 The CMF integration scheme becomes less accurate (and
takes smaller update steps) when the WF enters a CAP. The
CMF integration scheme should be modified and improved
such that it performs equally well when a CAP is
entered.
 If a propagation employs natural orbitals, the
initial wavepacket should be transformed to natural
orbitals after being read from file, to allow the use of
standard orbitals in the preceeding relaxation.
 Implement CDVR with CMF integrator.
 Implement usage of multidimensional surfaces with
CMF integrator.
 The MCTDH program should be able to dynamically
increase or decrease the numbers of singleparticle
functions during the propagation.
 In the CMF scheme, the equations of motion for the
singleparticle functions are not variationally optimal
for the given equations of motion of the coefficients. A
different projector and a symmetric propagation of
singleparticle functions and coefficients should be
implemented.
 The improved error estimate for the LanczosArnoldi
integrator should be implemented also in the Hermitian
Lanczos integrator.
 In a multipacket run one may increase the efficiency
of the propagation and decrease the memory needed by not
propagating the various packets simultaneously. This
would require to reorder the wavefunction vector and
select the correct Hamiltonian terms for the individual
packets.
 In a multipacket run it would be advantageous to use
different numbers of singleparticle functions for the
various packets. Note that the computation of the
crosscorrelation functions must be changed for
that.
 The MCTDH program may take a large amount of memory,
if there are large combined grids and many Hamiltonian
terms (e.g. a large natpot). The reason is the hpsi
array. The application of the Hamiltonian to the WF
should be reorganized (make the loop over k to the
outermost loop) such that this problem disappears.
