The Chaos Game Versus Uniform Rotation: From Sierpinski Gaskets to Periodic Orbits

Abstract

In this paper, we introduce a couple of dynamical systems that are related to the Chaos Game. We begin by discussing different methods of generating the Sierpinski gasket. Then we show how the transition from random to uniform selection reduces the Sierpinski gasket to simple periodic orbits. Next, we provide a simple formula for the attractor of each of the introduced dynamical systems based only on the contraction ratio and the regular $n$-gon on which the game is played. Finally, we show how the basins of attraction of a particular dynamical system can generate some novel motifs that can tile the plane.

1 The Chaos Game

Let $\mathbf{T}=ABC$ be an equilateral triangle in the plane. To play the Chaos Game (CG), start with any point in the plane, go halfway from that point to a randomly chosen vertex of $\mathbf{T}$, and then repeat the process, see [2, 4]. Skip** the first thousand or so iterates then plotting the points produced by the game, we obtain the Sierpinski triangle or gasket shown in Figure 1. The gasket, which is called the attractor of the game, was originally constructed by Waclaw Sierpinski [7] by removing the middle third of the original triangle and then repeatedly removing the middle third of the remaining triangles.

Figure 1: The Sierpinski gasket.

The Sierpinski triangle can be generated in yet another way described in [6]. Start with a point $P$ in $\mathbf{T}$ and double the distance from $P$ to the nearest vertex of $\mathbf{T}$ along the line from $P$ to that vertex. Clearly, most points will leave $\mathbf{T}$ under this game, but it just happens that points that stay trapped inside $\mathbf{T}$ are exactly those that belong to the Sierpinski gasket. This, however, is not a practical way to generate the gasket since the probability that a randomly picked point will stay in $\mathbf{T}$ under this game is virtually zero.

2 New Games

Now suppose that instead of doubling the distance away from the nearest vertex, we move halfway between the current point $P$ and that vertex of $\mathbf{T}$ which is farthest from $P$. Then, irrespective of where we start in the plane, the attractor will either be the vertices of the triangle $\mathbf{T}_{r}=A_{1}B_{1}C_{1}$ or those of the triangle $\mathbf{T}_{l}=A_{2}B_{2}C_{2}$, see Figure 2.

Figure 2: The attractors of the FVG.

We will show that the attractor reached depends solely on the location of the initial point in the plane and consequently where it first lands in $\mathbf{T}$ as the game is played.

It turned out the attractors $\mathbf{T}_{r}$ and $\mathbf{T}_{l}$ can be obtained by playing a uniform rotation game (URG). That is, instead of moving toward the farthest vertex, we simply go around $\mathbf{T}$ in a clockwise or counterclockwise direction. For counterclockwise rotation, first go halfway between the current point and the vertex $A$, followed by $B$, and then $C$. In this case, the attractor will be the vertices of the triangle $\mathbf{T}_{r}$. On the other hand, clockwise rotation leads to the vertices of $\mathbf{T}_{l}$.

It is no accident that attractors of the FVG coincide with those of the URG. But before we give the reason behind this, it should be noted that in the FVG, the attractor could be $\mathbf{T}_{r}$ or $\mathbf{T}_{l}$, depending on the starting point. This is in contrast with uniform rotation where the attractor depends only on the choice of rotation. Now to illustrate how the two dynamical systems lead to the same attractor, let us take a point that is attracted to $\mathbf{T}_{r}$ under the FVG and by counterclockwise rotation. If we follow the orbit of such a point we see that the two systems behave strikingly different outside $\mathbf{T}$, but once inside $\mathbf{T}$ they both converge to $\mathbf{T}_{r}$ quite fast, see Figure 3(a). Obviously, dark lines represent the FVG while faded lines represent the URG. A similar phenomenon occurs if we take a point that is attracted to $\mathbf{T}_{l}$ under clockwise rotation, see Figure 3(b).

Figure 3: (a) Counterclockwise rotation (b) clockwise rotation.

Indeed, taking a closer look at this, we find that the two dynamical systems are identical inside $\mathbf{T}$. To see this, let us divide $\mathbf{T}$ into dark and light regions as in Figure 4. Then once inside $\mathbf{T}$ (not on one of the bisectors), it is clear that the FVG is nothing but a clockwise rotation for points in the dark area and a counterclockwise rotation for points in the light area. If it happens that the current point is equidistant from two vertices, then we break the tie arbitrarily and return to the previous case.

Figure 4: Eventual behavior of the FVG.

3 Periodic Points

In this section, we shall study periodic points of the URG, which are in turn periodic points of the FVG. This is so since all points in the plane end up inside $\mathbf{T}$ under both games and since for points in $\mathbf{T}$ the two games are indistinguishable.

