[CPMD-list] how to get a equilibrium state at finite temperature?

[Chen Ying]

Dear axel,

Thanks for suggestions. i have seen your previous post on cpmd-list and find that
the average fluctuation in microcanonical simulation is about dT = sqrt[2/(3*N)] *
T_average. So in my system N=108, dT=0.79K when T_average=10K, is pretty small. i
have used a higher temperature at 30K (delta T=8K) and EKNIC increased quickly,
like this:

        18  7.81601    25.3     -41.14702     -41.13403     -33.31802   0.101E-02 
 63.58
        19105.72102    26.4     -65.38594     -65.37238      40.34863   0.113E-02 
 59.52
        20924.11893    29.2      21.53426      21.54922     945.66815   0.124E-02 
 43.09
        21*********    30.0      40.29756      40.31295  241151.80755   0.137E-02 
 48.98
        22*********    30.0     209.73502     209.75041**************   0.148E-02 
 58.08
        23*********    30.0    1079.45058    1079.46597**************   0.160E-02 
 58.31
        24???????????????????????????????????????????????????????????   0.159E-02
2467.54
        25???????????????????????????????????????????????????????????????????????
3276.60
 

So i increased emass to 1200 then EKINIC stayed around 0.02 a.u. during the
simulation time (1500 steps, TIMESTEP=5 a.u.), temperature oscillated in the first
20 steps then decreased and was stable at 25K after 750 steps, EKS is a little
larger than that of 10K. Is it meaningful to restart from that trajectory?

Or your suggestion was beginning with a higher temperature without tempcontrol,
however, i can't get a optimized geometry as a initial input. The total energy
kept increasing if i used pcg, odiis or linear scaling geometry optimizer, and the
energy oscillated when i employ VASP and GUASSIAN to optimize. Perhaps the PES of
my system is so mild that it's hard to find the minima. So i annealed ions and
electrons from 30K for 10000 steps(TIMESTEP=5.0a.u.), then re-optimize geometry
restarting from it, the energy also increased. Then i just begin with 30K
restarting from the annealing results. EKINIC increased quickly also, and i had to
set EMASS=2000 to get a stable EKINIC around 0.03a.u.. The temperature decreased
quickly in the first 10 steps then slowly and was around 15K after 500 steps. Is
it OK if i restart from that for TEMPCONT IONS md at 4K?

In fact my system is hydrogen molecular crystal at very low temperature and
pressure, with large zero point motion and high isothermal compressibility owing
to quantum effect. I know i have to employ PIMD and NPT simulation, but first of
all, i think i should get a somewhat equilibrium state through regular cpmd. 

Thanks very much.


> CY> Dear all
> 
> dear chen ying,
> 
> CY>  I used the following steps in order to get a equilibrium state at 4K.
> CY> 1. wavefunction optimization
> CY> 2. a short MD runs at 10K, because the phase transition happens at 13K.
> 
> please not that phase transitions are subject to finite size effects
> and that depending on how accurately your system can be described with
> planwave/pseudopotential + DFT, the transition temperature can be pretty
> far off.
> 
> CY>    TEMPERATURE
> CY>    10.0
> CY>    TEMPCONTROL IONS
> CY>    10.0 2.0
> 
> using a delta of 2K here is pretty pointless, as your 
> temperature is likely to flucuate in equilibrium much
> more. note, that for a system of 32 water molekules (~100 atoms)
> the fluctuation in the instantaneous temperature is typically
> around +/-50K at 300K. see my previous post on cpmd-list for
> a handwaving explanation.
> 
> CY> 3. a relatively long MD runs at 4K
> 
> what is 'long'? how many ps?
> 
> [...]
> 
> CY> the problem is that EKS keeps reducing (but very slowly) during the
> CY> MD runs. I wonder whether i have used the wrong method to relax the
> CY> configuration, or I should take more runs until EKS becomes stable.
> 
> no. you are basically caught between a rock and a hard place.
> equilibration at low temperature takes extremely long. there are
> little fluctuations to get the system across barriers to find the
> lower minima. the fact that you essentially kill all fluctuations
> with the small delta in TEMPCONTROL IONS makes it worse. your system
> is not in equilibrium. i suggest you try using a higher temperature
> (e.g. 30K or 50K) to see whether the phase transition actually does
> happen, run that for a while, perhaps keep a few restarts from that
> trajetory and try if one of them  gives you an reasonably 
> good starting point to get to an equilibrated configuration.
> 
> other than that, you may want to have a look at the tricks
> that people do, who simulate glasses whether there is a useful
> idea in it for you. most likely, you'd have to add the functionality
> to cpmd, though...
> 
> regards,
>     axel.
> CY> 
> CY> Thanks in advance
> CY> 
> CY> 
> CY> 
> CY> 
> CY> _______________________________________________
> CY> CPMD-list mailing list
> CY> CPMD-list at cpmd.org
> CY> http://cpmd.org/mailman/listinfo/cpmd-list
> CY> 
> 
> -- 
> =======================================================================
> 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.
> 
>