[aspect-devel] Internal heating in aspect (Ludovic Jeanniot)

[Rene Gassmoeller]

Dear all,

following our discussion last week I sat together with Max Rudolph here 
at UC Davis and discussed a few possibilities for modifying the current 
stabilization scheme in ASPECT to overcome the difficulties that were 
discussed, i.e. (apparent) imbalance of steady-state heat flux across 
the boundaries due to strong artificial diffusion through the boundary 
layers and thus a modified average temperature (all for models with a 
low to medium resolution of 3-5 global refinements in the boundary 
layer). I would think there are two possible solutions and went ahead 
with a few tests.

1. Reimplementing SUPG into ASPECT (as originally proposed in 
https://github.com/geodynamics/aspect/pull/412 and discarded). I have a 
rebased version of the original pull request open, but it is not tested 
rigorously. Everyone is welcome to contribute.

2. Modifying the entropy viscosity method, e.g. by applying the 
artificial diffusion in streamline direction only.

The second should not require significant changes to the code and so I 
implemented a experimental version (see 
https://github.com/geodynamics/aspect/pull/2649). In preliminary tests 
by myself and Max it seems to perform pretty well, essentially behaving 
like the existing EV method, except for drastically reducing the 
diffusion perpendicular to the flow (not surprising), and therefore 
having much less influence on boundary layer heat flow. I have written a 
summary in the discussion of the pull request, but in essence I feel 
like this might be a step forward and would appreciate if those of you 
who had problems with particular benchmarks could give this version a 
try. All tests I have run so far indicate more accurate global 
properties (average temperature, vrms), but I also see slightly bigger 
temperature oscillations, and would be interested if the method is 
stable in all cases. So if you have a bit of time, please let me know 
about the results of your tests. If the method performs well enough, I 
could see this becoming the new default for the stabilization scheme, 
but for that we need a lot of testing.

You can get my branch by running the following in your aspect git repo:

git checkout -b gassmoeller-suev master
git pull https://github.com/gassmoeller/aspect.git suev


Let me know how that works,

Best,

Rene


-- 

Rene Gassmoeller
https://gassmoeller.github.io/

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.geodynamics.org/pipermail/aspect-devel/attachments/20180904/12ab7464/attachment.html>