Some operator and trace function convexity theorems (Q898796)

A function \(f:\mathbb{C}^{n \times n} \times \mathbb{C}^{n \times n} \rightarrow \mathbb{R}\) is said to be \textit{jointly operator convex} if \[ f(\lambda A_1+(1-\lambda)A_2, \lambda B_1+(1-\lambda)B_2) \leq \lambda f(A_1, B_1)+(1-\lambda)f(A_2, B_2) \] for all \(A_1, A_2, B_1, B_2 \in \mathbb{C}^{n \times n}\) and \(\lambda \in [0,1]\). \(f\) is said to be \textit{jointly operator concave} if \(-f\) is jointly operator convex. Let \({\mathcal P}_n\), \(n \geq 2\), denote the set of all \(n \times n\) positive definite matrices. One of the main questions that is addressed here is: when is the mapping \((A,B) \mapsto \operatorname{Tr}[(A^{q/2}B^pA^{q/2})^s]\) convex or concave, for \(A, B \in {\mathcal P}_n\). If \(p \in (1,2]\) and \(q \in (-1,0]\) then they show that this map is jointly operator convex if \[ s \geq \min \left\{\frac{1}{p-1}, \frac{1}{q+1}\right\}. \] The map is shown to be jointly operator convex in the case \(p=2\) for all \(q \in [-1,0)\) whenever \(s \geq \frac{1}{q+2}\). The authors also show the following result: For arbitrary nonzero real numbers \(p,q\), the map \[ (A,B) \mapsto A^{q/2}B^pA^{q/2} \] is jointly operator convex if and only if \(q=2\) and \(-1 \leq p < 0\); the map is not jointly operator concave. As an application, the operator convexity/concavity of the related trace functional \(\operatorname{Tr}[A^{q/2}B^pA^{q/2}C^r]\) is studied. They also prove that, if either \(r\) is a negative number or a positive fraction and \(Y\) is a positive semidefinite matrix of rank one, then the map \(X \mapsto X^rYX^r\) is not operator convex. This result is useful in constructing some counterexamples in the work.