I want to study a material and first I need to optimize the structure with vasp. I know I need to relax the volume and I know I should do convergence analysis by plotting total energy vs k-points. My question is should I choose some reasonable k-points and do my volume relaxation calculations, and with the relaxed structure perform k-point convergence analysis? Is that an accurate order or should k-point convergence be done first and then volume relaxation calculations are done at the converged k-points?
ANy contribution would be great.
Dear Robert Appleton ,
Volume relaxation should come first, then k-point convergence. A safe starting point for insulators is 8 k-points per 4 Å and for metals is 12 k-points per 4 Å. Then, after you relax the volume, do the k-point convergence analysis. If you find that more than 8/12 k-points per 4 Å were needed to obtain total energy convergence (
Robert Wexler
Thanks that confirms my understanding. This was a question aimed at a more specific problem I am having. I posted a new question so as to be easier to follow.
I will put the link below if you wouldn't mind taking a look and seeing if you have insight.
https://www.researchgate.net/post/Volume_Relaxation_of_InAs_in_VASP_How_come_I_keep_getting_incorrect_lattice_parameter
Actually, I think you can’t really converge one without the other. As a matter of fact, you also need to bring into the picture the kinetic energy cutoff (ENCUT in vasp).
Specially, if your system has internal degrees of freedom where atomic relaxation is part of the volume optimization.
What you could do, as Rober Wexler, suggested is to set a fixed k-grid, I like to see something in the 2000-4000 points/atom, and converge the cutoff energy with respect to some real space quantity (not the energy). And then converge the k-grid.
In an ideal world, you would like to use very large cutoff and k-grids to ensure the accuracy of your predictions; however, the real world you want to select these quantities so that you balance accuracy and efficiency.
Just to finish this, I strongly suggest you take a look at “Designing meaningful density functional theory calculations in materials science—a primer” by Mattsson et. al. (http://dx.doi.org/10.1088/0965-0393/13/1/R01)
You can use a large k-grids in the volume relaxation, then plot total energy vs. k-points.
David Guzman
This explains why I am having trouble converging the ENCUT parameter. When I plot vs total energy I get plot that always increases even at extremely large ENCUT values as seen in the plot below.
You say to converge the cutoff energy with respect to a real space quantity, would this be something like Pullay stress?
David Guzman here is a plot of External Pressure vs Cutoff Energy. This looks more like convergence?
Robert Appleton ,
David Guzman is right that, in principle, both the equilibrium lattice constants and atomic positions depend on the energy cutoff.
It is worth noting that the total energy can converge rather slowly with respect to the energy cutoff. A better metric of convergence is the total energy per atom for which a value of 1 meV/atom is conservative. You can't really expect to do any better than this because there's an error associated with pseudopotentials on the order of 1 meV/atom, maybe even slightly larger.
The pressure converges faster than the total energy because it is related to the derivative of the total energy, i.e. energy differences. It is often the case that energy differences converge much more quickly than the total energy itself. Since phonon calculations involve energy differences, you should be able to get away with a much smaller energy cutoff than 1800 eV (P.S. that's a really large energy cutoff!).
David Guzman Robert Wexler
One of the materials I am studying is giving me weird results when doing kpoint convergence. I have been finding total energy convergence around 9x9x9 however for Ba3Zr2S7 in particular I am getting the following plot that shows the total energy fails to converge even at large kmesh (21x21x21).
I am using the following INCAR
ENCUT = 450 (1.3*ENMAX from PP)
ISTART = 0
ISMEAR = 0
SIGMA = 0.1
ICHARG = 2
Using PAW_PBE psuedopotentials.
I use a script to run vasp with increasing kmesh starting from 2 - 8 only evens and then 9 - 25 only odds to produce the plot.
Any suggestions on how to converge total energy with out extremely large kmesh?
David Guzman Robert Wexler
sorry if i mind you. I got a VASP work for calculation of change in energy of Bi2O3 with nitrogen ( nitrogen reduction reaction). I do not know when my geometric slab is optimization. i mean i set a EDIFFG = -0.01 for my INCAR and it works ! Does that means my structure is optimized. i attached some my result
Hi Nhut Nguyen ,
I took a look at your output and unfortunately your calculation crashed after the first SCF step completed with the following message: "Error EDDDAV: Call to ZHEGV failed." This is an issue with the diagonalization and you may want to try using a different algorithm (ALGO), e.g., "Normal" or "All", where the latter is the safest option. Good luck!
-Rob
- Badges
- Science method
- Similar topics
- Methodology
- Crystallography
- X-ray Diffraction