I know that any rotation of $\mathbb{R}^2$ can be expressed as the product of two reflections in non-parallel lines, and hence the product of two rotations can be written as $R_{L_4}R_{L_3}R_{L_2}R_{L_1}$, where $R_{L_i}$ denotes reflection in the line $L_i$ for $1 \leq i \leq 4$. It is then clear that a sufficient condition for this product of two rotations to yield another rotation is for $L_2 = L_3$ (in which case the product $R_{L_3}R_{L_2}$ reduces to the identity) and for $L_1$ and $L_4$ to be non-parallel. My question is whether this is also a necessary condition for the product of two rotations of the plane to yield another rotation - I am inclined to believe that it is, and it would be greatly appreciated if someone here can quickly confirm this for me.
Necessary condition for the product of two rotations of the plane to be a rotation
181 Views Asked by Bumbble Comm https://math.techqa.club/user/bumbble-comm/detail AtThere are 2 best solutions below
On
After doing a quick search on Mathematics Meta, I see that it is generally considered acceptable to post an answer to one's own question if no complete answers have been provided by other users, so here is what I have managed to work out.
Let $a$ and $b$ be points in the plane. A rotation about $a$ by angle $\theta \in [0,2\pi)$ can be expressed as a pair of reflections in any lines $L_1$ and $L_2$ that form an angle of $\frac{\theta}{2}$ at their point of intersection $a$. Likewise, a rotation about $b$ by angle $\phi \in [0,2\pi)$ can be expressed as a pair of reflections in any lines $L_3$ and $L_4$ that form an angle of $\frac{\phi}{2}$ at their point of intersection $b$. Insofar as we can always construct lines $L_1, L_2, L_3, L_4$ such that $L_2 = L_3$ passes through both points $a$ and $b$ and forms an angle of $\frac{\theta}{2}$ and $\frac{\phi}{2}$ with $L_1$ and $L_4$ at $a$ and $b$, respectively, this "condition" as stated in my earlier question will always be satisfied. That is, the pair of rotations is always expressible as the product of reflections $R_{L_4}R_{L_3}R_{L_2}R_{L_1}$ = $R_{L_4}R_{L_1}$, with $L_2 = L_3$.
The necessary condition is merely that $\theta + \phi \neq 2\pi$, since this is what is required for $L_1$ and $L_4$ to intersect at some point $c$. The product of rotations is then a rotation about $c$ by angle $\theta + \phi$. On the other hand, if $\theta + \phi = 2\pi$, then $L_1$ and $L_4$ will be parallel, and the product of reflections $R_{L_4}R_{L_1}$ will be a translation by twice the distance between $L_1$ and $L_4$ in the direction perpendicular to $L_1$.
I want to present another approach to the question.
$\newcommand\rg{\rm rg}$ $\newcommand\arho[2]{{\raise.5ex\rho}_{#1,\,#2}}$ $\newcommand\vrho[1]{{\raise.5ex\rho}_{#1}}$Let us consider two affine rotations of the affine plane $\mathcal A=\mathbb R^2$, say
$$ \arho{O_1}{\theta_1} \; : \; P\; \to \; O_1 + \vrho{\theta_1}(P-O_1) $$ $$ \arho{O_2}{\theta_2} \; : \; Q\; \to \; O_2 + \vrho{\theta_2}(Q-O_2) $$
around $O_1$ of angle $\theta_1$ the first, and around $O_2$ of angle $\theta_2$ the second, where $\vrho {\theta_1}$ and $\vrho {\theta_2}$ are the correspondig rotations of the vector space $E=\mathbb R^2$.
Their product of composition is given by \begin{align} \arho{O_2}{\theta_2}\circ\arho{O_1}{\theta_1}(P) &= \arho{O_2}{\theta_2}\big(O_1+\vrho{\theta_1}(P-O_1)\big) =\\[1.5ex] &= O_2 + \vrho{\theta_2}\big(O_1-O_2+\vrho{\theta_1}(P-O_1)\big)=\tag{*}\label{*}\\[1.5ex] &= O_2+\vrho{\theta_2}(O_1-O_2)+\vrho{\theta_1+\theta_2}(P-O_1). \end{align}
A fixed point $\;C\;$ of $\;\arho{O_2}{\theta_2}\circ\arho{O_1}{\theta_1}$ must satisfy the equation
$$ O_2+\vrho{\theta_2}(O_1-O_2)+\vrho{\theta_1+\theta_2}(C-O_1) = C $$
so
$$ \vrho{\theta_2}(O_1-O_2)+\vrho{\theta_1+\theta_2}(C-O_1) = (C-O_1)+(O_1-O_2) $$
so
$$ (\vrho{\theta_1+\theta_2}-1_E)(C-O_1) = (\vrho{\theta_2}-1_E)(O_2-O_1), \tag{**}\label{**} $$
where $\;1_E\;$ is the identity function of the vector space $E=\mathbb R^2$.
