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?

Thank you Alrik Stegmaier !