10

I want to confirm if a projector matrix is its own inverse. I have $x=Px$ and $Px=P^2x$, so premultiplying the second equation with $P^{-1}$ twice, I get $P^{-1}x=Px$ for all x, implying $P^{-1}=P$. Is this reasoning correct?

So are all projection matrices orthogonal too?

Bravo
  • 4,413
  • Not all projections are invertible. – JavaMan Nov 04 '12 at 22:21
  • But $tr(P)=rank(P)=k$ for a $k\times k$ projection matrix, right? – Bravo Nov 04 '12 at 22:23
  • Consider the $2\times 2$ projection matrix $$P=\left(\begin{array}{cc}1 & 0\0 & 0\end{array}\right)$$ for an easy example of a non-invertible projection matrix. – Cameron Buie Nov 04 '12 at 22:27
  • @Shyam It's true that for projections $\rm{tr}(P) = \rm{rank}(P)$ but it's not true that the matrix is always full rank. – EuYu Nov 04 '12 at 22:28
  • Thanks all of you. Yes, I have $P=X(X^TX)^{-1}X^T$, where $X$ is full-rank. The counterexamples are illuminating. – Bravo Nov 04 '12 at 22:29
  • It is true, however, that all invertible projection matrices are self-inverse and orthogonal. ;) It just happens that the only invertible projection is identity. In general, only orthogonal (in the geometric sense) projections are orthogonal, since no one seemed to answer your second question... – tomasz Nov 04 '12 at 23:57

3 Answers3

15

No, this reasoning isn't correct because you assumed that the projection matrix has an inverse without proving it. In fact a projection matrix is a good example of a matrix that doesn't have an inverse: Part of the vector it is applied to is projected out, and there's no way to reconstruct that part.

joriki
  • 238,052
5

Projector matrices are idempotent, and as a rule, need not have an inverse at all (since it will usually have a non-trivial nullspace). For $P$ to be its own inverse, we need $P^2=I$. Since $P^2=P$ for any projector matrix, then the only projector matrix that is its own inverse is the identity (which we can think of as the trivial projector of a space onto itself).

Cameron Buie
  • 102,994
2

If by "projector matrix" you mean the matrix of a projection onto a (proper) subspace, then the rank of such a matrix will be the dimension of that subspace, which is less than the number of columns. So the nullspace is nontrivial, and the matrix is not invertible.

Gerry Myerson
  • 179,216