The window is part of the light

Every viewing condition this collection has modelled is a spectral power distribution and an arrangement of surfaces. That is enough for a reflector, because a reflector multiplies whatever arrives and the arrangement decides what arrives.

For a sheet with a brightener in it a third quantity is needed, and it is one nobody records: what the light came through on its way in.

The claim #

A specification agreed under a standard light is a specification about a laboratory. For a fluorescent sample the journey from the laboratory to wherever the goods are actually looked at passes through glass, and the glass removes the part of the spectrum the specification was most sensitive to.

• Ordinary soda-lime window glass stops below about 310 nanometres and passes most of the rest of the excitation band.
• Laminated glass stops almost to 380. The polyvinyl butyral interlayer that holds a windscreen together in a crash absorbs the whole of the ultraviolet-A, which is a safety feature with an optical side effect nobody was designing for.
• A filter sold to protect a print removes the band entirely — that is what it is for.
• The same coated sheet measures CIE whiteness 142 outdoors, 140 behind window glass, 110 behind laminated glass and 93 behind a filter. The extremes are ΔE00 10.0 apart, which is larger than the range the four standard measurement conditions cover.
• And the D50 viewing booth in which the specification was signed sits in the middle of that range, at 122 — closer to the window than to either extreme, and matching neither.

Three glazings, three different edges #

The three materials in an ordinary building do not differ by degree. They cut in three different places, and the places bracket the band.

Soda-lime glass, which is nearly every window, has an absorption edge set by iron impurities and by the silica network itself. It is opaque below about 300 nanometres, which is why nobody gets sunburnt through a closed window — sunburn is ultraviolet-B, at 280 to 315 — and it passes the great majority of the ultraviolet-A above 340. A sheet behind an ordinary window receives 63 per cent of its excitation from below 380 nanometres, against 65 outdoors: a difference of nothing.

Laminated glass is two sheets of the above with a polymer interlayer between them, and the interlayer absorbs to about 380. Windscreens have been laminated since the 1930s, and since roughly the 1990s a great deal of architectural glazing has been too. Behind it the sheet’s excitation share falls to 35 per cent, and its whiteness by more than thirty points.

A conservation filter is a coating or film with an absorber tuned to cut at 400 or beyond, sold to stop prints and textiles fading. Behind it a sheet’s excitation share is 4 per cent — essentially the ultraviolet-excluded measurement condition, achieved by accident in a gallery.

Six places, one sheet #

Putting numbers on it needs a set of places rather than a set of filters, because the light differs too.

Outdoors under D65 with no glazing at all is the reference: the sheet reaches W 142.0, the highest number it will ever produce, because unfiltered daylight is the most excitation-rich thing it will meet. Behind ordinary glass, 140.2 — the two are three and a half units of ΔE00 apart, which is real but small.

The D50 viewing booth puts it at 122.5. That is not because the booth is filtered; it is because D50 is warmer than D65 and has proportionally less short-wave power, and because a booth’s ultraviolet content is a specified property of the booth rather than a property of daylight.

Behind laminated glass, 109.8. Behind a museum filter, 93.4. Under a tungsten lamp, 102.9.

The tungsten row is the one worth pausing on, because it is the case everybody already knows about and misattributes. A brightened sheet under an incandescent lamp looks yellower than it did in daylight, and the usual explanation is that the lamp is warmer — which is true and is not the whole of it. The lamp is also ten times poorer in ultraviolet: 0.65 per cent of illuminant A’s power lies below 380 nanometres against 6.75 per cent of D65’s. Part of the yellowing is chromatic adaptation failing to keep up, and part of it is the sheet simply not glowing, and the two are separable by measurement and never separated by eye.

Why the booth cannot fix it #

The obvious response is that this is what viewing standards are for, and the response is right about the laboratory and wrong about the room.

A viewing booth has a specified spectral power distribution including its ultraviolet content, and two booths meeting the same standard agree with each other. That is a real achievement and it is what makes a printed proof signable at all.

What it cannot do is predict any other place. The booth is one point in the range above and the range is ten ΔE00 wide. Agreeing on a point does not narrow the range; it fixes where in the range the agreement sits, which is a different and much weaker guarantee. A specification signed in a D50 booth promises that the goods will look like the proof in a D50 booth, and says nothing about the shop, the office or the car.

