On Aug 26, 2013, at 5:35 PM, John Assael <[email protected]> wrote:

> For example my smallest dimension in the mesh is a circle with diameter 6μ 
> (E-06). (where in order to get good results I have to define cellsizes of 
> 1E-07)

I am completely confused. Previously you said that your 1 cm x 2 cm mesh was 
"very small". Now you're saying that you want to resolve that mesh with 100 nm 
cells, so your mesh sounds very big to me.

If you are defining your entire .geo file with a resolution of 1e-7 m, FiPy 
will need about 4 TB of memory based on recent benchmarking. We need 
desperately to improve that, but I don't expect ten billion cells to ever be 
practical. 

If you are defining your .geo file with 1e-7 at some locations and 1e-3 
elsewhere, then I would say I have not seen Gmsh do a very good job with such 
big jumps in scale between neighboring geometric points.

As I said before, what I recommend is that you use the "background=" argument 
to iteratively refine your mesh based on an appropriate error metric for your 
problem. 

All said, though, I find that Gmsh is not very good with huge changes in scale. 
I have some experimental code that uses libMesh to successively bisect or merge 
cells with high or low error. The performance seems much better than Gmsh, but 
my FiPy-libMesh interface is not ready for release, yet.

> Now about the Gmsh2D I use a .msh generated file and import it do you think 
> it would be better to do it this way?

I don't understand what you mean.

_______________________________________________
fipy mailing list
[email protected]
http://www.ctcms.nist.gov/fipy
  [ NIST internal ONLY: https://email.nist.gov/mailman/listinfo/fipy ]

Reply via email to