0
$\begingroup$

I came across a claim that given $A\in\mathbb{R}^{n\times n}$ the following equality holds $$\left\Vert A\right\Vert _{2}=\sup\left\{ \text{trace}\left(ZA\right)\ :\ Z\succeq0\right\}$$ Where $\left\Vert A\right\Vert _{2}=\sup_{\left\Vert x\right\Vert _{2}=1}\left\Vert Ax\right\Vert _{2}$ w.r.t the Euclidian norm on $\mathbb{R}^{n}$. I can't figure out whether this is actually correct, if it is I would be happy to see a proof.

1 Answers 1

2

Do you have some restrictions on $Z$? otherwise if we take $Z$ to be the identity matrix, then your proposition would say $||A||_2 \geq Trace(A)$ which is untrue.

  • 0
    Yeap, that easily shows the claim is untrue. I think the person that wrote the claim thought the induced 2-norm is self-dual like the vector 2-norm.2017-01-18