Problem with eigen() and LAPACK
Dirk Eddelbuettel wrote:
Hi Hans, On Fri, Apr 18, 2003 at 11:55:33AM -0400, Hans Gardfjell wrote:
Hi all, when testing the new improvements in the new 1.7.0-version I stumbled over the following:
eigen(matrix(c(0,.3,2,.9),2,2))
Error in eigen(matrix(c(0,.3,2,.9),2,2)) :
LAPACK routine DGEEV gave error code -13
eigen(matrix(c(0,.3,2,.9),2,2),EISPACK=TRUE)
$values
[1] 1.3458236 -0.4458236
$vectors
[,1] [,2]
[1,] -1.1436890 -0.9760443
[2,] -0.7696018 0.2175718
My R-version
[1] "R version 1.7.0, 2003-04-16", running on Debian PC.
Am I correct in assumimg that you are using Debian 3.0 aka "stable", along with the R 1.7.0 package from CRAN?
No, I'm also running a "testing" Debian configuration. I first tried R from the CRAN testing depository, and then today installed the latest version from the unstable tree. So, just now I'm running the Debian precompiled R version 1.7.0-1 from the unstable branch at debian.org. I got the same result with both versions. However, when checking the dependencies I removed my atlas2-p3 package. And then it worked. I then realised that atlas2-p3 is replaced by atlas2-sse. So with an upgrade of atlas2 everything works fine. /hans
The example above works fine for me on Debian "testing". R on testing was built using the (almost) identical configuration at compile-time (gcc 3.0 and tcl/tk 8.3 on stable; 3.2 and 8.4 on testing) and the difference should therefore be attributable to the libraries and compilers used, which are, generally speaking, up to a year more recent on testing.
Anyone who has a clue why LAPACK refuses to solve the eigenvalues?
I suspect an problem with the LAPACK libs on the stable release. Matthias Burger reported a similar problem. We may just have to configure R to not use Lapack on that older platform unless we find a way to tame this. Dirk