6
$\begingroup$

I want to determine some structures of matrices that can be transformed into a symmetric matrices using similarity transformation, i.e.,

$B=T^{-1}AT$

where $T$ is the similarity transformation matrix. Here, $A$ is a non-negative matrix and the diagonal elements of $A$ is supposed to be the same $(diag(A)=[a, a, \ldots , a])$.

Any ideas?

$\endgroup$
3
  • $\begingroup$ The diagonal matrix in question belongs to the center of $GL(n,R)$ and the question itself does not belong to MO. Voting to close. $\endgroup$
    – Misha
    Jun 4, 2013 at 18:21
  • 3
    $\begingroup$ The matrix $A$ is not necessarily a diagonal matrix. The question may not be central, but still is reasonable. $\endgroup$ Jun 4, 2013 at 19:40
  • $\begingroup$ Oh, sorry, I misread the question. $\endgroup$
    – Misha
    Jun 4, 2013 at 21:46

2 Answers 2

3
$\begingroup$

We can say something on such matrices $B$ by characterizing its eigenvalues, which coincide with the eigenvalues of $A$. Since $A$ is a real symmetric matrix, it has real eigenvalues. Hence a necessary condition on $B$ is that it has real eigenvalues. But one can say more. There are necessary and sufficient conditions known for real numbers $\lambda_1,\ldots ,\lambda_r$ to be the eigenvalues of a non-negative symmetric matrix. See for example the paper of M. Fiedler, Eigenvalues of nonnegative symmetric matrices here: http://www.sciencedirect.com/science/article/pii/0024379574900317.

Update: We still have the additional condition that the diagonal of $A$ is $(a,a,\ldots ,a)$, but this is also discussed in the paper of Fiedler. Given $2n$ real numbers $\lambda_i$, one can say whether there is a nonnegative symmetric matrix with eigenvalues $\lambda_1,\ldots ,\lambda_n$ and diagonal $\lambda_{n+1},\ldots ,\lambda_{2n}$.

$\endgroup$
2
$\begingroup$

An addition to Dietrich Burde's answer: Since symmetric matrices are diagonizable, they are semisimple (each invariant subspace has an invariant complement). Thus $B$ has to be semisimple ($\iff$ the minimal polynomial is square free).

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.