That is a limitation the standards are explicit about and the industry is not. The reason it is tolerable for most work is that most colours are reflectances, and for a reflectance the booth’s promise transfers approximately — the illuminant changes, both samples change together, and adaptation removes most of what is left. For the paper white it does not transfer at all, because the paper’s excitation is not a property the adaptation has any access to — and the paper white is what everything printed is divided by.

The shop window, which is the worst case #

There is one place where the effect is largest and the stakes are commercial, and it is a shop window with goods displayed behind it.

The glass is often laminated, for security. Behind it is frequently a further ultraviolet film, to stop the merchandise fading — sold on exactly that basis, and correctly. The lighting inside is LED, which as a class emits nothing at all below about 400 nanometres, because the pump die is at 450.

So the goods in the window are lit by two sources with no excitation in either, while a customer standing on the pavement is in unfiltered daylight looking at their own clothes. A white shirt in the window and the same shirt on the customer’s back are two different whites, and the difference is a property of the building rather than of either shirt — a stronger version of what happens when the lamp in the shop decides a purchase.

The same argument applies to a print hung behind conservation glazing beside one hung bare — with the further irony that the filter is also slowing the rate at which the brightener is used up — and to a document held up to a window against one on a desk. In every case the direction is the same — filtered is duller and yellower — and the size runs to several ΔE00.

The window is one filter in a chain of three, and the last of them is inside the person doing the looking.

What would have to be written down #

The gap this leaves is a specification gap rather than a measurement one, and it is worth being concrete about what would close it.

A colour specification names an observer, an illuminant and a geometry, and since 2009 a printing one names a measurement condition. None of them names the place, and for a fluorescent sample the place carries more of the variance than the measurement condition does — ten ΔE00 against eight.

What a complete statement would need is a transmittance in front of the source: not a product name but an edge wavelength and a width, which are two numbers and describe every glazing well enough for this purpose. A specification saying measured under M₁; intended for viewing behind glazing cutting at 310 nanometres would be a complete sentence, and nothing prevents it being written.

The reason it is not written is that for every other quantity in the specification it would be redundant. A reflectance behind a window is the same reflectance; an ink’s density does not change; a geometry is a geometry. Glazing enters the answer for exactly one class of sample, so a field that is empty on ninety-nine specifications in a hundred does not survive a form design.

That is the general shape of the difficulty, and it is not solved by knowing about it. A parameter that matters for a small minority of cases is a parameter that gets dropped from the interchange format, and the minority then has no way to say what it means.

What was computed, and how #

Each glazing is a logistic edge at a stated wavelength with a stated width and a stated long-wave transmittance: 310 nanometres for soda-lime, 378 for laminated, 402 for a filter. Those are the standing figures for the materials and the shape is chosen — a real absorption edge is not a logistic and the fit is close enough for a band this wide.

The six places multiply an illuminant by a glazing and evaluate the bispectral sample under the product, each normalised to its own place’s perfect diffuser. So the numbers are what an observer adapted to that place would measure, not what an observer adapted to the booth would see there — which is the harder question and is the appearance one.

Whiteness is the CIE formula with the chromaticity of the perfect diffuser under the same light, so it too is computed in-place rather than referred back to D65. That matters: quoting whiteness against a fixed white point would fold the illuminant’s own colour into the answer and the tungsten row would be meaningless.

The assertion is a ratio rather than a level. The place supplying the most excitation must supply more than eight times what the place supplying the least does — a factor rather than a difference, because the claim is that glazing changes the kind of light rather than the amount of it.

Where the model stops #

The glazings are models of a material, not measurements of a product. Low-iron glass passes considerably more ultraviolet than ordinary float glass; low-emissivity coatings cut differently again; a modern windscreen with an infrared-reflecting layer has an edge nobody could guess from the material class.

Nothing here models the geometry of a window. Real daylight through a window is direct sun plus sky in a proportion that depends on orientation and time, and the sky is much bluer than the sun — which means the excitation available depends on whether the sheet is in a patch of sunlight or in the shade of the same room, by a factor this essay has not computed.

And the observer’s own adaptation is left out entirely. Every number here is a measurement. What a person makes of a sheet that is measurably duller in one room than another depends on what else is in the room, and that is an appearance question the colorimetric machinery cannot answer.

Who found it, and when #

That window glass blocks ultraviolet-B is common knowledge and is the reason a car does not give a driver sunburn. That it passes most of the ultraviolet-A is less well known and is why the driver’s left arm ages faster than their right in a country that drives on the right.