First, we prove that all points in the plane are attracted to the three vertices of $\mathbf{T}_{r}$ under counterclockwise rotation. To do so, let us assume (without loss of generality) that $\mathbf{T}$ is centered at the origin with $AB=1$. Then the vertices of $\mathbf{T}$ will have the coordinates

$$A\left(0,\frac{\sqrt{3}}{3}\right),\ B\left(-\frac{1}{2},-\frac{\sqrt{3}}{6}\right),\ C\left(\frac{1}{2},-\frac{\sqrt{3}}{6}\right),$$

and the vertices of $\mathbf{T}_{r}$ and $\mathbf{T}_{l}$ will be

$$A_{1}\left(\frac{1}{14}\,\raisebox{1.99168pt}{,}\ 5\frac{\sqrt{3}}{42}\right),\ B_{1}\left(-\frac{3}{14},-\frac{\sqrt{3}}{42}\right),\ C_{1}\left(\frac{2}{14},-4\frac{\sqrt{3}}{42}\right),$$

$$A_{2}\left(-\frac{1}{14},5\frac{\sqrt{3}}{42}\right),\ B_{2}\left(\frac{3}{14},-\frac{\sqrt{3}}{42}\right),\ C_{2}\left(-\frac{2}{14},-4\frac{\sqrt{3}}{42}\right).$$

It can be easily checked that

$$A_{1}\overset{B}{\longrightarrow}B_{1}\overset{C}{\longrightarrow}C_{1}\overset{A}{\longrightarrow}A_{1}\quad\text{and}\quad A_{2}\overset{C}{\longrightarrow}B_{2}\overset{B}{\longrightarrow}C_{2}\overset{A}{\longrightarrow}A_{2},$$

where $A_{1}\overset{B}{\longrightarrow}B_{1}$ means $A_{1}$ goes to $B_{1}$ by moving halfway toward $B$. In fact, it can easily checked that if $P_{0}(x_{0},y_{0})\overset{A}{\longrightarrow}P_{1}$ and $P_{1}\overset{B}{\longrightarrow}P_{2}$, then

$$P_{1}\left(\frac{x_{0}}{2},\frac{y_{0}}{2}+\frac{\sqrt{3}}{6}\right)\quad\text{and}\quad P_{2}\left(\frac{x_{0}}{4}-\frac{1}{4},\frac{y_{0}}{4}\right).$$

It follows that

$$|P_{1}A_{1}|^{2}-|P_{2}B_{1}|^{2}=\frac{3}{16}\left[\left(x_{0}-\frac{1}{7}\right)^{2}+\left(y_{0}+\frac{2}{7\sqrt{3}}\right)^{2}\right],$$

which is always positive except when $P_{0}=C_{1}$. Similarly, if $P_{2}\overset{C}{\longrightarrow}P_{3}$, then

$$|P_{2}B_{1}|^{2}-|P_{3}C_{1}|^{2}>0$$

unless $P_{0}=A_{1}$. In other words, once we start rotating inside $\mathbf{T}$, the distance between the current point and the nearest vertex of $\mathbf{T}_{r}$ will always be greater than the distance between its image and the nearest vertex of $\mathbf{T}_{r}$. As we keep iterating, the system will tend to $\mathbf{T}_{r}$, which makes it the sole attractor of the URG under counterclockwise rotation. Similarly, it can be shown that $\mathbf{T}_{l}$ is the attractor of the game under clockwise rotation.

Even though all points in the plane are attracted to $\mathbf{T}_{r}$ under counterclockwise rotation, it is still possible to obtain periodic orbits of periods other than three. Clearly, $A$ is the only fixed point of the game, assuming we start the counterclockwise rotation at $A$. Also, $P_{0}(-1/3,0)$ is a period 2 point since

$$P_{0}\left(-\frac{1}{3},0\right)\overset{A}{\longrightarrow}P_{1}\left(-\frac{1}{6},\frac{\sqrt{3}}{6}\right)\overset{B}{\longrightarrow}P_{0}\left(-\frac{1}{3},0\right),$$

where both $P_{0}$ and $P_{1}$ belong to the line $AB$. For a period $p>2$, the periodic points fall inside $\mathbf{T}$. For example, suppose that we want to find a point of period four. Starting with the initial point $P_{0}(x_{0},y_{0})$ then going counterclockwise, we get the following sequence of points

$$P_{0}(x_{0},y_{0}) \overset{A}{\longrightarrow}P_{1}\left(\frac{x_{0}}{2}\raisebox{2.84544pt}{,}\,\frac{y_{0}}{2}+\frac{\sqrt{3}}{6}\right)\overset{B}{\longrightarrow}P_{2}\left(\frac{x_{0}-1}{4}\raisebox{2.84544pt}{,}\frac{y_{0}}{4}\right)$$

$$\overset{C}{\longrightarrow}P_{3}\left(\frac{x_{0}+1}{8}\raisebox{2.84544pt}{,}\,\frac{y_{0}}{8}-\frac{\sqrt{3}}{12}\right)\overset{A}{\longrightarrow}P_{4}\left(\frac{x_{0}+1}{16}\raisebox{2.84544pt}{,}\,\frac{y_{0}}{16}+\frac{\sqrt{3}}{8}\right).$$

Hence, the desired point can be found by solving the linear system $P_{0}=P_{4}$. This yields $x_{0}=1/15$, $y_{0}=2\sqrt{3}/15$, see Figure 5(a).

Figure 5: A periodic point of (a) period 4 and (b) period 5.

