I'm really struggling to understand the definition of a "submanifold with boundary". Until now, I'm only familiar with the notion of a "submanifold of $\mathbb R^d$. I've defined this notion in the form I'm aware of below$^1$, but meanwhile I guess it should better be called "$C^1$-submanifold" and what I call "chart" should be called "$C^1$-chart", but in particular the latter might be totally wrong.

Let $M\subseteq\mathbb R^d$ for some $d\in\mathbb N$, $k\in\{1,\ldots,d\}$ and $\mathbb H^k:=\{x\in\mathbb R^k:x_k\ge0\}$.

Maybe it's easier to motivate the definition of a "submanifold with boundary" using the following equivalent characterization of a submanifold: Let $$\mathcal D_d:=\{(\Omega,\psi)\mid\Omega\subseteq\mathbb R^d\text{ is open and }\psi\text{ is a diffeomorphism from }\Omega\text{ onto }\psi(\Omega)\}.$$ Then $M$ is a $k$-dimensional embedded ($C^1$-)submanifold of $\mathbb R^d$ if and only if $$\forall x\in M:\exists(\Omega,\psi)\in\mathcal D_d:x\in\Omega\text{ and }\psi(M\cap\Omega)=\psi(\Omega)\cap(\mathbb R^k\times\{0\}).\tag2$$ Now let $\iota_k$ denote the canonical embedding of $\mathbb R^k$ into $\mathbb R^d$ with $\iota\mathbb R^k=\mathbb R^k\times\{0\}$. Then, it's easy to see that if $(\Omega_1,\psi)\in\mathcal D_d$ with $\psi(M\cap\Omega)=\psi(\Omega)\cap(\mathbb R^k\times\{0\})$ and $\Omega_2:=\psi(\Omega_1)$, then $U:=\iota_k^{-1}(\Omega_2)$ is open, $\psi^{-1}\in C^1(\Omega_2,\mathbb R^d)$, $$\phi:=\psi^{-1}\circ\left.\iota_k\right|_U\in C^1(U,\mathbb R^d)$$ and $$\phi(U)=M\cap\Omega_1$$ and hence $(U,\phi)$ is a $k$-dimensional chart of $M$.

Now, and hopefully I got it right, the definition of a "submanifold with boundary" should be as follows: $M$ is called $k$-dimensional embedded ($C^1$-)submanifold with boundary of $\mathbb R^d$ if for all $x\in M$ there is a $(\Omega,\psi)\in\mathcal D_d$ with $x\in\Omega$ and either

  1. $\psi(M\cap\Omega)=\psi(\Omega)\cap(\mathbb R^k\times\{0\})$; or
  2. $\psi(M\cap\Omega)=\psi(\Omega)\cap(\mathbb H^k\times\{0\})$ and $\psi_k(x)=0$.

I'm not sure, but maybe we can show that for each fixed $x$, either all choices of $(\Omega,\psi)$ satisfy (1.) or all choices satisfy (2.). I've asked for that separetegly: Show that these two diffeomorphisms cannot exist simultaneously.

How does the corresponding "chart definition" of a submanifold with boundary look like? Am I right that we can construct such a chart in the same way as before? Assuming $x\in M$ and $(\Omega,\psi)$ is as in the definition above satisfyin (2.). Then, if again $U:=\iota_k^{-1}(\Omega_2)$ and $\phi:=\psi^{1-}\circ\left.\iota_k\right|_U$, we should have $\psi(M\cap\Omega_1)=\iota(U\cap\mathbb H^k)$ and hence $M\cap\Omega_1=\phi(\mathbb H^k\cap U)$. $\phi$ is still an immersion from $U$ to $\mathbb R^d$ and a topological embedding of $U$ into $\phi(U)$. And $U\cap\mathbb H^k$ is open in $\mathbb H^k$; maybe that's what we need to replace.

