Thank you for your response!
> Depends on what you mean by "okay". If you integrate against a fine
> grid solution that's more refined on an element (as we generally do)
> then the main issue is quadrature error. Likewise if the grids aren't
> nested. If you integrate against a solution where the "fine grid"
> uses the same or a coarser element then there won't be any quadrature
> error... normally in that case there also won't be a decent error
> estimate, but it looks like you're doing something
> different/interesting with your first equation.
I see... There is a flag for computing error on parent... What exactly
does that do?
Something that could work for this particular situation would be to
compute this difference for the elements of the parent mesh only. i.e.
for each parent element loop over all the active child elements, sum
those int_{child_elem}u^2 values to get the int_{parent_elem}u^2
value, and then compute the difference
sqrt(int_{parent_elem}u^2) - sqrt(int_{parent_elem}(u')^2).
It doesn't seem like that is what the flag is telling it to do, or is it?
> The trouble is that afterward the sqrt call would still be applied to
> a single integrated result... I guess you could call the
> ExactErrorEstimator twice: once passing in an "exact solution" of 0 to
> get the u^2 integral and a second time passing in an approximate
> solution of 0 and an exact solution of the fine grid u' to get the
> u'^2 integral. But I'm afraid it would probably be more efficient to
> subclass ErrorEstimator, and just cut-and-paste code as applicable
> from ExactErrorEstimator.
This did work, the square root wasn't a problem, just comment it out
of the estimate_error routine and take care of it in the
find_squared_elem_error routine.
Thanks!
Andrea
------------------------------------------------------------------------------
This SF.net email is sponsored by
Make an app they can't live without
Enter the BlackBerry Developer Challenge
http://p.sf.net/sfu/RIM-dev2dev
_______________________________________________
Libmesh-users mailing list
Libmesh-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/libmesh-users