Observe that the nearest point on the orbit to $A_{1}$ is $P_{0}$, which lies on the line $P_{1}P_{3}$ since

$$P_{3}\overset{A}{\longrightarrow}P_{0}\overset{A}{\longrightarrow}P_{1}$$

This will be true whenever the period is one more than a multiple of 3. In Figure 5(b), we show a point of period five. In this case, the point $P_{0}$ must lie near $B_{1}$ on the line $B_{1}C$. This will always be true for periods that are two more than a multiple of 3. However, there are no points with a prime period that is a multiple of three. In this case, the point $P_{0}$ coincide with $C_{1}$. Otherwise, the image of $P_{0}$ will be closer to $A_{1}$ than $P_{0}$ is to $C_{1}$, which we know is impossible. Of course, we could obtain two more periodic orbits of period $p$ if we move first toward $B$ or $C$. This is equivalent to rotating the orbit obtained by moving first toward $A$ by an angle of $2\pi/3$ or $4\pi/3$.

We can prove the above statements more rigorously by finding a formula for $P_{n}$ in general. First, observe that if

$$M=\begin{array}[]{(}{rr})\frac{1}{2}&0\\[3.0pt] 0&\frac{1}{2}\end{array},\ T_{A}=\begin{array}[]{(}{c})0\\[3.0pt] \frac{1}{2\sqrt{3}}\end{array},\ T_{B}=\begin{array}[]{(}{c})\frac{-1}{4}\\[3.0pt] \frac{-1}{4\sqrt{3}}\end{array},\ T_{C}=\begin{array}[]{(}{c})\ \ \frac{1}{4}\\[3.0pt] \frac{-1}{4\sqrt{3}}\end{array}$$

then going halfway from a point $P(x,y)$ to $A$, $B$ or $C$ amounts to adding $T_{A}$, $T_{B}$ or $T_{C}$ to

$$MP=\begin{array}[]{(}{cc})\frac{1}{2}&0\\ 0&\frac{1}{2}\end{array}\begin{array}[]{(}{c})x\\ y\end{array}.$$

It follows that for $P_{0}(x_{0},y_{0})$

$$P_{3}=M^{3}P_{0}+M^{2}T_{A}+MT_{B}+T_{C}.$$

Let $T=M^{2}T_{A}+MT_{B}+T_{C}$. Then continuing this way we get

$$P_{3k}=M^{3k}P_{0}+\left(\sum_{l=0}^{k-1}M^{3l}\right)T$$ (1)

where $M^{0}$ is the identity matrix $I$. If $N=\left(I-M^{3k}\right)^{-1}$ and $S=\sum_{l=0}^{k-1}M^{3l}$, then solving $P_{3k}=P_{0}$ yields

$$P_{0}=NST$$ (2)

It happens that in our case

$$N=\begin{array}[]{(}{cc})\frac{2^{3k}}{2^{3k}-1}&0\\ 0&\frac{2^{3k}}{2^{3k}-1}\\ \end{array},\quad S=\frac{8}{7}\begin{array}[]{(}{cc})1-2^{-3k}&0\\ 0&1-2^{-3k}\\ \end{array}$$

and

$$T=\begin{array}[]{(}{cc})\ \ \frac{1}{8}\\[3.0pt] -\frac{1}{4\sqrt{3}}\\ \end{array}.$$

Therefore,

$$NS=\begin{array}[]{(}{cc})\frac{8}{7}&0\\ 0&\frac{8}{7}\\ \end{array}\quad\text{and}\quad NST=\left(\frac{1}{7},\,-\frac{2}{7\sqrt{3}}\right)=C_{1}.$$

This proves that $C_{1}$ is the only periodic point of period $3k$ for $k\in\mathbf{N}$. If $p=3k+1$, then solving $P_{3k+1}=P_{0}$, we obtain the unique point

$$\left(I-M^{p}\right)^{-1}(MST+T_{A})=\left(\frac{2^{p-1}-1}{7\left(2^{p}-1\right)},\,\frac{5\cdot 2^{p-1}+2}{7\sqrt{3}\left(2^{p}-1\right)}\right).$$

It can be easily checked that the point lies on the line $A_{1}C_{1}$. Similarly, for $p=3k+2$, we get the periodic point

$$\left(I-M^{p}\right)^{-1}(M^{2}ST+MT_{A}+T_{B})=\left(\frac{-3\ 2^{p-1}-1}{7\left(2^{p}-1\right)},\,\frac{2-2^{p-1}}{7\sqrt{3}\left(2^{p}-1\right)}\right),$$

which falls on the line $B_{1}C_{1}$.

It is important to note that all periodic orbits, except the period 3 attractor, are unstable and therefore they do not appear under iteration. In fact, they are not even periodic points in the true sense of the word. For instance, if we continue iterating the point of period 4 found earlier, we get

$$P_{0}\overset{A}{\longrightarrow}P_{1}\overset{B}{\longrightarrow}P_{3}\overset{C}{\longrightarrow}P_{4}\overset{A}{\longrightarrow}P_{5}=P_{0}\overset{B}{\longrightarrow}P_{6},$$

where $P_{6}$ is different from $P_{1}$. In other words, even if a point $P_{0}$ returns to itself after a certain number of iterates, it will then get off track and spiral toward $\mathbf{T}_{r}$. The reason behind this behavior is that the game does not only depend on the contraction ratio but also on the chosen vertex. For $P_{6}$ to be the same as $P_{1}$, we must move from $P_{5}$ toward $A$ instead of $B$, but then we would have violated the order of rotation. We conclude that if we play the URG on any point in the plane, including pseudo periodic points, we will always end up rotating around $\mathbf{T}_{r}$ or $\mathbf{T}_{l}$ jum** from one vertex to the next, depending on whether the rotation is counterclockwise or clockwise.

