[aspect-devel] Van Keken et al. 2008 subduction benchmark

[Jonathan Perry-Houts]

Hi Max,

I don't know what's going wrong, but one thing that comes to mind is
that constraining internal velocities is a bit weird in a finite element
sense because we're prescribing solutions on particular nodes (i.e.
setting rows in the mass matrix to the identity, and setting the RHS in
those rows to some arbitrary value that doesn't represent rho*g
anymore), but the rest of the assembly is still integrating over entire
cells and expecting the usual RHS values.

The reason the internal velocities thing is only described in the
cookbook and not included in the code is because the number of ways that
could go wrong (understandably) made Wolfgang nervous. It could be that
the discontinuous pressure elements in the "use locally conservative
discretization" option just make it easier to run into cases where one
of the assumptions here is violated.

Another way you might try the Van Keken benchmark would be making a
geometry model (mesh) with one 45 degree angle, and just prescribing
boundary conditions rather than internal velocities. I vaguely remember
hearing Wolfgang say he had tried something like this and found that
quadrilateral elements didn't work so well, though.

-JPH

On 10/27/2016 09:13 AM, Max Rudolph wrote:
> I was wondering if anyone has been successful in setting up the Van
> Keken et al. 2008 subduction benchmark using ASPECT? I have been using
> this as a starting point to set up a more complicated corner flow model
> and have not yet been successful.
> 
> This benchmark solution requires a rigid overriding plate and kinematic
> subducting plate. There is a discontinuity along the subduction
> interface. I tried imposing kinematic boundary conditions on the left
> and bottom of the box and using Jonathan Perry-Houts' prescribed
> velocity plugin (cookbooks/prescribed_velocity), but the Stokes solver
> fails to converge. Looking at the pressure field, it's not surprising to
> see very large oscillations in this region where the velocity field is
> discontinuous. It occurred to me that using the locally conservative
> discretization might help. However, when the locally conservative
> discretization is enabled, the internal prescribed velocities are no
> longer enforced anywhere. Is it obvious why this might be the case?
> 
> Regards,
> Max
> 
> 
> _______________________________________________
> Aspect-devel mailing list
> Aspect-devel at geodynamics.org
> http://lists.geodynamics.org/cgi-bin/mailman/listinfo/aspect-devel
>