3
$\begingroup$

Let $f$ be a non-invertible bounded outer function on the unit disk. Does $f$ has radial limit $0$ somewhere? Note that such a property holds for singular inner functions.

$\endgroup$
7
  • $\begingroup$ Look in Wiki and put $\varphi(e^{i\theta}):=|\theta| $ there. $\endgroup$
    – user64494
    Mar 26, 2021 at 14:20
  • $\begingroup$ Also see Hayman W. K. Meromorphic functions. Claredon Prees: Oxford. 1964, Par 6.7. $\endgroup$
    – user64494
    Mar 26, 2021 at 14:37
  • 1
    $\begingroup$ I don't think that this gives an answer. The radial limits exist almost everywhere. This information does not help though to get a positive answer to my question $\endgroup$
    – ray
    Mar 26, 2021 at 14:56
  • 1
    $\begingroup$ a zero-free bounded analytic function $f$ is called invertible if $1/f $ is bounded, too. "Such a property" refers to "having radial limit $0$ somewhere". $\endgroup$
    – ray
    Mar 26, 2021 at 16:28
  • 1
    $\begingroup$ Note that $f(z)=\exp(-(1+z)/(1-z))$ is a singular inner function. There is no contradiction although $f(r)\to 0$. $\endgroup$
    – ray
    Mar 26, 2021 at 16:36

2 Answers 2

1
$\begingroup$

The answer is negative: $f$ may have a non-zero radial limit everywhere.


Part 1. Let us start with definitions and notation. A holomorphic function $f$ in the unit disk $D$ is an outer function if and only if $$ -\log |f(r e^{i t})| = \int_0^{2\pi} \phi(s) K_r(t - s) ds $$ for some integrable function $\phi$ on $[0, 2\pi)$, where $K_r(t)$ is the Poisson kernel: $$ K_r(t) = \frac{1}{2 \pi} \frac{1 - r^2}{1 + r^2 - 2 r \cos t} \, . $$ Furthermore, every integrable $\phi : [0, 2\pi) \to \mathbb R$ corresponds in the above sense to an outer function (a unique one up to multiplication by a constant with modulus $1$). For convenience, let us extend $\phi$ to all of $\mathbb R$ in such a way that $\phi$ is $2\pi$-periodic.

If $\phi$ is continuous at $t$ (or, more generally, if $t$ is a Lebesgue point of $\phi$) and $f$ corresponds to $\phi$, then $$ \lim_{r \to 1^-} (-\log |f(r e^{it})|) = \phi(t) . $$ Due to symmetry of the Poisson kernel, we also have $$ \lim_{r \to 1^-} (-\log |f(r e^{it})|) = \frac{\phi(t^+) + \phi(t^-)}{2} $$ whenever $\phi$ has one-sided limits $\phi(t^+)$ and $\phi(t^-)$ at $t$.

(To clarify the notation: $e^{-\phi}$ is the radial boundary limit of $|f|$ almost everywhere. Thus, $\phi \geqslant 0$ a.e. if and only if $|f|$ is bounded by $1$; more generally, $\phi \geqslant -\log C$ if and only if $|f|$ is bounded by $C$. For $p > 1$, $f$ is in the complex Hardy space $H^p$ if and only if $e^{-\phi}$ is in $L^p$.)


Part 2. We are now ready to construct a counter-example. Let $$ \phi(t) = \begin{cases} 2^n & \text{if } t \in [\tfrac{1}{2^n}, \tfrac{1}{2^n} + \tfrac{1}{8^n}] \text{ for some } n = 1, 2, \ldots , \\ 0 & \text{otherwise.} \end{cases} $$ Let $f$ be an outer function corresponding to $f$. We claim that $f$ is a counter-example: its radial boundary limit is everywhere non-zero, and $1/f$ is not a bounded outer function.

