4

Let $f$ : $U\rightarrow V$ be a proper holomorphic map where $U$ and $V$ are open subsets of $\mathbb{C}$ with $V$ connected. Show that the cardinality of the fibres of $f$, i.e. $f^{-1}(\{z\})$ counted with the multiplicities are the same for each $z \in$ $V$. This looks like the property of covering maps and so I was trying to prove if $f$ is a local homeomorphism or a covering map, but to no avail. Thanks for any help.

Ester
  • 3,057
  • 2
  • 23
  • 34
  • What does "counted with the multiplicities" mean in this context? – Paul Frost Jul 14 '18 at 08:15
  • @PaulFrost If $f(a) = b$ then $f$ is locally around $a$ of the form $(z-a)^n+b + o(z-a)^{n+1}$, and $n$ is called the multiplicity of the point $a$ – Bart Michels Jul 14 '18 at 08:19
  • @barto I see. Nonconstant and proper implies surjective. Is it perhaps sufficient to assume that $f$ is surjective? In that case $f$ cannot be constant. – Paul Frost Jul 14 '18 at 08:26
  • Sorry! I have edited. f is a proper map. I am not sure if the other conditions like U is connected are actually required. Non-constant follows from properness. – Ester Jul 14 '18 at 08:33
  • You do not need $U$ connected. barto's proof doesn't use that. But assume that you have a proof only for connected $U$. Let $U_i$ be the components of $U$. Then each $f_i = f \mid_{U_i}$ is proper and holomorphic so that you know what happens with the fibres. Each $f_i$ must be surjective, therefore the preimage $f^{-1}(z)$ of any $z \in V$ is compact and intersects all $U_i$. This is only possible if you have a finite number of components which gives you the desired result for $f$. – Paul Frost Jul 14 '18 at 11:40

2 Answers2

3

This is very simple using some complex analysis.

Since $f$ is proper, given $p\in V$ and small enough $\epsilon>0$ there exists a cycle $\Gamma\subset U$ such that if $|p-q|<\epsilon$ then all the zeroes of $f-q$ lie "inside" $\Gamma$, and in fact such that if $z$ is a zero of $f-q$ then the index of $\Gamma$ about $z$ is $1$ (also the index of $\Gamma$ about any point of $\Bbb C\setminus U$ is $0$.).

Details added on request: If $\epsilon>0$ is small enough then $\overline{D(p,\epsilon)}\subset V$; since $f$ is proper this shows that $K=f^{-1}(\overline{D(p,\epsilon)})$ is a compact subset of $U$. Hence by a nameless result that appears in most books on complex analysis because it's needed a lot, there exists a cycle $\Gamma\subset U\setminus K$ with index $1$ about every point of $K$ and index $0$ about every point of $\Bbb C\setminus K$. (With apologies for knowing one particular book better than the others, this is Lemma 10.5.5 in Complex Made Simple.)

Hence if $|p-q|<\epsilon$ the number of zeroes of $f-q$ is $$\frac1{2\pi i}\int_\Gamma\frac{f'(z)}{f(z)-q}\,dz.$$That integral depends continuously on $q$...

  • Can you please elaborate ? – Ester Jul 14 '18 at 16:09
  • What part don't you get? – David C. Ullrich Jul 14 '18 at 16:18
  • How the zeros of f-q lie inside gamma and why it is of index 1?Finally what contradiction do we get? – Ester Jul 14 '18 at 16:30
  • 1
    @Ester David can correct me if I'm wrong, but I think the idea here is first that $f^{-1}({p})$ is simply number of zeroes of $f - p$. We then consider a neighborhood around $p$, and want to show $f^{-1}({q}) = f^{-1}({p})$ for each $q$ in the neighborhood. This of course means the number of zeroes of $f - q$ is equal to the number of zeroes of $f - p$. Since $f$ pulls back compact sets to compact sets, we should be able to surround the preimage with a cycle $\Gamma$ with simple winding number that goes about all the roots of $f - q$ for all $q$ in the neighborhood. – Brevan Ellefsen Jul 14 '18 at 16:40
  • @Ester I added details on the topology. II don't know what contradiction you're referring to. An integer-valued continuous function on a connected set is constant. – David C. Ullrich Jul 14 '18 at 16:43
  • 1
    @Ester because the integral continuously depends on $q$ (in fact, we can explicility evaluate it using argument principle - this is simply the argument principle with $g(z) = f(z) - q$) the number of roots of $f - q$ is the same for all $q$ in the neighborhood; in particular, $f - q$ has the same number of roots as $f - p.$ I'm not sure how David intends the argument finished, but it seems that by filling the set $V$ by overlapping neighborhoods we see $f - q$ has the same number of roots on each neighborhood and thus on the set. – Brevan Ellefsen Jul 14 '18 at 16:50
  • 2
    @BrevanEllefsen Say $n(p)$ is the number of zeroes of $f-p$. The argument given shows that $n$ is continuous at $p$, for every $p$. So $n$ is continuous in $V$, hence $n$ is constant in $V$ since $V$ is connected. – David C. Ullrich Jul 14 '18 at 16:57
  • @DavidC.Ullrich ah, I see! A clever way to finish the proof using the OP's assumption of connectedness. Thanks for elaborating. – Brevan Ellefsen Jul 14 '18 at 17:02
2

