1
$\begingroup$

Let $H$ be a (separable) Hilbert space and $B(H)$ be the space of bounded linear operators defined on $H$. Suppose $\{ e_i \}$ be an orthonormal basis of $H$, we say $T \in B(H)$ is a trace-class operator if

$$ \|T\|_1 := \sum_i \langle |T| e_i, e_i \rangle < \infty, $$

where $|T|$ is the positive operator such that $|T|^2 = T^*T$ and $T^*$ is the adjoint operator of $T$. If $T \in B(H)$ is a trace-class operator, we can define its trace as

$$ \newcommand{\Tr}{\mathrm{Tr}} \Tr(T) := \sum_i \langle Te_i, e_i \rangle. $$

From Wikipedia, or Conway's book A course in functional analysis, we know that trace-class operators have some nice properties:

  • If $T \in B(H)$ is a trace-class operator, then for any $A \in B(H)$, we know both $AT$, $TA$ are trace-class, and $\Tr(AT) = \Tr(TA)$, and $\|AT\|_1 \leq \|A\| \|T\|_1$.

However, I'm curious about the definition of trace-class operators between two different Hilbert spaces $H_1$ and $H_2$. Let $B(H_1, H_2)$ be the space of bounded linear operators from $H_1$ to $H_2$.

It seems the above definition can be generalized to this case as follows: We say an operator $T \in B(H_1, H_2)$ is trace-class if $\sqrt{T^*T} : H_1 \to H_1$ is trace-class and define its trace-norm as $\|T\|_1 := \| \sqrt{T^*T} \|_1$. This generalization looks reasonable to me, but I am not sure whether there exist some subtleties.

I am trying to prove the aforementioned properties of trace-class operators under such definition. Suppose $H_3$ is another Hilbert space, and $A \in B(H_2, H_3)$. To show that $AT$ is trace-class, by definition it suffices to show that $\sqrt{T^* A^* AT}$ is trace-class. But I don't know how to proceed.

I guess that the definition of trace-class operators can be generalized and the aforementioned properties also hold. I would appreciate it if you could provide some references about these.

(BTW: It seems that the Hilbert-Schmidt operator can be easily generalized to different Hilbert spaces.)

Update: Thanks for the answer of Ruy. There is another question about these trace-class operators: When $A \in B(H_2, H_1)$, can we conclude that $\Tr(AT) = \Tr(TA)$?

$\endgroup$

1 Answer 1

1
$\begingroup$

Notice that $$ |AT|^2 = T^*A^*AT \leq \|A\|^2T^*T = \|A\|^2|T|^2, $$ so, by Proposition 1.3.8 in [1] (square-root is an operator monotone function), one has that $$ |AT| \leq \|A\||T|, $$ and it follows that $AT$ is trace class according to the OP's definition.

[1] Pedersen, Gert K., C*-algebras and their automorphism groups, London Mathematical Society Monographs. 14. London - New York -San Francisco: Academic Press. X, 416 p. $ 60.00 (1979). ZBL0416.46043.

$\endgroup$
2
  • $\begingroup$ I was not aware the operator monotonicity of square-root. Thanks for pointing out it. I'm also curious whether $\mathrm{Tr}(AT) = \mathrm{Tr}(TA)$ holds when $H_3 = H_1$. Do you know some textbooks/papers about it? $\endgroup$
    – Barth2131
    Mar 22, 2021 at 2:55
  • $\begingroup$ It you consider all operators in sight as acting on $H_1\oplus H_2$, then $$ \pmatrix{0 & A \cr 0 & 0}\pmatrix{0 & 0 \cr T & 0}=\pmatrix{AT & 0 \cr 0 & 0} $$ while, $$ \pmatrix{0 & 0 \cr T & 0}\pmatrix{0 & A \cr 0 & 0}=\pmatrix{0 & 0 \cr 0 & TA} $$ so, yes, $\text{tr}(AT)=\text{tr}(TA)$. Kadison and Ringrose has a lot about traces in the second volume. $\endgroup$
    – Ruy
    Mar 22, 2021 at 3:52

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.