1

I am working on the following problem:

For $\Omega\in \mathbb{R}^N$ an open set and $K\subset\subset\Omega$ a compact subset, define the difference quotient in $j$ direction by $$D^h_ju(x)=\frac{u(x+he_j)-u(x)}{h}$$ prove that there exists a constant $C>0$ such that for every $u\in H^1(\Omega)$ and every small $|h|$, we have $$\|D^h_ju\|_{L^2(K)}\le C\|\nabla u\|_{L^2(\Omega)}$$

I imitate the procedure in proving Poincaré inequality that $$\frac{1}{h}(u(x+he_j)-u(x))=\frac{1}{h}\int_{x_j}^{x_j+h}\partial_ju(x_1,...,x_j',...,x_N){\rm d}x_j'$$ squre both side, integrate and employ the Cauchy-Schwarz inequality to get $$\|D_j^hu\|^2_{L^2(K)}\le\int_\Omega(\frac{1}{h}\int_{x_j}^{x_j+h}\partial_ju(x_1,...,x_j',...,x_N){\rm d}x_j')^2{\rm d}x$$ $$\le\frac{1}{h}\int_\Omega(\int_{x_j}^{x_j+h}|\partial_ju|^2{\rm d}x_j'){\rm d}x$$ However, I get stuck here since though $|\partial_ju|$ is square integrable, $\frac{1}{h}\int_{x_j}^{x_j+h}|\partial_ju|^2{\rm d}x_j'$ may not be bounded. I guess there must be some property of $u\in H^1(\Omega)$ that I haven't used. By the way, I am not familiar with the Sobolev space. Literally I just know the definition of it. I know there are already tons of developed inequalities in it which might be helpful, but I want to avoid them unless there is no other way.

Also there is another problem following this one that if $u\in L^2(\Omega)$ and there is a constant $C>0$ such that $\|D^h_ju\|_{L^2(K)}\le C$ for every small $|h|$, then $u\in H^1(K)$ and $\|\nabla u\|_{L^2(K)}\le C$.

This feels like a converse proposition to the one above. I thought about Dominated Convergence Theorem and the alike but didn't make much progress.

1 Answers1

1

Note that the integration domain for $x$ is $K$ not $\Omega$. After decoupling the integral regions, use Fubini to reverse the order of integration $$ \frac1h\int_K \int_{x_j}^{x_j+h} |\partial_j(x_j', x_i|_{i\ne j})|^2 dx_j' dx = \frac1h\int_K \int_0^h |\partial_j u(x+se_j)|^2 ds\, dx\\ = \frac1h\int_0^h \int_K |\partial_j u(x+se_j)|^2 dx\, ds\\ \le \frac1h\int_0^h \int_\Omega |\partial_j u(x)|^2 dx\, ds =\int_\Omega |\partial_j u(x)|^2 dx \\ $$

daw
  • 49,113
  • 2
  • 38
  • 76
  • Thank you so much! I indeed thought about changing the order of integration, but I didn't notice that the integral regions could be decoupled in this way, which makes it so simple to employ Fubini. By the way, do you have any ideas regarding the second quesiton? – Apocalypse Jul 03 '19 at 15:13