Yambo Community Forum

Dear Developers and Community,

I computed the G0W0 band structure of H-phase VSe2 using an 18×18×1 k-mesh.
First run (files/one; partially converged):
polband (BndsRnXp): 300
gbands (GbndRnge): 300
NGsBlkXp: 10 Ry
GTermKind = "none"
QP corrections: bands 17-25

Second run (files/second; converged):
polband: 500
gbands: 550
NGsBlkXp: 12 Ry
GTermKind = "BG"
QP corrections: bands 17-28
I also split the 37 IBZ q-points into five jobs (%QPkrange 1|37|17|28| → q001–008, q009–016, q017–024, q025–032, q033–037) and merged the results afterward.

The issue is a change in the top valence-band (VB) and lowest conduction-band (CB) topology.
In the first run, surprisingly, the VB maxima lies at K, matching earlier G0W0 results (DOI: 10.1038/srep32625 and VSe2 C2DB data base). However, on a more converged second run, the VBM unexpectedly shifts to Γ, and the band gap increases from 0.47 eV to 0.71 eV.

What could cause (1) the change in the VB/CB topology, specifically the shift of the VBM from K to Γ, and (2) the waviness in the second-run quasiparticle energies when moving from partially converged to fully converged G0W0 parameters?

I have attached the input gw.in files and the corresponding ndb.QP files from both the first and second runs for reference. I am also including the G0W0 band structures (left: first run, right: second run), and the PBE band structure.

Dear Zimmi,

it seems that VBM at gamma and kappa are "nearly" degenerate. This is already at PBE level, and this condition is maintained in your second GW run (higher parameters). Next, your GW bands are quite noisy due to the interpolations.
Probably the parameters of the first run are not yet converged.
My suggestion is to converge carefully the gap in a single k point, e.g. Gamma. This is faster, and you can look at the behavior of the gap wrt the relevant parameter. Next, using converged parameters, calculate for all the k points and interpolate.
I do not know if you are using ypp or yambo-py for the interpolation, i strongly suggest you to use yambopy.

Best,
Daniele

Dear Prof. Daniele,

Thank you very much for your time and for your helpful reply.
The suggestion worked perfectly.