Topological dimensions, Hausdor dimensions & fractals

Yuval Kohavi, Hadar Davdovich May 2006

Abstract
We present some basic facts about topological dimension, the motivation, necessary denitions and their interrelations. Finally we discuss the Hausdor dimension and fractals.

1 Topological dimension
1.1 Motivation
Topological dimensions denes the basic dierence between related topological sets such as I n and I m when n = m. The lack of that denition is especially highlighted because of the easy explanation of the geometric dimension. When trying to dene a dimension for a topological space, you might run into several diculties - Unlike a vector space, you can not state that the dimension is the maximum of linear independent vector. Therefore in order to get an intuitive denition for a topological dimension we should look for dierent properties that has the eects of the geometrical dimension.

1.2 General
A topological dimension has values in 1, 0, 1, 2, 3, . . . , and is topological, i.e: if X and Y are homeomorphic then they have the same "dimension". It will also be nice, if Rn gets value n, for each n. There are 3 commonly used denitions: 1. Small inductive dimension (ind) 2. Large inductive dimension (Ind) 3. Lebesgue covering dimension (dim) All these denitions have the required properties, and we will see that they are the same for separable metrizable spaces (ind = Ind = dim).

Remainder: metrizable space is a topological space that is homeomorphic to a metric space.

2 The inductive dimension

2.1 Intuition
Lets look at a cube. What is the (geometrical) dimension of the cube? 3, and what is the (geometrical) dimension of its boundary [square]? 2. Now let's look at the square. What is the dimension of its boundary? well its boundaries are lines, so their dimension is 1. And what about the boundary of the lines? Well their boundary is composed of dots, there dimension is 0. We get a pattern - the dimension of a spaces, is 1 + the dimension of its boundary. As a boundary is well dened in topology, this notion can be easily applied to topological spaces, to create a topological dimension.

3 The small inductive dimension

Denition 3.1. The small inductive dimension of X is notated ind(X ), and is dened as follows:
1. We say that the dimension of a space X (ind(X )) is -1 i X = 2. ind(X ) n if for every point x X and for every open set U exists an open V, x V such that V U , and ind(V ) n 1. Where V is the boundary of V . 3. ind(X ) = n if (2) is true for n, but false for n 1. 4. ind(X ) = if for every n, ind(X ) n is false.

Remark 3.1. The ind dimension is indeed a topological dimension: X is homeomorphic to Y implies that ind(X ) = ind(Y ).
This can be shown easily (and inductively) as the denition relays only of open\closed sets.

Remark 3.2. An equivalent condition to condition (2) is:

The space has a base B , and every U B has ind(U ) n 1 [You can construct this base using the sets U from condition (2)].

Theorem 3.1. As we would expect, if Y X then ind(Y ) ind(X ).

Proof. By induction, it is true for ind(X ) = 1. if ind(X ) = n, for every point y X there is an nbd V , with an open set U V , such that ind(V ) n 1. Note that VY = Y V is a nbd in Y , and UY is open in Y . By induction, it is enough to show that UY U , Because then ind(UY ) n 1, and therefore ind(Y ) n = ind(X ) (by denition). And indeed Uy U , UY = UY \ UY U \ U = U Example 3.1. Let's show that ind(R) is 1. for each x R, lets select a nbd, V and a set U = (a, b) V , ind(U ) = ind({a, b}) = 0. This implies that ind(R) 1.

So, it is enough to prove that ind(R) is not 0. But that is easy! if ind(R) = 0, that means that a set U exists, such that: ind(U ) = 1 U = U is clopen! but if U is clopen, R is disconnected! Because R is connected, we get that ind(R) > 0. So nally, ind(R) = 1 . The same proof can be done for the Sphere S 1 and the interval I 1 .

