Re: Successive Over-Relaxation ... what is wrong? Improvements?

[Sergei Steshenko]

----- Original Message -----
> From: Joza <address@hidden>
> To: address@hidden
> Cc: 
> Sent: Monday, November 5, 2012 11:30 PM
> Subject: Re: Successive Over-Relaxation ... what is wrong? Improvements?
> 
> Oh, and maybe I haven't made it clear enough...i've gotten it to work 
> fine to
> for n=100 , but it won't converge for n=1000?
> 
> Again, main diagonals are 2, off diagonals to left and right are -1, all
> other values are zero. I'm really lost with this one. I can't see why 
> the
> same set of numbers will work fine for one size matrix, and not for
> another!!
> 
> 
> 
> --
> View this message in context: 
> http://octave.1599824.n4.nabble.com/Successive-Over-Relaxation-what-is-wrong-Improvements-tp4646041p4646135.html
> Sent from the Octave - General mailing list archive at Nabble.com.
> _______________________________________________
> Help-octave mailing list
> address@hidden
> https://mailman.cae.wisc.edu/listinfo/help-octave
>

For example, I've tried to use 'invfreqs' function from 'signal' package. It 
does work - for narrow by my standards band fitting, say, 2 .. 3 octaves.

And I found it impossible to use it for 12+ octaves fitting of real curves. And 
I do not think it's a problem of Octave or the function - with low order 
accuracy isn't sufficient and with high order the matrices involved become 
singular, so the final result is even worse than with lower order.

This is numeric world, accuracy issues show themselves from time to time.

Regards,
  Sergei.