2
$\begingroup$

Let $\Omega\subset\mathbb{R}^N$ be a bounded smooth domain. Assume that $u\in W_0^{1,2}(\Omega)$, $u\ge 0$ and $-\Delta u\ge 0$ in the sense of distributions ($u$ is superharmonic).

The standard mollifier process, ensure the existence of a sequence $u_n\in C_0^\infty(\Omega)$, such that $u_n\ge 0$, $u_n$ is superharmonic (for large $n$) in every open set $\omega\subset\subset\Omega$ and $u_n\to u$ in $W^{1,2}(\Omega)$.

My question is: instead of the above regularization, I would like to find one where $u_n\in C_0^\infty(\overline{\Omega})$, $u_n\ge 0$, $-\Delta u_n\ge 0$ in $\Omega$ and $u_n\to u$ in $W^{1,2}(\Omega)$. Is it possible?

One possible way to solve the problem, is to extend $u$ outside $\Omega$, in a nieghbourhood of $\partial \Omega$, such that the extension is superharmonic. This is somehow related with the normal derivative of $u$ (if it exists). When $\Delta u$ is a bounded Radon measure, there is $f\in L^1(\partial \Omega)$, which will be denoted by $\frac{\partial u}{\partial \eta}$, such that $$\int_\Omega\varphi\Delta u=-\int_\Omega \nabla \varphi\nabla u+\int_{\partial\Omega} \varphi\frac{\partial u}{\partial \eta}, \forall \varphi\in C^\infty(\overline{\Omega})\tag{1}, $$

If $U_\delta=\{x\in \mathbb{R}^N\setminus\Omega:\ \operatorname{dist}(x,\partial\Omega)<\delta\}$ for small $\delta>0$ then, one reasonable extension would be $$\overline{u}(x)=\frac{\partial u}{\partial \eta}(\tau_\delta (x))\operatorname{dist}(x,\partial\Omega), \ x\in U_\delta,$$

where $\tau_\delta(x)$ is the unique pointe in $\partial \Omega$ such that $\operatorname{dist}(x,\partial\Omega)=\operatorname{dist}(x,\tau_\delta(x))$. In the classical case, by Hopf's lemma, $\frac{\partial u}{\partial \eta}<0$, so the above extension would be superharmonic, however, I fail to see anything in this generalized sense.

Another approach, which seems intuitive and can work for the general case, is to consider $$\Omega_\delta=\{x\in\Omega: \operatorname{dist}(x,\partial\Omega)\le \delta\}$$

and define $\overline{u}_\delta(x)=u(x)$ in $\Omega\setminus\Omega_\delta$ and $\overline{u}_\delta(x)=\frac{\operatorname{dist}(x,\partial\Omega)}{\delta}u(\tau_\delta (x))$ where $\tau_\delta (x)$ is the unique point in $\partial(\Omega\setminus\Omega_\delta)$, which minimizes $\operatorname{dist}(x,\partial(\Omega\setminus\Omega_\delta)$.

If $-\Delta u_\delta\ge 0$ and $u_\delta\to u$ in $W^{1,2}(\Omega)$ then, we could exnted each $u_\delta$ outside $\Omega$ in the obvius way and regularize it, to obtain the desired result.

$\endgroup$
5
  • $\begingroup$ (deleted comment) $\endgroup$ Jul 22, 2014 at 15:41
  • $\begingroup$ The second approach does not seems to work, if $\Omega$ is not convex. $\endgroup$
    – Tomás
    Jul 22, 2014 at 15:54
  • $\begingroup$ Why don't you mollify $u_\delta = u \chi_{\Omega_\delta}$ with a mollifier whose support lies wholly inside $B_{\delta/2}$, and then take $\delta \rightarrow 0$? Is this a problem somehow? $\endgroup$
    – Ray Yang
    Aug 4, 2014 at 12:18
  • $\begingroup$ @RayYang, this mollified sequence is not superharmonic in $\Omega$. $\endgroup$
    – Tomás
    Aug 6, 2014 at 15:24
  • $\begingroup$ It's true. See the proof of approximation theorems for BV functions. You can adapt the proof because $\nabla u$ is a BV vector. $\endgroup$
    – student
    Aug 12, 2015 at 1:42

0

Your Answer

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

Browse other questions tagged or ask your own question.