2
$\begingroup$

This is a variant of the question Compact Embedding of $W^{1,2}(0,T;ℝ^d)$ in $C(0,T;ℝ^d)$ where we had $X=\mathbb{R}^d$. Let now $X$ be some Banach space.

Question: Is $H^1(0,T;X)$ compactly embedded in $C([0,T];X)$?

We define $H^1(0,T;X)=\{u : (0,T) \to X ~|~ u \in L^2(0,T;X), u' \in L^2(0,T;X)\}$.

A proof for the continuous embedding can be found in a lot of books, e.g. Roubicek's book. Also, it is standard that $H^1(0,T,\mathbb{R})$ is compactly embedded in $C([0,T];\mathbb{R})$, e.g. Brezi's book. The main difficulty is that the classical Arzela-Ascoli theorem is only valid for handling something like $C(S;\mathbb{K}^d)$ and not for $C(S;Y)$ where $Y$ is some infinite-dimensional Banach space. But for this scenario there is the following version, which can be found in the book of Boyer.

Arzela-Ascoli: Let $E$ be a compact metric space and $F$ a metric space. Let $\mathcal{F}$ be a subset of $C(E,F)$. If $\{f(x) : f \in \mathcal{F} \}$ is relatively compact in $F$ for any $x\in E$ and $\mathcal{F}$ is equicontinuous, then $\mathcal{F}$ is relatively compact in $C(E,F)$.

I am translating the proof of Brezis to our setting.

Proof: Let $u \in B:=\{v \in H^1(0,T;X) : \|v\|_{L^2(0,T;X)}+\|v'\|_{L^2(0,T;X)} \leq 1\}$. In particular $u \in C([0,T];X)$. Then we have for all $t,s \in [0,T]$ $$\|u(t)-u(s)\|_X \leq \int_s^t \|u'(\tau)\|_X \, \text{d}\tau \leq \|u'\|_{L^2(0,T;X)} |t-s|^{1/2} \leq |t-s|^{1/2}$$ i.e. $u$ is equicontinuous. Further if we can show that $\{ u(t): u \in B\}$ is relatively compact in $X$ for any $t\in [0,T]$, then we would be finished.

Can someone help me with this last step written in bold font? I am not sure if this result is even correct. I think I have seen it in a article with $X=L^2$ or $X=H^1$, I don't remember clearly.


Before there was a picture of another Arzela-Ascoli theorem version which was written in an unclear/erroneous way. I've updated it with the version in Boyer's book which coincides with the notes provided in the comments.

$\endgroup$
  • $\begingroup$ As for the compactness subquestion: continuous images of compact intervals are compact; I guess what you mention is only an issue for non compact $S$ $\endgroup$ – Bananach Apr 5 '18 at 12:24
  • 1
    $\begingroup$ I believe his 'Warning' as written is actually incorrect, but there is a correct version of the warning: For any $x\in S$, the set $\{f(x) : f\in A\}$ must be precompact. This condition indeed follows from boundedness when $Y$ is finite-dimensional, but not when it is infinite-dimensional. See math.utk.edu/~freire/teaching/m447f16/AscoliArzelaNotes.pdf $\endgroup$ – Bananach Apr 5 '18 at 13:15
  • 1
    $\begingroup$ You may mean the correct thing but you certainly don't write it: $u(t)$ for a given $u\in H^1([0,T];X)$ is a single element of $X$ and thus certainly is precompact. You need to show that $\{u(t) : u \in B\}$ is precompact in $X$ for all $t\in [0,T]$ $\endgroup$ – Bananach Apr 5 '18 at 14:44
  • 1
    $\begingroup$ Now that we have clarified that, I don't think the statement to be proven actually holds: Let $B$ be the unit ball of $H^1([0,T];X)$, let $B_X$ be the unit ball of $X$, and let $\tilde{B}:=\{t\mapsto f : f \in B_X\}\subset B$ be the set of all constant functions in $H^1([0,T];X)$ whose constant value is in $B_X$. Then $\{u(0) : u \in B\}\supset \{u(0) : u \in \tilde{B}\} = B_X$ which is not precompact in $X$ unless $X$ is finite-dimensional $\endgroup$ – Bananach Apr 5 '18 at 15:04
  • 1
    $\begingroup$ Actually, this disproves your original statement, and doesn't only show that your planned proof doesn't work: $B$ is not pre-compact in $C([0,T];X)$ because $\tilde{B}$ isn't, because $\phi(\tilde{B})= B_X \subset X$ isn't, where $\phi$ sends elements of $C([0,T];X)$ to their values at $0$ (a continuous mapping) $\endgroup$ – Bananach Apr 5 '18 at 16:34
1
$\begingroup$

Having infinite-dimensional $X$ is problematic, because even the space of constant $X$-valued functions is not compactly embedded in $C([0, 1];X)$. Indeed, the image of its unit ball under the embedding into $C([0,1];X)$ is an isometric copy of the unit ball of $X$; not a totally bounded set.

The result you are looking for is the Aubin–Lions(-Simon) lemma where an additional compactness condition is imposed on the values taken by the functions in our space.

$\endgroup$
  • $\begingroup$ Thank you for your answer. Too bad the result is not valid. I am aware of Aubin-Lions lemma and its special case indeed has a compactness result in $C([0,T];X)$; for a Gelfand triple $X \hookrightarrow \hookrightarrow Y \hookrightarrow Z$ we have $$L^\infty(0,T;X) \cap W^{1,p}(0,T;Z) \hookrightarrow \hookrightarrow C([0,T];Y).$$ I thought there could be a trade-off, since in my case the derivative is in a better space even though I have no $L^\infty$ estimate for the function itself. But thank you for clarifying this. $\endgroup$ – Fritz Apr 5 '18 at 21:35

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.