41.3 Completely Positive and Completely Bounded Maps

Proposition 41.3.1.label Let $A$ be a $C^{*}$-algebra, $M_{n}(A) = M_{n}(\complex) \otimes A$ be the space of $n \times n$ matrices over $A$,

\[M_{n}(A) \times M_{n}(A) \to M_{n}(A) \quad (M \otimes a)(N \otimes b) = MN \otimes ab\]

and the involution

\[*: M_{n}(A) \to M_{n}(A) \quad M \otimes a \mapsto M^{*} \otimes a^{*}\]

then there exists a unique norm on $M_{n}(A)$ such that $M_{n}(A)$ is a $C^{*}$-algebra.

Proof. After applying a unitisation, assume without loss of generality that $A$ is unital.

By the Gelfand-Naimark Theorem, there exists a faithful representation $(H, \pi)$ of $A$. Under the identification that $H^{n} = \complex^{n} \otimes H$, $\pi$ induces a faithful unital *-representation

\[\pi_{n}: M_{n}(A) \to B(H^{n}) \quad \pi_{n}(M \otimes a)(\lambda \otimes \xi) = (M\lambda) \otimes (\pi(a)\xi)\]

whose image is a uniformly closed subalgebra of $B(H^{n})$. As such, $B(H^{n})$ induces a norm on $M_{n}(A)$, making it a $C^{*}$-algebra.

By Corollary 39.4.5, this norm is unique.$\square$

Definition 41.3.2 (Extended Map).label Let $A$ be a unital $C^{*}$-algebra, $B$ be a $C^{*}$-algebra, $S \subset A$ be an operator system, $\phi: S \to B$ be a linear map, and $n \in \natp$, then the mapping

\[\phi_{n} = \text{Id}_{M_n(\complex)}\otimes \phi: M_{n}(S) \to M_{n}(B)\]

is the extended map/amplification of $\phi$ to $M_{n}(S)$.

Proposition 41.3.3 (Tomiyama).label Let $A, B$ be unital $C^{*}$-algebras and $S \subset A$ be an operator system, then:

  1. (1)

    For any $a = \sum_{i, j = 1}^{n} E_{ij}\otimes a_{ij}\in M_{n}(A)$,

    \[\norm{a}\le \norm{\sum_{i, j = 1}^n\norm{a_{ij}}_AE_{ij}}_{M_n(\complex)}\le \braks{\sum_{i, j = 1}^n \norm{a_{ij}}_{A}^2}^{1/2}\le n\norm{a}_{M_n(A)}\]

    where all inequalities are sharp.

  2. (2)

    For any bounded linear map $\phi: S \to B$, $\norm{\phi_n}_{L(M_n(S); M_n(B))}\le n\norm{\phi}_{L(S; B)}$.

Proof, [Exercise 3.10, Po02]. (1): Using the Gelfand-Naimark Theorem, let $H$ be a complex Hilbert space such that $A$ is a $C^{*}$-subalgebra of $B(H)$.

For each $\xi = (\xi_{1}, \cdots, \xi_{n}), \eta = (\eta_{1}, \cdots, \eta_{n}) \in H^{n}$,

\begin{align*}\dpn{a\xi, \eta}{H^n}&= \sum_{i, j = 1}^{n} \dpn{(E_{ij} \otimes a_{ij})\xi, \eta}{H^n}= \sum_{i, j = 1}^{n} \dpn{a_{ij}\xi_j, \eta_i}{H}\\ |\dpn{a\xi, \eta}{H^n}|&\le \sum_{i, j = 1}^{n} \norm{a_{ij}}_{A}\norm{\xi_j}_{H}\norm{\eta_i}_{H} \\&= \sum_{i, j = 1}^{n}\dpn{\norm{a_{ij}}_AE_{ij}(\norm{\xi_1}_H, \cdots, \norm{\xi_n}_H), (\norm{\eta_1}_H, \cdots, \norm{\eta_n}_H)}{\complex^n}\\&\le \norm{\xi}_{H^n}\cdot \norm{\eta}_{H^n}\cdot \norm{\sum_{i, j = 1}^n\norm{a_{ij}}_AE_{ij}}_{M_n(\complex)}\end{align*}

The sharp case is given by the identity.

