[SciPy-user] linalg.lu is

[Pearu Peterson]

On Fri, 20 Feb 2004, Nils Wagner wrote:

> linalg.lu cannot handle complex equations, where the rhs is complex.

This is not generally true but you are right, something goes wrong.
For example,

In [4]: from scipy.linalg import *

In [5]: a = [[1,1],[1,0]] # ok

In [6]: x0 = solve(a,[1,0j])

In [7]: dot(a,x0)
Out[7]: array([ 1.+0.j,  0.+0.j])

In [8]: a = [[1,1],[1.2,0]] # not ok

In [9]: x0 = solve(a,[1,0j])

In [10]: dot(a,x0)
Out[10]: array([ 0.83333333+0.j,  0.        +0.j])

When I disable ATLAS, then the second case gives correct answer.
So, I suspect it is an ATLAS bug. I am using 3.6.0. 
What version of ATLAS are you using?

Pearu