Dear Naiwrit,

In general, temperature is a statistical property and only well-defined for large simulations; essentially "temperature" means a particular distribution of kinetic energies. This manifests itself in MD simulations (ab initio or otherwise) as fluctuations in the instantaneous kinetic energy, which is usually identified as the instantaneous "temperature".

 I should point out that these fluctuations are physical; temperature is maintained by coupling the atoms to a heat bath, so they randomly transfer energy to and from each atom. Whilst the transfer of energy averages to zero over long times, when only a few atoms are involved it's likely that there will be large, nonzero nett energy transfer; the more atoms there are in the simulation, the more the energy transfers tend to cancel out and the fluctuations are smaller.

The fluctuations (deltaT) are statistical such that:

deltaT/T = sqrt( 2/(3N) )

where T is the equilibrium temperature and N is the number of atoms in your MD simulation. Thus for a given N, the fluctuations are a constant fraction of the target temperature, and the way to reduce the fluctuations is to increase N (i.e. use a larger MD simulation systems with more atoms).

Hope that helps,

Phil

if the WAVECAR is not present in working folder than this, obviously will not be read. If you force the job continuation and setup ICHARG=0, there are no orbitals (waves) that the charge density can  be constructed from. you should have your WAVECAR and CHGCAR generated from previous run before attempting continuation job. then you need to read those files for next run (istart 1, icharge 1) and of course copy the contcar to poscar. don't forget to run in separate folder otherwise all the files will be overwritten so you will lose the previous trajectory and data.

sigma controls smearing width that, in other words, is the occupation around fermi level. so if you make it very small then there will be only little fluctuations. for this however you need denser k-point mesh.

Hi, Naiwrit.

To control the velocities in a MD simulation, you should use the SMASS option (http://cms.mpi.univie.ac.at/vasp/guide/node118.html). Otherwise, you would carry out a microcanonical simulation (it is the default value for SMASS, -3) and the total energy would be constant (not the temperature)

I recommend you to try different (positive) values of SMASS, in order to find a good Nosé mass for your system. In this way, the temperature would be constant.

To restart a MD simulation, you don't need the WAVECAR file. The CONTCAR file has also the velocities of the last step. You have to converge again the electronic density if you don't have WAVECAR, but in principle, it should work.

Best,

Fernando

Thanks to both. Generally WAVECAR and CHGCAR file are written at the last of the job. But if the job gets killed in between due to CPU time limit, both the WAVECAR and CHGCAR file would not be printed. Then I would be having difficulty in restarting using istart = 1. I found that copying CONTCAR to POSCAR is working, as Fernando said, but it takes a few cycles initially for the fluctuations to become similar as the last run.    

Dear Naiwrit,

In general, temperature is a statistical property and only well-defined for large simulations; essentially "temperature" means a particular distribution of kinetic energies. This manifests itself in MD simulations (ab initio or otherwise) as fluctuations in the instantaneous kinetic energy, which is usually identified as the instantaneous "temperature".

 I should point out that these fluctuations are physical; temperature is maintained by coupling the atoms to a heat bath, so they randomly transfer energy to and from each atom. Whilst the transfer of energy averages to zero over long times, when only a few atoms are involved it's likely that there will be large, nonzero nett energy transfer; the more atoms there are in the simulation, the more the energy transfers tend to cancel out and the fluctuations are smaller.

The fluctuations (deltaT) are statistical such that:

deltaT/T = sqrt( 2/(3N) )

where T is the equilibrium temperature and N is the number of atoms in your MD simulation. Thus for a given N, the fluctuations are a constant fraction of the target temperature, and the way to reduce the fluctuations is to increase N (i.e. use a larger MD simulation systems with more atoms).

Hope that helps,

Phil

Thank you for your elaborate answer Philip.

Hi Fernando, do you have a recommendation on how to make a good guess for SMASS?Thanks1