11
$\begingroup$

Assume that $X$ is a separable Banach space and $Y$ a closed subspace such that $Y$ and $X/Y$ are hereditarily indecomposable (HI). The general question is what is the possible structure of $X$. Obviously $X$ could be a HI space. Also $X$ could be the direct sum of $Y$ and $Z$ with both infinite dimensional HI subspaces.

Question I : Are the two alternatives the only possible answers?

Note that the space $X$ is always HI saturated (i.e., every infinite-dimensional closed subspace contains a HI subspace).

A weaker version of the above question is the following

Question * : Assume that $X$ is indecomposable is then HI.

Question II : If the second alternative occurs and W a closed subspace of $X$ such that $W$ and $X/W$ are HI, is $W$ essentially isomorphic to $Y$ or $Z$ (i.e. there are further subspaces of finite codimension which are isomorphic ) and $X/W$ essentially isomorphic to the other one? Namely the pair $W$, $X/W$ admits two alternatives when both are HI.

$\endgroup$
4
  • 1
    $\begingroup$ Something that is probably obvious for you but was not to me is that, under the given hypothesis on $X$, the space $X$ is HI if and only if the quotient mapping onto $X/Y$ is strictly singular. In other words, if $X$ is not HI, then there is an infinite dimensional subspace $W$ of $X$ s.t. the unit sphere of $W$ is of positive distance to $Y$. $\endgroup$ Commented Aug 22, 2019 at 20:55
  • $\begingroup$ Thanks Bill! Although it is rather obvious I did not have such a statement in my mind. $\endgroup$
    – S Argyros
    Commented Aug 23, 2019 at 7:57
  • 1
    $\begingroup$ Moreover if X is HI and Y of infinite dimension then the quotient map Q is necessarily strictly singular, This is a consequence of the following property of HI spaces. If X is HI then every bounded linear operator T from X to some space Y is either strictly singular or it is an essential embedding of X into Y. This is probably useful for the question II since W is essentially embedding either in Y or Z. but I did not see how to get something more. $\endgroup$
    – S Argyros
    Commented Aug 23, 2019 at 8:33
  • 2
    $\begingroup$ Let me clarify the following. When I saw Bill's comment I did not read the " only if" in th statement and I consider that it was only the "if". I noticed that when I read it again. The "only if" is not obvious as I explained in my second comment. $\endgroup$
    – S Argyros
    Commented Aug 23, 2019 at 12:41

2 Answers 2

3
$\begingroup$

The answer to Question II is negative. In particular the following general holds.

Theorem: Every separable HI space $Y$ is the quotient of the direct sum of two HI spaces $X$, $Z$ such that $Y$ is not essentially isomorphic to any of $X$, $Z$ and the kernel of the quotient map $Q$ is HI.

This result is a consequence of the known theorem that every separable Banach space $Y$ is the quotient of a separable HI space $X$ such that the quotient map is strictly singular.

Given this result we start with $Y$ a separable HI space and set $X$ the HI space that is mapped onto $Y$ through a strictly singular map $Q$. Next take a $Z$ subspace of $Y$ with infinite dim and codim. Consider the map $Q'$ that sends $(x,z)$ to $Qx+z$ which is onto the space $Y$ from the direct sum of $X$ and $Z$.

The space $W = \mathrm{Ker}Q'$ consists of the pairs $(x,z)$ satisfying the equation $z = -Qx$. The space $W$ is HI. Indeed in every $V$ subspace of $W$ there exists a normalized element $(x,z)$ with norm of $z$ very small. Therefore $(x,z)$ is very close to the space $X$ which is HI. Hence $W$ is also HI.

$\endgroup$
3
  • $\begingroup$ A correction: The known result concerns separable spaces not containing $ l_1 $ $\endgroup$
    – S Argyros
    Commented Aug 29, 2019 at 12:15
  • $\begingroup$ The HI property of W does not depend on the strict singularity of the map Q. indeed if (x,z) is in W then || x || > c || (x,z) || since Q is bounded. This yields that $P_X $ is an isomorphism on W. $\endgroup$
    – S Argyros
    Commented Aug 29, 2019 at 14:06
  • $\begingroup$ I wonder if the following reformulation of Question II has a positive answer.The new question has three alternatives. The two mentioned in Q. II and the additional one that X/W is a quotient either of Y or Z. There two evidences supporting such a case.Setting $ P_Y $, $ P_Z $ the corresponding projections onto Y, Z , since W is HI at least one of them is essential isomorphism on W. Further if $P_Y$ has this property then the quotient map Q from X onto X/W is an essential isomorphism on Z. Namely these remarks remind the construction presented in the answer of QII. $\endgroup$
    – S Argyros
    Commented Aug 30, 2019 at 11:22
2
$\begingroup$

There exists a separable reflexive Banach space $\hat X^*$ which is indecomposable but not HI, and admits a subspace $Y$ such that both $Y$ and $\hat X^*/Y$ are HI. This example provides a negative answer to Question I and Question $^*$.

The space $\hat X^*$ is obtained in Proposition 23 of a paper of Valentin Ferenczi [CJM-99], using the spaces provided by Proposition 25 in the same paper.

Note that $\hat X^*$ is not HI by construction, and it is indecomposable because its predual $\hat X$ is HI and reflexive (see Proposition 23).

A Banach space $X$ is called QHI if every infinite dimensional quotient of each subspace of $X$ is indecomposable. Since a quotient of a subspace is a subspace of a quotient, subspaces and quotients of a QHI space are QHI. Moreover, if $X$ is reflexive and QHI then the dual $X^*$ is QHI.

Proposition 25 provides two reflexive QHI Banach spaces $X_1$ and $X_2$ with a ``common" subspace $Z$ (some additional conditions are also satisfied), and the predual of $\hat X^*$ is defined in Proposition 23 as $\hat X =(X_1\oplus X_2)/D$ with $D=\{(z,-z): z\in Z\}$.

The expression $Jx_1 =(x_1,0)+D$ defines an embedding of $X_1$ into $\hat X$ such that $\hat X/J(X_1)$ is a quotient of $X_2$: $$ \hat X/J(X_1) = (X_1\oplus X_2)/(X_1+D)\simeq \frac{X_2}{(X_1+D)\cap X_2}. $$ Thus both $J(X_1)$ and $\hat X/J(X_1)$, as well as their dual spaces, are QHI.

Taking as $Y$ the annihilator $J(X_1)^\perp \equiv (\hat X/J(X_1))^*$, both $J(X_1)^\perp$ and $\hat X^*/J(X_1)^\perp$ are QHI, hence they are HI.

$\endgroup$
8
  • $\begingroup$ I posted all the questions expecting a possible positive answer. After thinking more Question* I convinced myself that the answer is negative and I have a possible example different from what you suggests which at the moment I do not understand. I have to look at Valentin' s paper. $\endgroup$
    – S Argyros
    Commented Sep 2, 2019 at 21:40
  • $\begingroup$ I have added some additional details. I hope the argument is more clear now. $\endgroup$ Commented Sep 3, 2019 at 6:46
  • $\begingroup$ Please avoid math mode for non-math (namely abbrevations). You can italicize using the asterisk symbol. $\endgroup$
    – YCor
    Commented Sep 3, 2019 at 7:32
  • $\begingroup$ In the statement of Prop. 23 and Prop. 25 the spaces $X_1 $, $X_2 $ are HI and not QHI. I suppose that the QHI property follows by the construction of the spaces and the earlier results of the paper. If this is correct then the answer in the question* is clear to me. $\endgroup$
    – S Argyros
    Commented Sep 3, 2019 at 8:53
  • $\begingroup$ The spaces $X_1$ and $X_2$ constructed in Prop. 25 are QHI and satisfy the hypothesis of Prop. 23, where only HI is required, but we need the spaces to be QHI to prove that $Y$ and $\hat X^*/Y$ are HI. $\endgroup$ Commented Sep 3, 2019 at 9:47

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .