2
$\begingroup$

Could any one give a proof for this inequility here? I just know its some kind of Gagliardo-Nirenberg inequility, but where does the second term come from? Thx~ $$ \int_{B_r}|u|^q\le C\left(\int_{B_r}|\nabla u|^2\right)^a \left(\int_{B_r}|u|^2\right)^{\frac{q}{2}-a}+\frac{C}{r^{2a}} \left(\int_{B_r}|u|^2\right)^{\frac{q}{2}} $$ $$ 2\le q\le 6, a=\frac{3}{4}(q-2) $$ Here $B_r$ is a ball of radius $r$ and $C$ is a constant independent of $r$. If complete detailed proof could be provided, that will help me a lot!!

$\endgroup$
4
  • 1
    $\begingroup$ The second term is needed to control constant functions. $\endgroup$ Mar 1, 2022 at 14:56
  • $\begingroup$ @WillieWong Thanks for your answer! I think that in the wiki link the second term should correspond to the case j=0. I think this is what you are referring to. $\endgroup$
    – Xeh Deng
    Mar 1, 2022 at 16:51
  • $\begingroup$ Yes, if you accept the GNS Interpolation inequality as generally true, then you just need Step 1 of my answer below, which is the scaling argument that shows the $r^{-2a}$ dependence. $\endgroup$ Mar 1, 2022 at 17:45
  • $\begingroup$ @WillieWong Yeah, I currently accept the GNS and now I know why the second term comes with the $r^{-a}$ dependence. It is just due to the scaling which ensures that the magnitude of the spatial integral will be proportional to the spatial domain size on both sides of the identity. Thanks a lot !~ $\endgroup$
    – Xeh Deng
    Mar 1, 2022 at 19:03

1 Answer 1

2
$\begingroup$

By context, I infer that you are working in 3 dimensions.

Step 1: reduce to case $r = 1$

Consider the mappings $S_r$ that map $$ S_r u(x) := r^{-3/q} u(r^{-1} x) $$ We have that $S_r: L^q(B_1) \to L^q(B_r)$ is an isometric bijection.

Note that $$ \|\nabla S_r u \|_{L^2(B_r)} = r^{-3/q - 1 + 3/2} \|\nabla u\|_{L^2(B_1)} $$ and $$ \| S_r u\|_{L^2(B_r)} = r^{3/2 - 3/q} \|u\|_{L^2(B_1)} $$

A direct computation shows that if we know

$$ \|u\|_{L^q(B_1)}^q \leq C \|\nabla u\|_{L^2(B_1)}^{2a} \|u\|_{L^2(B_1)}^{q-2a} + C \|u\|_{L^2(B_1)}^{q} $$

Then for $u\in H^1(B_r)$, applying the inequality to $S_r^{-1} u\in H^1(B_1)$ and using the scaling identity above we see that the desired inequality holds for all $r$.

Step 2: Extension and GNS

The unit ball has $C^1$ boundary and is a Sobolev extension domain. In particular, there exists a bounded linear operator $E: H^1(B_1) \to H^1_0(B_2)$.

Gagliardo-Nirenberg-Sobolev inequality says that

$$ \|Eu\|_{L^6(B_2)} \leq C \|\nabla Eu\|_{L^2(B_2)} \leq C \|Eu\|_{H^1(B_2)} $$

So (for a potentially different $C$)

$$ \|u\|_{L^6(B_1)} \leq \|Eu\|_{L^6(B_2)} \leq C\|u\|_{H^1(B_1)}$$

Step 3: interpolate

Apply the standard interpolation inequality $$ \|u\|_{L^q} \leq \|u\|_{L^2}^{1-\theta} \|u\|_{L^6}^{\theta} $$ for an appropriate $\theta\in [0,1]$ and $q\in [2,6]$, you get

$$ \|u\|_{L^q(B_1)}^q \leq C \|u\|_{L^2(B_1)}^{1-a} \|u\|_{H^1(B_1)}^{a} $$

Finally expand

$$ \|u\|_{H^1}^a \leq C'' \|u\|_{L^2}^a + C'' \|\nabla u\|_{H^1}^{a} $$

and you are done.

$\endgroup$
1
  • 1
    $\begingroup$ Thanks! Since my major is physics so I think I may need more time to think over it. $\endgroup$
    – Xeh Deng
    Mar 1, 2022 at 16:58

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.