Details about PyCDT original paper

[Sonu Singh]

Dear PyCDT team,

I am trying to replicate some of the original data shown in your paper https://doi.org/10.1016/j.cpc.2018.01.004.

In particular I would like to know the size of the supercells used to generate the data presented in SI2, in the table where you present the Freysoldt and Kumagai corrections. If you can give me this information for CdSe and GaAs then that will be enough to begin with.

Thanks a lot.

Sonu

[Bharat Medasani]

If I remember correctly, we used 64 atom unit cells. 

@Danny,

Since you ran the calculations, is that number correct?

Bharat

[Danny Broberg]

Late to the party with my response, but better late than never.

I believe the SI2 data all had a maximum cell size of 216 atoms - for CdSe and GaAs, I believe their supercell sizes were 216 atoms. Note that reproducing the data will depend on the POTCARs and relaxation settings you use, but hopefully you get the same ball park with basic settings.

[Steven Hartman]

I was just looking into this - from Figure 4 in the main text, it looks like the lattice parameter for GaAs is ~17 A.  Would that be the 128-atom cell with the 16.26 A lattice parameter?

[Danny Broberg]

Thanks - Yes, that seems to make sense. I suppose it would have been a 4x4x4 supercell of the primitive unit cell.

[Nigel Hew]

Hello, I'm also trying to replicate the results in the Charge Correction Data in Supplementary Information 2. Did you guys use PBE and not PBE_52 or PBE_54. Also I am having an issue where the --nmax keyword does not seem to be working for the command pycdt generate_input. The default appears to be 64 atoms, contrary to what is said in the paper which is 128 atoms. I have tried to use 'pycdt generate_input --mpid mp-2534 --nmax 128’, but it still produces supercells with only 64 atoms. Any response would be appreciated.

Regards,

Nigel 

[Nigel Hew]

I think I get it now. --nmax always rounds down to the nearest cube (nxnxn) where n is an integer >= 1. So the results from Fig. 4 must be from a 216-atom (3x3x3) supercell not a 4x4x4 primitive supercell. Please correct me if I am wrong. 

Using a 216-atom supercell for GaAs, I get similar results for Fig. 4a and Fig. 4b. The results I got compared to Fig. 4b appear to be a bit different, for example, there is only Vq/b not Vq/b for both As and Ga. If someone can tell me why that is that would be great. Also, it matches in charge correction data in SI2 fairly well. 

I've attached some of my plots below. The formation energies in Fig. 5 however appear to match my results more for the 64-atom (2x2x2) supercell rather than the 216-atom (3x3x3) supercell. 

[PyCDT forum]

On Tuesday, April 12, 2022 at 11:53:58 PM UTC-4 nigel...@gmail.com wrote:

I think I get it now. --nmax always rounds down to the nearest cube (nxnxn) where n is an integer >= 1.

It is not nearest cube (nxnxn), but nearest cube(lxmxn), where l, m, and n could be different. For the cases where lattice vectors are not identical, the code tries to generate a supercell that is nearest to a cube.

 

So the results from Fig. 4 must be from a 216-atom (3x3x3) supercell not a 4x4x4 primitive supercell. Please correct me if I am wrong. 

Using a 216-atom supercell for GaAs, I get similar results for Fig. 4a and Fig. 4b. The results I got compared to Fig. 4b appear to be a bit different, for example, there is only Vq/b not Vq/b for both As and Ga. If someone can tell me why that is that would be great. Also, it matches in charge correction data in SI2 fairly well. 

There are few differences with the plots you shared when compared to Figs. 4 in the paper. The paper shows results for Ga_As, where as you are showing data for As_Ga. Customized scripts were used to generate Fig 4, so Vq/b for  both As and Ga could be plotted.

[Nigel Hew]

Thanks for the clarification.