A third-drop wants its lengths repeated
Worth reading first: A bay window wastes less than a wall · A half-drop wastes less, cut in the right order · A repeat has to fit the panel, and the panel is cut.
A bay window wastes less than a wall turned a cutting room’s question into a map. A half-drop pattern has two classes of width, half a repeat apart; the cloth lost after each drop depends only on that drop’s length and on whether the next drop is of its own class or the other; so each drop has a price and a preference, and an order that gives every drop its preference is a walk across two islands that uses every bridge once. Mixing the lengths made such walks easier to find, and a bay window, with its long drops and short ones, came out cheaper to cut than a plain wall.
That essay named the next case and stopped. Before it, a repeat had to fit the width and then the panel, and every one of those questions was about a single class of drop. A third-drop has three classes, a third of a repeat apart, and its cutting order is a walk on three islands. The same reasoning should apply “with a more interesting walk.” It does, and the more interesting walk reverses the conclusion about mixing.
A third-drop’s three places to start
On a third-drop design the motif steps down a third of a repeat from one width to the next — the seam’s offset is the design’s lattice read across one width, and here that lattice has three rows to a repeat — so the widths of a wall fall into three classes — call them A, B and C — whose pattern starts a third of a repeat apart along the roll. On a 60 centimetre repeat a drop of class A may start at any whole repeat, a drop of class B twenty centimetres past one, a drop of class C forty past.
The cloth lost after a drop is the distance from where it ended to the next place the following drop may start. For a drop of length followed by a drop classes on, that is
and as before, nothing else enters: not the next drop’s length, not anything earlier on the roll. Each drop pays one of three amounts according to one fact about its successor — which class it is, counted from the drop’s own.
Three prices, and the cheapest
The three gaps are one saw-tooth shifted twice by a third of a repeat, so the cheapest of them is never more than a third of a repeat and averages a sixth:
The wall in the first figure was chosen so that its two lengths disagree. A 250 centimetre drop is ten centimetres short of the next place a class one step on may start, so it pays ten to be followed by the next class and fifty to be followed by its own. A 152 centimetre drop is eight short of the place two classes on may start, so it pays eight to be followed by the class one step back. One length wants the wall cut forwards and the other wants it cut backwards.
On a half-drop those two wishes would be the same wish. With two classes, one step on and one step back both mean “the other class”, and the half-drop essays never had to distinguish them. With three they are different moves, and that is where everything new comes from.
The two directions along a wall cost different amounts
The plainest consequence is one no half-drop cutter has met. A wall can be cut in the order its widths hang, and it can be cut from either end. On a half-drop the two directions alternate the classes in the same way and cost the same. On a third-drop they do not. From the left, the classes run A, B, C, A, B, C, and every drop is followed by the class one on; from the right, they run C, B, A, and every drop is followed by the class one back.
On the wall above, cutting from the left pays ten for each long drop and forty-eight for each short one, 126 centimetres in all; cutting from the right pays eight for each short drop and thirty for each long one, 84 centimetres. The difference, 42 centimetres, is two thirds of a repeat, and it is bought by nothing more than starting at the other end of the wall.
A design’s repeat is a unit of a plane pattern, and a third-drop’s lattice has a handedness a half-drop’s lacks. It has a near relative in the two directions of a twill, which are mirror images with identical counts that the yarn’s own twist tells apart. A third-drop’s walls are mirror images in the same way, and the drop length is what tells them apart — one more case of a plane pattern’s symmetry that the use of the cloth breaks.
An order is a walk on three islands, with one-way bridges
Draw the three classes as three islands. A drop that prefers its own class next is a loop at its island. A drop that prefers the next class is a bridge from its island to the next one round, and a drop that prefers the class one back is a bridge the other way. The bridges are now one-way: a drop of class A that prefers the class one on carries the cutter from A to B, and is no use for getting from B to A.
An order that gives every drop but the last its preference is a route along every arrow in turn, each taken from the island its drop belongs to. Euler’s condition for such a route, in the one-way form, is that the arrows all lie in one connected piece and that every island is left exactly as often as it is entered, except that the route may start one short at one island and finish one over at another. The last drop is excused, as before, since the cloth after the last cut is the roll’s, so its own arrow need not be walked at all — the route need only end at that drop’s island.
The floor is therefore the same as a half-drop’s: the sum of every drop’s price, less the largest price, since the most expensive drop can be cut last. On the wall that is three tens and three eights, less ten: 44 centimetres, and the route exists. The search through every order finds exactly that.
On two islands one count had to balance; on three, every island does
Here is the change that matters. On two islands, a walk needs the bridges from A to B and those from B to A to differ by at most one. That is one condition, and it is a loose one. On three islands with one-way bridges the condition applies to each island separately — arrows out of A against arrows into A, the same for B, the same for C — and the islands can be unbalanced in ways that cancel overall and still strand the cutter.
