Question regarding skip_interpolation with use_symmetries_* and PDSYEVX failure in BerkeleyGW

[Uday Pratap SIngh Kushwah]

Dear BerkeleyGW developers,

I am encountering an issue when attempting to use skip_interpolation together with symmetry unfolding in a BSE absorption calculation and would appreciate any guidance.

System:

• BerkeleyGW 4.0

• Semiconductor/insulator system

• 4×4×4 k-point grid

• 17 valence and 17 conduction bands

• Diagonalization approach (diagonalization)

• Momentum operator (use_momentum)

My WFN_co is generated from a 4×4×4 grid but contains 8 irreducible k-points (with symmetry weights summing to 64), i.e. a symmetry-reduced grid.

Kernel calculation:

number_val_bands 17 number_cond_bands 17 screening_semiconductor use_symmetries_coarse_grid verbosity 2

Absorption input:

energy_resolution 0.15 diagonalization write_eigenvectors 1 number_val_bands_coarse 17 number_cond_bands_coarse 17 number_val_bands_fine 17 number_cond_bands_fine 17 screening_semiconductor skip_interpolation use_symmetries_coarse_grid use_symmetries_fine_grid use_symmetries_shifted_grid eqp_corrections use_momentum degeneracy_check_override

I use:

• WFN_fi linked/copied from WFN_co

• eqp.dat generated from eqp_co.dat

BerkeleyGW recognizes skip_interpolation, as the output contains:

Skipping interpolation

However, diagonalization subsequently fails:

BLACS processor grid: 4 x 4; BLOCKSIZE = 64 Number of idle processors: 0 Using diagonalization routine: PDSYEVX Beginning ScaLAPACK diagonalization. Size: 18496 Done ScaLAPACK diagonalization ERROR: Diagonalization with PDSYEVX failed: Got info=2

Interestingly, the same 4×4×4 calculation completes successfully when using the standard interpolation workflow:

use_symmetries_coarse_grid no_symmetries_fine_grid no_symmetries_shifted_grid eqp_co_corrections

with the real WFN_fi.

I also tested skip_interpolation + no_symmetries_coarse_grid, but that fails earlier with:

Found a point that was not calculated before

which seems understandable since the current WFN_co contains only irreducible k-points.

My question is:

Is skip_interpolation expected to work with use_symmetries_{coarse,fine,shifted}_grid and symmetry-unfolded irreducible WFN_co files, or is it intended only for full explicit k-point grids (e.g. generated with nosym=.true.)?

Additionally, is the PDSYEVX info=2 failure in this context a known issue or expected behavior?

Thank you very much for your help.

Best regards,
Uday Pratap Singh Kushwah