Tiling by bars
by Martin Tassy
B.Sc., ENS Cachan 2007 M.A., Universit´ePierre et Marie Curie 2008
Adissertationsubmittedinpartialfulfillmentofthe requirements for the degree of Doctor of Philosophy in Mathematics at Brown University
PROVIDENCE, RHODE ISLAND
May 2014 c Copyright 2014 by Martin Tassy This dissertation by Martin Tassy is accepted in its present form by Mathematics as satisfying the dissertation requirement for the degree of Doctor of Philosophy.
Date
Richard Kenyon, Ph.D., Advisor
Recommended to the Graduate Council
Date
Je↵rey Brock, Ph.D., Reader
Date
Richard Schwartz, Ph.D., Reader
Approved by the Graduate Council
Date
Peter Weber, Dean of the Graduate School
iii Vitae
Martin Tassy was born in Marseille, France. After finishing high-school at Lyc´ee de Provence, he entered classes pr´eparatoires successively at Lyce Thiers and Lyc´ee Sainte-Genevieve. In 2006, he was admitted at Ecole´ normal sup´erieure de Cachan. During his four years as a ´etudiant normalien, he received a Bachelor of Science in Mathematics, a Master of Mathematics in Probabilities from Universit´eRennes 1, a Master in Financial Mathematics from Universit´eParis 6 in 2008, and the french Aggr´egation of Mathematics. After spending one year as an academic guest at ETH Zurich to work on Random Interlacements, he entered the Brown University graduate school in 2010.
iv Acknowledgements
This work is a PhD thesis supervised by Richard Kenyon whom I thank for his tremendous assistance. I am extremely grateful to him for introducing me to the subject of tilings and supporting my educational endeavors, as well as for the numer- ous hours he spent patiently discussing and reviewing my ideas despite my challenges with communicating them. Many thanks to Scott She eld for his intellectual advice and whose own thesis on random surfaces has been greatly inspirational to my work. My appreciation to Tim Austin who introduced me to cocycles and for his guidance on questions related to ergodic theory. To the members of my thesis jury, Je↵Brock and Richard Schwartz, and to the sta↵ of the Brown Mathematics department, thank you for your time and consideration. AParticularthankstoDoreenPappas,whoalwaysdidherutmosttoassistmedur- ing the four years of this program. Thanks to Emanuel Zgraggen and Philippe Eichmann who patiently assisted me when I needed help with programming related problems. Thanks to Sara Maloni, Tonya Spano and Pellumb Kelmendi who kindly reviewed parts of this work. Finally, the deepest thanks to my family, especially to my parents, for their infinite love and support.
v Abstract of “ Tiling by bars” by Martin Tassy, Ph.D., Brown University, May 2014
We study combinatorial and probabilistic properties of tilings of the plane by m 1 ⇥ horizontal rectangles and n 1verticalrectanglesalsocalledtilingsbybars. ⇥
In Chapter 2 We give a new criterion for the tilability of regions based on the height function in the Conway group. As a consequence we are able to prove that the tilability or regions with bounded number of holes and such that the boundary of those holes describes a trivial word in the Conway group is decidable in polynomial time. In the second part of Chapter 2 we generalize results on local moves connectiv- ity known for simply connected regions by giving a criterion to decide under which conditions tilings of a torus are connected by local moves.
