How can I show that the embedding of $W^{1,1}(a,b)$ in $C[a,b]$ is not compact?

I know that $W^{1,p}$ is compactly embedded in $C(\Omega \subset \mathbb R^n)$ for $p>n$. But since we now have $\Omega=[a,b]$ we have $n=1=p$. Why does it not work for $p=n$?


One way to show that the embedding is not compact it so find a sequence of function $(u_n)$ such that: $$\|u_n\|_{W^{1,1}} \leq C$$ $$\|u_n-u_m\|_\infty \geq 1 \text{ for } n \neq m$$ the second condition preventing the sequence to have any convergent subsequence in $C[a,b]$..

Without loss of generality let us take $[a,b]=[0,1]$. Let $\phi$ a nice function with compact support in $(0,1)$ such that $\|\phi\|_\infty=1$. Then: $$u_n(x)=\begin{cases}0 &\text{ for } 0 \leq x <\frac {1}{n+1}\\ \phi \left(n(n+1)\left(x -\frac{1}{n+1} \right)\right) &\text{ for } \frac{1}{n+1} \leq x <\frac{1}{n}\\ 0 & \text{ for } x > \frac{1}{n}\end{cases}$$ verifies:

  • $$\|u_n-u_m\|_\infty = 1 \text{ for } n \neq m$$
  • $$\|u_n\|_{L^1}= \frac{1}{n(n+1)}\|\phi\|_{L^1} \to 0$$ $$\|u_n'\|_{L^1}=\|\phi'\|_{L^1}$$ so $$\|u_n\|_{W^{1,1}} \leq C$$.

The idea behind this counter example is to use a gliding hump. But the reason why it work is because of the scaling, in $\Bbb R^n$ with $\psi_\lambda(t)=\psi(\lambda t)$:

  • $$\|\psi_\lambda \|_{L^p}=\lambda^{-\frac{n}{p}}\|\psi\|_{L^p}$$
  • $$\|\nabla \psi_\lambda \|_{L^p}=\lambda^{1-\frac{n}{p}}\|\nabla \psi\|_{L^p}$$
  • $$\| \psi_\lambda \|_{\infty}=\lambda^{0}\|\nabla \psi\|_{\infty}$$

this a way to understand why the exponents in the whole space, and in a bounded domain for compactly supported function you can take $\lambda \to + \infty$. This impose $1-\frac{n}{p} \leq 1$ but when you have equality, i.e the critical case, even if you have embedding you can find function as the previous counterexample (which does not works for $p>n$).


Here is a simpler example: On $\Omega=(-1,1)$ define $$ u_n(x) = \min(-1, \max(nx,1)). $$ Then $\|u_n\|_{L^1}\le 2$, $\|u'\|_{L^1} = 1$, $u_n \to v$ in $L^1$ with $v(x) = sign(x)$, which is not continous.

If the embedding $W^{1,1}(\Omega)\hookrightarrow C(\bar \Omega)$ would be compact, then the limit of a converging subsequence in $C(\bar \Omega)$ has to coincide with $v$, which is a contradiction.


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.