Degeneracy problem in response calculation

Running Abinit Response calculations
1

Hello all,

I am trying to learn the calculation of dielectric constant via the DFPT tutorial. After replacing the AlAs used in part 5 of the tutorial with SiC, I was able to complete the calculations and get very accurate results.

After that, I tried to scale up and calculate the effect of Al-Si antisite in SiC. I used a 2×2×2 cubic SiC supercell as a base and was able to complete the trial calculation when replacing two of the Si atoms with Al atoms. However, if only one is replaced, the calculation stopped at dataset 2, and the following error occurs:

--- !ERROR
src_file: m_occ.F90
src_line: 1692
mpi_rank: 7
message: |
    In a non-metallic case (occopt<3), for a RF calculation,
    if the eigenvalues are degenerate, the occupation numbers must also be degenerate.
    However, the following pair of states gave :
    k -state, band number 127, occ=    2.000000E+00eigenvalue=    2.630046E-01,
     kq-state, band number 128, occ=    1.000000E+00, eigenvalue=    2.630046E-01.
    Action: change occopt, consistently, in GS and RF calculations.
...

The input and output files are as below:
kpt-2x2x2.abi (5.9 KB)
kpt-2x2x2.abo (75.5 KB)

I tried to fix it with cellcharge 1, or occopt 0 and manually set occ and wtk, so that the band with the original occupation number of 1 (kq-state, band number 128) becomes 0, but the error remains.

I searched the old ABINIT forum and found several threads with similar problems, some replies said to switch to a different XC, but neither LDA nor PBE-GGA worked for me. So I would like to ask, how to correctly calculate the response in this kind of situation (1 electron less than normal)?

2

You have made the system metallic and need to set occopt >= 3, along with tsmear > 0. Both need to be converged, see the basic tutorials. I would recommend occopt 7 with tsmear 0.001 to start with.

Note there are some DFPT functionalities which only work for insulators, and the code will complain and/or stop if you have an effective metal.

Charging the cell is another matter, don’t know if you want to do that off the bat, and DFPT on top will be complicated : you need to add some Coulomb correction term, and its perturbation, which is not implemented afaik. Pertinent for the realistic defect, but complicated in PBC.

3

It seems that yes, I misunderstood the meaning of degenerate occ and thought that the missing electrons needed to be filled by an extra cellcharge. For my case, occopt 7, tsmear 0.01 and a denser k-point grid give much more reasonable results.

So, is there any way to determine whether this system behaves metallically or semiconductively in DFT? It still appears to have a band gap.

4

plot the DOS and band structure to see whether there is a gap. For the GS k-grid the code should give you the Fermi level, and often the top of valence band + lowest conduction band (could be that this only appears for semiconducting occopt…)