Towards Resolving Keller’s Cube Tiling Conjecture in Dimension Seven
Andrzej P. Kisielewicz
Wydział Matematyki, Informatyki i Ekonometrii, Uniwersytet Zielonogórski
ul. Z. Szafrana 4a, 65-516 Zielona Góra, Poland
[email protected]
Abstract
A cube tiling of Rd is a family of pairwise disjoint cubes [0,1)d+T={[0,1)d+t:t∈T} such that ⋃t∈T([0,1)d+t)=Rd. Two cubes [0,1)d+t, [0,1)d+s are called a twin pair if ∣tj−sj∣=1 for some j∈[d]={1,…,d} and ti=si for every i∈[d]∖{j}.
In 1930, Keller conjectured that in every cube tiling of Rd there is a twin pair. Keller’s conjecture is true for dimensions d≤6 and false for all dimensions d≥8. For d=7 the conjecture is still open. Let x∈Rd, i∈[d], and let L(T,x,i) be the set of all ith coordinates ti of vectors t∈T such that ([0,1)d+t)∩([0,1]d+x)=∅ and ti≤xi. Let r−(T)=minx∈Rdmax1≤i≤d∣L(T,x,i)∣ and r+(T)=maxx∈Rdmax1≤i≤d∣L(T,x,i)∣.
It is known that if r−(T)≤2 or r+(T)≥5, then Keller’s conjecture is true for d=7. In the paper we show that it is also true for d=7 if r+(T)=4. Thus, if [0,1)7+T is a counterexample to Keller’s conjecture, then r+(T)=3, which is the last unsolved case of Keller’s conjecture. Additionally, a new proof of Keller’s conjecture in dimensions d≤6 is given.
Key words: box, cube tiling, Keller’s conjecture, Keller graph, rigidity.
MSC: 52C22, 05C69, 94B25
1 Introduction
A cube tiling of Rd is a family of pairwise disjoint cubes [0,1)d+T={[0,1)d+t:t∈T} such that ⋃t∈T([0,1)d+t)=Rd. Two cubes [0,1)d+t, [0,1)d+s are called a twin pair if ∣tj−sj∣=1
for some j∈[d]={1,…,d} and ti=si for every i∈[d]∖{j}. In 1930, Keller [6] conjectured that in every cube tiling of Rd there is a twin pair. In 1940, Perron [15] proved that Keller’s conjecture is true for all dimensions d≤6 (see also [13]).
In 1992, Lagarias and Shor [10], using ideas from Corrádi’s and Szabó’s papers [2, 16], constructed a cube tiling of R10 which does not contain a twin pair and thereby refuted Keller’s cube tiling conjecture. In 2002, Mackey [14] gave a counterexample to Keller’s conjecture in dimension eight, which also shows that this conjecture is false in dimension nine. For d=7 Keller’s conjecture is still open.
Let [0,1)d+T be a cube tiling, x∈Rd and i∈[d], and let L(T,x,i) be the set of all ith coordinates ti of vectors t∈T such that ([0,1)d+t)∩([0,1]d+x)=∅ and ti≤xi (Figure 1).
For every x∈Rd and i∈[d] the set L(T,x,i) contains at most 2d−1 elements.
Figure 1: A portion of a cube tiling [0,1)2+T of R2. The number of elements in L(T,x,i) depends on the position of x∈R2. In the picture we have x=(23,23), y=(27,27), t=(1,1) and t′=(2,45). Then L(T,x,1)={1}, L(T,x,2)={1,45} and L(T,y,1)=L(T,y,2)={3}. This portion of the tiling [0,1)2+T shows that r−(T)=1 and r+(T)=2.
Let
[TABLE]
In 2010, Debroni et al. [3] computed that Keller’s conjecture is true for all cube tilings [0,1)7+T of R7 such that T⊂21Z7. This result shows that Keller’s conjecture is true for cube tilings of R7 with r−(T)≤2 ([7]).
In [7, 8] we showed that Keller’s conjecture is true for cube tilings [0,1)7+T of R7 for which r+(T)≥5.
In the presented paper we prove that
Theorem 1.1
Keller’s conjecture is true for all cube tilings [0,1)7+T of R7 for which r+(T)=4.
Thus, the above theorem resolves the penultimate case of Keller’s conjecture in dimension seven. To complete resolution of this conjecture remains to resolve the last case r+(T)=3. (Corollary 7.2).
Our proof of Theorem 1.1 is based on a structural result dealing with two systems of abstract words having n∈{13,...,16} words each (Theorem 7.1). It can be interpreted by means of systems of cubes in the flat torus Td={(x1,…,xd)(mod2):(x1,…,xd)∈Rd} as follows:
A set F⊂Td is called a polycube if F has a tiling by translates of the unit cube, that is, there is a family of pairwise disjoint translates of the unit cube F=[0,1)d+T, T⊂Td, such that ⋃t∈T[0,1)d+t=F.
The case r+(T)=4 is reduced to the following task: For d=6 determine all polycubes F⊂Td which have at least two twin pair free cube tilings F and G such that F∩G=∅ and ∣F∣=∣G∣=n, where n∈{13,...,16}. As we shall show (Theorem 7.1) no such F exists. This will imply Theorem 1.1 immediately.
To give an interpretation of Theorem 1.1 in the language of graph theory we define the fundamental concept of the paper: A polybox code.
A set S of arbitrary objects will be called an alphabet, and the elements of S will be called letters. A permutation s↦s′ of the alphabet S such that s′′=(s′)′=s and s′=s is said to be a complementation. Each sequence of letters s1…sd from the set S is called a word. The set of all words of length d is denoted by Sd. Two words u=u1…ud and v=v1…vd are dichotomous if there is j∈[d] such that uj′=vj. If V⊆Sd consists of pairwise dichotomous words, then we call it a polybox code (or polybox genome).
Two words u,v∈Sd form *a twin pair * if there is j∈[d] such that uj′=vj and ui=vi for every i∈[d]∖{j}.
If S is an alphabet with a complementation, then a d-dimensional Keller graph on the set Sd is the graph in which two vertices u,v∈Sd are adjacent if they are dichotomous but do not form a twin pair. It is a generalization of the well known d-dimensional Keller graph in which S={0,1,2,3} and the complementation is given by 0′=2 and 1′=3 ([2]).
As we shall show Theorem 1.1 for Keller graphs on Sd reads as follows:
Theorem 1.2
Let V⊂S7 be a clique in the Keller graph on S7 such that there is i∈[7] and there are four words va,vb,vc,vd∈V with via∈{a,a′},...,vid∈{d,d′}, where {a,b,c,d}⊂S. Then ∣V∣<27.
We also show that the mentioned above Theorem 7.1 together with Theorem 29 of [8] give us a clue about the form of a counterexample in the last case of the conjecture, if it exists:
Corollary 1.3
There is a counterexample to Keller’s conjecture in dimension seven if and only if there is a clique V in the Keller graph on {a,a′,b,b′,c,c′}7 having 128 words such that there are i∈[7] and three words va,vb,vc∈V with via∈{a,a′},vib∈{b,b′}, vic∈{c,c′} and ∣Vj,l∣>16 for every l∈{a,a′,b,b′,c,c′} and every j∈[7] such that Vj,l=∅, where Vj,l={v∈V:vj=l}.
At the end of the paper we shall also give a new proof of Keller’s conjecture in dimensions d≤6.
The presented paper is a continuation of two earlier papers [7, 8] devoted to Keller’s conjecture in dimension seven. Therefore, the first two sections containing the basic concepts have been limited to a necessary minimum. More comprehensive presentation of the notions dealing with the structure of polybox codes can be found in the mentioned two papers.
The outline of the paper will be presented after the following section:
2 Basic notions
In this section we present the basic notions on dichotomous boxes and words (details can be found in [7, 8, 9]). We start with systems of boxes.
In the whole paper, if X is a family of sets, then ⋃X=⋃A∈XA. Moreover, if Y is a set, then a partition of Y is a family Y of its pairwise disjoint subsets such that ⋃Y=Y.
2.1 Dichotomous boxes and polyboxes
Let X1,…,Xd be non-empty sets with ∣Xi∣≥2 for every i∈[d]. The set X=X1×⋯×Xd is called a d-box.
A non-empty set K⊆X is called a * box* if K=K1×⋯×Kd and
Ki⊆Xi for each i∈[d].
The box K is said to be * proper* if Ki=Xi for each i∈[d].
Two boxes K and G in X are called dichotomous
if there is i∈[d] such that Ki=Xi∖Gi. A suit is any collection of pairwise
dichotomous boxes. A suit is proper if it consists of proper boxes.
A non-empty set F⊆X is said to be a * polybox* if
there is a suit F for F, that is, if ⋃F=F. In other words, F is a polybox if it has a partition into pairwise dichotomous boxes.
A polybox F is rigid if it has exactly one suit, that is, if F and G are suits for a rigid polybox, then F=G.
A proper suit for a d-box X is called a *minimal partition * of X. In [5] we showed that a suit F is a minimal partition of a d-box X if and only if ∣F∣=2d.
Two boxes K,G⊂X are said to be a twin pair if Kj=Xj∖Gj for some j∈[d] and Ki=Gi for every i∈[d]∖{j}. Alternatively, two dichotomous boxes K,G are a twin pair if K∪G is a box. Observe that the suit for a rigid polybox cannot contain a twin pair.
The next concept is of particular importance in an analysis of the structure of suits.
Let X be a d-box, and let li={x1}×⋯×{xi−1}×Xi×{xi+1}×⋯×{xd}, where xj∈Xj for j∈[d]∖{i}. A set F⊆X is called an i-cylinder (Figure 2) if for every set li
one has
[TABLE]
Figure 2: The set on the left is a 3-cylinder in X=[0,1]3, and the set on the right is not, because the set l3={x}×{y}×[0,1] has a non-empty intersection with this set but l3 is not entirely contained in it.
Let F be a suit for a d-box X, A⊆Xi, and let Fi,A={K∈F:Ki=A}. Note now that the partition F has the “cylindrical” structure ([1, 11, 12]) (compare Example 2.3): Let i∈[d], and let A1,...,Aki⊂Xi be all sets such that Fi,Aj=∅ for j∈[ki] and An∈{Aj,(Aj)c} for n,j∈[ki],n=j, where (Aj)c=Xi∖Aj. Then the set ⋃(Fi,Aj∪Fi,(Aj)c) is a non-empty i-cylinder in X and F=⋃j∈[ki](Fi,Aj∪Fi,(Aj)c)∪Fi,Xi.
2.2 Cube tilings and dichotomous boxes
Every two cubes [0,1)d+t and [0,1)d+p in an arbitrary cube tiling [0,1)d+T of Rd satisfy Keller’s condition: There is i∈[d] such that ti−pi∈Z∖{0}, where ti and pi are ith coordinates of the vectors t and p ([6]). For any cube [0,1]d+x, where x=(x1,...,xd)∈Rd, the family Fx={([0,1)d+t)∩([0,1]d+x)=∅:t∈T} is a partition of the cube [0,1]d+x. Every two boxes K,G∈Fx are, by Keller’s condition, dichotomous: There is i∈[d] such that Ki and Gi are disjoint and Ki∪Gi=[0,1]+xi. Moreover, since cubes in cube tilings are half-open, every box K∈Fx is proper, and consequently the family Fx is a minimal partition. The structure of the partition Fx reflects the local structure of the cube tiling [0,1)d+T. Obviously, a cube tiling [0,1)d+T contains a twin pair if and only if the partition Fx contains a twin pair for some x∈Rd ([12, 15]) (see Figure 1). Observe also that if Fx=Fxi,A1∪Fxi,(A1)c∪⋯∪Fxi,Aki(x)∪Fxi,(Aki(x))c, then ∣L(T,x,i)∣=ki(x) (compare (1.1)).
2.3 Distribution of words in codes. Realizations of codes
In the presented paper suits will be encoded with polybox codes described in Introduction.
In what follows we assume that S is a finite alphabet with a fixed complementation.
If V⊆Sd, l∈S and i∈[d], then Vi,l={v∈V:vi=l}. If S={a1,a1′,...,ak,ak′}, then the representation V=Vi,a1∪Vi,a1′∪⋯∪Vi,ak∪Vi,ak′
will be called a distribution of words in V. Moreover, let Di(V)=((∣Vi,a1∣,∣Vi,a1′∣),…,(∣Vi,ak∣,∣Vi,ak′∣)) for i∈[d]. For example, if S={a,a′,b,b′} and V={bba,baa′,b′ab,b′a′a},
then
[TABLE]
A natural interpretation of a polybox code is a suit for a polybox: Let X=X1×⋯×Xd be a d-box.
Suppose that for each i∈[d] a mapping fi:S→2Xi∖{∅,Xi} is such that fi(s′)=Xi∖fi(s). We define the mapping f:Sd→2X by f(s1…sd)=f1(s1)×⋯×fd(sd).
If now V⊂Sd is a code, then the set of boxes f(V)={f(v):v∈V} is a suit for the polybox ⋃f(V). The set f(V) is said to be a realization of the set V. A code has infinitely many realizations. But to effectively study the structure of suits we shall use the following realization which has particular nice properties. Let
[TABLE]
Let V⊆Sd be a polybox code, and let v∈V. The equicomplementary realization of the word v is the box
[TABLE]
in the d-box (ES)d=ES×⋯×ES. The equicomplementary realization of the code V is the family
[TABLE]
If s1,…,sn∈S and si∈{sj,sj′} for every i=j, then
[TABLE]
The value of the realization E(V), where V⊆Sd, lies in the equality (2.1). In particular, boxes in E(V) are of the same size: ∣Evi∣=(1/2)∣ES∣ for every i∈[d] and consequently ∣v˘∣=(1/2d)∣ES∣d for v∈E(V). Thus, two boxes v˘,w˘⊂(ES)d are dichotomous if and only if v˘∩w˘=∅. The same is true for cubes in a cube tiling of a polycube F⊂Td and therefore working with the boxes v˘,v∈V, we can think of them as translates of the unit cube in Td.
2.4 Cover of a code, equivalent and rigid polybox codes
Let V,W⊆Sd be polybox codes, and let w∈Sd. We say that w is covered by V, and write w⊑V, if w˘⊆⋃E(V).
If w⊑V for every w∈W, then we write W⊑V, and the code V is called a cover of the code W. Every cover V of w such that w∈V has the following useful property ([9, Theorem 10.6]): If v∈V is a word such that v˘∩w˘=∅ and ∣{i∈[d]:vi=wi}∣≥∣{i∈[d]:ui=wi}∣ for every u∈V with u˘∩w˘=∅, then there is p∈V such that pi∈{vi,vi′} for every i∈[d], p˘∩w˘=∅, and moreover the number ∣{i∈[d]:vi=pi′}∣ is odd. This property simplifies the computations of covers of w∈Sd (compare algorithm CoverCode at the end of this section).
Another consequence of (2.1) dealing with covers of words is given in the following lemma ([8, Lemma 2]):
Lemma 2.1
Let V⊂Sd be a cover of a word w∈Sd such that w˘∩v˘=∅ for v∈V. If V does not contain twin pairs, then also the suit F={w˘∩v˘:v∈V} for w˘ does not contain such pairs.
□**
Polybox codes V,W⊆Sd are said to be equivalent if V⊑W and W⊑V. Thus, V and W are equivalent if and only if ⋃E(V)=⋃E(W). Obviously, if V and W are equivalent, then ∣V∣=∣W∣. Two codes V and W are *disjoint * if V∩W=∅.