In Chapter 3 we discuss the di↵erences between the space of ergodic Gibbs mea- sures for tilings by dominos and the space of ergodic Gibbs measures for tilings by longer bars. Using notions from ergodic theory we prove that when a measure µ on tiling of the plane is ergodic then the image of the height function must µ -almost surely stay close to a single geodesic. We also propose a characterization of those ergodic Gibbs measures based on the generalization of the domino slope and on the support of the height function. Contents
Vitae ii
Acknowledgments iii
1 Introduction 1 1.1 Tilingsandheightfunctions ...... 3 1.2 Translation invariant Gibbs measures on tilings ...... 8
2 Tiling of the plane 12 2.1 Definition of tilings of the plane ...... 13 2.1.1 Tilings and their Conway tiling group ...... 14 2.1.2 Tilings by bars ...... 16 2.2 Local flips on tiling by rectangles ...... 34 2.2.1 Tilings of simply connected regions in Z2 ...... 34 2.2.2 A criterion for local move connectivity of tori tilings ...... 38
3 Translation invariant measures on tilings by bars 43 3.1 Gibbs measures on tiling by bars ...... 44 3.1.1 Definition and basic properties ...... 44 3.1.2 Translation invariant Gibbs measures on tilings ...... 46 3.1.3 Ergodic Gibbs measure ...... 49 3.1.4 Support of the height function on ergodic Gibbs measures . . 51 3.2 Surface tension ...... 56 3.2.1 definition of the surface tension relative to a specific word . . 56 3.2.2 Minimizers of the entropy exist and are Gibbs measures . . . . 62 3.3 Construction of invariant measures with a given slope ...... 64
4 Conclusion 70 4.1 Existence and characterization of the ergodic Gibbs measures . . . . . 71 4.2 Variational principle and unicity of ergodics Gibbs measures . . . . . 73
v A Construction of the specific free energy 76 A.1 Entropyofamesure ...... 77 A.2 Constructionofthespecificfreeenergy ...... 80
B Notations 83
vi List of Figures
1.0.1 An aztec diamond randomly tiled by 3 1rectangles...... 3 ⇥ 1.1.1 The Cayley graph of Z3 Z3 ...... 5 1.2.1 Random tilings of a torus⇤ by 2 1and3 1rectangles...... 9 ⇥ ⇥ 2.1.1 Local picture of the set S! ...... 27 1 3.3.1 Torus with null vertical slope ...... 66 3.3.2 Torus with Extremal slope ...... 66 3.3.3 Combined configuration ...... 67 3.3.4 Combined configuration after shu✏ing ...... 68
vii Chapter One
Introduction 2
Tilings as mathematical objects have been the subject of a countless literature over the last fifty years. In certain specific cases, such has tilings of the square lattice by dominos or tilings of the triangular lattice by hexagons, their properties are fairly well understood, and questions concerning their limiting behaviors or estimates of the number of tilings of a given region can be answered accurately (see amongst others [16],[4]). In other cases even the question of tilability of a region have been proven to be NP complete (see [18]).
For most of the models for which precise results have been obtained, a tiling can be seen as the projection onto the plane of a 3-dimensional surface. Hence there exists a natural function which associates to every vertex of the tiling a number corresponding to the height of this surface; this is the so called height function. In [6] Conway introduced a generalized notion of height function for which the image space is not necessarily Z but could be a specific group associated to the tiling. When this group has a complicated structure a few results have been proven but they are still far from reaching the precision of the results obtained for height functions in Z.
This thesis an attempt to understand better those generalized height functions by studying the specific case of tiling by bars. These tilings have proven to be simple to study because even though the image group of the height function is complex, there exist a Glauber dynamic (see [13]) which allows to randomize tilings of finite regions R and study their limiting properties (see Figure 1.0.1). 3
Figure 1.0.1: An aztec diamond randomly tiled by 3 1 rectangles ⇥ 1.1 Tilings and height functions
AprototileoftheplanelatticeZ2 is a connected shape which covers exactly a finite number of squares of the lattice. A tiling of a region R of the plane Z2 by a finite set of prototiles t ,...,t up to translation is a collection of prototiles such that { 1 k} each square of R is covered by one and only one tile. In [6] Conway and Lagarias introduced a useful tool to describe tilings of a given region. The idea was to associate awordinthefreegroupF = a, b to every path contained in the tiling. The letter a,b h i 1 a would represent an horizontal step to the right, a an horizontal step to the left, b
1 averticalstepupandb averticalstepdown.Ifwedenote!i’s the words described be the boundaries of the prototiles t ,...,t we can define the Conway group of the { 1 k} tiling to be G = a, b ! = ... = ! = e , h | 1 n i 4
It is easy to see that two paths contained in the tiling with same starting and ending points must have the same projection into G. Hence this construction provides a necessary condition for the tilability of simply connected region of the plane since such regions should have a boundary path projecting into the identity word in G. This result is useful in itself but it appears that for certain types of tilings the height function contains much more information. In fact for some special sets of prototiles, tilings of a region R can be identified with the set of k-Lipschitz functions (with certain properties) from R to the Conway Group. The first result of this kind was obtained by J.C Fournier in [7] for the domino model. In this article Fournier showed that a simply connected region R would be tilable by dominos if and only if we had for all x,y in the boundary of R:
d(h(x),h(y)) 2 x y +1. || ||1
However the domino model has a very special place in the field of tilings due to its correspondence with the dimer model and the simple structure of its Conway group which can be embedded to Z.ThepurposeofChapter2istoproveageneralization of those results for tilings by bars and provide a method that we believe could be apply to other type of tilings.
