a grid of red, green and blue dots. There is no yellow subpixel. No cyan. No brown. The sunflower on the screen is being built entirely out of three colours of light, and the trick works because the eye receiving it has only three kinds of colour sensor to satisfy.

This is the quiet premise underneath every television, laptop, phone, cinema projector and stadium scoreboard made in the last century. A display is not reproducing the physical spectrum of a scene. It is producing a signal that the retina cannot tell apart from the original.

What the retina is actually doing

The human retina contains two families of photoreceptor: rods, which handle dim light and do not distinguish colour well, and cones, which handle colour vision in daylight. There are three cone types in a typical trichromatic eye, usually labelled S, M and L for short-, medium- and long-wavelength sensitive. Each contains a different opsin protein that responds most strongly to a particular band of light: roughly 420 nanometres for S cones (violet-blue), 530 nm for M (green), and 560 nm for L (yellow-green, despite the name “red”).

The peak sensitivities and the shapes of the cone response curves were measured directly by microspectrophotometry in the 1980s, most influentially by Dartnall, Bowmaker and Mollon at the University of Cambridge, who probed individual cones from human donor retinas. Their curves are broad and heavily overlapping. A single photon at 550 nm will trigger both M and L cones, just to different degrees.

The brain does not read a wavelength off the retina. It reads a ratio. Whatever combination of light hits the eye, the visual system compares how hard the three cone populations are being pushed and infers a colour from that ratio.

That inference is where the loophole lives.

Why three primaries are enough

Because colour perception depends on three numbers — the relative activation of S, M and L — any stimulus that produces the same three numbers will look identical, regardless of its actual spectrum. Two lights with completely different physical compositions can be visually indistinguishable. Colour scientists call these pairs metamers.

A ripe lemon reflects a broad band of wavelengths centred in the yellow. A screen showing that lemon emits no yellow light at all. It emits a carefully weighted mix of red and green from adjacent subpixels, and the L and M cones respond in almost the same ratio they would to the real lemon. The S cones are barely stirred in either case. Brain: yellow.

This is why three primaries suffice, and why they are red, green and blue rather than, say, orange, teal and purple. Choosing primaries near the peaks of the three cone types — with enough separation to independently drive each — gives the widest range of ratios, and therefore the widest range of reproducible colours. The formal framework for this dates to the 1931 CIE colour-matching experiments, in which observers adjusted red, green and blue lights to match test colours; the resulting CIE standard observer still underpins modern display specifications.

The gamut problem, and why some colours can’t be shown

Three real primaries can only make the colours that lie inside the triangle they define on a chromaticity diagram. Everything outside that triangle — the deepest spectral violets, the most saturated cyans and greens — is unreachable, no matter how bright the subpixels get. A sodium streetlamp’s yellow, a laser’s green, the blue of a morpho butterfly’s wing under sunlight: these sit at or beyond the edges of what a standard sRGB monitor can produce.

Display standards are essentially agreements about where to plant the three corners. The old sRGB standard, formalised by the IEC in 1999, covers a modest triangle suited to CRT phosphors of the era. DCI-P3, adopted by digital cinema and now common in phones and laptops, pushes the red and green corners outward. Rec. 2020, the ultra-high-definition television standard published by the ITU in 2012, uses primaries that are nearly monochromatic and covers a much larger fraction of visible colour — but no shipping consumer display currently reaches its full extent.

Widening the triangle usually means narrower-band primaries, which is why quantum-dot backlights and laser projectors have become interesting. They emit tighter spectral peaks, which sit closer to the corners of the visible gamut.

When the trick breaks

The three-primary shortcut assumes a standard trichromatic eye. About 8 per cent of men and 0.5 per cent of women of northern European ancestry have a variant form of one of the L or M opsin genes, both of which sit on the X chromosome, which shifts a cone’s peak sensitivity and collapses part of the red-green distinction. The genetic basis was worked out by Nathans, Thomas and Hogness in 1986. For these viewers, some pairs of colours that a display treats as different will look the same, and some metamers designed for standard observers will not quite match.

A smaller number of people appear to be functional tetrachromats, carrying a fourth distinct cone type. Standard RGB screens cannot present them with the full range of ratios their retinas can distinguish. To a true tetrachromat, a monitor is showing a slightly impoverished version of the world.

And the trick breaks in the other direction, too. Cameras and screens each have their own spectral responses, and a colour matched perfectly for the human eye can look wrong to a camera sensor, which is why product photography for paint, fabric and cosmetics is so difficult. The lemon on the screen fools the eye; it does not fool a spectrometer.

What the subpixels are really doing

Zoom in on an LCD and each pixel is three vertical stripes with colour filters over a white backlight. On an OLED panel, each subpixel is an organic compound that emits its own red, green or blue light when current flows through it. Some OLED layouts, such as Samsung’s PenTile arrangements, use different numbers of red, green and blue emitters per pixel, exploiting the fact that the eye resolves green detail more sharply than red or blue — a direct consequence of M and L cones being more numerous than S cones in the central retina.

The eye is being managed at every level. Anti-aliasing on text uses subpixel positioning to fake resolution the panel does not physically have. Video codecs throw away far more colour information than brightness information, because the visual system is much more forgiving of chroma loss than of luminance loss. The famous 4:2:0 chroma subsampling used in almost every streaming video keeps full-resolution brightness and quarter-resolution colour, and most viewers never notice.

None of this would work on a sensor with a different architecture. It works on the human eye because there are only three channels to feed, and they overlap enough that a small palette of primaries can drive them into any ratio the retina can produce on its own.

A three-channel world dressed as a full spectrum

The strangest consequence is what “colour” turns out to mean in practice. There is no yellow on the screen. There is no orange in the sunset photo on a phone. Those sensations exist in the visual cortex, assembled from the ratios of three protein activations, and a display’s entire job is to arrive at the correct ratios by whatever means the hardware allows.

The same principle explains why the technology has been so stable. Screens have changed from cathode-ray tubes to plasma to LCD to OLED to microLED, but the underlying bet has not moved since Thomas Young proposed in 1802 that the eye must contain three kinds of colour-sensitive particle. Every colour display since has been engineered around that guess, which turned out to be correct.

The physical world contains a continuous spectrum. The eye samples it in three channels. The screen just has to keep those three channels happy.