2
$\begingroup$

I am revising basic linear algebra concepts and I came across the concept of projecting vectors. Recall

The matrix that (orthogonally) projects the vector to which it is applied onto the vector $v$ is given by : $$ \mathcal{T}:=\frac{vv^{\top}}{\|v\|_{2}^{2}} $$ while : $$ \mathcal{S}:= I-\mathcal{T}=I-\frac{vv^{\top}}{\|v\|_{2}^{2}} $$ is the matrix that (orthogonally) projects the vector to which it is applied onto the space orthogonal to the vector $v$

I wanted to visualize both $\mathcal{S,T}$ for some $u,v,\in\mathbb{R}^{2}$. I have chosen $\mathbf{u}:=\begin{bmatrix}1&2 \end{bmatrix}^{\top}$ and $\mathbf{v}:=\begin{bmatrix}3&5\end{bmatrix}^{\top}$. On MATLAB I plotted $u,v,\mathcal{T}u, $ and $\mathcal{S}u$ all together and I obtained the following plot : $\hspace{2cm}$enter image description here

As seen above, $\mathcal{T}u$ (red) is a projection of $u$ to align on the same direction of $v$. Furthermore, $\mathcal{S}u$ (green) is an orthogonal projection of $u$. What I am curious about is how can we interpret the geometrical structure of $\mathcal{T}$ and $\mathcal{S}$ which I plotted both of them : enter image description here

The above surface is $\mathcal{T}$ which appears to be an oriented plane for some specific degree of orientation with respect to the $xy$ plane. Furthermore : enter image description here

is the surface $\mathcal{S}$ which I don't know how to describe it but I do notice symmetry in the middle. I would hope to know more about $\mathcal{S,T}$ as to how do they perform this reflection operation of $u$ and $v$ and what properties do they have they seem to be similar to the concept of Householder reflection.

Note that some textbooks define $\mathcal{T}u$ as $\widehat{u}$ and $\mathcal{S}u$ as $u^{\perp}$

Update : After doing some calculation, it appeared that $\mathcal{T}^{2}=\mathcal{T}$, which appears to be a property of $\mathcal{T}$ I believe this is meant to describe $\mathcal{T}$ geometrically, thus I hope someone can interpret this result for me.

$\endgroup$
8
  • 1
    $\begingroup$ Could you explain those second and third pictures? How have you taken these two matrices and turned them into these surfaces? $\endgroup$ Aug 4 at 8:04
  • $\begingroup$ I used MATLAB's command $\mathsf{surf()}$ to plot them @TheoBendit $\endgroup$
    – SPARSE
    Aug 4 at 9:07
  • $\begingroup$ Matlab is not one of my strong suits, but it sounds to me like you're plotting these matrices as functions of the form $x \mapsto Ax$. If this is the case, then something went wrong with your final picture. The function is linear, so the surface output should be a subspace. The fold that you're seeing is evidence that something has gone wrong. Or, it's evidence that I have totally misunderstood how surf would represent a matrix. $\endgroup$ Aug 4 at 18:06
  • 1
    $\begingroup$ $S-T$ is a reflection, namely the Householder reflection about the hyperplane orthogonal to $v$ ($v \neq 0$). So "geometrically", $S$ is a reflection plus a projection. A bunch of good things happens for $S$, and geometrically $T$ is a projection as $T^2=T$. $\endgroup$ Aug 6 at 7:32
  • 3
    $\begingroup$ I guessed and managed to recreate your plots. You are just doing surf(T) so matlab will interpret T as a heightmap over the indexed grid x=[1,2]; y=[1,2];. It's not a meaningful way to look at this data at all. $\endgroup$ Aug 7 at 0:40
1
+50
$\begingroup$

Let $V=\operatorname{span}(v)$ and $U=V^{\bot}=\{u\in \mathbb R^n\mid\langle u,v\rangle=0\}$. Then $V$ and $U$ are the eigenspaces of $\mathcal S$ as well as $\mathcal T$. We have $\mathcal S v=v-v\frac{v^\top v}{v^\top v}=0$ and $\mathcal T v=v\frac{v^\top v}{v^\top v}=v$. For $u\in U$, we get $\mathcal S u=u-v\frac{v^\top u}{v^\top v}=u$ and $\mathcal T u=v\frac{v^\top u}{v^\top v}=0$.

Hence, $\mathcal S$ is the orthogonal projection on $U$ and $\mathcal T$ is the orthogonal projection on $V$. Clearly, $\mathcal S^2=\mathcal S$ and $\mathcal T^2=\mathcal T$. Since after one orthogonal projection, we are already in the eigen space, so a second projection does not change anything. Further observe $\mathcal S\mathcal T=\mathcal T\mathcal S=0$, since $U\cap V=\{0\}$ and one operator maps on $U$ and the other maps on $V$.

You can visualize it, if you draw a plane and a line, which is orthogonal to the plane. The line has the direction $v$. Every point in the space will be mapped on the plane by $\mathcal S$. And it will be mapped on the line by $\mathcal T$.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.