dolfin team mailing list archive

[Johan Hake <johan.hake@xxxxxxxxx>]

On Tuesday September 6 2011 12:18:33 Anders Logg wrote:
> On Tue, Sep 06, 2011 at 11:31:03AM -0700, Johan Hake wrote:
> > On Tuesday September 6 2011 11:04:23 Anders Logg wrote:
> > > On Tue, Sep 06, 2011 at 05:45:33PM +0100, Garth N. Wells wrote:
> > > > On 6 September 2011 17:31, Johan Hake <johan.hake@xxxxxxxxx> wrote:
> > > > > On Monday September 5 2011 00:09:58 Anders Logg wrote:
> > > > >> On Sun, Sep 04, 2011 at 11:23:04PM -0700, Johan Hake wrote:
> > > > >> > On Friday September 2 2011 23:19:22 Anders Logg wrote:
> > > > >> > > On Fri, Sep 02, 2011 at 02:35:57PM -0700, Johan Hake wrote:
> > > > >> > > > What is the different between a MeshMarker and a
> > > > >> > > > MeshFunction? Is MeshMarker a MeshFunction but instead of
> > > > >> > > > storing the values in line with its global entity index it
> > > > >> > > > stores it wrt the global cell entity index together with
> > > > >> > > > its local entity index?
> > > > >> > > 
> > > > >> > > Yes, that and values don't need to be stored on the entire
> > > > >> > > mesh, only for a subset, so you can mark just 3 facets
> > > > >> > > without needing to store markers for a million facets.
> > > > >> > 
> > > > >> > ok, I will see what I can do.
> > > > >> 
> > > > >> Thanks!
> > > > >> 
> > > > >> > > Copy paste from the MeshMarker docstring:
> > > > >> > >   /// The MeshMarkers class can be used to store data
> > > > >> > >   associated
> > > > >> > > 
> > > > >> > > with /// a subset of the entities of a mesh of a given
> > > > >> > > topological /// dimension. It differs from the MeshFunction
> > > > >> > > class in two ways. /// First, data does not need to be
> > > > >> > > associated with all entities /// (only a subset). Second,
> > > > >> > > data is associated with entities /// through the
> > > > >> > > corresponding cell index and local entity number ///
> > > > >> > > (relative to the cell), not by global entity index, which
> > > > >> > > means /// that data may be stored robustly to file.
> > > > >> > > 
> > > > >> > > > Also, will this take over for the way we use MeshFunctions
> > > > >> > > > in the assembler, or will a MeshFunction be generated by a
> > > > >> > > > MeshMarker before assemble gets called?
> > > > >> > > 
> > > > >> > > I think we will do that as a first step (convert from
> > > > >> > > MeshMarker to MeshFunction) since then we don't need to touch
> > > > >> > > the assembler. Then later we can think about using
> > > > >> > > MeshMarkers directly.
> > > > >> > 
> > > > >> > Ok.
> > > > >> > 
> > > > >> > > > I think I also get confused with the naming here. If my
> > > > >> > > > explaination of what MeshMarker is doing is correct, a
> > > > >> > > > MeshMarker and a MeshFunction are essentially doing the same
> > > > >> > > > thing. What differs is the way the data is stored. This is
> > > > >> > > > not reflected in the naming of the classes
> > > > >> > > 
> > > > >> > > It was the best I could come up with. Feel free to suggest
> > > > >> > > something else. SubsetMeshFunction would also be confusing
> > > > >> > > since it's not really a MeshFunction.
> > > > >> > > 
> > > > >> > > Either way, I expect the MeshMarkers class to be used mostly
> > > > >> > > internally by the MeshDomains class.
> > > > >> > 
> > > > >> > Ok.
> > > > >> > 
> > > > >> > Not sure these are better, but they might reflect the difference
> > > > >> > between this guy and a MeshFunction in a slightly more intuitive
> > > > >> > way.
> > > > >> > 
> > > > >> >   MeshEntityFunction, LocalMeshEntityFunction,
> > > > >> >   LocalMeshFunction, SubMeshFunction
> > > > >> 
> > > > >> I'm not sure those are much better, and I don't think it would be
> > > > >> correct to call them something containing "Function" since they
> > > > >> are not really functions. With a MeshFunction, one can input x (a
