On 12/7/2016 6:14 PM, Roy Stogner wrote: > > On Tue, 29 Nov 2016, Michael Povolotskyi wrote: > >> I solve a system and I have to define some periodic boundary conditions >> and some Dirichlet boundary conditions. >> >> My application is an old one, it was developed before >> DirichletBoundaries and PeriodicBeoundaries existed in libmesh. So at >> that time (around 2005) I developed the following solution. I added >> constraints to the dof_map according to my boundary conditions. This was >> not a very easy piece of code, since I had to add constraints in terms >> of unconstrained DOFs only, but it was working. >> >> When I upgraded to libmesh 1.0.0, I found that in order to get correct >> results, I had to call constrain_element_matrix and >> constrain_element_vector with asymmetric_constraint_rows equal to false. >> >> Does it make sense to you? > > I've been running my head around this for a while but I can't think of > any obvious explanation. So let me try a bit harder to understand the > problem. > > Were all your Dirichlet boundaries homogeneous? If not, how did you > handle the offsets? > > What solver and preconditioner are you using? I assume it's something > that can handle asymmetric operators, like the PETSc default > GMRES+ILU? > > Is this a linear or nonlinear problem? If nonlinear, are you solving > for the delta between two Newton iterates or solving directly for each > Newton iterate? > > What are your linear solver tolerances? > > What's the smallest problem you can get this to replicate with? If > you can exhibit it with a small code or with only a handful of > elements then we might be able to dig into it ourselves. > --- > Roy
Thank you, Roy, I was trying to reduce the problem, and I found that it happens only for nonlinear system solved with trust region method of PETSc. Since with the new libmesh I also use new version of PETSc, there can be many reasons. Michael. ------------------------------------------------------------------------------ Developer Access Program for Intel Xeon Phi Processors Access to Intel Xeon Phi processor-based developer platforms. With one year of Intel Parallel Studio XE. Training and support from Colfax. Order your platform today.http://sdm.link/xeonphi _______________________________________________ Libmesh-users mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/libmesh-users
