Consider the following block matrix $$ X = \begin{bmatrix} A & C \\ C^\top & B\end{bmatrix}, $$ where $A\in\mathbb{R}^{n\times n}$, $B\in\mathbb{R}^{m\times m}$, and $C\in\mathbb{R}^{n\times m}$.

To Prove (or disprove): If $X$ is positive definite, i.e. $X>0$, then the following trace inequality holds $$ \left[\mathrm{tr}(CC^\top)\right]^2< \mathrm{tr}(A^2)\mathrm{tr}(B^2). $$

Some comments. Based on Theorem 2.3 of Horn and Mathias. "Cauchy-Schwarz inequalities associated with positive semidefinite matrices." Linear Algebra and Its Applications 142 (1990): 63-82, I think it is possible to prove the previous fact if we replace strict inequalities with non-strict ones.

  • 3
    $\begingroup$ By $X>0$, I guess you mean positive definite, not positive entries. $\endgroup$ – T. Amdeberhan Sep 18 '16 at 15:50
  • $\begingroup$ Yes, I just edited the question in order to specify it. $\endgroup$ – Ludwig Sep 18 '16 at 15:55

For any unitarily invariant norm it can be shown that \begin{equation*} \|X\| = \left\Vert \begin{bmatrix} A & C\\ C^* & B \end{bmatrix} \right\Vert \le \|A\| + \|B\|. \end{equation*} Thus, using the squared Frobenius norm on both sides and cancelling, we obtain \begin{equation*} \|C\|_F^2 \le \|A\|_F\|B\|_F, \end{equation*} as desired.

  • 2
    $\begingroup$ But, the desired inequality is supposed to be sharp $<$ and not $\leq$. Am I right? $\endgroup$ – T. Amdeberhan Sep 18 '16 at 16:46
  • 1
    $\begingroup$ You are right, the OP asks for a strict inequality; I believe that it should be possible to make it strict without much trouble. $\endgroup$ – Suvrit Sep 18 '16 at 17:08
  • $\begingroup$ Yes, indeed, I'm wondering whether this still holds true with strict inequality. $\endgroup$ – Ludwig Sep 18 '16 at 17:18
  • 1
    $\begingroup$ Alternatively, we know that $X > 0$ iff there exists a contractive matrix $K$ (i.e., $\|K\| < 1$) such that $C=A^{1/2}KB^{1/2}$. Thus, we see that \begin{equation*} \|C\|_F^2 = \mathrm{tr}(KAKB) < \text{rhs}, \end{equation*} because $K$ is a strict contraction. $\endgroup$ – Suvrit Sep 18 '16 at 17:22
  • 4
    $\begingroup$ To get the strict inequality it is enough to apply the above solution to $X-\epsilon I$, where $\epsilon$ is small enough so as to make $X-\epsilon I$ positive. $\endgroup$ – Ruy Sep 18 '16 at 18:01

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.