0
$\begingroup$

Consider the matrix $$B=\begin{pmatrix} 1 & i \\ -i & 0 \end{pmatrix}$$

Compute the eigenvectors and eigenvalues of this matrix viewed as a map from $\mathbb{C^2}$ to $\mathbb{C^2}.$ Determine whether it can be diagonalised.

I understand that in order to find the eigenvalues I must compute det$(B-\lambda I)$ and I have done this, giving me $\lambda=\frac{1+\sqrt{5}}{2} \text{ and } \lambda=\frac{1-\sqrt{5}}{2}.$ I also recognise that there are two distinct eigenvalues and so the matrix is diagonalisable.

How do I find the eigenvectors though?

So far I have done the following;

$$\begin{pmatrix} 1 & i \\ -i & 0 \end{pmatrix}-\begin{pmatrix} \frac{1-\sqrt{5}}{2} & 0 \\ 0 & \frac{1-\sqrt{5}}{2} \end{pmatrix}=\begin{pmatrix} \frac{1+\sqrt{5}}{2} & i \\ -i & \frac{-1+\sqrt{5}}{2} \end{pmatrix}$$

Then set $$\begin{pmatrix} \frac{1+\sqrt{5}}{2} & i \\ -i & \frac{-1+\sqrt{5}}{2} \end{pmatrix} \begin{pmatrix} v_1 \\ v_2 \end{pmatrix}=0$$ which gives $$v_1=v_2\bigg(\frac{-1-\sqrt{5}}{-2i}\bigg).$$ I have come to a bit of a sticking point here though.

$\endgroup$
  • $\begingroup$ The standard way is to find the nullspace of $B-\lambda I$, for each choice of eigenvalue. $\endgroup$ – vadim123 Mar 2 '16 at 15:59
  • $\begingroup$ Could you show how you would do this please? I have attempted that but I can't seem to find the solution. $\endgroup$ – Si.0788 Mar 2 '16 at 16:11
  • $\begingroup$ There are many examples of this process, on this website and elsewhere. If you share what you tried we can help diagnose where you went wrong. $\endgroup$ – vadim123 Mar 2 '16 at 16:21
  • $\begingroup$ @Si.0788 You are almost there: The relation between $v_1$ and $v_2$ gives the eigenvector associated with eigenvalue $\lambda_1=(1-\sqrt{5})/2$ (Note that if $v\in\mathbb{C}^2$ is an eigenvector of $B$, then so is $cv$, where $c\in\mathbb{C}$). Repeat the same process for $\lambda_2=(1+\sqrt{5})/2$. $\endgroup$ – Katie Imach Mar 2 '16 at 16:38
  • $\begingroup$ A tip: for any square matrix, the sum of the eigenvalues weighted by algebraic multiplicity equals the trace. Your eigenvalues add up to 1 but they should add up to 2, so your eigenvalues couldn't possibly be right. Also, the eigenvalues multiply to give the determinant; in this case you can see that the rows differ only by a factor of $-i$, so the determinant is $0$, which tells you both eigenvalues immediately. $\endgroup$ – Ian Mar 2 '16 at 17:57
1
$\begingroup$

For 2x2 matrices, the Cayley-Hamilton theorem is the easiest way to find eigenvectors. For 3x3, it is still very easy, and numerically more stable than solving for the null space.

The Cayley-Hamilton theorem says that if $P(x)$ is the characteristic polynomial of $A$, then $P(A) = 0$. Now if the eigenvalues of $A$ are $\{\lambda_i\}$ (with repetition for multiplicity), then the characteristic polynomial evaluated for $A$ is $$P(A) = (A - \lambda_1I)(A - \lambda_2I) \ldots (A - \lambda_nI) = 0$$ Therefore, the column vectors of $(A - \lambda_2I)(A - \lambda_3I) \ldots (A - \lambda_nI)$ lie in the null space of $A - \lambda_1I$. In particular, the non-zero column vectors are eigenvectors associated with $\lambda_1$.

In your case, letting $\varphi = \frac{1+\sqrt{5}}2$, and noting that the other eigenvalue is $1 - \varphi$, then $B - \varphi I$ contains eigenvectors of $1 - \varphi$ and $B - (1-\varphi)I$ contains eigenvectors of $\varphi$.

$\endgroup$
  • $\begingroup$ Sorry about that, I made mistake when I wrote the matrix, I've edited it now though. Is my answer so far now correct for the new matrix? $\endgroup$ – Si.0788 Mar 2 '16 at 18:04
  • $\begingroup$ @Si.0788 - No. You only corrected the top $B$. Your calculations have the same error in them. $\endgroup$ – Paul Sinclair Mar 2 '16 at 18:11

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.