The thread: Three numbers — page 12
What light is
A stricter fit declares a wider line
A two-nanometre table of a mercury lamp whose every line is self-reversed admits every reversed line as a Gaussian within half a per cent, and declares a width its own ceiling cannot stand on from a depth of two and a half. The obvious repair is to fit harder. The prediction was that at a tenth of a per cent the table refuses the reversed blue and green lines, admits the least-reversed yellow, declares unsafely from it between — and that a threshold that strict would refuse even unreversed lines on a real instrument's noisy table. The refusals come where predicted: the blue from a depth of three, the green from four. But the stricter threshold makes the declaration less safe, not more. It narrows the range of widths each line admits from below, so the yellow, never refused, declares a wider width at every depth, and the ceiling built on it falls below the true peak from a depth of two instead of two and a half. And a tenth of a per cent survives a quiet instrument: no unreversed line is refused until the table's noise reaches half a per cent — and then the first line refused is the yellow, the one that declares.
What a scene does
The slot's loss survives every finer division
Read on the painted faces alone, a satin finish on a room's grey faces makes the paint more colourful in twenty-eight of the painted census's thirty rooms and less colourful in two: the tall narrow slot with its floor painted and with its ceiling painted, where the paint loses about a twentieth of a per cent of its chroma. A loss that small could belong to how finely the slot's faces were divided, and the prediction was that it does not — that the sign holds at seven and eight strips a side. It holds. Solved from one strip a side to eight, the slot's paint gains at one, two and three strips and loses from four on, every one of seventy-two paints crossing at the same division; the loss deepens at every finer division, to −0.055 and −0.059 per cent at eight strips, and run out to no strip width it settles near −0.058 and −0.065. The coarse census had the sign wrong; the census at six strips had it right. The bright, broad paints lose most, as they gained most everywhere else.
What a scene does
The finish stops costing the paint six tenths deep
In the tall narrow slot a satin finish on the grey faces costs a painted floor or ceiling a little of its chroma; in the tall room, the same box a full unit deep, it adds some. Solved at eight depths between, the sign turns at 0.62 of the box's width with the floor painted and at 0.63 with the ceiling — later than the prediction of 0.45 to 0.55, and far enough along that the loss belongs to a range of shallow boxes rather than to the slot alone. A finer division moves the turn a little deeper, not back. Every one of seventy-two paints turns at the same depth, and each paint's share of the change is the same at every depth, loss or gain. Brightness does not decide it: the finish brightens the painted floor by between half and more than double while hardly touching the painted ceiling, yet the two turn together. What decides it is which reflectance the finish favours — whether it multiplies the light that has been back to the paint by more or by less than the light that has never touched it.
What a camera does
LEDs varied in their phosphors fall off every line
The merged smooth family estimates a camera's colour error from an ambient-light sensor's departures to ±0.127, and five new white LEDs, varied in their phosphors where the census's forty-four vary their pump, missed by ±0.196 because the family's line had learned almost nothing about the sensor's 640 nm channel. Rebuilt on forty-eight LEDs varied in pump, converter and red phosphor, the line learns it: its 640 nm weight goes from −0.13 to −0.74, most of the way to the −0.85 the new LEDs implied. The prediction was that the new LEDs would then land within ±0.13. They miss by ±0.229, worse than before, crossing to the other side of the line, and the census's own held-out error more than doubles, to ±0.294. No pool of LEDs restores it, and no line in the sensor's departures reads the grid's own LEDs better than ±0.29. Across the grid the camera's error is a V about the Planckian locus, rising away from it on the green side and on the pink — and every one of the census's LEDs sat on the green side.
What light is
Two nanometres hide the yellow's notch from any fit
A plain Gaussian fit admits the reversed tube's 578 nm yellow line at every threshold on a two-nanometre table, and the ceiling built on its width fails from a column depth of two and a half. A fit with the notch in it — an emitted Gaussian with an absorbed one taken out of its centre — might see what the plain fit cannot. The prediction was that two nanometres leave too few entries to determine the notch, and one nanometre finds it from a depth of two. Two nanometres hide it completely: the best notch fit carries a negative notch, a spike, at every depth, and the readings a two-standard-error rule calls reversed are the same noise draws at every depth, each fitting a notch that erases the whole line. One nanometre finds it sooner than predicted — in every reading from a depth of one and half of them at a half — and its fitted depth climbs with the column. But its standard error cannot be trusted with seven entries and four numbers: the rule calls a third or more of unreversed readings reversed. The notch's fitted depth itself separates them cleanly.
