Torus Cube Tilings

Mathieu Dutour Sikiric

Rudjer Boskovic institute, Zagreb

Yoshiaki Itoh
Institut Statistical Mathematics, Tokyo

Alexei Poyarkov
Moscow State University, Moscow

p.1/26

I. Torus tilings and packings

p.2/26

Torus Cube Tilings

A special cube tiling is a -periodic tiling of the space by integral translates of the hypercube .

A general cube tiling is a tiling of cube .

Special cube tilings can be lifted to the torus. There is only one special cube tiling in dimension . There are two special cube tilings in dimension :

by translate of the

p.3/26

Keller conjecture
Conjecture: for any general cube tiling, there exist at least one face-to-face adjacency. This conjecture was proved by Perron (1940) for dimension . Szabo (1986): if there is a counter-example to the conjecture, then there is a counter-example, which is special. Lagarias & Shor (1992) have constructed counter-example to the Keller conjecture in dimension

Mackey (2002) has constructed a counter-example in dimension .

p.4/26

Cube packings
A cube packing is a packing of by integral translates of cubes , which is -periodic.

If we cannot extend a cube packing by adding another cube, then it is called non-extendible. Non-extendible cube packing does not exist in dimension and .

There is an unique non-extendible cube packing in dimension . We are interested in the values of , for which there is a non-extendible cube packing with cubes.

p.5/26

non-extendible cube packing

p.6/26

II. Algorithms of generation

p.7/26

Clique formalism
Associate to every cube its center Two cubes with centers and are non-overlapping if and only if there exist a coordinate , such that .

$%

( $ '% 0

($

'

&

The graph is the graph with vertex set and two vertices being adjacent if and only if the corresponding cubes do not overlap.

A clique in a graph is a set of vertices such that any two vertices in are adjacent.

Cube tilings correspond to cliques of size graph .

All problems about those cube tilings are nite, since is nite, but the number of possibilities is huge.

"  #
in the

"  #

p.8/26

GAP enumeration
The graph acts on the has a symmetry group of size elements

"  #

Cliques are associated to subsets of those elements. GAP has extremely efcient techniques (backtrack search) for checking if two subsets are equivalent under a permutation group. We set and iterate from to : For every subset in , consider all vertices, which . are adjacent to all element in Test if they are isomorphic to existing elements in and if not, insert them into .

5 64 0 7 8""

9 @( 4 5

9 @( 4 5

&

##

, which

3

(4

(4

p.9/26

Results for
In dimension , there is a single non-extendible cube tiling and there are types of cube tilings.

p.10/26

Results for
In dimension , there is a single non-extendible cube tiling and there are types of cube tilings.

p.10/26

Results for
In dimension , there is a single non-extendible cube tiling and there are types of cube tilings.

In dimension , the repartition is as follows:

CF C D E G H P Q CR CC CD CE ST R R R R R R R EP H DF R CI R R R IF CG F CH

p.10/26

Flipping algorithm
-dimensional cube tilings suggest a ipping algorithm:

This algorithm uses face-to-face adjacencies to generate new tilings. We know that this ipping algorithm will not work in dimension , since there are cube tilings with no face-to-face adjacencies. This algorithm works in dimension

and .

p.11/26

Random generation
Another possibility is to take cubes at random and add them if and only if they do not overlap, until there is no space left any-more. Practical implementation: Generate elements at random and add them if they do not overlap. After the number of failures in the random generation, reach a certain threshold, enumerate all possible cubes, which do not overlap. Then, choose element at random in this list and update the list by removing cubes, which will overlap. Finish, when there is no choice left.

p.12/26

Greedy and Metropolis algorithm

If one wants to generate non-extendible cube packings with low density, then some other strategies are possible: Greedy algorithm In the choice of a cube at random, select the one, which covers the most important part of the space. Metropolis algorithm Take an existing non-extendible cube packing with low density and then: Remove some elements chosen at random in this cube packing. Do a random packing procedure on the left holes. If the density of the obtained non-extendible cube packing is too high, then discard it; otherwise, take it as basis for future computations.

p.13/26

III. non-extendible cube packings

p.14/26

Extension of cube packings

Theorem A cube packing with is extendible to a cube tiling. cubes,

tiles

Given a cube packing with cubes of coordinates , a coordinate and a value . The induced cube packing is the cube packing of obtained by taking all vectors with and removing the -th coordinate. Such cube packings have at least tiles.

! "  #

$V

c 8b d e f

9 5

The proof is by induction, using induced cube packings.

`$ V 0Y Y a 

9 5

$V
,

p.15/26

Extension of theorem
The complement of an non-extendible cube packing is the set .

