Knits and other structures

An atlas leans by its walk, not its runs

An underlap under tension pulls its course sideways by the sine of its angle, so a wale's sideways place is a running sum of those pulls, the same arithmetic as a beam's stroke with direction in place of length. An atlas of k laps out and k back leans k times a tricot. The trade's proxy, no three laps running one way, is exact at eight courses and passes 48 plans that lean three at twelve. And because the pull is a sine and not a shog, a lapping that closes on its needle can leave a pull that never comes back: 74% of all closed eight-course plans skew their cloth.

Worth reading first: A third bar has no say in a warp knit's first give · A patterned warp knit needs a stroke as long as its stripe · The shog is the anisotropy.

A third bar has no say in a warp knit’s first give found the cheapest way to pattern a warp knit. A bar that keeps one shog and changes only its direction costs a beam nothing and leaves the fabric’s first give exactly where its ground put it. Direction is a pattern the beam cannot see. That account then named the one cost it had left unpriced. A run of underlaps lapped the same way pulls the wales sideways, and the circles that describe a course’s first give treat it as a thing with a spacing and no lean. Its census gave plans that never lap three times running one way a row of their own, as a proxy for the lean, and said so.

This essay prices the lean. The pricing turns out to have exactly the arithmetic of the beam’s stroke, with direction where length was, and it says two things the proxy could not. The proxy is exact at eight courses a repeat and wrong at twelve. And a lapping whose needle places come back where they started can still leave its cloth a sideways pull that never comes back at all.

An underlap pulls sideways by its sine

An underlap is a straight run of thread from a needle on one course to a needle some spaces along on the next. The shog is the anisotropy built everything on that straightness: the run is a hypotenuse, of length u(s)=(sw)2+c2u(s) = \sqrt{(sw)^2 + c^2} for a shog of ss spaces, a wale spacing ww and a course spacing cc. At 28 needles an inch and 14 courses a centimetre, ww is 0.907 millimetres and cc is 0.714, and a tricot’s underlap is 1.155 millimetres long.

A straight run under tension pulls along itself. Its pull therefore has a part along the wale and a part across it, and the part across it is the thread’s tension times the sine of the underlap’s angle from the wale, sw/usw/u. For a tricot that sine is 0.786, an angle of about 52 degrees. The pull acts between the two courses the underlap joins: it draws the loops of the upper course towards the side the underlap came from, relative to the course below.

What this account assumes is a single number and nothing about its size. If each course’s loops give way sideways in proportion to the pull on them, with one stiffness that is the same on every course, then course kk sits displaced from course k−1k - 1 by an amount proportional to σk skw/u(sk)\sigma_k\, s_k w / u(s_k), where σk\sigma_k is the underlap’s direction. Measure that displacement in units of one tricot course’s displacement and the stiffness cancels. Every number below is in those units, and every one of them is exact under that assumption. How many millimetres one unit is depends on how stiffly a loop resists being pushed sideways, which nothing here computes.

The wale’s sideways place after kk courses is then the running sum of the pulls, and its lean is the range of that sum over a repeat: how far the wale wanders between its two extremes.

The atlas lapping. A warp-knit lapping drawn from the guide bar's movement. Each thread is coloured by the group of wales it belongs to, so a lapping that leaves the wales independent shows as several colours. This one joins them into 1 group.
Fig. 1 An atlas in the lapping notation, three spaces out and three back. Every underlap is one space long and every one pulls its course sideways equally hard; what changes is only which way, three courses running.

That is the same shape as the stroke. A patterned warp knit needs a stroke as long as its stripe found that a bar changing its shog from course to course takes yarn at a changing rate from a beam that gives at one, and the running sum of each course’s underlap less the mean is a slack whose range the warp must absorb. The lean is that sum taken over direction instead of length. A plan of constant shog has a stroke of nothing, because every course’s length is the mean, and its lean is whatever its directions add up to.

A tricot cancels at once and an atlas lets it run

The two simplest plans of constant shog are the two ends of the question. A tricot laps one space right and one space left, course after course, so its pulls alternate, +1,−1,+1,−1+1, -1, +1, -1, and the running sum never leaves the strip between nought and one. A tricot leans one unit, by definition, and the lean is visible only inside a single course, as a loop tipped one way and the next tipped back.

An atlas walks. One of four out and four back laps right on four courses running and then left on four, so its running sum climbs to four and comes down again. An atlas of kk laps out and kk back leans kk units, which is kk times a tricot’s, with a wavelength of 2k2k courses.

