I have a final exam tomorrow, am sure a 3x3 eigen value problem like the one below is there. But I find it very hard to find eigen values without zeros in the matrix

Show me how you do it quickly so that I can apply it tomorrow; thanks

enter image description here

  • $\begingroup$ There isn't a nice algorithm for doing it as far as I'm aware other than the usual way of computing determinants. $\endgroup$ – Cameron Williams Aug 9 '14 at 3:46
  • 3
    $\begingroup$ The numbers in your matrix are kind of large for hand calculation, so why do you think it is a good representative example of what could be on the exam? Is it taken from a real exam review or is it from elsewhere? Have you heard of the characteristic polynomial of a square matrix? $\endgroup$ – KCd Aug 9 '14 at 3:46
  • $\begingroup$ Use the condition: $\det(\lambda I-A)=0$. Good luck! $\endgroup$ – idiot Aug 9 '14 at 3:48
  • 1
    $\begingroup$ Just use the linearity of the determinant to reduce the polynomial to get something more easy to handle. $\endgroup$ – Jorge Ramos Aug 9 '14 at 3:51

You could (specially if you have a calculator at hand to crunch numbers) do this:

  1. Get the trace (or sum of the diagonal elements). It is $32-25-2=5$
  2. Compute the determinant. It is $-8$
  3. Compute the square of the matrix and get the trace of it. It is $21$

You now know that:

$$\lambda_{1} \lambda_{2} \lambda_{3} = -8 \space\space\space\space\space(1)$$ $$\lambda_{1} + \lambda_{2} + \lambda_{3} = 5 \space\space\space\space\space(2)$$ $$\lambda_{1}^{2} + \lambda_{2}^{2} + \lambda_{3}^{2} = 21 \space\space\space\space\space(3)$$

Where $\lambda_{1}, \lambda_{2}, \lambda_{3}$ are the eigenvalues to work out. Now, let's say you feel lucky and want to assume that all the eigenvalues are integer. Then, from equation $(3)$ you know the largest one could be only $3$ or $4$ in absolute value, in which case the second largest would have to be $\pm 2$ and then $\pm 1$. If the largest were $3$, the second would have to be $3$ and the third couldn't be an integer. Now, from $(1)$ you know either one is negative or all are. Now, at once you see from $(2)$ that $4$ got to be positive, and the last choice of whether $1$ or $2$ is negative is solved by mentally trying, to find that the answer is indeed $-1$, $2$, $4$.

Yes, we assumed the answer to be made of integers, but, from experience, this is a good assumption to make to give it a first go in exams questions, as it's usually the case!


I would not exactly call this quick, but a few tricks could simplify the steps:

  1. We have $\det(A)=\det(A^T)$
  2. If you add/subtract multiples of one row to another the determinant remains unchanged
  3. (1)+(2) means that you can add/subtract multiples of one column to another "for free" too
  4. You may want to modify the signs of a row or column which will change the sign of the determinant, but since we are trying to determine when $\det(A-\lambda I)=0$ we do not mind whether the sign has changed or not
  5. We do not need it for this example, but you could also easily scale a row or column by any factor $r\neq 0$ since $r\cdot\det(A-\lambda I)=0\iff\det(A-\lambda I)=0$. In fact (4) follows directly from this using $r=\pm 1$

so with all this in place we have $$ \begin{align} \begin{vmatrix}32-\lambda&-12&-2\\66&-25-\lambda&-4\\54&-21&-2-\lambda\end{vmatrix}&=\begin{vmatrix}32-\lambda&12&2\\66&25+\lambda&4\\54&21&2+\lambda\end{vmatrix}\\ \quad\\ &=\begin{vmatrix}32-\lambda&0&2\\66&1+\lambda&4\\54&9-6\lambda&2+\lambda\end{vmatrix} \end{align} $$ where the first equality was produced changing signs of the last two columns and the last equality is due to having subtracted $6$ times the last column from the middle column. Thus we got rid of annoying negative signs and are down to having to compute only two minors: $$ \begin{vmatrix}1+\lambda&4\\9-6\lambda&2+\lambda\end{vmatrix}=\lambda^2+27\lambda-34 $$ and $$ \begin{vmatrix}66&1+\lambda\\54&9-6\lambda\end{vmatrix}=540-450 \lambda $$ to finally arrive at $$ \begin{align} \det(A-\lambda I)&=(32-\lambda)(\lambda^2+27\lambda-34)+2(540-450 \lambda)\\ &=-\lambda^3+5\lambda^2-2\lambda-8 \end{align} $$ This is still tedious work! BTW the zeros for the above expression are $\lambda=-1,2,4$.

  • $\begingroup$ @GitGud: Hi! Sorry I just got the impression that noone seemed to care for my answer and waited for something more relevant to come along. I will undelete it again, no problem! Basically deleting answers is not such a good idea unless they are entirely irrelevant, after all. $\endgroup$ – String Aug 11 '14 at 1:28
  • $\begingroup$ I understand that. I often do the same. You're free to keep it deleted if you want. I can read it anyway. I'll do it later. $\endgroup$ – Git Gud Aug 11 '14 at 1:30

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.