2
$\begingroup$

Let $F$ be a Banach space with the closed unit ball $B$. Let $E\subset F^*$ be a total subspace such that $B$ is complete with respect to the norm $|||f|||=\sup \limits_{e\in E,~e\ne 0} \frac{|\left<f,e\right>|}{\|e\|}$. Note that $|||\cdot|||$ is the norm of $F^{**}/E^{\bot}$, restricted on the image of $F$, and so my condition tells that the canonical inclusion from $F$ into $F^{**}/E^{\bot}$ is a semi-embedding.

Does it follow that $E$ is norming?

i.e. there is $r>0$ such that $|||f|||\ge r\|f\|$ for all $f\in F$.

$\endgroup$
1
  • $\begingroup$ Have you checked whether the classical examples of total non-norming subspaces give a counterexample? $\endgroup$ Aug 21, 2017 at 20:21

2 Answers 2

1
$\begingroup$

I think that the answer is "Not necessarily" and can be shown as follows. Let $F=\ell_1=\left(\oplus_{n=1}^\infty \ell_1\right)_1$. In each summand $\ell_1$ we pick a unit vector $e_n$ (which can be regarded as the first vector of the unit vector basis in that space). The space $G:=\left(\oplus_{n=1}^\infty \ell_1^{**}\right)_1$ is canonically embedded into $F^{**}$

We construct $E^\perp$ as the weak$^*$ closure of the linear span of $e_n+p_n$, where $||p_n||\downarrow 0$ and $p_n$ has nonzero component only in $n$th summand of $G$ and the component is in $(\ell_1)^{**}\backslash \ell_1$.

Now consider any non-convergent sequence $\{f_k\}$ in the unit ball of $F$. We need to show that the limit of this sequence in $|||\cdot|||$, if exists, is in the unit ball of $F$. Passing to a subsequence we may assume that it is convergent coordinate-wise and subtracting its coordinate-wise limit we may assume that the coordinate-wise limit is $0$. By approximation we may assume that the sequence $\{f_k\}$ has finite disjoint supports. We need to show that the limit of the image of such sequence in $|||\cdot|||$ can be only $0$.

This can be done as follows. We split $f_k=s_k+t_k$, where $s_k$ is supported on $\{e_n\}$ (used above) and $t_k$ is supported on the rest of the unit vector basis of $\ell_1$. But since $\{f_n\}$ are uniformly bounded, it is easy to see that $|||s_k|||\downarrow 0$. Also it is not difficult to observe that if $|||t_k|||$ does not go to $0$, then the sequence $\{f_k\}$ is not convergent in $|||\cdot|||$.

$\endgroup$
4
  • $\begingroup$ I am sorry, but could you please elaborate on the second part of the answer? Namely, why can we approximate $\{f_{k}\}$ in the way that you suggest (and it has to be with respect to $|||\cdot|||$, right?), and also the very last claim about the convergence of $|||t_k|||$? $\endgroup$
    – erz
    Aug 23, 2017 at 10:49
  • $\begingroup$ As for $\{f_n\}$ it will suffice to approximate $f_n$, say, up to $1/n$ in $||\cdot||$, the new norm is only smaller. As for estimate of the norm of $t_k$, one can show that up to some constant $|||t_k|||$ and $||t_k||$ are equivalent. All this is standard techniques, which is developed in, for example, Lindenstrauss-Tzafriri, Classical Banach spaces, volume 1. Writing this in detail should take a rather nontrivial amount of time. $\endgroup$ Aug 24, 2017 at 4:44
  • $\begingroup$ @erz Why do not you accept the answer? $\endgroup$ Nov 19, 2017 at 18:52
  • $\begingroup$ @AugustCleaner I was not able to understand it, after applying some efforts, and so decided leave this thing aside for a while $\endgroup$
    – erz
    Nov 19, 2017 at 20:15
0
$\begingroup$

It is rather common for a Banach space to have a weaker norm for which the unit ball is complete. A historically important example is the $L^1$ norm on $L^\infty$. This provides a counterexample to your conjecture---take for $E$ the space $L^\infty$ regarded as a subspace of its own dual in the usual way. This example was used by Saks (more precisely, the fact that the unit ball of $L^\infty$ has the Baire property under the $L^1$ norm) in his proof of what is now known as the Vitali-Hahn-Saks theorem.

$\endgroup$
2
  • 1
    $\begingroup$ Could you please elaborate? Isn't $L^{\infty}$ dense in $L^1$? $\endgroup$
    – erz
    Aug 21, 2017 at 12:58
  • $\begingroup$ Your answer has nothing to do with the question posed. $\endgroup$ Aug 21, 2017 at 20:19

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.