The Jordan-Schönflies theorem states that any simple closed curve $C$ divides the plane into two connected regions $U$ and $V$, and that any point $x \in C$ is a boundary point of both $U$ and $V$. A stronger statement would be that for any point $x$ in $C$, the set $U \cup V \cup \{x\}$ is pathwise connected. Equivalently, any point in $U$ can be connected to $x$ by a continuous path lying entirely inside $U$ (except for its ending point, of course) and likewise for $V$.

I haven't run across any proof of this stronger statement. Has it been proved? If not, is there a counterexample?

  • 7
    $\begingroup$ The follows from Caratheodory's theorem: en.wikipedia.org/wiki/… $\endgroup$ – Autumn Kent May 9 '13 at 20:08
  • 6
    $\begingroup$ A better question is if the same is true for, say, codimension 1 topological spheres in $R^n$, since these can be wild (unlike for $n=2$). $\endgroup$ – Misha May 9 '13 at 21:53
  • 12
    $\begingroup$ Actually, what you state is not the full "Jordan Schonflies theorem". The full statement of the Schonflies theorem is that there is a homeomorphism of the plane taking $C$ to $S^1$, from which your property follows easily. $\endgroup$ – Lee Mosher May 9 '13 at 22:48

The answers by Kent and Eremenko are, of course, absolutely correct. Here is the more general statement: Let $M$ be a closed connected topological $n-1$-dimensional manifold embedded in ${\mathbb R}^n$ (note that $M$ can be wild for $n\ge 3$). Let $U$ be a component of ${\mathbb R}^n -M$. Then for every point $x\in M$ there exists a continuous path $p: [0,1)\to U$, so that

$$ \lim_{t\to 1} p(t)=x$$

In particular, $({\mathbb R}^n -M) \cup \{x\}$ is path-connected. The proof is based on the following observation:

Observation. There exists a function $\phi(r), r\in [0, \infty)$, satisfying $\phi(r)\ge r$ and
$$ \lim_{r\to 0+} \phi(r)=0$$ so that: For every $r>0$ the map $$ \tilde{H}_0(U \cap B(x, r))\to \tilde{H}_0(U \cap B(x, \phi(r))) $$ (induced by inclusion) is zero. In other words, any two points in $U \cap B(x, r)$ can be connected by a path in $U \cap B(x, \phi(r))$.

The above observation follows from the Poincare/Alexander duality in ${\mathbb R}^n$ (in conjunction with the fact that $M$ is a manifold).

Given the above observation, take a sequence $x_k\in U\cap B(x, \frac{1}{k})$ and construct the path $p$ by concatenating paths $p_i$ in $$ B(x, \phi(\frac{1}{i}))\cap U $$ connecting points $x_i, x_{i+1}$, $i\in {\mathbb N}$.


Yes. One way to prove this (and the Jordan theorem too) is to use Complex Variables:-) A good reference is Milnor, MR2193309 Dynamics in one complex variable. Third edition. Annals of Mathematics Studies, 160. Princeton University Press, Princeton, NJ, 2006.

A chapter in this book contans the best exposition of these questions that I know.


Let y be a point in, say, the exterior E of C, and x be a point in C. Then x belongs to the closure of E, hence, there is a sequence of points {x_n} in E which converges to x. As E is connected, there is a chain y --- x_0 --- x_1 --- x_2 --- of continuous curves lying entirely within E.

To see that their union connects y to x, we have to use a trick.

Namely, after a line s_0 connecting y to x_0 within E is defined, let r_0 be its distance to C, S_0 be the open (r_0/2) nbhd of s_0, and X_0 be the open (r_0/2) nbhd of x_0 - now we claim that the line s_1 connecting x_0 to x_1 be defined so that it does not contain points in S_0 except of those in X_0 - which is easily possible by path-connectedness.

A suitable iteration of this trick yields the result.

  • 4
    $\begingroup$ You did not use the fact that $C$ is a curve in your argument; for general compact subsets the statement is false (e.g., "topologist's sine curve"). $\endgroup$ – Misha May 11 '13 at 23:28

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.