In the case of tiling by m 1horizontalbarsandn 1verticalbars,R.Kenyon ⇥ ⇥ and C.Kenyon proved in [13] that the Conway group can be naturally projected into
Zm Zn.WecanalsoassociateanaturaldistancedZm Zn to this projection which is ⇤ ⇤ minimal number of letters used to write an element of Zm Zn. ⇤
This projection is maximally informative in the sense that it contains all the information about the tiling. More precisely Remila showed in [22] that such tilings of a region R could be identified with functions in Zm Zn such that: ⇤ 5
Figure 1.1.1: The Cayley graph of Z3 Z3 ⇤
h(x0)=eZm Zn where x0 is the lower left corner of the region R. • ⇤
For every pair of points x =(x ,x )andy =(y ,y )inR,ifwewrite: • 1 2 1 2
1 i j h(x) h(y)= a k b k , k 0 Y
we have ik =(x1 y1)modm and jk =(x2 y2)modn. If this condition k 0 k 0 is verifiedP we will say that x and y satisfyP the modulus condition.
For each pair (x, y)ofneighborverticesinRwehave: •
dZm Zn (h(x),h(y)) 3 ⇤
For every neighbor vertices x, y in the boundary R¯ of the region: •
dZm Zn (h(x),h(y)) = 1 ⇤
Using this characterization of tilings, we show that tilability of a region can be reduced to a set of conditions on the height of the points in the boundary R¯.Infact 6 we give a more general statement which get rids of the simply connected assumption. This is the main theorem of this chapter.
Theorem. Let x ,...,x be a set of points in the plane together with a set of heights { 1 n} 2 g1,...,gn in Zm Zn. There exist a tiling of Z with height function h such that { } ⇤ h(x )=g ,...,h(x )=g if and only if g ,...,g respects the modulus conditions { 1 1 n n} { 1 n} and for any i, j n we have:
x1 x1 / (m ↵ (i, j)) , ↵ (i, j) j i 2 k k k 0 k 0 ! X X or x2 x2 / (m (i, j)) , (i, j) j i 2 k k k 0 k 0 ! X X ↵k(i,j) k(i,j) 1 where a b is the only way to write gi gj in Zm Zn such that for every k 0 ⇤ triplet k,Q i, j we have 1 ↵ (i, j) m 1 and 1 (i, j) n 1. k k
The proof of this theorem relies on the construction of extremal tilings associated to a starting point and an asymptotic direction given by boundary point of the 1 group Zm Zn. These extremal tilings possess two fundamental properties: ⇤
1. They push the height function as fast as possible towards the boundary point in any given direction of the plane 1
2. The height of any point is within distance 2 of a half-geodesic of the group
which join the trivial word eZm Zn to . ⇤ 1
We show that by intersecting such tilings with two di↵erent starting points x and y it is possible to construct a new tiling if and only if x and y verifies the conditions Theorem. A generalization of this argument to n point allows us to obtain our result. 7
One of the advantages of this theorem over the Fournier version for dominos is that it does not make use of the notion of boundary points. This allows us to construct the same criterion for tilability independently of the number of holes in the region R.Themostnoteworthyconsequenceofthisconstructionconcernsthe complexity of deciding tilability of non simply connected regions.
Corollary 1. The tilability of a region with a bounded number of tilable holes is decidable in polynomial time.
