Interpolation for the CB using automated projection method.

[LIU-Binbin]

Hi,

I am encountering an issue with the quality of the Wannier interpolation for the Conduction and Valence Band when using the automated projection method in aiida-wannier90-workflows.

While the VB interpolation looks reasonably good, the Conduction Band shows significant deviation from the reference Quantum ESPRESSO (PW) band structure. Attached band structures and inputs for 3C-SiC.

Upon investigating the inputs generated by the workflow, I noticed that the number of bands calculated in the NSCF step (nbnd) is exactly equal to the number of Wannier functions (num_wann). Both are 16 in my case, which might be the cause for the not-so-good CB interpolation.

For CB entangled with higher energy states, accurate interpolation usually requires an outer window with nbnd > num_wann to allow for proper disentanglement. For metal it is true, but is it also implemented for semi-conductors and insulators?

Thanks!

auto_projections = .true.
bands_plot = .true.
conv_tol =   2.0000000000d-07
conv_window = 3
dis_conv_tol =   2.0000000000d-07
dis_froz_max =   1.2850850262d+01
dis_num_iter = 4000
dis_proj_max =   9.5000000000d-01
dis_proj_min =   1.0000000000d-02
fermi_energy =   1.0415500000d+01
mp_grid = 10, 10, 10
num_bands = 16
num_cg_steps = 200
num_iter = 4000
num_wann = 8
write_bvec = .true.