We have the following PDE in $U$ an open bounded subset of $\mathbb{R}^2$: $$\begin{align} -\Delta u(x,y) + u(x,y) & =f(x,y) & & (x,y) \in U \\[8pt] u(x,y)& = g & & (x,y) \in \partial U \end{align}$$

We suppose also for simplicity that $g \in C^1(\partial U)$

The objective is to show uniqueness of solution for the weak formulation in $H^1(U)$.

Brezis' Functional Analysis, Sobolev Spaces and Partial Differential Equations [Proposition 9.22] asserts a weak solution for this problem is a function $u \in K$ satisfying $$ \quad \quad \quad \quad \quad \quad \int_{U} \nabla u \nabla v + \int_{U}uv= \int_{U}fv \quad \forall v \in H_0^1(U), \quad \quad \quad $$ where $K$ is defined as the convex and closed subset of $H^1(U)$ $$K= \{v \in H^1(U) \ | \ v-\tilde{g} \in H_0^1(U) \},$$ where $\tilde{g}$ verifies $\tilde{g}=g$ in $\partial U$.

Then he shows that indeed there is a unique $u_0 \in K$ that verifies the weak formulation given above.

Now, Does this imply that the solution is unique in $H^1(U)$, because showing uniqueness in this subset does seem to imply uniqueness in the whole space, arguing something that any function that is outside $K$ does not verify the boundary condition, although I need to formalize this.


By definition, the set $$ K = \{ v \in H^1(\Omega) : v - \widetilde g \in H_0^1(\Omega) \}.$$ is precisely the set of functions in $H^1(\Omega)$ satisfying the boundary condition $v|_{\partial \Omega} = g$, and the purpose of the condition $v - \widetilde g \in H_0^1(\Omega)$ is to impose this boundary condition. But as you say, we need to think carefully about how we formalise the boundary condition $v|_{\partial \Omega} = g$ for $v \in H^1(\Omega)$: since a general element $v$ of $H^1(\Omega)$ is really an equivalence class of functions that are equal almost everywhere, it makes no sense to simply "evaluate $v$ on $\partial \Omega$".

For $v \in H^1(\Omega)$, the boundary condition $v = g$ on $\partial \Omega$ is to be understood in terms of the trace map. In Evan's PDEs book (which I'm more familiar with), it is explained that the trace map is a bounded linear map $T : H^1(\Omega) \to L^1(\partial \Omega)$ such that $Tu = u|_{\partial \Omega}$ if $u \in H^1(\Omega) \cap C(\bar \Omega)$. A weak solution obeying the boundary condition $v|_{\partial \Omega} = g$ is then defined to be any weak solution $v \in H^1(\Omega)$ such that $Tv = g$. This definition agrees with the usual definition when $v \in H^1(\Omega) \cap C(\bar \Omega)$, but this definition continues to make sense when $v$ is not in $C(\bar \Omega)$.

Finally, there is a general theorem that says that, for all $u \in H^1(\Omega)$, $Tu = 0$ iff $u \in H_0^1(\Omega)$. (This is Theorem 5.5.2 in Evans.) So the condition that $Tv = g$ is equivalent to the condition that $v - \widetilde g \in H_0^1(\Omega)$. And that is the motivation for the definition of the set $K$.

Edit: If you wish to avoid traces altogether (because Brezis avoids traces - see comment below), then take a look at Brezis Theorem 9.17. It says that, when $v \in H^1(\Omega) \cap C(\bar \Omega)$, the condition that $v|_{\partial \Omega} = g$ is equivalent to the condition that $v - \widetilde g \in H_0^1(\Omega)$. But of course, the statement $v- \widetilde g \in H_0^1(\Omega)$ continues to be meaningful even when $v$ is not in $C(\bar \Omega)$, which is why we prefer it.

  • $\begingroup$ You are right, $g$ is only defined in $\partial U$. Is this the only thing that I misquoted? $\endgroup$ – D1X May 30 '17 at 10:57
  • $\begingroup$ On the other hand, I am avoiding the use of traces because Brezis does, hoping there is a way to formalize it without using them. When you say $K$ is precisely the set of functions in $H^1(\Omega)$ satisfying the boundary condition v=g on $\partial \Omega$, If I am correct it is the set of functions whose traces verify it? $\endgroup$ – D1X May 30 '17 at 11:00
  • 1
    $\begingroup$ @D1X Yes, you're right. When people say that $v \in H^1(\Omega)$ obeys $v|_{\partial \Omega} = g$, what they really mean is that the trace $Tv$ is equal to $g$. That is just the definition! And of course, if $v$ is in $C(\Omega)$, then this definition reduces to the statement that $v|_{\partial \Omega} = g$. $\endgroup$ – Kenny Wong May 30 '17 at 11:17

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.