3
$\begingroup$

It is known that weak convergence implies norm convergence in $\ell^1(\mathbb{N})$, see e.g. here.

Because of the typical analogies of the Schatten ideals $C_p \subset B(H)$ (where $H$ is a Hilbert space), it seems natural to ask whether weak convergence in $C_1$ (the ideal of trace-class operators) already implies convergence with respect to the trace-class norm. (This question has been asked on math.stackexchange by another user but didn't receive an answer, see here).


It is known that weak convergence $T_n \rightarrow T$ implies strong convergence for a sequence in $C_1$ if one additionally knows that $\|T_n\|_1 \rightarrow \|T\|_1$, see here.

It seems to me that completely dropping the restriction $\|T_n\|_1 \rightarrow \|T\|$ might be a little too much to ask, even though it is possible in $\ell^1(\mathbb{N})$.

However, if the answer to the above question is negative, then one can ask if one can do better:

  • For example, is it sufficient to just have a uniform bound on $\|T_n\|$ (the operator norm) or $\|T_n\|_1$ (the trace class norm)?
  • Is it sufficient to know that $\mathrm{tr}(T_n) \rightarrow \mathrm{tr}(T)$ (this was claimed by user Folkmar Bornemann in this post)?

(Of course, all additionally to knowing that $T_n\rightarrow T$ in the weak topology.)

$\endgroup$

2 Answers 2

5
$\begingroup$

It's still false, however. Let $T_n: v \mapsto \langle v, e_1\rangle e_n$ where $(e_n)$ is an orthonormal basis of $H$. Then $\|T_n\|_1 = 1$ for all $n$, but the sequence converges weakly to zero. For any $A \in B(H)$, the sequence ${\rm Tr}(AT_n)$ reads off the entries of the first row of $A$, which has to lie in $l^2$ and hence goes to zero.

This example also falsifies the two suggested weakenings --- $\|T_n\|$ and $\|T_n\|_1$ are both constantly 1, and ${\rm Tr}(T_n)$ is constantly zero (after the first term).

$\endgroup$
1
$\begingroup$

We can assume $T=0$. Let $H$ be a HIlbert space with orthonormal basis $(e_1,e_2,\dots)$. For $j\in{\mathbb N}$ let $T_j:H\to H$ be given by $T_j e_k=\delta_{j,k}e_k$ (Kronecker-delta). Then $T_j$ is trace class of trace norm one, but the sequence $T_j$ tends to zero strongly (and hence weakly).

Also, insisting $tr(T_j)\to 0$ does not help for you can replace $T_j$ with $S_j$ where $S_je_k=e_k$ if $k=j$, $S_je_k=-e_k$ if $k=j+1$ and zero otherwise.

$\endgroup$
3
  • 2
    $\begingroup$ That's not correct --- your sequence $(T_j)$ converges to zero weak operator, but not weakly. Its pairing with the identity operator doesn't go to zero. $\endgroup$
    – Nik Weaver
    Jan 21, 2016 at 13:08
  • $\begingroup$ What does that mean? What is the pairingwith the identity? $\endgroup$
    – user1688
    Jan 21, 2016 at 13:13
  • 2
    $\begingroup$ ${\rm Tr}(IT_j) = 1 \not\to 0$. The map $B \mapsto {\rm Tr}(AB)$ is a bounded linear functional on $TC(H)$, for any $A \in B(H)$. $\endgroup$
    – Nik Weaver
    Jan 21, 2016 at 13:21

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.