I know that the extreme points of the closed unit ball of $\mathcal{H}^\infty(\mathbb{D})$, the space of all bounded holomorphic functions on the unit disc are the functions $f\in\mathcal{H}^\infty(\mathbb{D})$ such that $\int\limits_{-\pi}^\pi\log(1-|f(e^{i\theta})|)d\theta=-\infty$. What are the extreme points of the closed unit ball of the disc algebra, i.e. all functions continuous on $\overline{\mathbb{D}}$ and holomorphic on $\mathbb{D}$?
It is clear that all the extreme points of the closed unit ball of $\mathcal{H}^\infty(\mathbb{D})$, that are in the disc algebra are extreme points of the closed unit ball of the disc algebra. But are these the only extreme points?
Are there any extreme points of the closed unit ball of $\mathcal{H}^\infty(\mathbb{D})$ that is not in the disc algebra?
Is there any extreme point of the closed unit ball of the disc algebra that is not a finite Blaschke product?
- 21
-
Do you have a reference for the result on $H^ \infty$? – Ruy Dec 04 '20 at 21:21
-
@Ruy: there should be more modern references, but it appears for sure in Hoffman's 1962 Banach Spaces of Analytic Functions. – Martin Argerami Dec 05 '20 at 15:21
-
Thanks @Martin. I liked your answer below! – Ruy Dec 05 '20 at 15:35
-
Thanks. What I wrote is what Hoffman wrote, though he doesn't construct the $w$. The other implication, though, was harder for me to understand. The crucial step depends on subtle (to me, at least), properties of analytic functions on the boundary. Szego's Theorem seems essential. – Martin Argerami Dec 05 '20 at 16:35
1 Answers
$\newcommand\TT{\mathbb T}$ $\newcommand\AA{\mathbb A}$ $\newcommand\abajo{\\[0.3cm]}$ $\newcommand\e{\varepsilon}$ $\newcommand\DD{\mathbb D}$
The same condition characterizes the extreme points. Since $\AA\subset \mathcal H^\infty$, every $f$ satifying the integral condition is an extreme point.
Conversely, suppose that $\log(1-|f(e^{i\theta})|)$ is integrable. The argument is roughly the same as in $H^\infty$, with a twist to make the function $g$ stay in $\AA$. Let $w\in C([0,2\pi])$ with $w(2\pi)=w(0)$, $0\leq w(t)\leq1-|f(e^{it})|$, with $\log w\in L^1([0,2\pi])$, and $e^{iw}$ continuously differentiable on the open arcs in the set $\{|f|\ne1\}\cap\TT$ (details at the end). Define $$ g(z)=\exp\bigg[\frac1{2\pi}\,\int_0^{2\pi}\frac{e^{it}+z}{e^{it}-z}\,\log w(t)\,dt\bigg] $$ By the Schwarz formula, the expression inside the brackets is analytic on $\DD$ and has real part $\log w(t)$ when $z=e^{it}$; then $|g(e^{it})|=w(t)$ is continuous on $\TT$.
Since $|g(e^{it})|\leq 1-|f(e^{it})|$ we have, via the maximum modulus principle, that $\|f\|_\infty+\|g\|_\infty\leq 1$. Then $g\in\AA$, $f\pm g\in \AA$, and $$ f=\tfrac12\,(f+g)+\tfrac12\,(f-g) $$ is not extreme.
How to construct the function $w$: (not claiming that this is a smart way to do it)
The set $\{|f|\ne1\}\cap\TT=|f|^{-1}(-\infty,1)\cap\TT$ is open in $\TT$, so it is a countable union of disjoint arcs. Say (the union might be finite, but that only simplifies the argument) $$ \{|f|\ne1\}\cap\TT=\bigcup_n \exp(i\,(a_n,b_n)), $$ with $0<a_n<b_n<a_{n+1}<b_{n+1}<2\pi$, for all $n$. Let $h_n\in C[0,1]$ be continuously differentiable, with $0\leq h_n\leq1$, $h_n(0)=h_n(1)=0$, and $\int_{h_n<1}|\log t|<2^{-n}$ (this condition is achievable due to $\log t$ being integrable in $[0,1]$). Define $$ w(t)=(1-|f(e^{it})|)\,\sum_n h_n\big(\frac{t-a_n}{b_n-a_n}\big)\,1_{(a_n,b_n)}. $$ The sum is trivially defined because only one term is nonzero at a time; for the same reason, $w$ is continuously differentiable on each $(a_n,b_n)$. Now \begin{equation*} \begin{split} \int_0^{2\pi}|\log w(t)|\,dt &=-\sum_n\int_{a_n}^{b_n}\log h_n\big(\frac{t-a_n}{b_n-a_n}\big)\,dt =-\sum_n\int_{0}^{1}\log h_n\big(t\big)\,dt\abajo &=-\sum_n\int_{h_n\ne1}\log h_n\big(t\big)\,dt\abajo &\leq\sum_n2^{-n}<\infty \end{split} \end{equation*}
- 219,660
-
Can you give me an example of an extreme point that is in the disc algebra but not a finite Blascheke product? – Alexander Dec 05 '20 at 07:38
-
It's not easy to write everything explicitly. But for instance consider a continuous function $g\in C([0,2\pi])$ such that $m({g=1-\tfrac1n})=\tfrac1{2n(n+1)}$. Imagine the function $$\sum_n\tfrac1n,1_{\big[\tfrac1{n+1},\tfrac1n\big]},$$ just joining the end points of the segments to make it continuous; one also needs to tweak the construction so that it is periodic, but that's also easy to get. Then form the Schwarz Integral $$f(z)=\frac1{2\pi}\int_0^{2\pi} \frac{e^{it}+z}{e^{it}-z},g(t),dt$$ you get an $f$ that satisfies the log condition but $|f|\ne1$ often. – Martin Argerami Dec 05 '20 at 16:49
-
I am having some difficulty in showing that the function $f$ lies in the disc algebra. Can you please help? – Alexander Dec 08 '20 at 19:21
-
1What I wrote is not entirely right. One needs to make $g$ the log of function I wrote, and then make $f$ the exponent of the integral. What one is guaranteed is that the real part of the integral converges to $g$ in the boundary. The existence and continuity of the limits in the border is not trivial; Hoffman dedicates several pages to it. – Martin Argerami Dec 08 '20 at 20:06
-
1