I want to prove exactly what is written in the title, using the following idea:

Let $X$ be a separable Banach space. We have that $X$ is isometrically isomorphic to a subspace $M$ of $C(B_{X^\ast})$, where $B_{X^\ast} = \{x^\ast\in X^\ast:\ \|x^\ast\| = 1 \}$ and $C(B_{X^\ast})$ is the space of continuous functions from $B_{X^\ast}$ to $\mathbb{K}$ ( we have $\mathbb{K} = \mathbb{R}$ or $\mathbb{C}$).

I want to prove that $M$ is isometrically isomorphic to some subspace of $\ell_\infty$. Doing this it's sufficient to finish the proof, because then $X\cong M\cong $ some subspace of $\ell_\infty$. The first isometric isomorphism is already proven, the second one is the problem. I came up with a solution, but I'm not sure if it's right because the Banach hypothesis was not used (or I just missed where it's used). I appreciate your help, thanks.

My solution: $B_{X^\ast}$ is $w^\ast$-separable so let $(x^\ast_n)_{n\in\mathbb{N}}$ be a $w^\ast$-dense sequence in $B_{X^\ast}$ and define $\Phi:M\to\ell_\infty$ such that $\Phi(f) = (f(x^\ast_1),\ldots,f(x^\ast_n),\ldots)$. It's not hard to see $\Phi$ is well defined, for $|f(x^\ast_n)| \leq \|f\| \cdot \|x^\ast_n\| \leq \|f\|$ for all $n\in\mathbb{N}$, so $\Phi(f)\in \ell_\infty$ for all $f\in C(B_{X^\ast})$.

It's not hard also to prove that $\Phi$ is a linear injection, so the main problem is to prove that $\Phi$ is an isometry. Take any $f\in C(B_{X^\ast})$, we want to prove $\|\Phi(f)\| = \|f\|$, and for this we just need to prove that $\sup_{n\in\mathbb{N}}|f(x^\ast_n)| = \sup_{x^\ast\in B_{X^\ast}} |f(x^\ast)|$ ( for now, all we know is $\sup_{n\in\mathbb{N}}|f(x^\ast_n)| \leq \sup_{x^\ast\in B_{X^\ast}} |f(x^\ast)|$).

First of all, note that $B_{X^\ast}$ is $w^\ast$-compact (by Banach-Alaoglu theorem), then $\sup_{x^\ast\in B_{X^\ast}} |f(x^\ast)| = \max_{x^\ast\in B_{X^\ast}} |f(x^\ast)| = |f(y^\ast)|$ for some $y^\ast \in B_{X^\ast}$. Also, we know $f$ is continuous in $y^\ast$, so for any $\varepsilon > 0$ there is $\delta > 0$ such that $\|x^\ast - y^\ast\| < \delta \implies |f(x^\ast) - f(y^\ast)| < \varepsilon$.

We don't know if there is some $x^\ast_n$ close enough to $y^\ast$, for the separability is about the $w^\ast$ topology, not the norm, so let's work with that. Given any $w^\ast$-neighborhood $V$ of $0$ in $B_{X^\ast}$, we know $V$ is limited, so consider the collection of $w^\ast$-neighborhoods $(\frac{1}{m}V)_{m\in\mathbb{N}}$. Take $m\in\mathbb{N}$ such that $\frac{1}{m} < \delta$, since $(x^\ast_n)_{n\in\mathbb{N}}$ is $w^\ast$-dense in $B_{X^\ast}$, there is some $x^\ast_n$ such that $$x^\ast_n \in y^\ast+\frac{1}{m}V \implies x^\ast_n - y^\ast \in\frac{1}{m}V \subset \delta V \subset \delta B_{X^\ast} \implies \|x^\ast_n - y^\ast\| <\delta .$$

From this we conclude that $|f(x^\ast_n) - f(y^\ast)| < \varepsilon$, but $\varepsilon > 0$ is arbitrary, so there is functionals $x^\ast_n$ such that $f(x^\ast_n)$ is arbitrarily close to $f(y^\ast)$. With this we can conclude that

$$\sup_{n\in\mathbb{N}} |f(x^\ast_n)| = |f(y^\ast)|,$$

and this proves $\Phi$ is a isometry.

PS: There is this thread. But it looks like a different approach.

  • $\begingroup$ It doesn't matter for your main argument, but $X$ is of course again isomorphic to a subspace of $C(B)$, not to the whole space. $\endgroup$ – user138530 Nov 14 '15 at 0:58
  • $\begingroup$ Also, your question is answered in the question you linked too. $\endgroup$ – user138530 Nov 14 '15 at 1:02
  • $\begingroup$ You are right about $X$ being isomorphic to a subspace, not the whole space, I'm going to fix this. About the the other answer, it goes in another direction. They don't prove the way I want to prove. $\endgroup$ – Integral Nov 14 '15 at 17:58
  • $\begingroup$ Can you have the same argument by replacing $C(B_{X^*})$ with $X^{**}$? $\endgroup$ – Hashimoto Feb 17 at 4:46

I have two comments.

  1. I'm not sure that $\Phi$ is surjective. However, since you don't want to prove that $X$ is isometrically isomorphic to $\ell_\infty$ (but to a subspace of $\ell_\infty$), you can replace "linear bijection" by "linear injection".

  2. Every separable set has a countable norming set (Lemma 6.7 here). And any normed space with a countable norming set is isometric to a subspace of $\ell_\infty$ (Corollary 6.8 here). Thus, the completeness (that seems to be not used in your argument) is not really needed.

So, your solution looks fine to me.

  • $\begingroup$ Thank you for your observations! $\endgroup$ – Integral Nov 19 '15 at 1:27

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.