Pattern and colour

A bobbinet's twists light the faint fringe family

A net of three directions over a voile makes two families of fringes, one strong and one a fifteenth as strong, because its light is a product of three gratings and the slow family needs two of them at once. A real tulle is not three gratings. A bobbinet's holes are hexagons whose oblique threads stop at every twist, so its three directions are dashed lines, and a dashed line puts light where a product of gratings leaves none. Drawn as a bobbinet, the faint family is as strong as the other or stronger at every width tried — and the width of the twisted sides decides which wins.

Worth reading first: A net of three directions beats a voile one way at a time · A net over a voile beats through a harmonic · Two sheers make a moiré that walks with the viewer.

A net of three directions beats a voile one way at a time. Hung a few centimetres in front of a voile and seen from across a room, a tulle makes two families of fringes at right angles. Along the voile threads that lie parallel to one of the net’s directions it beats as a one-directional net does, through its fifth harmonic, in fringes six millimetres apart. Across them no direction of the net lies anywhere near, and the only slow beat comes from a line of points two of the net’s directions make together, at 3\sqrt{3} over the net’s pitch: fringes 27 millimetres apart, four times wider and a fifteenth as strong.

The weakness of that second family came from the model, and the essay said so. It drew the net as three sets of straight bars, so the light through it is the product of three gratings, and a point of its spectrum that two sets make together is the product of two small amplitudes — an order of the cover weaker than a point one set makes alone. It closed by asking what a real tulle does, since a bobbinet is not made of straight bars: its threads twist round one another at the corners of every hole.

The answer reverses the earlier finding. Drawn as the mesh a bobbinet actually has, the faint family is not faint.

A bobbinet's hexagonal mesh beside a straight-bar net of the same pitch. Left: a bobbinet tulle's mesh drawn as hexagonal holes, the two sides of each hole along the warp formed by a twisted pair 2 times as wide as the four oblique single threads, which cover 10 per cent of the net's pitch; it is 78 per cent open. Right: a net of three sets of straight bars at sixty degrees with the same pitch and the same bar width. The two have their rows of holes at the same spacing, so their spectra have points in the same places. They differ in that every bar of the bobbinet stops at a twist and the next starts offset, where the straight-bar net's bars run on unbroken: the bobbinet's three directions are dashed, and a dashed grating is not a grating. Both are idealised as sharp-edged threads lying flat.
Fig. 1 A bobbinet’s mesh, left, drawn as hexagonal holes whose two sides along the warp are a twisted pair twice as wide as the four oblique single threads; right, a net of three sets of straight bars at the same pitch and the same thread width. The two have their rows of holes at the same spacing, so their spectra have points in the same places. Every oblique thread of the bobbinet stops at a twist and the next starts offset: its three directions are dashed, not continuous.

A bobbinet is three dashed lines, not three gratings

A bobbinet tulle is made on a machine that twists a set of bobbin threads round the warp, crossing diagonally from one warp thread to the next — a twist, like a leno’s, being an arrangement no weave matrix can write down. The hole that results is a hexagon. Its two sides along the warp are twisted pairs — a warp thread and a bobbin thread wound round each other — and its four oblique sides are single bobbin threads on their way to the next twist.

Drawn flat, that is a honeycomb. Its holes sit on a triangular lattice, each the region nearer its own centre than any other, and its threads run along the boundaries. Seen from a distance it looks like a net of three directions, and its rows of holes repeat at the same three angles as the straight-bar net’s.

But no thread of it runs on. Each oblique side ends at a twist, and the next oblique side in the same direction starts half a hole along. A bobbinet’s three directions are three sets of dashed lines, and that is a different object, optically, from three gratings. A grating’s light lies on a line through the origin of its spectrum and nowhere else. A dashed grating’s is spread across the whole lattice, because the dashing is itself periodic in the other directions.

Why a dashed line lights the points between

The reason is one line of Fourier arithmetic, and it is worth having before the numbers. A dashed grating is a grating multiplied by a square wave that runs along its bars — on where a bar is present, off at the gap where it stops. A product in space is a convolution in the spectrum, so each of the grating’s points is copied along the dash direction, once for every harmonic of the dashing, with the dashing’s own amplitudes.

