More variations on the Sierpi ski sieve n

Kevin G. Hare
Department of Pure Mathematics, University of Waterloo, Waterloo, Ontario, Canada, N2L 3G1 E-mail: kghare@math.uwaterloo.ca Abstract. This paper answers a question of Broomhead, Montaldi and Sidorov upon the existence of gaskets of a particular type, related to the Sierpi skis sieve. n These gaskets are given by Iterated Function Systems that do not satisfy the Open Set Condition. Despite this, these gaskets are still simple enough to compute their dimension.

AMS classication scheme numbers: 28A80, 28A78, 11R06

1. Introduction and basic denitions This paper was motivated by the paper of Broomhead, Montaldi and Sidorov, [1]. In Broomhead, Montaldi and Sidorovs paper, the authors looked at a variation of the Sierpi skis sieve, that allowed some overlap between triangular subregions. They n showed that when the overlap was suciently nice, that the gaskets dimension could be computed. In the conclusions to their paper, they had raised the question of what happens if, instead of triangles, the maps are based on regular n-gons with n 5. This current paper answers this question, showing that it is possible to have gaskets of this type for n = 3, 6, 8 and 12. It also shows that these are the only values of n for which gaskets of this type can be constructed. We also show how the methods of [9] can be used to compute the dimension of these gaskets. Before we begin a detailed discussion of our results, we need to introduce some notation and denitions. Denition 1.1. Let f1 , f2 , , fn be a nite set of linear contractions from Rn to Rn . The Iterated Function System (IFS) associated with these contractions is the unique non-empty compact set K such that K = fi (K). Since their introduction, IFSs have been widely studied (see [4, 5] and the references listed there). One condition that useful in studying IFSs is the Open Set Condition. Denition 1.2. We say that an IFS, associated with f1 , f2 , , fn satises the Open Set Condition, (OSC), if there exists a non-empty open set O such that
Research of K. G. Hare supported, in part by NSERC of Canada

More variations on the Sierpi ski sieve n fi (O) O for all i.

fi (O) fj (O) = for all i = j.

What this condition intuitively says is, if K is the compact set associated to the IFS, then fi (K) and fj (K) have small overlap (dimension less than that of K). In practice the construction of this open set O can be quite complicated. If an IFS satises the OSC, then it is relatively straightforward to compute the dimension (see for example [4, 5, 7]). In our case, we examine IFSs that do not satisfy the OSC. There are a number of techniques that can be used if the overlapping sections look nice. We mean by nice that its construction satises the Finite Type condition that will be dened formally in Section 3. See for instances [8, 9]. In most cases, if the construction of our IFS is nice, then the box-counting dimension and the Hausdor dimension are the same. See for instance [2, 3]. We quote a result of Das and Ngai: Theorem 1.1 (Theorem 1.1(a) of [2]). Let G = (V, E) be a nite type graph-directed iterated function system of contractive similtudes on Rd , let = {i }q be a nite type i=1 condition family for G, and let A be the corresponding weighted incidence matrix. Then the associated graph self-similar set F satises dimH (F ) = dimB (F ) = ln /( ln ) where = min{e : e E} and = (A) is the spectral radius of A. We show in Sections 3 and 4 that the hypothesis of this Theorem are satised. The general construction of these gaskets is based on the contraction maps fi (x) = x + bi , where x R2 and bi are the vertices of our regular n-gon. If we wish to specify the value of n used in the construction, then we will say n-gasket, instead of gasket. Not all values of give interesting gaskets. If is too large, then the resulting object will be a lled-in n-gon of dimension 2. If is too small, then it will satisfy the OSC, and its dimension can be trivially computed. In Section 2 we will give a description of the contraction ratios used to nd ngaskets with nice overlap. We will also show that this method only works for triangles, hexagons, octagons, and dodecagon. In Section 3 a brief discussion of the algorithm of Ngai and Wang [9] is given, as well as how it relates to our problem. The interested reader is encouraged to go to the original article for a more more general discussion. (The algorithm can deal with problems of a more general nature that what we look at here.) In Section 4 we compute the dimension of these n-gaskets, using the contractions ratios given in Section 2. Some discussion of further directions for this research is given in Section 5. 2. Contraction Ratios In this section, we will show how to determine the contraction ratio associated to these gaskets. Before we begin though, we must determine the contraction ratio for the nonoverlapping connected n-gasket (i.e. where the corners touch but do not overlap). The

More variations on the Sierpi ski sieve n

. . . . . .