4 Uniform Rotation on Regular $n$-gon

The eventual behavior of the URG and FVG is the same only if the $n$-gon on which the games are played is a triangle. For $n>3$, the URG yields periodic orbits that form regular $n$-gons, while the FVG leads to star polygons. In this section, we will study the URG, leaving the FVG for the next section.

First, observe that the inner triangles $\mathbf{T}_{r}$ and $\mathbf{T}_{l}$ are tilted with respect to the original triangle $\mathbf{T}$ by an equal amount but in a direction opposite to the direction of the game, see Figure 2. That is, if $\mathbf{T}_{r}$ is rotated clockwise by an angle $\alpha$, then $\mathbf{T}_{l}$ is rotated counterclockwise by the same angle $\alpha$. Moreover, if the fraction of the distance traveled, henceforth called the contraction ratio, is changed from half the distance between the current point and the chosen vertex to any other number between $0$ and $1$, then the angle $\alpha$ changes as well. This means that $\alpha$ does not only depend on the number of vertices $n$, but also on the contraction ratio $r$, as shown Figure 6.

Figure 6: URG: (a) $n=4,\ r=1/3$ and (b) $n=5,\ r=2/3$.

In order to calculate $\alpha_{n,r}$, we first find the inner $n$-gon whose vertices are the attractor of the game. Suppose that $P(a,b)$ is a point on the attractor and that $Q$ is the image of $P$ as we move toward a vertex $V$ of the outer polygon $\mathbf{P}_{n}$, see Figure 7. If $r$ is the fraction of $PV$ traveled, then the point $Q$ is given by

$$Q=(1-r)P+rV$$ (3)

Figure 7: The angle $\alpha$ for the URG on a regular $n$-gon.

Let $M_{n}$ be the $2\times 2$ matrix that rotates a point in the plane by $2\pi/n$. Then we can find $P$ by solving the equation

$$M_{n}^{n-1}Q=P.$$ (4)

In particular, if $V$ has the coordinates $(x,y)$ and $\rho=r-1$, then the coordinates $(a,b)$ of $P$ are

$$(\rho+1)\left(\frac{\rho x+x\cos\frac{2\pi}{n}+y\sin\frac{2\pi}{n}}{1+\rho^{2}+2\rho\cos\frac{2\pi}{n}},\ \frac{\rho y-x\sin\frac{2\pi}{n}+y\cos\frac{2\pi}{n}}{1+\rho^{2}+2\rho\cos\frac{2\pi}{n}}\right)$$ (5)

Having found the coordinates of $P$ and thus $Q$, it is not difficult to see that the angle by which the inner $n$-gon is rotated with respect to the outer $n$-gon is simply $\alpha=\angle QOV$. Moreover, since $W=M_{n}^{n-1}V$, $P=M_{n}^{n-1}Q$, and $P$, $Q$, and $V$ are collinear, we must also have $\alpha=\angle PVW$. It follows that if we use the values of $a$ and $b$ obtained in (5), then both formulas for $\alpha$ should produce the same answer. The tedious calculation yields

$$\alpha_{n,\rho}=\cos^{-1}\left(\frac{1+\rho\cos\frac{2\pi}{n}}{\sqrt{1+\rho^{2}+2\rho\cos\frac{2\pi}{n}}}\right)$$ (6)

Note that as $r\to 1$, $\rho\to 0$ and so $\alpha_{n,\rho}$ goes to $\cos^{-1}(1)=0$. On the other hand, as $r\to 0$, $\rho\to-1$ and we have

$$\lim_{\rho\to-1}\alpha_{n,\rho} =\cos^{-1}\left(\frac{2\sin^{2}\frac{\pi}{n}}{\sqrt{4\sin^{2}\frac{\pi}{n}}}\right)=\cos^{-1}\left(\sin\frac{\pi}{n}\right)$$

$$=\cos^{-1}\left(\cos\left(\frac{\pi}{2}-\frac{\pi}{n}\right)\right)=\frac{(n-2)\pi}{2n},$$

which is half the angle of a regular $n$-gon.

Although the contraction ratio $r$ could be any number in $[0,1]$, it is more natural to choose a ratio that produces a fractal similar to the Sierpinski gasket under the CG. This particular ratio has been called the kissing ratio, see [5, 3]. In fact, it was shown in [1] that the kissing ratio is

$$r=k_{n}=\frac{\sin\frac{\pi+2\pi\left\lfloor\frac{n}{4}\right\rfloor}{n}}{\sin\frac{\pi+2\pi\left\lfloor\frac{n}{4}\right\rfloor}{n}+\sin\frac{\pi}{n}}=\begin{cases}\dfrac{1}{1+\tan\frac{\pi}{n}}&\text{if }n\equiv 0\bmod 4\\[15.0pt] \dfrac{1}{1+2\sin\frac{\pi}{2n}}&\text{if }n\equiv 1,3\bmod 4\\[15.0pt] \dfrac{1}{1+\sin\frac{\pi}{n}}&\text{if }n\equiv 2\bmod 4\end{cases}$$

