Exercise :

Let $H$ be a Hilbert space and $A \in \mathcal{L}_c(H)$. Show that $\exists x \in H : \|A(x)\| = \|A\|_\mathcal{L}$.

Attempt - Thoughts :

Note : The space $\mathcal{L}_c(H)$ is the space of all the Linear Compact Operators $L : H \to H$.

Let $A \in \mathcal{L}_c(H)$ such that :

$$\langle A(y),x\rangle = \langle y, A(x) \rangle, \quad \text{for} \quad x,y \in H$$

Then : \begin{align} \|A\|_\mathcal{L} &= \sup\left\{ \|A(y)\| : \|y\| \leq 1 \right\} \\ &= \sup\left\{\sup_{\|x\| \leq 1}|\langle A(y),x\rangle| : \|y\| \leq 1 \right\} \\ &= \sup\left\{\sup_{\|x\| \leq 1} |\langle y, A(x)\rangle| : \|y\| \leq 1 \right\} \\ &= \sup_x \|A(x)\| \end{align}

So for that $x$ that the supremum is achieved, it indeed is $\|A(x)\| = \|A\|_\mathcal{L}$.

But isn't that more like showing that $\exists A \in \mathcal{L}_c(X,Y)$ such that $\|A(x)\| = \|A\|_\mathcal{L}$ ? I got the intuition above from the way we defined the Dual-Operator $A^* \in \mathcal{L}(X,Y)$ over two Banach spaces.

If my approach is not legit, how would one show that $\exists x \in H : \|A(x)\| = \|A\|_\mathcal{L}$ ?

Any elaboration or hint will be greatly appreciated.

  • $\begingroup$ I'm guessing you want $\|x\| \le 1$, not just $x \in H$? Otherwise the result is trivial. $\endgroup$ – Theo Bendit Feb 24 at 10:52
  • $\begingroup$ @TheoBendit Hi, thanks for input. Well, the exercise says $x \in H$. $\endgroup$ – Rebellos Feb 24 at 10:52
  • 1
    $\begingroup$ In that case, two options: $\|Ax\| = 0$ for all $x$, in which case $\|A\| = 0$, so pick any $x$, or $\|Ax\| > 0$, for some $x$, in which case choose $y = \frac{\|A\|}{\|Ax\|} x$. Then $\|Ay\| = \|A\|$ as required. $\endgroup$ – Theo Bendit Feb 24 at 10:54
  • $\begingroup$ @TheoBendit Way simpler than what I tried. By the way, is my approach even legitimate? $\endgroup$ – Rebellos Feb 24 at 10:57
  • $\begingroup$ @user123 Well, the final part is $\sup_y \sup_x |\langle y, A(x)\rangle| = \sup_x \|A(x)\|_{y \in H}$ but $x \in H$ too so it is just $\sup_x \|A(x)\|$ and we are talking about a compact operator. This means that it transfers bounded sets from $H$ to sets with compact closure on $H$. Yes, I am sure that it is not $\| x\| \leq 1$. $\endgroup$ – Rebellos Feb 24 at 11:13

It seems that you assume that $A$ is self-adjoint, which is not necessarily the case.

As mentioned in the comments, it seems that we want $x$ to have a norm smaller or equal to one. Otherwise, we play with $\lambda y$, $\lambda \gt 0$ where $y$ is such that $Ay\neq 0$ (if it exists) and we are not, without needing the fact that we work on a Hilbert space or with a compact operator.

In order to see that the operator norm is reached by an element of the closed unit ball, we consider a sequence $\left(x_n\right)_{n\geqslant 1}$ of elements of the unit ball such that $\left\lVert Ax_n\right\rVert\geqslant \left\lVert A\right\rVert-n^{-1}$. We extract a weakly convergence subsequence to some $x$; by compactness, we extract from this subsequence a further subsequence, denotes $\left(x_{n_k}\right)_{k\geqslant 1}$, such that $\left(Ax_{n_k}\right)_{k\geqslant 1}$ is convergent to some $y$. Using adjoint, we can see that $\langle Ax_{n_k},z\rangle \to \langle Ax,z\rangle$ for all $z\in H$ hence $y=Ax$. From $$ \left\lVert Ax_{n_k}\right\rVert\geqslant \left\lVert A\right\rVert-n_k^{-1} $$ and the convergence of $\left(Ax_{n_k}\right)_{k\geqslant 1}$, we derive that $\left\lVert Ax \right\rVert=\left\lVert A\right\rVert$.


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.