This is a question from an old qualifying exam that I was trying to solve for practice:

Prove or disprove that there exist nonconstant holomorphic functions $f$ and $g$ on the open unit disk $D=\{z\in \mathbb{C}: |z|<1\}$ such that $e^{f(z)}+e^{g(z)}=1$ for all $z\in D$?

My initial thought was to show it is true on the real line and then use the uniqueness principle. But the examples I came up with on the real line use logarithms, which will not be continuous when I consider them as functions on the unit disk.

Since this is for practice, I was hoping only for hints and suggestions instead of complete answers. Thank you!

  • $\begingroup$ "uniqueness principle" would only apply if you had existence on an open set, and the real line is not open in $\mathbb{C}$. $\endgroup$ – John Fernley Jul 31 '14 at 0:26
  • 2
    $\begingroup$ I thought it was existence on a set with a nonisolated point, for which the real line would work. $\endgroup$ – user122916 Jul 31 '14 at 0:30
  • $\begingroup$ I think that's not right - to be honest uniqueness principle cannot show existence either way! It would show uniqueness $\endgroup$ – John Fernley Jul 31 '14 at 0:43
  • $\begingroup$ In practice, the idea of using the uniqueness is unlikely to be helpful. You would have to be in a scenario where (a) it was very obvious that $f,g$ are holomorphic on the whole disk, but (b) the computation to show the above formula is only "easy" to do on $\mathbb{R}$, which seems unlikely to me. $\endgroup$ – JHance Jul 31 '14 at 0:44
  • $\begingroup$ @ John Fernley Uniqueness theorem for holomorphic functions states that if they are equal on a set with a limit point they are equal on the entire domain encyclopediaofmath.org/index.php/…. Real line is more than enough. $\endgroup$ – Conifold Jul 31 '14 at 1:23

Hint: it is fine to use logarithms as long as they are applied to values from a set that doesn't circle the origin. Suppose you picked some $f(z)$, than you need to solve $e^{g(z)}=1-e^{f(z)}$. One solution is obviously $g(z)=\text{Ln}(1-e^{f(z)})$, where $\text{Ln}$ is the principal branch of the logarithm with the cut along the negative half-axis. So you need to make sure that $f(z)$ maps the unit disk into a region, which $1-e^{z}$ then maps into something not crossing the negative half-axis. A judiciously chosen Möbius transformation should do the trick.

  • $\begingroup$ Wonderful explanation, thank you. $\endgroup$ – user122916 Aug 9 '14 at 21:07

Just chose $g(x)$ such that $e^{g(z)}\neq 1$ for all $z$ in the unit disk. Then $f(x)=\ln (1-e^{g(x)})$ under some well defined branch of the log function.

  • $\begingroup$ $e^{g(z)}\neq 1$ is not enough. If $1-e^{g(z)}$ maps the disk onto some annulus surrounding $0$ there won't be globally defined branch of logarithm on it even if it misses $0$ itself. $\endgroup$ – Conifold Jul 31 '14 at 2:03
  • $\begingroup$ @Conifold: If you have a nowhere-vanishing holomorphic function $h(z)$ in a simply connected region $U$ then you can define a branch of $\log h(z)$ in $U$. Here $U$ is the unit disc, which is clearly simply connected, and $h(z)$ is $1 - e^{g(z)}$. $\endgroup$ – A. Barron Jul 31 '14 at 2:11
  • $\begingroup$ I was thinking of something like $z^2$ mapping a simply connected piece of annulus onto the whole annulus, there wouldn't be a branch of $\ln$ on that. $\endgroup$ – Conifold Jul 31 '14 at 2:18
  • $\begingroup$ Fix a point $z_{0}$ in $U$. You can define a branch by letting $L(z) = \int_{\gamma} \frac{h'(w)}{h(w)} \ dw + c_{0}$, where $\gamma$ is a path in $U$ connecting $z$ to $z_{0}$ (this is well-defined by the simple connectivity assumption), and $c_{0}$ is chosen so that $e^{c_{0}} = h(z_{0})$. It's then easy to verify that $e^{L(z)} = h(z),$ e.g. see Stein and Shakarchi's 'Complex Analysis,' end of Ch. 3. This isn't a "branch" in the normal sense as when you're considering all of $\mathbb{C}$, but the log is still well-defined in $U$. $\endgroup$ – A. Barron Jul 31 '14 at 2:22
  • $\begingroup$ Look up Monodromy theorem. $\endgroup$ – Rene Schipperus Jul 31 '14 at 2:26

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.