[CPMD-list] temperature Drift in CPMD

[Axel Kohlmeyer]

On Thu, 4 Oct 2007, Ho wrote:

YCH> Dear CPMD users:

dear ying-chieh hung,

since nobody cared to answer (and there are plenty people on 
this mailing list, that are more qualified than i am), a few
comments based on my personal experience.

YCH> I am performing a NVE CPMD simulation on a system with 105 atoms (35 O, 65
YCH> H, 3 F, 1C, 1 S). The system was initially equilibrated using NVT for about
YCH> 30 ps and then switched to NVE for another 30 ps. While the potential energy
YCH> (Eks) and the kinetic energy of ions (Temp) fluctuated around some mean
YCH> value, I observed a constant increase of electron kinetic energy (Ekinc) at
YCH> a rate of 0.0003 Hartree/ps. 
YCH> 
YCH> Can I safely ignore this and take the systems equilibrated? Or do I need to

you are mixing up two things: equilibration and adiabacity. for the
CP dynamics to work correctly adiabacity has to be preserved (= EKINC
should have ideally no drift, but with long trajectories this is not
always possible). whether your system is equilibrated (either the atomic
degrees of freedom or the electronic) is a different story. you never
know for sure, but if the potential energy and/or kinetic energy are
not drifting for a longer time, this is usually a good sign.

YCH> take other measures  to ensure the equilibration of the system? I have in
YCH> fact tried to quench the electron (QUENCH BOELECTRON) after the NVE
YCH> simulation but then found that the system temperature drops. Is there a

it has to be, since you take away all kinetic energy from the ficitious
dynamics (you set EKINC to 0.0 and then then system has to 'get it back' 
from the atomic degrees of freedom). you can compensate this if you run
ANNEALING IONS for a while with a value slightly > 1 (e.g. 1.001) until
you are back to normal.

YCH> better way to equilibrate the system? Following
YCH> is the detailed procedure that I used in my simulation. Any
YCH> comments/suggestions are greatly appreciated.
YCH> 
YCH> 
YCH> Here's the simulation procedure. Initially I set 
YCH> 
YCH> TEMPCONTROL IONS 
YCH> 400.0 80.0 
YCH> TIMESTEP
YCH>   5.0
YCH> and did about 1.5 ps simulation. Then I shifted whole systems to my desired
YCH> temperature 

you are missing two important pieces of information here.
the fictitious mass and whether you were using deuterium
or 'regular' hydrogen. good adiabacity is achieved with
a balanced choice of those values.


YCH> QUENCH BOELECTRON
YCH> TEMPERATURE
YCH>    353.0
YCH>  TEMPCONTROL IONS
YCH>    353.0 70.0

please keep in mind that for a system as small as you 
can afford with CPMD, temperature is only an approximation,
even more so with CP dynamics, since the 'drag-effect' of
the ficitious electron dynamics affects atoms of different
masses differently, and on top of that you have the deficiencies
of DFT that make the system also behave like it was at a
different temperature. for example DFT water freezes if you
run it long enough at 300K...

YCH> TIMESTEP
YCH>   5.0
YCH> for about 4.8ps simulations.  I observed the Ekinc is oscillated around
YCH> 0.013. I use the
YCH> value and shifted the system to NOSE NVT simulation and set
YCH>  
YCH> QUENCH BOELECTRON
YCH> NOSE IONS
YCH>      353.0d0   3000.0d0
YCH>  NOSE ELECTRONS
YCH>     0.0125d0  10000.0d0

i when turning on NOSE thermostats, you are usually best off, 
when the temperature is already close to the desired value.
for the electrons after a QUENCH BO ELECTRONS this is definitely
not the case, so i would first turn on the NOSE on the IONS
(and i would start with MASSIVE to facilitate faster 
equipartitioning and then drop back to regular) and keep the
electrons without a thermostat. this way you can see most
easily, if your system is adiabatic or something goes wrong.

finally, for the choice of the target temperature for the 
electron thermostat, i would go for a value slightly larger
than what you observe on average without the thermostat. this
way you make sure that you don't have an additional drag from
continously moving energy from one thermostat to the other.
you have to keep in mind, that with both thermostats on, you
are almost 'blind' against any problems with the wavefunction.
the fact that you saw a drift in EKINC even with the thermostat
on, is not a good sign. i would start checking this by reducing
the timestep (5.0 a.u. is pretty aggressive with default masses).

i guess checking all these suggestions and comments out,
will keep you busy for a while (...and i have to meet my
'babysitter' for dinner).

good luck with that,
    axel.


YCH> TIMESTEP
YCH>   5.0
YCH> for 12ps. Then I shifted the system into NOSE ION MASSIVE for about 12ps.
YCH> Later I released whole temperature control and tried to do a NVE simulation.
YCH> During a 30 ps simulation, I observed the TEMPP and EKS could maintain at a
YCH> mean value, but Ekinc
YCH> increased about 0.008 with a constant slope. 
YCH> 
YCH> 
YCH> 
YCH> 
YCH> Thanks and Regards
YCH> 
YCH> Ying-Chieh Hung
YCH> 
YCH> _______________________________________________
YCH> CPMD-list mailing list
YCH> CPMD-list at cpmd.org
YCH> http://cpmd.org/mailman/listinfo/cpmd-list
YCH> 

-- 
=======================================================================
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.