[petsc-users] TimeStepper norm problems. EMIL Please read this

[Emil Constantinescu]

On 3/24/15 5:31 AM, Lisandro Dalcin wrote:
> On 24 March 2015 at 06:15, Emil Constantinescu <emconsta at mcs.anl.gov> wrote:
>> On 3/23/15 4:44 AM, Lisandro Dalcin wrote:
>>
>> Lisandro, that's a neat idea. If you are basically moving in the multistep
>> realm for error estimation with one-step methods,
>
> I'm wondering if no one ever did it. I could not find any reference.

Not to my knowledge.

>> I think there are two
>> considerations:
>>
>> 1. The equations need to be continuous (no restarting, no events) or they
>> would have to reset the error estimator somehow.
>>
>
> Indeed. My generalized-alpha code does not handle events at all. I'll
> take a look into this. About restarting (not sure if we are talking
> about the same thing), I have some code that is specific to
> generalized-alpha (In the initial time step, I solve twice with
> backward-Euler using dt/2 and dt) that let me properly initialize
> generalized-alpha and also compute an initial error estimator.

Yes, that's sensible.

>> 2. Backward differences are sensitive to time stepping changes; unlike
>> one-step methods that can accommodate any step changes, B-D has strict
>> limits for stability. While this does not play a crucial role (it is just
>> the estimator);
>
> Of course, though my B-D is order 2, and as you said, it is the just
> for the estimator.
>
>> if the time step varies wildly (can be easily constrained),
>> it may create some problems with the estimator.
>>
>
> Indeed, the default clipping 0.1,10 is usually too wide, I usually do
> -ts_adapt_basic_clip 0.5,2.0

The limits are a bit more strict for BDF2, but that's for the worst case 
and no filtering. So that looks good too.

> Emil, is there any chance you can try my code with your problem? I
> really need some feedback to push this to PETSc, otherwise

Yes, we are a little backlogged, but we'll try it asap.

Emil