Let q∈Sd. We say that codes V,W⊂Sd are q-equivalent if
[TABLE]
In the paper q-equivalent codes appear in the strictly defined circumstances, which are explained at the beginning of the next subsection (see Lemma 2.2).
A polybox code V⊂Sd is called rigid if there is no code W⊂Sd which is equivalent to V and V=W. Thus, if polybox codes V,W are equivalent and one of them is rigid, then V=W. It can be checked that the code V given in Example 2.3 is rigid, while V={aaa,a′aa} is not rigid as W={baa,b′aa} is equivalent to V.
Observe that, rigid polybox codes cannot contain a twin pair.
2.5 Geometry of dichotomous boxes
In this subsection we describe the main techniques used in the paper.
Throughout the paper we consider two disjoint equivalent codes V,W⊂Sd.
If A={i1<...<ik}⊂[d] and v∈Sd, then vA=vi1…vik∈Sk and VA={vA:v∈V}. To simplify notation we let vic=v{i}c, that is, the word vic∈Sd−1 arises from v by skipping the letter vi in v. Moreover, Vic={vic:v∈V}.
∙ The structure of V from the suit for w˘, where w⊑V. Let w⊑V. Then w˘⊆⋃E(V) and the set of boxes F={w˘∩v˘:v∈V} is a suit for w˘. Assume that wi=b and the sets
[TABLE]
are non-empty (actually, if one of these sets is non-empty, then so is the second one).
The set of boxes F is a suit for the box w˘, which means that the set
[TABLE]
is an i-cylinder in the box w˘ (compare Lemma 12 of [8] and Example 2.3). Therefore, we have
Lemma 2.2
Let V,W⊂Sd be equivalent polybox codes. If there are i∈[d], l∈S and w∈W∖(Wi,l∪Wi,l′) such that the set Ui,l={v∈Vi,l:v˘∩w˘=∅} is non-empty, then the codes Uici,l,Uici,l′ are wic-equivalent. □
This is mentioned above context in which q-equivalent codes will appear in the paper. It is worth analyzing the following example in which we describe the typical situation encountered in the paper.
Example 2.3
In Figure 3 the five boxes on the left are a realization of the polybox code V={aaa,a′a′a′,baa′,a′ba,aa′b}, and the box in the middle is a realization of the word w=bbb. Since the set V is a cover of w, that is, w⊑V, we have w˘⊂⋃E(V). Thus, the 3-box w˘ is divided into pairwise dichotomous boxes of the form v˘∩w˘ for v∈V. (In other words, the set of boxes F={v˘∩w˘:v∈V} is a suit for w˘.) The set ⋃({v˘∩w˘:v∈U3,a}∪{v˘∩w˘:v∈U3,a′}),
where U3,l={v∈V3,l:v˘∩w˘=∅} for l∈{a,a′},
is a 3-cylinder in the box w˘. Therefore, ⋃{v˘3c∩w˘3c:v∈U3,a}=⋃{v˘3c∩w˘3c:v∈U3,a′}, and thus the codes U3c3,a,U3c3,a′ are w3c-equivalent.**
Figure 3: On the top: The light box (the box in the middle, which may be interpreted as w˘) is contained in the sum of five pairwise dichotomous boxes (the boxes on the left). These boxes determine a partition of the light box into pairwise dichotomous boxes (the partition on the right). On the bottom: The boxes in this partition are arranged into 3-cylinders. The sets Ea and Eb are identified with [0,21) and [41,43], respectively.
∙ The structure of W from the distribution of words in V. Below, in (P), (C) and (Co) we show how to use an information on a distribution of words in V to say something about the distribution of words in W.
Figure 4: A scheme of realizations E(V) (A), E(W) (B), where V=Vi,a∪Vi,a′ and W=Wi,a′∪Wi,b∪Wi,b′. We assume that the codes V and W are equivalent, and thus ⋃E(V)=⋃E(W).
Let V,W⊆Sd be polybox codes and assume that V and W are equivalent. Recall that Vici,l=(Vi,l)ic for every i∈[d] and l∈S. Moreover, xic=(x1,...,xi−1,xi+1,...,xd) for x∈(ES)d.
(P): Projections. Suppose that there is x∈⋃E(Vi,l′) such that xic∈⋃E(Vici,l) (see Figure 4A, where l=a). Since ⋃E(V)=⋃E(W) and (2.1), the point x can be covered only by a box w˘∈E(W) such that wi=l′. Thus, Wi,l′=∅. In particular, if Wi,l=∅ and Wi,l′=∅, then Wi,l′⊑Vi,l′ (compare Figures 4A and 4B). Note also that, by (2.1), the white box in Figure 4C cannot be intersected by a box v˘ for v∈V∖(Vi,a∪Vi,a′). Thus, the white box cannot be intersected by a box w˘ for w∈W. Finally, observe that if w˘∩v˘=∅ for some w∈W and v∈V with wi∈{vi,vi′}, then every point x∈v˘∖w˘ such that xic∈w˘ic is covered only by boxes u˘, u∈W, such that ui=wi′.
It follows from the above
Lemma 2.4
Let V,W⊂Sd be equivalent codes. Then
[TABLE]
for every i∈[d] and every l∈S. □
For example, it can be computed that if V,W⊂Sd, d=5,6, are disjoint equivalent codes without twin pairs and ∣W1,l∣=5 for l∈{a,a′} and ∣V1,l∣=1 for l∈{a,a′}, then, up to isomorphism (it will be defined in Subsection 2.6), there is only one pair (W1,a,W1,a′) and (V1,a,V1,a′) for which (2.4) is valid:
[TABLE]
[TABLE]
Note that v⊑W1,l for v∈W1,l′ and l∈{a,a′}. (We shall use this fact in Section 6.)
Let V,W⊂Sd be codes, and let
[TABLE]
The number m(V∖W) counts the fraction of boxes v˘,v∈V, which are not covered by boxes w˘,w∈W.
It follows from Lemma 2.4 that for every equivalent codes V and W we have
[TABLE]
For example, for the code V in Example 2.3 we have m(V3c3,l∖V3c3,l′)=43 for l∈{a,a′} (recall that Vici,l⊂Sd−1 for V⊂Sd.)
(C): Cylinders. Suppose that Vi,l∪Vi,l′=∅ and Wi,l∪Wi,l′=∅ for some l∈S. Then, by (2.4), ⋃E(Wici,l)=⋃E(Wici,l′), and hence the set ⋃E(Wi,l∪Wi,l′) in an i-cylinder in the d-box (ES)d, that is, ⋃E(Wici,l)=⋃E(Wici,l′) (compare Figures 4A and 4B, where l=b).
Thus, by the definition, the codes Wici,l and Wici,l′ are equivalent.
(Co): Covers. Suppose that polybox codes Vici,l and Wici,l∪Wici,s1∪⋯∪Wici,sk are equivalent, where sn∈{l,l′,sj,sj′} for every n,j∈[k],n=j. Then
[TABLE]
(In Figures 4A and 4B the codes Vici,a and Wici,a∪Wici,b are equivalent, where a=l,b=s1, and Wi,a=∅). Indeed, since boxes in E(V) are pairwise dichotomous and, by the definition of equivalent codes, w˘ic⊆⋃E(Vici,l) for every w∈Wi,s1∪⋯∪Wi,sk, it follows that each point x∈w˘∖⋃E(Vi,l) has to be covered by the set ⋃E(Vi,l′). Therefore, w˘ic⊆⋃E(Vici,l′) for every w∈Wi,s1∪⋯∪Wi,sk, and consequently wic⊑Vici,l′ for w∈Wi,s1∪⋯∪Wi,sk. Thus, Wici,s1∪⋯∪Wici,sk⊑Vici,l′. In the same manner we show that Vici,l⊑Wici,l∪Wici,s1′∪⋯∪Wici,sk′.
∙ The structure of W from the structure of Vi,l∪Vi,l′.
The following lemma describes a useful relationship between codes Vi,l∪Vi,l′ and Wi,s∪Wi,s′, where s∈{l,l′}.
Lemma 2.5
Let V,W⊂Sd be disjoint equivalent polybox codes, and let Ui,l={v∈Vi,l:v˘∩⋃E(W∖(Wi,l∪Wi,l′))=∅} be non-empty set for some i∈[d] and l∈S. Assume that there are disjoint sets A,B such that [d]∖{i}=A∪B and there are two words p,q such that:
(i) wA=p for every w∈W∖(Wi,l∪Wi,l′) with w˘∩⋃E(Ui,l∪Ui,l′)=∅.
(ii) (Ui,l∪Ui,l′)B={q}.
If P={w∈W∖(Wi,l∪Wi,l′):w˘∩⋃E(Ui,l∪Ui,l′)=∅},
then for every s∈S∖{l,l′} the codes Pici,s and Pici,s′ are r-equivalent, where rA=p and rB=q.
In particular, if ∣B∣≤1, then there is a twin pair in V or in W.
*Proof. *For simplicity, let i=1, l=a, s=b and A={2,...,k},B={k+1,...,d}. Moreover, xB=(xk+1,...,xd) and x1c=(x2,...,xd) for x∈(ES)d. To show that that P1c1,b and P1c1,b′ are r-equivalent it is enough to show, by (i), that
[TABLE]
Take a point xB∈⋃w∈P1,bw˘B∩q˘. Then xB∈w˘B∩q˘ for some w∈P1,b. By the definition of P1,b, there is v∈U1,a∪U1,a′ such that w˘∩v˘=∅. Assume that v∈U1,a and take y∈v˘∩w˘ such that yB=xB, which is possible by (ii). Then, by (P), a point z∈w˘ such that z1∈Ea′∩Eb and z1c=y1c must be covered by a box u˘ for some u∈U1,a′. We take one more point t∈u˘ such that t1∈Ea′∩Eb′ and t1c=z1c. Clearly, again by (P), t∈w˘1∩u˘ for some w1∈P1,b′. Since tB=xB and u˘B=q˘, we obtain xB∈w˘B1∩q˘⊆⋃w∈P1,b′w˘B∩q˘. Consequently, ⋃w∈P1,bw˘B∩q˘⊆⋃w∈P1,b′w˘B∩q˘. In the same manner we show the reverse inclusion.
To prove the second part of the lemma let first ∣B∣=1. If ∣P1,b∣≥2, then there are two words w,u∈W1,b such that w1=u1=b and wA=uA. Since w,u are dichotomous, wd=ud′, and thus w and u are a twin pair. If P1,b={w}, P1,b′={u} and v∈U1,a∪U1,a′ is such that w˘∩v˘=∅ and u˘∩v˘=∅, then, by Lemma 2.2 (in which we change the role of V and W) the codes P1c1,b and P1c1,b′ are v1c-equivalent, and hence, by (2.1), wd=ud. Thus, w,u form a twin pair.
If B=∅, then for every w∈P1,b and u∈P1,b′, we have wA=uA and w1=u1′, that is, the words w,u form a twin pair.
□
Let
g:Sd×Sd→Z be defined by the formula g(v,w)=∏i=1d(2[vi=wi]+[wi∈{vi,vi′}])
where \mbox[p\mbox]=1 if the sentence p is true and \mbox[p\mbox]=0 if it is false. Let w∈Sd, and let V⊂Sd be a polybox code. In [9] it was showed that
[TABLE]
∙ The structure of V from (2.7). Let V,W⊂Sd be disjoint equivalent polybox codes.
Then for every w∈W we have w⊑V,w∈V, and, by (2.7), ∑v∈Vg(v,w)=2d, where g(v,w)∈{0,1,2,…,2d−1} for every v∈V. Assume that w=b…b and let {v1,…,vk}⊆V be such that w˘∩v˘i=∅ for every i∈[k] and w⊑{v1,…,vk}. The solutions of the system of equations ∑i=0d−1xi2i=2d,∑i=0d−1xi=k, where xi are non-negative integers for i∈{0,1,…,d−1}, show the frequency of the letter b in the words from the set {v1,…,vk}. We explain this on the following example. Recall first that g(v,w)=2i if and only if vj=b for every j∈I⊂[d], ∣I∣=i and vj∈{b,b′} for j∈[d]∖I. In the example we assume that d=3, w=bbb and k=5. The above system has two solutions: x0=2,x1=3,x2=0 and x0=4,x1=0,x2=1. It follows from the first solution that in the cover {v1,…,v5} of w there are exactly three words such that each of them contains exactly one letter b and two words which have no letter b or, by the second solution, in the set {v1,…,v5} there is exactly one word with two letters b and the rest four words have no letter b. This observation is quite useful in the computations of covers of a word w∈W as it allows us to restrict the number of words which have to be considered during the computations (see algorithm CoverWord at the end of this section).
∙ The structure of V from a graph of siblings
In [7] we defined a graph on a polybox code V. We now recall the definition of it.
Two words v,u∈Sd such that vi∈{ui,ui′} for some i∈[d], and vic, uic is a twin pair are called i-siblings (the two top boxes in Figure 3 (on the left) are a realization of 1-siblings).
Let V⊆Sd be a polybox code. A graph of siblings on V is a graph G=(V,E)
in which two vertices v,u∈V are adjacent if they are i-siblings for some i∈[d]. We colour each edge in E with the colours from the set [d]: An edge (v,u)∈E has a colour i∈[d] if v,u are i-siblings.
Observe that if v,u are i-siblings in a polybox code V such that vi=l and ui=s, l∈{s,s′}, then the set Vici,l∪Vici,s contains the twin pair vic,uic. In Section 5 we shall use a subgraph of G to estimate the cardinalities of covers of w∈W by words from V.
2.6 Isomorphic polybox codes
If v∈Sd, and σ is a permutation of the set [d], then σ∗(v)=vσ(1)…vσ(d). For every i∈[d] let hi:S→S be a bijection such that hi(l′)=(hi(l))′ for every l∈S, and let h:Sd→Sd be defined by the formula h(v)=h1(v1)…hd(vd). We say that polybox codes P,Q⊂Sd are isomorphic if there are σ and h such that Q={h1(vσ(1))…hd(vσ(d)):v∈P}. The composition h∘σ∗ is an isomorphism between P and Q. Let V and W be disjoint, equivalent and twin pair free polybox codes, and let h∘σ∗(V) be an isomorphic code to V. It follows from the definition of the isomorphism h∘σ∗ that the codes h∘σ∗(V) and h∘σ∗(W) are also disjoint, equivalent and do not contain a twin pair. To show this, it is enough to notice that the definition of h∘σ∗ guarantees that h∘σ∗(w)⊑h∘σ∗(V), whenever w⊑V. Moreover, for every i∈[d] and l∈S there are j∈[d] and s∈S such that ∣Vi,l∣=∣(h∘σ∗(V))j,s∣.
