1

Now I'm reading the Young inequality. It says that if $f \in L^p(R)$, $g \in L^q(R)$, $1\leq p,q\leq \infty$, $1/p+1/q\geq 1$. Then how could we have the following inequalities:

$$\|f*g\|_q\leq \|g\|_q \|f\|_1$$

$$\|f*g\|_\infty\leq \|g\|_q \|f\|_{q^{'}},\quad (1/q+1/q^{'}=1)$$

$f*g$ means the convolution

The two inequalities seems not obvious, the second one is similar with the holder inequality, but actually not.

Could someone give me some hints or helpful links? Thanks a lot!

user39843
  • 957
  • In fact, there's a stronger inequality:

    $|f\ast g|_r\le |f|_p\cdot|g|_q$, where $1+1/r=1/p+1/q$ and $p,q,r\in [1,\infty]$. Unfortunately, I can't remember its name (I was taught it was Holder's inequality, but wiki doesn't agree with this name).

    – TZakrevskiy May 15 '13 at 14:42
  • @TZakrevskiy That's Young inequality for convolution, I believe. – Julien May 15 '13 at 14:42
  • @julien Indeed it is, thank you! Just found it on wiki. – TZakrevskiy May 15 '13 at 14:44
  • @TZakrevskiy yes, but Young inequality for convolution is based on the two inequalities. I have a idea to prove that now, meanwhile I am waiting for other elegant proof. – user39843 May 15 '13 at 14:53
  • @Norbert: the OP doesn't want to use Young's Inequality, since this is going to be used to show Young's Inequality. So, although they should state that in the question, that is an added requirement. – robjohn May 15 '13 at 23:16

3 Answers3

4

Here is the first inequality using Jensen instead of Young.

If $\|f\|_1=1$, we can use Jensen's inequality on the measure $\mathrm{d}\mu=f(x-y)\,\mathrm{d}y$ to get $$ \left|\int f(x-y)g(y)\,\mathrm{d}y\right|^p\le\int |f(x-y)||g(y)|^p\,\mathrm{d}y $$ Then integrate with respect to $x$ to get $$ \|f\ast g\|_p^p\le\|f\|_1\|g\|_p^p=\|f\|_1^p\|g\|_p^p $$ since $\|f\|_1=1$. This equation now scales nicely with $\|f\|_1$, so we can remove the restriction that $\|f\|_1=1$ to get $$ \|f\ast g\|_p\le\|f\|_1\|g\|_p $$


As mentioned by julien, the second inequality is simply Hölder's inequality.

robjohn
  • 353,833
3

The second inequality is easy. The first one is tricky, and was already proved by robjohn here as an answer to a different question. I just answer this for the sake of completeness because I couldn't find a proof of the second inequality here.

2) For the second inequality, I assume $1<q<\infty$ as the cases $q=1$ and $q=\infty$ are trivial. First note that $$ |(f\ast g)(x)|=\Big|\int f(x-y)g(y)dy\Big|\leq \int |f(x-y)||g(y)|dy $$ then by Hölder and variable change $$ \leq \left(\int|f(x-y)|^{q'} \right)^\frac{1}{q'} \left(\int|g(y)|^{q} \right)^\frac{1}{q}=\left(\int|f(y)|^{q'} \right)^\frac{1}{q'} \left(\int|g(y)|^{q} \right)^\frac{1}{q} $$ whence $$ \|f\ast g\|_\infty \leq \|f\|_{q'}\|g\|_q. $$

1) The first inequality is a particular case of Young's inequality $$ \|f\ast g\|_s\leq \|f\|_p\|g\|_q\qquad\mbox{when}\quad \frac{1}{p}+\frac{1}{q}=1+\frac{1}{s}, \quad 1\leq p,q,s\leq \infty $$ which is proved by robjohn in this thread. This does not cover the cases when $p,q,$ or $s$ are infinite. But $s=\infty$ is the case above, and your inequality has $p=1$, which yields $s=q$.