The consequence for brightened materials has been in the textile and paper literature for decades, usually stated as a caution about assessing whiteness near a window. The conservation literature approaches it from the other end: filters are specified for the fading they prevent, and the whiteness they also remove is a documented side effect that curators are advised about.

What is not standard is putting the two in the same table. The measurement conditions of ISO 13655 are a laboratory’s answer to which light; the six places above are the world’s, and the world’s range is wider.

The generalisation #

The pattern is a specification that fixes a point in a range it does not control, and then gets read as controlling the range.

A standard viewing condition is enormously valuable and its value is precise: it makes two observations comparable. Two laboratories with the same booth agree; a supplier and a customer with the same booth can settle a dispute. What it does not do is make either observation representative, and the difference is invisible while the quantity being observed transfers approximately from one condition to another.

The test is whether the standard’s own tolerance is smaller than the spread across the conditions the goods will actually meet. Here the standard’s spread is about 8 and the world’s is 10, so the standard is not narrowing anything — it is choosing where to stand. For a reflectance the same test gives a very different answer, which is why the arrangement works for almost everything and fails for this.

Whiteness is not proportional to the excitation share #

The four D65 rows can be read as a curve rather than as a list, because each carries both the share of excitation that arrives and the whiteness that results. The curve is not a straight line, and where it is steep is not where the essay puts its emphasis.

The middle segment is twice as costly per point of excitation as the bottom one. Removing the last thirty-one points of excitation — the whole step from laminated glass to a conservation filter, which is the most aggressive optical intervention in the list — costs 16.4 whiteness points. Removing the twenty-eight points before it costs 30.4.

That inverts which glazing a specification should worry about. A conservation filter is the dramatic case and it is operating on the flat tail of the curve: by the time the sheet is behind laminated glass, most of the whiteness a brightener can deliver has already gone, and the filter takes what is left at half rate. Laminated glass is the expensive one, and it is the case nobody thinks about, because it looks exactly like ordinary glass and has been standard in windscreens since the 1930s and in a great deal of architectural glazing since the 1990s.

The shape has a mechanism. At the top the brightener is close to saturated on the excitation available, so the last few per cent add little; at the bottom the sheet is approaching its own base reflectance, and no amount of further filtering can take it below a sheet with no brightener in it. The steep middle is the region where the emitted band is actually trading against the excitation, and it is exactly the region ordinary buildings occupy.

Where the booth sits, precisely #

Two statements about the booth’s position appear a few paragraphs apart and they do not agree: the claim section says it sits in the middle of that range, and the figure caption says it is not the middle of the range.

In whiteness units the booth’s 122.5 sits 59.9 per cent of the way from the filtered extreme at 93.4 to the unglazed one at 142.0 — 19.5 points below the top and 29.1 above the bottom. So it is above the midpoint and not by much, and both sentences overstate in opposite directions.

The practical reading survives either way and is worth stating in the form that does not depend on the rounding: every glazed viewing is a disappointment relative to the booth and every unglazed one is a bonus, with the disappointments running to twenty-nine points and the bonuses to twenty. The asymmetry is what matters commercially, because a specification is signed against the booth and the goods are then looked at somewhere with a window in it.

The standard does narrow the range, by a fifth #

The closing generalisation offers a test — is the standard’s own spread smaller than the spread across the conditions the goods will meet — and answers it with the standard is not narrowing anything.

On the essay’s own two figures it narrows it by 22 per cent: the four standard measurement conditions disagree by at most 7.8 ΔE00 across six stocks, and the six places disagree by 10.0 on one stock. Those are comparable quantities and 7.8 is the smaller.

That is a weak result and it is not nothing, and the difference matters for what the essay is recommending. If choosing a measurement condition achieved nothing, adding a glazing field would be the only repair worth making. Since it achieves about a fifth, the two are complementary: the measurement condition pins down what the instrument did, and the glazing field would pin down what the room will do, and the second is the larger term by about a quarter rather than being the whole of it.

Where the ladder goes next #

If the light a sample meets is a laboratory decision, then which laboratory decision is the question — and there are four standard answers. What an instrument brings with it is the census of those four and of what separates them.

The other direction is the camera, which sits in exactly this position and has no way to record which case it is in. A photograph cannot record the excitation, so two pictures of the same sheet under two lamps of the same colour differ by more than a white balance can remove.