Reflection matrix:

$$ \text{Reflection}(\theta) = \begin{bmatrix} \cos2\theta & \sin2\theta \\ \sin2\theta & -\cos2\theta \end{bmatrix}$$


Inspiration: enter image description hereenter image description here Speaking non-rigorously, it seems like the angle between the reflected vector and the original vector will be $2\theta$. Armed with this, let's consider how $e_1 = \begin{bmatrix}1\\0\end{bmatrix}$ and $e_2 = \begin{bmatrix}0\\1\end{bmatrix}$ change when we reflect them across an arbitrary line.

Let $$ \text{Reflection}(\theta) = \begin{bmatrix} a & b \\ c & d \end{bmatrix}$$


$$\begin{align} \text{Reflection}(\theta) \cdot e_1 &= \begin{bmatrix} a & b \\ c & d \end{bmatrix} \cdot \begin{bmatrix}1\\0\end{bmatrix} \\ &= \begin{bmatrix}a\\c\end{bmatrix} \end{align}$$

Using my assumption that reflected vectors have angle of $2\theta$ between itself and the original vector, $(\text{Reflection}(\theta)\cdot e_1) \cdot e_1 = \begin{bmatrix}a\\c\end{bmatrix}\cdot \begin{bmatrix}1\\0\end{bmatrix} = a = ||\text{Reflection}(\theta)\cdot e_1||\cdot||e_1|| \cos(2\theta)$ (dot product).

Simplifying the other side of the equation, we get: $$a = 1\cdot 1 \cos(2\theta) = \cos(2\theta)$$

Doing it similarly for $e_2$ yields $d=\cos(2\theta)$ which means our reflection matrix currently looks like this:

$$ \text{Reflection}(\theta) = \begin{bmatrix} \cos2\theta & b \\ c & \cos2\theta \end{bmatrix}$$

which is not correct.


  1. I suspect my implicit assumption that a reflection can be represented as a rotation of $2\theta$, where $\theta$ is the angle between the original vector and reflection line, is where I went wrong. Why is this wrong?

$$ \text{Rotation}(2\theta) = \begin{bmatrix} \cos2\theta & -\sin2\theta \\ \sin2\theta & \cos2\theta \end{bmatrix} \text{ for reference. }$$

  1. What's the correct way to do this?

I would normally ask one question per SE question but I think my two questions are tightly coupled.


The problem with your attempted derivation is that you keep changing the definition of $\theta$ since "the original vector" can be any input to the linear transformation.

Presumably, the $\theta$ in the definition of the reflection is such that the axis of reflection is along the vector $(\cos \theta, \sin \theta)$, which is to say that $\theta$ is the angle between the axis of rotation and the vector $e_1$. With that in mind, your derivation of the entry $a$ is fine, but your derivation of the entry $d$ is unjustified and ultimately incorrect.

Here's a geometric approach involving rotations that you might like. Let $F_\theta$ denote the matrix of the reflection through the axis at angle $\theta$, and let $R_\theta$ denote the matrix of the counterclockwise rotation by $\theta$. Note that the matrix of the reflection through the axis at $\theta = 0$ is given by $$ F_0 = \pmatrix{1&0\\0&-1}. $$ On the other hand, we can implement a reflection through the axis at angle $\theta$ by first rotating that axis to $\theta = 0$, then applying $F_0$, then rotating the axis back. In other words, we have $$ F_\theta = R_{\theta} F_0 R_{-\theta} = \pmatrix{\cos \theta & - \sin \theta\\ \sin \theta & \cos \theta} \pmatrix{1&0\\0&-1} \pmatrix{\cos \theta & \sin \theta \\ - \sin \theta & \cos \theta}. $$ After multiplying and simplifying with the double-angle laws, you should indeed find that $$ F_\theta = \pmatrix{\cos 2 \theta & \sin 2 \theta \\ \sin 2 \theta & - \cos 2 \theta}. $$

  • $\begingroup$ Hey @Omnomnomnom, can you elaborate more on why entry $d$ would be wrong? I was imagining the same reflecting line for both basic vectors (at least in my head). $\endgroup$ – Darius Oct 20 '19 at 9:27
  • $\begingroup$ @Darius my point against your explanation is that while $e_1$ moves through an angle $\theta$, $e_2$ moves an amount corresponding to its angle of separation from the axis, which would be $90^\circ - \theta$. Really, though, I can't be sure how exactly you got your entry $d$ because all you said is that you "did it similarly". $\endgroup$ – Ben Grossmann Oct 20 '19 at 9:35
  • $\begingroup$ I see my mistake now. I was under the impression that $\text{Reflection}(\theta)$ is how much I want to "rotate" my point ($e_2$ in this case) when $\theta$ is only referring to the angle that reflecting line makes with the $x$-axis. +1 for your explanation of where I went wrong and providing a nice solution that I like! $\endgroup$ – Darius Oct 20 '19 at 9:46

The angle parameterising a rotation is not how far a line moves when reflected, but the tilt of the line in which reflection occurs. The line $\ell$ of equation $y=x\tan\theta$ sends $p:=\binom{a}{b}$ to a point $p^\prime:=\binom{a^\prime}{b^\prime}$ for which $\vec{pp^\prime}$ is orthogonal to $\ell$ and hence of gradient $-\cot\theta$, so $\vec{pp^\prime}$ has equation $y-b=(a-x)\cot\theta$. The intersection with $\ell$ solves $x(\tan\theta+\cot\theta)=a\cot\theta+b$, so has $x$-coordinate $\cos^2\theta(a+b\tan\theta)$ and $y$-coordinate $\sin\theta\cos\theta(a+b\tan\theta)$. So $$a^\prime=2\cos^2\theta(a+b\tan\theta)-a=a\cos2\theta+b\sin2\theta,\\b^\prime=2\sin\theta\cos\theta(a+b\tan\theta)-b=a\sin2\theta-b\cos2\theta.$$This agrees with the reflection matrix here.

  • $\begingroup$ +1 for pointing out my error ("a rotation is not how far a line moves when reflected, but the tilt of the line in which reflection occurs") and giving an example method to prove :) $\endgroup$ – Darius Oct 20 '19 at 9:25

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.