For a function u in the Sobolev space $W_0^{1,p} (\mathcal O )$, ($p \in [ 1, n ]$), having $u > 0$ inside $\mathcal O$, where $\mathcal O$ is an open bounded connected set in $\mathbb R^n$, can one always find a precise reprentative of $u$ which is lower semicontinuous ?
1 Answers
For $n = 1$, this is obviously true.
I don't think it holds for $n > 1$. Let me sketch an example for $n = 2$. W.l.o.g. we assume that the unit square belongs to $O$. Let $\tilde F \subset \mathbb{R}$ be the fat cantor set. Then, $F = \tilde F \times (0,1)$ has positive measure. The set $U = (0,1)^2 \setminus F$ is open and dense in $[0,1]^2$, since $[0,1] \setminus \tilde F$ is dense in $[0,1]$. Hence, we find a countable set of points $\{x_n\} \subset U$, which are dense in $[0,1]^2$.
Since a point as capacity $0$, we can construct functions $f_n \in H_0^1(O) \cap C(\bar O)$ with
- $f_n \equiv 1$ on a small ball around $x_n$
- $0 \le f_n \le 1$ on $O$
- $\mathrm{supp} (f_n)$ lies in $U$
- $\| f_n \|_{H_0^1} \le 2^{-n}$
Then, for $f = \sum_{n=1}^\infty f_n$, we have
- $f \in H_0^1(O)$
- $f \ge 0$ on $O$
- $f \ge 1$ on a small ball around $x_n$
- $\mathrm{supp} (f)$ lies in $U$
Then, for any representative $\tilde f$ of $f$, and any $x \in [0,1]^2$, we find a sequence $\{y_n\}$, $y_n \to x$ and $\tilde f(y_n) \ge 1$. But $\tilde f$ is zero on $F$ a.e. This is a contradiction.
Edit: ups, this is actually lower semicontinuous. But you can consider something like $g - \min(f,1)$, for $g \in H_0^1(O) \cap C(\bar O)$ large enough.
- 31,359
-
Thanks a lot ! Please tell me: when you write Omega, you mean O ? – Antonio Kubiko Oct 24 '15 at 10:55
-
Yes, I am used to have $\Omega$ as the domain. I fixed the typo. – gerw Oct 24 '15 at 10:56
-
I need an hypothesis (e.g. to impose on the function f or on its level sets or on the boundary of its superlevel sets) so that a counterexample like yours is impossible. Would you care to suggest something ? For example, maybe ask that the set of lower semidiscontinuity points of f on each level set have zero (n-1)-dim Hausdorff measure ? (So that I could modify f there and turn f lsc.) I would prefer some hypothesis just on the boundary of the set where f > 0, if possible. – Antonio Kubiko Oct 24 '15 at 14:13
-
Why do you need lsc? Maybe it is sufficient, that $f$ is quasi-continuous? – gerw Oct 24 '15 at 17:15
-
But are Sobolev functions always quasi-continuous ? Can you give some reference on papers or books proving this ? – Antonio Kubiko Oct 25 '15 at 01:33
-
Some suggestions could be found here: http://math.stackexchange.com/questions/48776/capacity-theory-beginner-resources – gerw Oct 27 '15 at 14:27
-
Thanks a lot for your help ! – Antonio Kubiko Oct 28 '15 at 11:20
-
If you found the answer useful, you can vote for it (up arrow) and accept it (check mark). This is the concept of this site. – gerw Oct 28 '15 at 12:03