where $n$ is the number of sides in the regular polygon. Moreover, if we define

$$k_{x}=\frac{\sin\frac{\pi+2\pi\left\lfloor\frac{x}{4}\right\rfloor}{x}}{\sin\frac{\pi+2\pi\left\lfloor\frac{x}{4}\right\rfloor}{x}+\sin\frac{\pi}{x}}$$

for any positive real number $x$, then we can visually see how $k_{x}$ intersects with the functions

$$g_{1}(x)=\dfrac{1}{1+\tan\frac{\pi}{x}},\quad g_{2}(x)=\dfrac{1}{1+2\sin\frac{\pi}{2x}},\quad\text{and}\quad g_{3}(x)=\dfrac{1}{1+\sin\frac{\pi}{x}}$$

at only integer values, as illustrated in Figure 8.

[4.,0.58.,0.70712.,0.788] [5.00,0.6189.00,0.7427.00,0.69211.0,0.778] [6.00,0.66710.0,0.764]

Figure 8: The intersection of $k_{x}$ with $g_{1}$, $g_{2}$ and $g_{3}$.

Clearly, for $x\geq 3$ we have

$$g_{1}(x)<g_{2}(x)<g_{3}(x).$$

Note how $g_{1}$ touches $k_{x}$ only from below, $g_{2}$ crosses $k_{x}$, and $g_{3}$ touches $k_{x}$ only from above. If we set $\kappa_{n}=k_{n}-1$, then for a uniform rotation with a contraction ratio $k_{n}$, we have

$$\alpha_{n}=\cos^{-1}\left(\frac{1+\kappa_{n}\cos\frac{2\pi}{n}}{\sqrt{1+\kappa_{n}^{2}+2\kappa_{n}\cos\frac{2\pi}{n}}}\right)$$ (7)

Here $\alpha_{n}$ is a function of $n$ only since the kissing ratio is a function of $n$ as well. Curiously, the angle $\alpha_{n}$ increases as $n$ goes from $3$ to $4$ then decreases from $n=4$ on, tending to zero as $n$ goes to infinity, see Figure 9. Another curiosity is that on the way down from the maximum at $n=4$, when we reach $n=6$ we get $\alpha_{6}=\alpha_{3}$, which is the only case where two distinct integers yield the same value of $\alpha$.

Figure 9: Tilt angle for the URG with $r=k_{n}$.

Next, we will show that the sum of $\alpha_{n,r}$ diverges for $0<r<1$. Using L’Hôpital’s rule, it can be proved that

$$\lim_{n\to\infty}\frac{\alpha_{n,r}}{\frac{1}{n}}=\frac{0}{0}=2\pi\left(\frac{1}{r}-1\right),$$

which is positive for $r\in(0,1)$. By the limit comparison test, it follows that

$$\sum_{n=3}^{\infty}\alpha_{n,r}=\infty.$$

On the other hand, if we replace $r$ by the kissing ration $k_{n}$, then the sum $\alpha_{n}$ converges since in this case $k_{n}\to 1$ as $n\to\infty$. This can be proved by splitting the series into three series depending on the remainder of the division of $n$ by 4. It turned out

$$\lim_{n\to\infty}\frac{\alpha_{n}}{\frac{1}{n^{2}}}=\begin{cases}\frac{\pi^{2}}{8}&\text{if }n=2k\\[5.0pt] \frac{\pi^{2}}{2}&\text{if }n=2k+1\end{cases}$$

Hence, the sum of the whole series converges, again by the limit comparison test.

5 Farthest Vertex on Regular $n$-gon

For $n>3$, the FVG yields different attractors than the URG. In fact, the eventual behavior of the FVG depends on whether $n$ is even or odd. More precisely, for $n$ odd, the attractors are still periodic orbits of $n$ points each; but instead of drawing a regular polygon, the line segments joining consecutive vertices of an attractor now draw a star polygon, see Figure 10(a). On the other hand, for $n$ even, we get $n/2$ attractors, each consisting of two points that lie on a line joining opposite vertices of the original $n$-gon, as shown in Figure 10(b).

Figure 10: Attractors of the FVG (a) $n$ odd and (b) $n$ even.

Suppose that $n$ is odd, $P$ is a vertex of the attracting star polygon, and $Q$ is the image of $P$ as we travel toward the vertex $V(x,y)$ of the outer polygon $\mathbf{P}_{n}$, see Figure 11. First, note that if $n$ is odd and $V$ is chosen on the $y$-axis, then the two vertices that are farthest from $V$ lie on the horizontal side $WW^{\prime}$ of $\mathbf{P}_{n}$. In addition, if $m=(n-1)/2$, then the $W$ and $W^{\prime}$ are exactly $m$ vertices from $V$, depending on whether we move clockwise or counterclockwise. This means that moving counterclockwise from $V$, we reach $W^{\prime}$ after a rotation by $\pi-\pi/n$ and $W$ after a rotation by $\pi+\pi/n$. It follows that we can get back from $Q$ to $P$ by a rotation $M$ of $\pi+\pi/n$, and so the coordinates of $P$ can be found by solving the equation $P=MQ$. In this case, $P$ is given by

$$(\rho+1)\left(\frac{\rho x-x\cos\frac{\pi}{n}+y\sin\frac{\pi}{n}}{1+\rho^{2}-2\rho\cos\frac{\pi}{n}},\ \frac{\rho y-x\sin\frac{\pi}{n}-y\cos\frac{\pi}{n}}{1+\rho^{2}-2\rho\cos\frac{\pi}{n}}\right)$$ (8)

