The corridor of mirrors that turns green
Assumes: The absorption measured with a stopwatch · Why a litre of water is not blue for the reason the sky is
Stand between two large mirrors facing each other — in a lift, a changing room, a dance studio — and look into one of them. A corridor of reflections runs away into the glass, each image a copy of the last, smaller with distance. Two things happen along it that a single mirror never shows. The images get darker, quickly, so that after a few dozen they are lost. And they change colour: the deep images are distinctly green, as if the corridor were lit by a different lamp from the room.
Neither mirror looks green. A household mirror is the standard of a colourless surface; nobody would call one tinted. The colour has to be in every mirror, too small to see in one reflection, and the corridor must be a device for making it large. It is, and the device is simple: a weak preference multiplied by itself. What is being multiplied turns out to be not the silver, which is what one would guess, but the glass.
What one reflection keeps
A household mirror is a sheet of glass, usually four to six millimetres thick, with silver deposited on the back and protected by paint. Light reflected in it does three things. It meets the front surface of the glass, where about four per cent is reflected at once — the faint ghost image visible at the edge of a thick mirror — and the rest enters. It crosses the glass, is reflected by the silver, and crosses the glass again. It meets the front surface from inside, loses another four per cent, and leaves. The image in the mirror is made of what survives all of that.
Each step has a different dependence on colour. The front-surface reflection is set by the glass’s refractive index, which varies only slightly across the visible, the variation two glasses that cancel a derivative worked to cancel in a lens; for reflection it is negligible. Silver reflects about 97 per cent in the blue and a little more, about 98, in the red, falling off in the violet below about 400 nanometres. And the glass, crossed twice, absorbs a little — differently at different wavelengths.
Ordinary clear float glass contains about a tenth of a per cent of iron oxide, an impurity from the sand. Some of the iron is ferrous, Fe²⁺, which has a broad absorption band centred in the near infrared around 1,050 nanometres; its short-wavelength tail reaches into the red. Some is ferric, Fe³⁺, which with the glass’s own ultraviolet edge absorbs at the violet end. The visible window between them is centred in the green. Looking at the edge of a sheet of window glass, through many centimetres of it, shows exactly this colour; it is why glass edges are green and why glass shelves look greener the thicker they are.
The figure uses a stated model of this glass, four millimetres thick, built to match the absorption usually quoted for clear float. With it, one reflection keeps 87.7 per cent at its best, at 543 nanometres, and 86.8 per cent at 650 — a difference of about one per cent of the light. The eye cannot see a one per cent colour difference between two sides of a scene; it adapts to whatever illumination it is given and calls the mirror white.
Raising a spectrum to a power
In the corridor, the light that forms the $n$th image has been reflected times, and at each reflection it has been multiplied by the same curve. Its spectrum is the single reflection’s spectrum raised to the $n$th power.
Raising a function to a high power does a characteristic thing to it. Wherever the function is slightly less than its maximum, the power drives it towards zero relative to the maximum; the peak survives and everything else fades. A gentle hump becomes a sharp one. Near its peak any smooth curve is a parabola in its logarithm, so the single reflection’s spectrum is locally a broad Gaussian, and its $n$th power is a Gaussian whose width is smaller by . A hundred reflections make the band ten times narrower than the single reflection’s curvature would suggest, and the colours far from 543 nanometres — red, violet — are suppressed by factors of three or more.
This is the same arithmetic as the absorption measured with a stopwatch, where a pulse of light bouncing between two very good mirrors made a tiny absorption in the gas between them measurable by the time the pulse took to die away. There the mirrors were made as perfect as possible and the multiplication was used to weigh what lay between them. Here the mirrors are household ones, and what the multiplication weighs is the mirror. The corridor is a cavity ring-down instrument built from a bathroom, reading the iron content of its own glass.
The corridor dims faster than it colours
Most of what each reflection loses is not colour-dependent at all. Eight per cent goes at the two front-surface passes, two or three at the silver, two or three in the glass. The colour is a difference of one per cent riding on a loss of twelve.