Look at what kind of job passes. A length hung once in every class balances itself, whatever it prefers. If it prefers its own class, it is three loops, one at each island. If it prefers the next class, it is three arrows chasing each other round the triangle, A to B to C to A, and every island has one out and one in. If it prefers the class one back, it is the same triangle the other way round. In none of the three cases does it leave an island over or under. A job made of such lengths is balanced by construction, and it can fail only by not being connected: if every one of its lengths prefers to stay, the three islands hold loops and nothing joins them.
A drop of a length that appears only once is one arrow, from one island to another, and nothing balances it but other single arrows that happen to point the right way.
The numbers, length by length
The averages follow the islands. Over drop lengths taken across a whole repeat, in the best order, six drops lose these shares of a repeat a gap.
One length, two drops in each class, loses 0.211 of a repeat a gap and reaches its floor on exactly two lengths in three — the two thirds whose preference steps round the classes. The other third prefers to stay, and with three separate islands the cutter must cross twice against the drops’ preferences, where on a half-drop one crossing sufficed.
Two lengths, one of each in every class, lose 0.170 and reach the floor on exactly eight jobs in nine. The ninth is the job in which both lengths prefer to stay, a ninth of the time when the two remainders are independent.
Every drop a different length loses 0.189 and reaches the floor on only 36 per cent of jobs. Six single arrows, each pointing wherever its length sends it, balance at every island rarely.
On a half-drop the same three cases run the other way: 0.300, 0.258 and 0.245, with the all-different job the cheapest. On two islands mixing helps monotonically; on three it helps until the lengths stop repeating, and then it hurts.
Why the half-drop was the forgiving case
The half-drop’s monotone result was not a law about mixing; it was a fact about two islands. With two classes every arrow from A goes to B and every arrow from B goes to A, so imbalance has only one way to happen, and the excused last drop can absorb a difference of one. Six single arrows on two islands balance often enough — 62 per cent of all-different jobs reach the floor — that the larger excused price of a job with more lengths wins.
With three classes an arrow from A can go to B or to C, and two islands can each be one over while the third is two under. The excused drop absorbs one unit of imbalance at one island, not a deficit spread over two. So a job of single arrows reaches its floor about a third of the time, and the cost of missing it — at least one drop paying against its preference, often more — outweighs what extra lengths give back.
Repeated lengths spread across the classes do best
Two further arrangements on the figure separate the two things that matter, repetition and spread. Three lengths, each hung twice in two different classes, lose the least of all, 0.164, and reach their floor on 78 per cent of jobs: each length is two arrows, which do not balance by themselves, but two arrows of one length point the same way round and are matched more often than six unrelated ones; the larger excused price that comes with more lengths accounts for the rest of the difference.
Two lengths, each kept to its own classes — four long drops in A and B, two short in C — lose 0.252 and reach their floor on only 22 per cent. Repetition is there; spread is not. Every long drop sends the cutter the same way round from the same two islands, and the short drops cannot undo it from the third. It is the third-drop’s version of the half-drop’s lopsided wall, and worse, because there are now two directions in which to be lopsided.
So the rule is not “mix the lengths” but “let each length visit every class.” A wall where the long and short widths fall evenly across A, B and C is cheap to cut; a stairwell, or a wall where all the short widths sit in one class, is not.
Every order of the wall
The search walks the orders without visiting them singly, but on six drops of distinct kinds every order can be listed and costed, and doing so shows how much the choice is worth.
The 720 orders run from 44 to 246 centimetres. Only nine of them reach 44. Cutting from the right, at 84, is beaten or tied by 84 orders, so it is a good habit on this wall; cutting from the left, at 126, is beaten or tied by 366, about half. A cutter who knows nothing of the arithmetic but tries both directions recovers most of what reordering is worth here. A cutter who always starts at the left loses nearly three times the floor.
A stair has no repeated lengths
A stairwell is the case that makes the warning concrete. Each width of a stair hangs a little lower than the one before, so every drop is a different length — longer than the last by the rise the stair makes across one width — and the classes run A, B, C, A, B, C along it.
Averaged over rises from one centimetre to fifty-nine, the stair’s best order loses 53 centimetres, against a floor of 40, and reaches the floor at 31 of the 59 rises. Cutting along the stair loses 129 centimetres from the bottom and 144 from the top. At a rise of twelve centimetres, the case drawn on the island map, the floor is 42 and the best order 62: two islands take in more drops than they send out, the third sends out two more than it takes in, and no order lets every drop have its way.
The stair is not exceptionally expensive in its best order; it is exceptionally hard to cut well by habit. No direction along it is reliably good, because its drops do not agree on a direction, and the best order is one that jumps about the wall.
