11
$\begingroup$

Let $M$ be a closed subspace of $l^\infty$. Suppose that the quotient $l^{\infty}/M$ is isomorphic to $l^\infty$. Is it true that $M$ is complemented in $l^\infty$?

$\endgroup$
3
  • 4
    $\begingroup$ The "predual" version of this question -- namely, "if a subspace of $\ell^1$ is isomorphic to $\ell^1$, is it complemented in $\ell^1$? -- has a negative answer, see J. Bourgain, A counterexample to a complementation problem, numdam.org/item?id=CM_1981__43_1_133_0 I therefore suspect that the answer to your question is negative but I'm not sure. $\endgroup$
    – Yemon Choi
    Aug 27, 2013 at 16:25
  • 2
    $\begingroup$ By a result of Lindenstrauss, the OP really asks whether $M$ is isomorphic to $\ell_\infty$. $\endgroup$ Aug 27, 2013 at 16:28
  • 1
    $\begingroup$ If you are happy with @BillJohnson's answer then you should click the tick mark to "accept it". $\endgroup$
    – Yemon Choi
    Sep 3, 2013 at 4:00

1 Answer 1

11
$\begingroup$

Notice that the comment by Yemon gives a negative answer to the question. Bourgain constructs a short exact sequence

$0 \to \ell_1 \to \ell_1 \to X \to 0$

that does not split. The space $X$ is not $\mathcal{L}_1$ for then (since the kernel of the quotient map is a dual space and the quotient mapping onto any $\mathcal{L}_1$ space locally lifts) the quotient map would lift by a classical result of J. Lindenstrauss. Dualizing the diagram, you get a quotient mapping from $\ell_\infty$ onto $\ell_\infty$ whose kernel is not $\mathcal{L}_\infty$ and thus is not complemented in $\ell_\infty$.

Incidentally, from such an $\ell_\infty$ example you can deduce the $\ell_1$ result by localization; that is, modulo standard localization arguments, your question is equivalent to the question Bourgain answered. The moral is "don't look for an easy proof".

$\endgroup$
6
  • 1
    $\begingroup$ Thanks Bill: I'd forgotten the Lindenstrauss lifting result, which is why I couldn't see that X was not ${\mathcal L}^1$ $\endgroup$
    – Yemon Choi
    Aug 27, 2013 at 18:22
  • $\begingroup$ I'm a bit confused: doesn't Bourgain "only" construct a short exact sequence $0 \to \ell_1 \to L_1 \to X \to 0$? [This also yields the answer to the question by taking duals, using that $L_\infty$ and $\ell_\infty$ are isomorphic.] If I understand your last paragraph correctly, one can infer a short exact sequence $0 \to \ell_1 \to \ell_1 \to Y \to 0$ from this, but Bourgain doesn't seem to spell that out in his paper. $\endgroup$
    – Martin
    Aug 27, 2013 at 23:51
  • 1
    $\begingroup$ Martin, you have to localize Bourgain's theorem. You get from it that for every $n$ there is a finite dimensional subspace $E_n$ of $L_1$ with $E_n$ $C$-isomorphic to $\ell_1^{m_n}$ ($m_n$ is the dimension of $E_n$) and any projection from $L_1$ onto $E_n$ has norm at least $n$ (see next comment). Here $C$ is a constant independent of $n$. You get a superspace $F_n$ of $E_n$ in $L_1$ which is, say, $2$-isomorphic to $\ell_1^{k_n}$. Now take the $\ell_1$ sum of $E_n$ in the $\ell_1$ sum of $F_n$ to get an uncomplemented copy of $\ell_1$ in $\ell_1$. $\endgroup$ Aug 28, 2013 at 0:18
  • 1
    $\begingroup$ Here is how you get the $E_n$. Take $E$ an uncomplemented copy of $\ell_1$ in $L_1$ and let $G_n$ be the span in $E$ of the first $n$ unit vector basis elements. Suppose you have uniformly bounded projections $P_n$ from $L_1$ onto $G_n$. Endow $E=\ell_1$ with its weak$^*$ topology, which makes the unit ball compact and pass to an ultra limit of $P_n$. This gives a projection from $L_1$ onto $E$. $\endgroup$ Aug 28, 2013 at 0:26
  • 1
    $\begingroup$ Bourgain did not give this argument because it was well known to experts and he hates (or at least hated at that time) to write more than is absolutely necessary. $\endgroup$ Aug 28, 2013 at 0:32

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.