A camera turns a scene into a photograph. Change just a couple of dials — how wide the lens opens, where it focuses — and you get very different pictures. Drag the sliders below the render:
Initializing…
A camera turns a scene into a photograph. Change just a couple of dials — how wide the lens opens, where it focuses — and you get very different pictures. Drag the sliders below the render:
Left: the camera and scene from outside — drag to orbit. The bright amber frame is the focus plane — drag focus and it slides along the sight-lines. The two fainter frames bracketing it are the depth of field, the zone that comes out sharp: open the aperture and the bracket narrows to a sliver; stop down and it deepens. Right: the photograph that camera takes — only what falls inside that bracket is sharp. This is what the rest of the piece explains.
Behind that simple result is a chain of physics, and we'll build it from the bottom up — starting with the thing that actually catches the light.
A camera sensor is a dense grid of tiny buckets, and each one does something almost embarrassingly simple: it counts the individual photons that rain into it while the shutter is open. More photons, brighter pixel — that running tally is the exposure. But light doesn't arrive in a smooth stream; it comes as discrete grains, landing at random moments. Over a long exposure the randomness averages out, but catch only a handful of photons per bucket and the raw count is at the mercy of the luck of the draw. That statistical jitter is shot noise, and it isn't a flaw in the sensor — it is baked into light itself.
The mathematics is unforgiving. The noise grows as the square root of the count, so the signal climbs above it only as a square root too: gather four times as many photons and the graininess merely halves. Clean pictures are expensive in light, which is why a dim room grains up while daylight stays smooth — drag the slider on the figure below and watch the seething settle as the count climbs. At the bright end each bucket can hold only so much before it overflows and the highlight clips to featureless white; at the dim end the faintest tones drown under the noise. The gap between those limits is the sensor's dynamic range, and it is why a scene your eye reads at a glance can force a photograph to choose between a blown-out sky and shadows crushed to black.
Each bucket counts photons, and photons arrive at random. With little light gathered the count is pure speckle and the picture crawls; let more light pile up and the true image settles out of the noise — but slowly, since signal-to-noise climbs only as the square root of the light. Watch the bar target at lower left: the finest bars drown first, and detail is lost from the fine end inward long before the picture as a whole looks noisy.
A bucket, though, only measures how much light lands in it, never what colour. So each one is capped with a red, green, or blue filter, tiled in a fine checker — the Bayer mosaic, with twice as many greens as reds or blues because our eyes wring most of their detail from the green part of the spectrum. Every photosite now records just one channel; the missing two are reconstructed from its neighbours, a step called demosaicing. Done well it is invisible; pushed too far it invents colour that was never there — the faint rainbow shimmer across a finely striped shirt, the coloured staircase along a hard edge.
And underneath the colour sits a quieter trade. Make the buckets bigger and each drinks in more light for less noise, but the grid grows coarser and fine detail softens; shrink them for more resolution and every bucket starves, so the same scene grains up sooner. Turn up the gain to brighten a dark frame and you only amplify the noise along with the signal — the count of real photons never rises. Much of sensor design is the art of balancing those pulls, and the figure below lets you feel each one in turn.
Each photosite (drag the grid coarse to see them) measures just one colour through its filter — a speckled mosaic. Slide demosaic to let software reconstruct full colour from the neighbours; push exposure and the highlights clip to white.
Now every point can be recorded — brightness and colour alike. And yet aim that bare sensor at the world and you get nothing but a formless wash of light. The trouble is directional: each bucket faces the whole scene at once, catching rays from the lamp, the window, and every wall together, and simply adds them all up. Its neighbour sees almost the same jumble, so nothing changes from one bucket to the next — no edges, no shapes, only an average of everything, as the left panel below shows. To form an image the light has to be sorted by direction, each point of the scene steered onto its own bucket and no other. The surprising thing is that this takes no lens at all. You can start with nothing but a hole.
Left: point a bare sensor at the world and every bucket drinks in rays from every direction at once — the scene collapses to a formless wash. Right: force each bucket to see through a single tiny hole and an image snaps into being, upside down because the rays cross on their way through. Same scene, same sensor — the only difference is the hole.
