8
$\begingroup$

Recently, I asked a somewhat related question here. In the comment section, I found the formula $$ \lim_{r\to 0}\frac{1}{r}\int_{\Omega_r} f(x)\,dx = \int_{\partial \Omega}f(\sigma)\,d\mathcal{H}^{n-1}(\sigma), $$ where $\Omega_r=\{x\in\Omega: \text{dist}(x,\partial\Omega)<r \}$ is the ring of thickness $r>0$ near the boundary of a (bounded) Lipschitz domain $\Omega\subset \Bbb R^n$. It appears that the formula holds at least for all $f\in C(\overline\Omega)$.

Question: Does similar formula hold for $f\in W^{1,p}(\Omega)$ ? In particular, do we have $$ \lim_{r\to 0}\frac{1}{r}\int_{\Omega_r} |f(x)|^p\,dx = \int_{\partial \Omega}|f(\sigma)|^p \,d\mathcal{H}^{n-1}(\sigma) ? $$

Of course, $f$ on the right-hand side should be thought of as the trace of $f$ in $L^p(\partial\Omega)$.

I notice that the coarea formula almost gives us the result. If I let $d(x) := \text{dist}(x,\partial\Omega)$, then the coarea formula implies that $$ \int_{\Omega_r} |f(x)|^p\,dx = \int_0^r \int_{\{d=t\}} |f(\sigma)|^p \,d\mathcal{H}^{n-1}(\sigma) \,dt $$ hence it follows that $$\lim_{r\to 0}\frac{1}{r}\int_{\Omega_r} |f(x)|^p\,dx = \int_{\{d=0\}} |f(\sigma)|^p \,d\mathcal{H}^{n-1}(\sigma) = \int_{\partial\Omega} |f(\sigma)|^p \,d\mathcal{H}^{n-1}(\sigma) $$ if we know that $t\mapsto \int_{\{d=t\}} |f(\sigma)|^p \,d\mathcal{H}^{n-1}(\sigma) $ is approximately continuous at $t=0$, which I do not know.

Edit: I forgot to mention that I assume $p<n$ otherwise the solution follows easily from the first formula (and Morrey's theorem).

$\endgroup$
6
  • 1
    $\begingroup$ This is true with some regularity on $\partial \Omega$ and follows for $u \in W^{1,1}$ from the half-space case. If $Q$ is a cube in the $x$ variable, then $\int_Q |u(x,y_2)-u(x,y_1)| dx \le \int_{Q\times (y_1,y_2)}|u_y(x,y)|dxdy$ and the function $y \to \int_Q u(x,y)dx$ is continuous. $\endgroup$ Jan 5, 2021 at 13:20
  • 3
    $\begingroup$ If you look in Evans & Gariepy's "Measure theory and fine properties of functions", Sec. 5.3., you see that they define the trace on BV (which includes $W^{1,1}$) for Lipschitz domains in a similar fashion and prove a similar convergence. You should be able to adapt this proof. $\endgroup$
    – mlk
    Jan 5, 2021 at 13:21
  • $\begingroup$ @mlk Thank you, I'll have a look at that and perhaps just prove it myself. The formula in my question just seem so natural hence it surprised me a bit to learn that it's not already well-documented in the literatures. $\endgroup$
    – BigbearZzz
    Jan 5, 2021 at 13:54
  • 1
    $\begingroup$ @mlk since the OP tagged reference request, your comment is probably good as an answer. $\endgroup$ Jan 5, 2021 at 14:48
  • 1
    $\begingroup$ I agree. If he posts an answer based on this comment I'll accept it. $\endgroup$
    – BigbearZzz
    Jan 5, 2021 at 15:00

2 Answers 2

4
$\begingroup$

In Evans & Gariepy's "Measure theory and fine properties of functions", Sec. 5.3., they construct the trace operator on a bounded Lipschitz domain $\Omega$ for BV-functions (and thus by inclusion for the subspaces $W^{1,p}$) in a similar fashion, using averages. In their case, they simply consider each part of the boundary locally as a graph $x_n = \gamma(x_1,...,x_{n-1})$ and then average vertically in the form $$Tf(x_1,...,x_{n-1}) = \lim_{r\to 0} \frac{1}{r}\int_0^r f(x_1,...,x_{n-1},\gamma(x_1,...,x_{n-1}))dr$$ but your construction should be close enough to allow for a similar proof.

They also prove the useful theorem 2 which says that for $\mathcal{H}^{n-1}$-almost all $x\in \partial \Omega$ we have $$\lim_{r\to 0} \frac{1}{|B_r(x) \cap \Omega|} \int_{B_r(x) \cap \Omega} |f-Tf(x)| dy = 0 $$ which also might be helpful.

$\endgroup$
1
$\begingroup$

To add on the answer provided by mlk: I think the result can be more directly using Section 4.3 of Evans and Gariepy (revised edition).

First note that $f\in W^{1,p}(\Omega) \implies |f|^p \in W^{1,1}(\Omega)$, if $1\leq p$. So it suffices to prove it for the case $p = 1$. Now, in Section 4.3 of E&G you find:

Theorem 4.6 (part ii)
Let $\Omega$ be bounded with $\partial\Omega$ Lipschitz, and $1 \leq p < \infty$, and let $T:W^{1,p}(\Omega) \to L^p(\partial\Omega; \mathcal{H}^{n-1})$ be the trace operator. Then for all $\phi\in C^1(\mathbb{R}^n; \mathbb{R}^n)$ and $f\in W^{1,p}(\Omega)$, $$ \int_\Omega f ~\mathrm{div}(\phi)~dx = - \int_\Omega Df\cdot \phi~dx + \int_{\partial\Omega} (\phi\cdot\nu) Tf~d\mathcal{H}^{n-1} $$ with $\nu$ the outward normal to $\Omega$.

You should be able to prove that you want by choosing an appropriate $\phi$ and possibly mollifying. For example, if you let $U_r = \Omega \setminus \Omega_r$ and set $\phi = \frac1{2r} D( \mathrm{dist}(\cdot, U_r)^2)$, you find that for all sufficiently small $r$ the function $\phi$ is Lipshitz continuous, and that $\mathrm{div}(\phi) \approx \frac{1}{r} \mathbf{1}_{\Omega_r}$.

$\endgroup$

Your Answer

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