I am reading Functional Analysis, Sobolev Spaces and Partial Differential Equations, by Haim brezis, and I am a bit confused about the proof. The theorem is stated as:

Let $\Omega \subset \mathbb{R}^n$ be an open set of class $C^2$ with $\partial \Omega$ bounded. Let $f \in L^2(\Omega)$ and $u \in H_0^1(\Omega)$ satisfies the weak formulation (Laplacian): $$\int_\Omega \nabla u \nabla \varphi + \int_{\Omega}u \varphi=\int_{\Omega}f\varphi \quad \forall \varphi \in H_0^1(\Omega)$$ Then $u \in H^2(\Omega)$ and $||u||_{H^2} \leq C ||f||_{L^2}$.

I don't understand the part where Brezis uses tangential translations. First, he says: Brezis301

I understand this well, we can "translate" the domain $\mathbb{R}^n_+= \{(x_1,...,x_n) \ | \ x_n>0\}$ and it is clear that $\mathbb{R}^n_+ + h = \mathbb{R}^n_+$ for $h \in \mathbb{R}^{n-1} \times \{0\}$. Then he uses a trick to show the result for all the derivatives except for $\frac{\partial \varphi}{\partial x_n x_n}$. This is also clear since $h \in \mathbb{R}^{n-1} \times \{0\}$.

enter image description here

With the domain $Q_+=\{(x_1,...,x_n) \ | \ (\sum_{i=1}^{n-1}x_i^2)^{1/2} <1 \ and \ 0 < x_n < 1 \}$ and uses $ h \in Q_0= \{(x_1,...,x_n) \ | \ x_n=0 \ and \ \sum_{i=1}^{n-1}x_i^2)^{1/2} <1\}$ and argues for $h$ "small enough".

Then he also splits the case $\frac{\partial \varphi}{\partial x_n x_n}$. Why? If you can choose $h$ "small" so that for $w \in H_0^1(Q_+)$ $w(x+h) \in H_0^1(Q_+)$, Why do you need to pick it in $Q_0$ and then split for $\frac{\partial \varphi}{\partial x_n x_n}$? Why not choose $h$ in the boundary of $Q_+$ (Wich is a bounded "cilinder") and then argue it is "small enough" ...?

Note: This is what follows:

enter image description here

As you can see, it only shows the inequality for $(k,l) \neq (n,n)$:

enter image description here

enter image description here

Update: I paste here a drawing of the situation that helped me understand the proof, thanks to Sandwich:

enter image description here

Brown is the covering of the boundary of $\Omega$.

Red is the open set we are dealing with.

Green is the support of $\theta_i $, $\Rightarrow$ the support of $\theta_i u$.

Grey is $\Omega \cap U_i$.

  • $\begingroup$ It's not easy to answer without a book, since the part you quoted is not the part you have a question about. With Google Books, I couldn't even find on what page $Q_+$ appears. $\endgroup$ – user147263 May 23 '16 at 2:22
  • $\begingroup$ @sandwich You are right, I have added the next part. The set $Q_+$ is specified bellow. In the case $\mathbb{R}^n_+$ it is clear why he picks $h \in \mathbb{R}^{n-1} \times \{0\}$, as the domain does not change. But if you can pick $h$ "small enough" so that $w \in H_0^1(Q_+)$, which is bounded, Why do you need to do it for $h \in Q_0$ (i.e. lying on the plane $\{x_n=0\}$)? $\endgroup$ – D1X May 23 '16 at 10:46

The footnote 31 on page 305 explains the logic of the choice of $h$: the support of $w$ is at some distance from the top and side surface of the cylinder $ Q_+$, but it need not be separated from the bottom surface, which is $x_N=0$. This is how the chosen partition of unity works: in order to cover the domain by finitely many sets, we need these sets to touch the boundary. I sketched the situation below, where the blue square represents $Q_+$, the green square inside of it is the support of $w$, and the other similar squares are supports of other parts of the partition of unity.


So, translating $w$ left and right keeps it within the blue square, but vertical translation is constrained. Since the argument involves both $D_h$ and $D_{-h}$, it only works for $h$ parallel to the boundary.

The author eventually gets around this by using the PDE, which relates the second $x_N$ derivative to the other second partials.

  • $\begingroup$ Why does the support of $w$ "touch" $x_n=0$? $\endgroup$ – D1X May 23 '16 at 18:45
  • 1
    $\begingroup$ Why shouldn't it? The support of $u$ does. (You know that $H_0^1$ doesn't mean compact support, right?) $\endgroup$ – user147263 May 23 '16 at 18:46
  • $\begingroup$ Wow. I think only a book can answer this question. $\endgroup$ – ibnAbu May 23 '16 at 19:16
  • $\begingroup$ @sandwich Okay, now I understand it. Partition of the Unity $\{\theta_1,..., \theta_k\}$ gives us open sets $U_1,...,U_k$ covering the Boundary (The case we are studying now). $\{\theta_1,..., \theta_k\}$ are compactly supported in $U_i$ (So is $\theta_i u$), but the support contains "a portion" of the boundary of $\Omega$. As the boundary is $C^2$, we use $H$ to transfer $\Omega \cap U_i$ to $Q_+$ and $U_i \cap \partial \Omega$ is mapped to $Q_0$, and the support of $\theta_i u$ can touch $\partial \Omega \cap U_i$, but not the other parts of the boundary of $\Omega \cap U_i$ . $\endgroup$ – D1X May 24 '16 at 12:17
  • $\begingroup$ @sandwich I have added another drawing of $\Omega$ sumarizing the situation of the support of $\theta_i u$ in $\Omega$, explaining my previous comment. Could you check it? Thank you very much. $\endgroup$ – D1X May 24 '16 at 12:21

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.