Families of completely positive maps associated with monotone metrics

DOI10.1016/J.LAA.2013.05.012MaRDI QIDQ2435448zbMATH OpenOpenAlexFDO

Authors Fumio Hiai, Hideki Kosaki, Mary Beth Ruskai, Dénes Petz

Publication date 19 February 2014

Published in Linear Algebra and its Applications (Search for Journal in Brave)

Full work available at URL https://arxiv.org/abs/1212.1337

completely positive map complex Hermitian matrices geometric bridge infinite divisibility integral representation monotone Riemannian metric operator convex function operator monotone function positive definite kernel positive definite matrices positivity preserving quasi-entropy

Mathematics Subject Classification ID

Linear preserver problems (15A86) Positive matrices and their generalizations; cones of matrices (15B48) Convexity of real functions in one variable, generalizations (26A51) Positive definite functions in one variable harmonic analysis (42A82)

Abstract: An operator convex function on (0,infty) which satisfies the symmetry condition k(1/x) = x k(x) can be used to define a type of non-commutative multiplication by a positive definite matrix (or its inverse) using the primitive concepts of left and right multiplication and the functional calculus. The operators for the inverse can be used to define quadratic forms associated with Riemannian metrics which contract under the action of completely positive trace-preserving maps. We study the question of when these operators define maps which are also completely positive (CP). Although A --> D^{-1/2} A D^{-1/2} is the only case for which both the map and its inverse are CP, there are several well-known one parameter families for which either the map or its inverse is CP. We present a complete analysis of the behavior of these families, as well as the behavior of lines connecting an extreme point with the smallest one and some results for geometric bridges between these points. Our primary tool is an order relation based on the concept of positive definite functions. Although some results can be obtained from known properties, we also prove new results based on the positivity of the Fourier transforms of certain functions. Concrete computations of certain Fourier transforms not only yield new examples of positive definite functions, but also examples in the much stronger class of infinitely divisible functions.

Review text

Let \(\mathbb{M}_d,~\mathbb{H}_d\) and \(\mathbb{P}_d\) denote the sets of complex square matrices, complex Hermitian matrices and positive definite matrices of order \(d\), respectively. A real-valued function \(f\) on \((0,\infty)\) is said to be operator monotone if \(A \geq B\) implies that \(f(A) \geq f(B)\), where \(A\geq B\) stands for \(A-B\) being positive semidefinite with \(A,B \in \mathbb{H}_d\). A real-valued function \(k\) on \((0,\infty)\) is called operator convex if \(k(\lambda A + (1-\lambda)B) \leq \lambda k(A) + (1-\lambda) k(B)\) for all \(\lambda \in (0,1)\) and for all \(A,B \in \mathbb{P}_d\) for any positive integer \(d\). \(k\) is called operator concave if \(-k\) is operator convex. Let \(\mathcal K\) denote the class of functions \(k: (0,\infty) \rightarrow (0,\infty)\) which are operator convex, satisfying the symmetry condition \(xk(x)=k(x^{-1})\) and the normalization condition \(k(1)=1\). The authors prove an integral representation for functions that belong to \(\mathcal K\) using which necessary and sufficient conditions for a function to belong to \(\mathcal K\) are given. These statements are given in terms of operator convexity, operator concavity and operator monotonicity. A linear map \(\varPhi:\mathbb{M}_d \rightarrow \mathbb{M}_d\) is called positive if it is positivity preserving, namely \(\varPhi (\mathbb {P}_d) \subseteq \bar {\mathbb{P}_d}\), where \(\bar {\mathbb{P}_d}\) denotes the set of complex positive semidefinite matrices. \(\varPhi\) is called completely positive if \(\varPhi \otimes I\) is positive on \(\mathbb{M}_d \otimes \mathbb{M}_n\) for all positive integers \(n\). Given \(k \in \mathcal K\), consider the function \(\phi^k(x,y):=(1/y)k(x/y)\) for \(x,y >0\). Let \(D\) be unitarily diagonalizable so that there is a unitary matrix \(U\) such that \(D=U \operatorname{diag}(\lambda_1,\dots, \lambda_d) U^*\). Define \(\varOmega^k_D(X)=U([\phi(\lambda_i,\lambda_j)] \circ [U^*XU])U^*\), where \(\circ\) denotes the Hadamard entrywise product. The authors also study the problem of when these operators are completely positive. A complete analysis of the behaviour of one-parameter families for which either the map or its inverse is completely positive is presented.

Recommendations

Cites work

Cited in

(14)

This page was built for publication: Families of completely positive maps associated with monotone metrics

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2435448)