This page is still under construction.
                                                    You will find a text only version of the paper below.
                                                               Check back again for the figures.

A Difficult Job in the Tiling Industry

Aimee Johnson
Department of Mathematics
Swarthmore College

Kathleen Madden
Department of Mathematics and Computer Science
Drew University

Suppose that you are hired to tile a huge floor, so huge, in fact, that it goes on forever in every direction. The tiles you are to use are all equal sized squares, with various colored edges. Only a finite number of colored-edged tile types are available, but you may use as many tiles of each type as you want. So far, the task seems long, but not complicated. However, you then discover that your employer has two additional rules that you must follow: 1) When placing two tiles next to each other, edge colors must match. 2) You may not rotate your tiles. For example, you cannot place a tile on its side or upside down. And of course, you must leave no gaps.
    You might become concerned that these additional requirements may have made the task impossible; certainly you can envision a situation where the tile types available cannot tile an infinite floor. For instance, suppose the set of tiles types available are illustrated below:
Figure 1
    In this case, you would never be able to place one tile on top of another without placing red next to yellow, violating Rule 1. (Remember, Rule 2 does not allow you to turn the tiles upside down.)

Question: Can you look at the tile types available and determine, before you accept the tiling job, whether the task is possible or not?

    If it is possible, we will say that you can ``tile the plane".
    It is not always so easy to decide whether a set of tile types will tile the plane. If you are hired to tile an infinite floor with the four tile types shown below, should you accept the job? (In Figure 2, numbers represent colors.)

Figure 2
    In attempting to answer this question, you might begin by trying to tile 2×2 squares. If you are successful in doing this you might tile 3×3 squares and so on. Of course, if you find a size square that you are unable to complete, you can conclude that it will be impossible to use these tile types to tile the plane. (What size of square are you unable to complete for the previous example?) But this process might never end; even if you are able to tile larger and larger squares you cannot conclude that the tile types can be used to tile the plane. After all, even if you can tile a square with sides consisting of a million tiles, it might be the square with sides consisting of a million and one tiles that is the first size that you are unable to complete.
    This issue was studied by Hao Wang [7] in 1961. Wang speculated that this process would eventually end because either you would find a size square that you could not complete or you would find a ``periodic square". This idea is known as Wang's Conjecture:

Wang's Conjecture: Any set of tiles which can be used to tile the plane, can be used to tile the plane with periodic squares.

    An n×n square of tiles is periodic if its top row of tiles is the same as its bottom row of tiles and if its left column of tiles is the same as its right column of tiles. If you can construct such a periodic square with your tile types, you know you can tile the plane; periodic squares can be stacked end to end vertically and horizontally with the matching edges of each square overlapping.
    A tiling of the plane with periodic squares is said to be a periodic tiling. You can tell that a tiling is periodic if there are least two places where you can stand from which the resulting floor pattern looks exactly the same. (In fact there are many places you can stand and see the same pattern and you can move from one spot to another by a fixed translation.)
    Below is a portion of a periodic tiling. (Note in this simple example, the edges of the two tile types can meet all other edges) Can you see the periodic squares? What translations move you between spots where the pattern looks the same?

Figure 3

    Wang conjectured that any set of tile types which could be used to tile the plane could be used to tile a periodic square. This idea seems reasonable because Wang's Conjecture is true in one dimension. That is, suppose that instead of tiling a two dimensional floor you are tiling a one dimensional strip by laying the tiles end to end. If this can be done, since there are only finitely many tile types, you must use at least one tile type twice. But this means you have a block of tiles, beginning and ending with the same tile type, which you can use over and over to yield a periodic tiling of the strip. For instance, suppose the set of tile types is shown in figure 4. Then we can use the block in figure 5 to create a periodic tiling of the infinite strip.

