Most refinement algorithms do not set parent entities data

[Jan Blechta]

Check this and this question.

BisectionRefinement (serial) is only algorithm which sets parent cells and this code passes:

mesh = UnitSquareMesh(2, 2)

cf = CellFunction('bool', mesh, True)
mesh1 = refine(mesh, cf)

f = CellFunction("size_t", mesh)
f_refined = adapt(f, mesh1)

But UniformMeshRefinement

mesh = UnitSquareMesh(2, 2)

mesh1 = refine(mesh)

f = CellFunction("size_t", mesh)
f_refined = adapt(f, mesh1)

and also ParallelRefinementND and RegularCutRefinement raises

*** Error:   Unable to adapt mesh function.
*** Reason:  Unable to extract information about parent mesh entites.
*** Where:   This error was encountered inside adapt.cpp.

because they do not set parent_cells.

[Chris Richardson]

ParallelRefinementND cannot (easily) support parent entities if load balancing (repartitioning) is also used, as parent entities could be on different processes. It should be fine if there is no repartitioning.

[Chris Richardson]

I have some refinement code in progress that will address this for 1.5

[Chris Richardson]

New PlazaRefinementND method sets parent_cell, but the whole parent-child relation will probably be migrated to a different object, e.g. MeshHierarchy

[Chris Richardson]

Plaza refinement now sets parent_cell and parent_facet.

[Former user]

Hi, I ran into this issue in a freshly installed dolfin 1.6.0dev and found this bugreport.

As it should be solved, I doubt I am doing something wrong, so here's a brief summary of what I am doing.

I have two "related" problems, P1 and P2. They are related in the sense that I solve P1, then use its solution as a coefficient in P2 and solve it adaptively. Once this is done, I want to refine P1 to be run on the new mesh. To do so I set up a LinearVariationalProblem for P1 and try to call adapt(LinearVariationalProblem,Mesh) passing P1 and the mesh [if I pass mesh->leaf_node() it segfaults].

P1 and P2 are defined on the same mesh initially.

I set parameters["refinement_algorithm"] = "plaza_with_parent_faces"

I am running in serial al the moment.

Thanks for any help.

[Jan Blechta]

@massimiliano_leoni, could you provide a minimal working example to reproduce the problem?

[Former user]

I can try, but I have to move code through a few files so it might take a while :)

[Former user]

Ok, here should be everything you need to reproduce. I am sorry it is not small, but I think it's rather minimal.

While shrinking, I realised that if you delete the parts relative to the neumann BCs, it segfaults somewhere else. These parts are: Fracture2Dv.ufl:15 Fracture2Dv.ufl:25 Fracture2Du.ufl:15 Fracture2Du.ufl:25 fiber2D.cpp:143-155, 172,203-213,278-281

I am not including a CMakeLists as I made a mess of mine to produce this.

[Jan Blechta]

Can you strip it down more? This is quite long.

[Jan Blechta]

I get

$ ffc -ldolfin Fracture2Dv.ufl
*** FFC: Unable to extract all indices.

with FEniCS dev. Which FEniCS version do you use?

[Former user]

There's error control in that ufl, use flag -e.

I'll see if I can strip down further.

[Former user]

Also, I noticed there's a definition missing in the one I uploaded previously, due to crazy copy-pasting among files. Use this instead

[Former user]

I managed to make it shorter and added comments to make it more readable.

[Jan Blechta]

The stacktrace

#0  0x00007ffff75aea10 in dolfin::dolfin_error(std::string, std::string, std::string, ...)@plt ()
   from /usr/users/blechta/opt/fenics-dev/r0d0_fenics-dep/master/lib/libdolfin.so.1.6
#1  0x00007ffff76bc2db in dolfin::adapt (mesh_function=..., adapted_mesh=...)
    at /usr/users/blechta/fenics/fenics-dev-dbg/src/dolfin/dolfin/adaptivity/adapt.cpp:563
#2  0x00007ffff76bd211 in dolfin::adapt (form=..., adapted_mesh=..., adapt_coefficients=adapt_coefficients@entry=true)
    at /usr/users/blechta/fenics/fenics-dev-dbg/src/dolfin/dolfin/adaptivity/adapt.cpp:271
#3  0x00007ffff76bfd37 in dolfin::adapt (problem=..., adapted_mesh=...)
    at /usr/users/blechta/fenics/fenics-dev-dbg/src/dolfin/dolfin/adaptivity/adapt.cpp:308
#4  0x00000000004386cb in main (argc=<optimized out>, argv=<optimized out>) at /usr/users/blechta/fenics/issue105/fiber2d/fiber2D.cpp:320

[Jan Blechta]

@massimiliano_leoni, could you shorten the example? It is too long.

[Chris Richardson]

The mesh you are passing has no parent, so cannot extract parent entities. Try putting:

std::cout << "Has parent = " << mesh->has_parent() << std::endl;

before your call to adapt.

[Former user]

Hey, sorry for the late reply, I had guests this week.

@chris_richardson as you expected, I get "Has parent = 0", so this could be the problem. How do I take care of this? I tried uncommenting *mesh = mesh->leaf_node() but nothing changed.

I thought these issues were taken care of automatically, can you explain me what I am supposed to do in general to get adaptation to work?

@blechta I removed all the time stuff and tuned the parameters so that it breaks at the first solve, now it's even shorter. I attach only the file I modified

[Former user]

Hi again, I didn't get if Chris' comment contains the solution to my problem. How do I assign a parent to the mesh?

[Former user]

Ok, I figured out that I had to use child_shared_ptr() instead of leaf_node. I think I found another bug now, I'll file another bug report :)

[Martin Sandve Alnæs]

Removing milestone: 1.6 (automated comment)