PBE SOC calculations using HGH pseudopotentials

Running Abinit Ground state
1

Dear all,

I am calculating spin–orbit (SO) splitting for bulk ZnS in both wurtzite (WUR) and zinc blende (ZB) structures using HGH pseudopotentials.

The reference all-electron SO splittings are:

  • WUR: 0.032 eV
  • ZB: 0.059 eV

However, in ABINIT I obtain much smaller values:

  • WUR: 0.016 eV
  • ZB: 0.023 eV

This is unexpected, because in several other cases the same HGH pseudopotentials reproduce SO splittings very well compared to reference values.

My non-relativistic calculations look fine (correct total energies and band degeneracies). Also, when I use fully relativistic norm-conserving pseudopotentials from the PseudoDojo library, the SO splittings agree well with reference data.

I also tried setting all SOC-related parameters for S to zero, and in that case the splitting becomes zero which is unexpected (splitting due to Zn should be visible), so the setup seems internally inconsistent.

I am not sure whether this is an issue with my input setup or a possible ABINIT-related problem. I am attaching the input file for ZnS (WUR) and the pseudopotential files for reference.

Any help in identifying the issue would be greatly appreciated.

fyi: The states I am considering are the valence states at Gamma point, typically states 33 to 36. The ABINIT version is 9.6.2.

zns.abi (1.2 KB)

Goedecker pseudopotential for Zn
30 12 070301 zatom,zion,pspdat
10 11 2 0 2001 0 pspcod,pspxc,lmax,lloc,mmax,r2well
0.51000000 0 rloc nloc
3 nnonloc
0.40031644 3 11.53004133 -8.79189815 3.14508644 rs ns hs11 hs12 hs13
16.46577518 -8.12057827 hs22 hs23
6.44550918 hs33
0.54318233 2 2.59719512 -0.59426275 rp np hp11 hp12
0.70314117 hp22
0.10017431 -0.10693277 kp11 kp12
0.12652456 kp22
0.25095885 1 -14.46695795 rd nd hd11
0.01272732 kd11

Goedecker pseudopotential for S
16 6 070301 zatom,zion,pspdat
10 11 1 0 2001 0 pspcod,pspxc,lmax,lloc,mmax,r2well
0.42000000 1 -5.98626038 rloc nloc c1
2 nnonloc
0.36482035 2 13.14354448 -4.24183045 rs ns hs11 hs12
5.47617957 hs22
0.40948048 2 3.70089057 0.00000000 rp np hp11 hp12
0.00000000 hp22
0.03492359 -0.02660790 kp11 kp12
0.03148290 kp22

Eigenvalues ( eV ) for nkpt= 345 k points:
kpt# 1, nband= 50, wtk= 1.00000, kpt= 0.0000 0.0000 0.0000 (reduced coord)
-9.24347 -9.24347 -8.30687 -8.30687 -2.60265 -2.60265 -2.58224 -2.58224
-2.51383 -2.51383 -2.35172 -2.35172 -2.28707 -2.28707 -2.09115 -2.09115
-1.99815 -1.99815 -1.99735 -1.99735 -1.87305 -1.87305 -1.72559 -1.72559
-1.44924 -1.44924 3.00124 3.00124 3.01286 3.01286 3.81179 3.81179
3.91317 3.91317 3.92905 3.92905 6.03715 6.03715 6.99645 6.99645
10.11351 10.11351 10.22727 10.22727 10.47017 10.47017 11.53431 11.53431
11.61133 11.61133

The Fermi level is at 3.93 eV.

Thanks and Regards,
Bibek

2

Hello Bibek,

this is indeed a bit surprising, especially if the oncvpsp potentials work fine. Are you sure the structure and k convergence are exactly the same? The gap you find is tiny, so it will be very sensitive to everything.

The disappearance of the splittings when you set the S corrections to 0 may just mean that the specific bands you are pinpointing depend mainly on S. Have you checked the full band structure, and the semi core states? Also check the output - it may be that the spin orbit or nspinor is being turned off if both pseudos do not have SOC corrections. Then the Zn corrections are ignored as well. Try the converse with 0 Zn coefficients, as a test.

In passing, why do you want the HGH potentials, if you have alternatives?

Finally, can you try a more recent version of abinit, just in case?

best

MV

3

Dear MV,

thank you very much for your reply. I have already used converged parameters for the both the cases. Also, I have tried using the recent version of ABINIT and the issue persists. I will test by putting 0 Zn coefficients.

I was comparing the HGH pseudopotential with ONCVPSP potentials for molecules and bulk materials. This problem of HGH pseudopotential only appears in ABINIT.

Thanks,
Bibek

4

Dear MV,

I tried the tests you suggested, but unfortunately, nothing conclusive came out of them. The same problem appears for other bulk materials e.g. GaN [ZB]. Sometimes, this problem is associated with the state we are looking at, e.g., for GaN, the splitting of states above the Fermi energy is captured correctly. Still, for the same material, the splitting of valence occupied states is wrong. Do you think this is an implementation issue? These issues disappear while using ONCVPSP PPs from the pseudo-dojo library.

Thanks,
Bibek