Let $T : H \rightarrow H$ be a compact linear operator on a Hilbert space which is NOT necessarily self-adjoint. How would you prove that there exists $v \in H$, $\lvert\lvert v \rvert\rvert = 1$, such that $\langle Tv, Tv \rangle = \lvert\lvert T \rvert \rvert^2$ ?

I don't know how I should take on this problem. Of course, I can use the compacity assumption right at the beginning. But then, I'm lost.

  • $\begingroup$ What‘s the definition of the operator norm? $\endgroup$ – Fakemistake Jan 22 '18 at 16:35
  • $\begingroup$ @Fakemistake supremum of $\lvert\lvert Tv \rvert\rvert$ over elements $v$ of norm less or equal to $1$ $\endgroup$ – Desura Jan 22 '18 at 16:35

Apparently my original answer (below) overlooked an important detail.

With that being said, we can reduce this question to the case of self-adjoint compact operators if we note that $$ \langle Tv,Tv \rangle = \langle v, T^*Tv \rangle = \|\sqrt{T^*T}v\|^2 $$ and that $T^*T$ is self-adjoint and compact.

Hint: Consider the function $v \mapsto \|v\|^2$ as a continuous function restricted to the image of the closed unit ball under $T$.

  • $\begingroup$ I'm sorry, I still cannot see how to do it. $T(\overline{B_1(0)})$ is only relatively compact so (for example) I cannot immediately conclude that the max of the function $v \mapsto \lvert\lvert v \rvert\rvert^2$ is attained. $\endgroup$ – Desura Jan 22 '18 at 17:41
  • $\begingroup$ @Desura ah, that is a problem. I'll delete this answer if I can't think of a fix. $\endgroup$ – Ben Grossmann Jan 22 '18 at 18:06
  • $\begingroup$ @Desura does that answer let you get the rest of the way? If so, I'd like to know how $\endgroup$ – Ben Grossmann Jan 22 '18 at 18:16
  • $\begingroup$ @Omnomnomnom Do you need to take a detour through functional calculus to define the square root? $\endgroup$ – Cameron Williams Jan 22 '18 at 18:19
  • $\begingroup$ @CameronWilliams it suffices to define the square root by power series. Pedersen's Analysis Now uses something like that in his proof of the polar decomposition $\endgroup$ – Ben Grossmann Jan 22 '18 at 18:38

By definition of the operator norm, there is for each $n$ a vector $v_n$ with $\|v_n\|\le 1$ such that $$ \|Tv \|\ge \|T\|-1/n. $$ Hilbert spaces are reflexive, after extracting a subsequence if necessary we have $v_n\rightharpoonup v$ and (by compactness) $Tv_n\to Tv$. By weak lower semicontiuity of norms, $\|v\|\le1$. This shows $\|v\|\le1$ and $\|Tv\|=\|T\|$. If $T\ne0$ this implies $v\ne0$ and scaling argument shows $\|v\|=1$.


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.