For each $\xi = (\xi_{1}, \cdots, \xi_{n}) \in \complex^{n}$, by the Cauchy-Schwarz Inequality,

\begin{align*}\norm{\sum_{i, j = 1}^n \norm{a_{ij}}_AE_{ij}\xi}_{\complex^n}^{2}&= \sum_{i = 1}^{n} \abs{\sum_{j = 1}^n \norm{a_{ij}}_{A}\xi_j}^{2} \\&\le \braks{\sum_{i, j = 1}^n \norm{a_{ij}}_A^2}\braks{\sum_{j = 1}^n |\xi_j|^2}\\ \norm{\sum_{i, j = 1}^n \norm{a_{ij}}_AE_{ij}\xi}_{\complex^n}&\le \braks{\sum_{i, j = 1}^n \norm{a_{ij}}_A^2}^{1/2}\cdot \norm{\xi}_{\complex^n}\end{align*}

The sharp case is given by operators of the form $E_{ij}\otimes b$ for $b \in A$.

Finally,

\[\sum_{i, j = 1}^{n} \norm{a_{ij}}_{A}^{2} = \sum_{i, j = 1}^{n} \norm{E_{ii} \cdot a \cdot E_{jj}}_{M_n(A)}^{2} \le n^{2}\norm{a}_{M_n(A)}^{2}\]

and

\[\braks{\sum_{i, j = 1}^n \norm{a_{ij}}_{A}^2}^{1/2}\le n\norm{a}_{M_n(A)}\]

The sharp case is given by operators of the form $E = \sum_{i, j = 1}^{n} E_{ij}\otimes E_{ji}$[1]. As

\begin{align*}\braks{\sum_{i, j = 1}^n E_{ij} \otimes E_{ji}}^{*}\braks{\sum_{i, j = 1}^n E_{ij} \otimes E_{ji}}&= \sum_{i, j, k, l = 1}^{n} E_{ij}E_{kl}\otimes E_{ji}E_{lk}\\&= \sum_{i, j, l = 1}^{n} E_{il}\otimes E_{ji}E_{lj}= \sum_{i, j = 1}^{n} E_{ii}\otimes E_{jj}\\&= I_{M_n(\complex)}\otimes I_{M_n(\complex)}= I_{M_{n \times n}(\complex)}\end{align*}

the operator $E$ is of norm $1$, but $\sum_{i, j = 1}^{n} \norm{E_{ji}}_{M_n(\complex)}^{2} = n^{2}$.

(2): For any $a = \sum_{i, j = 1}^{n} E_{ij}\otimes a_{ij}\in M_{n}(S)$,

\begin{align*}\normn{\phi(a)}_{M_n(B)}^{2}&\le \sum_{i, j = 1}^{n} \norm{\phi(a_{ij})}_{B}^{2} \le \norm{\phi}_{L(S; B)}^{2}\sum_{i, j = 1}^{n} \norm{a_{ij}}_{A}^{2} \\&\le n^{2}\norm{\phi}_{L(S; B)}^{2}\norm{a}_{M_n(A)}^{2}\end{align*}

$\square$

Proposition 41.3.4.label Let $A$ be a unital $C^{*}$-algebra and $b = \sum_{i, j = 1}^{n} E_{ij}\otimes b_{ij}\in M_{n}(A)$, then the following are equivalent:

  1. (1)

    $b$ is positive in $M_{n}(A)$.

  2. (2)

    For each $\seqf{a_j}\subset A$, $\sum_{i, j = 1}^{n} a_{i}b_{ij}a_{j}^{*}$ is positive in $A$.

