[gpaw-users] Generating a core-hole setup with generator2

[Marcin Dulak]

Hi,

On 03/19/2015 07:16 PM, Eric Hermes wrote:
> Hello,
>
> I am attempting to generate a setup for Pd with a core hole in the 3d 
> state using generator2. If I generate the setup with '-f PBE -w -s Pd 
> --core-hole 3d,1.0', several new additional bound states appear that 
> did not exist in the ground-state calculation (2 s-states, 2 p-states, 
> and 3 d-states). Here is an example output from this calculation:
> https://gist.github.com/ehermes/6c85184e16d43acb6992
>
> If I instead generate the setup by specifying the electron 
> configuration with '-C' (as [Ar] 3s2 3p6 3d9 4s2 4p6 4d10 5s1), I 
> receive no error and get what appears to be a working core-hole setup 
> for Pd. In addition, the total all-electron scalar-relativistic energy 
> difference between the ground state and core-hole Pd atom is quite 
> close to the experimental metallic Pd 3d binding energy (334.2 eV vs 
> ~335.2 eV). This indicates to me that this procedure works. However, I 
> am not using the "--core-hole" argument to do this. Is what I am doing 
> valid? Here's the output file from the calculation I have described:
> https://gist.github.com/ehermes/3aec73bb427bbc6efe7c
>
> Also, I understand that the setups used for the ground state Pd and 
> core-excited Pd should be of comparable quality, and that the best way 
> to ensure this is to generate both setups as similarly as possible. 
> However, the default setup parameters for Pd in generator2 do not seem 
> to produce a good quality setup. While the official setups produce 
> lattice constants (3.946 for 10-electron and 3.937 for 18-electron) 
> that are in fairly good agreement with an all-electron FP-LAPW 
> calculation (3.948 angstrom), the default Pd setup generated by 
> generator2 predicts a lattice constant that is more than 0.02 angstrom 
> too large (3.972). If I use the cutoffs from the official 10-electron 
> setup ([2.3, 2.5, 2.2]) instead of generator2's default cutoff (2.4), 
> the lattice constant is in much better agreement (3.958). As I am not 
> an expert in generating pseudopotentials, I wonder what the best 
> procedure here is, and how to best measure the quality of a setup. Any 
> advice would be appreciated.
i would not use the non-semicore datasets (that's how we call setups 
now) like Pd 10-electron anymore.
Most modern pseudopotentials (http://www.physics.rutgers.edu/gbrv/ 
http://fpmd.ucdavis.edu/qso/potentials/index.htm 
http://www.abinit.org/downloads/PAW2/OLD-VERSIONS/JTH-TABLE-0.1/index.html)
dropped the idea of non-semicore ones because they are more difficult to 
create.
We don't have a working 10-electron Pd in generator2 (yet?). The 
parameters for non-semicore datasets in generator2.py are just so there 
is something there - they are not meant to be used.
On the other hand we have an "optimized" version of 18-electron Pd:
python -c "from gpaw.atom.generator2 import main as g; g(['Pd', '-f', 
'PBE', '-sw', '-P', '4s,5s,4p,5p,4d,0.5d,F', '-r', '1.98,2.5,2.05', 
'-0', '5,1.69', '-pl', 'spdfg,-1.5:1.5:0.01'])"
Try starting from the above.

Best regards,

Marcin
>
> Thank you,
> Eric Hermes
>