A braid is a third way to hold threads

Two mechanisms have been named. A woven cloth is held by friction, because its threads pass over one another and back, so their linking number is nought and nothing is threaded through anything. A knitted fabric is held by linking, because every loop passes through the loop below, so its linking number is one per wale and friction is optional.

A braid is neither, and putting it beside the other two is the fastest way to see what the two mechanisms actually are.

What a braid does #

A braid has no warp and no weft. Every strand is the same kind of thing, and every strand travels across the structure: it moves from one edge towards the other, passing alternately over and under the strands coming the other way, reaches the edge, turns, and comes back.

That is a third arrangement and it is worth being precise about how it differs.

In weaving, a thread stays in its own place across the width for the whole length of the cloth. That is what a reed does and it is why a weave can be written as a matrix of ups and downs.

In knitting, a thread travels across the fabric once per course and is drawn through the course below.

In braiding, a thread travels across and back repeatedly, and is drawn through nothing.

The number #

Take any two strands of a flat braid and ask the linking question.

Both strands wander back and forth across the structure, crossing one another several times per repeat. Some of those crossings are one strand over the other and some are under, and over a full repeat they cancel: each strand returns to the same side of the other as it started.

So the pairwise linking number is nought, exactly, and for a reason that has the same shape as the woven one — a round trip through nothing.

That is surprising the first time it is said, because a braid does not look like something with no linking in it. Its strands are visibly wound about one another in a way a woven cloth’s are not.

So what does hold it #

The honest answer is that a braid is held by something the pairwise linking number cannot see, and saying so precisely is the useful part.

Pairwise linking numbers do not determine a link. Two curves can be unlinked from one another and inseparable when a third is present — the standard example is a set of three rings, no two of which are linked, that cannot be taken apart. A braid is that situation with many components.

So a braid’s integrity is a collective property. No two strands are holding one another; the whole set is holding itself, and removing any one leaves the rest still held.

That is a genuinely different mechanism from the other two, and it is why a braid needs no reed, no beat-up, no sett and no minimum density: nothing about the holding depends on how hard the strands are pressed together or on how many crossings there are per unit length.

What a braid does when it fails #

Each mechanism has its own failure and a braid’s is neither of the other two.

A woven cloth frays: cut it and its threads slide out one at a time, over a short length, and stop, because friction accumulates.

A knitted fabric runs: break one loop and the loop threaded through it is free, and the failure propagates to the edge.

A braid unlays. Cut it and the strands at the cut end separate, and the separation travels — slowly, and only as far as it is worked. A cut braid does not shed a fringe and does not unfasten itself; it comes undone if it is pulled apart, one crossing at a time, and stops the moment it is left alone.

That is exactly what a collective hold predicts. There is nothing to slide out, because nothing is held by friction alone; and there is nothing to propagate, because no single strand was holding any other.

Three mechanisms, three densities #

Putting the three side by side settles a question this collection has treated separately in three places: what decides how open a structure can be.

Woven: friction, so there is a minimum sett. Below it the cloth slips. This collection has a whole field about where that number is.

Knitted: linking, so there is no minimum at all. A hand crochet with a centimetre between its loops is a fabric.

Braided: collective, so there is no minimum either — but for a different reason, and with a different consequence. A braid can be made as open as anybody likes and it will hold; what it cannot be is flat and wide, because the strands have to travel across the whole width and back, so the width is limited by how far a strand can be carried.

That is why braids are narrow. It is not a limitation of the machine; it is the mechanism.

Which is why a braid is the third system #

This collection has already argued that a braid is the third system — a way of making fabric that is neither warp-and-weft nor loop-and-course, with its own geometry and its own bias behaviour.

That rung made the case from the strand paths: a braid’s strands run at an angle to its axis, so a braid extends and contracts as its angle changes, which is why a braided sleeve grips and why a Chinese finger trap works.

