On Wed, Apr 16, 2008 at 9:27 AM, Roy Stogner <[EMAIL PROTECTED]> wrote:
>
>  On Wed, 16 Apr 2008, John Peterson wrote:
>
>
> > Just out of curiosity, what would be the "right thing" to do in order
> > to compare two meshes with different (still nested) refinement
> > patterns or two solutions on completely non-nested grids?
> >
>
>  Loop over one grid.  On each element, if the other grid is coarser,
>  you integrate here.  If the other grid is finer, you get the
>  corresponding ancestor element there and integrate on all its active
>  descendants.  That way you get the numerical integration right up to
>  floating point error, without worrying about quadrature error.

That should work great for the nested case.  For the arbitrarily
mismatched grids I think you must have to do something like Derek
suggested and introduce a "transfer" Mesh which is somehow the "union"
of both meshes you are comparing.

In the nested case, differing p-levels would also be interesting.
Should I then perform my integration over the mesh which has a single
high-p element or over the one that chose to split it into multiple
low-degree elements having small h?  The answer probably depends on
the smoothness of the true solution.

>
>
> > It's not immediately obvious to me what such a comparison would
> > actually tell you in any case...
> >
>
>  Oh, well I just meant that *what* to do to integrate between two such
>  grids was obvious; *why* to do such a comparison isn't obvious to me
>  either.  If you're contrasting two different refinement strategies
>  then you generally want to compare them to a completely finer
>  reference solution, not to each other.
>
>  I bet you'd usually still get a good error indicator for grid A in the
>  places where grid B was more refined, and vice versa; running into
>  pollution effects would be a nightmare, though.

That's where my line of thought was heading.  In problems with a
singularity the simple "is h finer here or there?" test can fail due
to non-local effects.

-J

-------------------------------------------------------------------------
This SF.net email is sponsored by the 2008 JavaOne(SM) Conference 
Don't miss this year's exciting event. There's still time to save $100. 
Use priority code J8TL2D2. 
http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone
_______________________________________________
Libmesh-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/libmesh-users

Reply via email to