I am studying the stability of a system using ab initio molecular dynamics simulation (as implemented in VASP) at temperature of 500 K in the NVT ensemble with 1 fs time step. I find that the total energy of the system is continously decreasing with time. I have attached the plot with 1500 steps (1.5 ps).
Can somebody please provide an insight to this graph and the reason behind such behavior?
Any help in this regard will be deeply appreciated.
Normally if there exist a non equilibrium stationary state then one expects to have an exponential decay to this final state under the isentropic and isochoric conditions (dS=0 and dV=0). But your graph almost up to 1400 steps shows linear decay without any leveling off. Therefore to be sure you should enlarge the testing time.
I recommend to use logarithmic time scale in order to get read off this initial transient stage. 1400 steps is too short for my taste!!!
I recommend you run until 10-15 ps, you should hopefully find the energy finaly converges.
The temperature fluctuation shows themselves as ripples on the energy versus time plots. I agree with Dr. Christian Jorgensen to extend your data much higher run numbers from 1.4 ps up to at least 10-20 ps in order to see the leveling off.
Use semilogarithmic plot: Log Abs(Energy) ve step number. The slop gives you inverse relaxation time in terms of step numbers.
Thank you all for your suggestions.
I will try to decrease the tolerance for electronic iteration as suggested by Alrik Stegmaier. Also I will definitely observe the behavior using large simulation time as suggested by Christian Jorgensen and Tarık Ömer Oğurtanı..
The drift in the energy is about 7 meV/500 steps on an average. Is this an acceptable value ? or precisely I would want to know if such a behavior is acceptable ?
I have used Real Space projection (LREAL = Auto). Could this be one of the causes of such a behavior?
- Badges
- Science topic