The ABINIT Users Mailing List ( CLOSED )

[Marc Torrent <marc.torrent@cea.fr>]

OK,

1) I think your energy for the second s- state is too close to the first 
one; you should try E=4 or 5 Ry
But that's not your problem !

2) If you look carefully at the log derivatives for l=1 (logderiv.1), 
you can see a little divergence around E=-3Ry... You should not have it 
! It could be the sign of a ghost state (not necessarily, but probably). 
This is generally due to a "too deep" pseudopotential. This is something 
known for high Z materials: the norm-conserving Trouillier-Martins 
scheme gives hard pseudopotentials for these materials. A solution is to 
change the pseudization scheme for the pseudopotential, in order too 
make it softer...
* You can try the "ultrasoft" scheme: the pseudopotential is generated 
with the same scheme as for TM but the norm-conserving constraint is 
suppressed.
* You can try the "bessel" scheme: no norm conserving constraint; a 
simple pseudization by a Bessel function; but, in the later case you 
have to decease the matching radius (r_vloc) in order to keep the 
pseudopotential "hard" enough. Try rvloc=2.0 bohr in your case... and 
"play" with this value to test its influence on physical results.

In any case, you have to test both schemes and select the one which 
gives better physical results and/or efficiency (size of plane-waves 
basis in Abinit).

Ultrasoft: simply add "ultrasoft" on the "3 0." line
Bessel: replace or add "bessel" on the "3 0." line

I enclose the input file for ultrasoft scheme.

Good luck.

I'm interessting by having some "feedback" from you: physical results of 
your PAW dataset, cut-off value for 10-3 or 10-4 mHa on energy , ....
If you PAW dataset is "good" enough, I could put it in Abinit table...


Marc Torrent



Pio Baettig a écrit :
> Dear Professor Torrent,
>
> thank you very much for your help. I managed to get good fits for the
> logarithmic derivatives for l=0, l=1 and for the occupied 5d-band, and
> not perfect fits for the unoccupied d-band and for l=3.
> The wavefunctions are now all <15 (p for wfn3, l=1 with minimum at -13
> and maximum at about 10), so the PAW should be usable for a first test.
>
> I converted it with atompaw2abinit using the standard parameter ("no"
> for real space optimization) and tried to perform a ground state
> calculation for Bismuth.
>
> Unfortunately, abinit stops with the following error:
> "
>   newocc : BUG -
>    It was not possible to find Fermi energy in 60 bissections.
>    nelecthi=  3.00000000002107E+01, and nelectlo=  2.99999999998639E+01.
>    Action : contact ABINIT group.
>
> .Delivered    0 WARNINGs and   0 COMMENTs to log file.
>
>   leave_new : decision taken to exit ...
> "
>
> using the same input-file and the USPP-PAW from the abinit-website, the
> same calculation runs without problems.
>
> Thank you very much for any hints.
>
> Best regards,
>
> Pio Baettig
>
>
> Marc Torrent wrote:
> > Now, as you have suppressed the f- partial waves, you should:
> > - put lmax=2
> > - put the l used to compute the pseudopotential back to l=lmax+1=3
> >
> > In principle (except for particular computations... high pressures, 
> > excited states, and so on) you should not need 3 partial waves for d 
> > electrons. I suggest that you suppress  the third d- partial wave and, 
> > consequently that you move the d- reference enrgy to get good log. 
> > derivatives again (for ex., try E=1 or 2 Ry)
> >
> > Marc Torrent
Bi 83
LDA-PW scalarrelativistic loggrid 2001
6 6 5 4 0 0 
6 1 3   !6p, 3 electrons
0 0 0   !last line of occupation
c !1s  1  core
c !2s  2
c !3s  3
c !4s  4
c !5s  5
v !6s  6 valence
c !2p  7
c !3p  8
c !4p  9
c !5p 10
v !6p 11 valence
c !3d 12
c !4d 13
v !5d 14 valence
c !4f 15
2        !lmax spdf, f electrons present, so 3
2.6 2.6 2.6 2.6      ! rpaw rshape rvloc rcore
y
4.
n
y
3
n
y
2
n
vanderbilt
3 0. ultrasoft
2.6
2.6
2.6
2.6
2.6
2.6
0
vanderbilt