[petsc-users] Suggestions for code with dof >> 1 ?

[Christophe Ortiz]

On Tue, Oct 8, 2013 at 2:25 PM, Jed Brown <jedbrown at mcs.anl.gov> wrote:

> Christophe Ortiz <christophe.ortiz at ciemat.es> writes:
>
> > On Tue, Oct 8, 2013 at 12:18 AM, Jed Brown <jedbrown at mcs.anl.gov> wrote:
> >
> >> Barry Smith <bsmith at mcs.anl.gov> writes:
> >> >> - In this type of problem, which TS scheme is recommended ?
> TSARKIMEX ?
> >> >
> >> >    Beats me.
> >>
> >> ARKIMEX should give you a decent integrator with adaptive error control.
> >> Use '-ts_arkimex_type 1bee' to use backward Euler with an
> >> extrapolation-based error estimator.
> >>
> >
> > Good to know. I tried TSBEULER but it has constant timestep.
> >
> > -Is there any other TS with adaptive timestep ?
>
> TSARKIMEX (nonlinearly implicit) and TSROSW (linearly implicit) have
> embedded error estimators and adaptive controllers.
>
> > -With ARKIMEX, is there a way to control the timestep ? For instance, is
> it
> > possible to control the max factor between two successive timesteps (dt'
> =
> > factor*dt), in order to avoid rejections ?
>
> -ts_adapt_basic_clip <shorten,lengthen> - Admissible decrease/increase in
> step size
> -ts_adapt_basic_safety <safety> - Safety factor relative to target error
> -ts_adapt_basic_reject_safety <rsafety> - Extra safety factor to apply if
> the last step was rejected
>


What are the names of the subroutines to access these options directly
within the code ? I see there is TSAdaptSetStepLimits but this is to set
the min and max timestep size.


>
> > - Is it possible to have Cranck-Nicholson
>
> Crank-Nicolson (spelling)
>
> > with adaptive timestep ? I tried TSCN but it seems timestep is
> > constant.
>
> What would you use as an error estimator?  The same approach as 1bee
> could be used to write an extrapolation-based method based on
> Crank-Nicolson.  Patches welcome if you do this (a simple exercise).
>
> I recommend using an A-stable ARKIMEX method.
>

Ok, thanks.


>
> > - I also tried TSROSW. Seems to work quite well in some cases. How does
> it
> > compare to ARKIMEX ?
>
> It is linearly implicit.  For problems with sometimes-"stiff"
> nonlinearities, ARKIMEX can often take longer time steps.  If ROSW is
> taking similar step sizes, it should be more efficient.
>

Good to know.

Christophe
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20131008/22887abb/attachment.html>