My Community

Hello,

once again I have a problem. As I started learning VASP I was told that positive free energies in the OUTCAR are a bad thing. Right now I have to deal with it because in my calculations I get posistve free energies during the calculations for the cutoff-convergency (approx. 80 eV). If I do not specify the cutoff, my calculations give me a negative free energy (about -70eV).

Where could be my mistake?

Thank you very much for your answers.

In general I wouldn't say a positive free energy nessicarily is a "bad thing". It just means that energy of your system is higher than the reference energy of the non-polarized atoms and the system will therefore not form.

Nevertheless, we need a bit more information in order to adress your issue. Which potential do you use? What's the default cutoff for your potential and which cutoff do you use when you specify it? What's your geometry (is it reasonable)? Are the energy values your refer to converged within the number of electronic steps (default 60)?

Regards,
/Dan Fors


<span class='smallblacktext'>[ Edited Tue Jul 15 2008, 05:47PM ]</span>

Hi,

I use the GGA-PW91 pseudopotential. The default cutoff is 180 eV. I start with 150 eV and then I increase it by 50 eV-steps up to 500 eV. I use supercells and within the supercell I have kind of a 2d-geometry with in fcc different lengths in (110) respectively (100)-direction but finite.The number of electronic steps is always 60 so it should not be convergedin this sense.

Can you help me out?

Thx

If all the runs reach 60 electronic steps (and you havn't changed NELM), then you havn't achieved self-consistency and you can't use the total energy values. To achieve self-consistency you will have to play around with your settings. First thing you can do is to increase the NELM to a higher value so that you get more steps before the cycle breaks (a bit naive approach but it works sometimes). However you should be able to achieve convergence within 20-40 steps so it's better to change other settings.

The k-points settings, mixing parameters, algorithm (RIM-DIIS/DAV) among other things can affect the convergence rate (I even had to switch to another exchange-correlation functional in one case to achieve convergence). For slabs and other 2D-structures it may however be helpful to switch to a linear mixing scheme. You can look at the tutorial slides:

http://cms.mpi.univie.ac.at/vasp-worksh ... ectron.pdf
pages 21-22 primarily and also the pages before. Try the scheme on page 22.

You may also consider to change the NELMDL value to -10 or -12 something. It will however depend on your system.

http://cms.mpi.univie.ac.at/vasp/vasp/node104.html

Best regards,
/Dan Fors





<span class='smallblacktext'>[ Edited Tue Jul 15 2008, 06:55PM ]</span>

positive total energies in that range (80eV) usually are due to a very bad guess of the starting geometry (system at extremely high pressure or under large tension), all other calculational parameters should not cause such large positive energies, even if they are not chosen very well. I would suggest to have a look at the interatomic distances of the NN and next-nearest neighbors as written in OUTCAR, and compare them with the experimental interatomic distances of the respective crystals (please remind that VASP uses Ã… as unit of length, not a.u.)

Is it really possible to use the energy values anytime during the self-consistency cycle to predict that you have a bad geometry? I understand that if we are close to the end of the convergence cycle and the energy differences are small between each electronic step we may use the energy as a guideline where large possible values corresponding to a bad geometry/large pressure as you (admin) pointed out.

In this case however I understood it as the default cutoff energy yielded the same range of energy but negative while other cutoff-energies gave the the same amount but positive. I would interpret it as that we have very large energy differences between each electronic step even when we reach NELM steps and thus very bad convergence. The geometry may off course be the reason to this behaviour (and a likely one) but is it that straightforward to draw the conclusion that is the most probable one? I'm just curious.

Best regards,
/Dan Fors
<span class='smallblacktext'>[ Edited Thu Jul 17 2008, 02:16PM ]</span>

hi Dan,
well, of course you are correct, the cutoff does influence the total anergy, but I have never experienced that simply changing the cutoff (supposed you started with a reasonable setup of your calculation!) turned the total energy from -70 to +80eV (the system cannot be too lagre with such a TOTEN!): For such a system, changes in the cutoff may change the total energy by a few eV at most (ie in the worst case some oscillations close to 0)
what may have happened (I guess) is the following combination of things (it is always hard to draw conclusion without seeing the actual numbers):
a system with a bad initial geometry guess (no electronic convergence within 60 steps + large forces), +
automatic geometry optimization (eg ISIF=3) +
too low precision (PREC=low instead of accurate) to calculate the force tensors reliably +
POTIM too large (overshooting correction of the first extrapolated ionic configuration)

However, if I got such behaviour of a calculation, the 2 things I would check first were to look into the OUTCAR file, to have a look whether the (1) interatomic distances of the starting geometry are too short and/or whether(2) the pressure on the cell seems reasonable.

Hello again,

well the geometry should not be the problem. I do not use my initial guess for the calculation. I first did the ionic relaxation without specifying my cutoff. I used ISIF=0 and I prefer using PREC=high. In the mean time since I have been starting this thread I found out that the reason for my problems could be the kpoint-mesh. I started for the 1d problem with a 1d kpoint-mesh. When I relaxed the geometry I used a 3d kpoint-mesh. Could this cause my problem?

Temporarily I'm checking any other point you mentioned.

Thank you very much.

Best regards

if the structure is 1d (linear chain) you have to keep the 1d-kmesh for relaxation as well. to relax the interatomic distances in the chain, you should then either
--) vary the cell dimension along the chain by hand and remain the other dimensions constant
--) or (if you want to do it automatically) use ISIF=4, to account for a possible relaxation of the interatomic distances along the chain but keep the separating vacuum large enough (otherwise your cell might collaps)

Well the structure is actually quasi-1d, so that it is in general 2d.

@admin: What do you mean by a reasonable pressure inside the cell?