[CPMD-list] PW91 implementation

[Juerg Hutter]

Hi
>
> I want to implement the PW91 (GGA) density functional into CPMD.
> I want to use Vanderbilt PW91 pseudo-potentials. I did not find
> a PBE pseudo-potentials for my atoms. So I thought that I can use
> PBE part of code in lsd_func.F as an example. The Perdew-Wang 1991
> GGA Correlation Functionals are the same in PBE and PW91. The PBE
> Exchange Functional and The Perdew-Wang 1991 GGA Exchange Functional
> are different in potential F(S) only.
>
> But I found that the description of potential F(S) in lsd_func.F is
> slightly different from the PBE functional described by John Perdew.
> In the PhysRevLett F(S)=1+R-R*(1+m*S^2/R)^-1
> But in code the first constant 1 is omitted: F(S)=R-R*(1+m*S^2/R)^-1

CPMD calculates LDA and GGA parts of functionals separately.
PBEX is defined as LDAX*(1 + ....) and calculated in CPMD as
LDAX(rho) + PBEX(rho,drho)
If you want to add a new functional you have only to program the second
part.


>
> Also it is not easy for me to understand how the partial derivatives
> V1X and V2X was obtained. I found in the code that V1X=dE/drho and
> V2X=1/grad(rho) * dE/grad(rho), but...
>
This is correct, using V1X and V2X you can calculate Vxc independently
of the details of the functional. See details in the paper by Bird and
White, I think PRB 1994/95.

regards

Juerg Hutter


> Any help is much appreciated!
>
> Sincerely yours,
>
> Alex Shepelyev
> _______________________________________________
> CPMD-list mailing list
> CPMD-list at cpmd.org
> http://www.cpmd.org/mailman/listinfo/cpmd-list
>