[aspect-devel] Nonlinear solver scheme

John Naliboff jbnaliboff at ucdavis.edu
Wed Nov 23 10:34:41 PST 2016

Hi Payman,

Typically the non-linear solver scheme is chosen based on the rheology 
and general processes you are trying to model, independent of the 
boundary conditions.  When using the 'visco plastic' material model I 
typically use the 'iterated IMPES' or 'iterated Stokes' non-linear solvers.

Trying the same model with each solver can be considered a good addition 
to a broader suite of sensitivity tests.  Likewise, testing how your 
results vary with the temperature polynomial degree is a good 
sensitivity test.  However, by default I typically use high-order 
elements for temperature/velocity/pressure/composition.

Hope this helps and when in doubt do lots of tests (preferably in 2D) to 
see how your model results vary with different parameters!


John Naliboff
Assistant Project Scientist, CIG
Earth & Planetary Sciences Dept., UC Davis

On 11/23/2016 10:17 AM, Payman Janbakhsh wrote:
> Hi
> For visco-plastic material model with fixed top surface , which non 
> linear solver scheme is recommended? Particularly in 3D where DOFs are 
> very large.
> To be more specific on where my doubts are I should say, to my 
> understanding velocity, temperature and viscosity are very much 
> influencing each other in this model. So I was thinking choosing a 
> scheme that solves temperature and velocity equations simultaneously. 
> Please correct me if I’m off the chart.
> As a second question, I know increasing temperature polynomial degree 
> increases accuracy. So is it recommended to have it at 3 or 2 is 
> sufficient for a viscoplastic 3D model.
> Thanks
> payman
> _______________________________________________
> Aspect-devel mailing list
> Aspect-devel at geodynamics.org
> http://lists.geodynamics.org/cgi-bin/mailman/listinfo/aspect-devel

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.geodynamics.org/pipermail/aspect-devel/attachments/20161123/8bd63375/attachment.html>

More information about the Aspect-devel mailing list