Suppose I have a vector $v$ which I want to orthogonally project onto a subspace $S.$ The subspace is defined as the space spanned by the columns of a matrix $A.$ I could do this by constructing a projection matrix $P$ via the common formula $P = A(A^TA)^{-1}A^T,$ and then computing the projection via $Px.$

However, we know that there exists some basis for the subspace $S$ such that the projection operator is diagonal, having diagonal elements equal to zero or one. We also know that orthogonal projection is independent of basis. So I should also be able to compute the projection via $Px,$ where $P$ is diagonal.

This seems strange to me - note that I didn't have to cast $x$ in any new basis, only $P.$ It would imply that projecting a vector onto a subspace can always be reduced to multiplication by a simple diagonal matrix.

I am guessing that my error is in asserting that $x$ need not be written in terms of the basis which diagonalizes $P.$ But I don't see why.

  • 3
    $\begingroup$ You write everyting in terms of the same basis: input vector, matrix, output vector. $\endgroup$ May 27, 2019 at 14:19
  • $\begingroup$ The fact that a projection is orthogonal is independent of basis. What the corresponding matrix looks like is not. $\endgroup$
    – user856
    May 28, 2019 at 5:31

1 Answer 1


By this reasoning a change-of-basis matrix should always just be an appropriately-sized identity matrix, since it represents the identity map. Just as a vector exists independently of a choice of basis, but its representation as a coordinate tuple is certainly basis-dependent, the linear map that is this projection exists independently of choice of basis, but its representation as a matrix is basis-dependent.


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.