Compound and figured cloths

A tapestry's total cut belongs to its cartoon

A slit tapestry is cut wherever two colours meet along the warp, and the weaver has three ways to make the cuts shorter: lay the cartoon somewhere else, step the outline, or break the join into hachures. None of them makes the cut smaller. Counted pass by pass, the total length of every slit is the cartoon's outlines measured along the warp — a 6 cm join is cut by 6 cm whether it is one slit, twelve steps or 168 hachure tips — and only turning the cartoon changes it, smoothly, between two projections whose average is Cauchy's. What the weaver controls is how the total is divided, and a picture of verticals and horizontals is divided worst at exactly the two angles tapestries are woven at.

Worth reading first: A tapestry's slit is the chord of one end · Woven on its side, a tapestry's slits carry its weight · The bias is a mechanism.

Two calculations have now been made about the cuts in a slit tapestry. The first found how long one slit is: a straight edge leaves the end spacing times its slope in every gap it crosses, and a curve’s vertical leaves the chord whose sagitta is one end, between Rd\sqrt{Rd} and 8Rd\sqrt{8Rd} however the cartoon is laid. The second turned the cartoon a quarter turn and found the cuts moving from the picture’s verticals to its horizontals, and from carrying nothing to carrying the cloth below them.

Both counted slits one at a time. This one counts them all at once, and the count turns out to be much simpler than either: the total length of cut in a slit tapestry is fixed by the cartoon and the direction of the warp, and nothing a weaver does at the loom changes it. The three things a weaver can do — where the cartoon is laid, how an outline is stepped, whether a join is broken into hachures — only decide how that fixed total is divided among slits. They are choices about the longest slit, never about the sum.

Three ways to join two colours in a tapestry, and the cut each leaves. A join 6 cm tall between two colour areas, woven at 8 ends and 32 passes a centimetre three ways and drawn end by end, every gap no weft crosses drawn heavy. A straight edge: 1 slit, the longest 6.00 cm, 6.00 cm in all; a kilim's staircase: 12 slits, the longest 0.50 cm, 6.00 cm in all; hachures along the weft: 168 slits, the longest 0.06 cm, 6.00 cm in all. The total is the join's height in every case; the joins differ only in how they divide it. The drawing cannot show how far any slit opens.
Fig. 1 A join six centimetres tall between two colours, woven three ways at eight ends and 32 passes a centimetre and drawn end by end, with every gap no weft crosses drawn heavy. A straight edge is one slit six centimetres long. A staircase that steps one end every half-centimetre is twelve slits of half a centimetre. Hachures — teeth eight passes thick reaching a centimetre and a half into the other colour, from alternate sides — are 168 slits, none longer than two passes. In all three the cut adds up to six centimetres.

Counting the cut the other way

Every slit is a run of passes in one gap. Add up the lengths of all the slits and the result is the number of (gap, pass) pairs that are open, divided by the pass density. Count the same pairs the other way — pass by pass rather than gap by gap — and on each pass the number of open gaps is the number of times the colour changes along it.

So the total cut is the number of colour changes along each pass, summed over every pass, divided by the pass density. That is the whole of the argument, and it is a re-counting rather than a model; the two counts agree to the last pass on every figure here.

What it says is that the total is decided pass by pass, by how many boundaries a horizontal line through the picture crosses at that height. A pass crosses a convex shape twice if it crosses it at all. Integrate over height and a convex shape contributes twice its height, whatever its outline does between top and bottom. A rectangle 30 cm tall is cut by 60 cm, a circle of radius RR by 4R4R, the robe in the standing figure by twice its 92 cm. In general the total is each outline’s height covered along the warp, counted with multiplicity — the total variation of its height as one walks round it.

