Compound and figured cloths

A tapestry's slit is the chord of one end

In a tapestry no weft runs from selvedge to selvedge. Each colour has its own bobbin and turns back at the edge of its area, so where two areas meet along the warp nothing crosses between the two ends either side, and the cloth has a cut in it. The cut's length is not a matter of the weaver's care. Where a curved outline turns vertical it stays between the same two ends for the length of the chord whose sagitta is one end spacing — at least √(Rd) and at most √(8Rd) wherever the cartoon is laid. A face's ten-centimetre curve at eight ends a centimetre leaves 1.1 to 3.2 cm, and doubling the warp takes less than a third off it.

Worth reading first: Balance, and what an unbalanced cloth does · Does it hang together · A fashioned edge has a quantised angle.

Almost every woven cloth has a weft that runs from one selvedge to the other. A pick may float, it may be bound in a satin or a twill, it may change colour by being a different pick, but it goes the whole way across, and so every pair of neighbouring warp ends is tied together by every pick. That is so basic that the criterion for whether a cloth hangs together never had to mention it.

A tapestry gives it up. The ground weave is the plainest there is, a weft-faced tabby, but no weft crosses the full width. Each area of colour has its own small bobbin, which is passed through the shed only as far as that colour reaches and then turned back. Where a red area meets a blue one, the red weft turns round the last end of the red and the blue weft round the first end of the blue, and on that pass nothing at all crosses the gap between those two ends. If the boundary stays in the same gap for a run of passes, the cloth has a cut in it: a line along the warp with whole cloth either side and no thread across.

That cut is the slit, and it is the most characteristic flaw of the oldest picture-weaving there is. Kilims are woven with slits left open, and their whole geometric vocabulary of stepped diagonals is often explained as a way of keeping the slits short. European tapestries of the great workshops were woven with slits and sewn up afterwards, and the sewing is what a conservator most often finds failed. What follows is what decides how long a slit is, and it turns out to be a short calculation with a bound in it that no weaver’s skill can get under.

A circle woven in slit tapestry, laid two ways against the warp. A circle of radius 3 cm woven as a slit tapestry at 8 warp ends and 32 passes a centimetre, drawn end by end at its true shape, with every gap no weft crosses drawn heavy. Laid 0.13 of an end spacing past an end centre, its longest slit is 0.63 cm; laid 0.97 of an end spacing past an end centre, its longest slit is 1.69 cm. The slits stand where the outline turns vertical, at its left and right, and no placement makes them shorter than the square root of the radius times the spacing, 0.61 cm here; the worst is 1.73 cm. The drawing cannot show how far a slit opens.
Fig. 1 One circle of radius 3 cm woven as a slit tapestry at eight ends and 32 passes a centimetre, drawn end by end at its true shape, with every gap the weft does not cross drawn heavy. The two panels differ only in where the cartoon was laid against the warp. In the first the longest slit is 0.63 cm and in the second 1.69 cm; the bounds that hold for every placement are 0.61 and 1.73 cm. The slits stand at the left and right of the circle, where its outline runs along the warp.

Why a weft-faced cloth can afford to stop its weft

The reason a tapestry can be woven this way at all is the reason it is weft-faced. In a balanced cloth both systems show about equally, so a picture made by changing the weft’s colour would be half-hidden behind an even screen of warp. A tapestry’s warp is set open, its weft is beaten down hard until it packs between the ends and covers them completely, and the warp becomes scaffolding that no one sees. Every visible mark is weft.

Once every visible mark is weft, there is no reason for any one weft to cross areas where it would be hidden anyway. It is also why a tapestry can show any colour anywhere: a woven cloth that makes its tones from the weave itself has only a handful of greys to work with, and a tapestry has as many as it has bobbins. A weft carried behind a red area to reach the next patch of blue would have to float on the back, and the face would show nothing of it. So each colour is woven only where it shows. That is what makes a tapestry a picture rather than a pattern: a repeat is no constraint at all, because nothing about one area is shared with any other.

Each colour area is, in effect, a small cloth with its own two selvedges, and a plain weave is the one weave whose edge end changes face between every pair of picks, so every turn round it is caught. That is one reason the ground of a tapestry is always a tabby. What it costs is the one property every other cloth here gets free. A pick that crosses the width is the thing that holds neighbouring ends together; take it away at a boundary and those two ends are held only through the cloth above and below the boundary’s run.

The cut is decided where the outline runs with the warp

Follow an outline up the cloth pass by pass. On each pass the boundary sits in one particular gap between ends, the gap where the colour changes. When the outline leans, it moves from gap to gap as the passes rise, and each gap is open only for the few passes the outline spends in it. When the outline runs straight up the warp, it stays in one gap for as long as it runs, and that gap is open the whole way.

