I'm currently working on this Dirichlet problem:

\begin{cases} div(\sigma |\nabla u|^{p-2} \nabla u) = f &\quad {in }~ \Omega\\ u = g &\quad in~\partial\Omega \end{cases}

with $\sigma \in L^{\infty}_{+}(\Omega)$, boundary $g \in W^{1,p}(\Omega)$ and $f \in L^2(\Omega)$.

I think I've already proved that this problem has a unique solution, but I don't know how to prove the following a priori estimate

\begin{align} ||{u}||_{W^{1,p}} \leq C ({||g||}_{W^{1,p}} + {||f||}_{L^2}^{\frac{1}{p-1}}) \end{align}

This estimate is mentioned in this paper https://pdfs.semanticscholar.org/7399/da07c625d51aa7ee72840789916b036019d2.pdf (second page; equation (1.4)) but without giving proof. I guess it's seen "easy", but I just don't know how to get there. I know how to conclude an a priori estimate in the case $f=0$ or $g=0$, but since p-Laplace operator isn't linear, this doesn't help and I'm really clueless know. I hope someone can help me with with.

As far as i thought i can use the Poincaré inequality since $u - g \in W^{1,p}_0$:

\begin{align} ||u||_{W^{1,p}} &= ||u||_{L^p} + ||\nabla u||_{L^p}\\ & \leq ||g||_{L^p} + ||u - g||_{L^p} + ||\nabla u||_{L^p}\\ & \leq ||g||_{L^p} + C||\nabla(u - g)||_{L^p} + ||\nabla u||_{L^p} \end{align}

The problem is that i can't find an upper bound for $||\nabla u||_{L^p}$ using $f$...

The inequality in the paper that i found implies that i should found something like

\begin{align} ||\nabla u||^p_{L^p} \leq ||f||_{L^2}~||\nabla u||_{L^p} \end{align}

which only be valid, when the solution u is in $W^{1,p}_0$...

I hope someone can help me with with this...

Edit: Using fedjas advice i got

\begin{align} \int_{\Omega} f~(u - g) &= \int_{\Omega} div(\sigma |\nabla u|^{p-2} \nabla u) (u - g)\\ &= \int_{\Omega} \sigma |\nabla u|^{p-2} \nabla u \nabla (u - g)\\ &\leq ||f||_{L^2} ||u - g||_{L^p} \\ &\leq ||f||_{L^2} ~C ||\nabla (u - g)||_{L^p} \end{align}

Now if i would have $||\nabla (u - g)||_{L^p}^p \leq \int_{\Omega} f~(u - g)$ it would be good... BUT I dont know how to get there...

$%\int_{\Omega} f~(u - g) \leq ||\nabla (u - g)||_{L^p}^p$

because the mapping $ x \mapsto |x|^p$ is convex we have always

$|x|^p \geq |y|^p + p |y|^{p-2}y (x-y)$ (characterization with first derivation)

and with $x = \nabla (u - g)$ and $y = \nabla u$ it follows

$|\nabla (u - g)|^p \geq |\nabla u|^p - p|\nabla u|^{p-2} \nabla u \nabla g$

don't know if it's useful here, but its valid for $p \geq 1$.

Furthermore i know that

$\frac{1}{p} (\int_{\Omega} |\nabla u|^p - |\nabla g|^p) \leq \int_{\Omega} f~(u - g)$

Because the solution u is the minimizer of the functional: $J(v) = \frac{1}{p} \int_{\Omega} |\nabla v|^2 - \int_{\Omega} fv$ for all $v \in W^{1,p}_g$ (space with needed boundary values) and therefore $J(u)\leq J(g)$

Nevertheless i still don't see how to put all of this together to achieve the desired result... :/

  • $\begingroup$ Play $u-g\in W_0^{1,p}$ against $f$, use the equation, and integrate by parts. I hope $p$ is good enough for the Sobolev embedding $W^{1,p}\subset L^2$. $\endgroup$
    – fedja
    Dec 11 '17 at 5:49
  • $\begingroup$ Thanks for comment! :) I tried to follow your advice, but i am not sure if i 100% understood it (espacally the sobolev embedding part?). I edited my post and tried to do what you meant, but as you can see I still dont know how to get there $\endgroup$
    – ANZM91
    Dec 11 '17 at 7:50

It looks like you are totally new to such estimates, so let me show you the old $\varepsilon$ vs $C_\varepsilon$ trick. If you need to estimate some product $xy$, you can write $|xy|\le |x|^2+|y|^2$, but sometimes you would prefer a smaller constant on $x$ (that you would later subtract from the other side) and do not really care how big one you get on $y$. In this case you can write $$ |xy|=|(\varepsilon^{1/2} x)(\varepsilon^{-1/2}y)|\le \varepsilon |x|^2+\varepsilon^{-1}|y|^2=\varepsilon |x|^2+C_\varepsilon |y|^2 $$ You can do it with any positive powers, so you can equally well write $$ |x|^a|y|^b\le\varepsilon|x|^{a+b}+C_\varepsilon |y|^{a+b} $$ (of course, $C_\varepsilon$ here depends on $a,b$ as well).

