0

Question: $\Omega$ Lipschitz bounded domain in $\mathbb{R}^n$, show that for $ q\geq 1$, $$ \|v\|_q \leq C(n)q^{1-1/n} |\Omega|^{1/q}\|v\|_{W^{1,n}} $$ for $ v\in W_0^{1,n} $.

I think it's correct, since when $n=1$, the exponent on $q$ vanish, which is consistent with the fact that $W^{1,1}\subseteq L^\infty$ in 1d.

My attempt: I tried to use potential estimate like $$ v(x)\leq C(n)\int_\Omega \frac{|Dv(y)|}{|y-x|^{n-1}}dy $$ as in the proof of Morrey inequality. I tried to use Minkovski inequality to bound the $L^q$ norm, but it failed, since the potential term $|y-x|^{-n+1}$ can't have too big exponent to be integrable with respect to $x$. I don't know what to do next.

I also tried to bound the $L^q$ norm by some $W^{1,p}$norm, $p<n$ and then use Holder. But in the first step I only get exponent 1 of $q$, not $1-1/n$ (I use Sobolev inequality in Evans PDE, where the constant is $\gamma=q(n-1)/n \cdot C(n)$).

My work on the 2nd attempt: Let $\gamma=q(n-1)/n$, and apply the inequality (in comment) to $|u|^\gamma$, \begin{align*} &\left(\int_\Omega |u|^{\gamma\frac{n}{n-1}}\,dx\right)^{(n-1)/n} \\ \leq &C(n)\int_\Omega |D(|u|^\gamma)|\,dx = C(n)\gamma \int_\Omega |u|^{\gamma-1}|Du|\,dx \\ \leq &C(n)\gamma\left(\int_\Omega |u|^{(\gamma-1)\frac{p}{p-1}}\right)^{(p-1)/p}\left(\int_\Omega |Du|^p \right)^{1/p} \end{align*} choose $p=\frac{nq}{n+q}$ and we get $$ \left(\int_\Omega |u|^q\,dx\right)^{(n-1)/n} \leq C(n)\gamma \left(\int_\Omega |u|^q\,dx\right)^{(nq-n-q)/nq} \|Du\|_p, $$ therefore $$ \|u\|_q \leq C(n)q\frac{n-1}{n} \|Du\|_p \leq C(n)q\frac{n-1}{n}|\Omega|^{1/q}\|Du\|_n, $$ where the last step use Holder. I still only get exponent 1 on $q$.

Would anyone like to help me or give me some hints? Thanks very much!

  • Last approach works. Try it again and make sure you’re using the sharp value of $p$. – Deane Sep 21 '24 at 18:10
  • Thanks @Deane, but I think I still can't get the result. The sharp value of $p=\frac{nq}{n+q}$, and I plug it in the Sobolev inequality( I refer to this link where I use its constant). I only get $q^1$ growth on $q$, instead of $q^{1-1/n}$. And after that I use Holder, I can only get the scaling factor $|\Omega|^{1/q}$, and nothing more about $q$. Would you like to give more hints about how to get a finer constant in the Sobolev inequality? Thanks! – MikeMichael_maths Sep 22 '24 at 00:52
  • Perhaps you could show your work. My suggestion is to start with the inequality $$ \left(\int_\Omega |u|^{\frac{n}{n-1}},dx\right)^{(n-1)/n} \le C(n)\int_{\Omega} |\nabla u|,dx$$ and use the Holder inequality to derive your inequality. – Deane Sep 22 '24 at 01:56
  • I just edited and showed my work in the last part of my question. Can you help me check where I inappropriately estimate the norm such that I get the wrong exponent? Thanks! – MikeMichael_maths Sep 22 '24 at 02:57
  • I confess that I misunderstood your question. Your calculation looks correct to me. Where did you see the inequality you're trying to prove? – Deane Sep 23 '24 at 03:02
  • It's from my homework of Finite Element Method, when we are studying Sobolev Inequalities. Since we always want the constants having more explicit dependence on parameters in numerical analysis, it's crucial to get a finer estimate like this. Maybe I will ask the professor for its source after the homework is due. Still thanks! – MikeMichael_maths Sep 23 '24 at 07:06
  • Please post solution when you get it. – Deane Sep 23 '24 at 13:17
  • Actually I should use the first approach, see the post. – MikeMichael_maths Oct 24 '24 at 03:52

0 Answers0