3
$\begingroup$

suppose I have a matrix $A_{1}\in C^{m*n}, m\geqslant n$ so the QR decomposition of $A_{1}$ is $A_{1}=Q_{1}*R_{1}$. Now, define the augmented matrix $A_{2}=\begin{bmatrix} A_{1}\\ I_{n*n} \end{bmatrix}$ where $I_{n*n}$ is $n*n$ identity matrix, so the QR decomposition of $A_{2}$ is $A_{2}=Q_{2}*R_{2}$.
Is there any relation between $Q_{1}$ and $Q_{2}$, $R_{1}$ and $R_{2}$, and especially between the diagonal elements of $R_{1}$ and $R_{2}$ knowing that $det(I_{n*n}+A_{1}^{H}*A_{1})=det(A_{2}^{H}*A_{2})$.

$\endgroup$
3
$\begingroup$

First consider the simple case, if $Q_1 = I$, then you want to find $Q_2$ and $R_2$ such that $Q_2 R_2 = \begin{bmatrix} R_1 \\ I \end{bmatrix}$. Split the columns of $Q_2^T$ into left and right halves such taht $Q_2^T = [U V]$. Note that $U^TU = V^TV = I$ and $U^TV = 0$. Then basically you want to find $R_2 = UR_1 + V$ which implies that $U^TR_2 = R_1$.

The more general case can be analyzed by considering the relation that $Q_2R_2 = \begin{bmatrix} Q_1R_1 \\ Q_1Q_1^T \end{bmatrix}$ and therefore $R_2 = UQ_1R_1 + V$ which leads to the result that $Q_1^TU^TR_2 = R_1$.

$\endgroup$
1
$\begingroup$

I think the relation is not that simple and here is why:

Since $A_1 = Q_1R_1$ we have $$ A_2 = \begin{bmatrix}A_1\\ I\end{bmatrix} = \begin{bmatrix}Q_1 & 0\\0 & I\end{bmatrix}\begin{bmatrix}R_1\\I\end{bmatrix}. $$ Now the QR decomposition of $\begin{bmatrix}R_1\\I\end{bmatrix}$ can be done by $n(n-1)/2$ Givens rotations and I don't see any reason, why this should lead to a simple relation of $R_1$ and $R_2$.

$\endgroup$

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.