Christophe.Geuzaine at ulg.ac.be
Tue May 22 09:08:23 CEST 2001
Are you talking about the accuracy of the spatial or time
discretizations? (Space and time are (unfortunately?) completely
decoupled in getdp.)
- for the space discretization, you may compute an error estimate in
post-processing, and use this error estimate to generate a refined mesh.
For h-type adaptation (i.e. modification of element sizes), you should
output a post-processing view which contains the target characteristic
lengths. Thus, you should
1) generate a first (coarse) mesh
2) compute target mesh sizes on this mesh, and output the results as a
Gmsh post-processing view
3) load this view in Gmsh, and apply it as the current background mesh
(with 'View->Apply as Background Mesh'), or load the view on the
command line, with the option '-bgm file'
Step 2) is of course the difficult part, and depends on your problem.
GetDP can help to generate the target size view if you have a mean to
compute an error field, i.e. some 'volumic' field on the mesh
representing (an approximation of) the error. If you can compute such a
field, then you can use 'Adapt H1' in a GetDP PostOperation on this
field, and you will obtain a valid background mesh for Gmsh.
For p-type adaptation (i.e. local modification of interpolation order),
you should output an adaptation file, and give it to GetDP on the
command line with the '-adapt file' option. The adaptation file can be
generated by GetDP by using 'Adapt P1' and by specifying 'Adaptation' in
the Format of a PostOperation.
- for the time discretization, there is no adaptative method implemented
yet. If you are interested, have some clear ideas about what you need,
and some time to implement it, we may start a collaboration on this
"Jean-Francois Remacle " wrote:
> > Jean-Francois,
> > We are still playing with getdp. Can you tell us how we control the
> > accuracy of the solution? Also, is adaptive method implemented in getdp?
> > Thanks,
> I forwarded this email to the conceptor of getdp, he'll tell you what is
> available. Bt the way, when you'll have an error estimate, I will be able
> to help you to refine the mesh
> > Ji, Lin
> > --
> > Dept. of Math., RPI
> > 518-276,2184(work)
> > 518-271,4486(home)
Tel: 32 (0) 4 366 37 10 http://geuz.org
Fax: 32 (0) 4 366 29 10 mailto:Christophe.Geuzaine at ulg.ac.be
More information about the getdp