[FLASH-USERS] [EXTERNAL] Re: Adaptive timestep

[Sean M. Couch]

It went beyond cocktail party discussion: https://link.springer.com/chapter/10.1007/3-540-27039-6_7

Basically, if ~50% of the number of blocks are at the highest refinement level then time refinement is not efficient. But lots of problems don’t meet this criterion, these days…

Sean

----------------------------------------------------------------------
Sean M. Couch, Ph.D.
Assistant Professor
Department of Physics and Astronomy
Department of Computational Mathematics, Science, and Engineering
Facility for Rare Isotope Beams
Michigan State University
567 Wilson Rd, 3260 BPS
East Lansing, MI 48824
(517) 884-5035 --- couch at pa.msu.edu --- www.pa.msu.edu/~couch
On Oct 1, 2019, 12:38 PM -0400, Carlo Graziani <carlo at oddjob.uchicago.edu>, wrote:
At the level of overheard cocktail-party chitchat, I recall a discussion
of subcycling in Flash in which the assertion was made that for oct-tree
AMR the efficiency gains from subcycling are expected to be minor.

The reasoning, which seemed plausible, was that in a typical 3-D
simulation the blocks at the highest level of refinement outnumber other
blocks by a large factor, so that reducing the computational load due to
less-refined blocks by updating them less frequently doesn't drop the
time-to-solution by very much. This was the reason adduced for not
making subcycling development a high priority.

Is this no longer believed to be the case?

Carlo

On 10/1/19 11:26 AM, Ricker, Paul Milton wrote:
It would be good to come up with a strategy for developing this feature in FLASH 5. It's been a desired feature for a long time, but it has essentially been triaged out of development since it is not easy and not the most urgent feature for most developers' science. The transition to AMReX is a good opportunity to revisit the question of how hard it would be to implement.

I'll just put that on my to-do list... :)

Paul


Paul M. Ricker
Professor of Astronomy
University of Illinois
https://urldefense.proofpoint.com/v2/url?u=http-3A__sipapu.astro.illinois.edu_-7Ericker&d=DwID-g&c=nE__W8dFE-shTxStwXtp0A&r=ISY2CKBuLFWqkB5oZl4IHg&m=IJiWOXgTCbRVoZ3UYUedmPBlkK6XeXJD2dGVYHz5O8A&s=ML9CnmlN_Vd7dGdoJ9nUYWKIidw-Zz-4xelZX125xpo&e=

On October 1, 2019 11:03:41 AM "Messer II, Bronson" <bronson at ornl.gov> wrote:

Hi Sean, Matuesz, et al.,

Sean is right both about not having sub-cycling in the current solvers and about the desire to change that in the future for FLASH5.

So, I guess I didn’t have a different opinion at all, just felt the need to chime in. :)

Bronson

On Oct 1, 2019, at 11:33 AM, Sean M. Couch <couch at pa.msu.edu> wrote:

Hi Mateusz,

I will venture my view of this and others can chime in if they have a different opinion. Also, I’m assuming by adaptive timestep you mean sub-cycling in time where different refinement levels take different time steps?

For FLASH5, we are transitioning to the AMReX package. AMReX is built on the idea of time sub-cycling so in principle, we could utilize that. For that matter, PARAMESH was also built to allow time sub-cycling, we just never used it. As of now, for the solvers we are porting to FLASH5/AMReX we are still making the assumption of a single global time step. This simplifies things considerably. But I am of the opinion that, eventually, we should implement time sub-cycling and take advantage of what AMReX offers in that regard. This, however, will not likely be a feature available from the outset.

Now, what might speed things along is capable developers willing to lend a hand in the implementation ;)

Sean

----------------------------------------------------------------------
Sean M. Couch, Ph.D.
Assistant Professor
Department of Physics and Astronomy
Department of Computational Mathematics, Science, and Engineering
Facility for Rare Isotope Beams
Michigan State University
567 Wilson Rd, 3260 BPS
East Lansing, MI 48824
(517) 884-5035 --- couch at pa.msu.edu --- http://www.pa.msu.edu/~couch
On Sep 30, 2019, 7:55 PM -0400, Mateusz Ruszkowski <mateuszr at umich.edu>, wrote:




Hi all,


Does anyone know if there a plan to extend FLASH to include adaptive timestep or if users would be interested in such a code feature?
Myself and a number of my collaborators would definitely be interested in using such a code capability.


Cheers,
Mateusz


--
Carlo Graziani (630) 252-1543 (Voice)
Argonne National Laboratory
"The less a statesman amounts to, the more he loves the flag."
-- Kin Hubbard


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://flash.rochester.edu/pipermail/flash-users/attachments/20191001/5e88f46c/attachment-0001.htm>