Corollary 3.2. A zero-dimensional space is disconnected.

ind(X ) = 0

Corollary 3.3. if for every nbd V of x X exists a clopen set U V then

Example 3.2. What is the dimension of C the cantor set? Lets look at C in R, C has no interior points (fact). That means, that for each nbd U = (a, b) of x C , we can nd two points c, d such that a c x d b, and c, d / C. Lets look at C as a subspace, V = (c, d) is open, and closed ind(C ) = 0
Note that these results are not exactly what we expect. The cantor set has the same 'size' as R: 20 , so we should expect it to have the same dimension. But then again, the cantor set has no interval in it. Its dimension should be then, somewhere between 0 and 1. We shall later learn of a dimension denition that suits our expectations. It is worth to mention, that the cantor set is universal 0-dimensional space of separable metrizable spaces. meaning that every other space separable metric space X, ind(X ) = 0, is homeomorphic to a sub-space of the cantor set. i Remember that a number is in the Cantor set It has the form iN x 3i , x i {0, 2}. 2 2 0 for example: 3 +0 9 + 27 + 81 + . We can display this number simply as a fraction of radix 3 0.2020.... This set of fractions has one-to-one and onto mapping to the set of binary fractions 0.1010.... We simply replaced all the 2's with 1's. The latter set of fractions is more familiar to topology, it is the space:D0 , D = {0, 1}, D with the discrete topology.

Proposition 3.4. The D0 is homeomorphic to the Cantor set.

Now it is enough to show that every zero dimensional separable metrizable space is homeomorphic to a sub-spaces of D0 .

Remark 3.3. Every zero dimensional spaces has a clopen base (by the alternative denition). If the space is also metrizable and separable, it has a countable clopen base (because then every base has a countable base contained in it).

Theorem 3.5. The Cantor set is a universal space for all the zero-dimensional
metrizable separable space. Proof. Let X be a separable metrizable space. It is enough to show that X is homeomorphic to a sub-set of D0 . Let B = {Bi } be its clopen countable base. We dene f to be: fi (x) = 1 x Bi . Dene f (x) = (f1 (x), f2 (x), . . .). f is the homeomorphism 0 otherwise that we wanted.
4

An interesting question we can ask about this dimension, is wether ind(Rn ) = n. The answer is yes,we will start by showing that ind(Rn ) n.

Proposition 3.6. The dimension of Rn , S n , I n n.

We can show this by induction. we have already shown that the dimension of R1 , S 1 , I 1 is 1. For every point x Rn , S n , I n , there is a nbd U with an open set V which is homeomorphic to S n1 , and by induction the proof is complete In order to show that ind(Rn ) n, We shall have some denitions and theorems. Let's start by dening a partition between sets:

Denition 3.2. A Partition L on between A and B exists, if there are open sets U, W , A U, B W , such that W U = and L = U c W c . Theorem 3.7. ind(X ) n For every point x and every closed set B there
is a partition L, such that:ind(L) n 1. Proof. () If x X and B is closed, Then there is a set V , such that V B = . By denition of ind, there is a U V , x U , ind(U ) n 1 now, let W = U and L = U . () Let x X and V an nbd of x. The set B = V c is closed, and therefore there are U, W such that x U , B W . Note that by denition of B , U V . Now W c is closed, therefore U W c , and obviously U U W c , and by = U \ U U c. denition U = U \ U Therefore U U c W c = L ind(U ) n 1 subspace of X , then for all closed set A, B of X A B = , there is a partition between them L, such that L Z =

Theorem 3.8. If X is a metric space and Z is a zero dimensional separable

Theorem 3.9. X is a separable metrizable space, then ind(X ) n X is the union of two sub spaces Y, Z such that ind(Y ) n 1, ind(Z ) 0.
From this we can easily see, that for a separable metrizable space X ind(X ) n X is the union of Z1 , . . . , Zn+1 , ind(Zi ) 0. A nice result of this theorem, gives us an estimate on the dimension of the sum of two spaces.

Theorem 3.10. (The addition theorem) If X, Y , ind(X ) n, ind(Y ) m are separable metric spaces, then X Y can be represented with Zx,1 , . . . Zx,n+1 , ZY,1 . . . , ZY,m+1 . And therefore ind(X Y ) n + 1 + m + 1 1 = m + n 1, meaning that ind(X Y ) ind(X ) + ind(Y ) 1. Theorem 3.11. (The separation theorem) For every closed sets A, B of
a separable metric space X ind(X ) n, There is a partition L, such that ind(L) n 1.

Proof. We can decompose X into two spaces Y, ind(Y ) n 1; Z, ind(Z ) = 0. There is a partition L between A, B and L Z = L Y and ind(L) n 1 as a sub-space of Y.
5

A U , B W are open in X and U W = . For every partition L of M U, M W there is a partition L of A, B , such that M L L .

