A float limit forbids almost every chart and costs almost nothing to break
Worth reading first: A tuck is the one stitch that links twice · What a tuck costs · Knit, tuck and miss.
Knit, tuck and miss wrote a weft knit as an array over three symbols, and the essays after it priced each one. A miss saves three quarters of a loop, because a knitted loop is about four wale spacings of yarn and a float past one needle is exactly one; a tuck is the one stitch that links twice, holding two loops in one head. All of that was written for one yarn.
Stranded colourwork is two. A Fair Isle yoke, a Norwegian mitten and a machine-knitted jacquard back all knit each row with two colours, and at every stitch one of them makes the loop while the other is carried behind it to wherever it is next wanted. In the notation that is exact and easy: a stranded row is two courses fed together, one colour knitting where the chart says it and missing everywhere else, the other the complement. So every stitch of the fabric is one K and one M, and a colour’s float is nothing more than a run of the other colour along the row.
That makes the stranded knitter’s oldest rule — carry no float further than five stitches, or catch it — a rule about the lengths of runs. Runs in a row can be counted exactly, and so can what it costs to break the rule.
A float is a run of the other colour
The floats lie on the side of the fabric that shows the loops’ heads, which in a jersey is its back, so a stranded fabric’s back is a second surface of nearly parallel yarns lying across the wales. Read the diamond chart one row at a time. Where a row is blue the blue yarn knits and the light yarn floats behind; where it is light the reverse. A run of eight blue stitches is a light float eight stitches long, and a run of ten light stitches is a blue float of ten. Nothing else in the chart makes a float: not the shape of the diamonds, not how one row sits on the next, only how long each row stays one colour.
That is also why the rule binds rows and never columns. A weft knit has no float along its wales, because the yarn travels along the course, and stranded knitting inherits that: a stripe of one colour twenty rows tall and one stitch wide carries no float at all, and a band one row tall and six stitches wide breaks the rule. The diamonds break it in their middle rows, where they are widest, and nowhere else.
A woven designer’s float limit is a different object, though the float decides as much in a woven cloth as here. In a woven cloth a float runs along a warp end or a weft pick, so a limit binds every thread in both directions at once, and the count of allowed drafts can only be taken by search. A stranded row’s limit binds one direction, and that is what makes everything below exact rather than counted.
Two rows in three pass a five-stitch limit
A row knitted in the round is a ring of n stitches, each one colour or the other, and the question is how many rings have no run longer than L. That is a closed walk on a small machine with one state for each colour and each run length so far, moving to the next run length or switching colour and starting again at one. Every row with both colours in it is exactly one closed walk of n steps, so the count is the trace of the machine’s matrix raised to the n-th power, which needs no enumeration and has no rounding in it. It agrees with a count by hand of every row at twelve and at sixteen stitches.
At a 24-stitch repeat a five-stitch limit passes 66 per cent of rows. A limit of seven passes 91; a limit of three, which some knitters keep for fine yarns, passes 13. Each curve falls by the same factor at every stitch added, so a long repeat costs more than a short one, but the five-stitch rule is gentle on any single row of a length anybody knits: half of all 40-stitch rows still pass.
The factor has a closed form. A row whose runs are at most L long grows, stitch by stitch, by the largest root of
which is 1.966 at a limit of five, against the 2 of a row with no limit. At a limit of two the root is the golden ratio, 1.618, because a row of runs of one and two is counted by the Fibonacci numbers; that the oldest number in pattern-making should turn up as the freedom of a two-stitch float is a fact about runs, not about knitting, but it is the same fact.
Read as information, a five-stitch limit leaves each stitch 97.5 per cent of a free choice. The rule hardly constrains a row at all. That is the first half of the result, and it is the half a knitter would expect.
Almost no chart passes
The second half is what happens to a chart. Because floats run only along rows, whether one row passes has nothing to do with whether the next does, and the share of whole charts that pass is the share of rows raised to the number of rows.
At 24 stitches and twenty rows, a five-stitch limit passes one chart in about 3,800. A four-stitch limit passes one in 49 million and a three-stitch limit one in 3 × 10¹⁷. The same rule that leaves each stitch 97.5 per cent of its freedom excludes all but a sliver of the charts a 24-by-20 grid can hold, and it does so by nothing more than multiplying a probability of two thirds by itself twenty times.
