The old and the new version have both been applied to
roughly 200 real world minimization problems (minimizing energy
function to detect membranes in QMEANBrane). The differences
are marginal and within the expected range, since the
internal SVD algorithm in eigen has changed.

changes:

-GetPrincipalAxes in Vec3List
 Results with Eigen3 have been verified with 10 randomly picked pdb structures.
 Differences only occur with changing signs of the vectors.

-CalcPrincipalAxis in BoundingBox
 No difference in algorithm, the SelfAdjointEigenSolver no lives in
 Eigenvalues. Does in principal the same thing as the function above,
 just using different functionality.

-SVD superposer
 Unit tests still run through with the changes introduced for Eigen3
 and also manual testing seems to work well.

-CartoonRenderer
 Principal axes calculation for rendering helices as cylinders had to be
 adapted. Rendering looks exactly the same...

-SuperposeFrames
 Not yet tested...

ToDo:
numeric.h and levenberg_marquardt.h in the img/alg module have to be
ported. This will be done in the next few days.

When an atom gets deleted, all related torsions get deleted from
the entities TorsionImplMap. The ResidueHandles keep their own vector
of the same TorsionHandles. If an atom gets deleted, only the internal
torsion list of the corresponding residue got updated. This had to be
extended to all residues containing torsions related to the deleted
atom.

export them to Python

some typos

an entry page for the mm module

not only check for the number of bond partners, but also whether
there is an atom called SG in the bonded atoms.

of threads in the cpu platform