The question is from Gilbarg and Trudinger Elliptic PDE of second order. Here's the statement:

Let $\Omega$ be a $C^1$ domain, prove the following interpolation inequality: \begin{eqnarray} \|D^{\alpha}u\|_{L^p(\Omega)} \le \epsilon\|u\|_{W^{k,p}(\Omega)} + C_\epsilon\|u\|_{L^p(\Omega)} \end{eqnarray} where $0<|\alpha|<k$, $1 \le p < \infty$, and $C_{\epsilon}$ is a constant independent of u.

Any help, hints, or approaches will be appreciated.

Also, could you explain to me what exactly interpolation means here? Interpolation in what sense? What I mean is that say, for example, the interpolation in $L^p$ and $L^q$, WLOG, say $p < q$. If certain thing holds for p and q, then interpolation would mean that it holds for all $p < x < q$. So what exactly interpolation means for this question?


Interpolation means here: you estimate the norm of some intermediate derivative by the $k$-th and the zeroth derivative, which is in some sense an interpolation between the orders of differentiation.

The correct statement should be: For every $\epsilon>0$ there is $C_\epsilon>0$ such that for all $u$ ...

Now, assume the claim is not valid. Then there is $\epsilon>0$ and a sequence $(u_n)$ such that $$ \|D^\alpha u_n \|_{L^p} \ge \epsilon \|u_n\|_{W^{k,p}} + n \|u_n\|_{L^p}. $$ We can rescale $(u_n)$ such that $\|u_n\|_{W^{k,p}}=1$. By compact embeddings (here we need boundedness of $\Omega$), $u_n$ converges to some $u$ in $W^{k-1,p}(\Omega)$ (after extracting a subsequence).

Dividing the above inequality by $n$ and passing to the limit yields $u=0$. Passing to the limit in $$ \|D^\alpha u_n \|_{L^p} \ge \epsilon \|u_n\|_{W^{k,p}} + n \|u_n\|_{L^p} \ge \epsilon $$ yields the final 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.