maxiter exceeded

[Zakhar Lyullin]

Dear all,
I am still trying to calculate interaction potential for various pairs of atoms, it is necessary for the calculation of scattering parameters. (My first question is here
 https://groups.google.com/g/nwchem-forum/c/4yx7eF6Msp0/m/Wu8FiAZgAQAJ)

I tried to calculate the interaction potential for He_Fe and encountered the error:
ccsd_energy_loc: maxiter exceeded 101

I found a recommendation on the NWChem forum to try increasing the maxiter parameter.
Now all calculations end correctly. The problem is in the data received, they do not meet expectations and instead of the type of potential, something wrong turns out.

What could be the reason for the phenomenon of such a jump in the received data? (For clarification, I attached a couple of figures. The first two graphs correspond to the results obtained using the attached input file for the Fe_He case. The graphs show discontinuities and unnatural behavior of the curves for the interaction potential.
The second pair of graphs corresponds to the Au_He case. The curves do not break, the values change smoothly and meet expectations.)

In the same answer on the forum, in which there was a recommendation to increase the maxiter parameter, it was written that perhaps the issue is in the input file. I can't figure out where the error may be hiding in the input file and how it can be improved. 

For your convenience, please find attached the input file used in calculations.
Looking to receive any suggestions how to solve it.


memory total 12000 mb

geometry
        Fe 0.00000        5.0        0.00000
        He 0.00000        -5.0      0.00000  
         symmetry cs
end

basis spherical
    Fe library aug-cc-pVDZ
    bqFe library Fe aug-cc-pVDZ
    He library aug-cc-pVDZ
    bqHe library He aug-cc-pVDZ
end

bsse
    mon metal 1
    mon gas 2
end

scf
  rohf
  singlet
end

tce
 freeze atomic
 ccsd(t)
 maxiter 5000
end
task tce


 first two graphs



second pair of graphs 
 

[Edoardo Aprà]

Are there linear dependencies showing up in some of the Fe outputfiles?

[Zakhar Lyullin]

Do you mean a linear solution? As far as I understand, yes, displayed.
For your convenience, I have attached one of the output files.

среда, 16 февраля 2022 г. в 04:05:22 UTC+3, Edoardo Aprà:

[Edoardo Aprà]

No, I am talking about basis set linear dependencies. This is likely to happen on the left side of your plots, i.e. at short bond lenghts.

You can check this with the following command

grep "linear dependencies" outputfile

My suggestion is to first try the calculate the curve with the cc-pv?z basis sets (i.e.g without the augmented functions) and see if that make any difference.

You can find  a related discussion in the following article

https://doi.org/10.1016/j.cplett.2016.12.047

[Zakhar Lyullin]

I have checked the files using the command you suggested. I didn't find any linear dependence in the output files.

Thank you for your suggestion! I read the article and listened to your suggestion.
I performed calculations using a non-extended cc-pV?Z basic set and the results still look bad. I attached a graph with new results (the prefix ver2 denotes the results of calculations using bases of the cc-pV?Z type, and without the prefix aug-cc-pV?Z).Actually, the change in the basis used did not produce strong changes.The results turned out to be so close that they are layered on top of each other almost everywhere.
I also attach an example of the output file.

Looking to receive any suggestions how to solve it.

среда, 16 февраля 2022 г. в 21:00:06 UTC+3, Edoardo Aprà:

[Edoardo Aprà]

What is on the Y axis of this plot?

[Zakhar Lyullin]

The Y-axis shows the difference between the energies at a specific point (for example, 2A) and the energy at the most distant point (in this case, at a distance between the particles of 15A). The energy is converted to eV.

I will try to describe the problem without using a graph, maybe there will be less confusion.
As a result of calculations, I get almost the same energy at distances that differ significantly from each other (for example, the energies coincide at 4, 5 and 6 A). And such results contradict the theory and other similar calculations.
Maybe there is something I can change in the input files to fix this error?

пятница, 18 февраля 2022 г. в 23:14:57 UTC+3, Edoardo Aprà:

[Edoardo Aprà]

Could you post the two pvdz output files next to the discontinuity, e.g. one to the left of x=7 and one to the right of x-7?

[Zakhar Lyullin]

Of course!
Only the energy gap often occurs not at one point, but in several in a row.
So in the attached files, the result at points 6.2 (Corrected energy = -1265.351455574872) and 6.7 (-1265.351455188471) is noticeably different from the results 6.4-6.6 (-1265.274536491128 ; -1265.274536757228 ; -1265.274536619677).
The result could not be obtained at a distance of 6.3 due to insufficient maxiter parameter in the input file. I have attached this result only for completeness.

понедельник, 21 февраля 2022 г. в 20:43:18 UTC+3, Edoardo Aprà:

[Edoardo Aprà]

Are you sure that the singlet closed shell state is the correct ground state? It seems very unlikely given what the ground state for iron atoms is. You should first tackle this problem before calculating the potential curve.

[Zakhar Lyullin]

You are absolutely right, thanks for bringing this to my attention!
I got the correct calculations for Fe when I changed the block bsse and scf:

bsse
    mon metal 1
    mult 9

    mon gas 2
end

scf
  rohf

  NOPEN 8
end

In addition, calculations no longer require a large maxiter parameter. I did the calculations using maxiter 500 (before that I used 5000).

tce
 freeze atomic
 ccsd(t)
 maxiter 500

end

вторник, 22 февраля 2022 г. в 21:17:10 UTC+3, Edoardo Aprà: