Calculating the energy for an atom

Question on input files/tags, interpreting output, etc.

Please check whether the answer to your question is given in the VASP online manual or has been discussed in this forum previously!

Moderators: Global Moderator, Moderator

Locked
Message
Author
zongtan_fang1
Newbie
Newbie
Posts: 14
Joined: Mon Dec 02, 2019 3:28 am

Calculating the energy for an atom

#1 Post by zongtan_fang1 » Fri Jun 04, 2021 10:44 pm

Hi,

I am having a question for calculating the electronic energy of an atom.

From the VASP manual, which says "In the GGA most atoms are characterized by a symmetry broken solution....", do it mean the cubic box does not work for all atom calculations ?

My second question is the accuracy for calculation the bond dissociation energy (BDE) from an atom and a dimer.
For example of O2, I am having:
Ee
O -1.90514
O2 -9.86272

The the BDE is predicted to be 6.05 eV, which is a little bit far away from the expt value of 5.15 eV ?

Where is the difference from ? ZPE may contribute a little, maybe 0.1 eV.

From Gaussian calculations, the BDE from GGA DFT is ~ 5.25 eV depending on the functional and basis sets.
I don't mean to compare two packages, just curious, where is the difference from ?

Thanks.

Best,

Zongtang

martin.schlipf
Global Moderator
Global Moderator
Posts: 455
Joined: Fri Nov 08, 2019 7:18 am

Re: Calculating the energy for an atom

#2 Post by martin.schlipf » Mon Jun 07, 2021 7:34 am

While it is hard to comment on the specific values you obtained without looking at the input and output files, I can provide some general guidance that may be helpful.

Regarding the symmetry of the system: You have to be aware that VASP will enforce the symmetry as it recognizes it. This is very beneficial for bulk systems as it can drastically reduce the computational effort. However it also means if you accidentally introduce a symmetry to the system that should not be there, you may obtain erroneous results. Specifically a single atom in a cubic box has cubic symmetry, which means all the p orbitals will be degenerate. The implications of this are studied in this tutorial. Please have a look at the first four steps in particular and compare them to your setup.

Regarding the dissociation energy: At convergence both Gaussian and VASP should yield the same results. Please carefully check that your Gaussian calculation is converged and that you use the same xc functional in VASP. Then you need to consider the following things to converge your calculation of atom/molecules in VASP: (i) Make sure your results are converged with respect to the energy cutoff (ENCUT). (ii) Make sure your cell size is sufficiently large. Because we use periodic boundaries, your system will interact with its periodic replica, so you need to check that it is sufficiently far away. (iii) Related to this, you may actually want to do some dipole corrections to reduce the interaction between the replica. (iv) If your calculation is very expensive, you can also try to extrapolate to infinite convergence by fitting energy-cutoff and energy-volume curves to there large scale limit.

Finally regarding the comparison to experiment: Dissociation is a hard problem for DFT, because fundamentally it does not account for nonlocal effects such as a charge localizing on one of the constituents or van der Waals interactions. To account for these, you need to go beyond local DFT, see e.g. here. What method to use strongly depends on your particular setup (system of interest, required accuracy, computational resources, ...).

zongtan_fang1
Newbie
Newbie
Posts: 14
Joined: Mon Dec 02, 2019 3:28 am

Re: Calculating the energy for an atom

#3 Post by zongtan_fang1 » Mon Jun 07, 2021 3:38 pm

Thanks a lot and that is helpful.

Best,

Zongtang

alex
Hero Member
Hero Member
Posts: 567
Joined: Tue Nov 16, 2004 2:21 pm
License Nr.: 5-67
Location: Germany

Re: Calculating the energy for an atom

#4 Post by alex » Tue Jun 08, 2021 7:02 am

Hello Zongtang,

~6 eV for O2 BDE and PBE or PW91 is fine. It reflects the (huge) error for this functionals. However, Gaussian should deliever the same. You'll get closer to experiment with e.g. hybrid functionals, if this is what you are after.

Cheers,

alex

Locked