The linking picture adds the other half. A braid is the third system not only in how its strands run but in what holds them, and the two facts are related: strands that travel across the structure and back can neither stay in place to be pressed together nor be drawn through one another, so the only mechanism left is the collective one.

What the maypole tells anybody watching #

The clearest picture of the mechanism is the machine that makes it, and a braiding machine is unusually legible.

A maypole braider carries its bobbins on carriers that run round two interlocking serpentine tracks, half going clockwise and half anticlockwise, weaving between one another as they go. Every carrier travels the whole way round; none of them stays put.

That is the mechanism made visible. There is no warp — nothing stationary — so there is nothing for a moving thread to be pressed against, and therefore nothing for friction to grip. And there is no drawing-through — no carrier passes its yarn through a loop of another — so there is nothing to link.

What there is, is every strand passing every other strand repeatedly in both senses, which is the collective condition described above and is exactly what the two counter-rotating tracks produce.

A loom and a knitting machine each have a fixed frame that the yarn is worked against: a reed, or a bed of needles. A braider has none, and its product is held by nothing that any part of the machine supplies.

Why a braid has no face #

A small consequence with a large reach, and it distinguishes braid from both the others at a glance.

A woven cloth has a face and a back, because the two thread systems are not interchangeable: the warp is on top in some places and the weft in others, and which is which is the weave. A knitted fabric has a face and a back for a stronger reason — the loops all point the same way — and this collection has argued that a cloth has an outside at some length.

A braid has neither. Every strand is the same kind of thing doing the same thing, and turning a braid over shows the same structure. That is a symmetry the other two do not have, and it comes from the same fact: there is no distinguished direction in a braid because there is no stationary system for a moving one to be worked against.

It has a practical edge. A braid cannot be one thing on one side and another on the other, so a braided structure that needs two different surfaces has to be made as two braids — which is why braided ropes have separate cores and covers rather than graded structures.

What was counted, and how #

The pairwise result is structural rather than numerical and is stated as such.

A flat braid’s repeat carries each strand from one edge to the other and back, and over that repeat every crossing with a given partner is matched by an opposite one. That is a property of the braid’s own periodicity: a strand that returns to its starting position relative to another strand has, by definition, unwound whatever it wound.

The linking integral is not run on a braid here, because this collection has no braid geometry to run it on — the same gap that stops it computing a leno. What is run is the check on the instrument: the same integral returns one for two rings threaded through one another and nought for the three arrangements that are not, to four parts in ten thousand.

The claim that pairwise unlinking does not imply separability is standard and is not this collection’s own. It is what makes the braided case interesting rather than a restatement of the woven one.

The tubular case, which is where braid earns its living #

Most braid is not flat. A rope, a shoelace, a cable sheath and a braided hose are tubular, and the tubular case sharpens everything above.

In a tubular braid the strands travel round the circumference rather than back and forth across a width, so they never turn: each strand spirals continuously in one sense, half of them one way and half the other. There is no edge, so there is nothing to reflect off.

Ask the linking question of two strands going the same way and the answer is that they are parallel helices and never cross at all. Ask it of two going opposite ways and they cross repeatedly — twice per revolution — with alternating sense, and cancel.

So the tubular result is the same nought, arrived at differently, and the structure is held by the same collective property. What the tubular case adds is a mechanism the flat one does not have: the braid can tighten on its contents. Pulling a tubular braid lengthwise reduces its diameter, because the helix angle steepens, and that is the whole of why a braided sleeve grips a cable and why a finger trap holds a finger.

That is a hold that is neither friction between strands nor linking: it is friction against something else, generated by the structure’s own geometry. A fourth mechanism, arguably, and one no flat fabric has.

Why braided rope replaced laid rope #

The comparison worth making is against the other way of making cordage, because it is the same distinction one level up.

A laid rope is a fold: three strands twisted about a common axis, each itself a fold of yarns twisted the other way. It is held the way a folded yarn is held, by the twist pressing the components together, and its components wind without ever unwinding.

