[CIG-SHORT] How does Pylith number vertices on fault

[Brad Aagaard]

On 10/04/2012 09:03 AM, Hongfeng Yang wrote:
> Thanks, Brad. Then that goes to my previous question: how should I
> locate the vertex on the fault using the v_fault label?

You can't unless you modify the source code.

> When I track the slip and stress on different points on the fault, I
> found the normal slip was indeed very large and the normal stress was
> decreased drastically to 0, as shown in the attached figure,
> tettrack.png. I've tried different stress drop and friction parameters
> which all lead to very large normal slip and small normal traction using
> a tetrahedron mesh.
>
> However, when I changed the mesh scheme from tetrahedron to hexahedra
> for the same fault geometry, I can see a rupture initiated and
> propagated on the fault. Tracking stress and slip history at one point
> on the fault would show reasonable stress change and slip.
>
> Varying the tolerance would make pylith run a longer simulation, but the
> results are not going to change based on my previous experience.

What time stepping scheme and solver settings are you using?

> This intrigues me to think of the mesh quality. But after a number of
> unsuccessful experiment in CUBIT of playing around different grid size
> and smoothing schemes, I want to verify which element on the fault is
> indeed cracking every time. If I can manually fixed it, I would expect
> to see a correct simulation or a different point that stops the run.

You should be using the condition number mesh quality metric in CUBIT 
and ParaView to examine the mesh quality. If the condition numbers in 
your hex and tet meshes are similar, then you should be able to get 
similar results. I aim for condition numbers around 1.8. In the SCEC 
dynamic spontaneous rupture benchmarks we find the tet meshes give 
slightly more accurate results with PyLith than the hex meshes.

Regards,
Brad