8

This is just a small query:

If $A$ is an $n\times n$ square matrix and $p(t)=(t-\lambda_1)(t-\lambda_2)\cdots(t-\lambda_m)$ be a polynomial (with $\lambda_i \in \mathbb{C}$ for all $i=1, \ldots, m$) such that $p(A)=0$, then is it necessary that $\lambda_1,\ldots,\lambda_m$ will be eigenvalues of $A?$

Now I know that if $p(A)=0$ then the minimal polynomial $m_A$ divides $p$; and as $m_A$ is a non-zero polynomial there should be at least one $\lambda_i$ which will be an eigenvector of $A$. But apart from that, I have no idea what to do for this question. The following question is something that I thought I might add with the first one in order to straighten up my understandings in this area.

If not, what extra condition can be imposed on the polynomial and/or on the matrix to make sure that the $\lambda_i's$ are necessarily eigenvalues of $A$?

Thanks and regards.

Sayantan
  • 3,418

1 Answers1

9

No, of course not. $A$ is a root of any polynomial $p$ divisible by $m_A$, and you can always give a polynomial more roots by multiplying it by something. For example, let $A$ be the zero matrix. This has minimal polynomial $t$, and so is a root of $p(t)=t(t-1)(t-2)$ for example, but $1$ and $2$ are not eigenvalues of $A$.

Chris Eagle
  • 33,306