Re: [Libmesh-users] need to implement error estimator?

[Roy S. <roy...@ic...>]

On Tue, 18 Jul 2006, li pan wrote:

> I'm not sure if I should ask this question here.

This is the right list.

> Do I need to implement my own error estimator for my
> special problem of a complexer geometric domain?

> I'm applying Total Largrange formulation to solve a
> nonlinear problem.

The Kelly error indicator isn't dependent on the domain.

If you have a more complex PDE, however, you may need a better
indicator.  The Kelly indicator is only really rigorous for linear
elements solving the Lagrange equation.

In particular, if you're using the error indicator as a stopping
criterion rather than just to guide refinement, you definitely need a
more rigorous error estimator.

> My incremental solution can run more than much more
> steps without mesh refinement based on
> kelly_error_estimator than with it.

This sentence hints at the existence of important additional
information without actually providing any.  If you'd like help
you need to fix that.  What happens to prevent your solution from
running for as many steps as you want?
---
Roy Stogner