You are on page 1of 32

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/305084244

Hausdorff Dimension of Cantor Sets

Thesis · June 2012


DOI: 10.13140/RG.2.1.2160.7920

CITATION READS

1 1,071

1 author:

Shaimaa Monem
Cairo University
2 PUBLICATIONS   2 CITATIONS   

SEE PROFILE

All content following this page was uploaded by Shaimaa Monem on 09 July 2016.

The user has requested enhancement of the downloaded file.


Hausdorff Dimension of Cantor Sets

Shaimaa Mohamed Abdelmonem (shaimaamonem@aims.ac.za)



Ѫ J Ü Ï @YJ. « YÒ m × ZAÒ J
ƒ
African Institute for Mathematical Sciences (AIMS)

Supervised by: Prof. Alan Beardon


Cambridge University, UK

17 May 2012
Submitted in partial fulfillment of a postgraduate diploma at AIMS
Abstract
This essay is a trial to determine conditions which characterize a local dimension function. In fact,
the problem reduces to the construction of a Cantor set with given local dimension function: the local
dimension at any point of the Cantor set is given by the value of the function at its homeomorphic
image in the universal Cantor set.

Declaration

I, the undersigned, hereby declare that the work contained in this essay is my original work, and that
any work done by others or by myself previously has been acknowledged and referenced accordingly.

Shaimaa Mohamed Abdelmonem, 17 May 2012

i
Contents

Abstract i

1 Introduction 2
1.1 Aims . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Organization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

2 Definitions and Preliminaries 3


2.1 Topological Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2 Hausdorff Measure and Dimension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.3 General Cantor Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.4 Universal Cantor Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

3 The Hausdorff Dimension of The General Cantor Set 10


3.1 The Global Hausdorff Dimensions of The General Cantor Set . . . . . . . . . . . . . . . 10
3.2 Local Hausdorff Dimension of a General Cantor Set . . . . . . . . . . . . . . . . . . . . 14

4 General Cantor Sets Correspond to Upper Semi-continuous Functions 17


4.1 Construction of a Cantor set Correspond to an Upper Semi-continuous Function . . . . . 17

5 Conclusion 23

References 26

1
1. Introduction
The name fractals are first proposed by Mandelbrot [AA08], the first to mention fractals in his paper
about self-similarity [Man67] followed by two books [Man83, Man04]. Fractals then have been widely
studied [Fal90, Fal02] and have been applied in many fields such as: Physics and cosmology [MC09],
analysis on manifolds [Str99], fluid mechanics [Tur88], even the natural shapes of our blood vessels and
the lung are fractals. Fractals are signed by their dimensions; their Hausdorff dimension exceeds the
topological dimension.
Felix Hausdorff was the founder of the important topics of Hausdorff measure and Hausdorff dimension;
also who improved the notion of fractal dimension in 1919. The concept that allows fractional dimensions
and that has been, intensively, used in fractal theory [Bea65a, McM84, Fal90, Fal02], self-similar sets
and Brownian motion [J.S09, Han]. Cantor set is one of the fractals that has a simple important
structure and that present many of the fractals’ properties and appears, frequently, in space-filling curves
[Sag86, Sag92, SW03] , Quantum applications [Nas94], modern function theory [For51, Tsu59, Bea65b].

1.1 Aims

The generalization of the Cantor set was, deeply, studied in detail by [Bea65a]; an accurate estimation of
the lower and upper bounds of the Hausdorff dimension of the general Cantor set was given. Moreover,
generalizations for some of the properties of the local dimension function which was introduced by
[RT59] were given. In [Bea65a], Beardon questioned about the sufficient condition on a function to
be a local dimension function for some general Cantor set. In this work we are trying to answer that
question after representing some of the work discussed in the paper.

1.2 Organization

The essay is divided into four chapters: Chapter 2 has four sections for the definitions and preliminaries
on which all the essay body is based. The topological spaces section defines the product topology;
followed by a section for the definition of the Hausdorff dimension that can be considered as the main
topic in this work; the last two sections give the definitions of the general and the universal Cantor Sets
and prove the homeomorphism between them after studying their constructions.
The next, chapter 3, consists of two sections: the first, studies the relation between the Hausdorff
dimension for the Cantor set and its fundamental intervals, also determines the Hausdorff dimension
for the self-similar Cantor set ; in the second, the local Hausdorff dimension is introduced, followed
by a short study of some its properties like continuity and upper semi-continuity with suitable counter
examples or proofs. In chapter 4, we try to determine the sufficient condition for a function to be a local
dimension function. Our result in this chapter is a particular case of a result in [Ols05] that we have
discovered after proving our results, and so the method we use is entirely independent of the methods
in the paper. And closing by the conclusion in chapter 5.

2
2. Definitions and Preliminaries
All basics which are needed for the presented work are given in this chapter. We move between two
mathematical fields in order to discuss both global and local dimensions of the general Cantor sets,
which is the type of fractals that the topic is speaking about. Measure theory presenting the definitions
of measures, convergence of limits and dimensions; as well as topology for using the concept of product
topology between topological spaces1 .
We start this chapter by a short review Section (2.1) which introduces some topological concepts. In
the following sections three concepts are covered, which are: The Hausdorff dimension which introduced
in the second Section (2.2) and both the general and universal Cantor sets in sections (2.3) and (2.4)
respectively.

2.1 Topological Spaces

2.1.1 Definition (Projection functions). Let {Xi }i∈A be a sequence of non empty sets for some A ⊆ N.
Define the set X as the product Y
X= Xi , (2.1.1)
i∈A

where any point x ∈ X can be presented as x = (x1 , x2 , . . . , xi , . . .). Then, the projection function
πi : X −→ Xi is defined for all x ∈ X by

πi (x) = xi ∈ Xi . (2.1.2)

2.1.2 Definition (Product


Q Topology). If (Xi , τi ) is a topological space for all i ∈ A, then the product
topology τ on X = i∈A Xi is the smallest topology on X for which all the projection functions πi are
continuous.
Q
It is worth mentioning that the set O = i Oi ⊂ X is basis if and only if each Oi is open in Xi and
Oi = Xi except for finitely many i.

2.2 Hausdorff Measure and Dimension

Measure is a generalization of the idea of length. Measures in R are the lengths of the one dimensional
intervals, in R2 are the areas of two dimensional subsets, in R3 are volumes of the three dimensional solid
objects and so on. Indeed, measure is determining the size of the objects. The Hausdorff dimension is a
big evolution in the measure concept; contrary to the Lebesgue measure, the Hausdorff measure value
is not restricted to integer values and can be defined on any metric space. The present section shows
the procedures of defining the Hausdorff dimension and is mainly influenced by [Bea65a] and [Fal90].
2.2.1 Definition (Outer Measure). An outer measure on a set X is a function ϕ : 2X −→ [0, ∞]
defined on the power set 2X of X, and satisfies the following:
1
It is important to note that many concepts as continuity, convergence, connectedness also some definitions like open
and closed sets and more can be stated mathematically using different methods depending on the description of the
underlying space (i.e. metric, topological, ...). In this topic both of the methods are used; depending on the what is needed
to reach at each piece

3
Section 2.2. Hausdorff Measure and Dimension Page 4

• ϕ(φ) = 0;

• For all subsets A and B of X, A ⊆ B implies ϕ(A) ≤ ϕ(B) “Monotonicity”;

• For all disjoint subsets Ai of X, ϕ(∪i Ai ) ≤ ∪i ϕ(Ai ) “Countable subaddivity”;

• A set E of X is called measurable if and only if for any set A of X, ϕ(A) = ϕ(A∩E)+ϕ(A\E).2

It is important to note that the considered space is Euclidean space RN of dimension N ; the general
concept of the outer measure with respect to any increasing real function h(t) with the property h(0) = 0
was defined by Hausdorff [Hau19]. The outer measure is continuous and increasing for t ≥ 0. In
particular for h(t) = tα , α, δ > 0 and any set E let
( )
X [
α α
mδ (E) = inf |In | |E ⊆ In , and for all n ∈ N, |In | ≤ δ , (2.2.1)
n∈N n∈N

where |In | = sup{|x−y| : x, y ∈ In } denotes the diameter of the set In ; also {In }n is called a δ-covering
for E, and the infimum is taken over all the arbitrary δ-coverings of E.

