MathOverflow is a question and answer site for professional mathematicians. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Let $f\colon \mathbb R^1\to \mathbb R^3$ be a continuous and injective map. Is $\mathbb R^3\setminus f(\mathbb R^1)$ a path-connected space?

share|cite|improve this question
up vote 14 down vote accepted

Yes. In fact, $\mathbb R^3$ could be any 3-manifold and $f(\mathbb R^1)$ any countable union of embedded segments.

Lemma 1. Let $U$ be an open ball in $\mathbb R^3$ and $f:I\to\mathbb R^3$ an embedding. Then $U\setminus f(I)$ is path-connected.

Proof. See Hatcher, Proposition 2B.1 on page 169. This proposition proves (among other things) that $S^3$ minus any embedded segment is path-connected. Extract the proof of this particular statement and repeat it word-by-word with $U$ in place of $S^3$. (The proof uses only the following facts about $U$: $U$ minus any closed sub-segment of $f(I)$ is open; $U$ minus any point of $\mathbb R^3$ is simply connected.)

Lemma 2. Let $f:I\to\mathbb R^3$ be an embedding and $s:I\to\mathbb R^3$ a path with endpoints outside $f(I)$. Then for any $\varepsilon>0$ there exists a path $s_\varepsilon:I\to\mathbb R^3$ which is $\varepsilon$-close to $s$ (in $C^0$), connects the same endpoints, and avoids $f(I)$.

Proof. Divide the domain of $s$ into intervals so small that the diameters of their images are less than $\varepsilon/10$. Let $p_0,p_1,\dots,p_n\in\mathbb R^3$ be the images of the division points ($p_0$ and $p_n$ are the endpoints of $s$). For each $p_i$, let $q_i$ be a point outside $f(I)$ such that $|p_i-q_i|<\varepsilon/10$ (for $i=0$ and $i=n$, choose $q_i=p_i$). Such a point $q_i$ exists since $f$ is injective and hence cannot cover a set with nonempty interior. Let $U_i$ be the ball of radius $\varepsilon/3$ centered at $p_i$. By Lemma 1, $U_i\setminus f(I)$ is path-connected. Therefore we can connect $q_i$ to $q_{i+1}$ by a path contained in $U_i$ and avoiding $f(I)$. The (suitably parametrized) product of these paths is the desired $s_\varepsilon$, q.e.d.

Now return to the original problem. The set $f(\mathbb R)$ is a countable union of sets $J_k$, $k\in\mathbb N$, where each $J_k$ is an injective image of a segment. We want to prove that any two points $p,q\in\mathbb R^3$ can be connected by a path avoiding $\bigcup J_k$.

Let $s_1$ be a path from $p$ to $q$ avoiding $J_1$ (such a path exists by Lemma 1). Let $\varepsilon_1$ be the minimum distance from $s_1$ to $J_1$, divided by 10. By Lemma 2, there exists a path $s_2$ which is $\varepsilon_1$-close to $s_1$ and avoids $J_1$. Let $\varepsilon_2$ be the minimum of $\varepsilon_1$ and the minimum distance from $s_1$ to $J_1$, divided by 10. Then there is a path $s_3$ which is $\varepsilon_2$-close to $s_2$ and avoids $J_2$, and so on. The resulting sequence $s_1,s_2,\dots$ converges (since $C^0$ is complete) to some path $s$ which avoids every set $J_k$.

Remark. The closure of $f(\mathbb R)$ can separate the space, for example, $f(\mathbb R)$ can be a dense subset of the torus.

share|cite|improve this answer
Excellent answer: plenty of instructive technique on display here. Probably belongs on the Tricki somewhere. The invocation of completeness of a path space (and use of dependent choice) would seem to be unavoidable. – Todd Trimble Nov 13 '10 at 12:03
Thanks, great answer! – Nikita Kalinin Nov 13 '10 at 21:22

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.