Given a Banach space $X$, we consider the space $B(X^*)$ of bounded, linear operators on $X^*$ with the weak*-topology from its canonical predual $B(X^*)_*=X^*\hat{\otimes}X$. What is $\overline{F(X^*)}^{wk*}$, the weak*-closure of the finite rank operators on $X^*$? Since this is rather vague, here are some concrete questions:

Q1: Do we always have $K(X^*)\subseteq\overline{F(X^*)}^{wk*}$, i.e., is every compact operator in the weak*-closure of finite-rank operators?

Q2: Is there a characterization when $B(X^*)=\overline{F(X^*)}^{wk*}$, i.e., when the finite-rank operators are weak*-dense?

Q3: Is there a characterization when $B(X^*)=\overline{K(X^*)}^{wk*}$, i.e., when the compact operators are weak*-dense?

Considering $B(X)$ as a subalgebra of $B(X^*)$ in the usual way, we may ask related questions in connection with $F(X)$ and $K(X)$. The principle of local reflexivity implies $\overline{F(X)}^{wk*}=\overline{F(X^*)}^{wk*}$ in $B(X^*)$. However, it is not clear to me if we always have $\overline{K(X)}^{wk*}=\overline{K(X^*)}^{wk*}$ (I guess not). Therefore, we may also ask:

Q4: Do we always have $K(X)\subseteq\overline{F(X^*)}^{wk*}$?


Regarding Q2: If and only if $X$ has the approximation property. I'll use Ryan's book "Introduction to tensor products of Banach spaces" as a reference, see Prop 4.6 (but this is all standard stuff).

Theorem: $X$ has the approximation property if and only if, whenever $u=\sum_{n\geq 1} \mu_n\otimes x_n \in X^*\hat\otimes X$ is such that $ \sum_n \mu_n(x) x_n = 0$ for all $x\in X$, then $u=0$.

As we have $\sum_n \|\mu_n\| \|x_n\|<\infty$, the condition is equivalent to $\sum_n \mu_n(x) \mu(x_n)=0$ for all $x\in X,\mu\in X^*$ (Hahn-Banach) and hence also equivalent to $\sum_n \mu(x_n)\mu_n = 0$ for all $\mu\in X^*$, and so finally also equivalent to $\sum_n \mu(x_n) f(\mu_n) = 0$ for all $\mu\in X^*, f\in X^{**}$.

This in turn is equivalent to $\langle u, F \rangle=0$ for all finite rank operators $F$ on $X^*$, under your dual pairing between $X^*\hat\otimes X$ and $F(X^*)$.

Finally, observe that $F(X^*)$ is weak$^*$ dense in $B(X^*)$ if and only if the only element of $X^*\hat\otimes X$ which annihilates all of $F(X^*)$ is $0$.

There is a related definition of the "compact approximation property". If I recall it correctly, then you can adapt the proof, and get

Thm: $X$ has the compact approximation property if and only if, whenever $u=\sum_{n\geq 1} \mu_n\otimes x_n \in X^*\hat\otimes X$ is such that $ \sum_n \langle\mu_n,T(x_n)\rangle = 0$ for all $T\in K(X)$, then $u=0$.

Thus, if $X$ has the compact approximation property, but not the approximation property (I think there is an example due to Willis) then we can find $u\in X^*\hat\otimes X$ which annihilates all of $F(X^*)$ but is not zero. There is then $T\in K(X)\subseteq K(X^*)$ with $\langle T,u\rangle \not=0$, and as $$ \overline{F(X^*)}^{wk^*} = \{ T\in B(X^*) : \langle T,u\rangle=0 \text{ for all }u\in X^*\hat\otimes X\text{ with } \langle S,u\rangle=0 \text{ for all } S\in F(X^*) \} $$ we conclude that $K(X)$ is not contained in the weak$^*$-closure of $F(X^*)$. So Q4 is a negative.


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.