2.2.2 Definition (Hausdorff Outer Measure). We define the Hausdorff outer measure mα (E) of a set
E with respect to h(t) = tα , by

mα (E) = lim mαδ (E). (2.2.2)


δ→0

The existence of the limit follows because mαδ (E) is a decreasing function of δ. The measure mα (E)
can be defined and computed from open coverings of E; in particular if E is compact3 , then we can
consider finite open coverings of E. The Hausdorff dimension of the set E is totally determined by
the Hausdorff outer measure mα (E) given by Definition (2.2.2); this is clearly stated in the following
theorem.

2.2.3 Theorem. [Fal90] There exists a unique positive number, d(E) called the Hausdorff dimension
of the set E such that (
∞ if α < d(E),
mα (E) = (2.2.3)
0 if α > d(E).

Proof. First, show that if ms (E) < +∞ and t > s then mt (E) = 0. Suppose that ms (E) < +∞, and
let M = 1 + ms (E).
Then there is a positive δ0 such that if 0 < δ < δ0 then msδ (E) < M .
By
P Equation (2.2.1), for every δ in the interval (0, δ0 ) there is a δ-covering {Ui } of E such that
|U |s < M.
n n

Now take t > s . Then


X X X
|Un |t = |Un |s |Un |t−s ≤ |Un |s δ t−s ≤ M δ t−s .
n n n
2
The ϕ-measurable sets form a σ-algebra Σ and ϕ restricted to the measurable sets is a countably additive measure,
and we call the triple (X, Σ, ϕ) a measure space.
3
If E is compact, then each of its open coverings has a finite open sub-covering
Section 2.3. General Cantor Sets Page 5

As t − s > 0 we let δ → 0 and we see that mt (E) = 0.


Now suppose that mr (E) > 0. If 0 < p < r then mp (E) = +∞ (because if mp (E) < +∞ we would
have mt (E) = 0 and this is not true).
It follows that there is a unique number d(E) such that mt (E) = +∞ if t < d(E), and mt (E) = 0 if
t > d(E).

2.2.4 Definition (Hausdorff Dimension). We define the Hausdorff dimension of a set E by

d(E) = inf{s : ms (E) = 0}, (2.2.4)


s
= sup{s : m (E) = ∞}. (2.2.5)

2.3 General Cantor Sets

This section is producing the definition of the general Cantor set which is one of the early fractals appear
in the worlds of mathematics, engineering, dynamical systems, physics and computer science. Henry
Smith discovered the Cantor set in 1875 and Georg Cantor introduced it in 1883 [G.C83].
Some fractals are described to be self-similar, in other words, detailed patterns with an iterating property.
Any magnified detailed part of the fractals pattern looks like the original whole pattern i.e. they repeat
themselves on the infinitesimal small scales. Following [Bea65a], this section introduces a fractal example
called general Cantor set.

2.3.1 Definition (General Cantor Set). A set E is called a general Cantor set if and only if E can be
written in the form
\∞
E= En , (2.3.1)
n=1

where En consists of kn mutually disjoint nonempty connected compact sets Ii1 ,...,in 4
[
En = Ii1 ,...,in , (2.3.2)
i1 ,...,in

and i1 , . . . , in ∈ {1, . . . , k} for a positive integer k ≥ 2 such that

(i) Ii1 ,...,in ,in+1 ⊂ Ii1 ,...,in and i1 , . . . , in , in+1 ∈ {1, . . . , k}.

(ii) There exists A and B real numbers, 0< A, B < 1 such that

|Ii1 ,...,in ,in+1 | ≥ A|Ii1 ,...,in |, (2.3.3)

and for t 6= s
ρ (Ii1 ,...,in ,s , Ii1 ,...,in ,t ) ≥ B|Ii1 ,...,in |, (2.3.4)
where ρ is the distance function: for any two sets A and B

ρ (A, B) = inf{|a − b| : a ∈ A and b ∈ B}. (2.3.5)


4
A general Cantor set in RN is called a spherical Cantor set if and only if for all sequences i1 , . . . , in Ii1 ,...,in is an
N -dimensional disk
Section 2.3. General Cantor Sets Page 6

Indeed, Ii1 ,...,in is called a fundamental interval for any possible sequence i1 , . . . , in ∈ {1, . . . , k}, also a
fundamental system, F , of E is a finite disjoint collection of fundamental intervals whose union covers
E.
A very important concept is the level of the fundamental interval; the level of Ii1 ,...,in is n, in other
words, the level gives the stage during the construction of the Cantor set in which the interval Ii1 ,...,in
appears. So, the upper and lower orders of a fundamental system are the maximum and minimum values
of the level of fundamental intervals in F . Moreover, if F consists of fundamental intervals of the same
level, then it shall be called En i.e. a fundamental system of level n.

2.3.2 Lemma. If we define δn = max |Ii1 ,...,in |, then for any general Cantor set δn → 0 as n → ∞.
Where the maximum is defined on all the possible sequences i1 , . . . , in ∈ {1, . . . , k} for the fixed level
n.

Proof. Let δn = |Ii∗1 ,...,in | be as defined above and let {jl }l be a sequence of indexes, which can be
defined inductively, such that for all m there exist infinitely many values of n for which

Ij1 ,...,jm ⊇ Ii∗1 ,...,in (2.3.6)

Now suppose that δn ≥ δ ≥ 0, then it follows that

|Ij1 ,...,jm | ≥ |Ii∗1 ,...,in | ≥ δn ≥ δ (2.3.7)

Because the sequence {jl }l can be determined, a sequence {il }l can be found such that for all l il 6= jl .
Pick xn ∈ Ij1 ,...,jn ,in+1 , then for all m > n with xm ∈ Ij1 ,...,jm ,im+1

|xn − xm | ≥ ρ Ij1 ,...,jn ,in+1 , Ij1 ,...,jm ,im+1 (2.3.8)

≥ ρ Ij1 ,...,jn ,in+1 , Ij1 ,...,jn ,jn+1

≥ B|Ij1 ,...,jn |

≥ Bδ (2.3.9)

{xn }n is a bounded infinite sequence or set, because Ij1 ,...,jn ,in+1 are infinite and bounded in [0, 1].
Then we see that δ = 0 and because E is compact 5 , the lemma follows on observing that

δ1 ≥ δ2 ≥ · · · ≥ δn ≥ · · · ≥ 0 (2.3.10)

5
Any point in E is a limit of a sequence in E
Section 2.4. Universal Cantor Sets Page 7

2.4 Universal Cantor Sets

The universal6 Cantor set is another point of view for the General Cantor set; one of the other methods
to represent the Cantor set points by using the infinite sequence of indexes of the fundamental intervals
that the point belongs to. In this section we prove the homeomorphism between the general and the
universal Cantor sets.

2.4.1 Definition (Universal Cantor Set). Let U be the space of all sequences x = (x1 , x2 , . . .) in which
each xj for all j can take one of the values in {0, 1}. The set U is called the universal Cantor set.

