Hello,Roy, I change the RKDG code to ExplicitSystem and I write a small Gassian
elimination solver, it works well and I can get high order results. I have tried the command line "--implicit_neighbor_dofs" and it no improvment. I ask Ben yesterday, can I write a GMRES+ILU(0) solver running in each processor and only exchange the solution between partitioned mesh, I think it is the same as the PETSC solver because its preconditioner also make ILU(0) in each processor, if that I can set the sparse matrix pattern by myself and I think the the programme can be runned faster. Can libmesh run without PETSC but only with mpi? I'd like to send a DG example after several days, libmesh is a so excellent library and thank you for your kindly reply. Best regards! Luyi 2008-5-14 Roy Stogner 写道: > > On Tue, 13 May 2008, luyi wrote: > >> The CONSTANT basis for both implicit and RK is OK but higher order need >> much more time because >> >> I use elem-based basis, maybe the problem is this? > > You're probably just not getting the correct sparsity pattern > preallocation because libMesh isn't expecting you to be doing a DG > method, and so isn't expecting direct coupling between the elem-based > DoFs in neighboring elements. I'm surprised CONSTANT worked okay if > that's the problem, though, unless we have some half-working > autodetection code in there that I don't know about. > > But my important quesiton: > > Did you try Ben's command line option suggestion yet? There ought to > be a better way to autodetect DG-like DoF coupling, but unless we do a > "fake" system assembly on a two-element mesh, I don't see how to make > it seamless. For now, the problem is probably just "you need to > explicitly tell libMesh you're doing DG". > >> Thank you very much and Best regards! > > If you're feeling especially grateful, you might think about > contributing a simple DG libMesh example once your own code is > straightened out. All the main developers are pretty swamped right > now, but that's something we've needed for a while. > --- > Roy > > ------------------------------------------------------------------------- This SF.net email is sponsored by: Microsoft Defy all challenges. Microsoft(R) Visual Studio 2008. http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/ _______________________________________________ Libmesh-users mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/libmesh-users
