[CPMD-list] EHAM always increases in H-containing system

[Axel Kohlmeyer]

On Fri, 19 Oct 2007, Paul Fleurat-Lessard wrote:

PL> Dear Axel and CPMDers,


dear paul,

PL> Is there a way to estimate what should be the correct kinetic energy 
PL> for the electrons ?

in theory, you should not even need a thermostat on the electrons.
but particularly for long trajectories, you'll be picking up some
noise and thus have a small increase. also, like in classical MD,
many people use a rather aggressive time step and compensate the
resulting energy drifts with a thermostat. since the CP method by
itself imposes some errors/artefacts, it is considered ok for as 
long as the error from the more aggressive parameters is small compared
to the systematic errors from the CP hamiltonian and DFT.

in practice, you can just run for a while (several picoseconds) without
a thermostat and then use the average EKINC value (or better a 
slightly higher value) of the last part as target kinetic energy.

now what value you will get depends on your system, the number of 
electrons, the fictitious mass, the time step etc. etc.
particularly critical are the first few thousand steps after you
start from a converged wavefunction. if you have a large band gap,
you can generally just start, will then see a peak in EKINC and it
then settles to some average, "natural" value for EKINC.

with a smaller band gap this will not always be the case and you 
run the risk of 'losing' the electrons. since your system may also
be equilibrating at the same time, it may be required to quench
the electrons a few times to the BO surface and restart of zeroed
electron velocities. in really tricky cases you need a more 
gentle way of getting the CP dynamics started, e.g. by a geometry 
optimization or cooling down via annealing and then a very slow and 
gentle heating up with annealing (this time using a coefficient
> 1.0). if you have a (near) metallic system, you could end up spending
almost as much cpu time on getting your system equilibrated and 
up to speed than for the production part of the run. 

depending on how you start your CP dynamics, you will end up with
a different "natural" EKINC value. the lower it is the better. :)

there have been attempts to seed the initial values for the electron
velocities from extrapolation from a couple of BO steps, and thus 
avoid the "overshoot" issue, but since the order of the states 
keeps changing it is not as easy as looks.

hope this helps,
    axel.


PL> 
PL> Regards,
PL> Paul.
PL> 
PL> 

-- 
=======================================================================
Axel Kohlmeyer   akohlmey at cmm.chem.upenn.edu   http://www.cmm.upenn.edu
   Center for Molecular Modeling   --   University of Pennsylvania
Department of Chemistry, 231 S.34th Street, Philadelphia, PA 19104-6323
tel: 1-215-898-1582,  fax: 1-215-573-6233,  office-tel: 1-215-898-5425
=======================================================================
If you make something idiot-proof, the universe creates a better idiot.