The guarantee does not move
A cloth order is placed against the worst case rather than the average. As on a half-drop, mixing leaves the worst case where it was: the largest loss a gap in the best order is 0.45 of a repeat for six drops of one length and 0.45 for two lengths one per class, and 0.47 for all different. Two lengths can always happen to share a remainder, which turns a mixed job back into a job of one length, so a cutter cannot order less cloth for a well-mixed wall; the saving appears as cloth left on the roll.
The worst order is worse on a third-drop than on a half-drop, too: about 0.8 of a repeat a gap on six drops, whatever the mixture. More classes give more ways to be right and more ways to be wrong, and an order chosen without thought is more dangerous here than on either of the matches the earlier essays priced.
The rule, written for a cutter
First, price each drop. Divide its length by a third of the repeat and note what is short of the next multiple; that is its price. Then see which of the three gaps gives it — to a drop of its own class, of the class one along the wall, or of the class one back — and that is its preference.
Second, check the islands. Count, for each class, how many drops leave it for each other class and how many arrive from each. If every class sends out as many as it receives, give or take one at two of them, the floor can be reached: cut in runs that follow the drops’ preferences, starting from the class with one spare to send.
Third, if the islands do not balance, pay the cheapest corrections, and leave till last the drop with the largest price. Where every length hangs once in every class the check almost always passes, and the order can be read straight off the wall in whichever direction the drops prefer — which on one length is simply the cheaper of the two directions.
Where the walk comes from
The two-island version was Euler’s, from 1736, and the one-way version on any number of islands is the form in which it is usually stated now: a route through every arrow exists when each node is entered as often as it is left, give or take the route’s two ends, and the arrows hang together. In 1946 de Bruijn and Good, independently, built the shortest sequences containing every word of a given length out of just such one-way walks, which is the best-known use of the directed form and a long way from a curtain workroom.
The sequencing view — each job a set of states required and states left, with a cost that depends only on the pair — is Gilmore and Gomory’s from 1964, as the bay-window essay recorded. What a third-drop adds is a second step size, which turns undirected bridges into one-way ones, and that is all it takes to change which jobs are easy.
How the orders were counted
The search is a memo over the drops still to cut and the drop just cut, so it reaches the exact best and worst without listing orders. On two classes it was required to agree, job for job, with the half-drop search the earlier essays used, which shares none of its code: it did, on 192 jobs. On three classes it was required to reach the floor on exactly the jobs the one-way trail condition accepts and on no other, and never to beat the floor: it did both, on 2,000 jobs across five arrangements of two and three lengths. The averages take each length’s remainder at the middle of each of thirty equal steps across the repeat — every combination for one or two lengths, a seeded stride through them for more — so that a remainder never sits exactly where two gaps tie. The wall’s 720 orders were listed and costed one by one.
What the arithmetic leaves out
The classes are fixed by the hanging. Which class a drop falls in is decided by where its width hangs, and nothing here moves a width; it only orders the cutting. And the classes are only as good as the cloth is square: a skewed cloth steps at every seam, which moves a drop’s true start whatever the arithmetic says. A designer choosing where the windows fall on a wall is choosing the arrangement, and the figure of arrangements is the only part of this that a design decision can reach.
A roll is assumed long enough. Every job is cut from one roll, as in the earlier essays. A job split across rolls has two excused drops and two roll starts, and how that trades against a remnant’s value is a separate question.
And the cloth is assumed to be cut where the arithmetic says. A real cutter allows for trimming and for the cloth drifting along a drop, and a third-drop’s narrower spacing between classes leaves less room for that than a half-drop’s. On a 60 centimetre repeat a third-drop’s classes are twenty centimetres apart, against a half-drop’s thirty, which is a tighter target.
Still open: four islands, and what a quarter-drop’s square does
A quarter-drop has four classes and three step sizes, one of which — two classes on — is its own reverse. On four islands, then, some bridges are one-way and some are not, and a length hung once in every class balances itself in each of three ways: four loops, a cycle round the square in either direction, or two pairs of two-way bridges across it. Whether that mixed case behaves like the half-drop, like the third-drop or like neither is a count of the same kind, and it would also say whether the third-drop’s reversal — mixing that stops helping — grows worse as the classes multiply or is peculiar to an odd number of them.
Shares its objects with
Essays naming at least two of the same things, that neither author linked.
- A larger skirt can cost nothing for its nap — both name cutting, specification, waste
- A nap is paid for by the taper of the pattern — both name cutting, specification, waste
- A bundle is weaker than its threads — both name population, specification
- A chenille is a yarn that is already a fabric — both name population, specification
- A cloth cannot be more even than its yarn — both name population, specification
- A designed thin place is kinder than an accidental one — both name population, specification
Named objects
A flat tag is an object no other essay names yet.