[aspect-devel] Benchmark

[Timo Heister]

Iris,

not sure this explains the result but:
> *p=x*y*z+x*x*x*y*y*y*z-5/32;
Here 5/32 gives 0, because it is an integer division. Write it as "5./32.".


On Mon, Jun 30, 2014 at 7:06 AM, I. van Zelst <i.vanzelst at students.uu.nl> wrote:
> Dear all,
>
> I have been trying to make a new benchmark for Aspect, based on the paper
> 'Large-scale adaptive mantle convection simulation' from Carsten Burstedde
> in 2013. Subsequently, the benchmark has been dubbed the 'Burstedde'
> benchmark.
> However, a problem I encounter is that the pressure error between the
> analytical solution and the numerical solution does not converge with
> resolution, as can be seen in the attached figure 'errors.pdf’.
> Furthermore, the minimum and maximum pressure in the system, for which an
> analytical value is available diverges more and more from the analytical
> solution as the resolution is increased, as can be seen in the attached
> figure 'pressure.pdf'.
> Could it be that the elements in the corners of the domain are constrained
> by too many (?) boundary conditions: these elements have three faces for
> which velocities are prescribed in the x, y and z directions.
> In this case, could it be that these corner elements lead to a matrix which
> the solver cannot handle well, resulting in excessive pressures at the
> corners and non-converging velocities ?
>
> Would anybody like to have a look at the benchmark to see whether this is
> indeed what goes wrong? I have included the .cc file and the input file,
> which could be put in the folder burstedde, which would be a folder of
> benchmark.
>
> Kind regards,
> Iris van Zelst
>
> _______________________________________________
> Aspect-devel mailing list
> Aspect-devel at geodynamics.org
> http://lists.geodynamics.org/cgi-bin/mailman/listinfo/aspect-devel