FLAME

I have also been looking at FLAME.  One operation that would be very
helpful in statistical computing is to have a QR decomposition with
pivoting only on approximate singularity.  That is, preserve the
original order of the columns unless a diagonal element would have a
magnitude less than some threshold, in which case the column is moved
to the last position.  That's what dqrdc2 in R and dqrdca in S do so
that the original order of terms is preserved in the partial sums of
squares but rank deficient model matrices can be accomodated.

Douglas Bates                            bates@stat.wisc.edu
Statistics Department                    608/262-2598
University of Wisconsin - Madison        http://www.stat.wisc.edu/~bates/