MathOverflow is a question and answer site for professional mathematicians. It's 100% free, no registration required.

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Let $A$ be a finite dimensional $C^*$ algebra and $S(A)$ the state space. Let $K\subset A$ be an intersection of $S(A)$ with a vector subspace $J\subset A$ and let $f$ be a positive affine functional on $K$.

I am dealing with the following questions: Can $f$ be extended to a positive linear functional on $A$? And if $F$ is a completely positive map on $J$, under which conditions can it be extended to a completely positive map on $A$? Is there anything else than Arveson's extension theorem?

Is there a good reference where these questions are treated in the finite dimensional case?

share|cite|improve this question
can you write more details, with what exactly do you intersect a subspace $J\subset A$? – Kate Juschenko Feb 20 '11 at 15:53
I guess Kate is asking: if $J\subset A$ then how can we intersect $S(A)$ with $J$, as $S(A)$ is a subset of the dual of $A$, not of $A$... – Matthew Daws Feb 20 '11 at 19:46
Since $A$ is finite dimensional, the states can be identified with density operator with respect to a fixed trace, that is, positive operators with trace 1. Then $S(A)\subset A$. In the general case, the question has to be formulated in the dual space, but in finite dimensions, this is all the same. – Anna Jenčová Feb 20 '11 at 20:36

The intersection of a subspace $J$ with the set of positive matrices of trace one is convex, say equal to $V$. Obviously, we can extend $f$ to a cone generated by $V$, denote $c(V)$ (in fact V is a convex combination of finite set of positive operators with trace equal to $1$). Then $f$ has natural positive extension to $c(V)-c(V)$ and also to $c(V)-c(V)+i(c(V)-c(V))$. Since now $f$ is defined on a self-adjoint subspace of a C*-algebra by Krein's theorem it has a positive extension to the whole C*-algebra. for the last 2 questions: positive functional on a C*-algebra is automatically completely positive.

EDIT: the proof of the extension theorem for the unital case is simple: if $\phi:S\rightarrow \mathbb{C}$ is positive then $\phi(1)\geq ||\phi||>0$. Let $\phi'=\phi \cdot \phi(1)^{-1}$ is positive and $||\phi'||=\phi'(1)=1$. By Hahn-Banach $\phi'$ has contractive extension $\overline{\phi}$ to the C*-algebra, since $\overline{\phi}$ is contraction and unital, it is automatically positive.

To get the question for the non-unital case one can argue as follows. If $1$ is not in the subspace generated by $V$, then the following extension is well defined:

$f(t1+(1-t)a)=t\lambda +(1-t)f(a)$ for $a\in V$.

In order to have a positive extension to $c(V)-c(V)$ one needs to choose $\lambda$ properly.

share|cite|improve this answer
Yes, thanks very much. For the last 2 questions: I was asking about a cp map $J\to B(H)$, for a Hilbert space $H$, not a functional. – Anna Jenčová Feb 21 '11 at 9:54
Does it mean that one can do this for any convex subset of $S(A)$? – Anna Jenčová Feb 21 '11 at 13:25
If you define positivity of $f$ on $M_n(J)$ as operator that sends positive matrix with coefficients in J to a positive operator, then this should not be enough for its extension, since $M_n(C(V)-c(V)+i(c(V)-C(V)))$ might have more positive elements, than those that come from matrix with coefficients in J. there should be an example, where there is no extension to completely positive map. – Kate Juschenko Feb 21 '11 at 13:48
I don't think one can find a positive extension for any convex subset. Let $V$ be a convex subset generated by two positive trace 1 elements $\rho_1$ and $\rho_2$. Then for any $f_1,f_2\ge 0$ there is a positive affine functionals on $V$ with $f_1=f(\rho_1)$ and $f_2=f(\rho_2)$. Suppose that $\rho_1\le M\rho_2$ for some $M>1$. Then $M\rho_2-\rho_1$ is positive, but one can always find $f_1,f_2\ge 0$ such that $Mf_2-f_1<0$. – Anna Jenčová Feb 21 '11 at 14:43
there is very good collection of positive/completely positive maps in the book of Vern Paulsen,… – Kate Juschenko Feb 21 '11 at 15:26

Kate's argument can be finished as follows:

Let us denote $L=c(V)-c(V)$. We suppose that $L$ is such that $L\cap A^+=c(V)$. Let $\tilde L$ be the smallest real vector subspace, containing $c(V)-c(V)$ and $1$.

Let us denote$f_m=\sup\{ f(x), x\in L, x\le 1\}$ and $f_M=\inf\{f(x), x\in L, x\ge 1\}$. Then $f_m\ge \sup \{f(a), a\in V\}$. If $x\in L$ is such that $x\ge 1$, then for any $y\in L$, $y\le 1\le x$ we have $f(y)\le f(x)$ and hence $f_m\le f_M$. If there is no elemment in $L$ greater than $1$, then $f_M=\infty$, so that we always have $0\le f_m\le f_M$.

Choose some $\lambda$ such that $f_m\le \lambda\le f_M$. We extend $f$ to a linear functional $\tilde f$ on $\tilde L$, by $\tilde f(x+t)=f(x)+t\lambda$, for $x\in L$, $t\in \mathbb R$. Then it is easy to see that if $-1\le x+t \le 1$, then $-\lambda \le \tilde f(x+t)\le \lambda$, so that $\|\tilde f\|\le \lambda=\tilde f(1)$. Extending $\tilde f$ to $\tilde L+i\tilde L$, $\tilde f$ must be positive and we are back at the unital case.

share|cite|improve this answer
now I see... from the fact that you intersect the subspace $J$ with $S(A)$ follows that $(c(V)-c(V))^+\subseteq c(V)$, this I've missed this fact for the non-unital case. – Kate Juschenko Mar 5 '11 at 17:08

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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