ENCUT and KPOINTS convergence test-large systems

[SKM]

hi Admin,
after sorting the issue with some error for a system at forum/viewtopic.php?f=4&t=5354&start=15#p19749
I Am raising the ENCUT convergence query here for the same system. I did run the the above model for a static calculation, now with VASP 6, and started from 400eV ENCUT to 700eV for convergence test. The graph ENCUT Vs Total Energy is not converging, rather the Energy is increasing/fluctuating. The graph picture is attached. Kindly suggest, a way out. Or do i need to test for further with higher ENCUTs?

Regards.

[martin.schlipf]

What do you actually want to calculate? You should converge the quantity that you want to investigate, which may or may not be the total energy. For example, band gaps or energy differences converge much faster than the total energy.

If I remember correctly, you had Cs in your system. When upgrading to Vasp.6 did you also use the new Cs PAW, which should be more reliable?

[SKM]

Thanks for the reply.

1. No i did not use any new Cs PAW. BY the way, you mean there is a new Pseudo-potential file in VASP site?

2. for the same system finally i want to do Band gaps in PBE and HSE functionals. and effective mass of hole and electron from band structure.

3. in such requirement as above, what should be the convergence test needed and how to do?

4. Do i need to select the desired ISIF tag (wish to use ISIF=3) during convergence test?

Thanks and regards

[martin.schlipf]

1. No i did not use any new Cs PAW. BY the way, you mean there is a new Pseudo-potential file in VASP site?

The principal investigator of your license should be able to access all current PAWs from the user portal where they also administrate who are their active users.

2. for the same system finally i want to do Band gaps in PBE and HSE functionals. and effective mass of hole and electron from band structure.
3. in such requirement as above, what should be the convergence test needed and how to do?

You should converge quantities that are related to what you want to study. In your particular case, look at band gaps and band dispersion. The trick is to not converge all parameters at the same time. You set all to loose settings except for one which you want to converge. Then you only run a few calculation with the fully converged setting checking if the experience you gained from the easier system translates to the more complicated one.

4. Do i need to select the desired ISIF tag (wish to use ISIF=3) during convergence test?

Unless you start very far from the relaxed structure, the convergence of the band gap should probably translate to the final structure. Keep in mind that relaxing is also determining an observable (the structural information) which typically has different convergence properties than band gaps. For relaxations you may start with smaller cutoffs initially and then increase the parameters once you are close to the minimum.

[SKM]

hi

The principal investigator of your license should be able to access all current PAWs from the user portal where they also administrate who are their active users.

I have checked it again, i am using latest Cs_sv POTCAR file only. So it should be fine with Cs.

In your particular case, look at band gaps and band dispersion. The trick is to not converge all parameters at the same time. You set all to loose settings except for one which you want to converge. Then you only run a few calculation with the fully converged setting checking if the experience you gained from the easier system translates to the more complicated one.

Does this mean we need to run, say PBE or HSE band structures for different ENCUTs and KPOINTS? or how do i actually do it?
a. i did not understand the point "set all to loose settings except for one which you want to converge". because what parameters that corresponds to loose settings if bandgap is my converging parameter.
b. also kindly elaborate on what you said "run few calulations with fully converged setting". I understand to do a full run, once the bandgap is converged, but do you mean we need an intermittent check?

Unless you start very far from the relaxed structure,....

i have initially relaxed the primitive cell, and this Supercell is created from that relaxed unit cell with one element doping. So i think i can expect its nearly a relaxed structure. So, ISIF=2 should be OK then.


Kindly help me in understanding the above.
Regards

[martin.schlipf]

You should probably discuss your research protocol a bit more in depth with your PI, I can only give suggestions here.

Typically you have more than one convergence parameter (e.g. ENCUT and k-point grid). You could now start increasing both of them at the same time until your result no longer changes. However, often it is more efficient to keep--let's say--the ENCUT fixed and increase the mesh and vice versa. Only when you know the more or less converged settings of each parameter individually you run full calculations.

In your particular example of band structures, you might for example never actually construct a band structure which is a bit involved with hybrid functionals and instead look at the band-energy difference between two k-points close to the band minimum to approximate the effective mass.