Poke a tiny hole in the front of a darkened box and each point in the scene sends a thin pencil of light straight through it to the back wall, painting a picture there. Block every ray but the one narrow bundle that lines up with the hole, and direction is restored: each bucket now sees a single point of the world. The picture lands upside down and flipped left-for-right, because the straight rays cross as they squeeze through the opening. This is the camera obscura, and it is ancient — Mozi wrote of it in China some twenty-four centuries ago, Alhazen studied it around the year 1000, and Renaissance painters traced the scene it threw onto the wall to get their perspective exactly right. No glass, no electronics; only a hole and the fact that light travels in straight lines.
Look closely at that image and the corners are already a little darker than the middle — and it has nothing to do with the hole's size. A patch of sensor off in the corner sees the opening from an angle: it sits farther away, the light rakes across it more steeply, and the hole itself looks like a squashed ellipse. Pile those three effects together and the brightness falls off as the fourth power of the cosine of the angle from the centre — the notorious cos⁴ law. It's why almost every photograph dims toward its corners, and why the widest-angle shots vignette the hardest.
This much is natural vignetting — pure geometry, present even in a flawless pinhole and unavoidable in the optics. Real lenses pile their own darkening on top: the barrel and the rims of the glass clip the slanting bundles bound for the corners, which is why a lens wide open often shades its edges and then brightens them as you stop it down. The geometric part can be lifted out afterwards in software, once you know the law it follows; the mechanical part is a matter of the lens simply letting less light reach the corner in the first place.
An evenly-lit card as the sensor sees it: the corners fall to cos⁴ of their angle off the axis. Widen the angle of view and the darkening bites deeper; the amber ring marks where the light has dropped by half — a full stop.
The whole scene through one tiny hole — and it lands upside down.
But there's a catch, and it's a hard one. Shrink the hole and each pencil of light narrows, so the picture sharpens — yet so little gets through that the exposure stretches to seconds or minutes, and the whole scene must hold perfectly still. Widen the hole for a brighter, faster picture and every point of the world spreads into a disc the size of the opening: the image blurs. Sharp or bright — a pinhole lets you pick one, never both.
Worse, you can't simply shrink your way to a perfect image. Squeeze the hole too far and the wave nature of light asserts itself: diffraction fans each pencil back out into a soft smear, so past a certain point a smaller hole makes things blurrier, not sharper. There is a single best pinhole size for a given box, and even it is only so-so. The way out is to stop blocking light and start bending it — an opening wide enough to gather a flood of photons, with something inside it that bends every ray from a point back to a single point. That something is a lens, and to build one we have to start with why light bends at all.
Light is a wave, and its speed depends on the material it moves through — slower in glass than in air. It slows because the passing wave jostles the electrons in the glass, and they re-radiate light a fraction out of step; original and echo add up to a wave that crawls along more slowly. The size of that slowdown is the refractive index n. When a wavefront meets glass at an angle, one edge slows before the other and the whole front pivots — the way a marching band wheels when its outer rank shortens stride. The frequency never changes, so as the wave slows, its wavelength compresses. And the slowdown depends faintly on colour — a loose thread we'll pull on later, when lenses start to smear white light into a rainbow:
The index is just a ratio: the speed in vacuum divided by the speed in the material. Air is 1.0003, near enough to nothing; water is 1.33; ordinary crown glass about 1.52, so light crosses a lens at roughly two-thirds of its usual pace. Diamond is 2.42, the hardest brake in common use. Small numbers, but every bend in every optical instrument is squeezed out of that single digit after the decimal point — which is why lens designers hunt exotic glasses the way metallurgists hunt alloys. Drag the incoming angle on the figure below and watch two things happen at once: the wavefronts crowd closer together as they cross into the glass, and the whole front swings toward the vertical.
A plane wave sweeping down into glass (lower half). The wavefronts crowd closer — the wave has slowed — and the whole front pivots toward the vertical as it crosses: the marching-band wheel that is Snell's law. Raise n₂ to slow it harder; set n₁ above n₂ and push the angle, and the wave can't get in at all — it decays into a thin evanescent skin, the near side of total internal reflection.
