[Pw_forum] problem with the fft <-> G-vector mapping in an el-ph calculation

[Samuel Poncé]

Dear all,

I have a problem while performing el-ph calculation with PWSCF .

I did 2 small tests:
1) A calculation at q=Gamma and with only 1 k-pt (also gamma).
2) A calculation at q=Gamma and a 2x2x2 k-kpt grid (only 3 kpt in the IBZ
with gamma).

In both cases I first computed the dvscf and dyn matrices on a 2x2x2 kpt
grid (they are the same in both cases and I read them).

The electron-phonon matrix elements (GKK) that QE gives me should be the
same for k= gamma. In the second case the GKK file is bigger since it also
contain the two other k-points. I verified this for other arbitrary
q-points and it work. I got the exact same GKK file.

The only problem I have is with q=Gamma. In that case you have "IF
(nksq.GT.1) THEN" flag that treat the calculation of dvscf_q*psi_k
differently.

After some debugging I've come to realize that the only difference was in
DO ig = 1, npw
    aux1 (nls (igk (ig) ) ) = evc (ig, ibnd)
ENDDO

The evc are the same in both cases (except that in the second cases you
have some additional "0"s at the end that come from the fact that the evc
vector is allocate with the biggest npw for all k-pts.

The difference come from the nls mapping. The mapping is strongly different
in both cases and as a result the GKK files outputed by PWSCF are
different.

I must point out that I'm using an old version of PWSCF (4.0.5) but I need
this version to be able to run Yambo afterward. The code of the 5.0 seems
somewhat similar ...

Could you help me figure out why the mapping is different in both cases?

Thank you,

Best regards,

Samuel Poncé.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20130423/199c9633/attachment.html>