1
(a) 9-gasket - no overlap Figure 1. 9-gon - no overlap
. . .

(b) 9-gasket - construction

. .

cos(2 ) cos()
(a) 9-gasket, width Figure 2. 9-gon - calculating

2 (1+cos( ) + ... + cos(N )) = 1

(b) 9-gasket - construction

result for the non-overlapping case is not new, but the technique used is useful for demonstrate, as a variation of this is used for nding the contraction ratios with nice overlaps. Consider an n-gasket with side length 1. We wish to put n smaller copies of the n-gasket, with side length , at each corner of the larger n-gasket, such that the corners of the smaller n-gasket will just touch. We will denote this ratio as = n, . (This choice of notation will be explained later, when we talk about n-gaskets with overlapping components.) See for example Figure 1. We notice that for an n-gasket with side length , we can determine the distance between where the smaller n-gasket touches the corner of the larger n-gasket, and where the smaller n-gasket touches the corner of the adjacent smaller n-gasket. See for example Figure 2. Let = 2 , and let N = n . We see that this length then is: n 4 + cos() + + cos(N). This then gives us that 1 = 2 ( + cos() + + cos(N)) = 1+
sin (2N + 1) 2 sin 2

More variations on the Sierpi ski sieve n n 3 4 5 6 7 8 9 10 11 12 tn 2 2 2 cos (1/5) + 1 3 sin(1/7)+sin(3/7) sin(1/7) 2+ 2 sin(1/9)+sin(4/9) sin(1/9)

2.618033989 3.246979604 3.414213562 3.879385241

5+3 51 5 sin(1/11)+sin( 11 ) sin(1/11)

4.236067979

3+

4.513337092 3 4.732050806
. . . . . .

Table 1. Table of tn values

. . . . . .

Figure 3. 2 sized gasket overlap

For ease of notation, let sin sin (2N + 1) 2 =1+ tn = 1 + sin 2 Hence 1 . (1) tn See Table 1 for values of tn for small n. We will use this tn in the description of what happens when we allow overlap. Consider an n-gasket where we scale by = n,m , and we wish that, after scaling m + 1 times by , that the m+1 sized copies of the n-gasket coming from adjacent parents overlap. See for example Figure 3 for 8,1 . We see that the width of a full n-gasket with side length m+1 is m+1 (tn 1). So we see that we have two gaskets, of side length , minus an overlapping gasket of side length m+1 , along an edge of length 1. This gives n, = m+1 (t 1) t + 1 = 0 (2) 2 n +1 4 sin n
n

More variations on the Sierpi ski sieve n

(a) gasket

(b) gasket - adjacent components and overlap

Figure 4. 3-gasket of order 2

(a) gasket

(b) gasket - adjacent components and overlap

Figure 5. 8-gasket of order 2

where (0, 1). It is worth observing that this gives the same equation as [1] in the case when n = 3. It is also worth observing that, as m that n,m n, , which explains our choice of notation from earlier. We will use the term n-gasket of order m to mean a n-gasket such that for adjacent gaskets of side length overlap at a gasket with side length m+1 . The next question that we need to consider is, what sort of additional overlap do we get? Consider an n-gasket of order m. We know that the two adjacent copies of this gasket of side length will overlap in a gasket of side length m+1 . The main question is, do we get any other overlapping sections of these gaskets of and if so, how nice is this overlap? In the case of n = 3, or n = 8, with m 2, there is no other overlap. In the case of n = 6, 12 or n = 8 with m = 1, there is overlap, but the overlap is nice in the sense that it is just the complete overlap of gaskets of side length k for some larger values of k. For all other n, the answer is not as nice, in that there is overlap, but the overlap is not complete overlap of smaller gaskets. This is summarized by the next three Theorems. The rst two Theorems are done by a careful observation of the relevant diagrams. The formalization of these results is left as an easy exercise to the reader. Theorem 2.1. If n = 3 with m 2 or n = 8 with m 2 then there is no additional

More variations on the Sierpi ski sieve n

(a) gasket

(b) gasket - adjacent components and overlap

Figure 6. 6-gasket of order 3

(a) gasket

(b) gasket - adjacent components and overlap

Figure 7. 8-gasket of order 1

(a) 12-gasket of order 2 Figure 8. Examples of n-gaskets

(b) 12-gasket of order 2

overlap. Proof. Consider two adjacent components of side length of a 3-gasket of order m. The only region of overlap for these two components is a triangle of side length m+1 . See Figure 4. Consider the two adjacent components of side length of an 8-gaskets of order m, for m 2 We see that the only overlap is 2m components of side length m+1 . See Figure 5.

