My Community

Hi

I was running a bandstructure calculation for a supercell. It ran quite long for sometime and it did not converged. Could you please cite out the mistakes.

I have now using ALGO=VeryFast but still this is diverging

N E dE d eps ncg rms rms(c)
RMM: 1 0.755692201211E+05 0.75569E+05 -0.28023E+04 35280 0.712E+02
RMM: 2 0.749983571465E+05 -0.57086E+03 -0.56792E+03 35280 0.146E+02
RMM: 3 0.748737363287E+05 -0.12462E+03 -0.13497E+03 35280 0.493E+01
RMM: 4 0.748437415908E+05 -0.29995E+02 -0.28831E+02 35280 0.281E+01
RMM: 5 0.748336089167E+05 -0.10133E+02 -0.76531E+01 35280 0.133E+01
RMM: 6 0.748302547878E+05 -0.33541E+01 -0.25171E+01 35280 0.743E+00
RMM: 7 0.748290202580E+05 -0.12345E+01 -0.99733E+00 35280 0.433E+00
RMM: 8 0.748284823691E+05 -0.53789E+00 -0.45745E+00 35280 0.280E+00
RMM: 9 0.748280701570E+05 -0.41221E+00 -0.38130E+00 95946 0.180E+00
RMM: 10 0.748279777064E+05 -0.92451E-01 -0.54932E-01 94707 0.572E-01
RMM: 11 0.748279689968E+05 -0.87095E-02 -0.63928E-02 95502 0.243E-01
RMM: 12 0.748279681575E+05 -0.83936E-03 -0.79527E-03 95962 0.864E-02 0.317E+01
RMM: 13 0.747997162836E+05 -0.28252E+02 -0.44651E+02141120 0.407E+01 0.219E+01
RMM: 14 0.748374822742E+05 0.37766E+02 -0.39271E+01141120 0.371E+01 0.109E+01
RMM: 15 0.748326056387E+05 -0.48766E+01 -0.25804E+01137502 0.164E+01 0.115E+01
RMM: 16 0.748469644058E+05 0.14359E+02 -0.14767E+01131014 0.983E+00 0.844E+00
RMM: 17 0.748504695999E+05 0.35052E+01 -0.34549E+00124807 0.404E+00 0.529E+00
RMM: 18 0.748535536487E+05 0.30840E+01 -0.30141E+00124535 0.333E+00 0.235E+00
RMM: 19 0.748490446424E+05 -0.45090E+01 -0.23582E+00121784 0.259E+00 0.101E+00

Hi,

What you did is incorrect. The k-mesĥ in your KPOINTS file is suitable only for a band-structure calculation (see https://www.vasp.at/wiki/index.php/KPOI ... lculations) and not for a SCF calculation. For a SCF calculation, the k-mesh has to be uniform.

Furthermore, the way to calculate the band structure with MGGA functionals (like MBJ that you are using) is more complicated than for a GGA functional, and actually similar to hybrid functionals. There are essentially two ways to calculate the band structure with MGGA/hybrid functionals, as explained here: https://www.vasp.at/wiki/index.php/Band ... unctionals.

Besides, this topic could be useful for you:
https://www.vasp.at/forum/viewtopic.php?f=4&t=18488

Thank you for the clarification.

I want to calculate the band structure with mbJ+SOC accurately of a structure (relaxed with GGA=PS) and so, I have read the links you have provided along with several other articles. I concluded with the following steps.

Initial: We have a relaxed structure (already relaxed with GGA=PS)

1. Run PBE-SCF calculation with SOC. (Regular k-mesh)
2. Run mbJ-SCF (mbj tags included) calculation with SOC (reading WAVECHAR from step 1), also including GGA=PS. (Regular k-mesh)
3. Run mbJ-band (mbj tags included) calculation with SOC (reading WAVECHAR from step 2), also including GGA=PS. [KPOINTS=Irreducible k points of the regular mesh (IBZKPT of step 1) + zero weighted k-points (from OUTCAR of step 2)].

Are these steps correct?

No, this is incorrect. The steps should be the following:

1. Choose a geometry. This can be the one obtained from a GGA like PBE (GGA=PE in INCAR) or PBEsol (GGA=PS in INCAR). Another choice could be an experimental geometry if available.

2. Run the MBJ calculation (METAGGA=MBJ in INCAR) with a KPOINTS file that contains two sets of k-points: (a) regular k-points (let's say the same as used in step 1 if you used a GGA) and (b) zero weighted k-points (the ones in the KPOINTS from your 1st post). Of course, for this calculation you should not have GGA=PS in INCAR (you are doing a MBJ calculation and not a PBEsol calculation! This does not make sense to have METAGGA=MBJ and GGA=PS at the same time!).

Besides, since the calculations with SOC are lengthy, it is recommended to check the convergence of the parameters (number of k-points, ENCUT, etc.) without SOC, and then use the parameters for the SOC calculation.

In general, it is a good idea to take time to understand things.

Thanks a lot for the enlightenment. I will proceed with your suggestions.

Thank you for the help.

The calculated band gap is underestimated from the experimental one. How can I choose the value of 'c' parameter to
get an accurate band gap?
Is there is any method to get an optimized 'c' parameter?

Usually, the band gap increases with c. So, you can try a larger value like c=1.3 ("CMBJ=1.3" in INCAR).

In general, a more elegant way to use MBJ is without specifying CMBJ in INCAR (i.e., c is not fixed). That means that c is calculated with Eq. (3) in http://dx.doi.org/10.1103/PhysRevLett.102.226401. But, that does not necessarily mean a better agreement with experiment.

Besides, there is currently no "exact" exchange-correlation functional in DFT that gives the correct answer. Beyond-DFT methods like self-consistent GW with vertex correction are more accurate than DFT methods, but can be tremendously expensive.

Thanks a lot again for your help.

Hi again

Calculation of mBJ with SOC yields appreciable band gaps for a pristine unit cell. However, when I calculate the band gap of
supercell containing 32 atoms, it shows an error as

"while reading WAVECAR, plane wave coefficients changed 41582 || 20791"

I have tried various workarounds as given in https://www.vasp.at/forum/viewtopic.php?t=8154. But it is not fixed. Please guide me.

Did you start the supercell calculation with a WAVECAR file already present in the directory? If yes, how was this WAVECAR file generated? In general, you can not start a calculation with a WAVECAR file that was generated from another structure.

Yes, I used a WAVECHAR from the same directory.

I ran a PBE-SCF calculation with KPOINTS (inclusive of ZERO WEIGHTED KPOINTS). The generated WAVECHAR was used to run another mBJ-SCF calculation with the same KPOINTS as previous one. In mBJ-SCF, I included SOC which was not there in the previous one. The structure is same in the two calculation.

I also tried to fix ISYM=1 in the latter calculation but same error.

The WAVECAR files generated from calculations with and without SOC are not compatible. Either you include SOC also in PBE or you delete the WAVECAR file before starting MBJ. The first solution is probably better.