New trace norm inequalities for \(2\times 2\) blocks of diagonal matrices (Q1887607): Difference between revisions

The matrix version of Hanner's inequality was proven for all \(p\geq1\) by \textit{K. Ball, E. A. Carlen} and \textit{E. H. Lieb} in [Invent. Math. 115, No. 3, 463--482 (1994; Zbl 0803.47037)]. It states that given \(n\times n\) complex-valued matrices \(X\) and \(Y\) with \(X+Y\) and \(X-Y\) positive semidefinite, then for \(1\leq p\leq2\) \[ \| X+Y\| _p^p+\| X-Y\| _p^p\geq(\| X\| _p+\| Y\| _p)^p+\left| \,\| X\| _p-\| Y\| _p\,\right| ^p \] with the reverse inequality holding for \(p\geq2\). The matrix version of Hanner's inequality can be expressed as \[ \left\| \begin{pmatrix} X&Y\\Y&X\end{pmatrix}\right\| _p \geq\left\| \begin{pmatrix}\| X\| _p&\| Y\| _p\\ \| Y\| _p&\| X\| _p\end{pmatrix}\right\|_p . \] The authors take this expression as the starting point to look out for generalizations. The first author showed in [Commun. Math. Phys. 242, No. 3, 531--545 (2003; Zbl 1049.15013)] that the inequality generalizes in the obvious way to positive semidefinite matrices of the form \(\left(\begin{smallmatrix} X&Y\\Y^*&Z\end{smallmatrix}\right)\). In the paper under review, the authors remove the condition of positive semidefiniteness under the additional assumption that the blocks are diagonal. Concretely, they prove that if \(A,B,C,D\) are \(n\times n\) diagonal matrices, then \[ \left\| \begin{pmatrix} A&B\\ C&D\end{pmatrix}\right\| _p \geq\left\| \begin{pmatrix}| A| &| B| \\ | C| &| D| \end{pmatrix}\right\| _p \geq\left\| \begin{pmatrix}\| A\| _p&\| B\| _p\\ \| C\| _p&\| D\| _p\end{pmatrix}\right\|_p \] if \(1\leq p\leq2\), with the inequalities reversed if \(p\geq2\). They also prove that in the case where the matrix is positive semidefinite and \(1\leq p\leq2\), the smallest (resp. biggest when \(p\geq2\)) possible value of the middle term of their inequality is achieved when its singular values are ordered in decreasing order along the respective diagonals.