ParaView doesn't support P3 elements.  Therefore, the data is
downsampled onto a p2 space.

[[Imported from SVN: r9863]]

[[Imported from SVN: r9862]]

For some reason, when numpy is used, I can't use ctrl-c to stop
the program.  But numpy is not really used anyway, so let's get
rid of it.

[[Imported from SVN: r9861]]

[[Imported from SVN: r9860]]

[[Imported from SVN: r9859]]

[[Imported from SVN: r9858]]

[[Imported from SVN: r9857]]

[[Imported from SVN: r9856]]

[[Imported from SVN: r9855]]

[[Imported from SVN: r9854]]

[[Imported from SVN: r9853]]

'Mixed' means that the spaces for deformations and microrotations can
be chosen separately.

At this point there hasn't been a lot of testing yet.  In particular,
parallelizing the assembly is currently not supported.

[[Imported from SVN: r9852]]

A few recent fixes haven't made it to Debian yet.

[[Imported from SVN: r9851]]

Instead, the FD approximation is now done in a separate class.

[[Imported from SVN: r9850]]

[[Imported from SVN: r9849]]

'Mixed' means that there are two different spaces used for the
deformation and the microrotations.

This implementation currently only does second order, and it
only writes the deformation, not the microrotations.

[[Imported from SVN: r9848]]

[[Imported from SVN: r9847]]

[[Imported from SVN: r9846]]

[[Imported from SVN: r9845]]

[[Imported from SVN: r9844]]

That projection is the zero map, not the identity.

[[Imported from SVN: r9843]]

This used to be in the base class as default implementations, but I think
it is nicer to have it in a derived class.  In particular, this will
eventually allow to do the FD approximation with high-precision number types.

[[Imported from SVN: r9842]]

It sometimes creates problems with ADOL-C.

[[Imported from SVN: r9841]]

[[Imported from SVN: r9840]]

Gauss-Seidel converges equally good/bad as IPOpt, but it is much faster.

[[Imported from SVN: r9839]]

[[Imported from SVN: r9838]]

Amazingly, this really does have a measurable effect on the overall computation
speed.  Assembly times for the global Hessian and gradient for the Cosserat shell
energy problem drop by about 5% (!)

[[Imported from SVN: r9837]]

That's obviously more efficient when solving one linear equation
with multiple right hand sides.

[[Imported from SVN: r9836]]

Previously the energy norm was used, but as we are solving non-convex
problems the Hessians may be indefinite, and then the EnergyNorm is not
a norm.

[[Imported from SVN: r9835]]

[[Imported from SVN: r9834]]

[[Imported from SVN: r9833]]

[[Imported from SVN: r9832]]

[[Imported from SVN: r9831]]

[[Imported from SVN: r9830]]

[[Imported from SVN: r9829]]

That makes it easier to keep them consistent.  Also, the code is shorter.

[[Imported from SVN: r9822]]

[[Imported from SVN: r9821]]

[[Imported from SVN: r9820]]

This used to be 'int', because dunepython didn't support the python-to-C
conversion for bool.

[[Imported from SVN: r9819]]

Previously we called hess_mat (vector-mode second derivatives) with the
canonical basis vectors, and then rotated the resulting matrix.
This is wasteful, as more directions are used than necessary.
This patch uses the minimal number of directions.

[[Imported from SVN: r9818]]