[gpaw-users] Problems with Zn and Cd setups?

[Marcin Dulak]

Hi,

On 05/08/12 21:10, Ask Hjorth Larsen wrote:
> Hi
>
> 2012/5/7 Jakob Schiøtz<schiotz at fysik.dtu.dk>:
>> Hi,
>>
>> I noticed an extreme dependence of the lattice constants of Zn alloys when I vary the lattice spacing, and the energy of a simple Zn3Mg alloy varies a factor two when I vary the grid spacing between h=0.1 and h=0.18 Å!
>>
>> I then looked at the PAW setup page on the web.  The bond length of Zn varies wildly with h, see the third graph here: https://wiki.fysik.dtu.dk/gpaw/setups/Zn.html  It does not look like it converges at all.
>>
>> I looked a bit around, and found the same problem for Cd.  It looks like something is very wrong for these two elements.  The other neighboring elements have almost flat lines in these plots.
>>
>> Best regards
>>
>> Jakob

the second case: bond length of Zn2 maybe can be explained by the fact 
that the elements in this row form very weak bonds.
When i run 
https://trac.fysik.dtu.dk/projects/gpaw/browser/trunk/doc/setups/make_setup_pages_data.py 
Zn, i get for example
$grep -E "1 Zn    [12]|Zero" Zn-dimer-0.220.txt
   1 Zn    1.8452    1.8452    1.8452
Zero Kelvin:     -0.517832
   1 Zn    1.8845    1.8845    1.8845
Zero Kelvin:     -0.516866
   1 Zn    1.9237    1.9237    1.9237
Zero Kelvin:     -0.516801
   1 Zn    1.9630    1.9630    1.9630
Zero Kelvin:     -0.511322
   1 Zn    2.0023    2.0023    2.0023
Zero Kelvin:     -0.510973
   1 Zn    2.0415    2.0415    2.0415
Zero Kelvin:     -0.512417
   1 Zn    2.0808    2.0808    2.0808
Zero Kelvin:     -0.508902

It's not the question of converging SCF better, it's already converged.
These numbers clearly cannot be used to make a reasonable fit of the 
bondlength,
and probably the energy eggbox be the source of them.
I remember seeing similar problems when calculating Hg2, or even Hg bulk 
structures.
However I have seen similar problems also in vasp for Hg2, and that 
cannot be explained by eggbox,
so it could be that part of the problem is in the setups.

By the way, a fitting outside of the range is also the source of 
slightly incorrect results
of some bondlengths at 
https://wiki.fysik.dtu.dk/gpaw/setups/molecule_tests.html

Best regards,

Marcin
> That is rather strange.  They (at least Cd) also seem to converge very
> slowly in SCF loops, which is surprising given that the DOS at the
> Fermi level is very low.  This weirdness goes away if the very short
> projector cutoffs are changed to something more normal (--radius=2.45
> for Cd).  So why would these elements need very short cutoffs?
>
> Regards
> Ask
>
> _______________________________________________
> gpaw-users mailing list
> gpaw-users at listserv.fysik.dtu.dk
> https://listserv.fysik.dtu.dk/mailman/listinfo/gpaw-users


-- 
***********************************

Marcin Dulak
Technical University of Denmark
Department of Physics
Building 307, Room 229
DK-2800 Kongens Lyngby
Denmark
Tel.: (+45) 4525 3157
Fax.: (+45) 4593 2399
email: Marcin.Dulak at fysik.dtu.dk

***********************************