💼 Case: consider the tent map
$$ x_{n+1}=\left \{ \begin{matrix} r x_n & \text{if} & x_n \le \frac 12 \\ r(1-x_n) & \text{if} & x_n >\frac 12 \end{matrix} \right . $$
If we consider $r=2$
All iterates of the map from the interval $[0,1]$ remain bounded within the interval $[0,1]$
Points that escape after one iteration have $f(x_0)>1$
$f(x)=1$ for $x=1/2,2/3$
if $x_0 \in (\frac 13, \frac 23)$ : $x_0$ escapes after 1 iteration
After two iterations we must have ${f(x_0)\in (\frac 12 ,\frac 23)}$
this happens for ${x_0\in \left ( \frac 19 , \frac 29 \right )} \text{ and } {x_0\in \left ( \frac 79 , \frac 89 \right ) }$
We can construct the cantor set as following
In total what we have removed is
$$ \small\begin{aligned} \sum^\infin_{n=1}\frac{2^{n-1}}{3^n}=\frac 13 \sum^\infin_{n=0}\left ( \frac{2}{3}\right )^n = \frac 13 \frac{1}{1-\frac 23}=1 \end{aligned} $$
After $n$ steps we have $2^n$ pieces of length $\frac 13^n$ so that the length $\frac 23^n \to 0$ as $n\to \infin$
Properties:
- There is structure on all length scales
- The Cantor set is self-similar
- The dimension of the Cantor set is not an integer
- The Cantor set has zero length
If we look at the length of the curve we have
$$ \begin{aligned} L_1&=\frac 43 L_0 \\ L_2&=\left ( \frac{4}{3} \right )^2 L_0 \\ &\; \; \vdots \\ L_\infin&=\infin
\end{aligned} $$
Total length of the Von Koch curve is infinite
The arclength between any two points in Euclidian space of the curve is infinite
So we cant express the positions of points using a 1D grid but 2D is redundant so it has non-integer dimensions between 1 and 2
<aside> <img src="https://prod-files-secure.s3.us-west-2.amazonaws.com/369dfa6b-d4d9-4cf2-a446-e369553b6347/949b372f-a131-4642-a322-a16a5b775eec/Self_similarity.gif" alt="https://prod-files-secure.s3.us-west-2.amazonaws.com/369dfa6b-d4d9-4cf2-a446-e369553b6347/949b372f-a131-4642-a322-a16a5b775eec/Self_similarity.gif" width="40px" />
Self similarity: Two objects are similar if they have the same shape, regardless of their size
</aside>
Similarity transformations:
Scaling
Rotation
Translation
A function $f(x)$ is scale-invariant if its shape is preserved under rescaling of the variable $x$
$$ f(\lambda x)=f(x)g(\lambda) $$
where $\lambda$ is the rescaling factor and $g$ is a function of $\lambda$
🗒️ Note: here $f(x)$ looks the same regardless of the scale on which we look at it
⚽ Goal: lets try to remove this abstract $g(\lambda)$ function
If we set $x=1$ on $f(\lambda x)=f(x)g(\lambda)$ we get
$$ \begin{aligned} g(\lambda )&=\frac{f(\lambda)}{f(1)} \\ \Rightarrow \qquad f(\lambda x)&=f(x) \frac{f(\lambda)}{f(1)} \end{aligned} $$
Now if we differentiate with respect to $\lambda$ we get
$$ xf'(\lambda x)=\frac{f(x)}{f(1)}f'(\lambda) $$