[Date Index] [Thread Index] [Author Index]

Re: Eigensystem[] bug in Mathematica 5.0 (fixed in 5.0.1)

• To: mathgroup at smc.vnet.net
• Subject: [mg48548] Re: Eigensystem[] bug in Mathematica 5.0 (fixed in 5.0.1)
• From: ivo.bleylevens at facburfdaw.unimaas.nl (Ivo bleylevens)
• Date: Fri, 4 Jun 2004 04:51:11 -0400 (EDT)
• References: <c9ccon$6ha$1@smc.vnet.net> <c9ec93$ga3$1@smc.vnet.net> <c9k49r$fhu$1@smc.vnet.net>
• Sender: owner-wri-mathgroup at wolfram.com

Hi everyone,

I had the same problem as Marcus Stollsteimer discovered. I had a
asymetric complex 512x512 matrix and the eigenvalues Mathematica 5.0
calculated were totally wrong.
Check also post http://groups.google.com/groups?q=bleylevens&hl=en&lr=&ie=UTF-8&selm=c8mu5a%248nu%241%40smc.vnet.net&rnum=1.

Wolfram reported to me that this is fixed in version 5.0.1. And it
was, but........

With a 1024x0124 matrix the results are still wrong !! I yet don't
know if the reason for that is that the matrix is badly scaled or that
it is a matter of numerical inprecizion or something else.......in
this 1024x1024 case the results of
 - Mathematica 4.1   on a Pentium 4
 - Mathematica 5.0   on a Pentium 4
 - Mathematica 5.0.1 on a Pentium 4
 - Mathematica 5.0   on an AMD
 - Mathematica 5.0.1 on an AMD        

In the 512x512 case only the eigenvalues of Mathematica 5.0 on a
Pentium 4 were different. So I conclude that in the 1024x1024 case
Mathematica is doing wrong by a matter of numerical inprecizion .....I
think.....

Any suggestions ??

Ivo Bleylevens
University of Maastricht