In: Statistics and Probability
3. Let P be an nxn projection matrix (which means that it satisfies P2=P and PT=P). The goal of this problem is to show that P as a linear transformation is the orthogonal projection onto the range of P.
(a) Show that if v is in the range of P, then Pv=v.
(b) Show that if v is orthogonal to the range of P then Pv=0. (Hint: PPv lies in the range of P, hence is orthogonal to v. Study the dot product of v with PPv.)
(c) The previous parts show that P and the projection map act the same way on vectors that lie either in the range of P or the orthogonal complement of it. Why does this imply that P and the projection map agree on all vectors?
a) Let
, then
such that
b) Let be orthogonal to the Range(P), then: <.> (denotes dot product)
For each . This we already proved.
Then,
Now, in particular :
Rewriting the dot product:
So,
This is only possible if .
This completes the proof.
c) If two linear transformations map a basis to the same image then they are equal!!
We can write any vector as a sum of a vector in Range(P) and in the Nullspace of P. This is true from the Rank-Nullity theorem.
Orthogonal complement of Range(P)
The projection map maps this to w. And:
Hence, the projection matrix and map agree on all vectors.