← Back to team overview

dolfin team mailing list archive

XML format for Higher Order meshes

 

Ok, here is what I suggest. If anyone sees anything wrong with this format, please say something!

=============== THE OLD WAY =============================

<?xml version="1.0" encoding="UTF-8"?>

<dolfin xmlns:dolfin="http://www.fenics.org/dolfin/";>
  <mesh celltype="triangle" dim="2">
    <vertices size="4">
      <vertex index="0" x="0.0" y="0.0"/>
      <vertex index="1" x="1.0" y="0.0"/>
      <vertex index="2" x="0.0" y="1.0"/>
      <vertex index="3" x="-1.0" y="0.0"/>
    </vertices>
    <cells size="2">
      <triangle index="0" v0="0" v1="1" v2="2"/>
      <triangle index="1" v0="3" v1="0" v2="2"/>
    </cells>
  </mesh>
</dolfin>

=============== THE NEW WAY =============================

# for an affinely mapped mesh; note the new variable `map_type'.

<?xml version="1.0" encoding="UTF-8"?>

<dolfin xmlns:dolfin="http://www.fenics.org/dolfin/";>
  <mesh celltype="triangle" dim="2" map_type="P1">
    <vertices size="4">
      <vertex index="0" x="0.0" y="0.0"/>
      <vertex index="1" x="1.0" y="0.0"/>
      <vertex index="2" x="0.0" y="1.0"/>
      <vertex index="3" x="-1.0" y="0.0"/>
    </vertices>
    <cells size="2">
      <triangle index="0" v0="0" v1="1" v2="2"/>
      <triangle index="1" v0="3" v1="0" v2="2"/>
    </cells>
  </mesh>
</dolfin>

# for a piecewise quadratic mesh; here map_type="P2".

<?xml version="1.0" encoding="UTF-8"?>

<dolfin xmlns:dolfin="http://www.fenics.org/dolfin/";>
  <mesh celltype="triangle" dim="2" map_type="P2">
    <vertices size="9">
      <vertex index="0" x="0.0"  y="0.0"/>
      <vertex index="1" x="1.0"  y="0.0"/>
      <vertex index="2" x="0.0"  y="1.0"/>
      <vertex index="3" x="-1.0" y="0.0"/>

      <vertex index="4" x="0.5"  y="0.05"/>
      <vertex index="5" x="0.6"  y="0.6"/>
      <vertex index="6" x="0.0"  y="0.5"/>
      <vertex index="7" x="-0.5" y="0.5"/>
      <vertex index="8" x="-0.5" y="0.0"/>
    </vertices>

    <cells size="2">
<triangle index="0" affine="false" v0="0" v1="1" v2="2" v3="4" v4="5" v5="6"/> <triangle index="1" affine="true" v0="3" v1="0" v2="2" v3="8" v4="6" v5="7"/>
    </cells>
  </mesh>
</dolfin>

# note the variable `affine' that indicates whether the triangle is actually higher order or not.

===============================================

- Shawn

On Tue, 19 Aug 2008, Anders Logg wrote:

On Tue, Aug 19, 2008 at 11:20:03AM -0400, Shawn Walker wrote:
I think in dolfin's mesh.xml format, I will need to put the extra
'curved' vertices into a separate list of data.  If I put all of the (say
2nd order polynomial mesh) vertices into the usual list of vertices, then
dolfin will think the matrix size is larger than it really is.  At least,
that is what the output looks like.  Does this seem correct.

One could either break the file format and introduce a <geometry> tag
that holds the geometry, or one could add a list of edge midpoints in
the data section. It could just be some <array> objects named "edge
midpoints x", "edge midpoints y" etc.

Also, I looked at this UFCCell.h file in Dolfin, and it looks like
quadrilaterals and hexahedrons are not implemented.  Is this true?

It's not supported anywhere except for in the UFC manual. :-)

--
Anders


- Shawn

On Mon, 18 Aug 2008, Kent-Andre Mardal wrote:

On ma., 2008-08-18 at 09:01 -0400, Shawn Walker wrote:
Sure, but I do want to have the option of some elements being curved and
others not.  This is actually practical since only the elements near the
boundary really need to be curved.

Sure, I agree totally on this. But this requires more. Neither UFC, the
form compilers (FFC or SFC) nor the FEM basis function generators,  FIAT
and SyFi, support different polynomial degree at different places.

Of course, in your case I guess it is only the geometrical mapping that
changes degree throughout the mesh and this may simplify alot eg. the
finite element basis functions will be the same throughout the mesh. The
dofs can also be reused.

I only suggest starting with the implementation of a higher-order mesh
because it seems as a feasible first step.

Kent



Also, it is possible to evaluate some forms exactly, even if the local map
is a polynomial.  Of course, this can be ignored for now...

- Shawn

On Mon, 18 Aug 2008, Kent-Andre Mardal wrote:

On s?., 2008-08-17 at 00:08 -0400, Shawn Walker wrote:
I think you are right.  I would propose the following:

1. Add a meshfunction to the mesh.xml file.  The name of it will
correspond to a new enum variable in ufc::cell.  For example, say we have
piecewise quadratics for the local map, so we call it "P2_map".  Maybe
later this could be made a more permanent xml data type, but for now this
should be ok.  The mesh function would just be a vector lagrange 2nd
order function.  And there would also be a boolean function on each cell
to say whether the element is straight or not.  This of course rules out
mixing curved types, i.e. having P3 and P2 in the same mesh.  But I am ok
with that for now.

2. In ufc::cell, we have another enum variable called "map_type":

enum map_type {P1_map, P2_map, P3_map, etc...}

Of course, this will only be for lagrange type polynomial maps.  But this
enum variable can have other types (i.e. iso-geometric).  We then create a
variable "map_type cell_map_type;"

Yes, we probably would need one extra variable for the map.

The way I see it, if one assume that all elements in the mesh has the
same order, the main work would be to create a dolfin mesh containing
a higher order geometry. Once this is done, you would need to ensure
that the coordinates in ufc::cell follows the UFC convention locally.
Finally, the computation of the Jacobian, the inverse of the Jacobian
and the determinant of the Jacobian has to be moved inside the
quadrature loop. This is the only modification to the form compiler, I
believe. The finite elements, its derivatives, dofs etc
stay the same.

Hence, once a higher order mesh is up an running, there should only
be minimal modifications to the form compilers and UFC.




3. When the mesh is read in, and the element is created, I think the
"coordinates" variable can store the vertex positions of the element,
including the extra ones for the curved sides.  If not, then some other
variable would need to be created to store it; actually, that may be
better.  Also, the "cell_map_type" will be set to either P1_map, or
P2_map, or etc...

4. The tabulate_tensor routine will have a switch to pick the correct
routine for computing the tensor coefs.  This will require modifying FFC
which I don't know yet.

5. Another thing that should change will be the basis function
evaluations.  This may be a pain.  Computing the inverse map to get the
location of the coordinates in the reference element may be inconvenient.
This can be ignored for now; this isn't really critical for what I would
like to do.

I am a little confused on where the ufc::cell data gets set by the data in
the mesh.xml file.  I'm still very new to this code.  Any help would be
appreciated.  If I am totally off on this, please say so!


Have a look in UFCCell.

Kent








Follow ups