[CPMD-list] Energy variations for calculations with torsional constraints

[Salomon Billeter]

Hi Veronique,

there have been a couple of small bug fixes in the constraints code
recently, but this problem is quite possibly related to the choice of the
geometry optimizer (default GDIIS/BFGS) which can be very efficient but
also sometimes gets trapped far from a minimum. You can either restart the
optimization from the last point, switch to the L-BFGS whose implementation
is very careful in choosing the size of the trust radius and rejecting
unsuccessful steps (keyword LBFGS; if you do that, choose also PRINT LSCAL
ON to get some diagnostics about the progress of the optimization), or you
can choose the conjugate gradient algorithm whose implementation performs a
line search to get a good step size.
Btw, do you want to do a periodic calculation? If not, ISOLATED MOLECULE
alone is not sufficient, SYMMETRY should be set to zero instead of one to
decouple the periodic images electrostatically from each other.

Best regards,
Salomon