Classic Iterated Function Systems

Larry Riddle, Agnes Scott College

Details about the Bottom of a Symmetric Binary Tree

Let the tree be defined by the angle $\theta$ and scaling factor $r$. We will take the base of the tree to be at the origin and the initial vertical trunk to be of length 1. By the "bottom" of the tree we will mean the bottom of the branch tips, that is, the smallest y-coordinate of the branch tips.

Let \(\alpha\) be the complex number \(\alpha = e^{\theta i}=\cos{\theta} + i \sin{\theta}\). We can represent the trunk of the tree by the complex number \(i\) (which we can also think of geometrically as a vector). Recall that multiplying a vector by r will scale the vector by r, that multiplying the vector by \(\alpha\) corresponds to a counterclockwise rotation by $\theta$, and multiplying by \(\alpha^{-1}\) corresponds to a clockwise rotation by $\theta$. We also have \(\alpha^n = e^{n\theta i}=\cos{(n\theta)} + i \sin{(n\theta)}\) and \(\alpha^{-n} = e^{-n\theta i}=\cos{(n\theta)} - i \sin{(n\theta)}\) for all integers $n$.

$R^k(LR)^\infty$
Branch

To get to the bottom of the branch tips, we need the initial branches to turn right until the last branch passes the vertical downward direction. We then want to go vertically downward as much as possible for each new branch in the construction of the tree. Thus we want to alternate left and right branches. This gives the path $R^k(LR)^\infty$, where $k$ is the smallest integer such that $k\theta \ge 180^\circ$. In terms of the complex numbers, this means we want to multiply by \(\alpha^{-1}\) for the first $k$ branches, then alternate multiplying the vectors representing the branches by \(\alpha\) and \(\alpha^{-1}\), while at the same time also multiplying by r. The branch tip for this path is therefore located at the point \((x,y)\) corresponding to the complex number

\( \left(i + i r \alpha^{-1} + i r^2 \alpha^{-2} + \ldots + i r^k \alpha^{-k}\right) + i r^k \alpha^{-k} \cdot \left( {r \alpha + r^2 + r^3 \alpha + r^4 + \dots} \right) \) \[ \begin{align} &= i \left( \sum_{n=0}^{k}{r^n \alpha^{-n}} + r^k \alpha^{-k} \cdot \left( \dfrac{ra}{1-r^2} + \frac{r^2}{1-r^2} \right) \right) \\ \\ &= i \left( \sum_{n=0}^{k}{r^n \alpha^{-n}} + \frac{r^{k+1}\alpha^{-(k-1)} + r^{k+2}\alpha^{-k}}{1-r^2} \right) = iz. \end{align} \]

The y-coordinate of this branch tip is \(\text{Im}(iz) = \text{Re}(z)\), or

\[ \sum_{n=0}^{k}{r^n \cos(n \theta)} + \frac{r^{k+1} \cos((k-1)\theta) + r^{k+2} \cos(k \theta)}{1-r^2}. \] Here is an example with $\theta$ = 55°, so $k$ = 4. The image on the left shows the $R^4(LR)^\infty$ path while the image on the right shows the actual tree. The y-coordinate of the $R^4(LR)^\infty$ branch tip is y = 0.53212.

$R^4(LR)^\infty$ path, $\theta$ = 55°, $r$ = 0.65

Example

The following image shows the symmetric binary tree for $\theta$ = 160° and $r$ = 0.8 along with the path $R^4(LR)^\infty$. The smallest integer satisfying $k\theta \ge 180^\circ$ is $k$ = 2, so this example shows that the path described above will not always produce the branch tip with the smallest y-coordinate. Here the $R^2(LR)^\infty$ branch tip has y = 0.27365 while the $R^4(LR)^\infty$ branch tip is at y = 0.22498.

The overlapping branches filling in the tree in the shape of a horizontal leaf
$\theta$ = 160°, $r$ = 0.8

In this case the black path is of the form $R^m(LR)^\infty$ where $m$ is the smallest integer satisfying $m\theta \ge 540^\circ$. In essence, the black path starts with enough right turns until the last branch has passed the downward vertical direction for the second time before it begins to alternate between left and right branch turns.

General
Case

Let $y_n$ be the y-coordinate of the $R^n(LR)^\infty$ branch tip. For a fixed $\theta$, this coordinate depends on the value of $r$. Let $k$ be the smallest integer satisfying $k\theta \ge 180^\circ$ and let $m$ be the smallest integer satisfying $m\theta; \ge 540^\circ$. Let \(f_\theta (r)=y_m - y_k\) be the difference between the y-coordinates of the two corresponding branch tips , so \[ \begin{align} f_\theta (r) &= \sum_{n={k+1}}^{m}{r^n \cos(n \theta)} \\ \\ &+ \frac{r^{m+1} \cos((m-1)\theta) + r^{m+2} \cos(m \theta) - r^{k+1} \cos((k-1)\theta) - r^{k+2} \cos(k \theta)}{1-r^2} \end{align} \] If \(f_\theta (r)\) is positive then $y_k$ is less than $y_m$, but if \(f_\theta (r)\) is negative then $y_k$ is greater than $y_m$.

