dolfin team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Tue, Nov 01, 2005 at 11:00:52PM +0100, Johan Jansson wrote:
> On Tue, Nov 01, 2005 at 09:20:03PM +0100, dolfin@xxxxxxxxxx wrote:
> > Commit from logg (2005-11-01 21:20 CET)
> > ----------------
> > 
> > ODE: reject time steps
> > 
> >  - Check residual at end of time step, reject step if too large
> >  - Add safety factor to adaptive time step selection. Can be controlled
> >    by the parameter "safety factor"
> > 
> > Only implemented for mono-adaptive solvers so far. Will fix multi-adaptive
> > solvers next.
> > 
> 
> Great, this could also be used for the case when the solution for the
> discrete system doesn't converge for some reason.

Yes, that's the plan.

It looks like the sequence of build(), solve(), check(), shift() works pretty
well now, at least for the mono-adaptive solvers, so that one can
repeat the first three steps (not doing shift()) repeatedly without
any side-effects.

I'm still struggling with fine-tuning the time-step selection for the
multi-adaptive solver to rebuild the time slabs properly when
rejecting time steps to ensure convergence.

/Anders