Math Forum Discussions

Dieter Britz

Posts: 14
Registered: 6/17/10

Re: RGG question
Posted: Oct 20, 2011 8:44 AM

Plain Text

Alois Steindl wrote:

> Am 20.10.2011 10:55, schrieb Dieter Britz:
>> Dieter Britz wrote:
>>
>>> While I am finding out how toactivate ARPACK (thanks, Peter!), I want to
>>> make a start doing it the brute force way, i.e. using whole matrices
>>> rahter than sparse. I have RGG for this, but am now unsure about how to
>>> use it. In case someone out there knows RGG, here is the commentary:
>> [...]
>>
>> Is there anyone out there who knows someone I can contact about RGG?
>> I tried it out on a matrix for which I know the results, and I do not
>> find in the matrix, that should contain the eigenvectors, any values
>> that resemble the known values. I must be misusing the routine but I
>> don't know how. I assume that the original programmer, Garbow, is no
>> longer in charge or even accessible.
>>
>> I know, I should update myself and install ARPACK but it's a biggish
>> job, especially for someone fairly inept like me.
> Hello,
> I do not regard me as an expert for RGG, but just 2 more hints:
> Did you get the right eigenvalues? This would be a first test, whether
> something is seriously wrong with your attempts.
> Are you aware, that complex eigenvectors may be multiplied by arbitrary
> complex numbers (except 0, of course), so that the real and complex
> parts look quite different?
> If your matrices are not too large, you could simply calculate the
> product inv(Z)*A*Z and look, whether it is similar to the Real Jordan
> Normal Form.
>
> Good luck
> Alois

Thanks. I did meanwhile realise your point about a multiplicity of
eigenvectors, which explains why the example presents different vectors
from those RGG returns - all differing by the same factor. I also now
worked out how to pick out the complex eigenvectors. It's slow going
with my old brain, but I got there in the end. Thanks for your patience.
--
Dieter Britz (dieterhansbritz<at>gmail.com)