ABINIT Discussion Forums

Hello,


I have recently been trying to construct PAW datasets for Ti and O in
order to calculate the electric field gradient in rutile within LDA.
I have had great difficulty accomplishing this.

I first tried the Ti and O from the "recommended Abinit table" on the
website, however, these PAWs yield EFGs that are a factor of 2 too
large. Using Abinit-6.8.1 and Abinit-6.6.2, both give the same
results.

I am using the structure given in Petrilli et al. (PRB v57 p14690) and
I have been able to reproduce the LAPW results given in this paper
with my own LAPW calculations.

I have been using atompaw v3.0.1.2 to generate my PAW datasets. I
have tried numerous combinations of PAW input parameters:

* different paw, partial wave, core, and local potential radii
* different construction methods: e.g. rrkj, Vanderbilt, Bloechl, etc.
* different numbers of projectors:

for Ti: at least 2 s, 2 p, and 2 d projectors (with 3s and 3p in the valence)
for O: at least 2 s, 2 p, and 1 d projectors

* different forms of the local potential and shape function

I have also converged all relevant Abinit input variables in the EFG
calculation including: ecut, pawecutdg, pawntheta, pawnphi, bxctmindg, etc.

In my most recent set of tests I computed the EFGs for Ti and O as
function of a the Ti and O PAW radii. My tests show that the Ti and O
EFGs are very sensitive to the choice of the Ti PAW radii (the Ti and
O EFGs are much less sensitive to the O PAW radius, between 1.0 and
1.4 au yields good results). In fact the O EFG doesn't converge until
the Ti PAW radii have been reduced to ~1.4 au and it converges to a
value that is ~5% smaller than the LAPW and PAW values given by
Petrilli et al. (see above). At this radius, the converged ecut is
~55 Ha.

If you analyze the convergence of the individual EFG tensor
contributions for the Ti and O EFG as a function of the Ti PAW radii
you will see that for the Ti EFG, the "efg_el" and "efg_paw" terms are
changing only slightly as a function of the Ti PAW radii. For the O
EFG, the "efg_paw" is flat while the "efg_el" contribution is changing
rapidly until r=~1.4 au.

As an additional check, I have used the O and Ti PAWs to compute the EFGs
for Ti in Ti hcp metal and O in solid O2. For both tests I get good agreement
with my own LAPW calculations.

It seems to me that the most likely explanation is an input file
problem. Perhaps there is another PAW input variable that I am not
setting correctly?

Does anyone have any advice?

Thanks,

Eric J. Walter
Senior Research Scientist
Department of Physics
College of William and Mary

-------------------------------------------------------------------------------------------------
Abinit INPUT file:
-------------------------------------------------------------------------------------------------
ndtset 10

ecut: 10
ecut+ 5
pawecutdg: 20
pawecutdg+ 10
getwfk -1

nstep 200

npulayit 4
ixc 7
kptopt 1
ngkpt 4 4 6
toldfe 1.0d-8
iscf 17

natom 6
nband 24
ntypat 2
typat 2*1 4*2

znucl 22 8

acell 3*8.6795

rprim 1.0 0.0 0.0
0.0 1.0 0.0
0.0 0.0 0.6442

prtefg 2

xred
0.0 0.0 0.0
0.5 0.5 0.5
0.304800000 0.304800000 0.000000000
-0.304800000 -0.304800000 0.000000000
0.804800000 0.195200000 0.500000000
0.195200000 0.804800000 0.500000000

Hi Eric, I am looking into this. With my own Ti and O PAW sets I get good values for Ti but not so good for O. This is puzzling to me because with this O set in other contexts (silicates) I routinely get extremely good values. I will let you know what I find out.

Joe