Therefore, usually we shall assume that V or W contains specific codes. In the following example we explain this statement presenting a typical situation that appears in the paper:
Example 2.6
Let V,W be two disjoint twin pair free equivalent codes. Suppose that V contains a code Q which is isomorphic to a code U={aabbbb,aa′abbb,a′babbb,a′aa′bbb}. If f is a isomorphism between Q and U, then U⊂f(V). Assume now that from this inclusion we are able to deduce that f(W) has to contain a code R such that: R=R1,b∪R1,b′, RA={bb} and the codes RB1,b, RB1,b′ are q-equivalent for q=bbb, where A={2,3} and B={4,5,6}. Let K and M be q-equivalent codes such that there is an isomorphism h with h∣B(q)=q, h∣B(RB1,b)=K, h∣B(RB1,b′)=M and moreover h∣H=id, where H={1,2,3}. Then U⊂h∘f(V) and h∘f(W) contains a code T such that T=T1,b∪T1,b′, TH=RH and TB1,b=K, TB1,b′=M. Clearly, h∘f(V) and h∘f(W) are disjoint twin pair free equivalent codes and for every i∈[d] and l∈S there are j∈[d] and s∈S such that ∣Vi,l∣=∣(h∘f(V))j,s∣ and ∣Wi,l∣=∣(h∘f(W))n,r∣ for some n∈[d] and r∈S. Therefore, we may assume at the very beginning that U⊂V and T⊂W.
2.7 Algorithms
Below we give two algorithms used in the paper.
Algorithm CoverWord.
Let u∈Sd, d=5,6, be a word, k≥5 be an integer, and let Ck be the family of all k-elements twin pair free covers of u such that u˘∩v˘=∅ for every v∈C and every C∈Ck. For simplicity let u=b...b. Let V3,0={aaaa...a,a′a′a′a...a}, V3,1={aaaba...a,a′a′a′ba...a}, V3,2={aaabba...a,a′a′a′bba...a}, V3,3={aaabbb,a′a′a′bbb} and V5,0={aaaaaa...a,a′a′a′a′a′a...a}, V5,1={aaaaab,a′a′a′a′a′b}. By the property given at the beginning of Subsection 2.4, every cover in Ck contains, up to isomorphism, a code Vn,i for some n∈{3,5}, i∈{0,...,3}. In the algorithm we use also the property discussed below (2.7). Our goal is to find the family Ck.
Input. The word b...b∈Sd and the number k.
Output. The family Ck.
-
Let Sk={(x0,...,xd−1)∈Nd:∑i=0d−1xi2i=2dand∑i=0d−1xi=k}, where N={0,1,2,...}.
-
For i∈{0,...,d−1} indicate the set Ai consisting of all words v∈Sd such that v contains precisely i letters b and v does not contain the letter b′.
-
Fix x∈Sk and let s(x)={i1<⋯<im} consists of all ij∈{0,...,d−1},j∈[m], for which xij>0. Fix Vn,i1⊂Ai1.
For i∈s(x) let Bi={v∈Ai:Vn,i1∪{v}isatwinpairfreecode}.
-
Let I be the multiset containing i1 with the multiplicity xi1−2 (recall that xi1≥2) and ij with the multiplicity xij for j∈{2,...,m}. By I[j] we denote the jth element of I.
-
Let D2={Vn,i1}.
-
For l∈{2,...,k−1} having computed Dl we compute the set Dl+1: For v∈BI[l] and for U∈Dl if U∪{v} is a twin pair free code, then we attach it to Dl+1.
-
Clearly, Dk=Dk(Vn,i1,x) so let Cˉk be the union of the sets Dk(Vn,i1,x) over x∈Sk and Vn,i1⊂Ai1 (recall that Vn,i1 depends on x).
-
Ck=⋃f∈Fbf(Cˉk), where Fb consists of all isomorphism f, defined in Subsection 2.6, such that f(b...b)=b...b.
Algorithm CoverCode.
Let U={u1,...,un} be a code and for every i∈[n] let Pi be a code such that u˘i∩p˘=∅ for every p∈Pi. Let Cui,Pi, i∈[n], be the family of all covers Cui of the word ui∈U such that Pi⊂Cui.
Our goal is to find the family CU of all covers CU of the code U such that Pi⊂Cui for i∈[n] and ∣CU∣≤m for a fixed m∈{1,2...}, where Cui⊂CU is the cover of the word ui.
Input. The codes U, Pi, i∈[n], the number m and the family (Cui,Pi)ui∈U.
Output. The family CU.
-
For C1∈Cu1,P1 and C2∈Cu2,P2 if the set C1∪(C2∖C1) is a twin pair free code (it is obviously a cover of the code {u1,u2}) and has at most m words, then it is attached to the set C1,2.
-
Assuming that the set C1,...,k has already been computed for 2≤k<n, we compute the set C1,...,k,k+1: for C∈C1,...,k and Ck+1∈Cuk+1,Pk+1 if the set C∪(Ck+1∖C) has at most m words and it is a twin pair free code (being a cover of {u1,...,uk+1}), then it is attached to the set C1,...,k,k+1.
-
CU=C1,...,n.
3 Covers of a code and q-equivalent codes
To show that every cube tiling [0,1)7+T of R7 such that r+(T)≥6 contains a twin pair, in [7, Theorem 2.7] we proved
Theorem 3.1
If V,W⊂Sd are disjoint equivalent codes without twin pairs, then ∣V∣≥12.
To prove the main theorem of the presented paper (Theorem 1.1, which resolves the case r+(T)=4) we have to settle whether there are two disjoint twin pair free equivalent codes V,W⊂Sd for d=6 such that ∣V∣∈{13,...,16}. We shall examine such codes V,W⊂Sd for d=5 and d=6 separately. The reason is that our method of analyzing the structure of equivalent codes V,W works better when for every i∈[d] the sets Vi,l and Wi,l are non-empty for at least two letters l∈S. Therefore, we shall exclude the case when V,W are flat, that is, they are of the form V=Vi,l and W=Wi,l for some i∈[d] and some l∈S. (A twin pair free disjoint equivalent codes V,W⊂Sd with ∣V∣=12 are not flat only for d=4 (see [8]).)
In the next four sections we show which codes V,W may be excluded from the considerations.
Those reductions will enable us to carry out the main computations, described in Section 7, using home PC equipped with processor Intel i7 and 32GB RAM memory.
In Section 3 we give covers of small codes U⊂Sd for d=4,5 as well as q-equivalent codes which will be used throughout the paper.
In the next section we eliminate some codes V,W based on their distributions of words (Lemma 4.5 and 4.6i). Additionally, in Lemma 4.6ii, we show how the distribution of words in V affects the position of some words in W relative to the words in V. In Section 5 we estimate the cardinalities of covers of a single word in V and W (Lemma 5.3). To do that, we first give a result on the rigidity of a code (Lemma 5.1). In Section 6 we show that codes V,W under consideration may be written down in the alphabet S={a,a′,b,b′}. Finally, in Section 7 based on the reductions made in the previous sections we prove Theorem 1.1.
3.1 Covers of a code
In the first part of the paper we shall need directly only two, up to isomorphism, twin pair free covers Cv of a word v∈Sd, d=4,5, such that ∣Cv∣∈{5,6} and v˘∩u˘=∅ for every u∈Cv. They can be easily computed using algorithm CoverWord and for v=bbbbb they are of the form:
[TABLE]
Table 1: Twin pair free covers of the word bbbbb with five and six words.
Obviously, C5c1,C5c2 are covers of v=bbbb.
Lemma 3.2
Let Ui,Ci⊂Sd, where S={a,a′,b,b′,c,c′}, be twin pair free disjoint codes such that Ui⊑Ci and ∣Ui∣=i for i∈[4]. Then
(i)* ∣C1∣≥5 for d≥3.*
(ii)* ∣C2∣≥7 for d=5 and there is precisely one, up to isomorphism, cover C2 of U2 with ∣C2∣=7. This is C2={ba′bbb,aabbb,a′aaaa′,a′aaba,a′aa′ab,a′aa′a′a,a′aba′a′}, where U2={bbbbb,b′abbb}.*
(iii)* ∣C2∣≥8 for d=4 and ∣C3∣≥9, ∣C4∣≥10 for d=4,5.*
*Proof. *We sketch the prove of the fact ∣C4∣≥10 for d=5. For n∈[5] let Pn be a set of all, up to isomorphism, twin pair free two elements codes {v,w}⊂S5 such that ∣{i∈[5]:vi=wi′}∣=n, and let P=⋃n∈[5]Pn.
Using algorithm CoverWord we compute the family Cv consisting of all twin pair free covers Cv of each word v that appears in P such that v˘∩u˘=∅ for every u∈Cv and ∣Cv∣≤9 for every Cv∈Cv. Next, based on algorithm CoverCode, we compute twin pair free covers CQ of Q∈P which are disjoint with Q and ∣CQ∣≤9. (Note that computing CQ we can omit all covers Cv, v∈Q, such that ∣Cv∣=9 and Q⊑Cv.) For Q∈P3∪P4∪P5 there are no such covers CQ, and every non-empty cover CQ of Q∈P2 has nine words. But for every such CQ if Q∪{v,w} is a twin pair free code with four words, v,w∈S5, and CQ∩(Q∪{v,w})=∅, then CQ is not a cover of Q∪{v,w}.
For every Q∈P1 such that ∣CQ∣=9 it can be computed that if Q∪{v,w} is a twin pair free code having four words which is disjoint with CQ, v,w∈S5, then CQ is not a cover of Q∪{v,w}. For Q∈P1 if CQ contains eight words, then the computations show that for every v,w,u∈Sd if CQ∪{u}, Q∪{v,w} are disjoint twin pair free codes containing nine and four words, respectively, then Q∪{v,w} is not covered by CQ∪{u}. Similarly, if ∣CQ∣=7, then for every v,w,u,q∈S5 if CQ∪{u,q}, Q∪{v,w} are disjoint twin pair free codes having nine and four words, respectively, then Q∪{v,w} is not covered by CQ∪{u,q}.
□
3.2 q-equivalent codes
From Lemma 3.2i, 2.1 and (2.1) we obtain
Corollary 3.3
For every d≥1 and every q∈Sd there are no q-equivalent twin pair free disjoint codes K,M⊂Sd such that ∣K∣=1 and ∣M∣∈[4].
□
The following q-equivalent codes will be used throughout the paper:
Lemma 3.4
Let K,M⊂Sd be disjoint and twin pair free q-equivalent codes.
(i)* If ∣K∣=1 and ∣M∣=5, then, up to isomorphism, qA=bbb, KA={bbb} and MA={aaa,a′a′a′,baa′,a′ba,aa′b},
where A={1,2,3}, KAc=MAc={b...b} and qAc=s4...sd, si=b′ for i∈{4,...,d}. If ∣K∣=1 and ∣M∣=6, then, up to isomorphism, qA=bbbb, KA={bbbb} and MA={aaaab,a′a′a′ab,baa′ab,a′baab,aa′bab,bbba′b},
where A={1,2,3,4}, KAc=MAc={b...b} and qAc=s5...sd, si=b′ for i∈{5,...,d}.*
(ii)* If ∣K∣=∣M∣=2, then, up to isomorphism, qA=bb, KA={ab,a′a} and MA={ba,aa′},
where A={1,2}, KAc=MAc={b...b} and qAc=s3...sd, si=b′ for i∈{3,...,d}.*
(iii)* If ∣K∣=2,∣M∣=3, then, up to isomorphism, qA=bbb and*
[TABLE]
where A={1,2,3}, KAc=MAc={b...b} and qAc=s4...sd, si=b′ for i∈{4,...,d}.
(iv)* In (i)-(iii) if rA=qA, where r∈Sd, then K and M are not r-equivalent.*
(v)* There are, up to isomorphism, seventeen disjoint and twin pair free q-equivalent codes K,M for q=bbb written down in the alphabet S={a,a′,b}. These are*
[TABLE]
Table 2: q-equivalent twin pairs-free disjoint codes K,M for q=bbb. Recall that two q-equivalent codes K,M are isomorphic to q-equivalent codes K1,M1 if there is an isomorphism f such that f(q)=q and f(K)=K1 and f(M)=M1.*
*Proof. *The part (i) follows from the forms of the covers C1 and C2 given in Table 1 and the definition of q-equivalent codes.
The simplest way to find all, up to isomorphism, q-equivalent twin pair free disjoint codes K,M for q=bbb, which are enumerated in Table 2, is to compute the family C3 of all covers of the word bbb having n∈{3,...,8} words. Having C3 for every C1,C2∈C3 if C1∖C1∩C2 and C2∖C1∩C2 are disjoint and twin pair free codes, then K=C1∖C1∩C2 and M=C2∖C1∩C2. (Clearly, codes in Table 2 can be found by the trial and error method.)
Since it is easy to show that the codes given in (ii) and (iii) are such that the set KAc is a singleton whose element belongs to Sd−2 for (ii) and to Sd−3 for (iii), it follows that from C3 we obtain also codes described in (ii) and (iii).
We prove the statement (iv) for the case (ii). (Along the same lines we prove this statement for (i) and (iii)). Let us consider the following boxes: K1=(Eb′∩Ea)×(Ea∩Eb)×G, K2=(Eb′∩Ea′)×Ea×G, K3=(Eb∩Ea)×(Ea∩Eb′)×G and K4=Ea×(Ea′∩Eb′)×G, where G=(Eb)d−2. Clearly, K1∪K2=⋃E(K)∖⋃E(M) and K3∪K4=⋃E(M)∖⋃E(K). It follows from the definition of q-equivalent codes that for every i∈[4] the box Ki cannot be intersected by the box q˘. Clearly, if qi∈{a,a′,b′} for some i∈{1,2}, then q˘ does not intersect all boxes in K or in M, which is impossible, by Corollary 3.3. Now, by (2.1), it is easy to check that if q is such that qi∈S∖{a,a′,b,b′} for some i∈{1,2}, then q˘∩Kj=∅ for some j∈[4], which is impossible.
□
4 Distribution of words in equivalent codes
In general, for equivalent codes V and W we have Di(V)=Di(W), i∈[d], (see Subsection 2.3 for the definition of Di(V)). However, in some cases to find the structure of equivalent codes V and W it is useful to compare the pairs (Vi,l,Vi,l′) and (Wi,l,Wi,l′), i∈[d],l,l′∈S. In this section we describe selected relationships between (Vi,l,Vi,l′) and (Wi,l,Wi,l′).
In [8] we proved the following two lemmas (Lemma 5 and Lemma 13 of [8]):
Lemma 4.1
Let S={a1,a1′,...,ak,ak′}, ε∈{0,1}k and Aε={a1ε1,...,akεk}, where an0=an,an1=an′ for n∈[k]. If V,W⊆Sd are equivalent polybox codes, then for every i∈[d] and every ε∈{0,1}k the codes ⋃l∈AεVici,l and ⋃l∈AεWici,l
are equivalent.
□
Lemma 4.2
Let V,W⊂Sd be equivalent codes. Suppose that w∈Wi,s∪Wi,s′ is a word such that the sets
[TABLE]
are non-empty, where l∈{s,s′}. If ∣Ui,l∣=1, then Uici,l⊑Uici,l′.
□
From Corollary 3.3 and Lemma 2.2 we obtain
Lemma 4.3
If V,W⊂Sd are disjoint equivalent polybox codes, the code V does not contain a twin pair and there are i∈[d] and l∈S such that ∣Vi,l∣=1 and ∣Vi,l′∣∈[4], then Vi,l⊑Wi,l and Vi,l′⊑Wi,l′.
□
Lemma 4.4
Let V,W⊂Sd be disjoint equivalent polybox codes without twin pairs. If there are i∈[d] and l∈S such that V=Vi,l∪Vi,l′ and Vi,l=∅, Vi,l′=∅, then ∣V∣≥24.
*Proof. *If W=Wi,l∪Wi,l′, then the codes Vi,l,Wi,l are disjoint and equivalent and similarly, Vi,l′,Wi,l′ are disjoint and equivalent. Thus, by Theorem 3.1, ∣Vi,l∣≥12 and ∣Vi,l′∣≥12.
