2

We know that if $\Omega\subseteq\mathbb R^d$ is open, $k\in\mathbb N$ and $p\in(1,\infty)$, then for every bounded $(u_n)_{n\in\mathbb N}\subseteq W^{k,\:p}(\Omega)$, there is a subsequence $\left(u_{n_k}\right)_{k\in\mathbb N}$ and some $u\in W^{k,\:p}(\Omega)$ s.t. $u_{n_k}\xrightarrow{k\to\infty}u$ weakly in $W^{k,\:p}(\Omega)$.

Does the same hold true with $W^{k,\:p}(\Omega)$ replaced by $W_0^{k,\:p}(\Omega)$? Moreover, if we know that $(v_n)_{n\in\mathbb N}\subseteq W_0^{k,\:p}(\Omega)$ is (uniformly) norm-bounded and almost everywhere pointwisely convergent to some $v:\Omega\to\mathbb R$, are we able to conclude that $v_n\xrightarrow{n\to\infty}v$ weakly in $W_0^{k,\:p}(\Omega)$ from the aforementioned fact?

0xbadf00d
  • 14,208

1 Answers1

2

Since $W^{k,p}_0$ is a closed subspace of $W^{k,p}$, all these results carry over to $W^{k,p}_0$.

If $v_n \rightharpoonup v$ in $W^{k,p}$ and $v_n \to u$ pointwise a.e., then $u=v$.

As the result ``weak equals pointwise limit'' is hard to track down, here is a sketch of a proof:

Let $(\Omega,\mathcal A,\mu)$ be a $\sigma$-finite measure space. Assume $v_n \rightharpoonup v$ in $L^p(\Omega)$, $1<p<\infty$, and $v_n \to u$ pointwise $\mu$-a.e. on $\Omega$. Then $v=u$ $\mu$-a.e.

Proof. By Fatou Lemma, $u\in L^p(\Omega)$. Let $B\subset \Omega$ be bounded. Let $\epsilon>0$. By Egorov's theorem, there is a subset $A\subset B$ such that $v_n \to u$ uniformly on $A$ and $\mu(B\setminus A)<\epsilon$. Let $\phi\in L^\infty(\mu)$. Then $\chi_Bv\in L^q(\mu)$ with $\frac1p+\frac1q=1$ and $1<q<\infty$. Then $$ |\int_B(v_n-u)\phi d \mu | = |\int_{A \cup (B\setminus A)} ...| \le \mu(A) \|v_n-u\|_{L^\infty(A)} \|\phi\|_{L^\infty} + \epsilon^{1/q} \|v_n-u\|_{L^p(B)} \|\phi\|_{L^\infty} $$ which implies $$ \limsup |\int_B(v_n-u)\phi d \mu| \le \epsilon^{1/q} 2M\|\phi\|_{L^\infty} $$ where $M$ is such that $\|v_n\|_{L^p}, \|u\|_{L^p}\le M$. Since $\epsilon>0$ was arbitrary, this proves $\int_B(v_n-u)\phi d \mu \to0$. Since $L^\infty(B)$ is dense in $L^q(B)$, this proves $v_n \rightharpoonup u$ in $L^p(B)$. This proves $u=v$ $\mu$-a.e. on $B$. Since $\Omega$ is a countable union of sets of bounded measure, $u=v$ $\mu$-a.e. on $\Omega$ follows. $\square$

I believe, this result is also true for $p=1$ (and $q=\infty)$. Then one has to use the Dunford theorem (see Diestel & Uhl), which is a characterization of weakly convergent sequences in $L^1$.

A proof by Banach-Saks property can be found here:

Weak and pointwise convergence in a $L^2$ space

daw
  • 54,637
  • 2
  • 44
  • 85
  • Thank you for your answer. Note that the answers in your link only treat the Hilbert space case $p=2$. If $\Omega$ would be bounded, we could conclude from $v_n\to v$ a.e. that $\varphi v_n\to\varphi v$ in $L^p(\Omega)$ for all $\varphi\in C_c^\infty(\Omega)$. So, since we already know that $v_{n_i}\to w$ weakly in $W^{k,:p}(\Omega)$ (hence weakly in $L^p(\Omega)$) for some $w\in W^{k,:p}(\Omega)$, it must hold $\int_\Omega\varphi v_{n_i}\to\int_\Omega\varphi w$ and hence $\int_\Omega\varphi(v-w)=0$ for all $\varphi\in C_c^\infty(\Omega)$. So, by the du Bois-Reymond lemma, $v=w$ a.e. – 0xbadf00d May 25 '21 at 07:05
  • But how do we need to argue without the boundedness assumption on $\Omega$? – 0xbadf00d May 25 '21 at 07:05
  • $p=2$ does not matter. Boundedness of $\Omega$ is not needed. The argument in your comment does not work: pointwise convergence does not imply $v_n \phi\to v\phi$ nor convergence of the integrals. Please check the answer in the linked question. An alternative uses Egorov theorem. – daw May 25 '21 at 07:18
  • If $\Omega$ is bounded, then (since $\varphi$ has compact support) $\operatorname{supp}(v_n\varphi)\subseteq K$ for all $n$ for some compact $K$ and hence $v_n\varphi\to v\varphi$ by the dominated convergence theorem. – 0xbadf00d May 25 '21 at 07:20
  • The answer in the linked question is talking about a Hilbert space. – 0xbadf00d May 25 '21 at 07:21
  • For dominated convergence, you need a dominating function ... – daw May 25 '21 at 07:23
  • see also https://de.wikipedia.org/wiki/Banach-Saks-Eigenschaft for $p\ne 2$ – daw May 25 '21 at 07:28
  • In the Wikipedia article, it is only claimed that this theorem holds for $L^p([0,1])$. Does it hold for $L^p(\Omega)$ as well? And if so, do you've got a reference? – 0xbadf00d May 25 '21 at 08:14
  • $L^p(\Omega)$ is isomorphic to $L^p([0,1]$. In any case, here (http://math.fau.edu/schonbek/realan/ra1sp16h3.pdf) the Banach-Saks property of $L^p$ is proven for arbitrary measure spaces. – MaoWao May 25 '21 at 09:30
  • I think your proof can be simplified since we already have the weak convergence. After your application of Egorov, you have $v_n \to u$ in $L^\infty(A)$ and $v_n \rightharpoonup v$ in $L^p(A)$. Hence, $v = u$ a.e. on $A$. Thus, $v = u$ a.e., since $\varepsilon > 0$ and the bounded set $B$ were arbitrary. – gerw May 25 '21 at 12:38
  • @gerw yes, you are right. – daw May 25 '21 at 12:47