Ahh, so basically the right and left hand sides in general were being scaled by the same wrong value?
I wonder if this was more likely to produce an error on flux boundaries, or do you expect still you'd be scaling by the same erroneous value on either side? Anyway, GOLD STAR Boyce! -Ben On Nov 8, 2011, at 8:16 PM, "Roy Stogner" <[email protected]> wrote: > > On Tue, 8 Nov 2011, Kirk, Benjamin (JSC-EG311) wrote: > >> None here. I was pretty much sequestered today - what's the gist? > > The short version is that it's a miracle I manage to dress myself and > feed myself despite gross mental handicaps. > > The long version is that my "has_affine_map" optimization, which > obtains the correct init_shape_functions result much faster on > triangles and tets, obtains incorrect dphidxi etc. maps on rectangluar > quads, rectangular prism hexes, and rectangular prisms. As far as I > can tell, the only reason we managed to avoid noticing the problem for > so long is that for most codes the errors can act as a self-consistent > weighting function in the weak formulation integrals, so e.g. > manufactured solution tests still converge at the right order and > exact solutions in the FEM space are still recovered. After the > change all our FIN-S regression test results are still coming in under > the 1e-15 tolerance, even! > > I think Boyce only managed to catch things by using different order > quadrature rules to integrate the matrix vs forcing function for an L2 > projection. > --- > Roy ------------------------------------------------------------------------------ RSA(R) Conference 2012 Save $700 by Nov 18 Register now http://p.sf.net/sfu/rsa-sfdev2dev1 _______________________________________________ Libmesh-devel mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/libmesh-devel
