5

Here is my effort to show this fact and I will use ball $X^*$ to denote the closed unit ball of $X^*$.

To show ball $X^*$ is weak star metrizable, we only have to show there is a metric $d$ on ball $X^*$ such that the topology induced by $d$ is the weak-star topology on ball $X^*$. Since $X$ is separable, ball $X$ is also separable. Thus there exists a countable dense subset $\{x_n\}$ in ball $X$. Now define the metric $d$ on ball $X^*$ by

$$d(x^*,y^*)=\sum_{n=1}^{\infty}\frac{|\langle x_n,x^*-y^*\rangle|}{2^n},$$

where $x^*,y^*\in\text{ball $X^*$}$. Let $T$ be the topology induced by $d$ and $wk^*$ be the weak-star topology on $\text{ball $X^*$}$. Then we need to show $T=wk^*$. And I try to use net convergence to show topology equivalence.

Let $x^*\in\text{ball $X^*$}$ and let $x_i^*$ be a net in $\text{ball $X^*$}$ such that $x_i^*\overset{wk^*}{\longrightarrow} x^*$. Then $\langle x_n,x_i^*\rangle\rightarrow\langle x_n,x^*\rangle$ for all $n$. Now Let $x^*\in\text{ball $X^*$}$ and let $x_i^*$ be a net in $\text{ball $X^*$}$ such that $x_i^*\rightarrow x^*$ in $(X,T)$. Then for each $\epsilon>0$, there exists $i_\epsilon\in\mathbb{N}$ such that $d(x_i^*,x^*)<\epsilon$ for all $i\geqslant i_\epsilon$; that is, for each $\epsilon>0$,

$$d(x_i^*,x^*)=\sum_{n=1}^{\infty}\frac{|\langle x_n,x_i^*-x^*\rangle|}{2^n}<\epsilon.$$

Here I want to show $x_i^*\overset{wk^*}{\longrightarrow} x^*$ if and only if $x_i^*\rightarrow x^*$ in $(X,T)$. But I forget some knowledge in Calculus. Can somebody help me to show this please? Thank you so much!!

John Griffin
  • 10,968
  • 3
  • 19
  • 37
Answer Lee
  • 1,069

3 Answers3

4

Let $I$ denote the identity map from $(X^*,\text{weak*})$ onto $(X^*,d)$. Since $I$ is a bijection from a compact space (Banach-Alaoglu) into a Hausdorff space, we only need to show that $I$ is continuous.

To this end, fix a net $(x^*_\alpha)$ in $B_{X^*}$ such that $x^*_\alpha\to x^*$ in the weak* topology on $B_{X^*}$ for some $x^*\in B_{X^*}$. We want to show that $x^*_\alpha=I(x^*_\alpha)\to I(x^*)=x^*$ in the metric $d$; that is, we want: $$ \sum_{n=1}^\infty \frac{|\langle x_n,x_\alpha^*-x^*\rangle|}{2^n} \longrightarrow 0. $$ To this end, fix $\varepsilon>0$. Since each $x_n$ is in the unit ball of $X$ and $x^*$ and each $x^*_\alpha$ are in the unit ball of $X^*$, we deduce $$ \frac{|\langle x_n,x_\alpha^*-x^*\rangle|}{2^n} \le \frac{|x_\alpha^*(x_n)|+|x^*(x_n)|}{2^n} \le \frac{2}{2^n} = \frac{1}{2^{n-1}}. $$ Since $\sum_{n=1}^\infty2^{1-n}$ converges, there exists $N\in\mathbb{N}$ such that $\sum_{n=N+1}^\infty 2^{1-n}<\varepsilon/2$. Now, using the fact that $x_\alpha^*\to x^*$ in the weak* topology on $B_{X^*}$, for each $n\in\{1,\ldots,N\}$ there exists $\alpha_n$ in the directed set such that $$ |\langle x_n,x_\alpha^*-x^*\rangle| < \frac{2^n\varepsilon}{2N} $$ whenever $\alpha\ge\alpha_0$. Taking $\alpha_0$ in the direct set such that $\alpha_0\ge\alpha_1,\ldots,\alpha_N$, we obtain \begin{align*} \sum_{n=1}^\infty\frac{|\langle x_n,x_\alpha^*-x^*\rangle|}{2^n} &= \sum_{n=1}^N\frac{|\langle x_n,x_\alpha^*-x^*\rangle|}{2^n} +\sum_{n=N+1}^\infty\frac{|\langle x_n,x_\alpha^*-x^*\rangle|}{2^n} \\ &\le \sum_{n=1}^N \frac{\varepsilon}{2N} + \varepsilon/2 \\ &= \varepsilon \end{align*} whenever $\alpha\ge\alpha_0$. This completes the proof.


A few comments on your post:

You write "Since $X$ is separable, ball $X$ is separable". This is a true statement, but only because $X$ is a metric space. In general topological spaces, subsets of separable spaces aren't necessarily separable.

It's important to show that $d$ is a metric. It is here that the density of $\{x_n : n\in\mathbb{N}\}$ in the unit ball of $X$ is actually needed.

