I am self-studying Brezis' functional analysis, and here is an exercise in it:

Exercise 2.7 Let $a=(a_n)$ be a given sequence sequence of real numbers and let $1\leq p\leq\infty$. Assume that $$\sum |a_n x_n|<\infty$$ for every $x=(x_n)\in \ell^p$. Prove that $a\in\ell^{p'}$, where $\frac{1}{p}+\frac{1}{p'}=1$.

My attempt (basically following the book's hint):

Write $E=\ell^{p}$. Define $$T_nx:=\sum_{i=1}^n a_ix_i$$ for all $n\in \mathbb{N}$. Note that $T_n\in E^\star$.

Since $\sum |a_n x_n|<\infty$, $T_nx$ tends to a limit $Tx\in\mathbb{R}$ as $n\rightarrow\infty$. By uniform boundedness principle, we have $$|T_nx|\leq C\lVert x\rVert_{\mathscr{l}^p}$$ for some $C>0$.

I stopped here.

It seems that I need to take specific $x$ to derive desired conclusion, but I only have a little experience with sequence spaces. Could anyone offer a further hint for me?

  • $\begingroup$ Please add that $ p' $ is defined so that $ \dfrac{1}{p} + \dfrac{1}{p'} = 1 $. $\endgroup$ – Transcendental Sep 5 '16 at 0:09
  • $\begingroup$ I have added that, but Brezis didn't mention that explicitly. $\endgroup$ – Fuhsuan Ho Sep 5 '16 at 6:06

I'll handle the case $1 < p < \infty$. Without loss of generality, assume $\mathbf{a}$ is nonzero. Let $N$ be an index for which $a_N$ is nonzero. For fixed $n\ge N$, let $\mathbf{x}^n = (x^n_i)_{i=1}^\infty$ where $x_i^n = \operatorname{sgn}(a_i)\lvert a_i\rvert^{p'-1}/(\sum\limits_{i = 1}^n\lvert a_i\rvert^{p'})^{1/p}$ for $i \le n$ and $0$ otherwise. Show that $\|\mathbf{x}^n\|_{\mathcal{p}} = 1$ and $T\mathbf{x}^n = (\sum\limits_{i = 1}^n \lvert a_i\rvert^{p'})^{1/p'}$. Then $$\left(\sum_{i = 1}^n \lvert a_i\rvert^{p'}\right)^{1/p'} \le \|T\|\quad \text{for all $n\ge N$},$$ whence $\mathbf{a}\in \mathcal{l}^{p'}$.

| cite | improve this answer | |
  • $\begingroup$ Nice answer, Kobe! $\endgroup$ – Fuhsuan Ho Sep 5 '16 at 6:02

Fix $x$, and fix $\varepsilon>0$. Choose $n$ such that $\sum_{j>n}|a_jx_j|<\varepsilon$. You have $$ |Tx|\leq|(T-T_n)x|+|T_nx|\leq\varepsilon+C\|x\|_p. $$ As this can be done for any $\varepsilon>0$, we conclude that $$ |Tx|\leq C\,\|x\|_p. $$

| cite | improve this answer | |
  • $\begingroup$ Thanks for replying! But I don't understand why this implies aa is in $\ell^{p'}$? @Martin Argerami $\endgroup$ – Fuhsuan Ho Sep 4 '16 at 16:08
  • $\begingroup$ I see. I misinterpreted your question. You mean $\ell^{p'}$, as in $1/p'+1/p=1$. I'll reply a little later, as I have to go now. $\endgroup$ – Martin Argerami Sep 4 '16 at 20:36
  • $\begingroup$ @Martin: You can always make edits on your iPhone, if you have one. :) $\endgroup$ – Transcendental Sep 5 '16 at 0:07
  • $\begingroup$ @Transcendental: years ago I have typed many answers on an iPad, and believe me it is something I don't have any intention of doing again. I still type on an Android tablet from time to time, and it is almost as bad. $\endgroup$ – Martin Argerami Sep 5 '16 at 3:38
  • $\begingroup$ @Fuxuan: the part that is missing from my answer is the one in kobe's answer. $\endgroup$ – Martin Argerami Sep 5 '16 at 3:39

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.