Okay, I think it has to do with machine precision issues; in the configuration 
where the code runs, I had barely satisfied the convergence tolerance for CG 
(in topology optimization, the 'void' regions worsen the stiffness matrix 
conditioning); in the other configuration, it never reaches that tolerance and 
ends up diverging.  If I slightly relax that tolerance in this configuration 
though, the analysis runs just fine.

Thanks!

-Julian

--- El lun, 12/27/10, Wolfgang Bangerth <[email protected]> escribió:

De: Wolfgang Bangerth <[email protected]>
Asunto: Re: [deal.II] Problem with CG
A: [email protected]
Cc: "Julian" <[email protected]>
Fecha: lunes, 27 de diciembre de 2010, 11:46 am


Julian,

> I have a code using preconditioned (SSOR) CG that converges (in <500
> iterations) in one system (32 bit, Ubuntu 9.10, gcc 4.4.1, no threading,
> deal.II 6.2.1), but fails to converge (in 10,000 iterations) in another
> system (4 bit, RedHat 4, gcc 4.4.2, no threading, deal.II 6.3.1).  Has
> anyone run into a similar problem or might otherwise know what the problem
> could be?

These are too many variables and too little details you haven't explained. 
What did you try so far to find out what the problem is? For example, did you 
try to use the same deal.II version on both machines? Did you try to write 
out the matrices and compare their values? What happens if you use a very 
small linear system?

Your mail just doesn't have enough detail for anyone to give useful feedback 
beyond stabbing in the dark...

Best
 WB

-- 
-------------------------------------------------------------------------
Wolfgang Bangerth                email:            [email protected]
                                 www: http://www.math.tamu.edu/~bangerth/




      
_______________________________________________
dealii mailing list http://poisson.dealii.org/mailman/listinfo/dealii

Reply via email to