> > > > >> mesh entity) and get y = f(x) (the value of the MeshFunction at
> > > > >> that entity). That's not possible with MeshMarkers; they are just
> > > > >> a collection of markers, not really a function since the value is
> > > > >> only defined on a subset and one would need to loop through the
> > > > >> list of values to get the value at any entity where the value is
> > > > >> actually defined.
> > > > > 
> > > > > What with MeshValueCollection? As it is a templated class I do not
> > > > > think Marker is an appropriated name.
> > > > 
> > > > Agree.
> > > > 
> > > > > 'Collection' says that the class is not
> > > > > defined over the whole Mesh.
> > > 
> > > I don't see what the templating has to do with the name "markers" but
> > > MeshValueCollection sounds good.
> > > 
> > > > > Two questions:
> > > > > 
> > > > > How can the following code work:
> > > > >      // Get marker data
> > > > >      const std::vector<uint>& marker = _markers[i];
> > > > >      const uint cell_index   = marker[0];
> > > > >      const uint local_entity = marker[1];
> > > > >      const T marker_value    = marker[2];
> > > > > 
> > > > > when _markers is declared as:
> > > > >    // The markers
> > > > >    std::vector<std::pair<std::pair<uint, uint>, T> > _markers;
> > > 
> > > The above code doesn't work. I suspect the code hasn't yet been
> > > instantiated so it wasn't detected by the compiler.
> > > 
> > > The markers need to be accessed as follows (from XMLMeshMarkers.h):
> > >   for (uint i = 0; i < mesh_markers.size(); ++i)
> > >   {
> > >   
> > >     pugi::xml_node entity_node = mf_node.append_child("marker");
> > >     const std::pair<std::pair<uint, uint>, T>& marker =
> > >     
> > >       mesh_markers.get_marker(i);
> > >     
> > >     entity_node.append_attribute("cell_index") = marker.first.first;
> > >     entity_node.append_attribute("local_entity") = marker.first.second;
> > >     entity_node.append_attribute("marker_value") = marker.second;
> > >   
> > >   }
> > >   
> > > > The above also permits multiple entries. Perhaps we want
> > > > 
> > > >     boost::unordered_map<std::pair<std::pair<uint, uint>, T> >
> > > >     _markers;
> > > 
> > > Yes, maybe but I'm not sure what the cost would be for the lookup on
> > > each cell during assembly.
> > > 
> > > > > What is the logic behind:
> > > > >    // Set all value of mesh function to maximum value (not all will
> > > > >    // be set) by markers below
> > > > >    mesh_function.set_all(maxval);
> > > > > 
> > > > > Isn't it more natural to initiate the values to zero? Also it makes
> > > > > no sense in conjunction with boundary markers. Then all boundary
> > > > > faces gets marked with the largest marker value. I cannot see how
> > > > > that could be correct.
> > > > 
> > > > I don't get ' mesh_function.set_all(maxval);' or the code comment.
> > > 
> > > The point is that one should be able to define a form with domains say
> > > dx(0), dx(1) and dx(2) and then have a mesh file that marks a subset
> > > of the cells with '0', '1' and '2'.
> > > 
> > > Then the conversion to MeshFunction inserts '3' for all other
> > > (unmarked) cells. This allows a user to specify only the interesting
> > > cells and no need to mark the rest with -1 or None or similar.
> > 
> > That would make sense if the code would be:
> >   mesh_function.set_all(maxval+1);
> 
> Yes, that is the intention! Thanks for proofreading my code before
> I've even had a chance to test it. :-)

:)

You gave the impression that it was test when you asked me to wrap it to 
Python. Give me a ping when it is ready and I will have a bite at the SWIG 
code.

Johan

> --
> Anders
> 
> > Johan
> > 
> > > > >> So MeshMarkers may not be that bad. I'm starting to get used to
> > > > >> it... :-)
> > > > > 
> > > > > That's what worries me :)
> > > > 
> > > > Me too (worried, that is).
> > > 
> > > Don't worry.