Hiding a moved end takes seven ends
Worth reading first: The window that names a cloth is half its repeat · A cloth's derivation class is its census of small patches · No cloth derives into more than three others.
The window that names a cloth is half its repeat walked every cloth of eleven small repeat shapes and found a rule. Slide a square window over a draft, count the patterns it sees, take the count up to the turnings, reversals and counterchange that cut a catalogue into derivation families, and the census names every family once the window is wider than half the repeat’s ends. A narrower window always confuses some.
The rule had one exception, and it was the more instructive result. At four picks by seven ends a window of four is wider than half, and one pair of cloths still defeated it. The pair is built from a stretch of two ends that a slide of two picks leaves unchanged, a copy of that stretch, and an end between them that sits two picks lower in one cloth than in the other. Every window that sees the moved end sees a view the other cloth also has, from the far side of the stretch.
Six by six is the first square repeat where the rule says a window of four is enough, and it is the first square that cannot be walked: 2³⁶ drafts, some hundred million families. So the question was left as a search rather than a census: can the mechanism that broke four at seven ends be built inside six by six? A random sample would not answer it. The four-by-seven pair is one draft in nearly half a million, and samples miss what is built by structure.
What the walk could reach and what it could not
A draft is a matrix, and the walk took every matrix of a shape that interlaces, so its answers are exact where it reached. Its reach was set by arithmetic. How many cloths there are grows as two to the power of the intersections, and the walk stopped at 28: five by five, four by seven, three by eight.
Read across the grid and the rule is plain, and so is its single red cell. Read down it and the gap is plain too: nothing above five picks, and nothing square above five by five. What a repeat repeats is a lattice of any size, and a weaver’s dobby repeat of six or eight is the ordinary case rather than the large one, so a rule about naming cloths that stops at five is a rule about the smallest cloths only. The six-by-six question is where the rule first has to be trusted rather than checked.
The pair has a shape, so the shape can be searched
The known pair is not an accident of seven ends. It has parts, and each part can be stated as a rule a draft either obeys or not.
A stretch is a run of consecutive ends, each of which reads the same after the draft is slid some whole number of picks along itself. On four picks the only interlacing ends with that property are the two plain ends, up-down-up-down and its counterchange, which a slide of two leaves alone. On six picks there are two such ends for a slide of two and six for a slide of three. A copy is the same run written again further along. The moved end stands between the stretch and its copy, and the second cloth is the first with that one end slid. Whatever ends are left over are free.
That is a family small enough to take whole. At six by six with a stretch two ends wide, a copy written in order or mirrored, the moved end any interlacing end, the one free end any interlacing end, and every slide of the moved end from one pick to five, it is 547,176 pairs in which both cloths interlace. Each pair’s two censuses are compared exactly as the walk compared them, with the same window codes and the same tables for the symmetries of the square. Only where the censuses agree is the slower question asked: whether some derivation carries one cloth to the other.
The search was first run where the answers are already known. At four by six the walk found that a window of four names every family, and the search, over 148,208 pairs built the same way, finds no confusion. At four by seven it finds 32 confused pairs, and all 32 are writings of the walk’s one exceptional pair. A search that missed that pair would have been useless here; one that invented others would have been wrong.
Six by six: forty-three thousand agreements and no confusion
At six by six, 43,800 of the 547,176 pairs share a census of four-by-four windows, and every one of them is a single cloth written twice. Some are the same cloth slid along the ends or the picks; some are its mirror; some need a counterchange or a quarter turn. None is a second family. The stretch builds plenty of pairs a window cannot tell apart, which is what a stretch is for, and at six ends every such pair turns out to be one cloth.
That is a stronger statement than a sample could make, and it is narrower than the question. It says the known mechanism cannot be built at six by six, in any of its variants: either period, copy in order or mirrored. It does not say that no other mechanism exists, and nothing short of a census of six by six could say that.
Every confusion writes the moved end three times
Why six ends defeat the stretch is visible in the confusions that do exist. Take the pairs the search finds at four by seven and at six by seven and look at the ends left over once the stretch, the moved end and the copy are placed.
In every one, the two free ends are the moved end itself and the moved end slid by the same amount. At four by seven, all 32 confused pairs have this form. At six by seven, all 336 do. The first cloth holds the moved end three times: between the stretches in one position, and beyond the copy in both. The second cloth holds the same three ends and differs only in which of its two positions stands between the stretches.
