1
$\begingroup$

Let $\Omega\subseteq\mathbb{R}^N$ be an open and bounded set, and let $\phi:\Omega\to\mathbb{R}^N$ be a $C^1$ function with the property that $\phi^{-1}(0)\neq\emptyset$, and $\nabla\phi(x)\neq 0,\ \forall\ x\in \phi^{-1}(0)$.

How can we prove or disprove that:

$$\mathcal{H}^{N-1}\left (\overline{\phi^{-1}(0)}\setminus\phi^{-1}(0)\right )=0$$

It is well-known that $\mathcal{L}^N(\phi^{-1}(0))=\mathcal{H}^N(\phi^{-1}(0))=0$ (Lebesgue measure of the zero level set is null).

I denote by $\mathcal{H}^{N-1}$ the $N-1$ dimensional Hausdorff measure on $\mathbb{R}^N$.

$\endgroup$
7
  • 3
    $\begingroup$ So what if I take $N=1$, $\Omega = (0,1)$, and $\phi(x) = \sin(1/x)$? Then $\overline{\phi^{-1}(0)} \setminus \phi^{-1}(0) = \{0\}$ whose 0-dimensional Hausdorff measure (counting measure) is not zero. I realize this is pretty trivial, but couldn't one make similar examples in higher dimensions? $\endgroup$ Jan 1, 2021 at 17:40
  • $\begingroup$ Does $\bar$ mean the closure, that is, the set you are interested in is contained in the boundary of $\Omega$? $\endgroup$ Jan 1, 2021 at 17:41
  • $\begingroup$ @NateEldredge for example, $N=2$, $\Omega=(0,1)^2$, the same function (not depending on the second coordinate). $\endgroup$ Jan 1, 2021 at 17:44
  • 1
    $\begingroup$ @Bogdan: How about $\Omega$ as in my previous example and $\phi(x) = x^3 \sin(1/x)$? Or $x^n \sin(1/x)$ if you want more derivatives. $\endgroup$ Jan 1, 2021 at 18:29
  • 1
    $\begingroup$ @bogdan I guess you mean the differential of $\phi$ at any zero x to be invertibile, not just nonzero $\endgroup$ Jan 1, 2021 at 19:46

1 Answer 1

3
$\begingroup$

The next result answers the question in the negative.

Theorem. There is $\phi:\mathbb{R}^n\supset\Omega\to\mathbb{R}^n$ of class $C^\infty$ such that $\phi$ is a local diffeomorphism in a neighborhood of $\phi^{-1}(0)$, but the Lebesgue measure of the following set is positive: $$ (*)\quad \mathcal{L}^n\left (\overline{\phi^{-1}(0)}\setminus\phi^{-1}(0)\right )>0. $$

Proof. Let $\Omega\subset\mathbb{R}^n$ be a open set such that $\mathcal{L}^n(\partial\Omega)>0$. It is well known that such sets exist and in fact they can be homeomorphic to a ball.

Let $E=\{x_i\}_{i=1}^\infty\subset\Omega$ be a countable set such that $\partial\Omega\subset\overline{E}$. Let $r_i>0$ be such that $$ \overline{B}(x_i,r_i)\subset\Omega \quad \text{and} \quad \overline{B}(x_i,r_i)\cap\overline{B}(x_j,r_j)=\emptyset. $$ Define $$ \phi:\bigcup_{i=1}^\infty\overline{B}(x_i,r_i)\to B(0,1) $$ as a similarity in each ball and extend it to $\Omega$ as a $C^\infty$ map. Then $E\subset\phi^{-1}(0)$ and hence $$ \partial\Omega\subset \overline{E}\setminus\Omega\subset\overline{\phi^{-1}(0)}\setminus\phi^{-1}(0). $$ proves ($*$). Clearly, $\phi$ is a local diffeomorphism in a neighborhood of $E\subset\phi^{-1}(0)$, but there might be points $x\in \phi^{-1}(0)\setminus E$ where the Jacobian $J_\phi=0$ equals zero. To avoid this problem we simply remove a small neighborhood of this set from $\Omega$. $\Box$

$\endgroup$
3
  • 1
    $\begingroup$ In fact one may take $\Omega$ to be the union of the family of balls, skipping the extension & restriction steps $\endgroup$ Jan 2, 2021 at 0:07
  • $\begingroup$ @PietroMajer You are right. I actually wanted to have a domain homeomorphic to a ball so the extension, but then I realized that I have to remove something. Perhaps I will modify my answer following your suggestion. $\endgroup$ Jan 2, 2021 at 12:57
  • $\begingroup$ I think one can do the extension as you do for any given $\Omega$ without introducing new zeros. Join the balls with a set of arcs so to form a tree (a Christmas tree indeed). Since this is contractible, we can map $\Omega$ to this tree by a smooth map $h$, that also map each ball in itself homeomorphically. We then compose $h$ with a map like you do, that maps each ball by a similarity to the unit ball, and each arc somewhere in the complement. $\endgroup$ Jan 2, 2021 at 13:44

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.