A grating of continuous bars has its points on one line through the origin, perpendicular to the bars. Dash it and every point on that line acquires a row of companions, spaced by the dash frequency, running along the bars’ own direction. In a honeycomb the dashes repeat exactly as often as the holes do, so the companions land precisely on the lattice points between the lines. The nearest of them is the dashing’s own fundamental copied onto the grating’s zero order, and for each of the three directions it is a point on the 3\sqrt{3} ring: the lattice vector perpendicular to a direction’s first point is exactly 3\sqrt{3} times as long. The next companions, copied onto the first points, land on the ring at two, and so on outwards, which is why the bobbinet’s circles are scattered across every ring in the figure below while the straight bars’ squares stay on the lines. A straight-bar net reaches those points only by multiplying two gratings together; a bobbinet reaches them directly, through its dashing, with amplitudes of the same order as its lines’ own.

This is the same mechanism that makes a moiré a vernier that magnifies the error too: a beat is a difference of frequencies, and anything that adds frequencies to either layer adds beats. The bobbinet adds frequencies the straight-bar model has no way to make.

The straight-bar net’s spectrum, for reference

The earlier calculation’s numbers are the ones to beat, so they come first.

The fringe families of a three-direction net over a square voileFor a 1.55 mm net of three directions in front of a 0.3 mm voile, 50 mm apart and seen from 3.0 m: one net set, with the voile's threads at right angles to it, fringes 0.58 mm apart at 100% of the strongest; two net sets together, with the voile's threads at right angles to it, fringes 27.37 mm apart at 7% of the strongest; one net set, with the voile's threads parallel to the aligned set, fringes 6.13 mm apart at 100% of the strongest; two net sets together, with the voile's threads parallel to the aligned set, fringes 1.72 mm apart at 10% of the strongest. What the bars cannot show is the voile's own second harmonic, which adds a fainter family of its own.The strong family is narrow and the widest family is faintfringe spacing on the near layer for each voile direction, from one net set and from two together, with each family's strength against the strongestone net set, with the voile's threads at right angles to it0.58 mm100% strengthtwo net sets together, with the voile's threads at right angles to it27.37 mm7% strengthone net set, with the voile's threads parallel to the aligned set6.13 mm100% strengthtwo net sets together, with the voile's threads parallel to the aligned set1.72 mm10% strengththe nearest net point to each voile first harmonic, among one set's points and among two sets'3.0 m · net in front
Fig. 2 The four fringe families of a 1.55-millimetre straight-bar net 50 millimetres in front of a 0.3-millimetre voile, seen from three metres: the spacing of each on the near layer, and its strength against the strongest. The slow family along the voile, six millimetres apart, is at full strength; the slow family across it, 27 millimetres apart, is at seven per cent.

Two families are slow enough to see from three metres. Along the voile’s threads parallel to one net set, the net’s fifth harmonic beats the voile’s first in fringes 6.1 millimetres apart. Across them, a point two sets make together beats in fringes 27.4 millimetres apart — and at seven per cent of the other’s strength, because its amplitude is the product of two sets’ harmonics, each a few hundredths.

Where a honeycomb puts its light

The bobbinet’s spectrum has points at exactly the straight-bar net’s places, because its holes repeat on the same lattice. What differs is how bright each point is, and that is a direct sum over one cell of the mesh, point by point.

The low-order points of a bobbinet's spectrum against a straight-bar net's. The Fourier amplitude at each point of the two nets' common lattice out to three times the first ring, for a straight-bar net of three sets covering 10 per cent of the pitch (squares) and a bobbinet of the same pitch whose twisted sides are 2 times as wide (circles). On the first ring the straight-bar net's points average 0.098 and the bobbinet's 0.087; on the ring at √3, where only two sets together reach, the straight-bar net's average 0.0097 and the bobbinet's 0.059. A product of gratings leaves the √3 ring an order of the cover weaker than the first; a honeycomb puts nearly as much there. The twisted sides also split each ring, weakening the points along the warp and strengthening others. Sharp-edged flat threads are assumed throughout.
Fig. 3 The Fourier amplitude at every point of the two nets’ common lattice out to three times the first ring, straight bars as squares and the bobbinet as circles, at a single-thread width of a tenth of the pitch and twisted sides twice as wide. On the first ring the two are alike, 0.098 against 0.087 on average. On the ring at 3\sqrt{3}, which only two straight-bar sets reach together, the straight-bar net averages 0.0097 and the bobbinet 0.059 — six times as much.

On the first ring, the points closest to the origin, the two nets are alike: an average amplitude of 0.098 for the straight bars and 0.087 for the bobbinet. On the ring at 3\sqrt{3} they part by a factor of six. The straight-bar net has nothing there of its own and only the product of two sets’ first harmonics, 0.0097; the bobbinet has 0.059, more than half of what it has on the first ring. A honeycomb’s dashing has moved light out of the single-direction lines and onto the lattice points between them.