Indeed, the limit of $-\log |f(r e^{it})|$ is equal to a finite number $\tfrac12(\phi(t^+) + \phi(t^-))$ whenever $t \in (0, 2 \pi)$, and so the radial boundary limit of $|f|$ at $e^{i t}$ is strictly positive (the radial limit of $f$ may fail to exist, though; see Part 3 below for more details). Thus, it remains to see what happens for $t = 0$. Clearly, $$ -\log |f(r)| = \int_0^{2\pi} \phi(s) K_r(-s) ds = \int_0^{2\pi} \phi(s) K_r(s) ds . $$ Note that if $r \geqslant \tfrac14$ and $s \in (0, \pi)$, we have $$ \begin{aligned} 0 \leqslant K_r(s) & = \frac{1}{2\pi} \, \frac{(1 + r) (1 - r)}{(1 - r)^2 + 4 r \sin^2 (s/2)} \\ & \leqslant \frac{(1 + r) (1 - r)}{2 \times (1 - r) \times 2 \sqrt{r} \sin(s/2)} \\ & = \frac{1 + r}{4 \sqrt{r} \sin(s/2)} \leqslant \frac{1}{\sin(s/2)} \leqslant \frac{\pi}{s} \, . \end{aligned} $$ Furthermore, $$ \begin{aligned} \int_0^{2\pi} \phi(s) \, \frac{\pi}{s} \, ds & = \sum_{n = 1}^\infty \int_{1/2^n}^{1/2^n+1/8^n} 2^n \times \frac{\pi}{s} \, ds \\ & \leqslant \sum_{n = 1}^\infty \int_{1/2^n}^{1/2^n+1/8^n} 2^n \times 2^n \pi \, ds \\ & = \sum_{n = 1}^\infty \frac{1}{8^n} \times 4^n \pi = \frac{\pi}{3} < \infty . \end{aligned} $$ Thus, by the dominated convergence theorem, $$ \begin{aligned} \lim_{r \to 1^-} (-\log |f(r)|) & = \lim_{r \to 1^-} \int_0^{2\pi} \phi(s) K_r(s) ds \\ & = \int_0^{2\pi} \phi(s) \lim_{r \to 1^-} K_r(s) ds = 0 . \end{aligned} $$ In other words, the radial boundary limit of $|f|$ at $0$ is $1$.

We have thus proved that at no boundary point the radial boundary limit of $f$ is equal to zero. On the other hand, the radial boundary limit of $-\log |1/f| = \log |f|$ at $e^{it}$ is equal to $\phi(t)$ for almost all $t$, and since $\phi$ is essentially unbounded, $1/f$ is not bounded in the unit disk. To be specific: the radial boundary limit of $|1/f|$ at $t = \tfrac{1}{2^n} + \tfrac{1}{2} \tfrac{1}{8^n}$ is equal to $e^{\phi(t)} = e^{2^n}$. In other words, $f$ is not invertible in the class of bounded outer functions.


Part 3. The above $f$ fails to have a radial boundary limits at $t = \tfrac{1}{2^n}$ and $t = \tfrac{1}{2^n} + \tfrac{1}{8^n}$. If we require $f$ to have a radial boundary limit everywhere, then we need to modify $\phi$ appropriately: we choose an auxiliary smooth "bump" function $\psi$ equal to $0$ outside of $(0, 1)$, equal to $1$ in $[\tfrac13, \tfrac23]$, and taking values in $[0, 1]$, and we set $$ \phi(t) = \begin{cases} 2^n \psi(8^n (t - \tfrac{1}{2^n})) & \text{if } t \in [\tfrac{1}{2^n}, \tfrac{1}{2^n} + \tfrac{1}{8^n}] \text{ for some } n = 1, 2, \ldots , \\ 0 & \text{otherwise.} \end{cases} $$ Now $\phi$ is smooth except at $t = 0$, and hence $f$ is continuous in $\overline{D} \setminus \{1\}$. The same calculation as in Part 2 shows that $-\log |f| = -\Re \log f$ again has a radial boundary limit $0$ at $1$. A similar calculation with the conjugate Poisson kernel shows that also $-\Im \log f$ has a radial boundary limit at $1$, and so $f$ has a (non-zero) radial boundary limit at $1$.

$\endgroup$
2
  • $\begingroup$ Could you give us additional details for $t=0$? I suppose you take integration on $[0,2\pi]$? Are you sure that the limit exists? $\endgroup$
    – ray
    Mar 27, 2021 at 15:22
  • $\begingroup$ With the previous definition of $\phi$ (that had $1/4^n$ instead of $1/8^n$) the radial boundary limit at $1$ indeed did not exist, but nevertheless it was not $0$. :-) In fact, the lower limit of $|f(r)|$ as $r \to 1^-$ was positive. I edited the definition so that the boundary limit exists at $1$ (see Part 2), and sketched how to get an example that has a boundary limit everywhere (see Part 3). Also, I gave more details in Part 2. Let me know if something is unclear. $\endgroup$ Mar 27, 2021 at 19:27
0
$\begingroup$

It is worth to mention that there exists a special class of outer functions where the answer is positive: let $u$ be a non-constant inner function and $|\alpha|=1$. Then $f:=u-\alpha$ is an outer function which has radial limit $0$ somewhere. This is a direct consequence of a theorem of Hoessjer-Frostman (see Noshiro's book, "Cluster sets" Theorem 6).

$\endgroup$

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.