More variations on the Sierpi ski sieve n

Theorem 2.2. If n = 6 with m 2, n = 8 with m = 1 or n = 12 with m 1, then there is nice additional overlap. Proof. Consider the two adjacent components of side length of a 6-gasket of order m. We see that the overlap between these two regions is 1 gasket of side length m+1 , 2 gaskets of side length 2m+1 , and in general 2k1 gaskets of side length km+1 . See Figure 6. Consider the two adjacent components of side length of an 8-gasket of order 1. We see that the overlap between these two regions is 2 gaskets of side length 2 , 4 gaskets of side length 3 , and in general 4 gaskets of side length k . See Figure 7. Consider the two adjacent components of side length of a 12-gasket of order m. If m = 1, then the overlap is 2 gaskets of side length 2 , and 4 gaskets of side length k for all k 3. If m 2, then the overlap is 2m gaskets of side length m+1 , and in general 2m+k1 gaskets of side length km+1 . See Figure 8. Theorem 2.3. For all other n and m there is overlap that is not complete overlap. Proof. Let n be odd, n 5. Consider the 2 adjacent components of an n-gasket of order m, where the components have side length , and we have an overlapping gasket with side length m+1 . See Figure 2. Consider the neighbours of the overlapping gasket of side length m+1 . What we wish to show is that these neighbours have additional overlap, and that this overlap is not the complete overlap of smaller copies of this gasket. As n 5 and n is odd, we see that two of these neighbours will overlap. We notice that the width of this overlap is (tn 2)m+1 + 2m+1 = m+1 (tn 2 + m+1 ). We see that the width of a gasket with side length k , (with k m + 1), is k (t 1). Letting k = k m + 1, it suces to show that there are no integer solutions to tn 2 + m = k (tn 1)

(3)

tn 1 This shows that the right hand side of (3) is bounded above by tn 1 = 1. For n 7, we see that the left hand side of (3) is bounded below by t7 2 1.24 > 1. Hence there are no solutions for n 7. So assume that n = 5. Then t5 = 5+3 . If k 2, then the left hand side is bounded 2 above by t51 = t5 2. The right hand side of (3) is strictly greater than t5 2, hence 1 there are no solutions if n = 5 and k 2. So assume that n = 5 and k = 1. Multiplying (3) by and using (2) we get

for m, k 1. Notice that 1 1 = m, m,3 m,2 m,1 = tn tn 1

(t5 2) +

Solving this equation gives = 5, = n 5.

t5 1 = 2 (t5 1) t5 1
1 . t5

Hence there are no solutions for n odd, and

More variations on the Sierpi ski sieve n

Overlap size Width

2m+1

m+1 2m+1 (t-2) +

Figure 9. n-gasket for n-odd

A similar argument can be made for when n 2 (mod 4) and n 10 and when n 0 (mod 4) and n 16. The dening equation for n 2 (mod 4) is tn 3 + 2m (tn + 1) = k (t 1) and for n 0 (mod 4) is

tn 3 2 cos + 2m (tn + 1) = k (t 1)

As before, there are no solutions if n 10 or n 16 respectively, as one side is bounded above by 1, and the other is strictly greater than 1.

3. Algorithm First, let us formally dene the maps used to construct our gaskets. Let j (x) = x + bj where x, bj R2 . Here bj are the vertices of a regular n-gonand = n,m is given in equation (2). The bj are scaled in such a way that the IFS has side length 1. We dene J {1, 2, , n} to be a nite word with symbols 1, 2, , n. Letting J = j1 j2 jk we dene J = j1 j2 jk . 1 Dene R = 1 . Let BR be the open ball of radius R around 0. We say that J1 and J2 are neighbours if |J1 | = |J2 | and J1 (BR ) J2 (BR ) = . In our case we can think of each J being associated with a particular n-gasket of side length |J| , and we say that two n-gaskets are neighbours if they overlap. We dene the neighbourhood set of J as (J) = {J : J is a neighbour to J}.