The twisted sides add a second effect the drawing shows. Each ring’s points are no longer equal: the points that read the rows along the warp, where the thick twisted sides lie, are weakened, and their neighbours strengthened. The first ring splits into 0.110 and 0.041, and the 3\sqrt{3} ring similarly. A bobbinet’s spectrum is not even six-fold; it has a direction, the warp’s, and the twist is what gives it one.

The two families, as a bobbinet makes them

The fringes beat through the net’s fifth harmonic, not its first, because the net’s pitch is about five times the voile’s. So the comparison that matters is out at the fifth ring, where the bobbinet’s amplitudes come from the same dashing.

The fringe families of a bobbinet tulle over a square voileFor a 1.55 mm bobbinet, its hexagonal holes' twisted sides 2 times as wide as its single threads, in front of a 0.3 mm voile, 50 mm apart and seen from 3.0 m: one net set, with the voile's threads at right angles to it, fringes 0.58 mm apart at 80% of the strongest; two net sets together, with the voile's threads at right angles to it, fringes 27.37 mm apart at 100% of the strongest; one net set, with the voile's threads parallel to the aligned set, fringes 6.13 mm apart at 95% of the strongest; two net sets together, with the voile's threads parallel to the aligned set, fringes 1.72 mm apart at 16% of the strongest. What the bars cannot show is the voile's own second harmonic, which adds a fainter family of its own.Drawn as a bobbinet, the wide family is the strongest of the fourfringe spacing on the near layer for each voile direction, from one net set and from two together, with each family's strength against the strongestone net set, with the voile's threads at right angles to it0.58 mm80% strengthtwo net sets together, with the voile's threads at right angles to it27.37 mm100% strengthone net set, with the voile's threads parallel to the aligned set6.13 mm95% strengthtwo net sets together, with the voile's threads parallel to the aligned set1.72 mm16% strengththe nearest net point to each voile first harmonic, among one set's points and among two sets'3.0 m · net in front
Fig. 4 The same four families for a bobbinet of the same pitch, its twisted sides twice the single threads’ width, at the same distance. The spacings are unchanged — they are set by the lattice — but the strengths are not: the slow family across the voile, 27.4 millimetres apart, is now the strongest of the four, and the slow family along it is at 95 per cent of it.

The spacings are identical to the straight-bar net’s, since they come from where the points are. The strengths change order. The family across the voile, 27.4 millimetres apart, is now the strongest of the four, and the family along it — the one a straight-bar net made strongest — is at 95 per cent of it. At three metres, 50 millimetres apart, a bobbinet over a voile makes the check of two equally visible families the earlier essay held out as the possibility its model could not reach.

Both families are weaker in absolute terms than a straight-bar net’s strong family: 0.0027 and 0.0028 against 0.0193. A honeycomb’s fifth harmonic is spread over more lattice points and is fainter on any one of them. What the bobbinet changes is the balance, from fifteen to one to one to one.

How wide the twist is decides which wins

The twisted sides’ width is the one property of the mesh with no counterpart in the straight-bar model, and the balance turns on it.

The two slow fringe families' strengths against a bobbinet's twist width. The strength of the two slow fringe families a 1.55 mm net makes over a 0.3 mm voile 50 mm behind it, seen from 3.0 m — the one beating along the voile's threads parallel to a net direction and the one beating across them through the point at √3 — for a bobbinet (solid) and a straight-bar net (dashed), against the twisted sides' width as a multiple of the single threads', at a cover of 0.1. twist 1: bobbinet 0.0036 along and 0.0092 across, bars 0.0193 and 0.0013; twist 1.5: bobbinet 0.0033 along and 0.0050 across, bars 0.0193 and 0.0013; twist 2: bobbinet 0.0027 along and 0.0028 across, bars 0.0193 and 0.0013; twist 2.5: bobbinet 0.0019 along and 0.0099 across, bars 0.0193 and 0.0013; twist 3: bobbinet 0.0011 along and 0.0119 across, bars 0.0193 and 0.0013; twist 3.5: bobbinet 0.0005 along and 0.0072 across, bars 0.0193 and 0.0013. The bobbinet's across family is the stronger of its two at 1, 1.25, 1.5, 2, 2.25, 2.5, 2.75, 3, 3.25, 3.5. Both families beat through the fifth harmonic of the net, where a sharp-edged thread's amplitude swings with its width, so the strengths oscillate; a real thread's soft edge would damp the swings and not their order of magnitude.
Fig. 5 The strengths of the two slow families against the twisted sides’ width as a multiple of the single threads’, at a single-thread width of a tenth of the pitch: bobbinet solid, straight bars dashed. The straight-bar net’s families do not depend on it and sit at 0.0193 and 0.0013. The bobbinet’s across family is the stronger at every twist but 1.75, from 0.0028 at a twist of two to 0.0119 at three; its along family falls steadily from 0.0036 to 0.0005.