That is not a paradox, and it is worth saying what it is instead. Most of the charts the grid holds are noise, and a designer was never going to draw them. A float limit is a statement about the catalogue, not about any design in it, which is the woven essay’s conclusion arrived at by a different road. Where the two roads differ is the shape: a woven limit couples every thread to every other, and a stranded one factorises row by row, so the knitter’s catalogue shrinks by an exact power and the weaver’s by whatever a search can find.
It also says how designers live with the rule. A chart drawn inside the limit from the start is not a random chart that happened to pass; it is drawn from a vocabulary of short runs — peeries, lice, small crosses, the narrow bands between the wide ones — and the wide motifs that would break the rule are kept to the rows where they can be caught. The diamond chart above breaks it in seven of its nine rows, which is about what a real chart of that size would do if it were not designed around the rule.
Breaking it costs one catch in sixty-two stitches
A float longer than the limit is not forbidden; it is caught. A hand knitter twists the carried yarn round the working one between two stitches, which pins it to the back of the fabric and divides the float in two without using a needle. A machine does the same thing with a tuck, holding the carried yarn on a needle for one stitch. So the cost of ignoring the rule entirely, drawing whatever chart one likes, is the number of catches it needs.
A float of r stitches needs ⌈r/L⌉ − 1 twists to bring every piece within L. A tuck uses a needle, which takes one stitch out of the float, so it needs ⌈(r − L)/(L + 1)⌉, sometimes one fewer: a float of eleven needs two twists or one tuck at a limit of five.
Take the worst case for the rule: a row in which every stitch’s colour is a fair coin, nothing designed at all. In a long row a run of exactly r stitches starts at any given stitch with probability one in 2^(r+1), so the expected catches are a sum over r, and the sum has a closed form. By twist, a row needs one catch every 2^(L+1) − 2 stitches; by tuck, one every 2^(L+1) − 1. At a limit of five that is one twist in 62 stitches. At three it is one in 14.
So a knitter who drew a chart with no regard for floats at all, and caught every float that broke a five-stitch limit, would make one catch per 62 stitches. The rule that excludes all but one chart in 3,800 is, from the other side, a rule that costs 1.6 per cent of the stitches to ignore. A knitter catches more than that for comfort, on long floats inside the limit, and every catch is visible from the front as a small speck of the carried colour — which is the real reason they are rationed, and the one thing here that is a judgement rather than a count.
What a stranded fabric spends is the same for every chart
The yarn accounting is even simpler, and it closes the loop with what a tuck costs. A stitch of stranded fabric is one loop of one colour, wale spacings in Munden’s constant, and one float of the other colour across one needle, which is exactly one wale spacing. A plain jersey’s stitch is the loop alone. So a stranded fabric spends
over a plain jersey of the same loop length, and the chart does not appear in it. Every arrangement of two colours on every grid costs the same.
That is 25 per cent in Munden’s dry-relaxed state, 24.4 wet-relaxed and 23.3 fully relaxed. A twisted catch adds no needle and is taken as adding nothing. A tucked one replaces a float of one wale spacing with a tuck of about a loop, so it adds some three wale spacings each time, and on an unplanned row at a limit of five that raises the excess by just over one percentage point; at a limit of three, by five.
So a stranded fabric is a quarter heavier than the jersey it is knitted on, whether its chart is a checkerboard or a single stripe, and almost all of that quarter is carried behind the fabric rather than in it. A plain knit’s weight nearly names its yarn, because a jersey’s weight is loops and nothing else; a stranded fabric’s is loops and one wale spacing more a stitch, so the same reading needs the quarter taken off first. The floats are why stranded knitting is warm: a second layer of yarn lying across the back, held a little off the loops, which is where a knit keeps its warmth, in the air its yarn encloses.
The floats also fix the width
A float’s yarn is laid at the width the fabric has on the needles. Once knitted it cannot get longer, so it sets how far the fabric can be stretched across the row, and a knitted float shortens its fabric rather than lengthening its thread. In a single-colour float structure that was a narrowing confined to the floated stitches. In stranded knitting every stitch has a float behind it, so the whole fabric is held to its knitted width by a second yarn that does not stretch with the loops.
