For which one $p \ge 1$ is with $$A(x_n)_{n=1}^{\infty}=\left(\frac{1}{m}\sum_{n=1}^{m}\frac{x_n}{\sqrt{n}}\right)_{m=1}^{\infty}$$ defined bounded linear operator $A:\ell^p \to \ell^p$? Find norm $\| A\|.$

First, I have to show that if $x = (x_n)_{n=1}^{\infty} \in \ell^p$ then $\displaystyle y=(y_m)_{m=1}^{\infty} \stackrel{\text{def}}{=}\left(\frac{1}{m}\sum_{n=1}^{m}\frac{x_n}{\sqrt{n}}\right)_{m=1}^{\infty} \in \ell^p$, that is $\displaystyle\sqrt[p]{\sum_{m=1}^{\infty}|y_m|^p} < +\infty.$

All right, $$\begin{align*} |y_m|=\left|\frac{1}{m}\sum_{n=1}^{m}\frac{x_n}{\sqrt{n}}\right|&\leq\frac{1}{m} \sum_{n=1}^{m} \frac{|x_n|}{\sqrt{n}}\\ &\stackrel{\star}{\leq}\frac{1}{m} \sqrt[p]{\sum_{n=1}^{m}|x_n|^p} \sqrt[q]{\sum_{n=1}^{m} n^{-q/2}} \\ &\leq \frac{\| x \|_p}{m}\sqrt[q]{\sum_{n=1}^{m} n^{-q/2}}.\end{align*}$$

So $$\begin{align*} \sum_{m=1}^{\infty} |y_m|^p &\leq \sum_{m=1}^{\infty} \left|\frac{\| x \|_p}{m}\sqrt[q]{\sum_{n=1}^{m} n^{-q/2}}\right|^p\\&=\sum_{m=1}^{\infty} \frac{1}{m^p} \| x\|^p_p \left(\sum_{n=1}^{m} n^{-q/2}\right)^{p/q}\\ &\leq \|x\|^p_p \sum_{m=1}^{\infty} \frac{1}{m^p} \left(\sum_{n=1}^{\infty} n^{-q/2}\right)^{p/q} \end{align*}$$

If I see well, last sum is convergent for $q/2 >1 \iff q>2$ and $p>1$ (but with condition $1/p + 1/q =1$).

I think my estimate is too sharp.

And we find that $$\| A \| \leq \sqrt[p]{\sum_{m=1}^{\infty} \frac{1}{m^p} \left(\sum_{n=1}^{\infty} n^{-q/2}\right)^{p/q}}$$

But for part (if my approximation is good enough) with $\geq$ I don't have idea (to be honest, I used Hölder inequality, so we want $x_n = c \frac{1}{\sqrt{n}}$, that will give us something divergent I think so).

$\star$ - one doubt also: I used here Hölder inequality, but that works only for $p,q >1$, but in my question we have $p \geq 1$, so basically I have to work case $p=1$ as separate?

  • $\begingroup$ I think Hölder inequality works for all $1\leq p,q\leq +\infty$ such that $\frac{1}{p}+\frac{1}{q}=1$. $\endgroup$
    – Neph
    May 31 '13 at 22:02

As mention by robjohn, only the case $p>1$ allows $A$ to be bounded.

In your last estimate, you changed $m$ to $\infty$, and there was no reason for that step, as you already had an upper bound.

One way to get a lower bound for the norm of $A$ is as follows. Fix $k\in\mathbb N$. Let $$ x_n=n^{-q/2p}. $$ Then, since $-q/2p-1/2=-q/2$, \begin{align} \|Ax\|_p^p&=\sum_{m=1}^\infty|(Ax)_m|^p =\sum_{m=1}^\infty \left|\frac1m\,\sum_{n=1}^m\frac{n^{-q/2p}}{n^{1/2} }\right|^p\\ \ \\ &=\sum_{m=1}^\infty \left(\frac1m\,\sum_{n=1}^m{n^{-q/2}}\right)^p\\ \ \\ &=\sum_{m=1}^\infty \frac1{m^p}\left[\left(\sum_{n=1}^m{|n^{-q/2p}|^p}\right)^{1/p}\left(\sum_{n=1}^m{n^{-q/2}}\right)^{1/q}\right]^p\\ \ \\ &=\|x\|_p^p\,\sum_{m=1}^\infty\frac1{m^p}\left(\sum_{n=1}^m{n^{-q/2}}\right)^{p/q}. \end{align} So $$ \|A\|\geq \left(\sum_{m=1}^\infty\frac1{m^p}\,\left(\sum_{n=1}^mn^{-q/2}\right)^{p/q}\right)^{1/p}, $$ as you already had the upper bound.

