Let $U\subset \mathbb{R}^n$ with smooth boundary $\partial U$. And consider the expression

$$\Delta u = f.$$

$$\text{+"convenient boundary conditions"}$$

In my specific case $f\in H^2_0$. Under what (boundary) conditions is it possible to conclude a weak solution which is also in $H^2_0$ of this problem?

The problem is mainly concerned with the subscript zero in $H^2_0$. For instance if I introduce the boundary condition

$$u=0 \text{ on } \partial U\ \text{ (trace-wise),}$$

then for any $f\in L^2$, there exists a weak solution in $H^1_0$. I can regularize the solution to be also in $H^2$, hence it is in $H^1_0\cap H^2$. (I use Evans for referrences concerning to Regularity theorems.) But as far as I know $H^1_0\cap H^2$ is a strictly bigger class of functions then $H^2_0$.

Help is very much appreciated.


Under what (boundary) conditions is it possible to conclude a weak solution to $\Delta u = f$ which is also in $H^2_0$ of this problem?

The answer is that: no boundary conditions can do this.

Rule of thumb: If the imposed boundary condition leads to a well-posed weak formulation (solution exists and is unique), then no matter what this boundary condition is, we can not get a weak solution $u\in H^2_0$ in general.

First notice $H^2_0$ means: $$ H^2_0(U) := \{u\in H^2: \; u\big|_{\partial U} = 0, \; \nabla u\big|_{\partial U} = 0\}. $$ Also note if $u$ is a constant on boundary, then $u$ does not change tangentially along any curve on boundary, so the directional derivative of $u$ along any tangent vector of $\partial U$ is 0. The only thing we wanna deal with is the normal derivative: $$ u\big|_{\partial U} = 0 \text{ and } \nabla u\big|_{\partial U} = 0 \implies u\big|_{\partial U} = 0\text{ and } \frac{\partial u}{\partial n}\Big|_{\partial U} = 0 $$

Now if $u$ is a weak solution of the Poisson equation, and $u\in H^2_0$, this means $$\left\{ \begin{aligned} -\Delta u &= f \quad \text{ in } U, \\ u &= 0 \quad \text{ on } \partial U, \\ \partial u/\partial n &= 0 \quad \text{ on } \partial U, \end{aligned}\right.\tag{$\star$} $$ for some $f$.

However, $(\star)$ is not a well-posed problem. Looking for a function $u\in H^2_0$ is like imposing Dirichlet and Neumann boundary conditions at the same time (a Cauchy boundary problem), which can be proved ill-posed in several ways.

A simple argument can be: if neglecting Neumann boundary at first, Poisson equation together with the Dirichlet boundary has a unique solution for any $f\in L^2(U)$. This means $\nabla u$ can be evaluated a.e. in $U$, and if $\partial u/\partial n|_{\partial U}$ is defined as a limit from the interior to the boundary in the direction of the normal , then $\partial u/\partial n$ is already determined in this case. In general we can't expect $\partial u/\partial n = 0$ coincides with what you obtained from the solution of Dirichlet boundary problem.

A counterexample would be: suppose we have a classical solution which coincides with a strong and weak solution to the Dirichlet boundary problem except a measure zero set, and $f\leq 0$. By maximum principle the maximum is achieved on boundary, $u|_{\partial U}=0> u(x)$ for $x\in U$, then $\partial u/\partial n$ must be strictly greater than 0 pointwisely (see Gilbarg and Trudinger Page 32).

Another argument would be if $\partial u/\partial n|_{\partial U} = 0$, by divergence theorem: $$ \int_U f\,dx = \int_U \mathrm{div}(\nabla u)\,dx = \int_{\partial U} \frac{\partial u}{\partial n} \,dS = 0, $$ which means $f$ has average 0 in $U$. Hence this says that, if $f$ does not have zero average, the $(\star)$ does not have a solution.

Some comment about your terminology, "regularization" doesn't mean what you have written in your question, what you applied is just elliptic regularity.


I think that you can't solve this equation in $H^2_0$. If a function $u$ is in $H^2_0$ then you have $u=0$ and $\nabla u=0$ at the boundary and these are too much boundary conditions for this equation. Let me think how one can prove the nonexistence.

  • $\begingroup$ It's the same thing that I was fearing. Restricting yourself to $H^2_0$ probably over determines the system. My question is motivated by the following proAblem: Consider two functions $u$ and $v$ in $L^2$, such that for every $\phi\in H^2_0$ it holds that $\int_U u\Delta \phi=\int_U v\Delta \phi$. Can we conclude that $u$ is equal to $v$ almost everywhere? $\endgroup$ – Aris May 20 '13 at 18:17
  • $\begingroup$ Notice that with these boundary conditions the laplacian is self-adjoint. Assume that $u-v$ is smooth. Then you obtain that $u-v$ is harmonic. In fact, u+1 verify that $\int_U (u+1)\Delta \phi=\int_U u\Delta \phi+\int_U\Delta \phi.$ Now you can apply the divergence theorem and you obtain $\int_U\Delta \phi$. Thus, if $U$ is bounded any constant function is in $L2$. Thus, the claim is false, right? $\endgroup$ – guacho May 20 '13 at 20:23

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.