Follow a single ray and that pivot becomes Snell's law — n₁·sin θ₁ = n₂·sin θ₂. Going into denser glass the ray bends toward the normal; coming back out, past a critical angle, it can't escape at all. For ordinary glass that escape hatch slams shut at about 42° from the surface's normal — steeper than that and every ray stays trapped, bouncing along inside. That trapped light is total internal reflection, the trick that pipes a beam down an optical fibre for kilometres and makes the facets of a cut diamond blaze.
Why bend at all? Because light takes the quickest route, not the shortest one — and when part of the journey runs through slow glass, the fastest path is a dog-leg, exactly as a lifeguard sprinting down the beach before entering the water beats the one who swims the straight line. That single principle, stated by Fermat, hands back the whole law. It also explains the everyday illusions: a straw snapped at the waterline, a pool floor sitting higher than it really is. Rays leaving the water pivot away from the normal, and your eye, which insists light travels straight, traces them back to a bottom that isn't there. And the critical angle varies sharply with the material — 49° for water, 42° for glass, but only 24° for diamond. That last number is why a diamond glitters: its facets are cut so that light entering the crown strikes the rear faces well beyond 24° and is flung back out the top rather than leaking away.
Incident, reflected, and refracted rays — computed per-frame. Push the angle past critical and the refracted ray vanishes.
And not all the light even crosses the boundary — some reflects. The split depends on the angle: looking almost straight through, glass transmits nearly everything; at a glancing angle the same glass turns into a mirror. It's why a lake is transparent at your feet and silver at the horizon. Those stray reflections are pure loss, and they add up: a real lens stacks a dozen glass surfaces, each skimming off a few percent, until the image dims and ghosts of bright lights bounce around inside. The cure is a whisper-thin anti-reflection coating on every surface — the faint purple-green sheen on a lens's front element is that coating at work.
The arithmetic is worth doing. A bare air-glass surface reflects about 4% of light striking it head-on, which sounds negligible until you count surfaces: a modest zoom has twenty of them, and 0.96 multiplied twenty times leaves barely half the light to reach the sensor — the rest ricocheting inside the barrel as veiling haze. The coating beats it with interference rather than absorption. Make the film a quarter of a wavelength thick and the reflection off its front face comes back exactly out of step with the reflection off its back face; the two cancel, and the light that would have bounced goes through instead. A single layer can only cancel one wavelength perfectly, which is why the leftovers tint the glass violet or amber, and why good lenses stack several layers to flatten the whole visible band. There is one more wrinkle in the reflection: at a particular angle — around 56° for glass — the reflected light comes off completely polarized. Turn a polarizing filter against it and the glare off water or a shop window simply vanishes, leaving what lies beneath.
A lake at dusk. Look steeply down near your feet and the water is a dim window onto the bed; let your eye slide to the grazing horizon and the very same water becomes a mirror, catching the low sun and the sky. That whole transition — transparent to reflective — is the Fresnel split, laid out from near to far. Raise the water's index and the mirror creeps closer.
That bending isn't just an abstraction — you can watch it. Hold a slab of glass at an angle and everything behind it shifts sideways, because each ray bends going in and bends back coming out. But the two faces are parallel, so the ray leaves travelling in exactly the direction it arrived — only nudged over. A flat slab displaces the world; it can't focus it. The thicker the glass and the steeper the tilt, the bigger the nudge. Here the right half of the view looks through glass; the left half doesn't.
The shift is small but never nothing: a centimetre of window glass tilted at 45° slides the view about three millimetres sideways. That is enough to matter, and camera designers have to pay for it. Every sensor sits behind a stack of flats — a cover glass, an infrared filter, sometimes an anti-aliasing layer — and each one lengthens the optical path, pushing the point of sharp focus back by roughly a third of its thickness. A lens computed for one stack will miss focus on a body with a different one, which is why adapted lenses sometimes refuse to go properly sharp. The same tax is paid by any flat port on an underwater housing, and by the aquarium glass between you and the fish.
