Updating wavefunctions in self-consistent GW

[irene]

Dear all,

Can someone explain how to update the wavefunctions in self-consistent GW with vasp 5.2?
Because I find contradictory some info at the manual:

- First, page 107 Sec. 6.66.1 says that "a full update of the wavefunctions can be performed by specifying ALGO = scGW"
- then the example on page 111 sets as recipe for selfconsistent GW calculations the variables: ALGO=scGW ; NELM=4 ; LWAVE=.TRUE.
- but finally, on page 110 Sec. 6.66.10 (about LWAVE=.TRUE.) it says "Note that only the energies are updated, whereas wavefunctions are kept constant on the DFT level."

Does this mean that LWAVE should be set to FALSE when using ALGO=scGW ? or did I misunderstand something?
Thank you.

[irene]

Concerning this self-consistent GW, I have also a question about NELM, since I find a weird behavior and NELM should no have such an influence...

In principle, if NELM controls only the number of iterations performed (as claimed by the manual), it should not have effect on the calculation in the sense that: the 4 first iterations should be the same if I set NELM to 4 or to 10. Am I right? But this is not the case. Since if I set NELM to 10, the gap is converged at the 4th iteration, but if I set NELM to 4 (with identical calculation) I get a wrong result, not converged at all, and with nonsense occupations changing apparently random from one cycle to another.

So my question is, does NELM do something else (non reported) apart from controlling the number of cycles?