Abstract
This research extends the well-known extremal characterization of the geometric mean of two $n \times n$ positive definite matrices, $A$ and $B$, to more complex $3 \times 3$ block matrix structures. By solving the matrix completion problem for a missing block in a $3 \times 3$ positive semidefinite matrix, the authors identify explicit maximum and minimum values in the Loewner order using Schur complements and the standard matrix geometric mean.
Introduction
In matrix analysis, the geometric mean of two positive definite matrices $A$ and $B$ (denoted as $A \# B$) is more than a simple average; it represents a fundamental concept used in diverse fields such as quantum information theory and medical imaging. A key property of this mean is its "extremal characterization": it is the largest possible matrix $X$ that ensures the $2 \times 2$ block matrix
$$\begin{pmatrix} A & X \\ X & B \end{pmatrix}$$
remains positive semidefinite. While this $2 \times 2$ case is a staple of linear algebra, generalizing this property to larger systems has remained a significant theoretical challenge. This paper, published in the Journal of the Korean Mathematical Society, provides a definitive solution for $3 \times 3$ block matrix structures.
The Problem: Generalizing the "Missing Link"
Consider a $3 \times 3$ block matrix where some blocks are known ($A, B, C, V, W$) and one block ($X$) is missing. The problem asks: what is the largest possible value for $X$ (in the Loewner partial ordering) that keeps the entire $3 \times 3$ system positive semidefinite?
Prior to this work, researchers struggled to provide an explicit formula for this "missing link" in larger systems. The authors address this by identifying the constraints necessary for such a maximum to exist and providing a closed-form solution for its calculation.
Methodology and Technical Innovation
The authors utilize rigorous mathematical proofs centered on Loewner ordering (a partial ordering for Hermitian matrices) and Schur complements. The technical approach involves three primary steps:
- Reduction via Schur Complements: The $3 \times 3$ block matrix is "condensed" into a form that mirrors the simpler $2 \times 2$ case, allowing the researchers to apply known properties of the geometric mean to a more complex structure.
- Hermitian Constraints: A critical finding is the identification of a specific symmetry condition: $V B^{-1} W = W^* B^{-1} V^*$. This condition ensures that the resulting "center" of the completion is Hermitian, which is a prerequisite for a well-defined maximum in the Loewner order.
- Handling Non-Commutativity: To address cases where matrices do not commute, the authors employ the concept of $\Gamma$-commuting families, extending the applicability of their results to more general operator structures.
Key Results: Theorem 3.6
The central contribution of the paper is Theorem 3.6, which provides the explicit formula for the maximum completion $X$. If the $3 \times 3$ partial matrix is partially positive semidefinite and satisfies the required symmetry condition, the maximum $X$ is defined as:
$$X_{max} = V B^{-1} W + (E/B) \# (G/B)$$
In this formula:
- $V B^{-1} W$ represents the "central completion," acting as a bridge between the known blocks.
- $(E/B)$ and $(G/B)$ are the Schur complements of the outer sub-blocks.
- $\#$ denotes the standard matrix geometric mean.
Symmetry and Corrections
The research reveals a striking symmetry in the extremal values: the maximum and minimum completions are equidistant from the central completion. Specifically, if the maximum is $Center + Mean$, the minimum is $Center - Mean$.
Additionally, the paper serves a corrective role in the field. The authors identify and rectify a minor error in previous literature (specifically Corollary 3.5 in a 2014 paper by Lawson and Lim), establishing a more accurate necessary condition for Hermitian completions.
Implications and Future Work
While this paper successfully solves the $3 \times 3$ case for a single missing block, it also highlights the boundaries of current matrix theory. The requirement that $V B^{-1} W$ must be Hermitian is a significant constraint that may not always be met in practical engineering or physics applications.
The authors suggest that the "multivariable" geometric mean—involving more than two matrices—remains an open frontier. Future research may focus on extending these extremal characterization techniques to $n \times n$ block matrices or exploring scenarios where multiple blocks are missing simultaneously.
Summary updated