If Wi,s=∅ for some s∈{l,l′}, then, by (C), Wici,s and Wici,s′ are equivalent. Since these codes are disjoint and twin pair free, again by Theorem 3.1, ∣Wici,s∣≥12 and ∣Wici,s′∣≥12.
□
Recall that a code V⊂Sd is flat if there are i∈[d] and l∈S such that V=Vi,l.
Lemma 4.5
Let V,W⊂Sd, where S={a,a′,b,b′,c,c′} and d=5,6, be disjoint equivalent polybox codes without twin pairs, ∣V∣∈{13,...,16}, and let
V be not flat. For every i∈[d] and every l∈S, if Vi,l∪Vi,l′=∅, then Vi,l=∅ and Vi,l′=∅.
*Proof. *Suppose on the contrary that there are i∈[d] and l∈S such that Vi,l′=∅ and Vi,l=∅. Then, by (P), Vi,l⊑Wi,l. We may assume that l=a.
If the codes Vi,a and Wi,a are equivalent, then, since they are disjoint and twin pair free, by Theorem 3.1, ∣Vi,a∣≥12. Observe now that the equivalence of the codes Vi,a,Wi,a implies the equivalence of the disjoint and twin pair free codes V∖Vi,a and W∖Wi,a. Then, again by Theorem 3.1, ∣V∖Vi,a∣≥12, and thus ∣V∣≥24, a contradiction.
It follows from the above that we can assume that Vi,l and Wi,l are not equivalent for l∈S.
Since the codes Vi,a and Wi,a are not equivalent, we have Wi,a′=∅. As Vi,a′=∅, it follows that, by Lemma 2.4, Wici,a′⊑Wici,a and the set Wici,a′∪Vici,a is a polybox code. In other words, taking into account that Vi,a⊑Wi,a, the codes Wici,a and Wici,a′∪Vici,a are equivalent. Thus, ∣Wi,a∣−∣Wi,a′∣=∣Vi,a∣ and, by Lemma 3.2iii, ∣Wi,a∣≥10. Hence,
[TABLE]
Assume first that Vi,b∪Vi,b′=∅ and Vi,c∪Vi,c′=∅.
It follows that, Wi,c∪Wi,c′=∅, otherwise, by (C), the codes Wici,c,Wici,c′ are equivalent, and thus, by Theorem 3.1, ∣Wici,c∣,∣Wici,c′∣≥12, a contradiction.
Then Vi,l=∅ for l∈{b,b′}, as if it is not so, in the same manner as (4.1), we show that ∣Wi,b∣+∣Wi,b′∣≥11, and consequently ∣W∣≥22, which is not true. Since ∣Wi,a∣≥10, we have ∣Wi,a′∣+∣Wi,l∣≤6 for l∈{b,b′}.
By Lemma 4.1, ∣Wi,a′∣+∣Wi,l∣=∣Vi,a′∣+∣Vi,l∣ for l∈{b,b′}, and thus ∣Vi,l∣≤6 for l∈{b,b′}.
Claim The code Vi,l is rigid.
To prove the claim let Vi,l=Q. If ∣Q∣≤5, then the rigidity of Q follows from Lemma 3.2i and Lemma 9 of [8]. Let ∣Q∣=6 and assume on the contrary that there is a code W⊂Sd which is equivalent to Q and disjoint with it.
By Theorem 3.1, there is a twin pair in W. Let u,w∈W be such a pair. We may assume that u1=w1′ and u1c=w1c. Let K=u˘∪w˘. Since K⊂⋃E(Q), the family of boxes F={K∩v˘:v∈Q} is a suit for K which, by Lemma 2.1, does not contain a twin pair. Let F1,s={K∩v˘:v∈Q1,s}, s∈S. Note that ∣F1,s∣=5 and ∣F1,s′∣=1, that is ∣Q1,s∣=5 and ∣Q1,s′∣=1. By Lemma 4.2, Q1c1,s′⊑Q1c1,s, and thus Q may be written down in the alphabet {a,a′,b,b′} (see the code C1 in Table 1). In Lemma 16 of [8] we showed that every code Q⊂{a,a′,b,b′}d with ∣Q∣≤9 is rigid. The claim has been proved
By Lemma 4.1, the codes Vici,l and Wici,a′∪Wici,l are equivalent for l∈{b,b′}. As V∩W=∅ and the code Vici,l is rigid for l∈{b,b′}, we have Vici,l=Wici,a′ for l∈{b,b′}. Then Vi,b∪Vi,b′ contains a twin pair. A contradiction.
Therefore, we may assume that Vi,l=∅ for l∈{b,b′,c,c′}. Then, as before, we show that Wi,l∪Wi,l′=∅ for l∈{b,c}.
Note that we may assume that Wi,l=∅ for l∈{b,b′}, because if on the contrary Wi,b=Wi,c=∅, then, in the same manner as (4.1), we show that ∣Vi,b∣+∣Vi,b′∣≥11 and ∣Vi,c∣+∣Vi,c′∣≥11 which is not possible.
Observe that ∣Vi,a∣≥5, otherwise ∣Wi,a′∣≥6, and then W=Wi,a∪Wi,a′ which, by Lemma 4.4, means that ∣W∣≥24. A contradiction. If Wi,c=∅, then ∣Vi,c∣+∣Vi,c′∣≥11, and consequently Vi,b∪Vi,b′=∅ which is not true.
Therefore, Wi,l=∅ for every l∈{b,b′,c,c′}. By (4.1), we may assume that ∣Wi,b∣+∣Wi,b′∣≤3. By Lemma 4.3, Wi,l⊑Vi,l for l∈{b,b′}, and by Lemma 3.2i, ∣Vi,b∣+∣Vi,b′∣≥10. But then Vi,c=∅ or Vi,c′=∅, a contradiction.
□
Lemma 4.6
Let V,W⊂Sd, where S={a,a′,b,b′,c,c′} and d=5,6, be two disjoint equivalent polybox codes without twin pairs, ∣V∣∈{13,...,16}, and let V be not flat. Then
(i)* {∣Vi,l∣,∣Vi,l′∣}={1,M}, where M∈{2,3,4}, and ∣Vi,l∣≤11 for every i∈[d] and l∈S.*
(ii)* If ∣Vi,l∣≥2 and ∣Vi,l′∣≥2 or ∣Vi,l∣=1 and ∣Vi,l′∣≥5
for some i∈[d] and l∈S, then there is w∈Wi,s∪Wi,s′, s∈{l,l′}, such that w˘∩⋃E(Vi,l∪Vi,l′)=∅.*
*Proof. *Fix i∈[d]. We first prove the lemma in the case Vi,l∪Vi,l′=∅ for every l∈{a,b,c}⊂S, which is an easy task. Observe that Wi,l∪Wi,l′=∅ for l∈{a,b,c}, otherwise, by (C), the codes Vici,l,Vici,l′ are equivalent for some l∈{a,b,c}, and thus, by Theorem 3.1, ∣Vi,l∪Vi,l′∣≥24, a contradiction. By Lemma 4.5, Vi,l,Wi,l=∅ for l∈{a,a′,b,b′,c,c′}.
We may assume on the contrary that ∣Vi,a∣=1 and ∣Vi,a′∣=M for some M∈{2,3,4}. Then, by Lemma 4.3, 3.2i and 3.2ii, ∣Wi,a∪Wi,a′∣=12 if M=2 and ∣V∣=16, and thus ∣Wi,l∣=1 for l∈{b,b′,c,c′}; if M=2 and ∣V∣≤15 or M∈{3,4} and ∣V∣≤16, then Wi,l=∅ for some l∈{b,b′,c,c′} . In the first case, again by Lemma 4.3 and 3.2i, ∣Vi,l∣≥5 for l∈{b,b′,c,c′} which contradicts the assumption ∣V∣≤16. The second case is not possible, by Lemma 4.5.
If ∣Vi,l∣≥12, then Vi,s=∅ for some s∈{a,a′,b,b′,c,c′}∖{l}, which is a contradiction, by Lemma 4.5.
Finally, if ∣Vi,l∣≥2 and ∣Vi,l′∣≥2 or ∣Vi,l∣=1 and ∣Vi,l′∣≥5 and in both cases
on the contrary Vi,l⊑Wi,l, Vi,l′⊑Wi,l′ , then, by Lemma 3.2 and 4.5, Wi,s∪Wi,s′=∅ for some s∈{a,b,c}∖{l,l′}, which is, as we showed, impossible. This completes the proof of the lemma for the case Vi,l∪Vi,l′=∅ for l∈{a,b,c}⊂S.
Therefore, in what follows we may assume that Vi,l∪Vi,l′=∅ for l∈{c,c′}
To prove the second part of (i) assume on the contrary that ∣Vi,l∣≥12 for some l∈S. Then, by Lemma 4.4 and 4.5, ∣Vi,s∣=∣Vi,s′∣=1 or ∣Vi,s∣=1, ∣Vi,s′∣=2, where s∈{l,l′}. Therefore, by Lemma 4.3 and 3.2i, ∣Wi,s∪Wi,s′∣≥10 from where ∣Wi,l∪Wi,l′∣≤6. Consequently, m(Wici,l∖Wici,l′)≤6. On the other hand m(Vici,l∖Vici,l′)≥10 which contradicts (2.6).
To prove the first part of (i) suppose on the contrary that ∣Vi,l∣=1 and ∣Vi,l′∣=M for some l∈S, where M∈{2,3,4}. Then, by Lemma 4.3, Vi,l⊑Wi,l and Vi,l′⊑Wi,l′.
Let ∣V∣=16, M=2 and d=6. By Lemma 3.2i and 3.2ii, ∣Wi,l∣≥5 and ∣Wi,l′∣≥7. Taking into account Lemma 4.1, 3.2 and (2.6), one can easily show that ∣Wi,l∣=6,∣Wi,l′∣=7, ∣Wi,s∣=1,∣Wi,s′∣=2 and ∣Vi,s∣=6,∣Vi,s′∣=7. The relations Vi,l⊑Wi,l and Vi,l′⊑Wi,l′ allow us to partially predict the structures of the codes Wi,l,Wi,l′.
Namely, if Wi,l contains six words and v˘∩w˘=∅ for v∈Vi,l and for every w∈Wi,l, then, up to isomorphism, Wi,l={wb:w∈C2}, where C2 is given in Table 1; if Wi,l contains six words but v˘∩w˘=∅ for five words w∈Wi,l, then, up to isomorphism, Wi,l={wb:w∈C1}∪{u}, where C1 is given in Table 1 and u∈Sd. Finally, by Lemma 3.2ii, we may assume that Vi,l′={l′bbbbb,l′b′abbb} and Wici,l′ is equal to the code C2 given in that lemma.
The computations show that there is only one, up to isomorphism, pair of the codes Vi,l,Vi,l′ and Wi,l,Wi,l′ with the above properties for which (2.4) is valid (we took i=1 and l=b):
[TABLE]
and
[TABLE]
[TABLE]
Since D3(W1,b∪W1,b′)=((11,1),(0,1),(0,0)), in the similar way as at the beginning of the proof of (i), we show that D3(W)=((11,1),(2,2),(0,0)).
Clearly, there is v∈V∖(V3,a∪V3,a′) such that v˘∩⋃E(W3,a∪W3,a′)=∅, otherwise W3,l⊑V3,l for l∈{a,a′}, and consequently, by Theorem 3.1 and Lemma 3.2i, ∣V3,a∣≥12 and ∣V3,a′∣≥5, which is not possible.
Let U=U3,a∪U3,a′, where U3,a=W1,b′∖{b′ba′bbb} and U3,a′={b′ba′bbb}. We shall use Lemma 2.5 in which i=3 and l=a but we change the role of V and W. We have U3,l⊆W3,l for l∈{a,a′}. Observe that, w˘3c∩u˘3c=∅ for every w∈W3,a∖U3,a, where {u}=U3,a′. For this reason, if v∈V∖(V3,a∪V3,a′) is such that v˘∩w˘=∅, where w∈W3,a∪W3,a′, then w∈U. Recall that for every v∈V∖(V3,a∪V3,a′) the set v˘∩⋃E(U) is a 3-cylinder in v˘. Therefore, for every v∈V∖(V3,a∪V3,a′) such that v˘∩⋃E(U)=∅ we have vA=uA=bbbb, where A={2,4,5,6}. Moreover, UB={b′}, where B={1}. By Lemma 2.5, in which p=bbbb and q=b′, the code V or W contains a twin pair. A contradiction.
This completes the proof of the case {∣Vi,l∣,∣Vi,l′∣}={1,2} for d=6 and ∣V∣=16
If d=5, M=2 and ∣V∣=16, then, by Lemma 3.2i and 3.2iii, ∣Wi,l∣≥5 and ∣Wi,l′∣≥8. By (2.6),
must be ∣Wi,l∣≥7. Indeed, if ∣Wi,l∣≤6, then, to preserve the equality (2.6), m(Wici,l′∖Wici,l)=2 and consequently, Wici,l⊑Wici,l′. Then, by Lemma 3.2iii, ∣Wi,l′∣≥10.
Therefore, by Lemma 4.5, W=Wi,l∪Wi,l′ and then ∣W∣≥24, by Lemma 4.4. A contradiction. Thus, ∣Wi,l∣≥7, which means that ∣W∣>16, because Wi,s,Wi,s′=∅ for s∈{a,a′,b,b′}∖{l,l′}. A contradiction.
Let now M∈{3,4}, d=5,6 and ∣V∣=16. By Lemma 3.2, ∣Wi,l∣≥5 and ∣Wi,l′∣≥9.
By Lemma 4.4 and 4.5, we have ∣Wi,l∣=5 and ∣Wi,l′∣=9. Since m(Wici,l′∖Wici,l)≥4 and m(Vici,l′∖Vici,l)≤4, by (2.6), m(Wici,l′∖Wici,l)=4 which means that m(Wici,l∖Wici,l′)=0. Thus, Vici,l⊑Vici,l′. A contradiction, by Lemma 3.2i. The proof of (i) is completed for ∣V∣=16.
Let ∣V∣=15 and d=6.
If M=2, then, by (2.6), Lemma 3.2ii, 4.5 and 4.1, it is easy to see that ∣Wi,l∣=6 and ∣Wi,l′∣=7, ∣Wi,s∣=∣Wi,s′∣=1 and ∣Vi,s∣=∣Vi,s′∣=6. In the similar way as before (the case M=2,∣V∣=16,d=6) we compute all pairs of codes Wi,s,Wi,s′ and Vi,s,Vi,s′ for which the last two equalities and (2.4) are valid. The computations show that there are no codes Wi,s,Wi,s′ and Vi,s,Vi,s′ with the above properties, where additionally, by the first part of the proof, ∣(Vi,s∪Vi,s′)j,l∣≤11 for every j∈[d]∖{i} and l∈S.
If M=3, then ∣Wi,l∣≥5, ∣Wi,l′∣≥9, and if M=4, then ∣Wi,l∣≥5, ∣Wi,l′∣≥10, by Lemma 3.2. In both cases the set Wi,s or Wi,s′ is empty, which is a contradiction, by Lemma 4.5 and 4.4. Along the same lines we prove (i) for ∣V∣=15 and d=5 as well as for ∣V∣∈{13,14} and d=5,6
To prove (ii), let ∣Vi,l∣=∣Vi,l′∣=2 and suppose on the contrary that ⋃E(Wi,s∪Wi,s′)∩⋃E(Vi,l∪Vi,l′)=∅ for every s∈S∖{l,l′}. Then Vi,l⊑Wi,l and Vi,l′⊑Wi,l′.