Joins only divide

That is why the three joins in the figure above come to the same six centimetres. The weaver’s choice at a join is what each pass does at the boundary, and on every pass of every one of the three, the boundary is crossed exactly once. A straight edge crosses it in the same gap every time, so the six centimetres are one slit. A staircase crosses it in the same gap for sixteen passes and then moves one end over, so the six centimetres are twelve slits of half a centimetre. Hachures move the crossing back and forth between gaps on nearly every pass, so the six centimetres are 168 slits too short to see.

The kilim’s staircase is the classic answer to a vertical line in slit weaving: the line is drawn as a staircase of short verticals, each offset by one end from the last, and nothing longer than a step is ever cut. Hachures are the classic way of grading one colour into another in European tapestry: interpenetrating comb-like teeth that run along the weft, so the eye mixes them into an intermediate tone at a distance. That the second also cuts a join into pinholes is not usually given as its reason, but it follows exactly.

How three joins divide the same cut between long and short slits. For a straight edge, a kilim's staircase and horizontal hachures joining two colours over 6.0 cm, the length of cut that lies in slits longer than a threshold, for thresholds from a sixty-fourth of a centimetre to 8.0 cm. All three start at the same 6.0 cm. The straight edge keeps it all until the threshold passes its own height; the staircase loses it all past 0.50 cm, its step; the hachures past 0.06 cm, two passes.
Fig. 2 For the same three joins, how much of the cut lies in slits longer than a threshold, for thresholds from a sixty-fourth of a centimetre to eight. All three start at six centimetres. The straight edge keeps all of it until the threshold passes six; the staircase loses all of it past half a centimetre, its step; the hachures lose three quarters of it past a single pass, where their teeth’s tips are, and the rest past two.

That figure is the result in one picture. Every curve starts at the same height, because the sum is the cartoon’s; the curves differ only in where they fall, because the division is the weaver’s. A join cannot reduce the cut, only cut it into shorter pieces.

Why shorter pieces are the whole of the gain

Since the sum is fixed, the case for a staircase or a hachure has to rest on what length does to a slit, and it does a great deal. A slit is a line with no thread across it, held closed only through the passes above and below its ends. A slit of two passes is held from both ends a sixteenth of a centimetre apart and cannot open at all; a slit of six centimetres is a buttonhole.

How far a slit gapes under a given pull grows faster than its length: in any sheet stiff enough to resist bending across the cut, the opening at the middle grows in proportion to the length, so the gap’s area grows as its square. That is a continuum argument, and a woven cloth is not a continuum — a cloth does not mind a hole the way a plate does — but the direction is not in doubt. Dividing a fixed total into kk equal slits divides the total area of gape by roughly kk, and 168 pieces of a join gape about as much in all as one piece did on its own.

So the staircase and the hachure are not ways of making less cut. They are ways of moving the same cut from the regime where it opens to the regime where it does not.

A finer weave divides it more finely, and sums the same

The first of these calculations found that a finer warp shortens a curve’s slits only as the square root of its sett, and leaves a straight vertical edge’s slit exactly as long as it was. The re-count explains both at once: a finer warp cannot change the total, which is a height, so it can only divide it.

Three ways to join two colours in a tapestry, and the cut each leaves. A join 6 cm tall between two colour areas, woven at 16 ends and 48 passes a centimetre three ways and drawn end by end, every gap no weft crosses drawn heavy. A straight edge: 1 slit, the longest 6.00 cm, 6.00 cm in all; a kilim's staircase: 12 slits, the longest 0.50 cm, 6.00 cm in all; hachures along the weft: 264 slits, the longest 0.04 cm, 6.00 cm in all. The total is the join's height in every case; the joins differ only in how they divide it. The drawing cannot show how far any slit opens.
Fig. 3 The same three joins woven at sixteen ends and 48 passes a centimetre. The straight edge is still one slit of six centimetres; the staircase, stepping one end every half-centimetre, is still twelve slits of half a centimetre, though it now leans half as far; the hachures are 264 slits of at most two passes, now 0.04 cm. Every one still adds up to six centimetres.

