Let $\mathcal{A}$ be a noncommutative $C^*$-algebra and $PS(\mathcal{A})$ be the set of its pure states.

Question 1. Is $PS(\mathcal{A})$ linearly independent (as vectors over $\mathbb{R}$)? (If $\mathcal{A}$ is commutative, $PS(\mathcal{A})$ is easily seen to be linearly independent (I learned this from Bernard Russo).)

Question 2. If $\mathcal{A}$ is unital, is $PS(\mathcal{A})$ always compact in the weak* topology? If not, is there any topology that makes $PS(\mathcal{A})$ compact?

  • $\begingroup$ Where is the commutativity of $\mathcal{A}$ used in the proof you already know? $\endgroup$ – Nate Eldredge Jul 24 '16 at 13:40
  • $\begingroup$ In what sense should $PS(\mathcal{A})$ should be "linearly indpendent"? It's not even a linear space. For example, for $\mathcal{A} = M_2(\mathbb{C})$, the pure states constitute a 2-sphere (the Bloch sphere). $\endgroup$ – Igor Khavkine Jul 24 '16 at 13:44
  • $\begingroup$ @IgorKhavkine: I assume we mean "linearly independent as a subset of the dual space $\mathcal{A}^*$." So your example seems to show that this is false. $\endgroup$ – Nate Eldredge Jul 24 '16 at 14:46
  • $\begingroup$ Thank you for your comments and a partial answer, Igor and Nate. In the commutative case, the set of pure states is the character space (= maximal ideal space) $\Omega(\mathcal{A})$ which is a locally compact Hausdorff space in the weak* topology, and $\mathcal{A}\cong C_0(\Omega(\mathcal{A}))$ $*$-isomorphically. Let $\lambda_1\tau_1+\cdots+\lambda_n\tau_n=0$, where $\lambda_1,\dots,\lambda_n\in\mathbb{C}$ and $\tau_1,\dots,\tau_n\in\Omega(\mathcal{A})$. (Continued below) $\endgroup$ – Masayoshi Kaneda Jul 24 '16 at 18:53
  • $\begingroup$ (Continuation) For any $i\in\{1,\dots,n\}$, by the complete regularity of $\Omega(\mathcal{A})$, $\exists\hat{a}\in C_0(\Omega(\mathcal{A}))$ such that $\hat{a}(\tau_i)=1$ and $\hat{a}(\tau_j)=0$ for $j\ne i$. Thus $0=(\lambda_1\tau_1+\cdots+\lambda_n\tau_n)(a)=\lambda_1\tau_1(a)+\cdots+\lambda_n\tau_n(a)=\lambda_1\hat{a}(\tau_1)+\cdots+\lambda_n\hat{a}(\tau_n)=\lambda_i$. $\endgroup$ – Masayoshi Kaneda Jul 24 '16 at 18:54

As pointed out in the comments, the answer to question 1 is "no". More explicitly, if $v$ is any unit vector in $\mathbb{C}^2$ then the map $A \mapsto \langle Av,v\rangle$ is a pure state on $M_2$, and the pure states arising from two vectors are distinct unless one vector is a scalar (of modulus 1) multiple of the other. Find five unit vectors, no two of which are scalar multiples of each other; the resulting pure states are distinct but cannot be linearly independent, since they all live in $M_2^*$ which is four dimensional.

The answer to question 2 is also "no". I think of compactness of the pure states as a commutative/finite dimensional phenomenon. A famous theorem of Glimm states that if $A$ is any C*-algebra which has an irreducible representation that contains no compact operators, then the pure states of $A$ are weak* dense in the state space of $A$. (Believe it or not.) There's no obvious topology that would make the set of pure states compact.


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.