Assume first that d=6. By Lemma 3.2ii, ∣Wi,l∣≥7 and ∣Wi,l′∣≥7.
By Lemma 4.4 and 4.5, ∣Wi,s∣=∣Wi,s′∣=1 for some s∈S∖{l,l′}.
Then, by Lemma 2.4, ∣Vi,s∣=∣Vi,s′∣=6. This, as it was showed in the proof of the part (i), is not possible.
Let now d=5. Since Vi,l⊑Wi,l and Vi,l′⊑Wi,l′, by Lemma 3.2iii, we have ∣Wi,l∣≥8 and ∣Wi,l′∣≥8 and consequently W=Wi,l∪Wi,l′. Then, by Lemma 4.4, ∣W∣≥24, a contradiction.
For the same reason, if ∣Vi,l∣≥3, ∣Vi,l′∣≥2 and Vi,l⊑Wi,l, Vi,l′⊑Wi,l′, then ∣W∣≥24.
Finally, let ∣Vi,l∣=1 and ∣Vi,l′∣≥5 and Vi,l⊑Wi,l, Vi,l′⊑Wi,l′. Then, by Lemma 3.2iii, ∣Wi,l∣≥5 and ∣Wi,l′∣≥10, and then, by Lemma 4.5, W=Wi,l∪Wi,l′ which gives ∣W∣≥24, a contradiction.
This completes the proof.
□
At the end of this section we summarize the above results in the following lemma which is a base for reductions that will be made in the final computations. Recall that if S={a1,a1′,...,ak,ak′}, Q⊂Sd and i∈[d], then Di(Q)=((∣Qi,a1∣,∣Qi,a1′∣),...,(∣Qi,ak∣,∣Qi,ak′∣)).
Lemma 4.7
Let V,W⊂Sd, where d=5,6, be two disjoint equivalent polybox codes without twin pairs, ∣V∣∈{13,...,16}, and let V be not flat. If U⊆V, CU⊆W is a cover of U and Q⊆CU, then:
(i)* If S={a,a′,b,b′,c,c′}, then n,m≤11 for every i∈[d] and every pair (n,m) in Di(Q).*
(ii)* If ∣Q∣=15 and S={a,a′,b,b′}, then for every i∈[d] and every pair (n,m) in Di(Q) we have n+m>0 and n≥1 if 2≤m≤4 (or conversely with respect to n and m).*
(iii)* If ∣Q∣=15 and S={a,a′,b,b′,c,c′}, then for every i∈[d] and every two pairs (n,m),(p,q) in Di(Q) if n+m=1, then p,q≤7.*
(vi)* If ∣Q∣=16 and S={a,a′,b,b′}, then for every i∈[d] and for every pair (n,m) in Di(Q) we have n⋅m>0 and n≥2 if 2≤m≤4 (or conversely with respect to n and m).*
(v)* If ∣Q∣=16 and S={a,a′,b,b′,c,c′}, then for every i∈[d] and every two pairs (n,m),(p,q) in Di(Q) if n=m=1, then p,q≤7.*
*Proof. *By Lemma 4.4, 4.5 and 4.6, only the third and the fifth statement needs an explanation. By Lemma 4.5, we assume that ∣V∣=16. Suppose on the contrary that there are (n,m) and (p,q) in Di(Q) for some i∈[d] such that n=1, m=0 and p≥8. Then, since ∣Q∣=15, adding a new word v to Q such that Q∪{v} is a twin pair free code, we obtain the pair (n′,m′), belonging to Di(Q∪{v}), which arises from (n,m) and (n′,m′)=(n,m) or n′=2 and m′=0 or n′=m′=1. The first two cases can be excluded by Lemma 4.5. In the third case, the distribution Di(Q∪{v}) contains the pair ((1,1),(p,q)), where p≥8, and thus q≤6. Since Q∪{v}=W, we may assume that ∣Wi,a∣=∣Wi,a′∣=1 and ∣Wi,b∣≥8 and ∣Wi,b′∣≤6. By Lemma 4.3 and 3.2i, Vi,l≥5 for l∈{a,a′}.
Suppose that ∣Vi,b∣≥2 and ∣Vi,b′∣≥2 or ∣Vi,b∣≥5 and ∣Vi,b′∣=1. Then, by Lemma 4.6ii, there are w∈Wi,a∪Wi,a′ and v∈Vi,b∪Vi,b′ such that w˘∩v˘=∅. Since the set {v˘∩w˘:w∈Wi,a∪Wi,a′} is an i-cylinder in v˘, it follows that Wici,a and Wici,a′ are vic-equivalent.
This means that the words in Wi,a∪Wi,a′ form a twin pair. A contradiction. Thus, by Lemma 4.6i, ∣Vi,b∣=∣Vi,b′∣=1. Then m(Vici,b∖Vici,b′)≤1. On the other hand m(Wici,b∖Wici,b′)≥2, which is a contradiction, by (2.6). The fifth statement is showed in the same manner.
□
5 A rigidity result
In this section we first prove a rigidity result, and next we use it to eliminate certain configurations of words in V and W (Corollary 5.3).
Lemma 5.1
Every polybox code V⊂Sd without twin pairs, where S={a,a′,b,b′} and d=4,5, having at most 10 words is rigid.
*Proof. *It follows from [8, Lemma 9] that the lemma is true for d≤3. We give the proof for d=4. (The proof for d=5 is nearly the same, however during the proof we give the differences between the case d=4 and d=5).
Assume on the contrary that there is a code W⊂Sd which is equivalent to V and disjoint with it. We now proceed as in the proof of Lemma 4.5: By Theorem 3.1, there is a twin pair, say u,w, in W. We may assume that u1=w1′ and u1c=w1c. Moreover, let K=u˘∪w˘ and F={K∩v˘:v∈V}. Recall that, by Lemma 2.1, the suit F for K does not contain a twin pair. Below we enumerate hypothetical structures of F. Let F1,l={K∩v˘:v∈V1,l}.
-
∣F∣=m,m∈{6,...,10},and∣F1,a∣=m−1,∣F1,a′∣=1
2. 2.
∣F∣=8and∣F1,a∣=2,∣F1,a′∣=2,∣F1,b∣=2,∣F1,b′∣=2
3. 3.
∣F∣=9and∣F1,a∣=2,∣F1,a′∣=2,∣F1,b∣=2,∣F1,b′∣=3
4. 4.
∣F∣=10and∣F1,a∣=2,∣F1,a′∣=3,∣F1,b∣=2,∣F1,b′∣=3
5. 5.
∣F∣=10and∣F1,a∣=2,∣F1,a′∣=2,∣F1,b∣=2,∣F1,b′∣=4
6. 6.
∣F∣=10and∣F1,a∣=2,∣F1,a′∣=2,∣F1,b∣=3,∣F1,b′∣=3
7. 7.
∣F∣=10and∣F1,a∣=2,∣F1,a′∣=2,∣F1,b∣=1,∣F1,b′∣=5
8. 8.
∣F∣=10and∣F1,a∣=5,∣F1,a′∣=5.
To decide whether the above cardinalities of F1,l, l∈S, are possible we first show that for every i∈[d] and every l∈S the sets Vi,l,Wi,l⊂Sd are nonempty. Suppose on the contrary that there is a letter l∈S, say l=b, such that Vi,b=∅ for some i∈[d]. We may assume Vi,a=∅. By Lemma 4.1, the codes Vici,a and Wici,a∪Wici,b are equivalent. The code Vici,a⊂S3 is rigid and therefore Vici,a=Wici,b, for otherwise Vi,a∩Wi,a=∅. Then, by (Co), Vici,a⊑Vici,a′. In the same way we show that Vici,a′⊑Vici,a, and consequently Vici,a and Vici,a′ are equivalent which gives ∣Vici,a∣≥12, by Theorem 3.1. A contradiction. Therefore, Vi,l=∅ for l∈S.
Now assume that Wi,b=∅ and ∣Wi,b′∣=k>0. Then, by (P), Wi,b′⊑Vi,b′.
If k≥3, then Wi,b′ contains at least two words which do not form a twin pair, and thus, by Lemma 3.2ii, ∣Vi,b′∣≥7. Then, ∣Vi,b′∣=7, ∣Vi,b∣=1 and ∣Vi,a∣=∣Vi,a′∣=1. By Lemma 4.3, Vi,l⊑Wi,l for l∈{a,a′} from where we obtain ∣Wi,l∣≥2 for l∈{a,a′} (recall that W contains a twin pair). Since m(Vici,b′∖Vici,b)=6, by (2.6), ∣Wi,b′∣=6. It is easy to check that then there are three words in Wi,b′ such that no two of them form a twin pair. Thus, by Lemma 3.2iii, ∣Vi,b′∣≥9, a contradiction.
If k=2 and the words in Wi,b′ do not form a twin pair, then, by Lemma 3.2ii and (2.6), ∣Vi,b′∣≥7 and ∣Vi,b∣≥5, a contradiction. If Wi,b′ contains a twin pair, then ∣Vi,b′∣≥6 and ∣Vi,b∣≥4, a contradiction, because Vi,a∪Vi,a′=∅.
Finally, if k=1, then, by Lemma 3.2i and (2.6), ∣Vi,b′∣≥5 and ∣Vi,b∣≥4, and then Vi,l=∅ for some l∈{a,a′} which is not possible.
We showed that Vi,l,Wi,l=∅ for every i∈[d] and l∈S.
It follows from the above that the distributions 4−8 are not possible. Indeed, since K=u˘∪w˘ and u1c=w1c (because u,w∈W form a twin pair in which u1=w1′), we have Vi,ui′=∅ for i∈{2,...,d}.
To show that the distribution 2 is impossible let U⊂V be the code that generates the partition F with the distribution given at the position 2, that is, U consists of all v∈V such that v˘∩K=∅ and ∣U1,a∣=∣U1,a′∣=2 and ∣U1,b∣=∣U1,b′∣=2. Consequently, by Lemma 2.2, the codes U1c1,l and U1c1,l′ are u1c-equivalent for l∈{a,b}, and hence they are, up to isomorphism, of the form given in Lemma 3.4ii. Thus, for l∈{a,a′,b,b′} there is k∈{0,...,d−3} such that
[TABLE]
Since the code U1c1,a∪U1c1,b is a cover of u1c, by (2.7), ∑v∈Qg(v1c,u1c)=2d−1, where Q=U1,a∪U1,b.
This, by (5.1), is not possible. A contradiction.
To show that also the rest of the distributions of F is not possible we now prove that ∣Vi,l∣≤5 for every i∈[d] and every l∈S. To do this, suppose on the contrary that there are i∈[d] and l∈S such that ∣Vi,l∣≥6. We may assume l=b. Since Vi,s=∅ for s∈{a,a′,b′}, we have m(Vici,b∖Vici,b′)≥4. For the same reason we may assume that ∣Vi,a∣=∣Vi,a′∣=1 or ∣Vi,a∣=1 and ∣Vi,a′∣=2, and then, by Lemma 4.3, Vi,s⊑Wi,s for s∈{a,a′}. Below we examine these two cases.
Case ∣Vi,a∣=∣Vi,a′∣=1. As W contains twin pairs, we have ∣Wi,s∣≥2 for s∈{a,a′}. Since m(Vici,b∖Vici,b′)≥4, it follows that, by (2.6), ∣Wi,b∣=5 and ∣Wi,b′∣=1 or ∣Wi,b∣=4 and ∣Wi,b′∣=2.
Let ∣Wi,b∣=5 and ∣Wi,b′∣=1. If there is v∈Vi,a∪Vi,a′ such that v˘∩⋃E(Wi,b∪Wi,b′)=∅, then, by Lemma 4.2, Wici,b′⊑Wici,b, and hence m(Wici,b∖Wici,b′)=4.
Since m(Wici,b∖Wici,b′)=m(Vici,b∖Vici,b′), ∣Vi,b∣≥6 and Vi,l=∅ for l∈S , we obtain Vici,b′⊑Vici,b and ∣Vi,b′∣=2, ∣Vi,b∣=6. This is not true, by Lemma 3.2ii. Thus, Wi,b⊑Vi,b and Wi,b′⊑Vi,b′, which gives, by Lemma 3.2i, ∣Vi,s∣≥5 for s∈{b,b′}. A contradiction, because Vi,l=∅ for l∈{a,a′}.
Let now ∣Wi,b∣=4 and ∣Wi,b′∣=2. Since m(Vici,b∖Vici,b′)≥4, by (2.6), m(Wici,b∖Wici,b′)=4 and then m(Vici,b∖Vici,b′)=4. Moreover, m(Wici,b′∖Wici,b)=2. Consequently, by (2.6), m(Vici,b∖Vici,b′)=6. A contradiction, by (2.6).
Case ∣Vi,a∣=1 and ∣Vi,a′∣=2. In this case we have ∣Vi,b∣=6 and ∣Vi,b′∣=1. Moreover, ∣Wi,a∣≥2 and ∣Wi,a′∣≥3, because, by Lemma 4.3, Vi,s⊑Wi,s for s∈{a,a′}. Thus, ∣Wi,b∣≤4, because Wi,b′=∅. Hence, m(Wici,b∖Wici,b′)≤4. This is not possible, because m(Vici,b∖Vici,b′)≥5.
We proved that ∣Vi,l∣≤5 for every i∈[d] and every l∈S. Therefore, the distributions ∣F1,a∣=m−1, ∣F1,a′∣=1 for m∈{7,...,10} are not possible. (Since in the same way we prove that ∣Vi,l∣≤5 for d=5, let us notice here that also the distribution ∣F1,a∣=5, ∣F1,a′∣=1 for d=5 is not possible. To explain it, note that, by Lemma 3.4i, the code U=U1,a∪U1,a′, where U1,l={v∈V1,l:v˘∩K=∅} for l∈{a,a′}, which generates the set of boxes F1,a∪F1,a′ (and thus, U1c1,a,U1c1,a′ are u1c-equivalent) has to be, up to isomorphism, of the form
[TABLE]
where we assumed that u1c=bbbb. Then, ∣U5,b∣=6, a contradiction.)
Skipping the last letter b in (5.2) we obtain the code U=U1,a∪U1,a′ which generates the partition F=F1,a∪F1,a with ∣F1,a∣=5, ∣F1,a′∣=1 for d=4. Computing all twin pair free codes V such that U⊂V, ∣V∣=10 and ∣Vi,l∣≤5 for i∈[4] and l∈S we always obtain Vi,l=∅ for some i∈[4] and some l∈S. Thus, F cannot have the distribution 1 for m=6.
If F has the distribution number 3, then there is U⊂V such that ∣U1,b∣=2,∣U1,b′∣=3 and U generates F1,b∪F1,b′ (this case is discussed for d=4,5). Since U1c1,b and U1c1,b are u1c-equivalent, by Lemma 3.4iii,
[TABLE]
or
[TABLE]
where s∈{a,a′,b} (it is assumed that u1c=b...b∈Sd−1,d=4,5). Now we compute all twin pair free codes Q such that F={K∩v˘:v∈Q}, that is, U⊂Q, ∣Q∣=9, ∣Q1,a∣=∣Q1,a′∣=2, and moreover the codes Q1c1,a∪Q1c1,b and Q1c1,a′∪Q1c1,b′ are covers of the word u1c=bbbb if d=5 and u1c=bbb if d=4 (for d=4 the codes U are obtained from (5.3) and (5.4) by skipping the letter s). The computations show that there is, up to isomorphism, one such code (we give the form for d=5; skipping the last letter in Q we get the form for d=4):
[TABLE]
but for every v∈Sd if V=Q∪{v} is a twin pair free code, then Vi,l=∅ for some i∈[d] and l∈S. A contradiction.