That is the mechanism the walk’s drawing described, now with its parts named. A window of four that holds the moved end can hold one whole copy of the stretch beside it and only half of the other. Looking left from the moved end it sees the stretch and then whatever stands before it; looking right, the copy and whatever stands after. In the first cloth the right-hand view runs into the moved end as it stands, and the second cloth’s left-hand view runs into the slid end — and because the stretch is unchanged by the slide, those are the same window, read from opposite sides. For the swap to balance, both versions of the moved end have to be there to be looked at.
Count the ends that needs. Two for the stretch, one for the moved end, two for the copy, and two for the moved end as it stands and slid: seven. A stretch one end wide would need only five, and at four by seven the search built 2,513,920 such pairs and found no confusion among them, because a window of four can then hold the moved end with both copies of the stretch whole, and sees the arrangement directly.
Six by six is one end short, and seven by seven has no stretch
At six by six the stretch, its copy and the moved end take five ends, and one is left. The two versions of the moved end cannot both stand beyond the copy, so the views from the two sides of the stretch cannot be exchanged. The repeat is one end too narrow for the mechanism, and the 547,176 pairs with no confusion are that arithmetic, checked.
The next square, seven by seven, has the ends, and the rule says a window of four is enough there too. It fails the other condition. A stretch needs ends that a slide leaves unchanged, and on seven picks the only slide that leaves an end unchanged is the slide of seven, which is no slide at all. Seven is prime, so every interlacing end of a seven-pick cloth looks different after every slide, and no stretch can be built. On a square repeat, then, the mechanism that broke four at seven ends has no room at six and no material at seven, and above seven the half-repeat rule no longer says a window of four is enough.
So on the squares the question was asked about, the answer from this family is that four does name every six-by-six cloth the stretch can build. The squares are where the mechanism is shut out. The rectangles are where it is not.
The rule’s exception is a family, not a pair
Six picks by seven ends is past what the walk could take, at 42 intersections against its limit of 28, and the rule says a window of four is wider than half of seven. Six picks allow a slide of two. Both conditions hold, so the shape should be open to the mechanism, and it is.
At six by seven the search finds 336 confused pairs among 2,949,396, and they are writings of four pairs of families. Each has the shape above: a stretch of two plain ends, its copy, a moved end slid two picks or four, and the moved end in both positions beyond. At their nearest writings, two of the four pairs differ in two intersections and two in four. Like the four-by-seven pair, each is a small figure on a plain ground — and a specification that named cloths by four-by-four windows would admit both members of each as one cloth.
That changes what the walk’s exception was. It looked like a single coincidence at one shape. It is the smallest member of a family of exceptions that exists wherever a repeat has at least seven ends, a number of picks with an even divisor, and a window too narrow to hold a stretch, the moved end and the copy at once. The six-by-seven search took only a slide of two with the copy in order; slides of three and mirrored copies at that shape were not run, and the four pairs found are a floor on that shape’s confusions, not a count.
So the half-repeat rule holds for every square repeat this family can reach, and fails on the rectangles beside them. For a reader specifying a cloth by its windows, the second clause the walk added — check that the cloth holds no repeated stretch long enough to defeat the window — is not a footnote for one pair. It is a check with a definite form: a stretch unchanged by some slide, its copy, and an end that appears beside them in two positions.
What the search says about the walk’s own reasoning
The walk compared its pair to genome assembly, where short reads cannot place the copies of a repeated sequence unless some read spans a copy and both its flanks. That analogy holds, and the search adds the part a genome does not need. A genome assembler is defeated by a repeat alone; a cloth’s census is defeated only when the flanks themselves are arranged so the two orders give the same reads. A cloth has to supply its own decoy, the moved end written out in both positions, and that is what costs the seventh end.
It also explains why the walk’s random sample would have missed it. Of every draft that could fill a six-by-seven repeat, the search’s family is a small corner, and the confused pairs are 336 in nearly three million even inside that corner. A structure that rare is found by naming it, as silence lives in the lifting plan found when a random sample of lifting plans said there was no silence and the twills held most of it. The same lesson ran through a mispick is one row in the wrong place, where the harm of one wrong shed depended entirely on the draft it fell in, and through a lifting plan says nothing without a threading: in this subject the interesting cases are built, not drawn at random.
