Linear matrix maps for which positivity and complete positivity coincide (Q1979391)

Let \(\mathbb{F}^{n \times m}\) denote the space of \(n \times m\) matrices over \(\mathbb{F} = \mathbb{R}\) or \(\mathbb F = \mathbb{C}\). The study of completely positive and positive linear maps defined on \(\mathbb F^{n\times n}\) have a profound importance to matrix/operator theory, mathematical physics, and many other interesting applications. Recall that a linear map \(\mathcal L: \mathbb F^{q\times q} \to \mathbb F^{n\times n}\) is \textit{positive} if \(\mathcal L(A)\) is positive semidefinite whenever \(A\) is positive semidefinite, and \textit{completely positive} if \(\text{id}_k \otimes \mathcal L\) is positive for all \(k \geq 1\) (where \(\text{id}_{k}\) is the identity operator on \(\mathbb F^{k\times k}\)). The well-known Choi matrix criterion in the influential paper [\textit{M.-D. Choi}, Linear Algebra Appl. 10, 285--290 (1975; Zbl 0327.15018)] establishes that every completely positive linear map admits a nice representation. Positive linear maps, on the other hand, have less tractable structure (see [\textit{I. Klep} et al., Int. Math. Res. Not. 2019, No. 11, 3313--3375 (2019; Zbl 1439.13069)]). The authors of the present paper identify a class of maps for which positivity and complete positivity coincide. Following the notation of the paper, let \(\mathcal L : \mathbb F^{q \times q} \to \mathbb F^{n\times n}\) be a linear map and associate to \(\mathcal L\) its \textit{Choi matrix} \(\mathbb L\) given by \[\mathbb L = [\mathbb L_{ij}] \in \mathbb F^{nq\times nq}, \quad \mathbb L_{ij} = \mathcal L(\mathcal E_{ij}^{(q)}) \in \mathbb F^{n\times n}, \text{ for } i = 1,\dotsc,n, j=1,\dotsc,q,\] where \(\mathcal E_{ij}^{(q)}\) is the matrix unit in \(\mathbb F^{q \times q}\) with 1 in position \((i,j)\) and zeros elsewhere. The authors also associate to \(\mathcal L\) its so-called \textit{matricization} \(L \in \mathbb F^{n^2 \times q^2}\), determined by the linear map \[L : \mathbb F^{q^2}\to \mathbb F^{n^2}, \quad L(\text{vec}_q(V)) = \text{vec}_n (\mathcal L(V)), \quad V \in \mathbb F^{q\times q}\] where \(\text{vec}_r : \mathbb F^{r \times r} \to \mathbb F^{r^2}\) is the vectorization operator sending an \(r \times r\) matrix to a column vector with \(r^2\) entries. Decompose \(L\) as a block matrix \(L = [L_{ij}]\) with \(L_{ij} \in \mathbb F^{n\times q}\) for \(i = 1,\dotsc,n, j = 1,\dotsc, q\). It is well-known that \(\mathcal L\) is completely positive if and only if \(\mathbb L\) is positive semidefinite, but less is known about the connection of \(\mathcal L\) to its matricization \(L\). The authors investigate structural properties of \(\mathcal L\) in terms of its matricization for which \(\mathcal L\) positive implies \(\mathbb L\) is positive semidefinite, hence \(\mathcal L\) is completely positive. Rather surprisingly, the relations among the blocks of the matricization \(L\) are all that matter, so that this result is independent of a specific linear map \(\mathcal L\) and thus applicable to a class of positive maps. The main result is Theorem 1.1, which states that any positive \(*\)-linear map \(\mathcal L\) (with matricization \(L = [L_{ij}]\) and Choi matrix \(\mathbb L\)) with \(m = \text{rank} \,\mathbb L\) linearly independent blocks \(L_{ij} \in \mathbb F^{n \times q}\) satisfying some technical assumptions is also a completely positive map. Indeed, the authors assume that \(k\) of the blocks \(L_{ij}\) either all have different indices \(j\) or different indices \(i\), and that the rest of the blocks are all equal to a single fixed matrix. Question 1.2 asks: does this theorem remain true one removes the assumption that the remaining blocks are equal to a fixed matrix? Counterexamples have not been found. In the special case that the remaining blocks are all equal to 0 (that is, when the \(k\) blocks \(L_{ij}\) form a linearly independent set), the authors do prove that \(\mathcal L\) positive implies completely positive (see Proposition 3.8). Key ideas for the proof of Theorem 1.1 use representations of linear matrix maps studied by \textit{R. D. Hill} [Linear Algebra Appl. 2, 131--142 (1969; Zbl 0186.33901); Linear Algebra Appl. 6, 257--262 (1973; Zbl 0252.15012)] (so-called \textit{Hill representations}) and a relationship between the surjectivity of a certain bilinear map and the complete positivity of a positive map \(\mathcal L\). The authors suggest that answering Question 1.2 may require a better understanding of surjective bilinear maps. At the end of the paper the authors demonstrate their results in the case \(q=n=2\).