□
Recall that a pair of dichotomous words u,v is an i-siblings if u,v are not a twin pair but uic,vic form a twin pair.
Corollary 5.2
Let V,W⊂Sd, where S={a,a′,b,b′} and d=5,6, be disjoint equivalent polybox codes without twin pairs, and let ∣V∣∈{13,...,16}. For every i∈[d] and every l,s∈S, l∈{s,s′}, if 4≤∣Vi,l∪Vi,s∣≤10, then the code Vi,l∪Vi,s contains an i-siblings.
*Proof. *Suppose on the contrary that there are no i-siblings in Vi,l∪Vi,s for some i∈[d],l,s∈S, l∈{s,s′}. By Lemma 4.4 and 4.5, the sets Vi,l,Vi,s are non-empty. By Lemma 5.1, the code Vici,l∪Vici,s is rigid, and thus, by Lemma 4.1, Vici,l∪Vici,s=Wici,l∪Wici,s. Since V∩W=∅, we have Vici,l=Wici,s and Vici,s=Wici,l. Then, by (Co), Wici,s⊑Vici,l′ and Wici,l⊑Vici,s′. Since ∣Vi,l∪Vi,s∣≥4, by Lemma 3.2, ∣Vi,l′∪Vi,s′∣≥14, a contradiction.
□
Corollary 5.3
Let V,W⊂Sd, where S={a,a′,b,b′} and d=5,6, be disjoint equivalent polybox codes without twin pairs, ∣V∣∈{13,...,16}, and let V be not flat.
If d=6, then ∣{v∈V:v˘∩w˘=∅}∣≤10 for every w∈W, and if d=5, then ∣{v∈V:v˘∩w˘=∅}∣≤11 for every w∈W.
*Proof. *Let d=6 and Vw={v∈V:v˘∩w˘=∅}.
Suppose on the contrary that ∣Vw∣≥11 for some w∈W. We assume that ∣V∣=16 and ∣Vw∣=11. Without loss of generality we may take w=b...b. Then every word from the set V∖Vw={v1,v2,v3,v4,v5} contains the letter b′ at some position i∈[d] and there is no word with the letter b′ in the set Vw.
Now we consider the graph of siblings G=(V,E).
In what follows we shall use the notation used in the proof of Lemma 22 of [8]: An edge between vetrices vi and vj, i,j∈[5], is called internal, while an edge joining v∈V∖Vw and u∈Vw is called external. An edge between v,u∈V is of the type b′ with the colour i, if vi=b′ and ui∈{a,a′} or ui=b′ and vi∈{a,a′} for some i∈[d]. Such an edge is denoted by (v,u)b′.
Note that each edge of the type b′ has to be incident to a vertex from the set V∖Vw, as there is no word with the letter b′ in the set Vw. It is easy to check that if v∈V∖Vw has more than two letters b′, then there is no external edge which is incident to v; if v has exactly two letters b′, then there is at most one external edge incident to v. Finally, if v∈V∖Vw has one letter b′, then all external edges of the type b′ which are incident to v are of the same colour.
Recall that, by Lemma 4.4 and 4.5, Vi,l=∅ for every i∈[d] and l∈S.
We show that for every i∈[d] there is an edge of the type b′ in G with the color i. To do this, we consider two cases: ∣Vi,l∪Vi,b′∣≤3 for some l∈{a,a′} and ∣Vi,l∪Vi,b′∣≥4 for l∈{a,a′}.
Let ∣Vi,l∣=1,∣Vi,b′∣=2. If there are no i-siblings in the set Vi,l∪Vi,b′, then we show, in the same way as in the proof of Lemma 5.2, that ∣Vi,l′∣≥5 and ∣Vi,b∣≥7. The code V has sixteen words, and therefore ∣Vi,l′∣≤6. Then, by Lemma 5.2, the code Vi,l′∪Vi,b′ contains an i-siblings.
Similarly we show that the set Vi,l∪Vi,b′ or Vi,l′∪Vi,b′ has to contain an i-siblings if ∣Vi,l∣=2 and ∣Vi,b′∣=1 or ∣Vi,l∣=∣Vi,b′∣=1 for some l∈{a,a′}.
Let now ∣Vi,l∪Vi,b′∣≥4 for l∈{a,a′}. Since 1≤∣Vi,b′∣≤5 and ∣V∣=16, we have 4≤∣Vi,a∪Vi,b′∣≤10 or 4≤∣Vi,a′∪Vi,b′∣≤10. Consequently, by Lemma 5.2, there is an i-siblings in Vi,a∪Vi,b′ or in Vi,a′∪Vi,b′.
Thus, for every i∈[d] there is an edge of the type b′ in G with the colour i.
It is easy to check that for every i,j∈[5],i=j, and every v,u∈Vw,v=u, the three edges (v,vi)b′,(vi,vj)b′ and (vj,u)b′ cannot have three different colours. Hence, if every two internal edges are not incident, then there is i∈[d] such that there is no edge of the type b′ in G with the colour i (recall that V is a twin pair free code). This, as we showed, cannot happen and therefore we may assume that there are at least two incident internal edges of the type b′. Let (v1,v2)b′ and (v2,v3)b′ be these edges. Note that we may assume that v1∈{b′aaaaa,...,b′b′b′b′b′b′}, and then we can enumerate all, up to isomorphism (of polybox codes), edges (v1,v2)b′. Having this, for every (v1,v2)b′ we can compute all edges (v2,v3)b′. (For every v1∈{b′aaaaa,...,b′b′b′b′b′b′} the number of twin pair free codes {v1,v2,v3} with the above properties ranges between 36 and 84.)
Now we consider two cases:
-
Let A={v1,v2,v3} and B={v4,v5} and suppose that every two vertices u∈A and v∈B are not joined by an edge of the type b′. Since the total number of edges of the type b′ with different colours which are incident to a vertex from B is less than three, to obtain six edges of the type b′ in all six colours, we need at least four edges of the type b′ with four different colours which are incident to vertices from A. The computations show that this is impossible.
-
Now we assume that there is no edge of the type b′ between every two vertices u,v from the sets A={v1,v2,v3,v4} and B={v5}, respectively, and moreover every two vertices in A are connected by a path consisting of edges of the type b′. Now we need at least five edges of the type b′ with five different colours which are incident to vertices from A. Also in this case the computations show that it is not possible.
It follows from the above that every two vertices in the set {v1,...,v5} are connected by a path consisting of edges of the type b′.
The number of the codes {v1,v2,v3} is low, and thus it is easy to compute all such sets {v1,...,v5}, which are, recall that, twin pair free codes. Since Vi,l=∅ for i∈[d] and l∈S, for every code {v1,...,v5} and every i∈[d], there is j∈[5] such that vij=b′.
(Note that we can eliminate from the further computations all codes {v1,...,v5} for which the sum of the number of internal edges of the type b′ with different colours and the number of words in {v1,...,v5} which have less then three letters b′ each, is at most five.)
Let e(vi)⊂Vw consist of all v∈Vw that form with vi an edge of the type b′ and {v1,...,v5,v} is a twin pair free code. Every set e(vi), i∈[5], has a few words, and therefore for every {v1,...,v5} we can compute twin pair free codes {v1,...,v5}∪P, where P⊂⋃i=15e(vi). In each computed code {v1,...,v5}∪P lacks an edge of the type b′ with some colour i∈[6]. A contradiction.
Clearly, we also obtain a contradiction if ∣V∣=16 and ∣Vw∣≥12 or ∣V∣≤15 and ∣Vw∣≥11
Along the same lines we prove the lemma for d=5.
□
6 The number of letters
In this section we show that the codes V and W under consideration can be written down in the alphabet S={a,a′,b,b′}. We begin with the following lemma:
Lemma 6.1
Let V,W⊂Sd, where S={a,a′,b,b′,c,c′} and d=5,6, be disjoint equivalent polybox codes without twin pairs, ∣V∣∈{13,...,16}, and let V be not flat. Assume that there is i∈[d] such that Vi,l∪Vi,l′=∅ for every l∈{a,b,c}. Then there is l∈S such that ∣Vi,l∣=∣Vi,l′∣=2 and ∣Wi,l∣,∣Wi,l′∣≤3 (or conversely with respect to V and W) or ∣Vi,l∣=2, ∣Vi,l′∣=3 and ∣Wi,l∣=2,∣Wi,l′∣=3. Moreover, every word in Vi,l∪Vi,l′ (or in Wi,l∪Wi,l′) is covered by at most eleven words from W (or from V).
*Proof. *By Lemma 4.5, (C) and Theorem 3.1, the sets Vi,l and Wi,l are non-empty for every l∈S.
Suppose that for some l∈S, say l=a, we have ∣Vi,a∣=∣Vi,a′∣=1. Then, by Lemma 4.3 and 3.2i, ∣Wi,a∣≥5 and ∣Wi,a′∣≥5. Since Wi,l=∅ for l∈{b,b′,c,c′}, we may assume, by Lemma 4.6i, that ∣Wi,a∣=∣Wi,a′∣=5, ∣Wi,b∣=∣Wi,b′∣=2 and ∣Wi,c∣=∣Wi,c′∣=1. Then ∣Vi,c∣=∣Vi,c′∣=5 and ∣Vi,b∣=∣Vi,b′∣=2.
Since the distribution ∣Vi,l∣=1 and 2≤∣Vi,l′∣≤4, by Lemma 4.6i, is not possible, if 4≤∣Vi,l∪Vi,l′∣≤5, then ∣Vi,l∣=∣Vi,l′∣=2 or ∣Vi,l∣=2, ∣Vi,l′∣=3. Clearly, ∣Vi,l∪Vi,l′∣≤5 for at least one l∈{a,b,c}, and thus ∣Vi,l∣=∣Vi,l′∣=2 or ∣Vi,l∣=2, ∣Vi,l′∣=3. We now consider these two cases assuming that l=a and i=1.
Let d=6.
If ∣V1,a∣=∣V1,a′∣=2, then, by Lemma 4.6ii, there is w∈W∖(W1,a∪W1,a′) such that w˘∩⋃E(V1,a∪V1,a′)=∅. Thus, by Lemma 3.3 and 2.2, the codes V1c1,a,V1c1,a′ are w1c-equivalent, and consequently, by Lemma 3.4ii, up to isomorphism,
[TABLE]
Observe that, m(V1c1,l∖V1c1,l′)=43 for l∈{a,a′} and therefore, by (2.6), ∣W1,a∣=∣W1,a′∣=m for some m≥1. By the previous case (that is, ∣V1,a∣=∣V1,a′∣=1) we assume m≤4. If m=4, then ∣W1,l∣=∣W1,l′∣=2 for l∈{b,c}, by Lemma 4.6i (compare the case ∣V1,a∣=∣V1,a′∣=1 to see that it cannot be ∣W1,l∣=∣W1,l′∣=1 for some l∈{b,c}). Thus, again by (2.6), ∣V1,l∣=∣V1,l′∣ for l∈{b,c}. Since ∣V∣≤16, we have ∣V1,l∣,∣V1,l′∣≤3 for l=b or l=c. Thus, if m=4, then ∣Wi,l∣=∣Wi,l′∣=2 and ∣Vi,l∣,∣Vi,l′∣≤3 for l=b or l=c, and if m≤3, then ∣Vi,a∣=∣Vi,a′∣=2 and ∣Wi,a∣,∣Wi,a′∣≤3.
Suppose now that ∣V1,a∣=2 and ∣V1,a′∣=3. It follows from the previous case that we may assume that ∣V1,l∪V1,l′∣≥5 and ∣W1,l∪W1,l′∣≥5 for every l∈S.
Since, by Lemma 4.6i, ∣V1,l∣=2, ∣V1,l′∣=3 for at least one l∈{b,c} and ∣W1,l∣=2, ∣W1,l′∣=3 for at least two letters l∈{a,b,c}, the proof of the first part of the lemma is completed.
To prove the second part of the lemma assume first that ∣V1,a∣=∣V1,a′∣=2 and ∣W1,l∣≤3 for l∈{a,a′}. Clearly, the code V1,a∪V1,a′ is of the form (6.1). Let v=aa′abbb and u=a′babbb. We shall consider covers Cu⊂W and Cv⊂W of u and v, respectively.
Figure 5: A scheme of the realization E(VH1,a∪VH2,a′), where H={1,2,3}. In the picture only the parts KH and GH of the boxes K and G are visible. In this scheme we identify Ea with the interval [21,1] on the first and second axis and with [0,21) on the third axis; Eb is identified with the interval [41,43] on all axes.
Assume on the contrary that the word u is covered by at least twelve words from W. Note that, by (P), the box K=Ea′×(Ea′∩Eb′)×Ea×(Eb)3 (Figure 5) cannot be intersected by a box w˘ for w∈V∪W. Therefore, w2=b for every word w∈Cu⊆W. Since ∣Cu∣≥12, we obtain a contradiction, by Lemma 4.6i.
Now we assume that v is covered by a code Cv⊂W having at least twelve words. By Lemma 4.6ii, the set P={w∈W∖(W1,a∪W1,a′):w˘∩⋃E(V1,a∪V1,a′)=∅} is non-empty.
Moreover, by Lemma 3.4iv, w2=w3=b for every w∈P.
Since, ∣W1,l∣≤3 for l∈{a,a′} and P⊆Cv, we have ∣P∣≥9. By Lemma 2.5 (in which we take A={2,3},B={4,5,6},p=bb,q=bbb and, by Lemma 3.3, U1,l=V1,l for l∈{a,a′}), the codes PB1,b∪PB1,c and PB1,b′∪PB1,c′ are q-equivalent.
Let K,M be such as in the first row of Table 2, K1={a′bb,aa′a′}, M1={ba′a′,a′ab,a′a′a}, K2={ba′a′,a′ba}, M2={a′a′b,a′aa,aa′a′} and K3,M3 be such as in the last row of Table 2 (the codes K1,M1 and K2,M2 are isomorphic to K,M in the third and the fourth row of Table 2, respectively).
Since ∣P∣≥9, an inspection of the codes in Table 2 show that, up to isomorphism, there are three possibilities:
-
PB1,b∪PB1,b′=K∪M and PB1,c∪PB1,c′=K1∪M1.
-
PB1,b∪PB1,b′=K∪M and PB1,c∪PB1,c′=K2∪M2.
-
PB1,b∪PB1,b′=K3∪M3 and PB1,c∪PB1,c′=∅.
