3

In John Lee's Introduction to Smooth Manifolds, a collar neighborhood of a smooth manifold $M$ with boundary is defined to be a neighborhood $V$ of $\partial M$ that is the image of a smooth embedding $\alpha: [0,1) \times \partial M \to M$ such that $\alpha(0,x)=x$ for all $x \in \partial M$. However, is it necessarily true that $\alpha$ is a diffeomorphism onto $V$? The following results in Lee seem to assume this, but I can't figure out how to show this.

The issue is that $[0,1) \times \partial M$ and $M$ both have boundaries, so we can't apply results that depend on the rank theorem. I believe this could be remedied by knowing that we can embed any smooth manifold with boundary into a smooth manifold without boundary, but this result in turns depends on what I'm currently asking being true, for Lee makes this diffeomorphism assumption in proving this fact (see Errata to Thm 9.26).

Any help would be much appreciated! Thank you :)

  • 2
    It is a nice exercise to check that the local diffeomorphism theorem holds for manifolds with boundary. – Moishe Kohan Sep 03 '24 at 23:01
  • @MoisheKohan To be precise, what is the result you are referring to? I know that, for example, it is not necessarily true for a smooth map $f:M \to N$ between manifolds with boundary that if $df_p$ is invertible, then $f$ is a local diffeomorphism at $p$. For example, look at the inclusion of half-space into Euclidean space (as Lee does). Is it true, however, that any smooth immersion which is a homeomorphism (between manifolds with boundary) is necessarily a diffeomorphism? How would one show this if so? – AJ LaMotta Sep 04 '24 at 02:17
  • You have to assume that boundary maps to boundary, of course. – Moishe Kohan Sep 04 '24 at 03:18

2 Answers2

1

Here is the desired theorem: Let $f: M \to N$ be a smooth map between smooth manifolds with boundary such that $f(\partial M) = \partial N$. If $df_p$ is invertible, then $f$ is a local diffeomorphism at $p \in M$.

Proof. If $p \in \mathrm{int}\, M$, then simply apply the result for manifolds without boundary to the smooth map $f: \mathrm{int}\, M \to \mathrm{int}\, N$. Otherwise, $p \in \partial M$. By taking smooth charts at $p$ and $f(p)$, we can assume that $M = N = \mathbb H^n$ and $p = f(p) = 0$. We can find a smooth function $g: U \to \mathbb R^n$ on a neighborhood $U$ of $0$ in $\mathbb R^n$ such that $g$ extends $f$ on $U \cap \mathbb H^n$. Then $dg_0$ is invertible, so we can assume $U$ is small enough that $g$ is a diffeomorphism onto an open ball $V \ni 0$ in $\mathbb R^n$.

Now let $\pi: \mathbb R^n \to \mathbb R$ denote projection onto the last coordinate, which is an open map. For a subset $S \subseteq \mathbb R^n$, we write $S^{>a} = S \cap \pi^{-1}((a,\infty))$ and similarly define $S^{\geq a}$, $S^{<a}$, $S^{\leq a}$, and $S^a$ (the latter being $S \cap \pi^{-1}(a)$). We know that $g$ maps $U^0$ into $V^0$ and $U^{>0}$ into $V^{>0}$, so it suffices to prove that $g$ maps $U^{<0}$ into $V^{<0}$ (for then $g^{-1}|_{V^{\geq 0}}$ is a local smooth inverse to $f$ at 0). Notice $$ V^{>0} = (V^{>0} \cap g(U^{\leq0})) \sqcup (V^{>0} - g(U^{\leq 0})). $$ Clearly $V^{>0} \cap g(U^{\leq0}) = V^{>0} \cap g(U^{< 0})$ is open, and $V^{>0} - g(U^{\leq 0}) \supseteq g(U^{>0})$ is nonempty open, so $V^{>0} \cap g(U^{\leq0}) = \varnothing$ because $V^{>0}$ is connected. Thus $g(U^{<0}) \subseteq V^{\leq 0}$. But $\pi(g(U^{<0})) \subseteq (-\infty,0]$ is open, so in fact $g(U^{<0}) \subseteq V^{<0}$, as desired. //

Edit: Thanks to Karthik Kannan for pointing out a subtlety in my original answer, which I have corrected by taking $V$ to be a ball. See also Is a map with invertible differential that maps boundary to boundary a local diffeomorphism?, where this subtlety is again overlooked.

