6
$\begingroup$

So, as a part of a problem, I've been asked to prove that if $H$ is a complex Hilbert space and $T\in L(H)$ is normal, then $\| T^2 \| =\| T \| ^2$ (Operator norm)

Context: This is part (b) in a three part problem obviously designed to build to a final result. The final result being this is good for any integer $n$, and the first part (already proved) was that $$T \text { normal } \iff \forall x\in H,\| T(x)\|=\|T^*(x)\|$$

So, here's my attempt at a proof of part b:
Since the operator norm is submultiplicative, we have $\|T^2\|\le \| T\| ^2$. Then from a theorem in Kreyszig's Introductory Functional Analysis with Applications (Thm 3.94(e), page 198) we have $$\|TT^*\|=\|T\|^2$$, and using submultiplicativity again, combined with them 3.92 (p 196) that states $\|T^*\|=\|T\|$, we have $$\|T\|^2\le \|T\|\cdot \|T^*\|=\|T\|^2$$ hence $\|T^2\|=\|T\|^2$ as desired.

Now, the problem is...none of theorems I have used require $T$ to be normal (Or the space to be complex). Nor did I use part a. So....either my proof is wrong, or this statement actually holds for all bounded linear operators on any Hilbert space...which I doubt, so I think I made a mistake somewhere. Any advice?

$\endgroup$
  • $\begingroup$ You didn't make any mistakes. The equalities $\|TT^*\| = \|T\|^2$ and $\|T\| = \|T^*\|$ hold for all bounded linear operators $T$ on a hilbert space. $\endgroup$ – Falko Mar 18 '15 at 23:48
  • $\begingroup$ @user49797 So the statement holds for any operator, not just normal ones, in any hilbert space (real or complex)? $\endgroup$ – Alan Mar 18 '15 at 23:50
  • $\begingroup$ Oh my bad, you did make a mistake. Your last inequality doesn't prove $\|T\|^2 \leq \|T^2\|$. $\endgroup$ – Falko Mar 18 '15 at 23:58
  • $\begingroup$ Woops. Yeah, sleep dep. Thanks. $\endgroup$ – Alan Mar 19 '15 at 0:05
  • $\begingroup$ Friend just walked in to the office and showed me the proof, I'm good. $\endgroup$ – Alan Mar 19 '15 at 0:07
5
$\begingroup$

So, just to formally complete this thing, the proper proof: Taking the square of the norms, we have

$$\|T^2x\|^2=\langle T^2x,T^2x\rangle=\langle T^*T^2x,Tx\rangle=\langle TT^*Tx,Tx\rangle=\langle T^*Tx,T^*Tx\rangle=\|T^*Tx\|^2$$ since this holds for all $x\in H$, we have $\|T^2\|=\|T^*T\|=\|T\|^2$ as desired

(And yeah, normalcy being used!)

For those playing along with our home game, to prove this extends to any arbitray positive power $n\in \mathbb N$, use forward/backward induction: First it proves for arbitrary powers of 2: $\forall k\in \mathbb N$ it holds for $2^k$ by induction on $k$. Then use backwards induction on $n$: if it holds for $n$, it holds for $n-1$. (this is only the second time I've seen forward/backwards induction used, the other time being in the proof of the arithmetic/geometric mean inequality....does anyone know of other uses?)

$\endgroup$

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.