Julien
  • 45,674
  • Thanks for your answer.The second inequality, in deed, can be showed by holder inequality, which I don't get it at first. And the first inequality, could you show that without using the Young's inequality? Because in my book, we have these two inequalities to prove the Young's inequality. If I use it, that will lead to argue in a circle. – user39843 May 15 '13 at 16:10
  • @user39843 In that case, just mimick robjohn's proof making $p=1$ and $s=q$. – Julien May 15 '13 at 16:11
  • @julien: there is a simpler method using Hölder for the case that $f\in L^1$. – robjohn May 15 '13 at 16:44
  • @robjohn Right, thanks for the note. You meant Jensen, I guess, as your answer shows. – Julien May 15 '13 at 16:47
  • @julien: Jensen indeed. I'm doing too much at once :-) – robjohn May 15 '13 at 16:50
  • @user39843: I assume you are talking about $$ \begin{align} \iint |f(x-y)||g(y)|^p,\mathrm{d}y,\mathrm{d}x &=\iint |f(x-y)||g(y)|^p,\mathrm{d}x,\mathrm{d}y\ &=\iint |f(x)||g(y)|^p,\mathrm{d}x,\mathrm{d}y\ &=\int |f(x)|,\mathrm{d}x,\int |g(y)|^p,\mathrm{d}y\ &=|f|_1|g|_p^p \end{align} $$ – robjohn May 15 '13 at 17:12
  • Sorrt that I make a mistake. Now I have known what you said. @robjohn, thank you! – user39843 May 15 '13 at 17:14
2

I'll give a proof just using Holder's Inequality so that it may easier for you to understand, but as the above comments said the proof for these inequalities is already somewhere else.

Proof for the first: By Holder's Inequality we have \begin{align*} \left|\int f(x-y)g(y)dy\right|\le&\int|f(x-y)|^{1/q}|g(y)||f(x-y)|^{1-1/q}dy \\ \le&\left(\int|f(x-y)||g(y)|^qdy\right)^{1/q}\left(\int|f(x-y)|dy\right)^{1-1/q} \\ =&\|f\|_1^{1-1/q}\left(\int|f(x-y)||g(y)|^qdy\right)^{1/q},\qquad\forall x. \end{align*} Then by Fubini Theorem we have \begin{align*} \|f*g\|_q^q\le&\|f\|_1^{q-1}\int\int|f(x-y)||g(y)|^qdydx \\ =&\|f\|_1^{q-1}\int\left(|g(y)|^q\int|f(x-y)|dx\right)dy \\ =&\|f\|_1^{q}\|g\|_q^q. \end{align*} Taking off $q$ we prove the first one.

The proof for the second one: By Holder's Inequality we have \begin{align*} \left|\int f(x-y)g(y)dy\right|\le\left(\int|f(x-y)|^{q'}dy\right)^{1/q'}\left(\int|g(y)|^qdy\right)^{1/q}=\|f\|_{q'}\|g\|_{q},\qquad\forall x. \end{align*} By taking the supremum we get the second one.

L. Xu
  • 882
  • Hölder is Jensen using the convex function $x^p$ where $p\ge1$. (+1) – robjohn May 15 '13 at 17:20
  • Yes, but the change of measure is needed. – L. Xu May 15 '13 at 17:36
  • In most applications of Jensen, a change of measure is needed:

    $$ \begin{align} \left(\int f g,\mathrm{d}x\right)^p &=|g|_q^{pq}\left(\int\frac{f}{g^{q-1}} \left(\frac{g}{|g|_q}\right)^q,\mathrm{d}x\right)^p\ &\le|g|_q^{pq}\int\left(\frac{f}{g^{q-1}}\right)^p \left(\frac{g}{|g|_q}\right)^q,\mathrm{d}x\ &=|g|_q^{pq-q}\int f^pg^{p+q-pq},\mathrm{d}x\ &=|g|_q^p|f|_p^p \end{align} $$

    – robjohn May 15 '13 at 18:42