The float in a knit

The float is this site’s most productive object. One number behind lustre, drape, snagging, abrasion and how closely a cloth can be set; five rungs of argument; and in every one of them the float is the same thing — a length of thread lying on the surface with nothing holding it down.

A weft knit has floats too. A missed needle takes no yarn, so the yarn passes straight across the back of the fabric between the loop on one side and the loop on the other. The trade calls it a float, and the word is not borrowed loosely: it is the same physical object, an unsupported run of yarn between two points where the structure holds it.

Whether the arguments transfer is a different question, and the answer sorts cleanly into two halves.

What transfers: everything about light and wear #

Lustre. A float shines because it is straight: a run of parallel, uninterrupted yarn reflects specularly along its own direction, where an interlaced surface scatters. That argument uses nothing about weaving. It uses a straight length of yarn and an eye, and a knitted float supplies both.

The consequence is real and visible: a float jersey’s back is glossier than its face, which is why the structure is used with the technical back outward in some sportswear and why single-jersey jacquards are made face-out for a matt appearance and back-out for a sheen.

Snagging. A float catches because it is unsupported over a length: something has to get under it, and the length of the float is the size of the thing that can. The woven argument is a geometric one about a hook and a gap and it transfers word for word. A knit float over four needles at a 1.8 mm needle pitch is a seven-millimetre unsupported run, and a woven float of four at twenty-five threads per centimetre is a 1.6 mm one — so knitted floats snag far worse than woven ones at the same float number, because the length that matters is millimetres and the machine pitches differ by an order of magnitude.

The two lengths in that sentence invite an obvious question and it has a clean answer. If a knitted float of four is seven millimetres of unsupported yarn, what woven float would expose the same length? At twenty-five threads per centimetre the arithmetic is seven millimetres times two and a half threads per millimetre, so a woven end would have to run over seventeen or eighteen picks to match it. No cloth anybody weaves does that. The float limit that governs a woven structure bites at a handful of crossings — the longest float in an ordinary furnishing satin is seven, and a shirting’s is three — so the woven family cannot reach the knitted family’s snag length at all, and it is not a matter of degree. The two structures occupy separate ranges of the one quantity the snagging argument depends on.

That is a stronger statement than the comparison it came from, and it is worth being clear about which part of it is measured. The geometric argument — something has to get under the float, and the float’s length is the size of the thing that can — is identical on both sides and is quantified on neither. What differs is the input, and the input is a machine constant: a needle bed is pitched in millimetres and a reed in fractions of one. A knit and a weave made of the same yarn to the same weight will differ by a factor of four or five in the length of their longest unsupported run before either structure has been chosen, which means the snagging comparison is settled by the gauge and not by the design.

Abrasion. Where the wear happens is at the high points, and a float is a high point. Same argument, same conclusion, and the same caution: on a knit the floats are on the back, so a float jersey wears at the loops on its face and at the floats on its back, and the two surfaces of the same fabric have different abrasion characters. That is not true of a woven cloth unless it is backed.

What does not transfer: everything about length #

Here the two objects behave in opposite directions, and the reason is one sentence.

A woven float lengthens the thread relative to the cloth. A knitted float shortens the fabric relative to the thread.

Take the woven case first. A thread that floats over four picks does not bend at three of them, so it takes a shorter path through the same length of cloth than a thread bending everywhere. Its crimp is lower, so per unit of cloth there is less thread in it, and the cloth is lighter and less extensible in that direction. The crimp is the whole distance between a woven thread’s length and its cloth’s, and the float is what reduces it.

The knitted case has no crimp in it at all. A missed needle contributes no loop, so the yarn between the two neighbouring loops is a straight segment where a loop would have been a bent one — and a loop is 4.3 wale spacings of yarn where the float is one. The float has not made the yarn straighter along its path; it has removed a loop from the fabric, and the wales on either side of it have nothing between them and close up.

So the woven float leaves the cloth’s width alone and lowers its weight per unit area by lowering crimp; the knitted float lowers the weight by removing yarn and narrows the fabric by removing a loop. A float jersey is narrower, longer and lighter than the plain jersey beside it. A satin is the same width as the plain weave beside it and lighter.

