How to calculate accurate binding energy between atom and large system?

[lahaye]

Hi,

I want to know how to accurately calculate the binding energy between
a single atom and another larger system (e.g. a metal atom attached
to an 80 C-atoms nanotube/CNT).

As far as I understand, one should do following:

Code: Select all

Ebind = Etotal(Atom-CNT) - Etotal(Atom) - Etotal(CNT)

where the Etotal is the E0-value in the bottom line of OSZICAR, right?

In the vasp guide, in "What kind of technical errors do exist, overview",
there is mentioning of this:

Code: Select all

[...] In conclusion, to minimize errors one should use the same setting
for ENCUT, ENAUG, PREC, LREAL and ROPT throughout all calculations,
and these flags should be specified explicitly in the INCAR file. In addition
it is also preferable to use the same supercell for all calculations whenever
possible.

How strictly should I follow these guidelines?
For example, for the large Atom-CNT system, I should use LREAL=.TRUE.
and a big supercell.
When using the same settings for the single atom (LREAL=.TRUE. and
big supercell), vasp gives serious warnings, that this will lead to
inaccurate results; it advices me to use LREAL=.FALSE. instead.
Then what should I do? Or is "LREAL=Auto" the best option in
all these cases?

And more generally, I like to know what is a reliable procedure to
calculate accurately the binding energy between an atom and a
much larger substrate, such as a carbon nanotube.

Thank you,
Rob.

[tjf]

One approach (not necessarily the best) is to ensure that all you calculations are converged w.r.t all parameters. In this limit you don't need to rely on technical tricks to ensure cancellation of errors.

[lahaye]

Tjf,

Sorry for me asking a rather simplistic question, but could you
eliaborate a little more on what you mean by "all calculations are
converge with respect to all parameters".

Of course I let each calculation converges self-consistently, but
that's probably not what you mean.

So your answer rather confused me.

Thanks,
Rob.

[admin]

I agree with you, tjf.
I put a little more explicitely what to care about:
1) free atom: please do it according to the handsonI.pdf example given in the VASP workshop tutorials (http://cms.mpi.univie.ac.at/vasp-workshop)
(1 k-point, verify if you really find the free atom's electronic ground state)
check the energy difference of a LREAL=True and False run,
In any case, the difference between these two energies should be at least a factor 100 smaller than the adsorption energy effects you are interested in.
The warning is given automatically if the number of atoms in the cell is to small to justify a LREAL=True run.
2) the clean and the adsorbed nanotube should be calculated with consistent aparameter sets (LREAL, k-mesh, box size, PREC, energy cutoffs,....)

[lahaye]

Concerning the free atom, I'd like to verify following:

1) The calculated EATOM in OUTCAR is 729.07002678, for both LREAL=F
and LREAL=T. EATOM as given in POTCAR is 729.1171 eV
The energy difference is about 0.047 eV. Is that OK?

2) The electronic ground state (E0 in OSZICAR) becomes
E0 = -0.24656986 (LREAL=F)
E0 = -0.23995976 (LREAL=T)
This energy difference is about 0.00661 eV.

I expect the ultimate binding energy between this atom and the carbon nanotube
to be around 2 eV, which is much bigger than the energy difference in 2). So I guess
I'm doing alright then. Yes?

Thank you,
Rob.

[bandy]

About (1). I think one needs to consider KE error(RRKJ; http://cms.mpi.univie.ac.at/vasp/vasp/node223.html)?
If anybody does not need, that line needs to be deleted.
About (2), LREAL=T and LREAL=F are in different energy scales (not important here though). From your statement, I believe expected BE is ca. 2 eV, but you are getting 0.246 eV. The discrepancy is huge. or I might be wrong to understand the problem.

[lahaye]

KE error ? RRKJ ? I have no idea what you're talking about. I do not
generate my own PP files; I take them from the vasp data base in pot_PBE/.

And what do you mean by "LREAL=T and LREAL=F are in different energy
scales" ?

When calculating binding energies, one should calculate like this:

Code: Select all

BE = E0(A+B) - E0(A) - E0(B)

and "B" happens to be a single atom. Hence my previous message
was about how to properly calculate this 'reference' energy of a free
atom. The number "0.246 eV" refers to E0(B), whereas the expected
binding energy BE of 2 eV refers to the result of the subtraction of
the three energies above. So I guess you got confused here and
I hope this resolves the confusion.

Still I hope someone can tell whether I'm doing the right thing,
concerning my aforegoing message.

Thanks,
Rob.

[bandy]

RRKJ KE error is included in EATOM in OUTCAR.Thus EATOM in OUTCAR(sum of all atoms) is related as: Sum over all atoms{EATOM(POTCAR) + KE error}.

About LREAL=T and LREAL=F, I found it in kresse_vasp_phonon.pdf, slide # 31 [http://wolf.ifj.edu.pl/workshop/files/t ... phonon.pdf] where it mentioned clearly that calculations with and without real space evaluations should not be compared. Thus I will assume you have done calculations for all three species A+B, A, B with either LREAL=T or LREAL=F (No mixing).
There could be a very small factor: the electronic state of the atom and real ground state. Second EATOM in PAW_PBE is spin-unpolarized calculation. If your atom is magnetic(has odd number of spins) then you need to subtract the difference also from BE. Hope somebody will elaborate more as I am perfectly sure about all those and also can tell something about 2 eV value.

<span class='smallblacktext'>[ Edited Mon Aug 07 2006, 07:45AM ]</span>

[admin]

I would keep LREAL the same for all 3 contributions to the binding energy, though, even the difference in the free atoms energies seems small .

[tjf]

(For the record: I was referring to cutoff, k-points etc. Get well converged w.r.t. all that and you are working in the DFT limit rather than crossing your fingers and hoping for cancellation of technical errors.)