A braided rope is a braid: its strands cross and recross and their winding cancels.

The consequences follow from that difference and are entirely practical. A laid rope has a residual torque and rotates under load, because its winding is real and unbalanced; a braided rope does not, because its winding cancels. A laid rope hockles — the rope-scale version of a yarn snarling — and a braided rope does not.

So the reason braided rope displaced laid rope for most uses is the linking number, stated in the working language of ropes: braid does not spin the load.

Where the model stops #

There is no braid geometry here. Every number in this collection about a woven cloth comes from a matrix and every number about a knit comes from a solved loop. A braid has neither, so nothing on this rung is computed from strand paths.

The collective hold is named rather than quantified. Saying that a braid is held by a property no pair of its strands has is correct and is not a mechanism anybody can put a force to. A proper account would need the whole braid’s configuration space and an argument about what deformations are available to it, which is a much larger piece of work.

And the unlaying is a description. How fast a braid unlays, and under what load, is not computed and would need friction as well as topology, because a braid’s strands do press on one another whatever the linking says.

The generalisation #

The rung’s value is in what it does to a classification, and the shape of that is worth extracting.

Two mechanisms had been identified and it was tempting to treat them as a dichotomy: held by friction, or held by linking. A braid does not fit either, and the reason it does not fit is instructive — the linking number is a pairwise quantity, and a structure can be held by something that is not a property of any pair.

That is a warning about the instrument rather than about braids. Every linking number in this work is computed between two curves, and a fabric is not two curves. The woven result — nought for every pair — was read as “friction holds it”, and that reading is correct for a cloth; but the braid shows that “every pair unlinked” and “held by friction” are not the same statement, and the first does not imply the second.

So the woven conclusion needs a footnote it did not have: a cloth is held by friction because its threads are also pairwise separable, which is a stronger condition than pairwise unlinked and which happens to be true for a weave because its threads run in two families of parallel lines. That extra fact was assumed silently, and the braid is what makes it visible.

What this collection owes the braid #

Three of this collection’s nine fields have essays that would read differently if the braid case had been in view when they were written, and it is worth saying which, because it is a queue rather than a complaint.

The integrity check asks whether a draft’s threads form one connected component and answers exactly. It is a pairwise question asked over a graph, and the braid is the standing reminder that a pairwise question can miss a collective property. Nothing about the integrity check is wrong — a woven cloth’s threads really are held pairwise — but the check’s justification assumes what the braid shows is not automatic.

The preform essays treat braided reinforcements alongside woven ones and compare them by fibre angle, drapability and shear locking. None of them says that the two hold by different mechanisms, which matters for what happens when the resin is left out.

And the bias essays treat a braid as a woven cloth already on the bias, which is right about the angles and silent about the holding.

None of those is a defect. All three are places where a rung written now would say one more thing, and recording that is cheaper than pretending the collection was complete.

Who found it, and when #

Braiding is prehistoric. The braid group as a mathematical object is Artin’s, from 1925, and it is the natural formalism for exactly this: a braid is a word in generators that say which strand crossed which, and two braids are the same if their words can be transformed into one another.

That the pairwise linking numbers of a braid’s strands vanish while the braid itself is nontrivial is elementary in that setting and is the standard first example of why linking numbers are a weak invariant.

What is this collection’s own is the placement: putting braiding beside weaving and knitting as a third holding mechanism, and noticing that the instrument used for the first two is blind to the third in a way that also weakens what it said about the first.

Where the ladder goes next #

The torsion ladder has run out of structures and closes with an accounting: what the second stiffness bought, what it cost, and which of this collection’s standing questions it did not touch. Where a torsion model stops is that reckoning.

After it this work turns from what a thread does about its own twist to what two threads do about occupying the same space, which begins with a measurement nobody had made: whether the fabric this collection computes with actually fits together. It does not.