Note that \(f_\theta (0) = 0\). It can be shown [Near 0 Details] that for $r$ close to 0, the graph of \(f_\theta (r)\) will initially begin increasing in a concave up fashion as $r$ increases, and thus \(f_\theta (r)\) will be positive for $r$ close to 0. Below are examples of the three things that can happen with the graph of \(f_\theta (r)\) for $0 < r < 1$:

$\theta$ = 60°
\(\lim_{r \rightarrow 1^{-}} f_\theta (r) = A_\theta > 0\)
$\theta$ = 135°
\(\lim_{r \rightarrow 1^{-}} f_\theta (r) = +\infty\)
θ = 146°
\(\lim_{r \rightarrow 1^{-}} f_\theta (r) = -\infty\)

To see why, we need to examine the behavior of the numerator of the fraction in the expression for \(f_\theta (r)\) when $r$ = 1 since the denominator is positive and goes to 0 as $r$ approaches 1 from the left. So we need to determine the sign of \[ N(\theta) = \cos((m-1)\theta) + \cos(m \theta) -\cos((k-1)\theta) -\cos(k \theta) \] where \(k = \lceil \frac{180}{\theta}\rceil\) and \(m = \lceil \frac{540}{\theta} \rceil\), in particular whether $N(\theta)$ is 0, positive, or negative (corresponding to the three graphs above).

Below is the graph of \(N(\theta)\) showing that \(N(\theta)\) alternates between intervals where it is positive and intervals where it is negative (with those intervals getting smaller and alternating more rapidly as $\theta$ approaches 0.)

Here are closer views of the graph on the intervals $40^\circ ≤ \theta ≤ 90^\circ$ and $90^\circ ≤ \theta ≤ 180^\circ$.

The places where the graph has a sharp corner are where \(\theta = \dfrac{540^\circ}{p}\) for an integer $p ≥ 4$.

The zeros of $N(\theta)$ occur at \(\theta = \dfrac{720^\circ}{p}\) for all integers $p ≥ 4$ [Zeros Proof]. Starting at the right end of the graph and going in decreasing order, the first four zeros are at 180°, 144°, 120°, and 102.85714°.

If $\theta$ is a zero of $N(\theta)$, then \(\lim_{r \rightarrow 1^{-}} f_\theta (r)\) exists and is positive [Positive Proof]. Where $N(\theta)$ is positive we have \(\lim_{r \rightarrow 1^{-}} f_\theta (r) = +\infty\). For these values of $\theta$, therefore, \(f_\theta (r) > 0\) for all $0 < r < 1$ and hence $y_k$ will be less than $y_m$.

The intervals where $N(\theta)$ is negative are of the form

\[ \frac{720^\circ}{4n+5} < \theta < \frac{720^\circ}{4n+4} \text{ and } \frac{720^\circ}{4n+4} < \theta < \frac{720^\circ}{4n+3} \] for \(n = 0,1,2, \ldots\). (For $n$ = 0, only the first interval is used. This gives \(144^\circ < \theta < 180^\circ\).)

On these intervals, we have \(f_\theta (r)\) positive for small $r$ and \(\lim_{r \rightarrow 1^{-}} f_\theta (r) = -\infty\), so there must be some some $r_B$ between 0 and 1 where \(f_\theta (r_B) = 0.\)

The equation \(f_\theta (r) = 0\) can be solved numerically to find the solution $r_B$. This equation will also have a solution on the intervals where $N(\theta)$ is positive, but for those values of $\theta$, $R_B$ will be greater than 1. The next two figures show the graph of $r_B$ for $0^\circ < \theta < 180^\circ$, and a close up of this graph just on the range $30^\circ < \theta < 102.85714^\circ (= 720^\circ/7)$.


The intervals where $N(\theta) < 0$ correspond to the intervals in these graphs where $r_B < 1$. On these intervals, the bottom of the symmetric binary tree will be $Y_m$ instead of $y_k$ for $r_B < r < 1$.

Notice, however, that when $\theta < 102.85714^\circ$, the value of the critical scaling factor $r_B$ is greater than 0.95 so that for all practical purposes the bottom of a "reasonable" symmetric binary tree will be determined by the $R^k(LR)^\infty$ path. Any choice of $r > 0.95$ will produce a tree that has a massive overlap of branches. Only on the interval $\theta > 144^\circ$ is there an opportunity for $r_B$ to be small enough that one can generate an interesting tree (see, for example, the tree above with $\theta$ = 160° and $r$ = 0.8.) We actually have \[ r_B = \frac{2\cos(\theta)}{1-4\cos^2 (\theta)} \] for $\theta > 144^\circ$ [Maple Calculation (pdf)]. So $r_B$ converges to 2/3 as $\theta$ approaches 180°.