Matrix monotonicity and concavity of the principal pivot transform (Q6185991)

Let \(A\in M_n\) be partitioned into a \(2\times 2\) block matrix, \begin{gather*} A=[A_{ij}]_{i,j=1,2}=\begin{bmatrix} A_{11} & A_{12}\\ A_{21} & A_{22} \end{bmatrix}, \end{gather*} where \(A_{ij}\in M_{n_i\times n_j},\,\, i,j=1,2\). The (generalized) principal pivot transform of \(A\) with respect to the \((2,2)\)-block \(A_{22}\) is defined by \[ \operatorname{PPT}(A):= \begin{bmatrix} A/A_{22} & A_{12}A_{22}^{+}\\ A_{22}^{+}A_{21} & -A_{22}^{+} \end{bmatrix}\in M_{n}, \] where \(A/A_{22}=A_{11}-A_{12}A_{22}^{+}A_{21}\) is the (generalized) Schur complement of \(A\) with respect to the \((2,2)\)-block \(A_{22}\) and \(A_{22}^+\) denotes the Moore-Penrose pseudoinverse of \(A_{22}\). In this paper, the authors prove the minimization variational principle for \(\operatorname{PPT}(\cdot)\) and employ it to prove the matrix concavity of this map, in the sense of the Löwner ordering. They also show that \(\operatorname{PPT}(\cdot)\) is matrix monotone under minimal hypotheses. Their results improve those of \textit{J. E. Pascoe} and \textit{R. Tully-Doyle} [Linear Algebra Appl. 643, 161--165 (2022; Zbl 1490.15009)].