Definition:
A linear map $f:\mathbb C^n\to \mathbb C^n$ is called positive if $\langle fa,a\rangle\ge0$ for all $a\in \mathbb C^n$. Equivalently, $f\in M_{n}(\mathbb C)$ is positive if it can be written in the form $g^*g$ for some $g\in M_{n}(\mathbb C)$. The unique positive element $g$ satisfying $g^2=f$ is denoted $\sqrt f$, and is called the square root of $f$.

Definition:
A linear map $\phi:M_{n}(\mathbb C)\to M_{m}(\mathbb C)$ is called completely positive if it sends positive elements to positive elements, and the same holds for $\phi\otimes id_{M_{k}(\mathbb C)}:M_{n\cdot k}(\mathbb C)\to M_{m\cdot k}(\mathbb C)$ for every $k\in\mathbb N$.


Let $e_{ij}\in M_{n}(\mathbb C)$ be the elementary matrix with a $1$ at $(i,j)$ and all other entries zero. Using the basis elementary matrices $\{e_{ij}\}$ to identify $M_{n}(\mathbb C)$ with $\mathbb C^{n^2}$, we can talk about a linear map $f:M_{n}(\mathbb C)\to M_{n}(\mathbb C)$ being positive (this has nothing to do with sending positive elements to positive elements).

Question:
Let $\phi:M_{n}(\mathbb C)\to M_{n}(\mathbb C)$ be a map which is both positive and completely positive. Is its square root $\sqrt\phi:M_{n}(\mathbb C)\to M_{n}(\mathbb C)$ completely positive?

Remark:
Maps which are both positive and completely positive are frequent. Indeed, for every completely positive map $\phi:M_{n}(\mathbb C)\to M_{m}(\mathbb C)$, its adjoint $\phi^{*}:M_{m}(\mathbb C)\to M_{n}(\mathbb C)$ is also completely positive. So $\phi^{*}\phi:M_{n}(\mathbb C)\to M_{n}(\mathbb C)$ is both positive and completely positive.

  • Just to clarify: in your last remark, when you identify $M_m({\bf C})$ with its dual, what is your pairing between $M_m({\bf C})$ and $T_m({\bf C})=M_m({\bf C})$? Are you taking ${\rm Tr}(AB)$ or ${\rm Tr}(AB^\top)$? – Yemon Choi Nov 13 at 0:43
  • The pairing is the one with respect to which the $e_{ij}$ form an an orthonormal basis. Namely, $\mathrm{Tr}(AB^*)$. – André Henriques Nov 13 at 15:50
up vote 6 down vote accepted

No. I use facts about Schur multipliers which can be found, for example in Paulsen's monograph on completely bounded maps. Basically you are searching for a positive semidefinite matrix with positive entries such that the pointwise square root of the matrix is not positive semidefinite. Specifically, let $A=\left[ \begin{array}{ccc} 1 & 2^{-1/2} & 0\\ 2^{-1/2} & 1 & 2^{-1/2}\\ 0 & 2^{-1/2} & 1 \end{array} \right].$ Let $S_A:M_3(\mathbb{C})\rightarrow M_3(\mathbb{C})$ be the Schur multiplier associated with $A$ (i.e. $S_A$ takes a matrix to its Schur product with $A$). Since $A$ is positive semidefinite, the Schur multiplier is completely positive. Since $S_A(e_{i,j})=r_{i,j}e_{i,j}$ for some $r_{i,j}\geq 0$ and the set $(e_{i,j})$ forms an orthonormal basis, this is a positive map. The positive square root is clearly $S_B$ where $B=\left[ \begin{array}{ccc} 1 & 2^{-1/4} & 0\\ 2^{-1/4} & 1 & 2^{-1/4}\\ 0 & 2^{-1/4} & 1 \end{array} \right].$ But $B$ is not a positive semidefinite matrix and therefore $S_B$ is not completely positive.

  • 1
    This is correct. – Nik Weaver Nov 13 at 3:20

Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

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