The straight edge does not divide at all: it crosses the same gap on every pass at any sett. The staircase divides into as many pieces as it has steps, which the designer chose and the sett did not. The hachures divide into more pieces because there are more passes to turn on, and each piece is two passes whatever a pass is. The sett decides how small the pieces of a hachured join are, and nothing else.

Only the warp’s direction changes the sum

There is one thing that does change the total, and it is the direction of the warp against the picture. The re-count makes that precise. The total is each outline’s height covered along the warp, so turning the cartoon replaces that height with the outline’s extent in another direction — its projection onto the warp.

For a rectangle ww wide and hh tall, turned so that its long side makes an angle θ\theta with the warp, the height it covers along the warp is w∣sin⁡θ∣+h∣cos⁡θ∣w|\sin\theta| + h|\cos\theta|, and the total cut is twice that.

A rectangle's total cut against the angle of the warp. A rectangle 12 by 30 cm woven as slit tapestry at 8 ends and 32 passes a centimetre, turned against the warp from 0 to 90 degrees: the total length of its slits, counted from the raster, beside the height its outline covers, 2(w·|sin θ| + h·|cos θ|), which they match within 0.26 cm. Upright it is cut by twice its height, 60.0 cm; turned, by twice its width, 24.0 cm; at its most it is cut by twice its diagonal. Averaged over every angle the cut is the perimeter times two over pi, 53.5 cm, which is Cauchy's formula for the mean width of a convex shape.
Fig. 4 A rectangle 12 by 30 cm woven at eight ends and 32 passes a centimetre, turned against the warp in steps of five degrees: the total cut counted from the woven raster (dots) and twice the height the outline covers along the warp, 2(w·sin θ + h·cos θ) (line). Upright it is cut by 60 cm, turned by 24; it is cut most, by twice its diagonal, near twenty-two degrees. Averaged over every angle it is cut by 53.5 cm, which is its perimeter times two over pi.

The average over every direction is worth stopping on. The mean of ∣cos⁡θ∣|\cos\theta| over a half-turn is 2/π2/\pi, so a straight segment of length ℓ\ell covers 2ℓ/π2\ell/\pi of height on average, and any outline covers its own length times 2/π2/\pi. A cartoon woven at a random angle is cut, on average, by the total length of its outlines times 2/π2/\pi. That is Cauchy’s formula for the mean width of a convex body, the same result that gives a needle’s chance of crossing a line in Buffon’s problem, and here it gives the expected cut of a tapestry from the length of the lines on its cartoon and nothing else.

Two cartoons turned through every angle

The two cartoons from the earlier calculation — a standing figure between two columns, and a landscape with a horizon and two hills — make the point at the scale of a picture.

A cartoon's total cut against the angle of its warp. A standing figure between two columns and a landscape with a horizon and two hills, woven as slit tapestry at 8 ends and 32 passes a centimetre with the cartoon turned against the warp in steps of five degrees, upright at 0 and on its side at 90. A standing figure between two columns: 700.0 cm upright, 588.0 cm at 45 degrees, 198.0 cm on its side; a landscape with a horizon and two hills: 112.0 cm upright, 242.1 cm at 45 degrees, 306.0 cm on its side. The total moves smoothly with the angle, because it is the outlines' height measured along the warp.
Fig. 5 The total cut of the standing figure and the landscape, woven at eight ends and 32 passes a centimetre with the cartoon turned against the warp in steps of five degrees, from upright (0) to on its side (90). The figure goes from 700 cm upright through 588 cm at 45 degrees to 198 cm on its side; the landscape from 112 cm through 242 to 306. Both move smoothly, because each is the outlines’ height measured along the warp.

The figure, made of verticals, is cut most upright and least on its side; the landscape, made of horizontals, the other way round. Averaged over every angle the figure is cut by 539 cm and the landscape by 230, in proportion to the total length of their outlines. Nothing about either curve is surprising once the total is known to be a projection, and that is the point: the sum is a fact about the drawing, readable off the cartoon before any warp is set.

