Sums and products of symplectic eigenvalues (Q2238861)

For a \(2n\times 2n\) positive definite matrix \(A\), denote by \(d_i(A)\), with \(i=1,\dots, n\), its symplectic eigenvalues. The author provides a symplectic analog of the Wielandt principle for canonical eigenvalues. More precisely, if we set \(1\le k\le n\), given the indices \(1\le i_1<\dots< i_k\le n\), then \[ \sum_{j=1}^k d_{i_j}(A)= \max_{\substack{\mathcal{W}_1 \supset\dots\supset \mathcal W_k\\ \dim\mathcal W_j=2n-i_j+1}}\ \min_{\substack{x_j,y_j\in\mathcal W_j\\ \text{sympl. orth.}}} \sum_{j=1}^k\frac{\langle x_j, A x_j \rangle+\langle y_j, A y_j \rangle}{2}, \] where ``sympl. orth.'' means ``symplectically orthonormal'', i.e., \[ \langle x_j, J x_j\rangle=\langle y_j, J y_j\rangle=0,\ \langle x_i, J y_j\rangle=\delta_{ij}, \] for any \(i,j=1,\dots,n\). Here, \(\delta_{ij}\) denotes the Dirac delta and \(J\) is the canonical symplectic identity: \[ J=\begin{pmatrix} O_n & I_n\\ -I_n & O_n \end{pmatrix}, \] \(I_n\) being the \(n\times n\) identity matrix and \(O_n\) the \(n\times n\) zero matrix. This principle has two consequences: first, it implies as a corollary the following Lindskii-type inequality: \[ \sum_{j=1}^k d_{i_j}(A+B) \ge \sum_{j=1}^k d_{i_j}(A) +\sum_{j=1}^k d_{j}(B). \] On the other hand, once we define the following geometric mean: \[ A\sharp B =A^{1/2} \left(A^{-1/2} B A^{-1/2} \right)^{1/2} A^{1/2}, \] then \[ \prod_{j=1}^k d_{i_j}(A) d_{j}(B)\le \prod_{j=1}^k d^2_{i_j}(A\sharp B) \le \prod_{j=1}^k d_{i_j}(A) d_{n-j+1}(B). \]