Now we compute all twin pair free covers Cv of v containing P. Observe that, by the condition ∣W1,l∣≤3 for l∈{a,a′} we have ∣Cv∣=12 in the cases 1-2; if P is such as in the third case, we have ∣Cv∣=12 or ∣Cv∣=13. By (P), every such Cv cannot contain a word w with w˘∩K=∅ or w˘∩G=∅, where G=Ea×(Ea∩Eb)×(Ea∩Eb′)×(Eb)3 (Figure 5). Easy computations (which can be made even by hand) show that every computed cover Cv with these properties is of the form C=C3,b∪C3,b′. Since Cv⊂W3,b∪W3,b′, we have ∣W3,b∪W3,b′∣≥12. But note that for every w∈P the box w˘ intersects u˘, that is, P⊂Cu for every cover Cu⊂W of u. For every P mentioned in 1-3 we compute twin pair free cover Cu such that ∣Cu∣≤11. In every such Cu there is a word w with w∈W3,b′ and w1=a′. Thus, w∈Cv for every above computed cover Cv of v. Therefore, ∣W3,b∪W3,b′∣≥13. Then, by Lemma 4.6i, ∣W3,a∣=∣W3,a′∣=1. This, by Lemma 4.7, is not possible, because P⊆W3,b and ∣P∣≥9.
Let now ∣V1,a∣=2 and ∣V1,a′∣=3 and ∣W1,l∣≤3 for l∈{a,a′}.
We shall show, by Lemma 3.4iii, that, up to isomorphism,
[TABLE]
or
[TABLE]
It follows from Lemma 4.6ii that there is a word w∈W1,l∪W1,l′, where l∈{b,c}, such that the sets Uw1,a={v∈V1,a:w˘∩v˘=∅} and Uw1,a′={v∈V1,a′:w˘∩v˘=∅} are non-empty, and by Lemma 2.2 and 3.3, ∣Uw1,a∣≥2,∣Uw1,a′∣≥2. To prove that the codes V1,a,V1,a′ have the form (6.2) or (6.3), it is enough to exclude the case ∣Uw1,a∣=∣Uw1,a′∣=2. To do this, we assume on the contrary that ∣Uw1,a∣=∣Uw1,a′∣=2 for every w∈W for which the sets Uw1,a,Uw1,a′ are non-empty. Clearly, by Lemma 2.2 and 3.4ii, up to isomorphism,
[TABLE]
Let {v}=V1,a′∖Uw1,a′. If v∈Uwˉ1,a′ for some wˉ∈W1,l∪W1,l′, wˉ=w, l∈{b,c}, then v=a′l1l2bbb (because Uwˉ1,a=Uw1,a=V1,a). It is easy to see that there are no letters l1,l2∈S such that Uw1,a′∪{v} is a twin pair free code. Therefore, v⊑W1,a′, and consequently, ∣W1,a′∣≥5, which contradicts the assumption ∣W1,a′∣≤3. Thus, V1,a,V1,a′ are of the form (6.2) or (6.3).
Let P={w∈W∖(W1,a∪W1,a′):w˘∩⋃E(V1,a∪V1,a′)=∅}. Similarly like above, P=∅ and, by Lemma 3.4iv, w2=w3=w4=b for every w∈P. Since now, by Lemma 2.5, the codes PB1,b∪PB1,c and PB1,b′∪PB1,c′ are q-equivalent, where B={5,6} and q=bb, we have ∣P∣≤4 (this means, that only one of the sets PB1,b∪PB1,b′ and PB1,c∪PB1,c′ is non-empty). Therefore, in every cover C⊆W of a word v∈V1,a∪V1,a′ there are at most four words with the first letter different form a or a′. Consequently, if ∣C∣≥12, then ∣C1,a∣≥4 or ∣C1,a′∣≥4. This contradicts the assumption ∣W1,a∣,∣W1,a′∣≤3.
If d=5 and ∣V1,a∣=∣V1,a′∣=2, where ∣W1,a∣,∣W1,a′∣≤3, then the codes V1,a,V1,a′ are obtained from (6.1) by skipping the last letter b. Since now w2=w3=b for every w∈P, where P={w∈W∖(W1,a∪W1,a′):w˘∩⋃E(V1,a∪V1,a′)=∅}, similarly like above we obtain ∣C1,a∣≥4 or ∣C1,a′∣≥4 in every cover C⊆W of a word v∈V1,a∪V1,a′ with ∣C∣≥12. A contradiction.
Skipping the last letter b in every word in (6.2) and (6.3) we obtain the codes V1,a,V1,a′ in the case ∣V1,a∣=2, ∣V1,a′∣=3 for d=5. Since now w2=w3=w4=b for every w∈P, where P={w∈W∖(W1,a∪W1,a′):w˘∩⋃E(V1,a∪V1,a′)=∅}, by Lemma 2.5 (in which we take A={2,3,4}, B={5}, p=bbb and q=b), the code V or W contains a twin pair, which is not true. The proof of the lemma is completed.
□
We are ready to prove the announced restriction on S:
Lemma 6.2
Let V,W⊂Sd, where d=5,6, be disjoint equivalent polybox codes without twin pairs, ∣V∣∈{13,...,16}, and let V be not flat. Then the codes V,W can be written down in the alphabet S={a,a′,b,b′}.
*Proof. *Suppose that the lemma is not true. We shall consider two cases: ∣S∣≥8 and ∣S∣=6.
Case ∣S∣≥8. Assume that there is i∈[d] such that Vi,l∪Vi,l′=∅ for every l∈S1={a,a′,b,b′,c,c′,d,d′}, where S1⊆S. Clearly, as we have seen before, Wi,l∪Wi,l′=∅ for every l∈S1.
We shall show that then Vi,l=∅ and Vi,l′=∅ for l∈S1.
To do this, we may assume on the contrary that Vi,a′=∅. Then, Vi,a⊑Wi,a. Note that if ∣Vi,a∣=1, then by Lemma 3.2i and (2.6), ∣Wi,a∣≥5 and ∣Wi,a′∣≥4 and if ∣Vi,a∣≥2, then, by Lemma 3.2ii and (2.6), ∣Wi,a∪Wi,a′∣≥8. Therefore, we may assume that ∣Wi,b∪Wi,b′∣≤3 and ∣Wi,c∪Wi,c′∣≤3. Now it is easy to show that ∣Vi,b∪Vi,b′∣≥8 and ∣Vi,c∪Vi,c′∣≥8. A contradiction. Thus, Vi,l=∅ and Vi,l′=∅ for l∈S1. Clearly, we have also Wi,l=∅ and Wi,l′=∅ for l∈S1.
Note that ∣Vi,l′∪Vi,l∣=4 for l∈S1. Indeed, if ∣Vi,l′∪Vi,l∣<4 for some l∈S1, then, by Lemma 4.3 and 3.2i, ∣Wi,l∣,∣Wi,l′∣≥5 and then Wi,s=∅ for some s∈S1, which cannot occur.
Consequently, ∣Vi,l′∪Vi,l∣=4 for every l∈S1, and then, again by Lemma 4.3 and 3.2i, ∣Vi,l′∣=∣Vi,l∣=2 for l∈S1. Similarly, ∣Wi,l′∣=∣Wi,l∣=2 for l∈S1.
Thus, for every l∈S1 there is w∈W with wi∈{l,l′} such that w˘∩⋃E(Vi,l∪Vi,l′)=∅ (as if it is not so for some l∈S1, then Vi,l⊑Wi,l and consequently, by Theorem 3.1, ∣Wi,l∣≥3. A contradiction.)
This means, by Lemma 2.2, that the codes Vici,l, Vici,l′ are wic-equivalent. In what follows we may assume that i=1, l=a and let d=6 (along the same lines we prove the lemma for d=5).
By Lemma 3.4ii, we may assume that the code V1,a∪V1,a′ is of the form (6.1). Then, by Lemma 3.4iv, w2=w3=b for every w∈P, where P={w∈W∖(W1,a∪W1,a′):w˘∩⋃E(V1,a∪V1,a′)=∅}. For every l∈S∖{a,a′} the codes PB1,l,PB1,l′, by Lemma 2.5, are q-equivalent for q=bbb, and then ∣PB1,l∣=∣PB1,l′∣=2, where B={4,5,6}, because ∣Wi,l∣=2 for l∈S1. In the similar way as in the middle part of the proof of Lemma 5.1 (that part where (5.1) is considered) we show that the code PB1,b∪PB1,c∪PB1,d cannot cover the word q.
Let (t4,t5,t6)∈q˘∖⋃{w˘B:w∈P}. Then, the boxes v˘H1×{t4}×{t5}×{t6} and v˘H2×{t4}×{t5}×{t6}, where {v1,v2}=V1,a and H={1,2,3}, must be covered by the boxes w˘1,w˘2, where {w1,w2}=W1,a. Clearly, it can be done only if vHi=wHi for i∈{1,2}, because ∣Wi,a∣=2. Since the structure of the code W1,a∪W1,a′ is, up to isomorphism, the same as V1,a∪V1,a′, we have wBi=r for i∈{1,2}, where r∈S3. (Obviously, q=r, because V and W are disjoint.) Note that r˘∩w˘B=∅ for every w∈P, otherwise w˘i∩w˘=∅ for some wi∈W1,a and some w∈P, which is not possible. Take (x4,x5,x6)∈w˘B∩q˘ for any fixed w∈P and, by (P), (x1,x2,x3)∈v˘H1∖⋃{w˘H:w∈W∖(W1,a∪W1,a′)}. The point x=(x1,...,x6) belongs to v˘1, because vB1=q. On the other hand x∈w˘1∪w˘2, because (x4,x5,x6)∈w˘B, where w∈P, and r˘∩w˘B=∅ for every w∈P. Clearly, x∈⋃E(W1,a′), because x1∈Ea. Moreover, by the manner of choosing of (x1,x2,x3), we have x∈⋃E(W∖(W1,a∪W1,a′)). Thus, x∈⋃E(W), a contradiction.
This proves that ∣S∣<8, that is, ∣S∣≤6.
Case ∣S∣=6. Let S={a,a′,b,b′,c,c′}, and let V1,l∪V1,l′=∅ for l∈S. Clearly, by (C) and Theorem 3.1, W1,l∪W1,l′=∅ for l∈S, and consequently, by Lemma 4.5, V1,l,W1,l=∅ for l∈S. By Lemma 6.1 we may assume that ∣V1,a∣=∣V1,a′∣=2 or ∣V1,a∣=2,∣V1,a′∣=3 and, in both cases, ∣W1,a∣,∣W1,a′∣≤3. In the proof of that lemma we showed that, up to isomorphism, the code V1,a∪V1,a′ is of the form (6.1), (6.2) or (6.3). In what follows we compute covers of V1,a∪V1,a′ by words from W. Clearly, we may assume that V1,a∪V1,a′ is of the form (6.1), (6.2) or (6.3).
Case ∣V1,a∣=∣V1,a′∣=2 and d=6. By Lemma 4.6ii and 2.5 we may assume that for every v∈V1,a∪V1,a′ every cover Cv of the word v by words from W contains a code P1,b∪P1,b′, where q-equivalent codes PB1,b and PB1,b′ for B={4,5,6} and q=bbb, are given in Table 2 at the positions 1-16 (that is, PB1,b=K,PB1,b′=M, where K,M are given in that table). Recall that, by Lemma 3.4iv, w2=w3=b for every w∈P1,b∪P1,b′. (Compare also Subsection 2.6.)
We use algorithm CoverCode in which we take U=V1,a∪V1,a′ and Pi=P1,b∪P1,b′ for every i∈[4] (recall that V1,a∪V1,a′ is of the form (6.1)). For v∈V1,a∪V1,a′ let Cv denote the set of all twin pair free covers Cv⊂Sd of v such that ∣Cv∣≤11, P1,b∪P1,b′⊂Cv and ∣Cv1,l∣≤3 for l∈{a,a′} (the last inequality steams from the condition ∣W1,a∣,∣W1,a′∣≤3). We have ∣Cv∣=9233 for v∈{aabbbb,a′babbb} and ∣Cv∣=2214 for v∈{aa′abbb,a′aa′bbb}. (There is no cover of the words aa′abbb and a′aa′bbb with eleven elements and containing PB1,b=K and PB1,b′=M, where K,M are given in the last row of Table 2.)
In the result of the computations we obtained seven non-empty covers C1,...,C7 of the code V1,a∪V1,a′ such that each of them has less than seventeen words and satisfies the condition ∣W1,a∣,∣W1,a′∣≤3.
Every code Ci, i∈[6], has thirteen words and Dk(Ci)=((0,1),(10,2),(0,0)) for some k∈[6] and every i∈[6]. In the similar way as in the proof of the second part of (i) in Lemma 4.6 we show that if W is a twin pair free disjoint equivalent code to V and Ci⊂W for i∈[6], then Dk(W)=((2,2),(10,2),(0,0)). Thus, ∣W∣=16.
The computations show that for every i∈[6] and for every three words v,u,w∈S6, if W=Ci∪{v,u,w} is a twin pair code which satisfies the condition ∣W1,a∣,∣W1,a′∣≤3 and the conclusions of Lemma 4.7 applied to Wi,l,Wi,l′ for i∈[d] and l∈{a,b}, then there are j∈[6] and l∈S such that ∣Wj,l∣=5 and ∣Wj,l′∣=1 or ∣Wj,l∣=6 and ∣Wj,l′∣=1 and in both cases Wjcj,l′⊑Wjcj,l. This, by Lemma 4.2, means that Wj,l⊑Vj,l and Wj,l′⊑Vj,l′. Then, by Lemma 3.2, ∣Vj,l∣≥10 and ∣Vj,l′∣≥5, and consequently, by Lemma 4.5, V=Vj,l∪Vj,l′. By Lemma 4.4, ∣V∣≥24, a contradiction.
The last code C7 has the form:
[TABLE]
and
[TABLE]
Now we show that every twin pair free code W such that C7⊂W and ∣W∣∈{13,...,16} cannot be equivalent to V. Suppose on the contrary that there is a twin pair free code W with C7⊂W which is equivalent to V.
Let w1=aa′b′bbb, v1=aa′abbb and w2=a′bb′bbb, v2=a′babbb. We have v1,v2∈V1,a∪V1,a′ and w1,w2∈C7⊂W. Since w3c1=v3c1 and w31=b′,v31=a, by (P), v3c1⊑V3c3,a′. For the same reason, v3c2⊑V3c3,a′. Thus, v3c1,v3c2⊑V3c3,a′. Therefore, by Lemma 3.2ii, ∣V3,a′∣≥7. Moreover, D3(C7)=((0,1),(7,2),(0,0)). Note that, by Theorem 3.1 and (P), V3,b=∅. Thus, by Lemma 4.5, the sets V3,l,W3,l are nonempty for l∈{a,a′,b,b′} and, by Lemma 4.1, ∣V3,a′∣+∣V3,b′∣+∣V3,c′∣=∣W3,a′∣+∣W3,b′∣+∣W3,c′∣ and ∣V3,a∣+∣V3,b∣+∣V3,c∣=∣W3,a∣+∣W3,b∣+∣W3,c∣. Since ∣V3,a′∣+∣V3,b′∣≥8 and ∣W3,a∣+∣W3,b∣≥8, we obtain V3,l,W3,l=∅ for l∈{c,c′}, ∣W3,b∣=∣V3,a′∣=7 and ∣W3,a∣=∣V3,b′∣=1. Hence, ∣W∣=16. We shall show that
[TABLE]
Suppose on the contrary that (∣W3,a∣,∣W3,a′∣)=(1,1). Since ∣W3,a∣=1, by Lemma 4.6i, we have ∣W3,a′∣≥5. Clearly, there is v∈V∖(V3,a∪V3,a′) such that v˘∩⋃E(W3,a∪W3,a′)=∅, otherwise W3,a′⊑V3,a′ and then, by Lemma 3.2iii, ∣V3,a′∣>7, a contradiction. By Lemma 4.2, W3c3,a⊑W3c3,a′, and then, by Lemma 2.4, V3c3,a⊑V3c3,a′. Therefore, by Lemma 3.2, ∣V3,a∣≤2. Consequently, it is easy to check that ∣W3,a′∣=6 and ∣V3,a∣=2. Thus, ∣W3,a∣=1, ∣W3,a′∣=6. We shall show that this is not possible. Since v˘∩⋃E(W3,a∪W3,a′)=∅, the codes U3c3,a, U3c3,a′ are, by Lemma 2.2, v3c-equivalent, where U3,l={w∈W3,l:v˘∩w˘=∅} for l∈{a,a′}. If ∣U3,a′∣=5, then, by Lemma 3.4i, 3.4iv and 2.5 the set P3,l={u∈V∖(V3,a∪V3,a′):u˘∩⋃E(U3,a∪U3,a′)=∅} contains two words for every l∈{b,b′}. A contradiction, because P3,b′⊂V3,b′ and ∣V3,b′∣=1. If ∣U3,a′∣=6, then, again by Lemma 3.4i, 3.4iv and 2.5, there is a twin pair in V or in W, which is not true. This completes the proof of the equalities (6.4).