Where the model breaks
A harder search pulls the band pigments away from the slope limit
Every count in the pigment-limit censuses — how many directions of the object-colour solid a slow pigment cannot reach within one per cent of the ideal — compares a slope limit solved exactly with two band models searched from a few structured starts. Under the mercury lamp a search from 680 starts had already found the band model reaching further in seventeen of twenty-five directions. Searched from 594 starts under all nine lamps, both band models reach further in between three fifths and nineteen twentieths of the directions they lost, and their counts fall by a sixth: 395 to 335 for bands in nanometres, 370 to 311 for bands in energy, while the slope limit stays at 463. The rankings the censuses reported hold where they were stated. What moves is everything measured against the slope limit: its lead over the band models grows from a sixth to more than a third, the coating's odd ranking — bands losing more than the slope limit — disappears, the triphosphor tube passes the three emitters for bands in nanometres, and under the mercury lamp the band models' deep losses halve.
What a scene does
A taller box must be deeper before the finish adds colour
In a box one unit wide and two and a half tall, a satin finish on the grey faces costs a painted floor a little of its chroma when the box is shallower than 0.62 of its width and adds some when it is deeper. Whether that turn is set by the width or by the height was the open question, and the prediction was that it scales with the height, because the satin walls pipe the lamp down to the floor and a taller box is a longer pipe. It does, and faster than in proportion. A box 1.5 tall gains at every depth from 0.35 and turns shallower than that; a box 4 tall loses at every depth to half again its width and turns near 1.67. As a share of the height the turn rises from about 0.17 to 0.25 to 0.42 — close to a depth growing as the square of the height. The taller box pipes far more of the lamp to the floor, tripling its light at 0.35 deep, and at every depth of every box the paint's colour still follows which reflectance the finish favours. Six strips a side keep both new boxes' signs.
What the eye does
Knowing the macula's density frees two sessions, not the line on its peak
Fitted with six numbers, an older observer's colour matches cannot fully split a lens band at 445 nm from a macula whose peak has moved: their correlation stays above 0.78 under every third session tried. The proposal was to measure the macular pigment's density from outside the matches, leaving five numbers, and the prediction was that a line on the macula's peak would then split the pair below 0.6 while two plain sessions stayed near 0.8. It runs the other way. With the density known exactly, two sessions fall from 0.83 to 0.62 and the line at 465 nm only from 0.86 to 0.74. The density was the road the band and the peak traded along — correlated with each at about −0.75 in two sessions' fit — and the designs that had already narrowed that road gain least from closing it. A realistic outside measurement closes almost none of it: the matches already know the density to 0.027, so a flicker measurement good to 0.05 moves the correlation by three hundredths. And far from the macula, at 415 nm, a known density binds the band to the peak more tightly than before.
Matching and measuring
The green primary decides how sharp the blue is
ΔEITP makes a display's blue its sharpest corner when the blue sits near 485 nm and nearly a face when it sits at 440, and the turn between them is where adding green stops raising the blue's L signal by the larger fraction. The prediction was that the turn follows the green — near 445 nm with a 545 nm green, near 470 with a 515 nm one. It follows the green, by less than half as much at the short end: 461 nm with a 515 nm green, 446 with a 545 nm one, and with a 555 nm green there is no turn left in the range at all. What the prediction did not ask is larger. The green sets how sharp the blue can be: with the blue at 485 nm the corner moves 2.4 times a margin against a 505 nm green and 8.2 times against a 555 nm one, while the sharpest blue stays at 485 nm for every green. The red primary, moved from 615 to 690 nm, leaves the turn where it was to a hundredth of a nanometre. The appearance model follows the green far more gently, under half of ΔEITP's move at every green.
What a scene does
The width moves the turn only when it is the narrower side
In a box one unit wide and two and a half tall, a satin finish on the grey faces costs a painted floor a little chroma when the box is shallower than about 0.64 and adds some when it is deeper. The prediction was that the width hardly matters — that the two end walls a depth apart form the pipe that carries the lamp down, and wider walls only add more of it — so the turn would move by under a tenth between widths. Solved at six strips a side, a box two units wide turns between 0.5 and 0.65, within the tenth or so predicted. A box half a unit wide still loses at 1.5 deep, and its quarter-turned twin says it still loses at 2: its turn is more than a unit and a quarter deeper. With only the floor painted a box w wide and d deep is the box d wide and w deep, so depth and width are one question, and the answer is that the floor's shorter side decides: under about half a unit the finish greys the paint, past about 0.65 it colours it. The scan's first pass, at four strips, said otherwise, and got four of seven signs wrong in the long, thin boxes.
