> > OK, so F_i would then be (f(u),\varphi_i), right? > > I think the question is whether you do or do not apply > ConstraintMatrix::condense to F. I never really know whether you do or > do not have to do that when you want to use F that way, so I typically > just assemble the (scalar number) > (f(u),v) > directly. >
Hi Wolfgang, a brief update -- before, I was using ConstraintMatrix::distribute_local_to_global in the assembly of both the derivative vector and the Hessian matrix. I can see how that would make the computed vector no longer represent the derivative of a functional, since you're changing entries more to make the hanging node constraints come out correctly. To see if this was the issue, I instead changed the code to add up contributions to the global vector/matrix directly, and only reconcile the constraints in the nonlinear solver routine. All the other unit tests still pass, so this change didn't break the nonlinear solver for the unrefined case. Still, the errors in the local linear approximation to P aren't decreasing for the refined mesh. I'll try your approach about assembling (f(u), v) directly and report back. If you have other suggestions, I'm all ears; this is a pretty serious roadblock for what I want to do next. -- The deal.II project is located at http://www.dealii.org/ For mailing list/forum options, see https://groups.google.com/d/forum/dealii?hl=en --- You received this message because you are subscribed to the Google Groups "deal.II User Group" group. To unsubscribe from this group and stop receiving emails from it, send an email to dealii+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