Proof. (1) $\Rightarrow$ (2): For any $\seqf{a_j}\subset A$, let $a = \sum_{i = 1}^{n} E_{1i}\otimes a_{i}$, then

\begin{align*}0 \le aba^{*}&= \braks{\sum_{k = 1}^n E_{k1} \otimes a_k^*}\braks{\sum_{j, k = 1}^n E_{jk} \otimes b_{jk}}\braks{\sum_{i = 1}^n E_{i1} \otimes a_i}\\&= \sum_{j, k = 1}^{n} (E_{1j}\cdot E_{jk}\cdot E_{k1}) \otimes (a_{j}b_{jk}a_{k}^{*}) = E_{11}\otimes \sum_{i, j = 1}^{n} a_{i}b_{ij}a_{j}^{*}\end{align*}

so $\sum_{i, j = 1}^{n} a_{i}b_{ij}a_{j}^{*} \ge 0$ in $A$.

(2) $\Rightarrow$ (1): For each state $\phi \in S(A)$, let $(H_{\phi}, \pi_{\phi}, \xi_{\phi})$ be the GNS triple associated with $\phi$. Let $H = [l^{2}(S(A)); H_{\phi}]$,

\[\pi : A \to B(H) \quad \pi(a)(\xi)_{\phi} = \pi_{\phi}(a)(\xi_{\phi})\]

be the universal representation of $A$, and identify $A$ as a $C^{*}$-subalgebra of $B(H)$ through $\pi$.

For any $\seqf{a_j}\subset A$ and $\phi \in S(A)$,

\begin{align*}0 \le \sum_{i, j = 1}^{n} \dpn{a_i^*b_{ij}a_j , \phi}{A}&= \sum_{i, j = 1}^{n}\dpn{\pi_\phi(b_{ij}a_j)\xi_\phi, \pi_\phi(a_i)\xi_\phi}{H_\phi}\\&= \sum_{i, j = 1}^{n}\dpn{\pi_\phi(b_{ij})\pi_\phi(a_j)\xi_\phi, \pi_\phi(a_i)\xi_\phi}{H_\phi}\end{align*}

Identify $H_{\phi}$ as a subspace of $H$ and let $\xi = (\pi_{\phi}(a_{1})\xi_{\phi}, \cdots, \pi_{\phi}(a_{n})\xi_{\phi}) \in H^{n}$, then

\[0 \le \sum_{i, j = 1}^{n}\dpn{\pi_\phi(b_{ij})\pi_\phi(a_j)\xi_\phi, \pi_\phi(a_i)\xi_\phi}{H_\phi}= \dpn{b\xi, \xi}{H^n}\]

As $(H_{\phi}, \pi_{\phi})$ is a cyclic representation, the above implies that $\dpn{b\eta, \eta}{H^n}\ge 0$ for all $\eta \in H_{\phi}^{n}$. Since this holds for all $\phi \in S(A)$, $b \ge 0$ in $M_{n}(A)$.$\square$

Definition 41.3.5 (Completely Positive Map).label Let $A, B$ be $C^{*}$-algebras with $A$ being unital, $S \subset A$ be an operator system, $\phi: S \to B$ be a linear map, and $n \in \natp$, then $\phi$ is $n$-positive if $\phi_{n}: M_{n}(S) \to M_{n}(B)$ is positive. The map $\phi$ is completely positive if $\phi$ is $n$-positive for all $n \in \natp$.

Definition 41.3.6 (Completely Bounded Map).label Let $A, B$ be $C^{*}$-algebras with $A$ being unital, $S \subset A$ be an operator system, and $\phi: S \to B$ be a linear map, then

\[\norm{\phi}_{\text{cb}}= \sup_{n \in \natp}\norm{\phi_n}_{L(M_n(S); M_n(B))}\]

is the completely bounded norm of $\phi$.

The map $\phi$ is completely bounded if $\norm{\phi}_{\text{cb}}< \infty$, and completely contractive if $\norm{\phi}_{\text{cb}}\le 1$.

Lemma 41.3.7.label Let $A, B$ be $C^{*}$-algebras with $A$ being unital, $S \subset A$ be an operator system, $\phi: S \to B$ be a linear map, and $n \in \natp$.