The longest slit is not smooth at all

The longest slit behaves completely differently, because it is a fact about the division rather than the sum.

A cartoon's longest slit against the angle of its warp. A standing figure between two columns and a landscape with a horizon and two hills, woven as slit tapestry at 8 ends and 32 passes a centimetre with the cartoon turned against the warp in steps of five degrees, upright at 0 and on its side at 90. A standing figure between two columns: 120.0 cm upright, 2.53 cm at 45 degrees, 90.0 cm on its side; a landscape with a horizon and two hills: 1.8 cm upright, 3.16 cm at 45 degrees, 150.0 cm on its side. The longest slit is long only where a straight line of the picture runs along the warp, which for a picture of verticals and horizontals is at 0 and 90 degrees and nowhere between.
Fig. 6 The longest slit in the same two cartoons, on a logarithmic scale, as the cartoon is turned against the warp in steps of five degrees. The figure’s is 120 cm upright, where its columns’ edges run along the warp, 5.8 cm at five degrees and 2.5 cm at forty-five; the landscape’s is under 2 cm upright and 150 cm on its side, where its horizon runs along the warp. Between the two ends neither is longer than about six centimetres.

The figure’s longest slit is 120 cm upright and 5.8 cm when the cartoon is turned by only five degrees. At that angle the columns’ edges are no longer along the warp; each crosses one gap every dtan⁡85∘d\tan 85^\circ, which is about 1.4 cm, and the 480 cm of cut along them is divided into hundreds of short slits rather than four long ones. The landscape does the same thing at the other end: 150 cm on its side, about 6 cm five degrees short of it.

So a picture built of verticals and horizontals is divided worst at exactly two angles, and those are the two angles at which tapestries are woven. Upright and on its side are the only directions in which a straight line of the picture can run along the warp, and a straight line along the warp is the one outline whose cut no weaver can divide.

Why nobody weaves at five degrees

That raises the obvious question, and the answer is not at the loom but on the wall. A tapestry woven with its cartoon turned five degrees has to hang with its warp five degrees off the vertical, and a cloth hung off its threads is a cloth loaded on the bias.

The bias is a mechanism: a woven cloth pulled along neither thread system deforms by its crossings rotating, with very little resistance until the threads lock. A hanging loaded five degrees off its warp would shear under its own weight, drifting its columns into parallelograms and its hem out of level, and the slit it saved at the loom would be paid for in a picture that slowly leans. Every tapestry is woven with its warp along a principal direction of the picture because the picture has to hang along one of the cloth’s two strong directions, and the price of hanging square is that the picture’s straight lines run along the warp, where their cuts cannot be divided.

The weaver’s tools then fall into place. The staircase and the hachure are ways of getting, locally and on purpose, the division that a tilted warp would give everywhere — leaning an outline over a few centimetres rather than leaning the whole cloth. The kilim’s stepped diagonals are a tilted warp confined to one line — the same staircase a knitter draws when a fashioned edge takes a quantised angle, and for the same reason: a lattice can only lean by whole steps.

The sum as a census

There is a second way to read the re-count, and it connects the tapestry to the oldest question there is about a cloth’s structure. A draft is a matrix, and asking how many times the colour changes along each pick of a colour-and-weave cloth is a census of the matrix’s rows. A tapestry’s cartoon, sampled at its ends and passes, is a matrix too, of colours rather than of over and under, and its total cut is the number of unequal neighbours along its rows — the same kind of count as a float census, read for a different purpose.

An ordinary woven cloth changes weft colour only between picks, which is why a weft stripe is counted in pairs of picks and never cuts anything; a tapestry changes colour within a pass, and every change is a gap. That puts the tapestry’s whole joinery problem on a footing a designer could compute before the loom is dressed. Sample the cartoon at the sett; count the unequal neighbours along each row; the sum, over the pass density, is the cut. Then look for long vertical runs of unequal neighbours, which are the slits that will gape, and step or hachure those. Everything else is short already.

