The set of $\psi$-bounded vectors for a vector state $\psi$ on a von Neumann algebra

51 Views Asked by At

Let $\mathfrak{M}$ be a von Neumann algebra acting on a Hilbert space $\mathcal{H}$, and let $\psi$ be a positive normal linear functional on $\mathfrak{M}$. According to [1, p. 69] the so-called lineal of $\psi$ (or set of $\psi$-bounded vectors) is defined to be $$D(\mathcal{H}, \psi) := \{x \in \mathcal{H} \ : \ \Vert ax \Vert_\mathcal{H} \le C_x \psi(a a^\ast) \ \text{for all $a \in \mathfrak{M}$}\} \ .$$ The book [1] continues with the observation that if there exists a vector $\varPsi \in \mathcal{H}$ such that $\psi(a) = \langle \varPsi, a \varPsi \rangle$ for all $a \in \mathfrak{M}$, that is, if $\psi$ is a vector state, then the lineal of $\psi$ takes the form $$D(\mathcal{H}, \psi) = \mathfrak{M}^\prime \varPsi = \{a^\prime \varPsi \ : \ a^\prime \in \mathfrak{M}^\prime\} \ ,$$ where $\mathfrak{M}^\prime$ denotes the commutant of $\mathfrak{M}$.

I have two questions about this:

  • Q1: Is the definition of $D(\mathcal{H}, \psi)$ given above indeed the most useful one? To me, it would seem more natural to set $D(\mathcal{H}, \psi) := \{x \in \mathcal{H} \ : \ \Vert ax \Vert_\mathcal{H}^2 \le C_x \psi(a^\ast a) \ \text{for all $a \in \mathfrak{M}$}\}$.

  • Q2: How can one prove the assertion that $D(\mathcal{H}, \psi) = \mathfrak{M}^\prime \varPsi$ for a vector state $\psi$? It is clear to me that $\mathfrak{M}^\prime \varPsi \subset D(\mathcal{H}, \psi)$, but I have trouble showing the reverse inclusion $D(\mathcal{H}, \psi) \subset \mathfrak{M}^\prime \varPsi$.

(Regarding the second set inclusion in Q2, my initial (maybe naive) idea was trying to show that $x \in D(\mathcal{H}, \psi)$ can be written as $\sqrt{C_x} \varPsi$, where $C_x > 0$ is as in the definition of the lineal of $\psi$.)

I would greatly appreciate any insights!


[1] Ohya, M. and Petz, D. Quantum Entropy and Its Use. Corrected Second Printing. Springer, Berlin, Heidelberg, 2004.

1

There are 1 best solutions below

1
On BEST ANSWER

Here's an outline how to do this. I'll let you check some of the detail.

Let $x\in D(\mathcal{H}, \psi)$ with $\psi(a) = \langle \varPsi, a \varPsi \rangle$. So $\forall a\in\mathfrak{M}: \Vert ax\Vert \le \sqrt{C_x}\Vert a\varPsi\Vert$.

Let $$H_\varPsi = \overline{\{a\varPsi: a\in \mathfrak{M}\}}$$ $$H_x = \overline{\{ax: a\in \mathfrak{M}\}}$$ Then $H_\varPsi, H_x$ are closed subspaces of $\mathcal{H}$. Since $\mathfrak{M}$ contains bounded operators, it is a routine matter to check that $H_\varPsi, H_x$ are $\mathfrak{M}$-invariant subspaces. Moreover, since $\mathfrak{M}$ contains the adjoint of any member of it - the orthogonal complements $H_\varPsi^\perp, H_x^\perp$ are also $\mathfrak{M}$-invariant. ($\varPsi, x$ are called cyclic vectors for $H_\varPsi, H_x$ respectively.)

Now define a linear operator $T$ on $\{a\varPsi: a\in \mathfrak{M}\}$ by $$ T(a\varPsi) = ax\in H_x$$

For start we need to check that $T$ is well-defined on this subspace. If $a\varPsi=0$ then $ax=0$ because of the inequality $\Vert ax\Vert \le \sqrt{C_x}\Vert a\varPsi\Vert$. I'll leave it to you to check that $T$ is a linear map and so it is well defined. Moreover we see that $\Vert T\Vert \le \sqrt{C_x}$ on the subspace $\{a\varPsi: a\in \mathfrak{M}\}$.

Since $T$ is a bounded linear operator, it may be extended uniquely (via Cauchy sequences) to $H_\varPsi$, and it still maps into $H_x$ on this larger domain because $H_x$ is closed.

Finally we extend $T$ to $\mathcal{H}$ by defining $T(v)=0$ for $v\in H_\varPsi^\perp$.

For $\tilde{a}\in \mathfrak{M}$ and $v\in H_\varPsi^\perp$, $\tilde{a}v\in H_\varPsi^\perp$, so $Tv = 0, \tilde{a}Tv = 0, T\tilde{a}v = 0$. In other words $\tilde{a}$ commutes with $T$ on $H_\varPsi^\perp$.

I'll leave it to you to show that $\tilde{a}$ commutes with $T$ on $H_\varPsi$ (starting from the subspace $\{a\varPsi: a\in \mathfrak{M}\}$ of course.)

So... $T\in \mathfrak{M}^\prime$ and ... $T(\varPsi) = x$. $\square$

PS: The spaces $H_\varPsi, H_x$ are isomorphic to the spaces constructed by the GNS construction from the vector states associated with $\varPsi, x$.