  1. (1)

    If $\phi$ is $n$-positive, then $\phi$ is $k$-positive for all $1 \le k \le n$.

  2. (2)

    For each $1 \le k \le n$, $\norm{\phi_k}_{L(M_k(S); M_k(B))}\le \norm{\phi_n}_{L(M_n(S); M_n(B))}$.

Proof. For any $n \in \natp$ and $1 \le k < n$, there is a natural *-homomorphism

\[e: M_{k}(\complex) \to M_{n}(\complex) \quad M \mapsto \begin{bmatrix}M & 0 \\ 0 & 0\end{bmatrix}\]

such that the following diagram commutes:

\[\xymatrix{ M_k(S) \ar@{->}[d]_{e \otimes \text{Id}_S} \ar@{->}[r]^{\phi_k} & M_k(B) \ar@{->}[d]^{e \otimes \text{Id}_B} \\ M_n(S) \ar@{->}[r]_{\phi_n} & M_n(B) }\]

(1): Since $e \otimes \text{Id}_{S}$ is the restriction of a *-homomorphism, it is positive. On the other hand, for any $x \in M_{k}(B)$, $x \ge 0$ if and only if $(e \otimes \text{Id}_{B})(x) \ge 0$. Given that $\phi_{n}$ is positive, $\phi_{k}$ must also be positive.

(2): The maps $e \otimes \text{Id}_{S}$ and $e \otimes \text{Id}_{B}$ are both isometric. As such, the norm of $\phi_{k}$ is bounded above by the norm of $\phi_{n}$.$\square$

Proposition 41.3.8.label Let $A, B$ be $C^{*}$-algebras with $A$ being unital, $S \subset A$ be an operator system, and $\phi: S \to B$ be a completely positive map, then

\[\norm{\phi}_{\text{cb}}= \norm{\phi}_{L(S; B)}= \norm{\phi(1_A)}_{B}\]

Proof. After taking a unitisation, assume without loss of generality that $B$ is unital.

It is sufficient to show that $\norm{\phi}_{\text{cb}}\le \norm{\phi(1_A)}_{B}$. To this end, let $T \in M_{n}(S)$ with $\norm{T}_{M_n(S)}\le 1$, then

\[\begin{bmatrix}1_{M_n(A)}&T \\ T^{*}&1_{M_n(A)}\end{bmatrix} \ge 0\]

in $M_{2}(M_{n}(S)) = M_{2n}(S)$ by Lemma 41.2.1. Since $\phi$ is completely positive,

\[\phi_{2n}\begin{bmatrix}1_{M_n(A)}&T \\ T^{*}&1_{M_n(A)}\end{bmatrix} = \begin{bmatrix}\phi_{n}(1_{M_n(A)})&\phi_{n}(T) \\ \phi_{n}(T)^{*}&\phi_{n}(1_{M_n(A)})\end{bmatrix} \ge 0\]

First suppose that $\phi_{n}(1_{M_n(A)})$ is invertible, then

\begin{align*}\phi_{n}(1_{M_n(A)})&\le \normn{\phi_n(1_{M_n(A)})}_{M_n(B)}\cdot 1_{M_n(B)}\\ 1_{M_n(B)}&\le \normn{\phi_n(1_{M_n(A)})}_{M_n(B)}\cdot \phi_{n}(1_{M_n(A)})^{-1}\\ \normn{\phi_n(1_{M_n(A)})}_{M_n(B)}^{-1}&\le \phi_{n}(1_{M_n(A)})^{-1}\end{align*}

By the Schur complement condition,

\begin{align*}\phi_{n}(1_{M_n(A)})&\ge \phi_{n}(T)\phi_{n}(1_{M_n(A)})^{-1}\phi_{n}(T)^{*} \\&\ge \normn{\phi_n(1_{M_n(A)})}_{M_n(B)}^{-1}\phi_{n}(T)\phi_{n}(T)^{*} \\ \normn{\phi_n(1_{M_n(A)})}_{M_n(B)}^{2}&\ge \normn{\phi_n(T)\phi_n(T)^*}_{M_n(B)}= \normn{\phi_n(T)}_{M_n(B)}^{2}\end{align*}

and $\normn{\phi_n(T)}_{M_n(B)}\le \normn{\phi_n(1_{M_n(A)})}_{M_n(B)}$. Therefore $\norm{\phi}_{\text{cb}}\le \normn{\phi(1_A)}_{B}$.

Now suppose that $\phi$ is an arbitrary completely positive map. Let $\psi \in S(A)$ be a state and $\eps > 0$, then $\phi + \eps 1_{B}\psi$ is also a completely positive map. In which case,

\[\norm{\phi}_{\text{cb}}\le \inf_{\eps > 0}\norm{\phi + \eps1_B\psi}_{\text{cb}}\le \inf_{\eps > 0}\norm{\phi(1_A) + \eps 1_B}_{B} = \norm{\phi(1_A)}_{B}\]

$\square$

Proposition 41.3.9.label Let $A, B$ be unital $C^{*}$-algebras, $S \subset A$ be an operator system, $\phi: S \to B$ be a unital $2$-positive map, then $\norm{\phi}_{L(S; B)}\le 1$.

Proof, [Proposition 3.2, Po02]. Let $x \in S$ with $\norm{x}_{A} \le 1$, then

\[\phi_{2}\begin{bmatrix}1&x \\ x^{*}&1\end{bmatrix} = \begin{bmatrix}1&\phi(x) \\ \phi(x)^{*}&1\end{bmatrix} \ge 0\]

By Lemma 41.2.1, $\norm{\phi(x)}_{B} \le 1$.$\square$

Proposition 41.3.10 (Kadison-Schwarz Inequality).label Let $A, B$ be unital $C^{*}$-algebras, $\phi: A \to B$ be a $2$-positive map, then for each $x \in A$,

\[\phi(x)^{*}\phi(x) \le \norm{\phi(1_A)}_{B} \phi(x^{*}x)\]

Proof, general case. Since

\[\begin{bmatrix}1_{A}&x \\ x^{*}&x^{*}x\end{bmatrix} = \begin{bmatrix}1_{A}&x \\ 0&0\end{bmatrix}^{*}\begin{bmatrix}1_{A}&x \\ 0&0\end{bmatrix} \ge 0\]

and $\phi$ is $2$-positive,

\[\begin{bmatrix}1_{B}\norm{\phi(1_A)}_{B}&\phi(x) \\ \phi(x)^{*}&\phi(x^{*}x)\end{bmatrix} \ge \begin{bmatrix}\phi(1_{A})&\phi(x) \\ \phi(x)^{*}&\phi(x^{*}x)\end{bmatrix} = \phi_{2}\begin{bmatrix}1_{A}&x \\ x^{*}&x^{*}x\end{bmatrix} \ge 0\]

By Lemma 41.2.1, $\phi(x)^{*}\phi(x) \le \norm{\phi(1_A)}_{B}\phi(x^{*}x)$.$\square$

Proof, completely positive case. Using the Gelfand-Naimark Theorem, let $H$ be a complex Hilbert space such that $B \subset B(H)$. Assume in addition that $\phi$ is completely positive. Let $(K, V, \pi)$ be a Stinespring representation for $\phi$, then for each $a \in A$ and $\xi \in H$,

\begin{align*}\dpn{\phi(a^*)\phi(a)\xi, \xi}{H}&= \dpn{V^*\pi(a^*)VV^*\pi(a)V\xi, \xi}{H}= \dpn{\pi(a^*)VV^*\pi(a)V\xi, V\xi}{K}\\&\le \norm{VV^*}_{B(K)}\dpn{\pi(a^*)\pi(a)V\xi, V\xi}{K}\\&\le \norm{V}_{B(K)}^{2} \dpn{\pi(a^*a)V\xi, V\xi}{K}= \norm{\phi(1_A)}_{B(H)}\dpn{\phi(a^*a)\xi, \xi}{K}\end{align*}

$\square$

Example 41.3.11.label Let $T: M_{2}(\complex) \to M_{2}(\complex)$ be the transpose map, then $T$ is positive, but not $2$-positive.

Proof. Let

\[C = \begin{bmatrix}1&0&0&1 \\ 0&0&0&0 \\ 0&0&0&0 \\ 1&0&0&1\end{bmatrix} = (1, 0, 0, 1) \otimes (1, 0, 0, 1) \ge 0\]

then

\[T_{2}(C) = \begin{bmatrix}1&0&0&0 \\ 0&0&1&0 \\ 0&1&0&0 \\ 0&0&0&1\end{bmatrix}\]

is invertible, with negative determinant, so $T_{2}(C)$ is not positive.$\square$

Example 41.3.12 (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 41.4.3, $\phi$ 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 $x \ne 0$, $-n(E_{11}\otimes x)$ is not positive. Therefore $\psi(X)$ is not positive, and $\psi$ is not a positive map.$\square$

Proposition 41.3.13.label Let $A$ be a $C^{*}$-algebra, $A^{\text{op}}$ denote $A$ with the same norm and *-operation, but with the product

\[A^{\text{op}}\times A^{\text{op}}\to A^{\text{op}}\quad (x, y) \mapsto x \circ y = y \cdot x\]

then:

  1. (1)

    $A^{\text{op}}$ is a $C^{*}$-algebra, $\text{Id}: A \to A^{\text{op}}$ is positive.

  2. (2)

    The transpose map $M_{n}(\complex) \to M_{n}(\complex)^{\text{op}}$ is a *-isomorphism.

  3. (3)

    $\text{Id}: M_{2}(\complex) \to M_{2}(\complex)^{\text{op}}$ is not 2-positive.

  4. (4)

    (Walter) For each $U, V \in U(A)$ and $X \in A$,

    \[\begin{bmatrix}1_{A}&U&X \\ U^{*}&1_{A}&V \\ X^{*}&V^{*}&1_{A}\end{bmatrix} \ge 0\]

    if and only if $X = UV$.

  5. (5)

    $\text{Id}: A \to A^{\text{op}}$ is completely positive if and only if $A$ is commutative.

Proof. (4, $\Leftarrow$): If $X = UV$, then

\begin{align*}\begin{bmatrix}1_{A}&U&UV \\ 0&0&0 \\ 0&0&0\end{bmatrix}^{*}\begin{bmatrix}1_{A}&U&UV \\ 0&0&0 \\ 0&0&0\end{bmatrix}&= \begin{bmatrix}1_{A}&0&0 \\ U^{*}&0&0 \\ V^{*}U^{*}&0&0\end{bmatrix}\begin{bmatrix}1_{A}&U&UV \\ 0&0&0 \\ 0&0&0\end{bmatrix} \\&= \begin{bmatrix}1_{A}&U&UV \\ U^{*}&1_{A}&U^{*}UV \\ V^{*}U^{*}&V^{*}U^{*}U&1_{A}\end{bmatrix}\\&= \begin{bmatrix}1_{A}&U&UV \\ U^{*}&1_{A}&U^{*}UV \\ V^{*}U^{*}&V^{*}U^{*}U&1_{A}\end{bmatrix}\end{align*}

so the given matrix is positive.

(4, $\Rightarrow$): Suppose that the given matrix is positive. By Lemma 41.4.8, there exists $\bracsn{a_{k, j}}_{j, k = 1}^{3} \subset A$ such that

\[\begin{bmatrix}1_{A}&U&X \\ U^{*}&1_{A}&V \\ X^{*}&V^{*}&1_{A}\end{bmatrix} = \sum_{k = 1}^{3}\begin{bmatrix}a_{k, 1}^{*}a_{k,1}&a_{k, 1}^{*}a_{k,2}&a_{k, 1}^{*}a_{k,3}\\ a_{k, 2}^{*}a_{k,1}&a_{k, 2}^{*}a_{k,2}&a_{k, 2}^{*}a_{k,3}\\ a_{k, 3}^{*}a_{k,1}&a_{k, 3}^{*}a_{k,2}&a_{k, 3}^{*}a_{k,3}\\\end{bmatrix}\]

