0
$\begingroup$

Let $P$ be a projection in an n-dimensional vector space $V$. Let $\{z_{1}, ..., z_{r}\}$ be a basis for range($P$). Now extend it to a basis for $V$, i.e., $\{z_{1}, ..., z_{r}, z_{r+1}, ..., z_{n}\}$. I want to show each $z_{j}$ ($j>r$) is in null($P$). I would like to derive the result from basic principles, e.g., definition of a projection, linear independence, etc. (not from rank-nullity theorem or something like that).

Each vector $x\in V$ can be written as $x = P(x) + (x - P(x))$. The first part $P(x)$ is in range($P$) by definition. Apply $P$ to the second part $x-P(x)$ and we get $P(x - P(x)) = P(x) - P(P(x)) = P(x) - P(x) = 0$, i.e. $x - P(x)$ is in null($P$).

Now suppose $x = u + v$ where $u$ is in range($P$) and v is in null($P$). Apply P to the equation and we get $P(x) = P(u) + P(v) = u + 0 = u$. This show that $u$ has to be $P(x)$, i.e., the representation is unique.

My first idea is to write $P(z_{j}) = c_{1}z_{1} + ... + c_{r}z_{r}$ ($P(z_{j})$ is in range($P$)), and show $c_{1} = ... = c_{r} = 0$, but I can not move one step further.

My second idea is to start with null($P$). $\{z_{1}, ..., z_{n}\}$ of course spans null($P$), and I would like to remove $z_{1}, ..., z_{r}$ and show the remaining set still spans null($P$). I fail to do that either. Even if I do, it does not immediately imply $z_{j}$ ($j>r$) is in null($P$).

Any guidance would be greatly appreciated.

$\endgroup$
1
  • 2
    $\begingroup$ you wrote "Let $\{z_{1}, ..., z_{r}\}$ be a basis for range($P$). Now extend it to a basis for $V$, i.e., $\{z_{1}, ..., z_{r}, z_{r+1}, ..., z_{n}\}$. I want to show each $z_{j}$ ($j>r$) is in null($P$)" ... but this is False in general so you will keep getting stuck. I suggest focusing on the $n=2$ case. $\endgroup$ Commented Jun 16 at 2:04

2 Answers 2

4
$\begingroup$

Let $P \colon \Bbb R^2 \to \Bbb R^2$ be the projection defined by $P(x,y) = (x,0)$.

Note that $\{(1,0)\}$ is a basis for $\text{range}(P)$, which we can extend to a basis for $\Bbb R^2$ in many different ways, e.g. $\{(1,0),(1,1)\}$. Of course, $(1,1) \notin \text{null}(P)$.

$\endgroup$
2
$\begingroup$

Your statement is false in general. To be clear, first review the definition of projection:

Let $V=U\oplus W$. Then for any $v\in V$, $v$ can be uniquely written as $v=u+w $ with $u\in U$ and $w\in W$. The projector on $U$ along $W$ is defined as: $$ \begin{aligned} \mathcal{P}_U: V&\rightarrow U\\ v&\mapsto u \end{aligned}$$

Consequently, $U=\operatorname{range}(\mathcal{P}_U)$, $W=\ker \mathcal{P}_U$ and your statement is true iff $z_{r+1},\dots, z_n\in W$.

However, we have the following statement:

Proposition: Let $V$ be a $n$-dimensional inner product space, $U\leqslant V$, let $\mathcal{P}$ be the orthogonal projector on $U$. (i.e. the projector on $U$ along $U^\perp$). Let $\{z_1,\dots,z_r\}$ be an orthogonal basis of $U$, extend it to an orthogonal basis $\{z_1,\dots,z_n\}$ of $V$, then $\{z_{r+1},\dots, z_n\}$ is a basis of $U^\perp=\ker \mathcal{P}$.

$\endgroup$

You must log in to answer this question.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.