My Community

Hi,

I am trying to perform an AIMD melt and quench simulation using the NVT ensemble and the Nose-Hoover thermostat but have noticed that I am getting a consistent drift in the total energy (E = F + EK + SP + SK). The drift is on the order 0.04 eV / ps. I have attached my INCAR, POSCAR, and POTCAR and a png plot of the drift. My INCAR is also copied below. When I calculated the slope of F, EK, SP, and SK only SP had a slope significantly above zero. This slope also matched the slope of E which suggests to me that my issue is related to the Nose thermostat but I am not sure why.

In other simulations I have tried changing ENCUT to 450 and 500 eV, smass to 2, 3, and 4, sigma to 0.1, changed the time step to 0.5 fs, changed lreal to false, and expanded the lattice constant to perhaps a more equilibrium value. These calculations also had a consistent drift in the total energy.

Also, FYI I am using VASP 6.3.2 compiled for use on gpus.

Is there something I am missing?

Thanks!

INCAR created by Atomic Simulation Environment
ENCUT = 400.000000
POTIM = 1.000000
SIGMA = 0.350000
SMASS = 0.000000
TEBEG = 4000.000000
TEEND = 4000.000000
EDIFF = 1.00e-06
ALGO = fast
GGA = PE
PREC = normal
IBRION = 0
ISIF = 2
ISMEAR = 0
ISYM = 0
KPAR = 1
NELM = 100
NELMIN = 5
NSW = 10000
NWRITE = 0
NCORE = 8
MDALGO = 2
LCHARG = .FALSE.
LREAL = Auto

Dear graham_pritchard,

When looking at your POSCAR file I could see you are starting from a very symmetric structure. I guess a relaxed 0K structure.
You are running your molecular dynamics simulations at 4000K. So the energy drift that you see in your simulation
is the equibrilation of your structure from the 0K to the 4000K structure. After the system reaches the equilibrium
structure for the desired temperature, the energy should start to fluctuate around the equilibrium energy.

I also saw that you are using a relatively small time step of 1fs for quite heavy atoms. You could use a larger time step to speed up
your equilibration time. For equilibration, it might be more suitable to use a stochastic thermostat such as the Langevin thermostat or the Andersen thermostat.
They are applying stochastic collisions with a heat bath to your particles which might speed up your equilibration time.


All the best Jonathan

Thank you for the feedback Jonathan!

Best,
Graham