Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

I am trying to prove Hardy's Inequality for integrals:

Let $f:(0,\infty) \rightarrow \mathbb R$ be in $L^p$. Let $F(x) = \frac{1}{x} \int_0 ^x f(t)dt$. Then $F \in L^p$ and $\|F\|_p \leq \frac{p}{p-1} \|f\|_p$.

I have seen proofs of this which use Haar measure and Fourier analysis, eg. here An inequality by Hardy. However, I am wondering if there is a more elementary proof.

I have tried to deduce this using Jensen's inequality and Fubini's Theorem as follows:

Jensen's inequality implies that $\left|\frac{1}{x} \int_0 ^x f(t)dt)\right|^p \leq \frac{1}{x} \int_0 ^x |f(t)|^p dt$. Therefore

$$\int_0 ^\infty |F(x)|^p dx \leq \int_0 ^\infty \frac{1}{x} \int_0 ^x |f(t)|^p dtdx = \int_0 ^\infty \int_t ^\infty \frac{1}{x} |f(t)|^p dx dt.$$

But, this last integral is infinite, and I'm not sure how to get a sharper bound on $|F(x)|^p$. Any suggestions?

share|cite|improve this question
up vote 5 down vote accepted

Let $p\in (1,+\infty)$ and $q$ its conugate exponent, that is, the real number $q$ such that $\frac 1p+\frac 1q=1$. Let $f\colon (0,+\infty)\to\mathbb R\in L^p(0,+\infty)$ and $F(x)=\frac 1x\int_{\left]0,x\right[}f(t)dt$. Then $F\in L^p(0,1)$ and $\lVert F\rVert_{L^p}\leq q\lVert f\rVert f_{L^p}$.

Pick $0<\alpha<\frac 1q$, we will specifie it later. We start using Hölder inequality \begin{align*} |xF(x)|&=\left|\int_{\left]0,x\right[}f(t)t^{\alpha}t^{-\alpha}dt\right|\\ &\leq \left(\int_{\left]0,x\right[}\left| f(t) t^{\alpha}\right|^pdt\right)^{\frac 1p} \left(\int_{\left]0,x\right[}t^{-q\alpha}\right)^{\frac 1q}\\ &=\left(\int_{\left]0,x\right[}\left| f(t)\right|^p t^{p\alpha}dt\right)^{\frac 1p} \left(\frac 1{1-q\alpha}x^{1-q\alpha}\right)^{\frac 1q}\\ &=(1-q\alpha)^{-\frac 1q}\left(\int_{\left]0,x\right[}\left| f(t)\right|^p t^{\frac 1{q}}dt\right)^{\frac 1p}x^{\frac 1q-\alpha}, \end{align*} hence \begin{align*} |F(x)|&\leq (1-q\alpha)^{-\frac 1q}\left(\int_{\left]0,x\right[}\left| f(t)\right|^p t^{p\alpha}dt\right)^{\frac 1p}x^{\frac 1q-\alpha -1}\\ &=(1-q\alpha)^{-\frac 1q}\left(\int_{\left]0,x\right[}\left| f(t)\right|^p t^{p\alpha}dt\right)^{\frac 1p}x^{-\alpha -\frac 1p} \end{align*} and $$|F(x)|^p\leq (1-q\alpha)^{-\frac pq}\int_{\left]0,x\right[}\left| f(t)\right|^p t^{p\alpha}dt x^{-p\alpha-1}.$$ Integrating and using Fubini's theorem, we get

\begin{align*} \int_{(0,+\infty)}|F(x)|^pdx&\leq (1-q\alpha)^{-\frac pq}\int_{(0,\infty)}\int_{(0,x)} \left| f(t)\right|^p t^{p\alpha}x^{-p\alpha-1}dtdx\\ &=(1-q\alpha)^{-\frac pq}\int_{(0,+\infty)}\int_{(t,+\infty)}\left| f(t)\right|^p t^{p\alpha}x^{-p\alpha-1}dxdt\\ &=(1-q\alpha)^{-\frac pq}\int_{(0,+\infty)}\left| f(t)\right|^p t^{p\alpha}\left(\int_{(t,+\infty)}x^{-p\alpha-1}dx\right)dt, \end{align*} thus $$\int_{(0,+\infty)}|F(x)|^pdx \leq(1-q\alpha)^{-\frac pq}\int_{(0,+\infty)}\left|f(t)\right|^p t^{p\alpha}\frac 1{p\alpha}t^{-p\alpha}dt$$ and finally $$\int_{(0,+\infty)}|F(x)|^pdx\leq (1-q\alpha)^{-\frac pq}(p\alpha)^{-1}\int_{(0,+\infty)}\left| f(t)\right|^p dt.$$ Now, we pick $\alpha:=\frac 1{pq}<\frac 1q$, to get $$(1-q\alpha)^{-\frac pq}(p\alpha)^{-1} =(1-\frac 1p)^{-\frac 1q}q= q^{\frac pq}q=q^{1+p\left(1-\frac 1p\right)}=q^p,$$ which gives $\lVert F\rVert_{L^p}\leq q\lVert f\rVert_{L^p}$.

This question deals with the case of the best constant.

share|cite|improve this answer
Hi Davide! I'm afraid you didn't prove that $q$ is the best constant. Indeed, if $f=\chi_{[0, 1]}$ then $\lVert F\rVert_p=q^{1/p}$ which is smaller than $q$. Also, any "best constant" argument for this inequality must involve some limiting process, because equality is never achieved non-trivially. – Giuseppe Negro Apr 1 '12 at 13:47

Show $$\left(\int_0^\infty \left| \frac1x \int_0^x f(t) \, dt\right|^p \, dx \right)^{1/p} \le \frac p{p-1} \left(\int_0^\infty |f(x)|^p \, dx \right)^{1/p}$$ Notice $$ \frac1x \int_0^x f(t) \, dt = \int_0^1 f(x t) \, dt $$ Apply Minkowski's inequality (integral version) to LHS, to see it is bounded above by $$ \int_0^1 \left(\int_0^\infty |f(xt)|^p \, dx \right)^{1/p} \, dt = \int_0^1 t^{-1/p} \, dt \left(\int_0^\infty |f(x)|^p \, dx \right)^{1/p} $$

share|cite|improve this answer
Upvoting: nice proof. – Did Dec 31 '11 at 12:39

Dualize - it becomes $$ \left(\int_0^\infty |F(x)|^p \, dx\right)^{1/p} \le p \left(\int_0^\infty |f(x)|^p \, dx \right)^{1/p} $$ where $$ F(x) = \int_x^\infty \frac{f(t)}t \, dt $$ W.l.o.g. $f(x) \ge 0$, and w.l.o.g. $F(x) \to 0$ as $x \to \infty$. $$ |F(x)|^p = - p \int_x^\infty F'(t) |F(t)|^{p-1} \, dt $$ Substitute this into the LHS of the first inequality, and reverse the order of integration $$ - p \int_0^\infty F'(t) |F(t)|^{p-1} \int_0^t \, dx \, dt = p \int_0^\infty f(t) |F(t)|^{p-1} \, dt \le p \|f\|_p \|F\|_p^{p-1} $$ where the last inequality is H\"older. Now assume $\|F\|_p$ is finite, and divide both sides by $\|F\|_p^{p-1} $.

share|cite|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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