2.4.2 Short Study of the Universal Cantor Set. The universal cantor set is uncountable. Assume
that U is countable. Then U can be written in the form U = {u1 , u2 , . . . , un , } where

u1 = u11 , u12 , . . . ,

u2 = u21 , u22 , . . . ,
..
.
un = (un1 , un2 , . . .) , (2.4.1)
..
.

now choose a sequence y = (y1 , y2 , . . .) such that for all j ∈ N yj ∈ {0, 1} and yj 6= ujj . Then y ∈ U
but y ∈
/ {u1 , u2 , . . . , un , }. Hence, U is uncountable.
But U has a countable subset U0 = {x ∈ U : such that ∃n for which xn = xn+1 = . . .} i.e. U0
contains the sequences have constant tails.
Let U m = {x ∈ U : xm = xm+1 = . . .}, then U m consists of exactly 2m elements. Therefore,
U0 = ∪Q∞
m=1 U
m is countable because it is a union of countable set. Also the set U is the infinite product

space n=1 Un , where each Un is the the discrete space {0, 1} which is the discrete topology7 . This
means that U can be given the product topology defined as follows:
(∞ )
Y
B= Ol : Ol open in Ul and Ol = Ul except for finitely many l . (2.4.2)
l=1
Q∞
2.4.3 Lemma. For all k ∈ N. The universal Cantor set U is homeomorphic to n=1 {1, 2, . . . , k}.

Q
It is worth mentioning that ∞ n=1 {1, 2, . . . , k}, equivalently, the base k representation of real numbers
in the interval [0, 1]. The next step is showing the strong normal connection between both the universal
and the general Cantor sets.

6
“The Cantor set is sometimes regarded as ”universal” in the category of compact metric spaces, since any compact
metric space is a continuous image of the Cantor set; however this construction is not unique and so the Cantor set is not
universal in the precise categorical sense” [Wil68].
7
In the discrete topology; every subset is both open and closed.
Section 2.4. Universal Cantor Sets Page 8

2.4.4 Theorem. For k ∈ N fixed, there is a natural homeomorphism φ from U to a general Cantor set
E constructed as in Definition (2.3.1), and this homeomorphism is defined for each x = (x1 , x2 , . . . )
by:

\
φ(x) = {Ii1 ,i2 ,...,in |x ∈ Ii1 ,i2 ,...,in } . (2.4.3)
n=1

Proof. First show that φ is a bijection. Assume that there exist two distinct points x and z in U such
that φ(x) = φ(z). Then

φ(x) = φ(z), (2.4.4)



\ ∞
\
⇒ Ii1 ,i2 ,...,in = Ii1 ,i2 ,...,in , (2.4.5)
n=1 n=1
x∈Ii1 ,i2 ,...,in z∈Ii1 ,i2 ,...,in

⇒ lim
n→∞
Ii1 ,i2 ,...,in = lim
n→∞
Ii1 ,i2 ,...,in , (2.4.6)
x∈Ii1 ,i2 ,...,in z∈Ii1 ,i2 ,...,in

which contradicts that x 6= z, because, there exists an integer m for which x ∈ Ii1 ,i2 ,...,im ,l and
z ∈ Ii1 ,i2 ,...,im ,t for some l, t ∈ {0, 1} and l 6= t. Therefore,φ is a bijection. It is obvious that x is given
by the representation x = (i1 , i2 , . . . , in , . . .), where for all n integer, x ∈ Ii1 ,i2 ,...,in . Recalling that U is
the product topology given by Equation (2.4.2) and E has the Euclidean topology given by

T = {E ∩ (a, b) : (a, b) ⊂ R}, (2.4.7)

Now to show that φ and φ−1 are continuous. Let (a, b) ∩ E ∈ T for some 0 < a < b < 1. Then

φ−1 ((a, b) ∩ E) = {x ∈ U : φ(x) ∈ (a, b) ∩ E}, (2.4.8)


 

 

 ∞
\ 
= x∈U : Ii1 ,i2 ,...,in ∈ (a, b) ∩ E ,

 

 n=1 
x∈Ii1 ,i2 ,...,in

= {x ∈ U : there exists m such that x ∈ Ii1 ,i2 ,...,im ⊂ (a, b) ∩ E},


( ∞
)
Y
= x ∈ U : x ∈ {i1 (x)} × {i2 (x)} × · · · × {im (x)} × Un , (2.4.9)
n=1
for some sequence {il (x)}m
1 and an integer m.

So φ−1 ((a, b) ∩ E) is open for any arbitrary open ((a, b) ∩ E) ∈ T which implies that φ is continuous.
Now to show that φ−1 is Qcontinuous we need to show that the image of each open set in B is also open
in T . Pick an open set ∞ l=1 Ol ∈ B. Then there exists a sequence j1 , j2 , . . . , jd ∈ {0, 1} for some
integer d such that

Y ∞
Y
Ol = {j1 } × {j2 } × · · · × {jd } × Un (2.4.10)
l=1 n=1
Section 2.4. Universal Cantor Sets Page 9

This implies

!
Y
φ Ol = {y ∈ E : y ∈ Ij1 ,j2 ,...,jd } , (2.4.11)
l=1
= {y ∈ E : y ∈ (a − δ, b + δ)} for some δ > 0,
= {y ∈ (a − δ, b + δ) ∩ E} , (2.4.12)

where a and b are the left and right end points for the interval Ij1 ,j2 ,...,jd respectively. So, φ−1 is
continuous. Hence, φ is a homeomorphism.

It should be noted that the points in U0 correspond to the points of E that are endpoints of some
fundamental interval. To show this, assume that x is an end point of some fundamental interval, then
there exists a sequence i1 , i2 , . . . , il ∈ {0, 1} such that

x ∈ Ii1 ,i2 ,...,il−1 ,1 or x ∈ Ii1 ,i2 ,...,il−1 ,k . (2.4.13)

If x ∈ Ii1 ,i2 ,...,il−1 ,1 , then x will remain a left endpoint for all the fundamental intervals Ii1 ,i2 ,...,it with
t ≥ l which means that il = il+1 = · · · = 1. Similarly, if If x ∈ Ii1 ,i2 ,...,il−1 ,k , then il = il+1 = · · · = 1
and x will remain a right endpoint for all the upcoming fundamental intervals contain it.
It follows that any map f : E −→ R can be represented by a map F = f φ from U to R. Moreover, f
is continuous if and only if f φ is continuous.
3. The Hausdorff Dimension of The General
Cantor Set
The measure of a fractal mathematical set can’t be determined the usual way because it is nowhere
differentiable. Fractals have an integer topological dimension1 [HW41] , although some fractals seems
to fill higher dimensional spaces. The Hausdorff dimension is used, intensively, in the theory of fractals.
Hausdorff dimension, unlike the Lebesgue one, can have fractional values and can also be defined locally
at each point of the set. This chapter has been influenced by [Bea65a].
The main studied points in the present chapter are: Starting by the global Hausdorff dimension Section
(3.1) ; firstly, a Lemma (3.1.1) to explore the possibility of determining the Hausdorff global dimension
of the general Cantor sets depending only on the fundamental systems. Secondly, the lemma is sup-
ported by a Theorem (3.1.2) that studies the summation given in 2.2.1 for the measure definition, but
specifically for fundamental intervals. This followed by a new presented theorem determining the global
Hausdorff dimension for a special case of general Cantor sets which have a condition on the scaling
of the fundamental intervals (3.1.3). Finally, the definition of the local Hausdorff dimension is given
followed by answering some questions related to its properties in Section (3.2).

