Note: This is one of the problems from Complex Analysis by Stein and Shakarchi.

Let $\Omega \subset \mathbb{C}$ be a bounded region (open, connected), and let $L \subset \mathbb{C}$ be a line that intersects $\Omega$, you may assume that $\Omega \cap L = I$ is an interval.

The line $L$ splits the complex plane in two halves, and since $\Omega$ is open, $\Omega$ will have a nonempty intersection with each half. Let's call these intersections $\Omega_l$ and $\Omega_r$. In particular, $\Omega$ will be the disjoint union $\Omega_l \cup I \cup \Omega_r$.

To prove: If $\Omega_l$ and $\Omega_r$ are simply connected, then so is $\Omega$.

Intuitively, this seems rather obvious: if $\Omega_l$ and $\Omega_r$ "have no holes", then $\Omega$ can't have any holes either. However, proving this rigorously has me stumped.

There seem to be a few possible approaches here:

  1. Use the following theorem/lemma:

    A bounded region is simply connected if and only if its complement is connected.

    This gives us that $\Omega^c \cup I \cup \Omega_l$ and $\Omega^c \cup I \cup \Omega_r$ are both connected, and that it suffices to prove that $\Omega^c$ is connected. I don't see how any of these help, though.

  2. Use the definition of simple connectedness:

    An open set is simply connected if any two curves with the same initial and final points inside the set are homotopic.

    We could take two curves inside $\Omega$ and try to deform one into the other. Inside $\Omega_l$ and $\Omega_r$ we can deform curves easily, but when they intersect $I$ I can't figure out how to do it.

  3. Use the winding number:

    A bounded region $\Omega$ is simply connected if and only if $W_\gamma(z) = 0$ for any closed curve $\gamma$ in $\Omega$ and any point $z \notin \Omega$.

    Ideally, we would take any curve $\gamma$ in $\Omega$ and somehow split this into two curves $\gamma_l, \gamma_r$ in $\Omega_l$ and $\Omega_r$ respectively, then write the winding number $W_\gamma(z)$ in terms of the winding numbers of these new curves. But, once again, I can't figure out how to handle this when $\gamma$ and $I$ intersect.

Any hints, suggestions or solutions are appreciated.

  • 1
    $\begingroup$ I suggest method 3. You can restrict to polygonal paths, with no segment parallel to $I$, to have an easy to handle set of possible intersections with $I$. And you can assume that a segment of the path either crosses $I$ or doesn't intersect $I$. Use segments on $I$ to get two closed curves in $\Omega_l \cup I$ and $\Omega_r \cup I$ (if $\gamma$ isn't contained in one part already). $\endgroup$ – Daniel Fischer Apr 14 '17 at 15:40
  • 1
    $\begingroup$ Actually, method 2 works pretty much the same, and if you have enough topology to throw at it, you can just mumble Seifert/van Kampen. $\endgroup$ – Daniel Fischer Apr 14 '17 at 15:47

Lets try method 2.

To simplify matters, we can make use of the following:

If $\Omega\subseteq \Bbb C$ is open, then any curve $\gamma$ in $\Omega$ is homotopic to a polyline.

In fact, we may even impose that the segments of the polyline run only parallel to any two directions of our choice. In particular, we can avoid segments parallel to $L$. Thus the worst that can happen for a closed polyline starting at $a\in \Omega_l$, say, is: It intersects $L$ finitely many times, and between any two consecutive intersections it runs all in $\Omega_l$ or all in $\Omega_r$.

Claim. Any closed polyline in $\Omega$ (with segments not parrallel to $L$) that intersects $L$ is homotopic to such a polyline with less intersection points with $L$.

Proof. Let $a_0a_1\ldots a_n$ (where $a_0=a_n$) be the closed polyline. We may assume wlog. that $a_0\notin $L$. We may assume that every intersection point is in fact a node of the polyline.

Case A: If there is an intersection point $a_i$ such that $a_{i-1}$ and $a_{i+1}$ are on the same side of $L$ (both in $\Omega_l$, say), we can do the following: Pick $r>0$ with $B_r(a_i)\subseteq \Omega$. Pick points $p$ within $[a_{i-1}a_i]$ and $q$ within $a_ia_{i+1}]$. Clearly $pa_iq$ is homotopic to $pq$, and hence $a_0\ldots a_n$ is homotopic to $a_0\ldots a_{i-1}pqa_{i+1}\ldots a_n$, which has one intersection less, as promised.

Case B: If the just treated case A does not occur, the polyline must switch sides at each intersection point. In particular, there must be at least two intersection points. If $a_i$ and $a_j$ are consecutive such intersection points, we can do as follows: The polyline stays on one side of $L$ (wlog. within $\Omega_l$) between $a_i$ and $a_j$ (in particular, $j>i+1$). For each $z\in I$, there is some $r=r(z)>0$ such that $B_r(z)\subseteq \Omega$. Since the closed line segment $[a_i,a_j]\subset I$ is compact, it is already covered by finitely many of such open disks $B_r(z)$ with $z\in I$. Conclude from this that for some $\epsilon>0$, we have $U:=U_\epsilon([a_i,a_j]):=\bigcup_{z\in[a_0,a_n]}B_\epsilon(z)\subseteq \Omega$. Now we can describe another polyline from $a_i$ to $a_j$ that runs inside $\Omega_l$ (except for the endpoints): Let $L'$ be the line parallel to $L$ at some distance $\delta<\epsilon$ on the $\Omega_l$ side. If $\delta$ is small enough, $L'$ intersects $[a_ia_{i+1}]$ in a point $p\in U$ and $[a_{j-1}a_j]$ in a point $q\in U$. Then the polyline $a_ipqa_j$ is homotopic (within $\Omega$) to $a_ia_{i+1}\ldots a_j$ because $pa_{i+1}\ldots a_{j-1}q$ is homotopic (within $\Omega_l$) to $pq$. By the simple shape of $U$, $a_ipqa_j$ is homotopic (in $\Omega$) in a straightforward manner to its reflection at $L$. So far we did not change the number of intersection points. But now $a_i$ (and $a_j$) are points where the polyline does not change sides, hence case A applies and we can reduce the number of intersection points, again. $\square$

As a corollary, every closed line in $\Omega$ (which we may assume to start/end outside $I$) is homotopic to a line living completely on one side of $L$. As $\Omega_l$ and $\Omega_r$ are simply connected, the closed line is ultimately homotopic to a single point. In other words, $\Omega$ is simply connected.


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.