13
$\begingroup$

A Hermitian diagonally dominant matrix $A$ with real non-negative diagonal entries is positive semidefinite.

Is it possible to have a Hermitian matrix be positive semidefinite/definite and not be diagonally dominant?

In other words, if I know that a matrix $M$ is symmetric positive definite then can I ensure $M - dI$, for a real number $d$, is positive definite only by ensuring $M - dI$ is diagonally dominant with non-negative diagonal entries?

I am aware that $d \le \lambda_{min}$, $\lambda$ being eigenvalue, for the matrix to remain semidefinite, but I need to avoid eigenvalue computation.

Thanks!

$\endgroup$
5
  • 14
    $\begingroup$ The matrix $\left(\begin{matrix} 1&-2 \\ -2&4\end{matrix}\right)$ is symmetric and positive semidefinite, but not diagonally dominant. You can change the "positive semidefinite" into "positive definite" by changing the $-2$'s to $-3/2$'s. Does this answer your question? (I am not totally sure what you are asking.) $\endgroup$ Sep 30 '15 at 22:54
  • 4
    $\begingroup$ Note: The converse direction holds. A symmetric matrix which is diagonally dominant is positive definite. I guess the same holds with "Hermitian" (no guarantees). $\endgroup$ Sep 30 '15 at 22:56
  • $\begingroup$ Thanks that answers my question $\endgroup$
    – muon
    Oct 1 '15 at 13:29
  • 1
    $\begingroup$ @darijgrinberg you wrote: 'A symmetric matrix which is diagonally dominant is positive definite.' This is only true if the values in the diagonal are positive. Diagonal dominance looks at the magnitude. For example $\begin{bmatrix} -3 & 1\\ 1 & 2 \end{bmatrix}$ is symmetric diagonally dominant but not positive definite. $\endgroup$
    – Hjan
    Dec 10 '18 at 15:53
  • 2
    $\begingroup$ @Hjan: thanks for catching this. I was using a nonstandard definition of diagonal dominance, in which the diagonal entries were required to be positive. $\endgroup$ Dec 10 '18 at 18:11
4
$\begingroup$

This was answered in the comments.

The matrix $\left(\begin{matrix} 1&-2 \\ -2&4\end{matrix}\right)$ is symmetric and positive semidefinite, but not diagonally dominant. You can change the "positive semidefinite" into "positive definite" by changing the $−2$'s to $−3/2$'s. Does this answer your question? (I am not totally sure what you are asking.) – darij grinberg Sep 30 '15 at 22:54

$\endgroup$

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.