My Community

Hi everybody!

My system is a monoclinic molecular crystal with C, H, O and Br. I am trying to relax the whole structure (ISIF=3) by using forces as relaxation criterium (EDIFFG=-0.001). The behaviour of the procedure is as follows:

-- First, I start from the experimental crystal structure and the calculated structures are successively being relaxed, the forces are lower and lower and the same is observed for the energy of the system (TOTEN).

-- Then, a point is reached where the energy is the minimum one but the value EDIFFG is not fulfilled. From that point on, despite the system is still relaxing the value of the energy begins to grow. Thus, my question is simple: Does it make sense that a more relaxed structure had a higher energy than another one less relaxed?

Thanks

Hi,

maybe you are trapped by Pulay stress.
Increasing the cutoff might help.
Or do a couple of fixed volume calculations with different volumes (ISIF=4) and check for the volume with the lowest total energy. Restart with this your ISIF=3 calculation.

Hth

Alex

I have just checked the Pulay stress data and they are 0.00

So, could it be Pulay stress anyway?

Hm,

what does external pressure say? How large is your cutoff energy? How large is the cell and how many k-points do you use?

alex

K-points = 1
External pressure for de energy minimum around -3 kB
cell (approximately) 7.5x10x10
cutoff 400

hmmmmm,

I'd try at least 800 eV and up to 6x4x4 k-points.
Should be not that costly.

External pressure shoud decrease to sth. like +/-0.xx kB

alex

If you're speaking purely pragmatically, how much does the geometry change in those last few steps? IF your aim is to determine the potential minimum geometry, AND the positions are changing by only small fractions of Aangstroms, then you needn't worry about a SMALL increase in the total energy. You will likely never reach the EXACT minimum energy anyway.

1) have you set PREC=Accurate and
ENCUT to at least 1.3*(max(ENMAX))?

if the cell volume and / or shape change due to ISIF=3, at some point the original basis sets (as calculated from the input cell) may be not the most appropriate ones:
--> in any case, after pre-relaxing for a few (ca 5-10) ionic steps, restart from CONTCAR without using the WAVECAR file, in order to adjust the basis set to the new cell (voulme, shape) which should be much closer to the real equilibrium structure than the one you started with.

I have neither set PREC to accurate nor ENCUT to 1,3*(ENMAX). Should I do it from the beggining or to pre-relax the system at less accuracy?

Also, I must say that I am interested in a reasonable calculated structure as compared with the experimental one. Thus, I am also considering tjf's option. Anyway, I am in learning period so I want to know as much as possible about the program.

1) It may speed up your calculation if you start form the final results of your relaxation run, probably these results are pre-converged to some degree already.
2) please read the vasp-workshop tutorials to learn about VASP, they give a good introduction to the basic theory, and valuable hints on how to do reliable calculations