How far a wale wanders, course by course, under three lappings. A wale's sideways place against the course, over 24 courses, if each course's loops give way sideways in proportion to the sideways pull of the underlap above them, measured in units of one tricot course's lean: tricot leans 1.00; atlas, three out and back leans 3.00; atlas, four out and back leans 4.00. Every plan has one shog throughout, so none of them costs the beam a stroke or moves the first give. What the figure cannot show is the size of one unit, which depends on how stiffly a loop resists being pushed sideways and is not computed here.
Fig. 2 A wale’s sideways place course by course under three plans of one shog, in units of one tricot course’s lean. A tricot cancels every course, an atlas of three out and back leans three units, an atlas of four leans four, and none of the three costs the beam a stroke.

That is the zigzag the trade describes in an atlas fabric, and the running sum says what its size is made of. It is not a property of the shog: every course in all three plans has the same pull. Nor is it a property of how far the thread travels across the wales, since an atlas of four takes its thread four spaces from home and its lean is four, but a plan that took its thread four spaces in eight courses by a different route could lean less, as the next section shows. It is the range of the walk the directions make, which in an atlas happens to equal how far the thread goes. Every course leans by the same amount in every one of these plans; what differs is how many of them lean the same way before one leans back.

The first-give circles are untouched by any of this. A course’s spacing at a stated width depends on its shog through s2s^2 alone, and a direction is invisible to a square. So the account here sits alongside the earlier one rather than replacing it. A course has a spacing, which the circles decide, and a lean, which the running sum decides, and the two are separate questions about the same course.

A longer shog buys yarn, not lean

Before the plans themselves, one fact about the shogs changes how the arithmetic reads. The pull is a sine, and a sine saturates. A tricot’s underlap runs at 52 degrees from the wale, a cord’s at about 68 and a satin’s at about 75, and their sines are 0.786, 0.930 and 0.967. In tricot units, a cord pulls 1.18 times as hard sideways and a satin 1.23, and no shog at this gauge can pull more than 1.27, which is one over a tricot’s own sine and is reached only by an underlap lying flat.

What a longer shog adds to a course's yarn, and to its lean. For shogs of 1 to 6 needle spaces at 28 gauge, the underlap's length and the sideways pull it puts on a course, both against a tricot's: shog 1, 1.00 and 1.000; shog 2, 1.69 and 1.184; shog 3, 2.44 and 1.231; shog 4, 3.20 and 1.249; shog 5, 3.98 and 1.257; shog 6, 4.75 and 1.262. The length grows with the shog and the pull cannot pass one over the sine of a tricot's angle, 1.273, because a long underlap already lies almost flat. The figure assumes the pull's sideways share is what leans a course; how stiff a loop is sideways it does not say.
Fig. 3 Per course, a longer shog’s underlap against its sideways pull, both against a tricot’s at 28 gauge. A satin’s underlap is 2.44 times as long and pulls only 1.23 times as hard, and six spaces is 4.75 times as long and pulls 1.26 times.

The length of the underlap does not saturate. A satin’s is 2.44 times a tricot’s and six spaces is 4.75 times. So a long shog is a large bill at the beam and in a course’s weight, as the stroke and weight accounts found, and only a slightly larger bill in the lean. An atlas walked in satin laps leans 23 per cent more than one walked in tricot laps, not 144 per cent more. If an atlas zigzag is to be kept small, the length of its run matters several times as much as the size of its shog.

The sine has a second consequence, which is the larger of the two and which the rest of this essay comes back to. A sine is not proportional to its shog. A thread that laps one space right twice and then two spaces left has come back to its needle, since 1+1−2=01 + 1 - 2 = 0. Its pulls have not, since 1+1−1.184=0.8161 + 1 - 1.184 = 0.816.

Where a run limit is exact, and where it is not

The census of a middle bar’s plans counted 211 plans with no stroke at eight courses a repeat: one chain, and seventy direction patterns at each of the shogs one, two and three, each pattern a choice of which four of the eight courses lap left. It then gave a separate row to the 91 that never lap three times running one way, as a proxy for the lean.

Counting the seventy patterns of one shog by their longest run and by their lean, both exactly, settles what the proxy was measuring.

