It offers a 'double' interface, but we don't really need that.

[[Imported from SVN: r9874]]

[[Imported from SVN: r9873]]

[[Imported from SVN: r9872]]

This fixes a dubious compiler error.

[[Imported from SVN: r9871]]

[[Imported from SVN: r9870]]

This code compares scalar and vector AD, and a finite difference
approximation.

[[Imported from SVN: r9869]]

[[Imported from SVN: r9868]]

It appears that there may be very small energy increases due to
rounding errors.

[[Imported from SVN: r9867]]

Solvers from dune-solvers expect the number type to be exported
as field_type.  This patch adds the missing export to RealTuple.

[[Imported from SVN: r9866]]

This bug has now influence as long as mu_c is zero, but for
non-zero mu_c there was one term missing.

[[Imported from SVN: r9865]]

[[Imported from SVN: r9864]]

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]]