A twist the width of a single thread — a mesh with all six sides alike — makes the across family 0.0092 and the along family 0.0036, two and a half to one in the faint family’s favour. At a twist of 1.5 it is 0.0050 against 0.0033; at two, level; at three, 0.0119 against 0.0011, ten to one. Across every twist from one to three and a half the across family is the stronger, except near 1.75 where the two pass each other. The along family weakens steadily as the twisted sides thicken, from 0.0036 to 0.0005, while the across family rises and falls.

That last point is the caution. Both families beat through the fifth harmonic, and a sharp-edged thread’s fifth harmonic swings through nulls as its width changes — the straight-bar net’s own strong family vanishes at a thread width of a fifth of the pitch. The bobbinet’s curves wave for the same reason. A real thread is round and its edge soft, which damps the swings, so the curves above give the order of each family’s strength and not its exact value at any one width.

And at every thread width tried

The single threads’ width is the other free input, and the result survives it.

The two slow fringe families' strengths against a bobbinet's thread width. The strength of the two slow fringe families a 1.55 mm net makes over a 0.3 mm voile 50 mm behind it, seen from 3.0 m — the one beating along the voile's threads parallel to a net direction and the one beating across them through the point at √3 — for a bobbinet (solid) and a straight-bar net (dashed), against the single threads' width, the twisted sides 2 times as wide. cover 0.050: bobbinet 0.0017 along and 0.0101 across, bars 0.0136 and 0.0006; cover 0.100: bobbinet 0.0027 along and 0.0028 across, bars 0.0193 and 0.0013; cover 0.150: bobbinet 0.0025 along and 0.0111 across, bars 0.0136 and 0.0005; cover 0.200: bobbinet 0.0009 along and 0.0033 across, bars 0.0000 and 0.0010; cover 0.250: bobbinet 0.0005 along and 0.0095 across, bars 0.0136 and 0.0012. The bobbinet's across family is the stronger of its two at 0.050, 0.075, 0.100, 0.125, 0.150, 0.175, 0.200, 0.225, 0.250. Both families beat through the fifth harmonic of the net, where a sharp-edged thread's amplitude swings with its width, so the strengths oscillate; a real thread's soft edge would damp the swings and not their order of magnitude.
Fig. 6 The same two families against the single threads’ width, from a twentieth of the pitch to a quarter, with the twisted sides twice as wide. The bobbinet’s across family is the stronger of its two at every width, from 0.0101 against 0.0017 at a twentieth to 0.0095 against 0.0005 at a quarter, level only near a tenth. The straight-bar net’s across family stays at a tenth or less of its along family, except at a fifth of the pitch, where the along family passes through its null.

At a thread width of a twentieth of the pitch, the bobbinet’s across family is six times its along family; at three twentieths, four times; at a quarter, nineteen times. Only near a tenth, the width the earlier essays used, are they level. The straight-bar net never comes close, except where its along family passes through a null at a fifth of the pitch and has nothing left to be compared with.

So the faint family is faint only in a model that a tulle is not. In the mesh a bobbinet has, it is as strong as the other family at the one width and twist examined first, and stronger — often several times stronger — almost everywhere else.

The null does not move

The faint family’s null, the distance at which its fringes become infinitely wide and vanish, was the earlier essay’s sharpest prediction, and it is set by the lattice rather than by the mesh.

The two slow fringe families against the eye's distance. Fringe spacing on the near layer for a 1.55 mm three-direction net 50 mm in front of a 0.3 mm voile, from 0.4 m to 30 m. The family one net set makes along the voile grows from 8.3 mm to 8.8 mm and has no null beyond 0.31 m; the faint family two sets make the other way passes through a null at 8.76 m. What the lines cannot show is whether the faint family, at its widest, is dark enough to see.
Fig. 7 The spacing of the two slow fringe families against the eye’s distance for the 1.55-millimetre net 50 millimetres in front of the 0.3-millimetre voile, from 0.4 to 30 metres. The family along the voile has no null beyond a third of a metre; the family across it passes through a null at 8.76 metres. The spacings belong to the lattice, so the bobbinet’s are the same.