As for the URG, the angle by which star polygon is rotated with respect to the outer $n$-gon is $\alpha=\angle QOV=\angle PVW$. But $W$ in this case is the farthest clockwise vertex from $V$, while it was the closest clockwise vertex to $V$ in the URG. Knowing that $W=MV$, we can calculate either $\angle QOV$ or $\angle PVW$ to obtain

Figure 11: The angle $\alpha$ for the FVG on a regular $n$-gon.

$$\alpha_{n,\rho}=\cos^{-1}\left(\frac{1-\rho\cos\frac{\pi}{n}}{\sqrt{1+\rho^{2}-2\rho\cos\frac{\pi}{n}}}\right)$$ (9)

Of course the other inner star will be rotated in a clockwise direction. As in the case of the triangle, the choice of rotation depends on where inside the $n$-gon the system lands first. In Figure 11, if we assume that $P$ is the first iterate that falls inside the pentagon, then clearly the game will continue in counterclockwise direction. On the other hand, the reflection of $P$ with respect to the $y$-axis will lead to clockwise rotation. Moreover, counterclockwise rotation will force the inner star to be rotated clockwise with respect to the outer polygon and vice versa.

Having found the formulas for $P$ and $\alpha$ for the FVG when $n$ is odd, it only is natural to compare them with those obtained for the URG. In fact, it is noticeable that one can get from the URG to the FVG simply by replacing every occurrence of $\cos 2\pi/n$ by $-\cos\pi/n$. From this we see that when $n=3$, the two games yield the same attractors since $\cos 2\pi/3=-\cos\pi/3$.

Another similarity between the two games is that when $r\to 1$, $\alpha$ tends to $0$ for all $n$. Also, as in the URG, the sum of $\alpha_{n,r}$ diverges for any fixed $r$, while the sum converges if we take $r=k_{n}$. However, as $r\to 0$, we get

$$\lim_{\rho\to-1}\alpha_{n,r}=\cos^{-1}\left(\frac{2\cos^{2}\frac{\pi}{2n}}{\sqrt{4\cos^{2}\frac{\pi}{2n}}}\right)=\cos^{-1}\left(\cos\frac{\pi}{2n}\right)=\frac{\pi}{2n},$$

which is different from the value of $\alpha_{n,r}$ in the URG. Moreover, in this case, $\alpha_{n+1}<\alpha_{n}$ for $n\geq 3$, as shown in Figure 12.

Figure 12: Tilt angle for the FVG with $r=k_{n}$.

Finally, if $n$ is even, then we get $n/2$ attractors, each consisting of a pair of points falling on a diagonal joining opposite vertices of the outer $n$-gon. In other words, if $Q$ is the image of $P$ as we move toward $V(x,y)$, then $P=MQ$ is the image of $Q$, where $M$ is rotation by $\pi$. In this case, the coordinates of $P$ are

$$\frac{\rho+1}{\rho-1}\left(x,y\right)$$ (10)

Note that the coordinates of $P$ are now independent of $n$. Also, the inner polygon whose vertices are the $n/2$ attracting pairs of points is not rotated with respect to the original $n$-gon, and so $\alpha=0$, see Figure 10(b). In fact, we could have obtained the formulas for $P$ and $\alpha$ from the formulas for $n$ odd by replacing $\pi/n$ by 0.

6 Transformation Matrix

The counterclockwise rotation game on $\mathbf{T}$ with contraction ratio $1/2$ yielded the triangle $\mathbf{T}_{r}$, whose sides are $1/\sqrt{7}$ the sides of $\mathbf{T}$. In general, if we take the regular $n$-gon to be centered at the origin with $(x,y)$ as one of its vertices, then that the point $P$ of the attractor that moves toward$(x,y)$ was given by (5). Also, if we calculate the length of the side of the inner polygon, which is the distance between $P$ and its image under a rotation by $2\pi/n$, we get

$$l_{n,\rho}=\frac{2c(\rho+1)\sin\frac{\pi}{n}}{\sqrt{1+\rho^{2}+2\rho\cos\frac{2\pi}{n}}}$$ (11)

where $c=\sqrt{x^{2}+y^{2}}$ is the radius of the circumscribing circle. Since the length of the side of a regular polygon in a circle of radius $c$ is $L_{n}=2c\sin\pi/n$, we get a scaling factor

$$\lambda_{n,\rho}=\frac{l_{n.\rho}}{L_{n}}=\frac{\rho+1}{\sqrt{1+\rho^{2}+2\rho\cos\frac{2\pi}{n}}}$$ (12)

Now knowing the angle $\alpha_{n,\rho}$ and the scaling factor $\lambda_{n,\rho}$, it follows that if

$$M_{n,\rho}=\begin{array}[]{(}{rr})\cos\alpha_{n,\rho}&-\sin\alpha_{n,\rho}\\ \sin\alpha_{n,\rho}&\cos\alpha_{n,\rho}\end{array}$$

then $\lambda_{n,\rho}M_{n,\rho}$ transforms the original $n$-gon to the attracting inner $n$-gon. The other inner $n$-gon is obtained y replacing $\alpha_{n,\rho}$ by $-\alpha_{n,\rho}$. In Figure 13, we show the inner polygons for a triangle and a square for

$$r=\frac{1}{4},\,\frac{1}{2}\,\frac{2}{3},\,\frac{6}{7}.$$

Figure 13: The inner polygons for different ratios.

