1
$\begingroup$

I have come across a problem which would simplify considerably if the following inequality

$$\left(\int |u|^2\,dx\right)^p \leq n(\left\Vert u \right\Vert_{1,2}^2 + \left\Vert u \right\Vert_{1,r}^r)$$

holds for some $n \in \mathbb N$, $0<p<1$, where $u$ ranges over $W^{1,2}_0 (\Omega)$ (zero trace) and $1<r<2$ is fixed.

My question is: Is it reasonable for this inequality to hold?

What I have so far: If we could further impose restrictions on dimension of $\Omega$, this would follow immediately by embedding results once we set $p=\frac r 2$. However, the general scenario seems to require some further estimates - and I would be curious if anyone could point me in the right direction.

$\endgroup$
1
  • $\begingroup$ This cannot be true for scaling reasons unless $2p \in [r,2]$: Plug in $\lambda u$, where $\lambda > 0$ is arbitrary and consider $\lambda \to 0$ and $\lambda \to \infty$. $\endgroup$
    – gerw
    May 21, 2021 at 5:52

1 Answer 1

1
$\begingroup$

First, we homogenize your inequality. I.e., we plug in $\lambda u$ with $\lambda > 0$ and consider $\lambda \to 0$ and $\lambda \to \infty$. This shows that the inequality cannot hold in case $2p \not\in [r,2]$.

In case $2p = r$, the first term on the rhs disappeards and your inequality is equivalent to $$ \|u\|_2 \lesssim \|u\|_{1,r}.$$ This holds under some relation of $r$ and the dimension $d$ (Sobolev embedding).

In the remaining case $2 p \in (r,2)$, you can divide by $\lambda^{2p}$ and minimize the resulting rhs w.r.t. $\lambda > 0$. This results in the homogeneous inequality $$ \|u\|_2 \lesssim \|u\|_{1,2}^{2\frac{1-r/(2p)}{2-r}} \|u\|_{1,r}^{r\frac{2/(2p)-1}{2-r}}. $$ Note that Young's inequality implies that this inequality is indeed equivalent to your original inequality; and the exponents on the right-hand side add up to one.

Now, we can try to produce this inequality by combining Sobolev embedding with interpolation (via Hölder's inequality).

From Sobolev, we get $$ \|u\|_q \lesssim \|u\|_{1,2}, \qquad \|u\|_s \lesssim \|u\|_{1,r} $$ for $ 1/q = 1/2 - 1/d$ and $ 1/s = 1/r - 1/d$. Thus, $$ \|u\|_{q}^{2\frac{1-r/(2p)}{2-r}} \|u\|_{s}^{r\frac{2/(2p)-1}{2-r}} \lesssim \|u\|_{1,2}^{2\frac{1-r/(2p)}{2-r}} \|u\|_{1,r}^{r\frac{2/(2p)-1}{2-r}}. $$ Interpolation implies $$ \|u\|_t \lesssim \|u\|_{1,2}^{2\frac{1-r/(2p)}{2-r}} \|u\|_{1,r}^{r\frac{2/(2p)-1}{2-r}}. $$ for $$ \frac1t = \frac{2\frac{1-r/(2p)}{2-r}}{s} + \frac{r\frac{2/(2p)-1}{2-r}}{q} . $$ Finally, we have $$ \|u\|_2 \le \|u\|_t $$ if $2 \le t$. Hence, in case $$ \frac12 \ge \frac{2\frac{1-r/(2p)}{2-r}}{s} + \frac{r\frac{2/(2p)-1}{2-r}}{q} . $$ we have shown your inequality. (If this relation is not satisfied, I expect that your inequality does not hold)

$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .