Hi Javier,
On Fri, Jan 13, 2012 at 3:32 PM, Javier Muñoz <[email protected]>wrote: > Hi Martin > > Thank you for the answer. Now I understand. I changed the maxit number > good . > to 2000 and the program advanced another cycle, but eventually gave the > same error. So, now my question would be, how many iterations are too > many? I mean, it is ok to increment maxit number to 3000, 4000, or more? > Typically, not so good. Each of those numbers usually involves at least one matrix vector multiplication and these quickly become expensive. > Regarding the preconditioning step, the program uses the Jacobi method. > You want to use a preconditioner that is tailored for your problem. For the discretization of an elliptic PDE you would typically use a multigrid preconditioner (geometric or algebraic). Very general preconditioners are based on incomplete factorisations and you could try one of those as the use of BiCGstab suggests that your problem is nonsymmetric. There are many other choices of preconditioners but it is hard to suggest the best without knowing what the underlying problem is. Ideally, you want the number of iterations to be mesh-independent so that if you refine (uniformly) the number of iterations stays constant but the time would increase by a factor of 4/8 depending on the setup. Best, Martin Is the selection of the preconditioner method mesh dependent? I'm not sure if these are naive questions, if so, I appreciate your > patience. > > Regards > Javier > > > El jue, 12-01-2012 a las 16:41 +0100, Martin Stoll escribió: > > Hi Javiear, > > > > I would guess that the solver cound not obtain the desired tolerance > > with the maximal number of iterations maxit=1000 number of iterations. > > In principal, you can increase the number of maxit as already > > suggested or you can make your system "nicer" by considering > > preconditioning. > > http://en.wikipedia.org/wiki/Preconditioner > > > > Best, > > Martin > > -- Martin Stoll Postdoctoral Research Fellow Computational Methods in Systems and Control Theory Max Planck Institute for Dynamics of Complex Technical Systems Sandtorstr. 1, D-39106 Magdeburg,Germany Email: [email protected] <[email protected]> URL: http://www.mpi-magdeburg.mpg.de/mpcsc/stollm/ Tel :+49 391 6110 805
_______________________________________________ dealii mailing list http://poisson.dealii.org/mailman/listinfo/dealii