70 direction patterns of 8 courses, by longest run and by lean. Every way of choosing which 4 of 8 courses lap left with one shog throughout, 70 plans, counted by the longest run of laps one way round the repeat and by the lean in tricot units. 30 lean two units or less; 30 have no run of three, and 0 of those lean further than two. Every count sits on the diagonal, so at this repeat the run limit is the lean exactly. The lean is a running sum under an assumed sideways stiffness the same on every course; its size in millimetres is not computed.
Fig. 4 The seventy direction patterns of one shog in eight courses, by longest run of laps one way and by lean in tricot units. Every count sits on the diagonal: at this repeat a plan leans exactly as far as its longest run.

At eight courses the run limit is the lean exactly. Two plans lean one, twenty-eight lean two, thirty-two lean three and eight lean four, and in every one of them the lean equals the longest run. So the census’s row of 91 was precisely the plans whose walk never climbs more than two laps from its lowest point, and the proxy did its job without knowing why.

The reason is a counting accident of the short repeat. A run of rr laps one way moves the wale rr units, so no plan can lean less than its longest run. The lean can exceed it when two runs in the same direction are separated by a single lap back: two right, one left, two right is a run of two and a climb of three. In eight courses, with only four laps each way, there is no room to do that and still close. In twelve there is.

924 direction patterns of 12 courses, by longest run and by lean. Every way of choosing which 6 of 12 courses lap left with one shog throughout, 924 plans, counted by the longest run of laps one way round the repeat and by the lean in tricot units. 126 lean two units or less; 174 have no run of three, and 48 of those lean further than two. Off the diagonal, the run limit passes plans it should not. The lean is a running sum under an assumed sideways stiffness the same on every course; its size in millimetres is not computed.
Fig. 5 The 924 direction patterns of one shog in twelve courses. Forty-eight have no run longer than two and lean three, and forty-eight more have runs of three and lean four: the run limit passes plans that lean further than it says.

At twelve courses the run limit lets through 48 plans that lean three units, half again what it promises, and 48 more whose longest run is three lean four. At sixteen courses it lets through 528 plans that lean more than two, sixteen of which lean four with no run longer than two. A run limit is a floor under the lean and not a ceiling on it, and the longer the repeat, the more room a plan has to climb in short steps.

The exact rule has a count of its own, which the walk reproduces at every repeat it was run on. A closed sequence of laps that never leaves a band three levels wide must alternate between the middle level and one of the edges, so it is fixed by which edge it visits on each of its n/2n/2 visits. The plans that lean two units or less number 2n/2+1−22^{n/2 + 1} - 2: thirty at eight courses, 126 at twelve and 510 at sixteen, against 30, 174 and 1,038 that have no run of three. At eight the two counts are the same number. After that the proxy overcounts by a margin that grows with the repeat, and every plan in the margin is a lean the proxy called safe.

A second bar cancels only by mirroring

What a second guide bar is for was a question about connectivity: two bars together join wales that neither joins alone. The lean gives a second bar a second job. Two bars on one course pull on the same loops, and if their pulls add, the cloth’s lean is the running sum of both.

An atlas front bar over four different back bars. A front bar lapping an atlas of 4 out and 4 back, over 3 repeats, with the cloth's sideways place the running sum of both bars' pulls, in tricot units: back chain, a lean of 4.00; back tricot, a lean of 4.00; back atlas, same way, a lean of 8.00; back atlas, mirrored, a lean of 0.00. Only a back bar lapping the same plan in the opposite direction on every course takes the lean out; a tricot adds its own alternation and leaves the atlas's run as it was. The two bars are taken to pull on the same loops and to add, which is the assumption the figure cannot test.
Fig. 6 An atlas of four out and four back on the front bar, over three repeats, with four different back bars. A chain or a tricot underneath leaves the lean at four units, an atlas lapping the same way doubles it to eight, and only the mirror image, lapping left on every course the front laps right, takes it to nothing.

A chain underneath leaves the atlas’s lean at four units, and so does a tricot. A chain has no pull at all. A tricot alternates, so it adds a unit on one course and takes it away on the next, which jitters the wale inside the atlas’s band without moving its ends. An atlas on the back bar lapping the same way doubles the lean to eight, and the same atlas lapped in the opposite direction on every course cancels it course for course.

That is what a two-bar atlas does in practice, and the arithmetic says it is the only arrangement that does so exactly. Partial cancellation is common and the sine says how partial. Locknit, one of the commonest two-bar warp knits, laps a cord on its front bar against a tricot on its back, in opposite directions. Its net pull per course is 1.184−1=0.1841.184 - 1 = 0.184 of a unit, alternating, so a locknit leans 0.18 of a tricot: very nearly upright, and not quite, because two different shogs cannot pull equally. A full tricot with two opposed bars leans nothing.

