I am trying to follow a step in a proof which uses the Hölder Inequality.

The setting is: $f\in L^{p}(\mathbb{R}^{d})$, $g\in L^{1}(\mathbb{R}^{d})$, with $1\leq p\leq \infty$.

The claim is that: By the Hölder Inequality, for $1\leq p < \infty$,

$$\left|\int_{\mathbb{R}^{d}}|f(x-y)g(y)|dy\right|^{p} \leq \left(\int_{\mathbb{R}^{d}}|f(x-y)|^{p}|g(y)|dy\right)\left(\int_{\mathbb{R}^{d}}|g(y)|dy\right)^{p-1}$$

But I am having a hard time seeing which Hölder Conjugates were used.

Is the Minkowski Inequality also used here? Thank you!

  • $\begingroup$ I think my answer to your question is fairly conceptual; have you read it? $\endgroup$ – Amitesh Datta Nov 18 '11 at 2:11

Note that when $\frac{1}{p}+\frac{1}{q}=1$, $q=\frac{p}{p-1}$. Therefore,

$$ \small\begin{align} \left|\int_{\mathbb{R}^n}|f(x-y)g(y)|\mathrm{d}y\right|^p &\le\left(\int_{\mathbb{R}^n}\left(|f(x-y)||g(y)|^{1/p}\right)\left(|g(y)|^{1-1/p}\right)\mathrm{d}y\right)^p\\ &\le\left(\left(\int_{\mathbb{R}^n}\left(|f(x-y)||g(y)|^{1/p}\right)^p\mathrm{d}y\right)^{1/p}\left(\int_{\mathbb{R}^n}\left(|g(y)|^{1-1/p}\right)^{\frac{p}{p-1}}\mathrm{d}y\right)^{\frac{p-1}{p}}\right)^p\\ &=\left(\int_{\mathbb{R}^n}|f(x-y)|^p|g(y)|\mathrm{d}y\right)\left(\int_{\mathbb{R}^n}|g(y)|\mathrm{d}y\right)^{p-1} \end{align} $$

Another approach:

Let $G=\int_{\mathbb{R}^n}|g(y)|\mathrm{d}y$. Then Jensen's inequality yields $$ \begin{align} \left|\int_{\mathbb{R}^n}|f(x-y)g(y)|\mathrm{d}y\right|^p &=\left|G\int_{\mathbb{R}^n}\left|f(x-y)\right|\frac{|g(y)|}{G}\mathrm{d}y\right|^p\\ &\le G^p\int_{\mathbb{R}^n}\left|f(x-y)\right|^p\frac{|g(y)|}{G}\mathrm{d}y\\ &=\left(\int_{\mathbb{R}^n}\left|f(x-y)\right|^p|g(y)|\mathrm{d}y\right)\left(\int_{\mathbb{R}^n}|g(y)|\mathrm{d}y\right)^{p-1} \end{align} $$

  • $\begingroup$ +1 I just figured out the exponents. You beat me to it. =) $\endgroup$ – Srivatsan Nov 17 '11 at 2:49
  • 1
    $\begingroup$ Neat as usual when you do $p,q$-gymnastics :) $\endgroup$ – t.b. Nov 17 '11 at 2:51
  • 3
    $\begingroup$ Oops, I just noticed that I proved this for $\mathbb{R}^n$, not $\mathbb{R}^d$! ;-) $\endgroup$ – robjohn Nov 17 '11 at 3:21

I think a conceptual manner of understanding this situation is to define a positive measure $\mu$ on $\mathbb{R}^n$ by the rule $d\mu = \left|g\right|dy$, where $dy$ is Lebesgue measure, i.e., by the rule $\mu(E)=\int_{E} \left|g(y)\right| dy$, where $E$ is a Lebesgue measurable subset of $\mathbb{R}^n$. (In the language of measure theory, $\left|g\right|$ is the Radon-Nikodym derivative $\frac{d\mu}{dy}$.)

If $x\in\mathbb{R}^n$ is fixed, then let us apply Hölder's inequality to the functions $y\to \left|f(x-y)\right|$ and $y\to 1$ on $\mathbb{R}^n$ with respect to the measure $\mu$ and with the exponents $p$ and $p'$:

$$ \begin{eqnarray*} \int_{\mathbb{R}^n} \left|f(x-y)g(y)\right| dy &=& \int_{\mathbb{R}^n} \left|f(x-y)\right|d\mu(y) \\ &\leq& \left(\int_{\mathbb{R}^n} \left|f(x-y)\right|^{p}d\mu(y)\right)^{\frac{1}{p}}\left(\int_{\mathbb{R}^n} 1^{p'}d\mu(y)\right)^{\frac{1}{p'}} \\ &=& \left(\int_{\mathbb{R}^n} \left|f(x-y)\right|^{p}\left|g(y)\right|dy\right)^{\frac{1}{p}}\left(\int_{\mathbb{R}^n}\left|g(y)\right|dy\right)^{\frac{1}{p'}}. \end{eqnarray*} $$

You should justify the equalities in the computation above (the inequality is simply a special case of Hölder's inequality). If we take the $p$th power of both sides of the inequality in the above computation, then we deduce the desired inequality.

The technique that I have used to answer your question is an important one; e.g., it is used in at least one proof of Young's inequality on convolutions (if I remember correctly).

I hope this helps!

  • $\begingroup$ I was just looking back at my answer again, and I was thinking that we could use $g$ as a weight. Then looking at your answer, I saw that is just what you did (+1). $\endgroup$ – robjohn Feb 4 '12 at 21:08

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.