[Pw_forum] xspectra calculation

[Matteo Calandra]

Dear Jiayu

 >And i think even for dipolar part, the direction of k vector should be 
 >important in some cases, such as the SiO2 in the example.

No, this is incorrect. The dipolar part cross section does not depend
on k.

 >And if we
 >want to compare the results with some experiments, we should cut the 
 >contribution of occupied states, that is, cut_occ_states = .true. 
 >should be used. But we should shift the position by hand in order to 
 >compare the results, cause i think it is very difficult to get the 
 >accurate position in DFT, or even in atomic physics, it is a difficult 
 >problem to determine the position of spectra. The "ef_r" should be the 
 >fermi energy in scf calculation with core-hole, i think.

On this second point you are right in the method but not on the reason
why one should cut at somewhat lower values with respect to e_f.
The problems that you can face are the following:

1) You have a system with a gap even in the presence of a core-hole
. In this case if you cut exactly at the bottom of the conduction band 
(empty states) then the cut_occ_states procecure can miss this peak.
So it is safer to cut at somewhat lower energy than the bottom of the
conduction band.

2) You have a system without gap in the presence of a core-hole
(either because it was a metal even without core-hole or because it
was a small gap semiconductor that becomes a metal when the core-hole
is inserted). In this case identification of where to cut is ill 
defined. In this case you should keep in mind that probably core hole
effects are not correctly described.
Of course, if DFT underestimates the gap you can easily end up in a 
metallic solution.

All the best, M.


Jiayu


-- 
* * * *
Matteo Calandra, Charge de Recherche (CR1)
Institut de Minéralogie et de Physique des Milieux Condensés de Paris
Université Pierre et Marie Curie, tour 16, case 115
4 Place Jussieu, 75252 Paris Cedex 05 France
Tel: +33-1-44 27 52 16       Fax: +33-1-44 27 37 85
http://www.impmc.jussieu.fr/~calandra