Observe how the angle $\alpha_{n,\rho}$ decreases as $\lambda_{n,\rho}$ increases so that each extended side of the inner polygons passes exactly through one vertex of the original polygon. This extremely simple yet powerful conclusion can be restated as an elegant

Theorem 1.

Let $\mathbf{P}_{n}$ be a regular $n$-gon. Then the attractor of a counterclockwise (clockwise) uniform rotation with contraction ratio $r$ is a periodic orbit of period $n$ that can be obtained by applying the linear transformation $\lambda M$ to the vertices of $\mathbf{P}_{n}$, where $M$ is a clockwise (counterclockwise) rotation by

$$\alpha_{n,\rho}=\cos^{-1}\left(\frac{1+\rho\cos\frac{2\pi}{n}}{\sqrt{1+\rho^{2}+2\rho\cos\frac{2\pi}{n}}}\right),$$

$\lambda$ is a scaling factor given by

$$\lambda_{n,\rho}=\frac{1+\rho}{\sqrt{1+\rho^{2}+2\rho\cos\frac{2\pi}{n}}},$$

and $\rho=r-1$.

We can derive a similar theorem for the FVG by thinking of the attractors, whether they are vertices of star polygons or endpoints of lines through the origin, as diagonals inside an $n$-gon, see Figure 14. If $n$ is odd, then the scaling factor $\lambda$ can be obtained from the URG by replacing $\cos(2\pi/n)$ by $-\cos(\pi/n)$, while if $n$ is even we replace $\cos(2\pi/n)$ by $-\cos(0)$. This yields

Figure 14: (a) Star polygon (b) rays from the origin.

Theorem 2.

Let $\mathbf{P}_{n}$ be a regular $n$-gon with $n$ odd. Then the attractor of the FVG with contraction ratio $r$ is a periodic orbit of period $n$ that can be obtained by applying the linear transformation $\lambda M$ to the vertices of $\mathbf{P}_{n}$, where $M$ is a clockwise or counterclockwise rotation by

$$\alpha_{n,\rho}=\cos^{-1}\left(\frac{1-\rho\cos\frac{\pi}{n}}{\sqrt{1+\rho^{2}-2\rho\cos\frac{\pi}{n}}}\right),$$

$\lambda$ is a scaling factor given by

$$\lambda_{n,\rho}=\frac{1+\rho}{\sqrt{1+\rho^{2}-2\rho\cos\frac{\pi}{n}}},$$

and $\rho=r-1$.

Whether the rotation is clockwise or counterclockwise depends on the location of the initial point in the plane and consequently the place where it first lands in $\mathbf{P}_{n}$.

Finally, if $n$ is even, we get $n/2$ attractors, each consisting of a period 2 orbit. In this case, there is no rotation and the scaling factor is simply

$$\lambda_{\rho}=\frac{1+\rho}{1-\rho},$$ (13)

which does not depend on $n$. Indeed, apart from the fact that the number of attractors is now $n/2$ instead of $2$, the formulas for $\alpha$ and $\lambda$ we derived for odd $n$ yields the corresponding formula for even $n$, if $\pi/n$ in the formula for odd $n$ is replaced by zero.

7 Basins of Attraction

For the URG, all points in the plane are attracted to $\mathbf{T}_{r}$ under counterclockwise rotation and to $\mathbf{T}_{l}$ under clockwise rotation. On the other hand, if we try to find the basins of attraction of $\mathbf{T}_{r}$ and $\mathbf{T}_{l}$ under the FVG, we get the carpet shown in Figure 15(a) with the triangle $\mathbf{T}$ in the middle. It is not difficult to see that light regions are attracted to $\mathbf{T}_{r}$ while dark regions are attracted to $\mathbf{T}_{l}$. More precisely, a point in a light (dark) region outside $\mathbf{T}$ will eventually reach a light (dark) region inside $\mathbf{T}$, and thus spiral counterclockwise (clockwise) toward $\mathbf{T}_{r}$ ($\mathbf{T}_{l}$). Like the original triangle $\mathbf{T}$, the basins of attraction are invariant under rotation by $2k\pi/3$ for $k=1,2,3$, which can be clearly seen in Figure 15(b).

Figure 15: Basins for $n=3$ (a) square of side 7 (b) circle of radius 7.

