OMEGAMAX, OMEGATL and CSHIFT

 OMEGAMAX = real number  (maximum frequency for dense part of frequency grid)
 OMEGATL  = real number  (maximum frequency for coarse part of frequency grid)
 CSHIFT    = complex shift

defaults: OMEGAMAX = outermost node in dielectric function $\epsilon(\omega)$/1.3
OMEGAMAX = energy difference between conduction and valence band minimum/1.3
OMEGATL = 10 $\times$ outermost node in dielectric function $\epsilon$
CSHIFT = OMEGAMAX *1.3 / max(NOMEGA,40)

For the frequency grid along the real and imaginary axis sophisticated schemes are used that are based on simple model functions for the macroscopic dielectric function. The grid spacing is dense up to roughly 1.3 OMEGAMAX and becomes coarser for larger frequencies. The default value for OMEGAMAX is either determined by the outermost node in the dielectric function (corresponding to a singularity in the inverse of the dielectric function) or the energy difference between the valence band minimum and the conduction band minimum. The larger of these two values is used. Except for pseudopotentials with deep lying core states, OMEGAMAX is usually determined by the node in the dielectric function.

The defaults have been carefully tested, and it is recommended to leave them unmodified whenever possible. The grid should be solely controlled by NOMEGA (see Sec. 6.66.3). The only other value that can be modified is the complex shift CSHIFT. In principle, CSHIFT should not be chosen independently of NOMEGA and OMEGAMAX: e.g. for less dense grids (smaller NOMEGA) the shift must be accordingly increased. The default for CSHIFT has been chosen such that the calculations are converged to 10 meV with respect to NOMEGA: i.e. if CSHIFT is kept constant and NOMEGA is increased, the QP shifts should not change by more than 10 meV; at least for LSPECTRAL=.TRUE. and the considered test materials this was the case. For LSPECTRAL=.FALSE. this does not apply, and it is recommended to set CSHIFT manually and to perform careful convergence tests in this case.

For LSPECTRAL=.TRUE. independent convergence tests with respect to NOMEGA and CSHIFT are usually not required, and it should suffice to control the technical parameters via the single parameter NOMEGA. Also note that too large values for NOMEGA in combination with coarse k-point grids can cause a decrease in precision (see Sec. 6.66.3).