Besides that I've read that we can always (i.e. for any submanifold with boundary, $x\in M$ and $(\Omega,\psi)\in\mathcal D_d$ with $x\in\Omega$) assume that $\psi(M\cap\Omega)=\psi(\Omega)\cap(\mathbb H^k\times\{0\})$; the only difference would be that if $x\in\partial M$ (I'm not sure if the topological boundary is meant), then $\psi_k(x)=0$ and if $x\in M^\circ$ (I'm not sure if the topological boundary is meant), then $\psi_k(x)>0$. Why is that the case?

$^1$ $(U,\phi)$ is called $k$-dimensional chart of $M$ if $U\subseteq\mathbb R^k$ is open, $\phi:U\to\mathbb R^d$ is an immersion and a topological embedding of $U$ into $M$ and $\phi(U)$ is $M$-open. Let $\mathcal C_k(M)$ denote the set of $k$-dimensional charts of $M$.

$M$ is called $k$-dimensional embedded submanifold of $\mathbb R^d$ if $$\forall x\in M:\exists(U,\phi)\in\mathcal C_k(M):x\in\phi(U)\tag1.$$

  • $\begingroup$ What is $u_d$? To show that the boundary is a manifold (in $\mathbb{R}^d$), a hint is to adapt the atlas on $M$ to an atlas on $\partial M$ - how can you change each chart? $\endgroup$ Jul 5, 2020 at 13:35
  • $\begingroup$ @OliverClarke Sorry, $u_d$ was supposed to be $u_k$; the $k$th component of $u$. $\endgroup$
    – 0xbadf00d
    Jul 5, 2020 at 13:41
  • $\begingroup$ @OliverClarke Please take note of my edit. $\endgroup$
    – 0xbadf00d
    Jul 5, 2020 at 13:56
  • $\begingroup$ By "manifold" I mean a $k'$ dimensional submanifold of $\mathbb{R^n}$, for some $k'$ - if you write down some charts for boundary of $M$ you'll see what the dimension of the boundary should be. $\endgroup$ Jul 5, 2020 at 13:56
  • $\begingroup$ According to Do Carmo's definition, a submanifold $N$ of dimension $k$ of a manifold $M$ of dimension $d$ is just considering open sets $W$ that cover $N$ and verify $\sigma \cap W$ is diffeomorphic to $\mathbb{R}^k$ for all charts $\sigma$ of $M$ that intersect $W$. I think in this context you can define it exactly in this way relative to $\mathbb{H}$ which is somewhat easier and inherits topological properties from $M$ in a more straightforward manner. $\endgroup$
    – astro
    Jul 5, 2020 at 14:19

1 Answer 1


A point $x$ of $M$ is an interior point if and only if there is an open neighbourhood of $x$ which is homeomorphic to an open subset of $\mathbb{R}^k$. So let's pick a chart $(U, \phi)$ where $x = \phi(u)$ for some $u \in U$.

If $u_k = 0$ then a $\mathbb{H}^k$-open neightbourhood of $u$ is not open in $\mathbb{R}^k$. So $x$ is not an interior point and so must lie on the boundary of $M$.

If $u_k > 0$ then we can restrict $U$ to $V = U \cap \{y \in \mathbb{H}^k : y_k > 0 \}$ and we get a chart $(V, \phi |_V)$. So a neightbourhood of $x$ is homeomorphic to $V$ which is an open subset of $\mathbb{R}^k$. So $x$ is an interior point.

The boundary of $M$ is a submanifold of $\mathbb{R}^n$ (without boundary). Hint: write down charts for $\partial M$ by using the charts for $M$. This will tell you what is the dimension of $\partial M$.

  • $\begingroup$ Thank you for your effort so far. I think I've already got a better understanding of it, but it still is not totally clear me. I've edited the question and hopefully it's clearer now what I'm struggling to understand. $\endgroup$
    – 0xbadf00d
    Jul 5, 2020 at 19:21

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.