2
$\begingroup$

The given exercise is, for all $\theta$ in $\mathbb{R}$, show that the matrix always has an eigenvector in $\mathbb{R^2}$ $$ A = \begin{pmatrix} \cos(\theta) & \sin(\theta) \\ \sin(\theta) & -\cos(\theta) \end{pmatrix} $$

The use of determinants isn't allowed, it is just allowed that I should use de definition of eigenvector and eigenvalue, and the fact that the eigenvectors are linearly independent if they are distinct (if needed).

I don't fully understand the problem because I can't see how is possible that this matrix can give me a multiple scalar of some vector.

$\endgroup$
  • 1
    $\begingroup$ Should the top right entry be $\sin \theta$? Also, are we allowing complex eigenvalues/vectors? $\endgroup$ – user88319 Jan 31 '15 at 22:22
  • $\begingroup$ Yes, sin(\theta). And we can use real eigenvalues. $\endgroup$ – HeMan Jan 31 '15 at 22:24
  • $\begingroup$ To be more precise: this matrix has no real eigenvector or eigenvalue if $\theta$ is not a multiple of $\pi$. $\endgroup$ – Rory Daulton Jan 31 '15 at 22:25
  • $\begingroup$ I didn´t know the counterexample. I ´ve been stuck for hours. It is one of my homework exercises. $\endgroup$ – HeMan Jan 31 '15 at 22:25
  • 2
    $\begingroup$ The OP's matrix is not a rotation. The rotation matrix doesn't (typically) have real eigenvalues, but this one does. $\endgroup$ – Joonas Ilmavirta Jan 31 '15 at 22:29
3
$\begingroup$

AFTERTHOUGHT: it occurs to me that there is something that does not need determinant, although the concept is implicit. As you can easily confirm, we have a matrix $A$ such that $\color{red}{A^2 = I}.$ Now, if you are willing to accept the proposition that every square matrix has an eigenvalue (possibly complex) then we can write $$ Av = \lambda v $$ for $v$ a nonzero column matrix, possibly complex as well. Then $$ v = Iv = A^2 v = \lambda Av = \lambda^2 v. $$ So, in fact, $\lambda = \pm 1.$ This argument holds for things that are not reflections or symmetric; consider $$ A = \left( \begin{array}{cc} 0 & 7 \\ \frac{1}{7} & 0 \end{array} \right) $$ Also, note that possession of an eigenvalue is not automatic in infinite dimension. From what I can see, in the space of one-sided infinite sequences with, say, complex entries, the right-shift operator does not have any non-zero eigenvector. See http://en.wikipedia.org/wiki/Shift_operator#Sequences

ORIGINAL: People seem confused. This matrix gives a REFLECTION. Determinant is $-1,$ trace is $0,$ so, no matter what $\theta$ might be, the characteristic polynomial is $$ \lambda^2 - 1 $$ and the eigenvalues are $1$ and $-1.$ there is no counterexample.

It saves a little writing if you write the matrix as $$ \left( \begin{array}{cc} a & b \\ b & -a \end{array} \right) $$ with the understanding that $$ a^2 + b^2 = 1. $$ So, the $+1$ eigenvector is a column vector sent to the zero vector by $$ \left( \begin{array}{cc} a -1 & b \\ b & -a -1 \end{array} \right). $$ Think about it. With $a^2 + b^2 = 1,$ why is the matrix immediately above singular? What is the actual eigenvector in terms of $a,b?$

Similar, the $-1$ eigenvector is a column vector sent to the zero vector by $$ \left( \begin{array}{cc} a +1 & b \\ b & -a +1 \end{array} \right). $$ Again, why is this singular? Oh, symmetric matrix, the eigenvectors have different eigenvalues and are perpendicular to each other.

You can tell that a matrix is a reflection if it can be written as $\color{red}{I - 2 v v^T},$ where the letter $v$ refers to a column vector of length $1.$ Oh, in the other order, $v^T v$ is the (squared) length of $v,$ it is a 1 by 1 matrix with entry $v \cdot v.$ In the order used above, $v v^T$ is a symmetric, rank one, positive semidefinite matrix, furthermore its trace is exactly $1.$ So, the determinant of $\color{red}{I - 2 v v^T}$ is $-1$ and its trace is $n-2,$ where $n$ is the dimension. In dimension 2, you need only check the trace and determinant to know for sure.