Notice what the slab cannot do. Because its faces are parallel, every ray leaves travelling exactly the direction it arrived — the picture is displaced but never gathered. Break that parallelism and everything changes: a wedge of glass sends the ray out at a genuinely new angle, deflected toward its thick end. Stack wedges of steadily increasing steepness, grind the steps smooth into a curve, and you have a surface that can aim every ray it catches at a single point. That is the whole idea behind the next section.
Watch the scene jump sideways at the seam — that shift is refraction. Tilt the glass or raise its index and it grows.
So how does a curved piece of glass focus? Think of it as a stack of prisms. A tilted slab just shifted the scene sideways; a prism, with its two faces set at an angle, bends light by a fixed amount that grows with the wedge. Stack prisms with steep wedges at the rim and gentle ones near the middle, and each ring bends its rays by just enough to aim them all at the same spot. Drag the segment count and watch a crude stack of facets sharpen into a true lens.
There is a quiet accident buried in that curve. The shape which would focus perfectly is not a sphere, yet almost every lens ever made is ground with spherical faces — because a sphere is what grinding naturally produces. Rub two glass discs together with grit between them and, whatever shape they started as, the high spots wear away until both surfaces settle into the one form that slides against itself in every direction: a sphere. Optics got the easy shape rather than the right one, and has been paying for the difference ever since. Almost every defect in the last section of this article is a bill for that convenience.
Parallel rays through a stack of glass prisms. With only a few segments each facet flings its rays to a different crossing — a smeared focus; add segments and the facets round into a smooth lens that pulls every ray to the one marked point.
In the limit — infinitely many infinitely thin prisms — the facets become a single smooth curve, and that curve is a lens. Now that every ray from a point lands back on one point, the messy stack collapses to a clean construction: trace just three principal rays and the image writes itself. Change the focal length or move the object and the image slides along the axis.
The three are worth knowing by name, because between them they pin the image down completely. A ray leaving the subject parallel to the axis must cross the focal point on the far side — that is what the focal point means. A ray aimed at the centre of the lens passes straight through undeviated, since there the two glass faces are parallel and we are back to the slab. And a ray that passes through the focal point on the near side leaves parallel, the first rule run backwards. Any two of them intersect at the image; the third is a free check that you drew it right.
Three principal rays converge at the image: parallel-in (amber) bends through the far focal point, the central ray (blue) passes straight through, and the near-focal ray (green) leaves parallel.
The bookkeeping collapses to one tidy equation — 1/f = 1/(object distance) + 1/(image distance) — and it explains the whole feel of a lens. A short focal length bends light hard and rakes a wide angle of the world onto the sensor; a long one bends gently and plucks a narrow slice from far away, which is why a telephoto seems to magnify and flatten. Because refocusing shifts the image distance, the framing breathes a little as you rack focus — and a fixed 'prime' lens, with one focal length, stays simpler and sharper than a 'zoom' that has to slide elements to change it.
Focal length itself is set by two things and nothing else: how sharply the faces curve, and how strongly the glass slows light. Curve them harder or choose a denser glass and the focus pulls in closer. The number quoted on a barrel is where the image of something very far away lands — 50 mm for the ordinary lens whose view roughly matches what your attention takes in, 14 mm for the wide that swallows a room, 400 mm for the telephoto that lifts a bird out of a distant tree. Focus, meanwhile, is a matter of distance rather than shape: your camera slides the whole assembly forward to bring near things in, while your eye, having no room to slide anything, squeezes its lens into a rounder shape instead.
And a lens doesn't image one point — it images all of them at once. Every point of the subject sends its own chief ray straight through the centre of the lens, undeviated, and lands on the far side; together those points paint the whole scene onto the image plane, flipped top-for-bottom and scaled. That scale is the magnification, −(image distance)/(object distance): watch the tick marks below stretch or shrink as you change the lens.
