I have the matrix $$\left( \begin{array}{ccc} -\alpha & \beta \\ \beta/K & -\alpha/K \end{array} \right)$$ for which the eigenvalues are $$\lambda_{1,2}=-\dfrac{\alpha}{2}-\dfrac{\beta}{2K}\pm\dfrac{\sqrt{(K\alpha-\beta)^2+4K\alpha\beta }}{2K}$$ I have a problem with finding the eigenvectors. It seems that the eigenvectors are zero vectors but then eigenvectors cannot be zero. What am I missing?

  • 2
    $\begingroup$ If $\lambda$ is an eigenvalue, then $A-\lambda I$ is a singular matrix, and the corresponding eigenvector is a non-zero vector in its nullspace. How are you calculating eigenvectors? $\endgroup$ – G Tony Jacobs Jul 4 '14 at 14:51
  • $\begingroup$ I believe you said eigenvectors when you meant so say eigenvalues so I have changed it for you. $\endgroup$ – Brad Jul 4 '14 at 15:09
  • 2
    $\begingroup$ I would say that after adjustment be: $\lambda_1=0,\,\,\lambda_2=-\frac{K\alpha+\beta}{K}$ $\endgroup$ – georg Jul 4 '14 at 15:12
  • $\begingroup$ When I calculate these eigenvalues, I get $\lambda_{1,2}=-\frac{(K+1)\alpha}{2K} \pm \frac{\sqrt{(K-1)^2\alpha^2 + 4K\beta^2}}{2K}$. Are you sure of your eigenvalues? $\endgroup$ – G Tony Jacobs Jul 4 '14 at 15:24
  • 1
    $\begingroup$ Are you sure you didn't exchange $\alpha$ and $\beta$ in the second row? $\endgroup$ – egreg Jul 4 '14 at 15:30

Your matrix is of the form $$M=\begin{pmatrix} a & b \\ b/k & a/k \end{pmatrix},$$ now let $x=(x_1,x_2) \neq (0,0)$ be such that $Mx = \lambda x$, then $$\begin{pmatrix} 0 \\ 0 \end{pmatrix} =Mx-\lambda x =\begin{pmatrix} (a-\lambda)x_1+bx_2 \\ \frac{1}{k}(bx_1 +(a-k\lambda)x_2) \end{pmatrix}, $$ This a two unknown linear system of two equations. Solving it shows directly that your eigenvectors must have the form $$\alpha\begin{pmatrix} 1 \\ b/(\lambda-a)\end{pmatrix}, \quad \alpha \in \mathbb{R}.$$ if $\lambda \neq a$ (use the first equation) and $$\alpha\begin{pmatrix} 1 \\ 0\end{pmatrix}, \quad \alpha \in \mathbb{R}.$$ if $\lambda = a$ (from the first equation you get $x_2=0$). Since you aleready computed the eigenvalues, it shouldn't be hard to recover the eigenvectors for your special case.

| cite | improve this answer | |
  • $\begingroup$ Your matrix is not quite the same as posters, regards the positions of $a$ and $b$. But I think the original question should be checked for mistakes, I am not confident in the calculation of the eigenvalues given. $\endgroup$ – Rene Schipperus Jul 4 '14 at 15:11
  • $\begingroup$ @ReneSchipperus It has been edited (thanks I mixed the components). Seems that georg checked the eigenvalues (see comments of the OP). $\endgroup$ – Surb Jul 4 '14 at 15:21
  • $\begingroup$ Yeah I thought something was off, I am deleting my answer which was based on the false computation. Thanks. $\endgroup$ – Rene Schipperus Jul 4 '14 at 15:46

To find the eigenvalues of a matrix $A$, we solve $\mathrm{det}(A-\lambda I)=0$ for $\lambda$. In this case:

$\begin{align} \left|\begin{array}{cc} -\alpha-\lambda & \beta \\ \frac{\beta}{K} & \frac{-\alpha}{K}-\lambda \end{array}\right| &= (-\alpha-\lambda)\left(\frac{-\alpha}{K}-\lambda\right)-\frac{\beta^2}{K} \\ &=\frac{\alpha^2}{K}+\left(\alpha+\frac{\alpha}{K}\right)\lambda + \lambda^2 - \frac{\beta^2}{K} \\ &=\lambda^2 + \left(\alpha+\frac{\alpha}{K}\right)\lambda + \frac{\alpha^2-\beta^2}{K} = 0, \end{align}$

implies that our eigenvalues are:

$\begin{align} \lambda_{1,2} &= -\frac{\alpha+\frac{\alpha}{K}}{2} \pm \frac{\sqrt{\left(\alpha+\frac{\alpha}{K}\right)^2 - 4\frac{\alpha^2-\beta^2}{K}}}{2} \\ &=-\frac{K\alpha+\alpha}{2K} \pm \frac{\sqrt{(K\alpha+\alpha)^2 - 4K(\alpha^2-\beta^2)}}{2K} \\ &=-\frac{(K+1)\alpha}{2K} \pm \frac{\sqrt{\left((K+1)^2-4K\right)\alpha^2 + 4K\beta^2}}{2K} \\ &=-\frac{(K+1)\alpha}{2K} \pm \frac{\sqrt{(K-1)^2\alpha^2 + 4K\beta^2}}{2K}. \end{align}$

Now, an eigenvector corresponding to the eigenvalue $\lambda_i$ is in the nullspace of $\left(\begin{array}{cc} -\alpha-\lambda_i & \beta \\ \frac{\beta}{K} & \frac{-\alpha}{K}-\lambda_i \end{array}\right).$

It appears that the vector $\bf{v}_i=\left(\begin{array}{c} \beta \\ \alpha + \lambda_i \end{array}\right)$ would work.

| cite | improve this answer | |
  • $\begingroup$ How are you getting v? Row operations give $\left( \begin{array}{ccc} 1 & -\frac{\beta}{\alpha+\lambda_i}\\ 0 & \frac{\alpha-K\lambda_i}{\beta} \end{array} \right)$ from which $v_2$ is zero $\endgroup$ – Vaolter Jul 4 '14 at 16:33
  • $\begingroup$ Since $A-\lambda I$ is singular, I'm not bothering with row operations, and just writing down a vector that kills row 1. If you multiply my $v_i$ by my matrix $A-\lambda I$, you get a $0$ on top. $\endgroup$ – G Tony Jacobs Jul 4 '14 at 22:40

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.