USPP, PW91 and PBE

[monty-wang]

Dear Admit and All
Thank you very much for you kind help.
I recently have to use USPP to calculate a large system. However there is only uspp-gga-91 POTCAR but not uspp-gga-pbe files? Can GGA=PE in INCAR be set in order to use GGA-PBE XC? Because I heart that uspp can not be with PBE. I tried GGA=PE and worked, but the results are a little different from PW91.
Any response is appreciated.

[martin.schlipf]

We recommend to use the PAW potentials even for large systems. Which version of VASP are you using?

[alex]

Dear monty-wang,

PAW potentials provide higher accuracy at similar cost compared to USPP. That's why they are so successful and widely used. I'd strongly recommend to use them instead of USPPs.
If you want to learn about the differencies I'd like to point you to Georg's workshop contribution https://www.vasp.at/vasp-workshop/pseudopp2.pdf

Cheers,

alex

[monty-wang]

Dear Martin, Alex and All
Thank you very much. I am engaged in a large system with almost one thousand atoms and our supercomputer is powerfully limited . I know the accuracy of PAW and there are still some cases in which USPP is used.
So may I use the GGA=PE to set PBE with USPP?
Any response is appreciated.

[martin.schlipf]

Well there are always some deficiencies when you change the xc potential. For PAW the system is flexible enough to deal with it so that it essentially amounts to just using a different core potential. I wouldn't recommend the same for USPP, though. Even for PAWs it is recommended that you use a PAW close to the desired functional, i.e. for a PBE0 calculation you would use the PBE PAW and not the LDA one.
To be completely honest, if your system size is such that being able to execute it depends on using USPPs over PAWs, it will be difficult to get reasonable results in either case. How would you study the convergence with respect to the energy cutoff ENCUT, which has typically a much larger computational cost associated with it?

[monty-wang]

We recommend to use the PAW potentials even for large systems. Which version of VASP are you using?

Dear Prof. Martin
I use vasp5.3.5
Thank you very much

[martin.schlipf]

Well your license also allows you to use VASP 5.4.4 and the PAW 54 potentials, so I'd recommend upgrading to those. Contact your PI who should have access to them.

[monty-wang]

Dear Prof. Martin
Thank you very much. We have checked the cutoff energy and it goes well. We will use USPP then.
Thank you very much.

[monty-wang]

Dear Prof. Martin, Prof. Alex and All
Thank you very much for your kind help. We ran into problems.
When using PAW to calculate the magnetic system (bulk bcc Fe as an example), turn on the spin polarization calculation ISPIN = 2, set LORBIT = 11 and the appropriate MAGMOM, the calculated single atom magnetic moment appears at the tail of OUTCAR, after the calculation is completed.
But when USPP is used to calculate the magnetic system, the same parameter settings, and finally the single-atom magnetic moment information at the tail of OUTCAR is all displayed as 0.
We have learned that LORBIT = 11 is not applicable to USPP, so we tested different LORBIT and RWIGS, and still did not get good results.
Any response is appreciated.

[henrique_miranda]

The USPPs are not updated in a long time and we don't provide support for them.
Martin has suggested that you use the PAW 54 pseudopotentials.