More variations on the Sierpi ski sieve n

Based on J we dene a map (x) := J (x) = |J| (x J (0)). This map re-scales and re-centers a neighbourhood set (J), allowing us to compare neighbourhood sets at dierent depths. We say two neighbourhood sets have the same neighbourhood type if there exists a rotational matrix U such that J1 ((J1 )) = UJ2 ((J2 )). We say that an IFS is of nite type if there are only a nite number of neighbourhood types. We now create a graph from neighbourhood types to neighbourhood types. To do this, consider a J = j1 j2 jk associated with a particular neighbourhood type. We then consider the sequences Jr = j1 j2 jk r where r takes on the values between 1 and n. We say an edge from J to Jr is acceptable if there are no other J with |J | = k + 1 such that J = Jr where J would have precedence. This concept of precedence is described in more detail in Section 4. In our implementation, we use a rotational precedence which we keep track of within each neighbourhood type. Then, if is the largest eigenvalue of the incidence matrix of this graph, then the dimension is dim(K) = log() . log()

In the next section, we will show that these gaskets have nite type, and construct their incidence matrix and compute their dimension. 4. Dimension In Section 2, we discussed the contraction ratio one would expect for an n-gasket of order m. In Section 3 we gave an overview of the algorithm of Ngai and Wang that we will be using. In this section, we show how to compute the dimension of such gaskets. We will do this separately for each n-gasket of order m, as the techniques are optimized for each case. Special attention is given to the case n = 6, after which we give the highlights only for n = 8 and n = 12, as the techniques are similar. We do not do the case of n = 3, as it is already done in [1]. 4.1. Hexagons Consider the case where we have overlapping hexagons. If our contraction ratio is = 6,m , then we have overlapping gaskets of side length m+1 . The trick to computing the dimension of this object, (or in fact any of these n-gasket of order m) is to carefully count how many gaskets of side length k there are as k . We need to worry about when gaskets overlap completely, as we need to avoid double counting. For each gasket of side length k , say A, consider the six hexagons of side length k+1 , say B1 , B2 , , B6 , of which A is composed. In this case we say that A is the parent of the Bi s, and the Bi s are ospring of the A. We similarly dene the notion of descendants and ancestors.

More variations on the Sierpi ski sieve n

D D D

10

A
valid offspring

B
non-valid offspring

C
Figure 10. Precedence diagram

Now consider some hexagon C of side length k such that it is the ospring to two dierent parents, say A and B. This will happen when C is an overlapping ospring gasket of two adjacent gaskets. These two parents have a common ancestor D. The nice property, which we can exploit in this case, (but is not true for general IFS), is that we can impose a priority on these parents that is rotational. Let DA be the ospring of D that contains A and DB the ospring of D that contains B. If DA is counter-clockwise to DB then we say that A has priority over B. As such, when counting ospring of A and B, we count C as an ospring of A, but not as an ospring of B. This avoids double counting. This is described in the ancestor diagram in Figure 10. At this point we need to introduce some notation for hexagonal gaskets of order m, (where m 3.) We will do this by way of example. Example 4.1. Consider the following example found by looking at the 6-gasket of order 2 in Figure 11. We will nd the 6 neighbourhood types that are ospring of the neighbourhood type of B. Here we mean in Figure 11(a), by the rst B that the neighbourhood type can be typied by a hexagon, where the upper left corner does not have precedence, and the lower left corner does. Moreover, the overlapping smaller hexagons in both cases with be 2 smaller than the hexagon in question. We can, based on this, determine the six hexagons that the parent hexagon is composed of. This is represented graphically in Figure 11(b) and 11(c) where blue is the hexagon we are interested in, green represents hexagons with which blue has precedence over, and red represents hexagons that have precedence over blue. If the hexagon has order m, then we see that each corner can take the values of empty, or 1 through m, with either positive or negative sign. This means that there are a total of (2m + 1)6 possible hexagons. (There are in fact much fewer in practice that we need to worry about.) Hence, these gaskets have Finite Neighbourhood Type. Further, this is a nite type graph-directed iterated function system, satisfying the conditions of Theorem 1.1. Below is our full description of the maps from the neighbourhood types to their ospring. The hexagon A is the neighbourhood type for the starting n-gasket