There is a third difference hiding inside that sentence, and it is about direction rather than length. A woven cloth has two systems and a float may belong to either: a warp-faced satin floats its ends over the picks, a weft-faced one floats its picks over the ends, and the same draft read the other way round turns one into the other. The choice is free, and a great deal of woven design is spent on making it.

A weft knit has one yarn path and it runs coursewise. A missed needle can only produce a float lying along a course, because that is the direction the yarn was travelling when it passed the needle. There is no wale-direction float in a weft knit and there cannot be one — a wale is not a continuous thread but a column of loops handed from one course to the next, so along it there is nothing to leave unsupported. The nearest object is a held loop, a loop that stays on the needle while several courses pass, and that is not a float at all: it is under tension rather than slack, it is drawn up rather than lying across, and it shortens its wale instead of removing one. The float’s asymmetry in a knit is therefore double — always on the back, and always one way — where a woven float is symmetric in both respects.

This is why the woven vocabulary of face and back effects has no clean translation. A woven designer choosing where a float shows is choosing which system carries it; a knitted designer has no such choice in single jersey and must go to a second bed to get one, which is the point at which the float stops being on a surface at all.

The extension reverses too, and it is the same reason #

A woven cloth extends in its own directions by crimp interchange: pull it warpwise and the warp straightens while the weft crimps more, and nothing in the yarn stretches. A float lowers the crimp available, so a satin extends less in the floated direction than a plain weave does. Longer float, less extension.

A knit extends by reconfiguring its loops, which is a much larger effect — a jersey stretches by tens of per cent where a woven cloth manages a few. A float removes loops, and a loop is where the extension comes from, so a float structure extends less across its width than plain. Longer float, less extension, again.

The conclusions agree and the mechanisms do not share a single step. That is worth noticing rather than glossing: the site’s habit is to ask whether an argument transfers, and here two entirely different arguments arrive at the same inequality, which is exactly the situation in which somebody concludes that one explanation covers both.

It does not. The woven statement is about the crimp reserve and it has a bound — a cloth cannot extend past its own crimp — computed from Peirce’s geometry. The knitted statement is about how many loops there are to reconfigure and it has no such bound in this site’s machinery. The two are the same shape and different quantities.

What was counted, and how #

The comparison is only as good as the two things being put side by side, so both were computed rather than described.

The knitted side. Yarn per needle position over each structure’s array, in wale spacings, with a loop taken as k_w and a float as one. The saving per missed needle is asserted to be exactly (k_w − 1)/k_w for every structure — not approximately, and not on average, because it is one subtraction repeated and any variation would mean something else had got into the count.

The woven side. Crimp from Peirce’s geometry at each weave’s own bending pitch, which is the site’s standard route, and the areal weight identity on top of it. A plain weave and an eight-end satin of the same yarns at the same setts differ by about eleven per cent in weight, all of it crimp.

Those two numbers — nineteen per cent from removing a quarter of the loops, eleven per cent from removing seven eighths of the interlacings — are the clearest statement of how differently the two structures spend yarn. A knit’s yarn is in its loops and a weave’s is in its bends, and the loop is much the more expensive of the two: a woven thread’s crimp adds five to fifteen per cent to its length, and a knitted loop is four hundred and thirty per cent of a wale spacing.

A table, because the two words have to be kept apart #

Read down the last two columns and the split is exactly where the essay says it is. Everything about the surface is shared and everything about the length is not — and the reason is in row four, which says the two are measured against different machines, and in row six, which says they remove different objects.

Row six is the one that does the real work and it is easy to read past. A woven float removes interlacings — crossings that were going to happen and now do not — and an interlacing is a small object: it costs a little crimp and a little firmness, and n of them removed leaves the two thread systems otherwise where they were. A knitted float removes a loop, and a loop is the fabric’s unit of area as well as its unit of yarn. Take one out and there is a hole in the plan that the neighbouring loops have to close, which is why the width moves and the woven case’s width does not.

Which is why the two quantities in row seven are not comparable even though both are percentages. A few per cent of weight is a correction to a cloth that is otherwise the same cloth; seventy-seven per cent of a loop is most of a structural unit, and four of them missed in sixteen is a quarter of the fabric’s loops gone. The first is a property that can be averaged over a fabric; the second changes what the fabric is.