You wrote "Then for each $\varepsilon>0$, there exists $i_\varepsilon\in\mathbb{N}$ such that $d(x_i^*,x^*)<\varepsilon$ for all $i\ge i_\varepsilon$." The problem is that $i_\varepsilon$ cannot be assumed to be a natural number, since your net may be indexed by some directed set other than $\mathbb{N}$.

John Griffin
  • 10,968
  • 3
  • 19
  • 37
  • Thank you so much for your help! Can you show me why $X$ is separable, ball $X$ is separable? And I sorry I don't see the density of ${x_n}$ is using in the proof of show $d$ is a metric. Can you prove it if possible!! – Answer Lee Sep 10 '17 at 15:43
  • @AnswerLee For metric spaces, separability is equivalent to second countability. This, and the fact that subsets of second countable spaces are second countable, shows that subsets of a separable metric space are separable. (We say that separability is hereditary for metric spaces, but not in general.) – John Griffin Sep 10 '17 at 15:46
  • @AnswerLee The density of ${x_n}$ is used in showing that $d(x^,y^)=0$ implies $x^=y^$. – John Griffin Sep 10 '17 at 15:47
  • Awesome!! Actually can I ask you some questions about my homework by sending you an email? Because your answer is so great and I can totally understand! Appreciate!! – Answer Lee Sep 10 '17 at 15:55
  • @AnswerLee I wouldn't feel comfortable communicating via email. Whenever I'm not busy with something else, I usually have MSE open and am answering questions, so you can just post here. One tip for getting better responses would be to accept answers you are satisfied with. I've noticed that you haven't accepted any answers in all of the questions you've asked here. This may prevent some people from putting a lot of effort into answering your questions. – John Griffin Sep 10 '17 at 16:13
  • Also, I think many people prefer their free contributions to be shared in a public setting so that everyone can be benefited. – clark Sep 10 '17 at 16:19
  • @John Griffin Thank you! Sorry, I am not very familiar with this but now I know! And from your answer, I am not getting how $\sum_{n=1}^{N}\frac{\epsilon}{2N}+\frac{\epsilon}{2}=\epsilon$? Thank you so much! – Answer Lee Sep 10 '17 at 16:23
  • @AnswerLee It's no problem! Note that $$ \sum_{n=1}^N c = cN $$ for any constant $c$. Thus $$ \sum_{n=1}^N\frac{\varepsilon}{2N} = \frac{\varepsilon}{2N}N=\frac{\varepsilon}{2}. $$ – John Griffin Sep 10 '17 at 16:25
  • @John Griffin Is that necessary to let ${x_n}$ in ball $X$? I think let ${x_n}$ be a countable dense subset of X. We will still get the conclusion. Thank you! – Answer Lee Sep 11 '17 at 20:09
  • @AnswerLee No. Without that assumption we wouldn't be able to get the estimate: $$ \frac{|\langle x_n,x_\alpha^-x^\rangle|}{2^n} \le \frac{|x_\alpha^(x_n)|+|x^(x_n)|}{2^n} \le \frac{2}{2^n} = \frac{1}{2^{n-1}}. $$ We also need it to make sure the metric is real (not infinite) valued. – John Griffin Sep 13 '17 at 01:40
  • @JohnGriffin I am sorry. There is one thing I don't quite understand and I forgot to ask: How to use $\sum_{n=1}^\infty2^{1-n}$ converges to get there exists $N\in\mathbb{N}$ such that $\sum_{n=N+1}^\infty 2^{1-n}<\varepsilon/2$? – Answer Lee Sep 27 '17 at 16:54
  • @AnswerLee It's known as the Cauchy's convergence test and it essentially follows from the definition of series convergence using partial sums and the completeness of the real line. – John Griffin Sep 27 '17 at 23:43
  • @JohnGriffin I am sorry for bothering you again. I don't understand a small thing in this proof. How to show the metric $d$ is well defined? Thank you so much!!! – Answer Lee Dec 07 '17 at 04:03
  • @AnswerLee The only thing we need to check to prove that $d$ is well-defined is that $d(x^,y^)<\infty$ for $x^,y^$ in the unit ball of $X^$. Since each $x_n$ is in the unit ball of $X$, for each such $x^,y^$ we have $$ \frac{|\langle x_n,x^-y^* \rangle|}{2^n} \le \frac{1}{2^{n-1}} $$ so that $d(x^,y^)\le 2$. – John Griffin Dec 07 '17 at 04:30
  • @JohnGriffin Thank for your reply!! Do I need to prove this: If $a^=c^$ and $b^=d^$ in $\text{ball $X^$}$, we have $|\langle x_n,a^-b^\rangle|=|\langle x_n,c^-d^\rangle|$ for each $n$. Thus $$d(a^,b^)=\sum\limits_{n=1}^{\infty}\frac{|\langle x_n,a^-b^\rangle|}{2^n}=\sum\limits_{n=1}^{\infty}\frac{|\langle x_n,c^-d^\rangle|}{2^n}=d(c^,d^*).$$ Hence, $d$ is well defined. – Answer Lee Dec 07 '17 at 04:40
  • @AnswerLee No, because that is obvious by the meaning of equality. I think you are confused becomes sometimes, in order to prove that something is well-defined, we need to prove that the definition does not depend on something else (like the representative of an equivalence class). Here there is nothing like that going on. – John Griffin Dec 07 '17 at 04:44
  • @JohnGriffin Thank you so much for your help. I understand now. But a special case in this proof comes to my mind today. What if the normed space $X={0}$, will the closed unit ball of $X^*$ weak-star metrizable? – Answer Lee Dec 08 '17 at 19:37
  • @AnswerLee Yes, it will be due to the theorem because ${0}$ is separable. However for this particular case it's easy enough to just compute $(B_{X^},\text{weak})$. (Hint: Since $X^={0}$, we have $B_{X^}={0}$. How many topologies are there on a singleton, and are they metrizable?) – John Griffin Dec 09 '17 at 07:40
  • @JohnGriffin I think one topology on it. And we equip it with discrete topology, it is metrizable. But in this special case we don't have ${x_n}$ as an infinite set. So we should put it out like a special case right? – Answer Lee Dec 09 '17 at 19:43
  • @AnswerLee Or one could take $x_n=0$ for every $n$. – John Griffin Dec 09 '17 at 20:38
  • @JohnGriffin Got it! So will $X={0}$ and $X\neq{0}$ cover all in this theorem? Thank you! – Answer Lee Dec 09 '17 at 21:12
2