Opposed bars even a warp knit by locking it found that opposition is what makes a patterned warp knit come out even in its give. Here opposition is also what makes it come out straight. The two findings concern different running quantities and come to the same design habit, which is part of why the habit is so old.

A closed walk can leave a pull that never returns

Back to the thread that laps one space right twice and two spaces left. Its walk closes, so the thread comes home, the fabric is stable on the machine and the beam sees a finite stroke. Its pull does not close. Every three courses the running sum gains 0.816 of a unit, and after the next three it gains the same again. The wale does not zigzag. It drifts, steadily, for as long as the plan runs.

How far a wale wanders, course by course, under three lappings. A wale's sideways place against the course, over 24 courses, if each course's loops give way sideways in proportion to the sideways pull of the underlap above them, measured in units of one tricot course's lean: tricot leans 1.00; two tricot right, a cord back leaves 0.816 of a unit unreturned every 3 courses and has wandered 7.71 by the 24th; two right, one left, twelve courses leans 3.00. A walk that returns to its needle need not return its pull, because the pull is the underlap's sine and not its shog. What the figure cannot show is the size of one unit, which depends on how stiffly a loop resists being pushed sideways and is not computed here.
Fig. 7 A tricot, a twelve-course plan of one shog that leans three units, and a plan that laps one space right twice and two spaces left. The first two return every repeat. The third gains 0.816 of a unit every three courses and has wandered 7.7 units by the twenty-fourth.

A steady drift of each course against the one below is a shear, and a sheared knitted fabric is a skewed one: its wales run straight and its courses cross them at an angle, or the other way round once it is relaxed and laid flat. This is a spirality with no twist in it. A jersey leans because its yarn still turns traced a weft knit’s skew to the torque in a singles yarn, and a circular machine leans its courses whatever the yarn added a helix that the machine makes by laying its courses on a cylinder. A warp knit has a third source, available to a perfectly balanced yarn on a flat machine: a lapping whose sines do not cancel.

The condition for cancelling is simple to state. A plan’s pull closes when, for every shog it uses, its laps to the right and to the left are equal in number. Closing the walk asks for less: only that the shogs to the right sum to the shogs to the left. The pull asks for the sum of each shog’s sines to balance, and only a plan that balances shog by shog can give that.

How common is the failure? Every closed plan of eight courses with shogs up to three either way can be enumerated, 398,567 of them, and each gives a stroke and either a lean or a skew.

Every closed plan of 8 courses, by its stroke and by its lean. All 398,567 closed lapping plans of 8 courses with shogs up to 3 either way at 28 gauge, binned by the stroke a beam must absorb and by the lean in tricot units. The two are different running sums of the same plan, of underlap length against its mean and of sideways pull, and they are free of each other. 295,288 plans, 74%, mix shogs so that their pull does not return at the end of the repeat and the cloth skews instead of leaning; 3 plans have neither a stroke nor a lean past a tricot's, the chain and the tricot started either way. The lean assumes one sideways stiffness on every course and has no size in millimetres here.
Fig. 8 Every closed plan of eight courses with shogs up to three, binned by its stroke at the beam and by its lean in tricot units. The 211 with no stroke split 3, 32, 88, 72 and 16 across the lean bins. 295,288, three quarters of all plans, mix their shogs so that their pull never returns, and skew.

295,288 of them, 74 per cent, skew. The stroke-free plans never do, because a constant shog balances itself as soon as the walk closes, and the 211 of them divide by lean into 3, 32, 88, 72 and 16. The plans that skew are also the ones that carry the larger strokes: nearly all of them sit at a millimetre or more. Only three plans in all have neither a stroke nor a lean beyond a tricot’s: the chain, and the tricot started on either foot.

That sharpens the earlier advice. Direction was the cheap way to pattern a bar because it costs no stroke and no give. It now has a price too, its lean, and the price is set by the shape of the direction walk and paid in a visible zigzag. Length patterning keeps its old bill and gains a new one: a mixed-shog plan must be balanced shog by shog, or its cloth will skew, and three quarters of the plans a designer could write down are not.

Why the trade’s lappings are the ones that close

There is a pattern in the lappings that have names. Chain, tricot, cord, satin and velvet each lap one shog, alternately right and left. Atlases lap one shog, out and back. The commonest two-bar fabrics are built in opposition: locknit, full tricot. Every one of those is a plan whose pull closes, and most of the plans whose pull does not close have no names at all.

