[Wien] La f-local orbital energy effects on spin polarized calculation

[Peter Blaha]

I'm not sure where it is documented that you should put a La-4f local orbital
at high energy, at least I do not agree for general purpose (only for specific
tasks where you need unoccupied states).

In any case, your input does NOT put a local orbital, but expands the La-f states
as APWs with a very high linearization energy (at least with 3.0 it means you do
NOT have any proper 4f basis functions near EF).
An LO would require that you repeat the same "l"-line twice (as for s and p in
your example below).

So most likely, the "3.0" calculation is worse; which you can also see from
the "variational principle", i.e. a better basis should give a lower E-tot
(-40468.398065 for 1.0 vs.    -40468.173848 for 3.0)

> I'm performing a series of calculations on a rhombohedral 12-atom unit
> cell LaCoO3 system (symmetry group:167 R-3c),  with a=10.351349 a.u.
> c=24.742106 a.u. (in hexagonal lattice parameters, as needed in the
> Wien code), and dx=0.0522 (0.25-dx, 0.25+dx, 0.75) as the input
> internal parameter for the O-atom.
> The purpose of my calculations is to find the optimal internal
> parameter (dx) for each magnetic configuration, non-magnetic(NM) and
> ferromagnetic(FM), at these specific external structural parameters.
> 
> As I said, the calculations were done for the NM (min -j "run_lapw -I
> -cc 0.00001 -ec 0.00001 -fc 0.10 -i 60") and FM (min -j "runsp_lapw
> -cc 0.00001 -ec 0.00001 -fc 0.10 -i 60") configurations with the
> PBE-GGA xc-functional.
> As documented in the Wien-mailinglist and the UG, it is recommended to
> include a local orbital for the La-f electrons at quite "high energy"
> (of the order of 1-3 Ry), to improve the description of unoccupied
> states and reduce linearization errors.
> I have tried 2 sets of calculations:
> 1) NM and FM with La f-local orbital at 1Ry (fixed)
>  0.30    6  0      (GLOBAL E-PARAMETER WITH n OTHER CHOICES, global APW/LAPW)
>  0   -2.56      0.010 CONT 1
>  0    0.30      0.000 CONT 1
>  1   -1.30      0.010 CONT 1
>  1    0.30      0.000 CONT 1
>  2    0.30      0.010 CONT 1
>  3    1.00      0.000 CONT 1  <==== La f-local orbital
> 2) NM and FM with La f-local orbital at 3Ry (fixed)
>  0.30    6  0      (GLOBAL E-PARAMETER WITH n OTHER CHOICES, global APW/LAPW)
>  0   -2.56      0.010 CONT 1
>  0    0.30      0.000 CONT 1
>  1   -1.30      0.010 CONT 1
>  1    0.30      0.000 CONT 1
>  2    0.30      0.010 CONT 1
>  3    3.00      0.000 CONT 1  <==== La f-local orbital
> all with the same structural and numerical parameters (given at the
> end of the message).
> 
> The results of the internal optimization  for each set of calculations
> (energy of f-local orbital) are the following (0.25-dx):
>               1 Ry                  3 Ry
> NM     0.18701389       0.20843091
> FM     0.19283528       0.21605310
> 
> and the energy results, which are quite interesting (units: Ry):
>                  1 Ry                     3 Ry
> NM     -40468.398065      -40468.173848
> FM     -40468.396233      -40468.180753
> Diff            -0.001832              0.006905
> 
> where the energy corresponds to the case once the optimization was
> done. Diff is defined as "E(NM)-E(FM)". Then, if I put the f-local
> orbital at 1 Ry, the NM is slightly more stable than the FM (~ 1.8
> mRy), but if I put the f-local orbital at higher energy, then the FM
> is the most stable one (by ~ 6.9 mRy).
> I was quite surprised by these results, since in principle this
> unoccupied state (or its position in energy) should not affect the
> overall description of the system, specially on the optimization
> (force optimization) of the Oxygen internal position.
> 
> So my main question is how should I realize which one is the "correct"
> one? and why the position in energy of the f-local orbital affects the
> total energy, in such a way that ordering or the magnetic phases are
> changed? Is there any criteria or energy range for choosing the
> f-local orbital position for La that maybe I'm not taking into account
> or misplaced?
> I've already checked the common problems in the calculations (ghost
> bands, high QTL values, spheres overlapping, leaking core charge,
> oscillation in energy/charge/forces, etc, etc,) but both sets of
> calculations seems reasonably good. Any comment to this issue will be
> highly appreciated.
> 
> The used numerical parameters were the following:
> Rmt(La)=2.40
> Rmt(Co)=1.90
> Rmt(O)=1.65
> RmtxKmax=7.0
> Lmax=6
> GMAX=16
> kpoints=14x14x14
> mixing=0.1 (case.inm "0.10  1.00      PW and CLM-scaling factors")
> force convergence= 0.1 (case.inM:  "PORT 0.1")
> energy cutoff=-7.7 Ry
> and no smearing was used (case.in2 "TETRA    0.000")
> 
> Regards,
> 
> Omar De la Peña Seaman
> 

-- 

                                       P.Blaha
--------------------------------------------------------------------------
Peter BLAHA, Inst.f. Materials Chemistry, TU Vienna, A-1060 Vienna
Phone: +43-1-58801-15671             FAX: +43-1-58801-15698
Email: blaha at theochem.tuwien.ac.at    WWW: http://info.tuwien.ac.at/theochem/
--------------------------------------------------------------------------