Now, IF the linear function $\;\vrho{\theta_1+\theta_2}-1_E\;$ is invertible, we deduce that there is a unique fixed point $C$ for $\arho{O_2}{\theta_2}\circ\arho{O_1}{\theta_1}$, given by
$$ C = O_1 + (\vrho{\theta_1+\theta_2}-1_E)^{-1}\circ(\vrho{\theta_2}-1_E)(O_2-O_1) $$
and we can verify that $\;\arho{O_2}{\theta_2}\circ\arho{O_1}{\theta_1}\;$ is a rotation centered on $C$ and of angle $\;\theta_1+\theta_2$. In fact, for all $P\in \mathcal A$, we have, from \eqref{*}
\begin{align} &\arho{O_2}{\theta_2}\circ\arho{O_1}{\theta_1}(P)-C = \arho{O_2}{\theta_2}\circ\arho{O_1}{\theta_1}(P)-\arho{O_2}{\theta_2}\circ\arho{O_1}{\theta_1}(C) =\\[1.5ex] &= \big(O_2+\vrho{\theta_2}(O_1-O_2)+\vrho{\theta_1+\theta_2}(P-O_1)\big)-\big(O_2+\vrho{\theta_2}(O_1-O_2)+\vrho{\theta_1+\theta_2}(C-O)\big)= \\[1.5ex] &= \vrho{\theta_1+\theta_2}\big((P-O_1)-(C-O_1)\big) = \vrho{\theta_1+\theta_2}(P-C). \qquad\blacksquare \end{align}
The (sufficient) condition seen above is equivalent to $\;\det(\vrho{\theta_1+\theta_2}-1_E)\neq0$, that is, with respect to the standard orthonormal basis of $E$,
$$\det\begin{bmatrix} \cos(\theta_1+\theta_2)-1 & \quad-\sin(\theta_1+\theta_2) \\[2ex] \sin(\theta_1+\theta_2) & \quad\cos(\theta_1+\theta_2)-1 \end{bmatrix}\neq0 $$
i.e.
$$ 2\big(1-\cos(\theta_1+\theta_2)\big)\neq0 $$
so that a sufficient condition for $\;\arho{O_2}{\theta_2}\circ\arho{O_1}{\theta_1}\;$ to be a rotation is that
$$ \theta_1+\theta_2 \neq 0. \tag{***}\label{***}$$
Conversely, suppose that $\;\arho{O_2}{\theta_2}\circ\arho{O_1}{\theta_1}$ is a rotation around the fixed point $C$. Let us consider two cases:
We have $\;\vrho{\theta_1+\theta_2} = 1_E,\;$ so $\;$[by $\;$\eqref{*}]
$$ \arho{O_2}{\theta_2}\circ\arho{O_1}{\theta_1}(P) = O_2+\vrho{\theta_2}(O_1-O_2) + (P-O_1) = P+ (O_2-O_1) + \vrho{\theta_2}(O_1-O_2) $$
i.e. a (effective) translation or the identity $\;1_{\mathcal A}$, depending on whether
$\hskip4cm\;(O_2-O_1) + \vrho{\theta_2}(O_1-O_2)\neq 0\;$ or $\;=0$.
The equation of fixed points $\;$\eqref{**}$\,$ admits infinite solutions $\;C-O_1\;$ if it is compatible, or no solutions otherwise. For each solution $\;C-O_1$, the point $\;C\;$ is a fixed point for $\arho{O_2}{\theta_2}\circ\arho{O_1}{\theta_1}$ and vice versa. Indeed, by $\eqref{*}$ and $\eqref{**}$, we have
\begin{align} \arho{O_2}{\theta_2}\circ\arho{O_1}{\theta_1}(C) &= O_2+\vrho{\theta_2}(O_1-O_2)+\vrho{\theta_1+\theta_2}(C-O_1) =\\ &= O_2 + \vrho{\theta_2}(O_1-O_2) + C-O_1 + (\vrho{\theta_1+\theta_2}-1_E)(C-O_1) =\\ &= O_2 + \vrho{\theta_2}(O_1-O_2) + C-O_1 +\vrho{\theta_2}(O_2-O_1) + (O_1-O_2) =\\ &= O_2 + (C-O_1) + (O_1-O_2) = C. \end{align}
Similarly for vice versa.
It follows that, if the rotation $\;\arho{O_2}{\theta_2}\circ\arho{O_1}{\theta_1}\;$ is not the identity, then $\rg(\vrho{\theta_1+\theta_2}-1_E)=2$, and therefore, as we have already seen, that the inequality $\;\eqref{***}\;$ holds.