Achieving Smoother and Faster SCF convergence for Lanthanide Complexes with DFT

[Ankur Gupta]

Hi all,

I am trying to compute the single-point energies of some lanthanide complexes (containing around 100 to 120 atoms) at the def2-TZVP basis set. However, the deltaE values for each of the SCF steps are quite large (~100 to 1000 Hartree), indicating high instability in the SCF procedure. I have the converged energies at a smaller basis set (def2-SVP), and I am using the corresponding "movecs" as a starting guess using the "vectors input project" option for computing the energies at the def2-TZVP basis set. I have also attached a sample input with my post. 

In addition, I have observed that each SCF step is around 1000-2000 seconds more expensive than the previous one, making each subsequent SCF step much more demanding. I am not sure if that's the expected behavior of NWChem.

I would like to know if there is anything else I could do to achieve smoother and faster convergence of the energies. I would be happy to provide more information if needed.

Best,

Ankur

[Edoardo Aprà]

Could you post the complete input and output files?

[Ankur Gupta]

Yes, sure. I have attached one of the jobs. I have other molecules as well, where I have seen similar behavior.

[Edoardo Aprà]

When you have linear dependencies showing up in the output file, the projected guess is not likely to be very successful.

Instead, the input file modifications below should get you a faster convergence for the tzvp basis set.

basis large spherical
  * library def2-tzvp
end

ecp
  Ce library "def2-ECP"
  I library "def2-ECP"
end
set "ao basis" large

dft
  vectors input atomic output CeTCMe2I_alt_xtbopt_tzvp_thf.movecs
  direct
  mult 2
  grid fine
  xc b3lyp
  disp vdw 4
  smear
  convergence damp 25 ncydp 0 dampon 1d99 dampoff 1d-4
end

cosmo
  do_cosmo_smd true
  solvent thf
  minbem 3
  ificos 1
end

task dft ignore

[Ankur Gupta]

Thank you for suggesting the input file modifications. I will rerun the computations with the proposed changes. 

From the output file, it seemed that the S eigenvalue threshold is 1e-05, so I thought all the S eigenvalues below 1e-05 would be ignored.

Best,

Ankur