Comments on Reflection in the plane with polar coordinates
Parent
Reflection in the plane with polar coordinates
My question comes from the academic paper: Symmetry of solutions to semilinear elliptic equations via Morse Index. The author is Filomena Pacella.
Let $S$ be the vector of the $n$-dimensional unit sphere in $\mathbb{R}^N$, $S=\{x \in \mathbb{R}^N : |x|=1\}$. For a unit vector $e \in S$, we consider the hyperplane $H(e)=\{x\in\mathbb{R}^N : x\cdot e=0 \}$ and the open half-domain of the $n$-sphere $B(e)=\{ x\in B\ : x\cdot e > 0 \}$. We define $\sigma_{e} : B \rightarrow B$ as the reflection with respect to $H(e)$, that is, $\sigma_{e}(x)=x-2(x\cdot e)e$ for each $x \in B$. Since $\sigma_{e}(x)$ is a reflection, the following properties hold \begin{equation*} H(-e)=H(e) \quad \text{y} \quad B(-e)=\sigma_{e}(B(e))=-B(e) \quad \text{for each} \; e\in S \end{equation*} A new vector $e^{\prime}=(\cos(\theta_{0}), \sin(\theta_{0}), 0, \ldots, 0)$ is defined for some $\theta_{0} \in \left(- \frac{\pi}{2}, \frac{\pi}{2}\right)$. Since only the first two coordinates are modified, the remaining coordinates are grouped into $\tilde{x}=(x_3, \ldots, x_N)$. We convert to cartesian coordinates to polar coordinates for the first two, $x_1= r \cos(\theta)$, $x_2= r \sin(\theta)$. We use the reflection with vector $e^{\prime}$ over the hyperplane $H(e^{\prime})$, so we name it $\sigma_{e^{\prime}}(x)$. The coordinates of the image after applying the reflection are \begin{equation*} \sigma_{e^{\prime}}(r\cos(\theta),r\sin(\theta),\tilde{x})=(r\cos(2\theta_{0}-\theta + \pi),r\sin(2\theta_0-\theta + \pi),\tilde{x}) \end{equation*} Do you know how it arrived at the expression $2\theta_{0} - \theta + \pi$? Is any trigonometric property being used here? One more thing, could someone share a link with a diagram explaining this application? Linear transformations that are reflections are usually defined with matrices and those that are not defined this way are not similar to this one.
Post
Perhaps the simplest way to get the formula is to think geometrically.
Let's say we wanted to reflect a point $x=(r\cos\theta,r\sin\theta)$ across the $x$-axis. In that case, we can simply negate $\theta$ giving $(r\cos(-\theta),r\sin(-\theta))=(r\cos\theta,-r\sin\theta)$ as expected.
If we want to reflect across a line at $\phi$ radians from the $x$-axis, we simply counter-rotate by $\phi$, i.e. subtract $\phi$ from the angles, to get into the above situation, negate to reflect, then rotate by $\phi$, i.e. add $\phi$ to the angles, to take things back to the original arrangement. Focusing just on the angles, this gives $-(\theta-\phi) + \phi = 2\phi - \theta$. The $\pi$ in the formulas in the OP is due to the fact that the $e'$ vector is the normal to the (hyper-)plane of reflection so $\phi$ is 90° or $\frac{\pi}{2}$ radians off from it, i.e. $\phi = \theta_0 \pm \frac{\pi}{2}$ leading to $2\phi-\theta = 2\theta_0 \pm \pi - \theta$. (The sign of $\pi$ doesn't matter here.)
If you'd prefer an algebraic approach, here's an approach using geometric algebra.
First, we can represent the reflection of the vector $\mathbf x$ in the (hyper-)plane orthogonal to a unit vector $\mathbf e$ by the simple expression $-\mathbf{exe}$. Using the basic geometric algebra identities $\mathbf{uv} = \mathbf u\cdot\mathbf v + \mathbf u\wedge \mathbf v$ and $\mathbf u\cdot(\mathbf v\wedge \mathbf w) = (\mathbf u\cdot \mathbf v)\mathbf w - (\mathbf u\cdot \mathbf w)\mathbf v$ (and, implicitly, $\mathbf u\wedge \mathbf u = 0$), we get: $$\begin{align} -\mathbf{exe} &= -(\mathbf x\cdot \mathbf e)\mathbf e - \mathbf e(\mathbf x\wedge \mathbf e) \\ &= -(\mathbf x\cdot \mathbf e)\mathbf e - \mathbf e\cdot(\mathbf x\wedge \mathbf e) \\ &= -(\mathbf x\cdot \mathbf e)\mathbf e - (\mathbf e\cdot \mathbf x)\mathbf e + (\mathbf e\cdot \mathbf e)\mathbf x \\ &= \mathbf x - 2(\mathbf x\cdot \mathbf e)\mathbf e \end{align}$$ reproducing the formula for $\sigma_{e'}$.
