2

Below is problem 4.28 from Lee's Introduction to Topological manifolds:

Suppose $M$ is a noncompact manifold of dimension $n \ge 1$. Show that its one-point compactification is an $n$-manifold if and only if there exists a precompact open subset $U \subseteq M$ such that $M \setminus U$ is homeomorphic to $\mathbb{R}^n \setminus \mathbb{B}^n$. [Hint: you may find the inversion map $f: \mathbb{R}^n \setminus \mathbb{B}^n \to \overline{\mathbb{B}^n}$ defined by $f(x)=x/|x|^2$ useful.]

I can establish the "only if" direction by considering a regular coordinate ball surrounding $\infty$. However, I have difficulty tackling the converse direction. After some searching on the site, I found the following answer:

$\color{red}{\text{Since } M\setminus U \text{ is homeomorphic to } \mathbb{R}^n\setminus \mathbb{B}^n, \text{ it follows that } M\setminus\overline{U} \text{ is homeomorphic to } \mathbb{R}^n\setminus \overline{\mathbb{B}^n}.}$ Call such a homeomorphism $g$. By using the inversion map $f$, one sees that $\mathbb{R}^n\setminus\overline{\mathbb{B}^n}$ is homeomorphic to $\mathbb{B}^n \setminus\{\vec{0}\}$. Composing $g$ and $f$, we have a homeomorphism between $M\setminus\overline{U}$ and $\mathbb{B}^n\setminus \{\vec{0}\}$. Try to prove that we can use these to find a homeomorphism between $M^\ast \setminus \overline{U}$ and $\mathbb{B}^n$.

It is exactly the red sentence that bothers me for a long time. Assuming the theorem on the invariance of boundary, since homeomorphism preserves interior points, $\text{int } (M\setminus U)\cong \text{int } (\Bbb R^n\setminus \Bbb B^n)= \Bbb R^n\setminus \overline{\Bbb B^n}$ (Here int should be interpreted as the manifold interior). If I can verify $\text{int } (M\setminus U) =M\setminus\overline{U}$, then I am done. But does it hold for arbitrary $U$ satisfying the hypothesis or do I have to choose $U$ cleverly? Of course, alternative proofs are welcome as well. Thanks in advance.

PatrickR
  • 4,247
YuiTo Cheng
  • 4,705
  • 1
    You're overthinking it. Since $U\subset\overline U$, also $M\setminus\overline U\subset M\setminus U$. The latter space is homeomorphic to $\mathbb{R}^n\setminus B^n$ by assumption. Adding in the point at infinity we have the composite ${\infty}\cup (M\setminus \overline U)\subset {\infty}\cup (M\setminus U)\cong {\infty}\cup(\mathbb{R}^n\setminus B^n)$ which takes its domain onto an open subspace of the target. Since we've (hopefully) already done the case $M=\mathbb{R}^n$, we're done. – Tyrone Sep 11 '20 at 16:54

1 Answers1

3

Lemma 1: Suppose $X$ is a manifold with boundary, $M$ is a manifold, and that $i:X\rightarrow M$ is a codimension $0$ embedding. If $Int(X)$ is the manifold interior of $X$, and $int$ refers to the topological interior then $i(Int(X)) = int(i(X))$.

Proof: Suppose $x\in Int(X)$. Then there is an open neighborhood $V\subseteq X$ of $x$ for which $U$ is disjoint from the manifold boundary of $X$. Then $i(V)$ is open in $M$ by invariance of domain. Since $i(x)\in i(V)\subseteq i(X)$, this proves that $i(Int(X))\subseteq int(i(X))$.

Conversely, assume $x\in int(i(X))$. Then there is an open set $V\subseteq i(X)$ containing $x$. Because $M$ is a manifold, the charts form a basis, so there is a neighborhood $x\in V'\subseteq V$ for which $V'$ is homeomorphic to a ball. Since $i$ is an embedding and $V'\subseteq V\subseteq i(X)$ it follows that $i^{-1}(V')$ is a ball around $i^{-1}(x)$. In particular, $i^{-1}(x)\in Int(X)$. That is, $x\in i(Int(X))$. $\square$

Using lemma 1, when you write "Int" we may interpret that as either the manifold interior, or the topological interior. Now, the claim follows from

Lemma 2: For any topological space $M$ and an subset $U$, $\operatorname{int}(M\setminus U) = M\setminus \overline{U}$. (There is no requirement that $M$ be a topological manifold, or that $U$ be open, or anything like that.)

Proof: If $x\in \operatorname{int}(M\setminus U)$, then there is an open set $V$ with the property that $x\in V\subseteq M\setminus U$. In particular, $x\notin\overline{U}$ because $V\cap U=\emptyset$ and $x\in V$. Thus, $x\in M\setminus \overline{U}$. This proves that $\operatorname{int}(M\setminus U)\subseteq M\setminus \overline{U}$.

On the other hand, suppose $x\in M\setminus \overline{U}$. Since $x\notin \overline{U}$, there is an open set $V$ for which $x\in V$ and $V\cap U = \emptyset$. But this means $V\subseteq M\setminus U$, so $x\in \operatorname{int}(M\setminus U)$. This shows that $M\setminus\overline{U}\subseteq \operatorname{int}(M\setminus U)$. $\square$

C.F.G
  • 8,523
  • But the intetior in your proof is the topological interior, right? The topological interior of the space is simply the space itself, which is quite different from the manifold interior as stated in the OP's attempt. – YuiTo Cheng Sep 11 '20 at 15:47
  • @YuiToCheng: Of course, you're right. My reading comprehension skills suck today. – Jason DeVito - on hiatus Sep 11 '20 at 15:51
  • @YuiToCheng: I've updated the answer. I think I'd addressed your (completely valid) objection. – Jason DeVito - on hiatus Sep 11 '20 at 20:44
  • Thanks for filling in the gap. I've never thought about relating the manifold interior to the topological interior, even though I am aware of lemma 2. – YuiTo Cheng Sep 12 '20 at 05:40
  • @ChanYunLi: I was unaware of Lemma 1 until YuiTo Cheng pointed out the gap (and my lack of reading comprehension skills!). That got me thinking about the difference between the two notions of interior, which led to Lemma 1. – Jason DeVito - on hiatus Sep 12 '20 at 17:29
  • @JasonDeVito: Are "$Int$" and "$int$" and "$\operatorname{int}$" all different? – C.F.G Sep 12 '20 at 17:43
  • @C.F.G: Well, $Int$ and $int$ are different in general (thought Lemma 1 asserts they coincide for codimension $0$ embeddings). For example, the closed interal $[0,1]$ thought of as a subset of $\mathbb{R^2}$ has $int([0,1]) = \emptyset$ and $Int([0,1]) = (0,1)$. The $\operatorname{int}$ is a left-over from the previous version of the answer. The original intent was that $\operatorname{int} = int$. I'm not sure it's worth an edit, but if the question gets bumped again, then changing all $\operatorname{int}$ to $int$ would be appropriate. – Jason DeVito - on hiatus Sep 12 '20 at 17:47