3.1 The Global Hausdorff Dimensions of The General Cantor Set

The following lemma shows that the Hausdorff dimension d(E) of a general Cantor set E is determined
by the diameters of the fundamental intervals used in the construction of E.
3.1.1 Lemma. Let E be a general Cantor set and suppose that Mδα (E) and M α (E) are defined as for
mαδ (E) and mα (E) in Equation (2.2.1) and Definition (2.2.2) respectively; where Mδα (E) and M α (E)
are computed using only the sets Ii1 ,i2 ,...,in in the coverings of E. Then

M α (E) ≥ mα (E) ≥ B α M α (E), (3.1.1)

and so M α (E) can be used to define d(E).

Proof. Without loss of generality the considerations can be restricted to finite open covering because
E is compact, say a δ-covering {G1 , G2 , . . . , Gq } of E, such that for all l = 1, . . . , q

Gl ∩ E 6= ∅. (3.1.2)

Because δ tends to zero we can say that

|Gl | ≤ min ρ (It , Is ) for all l = 1, . . . , q and forall t 6= s ∈ 1, . . . , k, (3.1.3)

which means that Gl intersects with precisely one fundamental interval of level one. From Lemma
(2.3.2) we see that for n large enough, an arbitrary neighborhood of any arbitrary point in Gl ∩ E
contains at least two fundamental intervals of level n.
Let s(l) be the largest integer such that Gl has a non-empty intersection with exactly one fundamental
interval of levels 1, 2, . . . , s(l). This means that for all s > s(l) Gl intersects at least two fundamental
1
Lebesgue covering dimension or topological dimension of a topological space is defined to be the minimum value of
n, such that any open cover has a refinement in which no point is included in more than n + 1 elements.

10
Section 3.1. The Global Hausdorff Dimensions of The General Cantor Set Page 11

intervals of level s. Assume that the latter fundamental interval is Ii1 ,...,in(l) of level s(l) . Gl has
a non-empty intersection with two fundamental intervals of level s(l) + 1, both of them contained in
Ii1 ,...,in(l) . Then
 
|Gl | ≥ ρ Ii1 ,...,in(l),t , Ii1 ,...,in(l),t+1 for some t = 1, . . . , k − 1; (3.1.4)
≥ B |Ii1 ,...,in(l) |. (3.1.5)

Therefore, E ∩ Gl ⊂ E ∩ Ii1 ,...,in(l) . Hence,

E ⊂ ∪ql=1 Ii1 ,...,in(l) , (3.1.6)

which means that the covering {G1 , G2 , . . . , Gq } of E can be replaced by a covering consisting of only
fundamental intervals {Ii1 ,...,in(l) }ql=1 and
q
X q
X
α α
|Gl | ≥ B |Ii1 ,...,in(l) |α . (3.1.7)
l=1 l=1

It can be deduced that,


mα (E) ≥ B α M α (E). (3.1.8)
Note that from the definition of mα (E) we always have

M α (E) ≥ mα (E). (3.1.9)

So, finally
M α (E) ≥ mα (E) ≥ B α M α (E). (3.1.10)

Lemma (3.1.1) shows that the measure M α (E) instead of mα (E) to define and determine d(E). Indeed,
the Hausdorff dimension may be determined using the coverings consisting of fundamental intervals i.e.
fundamental systems are enough.

3.1.2 Theorem. Let E be a general Cantor set. Then at any level n


Pk α
(i) If for all sequence i1 , . . . , in = 1, . . . , k, j=1 |Ii1 ,...,in ,j | ≤ |Ii1 ,...,in |α , then d(E) ≤ α.
Pk β
(ii) If for all sequence i1 , . . . , in = 1, . . . , k, j=1 |Ii1 ,...,in ,j | ≥ |Ii1 ,...,in |β , then d(E) ≥ β.

P
Proof. (i) The hypothesis says that kj=1 |Ii1 ,...,in ,j |α is bounded. Also from Lemmas (2.3.2) and
(3.1.1) for any general
 Cantor set δn tends to zero as n tends to infinity. So M α (E) =
limδn →0 Mδαn (E) is bounded. Therefore mα (E) is also bounded because mα (E) ≤ M α (E).
Hence, d(E) ≤ α, because mα (E) = ∞ only for α ≤ d(E).

(ii) Consider a fundamental system F = {Ii1 ,...,in(l) }ql=1 that covers E. Suppose m and n are the upper
and lower levels of F respectively. So, particularly, there exists a sequence i1 , . . . , im = 1, 2, . . . , k
such that Ii1 ,...,im ∈ F . By the disjointness of F

Ii1 , Ii1 ,i2 , . . . , Ii1 ,...,im−1 ∈


/ F. (3.1.11)
Section 3.1. The Global Hausdorff Dimensions of The General Cantor Set Page 12

Therefore, by the covering property of F

Ii1 ,...,im−1 ,j ∈ F ∀j = 1, 2, . . . , k. (3.1.12)

If not there will be a gap in the covering. This allows us to replace these k intervals by the set
Ii1 ,...,im−1 and this leads to a new covering F ∗ of E.
P
Hypothesis (ii) states that kj=1 |Ii1 ,...,in ,j |β ≥ |Ii1 ,...,in |β which implies that

X k
X X
|Ii1 ,...,in ,j |β ≥ |Ii1 ,...,in |β . (3.1.13)
i1 ,...,in =1,...,k j=1 i1 ,...,in =1,...,k

Hence, X X
|I|β ≥ |I|β . (3.1.14)
I∈F I∈F ∗
By repeating this step many times the new upper level may still m or be m − 1 for the new
covering F ∗ so we repeat the step and replace the k fundamental intervals which have the upper
level by its previous fundamental interval of level, and so on till we reach the lower level n for any
n < m. This application yields

X k
X
|I|β ≥ |Ii1 ,...,in |β . (3.1.15)
I∈F i1 ,...,in =1

Keep looking for the new upper level of the new covering at each step and replace the k funda-
mental intervals which have the upper level by its previous fundamental interval of level, and so
on till we reach the first level. Then
X k
X
|I|β ≥ |Ij |β > 0. (3.1.16)
I∈F j=1

So, X
M β (E) ≥ |I|β > 0. (3.1.17)
I∈F

Hence, β ≥ d(E), because M β (E) = 0 only for β ≤ d(E).

Pk α
The following theorem studies the case when j=1 |Ii1 ,...,in ,j | = |Ii1 ,...,in |α if and only if d(E) = α.
3.1.3 Theorem. Let E be a general Cantor set with the property that for all j ∈ {1, 2, . . . , k} and at
any level n there is a scaling condition on the fundamental intervals such that

|Ii1 ,...,in | = mj |Ii1 ,...,in ,j |. (3.1.18)

Then, the Hausdorff dimension d(E) of E is determined by


     d  
1 d 1 d 1 1 d
+ + ··· + + ··· + = 1, (3.1.19)
m1 m2 mj mk

and 0 < (mj )−1 < 1 is called the negative magnification scale used in the Cantor set construction. In
this case the Cantor set is describe as self-similar.
Section 3.1. The Global Hausdorff Dimensions of The General Cantor Set Page 13

Proof. Showing that d(E) ≤ d is easy, from Equation (2.2.1), also both the definitions (2.2.2) and
(2.2.4), it is obvious that d(E) is less than msδ (E) for any arbitrary δ-cover which allows us to choose
a fundamental system En for this propose. For proving that d(E) ≥ d Lemma (3.1.1) ensures allows to
determine d(E) using only fundamental intervals. Therefore, the task is to cover E with fundamental
intervals to obtain the correct bound. Recall that for any level n En is a covering of E, let Em be one
of them. Then Em consists of k m fundamental intervals Ii1 ,...,im i1 , . . . , im = 1, . . . , k and

X    d   !n
d 1 d 1 1 d
|Ii1 ,...,im | = + ··· + + ··· + , (3.1.20)
m1 mj mk
i1 ,...,im =1,...,k

= 1. (3.1.21)

Therefore, X
mα (E) ≤ |Ii1 ,...,im |d = 1 (3.1.22)
i1 ,...,im =1,...,k

Hence, by Theorem (3.1.2), it can be deduced that

d(E) ≤ d. (3.1.23)

Note that the summation in Equation (3.1.20) is a difficult one, Figure 3.1 follows to simplify the idea
of calculating the sum.

