[CPMD-list] Timestep problem

[Ari P Seitsonen]

Dear Younes,

On Sat, 14 May 2005, Younes Ansari wrote:

> I have decreased the timestep to 3 and I didn’t get any NAN on my
> results
>
> &CPMD
>  MOLECULAR DYNAMICS CP
>  RESTART WAVEFUNCTION COORDINATES LATEST
>  TRAJECTORY XYZ
>  TEMPCONTROL IONS
>  500 5
>  TIMESTEP
>
>  3
>
>  MAXSTEP
>  100
> &END
> &DFT
>  FUNCTIONAL LDA
> &END
>
> &SYSTEM
>  SYMMETRY
>    1
>  CELL
>    9.000  1.0  1.0  0  0  0
>  CUTOFF
>    70.0
>    ANGSTROM
> &END
>
> &ATOMS
> *Cs_q1_LDA
>   LMAX=S
>   22
>      -3.779452      -3.779452      -1.889726
>      -3.779452      -3.779452       1.889726
>      -1.889726      -1.889726       0.000000
>      -3.779452       0.000000      -1.889726
>      -3.779452       3.779452      -1.889726
>      -3.779452       0.000000       1.889726
>      -3.779452       3.779452       1.889726
>      -1.889726       1.889726       0.000000
>       0.000000      -3.779452      -1.889726
>       3.779452      -3.779452      -1.889726
>       0.000000      -3.779452       1.889726
>       3.779452      -3.779452       1.889726
>       1.889726      -1.889726       0.000000
>       0.000000       0.000000      -1.889726
>       3.779452       0.000000      -1.889726
>       0.000000       3.779452      -1.889726
>       0.000000       0.000000       1.889726
>       3.779452       3.779452      -1.889726
>       0.000000       3.779452       1.889726
>       3.779452       3.779452       1.889726
>       3.779452       0.000000       1.889726
>       1.889726       1.889726       0.000000
>
> &END
>
> I want to know if there is anyway to increase the timestep to 1000
> without getting any NANs or without any big physical changes.

1000???? That's certainly impossible: The atomic motion is already much
faster than what your period would be. As kind of rule-of-thumb the time
step in classical or Born-Oppenheimer molecular dynamics has to be at most
1/6th .. 1/10th of the highest ionic vibration. In addition in
Car-Parrinello dynamics the time step is even smaller because the
electronic motion must average out during the ionic vibrations. And if
your cut-off energy is high (like in your case: 70 Ry), the highest
electronic oscillations are dominated by the highest G vectors (kinetic
energy dominates), even further reducing the maximal time step. The brute-
force way is to increase the time step (and the fictitious electronic mass
correspondingly) until the dynamics is no longer stable/the energy is no
longer constant/conserved.

[Jürg Hutter already pointed out the following problem, but you maybe
didn't read it:
http://www.cpmd.org/pipermail/cpmd-list/2005-May/004157.html ]

   But in your case you have a further complication: You have a metallic
electronic structure, and you'll almost certainly have transfer of energy
between the electron structure and the ionic vibrations: because there's
no electronic gap the electrons can be excited _very_ easily, thus
enabling an easy coupling - in other words your electronic and ionic
vibrational spectra overlap, enabling the energy transfer - or still in
other words, the electrons and ions are not adiabatically decoupled, as
required by the Car-Parrinello method. One solution is to use a (Nose)
thermostat for the electrons ["Adiabaticity in first-principles molecular
dynamics", P. E. Blöchl and M. Parrinello, Phys. Rev. B 45, 9413-9416
(1992)].

     Greetings,

        apsi

-- 
-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-
   Ari P Seitsonen / Ari.P.Seitsonen at iki.fi / http://www.iki.fi/~apsi/
   CNRS & IMPMC, Université Pierre et Marie Curie
       4 place Jussieu, case 115 / F-75252 Paris
   Tel: +33-1-4427 7542, Fax: +33-1-4427 3785, GSM: +33-6-6736 3820