The float length itself is not the same quantity #

One more difference and it is the one most likely to cause a mistake in a specification.

A woven float of n is measured in crossings: the thread passes over n threads of the other system. Its length in millimetres is n divided by the other system’s sett, plus a little for the bends at each end.

A knitted float of n is measured in needles: the yarn passes n needles without taking any. Its length is n times the needle pitch, which is the machine’s gauge, and the fabric’s own wale spacing after relaxation is smaller than the needle pitch — because a knit contracts off the machine.

So a knitted float is laid at one length and lives at another, and the difference is slack. A relaxed float jersey’s floats are not taut; they are loose enough that the fabric can be pulled out to nearly the needle pitch before the float takes any load. That slack is why a float structure is not as inextensible as its arithmetic suggests, and it is a quantity that depends on the relaxation state rather than on the array.

The slack also creates a measurement problem with no woven equivalent. Ask how long a float is in a finished knitted fabric and there are two defensible answers: the yarn laid down, which is the needle pitch times the number of needles passed, and the distance it now spans, which is the relaxed wale spacing times the same number. The first is a property of the machine and the second of the fabric, and they differ by however much the fabric drew in. A specification quoting a float length without saying which it means has left a fifth of its own number undetermined — and because the two converge as the fabric is pulled out, the same piece returns different answers depending on how hard the person measuring it is pulling.

The woven case has no such ambiguity, and the reason is that one component fixes both ends of the comparison. A float laid at the reed’s spacing lives at very nearly that spacing, because the cloth’s sett is what the reed and the take-up between them decided; the only correction is the crimp taken up by the bends at each end of the float, which is a few per cent and is computed rather than guessed. A woven float length is a measurement on a cloth that agrees with a calculation from the loom, and a knitted one is two numbers that have to be told apart.

The woven float has no equivalent slack. It is laid at the reed’s spacing and the cloth’s spacing is set by the same reed, so a woven float is very nearly as long as it looks.

Where the model stops #

Nothing here computes the slack. How much shorter a relaxed fabric’s wale spacing is than the machine’s needle pitch is Munden’s arithmetic and it depends on the loop length; the float’s slack is the difference, and the essay names it and does not use it.

The optics are qualitative. Both the lustre argument and the snagging argument are geometric statements about a straight length of yarn, and neither is quantified on either side of the comparison. The site has never had a reflectance model and does not have one now.

The comparison holds one thing fixed that a mill would not. A woven satin and a plain weave are compared at the same setts and the same yarn, and a knitted float structure and a plain jersey at the same loop length and the same machine — which is the right comparison for isolating the structure and is not what anybody buys. Real fabrics are adjusted to a target weight, so a float jersey is knitted at a longer loop than the plain jersey it replaces, and every number here moves.

And the two-bed structures are absent. Everything above is single jersey, where a float is unambiguously on the back. A rib fabric’s float can lie between the two beds, inside the fabric, where it is neither on a surface nor visible — which breaks the lustre and abrasion arguments entirely and is the commonest place a real float actually sits.

Who found it, and when #

Nobody found this, and that is the point of the essay. The word “float” is used in both trades, for the same object, by people who mostly do not read each other’s literature — woven-structure books and knitted-structure books are written by different authors for different courses, and the overlap between them is close to nil.

The consequence is a body of shared rules of thumb that are right for different reasons. “Long floats snag” is true in both and provable in both from the same geometry. “Long floats make a lighter fabric” is true in both and provable in neither from the other’s argument. “Long floats reduce extension” is true in both from mechanisms with no step in common.

The one that does not survive is the one everybody assumes: that a float is a way of not interlacing. In a weave that is exactly what it is — the float is defined by the interlacing it omits, and the site’s firmness number counts those omissions directly. In a knit there is no interlacing to omit. A float is a way of not forming a loop, and a loop is a different object with a different cost, so the analogy that holds up under the light fails under the hand.

Where the ladder goes next #

Six rungs of this anchor have now taken the float through lustre, snagging, abrasion, tear, design limits and a change of fabric. What none of them has is a force: the site can say a float is longer, more exposed and less supported, and cannot say what load it carries or when it fails. That is recorded as not done, and it needs the same statistics of first failure that the tear arithmetic is still waiting for.