Figure 4 Figure 5

    If Wang's Conjecture is true in two dimensions as well, it gives you a general method for determining if a given set of tile types will tile the plane. You simply construct all tilings of 2×2 squares and then all tilings of 3×3 squares and so on. If the tiles can not be used to tile the plane, eventually you will find a square that you can not tile. If the tiles can be used to tile the plane, Wang's Conjecture says eventually you will find a periodic square. Either way, your method will terminate and allow you to decide whether your tile types can be used to tile the plane.
    Wang's Conjecture remained an open question for several years but, unfortunately for your tiling business, was shown to be false in the thesis work of one of Wang's students, Robert Berger [1]. Berger constructed an example of a set of tile types which could be used to tile the plane but for which any tiling constructed using these tiles was not periodic. Berger's original example involved a set of over twenty thousand tile types!
    So the answer to the original question is ``no"; there is no general method for you to use to determine if a certain finite set of tile types can be used to tile the plane. We recommend that you make sure you are paid for your tiling jobs in advance!
    Berger's example was published in 1966 and in 1971 Raphael Robinson [5] found a simpler example involving 28 tile types. Like Berger's tiles, Robinson's tiles can be used to tile the plane but no such tiling will be periodic, thus proving that Wang's Conjecture is false. We will discuss this example next. In the final section we will consider other familiar tilings, such as the Penrose tilings and the pinwheel tilings, and their connection with Wang's Conjecture.

Robinson's Example

Before introducing Robinson's example [5], we mention a technique that is quite common in the study of tiles and tilings. We have described tiles as squares with colored edges; these colored edges give us ``matching rules". We could equivalently use other types of markings to give us the desired matching rules; for example, we might use arrows that must match head to tail. The tiles in Figure 1 could then be replaced with the tiles shown below:

Figure 6

    Notice that the arrows allow the same horizontal matching, but because arrow heads must meet arrow tails there are no vertical matchings.
    The 28 tile types in Robinson's example are illustrated below:

Figure 7

    The four tiles with two arrows crossing each other (the far left tiles above) are called crosses. The remaining tiles are called arms. In addition to arrow heads meeting arrow tails, we will require the following: Rule A: crosses must appear in every other position in every other row as illustrated below (Figure 8). It is important to note that crosses may appear elsewhere as well. Rule A is not technically a matching rule, but we could increase the number of tile types used to 32 and include additional markings on these tiles to obtain essentially this result. This is in fact what Robinson does in [5]. (He also points out that if we wanted to use only tiles with colored edges we could do so with a set of 56 tile types.)

Figure 8

    In this section we will discuss how this example is able to disprove Wang's conjecture; that is, we will show that 1) these tile types can be used to construct a tiling of the plane and 2) any tiling of the plane constructed using these tiles is not periodic.
    Let's think about the consequences of the matching rule requiring arrow heads to meet arrow tails. First of all, a cross can not sit next to (vertically or horizontally) another cross since that would force arrow heads to meet arrow heads. Next, consider any two crosses with a sequence of arms in between them; they must either face each other (Figure 9) or be back to back (Figure 10). When they are back to back, the second cross may be a mirror image of the first or it may be inverted. These three scenario's are the only possibilities: any other arrangement of the crosses yields a configuration with a different number or different locations of arrow heads on the sides of the crosses nearest to one another. But then these cannot be connected by a sequence of arms alone; the number (and location) of arrow heads or arrow tails on opposite sides of any arm are always the same. For example, consider the two crosses below (Figure 11) and try to fill in the middle tile with an arm. It can't be done!

Figure 9

Figure 10

Figure 11

    A similar argument holds for crosses appearing vertically with a sequence of arms in between them.
    In particular, the crosses appearing in alternate positions in alternate rows must either face each other or be back to back. Thus, if we start with one of these crosses, we may assume that it is facing another cross two units to its left or right. Let's consider a 3×3 square with these two crosses in the corners of the bottom row. We know that we have crosses in the other corners of the 3×3 square (by Rule A) and that these crosses must face the crosses in the bottom row. Thus the 3×3 square looks like:

Figure 12

    We also know that positions a-d must be arms (crosses can't sit next to other crosses). We see there are three choices for the arm in position a, tiles 5, 6 and 11. We note that all these choices have their arrows pointing away from the center of the 3×3 square. Similarly there are three choices for the arms that go in each of the positions b-d and all these arms also have their arrows facing away from the center of the 3×3 square. That means that no matter how we choose the various arms for tiles a-d, the tile in position e will have arrow heads on all four sides, so it will have to be a cross.
    Some specific examples of possible 3×3 squares are illustrated below:

Figure 13

    We'll say that a 3×3 square faces in the direction of its center cross; for example, the first 3×3 square in Figure 11 faces up and to the left based on the directions of the double arrows of its center cross.
    We can extend any 3×3 square in the direction it faces; so we can extend the first 3×3 square in Figure 11 to a 7×7 square with itself in the lower right corner. We first note that because all the exterior edges of the 3×3 square have arrow heads and because crosses cannot meet edges with arrow heads, the top and left sides of the 3×3 square must meet arms. Secondly, the positions of the crosses in the 7×7 square must be as shown in Figure 14 (Rule A). Then, because crosses cannot sit next to crosses, we have arms between all these crosses.

Figure 14 Figure 15

    In fact the crosses will be oriented as in Figure 15. This is not obvious but if we play with our tiles a bit it becomes clear: there are five choices for the arm in Figure 14 labeled A, tiles 5, 6, 8, 11, or 21. All of these have arrow tails on their top edge. Since arrow heads must meet arrow tails, this forces the arm in the position labeled B to have its arrow head(s) pointing away from the center of the 7×7 square. Similarly, by considering the choices for arm f, we see arm e also points away from the center of the 7×7 square. But this means that the center tile of the 7×7 square must have arrow heads on at least two sides so it must be a cross.
    The cross labeled # 1 will be back to back with the cross in the upper left corner of the original 3×3 square. It will either be inverted or a mirror image. If it is inverted then we have the following in the center of our 7×7 square:

Figure 16

    Try filling in the arms and the central cross in Figure 16. It can't be done. Thus, the cross labeled # 1 must be the mirror image of the cross in the upper left corner of the original 3×3 square. You can make a similar argument for the cross labeled # 2. Thus the crosses labeled # 1 and # 2 are forced to appear as in Figure 15 and these two crosses force the remaining crosses in Figure 14 to be oriented as in Figure 15.
    Once we have Figure 15, we see that in each corner we will be forced to have a 3×3 square with a central cross determined by the central cross of our initial 3×3 square. (The central cross of the lower left 3×3 square must face the central cross of the initial 3×3 square. Then the central cross of the upper left 3×3 square must face the central cross of the lower left 3×3 square and so on.)
    Finally, because all the 3×3 sub-squares have arrow heads pointing away from their center cross, we see that the horizontal and vertical rows of tiles radiating away from the center cross of the 7×7 square must also have arrows pointing away from the center cross. A sample 7×7 square is illustrated below:

Figure 17

     In a similar way, a 7×7 square can be extended in the direction of its central cross to a 15×15 square. This 15×15 square will have a central cross with a row and column of arms pointing away from it. The four corners of the 15×15 square will consist of 7×7 squares whose central crosses face each other. The 15×15 square can then be extended to a 31×31 square with a central cross and with the central crosses of the 15×15 squares in each of its four corners facing each other, and so on.
    Because (2n-1) ×(2n-1) squares can be tiled for all values of n, it follows that the plane can be tiled with this set of tiles! However, none of the tilings can be periodic. To see this we need to show that no matter how we translate a tiling of the plane formed using these tiles, some tiles will fail to match up. (In other words, you can't stand at two different spots on your infinite floor and have the pattern look exactly the same.) So imagine you have tiled the floor using these tiles and you are now standing on one of the crosses found in alternate positions in alternate rows. As described before, this cross must be part of a 3×3 square with a pattern as in Figure 13, and this 3×3 square must lie in some 7×7 square as in Figure 17, and so on. Look at the 3×3 square you are standing on and memorize its pattern. Now, move 4 units horizontally and/or vertically, staying within the 7×7 square containing the 3×3 square. You are in a new 3×3 square, so the pattern will look similar. However, since the center cross of the new 3×3 square will face the center cross of your original 3×3 square, these 3×3 squares cannot look exactly the same! Thus the tiling is not invariant under a translation by 22. It is also not invariant under a translation by 3 or 2; if you move horizontally or vertically by these amounts, the pattern does not even look similar to your original 3×3 square.
    Now return to the cross you started on and memorize the pattern of the 7×7 square you are in. Move horizontally or vertically by 8, staying in the 15×15 square which contains your 7×7 square. Again you are in a 7×7 square, so the pattern looks similar to the one you memorized. But, again, the center cross of the new 7×7 square will face the center cross of the original 7×7 square, so these 7×7 squares cannot look exactly the same. So the tiling is not invariant under a translation by 23. It is also not invariant under a translation by 5, 6, or 7 because under these translations, the pattern is not even similar to your original 7×7 square.
    Similarly you can argue that the tiling is not invariant under translation by any value 2n or any value m ¹ 2n. This shows that the tilings as described in this section are not periodic.