For anything more than a few focal lengths away this collapses to something you can do in your head: a subject's height on the sensor is its real height times the focal length, divided by its distance. A person one and three-quarter metres tall, standing ten metres off, measures under nine millimetres on the sensor through a 50 mm lens — better than a third of the frame's height. Which exposes a stubborn myth. A long lens does not compress perspective; only distance does. Photograph a face from half a metre and the nose, being proportionally much nearer than the ears, looms; stand back three metres and the proportions settle. The telephoto merely lets you stand back and still fill the frame — the flattery comes from where your feet are, not from the glass.
The whole field at once: each chief ray runs straight through the lens centre, so the subject (green) lands inverted on the image plane (amber). The tick spacing there is the object's, scaled by the magnification −(image distance)/(object distance) — rack the focal length up and the image swells like a telephoto; push the subject away and it shrinks toward the focal point.
That flat sketch hides the fuller truth. A point of the scene doesn't send three tidy rays — it sprays a whole cone of light in every direction, and the lens catches a disc of that cone and folds it back to a single point. Orbit the figure below: the rays form a real double cone, diverging from the subject and converging to its image. Slide the subject in and its image races the other way along the axis — that's the conjugate relationship, made of geometry. And open the aperture: the lens simply drinks a fatter cone. How much of that cone it grabs is the next dial we turn.
The width of that cone is the whole reason big lenses are big. A point source throws its light out in every direction, and the lens intercepts only the sliver its front element happens to cover — so the light collected grows with the area of that disc, not its width. Double the diameter and you gather four times the photons. Read that against the first section and the expense of good glass stops being mysterious: photons are the currency noise is paid in, a wider opening is the only honest way to get more of them, and everything else — longer exposures, higher gain — is either a compromise with motion or no new light at all. A heavy fast lens is, in the end, just a bigger bucket.
A single point sprays a cone of light; the lens (blue ring) catches a disc of it and folds it to one image point. Drag to orbit — the rays are a real double cone. Slide the subject (green) closer and its image races outward along the axis; open the aperture and the lens simply gathers a fatter cone.
A lens gathers light through an opening — the aperture — and that hole does two jobs at once. Its size is written as an f-number, which is literally the focal length divided by the opening's diameter: f/2 is a hole half the focal length across. Each full 'stop' along the familiar chain — f/2, f/2.8, f/4, f/5.6 — is a step of √2 in diameter that halves the opening's area, and so halves the light. A smaller f-number is the wider hole and the brighter image — and the wider the hole, the harder the corners fall away under the cos⁴ law we met at the pinhole.
Dividing by the focal length looks like a strange way to describe a hole until you see what it buys. A long lens spreads its image over a wider area, thinning the light; a wide opening concentrates it. The f-number folds both effects into one number, so that f/4 delivers the same brightness on a 24 mm lens as on a 400 mm one, and a light meter need never ask which lens is mounted. That is also why a fast lens grows so expensive at long focal lengths: f/2 on a 400 mm lens demands a front element a full 200 mm across, and glass of that size, ground to that accuracy, is priced like jewellery. And the blades that form the hole give it more than a size — they give it a shape.
The iris. A smaller f-number opens it wider; the blade count sets the shape of the hole — and of the bokeh.
The aperture's other job is focus. Only objects at one exact distance land as true points on the sensor; everything nearer or farther spreads into a small disc — the circle of confusion. As long as that disc stays smaller than a sensor pixel the eye reads it as sharp, so 'in focus' is really a zone, not a plane. Open the aperture wide and the discs grow fast, shrinking that zone to a sliver — the depth of field. It's the whole reason a portrait shot wide open melts its background to a smooth wash while a landscape at f/16 holds everything from the near flowers to the far ridge.
Notice that 'small enough' is a human verdict, not a physical one. The tolerance in general use — about three hundredths of a millimetre on a full-frame sensor — was reverse-engineered from what an unremarkable eye can resolve in a modest print at arm's length. Print it larger, or pixel-peep at full magnification, and the same negative has less depth of field than the tables promise; the physics never moved, only the standard of proof. Focus distance matters as much as the aperture, too — the zone stretches quickly as you focus farther out, until at the hyperfocal distance the far edge reaches infinity and everything from half that distance to the horizon passes as sharp, which is the landscape photographer's oldest trick.