Lemma 3.12. M is a sub-space of metric X , A, B are closed and disjoint,

The last lemma says, that you can take a partition on M and extend it to a partition on X . We can use it to prove the following:

Theorem 3.13. (The second separation theorem) X is a metric space, M is a separable sub-space and ind(M ) n. Then For every disjoint closed sets A, B there is a partition L such that ind(L M ) n 1. Theorem 3.14. (Theorem on partitions) X is a separable metric space, ind(X ) n . Then For every sequence of (A1, , B1 ), . . . (An+1 , Bn+1 ) of closed disjoint sets, The partitions L1 , . . . , Ln+1 exists, such that they have an empty intersection.
Proof. We shall only prove () Let's look at A1 , B1 by the separation theorem, we can nd a L1 ,ind(L1 ) n 1. Lets look at A2 , B2 and let M = L1 . We can nd a partition L2 , such thatind(M L2 ) n 2. Lets look at Ai , Bi and let M = L1 L2 Li1 , we can nd a partition Li , such thatind(M Li ) = n i. When we get to toAn+1 , Bn+1 , we have ind(L1 L2 Ln+1 ) 1 L1 L2 Ln+1 = .
We are 2-3 theorems away from showing that ind(Rn ) = n

Theorem 3.15. (Brouwer xed point theorem) If S is a non-empty, compact, closed and convex sub set of Rn , then f : S S has a xed point.
Why do we need this theorem? you shall soon nd out!

Theorem 3.16. Let A1 B1 , . . . , An , Bn be the opposite faces of the I n cube

(meaning: the i-th coordinate of Ai is 0 and in Bi is 1). Now, if Li is a partition between Ai and Bi , then L1 Ln = .

Proof. As Li is a partition between Ai and Bi , by denition exists open sets Ui , Wi ; Ai Ui , Bi Wi . and Li = (Ui Wi )c . The following function is well dened: d(x,Li ) 1 1 x Wic 2 d(Ai ,x)+d(x,Li ) + 2 fi = d ( x,L ) 1 1 i 2 x Uic d(Bi ,x)+d(x,Li ) + 2
fi is continuous. We also have: f 1 ( 1 2 ) = Li and fi (Ai ) = {1} and fi (Bi ) = {0} (Note that the i-th coordinate of x Ai is 0, and of x Bi is 1). Lets dene f (x) = (f1 , . . . , fn ). 1 ,..., 1 if L1 Ln = then x f (x) = ( 2 2 ). 1 1 n Let p : I \ {( 2 , . . . , 2 )} be the projection from a point in the cube to its boundary (We stretch a line from the middle, to the point, until it reaches the boundary). Dene g = p f . We have that g (Ai ) Bi and g (Bi ) Ai and that g (I n )is on one of the Bi or Ai .That means that g (x) = x for every x. contradicting Brouwer xed point theorem.
6

Can you see where this is heading? We have already seen that I n n is true. let's show that I n n 1 is false. If I n n 1 is true, then it has we can nd n partitions of the faces that has an empty intersection. (the faces of I n are sequence of n disjoint closed sets, so we apply the Theorem on partitions) But by the previous theorem showed that any selection of partitions of the faces will always have a non-empty intersection! and we get:

ind(I n ) = n
And due to the sub-space theorem:

ind(Rn ) = n

This also shows us, that I n I m n = m.

4 The Large inductive dimension

One can see that for every separable metric space X with ind(X ) = n, for every closed subset F of every open subset U of X, there is an open V in between, such that the ind(V ) n 1, this suggest a modication in the denition of the small inductive dimension consisting in replacing the point x by a closed set A.

Denition 4.1. The large inductive dimension of normal space X is notated Ind(X ), and is dened as follows:
1. We say that the dimension of a space X (Ind(X )) is -1 i X = 2. Ind(X ) n if for every closed set C X and for every open set U exists an open V , C V such that V U , and Ind(V ) n 1. Where V is the boundary of V . 3. Ind(X ) = n if (2) is true for n, but false for n 1. 4. Ind(X ) = if for every n, Ind(X ) n is false.

Remainder: X is a normal space if, given any disjoint closed sets E and F, there are a neighborhood U of E and a neighborhood V of F that are also disjoint.

Proposition 4.1. A normal space X satises the inequality Ind(X ) n i for every pair A, B of disjoint closed subset of X exists a partition L between A and B such that Ind(X ) n 1
We have seen this before...

