numpy-discussion

[Numpy-discussion] Numeric eigenvectors problem on Linux

[Alexandre G. <gi...@sc...>]

Hi,
I would like to know if anybody as experience any problem in their 
result using eigenvector function on Linux

The same calculation return the correct value on Windows,SUN,SGI.
All the Numeric were built without out modifying the setup.py file. I 
just run python setup.py install  (python2.3, Numeric 23.0)

The test script we run is the following:
import Numeric
import LinearAlgebra
e = Numeric.array([[ 0.6875,-0.3125,0.3125],
                    [-0.3125 , 0.6875 , 0.3125],
                    [ 0.3125 , 0.3125 , 0.6875]])
evals, evecs = LinearAlgebra.eigenvectors(e)
print "----------------------------------"
print "evecs",evecs


on Linux we  get:
evals [ 1.      0.0625  1.    ]
evecs [[ 0.81649658 -0.40824829  0.40824829]
	 [-0.57735027 -0.57735027  0.57735027]
  	[-0.26840956  0.80201256  0.533603  ]]

On all others platforms we get:
evals [ 1.      0.0625  1.    ]
evecs [[ 0.81649658 -0.40824829  0.40824829]
  	[-0.57735027 -0.57735027  0.57735027]
	 [-0.22645541  0.79259392  0.56613852]]


Any idea? Is it a bug in Numeric?

Thanks
Alex


-- 
   o Alexandre Gillet               email: gi...@sc...
  /  The Scripps Research Institute,
o   Dept. Molecular Biology,  MB-5,
  \  10550  North Torrey Pines Road,
   o La Jolla,  CA 92037-1000,  USA.
  /  tel: (858) 784-2053
o   fax: (858) 784-2860

Re: [Numpy-discussion] Numeric eigenvectors problem on Linux

[Konrad H. <hi...@cn...>]

On Tuesday 03 February 2004 23:43, Alexandre Gillet wrote:

> The same calculation return the correct value on Windows,SUN,SGI.
> All the Numeric were built without out modifying the setup.py file. I
> just run python setup.py install  (python2.3, Numeric 23.0)
>
> The test script we run is the following:
> import Numeric
> import LinearAlgebra
> e =3D Numeric.array([[ 0.6875,-0.3125,0.3125],
>                     [-0.3125 , 0.6875 , 0.3125],
>                     [ 0.3125 , 0.3125 , 0.6875]])
> evals, evecs =3D LinearAlgebra.eigenvectors(e)
> print "----------------------------------"
> print "evecs",evecs
>
>
> on Linux we  get:
> evals [ 1.      0.0625  1.    ]
> evecs [[ 0.81649658 -0.40824829  0.40824829]
> =09 [-0.57735027 -0.57735027  0.57735027]
>   =09[-0.26840956  0.80201256  0.533603  ]]

I get the same results. Checking with

  print Numeric.dot(evecs, e)-evals[:, Numeric.NewAxis]*evecs

yields

array([[  0.00000000e+00,  -5.55111512e-17,   5.55111512e-17],
       [ -4.16333634e-17,  -4.16333634e-17,   4.16333634e-17],
       [  5.55111512e-17,   0.00000000e+00,   0.00000000e+00]])

which doesn't look so bad. What does the same test produce on your other=20
machines?

Konrad.
--=20
-------------------------------------------------------------------------=
------
Konrad Hinsen                            | E-Mail: hi...@cn...
Centre de Biophysique Moleculaire (CNRS) | Tel.: +33-2.38.25.56.24
Rue Charles Sadron                       | Fax:  +33-2.38.63.15.17
45071 Orleans Cedex 2                    | Deutsch/Esperanto/English/
France                                   | Nederlands/Francais
-------------------------------------------------------------------------=
------

Re: [Numpy-discussion] Numeric eigenvectors problem on Linux

[Alexandre G. <gi...@sc...>]

Thanks Konrad,

The test runs on our others machines return similar result.
Also Nils explanation  give us a better understanding of what was going on.

Thanks again
Alex

Konrad Hinsen wrote:
> On Tuesday 03 February 2004 23:43, Alexandre Gillet wrote:
> 
> 
>>The same calculation return the correct value on Windows,SUN,SGI.
>>All the Numeric were built without out modifying the setup.py file. I
>>just run python setup.py install  (python2.3, Numeric 23.0)
>>
>>The test script we run is the following:
>>import Numeric
>>import LinearAlgebra
>>e = Numeric.array([[ 0.6875,-0.3125,0.3125],
>>                    [-0.3125 , 0.6875 , 0.3125],
>>                    [ 0.3125 , 0.3125 , 0.6875]])
>>evals, evecs = LinearAlgebra.eigenvectors(e)
>>print "----------------------------------"
>>print "evecs",evecs
>>
>>
>>on Linux we  get:
>>evals [ 1.      0.0625  1.    ]
>>evecs [[ 0.81649658 -0.40824829  0.40824829]
>>	 [-0.57735027 -0.57735027  0.57735027]
>>  	[-0.26840956  0.80201256  0.533603  ]]
> 
> 
> I get the same results. Checking with
> 
>   print Numeric.dot(evecs, e)-evals[:, Numeric.NewAxis]*evecs
> 
> yields
> 
> array([[  0.00000000e+00,  -5.55111512e-17,   5.55111512e-17],
>        [ -4.16333634e-17,  -4.16333634e-17,   4.16333634e-17],
>        [  5.55111512e-17,   0.00000000e+00,   0.00000000e+00]])
> 
> which doesn't look so bad. What does the same test produce on your other 
> machines?
> 
> Konrad.

-- 
   o Alexandre Gillet               email: gi...@sc...
  /  The Scripps Research Institute,
o   Dept. Molecular Biology,  MB-5,
  \  10550  North Torrey Pines Road,
   o La Jolla,  CA 92037-1000,  USA.
  /  tel: (858) 784-2053
o   fax: (858) 784-2860