[CIG-SHORT] Meshing Schemes, Mixed Meshes, Mesh Output, & Initial Conditions

[Brad Aagaard]

Following up on Ravi's questions and my response yesterday-

I posted the PyLith poster from the 2007 AGU Fall Meeting in my folder on the 
CIG website (http://www.geodynamics.org/cig/Members/BradAagaard). The poster 
includes plots of the local error in the elastic solution of the strike-slip 
benchmark for both hex8 and tet4 cells at 1000 m, 500 m, and 250 m 
resolutions. We also show the runtime for the serial runs (1000 m and 500 m 
resolutions). We expect similar trends for comparison of linear (p1) quad4 
and tri3 cells.

Note that the largest local error occurs at the edge of the taper region where 
the slip transitions from a uniform value to a constant gradient. This causes 
a sharp gradient in the strain field along this edge, so finer cells are 
needed in this region to accurately capture the solution. For the rest of the 
area where slip is uniform or has a uniform gradient (taper region), the 
linear basis functions do quite a good job of capturing the displacement 
field (the local error is quite small).

These results indicate that for this problem the linear (p1) hex8 cells far 
outperform linear (p1) tet4 cells (factoring in both run time and the global 
error). We suspect that quadratic (p2) tet4 cells will be competitive with p1 
hex8 cells in terms of performance (comparable global error for the about the 
same run time). We have yet to implement automatic creation of p2 cells from 
a p1 mesh, so we have not tested this hypothesis.

Brad