What the re-count does not decide

It does not say how the cut should be divided, only that dividing is all there is. Nor does it touch the first question asked of any draft — every slit tapestry, however cut, passes the test of whether a cloth hangs together, because its warp is continuous; the cut is local, and only the division decides whether it shows. Whether a hachured join reads at a distance as a smooth grade or as a comb depends on the teeth’s size against the viewer’s acuity, which is a question about what the eye makes of a surface rather than about cloth. Whether a staircase’s jag is acceptable is a question about the picture.

And it does not apply to joins that are not slits. When two wefts are interlocked in the gap on every pass, a weft does cross the gap — two wefts do, linked — and the cut is not divided but removed, at the price of a doubled weft in one column. That is outside the slit count entirely, and it is the one join that a weaver who needs a long straight vertical has.

The model, and the numbers in it

A cartoon is a list of shapes in centimetres, painted in order, sampled at the centre of every end and every pass along each pass’s line; it is woven at eight ends and 32 passes a centimetre unless a figure says otherwise. A gap is open on a pass when the ends either side carry different colours. The total cut is counted twice — as the sum of every slit’s length, and as the number of colour changes on each pass summed over passes, over the pass density — and the two agree exactly. A turned cartoon is rotated about its centre onto a canvas large enough to hold it, and gaps against the empty canvas outside the picture are not counted. The three joins are a six-centimetre boundary drawn straight, as a staircase stepping one end every half-centimetre, and as triangular teeth eight passes thick reaching a centimetre and a half from alternate sides. The gape argument is a continuum’s and is stated only as a direction.

What the drawings cannot show

Gape. Every figure counts and measures cuts; none of them opens one. The case for dividing a cut rests on a slit’s gape growing faster than its length, which is true of a continuum and assumed rather than computed for a weft-faced cloth.

Tone. Hachures exist to grade colour, and the figures show them as cuts. What tone they make, and at what distance their teeth disappear into it, is not computed.

The interlocked join, which removes the cut and so has no place in a count of it.

The bias at a small angle. That a tapestry hung five degrees off its warp would shear is the bias argument applied by analogy; how far it would shear under its own weight is not computed.

Who found it and when

The staircase and the hachure are among the oldest devices in tapestry, the first characteristic of slit-woven kilims and the second of European tapestry from the Middle Ages on; both are described in every account of the technique. Cauchy’s formula for mean width is nineteenth-century and Buffon’s needle eighteenth. What is added here is the re-count — that the total slit length is the colour changes summed over passes, so that it is the outlines’ height along the warp and is fixed by the cartoon and the warp’s direction; that the straight edge, the staircase and the hachure divide one total; the rectangle’s 2(wsin⁡θ+hcos⁡θ)2(w\sin\theta + h\cos\theta) and its mean of perimeter times 2/π2/\pi; and the collapse of the longest slit, from 120 cm to under 6, as soon as no straight line of the picture runs along the warp.

Still open: whether hachures can be as fine as the eye

Hachures divide a join into pieces two passes long whatever their teeth, so their slits are settled. What is not settled is their teeth: a tooth eight passes thick and a centimetre and a half long reads as a comb from close to and as a tone from far enough away, and the distance between those two readings is a property of the tooth’s dimensions and the eye’s acuity.

The calculation would take a hachured band, compute the share of each colour across it as a function of position, and compare the comb’s spatial frequency with the frequency an eye can resolve at a given viewing distance. If a tooth short enough to vanish at the distance a tapestry is hung is also long enough to grade the colour over a useful width, then the hachure is a free join for any vertical edge a picture has — one that removes the cut’s gape and adds nothing a viewer can see — and the long sewn slits of the great tapestries were a choice about drawing rather than a necessity of weaving.

Shares its objects with

Essays naming at least two of the same things, that neither author linked.

Named objects

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

BiasCensusFigure and groundPlain weaveSlitTapestryWeft-faced