So the images fade much faster than they change colour. Each is an eighth dimmer than the one before; the thirty-fourth is a hundredth as bright as the scene, and the sixty-ninth a ten-thousandth. The eye’s ability to see detail across a range of brightness in one view is of that order, a thousand to ten thousand, and the corridor is lost into darkness somewhere between the fortieth and the seventieth image — sooner if the room is bright and the far images are swamped by glare from the near ones. The colour becomes clear in the same range, because by forty reflections the red has fallen to two-thirds of the green. The corridor is therefore seen to turn green just as it disappears, which is part of why the effect looks mysterious: the green images are always the faint ones.
There is a second geometric fact about the corridor that is easy to see and easy to misattribute. Mirrors on opposite walls are never quite parallel, and each pair of reflections turns the line of sight by twice the angle between them. A misalignment of a tenth of a degree turns the corridor by two-tenths of a degree per pair, and after thirty pairs by six degrees: the corridor curves away to one side. The two mirrors that hold a beam found the condition for a pair of curved mirrors to hold a ray near its axis indefinitely; flat mirrors are the marginal case of that condition, and any tilt walks the ray off. The curving corridor and the green corridor are both cumulative effects of something too small to see once.
The colour walks one way
The spectrum after many reflections can be turned into a colour by the standard rules of colorimetry, weighting it with the three colour-matching functions that summarise human colour vision and plotting the result on the chromaticity diagram.
The colour walks steadily away from white in one direction, towards a yellowish green, and the walk is nearly a straight line because each step is the same multiplication. The direction depends on the balance of losses at the two ends of the spectrum: the iron takes more red, and the silver and the glass’s ultraviolet edge take more violet and blue, and in this model the violet loss is large enough to pull the colour slightly towards yellow as well as green. Real mirrors, with different glass and different backings, walk in a direction somewhere between pure green and this one, and the deep images of real corridors are consistently described as green. The chart is drawn for light of equal energy at every wavelength; a room lit by warm lamps starts from a yellower white and walks from there.
The chromaticity walk is an instance of something why a litre of water is not blue described: a substance whose absorption is very weak in the visible looks colourless in small amounts and colours light only through great thicknesses. Water’s faint absorption of red, from overtones of its molecular vibrations, is invisible in a glass and makes a swimming pool cyan; glass’s faint absorption of red from iron is invisible in a window and makes a mirror corridor green. In both, the colour is a property of the material that only depth reveals.
The green is the iron’s
If the colour is the glass’s, mirrors made from different glass should make corridors of different colours. They do.
Low-iron glass, made from purer sand and sold for display cases and solar panels precisely because it has no green edge, makes a corridor whose spectrum after fifty reflections is nearly flat across the red and peaks well towards it, because with the iron gone the silver’s own slight preference for long wavelengths is what remains. Glass with three times the usual iron makes a corridor that is plainly green after a few dozen images. The colour of the corridor is a measurement of the iron in its glass, read by counting reflections, and a corridor of front-surface mirrors — silver or aluminium on the front, with no glass in the light’s path — would show a quite different, much weaker tint set by the metal alone.
The same reasoning tells a glazier how to read an edge. A sheet of glass seen face on is a few millimetres thick and looks clear; seen edge on it is tens of centimetres thick, the light having travelled along it, and the colour of that edge is the colour of the corridor at a comparable depth. A green edge predicts a green corridor, a nearly white edge a nearly neutral one, and the shop that sells mirrors made of low-iron glass is selling, among other things, a corridor that fades without changing colour.
That is the surprising part. The mirror’s reflecting layer, which is what makes it a mirror, contributes least to the colour. The colour comes from the protective sheet in front, the part of the mirror that is supposed to be optically absent, and it comes from the same impurity that is responsible for much of the green of bottle glass, which carries far more of it alongside other colouring metals.
Where the twelve per cent goes, and how a better mirror would spend it
The loss per reflection is worth taking apart once more, because each part has a remedy, and the remedies show what kind of instrument a corridor could become. The largest single loss, eight per cent, is the front surface: the glass’s two passes through a boundary between air and a material of index 1.5, each reflecting per cent. That loss is the one the layer that makes a reflection vanish removes from camera lenses with a coating a quarter of a wavelength thick, and a coated mirror would lose under one per cent there. It would make the corridor far deeper — at 95 per cent a reflection, a hundredth of the brightness is reached after ninety images instead of thirty-four — without changing its colour, because the coating’s loss is nearly flat across the visible.
