eigen

Ross Ihaka ihaka@stat.auckland.ac.nz
Sat, 10 Apr 1999 14:19:09 +1200 (NZST)


On 10 Apr 1999, Peter Dalgaard BSA wrote:

> Paul Gilbert <pgilbert@bank-banque-canada.ca> writes:
> 
> > Eigen seems to be giving different results in 0.64 :
> > 
> > eigen(matrix(
> >  c(0.00000000,  0.0000000,  0.00000000, 1.000000000,
> >    0.01262442, -0.1141633,  0.04989433, 0.006112607,
> >    0.04781087,  0.4177869, -0.15569656, 0.074242237,
> >    0.20167331,  0.2531109, -0.04973573, 0.034046467) ,4,4))$values
> > 
> > R:
> >  [1]  0.35519256 -0.29294606 -0.26637369 -0.03168619
> > 
> > S:
> > [1]  0.4689288205 -0.4246063863 -0.2810596564  0.0009238293
> > 
> > I am fairly sure that S and R previously gave the same results.
> 
> Indeedy. At least the pre 63.3 binary I had lying about did give the S
> results. The eigenvectors are also bogus as far as I can tell.
> 
> The weird thing is that eigen() goes via .Fortran to eigen.c
> (which was f2c'ed at some point) and neither eigen.c or eigen.R has
> been changed in 1999 according to the cvs logs. 
> 

Weirder still.  I get the problem results under Solaris with a clean
checkout of 0.65 but not under FreeBSD with a clean 0.65.
Solaris is using egcs and FreeBSD is using gcc 2.7.2.1.

I am beginning to suspect a compiler / library bug.

	Ross

-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-devel mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-devel-request@stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._