[SciPy-user] Incorrect results in linalg.solve and friends

Pearu Peterson pearu at scipy.org
Sun Feb 22 03:53:18 CST 2004


Hi,

This is heads-up for linalg crunchers. Incorrect results appear when
the following condition holds:

  an input array is a *slice* of a complex array.

> for i in arange(0,n):

For example, changing
 
>    r = R[:,i] 

to `r = R[:,i].copy()` will give correct results.

>    X[:,i] = linalg.solve(A,r) 

This gives a workaround to avoid this problem until it gets resolved.
Also, this is not due to some bug in Atlas but rather of copy_ND_array in
fortranobject.c or of PyArray_CopyFromObject, but most probably of
copy_ND_array.

Pearu



More information about the SciPy-user mailing list