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

Re: Eigenvalues broken?

• To: mathgroup at smc.vnet.net
• Subject: [mg102837] Re: [mg102827] Eigenvalues broken?
• From: Vadim Oganesyan <vadim.oganesyan at gmail.com>
• Date: Sat, 29 Aug 2009 06:30:35 -0400 (EDT)
• References: <200907310952.FAA19221@smc.vnet.net> <h50sjj$6pl$1@smc.vnet.net>

Windows XP Pro x64

Interestingly, 32 bit laptop executes this code flawlessly.




On Fri, Aug 28, 2009 at 9:47 AM, Sseziwa Mukasa <mukasa at jeol.com> wrote:

> On Aug 28, 2009, at 12:47 AM, vadim.oganesyan at gmail.com wrote:
>
> I have been doing full diagonalization of sparse symmetric real
>> matrices (Hamiltonians) of different sizes and just noticed that
>> Eigenvalues (and Eigensystem) break once the matrices are 256by256
>> (maybe smaller, but that's where i first saw it).
>>
>> i first noticed the problem when checking the quality of the solutions
>> by comparing eigenvalues with expectation of the matrix in each of
>> eigenvectors.
>>
>> even more dramatically, simply rerunning Eigenvalues twice on the same
>> matrix produces somewhat different results!!!
>>
>> If I chose to get only few eigenvales (thus forcing it to use Arnoldi
>> method) the problem goes away.
>>
>> I was recently told that the version of Lapack used by Mathemica was
>> recompiled by people at Wolfram for added functionality -- did they
>> break it too?
>> My trusty C++ code has no problems with these matrices....
>>
>> thoughts? suggestions?
>>
>
> What platform are you using?  I've had problems with the numeric linear
> algebra routines on OS X, the same code seems to work on Windows.
>
> Regards,
>
> Sseziwa
>