How the search was made
The family. Ends are written as columns read down the picks. A stretch is a run of w ends each unchanged by a slide of p picks, with p a divisor of the picks greater than one. The copy is the same run written again in order or reversed. The moved end is any end that interlaces, and so is every free end. The second cloth is the first with the moved end slid by each amount from one pick to one less than the repeat. A pair is kept only when every pick of both cloths interlaces.
The census. Each cloth’s four-by-four windows are read at every position of the repeat, coded, sent through each symmetry of the shape by a table, sorted, and the least of those lists is the census’s name — the walk’s own reading, unchanged, so that a pair the walk would call confused is one the search calls confused. The orbit test builds every writing of the first cloth under slides, reversals, counterchange and, on a square, the quarter turn, and asks whether the second is among them.
Required of it, not shown: that it find nothing at four by six, where the walk found nothing; that at four by seven its confusions be writings of the walk’s one pair; that it find nothing at six by six; that it find confusions at six by seven; that every confusion it finds hold the moved end as it stands and slid among its free ends; and that it refuse a window as wide as the repeat.
What the search does not reach
Six by six is not settled. The family searched is one mechanism, the only one known, and it is shut out. Whether some other arrangement defeats a window of four at six by six — not a stretch with a moved end, but something with no name yet — needs either a census of the whole shape or a proof, and neither is here. The half-repeat rule at six by six is now supported by more than its walked neighbours and remains observed rather than proved.
Stretches two ends wide were the ones taken. A stretch of one end fails at four by seven for the reason given, and at six by six a stretch of three leaves no room for anything else. Longer repeats would allow longer stretches and larger windows, and the arithmetic predicts the mechanism wherever the ends number at least the stretch’s two copies plus three and the window is too narrow to hold a stretch, the moved end and the copy at once. A window of five holds a stretch of two with its moved end and copy, so it needs a stretch of three, and nine ends; a window of six needs nine to eleven. None of that has been run.
The derivations are the walk’s. The families are the manuals’ operations — slides, reversals, counterchange, quarter turn — and a cloth’s census of small patches names those families. A cloth turned over, or woven with its warp and weft exchanged on a rectangular repeat, is a different question.
Who found which part
The derivations are the weaving manuals’, whose claim that the three basic weaves generate the rest started this line of counting. The census of windows comes from the essay on what a cloth’s derivation class is and the half-repeat rule from the walk over eleven shapes; the cloths themselves are drawn from the catalogue of four-by-four upward. The reads-and-repeats problem is Idury and Waterman’s and Pevzner’s from the 1990s.
What is added here is the stretch family searched exhaustively at four shapes; that it recovers the walk’s one pair and finds no confusion at six by six; that every confusion it finds holds the moved end in both positions beyond the stretch, so that the mechanism needs seven ends; that seven by seven cannot build a stretch at all; and four new confused pairs at six by seven, where the rule said a window of four was enough.
Still open: whether a window of five fails on nine ends
The seven-end count says where to look next. A window of five is wider than half of nine, and it holds a stretch of two with the moved end and the copy beside it, so the mechanism needs a stretch of three: three ends, the moved end, three for the copy and two for the moved end in both positions, nine ends exactly. The picks have to be at least five, for the window, and have a divisor, for the stretch. Six picks by nine ends should therefore hold a confused pair built the same way, and no shape between six by seven and it should.
It is a heavier search than any here, with tables for a five-by-five window running to thirty-three million codes a symmetry, but it is a definite one, and it is the first shape where the mechanism would be shown to grow with the window rather than to belong to a window of four. If it does not appear there, the seven-end count was not the whole of the condition, and something about four picks and six made the family possible that a wider window removes.
Shares its objects with
Essays naming at least two of the same things, that neither author linked.
- A proof plan has no slide in it — both name census, draft, symmetry
- What combining two weaves reaches — both name census, derivation, repeat
- A blind cell costs half the cloths at any real repeat — both name census, draft
- A crepe cannot be structureless — both name census, repeat
- A float limit gives a blind set its arrangement back — both name census, draft
- A profile draft is a notation whose alphabet is weaves — both name census, repeat
Named objects
A flat tag is an object no other essay names yet.