The second section of this chapter focuses on connectivity of tilings by local moves. It has been proven in [13] that any two tilings of a simply connected region can be connected by a sequence of local moves which consist in turning mn-vertical bars in nm-horizontal bars and vice versa. In the specific case of tiling by dominos, Saldanaha, Omei, Casarin Jr. and Romualdo proved in [24] that the same result holds on a torus provided that height di↵erence on the vertical and horizontal sections of the the torus for both tilings are the same. We prove the the analogous result in the
h v case of tiling by bars. Namely if we denote by !T and !T the words described by the vertical and horizontal paths around the torus for the tiling T ,wehave:
Theorem. For any couple of tilings T and T 0 of a torus such that there exist a
h h v v possible local move on T or T 0. If we have ! = ! and ! = ! then T and T 0 T T 0 T 0 T 0 are connected by local moves.
The main additional di culty compared to the simply connected case is that the height function is no longer well-defined. However, given a starting point with trivial height and under the conditions of this theorem it is still possible to define
1 the di↵erence of the height functions hT (x)hT 0 (x)foranypointx in the torus. Moreover this height function di↵erence remains the same if two points are in the same component of the clusters where the tilings agree. Using this observation and 8 the construction of maximal paths, which replaces global extremum for the height function, we are able to show that there is a necessarily a sequence of local moves that connect the two tilings.
1.2 Translation invariant Gibbs measures on tilings
Gibbs measures on tilings of the plane by a set of prototiles t ,...,t can be seen { 1 k} as a natural generalization of uniform measures on tilings of finite regions. In fact, given ⌦the space of tilings of Z2,wesaythatameasureµ is a Gibbs measure if for every finite connected region ⇤ Z2 we have: 2
Eµ [ ⇤c ]= ⇤c ( ) , ⇧|F ⇧
where ⇤c is the uniform measure on tilings of ⇤conditioned on the boundary tiles.
Since Gibbs measures are often studied as limiting measures of microscopical systems, it is natural to be interested in the set of translation invariant Gibbs mea- sures. This is a convex set whose extremal points are the ergodic Gibbs measures. Thus, if we are able to characterize completely what are the ergodic Gibbs measures are for a given type of tiling, we will have a full understanding of the translation invariant Gibbs measures on these tilings. One of the goals of this thesis is precisely to understand the set of ergodic Gibbs measures could be for tiling by bars.
In the specific case of domino tilings, Gibbs measures have been extensively studied and translation invariant Gibbs measures are now well understood. It has been proven by Kenyon, Okounkov and She eld in [16] and [26] that those ergodic Gibbs measures can be characterized uniquely by the expected variation of a one- 9 dimensional height function. In order to extend these results to the case of Bars, a natural approach would be to use multidimensional abelian height functions obtained by assigning a weight to edges depending on the type of squares they separate in a m n chessboard coloring of the plane as described in [25]. However, this approach ⇥ does not work as I am going to explain below.
Since ergodic Gibbs measures satisfy a zero-one law on the sigma-algebra of translation invariant events, a natural way to grasp a first intuition of the set of ergodic Gibbs measure is to construct translation invariant measures and look for the macroscopic characteristics of the picture. For tiling by bars there exists the Glauber dynamics introduced in Chapter 1 which randomize tilings of finite regions. We used this dynamics in order to simulate random tilings of tori such as the ones in Figure 1.2.1 and obtain translation invariant measures.
Figure 1.2.1: Random tilings of a torus by 2 1 and 3 1 rectangles ⇥ ⇥
The two pictures in Figure 1.2.1 have very similar appearance at first glance. However, upon closer examination, we see that for 3 1tilingswehavemultiples ⇥ choices in the pattern of the frozen red region, which is not the case for the domino tiling. Moreover, the pattern cannot be described by a commutative height function as it was expected since this pattern does not depend solely on the numbers of edges of each types described by the border of the dominos but also by the order in 10 which they appear. This means that, for max(m, n) 3, characterizing translation invariant Gibbs measures through a commutative height function cannot be su cient to describe the set of ergodic Gibbs measures entirely. This is why we need the full height function on Zm Zn. ⇤
As mentioned previously, tilings of the plane by bars can be seen as Lipschitz
2 functions from Z to the hyperbolic group Zm Zn.Suchobjectshavebeenstudied ⇤ by people working in ergodic theory.