E0 s| }s
0 1
{z
1
E1 |s }s s| }s
d d
{z {z m−1
1 + m−1
2 =1
m−1 m−1
E2 |s {z } s s| {z }s s| {z }s |s {z } s
1 2  d  d 2
m−1
1 + m−1
2 =1
m−2
1 m−1 −1
1 m2 m−1 −1
2 m1 m−2
2
.. .. .. ..
. . . .

Figure 3.1: The construction of the Cantor ternary set with k = 2, as an example of the value of the
sum (3.1.20) with the scale condition given in Theorem (3.1.3).

Now it is needed to show that d(E) ≥ d. Lemma (3.1.1) shows that d(E) can be determined using
only fundamental intervals. Consider a finite covering of E by fundamental intervals, say J1 , J2 , . . . , Jr .
Now we wish to obtain a lower bound of the sum

|J1 |d + |J2 |d + +|Jr |d . (3.1.24)

It can be assumed that this covering is a disjoint one because otherwise it can be refined by removing
some of the intervals and get a smaller sum. Assume the upper level of this cover is m + 1 that the
fundamental interval that has the largest level is say, Ij1 ,...,jm ,1 . Then the cover must include all other
fundamental intervals of the same level Ij1 ,...,jm ,2 , . . . , Ij1 ,...,jm ,k , because otherwise the points of E in
Section 3.2. Local Hausdorff Dimension of a General Cantor Set Page 14

this set would not be covered by the chosen collection of intervals J1 , J2 , . . . , Jr or else there will be
another interval in this cover which has larger level. So, our covering contains Ij1 ,...,jm ,l for all l =
1, . . . , k, Recalling that
|Ij1 ,...,jm |d = |Ij1 ,...,jm ,1 |d + · · · + |Ij1 ,...,jm ,k |d , (3.1.25)
says that the fundamental intervals Ij1 ,...,jm ,l for all l = 1, . . . , k can be replaced by the interval Ij1 ,...,jm
of level m without changing the value of the sum (3.1.24).
This process can be repeated and it is eventually found that the sum (3.1.24) is just |I1 |d +|I2 |d ++|Ik |d
which equals unity. Moreover, any picked collection J1 , J2 , . . . , Jr , the value d satisfies the condition

|J1 |d + |J2 |d + +|Jr |d = 1. (3.1.26)

Hence, H s (E) ≥ 1 which means that d(E) ≥ d.

Unfortunately, far from the studied case in Theorem (3.1.3); the Hausdorff dimension of the Cantor set
doesn’t have a general formula. But [Bea65a] showed that 0 < d(E) < N .

3.2 Local Hausdorff Dimension of a General Cantor Set

The local dimension of a set E at a point x was defined in order to examine the local structure of the
set E. A set has similar global and local dimensions if any neighborhood of any point of it has the same
global structure. In this section we discuss some properties of this promising function linked with simple
examples. But the door of questions never close. In this topic the most interesting question arises in
the next chapter.
3.2.1 Definition (Local Hausdorff Dimension). The local Hausdorff dimension of a set E at any of its
points x is defined by
dx (E) = lim d(E ∩ S(x, ε)), (3.2.1)
ε→0
where S(x, ε) is the closed disk of dimension N centered at x with diameter ε.

The existence of dx (E) follows because d(E ∩ S(x, ε)) is a decreasing function of ε as ε → 0, so the
limit as ε tends to 0 always exists.
In particular if E is a closed set 2 and x ∈
/ E, then dx (E) = 0. This is because there exists some ε0 > 0
such that for all ε < ε0 , E ∩ S(x, ε) = ∅.
Many questions can be asked about the Local Hausdorff Dimension function: Is it constant?, What
about its continuity?, How can this function describe the local structure of a set?, Which conditions
can be used to define a local dimension function?...etc
3.2.2 Is the local Hausdorff dimension function constant?. From Theorem (3.1.3) and the Definition
(3.2.1) it is clear that in case of k = 2 in the Euclidean space R the local dimension of a given general
Cantor set at any of its points is always constant and equals the global dimension of the Cantor set.
This is obvious because for all points x ∈ E any neighborhood of x still has the same structure of the
whole Cantor set. But from the definition of the local dimension it cannot be zero because this says that
as ε tends to zero the dimension of the neighborhood S(x, ε) becomes zero which means that S(x, ε)
contains only the singleton {x}.
2
i.e. E contains all its limit points
Section 3.2. Local Hausdorff Dimension of a General Cantor Set Page 15

Again the local dimension cannot equal one because this exceeds the value of the set E itself. So, the
local dimension function can be a constant a where 0 < a < 1. Now the question become, is the local
dimension function always constant?
The local dimension is not constant in general. To clarify this we construct a Cantor set D by induction:

• Start from D0 = I0 = [0, 1] ⊂ R;


• D1 = I1 ∪ I2 , where I1 = [0, 1/3] and I2 = [3/4, 1],
• Suppose that Dn has been constructed and that it consists of 2n compact intervals Ii1 ,...,in such
that i1 , . . . , in ∈ {0, 1}. Then Dn+1 consists of all the compact intervals Ii1 ,...,in ,1 and Ii1 ,...,in ,2
with i1 , . . . , in ∈ {0, 1} such that
– Ii1 ,...,in and Ii1 ,...,in ,1 have the same left hand endpoint;
– Ii1 ,...,in and Ii1 ,...,in ,2 have the same right hand endpoint;
– Ii1 ,...,in ,1 and Ii1 ,...,in ,2 are disjoint;
– The scaling condition: fundamental interval satisfies that
   
 1 1

 3 |Ii1 ,...,in | if Ii1 ,...,in+1 ⊂ 0, 3 ,

|Ii1 ,...,in+1 | =     (3.2.2)

 1 3

 |Ii1 ,...,in | if Ii1 ,...,in+1 ⊂ ,1 .
4 4

Surely, D = ∩∞
n=1 Dn , and the local Hausdorff dimension at a point x ∈ D depends on its position.
  
 ln 2 1

 ln 3 if x ∈ 0, 3 ,

dx (D) =   (3.2.3)

 ln 2 3

 if x ∈ ,1 .
ln 4 4
Hence, the local dimension is not a constant function.
3.2.3 Is the Local Dimension Continuous?. Again let Z has the same global structure of the general
Cantor set D but with another scaling condition. Let x ∈ Z such that the sequence representation of
x in the universal Cantor set is given by x = (0, 1, 0, 1, . . . ) i.e.


0 if r is odd,
xr = (3.2.4)

1 if r is even.

The set Z is compact and x is not an end point of Z, but x is an end point of at least one sequence
of endpoints in Z. The trick in this example is to construct two sequences of endpoints of Z such that
both of them have x as the limit but the sequences have different Hausdorff local dimensions:

• Define {yr }r such that y 0 = (0, 0, 0, . . . ), y 1 = (0, 1, 0, 0, . . . ), in general yjr agrees with the
sequence x till the digit number (2r − 1) then the rest of the sequence y r are zeros. Therefore,
{y r }r converges to x. 

xj if j ≤ 2r − 1,
r
yj = (3.2.5)

0 otherwise.
Section 3.2. Local Hausdorff Dimension of a General Cantor Set Page 16

• Define {z r }r such that z 0 = (0, 1, 1, . . . ), z 1 = (0, 1, 0, 1, 1, . . . ), in general z r agrees with the


sequence x till the digit number 2r then the rest of the sequence z r are ones. Therefore, {z r }r
converges to x. 

xj if j ≤ 2r,
r
zj = (3.2.6)

1 otherwise.