The next loss is the silver’s, two or three per cent, and its colour is faintly warm. Aluminium, the usual alternative, reflects about 90 per cent with a shallow dip in the near infrared and a slight bluish cast; a corridor of aluminium mirrors would walk, slowly, in a different direction. The glass’s loss, the one that carries the green, scales with the glass’s thickness: a six-millimetre mirror makes a greener corridor than a four-millimetre one, by half as much again per reflection, and a mirror whose glass is very thin or absent shows no green at all.
Each of these is a version of the rule that governs absorption everywhere, that the fraction surviving a path falls exponentially with its length, which how far the air lets a mountain be seen followed over tens of kilometres of air. The corridor folds a long path into a short room. Light that reaches the hundredth image has crossed eighty centimetres of glass — the thickness through which a pane’s edge is seen when one looks along it — and has been folded a hundred times to do it, the same trick by which the colour at which glass disappears used a glass rod’s faint dispersion to reveal it against oil: a property too weak to see directly, made visible by arranging for it to act many times.
What the model cannot show
The glass here is a model, not a measured sample: a flat background absorption, a Gaussian band for ferrous iron and an exponential edge in the violet, with magnitudes chosen to match the absorption usually quoted for clear float glass. Real glasses vary in iron content and in the balance between ferrous and ferric iron, and the silver’s reflectance depends on how it was deposited and protected. The direction of the colour walk is sensitive to the balance between the losses at the two ends of the spectrum, and the model puts slightly more violet loss into it than some real mirrors have, which is why its walk leans towards yellow.
The figures also treat every reflection as happening at normal incidence. In a real corridor the light that forms the deep images travels at a slight angle, crossing the glass on a slightly longer path, and the front surface’s reflection changes with angle and polarisation in the way the angle at which reflection picks a side described; at the small angles of a mirror corridor these are corrections of a per cent or so to a per cent effect. The ghost reflections from the front surface, which make their own fainter corridor slightly offset from the main one, are ignored. And the colour is computed for the standard observer under light of equal energy, which is a convention rather than a room.
Finally, the eye’s judgement of colour depends on context. A faint image seen beside bright ones is judged against them, and a dark green seen in isolation might be called grey. The chromaticity figure says what the light is; what an observer reports also depends on what surrounds it.
Still open: how well a corridor could measure a mirror
A ring-down measurement between laboratory mirrors resolves losses of a few parts per million per reflection. A household corridor, read by eye, resolves a colour shift of a few per cent after tens of reflections, which corresponds to a per-reflection spectral difference of a few tenths of a per cent. With a camera and a spectrometer looking down a corridor, the brightness and spectrum of each image could be measured and the single reflection’s spectrum recovered from their ratios, with the advantage that the multiplication is built into the geometry. How far such a measurement could be pushed — whether the ghost images, the angles and the mirrors’ small curvatures could be modelled well enough to read a glass’s iron content to a part in ten thousand from a hallway — has not, as far as is known here, been worked out.
One reflection in a silvered glass mirror keeps about 87.7% of green light at 543 nm and 86.8% of red at 650 nm — a one per cent bias the eye cannot see — but each image in a corridor of facing mirrors is the last one reflected again, so its spectrum is the single reflection raised to the nth power, narrowing as around the green until red is at two-thirds after forty reflections, just as the images fall to a hundredth of the scene’s brightness; and the bias is not the silver’s, which slightly favours red, but the glass’s, where ferrous iron absorbs the red, so that low-iron glass makes a corridor that stays nearly neutral. The corridor is a cavity built from a bathroom, and what it measures is the window glass in front of the mirror.
Part 10 of 10
This essay is one argument about Dispersion. The others:
What links here
Essays that reach for this one mid-argument — the half of a link its own author cannot write down.
The objects named here
The third axis, after the field and the reading path: the things themselves, and every essay that touches each one.
Absorption coefficientBeer lambert lawChromaticityFresnel reflectionReflectance