4
$\begingroup$

It is known that the space $H^{\infty}$ of bounded holomorphic functions on the unit disk $D$ is non-separable with respect to the supremum norm $\|\cdot\|_{\infty}^{D}$. Let $E\subset D$ be connected and not a singleton. Then, $\|\cdot\|_{\infty}^{E}$ (the supremum norm over $E$) is in fact a norm.

What are geometric conditions on $E$ equivalent to $H^{\infty}$ being non-separable with respect to $\|\cdot\|_{\infty}^{E}$?

If $\overline{E}\subset D$, then polynomials are dense with respect to $\|\cdot\|_{\infty}^{E}$, and so in this case $H^{\infty}$ is separable with respect to $\|\cdot\|_{\infty}^{E}$. Hence, we need $\overline{E}$ to intersect the unit circle $C$.

$\endgroup$
6
  • 2
    $\begingroup$ Carleson's interpolation theorem gives a sufficient geometric condition. $\endgroup$ Jul 27, 2020 at 17:50
  • $\begingroup$ The proof I know to show the non separability of $H^\infty$ uses the functions $g_\xi (z)=\exp \left ((z+\xi)/(z-\xi) \right )$, $|\xi|=1$ . These are in $H^\infty$ and their mutual distance is bigger than 1; however this last is seen by taking radial (or non tangential limits). So we could consider the set $\Gamma$ of all points in $\partial D$ which are non tangential limits of points in $E$. If this set is uncountable, then $H^\infty$ is not separable under the norm $\|\cdot \|_\infty^E$. $\endgroup$ Jul 27, 2020 at 18:27
  • $\begingroup$ @GiorgioMetafune I thought the same thing, but i don't have a good sense of the non-tangential limits. Also, perhaps this condition is too strong, it's not clear to me if $[0,1)$ works for my purpose.. $\endgroup$
    – erz
    Jul 27, 2020 at 18:33
  • $\begingroup$ You are right, the case of the interval is the first to be understood. $\endgroup$ Jul 27, 2020 at 19:01
  • $\begingroup$ @JochenWengenroth thank you for suggestion! It was enough to completely answer my question $\endgroup$
    – erz
    Jul 28, 2020 at 8:29

1 Answer 1

6
$\begingroup$

Jochen Wengenroth suggested to look at Carleson's interpolation theorem, and it seems like it completely answers my question. Namely, the following is true.

Let $E$ be a subset of $D$. Then $H^\infty$ is non-separable with respect to $\|\cdot\|_\infty^E$ if and only if $\overline{E}$ intersects the unit circle.

Necessity follows from the fact that if $\overline{E}\subset D$, the polynomials are dense in $H^\infty$ with respect to $\|\cdot\|_\infty^E$, and so $H^\infty$ is separable in this case.

To show sufficiency let us recall the following corollary from the Carleson's interpolation theorem (combine theorems 9.1 and 9.2 from the book Duren - Theory of Hp Spaces): Let $\{z_n\}_n$ be a sequence in $D$ such that there is $0<c<1$ such that $1-|z_{n+1}|\le c(1-|z_{n}|)$, for every $n$. Then $\{z_n\}_n$ is an interpolation sequence, i.e. the operator $T:H^\infty\to l^\infty$ defined by $Tf=\{f(z_n)\}_n$ is a surjection.

Now fix some $z_1\in E$ and $0<c<1$. If $z_1,...,z_n$ are already chosen then $F=\{z\in D, ~|z|\ge 1-c(1-|z_n|)\}$ intersects with $E$ as $\overline{E}$ intersects the unit circle. Choose $z_{n+1}\in F\cap E$.

The sequence $B=\{z_n\}_n$ constructed this way satisfies the condition above, and so the operator $T$ is surjective. Moreover, $\|\cdot\|_\infty^B\le\|\cdot\|_\infty^E$, and so $T$ is a continuous map from $(H^\infty, \|\cdot\|_\infty^E)$ onto a non-separable Banach space. Thus, $(H^\infty, \|\cdot\|_\infty^E)$ is non-separable.

Remark. Using this fact one can now show that $H^\infty(U)$ is non-separable for every bounded open connected set $U$. Indeed, $(H^\infty(V),\|\cdot\|_\infty^U)$ embeds isometrically into $H^\infty(U)$, where $V$ is the complement to the unbounded component of $\mathbb{C}\backslash U$, and the former is non-countable as we have shown.

$\endgroup$
4
  • 1
    $\begingroup$ I missed the connectedness assumption in your question. The Carleson interpolation theorem gives a sufficient condition also for disconnected sets. But then necessity seems to be difficult. $\endgroup$ Jul 28, 2020 at 9:33
  • $\begingroup$ I do not see the connectness assumption, too. If $E \subset D$ and its closure meets the boundary, you can always find a sequence $(z_n) \in E$ as above. Also, if $\bar E \subset D$, the density of polynomials follows without assuming connectness. $\endgroup$ Jul 28, 2020 at 10:37
  • $\begingroup$ @JochenWengenroth you are right. Thank you! $\endgroup$
    – erz
    Jul 28, 2020 at 21:13
  • 1
    $\begingroup$ @GiorgioMetafune you are also right! This is what happens when one has an extra nice condition - gets blinded somehow $\endgroup$
    – erz
    Jul 28, 2020 at 21:14

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.