Where the model breaks
The slope limit was exact all along
Every count of what a slow pigment cannot reach compares a slope limit, solved exactly on 161 levels of reflectance, with two band models found by searching. Searched harder, the band models lost a sixth of their losses, and the open question was whether the slope limit was being flattered by its own grid — whether 321 or 641 levels would lower its counts too. The prediction was a fall of under a twentieth. There is no fall at all: on every finer grid, under all nine lamps, every direction reaches exactly what it reached, and 463 directions are lost as before. The reason is the shape of the answer. The best reflectance that may change by at most an eighth per five-nanometre band is a staircase whose every step is a whole eighth, because a weighted sum is maximised at a corner of the allowed set and every corner is built from full steps off nought or one. A grid of only the nine eighths gives every answer to the last digit. The census's count is therefore not an estimate on a grid but an exact statement about every pigment that slow, and the slope limit's 1.38-fold lead over the searched band model is a floor that further search can only raise.
Where the model breaks
A slow pigment splits mercury's lines only if its swing fits between them
Under a clear mercury lamp a pigment whose reflectance may swing from nought to one over no less than forty nanometres loses 21 of the census's 148 directions, and every one of them asks it to reflect one of the lamp's green and yellow lines and absorb the other. The census samples the spectrum every five nanometres, which smears each six-nanometre line across a band, and the prediction was that a one-nanometre grid would let the pigment's cliff fall between the lines and cut the count by a third. It cuts it by nothing: 21 on either grid. Narrowing the lines to one and a half nanometres cuts it only to 17. What decides the losses is the lines' spacing, 31.5 nanometres, against the pigment's swing. With narrow lines a pigment whose swing is 30 nanometres loses no direction at all, and one whose swing is 34 loses fourteen; the step sits at the gap. The mercury lamp's cost to a slow pigment belongs to the lamp, not to how the census samples it.
What a camera does
LEDs built on the locus still fall off the line
A camera's colour error under a lamp can be estimated from three narrow channels of a small sensor, read as departures from a smooth lamp of the same colour, and the estimate held for the census's white LEDs, which varied only their blue pump. Rebuilt on LEDs varied freely in pump, converter and red phosphor, it failed: the error was a V about the Planckian locus, and no line in the departures read it. A maker does not let the white fall where it falls, so the proposal was to build the LEDs as a maker does — the converter tuned for each pump and red phosphor until the white sits on the locus — and the prediction was that the V would go and the line return, the census's held-out error to within ±0.15. Thirty of thirty-six such LEDs can be built. On the locus the V is gone and the line is not: the LEDs alone are missed by ±0.25 to ±0.28, the census's lamps by ±0.27. Taken one pump at a time, the departures read the LEDs pumped at 452 nm to ±0.05 and those at 440 and 465 nm no better than ±0.17. Within one pump the LEDs are readable, and across pumps they are not.
What it takes to deliver it
Off the grey axis a knee trades hue for chroma
Along the grey axis a total-ink cap approached through a knee quiets the cast a four-ink table prints between its nodes, because the knee slows the chromatic inks' retirement where black takes over. In a saturated shadow the chromatic inks carry the colour, and the prediction was that a knee would quiet the hue error there in the same way, a fifteen-unit knee halving the worst segment's. On two rays of dark colours, a deep blue and a brown, it does quiet the hue — every wider knee lowers it on both — and halves the worst segment's on the blue, to 0.28 of the hard cap's, but not on the brown, at 0.68, whose inks the knee slows far less. And the hue was never the main error. Between nodes the table prints these colours duller than asked, its chroma error five to eight times its hue error, and every wider knee makes that loss slightly worse, so the table's total error rises as its hue error falls. The cap itself, met hard, halves the blue's error, as it removed the grey's cast; on the brown it changes almost nothing.
