How to test whether k-points is enough

Message

fzcsydl · #1 Post by **fzcsydl** » Fri Aug 06, 2010 4:38 pm

Anyone knows an efficient way to test how many k-points will be enough for your calculation. e.g. I want to do a DOS calculation of a supercell structure of 10 atoms, 20 atoms or 30 atoms.

Thanks a lot.

metosa · #2 Post by **metosa** » Tue Aug 10, 2010 7:41 am

Suppose that you are studying graphene structure. In hexagonal lattice with 2 C atoms, KPOINTS-vs-ENERGY is like

1 -13.41311200
2 -20.93250200
3 -19.62164500
4 -20.16754800
5 -20.15429800
6 -20.18973500
7 -20.19344100
8 -20.18210000
9 -20.16419900
10 -20.17921000
11 -20.18077800
12 -20.18167200
13 -20.18239600
14 -20.18055800
15 -20.17698900
16 -20.18019900
17 -20.18071300
18 -20.18082800
19 -20.18103300
20 -20.18045400
21 -20.17924400
22 -20.18036400
23 -20.18058900
24 -20.18061200

When you plot this values, you will notice that the energy converges around 13x13x1, so that you can use this value for further structural relaxations. If you want to plot DOS, first you need to obtain a converged CHGCAR and then use that file for DOS calculations with ICHARG = 11 tag in INCAR and 41x41x1 KPOINTS.

In addition, if your system is a 2x2 graphene superlattice, wo can reduce your KPOINTS to 7x7x1.....

fzcsydl · #3 Post by **fzcsydl** » Sat Aug 14, 2010 12:26 am

[quote author= 11 tag in INCAR and 41x41x1 KPOINTS.

In addition, if your system is a 2x2 graphene superlattice, wo can reduce your KPOINTS to 7x7x1.....[/quote]

Thank you for your reply. If I am understanding correctly, you should first do a test of k-points versus total free energy till you find the k-points at which the total free energy converge.
But I don't understand why you need to increase the k-point to 41*41*1 in the DOS calculation. Why not use the relaxation value 13*13*1.

Looking forward to your reply.
Thanks a lot.

metosa · #4 Post by **metosa** » Sat Aug 14, 2010 9:59 pm

Refering the manual ( http://cms.mpi.univie.ac.at/vasp/vasp/A ... tions.html ) >>>> A high quality DOS requires usually very fine k-meshes.

Charge density converges very rapidly with number of KPOINTS. Namely, 13x13x1 KPOINTS CHGCAR is almost equal to 41x41x1 KPOINTS CHGCAR. But for DOS calculations, you should use as many as KPOINTS. The situation is reversed for relaxations. For finding enegy minimums, you should not use too much KPOINTS since energy does not change very much (as above values)...

limkr · #5 Post by **limkr** » Mon Aug 23, 2010 11:09 pm

Thanks metosa for your explanation.
I have a question similar to this topic here about total energy convergence.

When we say energy is converged, it may be acceptable to have 1meV/atom of energy difference between two consecutive energy calculations. I am wondering if 3-4 meV/atom of energy difference may also be acceptable..

For example, if I need to calculate a larger system of graphene (128 C atoms) with a Fe adsorbate (13 Fe atoms), it will be a burden to have larger k-points and ENCUT that give 1meV/atom of energy difference in a unit cell.

In a unit cell of graphene for a convergence test, if 460 eV gives 1meV/atom convergency in total energy and 400 eV gives 3-4 meV/atom convergency in toal energy, will it be a problem to choose 400 eV for ENCUT of the larger Fe+graphene system to reduce calculation burden?

Thanks!

metosa · #6 Post by **metosa** » Tue Aug 24, 2010 3:30 pm

Dear limkr,

if 400 ev is highest ENMAX in your POTCAR file. Using default value should not cause any problem. But do not forget to mention your cutoff setting in your publication.

deepika · #7 Post by **deepika** » Tue Oct 02, 2012 12:15 pm

Hello Metosa,
I am also doing sc calculations on graphene system but the total free energy in my case is -18.507690, difference of around 2 eV from your results. I used the lattice constant = 2.459, ENCUT = 400 eV and getting some band gap of 0.2 eV at k point. I don't know Why and how I vanish this band gap. Kindly help me to resolve this problem.

Looking forward for your reply.
Thanks
Deepika