[SciPy-dev] [SciPy-user] scipy.linalg.eig() returns transposed eigenvector matrix

[Travis Oliphant]

Robert Dick wrote:

>scipy.linalg.eig() returns transposed eigenvector matrix
>
>Results with old Numeric:
>  
>
>>>>import LinearAlgebra as la
>>>>from Numeric import *
>>>>la.eigenvectors(array([[1.0, 1.0], [1.0, 1.0]]))
>>>>        
>>>>
>(array([ 2.,  0.]), array([[ 0.70710678,  0.70710678],
>       [-0.70710678,  0.70710678]]))
>
>Results with svn current SciPy linked against AMD ACML BLAS/LAPACK.
>  
>
>>>>import scipy.linalg as la
>>>>from scipy import *
>>>>la.eig(array([[1.0, 1.0], [1.0, 1.0]]))
>>>>        
>>>>
>(array([ 2.+0.j,  0.+0.j]), array([[ 0.70710678, -0.70710678],
>       [ 0.70710678,  0.70710678]]))
>  
>
>>>>la.eig(array([[1.0, 1.0], [1.0, 1.0]]))[1].transpose()
>>>>        
>>>>
>array([[ 0.70710678,  0.70710678],
>       [-0.70710678,  0.70710678]])
>
>Can somebody else reproduce this?
>  
>

I can cofirm this. Notice that scipy_core basic eigenvalues work fine.  
I'm surprised this isn't being picked up by a test, though. 

import scipy.basic.linalg as sbl
sbl.eig([[1.0,1.0],[1.0,1.0]])

(array([ 2.,  0.]), array([[ 0.70710678,  0.70710678],
       [-0.70710678,  0.70710678]]))

import scipy.linalg as sl
sl.eig([[1.0,1.0],[1.0,1.0]])

(array([ 2.+0.j,  0.+0.j]), array([[ 0.70710678, -0.70710678],
       [ 0.70710678,  0.70710678]]))

It may be an issue with f2py and the new FORTRAN style arrays that can 
be created, because both functions return exactly the same data.  It's 
just that the hand-written routine in scipy_core (which is correct) has 
the CONTIGUOUS flag set, while the advanced routine generated 
automatically by f2py has the FORTRAN flag set.

-Travis