Now, you have your identity $$ \int_\Omega f(u-g)=\int_\Omega\sigma|\nabla u|^{p-2}\langle\nabla u,\nabla u-\nabla g\rangle $$ Rewrite it as $$ \int_\Omega\sigma|\nabla u|^{p}=\int_\Omega\sigma|\nabla u|^{p-2}\langle\nabla u,\nabla g\rangle+\int_\Omega f(u-g)\le \int_\Omega\sigma|\nabla u|^{p-1}|\nabla g|+\int_\Omega |f||u-g| $$ We certainly need $\sigma\ge \sigma_0>0$ in $\Omega$ (otherwise you can easily blow up everything). So, we have $$ \sigma_0\int_\Omega|\nabla u|^{p}\le \int_\Omega\sigma|\nabla u|^{p-1}|\nabla g|+\int_\Omega |f||u-g| $$ Now play $\varepsilon$ vs $C\varepsilon$ and recall that $\sigma\le\sigma_1<+\infty$: $$ \sigma|\nabla u|^{p-1}|\nabla g|\le \sigma_1(\varepsilon|\nabla u|^{p}+C_\varepsilon|\nabla g|^p)\,. $$ Assuming that $W^{1,p}\subset L^2$, we also have $$ \int_\Omega |f||u-g|\le \|f\|_2\|u-g\|_2\le \|f\|_2\|u-g\|_{1,p}=(\|f\|^{1/(p-1)}_2)^{p-1}\|u-g\|_{1,p} \\ \le \varepsilon \|u-g\|_{1,p}^p+C_\varepsilon \|f\|_2^{p/(p-1)}. $$ Recalling the bound $$ \|u-g\|^p_{1,p}\le C[\|\nabla u|^p_p+\|g\|_{1,p}^p]\,, $$ we get $$ \sigma_0\int_\Omega|\nabla u|^{p}\le \sigma_1[\varepsilon\int_{\Omega}|\nabla u|^{p}+C_\varepsilon\int_{\Omega}|\nabla g|^p]+\varepsilon C[\int_{\Omega}|\nabla u|^{p}+\|g\|_{1,p}^p]+ CC_\varepsilon \|f\|_2^{p/(p-1)} $$ Now choose $\varepsilon$ so small that after moving all the terms with $\int_\Omega|\nabla u|^p$ from the right to the left, you still have some positive coefficient on the LHS.

That's all. I presented it with all details and spelled out what exactly is needed for what step since it looks like you are seeing this standard mumbo-jumbo for the first time. Try to to it yourself next time.

  • $\begingroup$ Wow! Thank you very very much! Yes, I am new too such estimates and this functional analysis stuff, so I am very gratefull for your detailed answer. But I still have one question regarding your mentioned bound $||u - g||^p_{1,p} \leq C [|| \nabla u||^p + ||g||^p_{1,p}]$... Can you explain why it is true? Because in my eyes it's not - at least not for arbitrary u and g? :/ I only have $||u - g||^p_{1,p} \leq C ||\nabla u - \nabla g||^p_p \leq C ( ||\nabla u||_p + || \nabla g ||_p)^p$ $\endgroup$
    – ANZM91
    Dec 12 '17 at 13:29
  • $\begingroup$ @ANZM91 What do you mean? $\|g\|_{1.p}^p\approx \|g\|_p^p+\|\nabla g\|_p^p$, so my RHS is larger than yours (and $\|g\|_p$ matters, by the way: think of large constants). $\endgroup$
    – fedja
    Dec 12 '17 at 14:25
  • $\begingroup$ Why is it larger? I have $(||\nabla u||_p + ||g||_p + ||\nabla g||_p ) ^p$ - the p-th power of the sum. Or is $| x - y| ^p \leq |x|^p + |y|^p $ always true? Sorry, but I am a bit confused at the moment >.< $\endgroup$
    – ANZM91
    Dec 12 '17 at 14:49
  • $\begingroup$ @ANZM91 Analysis is not algebra. If $a,b,p>0$, then we have $(a+b)^p=a^p+b^p$ for all practical purposes. My RHS is larger because it also includes $\|g\|_p^p$, which you lost somewhere (and it matters!). $\endgroup$
    – fedja
    Dec 12 '17 at 16:50
  • $\begingroup$ Ahhhh, maybe you did mean this way? \begin{align*} ||u - g||^p_{1,p} &\leq C_p || \nabla u - \nabla g ||^p_{p} \\ &\leq C_p(||\nabla u||_p + ||\nabla g||_p)^p \\ &= C_p \sum_{i=0}^{p} \binom{p}{i} ||\nabla u||_p^{p - i}||\nabla g||_p^i \\ &\leq C_p \binom{p}{p/2} (||\nabla u||^p_p + ||\nabla g||^p_p) \\ &\leq C (||\nabla u||^p_p + ||\nabla g||^p_p + ||\nabla g||) \\ &= \leq C (||\nabla u||^p_p + ||\nabla g||^p_{1,p}) \end{align*} By the way: In my case I have $p\geq 1$ so $W^{1,p} \subset L^2$ isnt working, is it? What do i have to change then? $\endgroup$
    – ANZM91
    Dec 12 '17 at 16:53

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.