So a slit’s length is the height an outline spends in one gap. That is a property of two things only: the outline’s shape, and how far apart the gaps are, which is the end spacing dd. The pass density does not appear. Passes are many times denser than ends in a weft-faced cloth — 32 a centimetre against eight here, which is what beating a soft weft down hard enough to hide the warp buys, and which on a power loom would be a force budget of its own — and they decide how finely the slit’s two ends are placed, but not how long it is.

That is the first thing worth noticing. The quantity that gives a tapestry its fine vertical detail, the packed weft, buys nothing at all against its slits. The slits belong to the warp.

A straight edge leaves the same slit in every gap

The simplest outline is a straight edge at an angle θ\theta to the horizontal. To cross from one gap to the next it has to travel one end spacing sideways, and it rises dtan⁡θd \tan\theta while it does. So it spends exactly that height in every gap it crosses: every slit along a straight edge has the same length, dtan⁡θd\tan\theta.

At 45 degrees and eight ends a centimetre that is an eighth of a centimetre, four passes. That is nothing; a gap open for four passes is a pinhole at most. At sixty degrees it is about two millimetres. The length rises slowly and then, near the vertical, very fast, because the tangent does.

The slits along a straight edge, against the edge's angle. A straight boundary between two colour areas in slit tapestry, at an angle to the horizontal, leaves the same slit in every gap it crosses: the end spacing times the tangent of the angle. At 6 ends a centimetre each slit passes 1 cm only above 80.5 degrees; at 8 ends a centimetre each slit passes 1 cm only above 82.9 degrees; at 12 ends a centimetre each slit passes 1 cm only above 85.2 degrees; at 16 ends a centimetre each slit passes 1 cm only above 86.4 degrees. A 45-degree edge at 8 ends a centimetre leaves slits an eighth of a centimetre long, which is four passes at 32 passes a centimetre; a vertical edge leaves one slit as long as the edge.
Fig. 2 A straight boundary between two colour areas, at an angle to the horizontal, leaves the same slit in every gap it crosses: the end spacing times the tangent of the angle. The four curves are warps of 6, 8, 12 and 16 ends a centimetre. Each passes one centimetre only above 80.5, 82.9, 85.2 and 86.4 degrees respectively, and a vertical edge leaves a single slit as long as itself.

Take one centimetre as a yardstick for a slit long enough to matter. It is not a measured threshold, and the right number depends on the yarn, the weight of the hanging and how the weaver will finish it; it is a length at which a cut is plainly visible and plainly more than a pinhole. At eight ends a centimetre, a straight edge exceeds it only when it stands within about seven degrees of the vertical. At sixteen ends, within about three and a half.

So for straight edges the rule is sharp: anything visibly sloped is safe, and anything nearly vertical is one long cut. That matches what weavers of slit tapestry do. A kilim’s diagonals are steep enough to read as diagonals and never so steep as to stand, and where a vertical line is wanted, the design steps it — a staircase of short vertical runs offset sideways by one end, which converts a long slit into several short ones at the cost of a jagged line.

A curve cannot dodge its vertical

Curves are where it gets interesting, because a curve has to pass through the vertical somewhere if it closes, and there it cannot be stepped without changing the shape.

Near the point where a curve with radius of curvature RR runs vertically, its horizontal distance from the vertical tangent grows as the square of the height, x≈y2/2Rx \approx y^2/2R. The boundary stays in the gap the tangent sits in until it has moved inward by one end spacing — or, more exactly, until it has moved past the next end’s centre. A chord of a circle that stays within a strip of width δ\delta of its edge is the chord whose sagitta is δ\delta, and its length is 22Rδ2\sqrt{2R\delta} near enough.

A slit at a curve’s vertical is the chord whose sagitta is one end. That is the same arithmetic a lens grinder uses to measure a surface’s curve with a spherometer, or a surveyor to find the radius of a bend from the offset at its middle. Here it measures how far a weft-faced picture is cut along its outlines.

Where the cartoon lies decides which run is longest

The sagitta is not exactly one end spacing, because the tangent need not sit on an end. Call the distance from the last end centre inside the curve to the tangent a fraction ϕ\phi of a spacing. Then the outermost run — the gap the tangent itself sits in — is the chord with sagitta ϕd\phi d, which is 22Rϕd2\sqrt{2R\phi d}. The next run in, on either side of it, lasts while the edge moves from ϕd\phi d to (1+ϕ)d(1+\phi)d inside the tangent, which is 2Rd (1+ϕ−ϕ)\sqrt{2Rd}\,(\sqrt{1+\phi}-\sqrt{\phi}). Every run further in is shorter than that.

