There are 'canonical' examples of maps on operator spaces which are not completely bounded. Nevertheless, I couldn't produce any examples of bounded projections on relatively easy to understand operator spaces which are not c.b. In particular, I failed this task for $\mathcal{K}(H)$ and $\mathcal{B}(H)$, where $H$ is a Hilbert space (I tried projections onto certain subspaces isomorphic to $c_0$ and $\ell_\infty$, respectively).

I would appreciate any examples.


Symmetrisation (or anti-symmetrisation).

That is: let $T:K(H)\to K(H)$ be the transpose map and let $P=({\rm id}+T)/2$. Then $P:K(H)\to K(H)$ is a projection onto the subspace of symmetric [NOT self-adjoint] compact operators. Since $T$ is not completely bounded, $P$ is not completely bounded.

This argument also shows that the symmetrisation map $M_n\to M_n$ is a norm-one projection which has cb norm $\geq (n-1)/2$.

The examples you tried fail to work because of a general result: any bounded linear map into a minimal operator space will be automatically completely bounded. This is somewhere in the first third of Effros-Ruan's book, for example, though I don't have my copy here.

| cite | improve this answer | |
  • $\begingroup$ Thanks. How about (Banach) complemented subspaces (Banach) isomorphic to $\mathcal{K}(H)$ or $\mathcal{B}(H)$? $\endgroup$ – Olaf Kummers Apr 18 '12 at 6:41
  • $\begingroup$ @Olaf: without thinking about it too hard, I can at least point you to a result of Blower mentioned here mathoverflow.net/questions/12211/… which implies, in particular, that either $P(B(H))$ or $(I-P)(B(H))$ is isomorphic as a Banach space to $B(H)$. I rather suspect both are, but don't see at this moment how one would show this. $\endgroup$ – Yemon Choi Apr 18 '12 at 7:09
  • $\begingroup$ What do you mean both? One of them can be e.g. $\ell_\infty$. $\endgroup$ – Olaf Kummers Apr 18 '12 at 8:43
  • $\begingroup$ @Olaf: I meant the particular P in my answer, I thought that was obvious from context $\endgroup$ – Yemon Choi Apr 18 '12 at 10:03
  • $\begingroup$ Sure, sorry. :) $\endgroup$ – Olaf Kummers Apr 18 '12 at 10:53

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.