The following essentially implies the equivalence of Anantharaman-Delaroche's compact approximation property (page 337 of Link) and the Haagerup approximation property.

Let $M$ be a type ${II}_{1}$ factor with trace $\tau$. Let $\Omega$ denote the standard unit cyclic trace vector in $L^{2}(M)$ (associated to the element $1\in M$). If $\phi:M \rightarrow M$ is a normal completely positive map, we naturally associate an operator $T_{\phi}\in B(L^{2}(M))$ extending $T_{\phi}(x\Omega)=\phi(x)\Omega$.

If the map $x\mapsto \phi(x) \Omega$ is a compact linear map from $M$ with the operator norm into $L^{2}(M)$, is the operator $T_{\phi}$ compact?


1 Answer 1


(I removed my first answer as it contained an egregious mistake, pointed out by Yemon; here's a second attempt)

I think that $T_\phi$ may fail to be compact.

Fix a sequence of projections $\{p_k\}$ in $M$, pairwise orthogonal, with $\tau(p_k)=2^{-k}$ ($\tau$ the trace in $M$) and sum 1. Now define $$ \phi:M\to M,\ \ \ \mbox{ given by } \phi(x)=\sum_k 2^k\tau(xp_k)p_k. $$ (the series converges strongly because all of its terms are positive and any partial sum is bounded by $\|x\|$). This map is ucp (it is an infinite sum of cp), and since it commutes with $\tau$, it is normal. Let us also define the maps $$ \phi_n=\sum_{k=1}^n 2^k\tau(xp_k)p_k. $$ The maps $x\mapsto \phi_n(x)\Omega$ are all finite-rank.

If $\|x\|\leq1$, we have, using that the set $\{p_k\}$ is orthogonal in $L^2(M)$, $$ \left\|\phi(x)-\phi_n(x)\right\|_2^2=\left\|\sum_{k>n}2^k\tau(xp_k)p_k\right\|_2^2 =\sum_{k>n}|\tau(xp_k)|^2\leq\sum_{k>n}\tau(p_k)^2\leq\frac1{3\times 4^n} $$ This shows that $\|\phi-\phi_n\|<4^{-n}$ in $B(L^2(M))$: so the map $x\mapsto \phi(x)\Omega$ is compact.

Now consider the orthonormal set $\{2^kp_k\}$ in the unit ball of $L^2(M)$. Since $\phi(p_k)=p_k$, we get that $T_\phi(2^kp_k)=2^kp_k$; so the range of $T_\phi$ contains an orthonormal set: $T_\phi$ is not compact.

  • 1
    $\begingroup$ I don't think that the unit ball of L^\infty[0,1] is pre-compact as a subset of L^2[0,1] -- consider the sequence $e_n(t) =\exp(2\pi int)$. (This was pointed out to me, after a talk I gave, when I was momentarily hung up on an example of a non-compact map from $L^\infty[0,1]$ to $L^2[0,1]$. The map is completely continuous and weakly compact, though.) $\endgroup$
    – Yemon Choi
    Oct 9, 2010 at 5:00
  • $\begingroup$ Yeah, I could have thought of that example. Thanks, Yemon! $\endgroup$ Oct 9, 2010 at 5:20
  • $\begingroup$ Will find time to check, Martin. Thanks for updating! Thanks to Yemon for coming back to this ancient thing and finding the bug! (I had a weird premonition to come and check MO today for some reason...good thing I did!) $\endgroup$
    – Jon Bannon
    Jan 23, 2020 at 17:24
  • $\begingroup$ It's been a while :D $\endgroup$ Jan 23, 2020 at 17:26
  • 1
    $\begingroup$ No worries. It's been a while, though, this is my second answer ever on the SE network! $\endgroup$ Mar 1, 2021 at 17:31

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.