logo

Hermitian Matrix's Loewner Order 📂Matrix Algebra

Hermitian Matrix's Loewner Order

Definition

Loewner Order

Let’s assume two matrices A,BCn×nA, B \in \mathbb{C}^{n \times n} are Hermitian matrices. If (AB)\left( A - B \right) is positive semidefinite, it is denoted as ABA \ge B, and if (AB)\left( A - B \right) is positive definite, it is denoted as A>BA > B. This kind of partial order \ge, >> are called the Loewner order.

Explanation

Unlike commonly considered scalars, and indeed even with complex numbers, defining a natural order among them is not an easy task. Especially in the case of matrices, which have many more components, defining a valid order for them is difficult. However, the notion of positive definiteness in matrices provides a way to assign a kind of sign to matrices, enabling the establishment of an order among them.

Theorem

Hn:={ACn×n:A=A}Pn:={ACn×n:xAx>0}Pn:={ACn×n:xAx0} \begin{align*} \mathbb{H}_{n} :=& \left\{ A \in \mathbb{C}^{n \times n} : A = A^{\ast} \right\} \\ \mathbb{P}_{n} :=& \left\{ A \in \mathbb{C}^{n \times n} : \mathbf{x}^{\ast} A \mathbf{x} > 0 \right\} \\ \overline{\mathbb{P}_{n}} :=& \left\{ A \in \mathbb{C}^{n \times n} : \mathbf{x}^{\ast} A \mathbf{x} \ge 0 \right\} \end{align*} For the set of positive definite matrices in the vector space of Hermitian matrices Hn\mathbb{H}_{n}, denoted as Pn\mathbb{P}_{n}, let’s denote the set of positive semidefinite matrices as Pn\overline{\mathbb{P}_{n}}. Pn\overline{\mathbb{P}_{n}} is a partially ordered set with respect to \ge, but cannot be a totally ordered set.

Proof

It suffices to provide a counterexample. A:=[1000]B:=[0001] \begin{align*} A :=& \begin{bmatrix} 1 & 0 \\ 0 & 0 \end{bmatrix} \\ B :=& \begin{bmatrix} 0 & 0 \\ 0 & 1 \end{bmatrix} \end{align*} The two matrices A,BA, B given above are positive semidefinite; however, AB=[1001]BA=[1001] \begin{align*} A - B =& \begin{bmatrix} 1 & 0 \\ 0 & -1 \end{bmatrix} \\ B - A =& \begin{bmatrix} -1 & 0 \\ 0 & 1 \end{bmatrix} \end{align*} for any vector (x1,x2)R2\left( x_{1} , x_{2} \right) \in \mathbb{R}^{2}, its quadratic form always becomes of the form x12x22x_{1}^{2} - x_{2}^{2}, which cannot all be positive semidefinite. In this case, since neither A>BA > B nor B>AB > A holds, Pn\overline{\mathbb{P}_{n}} cannot form a totally ordered set.