[[Imported from SVN: r7908]]

[[Imported from SVN: r7907]]

[[Imported from SVN: r7906]]

[[Imported from SVN: r7905]]

[[Imported from SVN: r7904]]

[[Imported from SVN: r7903]]

[[Imported from SVN: r7902]]

Hence we want an interface more like a local basis, rather
than individual functions.

[[Imported from SVN: r7901]]

[[Imported from SVN: r7899]]

[[Imported from SVN: r7898]]

And use it to implement the Rotation<3> class.  This makes the
Rotation implementation easier to understand.  Previously we
have used a FieldVector<3> when a skew matrix was meant.
This leads to confusion now that we also use quaternions as
tangent vectors of SO(3).  In local coordinates, these also
have 3 entries.  Adding the new skew matrix class makes it
clearer what kind of mathematical object is meant.

[[Imported from SVN: r7897]]

It seems dangerous: I tend to get confused what these arguments
are.  Plus, the method is unused.

[[Imported from SVN: r7890]]

[[Imported from SVN: r7889]]

[[Imported from SVN: r7888]]

[[Imported from SVN: r7887]]

[[Imported from SVN: r7886]]

simplify the code by making the AverageDistanceAssembler accept the interpolation weights in a std::vector of FieldVectors

[[Imported from SVN: r7885]]

[[Imported from SVN: r7884]]

It's result is not necessarily a Rotation, hence such a method
would better be placed in the general 'Quaternion' class.

Also, nobody in this module used it anyways.

[[Imported from SVN: r7883]]

[[Imported from SVN: r7882]]

[[Imported from SVN: r7881]]

[[Imported from SVN: r7880]]

[[Imported from SVN: r7879]]

[[Imported from SVN: r7868]]

[[Imported from SVN: r7867]]

[[Imported from SVN: r7866]]

[[Imported from SVN: r7865]]

framework wants a global function space basis.

So far the P1 basis is hardwired for simplicity.

[[Imported from SVN: r7863]]

[[Imported from SVN: r7860]]

[[Imported from SVN: r7856]]

[[Imported from SVN: r7848]]

[[Imported from SVN: r7847]]

[[Imported from SVN: r7843]]

[[Imported from SVN: r7842]]

[[Imported from SVN: r7841]]

[[Imported from SVN: r7840]]

Computation of analytical energy gradients doesn't work, yet.

[[Imported from SVN: r7839]]

[[Imported from SVN: r7838]]

[[Imported from SVN: r7837]]

We need the extra flexibility for higher-order functions.
Run-time loss is around 1% -- that's okay.

[[Imported from SVN: r7836]]