A QR factorization of a $A - Qe_1 e_1^* R$.

220 Views Asked by At

I'm trying to solve the following problem of a 'perturbed' QR factorization in preparation for a test of mine.

Q: Let $A \in \mathbb{C}^{n \times m}, m \geq n$ have a thin $QR$ factorization $A = QR$. Determine the $QR$ factorization of $A - Q e_1 e_1^* R$ in terms of the $QR$ factorization of $A$.

Attempted Solution: This new matrix has an interesting structure, that is: $$ Q e_1 e_1^* R = \begin{bmatrix} Q(:,1)R(1,1) & \dots & Q(:,1)R(1,n) \end{bmatrix} $$ where $Q(:,1)$ is referring to the first column of $Q$. This tells us that in fact: $$ A - Q e_1 e_1^* R = \begin{bmatrix} 0 & A(:,2) - Q(:,1)R(1,2) & \dots & A(:,n) - Q(:,1)R(1,n) \end{bmatrix} $$ So this perturbed matrix is $A$ with the first column zeroed out and with some '$QR$' pieces removed. However, I don't see an easy way to directly find the $QR$ factorization of this? Is there perhaps an argument I can make from the orthogonality of $Q(:,1)$ with respect to the rest of $Q$, and then perhaps there's a argument of how to augment the original $Q$ to triangularize this?

Any thoughts?

EDIT: I have discovered via running an example in octave, that $Q^*(A - Qe_1 e_1^* R)$ actually seems to produce an upper triangular matrix, will report back if this results in something interesting.

1

There are 1 best solutions below

0
On BEST ANSWER

I'm foolish, this was actually easy.

Going off of my edit, we see that: \begin{align*} Q^*(A - Q e_1 e_1^* R) &= R - e_1 e_1^* R \end{align*} which is simply $R$ with its first row zeroed out, and critically is upper triangular. Thus $A - Qe_1 e_1^* R = Q (R - e_1 e_1^* R)$ is a QR factorization (which I could have alternatively seen by factoring a $Q$ out of the original matrix...)

Whoops.