2
$\begingroup$

Let us consider a smooth bounded domain $\Omega \subset \mathbb R^n$ and the problem $$ \begin{cases} -\Delta u = 0 & x \in \Omega \\ u = 1 & x \in \partial \Omega \end{cases} $$ Does it make sense to use the change of variables $v = u-\mathbf{1}_{\partial \Omega} $ to reduce the problem to the following one (with a source term but homogeneous boundary data)? $$ \begin{cases} -\Delta v = \Delta \mathbf{1}_{\partial \Omega} & x \in \Omega \\ v = 0 & x \in \mathbb R^n \setminus \Omega \end{cases} $$ How can this change of variable be made rigorous in the context of viscosity solutions?


Because of the answer below, let us consider the same question for $$ \begin{cases} -\Delta u +\lambda u= 0 & x \in \Omega \\ u = 1 & x \in \partial \Omega \end{cases} $$ with $\lambda >0$.


This question is motivated by Viscosity solutions of $(-\Delta)^s u = 0$ in $\Omega $ with non-homogeneous data $u = 1$ in $\mathbb R^n \setminus \Omega$

$\endgroup$

1 Answer 1

2
$\begingroup$

There seems to be a weakness to the approach with your (generalized) differential equation for $v$, because the Laplacian of the indicator function $\Delta\mathbf{1}_{\partial\Omega}$ is not actually singular enough to produce a meaningful boundary condition on the surface $\partial\Omega$. We have that $\Delta\mathbf{1}_{\partial\Omega}=0$ wherever it is defined, and where it is undefined (that is, on $\partial\Omega$) it is merely undefined (with a removable singularity), not singular like a $\delta$-distribution (or derivative thereof). You can see this by integrating $\Delta\mathbf{1}_{\partial\Omega}$ times a test function and applying Cauchy's second formula (which is integration by parts twice), $$\int_{\mathcal{V}}dV\, \phi\left(\Delta\mathbf{1}_{\partial\Omega}\right)= \int_{\mathcal{V}}dV\, (\Delta\phi)\mathbf{1}_{\partial\Omega} +\int_{\partial\mathcal{V}}dA\left(\phi\frac{\partial\mathbf{1}_{\partial\Omega}}{dn}-\mathbf{1}_{\partial\Omega}\frac{\partial\phi}{\partial n} \right)=0,$$ where $\mathcal{V}$ is a larger volume containing $\Omega$. The integral is exactly the same as if $\mathbf{1}_{\partial\Omega}$ were replaced by $0$.

If the source term on the right-hand-side of the Poisson equation contained a singular distribution, that could yield a condition on a discontinuity of $v$ or its derivatives. However, with only $\Delta\mathbf{1}_{\partial\Omega}$, we are left with merely, $$\Delta v=0\quad \textrm{almost everywhere},$$ to which the generalized solution satisfying the boundary condition that $v|_{\partial\Omega}=0$ is just $v(x)=0$ everywhere. This is not actually giving a wrong answer; it just corresponds to the fact that the original boundary value problem for $u$ has the unique solution $u(x)=1$ on $\Omega$. However, your revised source is not actually contributing anything to the solution, so it is questionable whether this approach can be usefully generalized.

$\endgroup$
1
  • $\begingroup$ Thanks. What if we consider $-\Delta u +\lambda u= 0$? Then $u \equiv 1$ is no longer a solution. $\endgroup$
    – Zac
    Nov 3, 2021 at 1:02

Your Answer

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

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