Matrix power means and the information monotonicity (Q513248)

For positive definite complex matrices \(A\) and \(B\), define \(A {\sharp}_t B:=A^{1/2}(A^{-1/2}BA^{-1/2})^tA^{1/2}\) denote the \(t\)-weighted geometric mean of \(A\) and \(B\). Let \(A_1,\dots,A_k\) be positive definite matrices and \(\omega=(\omega_1,\dots,\omega_k)\) be a weight vector so that \(\omega_i \geq 0\) for all \(i\) and \(\sum_{i=1}^k \omega_i=1\). Let the matrix power mean \(P_t(\omega; \mathbb{A})\) be defined to be the unique positive definite solution of the nonlinear equation: \(X=\sum_{i=1}^k {\omega}_i(X {\sharp}_t A_i)\) for \(t \in [0,1]\). A~mapping \(\Phi: \mathbb{C}^{n \times n} \to \mathbb{C}^{p \times p}\) is called \textit{positive} if \(\Phi(A) \geq 0\) whenever \(A \geq 0\). Here, \(A \geq 0\) denotes that \(A\) is positive definite. \(\Phi\) is called \textit{unital} if \(\Phi(I) =I\). Let the \(k\)-tuple of matrices \(A_i\) satisfy the condition that \(0 \leq m \leq A_i \leq M\) for each \(i\). The authors show that, if \(\Phi\) is a unital positive linear mapping, then, for every \(t \in [-1,1]\setminus \{0\}\), \[ P_t^2(\omega; \Phi(\mathbb{A})) \leq (\frac{(m+M)^2}{4mM})^2 {\Phi}^2(P_t(\omega; \mathbb{A})). \]