I would like to figure out how to properly define the outward unit normal vector $\nu$ for a bounded domain $\Omega \subset \mathbb{R}^n$ with smooth boundary $\partial \Omega$ ($n \ge 2$). I am following Gilbarg and Trudinger's Elliptic Partial Differential Equations of Second Order.

This function $\nu$ doesn't seem to be defined anywhere in Gilbarg and Trudinger, but it is used often, and even at the beginning of the text, when the divergence theorem is stated.

First, I will start with a definition from section 6.2 in Gilbarg and Trudinger:

Let $\Omega \subset \mathbb{R}^n$ be a bounded domain. We say $\partial \Omega$ is of class $C^k$ if and only if, for each $x_0 \in \partial \Omega$, there is a open ball $B(x_0) = B$ and an bijective function $\psi \colon B \to D \subseteq \mathbb{R}^n$ so that:

  1. $\psi(B \cap \Omega) \subseteq \mathbb{R}^n_+ = \{x \in \mathbb{R}^n | x_n > 0\}$,

  2. $\psi(B \cap \partial \Omega) \subseteq \partial \mathbb{R}^n_+ = \{x \in \mathbb{R}^n | x_n = 0\}$,

  3. $\psi \in C^k(B), \psi^{-1} \in C^k(D)$.

Note: D is open in $\mathbb{R}^n$ by invariance of domain.

Now, starting with this definition, I would like to come to a suitable definition for the outward unit normal (i.e., a definition that makes the divergence theorem hold).

Here's what I have so far. For $x_0 \in \partial \Omega$, we can let $\psi = (\psi_1, \dots \psi_n) : B \to D$ be as in the definition above, and then notice that $B \cap \partial \Omega$ is a level set of the function $\psi_n$ ($\psi_n \equiv 0$ on $B \cap \partial \Omega$). Now, from multivariable calculus, I remember learning that $\nabla \psi_n(x_0)$ "points perpendicular" to the level set $B \cap \partial \Omega$. Intuitively, this seems to be to the property that we want $\nu$ to possess, so it seems like a good attempt to define $\nu(x_0) = \frac{\nabla \psi_n(x_0)}{||\nabla \psi_n(x_0)||}$.

However, I see two issues with trying to make this our definition for $\nu$.

  1. How do we know that $\nabla \psi_n (x_0) \neq 0$, so that we can make it into a unit vector?
  2. Since the map $\psi$ corresponding to a particular ball is not necessarily unique, how do we know that $\nu$ is well defined?

Any comments are greatly appreciated!


By property 1, the function $\psi_n$ increases going inside the domain. You should set $$\nu(x_0) = - \frac{\nabla \psi_n(x_0)}{||\nabla \psi_n(x_0)||}$$ if you want the outward normal.

How do we know that $\nabla \psi_n (x_0) \neq 0$?

By assumption, $\psi$ is a diffeomorphism (its inverse is also smooth). Therefore, the Jacobian matrix of $\psi$ is invertible. Consequently, every row of this matrix is a nonzero vector. Including the last row, which is $\nabla \psi_n$.

how do we know that $\nu$ is well defined?

Suppose $\phi$ is another map like $\psi$. It's convenient to normalize it (by adding a constant) so that $\phi(x_0)=\psi(x_0)$. Then $\phi = F\circ \psi$ where $F = \phi\circ \psi^{-1}$ is a diffeomorphism of a neighborhood of $\psi(x_0)$. Note that $F$ leaves the hyperplane $x_n=0$ invariant. Therefore, the partial derivatives of its $n$th component with respect to $x_1,\dots,x_{n-1}$ are zero. In matrix terms, this means the $n$th row of the Jacobian matrix $DF$ is of the form $(0,0,\dots, 0, * )$ where $*$ stands for a nonzero number. This number is positive, because $F$ sends upper halfspace into itself.

The chain rule, applied to $\phi = F\circ \psi$, shows that $\nabla \phi_n(x_0)$ is a positive multiple of $\nabla \psi_n(x_0)$. Therefore, the normalization of these vectors produces the same result.


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.