a separate name.  We cannot rely on overloading for different arguments,
because embedded and non-embedded tangent vectors may have the same type.

[[Imported from SVN: r7098]]

... instead of coordinates in the embedding space.  The usual problem
is that when working in the embedding space, a projection onto the
trust-region makes the iterate leave the tangent space.

[[Imported from SVN: r7035]]

[[Imported from SVN: r5931]]

[[Imported from SVN: r5667]]

remove old code that uses the derivative of the distance functions (instead of the derivative of the squared distance function)

[[Imported from SVN: r5666]]

[[Imported from SVN: r5664]]

[[Imported from SVN: r5599]]

[[Imported from SVN: r5597]]

[[Imported from SVN: r5595]]

rewrite the gradient assembler such as to demand the derivative of the _squared_ distance from the target space.  Unlike the distance itself, the squared distance is differentiable at zero

[[Imported from SVN: r5546]]

implement matrix assembly by simply returning the identity matrix.  This should a sort of projected gradient descent.  It is too late at night for me to compute the actual matrix

[[Imported from SVN: r5545]]

[[Imported from SVN: r5542]]

[[Imported from SVN: r5531]]

[[Imported from SVN: r5529]]

[[Imported from SVN: r5527]]

[[Imported from SVN: r5516]]

[[Imported from SVN: r5513]]

[[Imported from SVN: r5324]]

[[Imported from SVN: r4138]]