I am studying the following application which goes from $H^1(\Omega)$ to $\mathbb{R}$ with $\Omega$ a bounded regular subset of $\mathbb{R}^3$ :

$$H : u \mapsto \int_{\Omega} \left(|\nabla u|^2 - a \cdot \nabla u \right) \ \mathrm{d} x$$

where $a$ is a fixed vector in $\mathbb{R}^3$. I would like to prove that $H$ verifies the lower semi-continuity definition.

I know 3 equivalent ways of defining lower semi-continuity, denoting $X=H^1(\Omega)$ :

  1. for all $u_0 \in X$, for all $t<H(u_0)$, there exists a neighbor of $u_0$ such that :

$$\forall u \in X, H(u) \geq t$$

  1. for all real $\alpha$, the set $\{x \in X \ | \ H(u) \leq \alpha \}$ is closed in X.

  2. the epigraph $\{(u,\alpha) \in X \times \mathbb{R}, \ | \ H(u)\leq \alpha  \}$ is closed in X.

As I am very unfamiliar with this notion, I'm not sure which definition should I use to show the lower semi-continuity of $H$.

Any advices are welcomed ! Maybe we'll have to change $X$ to $H^1_0(\Omega)$ to use Poincaré inequality and change the norm.

  • $\begingroup$ What is the value of $H(u)$ if $\nabla u$ is a singular distribution? $\endgroup$ – MaoWao Jan 13 at 17:00
  • $\begingroup$ Let's take $u$ in $H^1(\Omega)$ so that $\nabla u$ is in $L^2(\Omega)$ $\endgroup$ – Velobos Jan 13 at 17:21
  • $\begingroup$ Therefore i need to change the beginning space of my functional $H$ $\endgroup$ – Velobos Jan 13 at 17:22
  • 2
    $\begingroup$ This question does not make any sense right now. The gradient $\nabla u$ does not always exist for functions $u\in L^2(\Omega)$. $\endgroup$ – supinf Jan 13 at 19:14
  • $\begingroup$ @supinf yes it is wrong but I changed the question. (see the comment above also). $\endgroup$ – Velobos Jan 13 at 20:51

Your functional is actually well-defined on $\mathrm{H}^1(\Omega)$. It is also strongly lower-semicontinuous with respect to the topology of this space. Indeed, define $f(z):= \vert z \vert^2 -a \cdot z$ for $z \in \mathbb{R}^{3}$. You then have $H(u)=\int_{\Omega}f(\nabla u)$.

In order to show that $H$ is strongly l.s.c on $\mathrm{H}^1(\Omega)$, I use the following characterization (valid on metric spaces) : $H$ is l.s.c if and only if for all $u \in \mathrm{H}^1(\Omega)$ and for all sequence $(u_n) \in \mathrm{H}^1(\Omega)^{\mathbb{N}}$ such that $u_n \rightarrow u$ in $\mathrm{H}^1(\Omega)$, we have $H(u) \leq \liminf H(u_n)$.

Thus, take $u \in \mathrm{H}^1(\Omega)$ and $(u_n) \in \mathrm{H}^1(\Omega)^{\mathbb{N}}$ such that $u_n \rightarrow u$ in $\mathrm{H}^1(\Omega)$. There is nothing to prove if $\liminf H(u_n)=+\infty$. If $\liminf H(u_n)<+\infty$, we can pick a subsequence $(u_{n_k})$ of $(u_n)$ such that $$\nabla u_{n_k} \rightarrow \nabla u \ \ a.e \ \ \text{and} \ \ H(u_{n_k}) \rightarrow \liminf H(u_n), \ \ \text{when} \ \ k \rightarrow + \infty.$$

Since for almost any $z$, we have $f(z) \geq -a \cdot z$, we can use the Fatou lemma to write $$ \liminf \int_{\Omega} (f(\nabla u_{n_k})+a \cdot \nabla u_{n_k} ) \geq \int \liminf (f(\nabla u_{n_k})+a \cdot \nabla u_{n_k})=\int_{\Omega} (f(\nabla u)+ a \cdot \nabla u),$$ because $f:\mathbb{R}^3 \rightarrow \mathbb{R}$ is continuous. Since $\nabla u_{n_k} \rightarrow \nabla u$, we also have $$\liminf \int_{\Omega} (f(\nabla u_{n_k})+a \cdot \nabla u_{n_k} ) \leq \lim \int_{\Omega} f(\nabla u_{n_k}) + \int_{\Omega} a \cdot \nabla u.$$ We end up with $$ H(u)=\int_{\Omega} f(\nabla u) \leq \lim \int_{\Omega} f(\nabla u_{n_k})=\lim H(u_{n_k})=\liminf H(u_n),$$ and this completes the proof.

Note that since $f$ is convex, the functional $H$ is convex. This implies that $H$ is also weakly lower semicontinuous on $\mathrm{H}^1(\Omega)$.


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.