[SciPy-user] small error problem

[Travis Oliphant]

Ryan Krauss wrote:

>I have a script that uses row reduction to mind the null space of a
>matrix.  I am experiencing some strange behavior.  At the begining of
>the script, I have a 2x2 matrix, but it is of rank 1.  (I realize that
>find the null space of a 2x2 could be much simpler, but I will
>generally have higher ranked matrices.  I have also found the
>nullspace using svd, but have gotten the best results from rref in the
>past.)
>  
>
which version of scipy are you using?

>The fact that the matrix is rank one can be verified early in the
>script.  Row 2 is a multiple of row 1:
>(Pdb) mat[1,1]/mat[1,0]-mat[0,1]/mat[0,0]
>Out[2]: NumPy array, format: long
>0.0
>(Pdb) '%0.30f'%(mat[1,1]/mat[1,0]-mat[0,1]/mat[0,0])
>Out[2]: '0.000000000000000000000000000000'
>
>In order to do the row reduction, I append a column of zeros:
>(Pdb) vect
>Out[2]: NumPy array, format: long
>[[ 0.,]
> [ 0.,]]
> 15  ->     mat=c_[mat,vect]
>(Pdb) mat
>Out[2]: NumPy array, format: long
>[[ 1.50155914,-1.24770975,]
> [-2.98977017, 2.48432803,]]
>
>But now there is some small error:
>(Pdb) '%0.30f'%(mat[1,1]/mat[1,0]-mat[0,1]/mat[0,0])
>Out[2]: '-0.000000013035565737951060327759'
>(Pdb) mat[1,1]/mat[1,0]-mat[0,1]/mat[0,0]
>Out[2]: -1.303556573795106e-08
>
>and the output is now a scalar instead of NumPy array, format: long
>  
>
I'm not sure what NumPy array, format: long means

It does seem strange that the expression is different things in 
different places. 

Could you post more information.  What is the result of type(mat), for 
example.

Thanks,

-Travis