R devel - Apr 2003 - qr(x,LAPACK=TRUE) (PR#2867)

Hi, 

I think there is a problem with the LAPACK version of qr() in version
1.7.0. (version below).

1. The documentation states that LAPACK=TRUE is the default, but the code
has LAPACK=FALSE.

2. With LAPACK=TRUE qr() is never pivoting, even in cases where it very
clearly should be. e.g.

set.seed(0)
X<-matrix(rnorm(40),10,4);X[,1]<-X[,2]
qrx<-qr(X,LAPACK=TRUE)
qrx$pivot # note, no pivoting
qr.R(qrx) # note zero on leading diagonal is not in trailing block

3. ?qr includes the statement:

Using LAPACK (including in the complex case) uses column pivoting

4. If I call F77_NAME(dgeqp3) directly I get correct pivoting.

platform i686-pc-linux-gnu
arch     i686             
os       linux-gnu        
system   i686, linux-gnu  
status                    
major    1                
minor    7.0              
year     2003             
month    04               
day      16               
language R     

Apologies if I'm doing something dumb, but I can't see what it is at the
moment.

best,
Simon

_____________________________________________________________________
> Simon Wood simon@stats.gla.ac.uk        www.stats.gla.ac.uk/~simon/
>>  Department of Statistics, University of Glasgow, Glasgow, G12 8QQ
>>>   Direct telephone: (0)141 330 4530          Fax: (0)141 330 4814

Thanks for this.

On Mon, 28 Apr 2003 simon@stats.gla.ac.uk wrote:
> I think there is a problem with the LAPACK version of qr() in version
> 1.7.0. (version below).
> 
> 1. The documentation states that LAPACK=TRUE is the default, but the code
> has LAPACK=FALSE.
The code is the intended. Originally I intended to change the default, but 
discovered that quite a bit of code depended on the (generous) tolerance 
in the previous qr, not least to determine the rank.
> 2. With LAPACK=TRUE qr() is never pivoting, even in cases where it very
> clearly should be. e.g.
Well, it did pivot sometimes.  I had assumed that jpvt was zeroed, but it
was not guaranteed.  In fact I got pivoting sometimes in your example,
which threw me for a while.  It is now explicitly zeroed in R-patched, and
seems to be solved.

-- 
Brian D. Ripley,                  ripley@stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595