Sensor size quietly sets the scale of all of it. Fill the frame with the same face on a small sensor and you must either stand back or fit a shorter lens, and both flatten the cone of light reaching any given point — so a phone at its widest aperture still holds far more in focus than a full-frame camera at the same f-number. This is why phones fake the effect: they estimate depth, then blur by arithmetic. It also explains the opposite complaint in macro work, where the lens sits so close that depth of field collapses to millimetres and an insect's eye is sharp while its wings are already gone.
Objects at different depths (left → far, right → near). Only those near the focus plane stay sharp; a wider aperture shrinks that zone. Slide focus to choose what's crisp.
And the points of light that fall well outside focus don't just blur — they bloom into bright discs the exact shape of the aperture. That's bokeh: a hexagon of blades paints hexagons of light; round blades paint circles. The reason is almost too simple — the blur disc is a picture of the opening itself, cast by a point too far out of focus to converge.
Which means bokeh reports on the lens's flaws as faithfully as on its blades. A disc with a bright rim and a hollow middle betrays residual spherical aberration and reads as busy and nervous; one that fades softly from the centre reads as creamy, and lens designers chase that quality knowing no measurement will ever capture it. Near the frame's edges the barrel itself clips the bundle and the circles are shaved into cat's-eye slivers leaning away from centre, which is what gives some fast lenses their swirling backgrounds. Stop down and those same straight blades turn bright points into stars, each blade edge diffracting light into a pair of spikes — an even number of blades overlaps its pairs into that many rays, an odd number cannot, and gives you twice as many.
Out-of-focus highlights take the aperture's shape. Open up for big, round bokeh; stop down or drop blades for small, polygonal ones.
A perfect lens would bend every ray from a point back to a single point. Real glass never quite manages it, and the ways it fails each have a name. Bend a lens from a simple spherical surface and the rays near its edge bend a touch too much — they cross the axis ahead of the central rays, smearing a point into a soft glow.
This is the grinding accident coming due. A sphere is not the surface that would focus a point perfectly, only the surface that is easy to make, and spherical aberration is the size of that discrepancy. Two escapes exist. Stop the lens down and the offending rim rays are simply blocked, which is why almost any lens sharpens up at f/8 — a fact so reliable that photographers treat it as a rule of thumb without knowing its cause. Or grind a surface that isn't spherical at all, flattening the profile toward the rim just enough to hold the rays back; aspheric elements were once ruinous to make and are now moulded by the million, which is much of why a modern kit lens outperforms a fine old prime wide open. Not everyone considered it a defect — portrait lenses were once prized for exactly this glow, and a few are still built to keep it.
Spherical aberration: edge rays bend too much and cross the axis early, so a point never comes to one sharp focus.
Slide off the axis and a new failure appears. Rays from a point out near the corner of the frame strike the lens lopsided, and each annular zone of the glass images them to a slightly different place and scale. The zones stack into a little comet — a bright head where the inner rays pile up, trailing a flare toward the frame edge. That's coma, and it's why cheap lenses smear the corners while the centre stays crisp.
The comet is where the name comes from, and astronomers are the ones who mind it most. Their subjects genuinely are points, so there is nothing else in the frame to hide behind: a field of stars with coma shows every one outside the centre wearing a little tail, all of them pointing outward like iron filings. It worsens sharply as the opening widens and as you move further off-axis, so stopping down tames it in the same stroke that tames spherical aberration — a poor bargain under a night sky, where the light being thrown away is the whole point. Lens designers instead fight it with symmetry, arranging elements front and back so that each half's coma cancels the other's.
Coma: an off-axis point images not as a dot but as a comet, each zone of the lens landing its own offset ring. The blue mark is where a perfect lens would focus it; push the point off-axis and the flare grows toward the corner.
