11. Hermitian Matrices#

We now know the definition of adjoints. Let’s now look at self-adjoint maps

Self-adjoint maps#

Consider \((H, \mathbb{F}, \langle\cdot,\cdot\rangle _H)\). Consider linear and continuous map \(\mathcal{A}:H \rightarrow H\). Meaning that \(\mathcal{A}^*:H\rightarrow H\). \(\mathcal{A}\) is self-adjoint if \(\mathcal{A} = \mathcal{A}^*\). This also means that:

\[ \langle x, \mathcal{A}y \rangle = \langle \mathcal{A}x, y \rangle \; \forall x,y \in H \]

Hermitian matrices#

Let \(\mathcal{A} \simeq A = (a_{ij}), i, j \in 1\dots n\), with \(\mathcal{A}\) the operator (map) and \(A\) the matrix representation \(\in \mathbb{F}^{n\times n}\).

The map \(\mathcal{A}\) is self-adjoint iff \(A\) is hermitian, i.e. \(A = A^*\). The \(A^*\) in matrix means complex conjugate transpose, i.e. \bar{a_{ji}}$.

Remark

Hermitian matrices always have real eigenvalues.

Unitary matrices#

Definition: Unitary matrix

A matrix \(U\) is unitary iff

\[ U^*U = UU^* = I \]

Singular values#

Let us have the matrix \(A \in \mathbb{C}^{m\times n}\). We will have \(AA^* \in \mathbb{C}^{m\times m}\) a square matrix. Let \(\lambda_i , i=1\dots m\) be the eigenvalues of \(AA^*\). This also means that \(\lambda_i\) is real and non-negative.

If we have \(\lambda_1 \ge \lambda_2 \ge \dots \ge \lambda_r \dots \ge \lambda_m\). If \(r = \text{rank}(A) = \text{rank}(AA^*)\), then \(\lambda_1 \ge \lambda_2 \ge \dots \ge \lambda_r \ge 0\) and all the remaining eigenvalues are all zeros.

We define the singular values of the matrix to be the square root of the eigenvalues.

Definition: Singular values

The non-zero singular values of A are

\[ \lambda_i^\frac{1}{2}, i = 1\dots r \]