[[Imported from SVN: r7369]]

[[Imported from SVN: r7368]]

[[Imported from SVN: r7367]]

[[Imported from SVN: r7366]]

[[Imported from SVN: r7365]]

Solve the minimization problems for the evaluation of gfe functions really down to machine precision.  This appears to make the overall algorithm much more stable

[[Imported from SVN: r7355]]

[[Imported from SVN: r7354]]

[[Imported from SVN: r7353]]

[[Imported from SVN: r7352]]

[[Imported from SVN: r7351]]

[[Imported from SVN: r7350]]

[[Imported from SVN: r7349]]

[[Imported from SVN: r7348]]

[[Imported from SVN: r7347]]

[[Imported from SVN: r7346]]

[[Imported from SVN: r7345]]

[[Imported from SVN: r7344]]

[[Imported from SVN: r7343]]

[[Imported from SVN: r7342]]

[[Imported from SVN: r7341]]

using the ideas of Patrizio Neff.

In particular, this code should handle real 3d continua, as well
as Neff-style Cosserat shells.

This initial commit only adds a lot of infrastructure.

[[Imported from SVN: r7339]]

[[Imported from SVN: r7338]]

[[Imported from SVN: r7314]]

[[Imported from SVN: r7313]]

[[Imported from SVN: r7312]]

[[Imported from SVN: r7311]]

[[Imported from SVN: r7310]]

[[Imported from SVN: r7309]]

[[Imported from SVN: r7308]]

[[Imported from SVN: r7307]]

[[Imported from SVN: r7306]]

[[Imported from SVN: r7305]]

[[Imported from SVN: r7304]]

[[Imported from SVN: r7303]]

[[Imported from SVN: r7302]]

[[Imported from SVN: r7301]]

[[Imported from SVN: r7300]]

[[Imported from SVN: r7286]]

don't compute and measure on the last grid.  Since we are doing uniform refinement that last grid is equal to the reference grid.  Hence computing there will lead to an error of zero and is a waste of time

[[Imported from SVN: r7233]]

instead of a singular value decomposition.  This makes computing
a Hesse matrix roughly 20% faster.

[[Imported from SVN: r7232]]