1
$\begingroup$

Rosenthal introduce the notion of the extendable local reflexivity for Banach spaces as follows: Let $X$ be a Banach space and let $\lambda\geq 1$. $X$ is said to be $\lambda$-extendably locally reflexive ($\lambda$-ELR) if for every finite-dimensional subspaces $E \subset X^{**}$ and $F \subset X^{*}$, and for every $\varepsilon > 0$, there exists a bounded linear operator $T:X^{**}\rightarrow X^{**}$ with $\|T\|\leq \lambda+\varepsilon$ such that $T(E)\subset X$ and $(Tx^{**})(x^*)=x^{**}(x^*)$ for every $x^{**} \in E$ and $x^* \in F$. A Banach space $X$ is said to be ELR if $X$ is $\lambda$-ELR for some $\lambda \geq 1$. Rosenthal, Johnson and Oikhberg proved that a Banach space $X$ has the BAP and is ELR if and only if $X^{*}$ has the BAP. Now if we replace the "$(Tx^{**})(x^*)=x^{**}(x^*)(x^{**} \in E, x^* \in F)$" in the definition by "$|(Tx^{**})(x^*)-x^{**}(x^*)| \leq \varepsilon \|x^{**}\|\|x^*\|(x^{**} \in E, x^* \in F)$", we have a formally weaker form of the ELR. Moreover, we can prove that this formally weaker form of the ELR also permits to lift the BAP from a Banach space to its dual space. My question is: Is this formally weaker form of the ELR equivalent to the ELR? The answer,at least, is yes whenever the Banach space $X$ has the BAP.

$\endgroup$
2
  • $\begingroup$ @BillJohnson. I am sorry that I do not reply you as soon as possible because I am too busy to log in Mathoverflow when I come back to China. I do not know how to accept your answer to keep this thread from coming to the front. Could you tell me how to do? $\endgroup$ Aug 27, 2014 at 3:18
  • $\begingroup$ Each answer has a checkmark, visible only to the OP, below it. Click the checkmark to accept the answer. $\endgroup$ Aug 28, 2014 at 4:10

1 Answer 1

1
$\begingroup$

The answer is yes by a small perturbation argument. For a small $\epsilon$, enlarge $F$ so that $F$ $1+\epsilon$-norms $E$. Get $T$ from the condition. Choose any Auerbach basis $u_k$ for $E$ and take biorthogonal functions $x_k^*$ to $u_k$ with $x_k^*$ in $F$. So $\|u_k\|=1$ and by the norming condition you can have $\|x_k^*\| \le 1 + \epsilon$. There are $x_k$ in $X$ biorthogonal to $x_k^*$ with $\|x_k\| \le (1 +\epsilon)^2$ (for example, by usual local reflexivity, or just by Helly's theorem). Define $S:X^{**} \to X$ by $$ Sx^{**} = \sum_k \langle x^{**},x_k^*\rangle (1-\langle x_k^*, Tu_k \rangle ) x_k. $$

Then $S$ has small norm if $\epsilon$ times the dimension of $E$ is small. Replace $T$ with $T+S$.

EDIT, 7/20/14. I wrote down the wrong expression for $S$. Rather than giving the most natural correct $S$, I'll explain in a more conceptual manner how to get $S$. Having gotten $u_k$, $x_k^*$, and $x_k$ as above, you define $S$ on the span of $u_k$ so that for each $k$,

$$ Su_k = (1-\langle x_k^*, Tu_k \rangle ) x_k - \sum_{j\not= k} \langle x_j^*, Tu_k \rangle x_j $$

so that for all $k$ and $j$ you have $\langle (T+S)u_k, x_j^* \rangle = \langle u_k, x_j^* \rangle$. The coefficients of each $x_j$ in the expression for $S$ are of order $\epsilon$, so, since $u_k$ is Auerbach and $x_k$ is almost Auerbach, The norm of $S$ on $E$ is of order at most $\epsilon N$, where $N$ is the dimension of $E$. So $S$ can be extended to an operator from $X^{**}$ to $SE$ having norm of order at most $\epsilon N^{1/2}$.

$\endgroup$
8
  • 1
    $\begingroup$ I am not sure that the operator $T+S$ satisfies $<(T+S)x^{**},x^{*}>=<x^{**},x^{*}>(x^{**}\in E, x^{*}\in F)$ and the norm of $x^{*}_{k}$ is less than and equal to 1+ϵ . $\endgroup$ Jul 20, 2014 at 16:47
  • $\begingroup$ The norm condition on $x_k^*$ follows from the fact that $F$ almost norms $E$, which means that the natural mapping from $F$ to $E^*$ is almost (i.e., up to $1+\epsilon$) a quotient map. $\endgroup$ Jul 20, 2014 at 20:27
  • $\begingroup$ I gave an incorrect formula for $S$ and so have made an edit to my answer. $\endgroup$ Jul 20, 2014 at 20:28
  • $\begingroup$ The equality $<(T+S)u_{k},x^{*}_{j}>=<u_{k},x^{*}_{j}>$ only implies that $<(T+S)x^{**},x^{*}>=<x^{**},x^{*}>(x^{**}\in E, x^{*}\in span\{x^{*}_{k}:k=1,2,...n\})$. Is $(x^{*}_{k})_{k=1}^{n}$ a basis for $F$ ? $\endgroup$ Jul 21, 2014 at 21:07
  • $\begingroup$ You are right. I was trying to do it so that you can get estimates depending on the dimension of $E$. Forget that and just enlarge $F$ and $E$ so that each is total over the other and choose $u_k$ Auerbach for $E$ and the unique biorthogonal $x_k^*$ in $F$, which must span $F$. Now you have no good estimate on the norms of the $x_k^*$. That does not matter because $\epsilon $ can be chosen small enough to compensate. $\endgroup$ Jul 22, 2014 at 1:03

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.