Then the condition on the fundamental intervals of Z is such that at any level n if x ∈ Ii1 ,...,in+1 , then
 
 1

 3 |Ij1 ,...,jn | if Ij1 ,...,jn+1 lays on the left side of Ii1 ,...,in+1 ,

|Ij1 ,...,jn+1 | =   (3.2.7)

 1

 |Ij1 ,...,jn | otherwise.
4

Then this leads to restrict the sequence {y r }r in the intervals has the 3−1 scaling so that any point
in this sequence has a local dimension ln 2/ ln 3. While {z r }r in the intervals has the 4−1 scaling and
so this sequence has local dimension ln 2/ ln 4. But any neighborhood of x contains points of local
dimension ln 2/ ln 3 as well as of dimension ln 2/ ln 4.

3.2.4 Definition (Upper Semi-continuous Functions). A function f is called upper semi-continuous at


a point x0 if and only if for all δ > 0 there exists ε > 0 such that for all the other points x in the
neighborhood of x0 if |x − x0 | < ε, then f (x) ≤ f (x0 ) + δ. This can be expressed as:

f (x0 ) = lim sup f (x). (3.2.8)


x→x0

Therefore, an upper semi-continuous function satisfies this condition at all points.

3.2.5 Theorem. The function f which takes a point x in a Cantor set E to its local dimension is an
upper semi-continuous function.

Proof. Suppose that the local dimension function is not upper semi-continuous. Then there exists a
point x0 for which there exists δ > 0 such that for all ε > 0 there exists a point y such that |y − x0 | < ε
and dy (E) > dx0 (E) + δ. This means that any neighborhood S(x0 , ε) of x contains such a point y
which means that
d(E ∩ S(x0 , ε)) ≥ dy (E) − δ > dx0 (E). (3.2.9)
contradicts the definition of the local dimension. Hence, the local dimension function is upper semi-
continuous.
4. General Cantor Sets Correspond to Upper
Semi-continuous Functions
Given a Cantor set E, the local dimension at any point x ∈ E is determined by Equation (3.2.1). An
emerging question is: for which kind of functions g does a Cantor set exist, for which the local dimension
value at any of its points can be determined by the function g? This question becomes the focus of the
essay in the present chapter.

4.1 Construction of a Cantor set Correspond to an Upper Semi-continuous


Function

4.1.1 Definition (Local Dimension Functions). A function g : U −→ R from the universal Cantor set
U defined by (2.4.1) is called a local dimension function if there exists a Cantor set E such that

g(x) = dφ(x) (E), (4.1.1)

where φ is given by Equation (2.4.3).

4.1.2 Definition (Simple Functions). A function f : X −→ R is called a simple function if it takes


only a finite number of different values. If Aj = f −1 (pj ), for some finite set of pj ’s , then the function
f can be written as X
f= pj χAj , (4.1.2)
j

where χA is the characteristic function of A, defined using Boolean right-hand side

χA (x) = (x ∈ A) (4.1.3)

4.1.3 Theorem. Every upper semi-continuous function g : U −→ (0, 1) is a limit of a sequence of


simple upper semi-continuous functions.

Proof. Assume that g : U −→ (0, 1) ⊂ R is an upper semi-continuous function. Q We construct {gl }l ,


gl : U −→ (0, 1) as follows, let x ∈ Oj1 ,j2 ,...,jl = {j1 } × {j2 } × · · · × {jl } × ∞n=1 n for some sequence
U
j1 , j2 , . . . , jl ∈ {0, 1}, then define

gl (x) = max g(y). (4.1.4)


y∈Oj1 ,j2 ,...,jl

It is obvious that for all l the function gl is simple because it is constant on Oj1 ,j2 ,...,jl .
The function g is upper semi-continuous by the assumption, therefore by definition (4.1.4) it follows
that gl is upper semi-continuous for all l values.
Now {gl }l is a a sequence of simple upper semi-continuous functions. We show that it converges to the
upper semi-continuous function g. From Lemma (2.3.2) δn = maxn |Ii1 ,...,in | tends to zero as n tends

17
Section 4.1. Construction of a Cantor set Correspond to an Upper Semi-continuous Function Page 18

to infinity. This condition can be used as y → x.

g(x) = lim sup g(y)


y→x

= lim max g(y)


l→∞ y∈Oj1 ,j2 ,...,jl

= lim gl (x). (4.1.5)


l→∞

Hence, the sequence {gl }l converges pointwise to g.

4.1.4 Lemma. The sequence of simple upper upper semi-continuous functions gl : U −→ (0, 1) defined
by Equation (4.1.4) are local dimension functions.

Proof. It is shown in Section (3.2.2) that the local dimension function can be a constant a, where
0 < a < 1. Each function gl is locally constant on the intervals Oj1 ,j2 ,...,jl for all possible sequences
j1 , j2 , . . . , jl ∈ {0, 1}; and for all l we have
[
Oj1 ,j2 ,...,jl = U, (4.1.6)
j1 ,j2 ,...,jl

So, gl is a local dimension function on each Oj1 ,j2 ,...,jl which means that there is a Cantor set Cj1 ,j2 ,...,jl
corresponding to gl : Oj1 ,j2 ,...,jl −→ (0, 1), For all l and therefore the Cantor set Cl that is corresponding
to gl : U −→ (0, 1) is the union of these Cantor sets on all sequences j1 , j2 , . . . , jl ∈ {0, 1}, For all l.
[
Cl = Cj1 ,j2 ,...,jl , (4.1.7)
j1 ,j2 ,...,jl

and from the way of constructing Cl it follows that for all x ∈ U

gl (x) = dφ(x) (Cl ). (4.1.8)

Now a sequence {Cl }l of general Cantor sets has been constructed; this sequence satisfies the theorem
for the sequence of functions {gl }l . Indeed, Equations (4.1.5) and (4.1.8) say that any upper semi-
continuous function is a limit of a sequence of functions that satisfy the theorem. In other words, the
space of the solutions for the theorem is a dense1 subspace in the space of all upper semi-continuous
functions.
The next step is to characterize the Cantor set E using the scales of its fundamental systems En at
each level n. Obviously, the fundamental system En at each level n is totally determined by k n pairs
of points, which are the endpoints of each fundamental interval. In other words, one can say that En
is determined by k n scales as shown in Figure 4.1 for a general Cantor set with k = 3 as an example.
1
A subset S of a space X is called dense (in X) if any point x in X belongs to S or is a limit point of S.
Section 4.1. Construction of a Cantor set Correspond to an Upper Semi-continuous Function Page 19

E0 s| {z }s
a0

E1 s| {z }s s| {z }s s| {z }s
a1 a2 a3

E2 s| {z } s s| {z } s |s {z }s s|{z}s s|{z}s s|{z}s


..
.
a11 a12 a13 a21 a23 a23
.. .. .. .. .. .. ..
. . . . . . .

Figure 4.1: The construction of a general Cantor set with k = 3, as an example of the scale characteri-
zation.

Therefore, we say that En is totally determined by a k n -dimensional vector vn


E0 : v0 = (a0 ) ,
E1 : v1 = (a1 , a2 , a3 )
E2 : v2 = (a11 , a12 , a13 , a21 , a22 , a23 , a31 , a32 , a33 ) ,
..
.
 

