dolfin team mailing list archive

[Shawn Walker <walker@xxxxxxxxxxxxxxx>]

DISCLAIMER: the gzipped file will be sent in a separate email.

I have done an example implementation that modifies the FFC generated
form.h file so that P2 iso-parametric elements can be computed.  To look
at this, do the following:

1. uncompress and look at the PDF; it gives a short discussion.

2. compile and run it.  it will generate a decent amount of output which
is self-explanatory.  The most interesting part is at the end, which
indicates the max error between the entries of the matrix that DOLFIN
generates and an almost identical matrix that I created with my own code.
The error I get is 2.72E-9 using a 25 point quad rule in FFC.  The matrix
created with my code uses a 37 point quad rule (which might be different
than what FFC uses; I didn't check).  Note that there are only 2 triangles
in the example mesh (and only one is curved); the triangles only have a
diameter of approx 0.5.

3. The main.cpp is the main part of the demo.  If you want to see what I
did, open the Poisson.h file and do a text search for "SWW".  Please keep
in mind that this is just a first cut.  The coding I inserted is not
efficient and contains some non-OOP pieces.

I think that, eventually, implementing non-linear maps should be done with
a class.  I didn't do that here because this is just a test.  And this
class should eventually be integrated (maybe abstractly) into UFC.  I'm
not completely clear yet on how to do it.  Any suggestions would be
welcome.

- Shawn