How to calculate the binding energy of gas with odd electron (NO2) on graphene?

Message

dinhloc1984 · #1 Post by **dinhloc1984** » Sat Sep 12, 2009 1:18 am

Dear all,
I am newbie of VASP. I try to calculate the binding energy of NO2 on graphene layer by LDA and GGA-PBE with spin polarization (ISPIN=2).

The binding energy is calculated by:
E(binding)= E(Graphene_NO2)-E(Graphene)-E(NO2)
All calculations is relaxed with the same parameter with Ecut_off, cell size,...

The problem is the binding energy is quite small compared to experiment and other paper (PRB 77, 125416, 2008; CPL 387, 2004, 271). My results is -0.16eV for LDA and -0.035 eV for GGA-PBE. The value in CPL (using VASP) is ~0.40 eV.

Could you please give me some reasons that make my results so strange?
I great appreciate any your help.

Sincerely,
Loc

Ps: This is my POSCAR, INCAR and KPOINTS file.

POSCAR
Graphene_NO2
1.00000000000000
8.519 0.000 0.00
0.000 7.376 0.000
0.000 0.000 15.00
24 1 2
cart
-0.00000 7.37700 4.80000
-0.00000 9.83600 4.80000
-0.00000 12.29500 4.80000
0.71000 6.14800 4.80000
0.71000 8.60700 4.80000
0.71000 11.06600 4.80000
2.13000 6.14800 4.80000
2.13000 8.60700 4.80000
2.13000 11.06600 4.80000
2.84000 7.37700 4.80000
2.84000 9.83600 4.80000
2.84000 12.29500 4.80000
4.26000 7.37700 4.80000
4.26000 9.83600 4.80000
4.26000 12.29500 4.80000
4.97000 6.14800 4.80000
4.97000 8.60700 4.80000
4.97000 11.06600 4.80000
6.39000 6.14800 4.80000
6.39000 8.60700 4.80000
6.39000 11.06600 4.80000
7.10000 7.37700 4.80000
7.10000 9.83600 4.80000
7.10000 12.29500 4.80000
4.97000 11.06600 8.20000
5.08700 9.97700 8.68400
5.01000 12.17700 8.64400

INCAR

SYSTEM = Graphene_NO2

# Control tags
ISTART = 0
LREAL = .TRUE.
LWAVE = .FALSE.
NELMDL = -12
PREC = Medium
ENCUT = 500
NWRITE = 2
AMIX = 0.2
BMIX = 0.0001
AMIX_MAG = 0.8
BMIX_MAG = 0.0001
LAECHG = .TRUE

# Magnetic parameters
ISPIN = 2
MAGMOM = 24*0 1*1 2*0

# Electronic relaxation
EDIFF = 1E-04
ALGO = Fast

# Ionic relaxation
IBRION = 2
NSW = 150
EDIFFG = -0.02
ISMEAR = 0

KPOINTS

Automatic mesh
0
Monkhorst Pack
4 4 1
0. 0. 0.

alex · #2 Post by **alex** » Mon Sep 14, 2009 2:42 pm

Loc,

it seems you are also rather new to DFT. So it helps reading some textbook on that.
DFT does not account for dispersion. So you are lacking this energy contribution. That's the difference to experiment.

Alex

dinhloc1984 · #3 Post by **dinhloc1984** » Mon Sep 14, 2009 5:16 pm

Dear Alex,

Thank you for your reply, Alex. I look forward the reply for a long time.

Actually, the papers I mention above used DFT to calculate binding energy. One of them used VASP code. Another used ABINIT code ( also plane wave basic set).

To make clear the problem i get, I also calculate for H2O and SO2. The value I get is quite reasonable and comparable to above papers.

I guess the problem may be spin state of NO2. I am quite new in this type of calculation.

Have any one give me some instructions for do this type calculation?
I appreciate any your help.

Sincerely,
Loc.

alex · #4 Post by **alex** » Tue Sep 15, 2009 7:39 am

Hello,

I missed the VASP value (0.4 eV). Sorry for that.
Suggestions: Start as simle as posssible:
- remove MAGMOM and use NUPDOWN = 1
- remove all the adjusted mixing parameters *MIX(_MAG) VASP's mixing is good enough except for real suckers (like f-states in rare earths)
- what does LAECHG do. I could not find it in the manual

But don't forget: You'll never get dispersion right with DFT.

cheers

alex

dinhloc1984 · #5 Post by **dinhloc1984** » Wed Sep 16, 2009 2:10 am

Dear Alex,

I try to follow your instructions but I can not get better result.

For the graphene_NO2 system, the spin constraint (NUPDOWN=1) make the energy of the system is higher. The energy of the NO2 have a little change but it is negligible. This make the binding energy even smaller than using MAGMOM case.

The option LAECHG is use for charge analysis using Bader code.

Do you have another opinion?

Sincerely,
Loc.

alex · #6 Post by **alex** » Wed Sep 16, 2009 12:45 pm

Please perform a
grep TIT POTCAR
and post your output, please.

Thx

alex

dinhloc1984 · #7 Post by **dinhloc1984** » Wed Sep 16, 2009 2:23 pm

THis is the output of grep TIT POTCAR:

TITEL = PAW C 31May2000
TITEL = PAW N 31May2000
TITEL = PAW O 31May2000

Best regards,
Loc.

alex · #8 Post by **alex** » Thu Sep 17, 2009 8:45 am

Hm, looks ok. I performed your input, which I slightly modified for speed, with PBE (yours looks like PW91) and I got -2.6 kJ/mol interaction energy. Which is sth I would expect since the NO2 is quite far away. So I'm rather confident that the cited number is wrong and we are right.
If you want to continue with this kind of studies you may wish to look into Stefan Grimme's work on DFT+D (=dispersion), a rather straight forward approach to account for van der Waals interaction.

Cheers

Alex

dinhloc1984 · #9 Post by **dinhloc1984** » Thu Sep 17, 2009 11:58 pm

Dear Alex,

I try to increase K-point sampling, energy cut-off and grip size. All of them give the similar value.

Do you think PP is important in this calculation? Should I try with ultra-soft PP?

Best,
Loc.

alex · #10 Post by **alex** » Fri Sep 18, 2009 7:28 am

Hi Loc,

you can try, but I don't think you'll get sth. completely different. PAWs have a conceptual advantage vs US-PPs, so I'd stick to PAWs.

Try to find out, if the published work contains some hints which shows that they made a mistake. But I'd assume, if there would be a hint, they'd recognised by their own.

You could try the following to boil down their error: leave out ENCUT in the INCAR of the graphene only calculation. You'll make an error due to different cutoff energies in the three calculations. This adds up to larger energies than 0.02 eV.

Word of advice: It's always good to compare with published data, but stop trying to figure out other people's error (here we apparently have such a situation). You'll probably learn a lot about how you can do things wrong, but this is not the objective, isn't it? Try to improve your calculation with e.g. accounting for dispersion and you'll get closer to experiment. Or check for other and more reliable published data on this subject.

Cheers

Alex