3
$\begingroup$

Prove the dual space of $\ell^p$ is isomorphic to $\ell^q$ if $\frac{1}{q}+\frac{1}{p}=1$ ($1<p<\infty$)

Define a map $J:\ell^q \to (\ell^p)'$ such that $Jy(x)=\sum_{k=1}^\infty x_ky_k,x\in \ell^p,y\in \ell^q$

I have verified that $Jy\in (\ell^p)'$, $J$ is linear and $\lVert Jy \rVert\leq \lVert y \rVert_q$.

How to show $\lVert Jy \rVert \geq \lVert y \rVert_q$ and $J$ is surjective?

$\endgroup$
  • $\begingroup$ @Aweygan It's $1<p<\infty$. I have edited the question. $\endgroup$ – Kenneth.K Feb 23 '17 at 16:16
2
$\begingroup$

Given $y=\{y_k\}\in\ell^q$ nonzero, consider the sequence $x=\{e^{-i\theta_k}\|y\|_q^{1-q}|y_k|^{q/p}\}$, where $\theta_k=\arg(y_k)$ (define $\theta_k$ arbitrarily if $y_k=0$). Then $x\in\ell^p$ with $\|x\|_p=1$ (it's not too hard to see), and \begin{align} Jy(x)&=\sum_{k=1}^\infty y_ke^{-i\theta_k}\|y\|_q^{1-q}|y_k|^{q/p}\\ &=\|y\|_q^{1-q}\sum_{k=1}^\infty|y_k|^{1+q/p}\\ &=\|y\|_q^{1-q}\sum_{k=1}^\infty|y_k|^q \\ &=\|y\|_q. \end{align} Thus $\|Jy\|\geq|Jy(x)|=\|y_p\|$.

To show that $J$ is surjective, fix $f\in(\ell^p)^*$ and look at what it does to the (Schauder) basis vectors $e_n=\{\delta_{k,n}\}$ for each $n\in\mathbb{N}$. This gives you a sequence $y=\{y_k\}$, and show that this sequence is in $\ell^q$ and $Jy=f$.

$\endgroup$
  • $\begingroup$ why the p-norm of $x$ is 1? $\sum_k e^{-pi\theta_k}\lVert y\rVert_q^{p-pq}|y_k|^q=1$? $\endgroup$ – Kenneth.K Feb 23 '17 at 16:39
  • $\begingroup$ @Kenneth.K You take modulus of the entries, so $e^{-i\theta_k}$ drops out. After that, mess arround with the equation $\frac{1}{p}+\frac{1}{q}=1$ to get some formulas that help simplify this expression. $\endgroup$ – Aweygan Feb 23 '17 at 16:43
  • $\begingroup$ To prove it's surjective, I have $y_k = f(e_k)$. Then $\lVert f \rVert \geq |f(e_k)|=|y_k|$. Hence $(\sum |y_k|^q)^{1/q} \leq (\sum \lVert f \rVert^q)^{1/q}$. Is this right? $\endgroup$ – Kenneth.K Feb 23 '17 at 18:04
1
$\begingroup$

Surjectivity of $J$.

Let $\varphi\in (\ell^p)'$, with $\|\varphi\|_*=1$, and set $y_n=\varphi(e_n)$, where $e_n=(0,\ldots,0,1,0,\ldots)$, with exactly one $1$ in the $n$th place. We shall show that $\{y_n\}\in \ell^q$, $\|\{y_n\}\|=1$ and $\varphi(\{x_n\})=\sum x_ny_n$.

I. $\{y_n\}\in \ell^q$. For every $x_1,\ldots,x_n$, with $|x_1|^p+\cdots+|x_n|^p\le 1$, we have $\|(x_1,\ldots,x_n,0,0,\ldots)\|_p\le 1$ and hence $$ 1\ge \varphi(x_1e_1+\cdots+x_ne_n)=x_1y_1+\cdots+x_ny_n. $$ But $$ \sup_{|x_1|^p+\cdots+|x_n|^p\le 1}x_1y_1+\cdots+x_ny_n=\big(|y_1|^q+\cdots+|y_n|^q\big)^{1/q}, \tag{1} $$ and hence $$ |y_1|^q+\cdots+|y_n|^q\le 1, $$ and since this holds for every $n$, then $\|\{y_n\}\|_q\le 1$ and $\{y_n\}\in \ell^q$.

II. $\varphi(\{x_n\})=\sum x_ny_n$. Let $x=\{x_n\}\in\ell^p$ and set $x^n=(x_1,x_2,\ldots,x_n,0,0,\ldots)$. Then $\|x^n-x\|\to 0$ and hence $\varphi(x^n-x)\to0$. But $$ \varphi(x^n)=\sum_{k=1}^n x_ky_k, $$ and as the series $\sum_{n=1}^\infty x_ny_n$, converges then $$ \varphi(x)=\lim_{n\to\infty}\varphi(x^n)=\sum_{n=1}^\infty x_ny_n=(x,y). $$ The fact that $\|y\|_q=1$ can be readily shown.

Proof of (1). Clearly, (Hölder) $$ |x_1y_1+\cdots+x_ny_n|\le \left(|x_1|^p+\cdots+|x_n|^p\right)^{1/p} \left(|y_1|^q+\cdots+|y_n|^q\right)^{1/q}\le \left(|y_1|^q+\cdots+|y_n|^q\right)^{1/q} $$ It is easily shown that equality is obtained for $$ x_i=\frac{|y_i|^{q/p}\mathrm{sgn}(y_i)}{\left(|y_1|^q+\cdots+|y_n|^q\right)^{1-1/q}}, \quad i=1,\ldots,n. $$

$\endgroup$
  • $\begingroup$ why the supremum of $x_1y_1+\dots+x_ny_n$ is $(|y_1|^q+\dots+|y_n|^q)^{1/q}$? I know by Minkowski's inequality, the supremum is less than the right, but why could the equality be achieved? $\endgroup$ – Kenneth.K Feb 23 '17 at 18:13
  • $\begingroup$ Short answer: That's the case in $\mathbb R$ with $p-$norm. On the other hand, take a look at my edited answer. $\endgroup$ – Yiorgos S. Smyrlis Feb 23 '17 at 23:26

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.