CO on Cu using DFT+U

[kpkuhl]

Hi,

I've would like to calculate CO adsorbtion energy to Cu using DFT+U as they did in the paper M. Gajdosˇ, J. Hafner / Surface Science 590 (2005) 117–126 in order to get the correct site preference and get close to experimentally known values. However, I'm having some trouble reproducing their work. I'm using vasp 4.6. I get an incorrect site preference and much lower HOMO-LUMO gap when I use U=1.25 (the authors report setting U to 1.25 eV). To get close to the HOMO-LUMO gap that they report I need to do U=5. Are the units of U eV, so that this is a unit problem? Or are there some othere parameters I'm missing? I'm using an 8x8x8 kpoints grid, pw91 POTCARS, and POSCAR and INCAR as follows:

INCAR:

general:
SYSTEM = CO
ISTART = 0
ICHARG = 2
PREC = High
LREAL = FALSE
IDIPOL=3

Electronic Relaxation:
ISMEAR = 2
SIGMA = 0.1
ENCUT = 450
GGA=91

Ionic Relaxation:
IBRION = 2
NSW = 100
LDAU=TRUE
LDAUTYPE=2
LDAUL=1 1
LDAUU=1.25 1.25
LDAJJ=0 0
VOSKOWN=1

POSCAR:

CO
1.00
10.0 0.0 0.0
0.0 10.0 0.0
0.0 0.0 10.0
1 1
Direct
0 0 .444
0 0 .556


If I could get any advice on how to better reproduce this paper, that would be great. Thanks.

Kendra

[alex]

They report a gap for a metal? Are you sure? Even if it is a metal surface, it stays metallic (no gap).

Do you speak about CO only eventually?

Alex

[kpkuhl]

Sorry to be unclear above; I'm referring to the HOMO-LUMO gap in the CO molecule and the problems with site preference are calculating CO adsorbed on the metal surface. Since getting the correct site preference depends on getting the HOMO-LUMO gap of CO correct, I've been trying to get that part of the calculation to work.

[admin]

please have a look at Stroppa, Termentzidis, Paier, Kresse and Hafner, PRB 76, 195440 (2007): you need PBE0 or HSE (screended exact exchange contributions) potentials (as included in vasp.5) to obtain correct site preferences of CO on Cu.
the DFT+U method restricted to certain MOs is not included in the standard implementations of vasp.

[kpkuhl]

Right now I am interested in restricting the simulations to a purely DFT+U formalisms and want to avoid the higher level exchange-correlation functionals that are in vasp5. That said, I am confused by what you mean when you say that the current implementation of DFT+U (vasp 4.6) is restricted to certain MOs. This is definitely not explained well in the vasp manual where it clearly states that one can specify LDAUL to account for different MOs for different atoms. Is this not the case? Also, the units are not given for the U parameter. I assumed they were in eV, but find that I need to include a U value that is approximately 4 times larger than the one given in the Gajdos and Hafner 2005 paper (regardless of whether I ran with or without spin). Please advise on how to properly use DFT+U. Thanks, Kendra