More variations on the Sierpi ski sieve n

-2 -1 -2 -2 2 2

11 B
2

B
2 -2

D1 2

-2

B
2 2

B
1

C1

-2

B
-2

(a) Precedence structure

(b) Parent

(c) Children

Figure 11. Precedence structure and graphical interpretation

of side length 1.
-m -m m m

A
-m -m -(m-1)

-m

B
m m -m

B
-m m

-m

B
-m m

B
m -m

m Dm1 -m -(m-1)

-m

B
m m -m

B
-m m

-m Cm1 m-1

B
-m m

Cn
m -m

m Dm1 -m -(m-1)

-m

B
m m

n-1 Cn1

B
-m m

-m Cm1 m-1

B
-m m

-m

Dn -n
m -m

m Dm1 -m -(m-1) m Dm1 -m

-m

B
m m

1-n Dn1

B
-m m

-m Cm1 m-1

B
-m m

-m

C1
m

-m

m m m

F
m

-m Cm1 m-1

B
-m

More variations on the Sierpi ski sieve n

-m -(m-1) m Dm1 -m -(m-1) -m m -m Cm1 m-1 -m -m m -m Cm1 m

12
m

D1
m -m

-1

-m

-m -m

-m

B
-m m

E
m -m m m

m Dm1 -m -(m-1) m Dm1 m m -m -(m-1) m Dm1 -m -m -(m-1) m Dm1 -m -m -(m-1) m Dm1 -m

-m

B
m m m-1 -m

Cm1
-m m

B
-m m

m-1 m-1

F
m -m

-m

Cm1
m m m-1 -m

Cm1
-m m

-m Cm1

B
-m m

m-1 m-1

G
m -m

-m

Cm1
m m 1-m -m

B
-m m

-m Cm1 m-1

B
-m m

H
m -m

-m

Dm1
m m 1-m -m

B
-m m

-m Cm1 m-1

B
-m m

I
-m

-m

Dm1
m m

Dm1
-(m-1)

-m

B
-m

We notice that because our priority is rotational, we can rotate the existing ospring and treat a number of these gaskets as equivalent. From this network diagram, we can compute an adjacency matrix. A B Cm1 Dm1 . . . C2 D2 C1 D1 E F G H I A 0 0 0 0 0 0 0 0 0 0 0 0 0 B Cm1 Dm1 Cm2 Dm2 C1 D1 E F G H I 6 4 1 1 3 1 1 1 3 1 1 0 1 .. . 3 1 1 0 3 1 1 0 0 1 3 1 1 0 0 0 1 1 1 1 0 0 0 0 1 1 1 1 1 1 0 0 0 0 0 0 0 1 1 3 2 1 2 3 1 3 2 1 3 1 2 3 0 3

We see that (k) := [1, 1, , 1]Ak [1, 0, 0, , 0]T gives the number of hexagons of side length k . At this point it suces to nd the maximal eigenvalue of A to tell us

More variations on the Sierpi ski sieve n

13

what the growth of (k) looks like. Let a, b, , be the frequency associated with A, B, . Let be the eigenvalue of A. This tells us that a = 0 b = 4b + 3cm1 + 3dm1 + + 3c2 + 3d2 + c1 + d1 +3e + 2f + 3g + 3h + 3i cm1 = b + cm1 + dm1 + + c1 + d1 + 2e + 3f + 2g + h dm1 = b + cm1 + dm1 + + c1 + d1 + e + f + g + 2h + 3i cm2 = cm1 dm2 = dm1 e = c1 h = d1 . . . . . . f = c1 i = d1 . c1 = c2 By using the observations that ck = k g f = g b dk = k h i = h e = g f = g. d1 = d2 g = c1

we can rewrite this system to get the relevant equations + 3g + 2g + 3g + 3h + 3h m 2 (g + h) + g + h + 8g + 6h = 4b + 3 1 m g = b + (m1 + + )(g + h) + 2g + 3g + 2g + h m (g + h) + 7g + h = b+ 1 m h = b + (m1 + + )(g + h) + g + g + g + 2h + 3h m (g + h) + 3g + 5h. = b+ 1 Subtracting equation (4) from (5) gives us 0 = (m 4)(h g). Either = m 4 or h = g. Here we want to be the largest eigenvalue, so assume for the moment that h = g, to nd the other eigenvalues. This then gives us m 2 b = 4b + 6 g + 2g + 14g (6) 1 m 8g (7) b = m g 2 1 Combining (6) and (7), and noticing that g = 0 gives the equation for .
(m 2 1 8) (4(m 2 1 8) + 6 1 + 2 + 14) m+1 +22 m 8 m +62 = m+1 + 2 1 8 4m + 81 + 32 + 61 2 14 m+1 m 2 = m+1 + 2 +2 +8 8 4m 10 + 18 1 = m+1 6m 2 + 18
m m m 2