En : vn = a11 . . . 1 , a11 . . . 12 , . . . , a33 . . . 3 , (4.1.9)


| {z } | {z } | {z }
n digits n digits n digits
| {z }
3n possibilities
..
.

where
|Ii1 ,i2 ,...,in | = ai1 ai1 ,i2 ai1 ,i2 ,i3 . . . ai1 ,i2 ,...,in , (4.1.10)
and
|Ii1 ,i2 ,...,in ,in+1 |
ai1 ,i2 ,...,in ,in+1 = (4.1.11)
|Ii1 ,i2 ,...,in |
Hence, the Cantor set E is totally determined by its fundamental systems and scales as seen in Equations
(2.3.1) and (4.1.11). Thus, E is totally determined by the k n -dimensional vectors.
Now I want to construct the Cantor set that satisfies the theorem for the function g. Define new
parameter bi1 ,i2 ,...,in on each Oi1 ,i2 ,...,in ⊂ U by:
bi1 ,i2 ,...,in = max g(y). (4.1.12)
y∈Oi1 ,i2 ,...,in

From Theorem (3.1.3) for k = 2, if a Cantor set C which constructed such that it has a Hausdorff
dimension equals bi1 ,i2 ,...,in , then its scales in Equation (4.1.11) are given by:
 

−1 
 
bi1 ,i2 ,...,in
ai1 ,i2 ,...,in = 2 . (4.1.13)
Section 4.1. Construction of a Cantor set Correspond to an Upper Semi-continuous Function Page 20

To show that this Cantor set is the desired one, we divide the problem into two parts: the first, shows
that g is a local dimension function for C for all the points x ∈ U at which the function is locally
continuous; the second discusses the cases of x at which the function jumps.

4.1.5 Theorem. If the upper semi-continuous function g : U −→ (0, 1) is continuous at a point x, then

dφ(x) (C) = g(x). (4.1.14)

Proof. Because U and C are homeomorphic, g can be considered as acting on C. The point x can be
used instead of its image φ(x) in C. Assume that g is continuous at x ∈ C and take ε > 0 such that g
is continuous on S = S(x, ε) ∩ C. Let

gmax,ε = max{g(y) : y ∈ S} (4.1.15)


gmin,ε = min{g(y) : y ∈ S}. (4.1.16)

Also define
 

−1 
 
gmax,ε
amax,ε = 2 , (4.1.17)
 

−1 
 
gmin,ε
amin,ε = 2 . (4.1.18)

Let Iε be one of the fundamental intervals contained in S. So, Iε has a scale between amin,ε and amax,ε .
After n levels, each fundamental interval of Iε has a scale between anmin,ε and anmax,ε . Hence, for all
α>0

X
2n anα
min,ε ≤ |I|α ≤ 2n anα
max,ε . (4.1.19)
I⊂Iε

Therefore,
gmin,ε ≤ d (S(x, ε) ∩ C) ≤ gmax,ε . (4.1.20)
By continuity of g at x
lim gmin,ε = lim gmax,ε = g(x). (4.1.21)
ε→0 ε→0

Hence,
dx (C) = g(x), (4.1.22)
or strictly written:
dφ(x) (C) = g(x). (4.1.23)

So, the function satisfies the theorem wherever it is continuous at the point. Because this construction
fails for the cases in which the function g is not locally constant. The following Example (4.1.6) shows
that this can be solved at any local maximum point x; that is, changing the scales slightly around x
does not affect the local dimension of any point different from x.
Section 4.1. Construction of a Cantor set Correspond to an Upper Semi-continuous Function Page 21

4.1.6 Local dimension at local maximum points. This example is to show the possibility of finding
a suitable Cantor set in the case of single jumps at certain points. For the function f shown in Figure
4.2 there is a problem at the point x0 ∈ U and we want to construct a Cantor set for which the local
dimension given by the function f

 ln 2
 ln 3 if x = x0 ,

f (x) = (4.1.24)


 ln 2 otherwise.
ln 4
 n
2
We construct a Cantor set with k = 2 such that at the level n there are at least fundamental
n

f
6

ln 2 r
ln 3

ln 2 c
ln 4
- x
x0

Figure 4.2: An example of a function f has a single jump at one point.

intervals of size 3−n where x0 included in one of them. The rest of the fundamental intervals
 n  are (1/4)
2
of the previous interval size. At each level these interval can be found because the intervals
 n+1   n+1  n
2 2
generate intervals at the next level, therefore we can always choose of them to have
n n+1
size 3−(n+1) . Hence any small neighborhood of x0 contains many intervals of size 3−n for n large
ln 2
enough, which ensures that the local dimension at x0 equals .
ln 3
Let y 6= x0 , this means that at some level n they will split into different intervals. Assume the y and
x0 are separated by one fundamental interval. Then after additional t levels they will be separated by
2t fundamental intervals which allows to choose the scaled intervals far from the point y.
By Definitions (2.2.1) and (2.2.2) the dimension of the constructed Cantor set can be checked using
the diameters of the fundamental intervals:

   n−1
X
2n 1 1 1
+ cj , (4.1.25)
n 3nd 3jd 4(n−j)d
j=1

where
n−1
X  
2n
n
cj = 2 − . (4.1.26)
n
j=1
Section 4.1. Construction of a Cantor set Correspond to an Upper Semi-continuous Function Page 22

ln 2 ln 2
If d > , then all terms in Equation (4.1.25) tends to zero as n tend to infinity. While if d <
ln 3 ln 3
then the sum term tends to zero but the first term tends to infinity when n grows. The same example
can be done for any number of intervals k instead of 2 if required.

4.1.7 Theorem. The function g : U −→ (0, 1) ⊂ R, is a local dimension function if and only if g is
upper semi-continuous with the property that there is no sequence {xj }j such that g(xj ) tends to zero
as j tends to infinity.

g(x) = lim sup g(y). (4.1.27)


y→x

The proof can be done by dividing the 2n sets to n groups {Fj }nj of sequential sets (i.e corresponding
 n
2
to sequential fundamental intervals in the Cantor set). So each group Fj consists of and then
n
redefining the parameter bi1 ,i2 ,...,in such that

bi1 ,i2 ,...,in = max g(y). (4.1.28)


y∈Fj
Oi1 ,i2 ,...,in ⊂Fj

Therefore, the scales in Equation (4.1.11) of the Cantor set C are given by:
 

−1 
 
bi1 ,i2 ,...,in
ai1 ,i2 ,...,in = 2 . (4.1.29)

The last scaling condition given by Equation (4.1.29) is suitable in case of all point at which the function
is constant, as well as, at the local maximum points.
5. Conclusion
The Hausdorff dimension of the general Cantor set has been studied. It has been proved that the
Hausdorff dimension of the general Cantor set can be determined using only its fundamental systems
as special coverings. Then the Hausdorff dimension of the general self-similar Cantor set has been
calculated. In case of self-similarity Hausdorff dimension does not depend on the dimension of the space
although it is bounded by it. It is a function of the number of intervals we divide into and the scale of
each division only.
The local dimension has been introduced and we have discuss its constancy and continuity properties.
In general, the local dimension function is not constant and not continuous and some suitable examples
have been shown. Indeed, as shown the local dimension is an upper semi-continuous function.
Finally, we have shown a necessary and discussed the sufficient condition for a function g to be a local
dimension function: upper semi-continuity. The proof was done by showing first that the solution set is
a dense subspace of the upper semi-continuous space, i.e. any upper semi-continuous function is a limit
of local dimension functions {gl }l . Secondly, the sequence {gl }l was used to construct a Cantor set C
depending on the idea of scaling characterization. Finally, the proof was finished by showing that the
Cantor set C is the desired one for all points at which the function g is locally continuous. The cases
of the points at which g is not locally continuous are discussed separately and it is shown that we can
always modify C slightly around those points without changing the values of the local dimension at any
other points close to them or affecting the Hausdorff dimension of the Cantor set itself.