Moreover, the dark and light regions are reflections of each other with respect to the axes of symmetry of $\mathbf{T}$, where an axis of symmetry of a regular $n$-gon with odd $n$ is a line connecting a vertex to the midpoint of the opposite side. In other words, the two basins of attraction are invariant under the rotational symmetries of the dihedral group $\mathbf{D}_{3}$, and are the images of each other under the reflection symmetries of $\mathbf{D}_{3}$. Of course, the same is true for the attractors $\mathbf{T}_{r}$ and $\mathbf{T}_{l}$.

If we do the same for a regular pentagon, we obtain the wallpaper shown in Figure 16. Again, the basins of attraction have the rotational symmetries of $\mathbf{D}_{5}$ and are mirrors of each other with respect to reflection symmetries of $\mathbf{D}_{5}$.

Figure 16: Basins for $n=5$ (a) square of side 17 (b) circle of radius 7.

This is particularly interesting as it provides us with quasi pentagonal tiling of the plane, since a truly pentagonal tiling is forbidden. Clearly, we can generate similar tessellations of the plane by finding the basins of attractions of the FVG played on any regular $n$-gon with odd $n$.

For $n$ even, the number of attractors is $n/2$ and so the number of basins increases with $n$ leading to more intricate wallpapers. Moreover, there are now $n/2$ axes of symmetry connecting the midpoints of opposite sides and $n/2$ axes of symmetry connecting opposite vertices. Taking $n=4$, we get the tiling in Figure 17(a). Observe that the basins are now invariant only under a rotation by $k\pi$, $k=1,2$, and not the full rotational symmetries of the dihedral group $\mathbf{D}_{4}$. This loss of rotational symmetries is compensated by a gain in reflection symmetries, since in this case each basin is symmetric with respect to the axes joining opposite vertices.

Figure 17: Basins for $n=4$ in a square of side 15 (a) $r=0.5$ (b) $r=0.4$.

That is, each basin has both rotational and reflection symmetries. Moreover, the basins are the images of each other under a rotation by $k\pi/2$, $k=1,3$, and when reflected with respect to the axes of symmetries obtained by joining the midpoints of opposite sides.

For the regular hexagon $\mathbf{P}_{6}$, we get the wallpaper shown in Figure 18(a). From darkest to lightest, let the three basins of attraction be called $\mathcal{B}_{1},\ \mathcal{B}_{2}$ and $\mathcal{B}_{3}$. Also, let $l_{i}$ be the axis of symmetry joining the opposite vertices of $\mathbf{P}_{6}$ that are contained in $\mathcal{B}_{i}$, and let $m_{i}$ be the axis of symmetry perpendicular to $l_{i}$. Clearly, $m_{i}$ passes through the midpoints of opposite sides of $\mathbf{P}_{6}$ and the symmetry of the tile in Figure 18(b) can be described as follows:

1. $\mathcal{B}_{i}$ is invariant under a rotation by $2k\pi/6$ for $k=3,6$.

2. $\mathcal{B}_{i}$ is invariant under a reflection with respect to $l_{i}$ and $m_{i}$.

3. If a symmetry of the dihedral group $\mathbf{D}_{6}$ does not preserve $\mathcal{B}_{i}$, then it will map it onto another basin $\mathcal{B}_{j}$.

Figure 18: Basins for $n=5$ (a) Square of side 10 (b) $(x,y)$ with $x\in[-5,5]$.

The tile in Figure 18(b) was generated by taking all points in $\mathcal{B}_{1}$ whose $x$-coordinates are in $[-5,5]$, and then rotating those points by $\pi/3$ and $2\pi/3$. Of course, the symmetries of the tile, as listed above, can be extended to the tessellation of the full plane. Moreover, these symmetries can be similarly described for any regular $n$-gon with $n$ even. In fact, we can do so without even knowing how the $n/2$ basins look like, since the symmetries of the basins are directly inherited from the symmetries of the $n/2$ attractors.

Finally, changing the contraction ratio $r$ amounts to scaling the basins of attraction, but does not change the symmetries of the tessellation. In Figure 17(b), we changed the contraction ratio from $k_{4}=0.5$ to $0.4$. Observe how the overall structure of the wallpaper is preserved, but everything is now scaled down.

8 Conclusion

We have seen how random and uniform motion around a regular $n$-gon produced strikingly different results. Also, we were able to fully describe the behavior of the URG game and the FVG and in the process identify their similarities and differences. In particular, we have shown how for $n=3$ the two games lead to the same eventual behavior. That is, they have the same attractors. But what is most important about this work is that starting with any initial point in the plane, we can tell exactly where each dynamical system will end up.

In the URG, there are two attractors for each $n$ and $r$, one for clockwise rotation and the other for counterclockwise rotation. The attractors can be determined without iterating the system since they are merely a rotation of the original $n$-gon by angle $\alpha$ and a scaling down by a factor $\lambda$, where $\alpha$ and $\lambda$ can be fully calculated using only $n$ and $r$.

For the FVG, the situation is slightly more involved. If $n$ is odd, we still get two attractors, but we will not know where the initial point will end up until it enters the original $n$-gon, as the game unfolds. On the other hand, if $n$ is even, we get $n/2$ attractors, each consisting of a period 2 orbit. Again, we cannot know in advance to which of the attractors will the initial point converge. To overcome this lack of knowledge about the final attractor, we iterated a grid of points around the original $n$-gon, which led us to the discovery of some beautifully intricate patterns that tile the plane.