Some flaws leave every point sharp yet still bend the picture. If the magnification isn't quite constant across the frame, straight lines stop being straight: when the centre is magnified more than the edges the frame swells into a barrel, and when the edges win it pinches into a pincushion. The grid below is perfectly focused — only its geometry is wrong.
Wide lenses tend to barrel and long ones to pincushion, and a zoom usually does both, barrelling at one end of its range and pinching at the other, passing through near-neutral somewhere in the middle. But distortion is the one defect on this list that software can genuinely undo, because nothing was lost: each point is still sharp and still somewhere definite, so the correction is a matter of moving pixels back where they belong. Blur destroys information; misplacement only hides it. Which is why manufacturers now design lenses knowing the profile will be applied in the camera — accepting visible barrelling as the price of a smaller, faster, cheaper design, and paying only in slightly stretched corners once the grid is straightened. Push the effect deliberately and it stops being a defect at all: a fisheye is barrel distortion embraced, trading straight lines for a field of view no rectilinear lens could reach.
Distortion warps geometry without blurring: drag toward barrel and the grid bulges out against the straight frame; drag toward pincushion and it pinches in. Every intersection is still a sharp point — only its place is wrong.
And colour has its own failure. Glass bends different wavelengths by different amounts — blue more than red — so a single white point becomes a tiny smear, its reds and blues focusing at different places. That's chromatic aberration, the coloured fringing along high-contrast edges — and it is the loose thread from the very first section, where the refractive index turned out to depend faintly on colour. A lens cannot help but be a weak prism.
It arrives in two flavours, and they behave quite differently. When the colours focus at different distances along the axis, the fringes sit in front of and behind the plane of focus — magenta on one side, green on the other — haloing the specular glints in a backlit scene. Stopping down helps, since a narrower cone lands a smaller disc whatever colour it is. The other kind spreads the colours sideways instead, so each wavelength is imaged at a slightly different size: the centre of the frame stays clean while the corners grow red and cyan seams along every hard edge. Closing the aperture does nothing to that one, because the geometry is unchanged — but software can lift it out, by rescaling the colour channels until the edges line up again.
Chromatic aberration: blue bends more than red, so one white ray fans into colours — the coloured fringing at high-contrast edges.
And here is the classic cure at work. One converging element of crown glass, cemented to a diverging element of flint, each chosen so its colour-spreading almost exactly cancels the other's. A single lens throws red and blue to different focal points; the doublet folds them back onto one. Slide the correction down to nothing and watch the coloured foci fan apart — that's the plain lens; slide it up and they collapse together.
The trick is subtler than it looks, because the two elements must disagree in just one respect. Flint glass spreads colour more than crown does relative to how strongly it bends light at all, so a weak flint element running backwards can undo the crown element's colour spread while cancelling only part of its focusing power. You are left with a lens that still converges, but converges every colour to nearly the same place. Nearly — two wavelengths can be brought together exactly, and the rest land close but not on top of them, a residue called the secondary spectrum. Chasing it further means three-element apochromats and exotic materials — fluorite, or the extra-low-dispersion glasses whose initials get printed on telephoto barrels precisely because buyers have learned to look for them.
Fighting all of these at once — spherical, coma, distortion, colour, and the curving of the focal plane itself — is most of what separates a cheap lens from an expensive one, and the reason a serious lens holds a dozen elements rather than one. Every surface added to cancel one defect perturbs the others and skims off a few more percent of the light, so the design is never solved, only balanced. Which is the quiet joke at the heart of the whole enterprise: we started with a hole in a wall, which had no aberrations whatsoever, and everything since has been an elaborate apology for making it bigger.
The achromatic doublet: parallel white light splits into colours, but the crown-and-flint pair bends them back toward a common focus. Turn the correction down and red and blue fan to separate focal points (a plain lens); turn it up and they meet.
So that's a camera, from the bottom up: light is a wave that slows in glass and bends; a lens shapes that bending to gather a scene onto a sensor; the aperture trades brightness against depth of field; and the whole art of lens design is fighting the ways glass refuses to cooperate. Everything else — autofocus, coatings, stabilisation — is refinement on top of these few ideas.