Matching and measuring
A broad green sharpens the blue; a broad blue dulls it
ΔEITP makes a display's blue corner sharp or blunt according to its primaries, and the census of single-wavelength primaries found the green deciding how sharp — the 485 nm blue's corner moving 4.37 times a margin against a 530 nm green and 8.19 against a 555 nm one. Real primaries have widths, and the prediction was that a broad green acts as a longer green, raising the spike and moving the turn towards the violet, while a broad blue barely matters. The first half holds: a green forty nanometres wide at 530 nm sharpens the 485 nm corner to 5.05 and moves the turn from 453.7 nm to 451.2, as a single-wavelength green at 535.8 nm would, though its dominant wavelength has moved only to 533. The second half does not: a blue twenty nanometres wide moves the turn by a nanometre but dulls the corner by a tenth, to 3.94. The two widths pull the corner opposite ways, and a display whose green and blue are both broad lands near where its line primaries were. The red, at forty nanometres, still changes nothing that matters.
What the eye does
A density a tenth wrong costs the band more than fitting it
An older observer's colour matches carry a lens band and a macula that can trade against each other, and fixing the macular pigment's density from an outside measurement breaks much of that trade when the density is right. A flicker measurement and a colour match average the pigment over different fields, so the fixed density can be wrong by a tenth or more, and the prediction was that a tenth too high takes about a third of a standard error off a band at 445 nm and moves the macula's peak by under half a nanometre. It takes 1.4 standard errors off the band and moves the peak 1.25 nm short. At 415 nm the same error pushes the band the other way, up by 2.4 standard errors, and runs its fitted centre 5.6 nm shorter and the lens's age three years younger. Bias and error together, a density fixed a tenth wrong gives the band at 445 nm a larger error than leaving the density free; the break-even is about eight per cent. Given as a measurement with its own standard error of 0.03 instead of fixed, the same wrong density biases the band by under half its error.
Where the model breaks
A band pigment must be narrower than the gap it splits
Under a clear mercury lamp a pigment built from absorption bands no narrower than forty nanometres falls short in fifteen of the twenty-five directions that ask it to reflect the green line and absorb the yellow, or the reverse — the count an earlier search left short however deep it cut its cliff. The slope-limited pigment's losses turned out to be the lines' 31.5 nm spacing against its swing, and the prediction was that the band pigment's are the same arithmetic, falling to nought near a band thirty nanometres wide. Searched on a one-nanometre grid with the lines narrowed, the count falls at every narrower band — fifteen at forty, eleven at thirty-five, nine at thirty — and reaches nought at twenty-five, not thirty. Every loss already splits the pair and falls short on both lines at once: a Gaussian band deep enough to absorb one line keeps absorbing a little of the other. One band of depth twelve, placed as well as it can be, errs by 3.2 per cent at forty nanometres, 0.8 at thirty and 0.2 at twenty-five; doubling its depth buys less than narrowing it by five nanometres.
What a scene does
A square floor needs more room than a long one
A satin finish on a room's grey faces costs its painted floor a little chroma when the room is narrow and adds some when it is wide, and in a box two and a half tall the turn was found on the floor's shorter side — 0.64 when the longer side is one unit, about 0.51 when it is two. The prediction for a square floor, with no longer side to speak of, was a later turn, near 0.7, because both pairs of walls are then equally close and both carry the lamp down. Solved at six strips a side, the square floor loses for every one of seventy-two paints at 0.5, 0.6 and 0.7 on a side and gains for every one from 0.8, turning near 0.78. A floor 0.7 by 0.7 loses where a floor 0.7 by one gains. The finish brightens the square's darkest floor by more than an oblong's at every shorter side — by a hundred per cent against forty-six at half a unit — and the turn falls steadily as the longer side grows: 0.78, 0.64, 0.51. The longer side matters, and the light-pipe account says why.
What it takes to deliver it
More nodes quarter the dulling and leave the knee a larger share
A four-ink table printing a dark saturated colour between its nodes prints it a little duller than asked, and a knee in the ink limit, which quiets the hue error, deepens that dulling by about a hundredth of a unit. The prediction was that doubling the table's nodes quarters the dulling — the square of the spacing — and the knee's added cost with it, so that on a dense table the knee's hue gain is all that remains. On tables of 9, 17, 33 and 65 nodes along a deep-blue and a brown ray, the dulling does fall about fourfold with each doubling, capped or not, to under a hundredth at 65 nodes. The knee's added dulling does not keep pace: on the blue it is 13 per cent of the hard cap's at seventeen nodes, 23 at 33 and 35 at 65, and the knee's whole error, an eighth more than the hard cap's at seventeen nodes, is a quarter more at 65. The knee keeps its hue gain at every density. A denser table shrinks every error and makes the knee's trade worse.
