Large total energy difference for Sc_sv_GW vs. Sc_sv

[jl_wolf]

Hello,
when calculating a primitive rocksalt ScN cell, I observed a substantial difference in the total energy, when using the Sc_sv (07Sep2000) and Sc_sv_GW (05Dec2013) PAW pseudopotentials for Sc.

All calculation settings were kept identical (INCAR/POSCAR below), using a dense 15×15×15 k-point mesh and the same N potential (N 08Apr2002).

Results:

Sc_sv_GW: TOTEN = -20.90995078 eV
Sc_sv: TOTEN = -18.41329565 eV

The ~2.5 eV difference in absolute total energy seems unusually large compared to what I observed for other elements when switching between standard and GW PAW pseudopotentials.

Is this expected due to differences in the PAW construction/reference energies, or could this indicate an issue with my calculation?

Thank you.

INCAR:
ENCUT= 600.0
EDIFF= 1e-6
ISMEAR= -5
IBRION= -1
NSW= 0
ISTART= 0
ICHARG= 1
ALGO= Normal
LREAL= False
LASPH= True
PREC= Accurate
KPAR= 2
NCORE= 1

POSCAR:
N Sc
1.0000000000000000
-0.0000000354204632 2.2593157490777167 2.2593157332309279
2.2593157391680339 -0.0000000240437520 2.2593157218542208
2.2593157527175523 2.2593157512505253 -0.0000000375932775
N Sc
1 1
Cartesian
2.2593157624633875 -0.0000000155167459 -0.0000000312831107
-0.0000000029260828 0.0000000006475571 0.0000000066773011

[martin.schlipf]

Only relative energies are meaningful in VASP. Usually, the energies are somewhat comparable but that is only the case if the atomic energy (single atom in the cell) is the same. In your case, you will see a similar energy difference for the bare atoms. I got 2.41 eV or 2.44 eV for the atom depending on whether I use a nonmagnetic or collinear setup. For the bulk I get 2.49 eV. The remaining difference of 0.05 to 0.08 eV is still on the large side but I did not investigate carefully the convergence of these energies with the system size.