The first-give account cannot tell a skewing plan from a straight one, since a direction is invisible to its squares. The stroke cannot either, since an atlas and a tricot both have none. The lean can. A mixed-shog plan that skews its cloth would have been knitted, found to come off the machine twisted, and abandoned without anyone needing a reason more precise than that. The arithmetic says which plans those were. It is not evidence of how the named lappings were chosen, which no record here describes, but it is consistent with it.

How it was computed

A course’s pull is the sine of its underlap’s angle from the wale, sw/u(s)sw/u(s), signed by direction and divided by a tricot’s at the same gauge, 28 needles an inch and 14 courses a centimetre. A wale’s place is the running sum of the pulls, and the lean is that sum’s range over one repeat; the skew is what the sum has gained at the end of the repeat. Two bars on one course add their pulls.

The direction census walks every choice of which half of nn courses lap left, (nn/2)\binom{n}{n/2} of them, at eight, twelve and sixteen, and counts each by its longest run round the repeat and its lean. The closed form 2n/2+1−22^{n/2+1} - 2 for leans of two or less is checked against the walk at each of those repeats. The plan census walks every sequence of eight signed shogs from −3 to +3 whose sum is nought, 398,567 of them, as the middle-bar census did, and records each one’s stroke and its lean or skew.

Required of it and not shown: that a tricot lean one unit and an atlas of kk out and back lean kk for kk from two to six; that a mirrored back bar take each of those atlases to nothing; that no plan lean less than its longest run; that the run limit be the lean at eight courses and not at twelve; and that a walk off the needles be refused. Each of those fails if the pull is made proportional to the shog rather than to its sine, and the skew vanishes entirely under that change.

What the running sum leaves out

The size of a unit. Everything above is a ratio to a tricot course’s lean, and the lean itself is a balance between the pull and a loop’s resistance to being pushed sideways, which is a question about the loop’s own shape and its contacts. Every crossing is a force, and a loop pressed by its neighbours resists differently from one held loosely. Without that number, the account says that an atlas of four leans four times a tricot, and not whether either is visible.

One stiffness for every course. A course whose underlap is long may hold its loops differently from one whose underlap is short, and in that case a mixed plan’s lean would be weighted by more than its sines. That could move the skew of any mixed plan; it could not move the cancellation of a constant-shog plan, which depends only on its rights and lefts being equal.

Tension the same in every underlap. The pull is the thread’s tension times a sine, and a thread’s tension varies along a course where the stroke is not zero, which is exactly where the skew lives. A plan with a large stroke may skew more or less than its sines say, and the census above takes the tension as one number.

Relaxation. A fabric off the machine settles, and a skew that is a steady shear on the machine may partly recover or partly grow as the loops find their relaxed shapes. The bias is a mechanism in a woven cloth, a shear with no resistance at first, and whether a warp knit’s courses behave similarly under their own small shear is not computed.

And it is the first give only, again. The lean is drawn as a displacement under the pull the fabric already holds, not under a load anyone applies. What a load added along the wale does to it is a further question.

Who worked out which part

The atlas lapping, its zigzag, the opposed two-bar fabrics and their upright loops are ordinary warp-knitting knowledge, set out in the trade’s texts along with the notation used here. That an underlap’s tension tips a loop towards it is part of the same body of practice.

What is added here is the reduction of the lean to a running sum of sines, which makes it the stroke’s arithmetic with direction in place of length; the exact counts of direction patterns by run and by lean, with the closed form for those that lean two or less, and the finding that the run limit is exact at eight courses and loses 48 plans at twelve; the pull of a second bar, with locknit’s lean as a worked case; and the skew of a closed plan whose shogs do not balance one by one, with its count among all closed plans of eight courses.

Still open: how many millimetres one unit is

Every number here is in units of one tricot course’s lean, and that unit is the one measurement the account needs. A single-bar tricot and a single-bar atlas of four, knitted in one yarn at one gauge and relaxed alike, would give it twice over. The tricot’s loops tip by one unit, the atlas’s wales zigzag by four, and a ratio far from four would say that a course’s lean is not proportional to its sine.

The sharper test is the skew. A single bar lapping one space right twice and two spaces left, and the same bar lapping a plain cord, would come off the machine differently if the pull is a sine: the cord straight, the mixed plan skewed by 0.816 of a unit every three courses. If the mixed plan comes off straight, the pull is proportional to the shog, the sine is wrong, and three quarters of the closed plans are safer than this account says.

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.

CensusLappingSkewSpiralityTricotWarp knitting