Given G y X be an action of a group G on a set ⌦and given another group L ,a cocycle for the action of G on L is a map a : G X L such that for all , ⇥ ! 1 2 2 and all ! ⌦wehave 2
↵ ( 1 2,x)=↵ ( 2,x) ↵ ( 1, 2x) .
If there exists a measure µ on ⌦such that the action of on µ is ergodic we say that ↵ is an ergodic cocycle.Fromthisdefinitionweobtainthatforanyergodic Gibbs measure on tilings on the plane, the set of height functions define an ergodic cocycle from Z2 to the Conway group of the tiling. When this group is hyperbolic, as
Zm Zn,thisimposessomestrongrestrictionsontheimagesetoftheheightfunction ⇤ and we have the following result:
Theorem. Let µ be an ergodic Gibbs measure on Z2. Then for µ-almost every tiling
2 T there exists a geodesic ray ! such that for any vector v R we have: 2
1 lim dZm Zn hT (x kv ), ! =0, k ⇤ b c !1 k 2 where h x k ,! is the height of the closest point of k in Z and dZm Zn is the b c b c ⇤ distance in the Cayley graph of Zm Zn. ⇤ 11
The proof uses the same kind of technique as the Karlsson-Ledrappier theorem in [11]. Applying the subadditive ergodic theorem, we can define, for any ergodic measure µ,asemi-norminR2 which encodes the asymptotic speed at which the height function is travelling on the Cayley graph of Zm Zn in any given direction. ⇤ The translation invariance of the measure imposes that the set X of the zeros of this semi-norm can only be an hyperplane. In particular the complement Xc of this set has exactly two connected components. Now, if we denote v,! the half- geodesic which supports the height function in the direction v,onecanprovethat the function v is almost surely continuous on Xc for the discrete topology, and 7! v,! thus constant on each component of Xc.Thismeansthat,foranygivenconfiguration !,theimageoftheheightfunctionissupportedonexactlytwohalfgeodesicswhich together form a single geodesic of Zm Zn. Another consequence of this theorem ⇤ is that the zeros of the height function have a preferred direction in Zm Zn which ⇤ corresponds to the direction of the white paths in Figure 1.2.1.
Morever the construction of this semi-norm tells us that the limits
1 sh =lim dZm Zn ((0,n), 1Zm Zn )) n ⇤ ⇤ !1 n and 1 sv =lim dZm Zn ((n, 0), 1Zm Zn )) n ⇤ ⇤ !1 n both exists and are µ-a.s surely constants. Those limits should be seen as the anal- ogous for tiling by bars of the slope in the domino model.
It is now natural to attempt to construct ergodic Gibbs measures by fixing a periodic geodesic and an asymptotic vertical and horizontal speed. In the last section of Chapter 3 we describe a qualitative construction of those Ergodic Gibbs measures. Chapter Two
Tiling of the plane 13
In this chapter we will be interested in tilings of the plane and their height functions as introduced by Conway and Lagarias. The main result we want to emphasize is that for certain class of tilings, the height function contains somehow all the information on the tilings. This means that height functions do not only provide a necessary condition on the tilability of a region as described in [6] but one can also obtain a su cient condition for tilability based on the distance of the height of the boundary points in the Conway group. We will cite several consequences of this result for the tiaibility of subsets of the plane.
The second section of this chapter deals with local move connectivity for regions of the planes. We extend the result first obtain my R. Kenyon and C.Kenyon in [13] to the case of tilings of a torus.
2.1 Definition of tilings of the plane
Throughout the rest of this thesis we will always identify the plane with Z2.A prototile t of the plane is a simply connected figure in the square lattice of Z2.
2 For any region R in Z and any finite set of prototiles t1,...,tk ,atilingofR by { } t ,...,t is a collection of prototiles such that each square of R is covered by one { 1 k} and only one tile. If there exist such a tiling of R by t ,...,t ,wesaythatthe { 1 k} region R is tileable by t ,...,t . A tiling problem consist of a region R and a finite { 1 k} set of tiles. 14
2.1.1 Tilings and their Conway tiling group
1 Let t ,...,t be a set of tiles. Let a be an horizontal step on the right and a be { 1 k} 1 an horizontal step on the left. Equivalently, let b be a vertical step up and b be a vertical step down.
