## Introduction
In **differential geometry** we develop the idea of a *[differentiable](/page/Derivative) manifold* so that the familiar tools of calculus extend to curved spaces that need not sit inside any Euclidean space.
Nineteenth-century results of **Stokes, Green, and Gauss** showed that, when a sufficiently smooth hypersurface $S \subset \mathbb{R}^{n}$ bounds a solid region $U$, certain [integrals](/page/Integral) depend only on data along the *[boundary](/page/Boundary)* $\partial U$.
To make such theorems rigorous we must specify **how smooth** that boundary is; differential geometry solves the problem elegantly by defining smoothness through coordinate charts—the very language used to describe manifolds themselves.
## Formal Definition (Differential-geometry version)
[definition: $C^{k}$ boundary via sub-manifolds]
Let $k \in \mathbb{N} \cup \{\infty\}$ and let $U \subset \mathbb{R}^{n}$ be an [open set](/page/Open%20Set).
We say that **$U$ has a $C^{k}$ boundary** if its [topological](/page/Topology) boundary $\partial U$ is a $C^{k}$ embedded sub-manifold of codimension $1$ in $\mathbb{R}^{n}$.
Concretely, around every point $x^{0} \in \partial U$ there exist an open neighbourhood $V \subset \mathbb{R}^{n}$ and a $C^{k}$ diffeomorphism
\begin{align*}
\Phi : V &\longrightarrow \Phi(V) \subset \mathbb{R}^{n},
\end{align*}
such that
\begin{align*}
\Phi(x^{0}) = 0, \qquad
\Phi\bigl(V \cap \partial U\bigr) = \{\, (y',0) : y' \in \mathbb{R}^{\,n-1} \}.
\end{align*}
[/definition]
## PDE-friendly Graph Characterisation
When we deal with pdes the differential geometry definition is not so easy to work with.
[proposition]
For $k \ge 1$ the following statements are equivalent for an open set $U \subset \mathbb{R}^{n}$.
1. $\partial U$ is a $C^{k}$ embedded sub-manifold of codimension $1$ in $\mathbb{R}^{n}$.
2. For every $x^{0} \in \partial U$ there exist
* a radius $r > 0$,
* an orthonormal coordinate system $(x_{1},\dots,x_{n})$ with origin at $x^{0}$,
* a function
\begin{align*}
\gamma \in C^{k}\!\bigl(B'_{r}(0)\bigr), \qquad B'_{r}(0) \subset \mathbb{R}^{\,n-1},
\end{align*}
such that
\begin{align*}
U \cap B_{r}(x^{0})
= \bigl\{\,(x',x_{n}) \in B_{r}(x^{0}) : x_{n} > \gamma(x')\,\bigr\},
\qquad x' = (x_{1},\dots,x_{n-1}).
\end{align*}
[/proposition]
[proof]
$(1) \;\Rightarrow\; (2)$ From manifold chart to graph
Because $\partial U$ is a $C^{k}$ embedded sub-manifold, pick a $C^{k}$ chart
\begin{align*}
\psi : W \longrightarrow \mathbb{R}^{\,n-1}, \qquad \psi(x^{0}) = 0,
\end{align*}
whose coordinate [functions](/page/Function) parametrize $\partial U$ near $x^{0}$.
Let $\nu$ be the outward unit normal to $\partial U$ at $x^{0}$ and rotate the ambient space so that $\nu = e_{n}$.
Write $x = (x',x_{n})$ in this basis and define
\begin{align*}
F(x',x_{n}) = \psi(x',x_{n}).
\end{align*}
Since $\partial_{x_{n}}F(x^{0}) \neq 0$, the implicit-function theorem yields a $C^{k}$ function $\gamma$ on a ball $B'_{r}(0)$ with
\begin{align*}
\partial U \cap B_{r}(x^{0}) &= \{\, (x',\gamma(x')) : x' \in B'_{r}(0) \}, \\
U \cap B_{r}(x^{0}) &= \{\, (x',x_{n}) : x_{n} > \gamma(x'),\; x' \in B'_{r}(0) \},
\end{align*}
establishing (2).
---
$(2) \;\Rightarrow\; (1)$ — From graph patches to manifold atlas
Assume (2). Define
\begin{align*}
\varphi : B'_{r}(0) \longrightarrow \partial U, \qquad \varphi(x') = (x',\gamma(x')).
\end{align*}
The Jacobian of $\varphi$ has rank $n-1$, so $\varphi$ is a $C^{k}$ embedding.
Taking all such graph patches gives an atlas $\{\varphi_{j}\}$ for $\partial U$; transition maps are
\begin{align*}
\varphi_{i}^{-1}\circ\varphi_{j}(x') = x',
\end{align*}
hence $C^{k}$. Therefore $\partial U$ is a $C^{k}$ embedded sub-manifold, verifying (1).
[/proof]
There are a couple reason analysists prefer this version. In particular:
- **Inside vs. outside.**
The inequality $x_{n} > \gamma(x')$ records which side of the surface belongs to the domain.
- **Boundary flattening.**
The change of variables
\begin{align*}
(y',y_{n}) \;\longmapsto\; (\,y',\,y_{n} + \gamma(y')\,)
\end{align*}
sends a flat half-ball to $U$, turning boundary estimates into interior ones.
- **Coefficient control.**
If $\gamma$ is $C^{k}$, then normals, Jacobians, and metric factors inherit $C^{k-1}$ regularity—exactly what elliptic and parabolic theory demands.
Thus this seemingly technical definition becomes indispensable for energy estimates, trace theorems, divergence-form identities, and countless other foundational tools in the analysis of PDEs.
