[aspect-devel] Weird pressure with periodic BC's + direct solver?

[Jonathan Perry-Houts]

Thanks for the suggestion Wolfgang! It does indeed only show up at 
hanging nodes (figure attached). And it apparently only happens with 
both the direct solver, and 'remove null space' enabled at the same time 
(also attached).

When those two options are both enabled the pressure solution is zero 
everywhere, but gets these little (~ 1e-6) jitters at hanging nodes. 
Definitely sounds like a ConstraintsMatrix problem.


On 09/05/2017 06:10 PM, Wolfgang Bangerth wrote:
> 
> Jonathan,
> 
>> I just noticed that a bunch of models I've run recently have strange
>> pressure fields (see attached). It only happens with a periodic boundary
>> condition, and the direct solver enabled at the same time. Any other
>> combination of one or the other setting doesn't do this.
>>
>> The velocity fields look correct (as compared to models with direct
>> solver turned off), so *hopefully* this is somehow just a
>> post-processing problem. Anyone have insight here?
> 
> It looks like the problems are all located along specific lines. Is this 
> where processor boundaries lie? Or hanging nodes? Can you produce the 
> same picture with the mesh overlaid?
> 
> I'm wondering whether we properly call ConstraintMatrix::distribute() on 
> the solution vector after calling the direct solver. This would explain 
> artifacts at hanging nodes.
> 
> Best
>   W.
> 

-- 
Jonathan Perry-Houts
Ph.D. Candidate
Department of Earth Sciences
1272 University of Oregon
Eugene, OR 97403-1272
-------------- next part --------------
A non-text attachment was scrubbed...
Name: direct solver pressure problems - closeup.png
Type: image/png
Size: 37577 bytes
Desc: not available
URL: <http://lists.geodynamics.org/pipermail/aspect-devel/attachments/20170906/65959bb0/attachment-0001.png>