revPBE Pseudos

[aromero]

We are doing some testing on different pseudopotentials and I have found that revPBE give
some cases superior results than normal PBE but I do not find any pseudo or code to
create revPBE pseudos to be read from Abinit. Does anybody knows any code or place
to find those?

Thanks!

-aldo.

[pouillon]

Did you try APE?

http://www.tddft.org/programs/APE/

[torrent]

Yann,
I don't think APE is for PAW (NC-pseudopotentials only).

Aldo,
if you want to use rePBE with atompaw or uspp2abinit you have to hack these codes (replace Pbe routines by revPbe ones). But this probably needs some work.
I have begun to include libXC in atompaw but this work is not finished and it is (unfortunately) not on top of my stack...

Cheers,
Marc

[proffess]

Dear all,

Would it be correct, that in order to use any functional from libxc library (for example, AM05) one need to have a pseudopotential generated for this specific type of functional? Can I use PBE (PAW) pseudopotential and ixc value, that activates an AM05 functional? From some papers (unfortunately, where VASP was used), people use PBE datasets for PBEsol or AM05 functionals. Can we do the same in ABINIT?

Thanks,
P.

[mverstra]

Dear all,

Would it be correct, that in order to use any functional from libxc library (for example, AM05) one need to have a pseudopotential generated for this specific type of functional? Can I use PBE (PAW) pseudopotential and ixc value, that activates an AM05 functional? From some papers (unfortunately, where VASP was used), people use PBE datasets for PBEsol or AM05 functionals. Can we do the same in ABINIT?

Thanks,
P.

This should be a new thread - could someone detach it?

I agree with your critique: formally it is completely wrong and evil to mix XC types between pseudopotential and GS, or between GS and GW and so on. In some cases it's harmless, but as AM05 adds non-trivial things compared to PBE, I think it's not a good idea. To some extent you are counting on a cancellation of the xc contribution from core electrons with the valence, and if these have been included in the pseudopotential using a different functional, then the results will be wrong.

You can do the same in abinit: just set ixc by hand and it will override the pseudopotential input xc type.

Matthieu

[proffess]

Hi Mattiew,

Thanks for reply.

I indeed try to active AM05 functional from libxc while using PBE-PAW pseudopotential. ABINIT warns me about that:

pspatm: WARNING -
Pseudopotential file pspxc= 11,
not equal to input ixc= -120135.
These parameters must agree to get the same xc
in ABINIT code as in psp construction.
Action : check psp design or input file.
Assume experienced user. Execution will continue.
.....

but soon crashed with error message :

.....

iter Etot(hartree) deltaE(h) residm nres2 diffor maxfor

FFT (fine) grid used in SCF cycle:

getcut: wavevector= 0.0000 0.0000 0.0000 ngfft= 30 30 30
ecut(hartree)= 50.000 => boxcut(ratio)= 2.07678
scfcv : before setvtr, energies%e_hartree= 0.000000000000000E+000

ewald : nr and ng are 4 and 11
setvtr : istep,n1xccc,moved_rhor= 1 1 0

drivexc : ERROR -
At least one of the functionals is a GGA or a MGGA,
but not all the necessary arrays are present.
ixc=-120135 nvxcdgr= 0 ngr2= 1
-P-0000
-P-0000 leave_new : decision taken to exit ...

Did I forget something?

Thanks,
P.

[gonze]

Seems a bug. Indeed, the exchange AM05 is working properly, and also the correlation AM05,
see the test in tests/libxc/Input/t04.in , but they are not tested together in this automatic test ...
Strange ...