> > You know, we could and probably *should* use a similar trick to your > DG code elsewhere in the library. Instead of seeing whether element > vs neighbor sides are conforming via "if (refinement_type == "p")", > could we test "if (elem->level() == neighbor->level())"? Then we only > have to fall back on inverse_map when the particular elem/neighbor > interface is non-conforming, not when any interface anywhere in the > mesh might be. I haven't tested that idea on ex21 yet; would you?
I haven't yet, but I think it's the best option. > >> In FEBase I added a method that returns the nodes coordinates for reference >> frame elements. >> This method could also be used to speed-up the solution output when using >> reference frame finite elements. >> I did it based on FEBase::on_reference_element() >> would you please check that everything is correct. This kind of coding is >> error-prone and I'm not sure about the PIRAMID5. > > John, double-check me on this: > > I believe the PYRAMID5 in Lorenzo's patch is wrong - the Pyramid5 we > use, going by both my own hazy memory and by the ::shape() equations > in fe_lagrange_shape_3D.C, has points 0-3 in the zeta=0 plane, not > zeta=-1. Since the PYRAMID5 case is not implemented in FEBase::on_reference_element(), (looking at the code it seems that Ben has been lazy :->) I didn't know where to find info about the coordinates. Thanks to you and John for checking and double-checking. > > Only trouble is, if we make the library depend on this code > internally, I don't want AMR to break for the less common > configurations. So that would mean doing better than a > libmesh_not_implemented() for InfFE: we'd need to add the InfFE > entries to FEBase::get_refspace_nodes() and either move side_map up > the inheritance tree or duplicate it. > > I don't have any time to do that myself right now, though. And > otherwise the patch looks good. So (assuming it passes my app tests, > and unless anyone has objections) I'm planning to commit things nearly > as-is, with InfFE as libmesh_not_implemented() and with the PYRAMID5 > case fixed. > > And if anyone wants faster jump error estimators any time soon and has > more time to play with it, that should soon be low-hanging-fruit. I hope I'll have some time to improve the error estimators code in the near future and also avoid inverse maps in the solution output using FEBase::get_refspace_nodes(). The InfFE stuff is a bit harder for me since I have never used it and at least I would need to understand how it works... Thanks a lot. Lorenzo ------------------------------------------------------------------------------ BlackBerry® DevCon Americas, Oct. 18-20, San Francisco, CA The must-attend event for mobile developers. Connect with experts. Get tools for creating Super Apps. See the latest technologies. Sessions, hands-on labs, demos & much more. Register early & save! http://p.sf.net/sfu/rim-blackberry-1 _______________________________________________ Libmesh-users mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/libmesh-users