Theorem 4.2. For every separable space X we have ind(X ) = Ind(X )

Proof. For every normal space X we have ind(X ) Ind(X ) by denition. To show that Ind(X ) ind(X ) we will use induction with respect to ind(X ), clearly one can suppose that ind(X ) < . If ind(X ) = 1ind(X ) Ind(X ). Assume that the inequality is proven for all separable metric space X of ind(X ) < n and consider a separable metric space X such that ind(X ) = n. Let A and B be a pair of disjoint closed subset of X , according to the rst separation theorem there exists a partition L between A and B such that ind(L) n 1, by the inductive assumption for every k < n Ind(L) n 1 and according to the previous proposition Ind(X ) n and nally we got Ind(X ) ind(X ) Ind(X ) = ind(X ).

5 The Lebesgue covering dimension

Lebesgue covering dimension or topological dimension of a topological space is dened to be the minimum value of n, such that any open cover has a renement in which no point is included in more than n+1 elements, if a space does not have Lebesgue covering dimension m for any m, it is said to be innite dimensional. In this context, a renement is a second open cover such that every set of the second open cover is a subset of some set in the rst open cover. It is named after Henry Lebesgue, although it was independently arrived at by a number of mathematicians.

Example 5.1. consider some arbitrary open cover of the unit circle. This open cover will have a renement consisting of a collection of open arcs. The circle has dimension 1, by this denition, because any such cover can be further rened to the stage where a given point x of the circle is contained in at most 2 arcs. That is, whatever collection of arcs we begin with some can be discarded, such that in the remainder still covers the circle, but with simple overlaps. Similarly, consider the unit disk in the two-dimensional plane. It is not hard to visualize that any open cover can be rened so that any point of the disk is contained in no more than three sets.

5.1 Properties
Property 5.1. Lebesgue covering dimension is a topological property
two homeomorphic spaces have the same dimension.

Property 5.2. Rn has dimension n

Theorem 5.1. For every closed subspace M of a normal space X we have

dimM dimX

Proof. The theorem is obvious if dimX = , so that we can assume that dimX = n < . Consider a nite open cover U = Ui k i=1 of the space M. For i=1,2...,k let Wi be an open subset of X such that Ui = M Wi . The family X \M Wi k i=1 is an open cover of the space X and since dimX n it has a nite open renement which no point is included in more than n+1 elements of . one easily see that the family \M is a nite open cover of space M, renes U and has no point of M is included in more than n+1 elements of \M , so that dimM n = dimX .

Theorem 5.2. For X metrizable space IndX = dimX . Theorem 5.3. For X normal space dimX IndX

5.2 Some topological constructions

The denition of the Lebesgue covering dimension can be used to build some topological sets, such as the Sierpinski carpet. A construction can proceed as follows: Consider, for example, a nite open covering for the two-dimensional unit disk. This covering can always be rened so that no point in the disk belongs to more than three sets. Now, we will remove all of the points in the disk that belong to three sets. Depending on the renement, this will leave possibly one or more 9

holes in the disk. The remaining object is again two-dimensional, and again has a nite open cover. The process of selecting a cover and rening, and then punching out holes can be repeated, ad innitum. The resulting object is homeomorphic to the Sierpinski carpet. What is curious about this construction is that the carpet has a Lebesgue covering dimension of one, and not two, although at any step of the creation the shape had dimension of two. The proof of this is essentially by contradiction: were there a covering which required membership to three sets, then the aected area would have been punched out during the construction phase. Similar constructions can be performed in higher dimensions; the three-dimensional analogue is called the Menger sponge. Curiously, the Lebesgue covering dimension of the Menger sponge is again one.

10

6 Fractals
The word "fractal" denotes a shape that is recursively constructed or self-similar, that is, a shape that appears similar at all scales of magnication and is therefore often referred to as "innitely complex". There is also a mathematics fractal denition that we intrudes later. Some examples that are: 1. Cantor set 2. Sierpinski carpet Which we already see, more known examples are: 3. The Koch curve: one can imagine that it was created by starting with a line segment, then recursively altering each line segment as follows: (a) divide the line segment into three segments of equal length. (b) draw an equilateral triangle that has the middle segment from step 1 as its base. (c) remove the line segment that is the base of the triangle from step 2.

