Atomization energy of O2 [SOLVED]

[marcindulak]

Hi,

i'm calculating atomization energy of O2 on G2_1 geometry,
PBE, GGA FHI pseudopotentials, abinit 5.4.4, and getting strange results.
The total energy (Total energy(eV)) of O: -4.302379E+02 and O2: -8.663260E+02
what gives atomization energy of 5.850 eV.
The reference value should be around 6.2 eV:
(see Table II from http://chem.hamilton.edu/~gshields/publ ... MK2005.pdf).
I have verified "Magnetisation (Bohr magneton)= 2.00000000E+00",
and that there are no fractional occupancies.
I use nband 20, toldfe 1e-08, tsmear 0.001 eV, ecut 800 eV,
and a very large cell ~(18 A)**3 with broken symmetry for the atom.

Best regards,

Marcin

Code: Select all

diemac 1.0
occopt 3 # Fermi-Dirac smearing
spinat
0.00000000000000 0.00000000000000 2.00000000000000
charge 0
ecut 800 eV
npulayit 5
diemix 0.1
natom 1
ixc 11
nsppol 2
tsmear 0.001 eV
nstep 950
nband 20
#Definition of the unit cell
acell
1.00000000000000 1.00000000000000 1.00000000000000 Angstrom
rprim
18.00000000000000 0.00000000000000 0.00000000000000
0.00000000000000 18.01000000000000 0.00000000000000
0.00000000000000 0.00000000000000 18.02000000000000
chkprim 0 # Allow non-primitive cells
#Definition of the atom types
ntypat 1
znucl 8
#Enumerate different atomic species
typat
 1
#Definition of the atoms
xangst
9.00000000000000 9.00500000000000 9.01000000000000
kptopt 1
ngkpt 1 1 1
#Definition of the SCF procedure
toldfe 1e-08
chkexit 1 # abinit.exit file in the running directory terminates after the current SCF

Code: Select all

diemac 1.0
occopt 3 # Fermi-Dirac smearing
spinat
0.00000000000000 0.00000000000000 1.00000000000000
0.00000000000000 0.00000000000000 1.00000000000000
charge 0
ecut 800 eV
npulayit 5
diemix 0.1
natom 2
ixc 11
nsppol 2
tsmear 0.001 eV
nstep 950
nband 20
#Definition of the unit cell
acell
1.00000000000000 1.00000000000000 1.00000000000000 Angstrom
rprim
18.00000000000000 0.00000000000000 0.00000000000000
0.00000000000000 18.00000000000000 0.00000000000000
0.00000000000000 0.00000000000000 19.24595600000000
chkprim 0 # Allow non-primitive cells
#Definition of the atom types
ntypat 1
znucl 8
#Enumerate different atomic species
typat
 1 1
#Definition of the atoms
xangst
9.00000000000000 9.00000000000000 10.24595600000000
9.00000000000000 9.00000000000000 9.00000000000000
kptopt 1
ngkpt 1 1 1
#Definition of the SCF procedure
toldfe 1e-08
chkexit 1 # abinit.exit file in the running directory terminates after the current SCF

[Boris]

Hi

5.9 eV is actually pretty good, compared to the experimental 6.2 eV.

The largest part of your error comes from the PBE functional that fails to describe the O2 molecule [J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev. Lett. 77, 3865 (1996)].

Regards

[marcindulak]

I'm sorry, but i'm comparing PBE from abinit to PBE from vasp, gaussian and Wien (they all 3 agree).
For me this looks like a flaw in FHI pseudopotential, unless i'm doing something wrong in the abinit inputs.

[Boris]

So you're saying that abinit is not giving the same result as the three others?

It's not surprising that abinit has the least accurate result since you're using pseudopotential. They are less accurate than the PAW method that you're probably using with vasp. As for gaussian, it's a localized orbital code, so it's supposed to be more efficient for molecule and atom calculations. Finally, it seems to me Wien2k is an all electron code, so it should yield the most accurate result.

If you use the PAW method as implemented in abinit (together with a proper PAW dataset), you should obtain the same result as vasp.

Regards

[marcindulak]

Thanks, yes abinit GGA FHI pseudopotentials give me a different PBE result than the other codes I mention.
I know there there is a possibility of getting nonsense with pseudopotentials
(with PAW potentials too, but we all believe it's smaller).
Still I would like to know, if i'm not doing something wrong in my inputs, so I can make a conclusions.
By the way i get a similarly bad results with HGH pseudopotentials
from http://cvs.berlios.de/cgi-bin/viewvc.cg ... binit/pbe/
and other two HGHs: http://www.abinit.org/downloads/psp-links/psp-links/gga
and http://www.abinit.org/downloads/psp-links/psp-links/hgh
For all HGHs i used ecut 1500. The last HGHs are in fact LDA's but the result
is very close to the HGH PBE's.

[marcindulak]

Hi,

i have tried the following PAW:
http://www.abinit.org/downloads/PAW2/TA ... -hard-uspp
with result of 5.93 eV (still far from 6.2 eV).
ecut 1000 eV, pawecutdg 2000 eV.
I see many "Density went < 0 at".
Seems anything wrong in the above inputs?

[marcindulak]

Anobody?

[leon]

I'm not sure, but may be it worth to try opposite spin directions instead collinear?
like:

Code: Select all

spinat
0.00000000000000 0.00000000000000 1.00000000000000
0.00000000000000 0.00000000000000 -1.00000000000000

[marcindulak]

I'm not sure, but may be it worth to try opposite spin directions instead collinear?
like:

Code: Select all

spinat
0.00000000000000 0.00000000000000 1.00000000000000
0.00000000000000 0.00000000000000 -1.00000000000000

This is incorrect, it has to be triplet state.

The problem indeed is in the pseudopotentials/PAW used.
The latest potentials from
http://arxiv.org/pdf/1309.7274.pdf
give correct PBE atomization energy of O2 (+- 0.05 eV).