Now we need to talk about rotation. The general formula for rotating a vector $\mathbf x$ by a $\theta$ radians in the plane spanned by the unit vectors $\mathbf u$ and $\mathbf v$ is $e^{-B\theta/2}\mathbf xe^{B\theta/2}$ where $B=(\mathbf u\wedge \mathbf v)/\vert \mathbf u\wedge \mathbf v\vert$ and, since $B^2 = -1$, the exponential expands a la Euler's formula $e^{B\theta} = \cos\theta + B\sin\theta$. In the special case that $\mathbf x$ is in the span of $\mathbf u$ and $\mathbf v$, then $\mathbf x(\mathbf u\wedge \mathbf v) = -(\mathbf u\wedge \mathbf v)\mathbf x$ and the rotation formula simplifies to $e^{-B\theta}\mathbf x$. But if we compare this to $\mathbf{vu}$ we get: $$\mathbf{vu} = \mathbf v\cdot \mathbf u + \mathbf v\wedge \mathbf u = \mathbf v\cdot \mathbf u - \vert \mathbf u\wedge \mathbf v\vert B = \cos\theta - B\sin\theta = e^{-B\theta}$$ where $\theta$ is the angle between $\mathbf u$ and $\mathbf v$. In other words, $\mathbf{vu}$ represents the rotation that would rotate $\mathbf u$ into $\mathbf v$, evident from the fact that $\mathbf{vuu}=\mathbf v$.
Thus we can view $-\mathbf{exe}$ as "rotate $\mathbf e$ by the angle between $\mathbf e$ and $\mathbf x$ and then negate". In terms of angles, this means add the angle between $\mathbf e$ and $\mathbf x$, i.e. $\theta_0-\theta$, to the angle $\mathbf e$ is at, i.e $\theta_0$, and then negate which is equivalent to adding $\pm\pi$ to the angle. In symbols, $\theta_0 + (\theta_0-\theta) \pm \pi = 2\theta_0 - \theta \pm \pi$.
But screw intuition, I want to compute. Write $\mathbf x=e^{-B\theta}\mathbf{e}_0$ where $\mathbf{e}_0$ is the unit vector in the $x_0$ direction, and $\mathbf e = e^{-B\theta_0}\mathbf{e}_0$ where $B=\mathbf{e}_0\mathbf{e}_1$. We now get the above paragraph in symbols: $$\begin{align} -\mathbf{exe} &= -e^{-B\theta_0}\mathbf{e}_0 e^{-B\theta}\mathbf{e}_0 e^{-B\theta_0}\mathbf{e}_0 \\ &= -e^{-B\theta_0}e^{B\theta}\mathbf{e}_0\mathbf{e}_0 e^{-B\theta_0}\mathbf{e}_0 \tag{$\mathbf{e}_0^2 = 1$}\\ &= -e^{-B(\theta_0-\theta)}e^{-B\theta_0}\mathbf{e}_0 \\ &= -e^{-B(2\theta_0-\theta)}\mathbf{e}_0 \\ &= e^{\mp B\pi}e^{-B(2\theta_0-\theta)}\mathbf{e}_0 \tag{$-1 = e^{\mp B\pi}$} \\ &= e^{-B(2\theta_0-\theta\pm\pi)}\mathbf{e}_0 \\ &= \cos(2\theta_0-\theta\pm\pi)\mathbf{e}_0 + \sin(2\theta_0-\theta\pm\pi)\mathbf{e}_1 \end{align}$$
As for matrices, you can recover them if you want by introducing a basis and seeing how each basis vector is transformed. But there's no reason to do this. Talking about matrices (as opposed to linear transformations) necessarily implies introducing a basis and coordinates, and that's simply not necessary most of the time. Often it will be necessary to introduce a basis to concretely specify a vector, but the formulas can be manipulated and derived without doing that. Usually the formulas are simpler and more general by avoiding coordinates. You can see above that the only time I introduced a basis in the second part was to connect to the coordinate-based expressions in the question.
0 comment threads