[petsc-dev] How come TSBDF has its own flag for turning on adaptivity?

[Jed Brown]

Lisandro Dalcin <dalcinl at gmail.com> writes:

> On 19 April 2017 at 02:36, Barry Smith <bsmith at mcs.anl.gov> wrote:
>>
>>    Lisandro,
>>
>>      How come TSBDF has its own flag for turning on adaptivity?
>>
>
> Well, it is there from the time this code was inside PetIGA. I started
> BDF following my ALPHA implementations, where the error estimator is
> not sharp but approximate (and this strategy was also implemented in
> THETA).
>
> In the case of BDF, the error estimator is sharp (I mean, it is
> computed of BDF formulas of lower order), so I think you are right,
> there is no point on having this flag.
>
> Finally, I would like to mention that adaptivity with BDF has not been
> tested exhaustively by me or my collaborators, so I was not completely
> confident to turn it on by default.
>
> Anyway, some day we should also consider defaulting to use adaptivity
> in ALPHA and THETA, even if the error estimator is approximate.
>
>>
>> Is there anyway this can be removed so the standard TS flags can be used to control turning on adaptivity for TSBDF?
>>
>
> Yes, the changes should be quite easy to implement.
>
>>
>> Even if TSBDF needs its own custom adapt code that is fine, but it should be able to be turned on (probably should be on by default anyways) without a special switch.
>>
>
> I would like your opinion on code located at TSSetUp_BDF() related to
> clipping factors. I'm restricting the high clipping factor to 2.0, as
> the default (10.0) is too much in practice, 

I definitely believe that for BDF.  Not sure I believe it for some
one-step methods.  Are you intending to change it for all methods or
just for BDF?

> there are theoretical reasons related to stability (although the value
> depends on the order of the BDF), and this factor of 2.0 is used also
> on Sundials (regardless of the order).

Cool.

> I'm not going to object the default factor of 10.0, otherwise Jed
> Brown will get mad and start MOABing me from the highs of a
> mountain...

I'll just threaten you with an aircraft carrier that is in a different
ocean and going the other direction.  ;-)

> If you are OK with that piece of code, then I can put a PR for your review.
>
>
> -- 
> Lisandro Dalcin
> ============
> Research Scientist
> Computer, Electrical and Mathematical Sciences & Engineering (CEMSE)
> Extreme Computing Research Center (ECRC)
> King Abdullah University of Science and Technology (KAUST)
> http://ecrc.kaust.edu.sa/
>
> 4700 King Abdullah University of Science and Technology
> al-Khawarizmi Bldg (Bldg 1), Office # 0109
> Thuwal 23955-6900, Kingdom of Saudi Arabia
> http://www.kaust.edu.sa
>
> Office Phone: +966 12 808-0459
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 832 bytes
Desc: not available
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20170419/a7f09b5f/attachment.sig>