The lamp that stopped emitting ultraviolet

Every argument in the essays around this one turns on how much light a source supplies below 380 nanometres. That quantity is a property of the lamp, and over the past two decades the population of lamps in the world changed almost completely.

The replacement was chosen for efficacy and for lifetime. Its short-wave behaviour was nobody’s design criterion and is the most abrupt discontinuity in the quantity.

The claim #

A phosphor-converted white LED emits nothing below about 420 nanometres, because its pump die is at 450. That is not a small number or a design variation — it is a structural zero, and it removed the excitation supply from indoor lighting in about fifteen years.

• Daylight puts 6.75 per cent of its power below 380 nanometres. A tungsten lamp puts 0.65 — ten times less — and a blue-pumped white LED puts none.
• The zero is architectural. The die emits at 450; a phosphor converts some of that to longer wavelengths. Nothing in the device produces a shorter photon than the die does.
• So a brightened sheet under LED light is measurably the sheet with its brightener switched off, at b* +0.6 against −11.7 under a xenon flash of nearly the same chromaticity.
• Every material designed against the old supply is affected and none of them was reformulated, because nothing failed — the sheets simply stopped doing indoors what they were bought for.
• And the change is invisible to every measurement anybody makes of a lamp: colour temperature, colour rendering index and efficacy are all computed on the visible band and are identical for a lamp with the ultraviolet and one without.

What each generation supplied #

The sources are worth setting out in order, because each has a different reason for its short-wave output.

Daylight supplies the most, and supplies it because the sun is hot and the atmosphere is only partly opaque. Of D65’s power, 6.75 per cent lies below 380 nanometres; of D50’s, 3.35. The excitation-weighted share is higher still — 66 per cent of what a brightener absorbs under D65 comes from below 380 — because the fluorophore’s absorption is concentrated exactly there.

An incandescent lamp supplies little and supplies it because Planck’s law has a tail. At 2856 K the spectral radiance at 350 nanometres is about a twentieth of what it is at 560, and 0.65 per cent of the lamp’s total falls below 380. That is small, non-zero, and computable — it is the one lamp in the list whose ultraviolet follows from a formula rather than from a device decision.

A fluorescent tube is the interesting historical case and this collection cannot compute it. The tube works by exciting mercury vapour, which emits hard in the ultraviolet at 254 nanometres, and coating the glass with a phosphor that converts that to visible light. The conversion is not complete and, as a lamp that is not a blackbody shows, the mercury also emits at 313, 334 and 365 nanometres — lines that are partly transmitted by the glass envelope. How much escapes is a property of the glass and the phosphor coating rather than of the physics, so it varies between tubes and is a manufacturer’s number. This collection’s fluorescent lamp model is built on the visible grid and has no ultraviolet in it at all, which understates a real tube.

A phosphor-converted white LED supplies none, and the zero is structural rather than incidental. An indium gallium nitride die emits at about 450 nanometres; a phosphor absorbs part of that and re-emits broadly through the yellow. Every photon in the output either came from the die at 450 or from the phosphor at longer wavelengths. There is nothing in the device that makes a photon shorter than 450, so the emission below 420 is not small — it is zero to whatever precision anybody cares to measure.

Neither of the artificial sources can be drawn there at all, and that is a fact about the tables rather than about the lamps.

Why no lamp measurement notices #

The striking thing about the change is that every standard figure of merit for a lamp is blind to it.

Correlated colour temperature is computed from the chromaticity, which is an integral against the colour-matching functions, which are zero below 360. Two lamps identical above 400 and differing arbitrarily below it have exactly the same colour temperature.

Colour rendering index compares how eight or fifteen test samples appear under the lamp against a reference. The test samples are reflectances — and none of them fluoresces, which is a documented limitation of the method and one of several reasons it is unsatisfactory. A lamp with no ultraviolet renders every one of them identically to a lamp with plenty.

Luminous efficacy is lumens per watt, and lumens are the visible integral. Ultraviolet output counts against efficacy in the denominator, which means the incentive runs one way: a lamp that emits ultraviolet is penalised for it.

So there is no figure on a lamp’s data sheet that would have flagged the change. Nothing was measured, nothing was reported and nothing regressed, because the quantity had never been on the sheet at all.

The one lamp with a formula, checked #

Illuminant A is the only source in the list whose short-wave output follows from an equation rather than from a device, which makes it the one row of this essay that can be recomputed from outside. Evaluating Planck’s law at 2856 K:

The ratio at the two wavelengths comes out at 1 : 21.1, against the essay’s about a twentieth — the spectral radiance at 350 nanometres against 560. And the power below 380 nanometres, as a share of 300 to 780, comes out at 0.75 per cent against the published 0.65, which is the sort of gap a rectangle sum and a trapezoid produce between them and is not a disagreement about the lamp.

