[Wien] Strange discontinuity for GGA calculations

[Peter Blaha]

A better FFT-mesh is necessary for all GGAs (and even more for meta-GGAs)
AND highest precision calculations of lattice parameters (because the stupid code
calculates the gradients/laplacian numerically (and not analytically from the Fourier expansion).

For such high-precision results you have to consider:

large RMTs (which are computationally efficient) introduce some linearization error
as well as some cutoffs in LM series-expansion due to large non-sphericity near RMT.
In part one may improve by "additional LOs", but this requires careful considerations
and some deeper understanding.

small RMT are computationally "expensive", but do not suffer from linearization
errors. As long as all core states are still confined in the sphere, the basis
should be virtually "exact" inside the spheres. However, due to the large
interstitial with rapidly varying wavefunctions/densities close to the atomic
spheres one needs large RKMAX and also a "good" FFT mesh for XC.

I cannot make a more general statement, but you have to test it for every case
(or make sure you get results close to Haas et.al, PRB 79, 085104 and  PRB 79, 209902(E))


Am 07.09.2010 22:02, schrieb Natalie Holzwarth:
> Dear Peter,
>       Thanks so much for solving this problem.   Now I am wondering
> how many calculations I will need to redo to make sure that everything is
> well-converged.  For this example of fcc Sc, I added the following lines
> to Sc.in0:
> NR2V   FFTopt IFFT
>    45   45   45 5.000  1
> Perhaps it would have worked as well to replace the last line by
>    -1  -1 -1  5.000   1              ?
> Do you generally add similar lines to all of your calculations or only
> those using PBE?    In your experience is IFFTfact=5.0 always necessary or
> is Sc particularly problematic?  Or is it because we have chosen to use
> such small values of RMT?   I would have thought that as long as Gmax is
> large enough, and the linearization energies are good enough, the
> calculations should be insensitive to RMT.   Normally this seems to be the
> case.   It would be nice if there were a universal recipe for always
> getting converged results...
>      THanks again for all of your advice,
>                    Natalie
>
>
>
> On Tue, 7 Sep 2010, Peter Blaha wrote:
>
>> Sorry, but I cannot reproduce your results.
>>
>> With good parameters I get nice and smooth curves.
>> The lattice parameter (PBE) is near 8.73, which is smaller
>> than in your plots.
>> The RMT dependency is quite small, provided you use good parameters:
>>
>> RKmax=10 (9 is probably almost ok)
>> NUMK: I used 11000 (22x22x22) points in full-BZ.(16x16x16 leads to some noise)
>> Most important for GGA calculations: You have to use a VERY good FFT mesh for
>> the xc-energy, in particular for smaller spheres. I used IFFT 5 (instead of
>> the default 2) in case.in0
>>
>> I attach some plots.
>>
>>
>> Am 07.09.2010 04:58, schrieb Natalie Holzwarth:
>>> Dear Wien2k mailing list,
>>>        For many years, we have been using the wien2k code as a standard for
>>> comparing our PAW calculations.   Recently, we noticed some difficulties
>>> particularly with GGA calculations.  The attached example is for Sc in an
>>> fcc structure using what we think are well converged parameters
>>> (RMT*KMAX=9, GMAX=14, 16x16x16 kpoints).  We calculate the total energy
>>> versus lattice constant both for lda and gga varying RMT to make sure that
>>> there is no sensitivity to the linear LAPW/APW parameters.  RMT = 2.0,
>>> 1.6, and 1.5 for lda/gga, lda1/gga1, and lda2/gga2, respectively.  The
>>> binding energy curves are identical for the 3 lda calculations.  But for
>>> the gga calculations, there seems to be a very strange discontinuity at
>>> the lattice constant of 9.0 which is very sensitive to the choice of RMT.
>>> The Sc.in1 file that we used for gga is attached and was obtained from
>>> running lstart.  Of course I will be very embarrassed if there is a typo
>>> in our files, but we have not been able to discover it for several days.
>>> Any advice will be very much appreciated.    Thanks, kindly,
>>>                                   Natalie Holzwarth
>>>
>>> N. A. W. Holzwarth			email: natalie at wfu.edu
>>> Department of Physics			www: http://www.wfu.edu/~natalie
>>> Wake Forest University		        voice: 336-758-5510
>>> Winston-Salem, NC 27109-7507            fax: 336-758-6142
>>> U. S. A.
>>>
>>>
>>>
>>> _______________________________________________
>>> Wien mailing list
>>> Wien at zeus.theochem.tuwien.ac.at
>>> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
>>
>> --
>>
>>                                         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/
>> --------------------------------------------------------------------------
>>
>
> N. A. W. Holzwarth			email: natalie at wfu.edu
> Department of Physics			www: http://www.wfu.edu/~natalie
> Wake Forest University		        voice: 336-758-5510
> Winston-Salem, NC 27109-7507            fax: 336-758-6142
> U. S. A.
> _______________________________________________
> Wien mailing list
> Wien at zeus.theochem.tuwien.ac.at
> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien

-- 

                                       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/
--------------------------------------------------------------------------