Matching and measuring
A green's long tail sharpens the blue; its short tail does not
A quantum-dot green falls steeply on one side and trails on the other, and the prediction was that a tail on its long side, holding a third of its power, would make ΔEITP's blue corner treat the green as a line past 538 nm, further than a symmetric forty-nanometre green's 535.8, because all its spread is on the L side. It does: 540.7 nm, the 485 nm corner rising from ×4.51 to ×5.70. The control is the larger finding. The same tail on the short side costs the green far more purity — 0.77 against the long tail's 0.95 — and moves its equivalent line by under a nanometre and a half. The corner reads the green's ratio of L to M signal, to within a third of a nanometre for every long tail, and not its purity, its centroid or its dominant wavelength; a short tail lowers that ratio less than its colour suggests and brings S signal that pushes the corner back.
Matching and measuring
Brightness and the green sharpen the blue almost independently
ΔEITP's blue corner is sharper the longer a display's green, and sharper the brighter the display. The prediction was that the two compound — that the 555 nm green's lead over the 505 nm one, 3.42 times at 320 cd/m², would pass four by 4,000, because a brighter display narrows every angle and the long green's was narrowest — and that the turn, where adding green starts raising the blue's L by the larger fraction, would not move with the peak at all. Both halves fail. The lead grows by a twentieth, to 3.60 at 4,000 and 3.63 at 10,000; across thirty-five displays the corner is a green factor times a brightness factor to within 4.5%. And the turn moves at every step, two to six nanometres towards the violet from 100 to 10,000 cd/m², crossing the blue at which green raises L and M equally near 1,000. The quantiser weighs a fractional rise in L against one in M by their encoded levels, and brightness evens the scales.
What the eye does
A wrong density shifts every experiment alike near the macula
Read from the expected matches, a macular density fixed a tenth too high lowers an older observer's lens band at 445 nm by 1.4 of its standard errors and raises a band at 415 nm by 2.4. Real experiments carry noise, and the prediction was that their fits scatter about those biases with the fixed fit's own error, so that a tenth's mistake puts the band beyond twice its error in about a third of experiments at 445 nm and two thirds at 415. Twenty simulated experiments per case, each fitted with the density free and held at 0.9, 1 and 1.1 of its truth, give five of twenty at 445 nm and sixteen at 415 under two sessions — eight and eighteen with a line on the macula's peak. Near the macula the bias is a rigid shift: every experiment moves by the first-order amount to within a tenth of a standard error. Far from it the bias is lopsided — up by 0.058 for a density too high, down by 0.032 for one too low — and a density too low erases the band altogether in one experiment of twenty.
What a camera does
A channel at each pump cannot see the recipe
Thirty white LEDs built onto the Planckian locus defeat a line in a small sensor's three departures, except within one pump: the 452 nm LEDs are read to ±0.047 by the 450 nm channel, the others no better than ±0.15. The proposal was a channel at each pump — 440 and 465 nm beside the 450 — and the prediction that each pump's LEDs would be read from their own channel within ±0.08. The 465 nm LEDs' own channel halves their miss, to ±0.135; the 440 nm LEDs' moves it from ±0.171 to ±0.149. No line in any three of the five departures reads either pump better than ±0.1. Yet the camera's error is simple: a line in what each LED is made of — its red phosphor, its colour temperature and their product — reads every pump within ±0.07, the 440 nm LEDs to ±0.043. The error is easy; the channels cannot see the recipe that sets it. Knowing each LED's pump halves the pooled miss, to ±0.119, and two blue channels can tell only one pump of three from the others.
What a camera does
The converter gives the red phosphor away
A camera's error under white LEDs built onto the Planckian locus is simple in what each LED is made of — its red phosphor and its colour temperature — and a small sensor's narrow departures cannot read the red phosphor. The proposal was a channel at 610 nm, where a garnet converter falls and a red phosphor rises, and the prediction that it reads the red-phosphor strength within a tenth of its range. Within a pump it reads nothing: ±0.36 to ±0.43 at 440 and 452 nm against a spread of ±0.36, and no red channel from 600 to 680 nm does better than ±0.22. The reason is the locus. To keep the white's colour, each step of red phosphor pushes the converter's peak shorter, by 29 to 55 nanometres from none to the most, and in every red channel the two changes cancel. What survives is the converter: the 600 nm departure reads its peak to ±5 nm. Read across all thirty LEDs at once, the converter gives the red phosphor away — ±0.12 — and the error read through that estimated recipe comes within ±0.08 at every pump, the 440 nm LEDs at ±0.078, half their own channel's miss.
308 essays on this thread, page 12 of 13 · all threads · all essays