So the longest slit is the larger of two numbers, one growing with ϕ\phi and one shrinking, and the weaver controls ϕ\phi only by sliding the cartoon sideways behind the warp.

The longest slit at a curve's vertical tangent, against where the cartoon is laid. For a circle of radius 10 cm at 8 ends a centimetre, the length of the outermost slit (the run in the gap the tangent sits in) and of the first flanking slit, as the tangent moves across one end spacing. The longest slit is the larger of the two. It is least, 1.12 cm, with the tangent an eighth of a spacing past an end, where the two are equal and both are the square root of the radius times the spacing (1.12 cm); it is greatest, 3.16 cm, just before a new end enters the circle. Averaged over placements it is 2.17 cm.
Fig. 3 For a circle of radius 10 cm at eight ends a centimetre, the outermost slit (thin, rising) and the first flanking slit (thin, falling) as the tangent moves across one end spacing, and the longest of all (heavy). The longest is least, 1.12 cm, when the tangent sits an eighth of a spacing past an end, where the two runs are equal; it is greatest, 3.16 cm, just before another end enters the circle. Averaged over placements it is 2.17 cm.

The two are equal where 2ϕ=1+ϕ−ϕ2\sqrt{\phi} = \sqrt{1+\phi}-\sqrt{\phi}, which is 9ϕ=1+ϕ9\phi = 1+\phi, so ϕ=1/8\phi = 1/8. At that placement both runs are 22Rd/8=Rd2\sqrt{2R d/8} = \sqrt{Rd}. As ϕ\phi approaches one, the outermost run approaches 22Rd=8Rd2\sqrt{2Rd} = \sqrt{8Rd}. So whatever the placement, the longest slit at a vertical tangent lies between Rd\sqrt{Rd} and 8Rd\sqrt{8Rd} — a factor of 222\sqrt 2 between the luckiest cartoon and the unluckiest, and no luck at all below the floor.

The floor is the part worth remembering. It says that the weaver who slides the cartoon to the best position has not removed the slit; they have split it into three roughly equal runs instead of one long one, each about Rd\sqrt{Rd}.

Every slit round a circle, by its slope

Put the straight-edge rule and the curve rule together and the whole outline of a shape is accounted for. Away from its verticals, every part of a curve behaves locally like a straight edge at its own angle, and leaves slits of about dtan⁡θd\tan\theta. Near the verticals, tan⁡θ\tan\theta runs to infinity and the curvature takes over, capping the slit at the sagitta chord.

Every slit round a woven circle, by the slope of the outline. The 282 slits a circle of radius 10 cm leaves at 8 ends and 32 passes a centimetre, each plotted at the angle the outline makes with the horizontal where it sits. Along most of the outline each slit is the one a straight edge at that angle would leave, the end spacing times the tangent of the angle (0.13 cm at 45 degrees). Near the vertical the tangent runs to infinity and the circle's curvature caps it: the longest is 2.25 cm, inside the bounds 1.12 to 3.16 cm. A straight edge passes 1 cm only above 82.9 degrees.
Fig. 4 The 282 slits a circle of radius 10 cm leaves at eight ends and 32 passes a centimetre, each plotted at the angle the outline makes with the horizontal where it sits. Along most of the circle every slit lies on the straight-edge rule, the end spacing times the tangent of the angle. Near the vertical the rule runs off to infinity and the circle’s curvature caps it: the longest slit here is 2.25 cm, inside the bounds of 1.12 and 3.16 cm.

Two things show in that picture. One is how little of the outline matters: of 282 slits on the circle only ten are longer than half a centimetre, and only two longer than a centimetre — one at each side, exactly where the outline runs vertical. The other is that the circle’s long slits sit below where the straight rule would put them at the same angle, because a curve passes through the vertical without staying there. The straight rule is a ceiling for any curve, and only a straight vertical line reaches it.

That is the precise sense in which curves are the hard case and straight lines the easy one. A straight edge can always be leaned or stepped. A closed curve cannot avoid the vertical, and at the vertical it is the curvature, not the weaver, that decides.

A curve’s slit grows as the square root of its size

Both bounds go as R\sqrt{R}. A curve four times larger leaves a slit twice as long. A face’s oval, a fold of drapery, the round of a shield: each has a vertical somewhere, and each leaves a slit there that grows with the square root of its size.