The Koch curve is in the limit approached as the above steps are followed over and over again
4. The Mandelbrot set: this set can be dened as the set of parameters c for which the critical point 0 of fc : C C; z z 2 + c does not tend to n n , That is: fc (0) where fc is the n-fold composition of fc with itself. This denition lends itself immediately to the production of computer generated renderings. Example to approximate fractals are easily found in nature. These objects display self-similar structure over an extended, but nite, scale range. Examples include clouds, snow akes, mountains, river networks, and systems of blood vessels. Famous example in this class is 5. The coast line of Britain.

6.1 Fractals dimension

We can observe the Cantor set as a key example to understanding fractals dimensions. The Cantor set has topological dimension of zero, but yet it has the same cardinality as the real line - in that sense we'd expect its dimension to be one. But the Cantor set has no interval in it - and in that sense we'd expect its dimension to be zero. The answer then, lies somewhere in the middle. The Cantor set should have a dimension greater than zero, but smaller than one.

11

7 Hausdor dimension
7.1 Introduction
The topological dimensions that we saw gives us a notion of geometrical dimension for topological space. And as we expect the topological dimension of Rn is indeed n. But there are some complex sets as we seen before that the topological dimensions seems too naive for. For this purpose, the Hausdor dimension (fractal dimension) was invented. The topological dimension were built on some topological notions, like the notion that boundary of the space has a dimension smaller by 1. This dimension has a dierent notion that is not topological. We will now look at Rn and observe another feature that involves it's geometrical dimension, to get an intuition of the Hausdor dimension. Imagine a square in R2 with side length of 1. its area is also 1. let's see what happens when we 'zoom-in':

2 = Comparing the two squares: Zoom Side Area Factor:

1 2 3 4 1 2 3 4 1 4 9 16

log (Area) log (Side)

2 2 2

Imagine a cube in R3 with side length of 1. its volume is also 1. let's see again what happens when we 'zoom-in' (Area) Zoom Side Volume Factor: log log (Side) 1 2 3 4 1 2 3 4 1 8 27 64 3 3 3

More generally, if we take the cube I n , and we 'zoom in' the space k times, the cube will grow by k n . As we can see this is another eect of the geometrical dimension. The Hausdor dimension denition is based on this notion of the dimension, and that's why it requires a metric (so we can measure the growth of the space). This dimension gives nice results for fractals, as we will demonstrate, but rst lets formalize the denition.

7.2 Formal denition

Note that the Hausdor dimension is dened only for metric spaces - it uses concepts like length and volume, that require a metric.

Remainder: Let X be a metric set and C be a collection of sub-sets. The

12

mesh of C is mesh(C ) = sup{diam(B ) : B C }.

The denition of the Hausdor dimension is done using a measure on the space, so we shall rst dene the Hausdor measure:

Denition 7.1. Let A be a subset of X , we annotate Hd (A, ) = inf {

C = {Bi }, mesh(C ) < } where C is a countable cover of A.

diam(Bi )d :

Remember that we are trying to use the geometrical notion to dene a new dimension. The sum diam(Bi ) can be thought of as the 'size' of a 1-d segment in the space. The sum diam(Bi )d is the volume of the number of d-dimensional cubes needed to cover the space. Now, to be able to use Hd comfortably, we dene a measure:

Denition 7.2. The Hausdor p-measure Md is Md (A) = sup{Hd (A, )} Note: a < b Hd (A, a) Hd (A, b), so the following is also true:
Md (A) = lim
0 Hd (A,

. This measure 'tells' the 'volume' of A if it was in a d dimensional space.

Denition 7.3. The Hausdor dimension of A, is dimH (A) = sup{d : 0 <

Md (A) < }.
This denition is quite natural - the dimension of A is the highest dimension that A has a nite 'volume'. Note that if p d, then Mp (A) Md (A).

Theorem 7.1. If 0 < Md (A) < then dimH (A) = d.

Proof. We shall rst show that if p < d then Mp (A) = . Md (A) > 0 there is a 1 > > 0, t > 0 such that Hd (A, ) = t, This is due to the denition of Hd . And by denition, for every < we have Hd (A, ) t. Let's choose a , so the following inequality holds: dp < t/N ; < , Where N is an arbitrary number. Let C = {Bi }, mesh(C ) < . Now, () Remember that x < y x1 > y 1 . diam(Bi )p = diam(Bi )pd diam(Bi )d pd diam(Bi )d pd Hd (A, ) (N/t)t = N . For every N , there is an , diam(Bi )p does not converge, for any cover C with mesh(C ) . This gives us that Mp (A) = lim 0 Hd (A, ) = . Now if d < p, then by the above proof, it implies that Md (A) = , contradicting the assumption. Therefore, the proof is complete.
This makes sense - if we were to try to cover a square (2-d shape) with lines (1-d shape) we would need innity of lines. As we said, the measure gives us the 'volume' of the set. If d is the dimension, Then if we try to measure a set A with a dimension p < d then its p dimensional volume will be . The last theorem gives us quite a lot help on determining the dimension of a space. when we nd one d that has a non-zero Hausdor measure we found the dimension. 13

