[[Hermitian Adjoint]]: Needs to know what conjugate transpose is doing to the matrix.
Theorem 2.1.9 in book \<Matrix Analysis\>
---
### **Intro**
> This is the theroem statement:
> $
> A\in \mathbb{M}_n \quad A^{-1} \sim A^H \iff
> B\in \mathbb{M}_n: A = B^{-1}B^H
> $
> The inverse of matrix $A$ is similar to $A^H$ if and only if $A$ can be written as the product of $B^{-1}B^H$.
The assumptions are:
* Matrix $A$ is invertible.
* Matrix $B$ is invertible.
**Proof $\leftarrow$**
$
\begin{aligned}
A &= B^{-1}B^H
\\
A^H &= (B^{-1}B^H)^H = (BB^{-H})
\\
A^{-1} &= B^{-H}B
\\
(B^H)\underbrace{B^{-H}B}_{A^{-1}}(B^{-H}) &= \underbrace{BB^{-H}}_{A^H}
\end{aligned}
$
If $A = B^{-1}B$ then $A^{-1}$ is similar to the matrix $A^H$ by the transformation represented by $B^H$
$\blacksquare$
**Proof $\rightarrow$**
Note: This direction is way harder to proof. We wish to show that:
> $
> A^{-1}\sim A^{H}\implies A = B^{-1}B^H
> $
Our task is to look for the matrix $B$. It's procedurals. Firstly we would like to find this representation of:
$
H = A^HHA \quad \text{ Where: }\quad H^H = H \wedge \exists \; H^{-1} \tag{1}
$
Let's start by considering that:
$
\begin{aligned}
& A \sim A^{H}
\\
\implies & SAS^{-1} = A^H \wedge S = A^HSA
\\
\text{Let: } & S_\theta = e^{i\theta}S
\\
\implies & S_\theta + S_\theta^H
=
A^HS_\theta A + A^H S_\theta^H A
\\
& S_\theta + S_\theta^H
=
A^H(S_\theta + S_\theta^H)A
\end{aligned}\tag{2}
$
Consider the choice of $\theta$ to make $S_\theta + S_\theta^H$ invertible, then:
$
\begin{aligned}
& \exists \; x \neq \mathbf{0}:
H_\theta x = \mathbf{0}
\\
\implies & (S_\theta + S_\theta^H)x =\mathbf{0}
\\
& x = -S^{-1}_\theta S_\theta^{H}x
\\
& x = e^{-2i\theta} S^{-1}S^Hx
\\
& S^{-1}S^{H}x = -e^{2i\theta}x
\end{aligned}\tag{3}
$
Therefore, to make it invertible we will have to choose $\theta\in (0, \pi]$ such that it's not an eigenvalue for the matrix $S^{-1}S^H$
Choose that $\theta$, then we consider $S$ using that $\theta^+$ which is then used to construct the matrix $H$:
> $
> H = A^HHA\tag{4}
> $
We are halfway there, let's talk about the motivation for the second part of the proof.
**Motivations and Tricks to Reach there**
Let's consider:
$
\begin{aligned}
B &= (I - A^H)H
\\
B^H &= H(I - A)
\end{aligned}\tag{5}
$
Then:
$
\begin{aligned}
A &= B^{-1}B^H
\\
BA &= B^H
\end{aligned}\tag{6}
$
To this regard, it would mean that:
$
\begin{aligned}
BA &= (I - A^H)HA
\\
&= (HA - A^HHA)
\\
&= (HA - H)
\\
&= H(A - I)
\\
B^H &= H(I - A)
\end{aligned}\tag{7}
$
Notice that, we are one constant way by expression (7), however, $B$ is invertible when the eigen value of $A$ contains 1, in that case, we are farther away from where we want to get, hence, let's consider:
**Some Extra Ingredients**
> $
> \begin{aligned}
> B &= \beta (\alpha I - A^H)H
> \end{aligned}\tag{8}
> $
> Where we need $|\alpha| = 1$ and $\alpha \in \mathbb{C}$, and we have the free choice for $\beta \in \mathbb{C}$, which will reveal itself after a while.
Let's consier again on the values before which is going to be:
$
\begin{aligned}
BA &= \beta(\alpha I - A^H)HA
\\
&= \beta(\alpha HA - A^HHA)
\\
&= \beta(\alpha HA - H)
\\
&= H(\beta\alpha A - \beta I)
\\
B^H &= H(\bar{\beta} \bar{\alpha}I - \bar{\beta}A)
\\
\implies
H(\bar{\beta} \bar{\alpha} - \bar{\beta}A) &= H(\beta\alpha A - \beta I)
\\
\implies \beta &= -\bar{\beta}\bar{\alpha}
\end{aligned}\tag{9}
$
If we were to assume the choice of $\alpha = e^{i\varphi}$ then it means that:
$
\beta = \exp \left(
\frac{i(\pi - \varphi)}{2}
\right)
$
And that concludes the proof both ways.
$\blacksquare$