Definition 2. afinitesequencesofverticesp =(v ,...,v )isapathifforeveryi n 0 n we have v v 1Ifp is a path we can define a path word ! (p)=u ...u k i i+1k1 0 n 2 1 1 F = a, b ,suchthatoreveryi n such that u ⌃= a, a ,b,b represents a,b h i i 2 { } the step to go from vi to vi+1.
To every tile ti and any arbitrary chosen starting point on the tile, we can asso- ciate a word ! in the free group F = a, b corresponding to the path described by i a,b h i the perimeter of the tile starting at this point. In [6] Conway and Lagarias introduced agroupassociatedeverysetofprototiles.
Definition 3. The Conway group of the tiling t ,...,t is the quotient group { 1 k}
G = a, b ! = ... = ! = e . h | 1 n i
We will also use the notation CG to mention the Cayley graph of G
The first remark me can make is that the definition of the group G does not depend on the starting point we choose for each !i. Indeed, two tiles words which describe the same prototile boundary but with di↵erent starting points are conjugate with each other. Thus one has a trivial representation in G if and only if the other one has also a trivial representation.
The first property of this group is that it provides a necessary condition for 15 tilability of a simply connected region R.
Lemma 4. Let R be a connected region with boundary R¯, if R is tilable then we have projG(!R¯ =1G) with !R¯ being the word in Fa,b associated to the boundary path and projG being the projection to the Conway group G
Proof. The proof is realized by induction on the number n of tiles of a tiling T of R.Thecasewheren =1followsfromthedefinitionofG and by definition one does not change the the boundary word of a given tiling by removing a tile on the boundary.
Now given a finite connected region R,thelowerleftcornerx0 of R and a tiling T of this region, we can associate to T the function f : R G in the following R R TR,x0 ! way:
f (x )=e where e is the trivial word in G. • 0
If x, y is a horizontal right edge (resp. vertical down edge) in T ,wehave • { } R f (y)=f (x) a (resp. f (y)=f (x) b).
When there is no ambiguity in the region R and in the starting point we will use
fT for fTR,x0 .
This tells us that the function fTR is well defined since two connected paths with the same starting and ending points go around a tiling.
Remark. The tiling function is defined for simply connected regions but can natu- rally be extended to regions with holes such that each hole is tilable. 16
f (x)=baaba
It is important to notice that this is just a necessary condition and in general even if the projection of any closed path to G is trivial, this does not necessarily mean that any simple connected region is tilable. For example if we try to tile the figure below by 3 1rectangles,itisstraightforwardtocheckthattheboundarywordis ⇥ trivial in the Conway group but the region is not tilable. This happens because one can add vertical tiles on the corner of the region R to obtain a tilable 3 4rectangle ⇥ without changing the boundary word. In fact the triviality of the boundary word means that there exist a finite sequence of move of adding and removing tiles which ends in the empty set.
2.1.2 Tilings by bars
We will now be specifically interested in the case of tilings by m 1-rectanglesand ⇥ a n 1-rectangleswhichwewillcalltilingbybarsfromnowon.Thiscasehasbeen ⇥ primarily studied by C. Kenyon and R. Kenyon in [13] and subsequently by Remila 17
3 2 1 1 2 p (f ¯)=bab a b a b 3a =1 G R G in [1] and [22]. In the rest of this section we will mention several results from those papers.
m m 1 n n 1 In the case of tiling by bars we have !1 = a ba b and !2 = b ab a ,and thus the corresponding Conway group becomes
m m 1 n n 1 G = a, b a ba b = b ab a = e . h | i
Due to its complicated structure, this groups is hard to work with. In order to simplify the problem we can look for projection into groups with structure which can be more easily manipulated. One option could consist in using an abelian height function as suggested in She eld [25]. An e cient mean to construct such a height function is to use 3-color chessboard coloring of Z2 and count how many times a path is winding left or right of squares with a given color. However the rest of this chapter shows that such a height function cannot capture the full complexity of the space of tiling and this is why we will use a more complicated non-abelian height function. 18
We consider the proposition introduced by Kenyon and Kenyon in [13] which consist in projecting this group in
m n Zm Zn = a, b a = b = e . ⇤ h | i
Even though this group is not commutative, Its structure is much easier to under- stand than the one of the original Conway group and its Cayley graph is very similar to a tree as shown in the picture below.