  • Your proof is the same as the one in the linked answer. – Moishe Kohan Sep 04 '24 at 13:37
  • 1
    @A.J. LaMotta how do you know that $g$ maps $U\cap\partial\mathbb{H}^{n}$ bijectively onto $V\cap\partial\mathbb{H}^{n}$? If $q\in V\cap\partial\mathbb{H}^{n}$ is it not possible that there exists a $p\in U\setminus\mathbb{H}^{n}$ (i.e. a point in the lower half plane) such that $g(p) = q$? – Karthik Kannan Jan 27 '25 at 16:53
  • My point is that even though $f(\partial\mathbb{H}^{n}) = \partial\mathbb{H}^{n}$ it may be the case that $f^{-1}({q})\cap (U\cap\partial\mathbb{H}^{n}) = \varnothing$ whereas $g^{-1}(q)$ may lie in $U\setminus\partial\mathbb{H}^{n}$. – Karthik Kannan Jan 28 '25 at 07:33
  • @KarthikKannan Thanks for pointing this out! If we take $V$ to be a ball, then connectedness considerations prevent this. I've edited my original answer. – AJ LaMotta Apr 06 '25 at 19:39
  • @A.J.LaMotta I'll try to read your proof carefully when I review this material. – Karthik Kannan Apr 12 '25 at 19:26
  • @AJLaMotta I think your argument is correct and is indeed much simpler than my answer below. Thanks! – Karthik Kannan Jun 27 '25 at 06:06
0

As in the previous answer, let $U$ and $V$ be open neighborhoods of $p$ and $f(p)$ respectively in $\mathbb{R}^{n}$, and let $g : U\rightarrow V$ be a diffeomorphism such that $g|_{U\cap \mathbb{H}^{n}} = f|_{U\cap\mathbb{H}^{n}}$. The problem is that even though $f(\partial\mathbb{H}^{n}) = \partial\mathbb{H}^{n}$, it might happen that$ f(U\cap\partial\mathbb{H}^{n}) = g(U\cap\partial\mathbb{H}^{n})$ is a proper subset of $V\cap\partial\mathbb{H}^{n}$. More specifically, it might be the case that for some $q\in V\cap\partial\mathbb{H}^{n}$, we have $f^{-1}(\{q\})\cap (U\cap\partial\mathbb{H}^{n}) = \varnothing$ (and thus $g^{-1}(q)\in U\setminus\partial\mathbb{H}^{n}$).

This situation can be remedied as follows. If $h = f|_{\partial\mathbb{H}^{n}}: \partial\mathbb{H}^{n}\rightarrow\partial\mathbb{H}^{n}$ then $\text{d}h_{p}: T_{p}\partial\mathbb{H}^{n}\rightarrow T_{f(p)}\partial\mathbb{H}^{n}$ is also an isomorphism. By the Inverse Function Theorem, $h$ is a diffeomorphism on a neighborhood of $p$ in $\partial\mathbb{H}^{n}$. Let $C$ be small open cubical neighborhood of $f(p)$ contained in $V$ such that $h: h^{-1}(C\cap\partial\mathbb{H}^{n})\rightarrow C\cap\partial\mathbb{H}^{n}$ is a diffeomorphism. In particular, $g : g^{-1}(C)\rightarrow C$ is a diffeomorphism that maps $g^{-1}(C)\cap\partial\mathbb{H}^{n} = h^{-1}(C\cap\partial\mathbb{H}^{n})$ onto $C\cap\partial\mathbb{H}^{n}$. Using a connectedness argument, we may conclude that $f(g^{-1}(C)\cap\text{Int}(\mathbb{H}^{n})) = C\cap\text{Int}(\mathbb{H}^{n})$.