I am looking for a counterexample of Sobolev Embedding Theorem, i.e. I am seeking for a sobolev function $u\in W_0^{1,p}(\Omega),\,p\in[1,n)$, $\Omega$ is a bounded domain in $\mathbb{R}^n$ such that the inequality: $||u||_{L^q}\leq C||\nabla u||_{L^p}$ does not hold, where $q>p^*:=\frac{np}{n-p}$. A similar question was asked in the past, but with $\mathbb{R}^n$ as a domain and the space used was $W^{1,p}(\Omega)$, see link

  • $\begingroup$ You can use the exact same example in that question (suppose $0\in \Omega$), just use a mollifier to cut if off from $\Omega$, the singularity is near $0$. $\endgroup$ – Shuhao Cao Mar 2 '16 at 23:00

You don't need any special function to reach this conclusion.

Any domain contains a ball; so it's enough to consider a ball, which may as well be centered at $0$. Take any function $f$ with nonzero $L^q$ norm supported in this ball. Consider the sequence $f_k(x)=f(kx)$ and use the change of variables $y=kx$ to show that $$ \|f_k\|_{L^q} = k^{-n/q} \|f\|_{L^q},\quad \|\nabla f_k\|_{L^p} = k^{1-n/p} \|f\|_{L^p} $$ The conclusion follows since $-\frac{n}{q} > 1-\frac{n}{p}$.

  • $\begingroup$ Thanks, @Sally! This really helps! Shouldn't it be $\|f_k\|_{L^q} = k^{-n/q} \|f\|_{L^q},\quad \|\nabla f_k\|_{L^p} = k^{1-n/p} \|f\|_{L^p}$ (And the conclusion follows since $-\frac{n}{q} >1 -\frac{n}{p}$)? $\endgroup$ – user315279 Mar 6 '16 at 10:04
  • $\begingroup$ Yes, I corrected this. $\endgroup$ – user147263 Mar 6 '16 at 16:11
  • $\begingroup$ Thanks again, @Sally. One more question - why is it necessary to work in a ball? Can't I just take a function $f\in\Omega$ with the same procedure? $\endgroup$ – user315279 Mar 7 '16 at 8:00
  • $\begingroup$ If $\Omega$ is not star-shaped about $0$, the support of the function $f(kx)$ might fall outside of the domain. $\endgroup$ – user147263 Mar 7 '16 at 12:03

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.