3 3 H = a, b a = b = e = Z3 Z3 | ⇤ ⌦ ↵ One can notice from Remila [22] that this is not the only relevant projection however we will prove that it is a maximally informative one in a certain way.
An important property of tilings by bar is that they are “local tilings” in the sense that there exist a set of local conditions to verify on the height functions which are su cient to assure that the function describe a tiling of a region R.Inother words if the height function satisfies those conditions everywhere then the global height function represents a unique tiling. This has been proved by Remila in [22] 19 but for the sake of completeness we will recall and prove this theorem.
Theorem 5. For any g1,g2 in Zm Zn, let dZm Zn (g1,g2) be the minimal number of ⇤ ⇤ 1 letter used to write the word g1 g2 in Zm Zn. Let R be a simply connected region ⇤ of the plane, x0 be its lower left corner and h be a function from R to Zm Zn = ⇤ a, b am = bn = e such that we have: h | i
h(x0)=1Zm Zn • ⇤
For every pair of points x =(x ,x ) and y =(y ,y ) in R, if we write: • 1 2 1 2
1 i j h(x) h(y)= a k b k , k 0 Y
we have ik =(x1 y1)mod[m] and jk =(x2 y2)mod[n] If this k 0 k 0 conditionP is verifies we will say that x and yPsatisfy the modulus condition.
For each pair (x, y) of neighbors vertices in R we have: •
dZm Zn (h(x),h(y)) 3 ⇤
where dZm Zn is the distance in the Cayley graph for the set of generators a, b . ⇤ { }
For each pair (x, y) of neighbors vertices in the boundary R¯ we have: •
dZm Zn (h(x),h(y)) = 1 ⇤
Then there exist a unique partial tiling T of R such that h = hT
Proof. We begin by proving the following claim: 20
Claim: If x and y are neighboring vertices with y = x+(0, 1) (resp. y = x+(1, 0)) and dZm Zn (x, y) 3 then there exist j with 0 j n 1(resp.iwith0 i m 1) ⇤ j j i i such that h(y)=h(x)b ab (resp. h(y)=h(x)a ba ).
Moreover for every j0 such that j n j0 j (resp. i0 s.t i m i0 i)wehave
j0 h(x +(0,j0)) = h(x)b and
j0 h(y +(0,j0)) = h(y)b
i0 i0 (resp. h(x +(i0, 0)) = h(x)a and h(y +(i0, 0)) = h(y)a ).
This first part of the claim comes from the observation that we must have
i1 i2 i3 dZm Zn (x, y)=1ordZm Zn (x, y)=3.IndeedIfwewriteh(y)=h(x)b a b ⇤ ⇤ (resp. h(y)=h(x)aj1 bj2 aj3 ), the modulus condition tells us that we must have
i1 i1 i1 + i3 =0modm and i2 =0modn.Thuswecanwritethath(y)=h(x)b ab
j1 j1 (resp. h(y)=h(x)a ba )whichgivesusthefirstpartoftheproof.
We prove the second part of the claim for the case where y = x +(0, 1), the case y = x +(1, 0) follows directly by symmetry. Let x0 = x +(1, 0), y0 = y +(1, 0), u =
1 1 h(x) h(x0)andv = h(y) h(y0). By assumption we know that dZm Zn (x0,y0) 3. ⇤ 1 j j This means that we must have l(u b ab v) 3. Since the first part of the claim k k l l tells us that there exist k and l such that u = a ba and u = a ba ,Itisdirectto check that this is only possible if u = b and v = b.
This gives us the proof for j0 =1.Byrepeatingthisargumentsuccessivelyfor x0 = x +(0,j0)whilej0 j in one direction and j0 n j in the other direction, we obtain the complete statement. 21
Me now have enough elements to define the tiling T associated to the height function h.Wesayshatthatacelliscoveredbyanhorizontaltile(resp.verticaltile) if there is a vertex x on the boundary of the cell such that h(x+(1, 0)) = h(x)bjab j with 0