The series converges for any $p>1$. Indeed, for $m$ large enough we have $$ \sum_{n=1}^mn^{-q/2}\leq 1+\int_1^mt^{-q/2}=1+\frac{m^{1-q/2}-1}{1-q/2}\leq c\,m^{1/2} $$ for some $c>0$. So $$ \frac1{m^p}\,\left(\sum_{n=1}^mn^{-q/2}\right)^{p/q}\leq c^{1/q}\,m^{p/2q-p} =c^{1/q}\,m^{-p/2-1/2}. $$ Since $p>1$, the series $\sum_{m=1}^\infty\frac1{m^p}\,\left(\sum_{n=1}^mn^{-q/2}\right)^{p/q}$ converges.

  • $\begingroup$ Thank you Martin for this. Could you just put a line or two of explanation for $\| Ax \|_p^p \geq \ldots$, because I am having trouble to understand it well. $\endgroup$
    – Cortizol
    Jun 1 '13 at 10:24
  • $\begingroup$ I have added a few lines. I don't know if that is what you were looking for. $\endgroup$ Jun 1 '13 at 14:45
  • $\begingroup$ Do we need here $\sum_{m=1}^k \left| \frac 1m \sum_{n=1}^m \frac {n^{-q/2}}{\left(\sum_{n=1}^kn^{-q/2} \right)^{1/p}} \right|^p = \sum_{m=1}^k \frac 1{m^p} \, \left( \sum_{n=1}^m n^{-q/2} \right)^{p-1}$ sign $\leq$ instead $=$, because, if I see good, $m \leq k$, so $\sum_{n=1}^k n^{-q/2} \geq \sum_{n=1}^m n^{-q/2}$. And that might be problem. I am probably wrong, I am very tired. $\endgroup$
    – Cortizol
    Jun 1 '13 at 15:40
  • $\begingroup$ I am sorry, but something is wrong with Comment, and I can't edit him because it's too long $\endgroup$
    – Cortizol
    Jun 1 '13 at 15:43
  • 1
    $\begingroup$ @ArthurFischer Thank you very much (at least, I voted for you, so I have some use of it :-)) $\endgroup$
    – Cortizol
    Jun 1 '13 at 16:14

Brute force and Hölder give us, with $\frac1p+\frac1q=1$, $$ \begin{align} \sum_{m=1}^\infty\left|\frac1m\sum_{n=1}^m\frac{x_n}{\sqrt{n}}\right|^p &\le\sum_{m=1}^\infty\frac1{m^p}\left(\sum_{n=1}^m\frac{|x_n|}{\sqrt{n}}\right)^p\\ &\le\sum_{m=1}^\infty\frac1{m^p}\|x_n\|_p^p\left(\sum_{n=1}^m\frac1{n^{q/2}}\right)^{p/q}\\ &\le\|x_n\|_p^p\sum_{m=1}^\infty\frac1{m^p}\left(2\sqrt{m}\right)^{p-1}\tag{$q\ge1$}\\ &=2^{p-1}\|x_n\|_p^p\zeta\left(\frac{p+1}{2}\right) \end{align} $$ Thus, $A$ is bounded for $p\gt1$ with norm no greater than $2^{1-1/p}\zeta\left(\frac{p+1}{2}\right)^{1/p}\le2\,\zeta\left(\frac{p+1}{2}\right)$.

For $(x_n)=(1,0,0,\dots)\in\ell^1$, $$ \frac1m\sum_{n=1}^m\frac{x_n}{\sqrt{n}}=\frac1m $$ Thus, we cannot have that $A$ is bounded on $\ell^1$.

  • $\begingroup$ How did you find that $\displaystyle\sum_{n=1}^{m} n^{-q/2} \leq 2 \sqrt{m}$ (if I see good, because $p/q=p-1$, so you didn't change that). It's probably something obvious, but I can't see it. And thanks for example that $A$ is unbounded on $\ell^1$. $\endgroup$
    – Cortizol
    Jun 1 '13 at 7:47
  • $\begingroup$ @Cortizol: since $q\ge1$, we have that $$ \sum_{n=1}^m n^{-q/2}\le\sum_{n=1}^m n^{-1/2}\le\int_0^m\frac{\mathrm{d}x}{\sqrt{x}}=2\sqrt{m} $$ $\endgroup$
    – robjohn
    Jun 1 '13 at 11:27
  • $\begingroup$ Now everything is clear. Thank you Rob for your help. I accepted Martin answer because he found sequence $(x_n)$ such that $\| A \| \geq \ldots$, I hope that is all right with you. $\endgroup$
    – Cortizol
    Jun 1 '13 at 16:36

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.