Clarification on HSE band structure workflow and training data in DeepTB

[Rizwan130]

Details

Hello,

I have a question regarding HSE band structure calculations in DeepTB. As we know, an HSE calculation usually involves two steps:

SCF (self-consistent field) calculation.

NSCF (non-self-consistent) calculation with zero-weighted k-points along the band path.

Since DeepTB uses the ABACUS input format, where the SCF calculation typically uses only a simple k-point mesh, I wanted to confirm whether it is reasonable to train the model using only the SCF data.

I tried training with data generated from the SCF step alone, but the resulting band structure is not accurate.

So my question is:

Should we perform both SCF and NSCF calculations (including the zero-weighted band path) before generating the DeepTB training data for HSE?

Or is there a specific recommended workflow for HSE-based band training in DeepTB?

Any guidance or examples on the correct setup for HSE training would be very helpful.

Thank you!