Lecture 19 - Cantor Set

We construct the Cantor Set in the following way:

C0=[0,1]C1=[0,13][23,1]=C0(13,23)C2=C1((19,29)(79,89))

And so on:

Notice that:

C0C1C2

We define the final set as:

C=i=0Ci

What is in C? Well notice that:

0,1,13,23,C

C has all the endpoints each of the form:

m3nC

So does this imply that C is countably infinite? Consider instead asking how much length of the interval is there. We have C0 has a length of 1:

13+29+=13(1+23+(23)2+)

This is a geometric series:

=131123=1

So this implies that C should have no length. Maybe it is indeed countably infinite?

Another Perspective

Let xC. Construct a sequence of 0's and 1's as follows (the example uses the case where x=79C):

79:101=101111111

Where the first digit says:

We repeat this process for each cut, adding digits. As another example:

19:001=0011111111

Now consider an arbitrary sequence of these digits. We want to do Cantor's Diagonalization Method to try to construct some new digit that isn't in the list:

(b1,b2,b3,)

We can change the first digit of b1, the second of b2, the third of b3, and so on to make a new digit x.

Another way of thinking of this is that each CiCi+1. This implies that we can use the Ci's with the Nested Interval Property to show that:

xi=1Ci=C

But in both cases, there are uncountable many such sequences! So then C must not be countable.

Dimensions of A Set

To set the stage, consider a 0-dimensional object, 1-dimensional object, ...

The idea from geometry is if you multiply an n-dimensional object by c, then the area/volume/... of that object is scaled by cn. In this case c=3 so it magnifies by 3n.

For the case of Cantor's set is if we scale each line by 3:

So somehow we had some n-dimensional object, scaled by 3, where the area/volume/length here got scaled by 2! Not 3 as we expected. Here:

size2=3nn=log3(2)

So somehow we have fractional dimension. This is where the word fractal comes from (fracttional dimension). This brings us into talking about the start of the new section:

Topology of R

Recall the definition of the Epsilon-Neighborhood of x:

ε-neighborhood of a

Vε(a)=(aε,a+ε)={xR||xa|<ε}

open set

Let UR. Then U is open if:
xUε>0(Vε(x)U)
ie: each xU has some ε-neighborhood contained in U.

Example: (a,b)

(a,b) is open.

Since choose ε=min(|xa|,|xb|) implies that Vε(x)(a,b).

By constrast if we considered even a one-sided interval like (a,b] then it is NOT open:

Notice if $x = b$ then any $\varepsilon > 0$ would have some element in $V_\varepsilon(b)$ be outside the interval.

See also Union & Intersection of Collections of Open Sets.

Connecting it to Sequences

Definition

Let AR and xR. Then
x is a limit point of A if every Epsilon-Neighborhood of x intersects A at a point other than x. Namely:
ε>0(Vε(x)A{x})

For example, if A=(a,b], then what are the limit points of A? The limit points of A are:
=a,b

Alternative Definition of Limit Point

Alternative Definition of Limit Point

Let AR and xR. Then x is a limit point of A iff (an):
anA{x}
for all n and anx.

Proof

(): Suppose x is a limit point of A. Let's apply a sequence of neighborhoods such that we can construct an. To do this, let nN. Then choose:
an(V1n(x){x})A
we claim that anx. To prove this via the definition let ϵ>0. Choose NN such that 1N<ϵ. Then suppose nN then:
|anx|<1n(anV1n(x))1N<ϵ
(): Suppose an where anA{x} for all n and that anx. Let ϵ>0 so that Vϵ(x) is an epsilon-neighborhood of x. We want to show that (Vϵ(x){x})A. We can choose a NN such that:
|anx|<ϵ
by the definition and that anx. Then definitely |aNx|<ϵ, so then NVϵ(x). Furthermore clearly aNA while aNx so then aN(Vϵ(x){x})A is not empty as desired.

To illustrate this, consider:
A={1n|nN}
Then the set of all limit points of A is just {0}. Notice other points like 12 are not in this set since there's no sequence of numbers constructing from the set cannot approach 12.

Another example is when A=Q. The set of limit points of A are now R.