That matters more than a checked number usually does, because the whole essay rests on a comparison between a quantity that can be computed for one lamp, is quoted for another, and is asserted to be zero for a third. The one that can be computed is right.

Where a tungsten lamp’s excitation actually comes from #

Running the same source against a brightener’s absorption profile gives a figure the essay computes for daylight and not for tungsten, and the two do not agree in the way the power shares would suggest.

Under illuminant A, 89 per cent of what a brightener absorbs comes from below 380 nanometres. Under D65 the published figure is 66. So the lamp supplying ten times less short-wave power draws a substantially larger share of its excitation from below the 380 boundary — the opposite of the ordering the power shares give.

The reason is the atmosphere, and it is visible as soon as a blackbody is used as a control. A 6504 K Planck radiator — daylight’s colour temperature with no sky in front of it — gives 94 per cent, even further above D65’s 66. Real daylight is cut off at the short end by ozone and scattering and has very little left below about 320 nanometres, while a blackbody’s tail runs smoothly down through the whole range. So daylight’s excitation is squeezed into a narrower window near the boundary, and a third of it ends up above 380.

Two consequences follow, and both sharpen the essay rather than qualify it.

The 380-nanometre line is a good boundary for a thermal source and a poor one for daylight. A third of what a brightener absorbs outdoors arrives in the violet, inside the visible band, where every one of this collection’s other calculations already runs. The band that was “assumed away” is not entirely outside the visible after all.

And that makes the LED’s zero worse rather than better. A die at 450 nanometres supplies nothing below about 420, which removes not only the ultraviolet the essay is named for but also the violet third that daylight actually delivers most of its excitation through. The relevant statement is not that a white LED has no ultraviolet; it is that it has nothing anywhere in the brightener’s band, and the two are different claims with different boundaries.

The Stokes cost tracks the lamp, and by about a point #

One more quantity falls out of the same integral, and it closes a loop with the neighbouring essay on the Donaldson matrix.

The photon-weighted mean wavelength at which a brightener actually absorbs is 364.6 nanometres under illuminant A and 359.5 under a 6504 K blackbody — the hotter source pulling the absorbed photons towards the short end of the band, as it must. Against an emission at 435 nanometres those give energy returns of 83.8 and 82.6 per cent.

So the Stokes loss is not one number: it runs from about 16 per cent under a hot source to about 14 under a warm one, purely through which part of its own band the fluorophore is being fed from. The site’s own computed figure of 84.3 per cent sits just beyond the tungsten end, which is what real daylight’s short-wave cutoff would produce — the atmosphere removing the very photons that would have cost the most energy to convert.

A brightener is therefore slightly more efficient, in energy terms, under a lamp that excites it less. That is a small effect and an entirely real one, and it is the same lesson this essay is about arriving from underneath: a number that reads as a constant of the material is a property of the material and whatever is shining on it.

What it changed #

Brightened materials stopped working indoors. This is the largest effect and the hardest to see, because it happened everywhere at once and gradually. A shirt, a sheet of office paper, a laminate worktop and a plastic housing all contain brighteners chosen against a lighting environment that has since changed. Under LED light they are the base material with an inert additive in it, and the lamp in the shop decides what a buyer ever sees of the difference.

The trade’s own inspection changed. A quality-control lamp that supplies no excitation cannot show a brightening defect. An inspector who has always judged whiteness under a fluorescent tube and now judges it under an LED panel is judging a different quantity, and would need a lamp with a specified ultraviolet content to get the old one back.

Ultraviolet security features became invisible without a dedicated lamp, which they always partly needed and now need entirely. A banknote’s fluorescent fibres and a passport’s printed features are designed to be checked under a purpose-built 365-nanometre source; what changed is that ambient light no longer provides even a hint of them, so a feature that used to glimmer faintly under a fluorescent tube now shows nothing at all until the dedicated lamp is switched on.

And photography inherited all of it. A camera cannot record the excitation, so a photograph taken with a xenon flash and one taken under LED panels differ on white materials by more than any white balance can remove — and the two lighting kits are indistinguishable by every measurement a photographer routinely makes.

What would put it back #

Three things exist and none is common.

A violet-pumped LED uses a die at about 405 nanometres with a full phosphor set, and produces a little excitation — not much, because the die is above the brightener’s peak, but not zero. Such lamps are sold for colour-critical work and for their better rendering of deep blues.

An added ultraviolet emitter is what a viewing booth does: a separate small source with its output specified as a fraction, so that the booth can meet a standard for measuring fluorescent samples. That is a deliberate design and it costs a component.

And a fluorescent tube still does it by accident, which is one of the few respects in which the technology being replaced was better at something.

The honest summary is that the excitation supply is now a specified quantity in the few places anybody specifies it, and absent everywhere else — which is a change from a world where it was unspecified everywhere and present nearly everywhere.

