Example 41.4.9 (Choi).label Let $A$ be a $C^{*}$-algebra, $\lambda \in \complex$ with $|\lambda| = 1$, and $U_{\lambda} = \text{diag}(\lambda^{1}, \cdots, \lambda^{n})$, then:

  1. (1)

    For each $X = \sum_{i, j = 1}^{n} E_{ij}\otimes x_{ij}\in M_{n}(A)$,

    \[U_{\lambda}^{*} X U_{\lambda}= \sum_{i, j = 1}^{n} E_{ij}\otimes (\lambda^{j - i}x_{ij})\]

  2. (2)

    The mapping $\phi: M_{n}(A) \to M_{n}(A)$ defined by $\phi(X) = n\text{diag}(X) - X$ is completely positive.

  3. (3)

    The mapping $\psi: M_{n}(A) \to M_{n}(A)$ defined by $\psi(X) = (n-1)\text{diag}(X) - X$ is not completely positive.

Proof, [Exercise 3.6, Po02]. Throughout this proof, if $A$ is not unital, view $A$ as an operator system in its unitisation, and $M_{n}(A)$ as an operator system in $M_{n}$ of its unitisation.

(1): Let $X = \sum_{i, j = 1}^{n} E_{ij}\otimes x_{ij}\in M_{n}(A)$, then

\begin{align*}U_{\lambda}^{*} X U_{\lambda}&= \sum_{i, j, k, l = 1}^{n} (E_{ii}\otimes \lambda^{-i})(E_{jk}\otimes x_{jk})(E_{ll}\otimes \lambda^{l}) \\&= \sum_{j, k = 1}^{n} (E_{jj}\cdot E_{jk}\cdot E_{kk}) \otimes (\lambda^{-j}\cdot x_{jk}\cdot \lambda^{k}) \\&= \sum_{j, k = 1}^{n} E_{jk}\otimes (\lambda^{k-j}x_{jk})\end{align*}

(2): Let $\omega = e^{2\pi i /n}$, then the $n$-th roots of unity are $\bracsn{\omega^k}_{k = 1}^{n}$. By (1), for any $X = \sum_{i, j = 1}^{n} E_{ij}\otimes x_{ij}$,

\[\sum_{k = 1}^{n}U_{\omega^k}^{*}XU_{\omega^k}= \sum_{i, j, k = 1}^{n} E_{ij}\otimes \omega^{k(j - i)}x_{ij}\]

where

\[\sum_{k = 1}^{n} \omega^{k(j-i)}= \begin{cases}0 &j \ne i \\ n &j = i\end{cases}\]

so

\begin{align*}\sum_{k = 1}^{n} U_{\omega^k}^{*}XU_{\omega^k}&= \sum_{i = 1}^{n} E_{ii}\otimes (nx_{ii}) = n\text{diag}(X) \\ \phi(X) = \sum_{k = 1}^{n-1}U_{\omega^k}XU_{\omega^k}&= n\text{diag}(X) - U_{1}^{*}XU_{1} = n\text{diag}(X) - X\end{align*}

By Proposition 41.5.3, $\phi$ is completely positive.

(3): Let $x \in A$ with $x \ge 0$ and $x \ne 0$, and $X = \sum_{i, j = 1}^{n} E_{ij}\otimes x$, then

\[X = \sum_{i, j = 1}^{n} E_{ij}\otimes x = \braks{\sum_{i = 1}^n E_{1i} \otimes x^{1/2}}^{*}\braks{\sum_{i = 1}^n E_{1i} \otimes x^{1/2}}\ge 0\]

However, if $V = \sum_{i = 1}^{n} E_{i1}\otimes 1_{A}$, then

\begin{align*}V^{*}\psi(X)V&= (n-1)V^{*}\text{diag}(X)V - V^{*}XV \\ V^{*}\text{diag}(X)V&= \sum_{i, j, k = 1}^{n} (E_{1i}\otimes 1_{A})(E_{jj}\otimes x)(E_{k1}\otimes 1_{A}) \\&= \sum_{i = 1}^{n} (E_{1i}\otimes 1_{A})(E_{ii}\otimes x)(E_{i1}\otimes 1_{A}) = n(E_{11}\otimes x) \\ V^{*}XV&= \sum_{i, j, k, l = 1}^{n} (E_{1i}\otimes 1_{A})(E_{jk}\otimes x)(E_{l1}\otimes 1_{A}) \\&= \sum_{i, j = 1}^{n} (E_{1i}\otimes 1_{A})(E_{ij}\otimes x)(E_{j1}\otimes 1_{A}) = n^{2}(E_{11}\otimes x) \\ V^{*}\psi(X)V&= (n-1)n(E_{11}\otimes x) - n^{2}(E_{11}\otimes x) = -n(E_{11}\otimes x)\end{align*}

As $x \ne 0$, $-n(E_{ii}\otimes x)$ is not positive. Therefore $\psi(X)$ is not positive, and $\psi$ is not a positive map.$\square$

Post a Comment

Name:Email:
Please enter the tag of the current page (1L2) to post the comment.
Tag: