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

[[Imported from SVN: r7232]]

[[Imported from SVN: r7231]]

using a finite-difference approximation only for the derivative
of the gfe gradients wrt coefficients.

[[Imported from SVN: r7230]]

[[Imported from SVN: r7229]]

[[Imported from SVN: r7228]]

[[Imported from SVN: r7227]]

[[Imported from SVN: r7226]]

[[Imported from SVN: r7225]]

[[Imported from SVN: r7206]]

[[Imported from SVN: r7205]]

[[Imported from SVN: r7201]]

[[Imported from SVN: r7196]]

[[Imported from SVN: r7195]]

Also: fix more bugs in the implementation of the derivative of gfe
function gradients wrt the coefficients.  This appears to work now,
but inexact evaluations remain an issue.

[[Imported from SVN: r7180]]

[[Imported from SVN: r7179]]

[[Imported from SVN: r7177]]

[[Imported from SVN: r7176]]

rename method assembleHessian of AverageDistanceAssembler to assembleEmbeddedHessian, to make clearer what it does

[[Imported from SVN: r7175]]

[[Imported from SVN: r7174]]

method to compute the actual Hesse matrix.  For some strange reason this slows down convergence, rather than speeding it up.  to be investigated

[[Imported from SVN: r7173]]

[[Imported from SVN: r7172]]

[[Imported from SVN: r7141]]

[[Imported from SVN: r7140]]

[[Imported from SVN: r7138]]

[[Imported from SVN: r7137]]

[[Imported from SVN: r7136]]

[[Imported from SVN: r7135]]

[[Imported from SVN: r7134]]

[[Imported from SVN: r7133]]

[[Imported from SVN: r7132]]

[[Imported from SVN: r7131]]

add a method with an fd approximation of the derivative of the function gradient.  only for testing purposes

[[Imported from SVN: r7130]]

[[Imported from SVN: r7129]]

[[Imported from SVN: r7128]]

[[Imported from SVN: r7127]]

[[Imported from SVN: r7126]]

[[Imported from SVN: r7125]]

[[Imported from SVN: r7124]]

[[Imported from SVN: r7123]]

[[Imported from SVN: r7122]]