SAS function EIGVEC and R software function eigen

[LineMoon]

Hello expert,

My question is about EIGVEC function

Why for the same Matrix : M

in SAS  IML

H=eigvec(M) gives this error:

ERROR :Floating zero divide

In R software

if we write

eigen(V) gives the rusults.

We do not get an error.

So, why we have an error in sas, but we do not have an orror in R

Thank you

[Reeza]

Not to sound like a broken record, but please post data so we can replicate your issue. 

[Rick_SAS]

Eigenvectors are not unique, so in general you shouldn't expect that two software packages will compute the same set of eigenvectors.

However, it is interesting that one package stops with an error.

There are several reasons why SAS/IML could fail to give an answer, but it is impossible to guess without seeing the data. If you think the computation is incorrect, please open a track with SAS technical support and provide your version of SAS, the data, and the program.

[LineMoon]

thank that s very kind from you.
please,can you give somme of them

[Rick_SAS]

Anything I say would be purely conjecture. The reasons are typically related to the data (matrix). Numerical linear algebra computations are sensitive to the condition number of a matrix. So, for example, if your matrix is ill conditioned then numerical computations could be unstable and small roundoff errors could  accumulate.

[LineMoon]

Thank you.

SAS still usually uses this method to calculate eigvec , in page 518 ?

http://www.okstate.edu/sas/v7/saspdf/iml/chap17.pdf

For dbbeugging the IML program,

when I make call for eigvec(A)

I will possible to have somme options in sas 9.2 under unix to do this:

- stop running after the error message.

-debeugging the call of eigvec or making a break point

[Rick_SAS]

A few comments:

1. You cannont conclude that "SAS uses this method."  Different parts of SAS use different eigenvector computations, including sparse algorithms, multithreaded algorithms, and symmetric vs unsymmetric algorithms.

2. SAS 9.2 is very old... from 2008. It is possible that this problem does not exist in more recent releases.

3. You should open a track with Technical Support so that the SAS developers can determine the source of the problem.  Without seeing your matrix, there is no way to proceed.