[CIG-SHORT] general mesh optimization

[Brad Aagaard]

Rebecca Bendick wrote:
> Thanks for the advice and the examples.  Just so I have a better sense  
> of strategy, what do you do when a run does fail for meshing reasons?   
> So, you advised me to start with a vertical fault, then change it to a  
> dipping one.  If that single change makes pylith exit before  
> calculating a solution, how would you refine the problem?  Using a  
> finer mesh?  A tiny change to the geometry?  What kinds of changes  
> work best?  For most dislocation problems that I can imagine, the  
> fault geometry is pretty constrained by a priori slip inversions, or  
> at least the moment tensor, so there isn't a whole lot of latitude to  
> modify things to improve mesh characteristics.

There are two general reasons why PyLith encounters errors related to 
the mesh.

(1) The mesh may not conform to what PyLith expects. In this case it 
either detects an error and attempts to give a meaningful error message 
or it crashes. Usually the error message and the place where the error 
occurs provide a good indication of what is wrong. Remedying the error 
may involve fixing errors in the .cfg file or fixing an inconsistency in 
the mesh. Some examples include fixing errors in the specification of 
the groups of nodes or fixing errors in how the mesh is generated (e.g., 
inadvertently creating a mesh with cells that share the same volume).

(2) The mesh quality is very poor. This usually involves additional 
processing to take a poor quality mesh and improve the aspect ratio of 
the cells. LaGriT and CUBIT contain various tools for doing this. Nodes 
within a volume are moved around, merged, etc to make nice, regular 
cells. In most cases, the geometry of the problem does not need to be 
altered; however in some cases a slight tweak in the geometry on a small 
length scale can eliminate distorted cells and is worthwhile and reasonable.

Brad