5
$\begingroup$

I have seen the following inequality here but I don't know where I can find a proof for it. Could somebody give me a hint to understand it or guide me to a reference please?

$\|AB\|_{HS} \leq \|A\|_{\mathrm{op}} \|B\|_{HS} \ {\rm for}\ A, B ∈ \mathcal{M}_n$.

$\endgroup$
4
$\begingroup$

The key are these properties about positivity of operators: if $0\leq C\leq D$, then $B^*CB\leq B^*DB$; and $C\leq \|C\|_{op}\,I$. You have then, using the positivity of the trace, $$ \|AB\|_{HS}^2=\text{Tr}(B^*A^*AB)\leq \|A^*A\|_{op}\,\text{Tr}(B^*B)=\|A\|^2_{op}\,\|B\|_{HS}^2. $$

$\endgroup$
  • $\begingroup$ Thank you very much. But I find hard to understand what you wrote? - What does positivity mean? - Are you talking about linear operators? $\endgroup$ – Cupitor Nov 6 '14 at 16:53
  • 1
    $\begingroup$ Yes, of course I'm talking about linear operators. That's where the "op" in the operator norm comes from :) If you want to think matrices, what I call "positive" should be "positive semidefinite". And the trace is "positive" in the sense that the trace of a positive semidefinite matrix is non-negative. $\endgroup$ – Martin Argerami Nov 6 '14 at 18:52
  • $\begingroup$ Thank you very much for your kind reply. Glancing over @elexhobby's answer I better understand the whole situation :) $\endgroup$ – Cupitor Nov 7 '14 at 10:27
3
$\begingroup$

Like the OP (see comment to Martin's answer), I had to convince myself of the correctness of Martin's excellent solution, and I worked it out. I'll just copy it here, in case it helps anybody else.

First, an operator (matrix) $A$ is semipositive (positive semi-definite) if $x^T A X \geq 0 \,\forall x$.

$C \leq D$ just means $D-C$ is positive semi-definite.

The first property to note is that if $C \leq D$, then $B^T C B \leq B^T D B$. This is easy. Consider $x^T (B^T D B - B^T C B) x = (Bx)^T (D-C) (Bx) \geq 0$ because $C \leq D$.

Secondly, if $C \leq D$, $Tr(C) \leq Tr(D)$. This follows because the trace is the sum of eigenvalues, hence $Tr(D-C)\geq 0$, and the linearity of trace.

Finally, the third property to note is that for any matrix $A$, we have $A \leq \|A\|_{op}I$. Why? \begin{align} x^T A x &\leq \|x\|\|Ax\| \quad\text{(Cauchy Schwarz)} \\ &= x^T x \frac{\|Ax\|}{\|x\|} \\ &\leq (x^T x) \|A\|_{op} \end{align}

So for any matrix $A$ we have $$A \leq \|A\|_{op}I \Rightarrow B^T A B \leq B^T \|A\|_{op} A \Rightarrow Tr(B^T A B) \leq Tr(B^T \|A\|_{op} A)$$ Finally, \begin{align*} \|AB\|_{HS}^2 &= Tr((AB)^T (AB)) \\ &= Tr(B^T A^T A B) \\ &\leq Tr(B^T (\|A^TA\|_{op} I) B) \\ &= \|A^T A\|_{op} Tr(B^T B) \\ &= \|A\|_{op}^2 \|B\|_{HS}^2 \end{align*}

$\endgroup$

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.