ionic relaxation: GGA vs. LDA

Message

lahaye · #1 Post by **lahaye** » Sun Sep 10, 2006 7:57 am

Hi,

I have two questions:

1) I am told that for ionic relaxation LDA is better than GGA.
Is that so?

2) How do I switch from GGA to LDA?
I usually use the potentials that are in the pot_PBE/ directory.
I assume that this induces GGA, right? Should I instead use the
potentials in pot/ directory for doing LDA?

Thanks,
Rob.

bandy · #2 Post by **bandy** » Mon Sep 11, 2006 4:41 am

You are using PBE/PE potential. PBE is a special case(simplified) of PW91. The possible GGA tags are: PW/PB/LM/91/PE/RP. That means you can use say RP functional by defining GGA=RP in INCAR. GGA potential are LDA + GGA corrections. The LDA part is parametrized by Perdew Zunger method, not Perdew's pade approximation (mentioned in
http://cms.mpi.univie.ac.at/vasp-worksh ... tabase.pdf ). You can switch to LDA by keeping GGA= (blank) in INCAR as well as in POTCAR and in POTCAR LEXCH=CA; these potential dataset are provided in different directory. for paw PPs, there are 3 XC types available by default, LDA, PW91 and
PBE. only RP has to be specified via INCAR (GGA=RP).
You can switch between ULTRA/PAW by LPAW/LULTRA in POTCAR.

The general statistics about the comparison of performance of LDA(LSDA)/GGA(GGS):

(1) Cohesive Energy:
GGA: 4% larger (~0.3 eV)
LDA: 15% larger.

(2) Structure:
GGA: 1% larger
LDA: 1% shorter

(3) Energy Barrier:
GGA: 30% low
LDA: 100% low

(4) Ex
LDA: 5% larger

(5) Ec
LDA: 100% larger

The manifestations (LDA) are: it favors close-packed structure. Fe LDA is fcc (experimentally bcc ferromagnetic)
energy barriers too low (no chemical accuracy)
negative ions are unstable

lahaye · #3 Post by **lahaye** » Mon Sep 11, 2006 11:22 am

Bandy wrote:

Code: Select all

You can switch to LDA by keeping GGA= (blank) in INCAR as well as in POTCAR and in POTCAR LEXCH=CA;

I have done this:

$ grep GGA POTCAR INCAR
POTCAR: GGA =
INCAR: GGA =
$ grep LEXCH POTCAR INCAR
POTCAR: LEXCH = CA

However, I then get this warning (error?) message:

Code: Select all

----------------------------------------------------------------------------- 
|                                                                             |
|  ADVICE TO THIS USER RUNNING 'VASP/VAMP'   (HEAR YOUR MASTER'S VOICE ...):  |
|                                                                             |
|      You enforced a specific xc-type in the INCAR file,                     |
|      a different type was found on the POTCAR file                          |
|          I HOPE YOU KNOW, WHAT YOU ARE  DOING                               |
|                                                                             |
 -----------------------------------------------------------------------------

I find this quite scary, because I don't know exactly what I'm doing.
What is this warning about? Is it serious problem?
Or does one have to take these warnings for granted when
switching to LDA?

Rob.

bandy · #4 Post by **bandy** » Mon Sep 11, 2006 7:29 pm

Did you use LDA in specific directory in POTENTIAL database?

lahaye · #5 Post by **lahaye** » Tue Sep 12, 2006 5:19 am

Bandy wrote:

Code: Select all

Did you use LDA in specific directory in POTENTIAL database?

I have no idea what you are asking in this sentence.

Maybe I'm too innocent, but could you please rephrase it to make
it easier for me to understand.

By the way: I do not generate my own POTCAR files; I use the ones
provided by VASP (usually the ones from pot_PBE/ directory).

Thanks,
Rob.

bandy · #6 Post by **bandy** » Tue Sep 12, 2006 5:39 am

You should have LDA PP in a separate directory (PAW LDA) when you have downloaded PPs from vasp website
ftp://cms.mpi.univie.ac.at. I have PPs in 5 different directories:
paw_pbe, pot_GGA, pot_LDA, pawGGA, pawLDA

paw_pbe, pawGGA, pawLDA are PAW PP
pot_LDA, pot_GGA ultrasoft PP

bandy · #7 Post by **bandy** » Tue Sep 12, 2006 6:34 am

Hi Rob,

Let me tell you in detail. In fact the possible LDA tags (LEXCH) are: Slater-XC, HL (Hedin Lundquist [1971]), CA (Ceperly Alder parametrized by Perdew Zunger ), WI (Wigner perturbation), PB (Perdew Becke), PW (Perdew-Wang 86), LM (Langreth-Mehl-Hu), 91(Perdew-Wang 91). Similarly possible GGA tags are:
PW (Perdew-Wang 86), PB, LM, 91, PE, RP.
Also you can define LEXCH by numbers like LEXCH=0(local format), 1(CA; default), 2, 3, 4(HL Hedin Lundquist), 5, 6(Wigner), 7 (Pade approximation of Perdew) etc. similarly LEXCHV=0, 1, 2, 3, 8,9

Hope somebody who has tried these PPs can expalin well their usage as I am restricted to 1-2 PPs only.

#8 Post by **admin** » Wed Sep 13, 2006 12:55 pm

I would strongly suggest to choose the most appropriate pseudopotential type from the data base instead of setting the GGA flag in INCAR. Please have a look at the pseudoppdatabase.pdf file which you can download from http://cms.mpi.univie.ac.at/vasp-workshop in order to check which PP is most appropriate for the calculations you want to perform.

stolis · #9 Post by **stolis** » Wed Oct 30, 2013 10:24 pm

Hi,
I have made several calculations with POTCAR PAW LDA. In what way the results of my calculations would be different if i had used POTCAR PAW GGA? Is the lattice optimization of a structure calculated with PAW LDA different of the SAME strucrure calculated with PAW GGA?
Thanks