Here's a useful lemma:

Lemma. Let $X,Y$ be locally compact Hausdorff topological spaces, $f : X \to Y$ continuous, open, proper, surjective and with discrete fibers. Let $K$ be a neighborhood of $y \in Y$, $y_1, \ldots, y_n$ its preimages (a finite number, by properness and discreteness) and $K_i$ a neighborhood of each $y_i$. Then there exist disjoint open neighborhoods $V_i \subseteq K_i$ of the $y_i$ and an open neighborhood $V \subseteq K$ of $y$ such that $f^{-1}(V)$ is the disjoint union of the $V_i$.

It is nothing deep (you can prove it). It is similar to the proof that a proper surjective local homeomorphism is a covering map. The only difference is that we don't have local injectivity here.

Note that proper implies closed, so proper+open+ $Y$ connected implies surjective.

Now let $X=U$, $Y=V$ as in the question. By the lemma, outside of the set $B$ of branch points (images of points where the derivative $f'$ vanishes) $f$ is a covering map (by the lemma + we now have local injectivity at the preimages). In particular, the size of fibers is constant on connected components of $Y - B$, say equal to $n$.

If $f$ proper, the set of branch points is closed (by closedness) and discrete (by properness) in $V$. In particular, $Y-B$ is connected.

It remains to check what happens in a neighborhood of a branch point $y$. Take $V$ and $V_i$ as in the lemma. It suffices to check that the number of preimages is constant for each of the restrictions $f : V_i \to V$. On such $V_i$, $f$ has the form $y + a_{m_i}(z-y_i)^{m_i} + a_{m_i+1} (z-y_i)^{m_i+1} + \cdots$ by Taylor-expansion, for some $m_i$ ($a_{m_i} \neq 0$). We want $\sum m_i=n$. W.l.o.g. we may assume $y_i=y=0$. We have that $$a_{m_i} z^{m_i} + a_{m_i+1} z^{m_i+1} + \cdots = g(z)^{m_i}$$ for some holomorphic $g$ which is a homeomorphism between open neighborhoods of $0$. Replacing $f(z)$ by $(f \circ g^{-1}) (z) = z^{m_i}$ does not change the number of preimages. So the fibers have cardinality $\sum m_i$ in a neighborhood of $y$, including at $y$. Comparing this to any point different from $y$ in that neighborhood, we conclude that $\sum m_i=n$.

Bart Michels
  • 26,355
  • I have some problems understanding your solution.How do you conclude that f is a covering map? Covering maps also need to satisfy the homeomorphism property with the slices. Finally how does it reduce to the case z---->z^n? – Ester Jul 14 '18 at 08:50
  • Thanks for the clarification. I shall be accepting your solution a little later in case someone comes up with something else. – Ester Jul 14 '18 at 09:08
  • You're welcome. If you want, I have some notes where I wrote down full proofs of this, but they're in French. I'm having issues to upload files to my site right now. – Bart Michels Jul 14 '18 at 13:05
  • 1
    Got it: http://planzeta.be/files/surfaces_de_riemann_2017.pdf Prop 4.5 - Prop 4.11 – Bart Michels Jul 14 '18 at 13:23