0
$\begingroup$

Let $A \in \mathbb{R}^{n \times n}$ be a normalized non-strict column diagonally dominant matrix, that is:

$$a_{j,j} = \sum_{i \ne j} \left|a_{i,j}\right|$$

where $0 \le a_{j,j} \le 1$ and $-1 \le a_{i,j} \le 0$ for all $i \ne j$. Is it possible to find a symmetric, positive definite matrix $S$ such that $\langle A x, x \rangle \le \langle S x, x \rangle$ for all $x \in \mathbb{R}^n?$

Improve this question
$\endgroup$
4
  • 1
    $\begingroup$ Are you sure you wrote what you meant? Just take $S = k I$ for sufficently large $k$. $\endgroup$
    – Robert Israel
    Commented Jul 24, 2016 at 17:48
  • $\begingroup$ @RobertIsrael Thanks, I got the suggestion to take $k=\lambda_{max}(A+A^T)/2$ but do I not need $\left< A \cdot, \cdot \right>$ to be an inner product for that? $\endgroup$
    – Astor
    Commented Jul 24, 2016 at 17:53
  • $\begingroup$ $\langle Ax, x \rangle$ is bounded on the unit sphere $\{x: \langle x, x \rangle = 1\}$. Its maximum there is $k$. $\endgroup$
    – Robert Israel
    Commented Jul 24, 2016 at 18:02
  • $\begingroup$ @RobertIsrael Thanks a lot! I got it! $\endgroup$
    – Astor
    Commented Jul 24, 2016 at 18:17

1 Answer 1

Reset to default
1
$\begingroup$

Take $S = k I $ with $k = \lambda_{max}(\frac{A + A^T}{2})$.

Because

$A = \frac{A + A^T}{2} + \frac{A - A^T}{2}$

But

$\left< \frac{A - A^T}{2}x,x \right> = 0$

Improve this answer
$\endgroup$

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.