= 4b + 3(m1 + + 2 )(g + h) + g + h

(4)

(5)

More variations on the Sierpi ski sieve n

14

-2 1

(a) Precedence Structure

(b) Graphical Interpretation

Figure 12. Precedence structure and graphical interpretation

By noticing that 5 for all m 3, gives that = result:

4. This gives the nal

log() Theorem 4.1. The 6-gasket of order m has dimension log() where is the largest root of m+1 6m 2 + 18 and is the root in (0, 1) of 2m+1 3 1.

It is worth observing that as m that 6, and 6,m 6, , and further log(6) the dimension of the non-overlapping case is log(6, ) . 4.2. Octagon The case of octagons is done in a similar way. The main dierence is that they are denoted dierently, as the overlaps are on the edges, not the corners, and will always eect two octagons. The case of m = 1 is a special case, as its overlap structure is unusual (see Theorem 2.2). Its dimension was computed directly by a computer implementation of [9], (see [6] for the implementation). Using the same colour notation as Figure 11, we give an example of a particular 8-gasket of order 2 (see Figure 12). This means that the left side will overlap an adjacent octagon in 1 step, and this octagon will have priority. This also means that the upper right side will overlap an adjacent octagon in 2 steps, and this octagon will not have priority. Lastly, all other sides do not overlap other octagons. All of the ospring octagons can be determined by the parents with this information. This is done in much that same way as the hexagon case. For easy of notation, we will not give a pictorial representation as we did for the hexagons. Instead we will list only what is meant by each 8-gasket, and the transitions. See Figure 13. The maps are:

More variations on the Sierpi ski sieve n

n n -n -m m -(m-1) -n

15

-m

B
m

-m

Cn

Dn
m

-m

En
m

-m

Fn
m

-m

G
-m

-m

H
-m

Figure 13. Pictorial Representations

A B Cn Dn C1 D1 En Fn E1 F1 G H H

8B 4 B, Cm1 , Dm1 , Em1 , Fm1 2 B, Cn1 , Dn1 , Cm1 , Dm1 , Em1 , Fm1 2 B, Cn1 , Dn1 , Cm1 , Dm1 , Em1 , Fm1 2 B, Cm , Dm , Cm1 , Dm1 , Em1 , Fm1 2 B, Cm , Dm , Cm1 , Dm1 , Em1 , Fm1 2 B, En1 , Fn1 , Cm1 , Dm1 , Em1 , Fm1 2 B, En1 , Fn1 , Cm1 , Dm1 , Em1 , Fm1 B, G, Cm1 , Dm1 , Em1 , Fm1 2 B, H, Cm1 , Dm1 , Em1 4 B, 2 Em1 , 2 Fm1 2 B, 2 Em1 , 2 Fm1 , Em2 , Fm2 2 B, 2 Em1 , Fm1 , H

(n = 1) (n = 1)

(n = 1) (n = 1)

(m = 2) (m = 2)

Using a similar technique to before, we get satises (for m 2) m+1 8m + 2m1 + 2m+2 Theorem 4.2. The 8-gasket of order m has dimension root of m = 1, 3 62 + + 12
log() log()

where is the largest

m 2, m+1 8m + 2m1 + 2m+2 and is the root in (0, 1) of ( 2 + 1)m+1 (2 + 2) 1. 4.3. Dodecagon We proceed the same as for the octagon case, for m 2. The case dodecagon with m = 1 is somewhat special, and is calculated using the code provided at [6]. Using Figure 14 for a representation of which each symbol represents we get the maps for m 2 are: A 12 B

More variations on the Sierpi ski sieve n -m A -m m Fn -n -m m m -m

Figure 14. Pictorial Representations

16 -m n m -m K -m m Dn n -n m -m m -m -m L -m m En

Cn

-m G m

Cn 6 B, Cn1 , Dn1 , Cm1 , Dm1 , Em1 , Fm1 En 6 B, En1 , Fn1 , Cm1 , Dm1 , Em1 , Fm1