Using an elementary column operation,

\[\begin{bmatrix}1_{A}&U - U&X \\ U^{*}&1_{A} - U^{*}U&V \\ X^{*}&V^{*} - X^{*}U&1_{A}\end{bmatrix} = \sum_{k = 1}^{3}\begin{bmatrix}a_{k, 1}^{*}a_{k,1}&a_{k, 1}^{*}(a_{k,2}- a_{k, 1}U)&a_{k, 1}^{*}a_{k,3}\\ a_{k, 2}^{*}a_{k,1}&a_{k, 2}^{*}(a_{k,2}- a_{k, 1}U)&a_{k, 2}^{*}a_{k,3}\\ a_{k, 3}^{*}a_{k,1}&a_{k, 3}^{*}(a_{k,2}- a_{k, 1}U)&a_{k, 3}^{*}a_{k,3}\\\end{bmatrix}\]

and an elementary row operation,

\[\begin{bmatrix}1_{A}&0&X \\ 0&0&V - U^{*}X \\ X^{*}&V^{*} - X^{*}U&1_{A}\end{bmatrix}\]

is equal to

\[\sum_{k = 1}^{3}\begin{bmatrix}a_{k, 1}^{*}a_{k,1}&a_{k, 1}^{*}(a_{k,2}- a_{k, 1}U)&a_{k, 1}^{*}a_{k,3}\\ (a_{k, 2}- a_{k, 1}U)^{*}a_{k,1}&(a_{k, 2}- a_{k, 1}U)^{*}(a_{k,2}- a_{k, 1}U)&(a_{k, 2}- a_{k, 1}U)^{*}a_{k,3}\\ a_{k, 3}^{*}a_{k,1}&a_{k, 3}^{*}(a_{k,2}- a_{k, 1}U)&a_{k, 3}^{*}a_{k,3}\\\end{bmatrix}\]

In particular,

\[\sum_{k = 1}^{3} (a_{k, 2}- a_{k, 1}U)^{*}(a_{k,2}- a_{k, 1}U) = 0\]

so $a_{k, 2}= a_{k, 1}U$ for $1 \le k \le 3$. Thus

\[\begin{bmatrix}1_{A}&0&X \\ 0&0&V - U^{*}X \\ X^{*}&V^{*} - X^{*}U&1_{A}\end{bmatrix} = \sum_{k = 1}^{3}\begin{bmatrix}a_{k, 1}^{*}a_{k,1}&0&a_{k, 1}^{*}a_{k,3}\\ 0&0&0 \\ a_{k, 3}^{*}a_{k,1}&0&a_{k, 3}^{*}a_{k,3}\\\end{bmatrix}\]

and $V = U^{*}X$, so $X = UV$.

(5): Let $U, V \in U(A)$, then by (4),

\[\begin{bmatrix}1_{A}&U&UV \\ U^{*}&1_{A}&V \\ V^{*}U^{*}&V^{*}&1_{A}\end{bmatrix} \ge 0\]

If the identity is completely positive, then

\[\begin{bmatrix}1_{A}&U&V \circ U \\ U^{*}&1_{A}&V \\ U^{*} \circ V^{*}&V^{*}&1_{A}\end{bmatrix} = \begin{bmatrix}1_{A}&U&UV \\ U^{*}&1_{A}&V \\ V^{*}U^{*}&V^{*}&1_{A}\end{bmatrix} \ge 0\]

in $M_{3}(A^{\text{op}})$. By (4), $VU = U \circ V = V \circ U = UV$. Therefore $A$ is commutative.$\square$

  1. A minimum ”complexity” of $M_{n}$ is required on $A$ for the sharp case to hold.keyboard_return

Post a Comment

Name:Email:
Please enter the tag of the current page (1J9) to post the comment.
Tag: