R devel - Dec 2003 - La.eigen hangs R when NaN is present (PR#6003)

Full_Name: Sundar Dorai-Raj
Version: 1.8.1
OS: Windows 2000 Professional
Submission from: (NULL) (12.64.199.173)


I discovered this problem when trying to use princomp in package:mva when a
column in my matrix was all zeros and I set cor = TRUE (thus division by 0).
Doing so hangs R, never to return. I have to shut down Rterm in the Task Manager
and lose all work from the current image. I tracked down the problem to using
La.eigen when the input matrix has one or more NaN. Note that eigen(x, sym TRUE)
gives a sensible result without hanging R.
> version
         _              
platform i386-pc-mingw32
arch     i386           
os       mingw32        
system   i386, mingw32  
status                  
major    1              
minor    8.1            
year     2003           
month    11             
day      21             
language R              
>
> x <- matrix(c(1, 1, 1, 0.5), 2, 2)
> x[1, 1] <- NaN
> x
     [,1] [,2]
[1,]  NaN  1.0
[2,]    1  0.5
> eigen(x, sym = TRUE)
$values
[1] NA NA

$vectors
[1] NA NA NA NA
> La.eigen(x, TRUE)
La.eigen(x, TRUE)

>From the NEWS for R-patched:
    o	Non-finite input values to eigen(), La.eigen(), svd() and
	La.svd() are now errors: they often caused infinite
	looping.  (PR#5406, PR#4366, PR#3723: the fix for 3723/4366
	returned a vector of NAs, not a matrix, for the eigenvectors.)

so this is already fixed (and your example works there).

On Mon, 22 Dec 2003 sundard@pdf.com wrote:
> Full_Name: Sundar Dorai-Raj
> Version: 1.8.1
> OS: Windows 2000 Professional
> Submission from: (NULL) (12.64.199.173)
> 
> 
> I discovered this problem when trying to use princomp in package:mva when a
> column in my matrix was all zeros and I set cor = TRUE (thus division by
0).
> Doing so hangs R, never to return. I have to shut down Rterm in the Task
Manager
> and lose all work from the current image. I tracked down the problem to
using
> La.eigen when the input matrix has one or more NaN. Note that eigen(x, sym
> TRUE) gives a sensible result without hanging R.
It did give an incorrect result: the eigenvector `matrix' wasn't a
matrix.

> > version
>          _              
> platform i386-pc-mingw32
> arch     i386           
> os       mingw32        
> system   i386, mingw32  
> status                  
> major    1              
> minor    8.1            
> year     2003           
> month    11             
> day      21             
> language R              
> >
> > x <- matrix(c(1, 1, 1, 0.5), 2, 2)
> > x[1, 1] <- NaN
> > x
>      [,1] [,2]
> [1,]  NaN  1.0
> [2,]    1  0.5
> > eigen(x, sym = TRUE)
> $values
> [1] NA NA
> 
> $vectors
> [1] NA NA NA NA
> > La.eigen(x, TRUE)
> La.eigen(x, TRUE)
> 
> ______________________________________________
> R-devel@stat.math.ethz.ch mailing list
> https://www.stat.math.ethz.ch/mailman/listinfo/r-devel
> 
> 
-- 
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