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ Yes, I understand that. But the exercise is previous of the section of caracteristic polynomial so I can´t use the determinant for caracteristic polynomial and I can´t see a way to get the eigenvector without it. $\endgroup$ – HeMan Jan 31 '15 at 22:34
  • $\begingroup$ Just write the system with a third parameter, $\lambda,$ and a column vector with entries $x,y$ such that, for example, $x^2 + y^2 = 1$ as well. Call the 2 by 2 matrix $A,$ call the column vector $v,$ solve the system $Av = \lambda v$ by hand. $\endgroup$ – Will Jagy Jan 31 '15 at 22:37
  • $\begingroup$ oh! but if that matrix is a reflection, then the eigenvector that I´m looking for is the one who isn´t "altered" after the reflection! kinda fixed point I think. I´m going to try it again! Thank you so much! $\endgroup$ – HeMan Jan 31 '15 at 22:38
4
$\begingroup$

the matrix $\pmatrix{\cos \theta & \sin \theta\\ \sin \theta & -\cos \theta}$ represents the reflection on a mirror along the line $y = \tan (\frac{\theta}{2})\ x.$ therefore $\pmatrix{\cos (\theta/2) \\\sin(\theta/2)}$ is an eigenvector corresponding to the eigenvalue $1$ and $\pmatrix{\sin (\theta/2) \\-\cos(\theta/2)}$ is an eigenvector corresponding to the eigenvalue $-1$.

| cite | improve this answer | |
$\endgroup$
1
$\begingroup$

Suppose $Av=\lambda v$. We can scale $v$ to have unit norm, so that $v=(\cos\alpha,\sin\alpha)$. Writing out $Av=\lambda v$ gives $$ (\cos\theta\cos\alpha+\sin\theta\sin\alpha,\sin\theta\cos\alpha-\cos\theta\sin\alpha) = \lambda(\cos\alpha,\sin\alpha). $$ We can simplify the LHS: $$ (\cos(\theta-\alpha),\sin(\theta-\alpha)) = \lambda(\cos\alpha,\sin\alpha). $$ If we calculate the norm of both sides, we get $$ 1=\lambda^2\cdot1, $$ so $\lambda=\pm1$.

If $\lambda=1$, we have $\cos(\theta-\alpha)=\cos\alpha$ which implies $\theta-\alpha=\pm\alpha+2\pi n$ for some $n\in\mathbb Z$. The negative sign is only possible if $\theta$ is a multiple of $2\pi$ (this case is simple to solve separately). For the positive sign we get $\alpha=\theta/2-\pi n$ for some $n\in\mathbb Z$. Shifting $\alpha$ by $\pi$ changes the sign of $v$. Therefore, up to sign, the normalized eigenvector corresponding to $\lambda=1$ is $(\cos(\theta/2),\sin(\theta/2))$.

A similar calculation for $\lambda=-1$ gives $(\sin(\theta/2),-\cos(\theta/2))$.

| cite | improve this answer | |
$\endgroup$
1
$\begingroup$

That transformation $A$ is a reflection $P_x$ along the $x$-axis followed by a rotation $R_\theta$ by an angle $\theta$ : $$ A = \left( \begin{matrix} \cos\theta & \sin\theta \\ \sin\theta & -\cos\theta \end{matrix} \right) = \left( \begin{matrix} \cos\theta & -\sin\theta \\ \sin\theta & \cos\theta \end{matrix} \right) \left( \begin{matrix} 1 & 0 \\ 0 & -1 \end{matrix} \right) = R_\theta \, P_x $$ The length of the vector will be preserved by the transformation, because reflections and rotations do. We want $A x = \lambda x$ for some non-zero vector $x$ and real number $\lambda$. Thus $$ \lVert x \rVert = \lVert A x \rVert = \lvert \lambda \rvert \, \lVert x \rVert $$ The preserved length enforces $\lambda = \pm 1$. So a vector $u_\phi$ with angle $\phi$ and non-zero length $r$ $$ u_\phi = \left( \begin{matrix} r \cos \phi \\ r \sin \phi \end{matrix} \right) \quad (*) $$ and $$ A u_\phi = R_\theta \, P_x u_\phi = R_\theta u_{-\phi} = u_{-\phi + \theta} = u_\phi \iff \\ -\phi + \theta = \phi \iff \phi = \frac{\theta}{2} $$ will have an eigenvalue $1$. For the eigenvalue $-1$, $$ A u_\phi = - u_\phi = u_{\phi + \pi} $$ one needs a vector with $$ -\phi + \theta = \phi + \pi \iff \phi = \frac{\theta-\pi}{2} $$ Both vectors exist for any $\theta$, just use the above determined $\phi$ for any non-zero $r$ in equation $(*)$.

| cite | improve this answer | |
$\endgroup$

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.