The null sits at 8.76 metres for the straight-bar net and for the bobbinet alike, because it comes from where the net’s points are and the two meshes share a lattice. What changes is whether anybody would see it. A null in a family at seven per cent of the strongest is a null in something already hard to see; a null in the bobbinet’s strongest family is the disappearance of the most visible fringes in the room, at a definite distance, while the other family carries on unchanged. Walking back from a curtain of bobbinet over voile, an observer should see the wide check’s bars in one direction swell and vanish at about nine metres while the other direction’s stay put — the same parallax that makes two sheers walk with the viewer, with a definite distance in it.

What the mesh is, and what it is not

The bobbinet is drawn flat and sharp-edged. Its holes are exact Voronoi hexagons, its threads are bands of uniform width with clean edges, and its twists are the thicker sides and nothing else. A real twist is a helix of two threads whose outline varies along the side, and its corners, where four threads meet, are knots of yarn rather than clean junctions. All of that changes the high harmonics more than the low ones.

The voile is the earlier essays’ straight-bar voile, two sets at right angles covering two fifths of its pitch, so its own amplitudes are unchanged and the comparison is the net’s alone.

Nor is the curtain behind glass. A pane in front changes what reaches the eye by its own reflection, which glass hides a black net and not a white voile priced; nothing here includes it.

And strength is not visibility. A family’s strength here is the product of the two points’ amplitudes, the quantity a net over a voile used; whether a fringe of a given strength and spacing is visible depends on the eye’s sensitivity at that spacing, which the reed leaves its own mark and a slub finds the width of the cloth handled for their own patterns. The wide family is at a spacing where the eye is less sensitive, so its visibility relative to the narrow one is less than its strength says.

How the amplitudes were computed

The mesh is a honeycomb in units of the hexagon’s side: hole centres on a triangular lattice with spacing 3\sqrt{3} along the rows and rows 1.5 apart, a point counted as thread if it lies within half a thread’s width of the boundary between its own hole and a neighbour’s. Boundaries between holes side by side along a row are the twisted sides, a stated multiple of the oblique ones’ width, and the oblique ones are the stated share of the pitch — the row spacing — wide. Each Fourier amplitude is a direct sum of e−2πi G⋅re^{-2\pi i\,G\cdot r} over the open points of one sampled unit cell, 240 points across, divided by the number of points.

The lattice is shared. The bobbinet’s reciprocal lattice is the straight-bar net’s triangular one when the net’s pitch is the honeycomb’s row spacing, so every point the earlier families used exists for the bobbinet too, with the bobbinet’s own amplitude in place of the product of gratings. The families and their spacings are the earlier calculation’s, unchanged: the far layer projected onto the near by the eye’s perspective, each voile harmonic beaten against its nearest net point. Required of it, and checked: the bobbinet’s points are found on its lattice to a millionth; a twist narrower than a single thread, a cover leaving no hole and a mesh that is neither bars nor bobbinet are refused.

Who found what

Bobbinet was mechanised by John Heathcoat, whose machine of 1808–1809 twisted bobbin threads round a warp to make the hexagonal mesh still called by the name; moiré as a beat between two periodic layers is older than any account of it here, and watered silk is a beat of the oldest kind. That a honeycomb’s spectrum differs from a product of gratings is ordinary Fourier optics. What is added here is the comparison at the points two directions of a net reach together: that a bobbinet’s dashed threads put six times as much light on the 3\sqrt{3} ring as straight bars do; that over a voile its wide fringe family is as strong as the narrow one or stronger, at nearly every thread width and twist, where a straight-bar net leaves it at a fifteenth; and that the twist’s width, which a straight-bar model has no place for, decides which of the two families is the brighter.

Still open: a round thread’s fifth harmonic

Everything that decides the balance happens at the net’s fifth harmonic, and the fifth harmonic of a sharp-edged band swings through nulls as the band widens. A round thread seen from in front is not a band: its transmittance falls off towards its edges as the thread’s section thins, and its high harmonics fall away smoothly rather than oscillating. The prediction is that a round-thread bobbinet keeps the across family stronger than the along family over the whole range — since the along family’s decline with twist is a trend, not a swing — but at strengths a few times lower than the sharp-edged ones.

The calculation is the same sum with a thread profile in place of a band: a half-ellipse in transmittance across each side, wider at the twists. And the check needs no instrument: a bobbinet curtain over a voile, walked back from in a straight line, should show its wide check’s bars vanish near nine metres.

Named objects

A flat tag is an object no other essay names yet.

BeatHarmonicMagnificationMoirePlane group