How long the slit at a curve's vertical tangent is, against the curve's radius and the sett. The longest slit a slit tapestry leaves where a curved outline turns vertical, against the outline's radius of curvature there, for warps of 6, 8, 12, 16 ends a centimetre. Every curve lies between the square root of the radius times the end spacing and the square root of eight times that, whatever the placement: at 8 ends a centimetre a radius of 10 cm gives 1.12 to 3.16 cm. At 6 ends a centimetre every curve of radius over 6.0 cm carries a slit longer than 1 cm wherever it is laid; at 8 ends a centimetre every curve of radius over 8.0 cm carries a slit longer than 1 cm wherever it is laid; at 12 ends a centimetre every curve of radius over 12.0 cm carries a slit longer than 1 cm wherever it is laid; at 16 ends a centimetre every curve of radius over 16.0 cm carries a slit longer than 1 cm wherever it is laid. Doubling the sett shortens every slit by the square root of two.
Fig. 5 The longest slit at a vertical tangent against the outline’s radius of curvature there, for warps of 6, 8, 12 and 16 ends a centimetre: each line is the mean over every placement of the cartoon, and the shaded band for eight ends runs from the best placement to the worst. At eight ends a centimetre every curve with a radius over 8 cm carries a slit longer than one centimetre wherever it is laid; at sixteen ends the same holds from 16 cm.

Set the floor equal to the yardstick and it gives a radius: Rd>L\sqrt{Rd} > L whenever R>L2/dR > L^2/d. With LL one centimetre, that is RR greater than the sett in ends a centimetre, read as centimetres. At eight ends a centimetre, every curved outline with a radius over eight centimetres carries a slit longer than a centimetre wherever it turns vertical, and no placement of the cartoon prevents it. At sixteen ends the limit is sixteen centimetres.

That is a fact about every large tapestry. Their figures are life-size or larger, and the outlines of a body, a robe or a tree trunk run vertical over radii of tens of centimetres. At the setts the great workshops used — from about five ends a centimetre for coarse work to a dozen or more for the finest — such outlines leave slits of several centimetres, and they did. That is why those tapestries were sewn.

What a finer warp buys

The natural remedy is a finer warp, and the arithmetic says how little it does against slits.

The finest horizontal step a tapestry can draw is one end; halve the spacing and it can draw details half as wide. But the slit goes as d\sqrt{d}, not as dd. Doubling the sett halves the resolution’s step and takes only a factor of 2\sqrt 2 off every slit, which is less than a third.

What a finer warp buys against a tapestry's slits. For a curve of radius 10 cm, the longest slit at its vertical tangent between the best and worst placement, and its mean over placements, against the warp's sett; beside it the width of one end, which is the finest horizontal step the cloth can draw. From 8 to 16 ends a centimetre the step halves, 0.13 to 0.06 cm, and the mean slit falls only from 2.17 to 1.54 cm. A curve of this radius carries a slit longer than 1 cm wherever it is laid up to 10.0 ends a centimetre.
Fig. 6 For a curve of radius 10 cm, the longest slit at its vertical tangent between the best and worst placements (shaded) and its mean over placements, against the warp’s sett, beside the width of one end, which is the finest horizontal step the cloth can draw. From eight to sixteen ends a centimetre the step halves, from 0.13 to 0.06 cm, and the mean slit falls only from 2.17 to 1.54 cm. A curve of this size carries a slit over a centimetre wherever it is laid at any sett up to ten ends a centimetre.

So a finer warp is a far better buy for detail than for joinery. A workshop that doubled its sett to make its faces finer got faces twice as fine and slits only 29 per cent shorter, and paid for it in twice the warp and nearly twice the weaving time, since twice as many ends must be wrapped on every pass. The slits remained a sewing job.

This is also why the trade’s solutions to the slit are not finer setts but different joins. A weaver can share an end between two colours, turning each round it in alternating groups of passes, which bounds every slit at the group’s height at the price of a serrated edge; or link the two wefts round each other in the gap on every pass, which closes the slit altogether at the price of a small ridge on the back. Neither is modelled here. Both exist because the geometry above leaves no other remedy that keeps the picture’s shape.

A cut that the integrity criterion does not see

It is worth being exact about what a slit is not. It is not a failure to hang together in the sense the connectivity test uses. The warp runs continuously from the bottom of the tapestry to the top, every colour area is woven onto it, and the whole cloth is one connected thing; the connectivity test would pass it at once.

What a slit removes is not connection but local connection: along its length, two neighbouring ends have no thread between them, so the cloth is held across that line only through its two ends, above and below. That makes a slit closer kin to the hole in a mock leno, where threads slide together and leave space with nothing crossing it, than to a separated layer. And like a hole, a slit is relatively harmless in a cloth that carries load thread by thread — a cloth does not mind a hole the way a plate does — so long as no load has to cross it. Whether load does cross it depends on how the tapestry hangs, which is a different question from how it was woven.