Clearly, we have
[TABLE]
Moreover, m(W3c3,l∖W3c3,l′)≤1 for l∈{b,b′}, by (2.6) and (6.4).
Now we show that ∣W1,c∪W1,c′∣≥4. If it is not true, by Lemma 4.6i, ∣W1,c∣=∣W1,c′∣=1, and then D1(W)=((2,2),(5,5),(1,1)) which gives, by Lemma 4.3 and 3.2i,
D1(V)=((2,2),(1,1),(5,5)) (compare the beginning of the proof of Lemma 6.1).
Note now that w1c⊑W1c1,b, where w=b′bbbbb. Since w∈W1,b′, by (2.5), there is no twin pair free code W1,b∪W1,b′ such that ∣W1,l∣=5 for l∈{b,b′} and the equality (2.4) is valid, where ∣V1,l∣=1 for l∈{b,b′}. A contradiction, by Lemma 2.4.
Since ∣W1,c∪W1,c′∣≥4, it follows that at least two words from the set W1,c∪W1,c′ have to belong to the set W3,b′. But it is easy to compute that for every two words w,u∈Sd∖C73,b′ such that w1,u1∈{c,c′} if Q=C73,b′∪{w,u}, where recall C73,b′⊂W3,b′, is a twin pair free code, then m(W3c3,b∖Q3c)>4.
This shows that the inequality m(W3c3,b∖W3c3,b′)≤1 does not hold. Thus, (6.4) is not possible, and consequently V cannot contain a code V1,l∪V1,l′ such that ∣V1,l∣=∣V1,l′∣=2 and ∣W1,l∣,∣W1,l′∣≤3 for l∈S and d=6.
Case ∣V1,a∣=∣V1,a′∣=2 and d=5. The code V1,a∪V1,a′ is obtained from (6.1) by skipping the last letter b in every word of (6.1). By Lemma 4.6ii, 4.6iv and 2.5 we may assume that for every v∈V1,a∪V1,a′ every cover Cv of the word v by words from W contains the code
P={bbbab,bbba′a,b′bbba,b′bbaa′}. Moreover, by Lemma 6.1, ∣W1,l∣≤3 for l∈{a,a′} and ∣Cv∣≤11 for every v∈V1,a∪V1,a′. We use algorithm CoverCode in which we take U=V1,a∪V1,a′ and Pi=P for every i∈[4]. Our aim is to compute all covers CU⊆W of U with the above properties.
We have ∣Cv∣=2159 for v∈{aabbb,a′babb} and ∣Cv∣=1130 for v∈{aa′abb,a′aa′bb}.
The computations show that every such cover CU of U which satisfies the condition ∣CU1,l∣≤3 for l∈{a,a′}
has more than sixteen words. Thus, V cannot contain a code V1,l∪V1,l′ such that ∣V1,l∣=∣V1,l′∣=2 and ∣W1,l∣,∣W1,l′∣≤3 for l∈S and d=5.
Case ∣V1,a∣=2,∣V1,a′∣=3 and d=6. Recall that the code V1,a∪V1,a′ has the form (6.2) or (6.3).
For the same reasons as in the previous case we compute all twin pair free covers of V1,a∪V1,a′ such that for v∈V1,a∪V1,a′ every cover Cv of the word v by words from W contains the code P={bbbbab,bbbba′a,b′bbbba,b′bbbaa′}, ∣Cv∣≤11 and ∣W1,a∣,∣W1,a′∣≤3. We use algorithm CoverCode in which we take U=V1,a∪V1,a′ and Pi=P for every i∈[5].
We have ∣Cv∣=6799 for v∈{aabbbb}, ∣Cv∣=1946 for v∈{a′ba′a′bb,a′ababb} and ∣Cv∣=222 for v∈{aa′a′a′bb,a′aaa′bb} if V1,a∪V1,a′ is of the form (6.2). If it is of the form (6.3), then ∣Cv∣=1946 for v∈{aaabbb,aba′a′bb,a′aba′bb} and ∣Cv∣=222 for v∈{a′aaabb,a′a′a′a′bb}.
The computations show that every such cover CU of U which satisfies the condition ∣CU1,l∣≤3 for l∈{a,a′} has more than sixteen words. Thus, V cannot contain a code V1,l∪V1,l′ such that ∣V1,l∣=2,∣V1,l′∣=3 and ∣W1,l∣,∣W1,l′∣≤3 for l∈S.
Case ∣V1,a∣=2,∣V1,a′∣=3 and d=5. Skipping the last letter in (6.2) and (6.3) we obtain the set V1,a∪V1,a′ for d=5. Since, by Lemma 3.4iv, w2=w3=w4=b for every w∈W∖(W1,a∪W1,a′) such that w˘∩⋃E(V1,a∪V1,a′)=∅, by Lemma 2.5, there is a twin pair in V or in W. A contradiction.
This, together with the conclusions of the previous cases, contradicts Lemma 6.1. The proof of the lemma is completed.
□
7 The proof of Theorem 1.1
A code U⊂Sd with 2d words is called a partition code. Any realization f(U) of a partition code U is a minimal partition (compare Subsection 2.3).
The main theorem of the paper steams from the following
Theorem 7.1
There are no two disjoint equivalent polybox codes V,W⊂Sd without twin pairs for d≤6, where ∣V∣∈{13,...,16}. In particular, if U⊂Sd, d≤7, is a partition code such that there is i∈[d] for which the set Ui,l is non-empty for every l∈{a,b,c,d}⊂S, then U contains a twin pair.
*Proof. *To prove the first part of the theorem assume on the contrary that V and W are disjoint twin pair free equivalent polybox codes, where ∣V∣∈{13,...,16}.
Let d=4. We give the proof for the case ∣V∣=13. It was computed in [2] that every code Q⊆{a,a′,b,b′}4 having more than twelve words contains a twin pair. It follows from Theorem 2 of [4], that there are three words v,w and u such that U=V∪{v,w,u} is a partition code. Clearly, there are i∈[4] and l∈S such that Ui,l=Vi,l and Ui,s=∅ for s∈{a,a′,b,b′,c,c′}⊆S. Since ∣Ui,l∣≤6, the code Ui,l is rigid (see the claim in the proof of Lemma 4.5). Thus, if Ui,l′∖{v,w,u}=∅, then (Ui,l′∖{v,w,u})ic∩Uici,l=∅, and consequently there is a twin pair in V. A contradiction. If Ui,l′∖{v,w,u}=∅, then Ui,s=Vi,s and Ui,s′=Vi,s′ for some s∈S∖{l,l′}. Since Ui,s is rigid (we have ∣Ui,s∣≤6), there is a twin pair in Ui,s∪Ui,s′, a contradiction.
In the case d∈{5,6} we shall made computations for d=6 and d=5 separately. By Lemma 6.2 we may assume that V,W⊂Sd, where S={a,a′,b,b′}.
Let d=6. By Corollary 5.3, we assume that every word v∈V is covered by at most ten words from W. We may assume that the word v=bbbbbb belongs to V (compare Subsection 2.6) and let Cvn be the family of all twin pair free covers of the word v having n words for n∈{5,...,10}. The family C6(v)=⋃n=510Cvn contains 2058920 covers, and 104 of them are non-isomorphic codes. We denote the set of such codes by N6.
Now we shall proceed as follows: For every cover U∈N6 of v, based on algorithm CoverCode, we compute all twin pair free covers CU of the code U which satisfy the conclusions of Lemma 4.7, CU∩U=∅, v∈CU and ∣CU∣≤16.
In the next step we compute all twin pair free covers CCU of CU which satisfy the conclusions of Lemma 4.7, CCU∩CU=∅, U⊂CCU and ∣CCU∣≤16.
For U∈N6 by CU we denote the family of all covers CU, and CU2 stands for the family of all CCU. As as we shall see CU2=∅ for every U∈N6. This implies
that V and W cannot be equivalent. Indeed, if V and W are equivalent then we may assume that there is U∈N6 such that U⊂W (compare Subsection 2.6) and there is a cover CU⊂V of U such that the cover CCU={w∈W:w˘∩⋃E(CU)=∅} has at most sixteen words (because ∣W∣≤16), U⊂CCU (because U⊂W), CCU∩CU=∅ (because V∩W=∅) and CCU satisfies the conclusions of Lemma 4.7 (because W satisfies them). Thus, CU2=∅.
To compute CU, U∈N6, using algorithm CoverCode we need also initial configurations Pi, i∈[∣U∣] described in this algorithm. Let U={u1,...,un}. In our case Pi={v} for every ui∈U, where recall v=bbbbbb. This means that every word ui∈U is covered by codes Cui∈Cui,Pi, where Cui,Pi is the family of all codes Cui∈C6(ui) such that Pi⊂Cui.
To indicate codes Pi for the computations of a cover CCU of the code CU, CU∈CU, let CU={v1,...,vk} and Pi={u∈U:u˘∩v˘i=∅} for i∈[k]. Clearly, if Cvi⊂CCU is a cover of vi, then Pi⊂Cvi. Thus, computing CCU for every i∈[k] we use only covers from the family Cvi,Pi consisting of all codes Cvi⊂C6(vi) such that Pi⊂Cvi.
As the result of the computations we obtained CU=∅ for all, but five, codes U∈N6. These five families CU contains 42, 48, 48, 24 and 24 covers CU, and the corresponding codes U contain 5, 6, 7, 8 and 9 words, respectively. Computing in all five cases the sets CU2 we obtain each time CU2=∅. This completes the proof of the first part of the theorem for d=6.
In the same way we prove the theorem for d=5. In this case, by Corollary 5.3, it is assumed that every word V is covered by at most eleven words from W. As for d=6 we may assume that the word v=bbbbb belongs to V
The family C5(v)=⋃n=511Cvn contains 738680 covers, and 232 of them are non-isomorphic codes. The set of these non-isomorphic codes is denoted by N5.
The computations show that CU=∅ for all, but two, codes U∈N5. These two families CU contains 324 and 8 covers CU, and the corresponding codes U contain five and seven words, respectively. Computing in these two cases the sets CU2 we obtain each time CU2=∅. This completes the proof of the first part of the theorem for d=5.
To prove the second part of the theorem, assume that there is l∈{a,b,c,d} such that ∣Ui,l∣≤12. Then, as we showed in Theorem 29 of [8], there is a twin pair in U. Therefore, we may assume that ∣Ui,l∣≥13 for every l∈{a,b,c,d}. Then there is l∈{a,b,c,d} such that ∣Ui,l∣∈{13,...,16}. The codes Uici,l and Uici,l′ are equivalent, and hence, by the first part of the theorem, Uici,l∩Uici,l′=∅ or there is a twin pair in Uici,l or in Uici,l′. In both case the set Ui,l∪Ui,l′ contains a twin pair.
□
We are ready to prove the main theorem of the paper.
Proof of Theorem 1.1. Since r+(T)=4, it follows that there are x∈R7 and i∈[7] such that there are sets A1,...,A4⊂[xi,1+xi] with Aj∈{Ak,(Ak)c} for k,j∈[4],k=j, where (Ak)c=[xi,1+xi]∖Ak and Fx=Fxi,A1∪Fxi,(A1)c∪⋯∪Fxi,A4∪Fxi,(A4)c (see Subsection 2.2).
Thus, we may write down a partition code U, whose realization is the minimal partition Fx in the alphabet S={a,a′,b,b′,c,c′,d,d′}, where Ui,l=∅ for above mentioned i∈[7] and every l∈S. (A way of receiving of such code from a minimal partition is given at the end of Subsection 2.7 of [8]). From Theorem 7.1 we infer that U contains a twin pair. Then Fx contains such a pair, and consequently there is a twin pair in [0,1)7+T.
□
Theorem 1.1 together with Theorem 1.1 of [7] and [8] show that Keller’s conjecture is true for cube tilings [0,1)7+T with r+(T)≥4. On the other hand the result of Debroni et al. ([3]) shows that it is true for cube tilings [0,1)7+T with r−(T)≤2 ([7, Introduction]). Thus, we obtain
Corollary 7.2
If [0,1)7+T is a counterexample to Keller’s conjecture, then r+(T)=3.
□
Theorem 7.1 provides also the following
Proof of Theorem 1.2. It is straightforward: If V is a clique with ∣V∣=2d, then V is a partition codes without twin pairs. This, by Theorem 7.1, is impossible. Thus, ∣V∣<2d.
□
Proof of Corollary 1.3. If [0,1)7+T is a counterexample, then by Corollary 7.2, there is x∈R7 such that a partition code U of the minimal partition Fx can be written down in the alphabet S={a,a′,b,b′,c,c′}. Clearly, V=U is a clique in the Keller graph on {a,a′,b,b′,c,c′}7 with ∣V∣=128, as it is a partition code. Since r+(T)=3, there is i∈[7] such that Vi,a=∅, Vi,b=∅ and Vi,c=∅. If 0<∣Vj,l∣≤16 for some j∈[7] and some l∈S, then the similar arguments as those used at the end of the proof of Theorem 7.1 show that V contains a twin pair. Then the tiling [0,1)7+T contains a twin pair, a contradiction.
If V is a clique with 128 elements, then for i∈[7] let fi(a)=[0,1)+2Z,fi(a′)=[1,2)+2Z, fi(b)=[31,34)+2Z,fi(b′)=[34,37)+2Z and fi(c)=[21,23)+2Z,fi(b′)=[23,25)+2Z. The realization f(V) of V (see Subsection 2.3) is a twin pair free minimal partition of the 7-box R7 (see Subsection 2.1). It is easy to see that at the same time it can be viewed as a 2-periodic cube tiling of R7 without tiwn pair.
□
Finally, we can give a new proof of Keller’s conjecture in dimensions d≤6 (recall that it was originally proved by Perron in 1940; another proof is given in [7]).
Theorem 7.3
In every cube tiling [0,1)d+T of Rd, where d≤6, there is a twin pair.
*Proof. *It is enough to show that in every partition code U⊆Sd, d≤6, there is a twin pair. It is obvious for a partition code U such that U=Ui,li∪Ui,li′ for every i∈[d], where li∈S for i∈[d] (as it is isomorphic to the binary code {0,1}d). Let U=Ui,l∪Ui,l′ for some i∈[d] and every l∈S. Then there is s∈S such that 0<∣Ui,s∣≤16. This, as we have seen in the proof of Theorem 7.1, implies that U contains a twin pair.
□