[FLASH-USERS] split solver, timestepping/refinement and the time accuracy order

[Andrea Gatto]

Dear Michael,

yes this is another option.
Thank you.

Best,

Andrea

> you may want to consider introducing a new timestep limiter that, at the
> start of the cycle, evaluates the source term, and returns a timestep
> estimate such that the energy released over the step is small compared to
> the internal energy in the cell.  You would then take the smaller of this
> and the Courant step, and do the reactions and hydro with this step.  If
> you mess around with the order of operations or the size of the timestep,
> you will affect the temporal order of accuacy.
>
>
> On Wed, Apr 24, 2013 at 8:40 AM, Andrea Gatto
> <andreag at mpa-garching.mpg.de>wrote:
>
>> Dear Klaus,
>>
>> thank you for your answer.
>> Yes using an unsplit solver would definitively solve the problem
>> (because
>> there isn't such problem anymore), but unfortunately I have to use the
>> split one.
>> Basically this timestep reduction is performed due to an enormous amount
>> of energy injected (Supernova) in the first call of the source terms. At
>> this point the only safe thing to do is to allow this injection of
>> energy
>> only in the second call of the source terms.
>> Many thanks.
>>
>> Best,
>>
>> Andrea
>>
>> > On Tue, 23 Apr 2013, Andrea Gatto wrote:
>> >
>> >> for stability reasons I find myself forced to reduce the timestep
>> after
>> >> the first call (in the first half of the Strang splitting loop) of
>> the
>> >> source terms when using a split solver in Driver_evolveFlash.F90 .
>> >> This means that the second call of the solver (in the second half of
>> the
>> >> loop) is performed on a different (smaller) timestep.
>> >
>> > Hello Andrea,
>> >
>> > I am wondering why you want to use the split solver?  It seems that
>> > your questions would not even arise if you used the Unsplit variant of
>> > Driver_evolveFlash.F90 .
>> >
>> >> Could this cause the Strang splitting scheme accuracy order to be
>> >> reduced
>> >> from 2 to 1?
>> >
>> > Yes, I think so.
>> >
>> >> Are there any other problems with this timestep change?
>> >> Could this problem happen also when I force a refinement in a certain
>> >> region during the first call of the source terms?
>> >
>> > I think changing the Grid in the middle of one "split timestep" (which
>> > should just be two steps of the same time interval) is asking for
>> trouble;
>> > again, why not go with the simpler Unsplit scheme?
>> >
>> > Klaus
>> >
>>
>>
>
>
> --
> Michael Zingale
> Associate Professor
>
> Dept. of Physics & Astronomy • Stony Brook University • Stony Brook, NY
> 11794-3800
> *phone*:  631-632-8225
> *e-mail*: Michael.Zingale at stonybrook.edu
> *web*: http://www.astro.sunysb.edu/mzingale
>