The same staircase a knitter draws

The arithmetic here has a close cousin in the other construction that builds a shape on a lattice. A fully fashioned knitted edge can only take angles whose tangents are fractions of the loop’s aspect, because it moves by whole wales at whole courses. A tapestry’s edge also moves by whole ends, and the slit is the length of each tread of its staircase.

The difference is in what the treads cost. A knitter’s staircase is invisible once the fabric relaxes; a tapestry’s leaves a cut at every tread, long where the outline is steep and longest where it is vertical. Both constructions draw curves out of steps, and only one of them is cut at every step.

When the weft does cross: a brocade

It helps to compare the opposite arrangement, where a figure is made by a weft that does cross. In a brocade a supplementary weft is carried behind the ground between figures, and its cost is a float on the back as long as the gap between motifs. A tapestry’s weft never crosses the gap, so it has no float at all, and its cost is the slit.

Those are the two prices for putting a colour where it is wanted and nowhere else. A brocade pays in a long float behind the figure, which can snag; a tapestry pays in a cut along the figure’s edge, which can open. Neither price can be removed without changing the construction, and both are set by geometry a designer could compute before the loom is dressed.

The model, and the numbers in it

The cloth is a raster: a cartoon of shapes in centimetres, painted in order over a ground, sampled at the centre of every end and every pass, at eight ends and 32 passes a centimetre unless a figure says otherwise. A gap between neighbouring ends is open on a pass when the two carry different colours, and a slit is a maximal run of open passes in one gap. For the curve results the runs were also computed exactly, from the chord of a circle, and the raster’s longest slit agrees with the exact one to within two passes, at 400 passes a centimetre, at the three placements tested. The bounds Rd\sqrt{Rd} and 8Rd\sqrt{8Rd} were found in the parabolic approximation and checked against a sweep of 800 placements of the exact circle, whose least lies at an eighth of a spacing. The one-centimetre yardstick is a choice, not a trade figure.

What the picture cannot show

The gape. A slit is drawn as a line because its length is what the model computes. How far it opens depends on the weft’s packing, the warp’s tension and above all on the load across it, and nothing here models a cloth deforming round a cut.

The pass pair. Weavers turn at an area’s edge on the pass and its return together, so the slit’s two ends are placed to the nearest pair of passes rather than the nearest pass. That moves each end by a pass and leaves the length unchanged at the precision quoted.

The edge end’s draw. A weft turning round the same end on every pass pulls it slightly towards its own area, and an end at a long slit is pulled both ways at once from its two sides. Weavers know the slit’s ends draw in and the cut becomes a narrow lens; the model keeps every end straight.

The joins. Dovetailing and interlocking exist to remove slits and are not modelled. The bound here is for the slit tapestry, which is the case they are measured against.

Who found it and when

Slit tapestry is among the oldest woven picture techniques, found in Egyptian linen of the second millennium BC, in Coptic and Andean work, in the kilims of Anatolia and Persia, and in the tapestries of the European workshops from the fourteenth century on, where slits were sewn as the tapestry came off the loom. That kilim design favours steep diagonals and stepped verticals because they keep slits short is the common explanation in the literature on rugs. The sagitta is as old as geometry. What is added here is the slit as the chord whose sagitta is one end; the exact bound between Rd\sqrt{Rd} and 8Rd\sqrt{8Rd} over every placement, with its floor at an eighth of a spacing; the straight edge’s dtan⁡θd\tan\theta and its critical angles; the independence from the pass density; and the factor of 2\sqrt 2 that a doubled warp buys.

Still open: whether turning the cartoon helps

Every result here is for a cartoon woven upright, with the picture’s vertical along the warp. The great European workshops more often wove theirs on its side, with the picture’s vertical across the warp, so that a figure’s long verticals became horizontals on the loom and stopped being slits at all.

That is not a free exchange, and its terms are computable from what is here. A tall shape’s verticals are its flattest parts, so turning it shortens their slits by a power of its aspect; but every horizontal in the picture becomes a vertical on the loom and is cut instead, and a tapestry turned to weave hangs with its warp horizontal, so those cuts run across the direction its weight pulls. Woven on its side, a tapestry’s slits carry its weight counts what is gained and what is put under load.

What links here

Computed from the collection rather than written here: the essays that point at this one.

Reads more easily once this is understood

Essays that name this one as worth reading first.

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.

Cloth integrityFigure and groundPlain weaveSettSlitTapestryWeft-faced