K-poit selection

[Om Prakash Upadhyay]

 I Want to optimize Cu2O transition metal oxide semiconductor nanowire using ATK_VNL (SGGA RPBE).
I feel some what trouble in k point sampling because while ploting the graph of Total energy vs k-points along the growth of wire (z-axis),
the minimum energy curve shows different stable values of k-poits for different structures k=1*1*10 , 1*1*20 ,1*1*30 , 1*1*40  , 1*1*50.........
So, how the k point selection is done? & which value of k-point must be suitable for the optimization as well as whole calculation process for Cu2O nanowire? What is the differece on choosins lesser & higher values of K-poit selection? Please clearify this..................
                          Thanks in Advance...................................

[Anders Blom]

Higher values is always more accurate, except in certain cases (like graphene) when it's important to sample special points in the Brillouin zone.
However, the total energy is not variational with respect to k-points so it will typically oscillate a bit when you increase the number of k-points. What  you want to find is a reasonable number that give a value not too far from the asymptot of infinite number of points.

To assist you, it's always a lot easier if you e.g. include a plot of the total energy, so we have a better picture of what the issue is.

[Om Prakash Upadhyay]

While calculating band gap from band structure & DOS taking k=1*1*30 & k=1*1*40, the result is different (half metal/semiconductor) for same structure. What is the reason behind it? Please make me confusion free.......
                                 thanks in Advance...........................

[Anders Blom]

Show input file and results, else we can only guess

[Om Prakash Upadhyay]

I have attached the band structure & DOS files with k=1*1*30 & k=1*1*40 in which the band structure shows Cu2O as half metal for k=1*1*30 & semiconductor for k=1*1*40. I have used the respective k-point sampling for given structure with ATK-DFT , SGGA-RPBE , Cu-DZDP , O-DZP. Why is this being so??? What is the solution for this problem???
             Waiting for reply.................
                                     Thnks............................

[Om Prakash Upadhyay]

Contd.......................

[Om Prakash Upadhyay]

........................

[Anders Blom]

It would help if you submit the Python input. You are also using an old version of ATK, I see.

[Om Prakash Upadhyay]

....................................
Now please solve my problem ....................

[Anders Blom]

A couple of thoughts. You don't have to evaluate the DOS at the same k-point sampling as used in the calculation. If you calculate the 30-kpoint case with 40 DOS k-points, do you see a difference? That is, it might not be a matter of k-point sampling for the calculation (30 or 40 should both be plenty enough), but that of the DOS evaluation.

For a spin-polarized calculation (and it really looks like there is a strong polarization) you also should take care that both calculations converge to the same spin state. You don't control the initial spin, so maybe they converge to very different states?

Finally, just to be safe, I would also make the unit cell a bit larger in X and Y to avoid interactions with the periodic images.

Not saying that any of this will magically solve the puzzle, but it's some points you can investigate.

[Om Prakash Upadhyay]

I have taken k-point for optimization as well as DOS same not different.
also, please help me through tuturial that haw I  go forward by taking initial state....
Should I have to take initial state initially from optimization process or after optimization while calculating band structure and DOS???
                 Thank u for reply................

[Jess Wellendorff]

1) Post-SCF analyses such as DOS are always done keeping the converged ground state Hamiltonian fixed. You can therefore increase the k-point sampling significantly without the DOS analysis to become very expensive as compared to the SCF. For example, I usually double the k-point density for DOS analysis, e.g. 3x3x1 k-points for SCF and 6x6x1 k-points for DOS analysis.
2) The geometry optimization gives you the minimal-energy (equilibrium) geometry of the structure. You should therefore do bandstructure and DOS analysis on the ground state that is saved after geometry optimization has finished.

[Daniele Stradi]

Hi,

may I ask you what happens if you try to perform the same DOS vs. k-points analysis but for the non-spin-polarized system (even without relaxing)?

Your results may indicate that the system has a complex energy landscape, with several different magnetic configuration very close in energy. The energy differences might be so tiny that their relative order changes if the k-point sampling is changed.

Daniele

[puspa raj adhikari]

yes, absolutely mr.  Daniele Stradi. i also try BS Dos and total energy vs. changing k points with remainng all parameter same.. it shows variation in result. nature of material also changes as band structure...

[Petr Khomyakov]

The Cu2O nanowire is modeled with too little vacuum separation layer between the image cells. Before doing the k-point convergence test, I would do the convergence test for the separation distance  (it is now of 4.7 Angs only) as also suggested by Anders. Assuming a given number of k-points (e.g., 1x1x40), I would set the separation distance to 10 and 15 Angs to see if it makes any difference to the results.

As proposed by Daniele, this kind of tests can first be done for a non-spin-polarized system.