[CPMD-list] Lack of energy conservation...

[Nick Cooper]

Hello once again,

I am still having problems getting any sort of energy conservation for an
elevated temperature run of Si, C and H. This is meant to be the
equilibration of a liquid obtained by melting a random collection of atoms
but I think there must be something drastic wrong.

I utilised the NOSE PARAMETERS to increase the order of the Suzuki
integrator and the decomposition ratio for the thermostat but the energy
drift was still of the order of 0.1eV/ps, when I believe that one should
be looking at a drift of no more than 10E-5eV/ps?

The relevant part of my input file is:
&CPMD
  MOLECULAR DYNAMICS
  COMPRESS READ WRITE
  RESTART WAVEFUNCTION COORDINATES CELL LATEST
  QUENCH BO
  TEMPERATURE
   3500
  NOSE IONS
   3500 3000
  NOSE ELECTRONS
   0.035 10000
  NOSE PARAMETERS
   4 4 4 6 125 4
  MAXSTEP
   13780
  TIMESTEP
   3.
  EMASS
   200.
&END

I noticed that the ion temperature was fluctuating by +/- 1000 of so
(which is VERY large for a 64 atom sample, as here), so I thought my
choice of frequency for the ions (3000cm-1, which is the characteristic
frequency for a C-H stretch) might not be coupling paticularly well.
Thus I performed a VIBRATIONAL ANALYSIS run to find the harmonic
frequencies, only to discover that the maximum frequency is 4636cm-1,
which is well below the characteristic freqency for the electron
thermostat.

Finally, I decreased the timestep to 1a.u. but all this did was decrease
the gradient of the monotonically decreases conserved quantity.

Any ideas of what I'm doing wrong or suggestions for what to try next!?

Help would be very much appreciated.

Regards,

Nick Cooper

PhD Student,
University of Sydney