Calculate the stable adsorption site of CO on Cu(111) with ATK.

[fangyongxinxi]

Dear Sir,
I want to get the stable adsorption site of CO on Cu(111) with ATK14.
the experimental result is the most stable site for CO on Cu(111) is the "top site"
However, my calculation shows the most stable site is the "fcc hollow"
(the input.py file is shown in the attachments)

I checked some papers, like:
1.
https://scirate.com/arxiv/cond-mat/0408394
CO adsorption on Cu(111) and Cu(001) surfaces: improving site preference in DFT calculations
2.
CO adsorption on the Cu(111) surface: A density functional study
http://www.sciencedirect.com/science/article/pii/S0039602806000045

they said the calculation result of the stable site is indeed the "fcc hollow" with tranditional exchange-correlation (exch-corr) function (eg. lda, pw91), as the attachment show (3.bmp).  However, this result is not agreed with experimental observation.

Two possible choice can make the calculation results right.
the first one is to use "not popular" exch-corr function, such as B3LYP (As 3.bmp shows),
the second one is use : "pw91+u, u=1.25"
However, atk14 seems not have the B3LYP. (ATK has blyp, but my test calculation shows it could not give the result like "top site is the stable site")

So, my question is:
1. could you give me some suggestion on how to choose the exch-corr for CO/Cu(111) ? I want to get the stable site is "top site".
I tested all the exch-corr with GGA, but the results are not agree with the experiment.

2. How to set "+U" in my calculation.
As the attachment show (2.bmp), if I want to set U=1.25eV (as the attachment 3.bmp show),
How to set the parameters in details?
could you give an example about the output file (generated by atk, with U=1.25, for CO/Cu(111)) for my system ?

Thanks.

[Umberto Martinez]

This is indeed an interesting phenomena that, as you report, is pushing the limits of these theories.
Again, as you show ATK compares well with other codes by using the same level of theory.
B3LYP is an hybrid functional and not implemented in ATK.

To answer to your questions:
1. there is no exact exch-correlation functional. you can indeed try some GGA+U one as you found.
Since the +U parameter is kind of arbitrary and you can pick whatever you want, you will most probably be able to find the correct absorption site.
Just a note about accuracy, since this is a delicate case, you may want to use a more complete DZP basis set.

2. you can add the U parameter to specific orbitals.
If you want to compare exactly with your reference you should find out to which orbitals the authors added the U correction.
Probably the d orbitals of Cu, but you have to check.
Once you know this you can reproduce the results with ATK.
See also this tutorial for more info: http://quantumwise.com/documents/tutorials/latest/NiO_LDA_U/index.html/
Note that you may not be able to compare the absolute number of 1.25 eV of your reference to the one you need to use in ATK.
This is because this value depends on the implementation of the Hubbard correction.
However, you can plot absorption energy versus U and you should be able to see the same trend.

[fangyongxinxi]

This is indeed an interesting phenomena that, as you report, is pushing the limits of these theories.
Again, as you show ATK compares well with other codes by using the same level of theory.
B3LYP is an hybrid functional and not implemented in ATK.

To answer to your questions:
1. there is no exact exch-correlation functional. you can indeed try some GGA+U one as you found.
Since the +U parameter is kind of arbitrary and you can pick whatever you want, you will most probably be able to find the correct absorption site.
Just a note about accuracy, since this is a delicate case, you may want to use a more complete DZP basis set.

2. you can add the U parameter to specific orbitals.
If you want to compare exactly with your reference you should find out to which orbitals the authors added the U correction.
Probably the d orbitals of Cu, but you have to check.
Once you know this you can reproduce the results with ATK.
See also this tutorial for more info: http://quantumwise.com/documents/tutorials/latest/NiO_LDA_U/index.html/
Note that you may not be able to compare the absolute number of 1.25 eV of your reference to the one you need to use in ATK.
This is because this value depends on the implementation of the Hubbard correction.
However, you can plot absorption energy versus U and you should be able to see the same trend.

thanks for your reply,
I will try as you said.

[zh]

The adsorption of CO molecule on metal surface has long story. You had better dig into the literature and choose proper method (including different exchange-correlation functional, Van der Waals interaction correction, or more advanced technique such as random phase approximation). 
1. arxiv:cond-mat/0408394v1
2. www.nanomat-master.eu/pdf/CO_ads.pdf
3. Nature Materials 9, 741-744(2010)

[fangyongxinxi]

The adsorption of CO molecule on metal surface has long story. You had better dig into the literature and choose proper method (including different exchange-correlation functional, Van der Waals interaction correction, or more advanced technique such as random phase approximation). 
1. arxiv:cond-mat/0408394v1
2. www.nanomat-master.eu/pdf/CO_ads.pdf
3. Nature Materials 9, 741-744(2010)

Thanks for your advice!