Monotonicity of the matrix geometric mean (Q443939): Difference between revisions

Let \(A_{1},A_{2},\dots,A_{m}\) be elements of \(\mathbb{P}\), let \(G=G(A_{1},A_{2},\dots,A_{m})\) and \(\alpha =\sum \frac{1}{m} \delta^{2}_{2}(G,A_{j})\). For \(n\geq1\), let \(\mathcal{J}_{n}\) be the set of all ordered \(n\)-tuples \((j_{1},j_{2},\dots,j_{n})\), with \(j_{k}\in \{1,2,\dots,m \}\). This is a set with \(m^{n}\) elements. For each element of this set a ``mean'' \(\mathcal{M}_{n}(j_{1},j_{2},\dots,j_{n})\) of the given matrices \(A_{1},A_{2},\dots,A_{m}\) is defined by the following inductive procedure: \(\mathcal{M}_{1}(j)=A_{j}\) for all \(j\in\mathcal{J}_{1}\), \(\mathcal{M}_{n+1}(j_{1},j_{2},\dots,j_{n},k)=\mathcal{M}_{n}(j_{1},j_{2},\dots,j_{n})\sharp_{\frac{1}{n+1}}A_{k}\) for all \((j_{1},j_{2},\dots,j_{n})\) in \(\mathcal{J}_{n}\) and \(k\) in \(\mathcal{J}_{1}\). In this paper, the authors give an elementary proof of the below theorem, which was first proved by \textit{J. Lawson} and \textit{Y. Lim} [Math. Ann. 351, No. 2, 267--279 (2011; Zbl 1229.15024)], using standard matrix analysis and some counting arguments: Let \(A_{1},A_{2},\dots,A_{m}\) and \(A_{1}^{'},A_{2}^{'},\dots,A_{m}^{'}\) be positive definite matrices such that \(A_{j}\leq A_{j}^{'}\) for \(1\leq j\leq m\). Then \(G(A_{1},A_{2},\dots,A_{m})\leq G(A_{1}^{'},A_{2}^{'},\dots,A_{m}^{'})\). The authors arguments give simplifications of the proofs of joint concavity and continuity, which are the two important properties of the geometric mean. They give some new inequalities and results. The results can also be proved for the weighted geometric mean and their argument offers an easy transition from the equal-weights case to the general one.