Why nobody chose it #

The change was not a decision about ultraviolet at any point, and it is worth tracing how a quantity gets removed from the world without anybody removing it.

An incandescent lamp emits short-wave light because it is hot and Planck’s law has a tail; nobody put it there. A fluorescent tube emits it because mercury does and the phosphor coating is imperfect; the design intent was to convert all of the 254-nanometre line, and the lines that escape are a shortfall rather than a feature. A white LED emits none because its die is at 450 and a down-converting phosphor cannot make a shorter photon.

So the quantity went from an accident, through a shortfall, to a structural zero, and at no stage was it a specification. Each generation of lamp was chosen against efficacy, lifetime, colour temperature and rendering, and every one of those four is computed on the visible band.

The materials on the other side of the transaction were formulated against the world as it was. A brightener’s excitation band was chosen in the 1950s against sunlight, incandescent and fluorescent light — all of which supply some — and the choice was optimal against that environment. Nothing recorded the assumption, because it was not an assumption.

What is left is a coupling nobody owns. The lamp industry’s specifications do not mention the band; the additive industry’s do not mention the lamp; and the property that depended on both is a property of the pair.

What was computed, and how #

The ultraviolet shares are integrals of each illuminant below 380 nanometres divided by its integral to 780, on the wide grid. Daylight is the CIE reconstruction from the basis functions the CIE publishes from 300; illuminant A is Planck’s law at 2856 K, which extends exactly because it is a formula.

The excitation-weighted share is the same integral with the brightener’s own absorption as a weight, which is the quantity that decides how much a sheet glows and is different from the power share — 66 per cent against 6.75, because the fluorophore’s absorption is concentrated in exactly the band the power share dilutes.

The white LED is this collection’s own lamp model with its phosphor conversion set to land near daylight chromaticity, and with nothing below 380. That zero is asserted rather than computed and is a claim about the device: a die at 450 emits no shorter photons, and the phosphor is a down-converter.

The fluorescent tube’s ultraviolet is declared missing rather than estimated. The model is built on the visible grid and the escaping mercury lines are a property of a particular envelope and coating. Putting a number on them would mean inventing one, and an argument whose whole subject is a band that was assumed away is the wrong place to assume a band.

Where the model stops #

The LED zero is a structural claim, not a measurement. Real packages have a small amount of short-wave emission from imperfect filtering and from the die’s own tail, and a few per cent of the die’s output can sit below 440. It is not enough to excite a brightener meaningfully and it is not exactly zero.

Fluorescent tubes are not in the comparison for the reason above, which leaves a hole in the middle of the historical argument. What the transition removed is bracketed rather than measured: more than an LED supplies and less than daylight does.

And the population claim is not computed at all. That LEDs replaced most indoor lighting between 2005 and 2020 is an observation about the world, not a result from this collection’s machinery, and nothing here quantifies how much of the world’s indoor light is now blue-pumped.

One filter in this chain sits inside the observer rather than in the room, and it is the only one nobody can specify.

Who found it, and when #

Blue-pumped white LEDs date from Nakamura’s gallium nitride work in the early 1990s and reached general lighting in the 2000s. The regulatory phase-out of incandescent lamps ran through the 2010s in most of the world, and fluorescent tubes followed.

That brightened materials look different under LED light is well known in the paper and textile trades and is usually described as LEDs “not having UV”, which is exactly right and is not usually followed up with a number. The viewing-booth standards responded by specifying ultraviolet content, which is the same recognition arriving in the one place where it could be written down.

The security-printing case is the one where the change was noticed immediately, because a feature stopping working is a failure and a sheet of paper looking slightly duller is not.

The generalisation #

The pattern is an environmental constant that was never a specification and then changed.

Materials get designed against the world as it is. A brightener’s excitation band was chosen in the 1950s against sunlight and incandescent and fluorescent light, all of which supply some short-wave power, and the choice was optimal against that environment. Nothing recorded the assumption because it was not an assumption — it was the world.

When the environment changed, nothing failed. The materials still meet their specifications; the lamps still meet theirs; every measurement on both sides is unchanged. What was lost is a quantity that lived in neither specification, because it was a property of the pair.

The diagnostic worth carrying is to ask, of any long-standing product, which environmental quantities it depends on that nobody measures — and then whether anything is currently changing them. The answer is usually short and occasionally alarming, and it is never available from either party’s data sheet.

Where the ladder goes next #

If the supply is gone from the room, the remaining supply is outdoors and behind whatever glass a building has, which makes the window part of the light in a stronger sense than before — it is now the only route by which many materials receive any excitation at all.

The other direction is time. The brightener that is no longer excited is also no longer being destroyed, so the clock it runs on has slowed with the lighting, and a sheet stored under LED light keeps a whiteness it can no longer display.