Inequalities for the Heinz mean of sector matrices (Q826590): Difference between revisions

The numerical range of \(A\in {\mathbb M}_n\) is defined by \(W(A)=\{x^*Ax:~x\in{\mathbb C}^n,~x^*x=1\}.\) A matrix \(A\in {\mathbb M}_n\) is called a sector matrix if its numerical range is a subset of the sector \(S_\alpha\), \(\alpha\in[0,\pi/2)\), and \[ S_\alpha=\{z\in{\mathbb C}: \mbox{Re}\, z>0,~| \mbox{Im}\, z|\leq (\mbox{Re}\, z)\tan\alpha\}.\] The Heinz mean of two positive definite matrices \(A,B\in {\mathbb M}_n\) is defined by \[{\mathcal H}_\nu (A,B)=\dfrac{A\sharp_\nu B+A\sharp_{1-\nu}B}{2},\qquad 0\leq \nu\leq 1,\] where \(A\sharp_\nu B=A^{1/2}(A^{-1/2}BA^{-1/2})^\nu A^{1/2}\) is the weighted geometric mean. Due to the extension of the weighted geometric mean to sector matrices given by \textit{M. Raïssouli} et al. [C. R., Math., Acad. Sci. Paris 355, No. 6, 687--693 (2017; Zbl 1381.47011)], the Heinz mean of two sector matrices is then defined in the same way. In this paper, the authors extend some classical inequalities for the Heinz mean of positive definite matrices to the class of sector matrices. More precisely, the authors show that for sector matrices \(A,B\in {\mathbb M}_n\) with \(W(A),~W(B)\subseteq S_\alpha\) and \(\nu\in[0,1]\), \[\cos^2(\alpha) \mbox{Re}\,(A\sharp B)\leq \mbox{Re}\, {\mathcal H}_\nu(A,B)\leq \sec^2(\alpha)\mbox{Re}\,\frac{A+B}{2}. \tag{1}\] Here for \(X,Y\in{\mathbb M}_n\), \(X\leq Y\) if \(Y-X\) is positive semidefinite and \(\mbox{Re}\, X=(X+X^*)/2\) is the real part of \(X\). In addition, singular value inequalities and norm inequalities of the Heinz mean of two sector matrices are obtained as consequences of \(\mbox{Re},f (1)\). It is clear to see that every positive definite matrix is a sector matrix with \(\alpha=0\). Therefore, these inequalities reduce to classical cases when \(A,B\) are both positive definite. The techniques in the paper are pretty standard. One first observes that the real part of a sector matrix is positive definite. Two-sided inequalities on sector matrices \(A\in{\mathbb M}_n\) with \(W(A)\subseteq S_\alpha\) are known, as for example, \[\cos^2 (\alpha) (\mbox{Re}\, A)^{-1}\leq \mbox{Re}\,A^{-1}\leq (\mbox{Re}\, A)^{-1}\] and \[\cos^2 (\alpha)\sigma_j(A) \leq \sigma_j(\mbox{Re}\, A)\leq \sigma_j(A),\qquad j=1,\dots, n,\] where \(\sigma_j(A)\) is the \(j\)-th largest singular value of \(A\). Therefore, extensions can be obtained by applying the classical inequalities of positive definite matrices on the real part of sector matrices and the two-sided inequalities. Note that the geometric mean of two sector matrices has been defined for the first time in [\textit{S. Drury}, Linear Multilinear Algebra 63, No. 2, 296--301 (2015; Zbl 1310.15032)].