Other Facts And Examples

In disproving Wang's Conjecture, we constructed an example of a set of tiles which could be used to tile the plane but for which no such tiling is invariant under a translation. In fact, no tiling constructed using this set of tiles is invariant under a rotation [5].
    It is interesting to note that although no tiling constructed with this set of tiles is periodic, they are ``almost periodic"; that is, when choosing a period large enough, an arbitrarily large percentage of the tiles will repeat. For example, the crosses making up the corners of the 3×3 squares will repeat horizontally and vertically with period 4. These crosses make up one quarter of the total tiles; to see this, divide any 2n×2n block into 2×2 blocks. Each will contain exactly one such cross. So one quarter of the tiles repeat with period 4.
    Similarly, we can divide any 2n×2n block into 4×4 blocks, each containing exactly one 3×3 square. The 3×3 squares are determined by their central cross and thus repeat with period 8. So 9/16th of the tiles repeat with period 8. In general, ([(2n -1)/( 2n)])2 of the tiles repeat with period 2n+1.
    Another famous tiling of the plane is the Penrose tiling [4],[2]. First discovered by Roger Penrose in 1973, a modified version done in 1974 uses only two tile shapes, commonly referred to as ``kites" and ``darts", and finitely many different rotations of these shapes. A tiling must satisfy the matching rule that colored vertices, indicated below by shadings, must match (Figure 18).

Figure 18 Figure 19

    The Penrose tilings (Figure 19) are not periodic under any translation, however they can be periodic under rotations by a fifth root of unity. The Penrose tilings are almost periodic in the sense that the pattern seen in any arbitrarily large block will repeat within a bounded translation. The Penrose tiles are actually patented and are useful in understanding the geometric properties of quasicrystals [6].
    In the Pinwheel tilings (Figure 20) there is a single tile shape, a right triangle with sides 1-2-Ö5. This example differs from our previous ones in that it involves infinitely many different tiles because in each tiling the triangle occurs in infinitely many different orientations [3]. The Pinwheel tilings are nonperiodic under translations and rotations. They are almost periodic in the sense described above.

Figure 20

    There are many other interesting tiling examples and we refer the reader to [2] for further exploration.

                                                                                    References

[1] Berger, R., ``The Undecidability of the Domino Problem", Mem. Amer. Soc. 66, (1966).
[2] Grunbaum, B., and G. C. Shephard, ``Tilings and Patterns", W. H. Freeman and Company, (1987).
[3] Radin, C., ``The Pinwheel Tilings of the Plane", Annals of Mathematics, (to appear).
[4] Robinson, E. A., ``The Dynamical Properties of Penrose Tilings", (preprint).
[5] Robinson, R. M., ``Undecidability and Nonperiodicity for Tilings of the Plane", Invent. Math. 12 (1971), 177-209.
[6] Senechal, M., ``Crystalline Symmetries", Adam Hilger (1990).
[7] Wang, H.,``Proving Theorems By Pattern Recognition-II", Bell System Tech. J. 40, 1-41 (1961).