I need a way to decide if a given affine transformation preserves arbitrary rectangles in $\mathbb{R}^2$, meaning after applying it to any rectangle, it is still a rectangle afterwards.
Thought process
Given a matrix of the affine transform in the form $$ T_{2 \times 3} = \begin{pmatrix} t_{11} & t_{12} & t_{13}\\ t_{21} & t_{22} & t_{23} \end{pmatrix} $$ Translations ($t_{13}$ and $t_{23}$) preserve orthogonality and they don't have any influence over the orthogonality of the whole transformation
Formally, let's consider the full $3 \times 3$ affine matrix $$ T = \begin{pmatrix} t_{11} & t_{12} & t_{13}\\ t_{21} & t_{22} & t_{23}\\ 0 & 0 & 1\\ \end{pmatrix} $$ and apply Laplace expansion of its determinant along the third row. That gives us $$\det{T} = 1 \det{T_{\mathrm{lin}}}$$ where $$ T_{\mathrm{lin}} = \begin{pmatrix} t_{11} & t_{12} \\ t_{21} & t_{22} \end{pmatrix} $$ so we can only consider the matrix of a linear transformation $T_{\mathrm{lin}}$ because the determinant determines othogonality and if orthogonality is preserved, so is rectangularity.
Informally, we can always apply a purely translational transformation with $-t_{13}$ and $-t_{23}$ which does not change the (non-)orthogonality leaving us only with a linear transformation on the linear 2-dimensional subspace $$ T = \begin{pmatrix} t_{11} & t_{12} & 0\\ t_{21} & t_{22} & 0\\ 0 & 0 & 1\\ \end{pmatrix} $$
For this case it is known that the transformation preserves orthogonality if and only if $|\det T| = |\det T_{\mathrm{lin}}| = 1$ so we can use this condition to decide whether $T$ preserves rectangles.
Questions
Is this reasoning correct?
I do have the impression that it is still quite informal. Meaning that I did convince myself that only considering the $|\det T_{\mathrm{lin}}| = 1$ is the solution but generally speaking I did not consider everything that would need to be considered in order to accept it as a formal proof. What needs to be further considered in order to formally prove the above statement? Perhaps that orthogonal and only orthogonal transformations preserve rectangles so that just orthogonality of the transformation needs to be considered? Also the argument reducing the affine space to the linear subspace seems quite hand-wavy.