That is why stranded fabric is famously less elastic across the row than plain knitting, and why knitters spread the stitches on the needle before carrying a float: the float’s length is set when it is laid, and laying it slack is the only way to give the fabric room. It is also why the length of a float matters less to the fabric than it seems. A float of five and a float of one hold the fabric to the same width per stitch; what a long float adds is a loop of yarn that can be caught by a finger or a ring, which is the hazard the rule exists for, and which no count in the notation can price.
How the rows were counted
The fabric is written in this notation as two courses a row: colour A knits where the chart has A and misses elsewhere; colour B the complement. A row is a ring of n stitches knitted in the round, so a run that wraps from the last stitch to the first is one run. Rows of one colour are left out, since in them the other colour floats round the whole repeat.
The count is the trace of the n-th power of a matrix on 2L states, one for each colour and each run length up to L, taken in exact integer arithmetic. The growth is the largest root of x^L = 1 + x + … + x^(L−1), found by bisection. The catches for a random row are the sum over run lengths of 2^−(r+1) times the catches a run of r needs, which closes to 1/(2^(L+1) − 2) by twist. The yarn uses Munden’s for the three relaxed states, the same constants the yarn accounting of the single-colour structures uses, with a tuck at one loop.
Required of it, not shown: that the matrix count agree with a count by hand of every row at twelve and sixteen stitches at limits of two, three and five; that a limit of two grow by the golden ratio; that a stranded fabric’s excess over plain be one over at every relaxation state; that a twist and a tuck agree on a float as long as the limit and differ by one on a float of eleven; and that the count refuse a limit as long as the repeat.
What the counting leaves out
Rows are counted, not designs. The shares are over every chart a grid holds, and a designer’s charts are a vanishing corner of that set. The useful readings are the two that do not depend on what is drawn: what a limit does to the freedom of a stitch, and what breaking it costs on a row with no plan, which is the worst case and so an upper bound on any designed row’s catches.
Tension is absent, and it is where stranded knitting actually goes wrong: a float pulled tight puckers the fabric, one left loose snags and lets the stitches either side of it spread. Nothing here measures how much slack a float carries, and the width argument above is a direction, not a number.
Yarn of both colours is taken as the same count, the loops as a plain jersey’s, and the dimensions as unchanged by the second yarn. A stranded fabric relaxes to a different shape from plain, and its weight per square metre moves with it by an amount not computed here. A catch is costed only as yarn and as a count; how visible it is through the front depends on the yarns and the gauge.
Who knew which part
Stranded colourwork and its float rule are old craft knowledge, written down in every Fair Isle and Scandinavian pattern book, and the twisting and tucking of floats are standard techniques by hand and on the machine. Counting binary strings with bounded runs is standard combinatorics: the transfer matrix and the generalised Fibonacci recurrence behind it go back at least to the work on run-length-limited codes for magnetic recording in the 1960s and 1970s, which faced the same problem with different symbols. Munden’s constants are from 1959.
What is added here is the reading of a stranded row as two courses of knit and miss, so that a float limit is a run limit; the exact share of rows each limit passes and the closed form for its growth; the chart share as a power, which the woven case does not have; the catches an unplanned row needs, one in 2^(L+1) − 2 stitches; and the yarn excess of one over that no chart can change.
Still open: whether a catch shows where a tuck would not
Every catch has been costed by its count and, for a tuck, by its yarn. What decides how many a knitter will accept is how much of the carried colour shows through the front, and the two kinds of catch differ there in a way the notation can describe. A twist lays the carried yarn across the back of one loop’s legs, between two stitches. A tuck takes it into a loop’s head, and a tuck holds two loops in one head, so its yarn sits in the front of the fabric for the width of the head.
So a tucked catch should show more and a twist less, and the difference is a length of carried yarn exposed on the face, which a loop model with a second yarn in it could compute. If the twist shows little enough, the five-stitch rule is a habit inherited from machines and a hand knitter could relax it; if the twist shows too, the rule is a fair price for a clean face.
Shares its objects with
Essays naming at least two of the same things, that neither author linked.
- A knit's dimensions come from its loop — both name knit, loop length
- A second bed changes what a float is — both name miss, stitch notation
- A tuck decides whether a third of two-bed fabrics lean — both name census, tuck
- A tuck held two courses is a second unknown — both name census, tuck
- Does a double jersey hang together — both name census, stitch notation
- Where a two-bed fabric's yarn is — both name miss, tuck
Named objects
A flat tag is an object no other essay names yet.