Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

Let $P$ is an $n \times n$ matrix such that $P^2 = P$ and $P^t = P$. With this I am supposed to do two show/prove two conditions:

  1. Let $E = \operatorname{Col}(P)$. Show that $E^{\perp} = \operatorname{Nul}(P)$. (Hint: This is true for all symmetric matrices, ie. $P^2 = P$ is not necessary.)
  2. Show that $P$ is the matrix (with respect to the standard basis) of the orthogonal projection onto $E = \operatorname{Col}(A)$. That is, for each $x$ we have $\hat{x} = Px$. (Hint: $x = Px + (x - Px)$.)

What I know:

  • $\operatorname{Nul}(A^t) = \operatorname{Col}(A)^{\perp}$
  • Since $E=\operatorname{Col}(A)$ then the number of pivots of both $A^t$ and $A$ equals the number of columns of $A$.
  • $\dim E + \dim E^{\perp} = \dim\mathbb{R}^n = n $ so $\dim E^{\perp} = \dim\operatorname{Nul}(A) = n - p$

I just don't know how to approach answering this.

share|improve this question
Hint for the hint: What is $P(x-Px)$? –  EMS Apr 3 '12 at 0:24
I have no clue.... that was a homework question. All I know are the points under "what I know". :) –  Kyra Apr 3 '12 at 1:35
I bet you know a lot more than what you've listed under "what I know". I bet you know that if $A,B,C$ are matrices then $A(B+C)=AB+AC$. Now look at EMS' hint again. –  Gerry Myerson Apr 3 '12 at 1:42

1 Answer 1

up vote 2 down vote accepted

If $A$ is any matrix, then $\mathbf{v}\in\mathrm{Null}(A)$ if and only if $A\mathbf{v}=\mathbf{0}$. If you think about how you multiply a matrix by a column vector, you will see that this requires that $\mathbf{v}$ be orthogonal to each row of $A$ (I'm conflating a bit row and column vectors; if your vectors are columns, then transpose the rows of $A$ before taking the inner product with $\mathbf{v}$). Conversely, if $\mathbf{v}$ is orthogonal to every row of $A$, then $A\mathbf{v}=\mathbf{0}$. That is: $$\mathrm{null}(A) = (\mathrm{Rowspace}(A))^{\perp}.$$ This is true for any matrix.

If $A$ is symmetric, so that $A=A^t$, then the rowspace of $A$ equals its columnspace (because the columnspace of $A$ equals the rowspace of $A^t$). So if $A$ is symmetric, then $$\mathrm{null}(A) = (\mathrm{Rowspace}(A))^{\perp} = (\mathrm{columnspace}(A))^{\perp}.$$ So for $P$, since $P=P^t$, we have: $$\mathrm{null}(P) = (\mathrm{columnspace}(P))^{\perp} = E^{\perp},$$ which proves 1.

For part 2, we want to show that (i) $\mathrm{null}(P) = \mathrm{Im}(P)^{\perp}$; (ii) $\mathrm{Im}(P) = E$. The second clause, together with the fact that $P^2=P$, tells us that $P$ is a projection onto $E$. The first tells us that it is in fact an orthogonal projection.

But since $\mathrm{null}(P) = E^{\perp}$, in order to show the first clause we need to show that $\mathrm{Im}(P)=E$; that is, the second clause will give us both the first and second clauses, since we have already proven 1.

So we need to show that the image of $P$ is precisely $E$. But this is immediate: the image of any matrix is precisely its columnspace.

If you don't already know that $P^2=P$ implies that we have a projection, then we also need to show that $P$ is a projection. If $x\in V$, then I claim that $x-Px\in\mathrm{nullspace}(P)$: indeed, $$P(x-Px) = Px - P^2x = Px - Px = \mathbf{0},$$ where the equality $Px-P^2x = Px-Px$ is justified because $P^2=P$.

Now given any element $x\in V$, we can write $x = x + (Px - Px) = Px + (x-Px)$. This shows that we can write $x$ as an element in $\mathrm{Im}(P)$ plus an element in $\mathrm{nullspace}(P)$. That is, $$V = \mathrm{Im}(P) + \mathrm{nullspace}(P).$$ By the rank-nullity theorem, $\dim(V) =\mathrm{rank}(P) + \mathrm{nullity}(P)$, so we must have $\mathrm{Im}(P)\cap\mathrm{nullspace}(P) = \{\mathbf{0}\}$, so the sum is a direct sum; that is, $$V = \mathrm{Im}(P) \oplus \mathrm{nullspace}(P)$$ and for every $x\in V$, to write $x$ as a sum of a vector in $\mathrm{Im}(P)$ and a vector in $\mathrm{nullspace}(P)$, we write $x = Px + (x-Px)$; so $P$ is the projection onto the first component, i.e., $P$ is the projection onto its columnspace. Since the nullspace is orthogonal to the image, $P$ is an orthogonal projection.

share|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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