Conjecture: If is an non-extendible cube packing with tiles, then its complement is of the same shape, as the one in dimension .

hg

hg

Conjecture: If is a cube packing with then it is extendible by at least one cube.

cubes,

hg

A complement of a cube tiling is called irreducible if it is not the union of cube tilings on different layers. Conjecture: For a given , there is a nite number of irreducible complements of volume .

hg

p.16/26

Low density cube packings

Denote by the smallest number of cubes of non-extendible cube packing.

f

pb

and

f pb 0

pb

f

pb

The following inequality holds:

 qf  pb

The cube packing realizing this is constructed by and product of two cube packings of

rqf t

f

pb

pb uv f

So,

f

pb

But no random algorithm manage to nd such a packing!



f pb

p.17/26

Covering sets
A covering set is a set of cubes (possibly, overlapping), such that we cannot add a cube without overlapping with at least one cube in .

gw

Let us denote by covering sets.

xb

xb

f

if and only for every set of cubes (possibly overlapping), there exist a cube, which do not overlap with them. Theorem. One has the relation

y

xb

f

f

pb

f

the smallest number of cubes in

x xb f  )  r a a  

b

gw

p.18/26

Proof of theorem
y
Lemma. If one has satises the inequality .

xb

xb

f

Take vectors in coordinate. At least . some

"  #

and consider its last vectors satisfy to for and

In illustrated proof below, one has

! "  #

5 $

 0

f

, then

p.19/26

Proof of theorem
0 0 1 3 1 0 2 1 2 3 0 2 3 1 2 0 0 3 3 3
0 0 1 3 1 0 2 1 2 3 0 2 3 1 2 0 0 3 3 3

Eliminate those vectors and, for the remaining vector, their last coordinate

p.19/26

Proof of theorem
0 0 1 3 1 0 2 1 2 3 0 2 3 1 2 0 0 3 3 3

1 0 2 2 3 0 3 1 2 0 2 0

y
(

We have

xb

) remaining vectors; so, one can nd a not overlapped cube.

f

p.19/26

Proof of theorem
0 0 1 3 1 0 2 2 3 0 3 1 2 0 2 0
c 8b d e f

1 0 2 1 2 3 0 2 3 1 2 0 0 3 3 3 0 2 0 1
a

By the basic assumption, setting the last coordinate to gives a cube, which does not overlap with the preceding ones.

p.19/26

Set

Then it holds:

 9 5
.

So, by the Lemma,

xb a   )  r a x b f f a  f 
, i.e.:

Proof of theorem

 xb

0 de  9 5 g f f ef  x  bf )  y r f
p.19/26

xb

Values of
f xb
, while

f if 0

pb

Above theorem yields

pb

 pb

and

f

We found (by random method) many non-extendible cubes, but not with less than cubes. packings with

r  

Could it be that

if

pb

It seems also likely, that cube packing in dimension

f 

pb

and that the best is unique.

An interesting question if to estimate the asymptotic behavior of or .

xb

f

f

pb



p.20/26

IV. Second moment

p.21/26

The counting function

Take a cube packing and is the cube of corner the number of cubes inside .

hg

j j  k

i a

j 

b hg hg j  j  j 
.

i hg f

is -periodic. Denote by operator.

r lb

j 

First moment

m f m 

lb

j 

the averaging

p.22/26

Maximal second moment

We want to nd minimal values of the second moment, i.e. . One denes the space

q p

j n f

lb

$w ! "  t  # up "  #

one has and

$f

pb x {|z y 5 }m ! p o

r sq

0 o  sq

Cube tilings correspond to

We will prove that , with , is maximal for associated to a regular cube tiling.

b j  b f pf fn

lb

$f 

pb

-vectors of .

q ~

r v

p.23/26

Maximal second moment

Given , dene:

$f a pb

pb

(f

if if

or or

($ 0 ($ 0

b $

pf

pf

belongs to .

One proves .

If , then cube tiling.

rr

So, the second moment tiling.

pf

j  b ( b pf fn f  lb j  b pf fn

lb

"

for every

is the function of the regular is maximal for regular cube

r 

p.23/26

Lower bound
Theorem. If is a cube packing with cubes, then its second moment satises the inequality:

hg

8 " b

b

f

f a n 

with

and

For and , the cube packing minimizing the second moment is the non-extendible one.

r 

p.24/26

Proof of lower bound

Take a cube packing If for then every

,. . . ,

is the collection of all is contained in cubes .

. -cubes,

`

`

Denote by , for contained in .

, the number of cubes

`

`


and

`
Denote by the number of cubes and . One has: , containing

V

` ( ` 0

` ) r f

`

r` n

` 0

p.25/26

Proof of lower bound

Denote . the number of equal coordinates of

and

Then one has:

with equality for So, one gets:

( ` 0

9 5

 a 9 5 ( ` r` (

b ` ( `  ( 

f

b f   a 9 5 ( r`
or .

( ` 0

p.25/26

Proof of lower bound

Denote has: the number of equal pairs in column . One

Fix ; if is the number of vectors of value , then it holds:

V 5

rV

( ` 0

in column

b  ) f

and

 

Writing

and minimizing over

, one gets:

V  b

f

 0 r 

V 0

p.25/26

THANK YOU

p.26/26