The concept of a boundary is fundamental to distinguishing between the interior of a domain and the exterior. While the [topological](/page/Topology) definition is sufficient for general set theory, the study of Partial Differential Equations (PDEs) requires boundaries to possess specific geometric smoothness. This regularity allows us to define normal vectors, perform [integration](/page/Integral) by parts, and establish the regularity of solutions.

## Formal Definition

In general topology, the boundary is defined in relation to the closure and interior of a [set](/page/Set).

[definition: Topological Boundary] Let $U \subset \mathbb{R}^n$ be a subset. The **boundary** of $U$, denoted by $\partial U$, is the set of points in the closure of $U$ that do not belong to the interior of $U$: \begin{align*} \partial U := \bar{U} \setminus U^\circ. \end{align*} [/definition]

While this definition identifies the "edge" of a set, it allows for extremely rough geometries (e.g., fractals) where calculus is impossible. For PDE theory, we restrict our attention to "smooth" boundaries.

## Smooth Boundary

To perform analysis on the boundary—such as defining the trace of a [function](/page/Function) or the outer unit normal—we require the boundary to locally resemble a flat Euclidean space.

### Motivation

Standard calculus operations are well-defined on the flat half-space $\mathbb{R}^n_+ = \{z \in \mathbb{R}^n \mid z_n > 0\}$. To extend these operations to arbitrary domains, we need a rigorous way to say that a curved boundary "looks like" a flat half-space locally.

To avoid ambiguity regarding coordinate choices, we first define the boundary using the **Implicit (Level Set)** method. This definition is intrinsic and does not rely on the orientation of the axes.

[definition: $C^k$ Boundary (Implicit)] Let $U \subset \mathbb{R}^n$ be an open, bounded set. We say $\partial U$ is $C^k$ ($k \in \mathbb{N}, k \geq 1$) if for each point $z \in \partial U$, there exists a radius $r > 0$ and a $C^k$ function $\Psi: \mathbb{R}^n \to \mathbb{R}$ such that: 1. **$U$ is the sublevel set:** \begin{align*} U \cap B(z, r) = \{ x \in B(z, r) \mid \Psi(x) < 0 \} \end{align*} 2. **$\partial U$ is the zero set:** \begin{align*} \partial U \cap B(z, r) = \{ x \in B(z, r) \mid \Psi(x) = 0 \} \end{align*} 3. **Non-degeneracy:** \begin{align*} \nabla \Psi(x) \neq 0 \quad \text{for all } x \in B(z, r) \end{align*} [/definition]

If the boundary is $C^\infty$, we call $U$ a **smooth domain**. If $\Psi$ is merely Lipschitz [continuous](/page/Continuity), we refer to $U$ as a **Lipschitz domain**.

### Local Graph Representation

While the level set definition is mathematically robust, explicit calculations (like those found in Evans Appendix C) are often easier if we view the boundary as a graph. We now show that the implicit definition allows us to construct a local coordinate system where the boundary is a graph.

[quotetheorem:6]

### Straightening the Boundary

A primary technique in PDE regularity theory is to "flatten" the boundary locally via a change of variables. We define a map $\Phi$ that transports the problem from the physical set $U$ to a simplified target set.

[quotetheorem:50]