B 8 B, Cm1 , Dm1 , Em1 , Fm1

(n = 1) (n = 1) (n = 1) (n = 1)

Dn 6 B, Cn1 , Dn1 , Cm1 , Dm1 , Em1 , Fm1 Fn 6 B, En1 , Fn1 , Cm1 , Dm1 , Em1 , Fm1

C1 4 B, 2 G, 2 H, Cm1 , Dm1 , Em1 , Fm1 E1 4 B, K, L, Cm1 , Dm1 , Em1 , Fm1

D1 4 B, 2 G, 2 H, Cm1 , Dm1 , Em1 , Fm1 F1 4 B, K, L, Cm1 , Dm1 , Em1 , Fm1

G 6 B, 2 Cm1 , 2 Dm1 , Em1 , Fm1 K 6 B, 3 Em1 , 3 Fm1

H 6 B, 2 Cm1 , 2 Dm1 , Em1 , Fm1 L 6 B, Cm1 , Dm1 , 2 Em1 , 2 Fm1

log() log()

Using the same techniques as before, this gives us the Theorem. Theorem 4.3. The 12-gasket of order m has dimension root of m = 1, 2 11 + 16. where is the largest

m 2, m+1 12m 2m1 z + 18 2m and is the root in (0, 1) of ( 3 + 2)m+1 (3 + 3) 1. 5. Conclusions This paper answers a question of Broomhead, Montaldi and Sidorov, as to the existence and construction of n-gaskets. It shows how to use the algorithm of Ngai and Wang to computed these gasket dimensions. There are a number of observations that are worth making at this point. First, the gaskets studied in this paper all have nite type. It can

More variations on the Sierpi ski sieve n

17

be shown that if 1/ is a Pisot number, with Q() Q(cos(2/n)) then the n-gasket with contraction will also have nite type. (If is too big, then this will become a trivial full set, and if it is too small, then it will satisfy the OSC.) Examples would be 0.5698402911, the root of x3 x2 + 2x 1, for n = 3 or 0.3819660113 the root of x2 3x + 1 for n = 5. These objects in general will be much more complicated that the n-gaskets of order m that we studied here. It is not immediately obvious whether the existence of such a in the relevant range is guaranteed for all n-gons. Moreover, the computational aspects in even this simple case of this problem are immense. For example, the resulting minimal polynomial for , for the 3-gon with the contraction 0.5698402911, is of degree 445. How these computations could be, or should be done for more complicated objects would require techniques from sparce matrix theory. Variations of these sorts of objects can also be studied in higher dimensions. It is not clear if there is an equivalent trick to rotational precedence in in higher dimensions. It is possible to ignore this, and simply do a larger computation, but that seems computational inecient. Acknowledgments This work would not have been possible without the useful suggestions and comments of Nikita Sidorov. References
[1] Dave Broomhead, James Montaldi, and Nikita Sidorov. Golden gaskets: variations on the Sierpi ski n sieve. Nonlinearity, 17(4):14551480, 2004. [2] Manav Das and Sze-Man Ngai. Graph-directed iterated function systems with overlaps. Indiana Univ. Math. J., 53(1):109134, 2004. [3] K. J. Falconer. Dimensions and measures of quasi self-similar sets. Proc. Amer. Math. Soc., 106(2):543554, 1989. [4] Kenneth Falconer. Techniques in fractal geometry. John Wiley & Sons Ltd., Chichester, 1997. [5] Kenneth Falconer. Fractal geometry. John Wiley & Sons Inc., Hoboken, NJ, second edition, 2003. Mathematical foundations and applications. [6] K. G. Hare. Home page. http://www.math.uwaterloo.ca/kghare, 2004. [7] John E. Hutchinson. Fractals and self-similarity. Indiana Univ. Math. J., 30(5):713747, 1981. [8] Steven P. Lalley. -expansions with deleted digits for Pisot numbers . Trans. Amer. Math. Soc., 349(11):43554365, 1997. [9] Sze-Man Ngai and Yang Wang. Hausdor dimension of self-similar sets with overlaps. J. London Math. Soc. (2), 63(3):655672, 2001.