23
Acknowledgements
This work has been supported by AIMS the African Institute for Mathematical Science. I would like
to thank my supervisor Prof. Alan Beardon for his motivation and supervision; all my appreciation
to Prof. Jeff Sanders for the revision and open minded questions; also I’m grateful to my tutor Dr.
Dimbinaina Ralaivaosaona for his following up, interest and supportive discussions and also thank Mr.
Jan Groenewald for all the IT organization.
I would love to thank Mr. Mostafa Eldemery for revising the language and also for raising my morale.
Finally but the most important, even a small piece of this essay couldn’t be done without the praise and
continuous encouragement given from my parents, to whom I completely dedicate this work.

24
References
[AA08] D.J. Albers and G.L. Alexanderson, Mathematical People: Profiles and Interviews, AK Peters,
2008.

[Bea65a] A. F. Beardon, On the Hausdorff Dimension of General Cantor Sets, Proc. Camb. Phil. Soc.
61 (1965), 679–693.

[Bea65b] A. F. Beardon, The Hausdorff Dimension of Singular Sets of Property Discontinuous Groups
in N-Dimensional Space1.

[Fal90] K. J. Falconer, Fractal Geometry, John Wiley & Sons Ltd., 1990.

[Fal02] Kenneth J. Falconer, Handbook of Measure Theory, Elsevier Science, 2002.

[For51] L. R. Ford, Automorphic Functions.

[G.C83] G.Cantor, ¨ber unendliche, lineare Punktmannigfaltigkeiten v, Mathematische Annalen 21


(1883), 545591.

[Han] P. Hansen, Brownian Motion and Hausdorff Dimension.

[Hau19] F. Hausdorff, Dimension und äusseres Mass, Math. Ann 79 (1919), 157–179.

[Hut81] J. E. Hutchinson, Fractals and Self Similarity, Indiana University Mathematics Jornal 30
(1981), 713–747.

[HW41] W. Hurewicz and H. Wallman (eds.), Dimension Theory, Princeton Univ. Press, 1941.

[J.S09] J.Shah, Hausdorff Dimension and its Applications, Computing (2009), 1–10.

[Man67] B. B. Mandelbrot, How Long is the Coast of Britain? Statistical Self-similarity and Fractional
Dimension, Science 156 (1967), 636.

[Man83] , The Fractal Geometry of Nature, W.H. Freeman, 1983.

[Man04] B.B. Mandelbrot, Fractals and Chaos: The Mandelbrot Set and Beyond : Selecta, Springer,
2004.

[MC09] L. Marek-Crnjac, A Short History of Fractals-Cantorian Space-time, Elservier Ltd. 41 (2009),


2697–2705.

[McM84] C. McMullen, The Hausdorff Dimension of General Sierpinski Carpets, Nagoya Math. J. 96
(1984), 1–9.

[McM98] C. T. McMullen, Hausdorff Dimension and Conformal Dynamics, iii: Computation of Dimen-
sion, American Journal of Mathematics 120 (1998), 691–721.

[Nas94] M. S. El Naschie, Is Quantum Space a Random Cantor Set with a Golden Mean Dimension
at the Core?, Elservier Science Ltd. 4(2) (1994), 177–179.

[Ols05] L. Olsen, Characterization of Local Dimension Functions of Subsets of Rd , Colloq. Math. 103
(2005), 231–239.

25
REFERENCES Page 26

[Pau02] Djura Paunic, Handbook of Measure Theory, Elsevier, 2002.

[RT59] C. A. Rogers and S. J. Taylor, Additive Set Function in Euclidean Space, Acta Math 101
(1959), 273–302.

[Sag86] H Sagan, Approximating Polygons for Lebesgues and Schoenbergs Space-filling Curves, Amer.
Math. 93(5) (1986), 361–368.

[Sag92] , An Elementary Proof that Schöenberg’s Space-Filling Curve is Nowhere Differen-


tiable, Math. Mag. 65(2) (1992), 125–128.

[Sch05] Dierk Schleicher, Hausdorff Dimension, Its Properties, and Its Surprises, JSTOR 114 (2005),
509–528.

[Str99] Robert S. Strichartz, Analysis on Fractals, AMS 46 (1999), 1199–1208.

[SW03] L. Sadun and R. F. Williams, Tiling Spaces are Cantor Set Fiber Bundles, Ergod. Th. &
Dynam. Sys. 23 (2003), 307316.

[Tsu59] M. Tsuji, Potential Theory in Modern Function Theory.

[Tur88] D. L. Turcotte, Fractals in Fluid Mechanics, Ann. Rev. FIuM Mech 20 (1988), 5–16.

[Wil68] S. Willard, General topology, Addison-Wesley Publishing Company, 1968.


Errata: Hausdorff Dimension of Cantor Sets

Shaimaa Mohamed Abdelmonem Abdelhafez (shaimaamonem@aims.ac.za)

1 Errata by author

1. Page 1, first paragraph: over should read our

2. Page 2, first paragraph: on which the whole work is based should be removed

3. Page 2, Definition 2.1.1: x =∈ X should read x ∈ X

4. Pages 3, last paragraph and page 12, first paragraph: δ − ms (E) should read msδ (E)

5. Page 3, second footnote, If E should read If E is compact

6. Pages 5, 9 and 13, ∀ should read for all


Q∞
7. Pages 6, last paragraph: the Unirversal Cantor set should read n=1 1, 2, . . . , k

8. Page 7, second paragraph in the proof: ans should read and

9. Page 8, third paragraph: il = il+1 = · · · = 0 should read il = il+1 = · · · = 1

10. Page 8, third paragraph: il = il+1 = · · · = 1 should read il = il+1 = · · · = k

11. Page 9, Lemma 3.1.1: δ − M α (E) should read Mδα (E)

12. Page 10, Proof (i): δn − M α (E) should read Mδαn (E)

13. Page 11, line before Theorem 3.1.3: Therefore one can conclude that should read The following
theorem studies the case in which

14. Page 12, first paragraph: k n should read k m

15. Page 17, last paragraph: in order to find the limit of the sequence {Cl }l should read be removed

16. Page 19, Theorem 4.1.5: upper upper should read upper

17. Page 20, third paragraph: which is still ... should read which allows to choose the scaled intervals
far from the point y

1
18. Page 20, Figure 4.2: should be

f
6

ln 2 r
ln 3

ln 2 c
ln 4
- x
x0

Figure 1: An example of a function f has a single jump at one point.

2 Errata requested by examiners

1. Page 1, first paragraph: fractals is should read fractals are

2. Page 1, first paragraph: fields as should read fields such as

3. Page 1, third paragraph: in details should read in detail

4. Page 2, last paragraph: on integers should read to integer values

5. Page 3, fourth paragraph: sub-coverings should read coverings

6. Page 4, Definition 2.3.1: is a set of should read consists of

7. Page 6, fourth paragraph: U has should read U is

8. Page 13, second paragraph: The the should read The

9. Page 16, Definition 4.1.2: Aj = f −1 (pj ) should read Aj = f −1 (pj ), for some finite set of pj ’s

10. Page 16, Definition 4.1.2: that is should read defined using a Boolean right-hand side

11. Page 21, last paragraph: csaling should read scaling

View publication stats

You might also like