I'm trying to wrap my head around how QR decomposition can be directly used to find orthogonal bases for the fundamental subspaces.

I know that the bases can be obtained directly from the full QR factorization of a matrix, but I am unsure how to "extract" that information.

What is the best approach for doing so? To make things easier, assume that the starting matrix is $A$ and that the QR decomposition of $A^T=[Q_1 Q_2]R$ and, if needed, $A=[Q_3 Q_4]R$.


1 Answer 1


The four fundamental spaces are laid out completely with $A^T=[Q_1 Q_2]R$ and $A=[Q_3 Q_4]R$. I am supposing that you know this yourself since you actually split the $Q = [Q_1 Q_2]$ in the QR factorization.

The $R$ is the column mix of $Q$. $R$ shows exactly what columns from $Q$ form $A$ (and $A^T$ for the other factorization), and which columns don't, ie the null space of $A$. If $R$ is full rank, then $A$ is full rank.

$R$ will be triangular of course, and if $A$ has null space, $R$ will have rows that are zero. These rows if non-zero would select the columns from $Q$ in the formation of $A$, but since they are zero, then those respective columns from $Q$ are the null space of $A$. Row space if looking at $A^T$

  • $\begingroup$ I'm still a bit confused. I believe that the row space would be the span of the columns of $Q_1$, but what would the null space be? I believe it is related to $Q_2$, but I don't know how to write that basis given Q and R. $\endgroup$ Jan 30, 2013 at 4:31
  • 1
    $\begingroup$ Say the last row of $R$ is zero. This means that the null space is the last column of $Q$, because with $A = QR$, $R$ is a right multiplication of $Q$, and thus is a column mix of $Q$. The case of a zero row in $R$ then gives the last column of $Q$ is not used in the composition of $A$, so it is the column null space of $A$. $\endgroup$
    – adam W
    Jan 30, 2013 at 4:42

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.