[CPMD-list] "NaN" in path integral molecular dynamics?

[Juerg Hutter]

Hi

your calculation is obviously non-stable. There is an onset of instability
starting at step 6 that leads within a couple of addtional step to a full
divergence. Reasons for this could be a problem with equilibration,
meaning you have to choose system parameters that allow for more
gentle approach to equilibrium.
I would try the following changes

- reduce the timestep
- change the value for NOSE IONS (10000) to a more physical value
- increase the taget value of the electronic kinetic energy by at least
  an order of magnitude
- decrease the value for NOSE ELECTRONS to 8000 - 12000
- do you really need a Nose-Hoover chain of length 10 for the ions?
- ramp up the Suzuki/Yoshida integrator
  NOSE PARAMETER
  10 4 4 6. 25 4

regards

Juerg Hutter


On Mon, 11 Jul 2005, Lan-Feng Yuan (lyuan at Princeton.EDU) wrote:

>   Hi, dear community,
>
>   I am practising path integral calculations in CPMD. To my surprise, sometimes the dynamics diverge in several steps generating "NaN" (not a number). For example:
>
>     NFI  EKINC/P  TEMP   EKINP(PRI)  EKINP(VIR)      EKS/P      EQUANT    ECLASSIC        EHAM     TCPU
>       1 0.00000    0.8     0.00033     0.04348   -17.17017   -17.16984   -16.79743   -16.79743    15.06
>       2 0.00000    3.3     0.00040     0.04295   -17.17038   -17.16998   -16.79743   -16.79743    15.02
>       3 0.00000    7.5     0.00051     0.04225   -17.17072   -17.17021   -16.79743   -16.79743    14.99
>       4 0.00001   13.3     0.00067     0.04154   -17.17120   -17.17053   -16.79744   -16.79743    14.93
>       5 0.00002   20.9     0.00087     0.04055   -17.17180   -17.17093   -16.79744   -16.79742    14.96
>       6 0.00009   30.1     0.00112     0.04018   -17.17242   -17.17130   -16.79734   -16.79726    14.98
>       7 0.00092   40.9     0.00140     0.03735   -17.17169   -17.17029   -16.79583   -16.79491    14.92
>       8 0.01155   53.7     0.00173     0.04254   -17.15014   -17.14841   -16.77265   -16.76110    14.92
>       9 0.06071   66.2     0.00210     0.02090   -16.86580   -16.86370   -16.11061   -16.04989    14.98
>      10 NaN       86.1     0.00249     0.08526   -14.48683   -14.48433 NaN         NaN            15.00
>      11 NaN     NaN        0.00295 NaN         NaN         NaN         NaN         NaN            11.98
>
>   I also note that "EHAM", which is supposed to be conserved to machine accuracy, changed before "NaN" appears. This calculation is for a H2O molecule in a 10 A cubic cell (SYMMETRY 1), cutoff 100 Ry, Trotter Dimension 8, FUNCTIONAL PBE (NEWCODE), and in &CPMD section:
>
> MOLECULAR DYNAMICS
> PATH INTEGRAL
> NOSE IONS MASSIVE
>    300.0  10000.0
> NOSE PARAMETERS
>    10  3  3  6.0  7  1
> NOSE ELECTRONS
>    0.001   20000.0
>
>   The initialization of the 8 beads is done by "DEBROGLIE CENTROID 300.0" in &PIMD and "ISOLATED MOLECULE",
> "OPTIMIZE WAVEFUNCTION", "PATH INTEGRAL" in &CPMD. What could be the reason? Thank you very much! Regards,
>
>   Lan-Feng
>
> ----------------------------
> Dr. Lan-Feng Yuan,
> Postdoctoral Associate,
> Department of Chemistry,
> Princeton University,
> Princeton, NJ 08544, USA
> Office Phone: 1-609-258-1834
> Fax: 1-609-258-6746
> ----------------------------
>
> _______________________________________________
> CPMD-list mailing list
> CPMD-list at cpmd.org
> http://cpmd.org/mailman/listinfo/cpmd-list
>