I do not believe the argument below is correct, but I am having quite a bit of trouble finding where I went wrong with this, so perhaps someone with more expertise in this area can push me in the correct direction.

It seems we may prove the following: Let $X$ be a Banach space and suppose every separable quotient of $X$ is reflexive. Then $X$ is a Grothendieck space.

Proof: Fix a separable Banach space $Y$ and $T \in B(X,Y)$. Then $Z \doteq X/\ker T$ is a separable quotient of $X$, hence is reflexive. There is a unique bounded linear operator $\hat{T} \in B(Z,Y)$ such that $T= \hat{T} \circ \pi$ (where $\pi: X \to Z$ is the canonical projection). Note that since $Z$ is reflexive, thus Grothendieck, and $Y$ is separable, $\hat{T}$ is weakly compact.

But then it follows that $T$ is weakly compact. Indeed, let $C \subseteq X$ be norm-bounded. Then $\pi(C)$ is as well, so $T(C)=\hat{T}(\pi(C))$ is weakly compact in $Y$. Since $T$ and $Y$ were arbitrary, it follows that $X$ is Grothendieck. $\Box$

(The converse of the above holds, but this is not important right now). Now here is the source of my skepticism. If the above is true, we can go a bit further: Let $X$ be Banach. Then $X$ has a separable quotient of infinite dimension.

$\textit{Proof:}$ If $X$ is reflexive we are done, so assume $X$ is not reflexive. Further, if $X$ has a separable quotient that is not reflexive, then we are done since such a quotient must be infinite dimensional. We may therefore assume $X$ is a non-reflexive space such that every separable quotient of $X$ is reflexive. Then $X$ is a non-reflexive Grothendieck space by the above.

We now note that $X^*$ must contain a copy of $\ell_1$. Indeed, suppose not. Then by a result presented in the abstract here or here, any bounded sequence $(\phi_n)$ in $X^*$ has a weakly Cauchy subsequence $(\phi_{n_i})$. This subsequence is weak-* convergent, thus weakly convergent since $X$ is Grothendieck. It follows immediately that the unit ball in $X^*$ is weakly sequentially compact, hence weakly compact, so $X$ is reflexive, a contradiction.

But if $X^*$ contains a copy of $\ell_1$, then $X$ admits an infinite dimensional, separable quotient (see the answer given here by @Bill Johnson). This concludes the proof. $\Box$

I suspect the error is in the first claim, but I've looked at this over a period of a few days and I can't find anything. Any input would be much appreciated.

  • $\begingroup$ Why is it true that "if $X$ is reflexive we are done"? $\endgroup$
    – Yemon Choi
    Jan 5, 2018 at 2:56
  • $\begingroup$ If $X$ is reflexive, then $X$ is the dual of a Banach space. It is a theorem of Argyros, Dodos and Kanellopolous that every dual of a Banach space has separable quotient: users.uoa.gr/~pdodos/Publications/13-Unconditional.pdf $\endgroup$
    – M10687
    Jan 5, 2018 at 3:19
  • 5
    $\begingroup$ My previous comment had an embarrassing mistake, and I now think that your first claim is not justified (as you yourself suspected). The point is that althoug $X/\ker T$ injects continuously into something separable, that doesn't necessarily mean it is separable. Consider $\ell_\infty \to c_0$ obtained by multiplying the $n$th entry with $1/n$. $\endgroup$
    – Yemon Choi
    Jan 5, 2018 at 4:49
  • 2
    $\begingroup$ @YemonChoi thank you, that certainly exposes the error. $\endgroup$
    – M10687
    Jan 5, 2018 at 5:12
  • $\begingroup$ You may find this useful arxiv.org/abs/2102.03838 $\endgroup$ Aug 15, 2021 at 17:26

1 Answer 1


(Posting this comment as an answer, just so the question doesn't still show up as "unanswered".)

Your first claim is not justified (as you yourself suspected). If $T:X\to Y$ has separable range, this implies $X/\ker T$ injects continuously into something separable — but this doesn't necessarily imply $X/\ker T$ is itself separable. (Consider $\ell_\infty \to c_0$ obtained by multiplying the $n$th entry with $1/n$.)


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.