You can do this directly: after checking that $d$ is a metric (it is), let $\tau$ be the topology induced by it. Then,

$(1). \ \sigma (X^*,X)\subseteq\tau :$

Let $S$ be the closed unit ball in $X^*,\ $ fix $f_0\in S$ and take $\left \{ f\in S:d(f_o,f) <\epsilon \right \}\subseteq \tau.$ There is an integer $k$ such that $\frac{1}{2^{k}}<\frac{\epsilon }{2}.$ Then, $\ U=\left \{ f\in S:|(f-f_0,x_j)|< \frac{\epsilon }{2};\ 0\le j\le k \right \}$ is weak* open in $S,$ contains $f_0$ and satisfies:

$\forall f\in U,\ d(f,f_0)=\sum_{n=0}^{k}\frac{(f-f_0,x_n)}{2^{n}}+\sum_{n> k}\frac{(f-f_0,x_n)}{2^{n}}<\frac{\epsilon }{2}+\frac{1}{2^{k}}<\epsilon.$

$(2).\ \tau \subseteq\sigma (X^*,X) :$

Take $U=\left \{ f\in S:|(f-f_0,y_j)|<\epsilon;\ 0\le j\le k \right \}$, a weak* neighborhood of $f_0.$ Let $M=\max \left \{ \|y_j\|:0\le j\le k \right \}.$ There are integers $n_j$ such that for any $\delta>0,\ \|x_{n_j}-y_j/M\|<\delta.$ And there is an $r>0,$ such that $M2^{n_j}r<\frac{\epsilon }{2}.$

Then, if $d(f,f_0)<r,$ we have of course, $\sum_{n=0}^{\infty }\frac{|(f-f_0,x_n)|}{2^{n}}<r$ so that $|(f-f_0,x_{n_j})|\le 2^{n_j}r$ and therefore,

$|(f-f_0,y_j)|=M|(f-f_0,y_j/M)|\le M|(f-f_0,x_{n_j})|+M|(f-f_0,x_{n_j}-y_j/M)|\le M(2^{n_j}r+2\delta)\le \epsilon /2+2M\delta.$

Thus, $f\in U$ as soon as $\delta$ is small enough.

Matematleta
  • 30,081
0

Shorter, but less direct, proof. We show that dual ball $B_{X'}$ is homeomorphic to a subset of a metric space. Let $D$ be a countable dense subset in $X$. All continuous functions from $X$ to Hausdorf space (in particular to $ℝ$) are determined on $D$, hence $$ (B_{E'}, \sigma(E',E)) \ni f \mapsto f|_D \in ℝ^ℕ $$ is injective. This map is continuous ($ℝ^ℕ$ is equipped with the product topology, i.e. convergence = pointwise convergence). Indeed, let $f_\alpha \to f$ be a convergent net in $B_{E'}$ with weak* topology. This means that $$ f_\alpha(x) \to f(x) \quad (\forall x \in E). $$ In particular, for any $x \in D \subset E$ we have $$ (f_\alpha)|_D(x) = f_\alpha(x) \to f(x) = f|_D(x). $$ Hence $f\mapsto f|_D$ is a continuous injection. Moreover, $B_{E'}$ is weak* compact (Banach–Alaoglu theorem), hence this map is a homeomorphism onto its image which is a subset of a metrizable space $ℝ^ℕ$ (countable product of metric spaces is a metric).


Comment. We swept under the rug the fact that countable product of metric spaces is a metric. This fact is usually shown by defining a metric analogous to the one described in question.

Kamil
  • 861