7.3 Examples and Computations

7.3.1 What is the Hausdor dimension of a countable set?
Let A = a1 , a2 , ..., an be a nite subset of a pseudometric space (X, d). Suppose that d(aj , ai ) > 0 for i = j . Then, we can see that M0 = n dimH (A) = 0.

7.3.2 What is the Hausdor dimension of an interval ?

Let X be the space of the real numbers with the usual pseudometric, and let A = [a, b]. We will show that M1 (A) = b a which will leads us to dimH (A) = 1. We rst show that M1 (A) b a by showing that M1 b a + for a) every > 0. Let > 0, and let N be an integer such that h = (bN < 2 . For i = 1, 2, 3..., N , let xi = a + ih. We dene an open cover C = {Ci : i = 1, 2..., N } for A as follows (we may assume < b a): C1 = [a, x1 + N ) ) Ci = (xi (2N ) , xi+1 + (2 N) ) C N = (x N 1 N , b] for each i we have diam(Ci ) = h + N < 2 + 2 = . Thus m(C ) < and ) + (h + N ) = N h + = b a + . diam(Ci ) = (h + N ) + (N 2)(h + N Therefore, M1 (A) b a + for every > 0, which implies that M1 (A) b a. Now we will show that M1 (A) b a. Let C = {Cj : j Z+ } be an open cover of A. Because A is compact there is a Lebesgue number > 0 for C ; that is, whenever x, y A such that |x y | < , than Ci C that x, y Ci . Let N be a positive integer such that a) h = (bN < , and dene xi = a + ih for i = 0, 1, 2..., N . Then for each i there is a Cj (i) so that xi1 and xi in Cj (i) ; this implies that xi xi1 = h diam(Cj (i) . Hence, N diam(Cj (i) ). b a = i=1 (xi xi1 ) < were the summation is over distinct values j (i). Therefore ba i=1 diam(Cj ) for every countable open cover C ; consequently, M1 (A) b a. Thus M1 (A) = b a.

And in general the Hausdor dimension of a n-dimension surface is n.

As we can see, The Hausdor dimension of these sets is intuitive, and resembles the geometrical dimension and the topological dimension. These examples are important, because they give us the 'right' to call the Hausdor dimension a dimension (and not 'The Hausdor strange function')

7.4 Fractals
Now that we grasped the Hausdor dimension, it's time to see some uses of it in the fractals eld. The Hausdor dimension used to formalize the denition of fractals. is dierent from its topological dimension (dim). Some dene a fractral as a set with non-integer Hausdor dimension.

Denition 7.4. The set A is a said to be a Fractal if its Hausdor dimension

14

7.4.1 Examples
If we nd a nice enough fractal, we can calculate its Hausdor dimension with ease using the geometrical notion of the Hausdor dimension. The dimension of a fractal is intuitively, The factor between the zoom, and self resmbling parts) the number of self resembling parts after the zoom log(Smallerlog . (Zoom) Lets give an example with a common fractal The Cantor set. self resmbling parts) Zoom Number of smaller self resembling parts Factor: log(Smallerlog (Zoom) 1 3 9 1 2 4
log (N ) log (r )

0.630929754 0.630929754

D=

D=

log (2) log (3)

= 0.63.

We got an object with Hausdor dimension dierent from his topology dimension. We can also notice The Cantor set dimension is between a point (dimensionality 0) and a line (dimensionality 1) just like we would expected. Another well known and loved fractal example is The Koch curve We can see that the length of the curve increases with each iteration, so it has innite length. This is a good example of a bound shape of innite length. But (N ) we still can obtain the Hausdor dimensions from the formula D = log log (r )

D=

log (4) log (3)

D = 1.26

15

References
[1] R. Engelking, Dimension theory, 1978. [2] Wikipedia, www.wikipedia.org [3] G.L. Cain, Introduction to general Topology, 1993.

16