I was trying to obtain electronic bandstructure of InSb using Quantum Espresso (QE). At first, I used PBE pseudopotentials (with/without spin-orbit-coupling [SoC] and ultrasoft/PAW) available on the QuantumEspresso website. But in any case, I didn’t get an accurate band structure, there is no bandgap and also with SoC heavy-hole (HH) and light hole (LH) bands are separated on GAMMA high-symmetry points (InSb-PBE-PAW-SoC-conf1.png ). All those pseudopotentials (PP) of In and Sb available on the QE have taken all 10 electrons of 4d orbital as their valence electrons (for In: 4d10 5s2 5p1, for Sb: 4d10 5s2 5p3) along with 5s and 5p and rests as the core electrons.

But when I generated PBE PPs by myself without taking 4d electrons as valence electrons (for In: 5s2 5p1, for Sb: 5s2 5p3), those PPs gave me quite an accurate bandstructure (PBE-PAW-SOC-conf2.png).

My question is, why the inclusion of 4d electrons in the valence cell in the PP of In and Sb (which are the default and believed to be the most optimized configuration for PP) is not giving accurate bandstructure of InSb?

Thank you in advance.