1

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

  1. Is this reasoning correct?

  2. 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.

VaNa
  • 163
  • 8
  • 2
    Reflections also preserve orthogonality so determinants valued $\pm 1$ should be considered not merely $1$. More so scaling will preserve rectangles as well so matrices with determinants of all values can be valid. You need to exclude shear matrices since those break orthogonality. https://en.wikipedia.org/wiki/Shear_matrix – CyclotomicField Sep 04 '23 at 18:31
  • @CyclotomicField I do consider $\pm 1$ by taking the absolute value of the determinant. True, even scaling does preserve rectangles and has non-unit determinant. But uneven scaling (different in the $x$ and $y$ direction) does not seem to preserve arbitrary rectangles. Shear obviously doesn't preserve them. – VaNa Sep 04 '23 at 18:44
  • 1
  • Your reasoning about focusing on the "linear part" looks all right. 2. As CyclotomicField's comment notes, "preserves orthogonality" if and only if $|\det T| = 1$ is wrong in both directions. ($|\det T| = 1$ doesn't imply "preserves orthogonality" and "preserves orthogonality" does not imply $|\det T| = 1$.)
  • – Andrew D. Hwang Sep 04 '23 at 19:01
  • Let $S$ be the translation you mention. If you know that $S$ (and $S^{-1}$) preserves the orthogonality, what do you can say of relations between orthogonality-preserving properties of the mappings $T\circ S$, $T \circ S$, $T\circ S^{-1}$, $ S^{-1}\circ T$. Here I write $T\circ S$ for composition, may be you write the same as $TS$. May something of this help you to convert some informal part of the argument? – minorChaos Sep 04 '23 at 19:40
  • 1
    @AndrewD.Hwang Now it struck me. Shear transformation has $\det = 1$ and it does not preserve angles ($\det$ is about volume preservation). Similarly, uniform scaling with scaling factor $s$ has $\det = s^2$ and does preserve angles. So yes, the condition is insufficient in both ways. – VaNa Sep 05 '23 at 07:58