[aspect-devel] aspect release planning

[Wolfgang Bangerth]

On 01/14/2015 05:13 PM, Wolfgang Bangerth wrote:
> On 01/14/2015 05:11 PM, Timo Heister wrote:
>> @Wolfgang, did you have a chance to look into the two crashes that are
>> happening there?
>
> I'm on it. I can only reproduce one of them, the other one seems to be a
> misreferenced input file, but I haven't understood yet where the first
> fail comes from. Will report back.

I think there is nothing inherently mysterious about either
   steinberger-viscosity
   steinberger-viscosity-adiabatic
Their respective first time steps previously already took 30+353 and 
30+385 iterations, and on the new mesh the first one now needs 30+1761 
iterations -- and the second hits the ceiling of 1891 and explodes. 
(Subsequent time steps are significant cheaper.)

The thing is that I believe this is entirely coincidental. The viscosity 
of both is very inhomogeneous (a picture for one of the cases is 
attached): the viscosity changes by 4 orders of magnitude over 1.5 
cells. You have to expect that this is slow to converge, we just don't 
have a particularly good solver for these cases (we have open bug reports).

I don't know what to do with this particular test, but I don't think it 
presents an underlying problem with the patch -- just an underlying 
problem with the solver.

You currently hold the master branch for this patch. It doesn't apply 
cleanly to master -- can you rebase it and see which tests still fail?

Cheers
  W.

-- 
------------------------------------------------------------------------
Wolfgang Bangerth               email:            bangerth at math.tamu.edu
                                 www: http://www.math.tamu.edu/~bangerth/

-------------- next part --------------
A non-text attachment was scrubbed...
Name: visit0000.png
Type: image/png
Size: 187954 bytes
Desc: not available
URL: <http://lists.geodynamics.org/pipermail/aspect-devel/attachments/20150114/d474d607/attachment-0001.png>