Next: - CP-dynamics with Damped
Up: Methods to Compute the
Previous: - Davidson Diagonalization:
Contents
Index
One can actually perform a wavefunction optimization by
running a CP-dynamics with all atom positions frozen and
then gradually removing kinetic energy from the fictitious
degrees of freedom by multiplying velocities with a factor
(in the example here it is set to 0.9, so 10% of the kinetic
energy will be removed in every step).
Particularly for systems, that are very difficult to converge
or where the ``fast'' methods like DIIS or PCG converge to
the wrong state, this may be the only way to get to the
(desired) ground state. To do this a few more modifications
of the input file are needed. The &CPMD section
becomes:
&CPMD
MOLECULAR DYNAMICS CP
ANNEALING ELECTRONS
0.90
MAXSTEP
200
EMASS
800.0
TIMESTEP
6.0
&END
and we need to add to the &ATOMS section the following:
CONSTRAINTS
FIX ALL ATOMS
END CONSTRAINTS
Note: Unlike with the regular wavefunction optimization methods
there is no automatic stop when the wavefunction is converged, so the
number of steps has to be either overestimated or the run restarted
several times. Also, the annealing factor has to be chosen carefully.
If it is too close to 1.0, it will take a very long time to converge,
but if it is too small (0.90 is quite aggressive), the calculation will
get ``frozen out'' before it fully reaches the minimum.
Next: - CP-dynamics with Damped
Up: Methods to Compute the
Previous: - Davidson Diagonalization:
Contents
Index
Costas Bekas
2008-09-04