7
$\begingroup$

I am not really aware what's going on in this question. I appreciate your help.

Let $U$ be a vector space over a field $F$ and $p, q: U \rightarrow U$ linear maps. Assume $p+q = \text{id}_U$ and $pq=0$. Let $K=\ker(p)$ and $L=\ker(q)$.

Prove $p^2=p$ and $qp=0$.

$$p(p+q) = p(\text{id}_U) \Rightarrow p^2+pq=p \Rightarrow p^2 =p.$$ I actually first found this by letting $(\text{id}_U) =1$, although probably a wrong to do it.

Since $p^2 =p$ and $q$ is also a linear map defined by $q: U \rightarrow U$ then $q^2=q$. Again, I'm not sure if this actually is the right way to do it.

Then we have $$q(p+q) = q(\text{id}_U) \Rightarrow q^2+qp=q \Rightarrow q^2+ qp= q^2 \Rightarrow qp=0.$$

Prove $K=\text{im}(q)$.

For this question, I honestly do not know how to tackle it from the following definitions.

$K = \ker(p)=\left\{u \in U \ |\ p(u)=0_U\right\}$ and $\text{im}(q)=\left\{q(u) \ | \ u \in U \right\}$.

What's the correct way of doing these questions? Thank you for your time.

$\endgroup$
  • $\begingroup$ I can't really tell when you're multiplying functions and when you're composing them. $\endgroup$ – Git Gud May 30 '13 at 1:01
  • 1
    $\begingroup$ First treating $id_U=1$ is correct. Because when you multiply a matrix by identity matrix, you got the matrix. So you can just ignore identity matrix during matrix multiplication, as you ignore 1 during multiplication. $\endgroup$ – 1LiterTears May 30 '13 at 1:03
  • $\begingroup$ @MathSnail That would depended on what the OP means by $1$. $\endgroup$ – Git Gud May 30 '13 at 1:04
  • $\begingroup$ @GitGud Hi, I simply answered the earlier question "I actually first found this by letting (idU)=1, although probably a wrong to do it." $\endgroup$ – 1LiterTears May 30 '13 at 1:05
  • 1
    $\begingroup$ To answer the following (sub) question, this statement is correct: "Since $p^2 =p$ and $q$ is also a linear map defined by $q: U \rightarrow U$ then $q^2=q$. But I believe the reasoning is the following computation similar to the previous one. The claim you made "because $q$ is also a linear map" justified the commutativity of addition for linear maps, a.k.a $p+q = q+p$: $$q(q+p) = q(p+q) = q(\text{id}_U) \Rightarrow qp + q^2=q \Rightarrow q^2 =q.$$ $\endgroup$ – 1LiterTears May 30 '13 at 1:16
3
$\begingroup$

Since $\mbox{id}_U$ is the unit of the ring $L(U)$, it is standard to denote it by $1$.

So you know that $p$ is idempotent ($p^2=p$). It follows automatically that $q=1-p$ is idempotent as $q^2=(1-p)^2=1^2-p-p+p^2=1-2p+p=1-p=q$. And then $qp=(1-p)p=p-p^2=0$ as well.

Such a pair $(p,q)$ is the $n=2$ case of what is called an orthogonal decomposition of the unit. When such a thing happens, we have the Peirce decomposition which is very handy for many purposes. In operator theory/algebras in particular.

For any given idempotent $p$, the range and the nullspace yield a direct sum decomposition of the vector space $V=\mbox{im} p\oplus \ker p$. Conversely, a direct sum decomposition of $V=I\oplus K$ defines a unique idempotent such that $\mbox{im} p=I$ and $\ker p=K$. Cf. remark below. Now let us answer your question.

The relation between the idempotents $p$ and $q=1-p$ is $$ \mbox{im }(1-p)=\ker p\qquad \ker (1-p)=\mbox{im} \,p. $$

Proof: by symmetry, it suffices to prove one of these two properties. Let us do the lhs. Since $0=p-p^2=p(1-p)$, we get $p(1-p)x=0$ for every $x$ whence $\mbox{im}(1-p)\subseteq \ker p$. Now take $x\in \ker p$. Then $x=x-px=(1-p)x$ belongs to $\mbox{im} (1-p)$. QED.

Remark: it is also useful to remember that the idempotents are exactly the diagonalizable operators with spectrum in $\{0,1\}$. If $p$ is nontrivial ($p\ne 0,1$), we have $\ker p$ the eigenspace of $0$, and $\mbox{im} p=\ker(1-p)$ the eigenspace of $1$. That is $\mbox{im} p=\{x\in V\,;\,px=x\}$, as opposed to $\ker p=\{x\in V\,;\,px=0$.

$\endgroup$
  • $\begingroup$ Thank you for your answer. Would it be a better way to answer the first question if I write something as follows? Let $u$ be some vector where $u \in U$. Then start the proof by writing $(p(p+q))(u)$ or $p(u)(p(u)+q(u))$? Is that necessary, and which of them is the correct way to write it? $\endgroup$ – user4167 May 30 '13 at 1:35
  • 1
    $\begingroup$ @user4167 If you don't need to take vectors and can do it algebraically with the operators, that's better in general. So $p=p1=p(p+q)=p^2+pq=p^2$ is the best possible argument here. Nicely done. $\endgroup$ – Julien May 30 '13 at 1:37
2
$\begingroup$

First treating id$_U$=1 is correct. Because when you multiply a matrix by identity matrix, you got the matrix. So you can just ignore identity matrix during matrix multiplication, as you ignore 1 during multiplication.

Since $p^2 =p$ and $q$ is also a linear map defined by $q: U \rightarrow U$ then $q^2=q$.

But I believe the reasoning is the following computation similar to the previous one. The claim you made "because $q$ is also a linear map" justified the commutativity of addition for linear maps, a.k.a $p+q = q+p$:

$$q(q+p) = q(p+q) = q(\text{id}_U) \Rightarrow qp + q^2=q \Rightarrow q^2 =q.$$

Then following your own computation we're done.

$\endgroup$
1
$\begingroup$

You did the questions correctly (see below). It might be slightly more common to capitalize $P$ and $Q$, and it would be useful to see these as projections onto orthogonal subspaces to ge geometric intution. For instance, you could take $P : R^2 \rightarrow R^2$ to be $P(x,y) = (x,0)$ and $Q(x,y) = (0,y)$. In this case, the vector space is finite dimensional and the operators are defined on a basis. I find the best way to convince myself of the kind of operations you're carrying out is to say, "if $P$ and $Q$ both do the same thing to any $u \in U$, then $P = Q$."

To finish your question, we show $im(Q) = K = ker(P)$.

i) $im(Q) \subset ker(P)$. Let $v \in im(Q)$. Then $ v = Qw$ for some $w \in U$. Then $Pv = PQw = 0w = 0.$

ii) $ker(P) \subset im(Q)$. Let $v \in ker(P)$. Then $v = (P+Q)v = Pv + Qv = 0 + Qv$, so in fact $v = Qv$.

To address something mentioned in the comments: you might be treating the linear maps as matrices and implicitly using the fact that matrix multiplication corresponds to operator composition. In other words, I can define an operator $P$ on $\mathbb{R}^2$ that, say, rotates clockwise by $\pi/4.$ This is a linear map on $\mathbb{R}^2$, so if I pick a basis $\mathcal{B}$, I can represent $P$ by its standard matrix $[P]_\mathcal{B}$. Then if I have any other operator $Q$, $(P \circ Q)(v) = [P]_\mathcal{B}[B]_\mathcal{B}v_\mathcal{B}$.

$\endgroup$

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.