Hi Roy, I followed the procedure as discussed earlier to initialize ParallelMesh. Elements are partitioned and all relevant nodes are added to each processor with the ids of the elements and nodes set before addition.
I am tripping the assert on line 771 of parallel_mesh.C: libmesh_assert ( !obj || procid == min_procid ); It is likely that I have a bug in my initialization, but I wanted to see if you had any quick thoughts on this error. Thanks, Manav On Thu, Jun 6, 2013 at 3:18 PM, Roy Stogner <[email protected]>wrote: > > On Thu, 6 Jun 2013, Manav Bhatia wrote: > > I am curious if there is an advisable strategy to add nodes and >> elements to the ParallelMesh? >> > > Usually what we do in mesh refinement is give each processor > "ownership" of dof ids which are equal to the processor id modulo > n_processors+1. That way, no overlap. > > > If an initial approximate mesh partition is available from a >> space-filling algorithm, then is it OK to add the partitioned nodes to the >> ParallelMesh on its respective processor? Following this, how does one add >> the elements? Is it safe to add all elements that a node on a processor >> connects to? If so, what does one do about nodes that the element might be >> connected to, but are not stored on the processors? >> >> On the other hand, is it better to have an initial partition of >> elements (as opposed to nodes), and then one can add elements and only >> relevant nodes to the ParallelMesh on a specific processor? >> > > Since libMesh defines node processor ids in terms of the connected > element processor ids, it's better to partition the elements and then > derive the node partitioning from that. If you do that and you make > sure the interfacial nodes got added consistently on both processors, > then MeshCommunication::gather_**neighboring_elements() should get you > the ghost objects. > > > In either case, if I am assigining consistent IDs to the nodes and >> elements before adding to ParallelMesh, would this be enough to ensure >> consistency of node and element data? >> > > I believe so, yes. You'll want to do a renumbering afterwards for > efficiency, though. > --- > Roy > ------------------------------------------------------------------------------ This SF.net email is sponsored by Windows: Build for Windows Store. http://p.sf.net/sfu/windows-dev2dev _______________________________________________ Libmesh-users mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/libmesh-users
