In the Ahlfors' Complex Analysis chapter about the Riemann Mapping Theorem section 6.1.3, page 233, he states and proves this theorem:

Theorem 3. Suppose that the boundary of a simply connected region $\Omega$ contains a line segment $\gamma$ as a one-sided free boundary arc. Then the [Riemann mapping] function $f(z)$ which maps $\Omega$ onto the unit disk can be extended to a function which is analytic and one-to-one on $\Omega \cup \gamma$. The image of $\gamma$ is an arc $\gamma'$ on the unit circle.

In this particular context by "free boundary arc" he means an open interval $a < x < b$ of the real line.
I understand the proof of the existence of the extension: use the version of Schwarz reflection which only requires continuity of $\text{Re} \log f(z)$ at the boundary.

What I don't understand is his argument that $f$ is injective on $\gamma$. Here's how it goes:

We note further that $f'(z) \ne 0$ on $\gamma$. Indeed, $f'(x_0) = 0$ wold imply that $f(x_0)$ were a multiple value, in which case the two subarcs of $\gamma$ that meet at $x_0$ would be mapped on arcs that form an angle $\pi/n$ with $n \ge 2$; this is clearly impossible. If, for instance, the upper half disks are in $\Omega$, then $$\partial \log |f| / \partial y = - \ \partial \arg f / \partial x < 0$$ on $\gamma$, and $\arg f$ moves constantly in the same direction. This proves that the mapping is one-to-one on $\gamma$.

My attempt to clarify: In order to show that $f'(x_0)$ is not zero, I figured that if it were, we could factor $f(z)-f(x_0) = [(z-x_0)g(z)]^n$ where $g(x_0) \ne 0$. I think if you pick the right arc in the domain $\Omega \cup \gamma$, the image of that arc contains a point outside the closed unit disk.
Even if $f'(x) \ne 0$ on $\gamma$, I still don't see why the image of $\gamma$ can't "wrap all the way around the circle" and come back to the same point, and so be locally injective without being globally one-to-one.

Any help is greatly appreciated.


Concerning the non-vanishing of $f'$ on $\gamma$, Ahlfors' argument is incorrect (even the best make mistakes sometimes), $f^{-1}$ would map two arcs of the unit circle to two arcs at an angle of $\pi/n$, not $f$ maps two subarcs of $\gamma$ to two arcs intersecting at an angle of $\pi/n$, cf. also this question.

Once the non-vanishing of $f'$ on $\gamma$ is established, the injectivity follows thus:

Suppose there were $x_1 \neq x_2$ in $\gamma$ with $f(x_1) = f(x_2) = w$. Then $f$ maps small disks $D_{\rho_k}(x_k)$ conformally to neighbourhoods of $w$. If $\rho_k$ is small enough, we have

$$f\bigl(D_{\rho_k}(x_k) \cap \Omega\bigr) = \mathbb{D}\cap f\bigl(D_{\rho_k}(x_k)\bigr)$$

for $k\in \{1,2\}$. But then every point in the nonempty open set

$$\mathbb{D}\cap f\bigl(D_{\rho_1}(x_1)\bigr) \cap f\bigl(D_{\rho_2}(x_2)\bigr)$$

would have at least two preimages in $\Omega$, thus $f$ would not be injective on $\Omega$.

Since $f$ is by assumption injective on $\Omega$ (it is, after all, a biholomorphic map between $\Omega$ and $\mathbb{D}$), the assumption that $f(x_1) = f(x_2)$ for some $x_1 \neq x_2$ in $\gamma$ must have been false.

  • $\begingroup$ Thank you - again. I blame the error on his typist. ;) $\endgroup$ – bryanj Jun 15 '14 at 1:48
  • $\begingroup$ So it boils down to: locally one-to-one, plus open mapping. Got it. $\endgroup$ – bryanj Jun 15 '14 at 1:52

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.